Merge branch 'tracking-gator' into merge-linux-linaro-core-tracking

218 files changed, 42730 insertions, 0 deletions

diff --git a/drivers/Kconfig b/drivers/Kconfig
index e451dbab0964..87a2552cafea 100644
--- a/drivers/Kconfig
+++ b/drivers/Kconfig
@@ -184,4 +184,6 @@ source "drivers/ras/Kconfig"
 
 source "drivers/thunderbolt/Kconfig"
 
+source "drivers/gator/Kconfig"
+
 endmenu
diff --git a/drivers/Makefile b/drivers/Makefile
index ba20c2a8bc06..48a5395ba400 100644
--- a/drivers/Makefile
+++ b/drivers/Makefile
@@ -162,3 +162,5 @@ obj-$(CONFIG_POWERCAP)		+= powercap/
 obj-$(CONFIG_MCB)		+= mcb/
 obj-$(CONFIG_RAS)		+= ras/
 obj-$(CONFIG_THUNDERBOLT)	+= thunderbolt/
+
+obj-$(CONFIG_GATOR)		+= gator/
diff --git a/drivers/gator/Kconfig b/drivers/gator/Kconfig
new file mode 100644
index 000000000000..a9503a9a4e9f
--- /dev/null
+++ b/drivers/gator/Kconfig
@@ -0,0 +1,38 @@
+config GATOR
+	tristate "Gator module for ARM's Streamline Performance Analyzer"
+	default m if (ARM || ARM64)
+	depends on PROFILING
+	depends on HIGH_RES_TIMERS
+	depends on PERF_EVENTS
+	depends on HW_PERF_EVENTS || !(ARM || ARM64)
+	select TRACING
+	help
+	  Gator module for ARM's Streamline Performance Analyzer
+
+config GATOR_WITH_MALI_SUPPORT
+	bool
+
+choice
+	prompt "Enable Mali GPU support in Gator"
+	depends on GATOR
+	optional
+	help
+	  Enable Mali GPU support in Gator
+
+config GATOR_MALI_4XXMP
+	bool "Mali-400MP or Mali-450MP"
+	select GATOR_WITH_MALI_SUPPORT
+
+config GATOR_MALI_MIDGARD
+	bool "Mali-T60x, Mali-T62x, Mali-T72x or Mali-T76x"
+	select GATOR_WITH_MALI_SUPPORT
+
+endchoice
+
+config GATOR_MALI_PATH
+	string "Path to Mali driver"
+	depends on GATOR_WITH_MALI_SUPPORT
+	default "drivers/gpu/arm/mali400mp"
+	help
+	  The gator code adds this to its include path so it can get the Mali
+	  trace headers with: #include "linux/mali_linux_trace.h"
diff --git a/drivers/gator/LICENSE b/drivers/gator/LICENSE
new file mode 100644
index 000000000000..d159169d1050
--- /dev/null
+++ b/drivers/gator/LICENSE
@@ -0,0 +1,339 @@
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 2, June 1991
+
+ Copyright (C) 1989, 1991 Free Software Foundation, Inc.,
+ 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The licenses for most software are designed to take away your
+freedom to share and change it.  By contrast, the GNU General Public
+License is intended to guarantee your freedom to share and change free
+software--to make sure the software is free for all its users.  This
+General Public License applies to most of the Free Software
+Foundation's software and to any other program whose authors commit to
+using it.  (Some other Free Software Foundation software is covered by
+the GNU Lesser General Public License instead.)  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+this service if you wish), that you receive source code or can get it
+if you want it, that you can change the software or use pieces of it
+in new free programs; and that you know you can do these things.
+
+  To protect your rights, we need to make restrictions that forbid
+anyone to deny you these rights or to ask you to surrender the rights.
+These restrictions translate to certain responsibilities for you if you
+distribute copies of the software, or if you modify it.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must give the recipients all the rights that
+you have.  You must make sure that they, too, receive or can get the
+source code.  And you must show them these terms so they know their
+rights.
+
+  We protect your rights with two steps: (1) copyright the software, and
+(2) offer you this license which gives you legal permission to copy,
+distribute and/or modify the software.
+
+  Also, for each author's protection and ours, we want to make certain
+that everyone understands that there is no warranty for this free
+software.  If the software is modified by someone else and passed on, we
+want its recipients to know that what they have is not the original, so
+that any problems introduced by others will not reflect on the original
+authors' reputations.
+
+  Finally, any free program is threatened constantly by software
+patents.  We wish to avoid the danger that redistributors of a free
+program will individually obtain patent licenses, in effect making the
+program proprietary.  To prevent this, we have made it clear that any
+patent must be licensed for everyone's free use or not licensed at all.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                    GNU GENERAL PUBLIC LICENSE
+   TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION
+
+  0. This License applies to any program or other work which contains
+a notice placed by the copyright holder saying it may be distributed
+under the terms of this General Public License.  The "Program", below,
+refers to any such program or work, and a "work based on the Program"
+means either the Program or any derivative work under copyright law:
+that is to say, a work containing the Program or a portion of it,
+either verbatim or with modifications and/or translated into another
+language.  (Hereinafter, translation is included without limitation in
+the term "modification".)  Each licensee is addressed as "you".
+
+Activities other than copying, distribution and modification are not
+covered by this License; they are outside its scope.  The act of
+running the Program is not restricted, and the output from the Program
+is covered only if its contents constitute a work based on the
+Program (independent of having been made by running the Program).
+Whether that is true depends on what the Program does.
+
+  1. You may copy and distribute verbatim copies of the Program's
+source code as you receive it, in any medium, provided that you
+conspicuously and appropriately publish on each copy an appropriate
+copyright notice and disclaimer of warranty; keep intact all the
+notices that refer to this License and to the absence of any warranty;
+and give any other recipients of the Program a copy of this License
+along with the Program.
+
+You may charge a fee for the physical act of transferring a copy, and
+you may at your option offer warranty protection in exchange for a fee.
+
+  2. You may modify your copy or copies of the Program or any portion
+of it, thus forming a work based on the Program, and copy and
+distribute such modifications or work under the terms of Section 1
+above, provided that you also meet all of these conditions:
+
+    a) You must cause the modified files to carry prominent notices
+    stating that you changed the files and the date of any change.
+
+    b) You must cause any work that you distribute or publish, that in
+    whole or in part contains or is derived from the Program or any
+    part thereof, to be licensed as a whole at no charge to all third
+    parties under the terms of this License.
+
+    c) If the modified program normally reads commands interactively
+    when run, you must cause it, when started running for such
+    interactive use in the most ordinary way, to print or display an
+    announcement including an appropriate copyright notice and a
+    notice that there is no warranty (or else, saying that you provide
+    a warranty) and that users may redistribute the program under
+    these conditions, and telling the user how to view a copy of this
+    License.  (Exception: if the Program itself is interactive but
+    does not normally print such an announcement, your work based on
+    the Program is not required to print an announcement.)
+
+These requirements apply to the modified work as a whole.  If
+identifiable sections of that work are not derived from the Program,
+and can be reasonably considered independent and separate works in
+themselves, then this License, and its terms, do not apply to those
+sections when you distribute them as separate works.  But when you
+distribute the same sections as part of a whole which is a work based
+on the Program, the distribution of the whole must be on the terms of
+this License, whose permissions for other licensees extend to the
+entire whole, and thus to each and every part regardless of who wrote it.
+
+Thus, it is not the intent of this section to claim rights or contest
+your rights to work written entirely by you; rather, the intent is to
+exercise the right to control the distribution of derivative or
+collective works based on the Program.
+
+In addition, mere aggregation of another work not based on the Program
+with the Program (or with a work based on the Program) on a volume of
+a storage or distribution medium does not bring the other work under
+the scope of this License.
+
+  3. You may copy and distribute the Program (or a work based on it,
+under Section 2) in object code or executable form under the terms of
+Sections 1 and 2 above provided that you also do one of the following:
+
+    a) Accompany it with the complete corresponding machine-readable
+    source code, which must be distributed under the terms of Sections
+    1 and 2 above on a medium customarily used for software interchange; or,
+
+    b) Accompany it with a written offer, valid for at least three
+    years, to give any third party, for a charge no more than your
+    cost of physically performing source distribution, a complete
+    machine-readable copy of the corresponding source code, to be
+    distributed under the terms of Sections 1 and 2 above on a medium
+    customarily used for software interchange; or,
+
+    c) Accompany it with the information you received as to the offer
+    to distribute corresponding source code.  (This alternative is
+    allowed only for noncommercial distribution and only if you
+    received the program in object code or executable form with such
+    an offer, in accord with Subsection b above.)
+
+The source code for a work means the preferred form of the work for
+making modifications to it.  For an executable work, complete source
+code means all the source code for all modules it contains, plus any
+associated interface definition files, plus the scripts used to
+control compilation and installation of the executable.  However, as a
+special exception, the source code distributed need not include
+anything that is normally distributed (in either source or binary
+form) with the major components (compiler, kernel, and so on) of the
+operating system on which the executable runs, unless that component
+itself accompanies the executable.
+
+If distribution of executable or object code is made by offering
+access to copy from a designated place, then offering equivalent
+access to copy the source code from the same place counts as
+distribution of the source code, even though third parties are not
+compelled to copy the source along with the object code.
+
+  4. You may not copy, modify, sublicense, or distribute the Program
+except as expressly provided under this License.  Any attempt
+otherwise to copy, modify, sublicense or distribute the Program is
+void, and will automatically terminate your rights under this License.
+However, parties who have received copies, or rights, from you under
+this License will not have their licenses terminated so long as such
+parties remain in full compliance.
+
+  5. You are not required to accept this License, since you have not
+signed it.  However, nothing else grants you permission to modify or
+distribute the Program or its derivative works.  These actions are
+prohibited by law if you do not accept this License.  Therefore, by
+modifying or distributing the Program (or any work based on the
+Program), you indicate your acceptance of this License to do so, and
+all its terms and conditions for copying, distributing or modifying
+the Program or works based on it.
+
+  6. Each time you redistribute the Program (or any work based on the
+Program), the recipient automatically receives a license from the
+original licensor to copy, distribute or modify the Program subject to
+these terms and conditions.  You may not impose any further
+restrictions on the recipients' exercise of the rights granted herein.
+You are not responsible for enforcing compliance by third parties to
+this License.
+
+  7. If, as a consequence of a court judgment or allegation of patent
+infringement or for any other reason (not limited to patent issues),
+conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot
+distribute so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you
+may not distribute the Program at all.  For example, if a patent
+license would not permit royalty-free redistribution of the Program by
+all those who receive copies directly or indirectly through you, then
+the only way you could satisfy both it and this License would be to
+refrain entirely from distribution of the Program.
+
+If any portion of this section is held invalid or unenforceable under
+any particular circumstance, the balance of the section is intended to
+apply and the section as a whole is intended to apply in other
+circumstances.
+
+It is not the purpose of this section to induce you to infringe any
+patents or other property right claims or to contest validity of any
+such claims; this section has the sole purpose of protecting the
+integrity of the free software distribution system, which is
+implemented by public license practices.  Many people have made
+generous contributions to the wide range of software distributed
+through that system in reliance on consistent application of that
+system; it is up to the author/donor to decide if he or she is willing
+to distribute software through any other system and a licensee cannot
+impose that choice.
+
+This section is intended to make thoroughly clear what is believed to
+be a consequence of the rest of this License.
+
+  8. If the distribution and/or use of the Program is restricted in
+certain countries either by patents or by copyrighted interfaces, the
+original copyright holder who places the Program under this License
+may add an explicit geographical distribution limitation excluding
+those countries, so that distribution is permitted only in or among
+countries not thus excluded.  In such case, this License incorporates
+the limitation as if written in the body of this License.
+
+  9. The Free Software Foundation may publish revised and/or new versions
+of the General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+Each version is given a distinguishing version number.  If the Program
+specifies a version number of this License which applies to it and "any
+later version", you have the option of following the terms and conditions
+either of that version or of any later version published by the Free
+Software Foundation.  If the Program does not specify a version number of
+this License, you may choose any version ever published by the Free Software
+Foundation.
+
+  10. If you wish to incorporate parts of the Program into other free
+programs whose distribution conditions are different, write to the author
+to ask for permission.  For software which is copyrighted by the Free
+Software Foundation, write to the Free Software Foundation; we sometimes
+make exceptions for this.  Our decision will be guided by the two goals
+of preserving the free status of all derivatives of our free software and
+of promoting the sharing and reuse of software generally.
+
+                            NO WARRANTY
+
+  11. BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY
+FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW.  EXCEPT WHEN
+OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES
+PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED
+OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.  THE ENTIRE RISK AS
+TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU.  SHOULD THE
+PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING,
+REPAIR OR CORRECTION.
+
+  12. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY AND/OR
+REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES,
+INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING
+OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED
+TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY
+YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER
+PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE
+POSSIBILITY OF SUCH DAMAGES.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+convey the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+Also add information on how to contact you by electronic and paper mail.
+
+If the program is interactive, make it output a short notice like this
+when it starts in an interactive mode:
+
+    Gnomovision version 69, Copyright (C) year name of author
+    Gnomovision comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, the commands you use may
+be called something other than `show w' and `show c'; they could even be
+mouse-clicks or menu items--whatever suits your program.
+
+You should also get your employer (if you work as a programmer) or your
+school, if any, to sign a "copyright disclaimer" for the program, if
+necessary.  Here is a sample; alter the names:
+
+  Yoyodyne, Inc., hereby disclaims all copyright interest in the program
+  `Gnomovision' (which makes passes at compilers) written by James Hacker.
+
+  <signature of Ty Coon>, 1 April 1989
+  Ty Coon, President of Vice
+
+This General Public License does not permit incorporating your program into
+proprietary programs.  If your program is a subroutine library, you may
+consider it more useful to permit linking proprietary applications with the
+library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.
diff --git a/drivers/gator/Makefile b/drivers/gator/Makefile
new file mode 100644
index 000000000000..28d2070b11d5
--- /dev/null
+++ b/drivers/gator/Makefile
@@ -0,0 +1,79 @@
+ifneq ($(KERNELRELEASE),)
+
+# Uncomment the following line to enable kernel stack unwinding within gator, or update gator_backtrace.c
+# EXTRA_CFLAGS +=	-DGATOR_KERNEL_STACK_UNWINDING
+
+CONFIG_GATOR ?= m
+obj-$(CONFIG_GATOR) := gator.o
+
+gator-y :=	gator_main.o \
+		gator_events_block.o \
+		gator_events_irq.o \
+		gator_events_meminfo.o \
+		gator_events_mmapped.o \
+		gator_events_net.o \
+		gator_events_perf_pmu.o \
+		gator_events_sched.o \
+
+# Convert the old GATOR_WITH_MALI_SUPPORT to the new kernel flags
+ifneq ($(GATOR_WITH_MALI_SUPPORT),)
+  CONFIG_GATOR_WITH_MALI_SUPPORT := y
+  ifeq ($(GATOR_WITH_MALI_SUPPORT),MALI_MIDGARD)
+    CONFIG_GATOR_MALI_4XXMP := n
+    CONFIG_GATOR_MALI_MIDGARD := y
+  else
+    CONFIG_GATOR_MALI_4XXMP := y
+    CONFIG_GATOR_MALI_MIDGARD := n
+  endif
+  EXTRA_CFLAGS += -DMALI_SUPPORT=$(GATOR_WITH_MALI_SUPPORT)
+  ifneq ($(GATOR_MALI_INTERFACE_STYLE),)
+    EXTRA_CFLAGS += -DGATOR_MALI_INTERFACE_STYLE=$(GATOR_MALI_INTERFACE_STYLE)
+  endif
+endif
+
+ifeq ($(CONFIG_GATOR_WITH_MALI_SUPPORT),y)
+  ifeq ($(CONFIG_GATOR_MALI_MIDGARD),y)
+    gator-y +=	gator_events_mali_midgard.o \
+		gator_events_mali_midgard_hw.o
+    include $(src)/mali_midgard.mk
+  else
+    gator-y +=	gator_events_mali_4xx.o
+  endif
+  gator-y +=	gator_events_mali_common.o
+
+  ifneq ($(CONFIG_GATOR_MALI_PATH),)
+    ccflags-y += -I$(CONFIG_GATOR_MALI_PATH)
+  endif
+  ccflags-$(CONFIG_GATOR_MALI_4XXMP) += -DMALI_SUPPORT=MALI_4xx
+  ccflags-$(CONFIG_GATOR_MALI_MIDGARD) += -DMALI_SUPPORT=MALI_MIDGARD
+endif
+
+# GATOR_TEST controls whether to include (=1) or exclude (=0) test code.
+GATOR_TEST ?= 0
+EXTRA_CFLAGS +=	-DGATOR_TEST=$(GATOR_TEST)
+
+# Should the original or new block_rq_complete API be used?
+OLD_BLOCK_RQ_COMPLETE := $(shell grep -A3 block_rq_complete $(srctree)/include/trace/events/block.h | grep nr_bytes -q; echo $$?)
+EXTRA_CFLAGS += -DOLD_BLOCK_RQ_COMPLETE=$(OLD_BLOCK_RQ_COMPLETE)
+
+gator-$(CONFIG_ARM) +=	gator_events_armv6.o \
+			gator_events_armv7.o \
+			gator_events_l2c-310.o \
+			gator_events_scorpion.o
+
+gator-$(CONFIG_ARM64) +=
+
+else
+
+all:
+	@echo
+	@echo "usage:"
+	@echo "      make -C <kernel_build_dir> M=\`pwd\` ARCH=arm CROSS_COMPILE=<...> modules"
+	@echo
+	$(error)
+
+clean:
+	rm -f *.o .*.cmd modules.order Module.symvers gator.ko gator.mod.c
+	rm -rf .tmp_versions
+
+endif
diff --git a/drivers/gator/gator.h b/drivers/gator/gator.h
new file mode 100644
index 000000000000..5cc73a388c4f
--- /dev/null
+++ b/drivers/gator/gator.h
@@ -0,0 +1,152 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef GATOR_H_
+#define GATOR_H_
+
+#include <linux/version.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/list.h>
+
+#define GATOR_PERF_SUPPORT      (LINUX_VERSION_CODE >= KERNEL_VERSION(3, 0, 0))
+#define GATOR_PERF_PMU_SUPPORT  (GATOR_PERF_SUPPORT && defined(CONFIG_PERF_EVENTS) && (!(defined(__arm__) || defined(__aarch64__)) || defined(CONFIG_HW_PERF_EVENTS)))
+#define GATOR_NO_PERF_SUPPORT   (!(GATOR_PERF_SUPPORT))
+#define GATOR_CPU_FREQ_SUPPORT  ((LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 38)) && defined(CONFIG_CPU_FREQ))
+#define GATOR_IKS_SUPPORT       defined(CONFIG_BL_SWITCHER)
+
+/* cpu ids */
+#define ARM1136     0xb36
+#define ARM1156     0xb56
+#define ARM1176     0xb76
+#define ARM11MPCORE 0xb02
+#define CORTEX_A5   0xc05
+#define CORTEX_A7   0xc07
+#define CORTEX_A8   0xc08
+#define CORTEX_A9   0xc09
+#define CORTEX_A15  0xc0f
+#define CORTEX_A17  0xc0e
+#define SCORPION    0x00f
+#define SCORPIONMP  0x02d
+#define KRAITSIM    0x049
+#define KRAIT       0x04d
+#define KRAIT_S4_PRO 0x06f
+#define CORTEX_A53  0xd03
+#define CORTEX_A57  0xd07
+#define AARCH64     0xd0f
+#define OTHER       0xfff
+
+/* gpu enums */
+#define MALI_4xx     1
+#define MALI_MIDGARD 2
+
+#define MAXSIZE_CORE_NAME 32
+
+struct gator_cpu {
+	const int cpuid;
+	/* Human readable name */
+	const char core_name[MAXSIZE_CORE_NAME];
+	/* gatorfs event and Perf PMU name */
+	const char *const pmnc_name;
+	/* compatible from Documentation/devicetree/bindings/arm/cpus.txt */
+	const char *const dt_name;
+	const int pmnc_counters;
+};
+
+const struct gator_cpu *gator_find_cpu_by_cpuid(const u32 cpuid);
+const struct gator_cpu *gator_find_cpu_by_pmu_name(const char *const name);
+
+/******************************************************************************
+ * Filesystem
+ ******************************************************************************/
+struct dentry *gatorfs_mkdir(struct super_block *sb, struct dentry *root,
+			     char const *name);
+
+int gatorfs_create_ulong(struct super_block *sb, struct dentry *root,
+			 char const *name, unsigned long *val);
+
+int gatorfs_create_ro_ulong(struct super_block *sb, struct dentry *root,
+			    char const *name, unsigned long *val);
+
+/******************************************************************************
+ * Tracepoints
+ ******************************************************************************/
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
+#	error Kernels prior to 2.6.32 not supported
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 35)
+#	define GATOR_DEFINE_PROBE(probe_name, proto) \
+		static void probe_##probe_name(PARAMS(proto))
+#	define GATOR_REGISTER_TRACE(probe_name) \
+		register_trace_##probe_name(probe_##probe_name)
+#	define GATOR_UNREGISTER_TRACE(probe_name) \
+		unregister_trace_##probe_name(probe_##probe_name)
+#elif LINUX_VERSION_CODE < KERNEL_VERSION(3, 15, 0)
+#	define GATOR_DEFINE_PROBE(probe_name, proto) \
+		static void probe_##probe_name(void *data, PARAMS(proto))
+#	define GATOR_REGISTER_TRACE(probe_name) \
+		register_trace_##probe_name(probe_##probe_name, NULL)
+#	define GATOR_UNREGISTER_TRACE(probe_name) \
+		unregister_trace_##probe_name(probe_##probe_name, NULL)
+#else
+#	define GATOR_DEFINE_PROBE(probe_name, proto) \
+		extern struct tracepoint *gator_tracepoint_##probe_name; \
+		static void probe_##probe_name(void *data, PARAMS(proto))
+#	define GATOR_REGISTER_TRACE(probe_name) \
+		((gator_tracepoint_##probe_name == NULL) || tracepoint_probe_register(gator_tracepoint_##probe_name, probe_##probe_name, NULL))
+#	define GATOR_UNREGISTER_TRACE(probe_name) \
+		tracepoint_probe_unregister(gator_tracepoint_##probe_name, probe_##probe_name, NULL)
+#endif
+
+/******************************************************************************
+ * Events
+ ******************************************************************************/
+struct gator_interface {
+	/* Complementary function to init */
+	void (*shutdown)(void);
+	int (*create_files)(struct super_block *sb, struct dentry *root);
+	int (*start)(void);
+	/* Complementary function to start */
+	void (*stop)(void);
+	int (*online)(int **buffer, bool migrate);
+	int (*offline)(int **buffer, bool migrate);
+	/* called in process context but may not be running on core 'cpu' */
+	void (*online_dispatch)(int cpu, bool migrate);
+	/* called in process context but may not be running on core 'cpu' */
+	void (*offline_dispatch)(int cpu, bool migrate);
+	int (*read)(int **buffer, bool sched_switch);
+	int (*read64)(long long **buffer);
+	int (*read_proc)(long long **buffer, struct task_struct *);
+	struct list_head list;
+};
+
+int gator_events_install(struct gator_interface *interface);
+int gator_events_get_key(void);
+u32 gator_cpuid(void);
+
+void gator_backtrace_handler(struct pt_regs *const regs);
+
+void gator_marshal_activity_switch(int core, int key, int activity, int pid);
+
+#if !GATOR_IKS_SUPPORT
+
+#define get_physical_cpu() smp_processor_id()
+#define lcpu_to_pcpu(lcpu) lcpu
+#define pcpu_to_lcpu(pcpu) pcpu
+
+#else
+
+#define get_physical_cpu() lcpu_to_pcpu(get_logical_cpu())
+int lcpu_to_pcpu(const int lcpu);
+int pcpu_to_lcpu(const int pcpu);
+
+#endif
+
+#define get_logical_cpu() smp_processor_id()
+#define on_primary_core() (get_logical_cpu() == 0)
+
+#endif /* GATOR_H_ */
diff --git a/drivers/gator/gator_annotate.c b/drivers/gator/gator_annotate.c
new file mode 100644
index 000000000000..ff9a3cef7b2e
--- /dev/null
+++ b/drivers/gator/gator_annotate.c
@@ -0,0 +1,189 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#include <linux/slab.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/sched.h>
+#include <linux/uaccess.h>
+#include <asm/current.h>
+#include <linux/spinlock.h>
+
+static DEFINE_SPINLOCK(annotate_lock);
+static bool collect_annotations;
+
+static int annotate_copy(struct file *file, char const __user *buf, size_t count)
+{
+	int cpu = 0;
+	int write = per_cpu(gator_buffer_write, cpu)[ANNOTATE_BUF];
+
+	if (file == NULL) {
+		/* copy from kernel */
+		memcpy(&per_cpu(gator_buffer, cpu)[ANNOTATE_BUF][write], buf, count);
+	} else {
+		/* copy from user space */
+		if (copy_from_user(&per_cpu(gator_buffer, cpu)[ANNOTATE_BUF][write], buf, count) != 0)
+			return -1;
+	}
+	per_cpu(gator_buffer_write, cpu)[ANNOTATE_BUF] = (write + count) & gator_buffer_mask[ANNOTATE_BUF];
+
+	return 0;
+}
+
+static ssize_t annotate_write(struct file *file, char const __user *buf, size_t count_orig, loff_t *offset)
+{
+	int pid, cpu, header_size, available, contiguous, length1, length2, size, count = count_orig & 0x7fffffff;
+	bool interrupt_context;
+
+	if (*offset)
+		return -EINVAL;
+
+	interrupt_context = in_interrupt();
+	/* Annotations are not supported in interrupt context, but may work
+	 * if you comment out the the next four lines of code. By doing so,
+	 * annotations in interrupt context can result in deadlocks and lost
+	 * data.
+	 */
+	if (interrupt_context) {
+		pr_warning("gator: Annotations are not supported in interrupt context. Edit gator_annotate.c in the gator driver to enable annotations in interrupt context.\n");
+		return -EINVAL;
+	}
+
+ retry:
+	/* synchronize between cores and with collect_annotations */
+	spin_lock(&annotate_lock);
+
+	if (!collect_annotations) {
+		/* Not collecting annotations, tell the caller everything was written */
+		size = count_orig;
+		goto annotate_write_out;
+	}
+
+	/* Annotation only uses a single per-cpu buffer as the data must be in order to the engine */
+	cpu = 0;
+
+	if (current == NULL)
+		pid = 0;
+	else
+		pid = current->pid;
+
+	/* determine total size of the payload */
+	header_size = MAXSIZE_PACK32 * 3 + MAXSIZE_PACK64;
+	available = buffer_bytes_available(cpu, ANNOTATE_BUF) - header_size;
+	size = count < available ? count : available;
+
+	if (size <= 0) {
+		/* Buffer is full, wait until space is available */
+		spin_unlock(&annotate_lock);
+
+		/* Drop the annotation as blocking is not allowed in interrupt context */
+		if (interrupt_context)
+			return -EINVAL;
+
+		wait_event_interruptible(gator_annotate_wait, buffer_bytes_available(cpu, ANNOTATE_BUF) > header_size || !collect_annotations);
+
+		/* Check to see if a signal is pending */
+		if (signal_pending(current))
+			return -EINTR;
+
+		goto retry;
+	}
+
+	/* synchronize shared variables annotateBuf and annotatePos */
+	if (per_cpu(gator_buffer, cpu)[ANNOTATE_BUF]) {
+		u64 time = gator_get_time();
+
+		gator_buffer_write_packed_int(cpu, ANNOTATE_BUF, get_physical_cpu());
+		gator_buffer_write_packed_int(cpu, ANNOTATE_BUF, pid);
+		gator_buffer_write_packed_int64(cpu, ANNOTATE_BUF, time);
+		gator_buffer_write_packed_int(cpu, ANNOTATE_BUF, size);
+
+		/* determine the sizes to capture, length1 + length2 will equal size */
+		contiguous = contiguous_space_available(cpu, ANNOTATE_BUF);
+		if (size < contiguous) {
+			length1 = size;
+			length2 = 0;
+		} else {
+			length1 = contiguous;
+			length2 = size - contiguous;
+		}
+
+		if (annotate_copy(file, buf, length1) != 0) {
+			size = -EINVAL;
+			goto annotate_write_out;
+		}
+
+		if (length2 > 0 && annotate_copy(file, &buf[length1], length2) != 0) {
+			size = -EINVAL;
+			goto annotate_write_out;
+		}
+
+		/* Check and commit; commit is set to occur once buffer is 3/4 full */
+		buffer_check(cpu, ANNOTATE_BUF, time);
+	}
+
+annotate_write_out:
+	spin_unlock(&annotate_lock);
+
+	/* return the number of bytes written */
+	return size;
+}
+
+#include "gator_annotate_kernel.c"
+
+static int annotate_release(struct inode *inode, struct file *file)
+{
+	int cpu = 0;
+
+	/* synchronize between cores */
+	spin_lock(&annotate_lock);
+
+	if (per_cpu(gator_buffer, cpu)[ANNOTATE_BUF] && buffer_check_space(cpu, ANNOTATE_BUF, MAXSIZE_PACK64 + 3 * MAXSIZE_PACK32)) {
+		uint32_t pid = current->pid;
+
+		gator_buffer_write_packed_int(cpu, ANNOTATE_BUF, get_physical_cpu());
+		gator_buffer_write_packed_int(cpu, ANNOTATE_BUF, pid);
+		/* time */
+		gator_buffer_write_packed_int64(cpu, ANNOTATE_BUF, 0);
+		/* size */
+		gator_buffer_write_packed_int(cpu, ANNOTATE_BUF, 0);
+	}
+
+	/* Check and commit; commit is set to occur once buffer is 3/4 full */
+	buffer_check(cpu, ANNOTATE_BUF, gator_get_time());
+
+	spin_unlock(&annotate_lock);
+
+	return 0;
+}
+
+static const struct file_operations annotate_fops = {
+	.write = annotate_write,
+	.release = annotate_release
+};
+
+static int gator_annotate_create_files(struct super_block *sb, struct dentry *root)
+{
+	return gatorfs_create_file_perm(sb, root, "annotate", &annotate_fops, 0666);
+}
+
+static int gator_annotate_start(void)
+{
+	collect_annotations = true;
+	return 0;
+}
+
+static void gator_annotate_stop(void)
+{
+	/* the spinlock here will ensure that when this function exits, we are not in the middle of an annotation */
+	spin_lock(&annotate_lock);
+	collect_annotations = false;
+	wake_up(&gator_annotate_wait);
+	spin_unlock(&annotate_lock);
+}
diff --git a/drivers/gator/gator_annotate_kernel.c b/drivers/gator/gator_annotate_kernel.c
new file mode 100644
index 000000000000..69471f99e5fb
--- /dev/null
+++ b/drivers/gator/gator_annotate_kernel.c
@@ -0,0 +1,200 @@
+/**
+ * Copyright (C) ARM Limited 2012-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#define ESCAPE_CODE 0x1c
+#define STRING_ANNOTATION 0x06
+#define NAME_CHANNEL_ANNOTATION 0x07
+#define NAME_GROUP_ANNOTATION 0x08
+#define VISUAL_ANNOTATION 0x04
+#define MARKER_ANNOTATION 0x05
+
+static void kannotate_write(const char *ptr, unsigned int size)
+{
+	int retval;
+	int pos = 0;
+	loff_t offset = 0;
+
+	while (pos < size) {
+		retval = annotate_write(NULL, &ptr[pos], size - pos, &offset);
+		if (retval < 0) {
+			pr_warning("gator: kannotate_write failed with return value %d\n", retval);
+			return;
+		}
+		pos += retval;
+	}
+}
+
+static void marshal_u16(char *buf, u16 val)
+{
+	buf[0] = val & 0xff;
+	buf[1] = (val >> 8) & 0xff;
+}
+
+static void marshal_u32(char *buf, u32 val)
+{
+	buf[0] = val & 0xff;
+	buf[1] = (val >> 8) & 0xff;
+	buf[2] = (val >> 16) & 0xff;
+	buf[3] = (val >> 24) & 0xff;
+}
+
+void gator_annotate_channel(int channel, const char *str)
+{
+	const u16 str_size = strlen(str) & 0xffff;
+	char header[8];
+
+	header[0] = ESCAPE_CODE;
+	header[1] = STRING_ANNOTATION;
+	marshal_u32(header + 2, channel);
+	marshal_u16(header + 6, str_size);
+	kannotate_write(header, sizeof(header));
+	kannotate_write(str, str_size);
+}
+EXPORT_SYMBOL(gator_annotate_channel);
+
+void gator_annotate(const char *str)
+{
+	gator_annotate_channel(0, str);
+}
+EXPORT_SYMBOL(gator_annotate);
+
+void gator_annotate_channel_color(int channel, int color, const char *str)
+{
+	const u16 str_size = (strlen(str) + 4) & 0xffff;
+	char header[12];
+
+	header[0] = ESCAPE_CODE;
+	header[1] = STRING_ANNOTATION;
+	marshal_u32(header + 2, channel);
+	marshal_u16(header + 6, str_size);
+	marshal_u32(header + 8, color);
+	kannotate_write(header, sizeof(header));
+	kannotate_write(str, str_size - 4);
+}
+EXPORT_SYMBOL(gator_annotate_channel_color);
+
+void gator_annotate_color(int color, const char *str)
+{
+	gator_annotate_channel_color(0, color, str);
+}
+EXPORT_SYMBOL(gator_annotate_color);
+
+void gator_annotate_channel_end(int channel)
+{
+	char header[8];
+
+	header[0] = ESCAPE_CODE;
+	header[1] = STRING_ANNOTATION;
+	marshal_u32(header + 2, channel);
+	marshal_u16(header + 6, 0);
+	kannotate_write(header, sizeof(header));
+}
+EXPORT_SYMBOL(gator_annotate_channel_end);
+
+void gator_annotate_end(void)
+{
+	gator_annotate_channel_end(0);
+}
+EXPORT_SYMBOL(gator_annotate_end);
+
+void gator_annotate_name_channel(int channel, int group, const char *str)
+{
+	const u16 str_size = strlen(str) & 0xffff;
+	char header[12];
+
+	header[0] = ESCAPE_CODE;
+	header[1] = NAME_CHANNEL_ANNOTATION;
+	marshal_u32(header + 2, channel);
+	marshal_u32(header + 6, group);
+	marshal_u16(header + 10, str_size);
+	kannotate_write(header, sizeof(header));
+	kannotate_write(str, str_size);
+}
+EXPORT_SYMBOL(gator_annotate_name_channel);
+
+void gator_annotate_name_group(int group, const char *str)
+{
+	const u16 str_size = strlen(str) & 0xffff;
+	char header[8];
+
+	header[0] = ESCAPE_CODE;
+	header[1] = NAME_GROUP_ANNOTATION;
+	marshal_u32(header + 2, group);
+	marshal_u16(header + 6, str_size);
+	kannotate_write(header, sizeof(header));
+	kannotate_write(str, str_size);
+}
+EXPORT_SYMBOL(gator_annotate_name_group);
+
+void gator_annotate_visual(const char *data, unsigned int length, const char *str)
+{
+	const u16 str_size = strlen(str) & 0xffff;
+	char header[4];
+	char header_length[4];
+
+	header[0] = ESCAPE_CODE;
+	header[1] = VISUAL_ANNOTATION;
+	marshal_u16(header + 2, str_size);
+	marshal_u32(header_length, length);
+	kannotate_write(header, sizeof(header));
+	kannotate_write(str, str_size);
+	kannotate_write(header_length, sizeof(header_length));
+	kannotate_write(data, length);
+}
+EXPORT_SYMBOL(gator_annotate_visual);
+
+void gator_annotate_marker(void)
+{
+	char header[4];
+
+	header[0] = ESCAPE_CODE;
+	header[1] = MARKER_ANNOTATION;
+	marshal_u16(header + 2, 0);
+	kannotate_write(header, sizeof(header));
+}
+EXPORT_SYMBOL(gator_annotate_marker);
+
+void gator_annotate_marker_str(const char *str)
+{
+	const u16 str_size = strlen(str) & 0xffff;
+	char header[4];
+
+	header[0] = ESCAPE_CODE;
+	header[1] = MARKER_ANNOTATION;
+	marshal_u16(header + 2, str_size);
+	kannotate_write(header, sizeof(header));
+	kannotate_write(str, str_size);
+}
+EXPORT_SYMBOL(gator_annotate_marker_str);
+
+void gator_annotate_marker_color(int color)
+{
+	char header[8];
+
+	header[0] = ESCAPE_CODE;
+	header[1] = MARKER_ANNOTATION;
+	marshal_u16(header + 2, 4);
+	marshal_u32(header + 4, color);
+	kannotate_write(header, sizeof(header));
+}
+EXPORT_SYMBOL(gator_annotate_marker_color);
+
+void gator_annotate_marker_color_str(int color, const char *str)
+{
+	const u16 str_size = (strlen(str) + 4) & 0xffff;
+	char header[8];
+
+	header[0] = ESCAPE_CODE;
+	header[1] = MARKER_ANNOTATION;
+	marshal_u16(header + 2, str_size);
+	marshal_u32(header + 4, color);
+	kannotate_write(header, sizeof(header));
+	kannotate_write(str, str_size - 4);
+}
+EXPORT_SYMBOL(gator_annotate_marker_color_str);
diff --git a/drivers/gator/gator_backtrace.c b/drivers/gator/gator_backtrace.c
new file mode 100644
index 000000000000..76c941d009a9
--- /dev/null
+++ b/drivers/gator/gator_backtrace.c
@@ -0,0 +1,208 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+/*
+ * EABI backtrace stores {fp,lr} on the stack.
+ */
+struct stack_frame_eabi {
+	union {
+		struct {
+			unsigned long fp;
+			/* May be the fp in the case of a leaf function or clang */
+			unsigned long lr;
+			/* If lr is really the fp, lr2 is the corresponding lr */
+			unsigned long lr2;
+		};
+		/* Used to read 32 bit fp/lr from a 64 bit kernel */
+		struct {
+			u32 fp_32;
+			/* same as lr above */
+			u32 lr_32;
+			/* same as lr2 above */
+			u32 lr2_32;
+		};
+	};
+};
+
+static void gator_add_trace(int cpu, unsigned long address)
+{
+	off_t offset = 0;
+	unsigned long cookie = get_address_cookie(cpu, current, address & ~1, &offset);
+
+	if (cookie == NO_COOKIE || cookie == UNRESOLVED_COOKIE)
+		offset = address;
+
+	marshal_backtrace(offset & ~1, cookie, 0);
+}
+
+static void arm_backtrace_eabi(int cpu, struct pt_regs *const regs, unsigned int depth)
+{
+#if defined(__arm__) || defined(__aarch64__)
+	struct stack_frame_eabi *curr;
+	struct stack_frame_eabi bufcurr;
+#if defined(__arm__)
+	const bool is_compat = false;
+	unsigned long fp = regs->ARM_fp;
+	unsigned long sp = regs->ARM_sp;
+	unsigned long lr = regs->ARM_lr;
+	const int gcc_frame_offset = sizeof(unsigned long);
+#else
+	/* Is userspace aarch32 (32 bit) */
+	const bool is_compat = compat_user_mode(regs);
+	unsigned long fp = (is_compat ? regs->regs[11] : regs->regs[29]);
+	unsigned long sp = (is_compat ? regs->compat_sp : regs->sp);
+	unsigned long lr = (is_compat ? regs->compat_lr : regs->regs[30]);
+	const int gcc_frame_offset = (is_compat ? sizeof(u32) : 0);
+#endif
+	/* clang frame offset is always zero */
+	int is_user_mode = user_mode(regs);
+
+	/* pc (current function) has already been added */
+
+	if (!is_user_mode)
+		return;
+
+	/* Add the lr (parent function), entry preamble may not have
+	 * executed
+	 */
+	gator_add_trace(cpu, lr);
+
+	/* check fp is valid */
+	if (fp == 0 || fp < sp)
+		return;
+
+	/* Get the current stack frame */
+	curr = (struct stack_frame_eabi *)(fp - gcc_frame_offset);
+	if ((unsigned long)curr & 3)
+		return;
+
+	while (depth-- && curr) {
+		if (!access_ok(VERIFY_READ, curr, sizeof(struct stack_frame_eabi)) ||
+				__copy_from_user_inatomic(&bufcurr, curr, sizeof(struct stack_frame_eabi))) {
+			return;
+		}
+
+		fp = (is_compat ? bufcurr.fp_32 : bufcurr.fp);
+		lr = (is_compat ? bufcurr.lr_32 : bufcurr.lr);
+
+#define calc_next(reg) ((reg) - gcc_frame_offset)
+		/* Returns true if reg is a valid fp */
+#define validate_next(reg, curr) \
+		((reg) != 0 && (calc_next(reg) & 3) == 0 && (unsigned long)(curr) < calc_next(reg))
+
+		/* Try lr from the stack as the fp because gcc leaf functions do
+		 * not push lr. If gcc_frame_offset is non-zero, the lr will also
+		 * be the clang fp. This assumes code is at a lower address than
+		 * the stack
+		 */
+		if (validate_next(lr, curr)) {
+			fp = lr;
+			lr = (is_compat ? bufcurr.lr2_32 : bufcurr.lr2);
+		}
+
+		gator_add_trace(cpu, lr);
+
+		if (!validate_next(fp, curr))
+			return;
+
+		/* Move to the next stack frame */
+		curr = (struct stack_frame_eabi *)calc_next(fp);
+	}
+#endif
+}
+
+#if defined(__arm__) || defined(__aarch64__)
+static int report_trace(struct stackframe *frame, void *d)
+{
+	unsigned int *depth = d, cookie = NO_COOKIE;
+	unsigned long addr = frame->pc;
+
+	if (*depth) {
+#if defined(MODULE)
+		unsigned int cpu = get_physical_cpu();
+		struct module *mod = __module_address(addr);
+
+		if (mod) {
+			cookie = get_cookie(cpu, current, mod->name, false);
+			addr = addr - (unsigned long)mod->module_core;
+		}
+#endif
+		marshal_backtrace(addr & ~1, cookie, 1);
+		(*depth)--;
+	}
+
+	return *depth == 0;
+}
+#endif
+
+/* Uncomment the following line to enable kernel stack unwinding within gator, note it can also be defined from the Makefile */
+/* #define GATOR_KERNEL_STACK_UNWINDING */
+
+#if (defined(__arm__) || defined(__aarch64__)) && !defined(GATOR_KERNEL_STACK_UNWINDING)
+/* Disabled by default */
+MODULE_PARM_DESC(kernel_stack_unwinding, "Allow kernel stack unwinding.");
+static bool kernel_stack_unwinding;
+module_param(kernel_stack_unwinding, bool, 0644);
+#endif
+
+static void kernel_backtrace(int cpu, struct pt_regs *const regs)
+{
+#if defined(__arm__) || defined(__aarch64__)
+#ifdef GATOR_KERNEL_STACK_UNWINDING
+	int depth = gator_backtrace_depth;
+#else
+	int depth = (kernel_stack_unwinding ? gator_backtrace_depth : 1);
+#endif
+	struct stackframe frame;
+
+	if (depth == 0)
+		depth = 1;
+#if defined(__arm__)
+	frame.fp = regs->ARM_fp;
+	frame.sp = regs->ARM_sp;
+	frame.lr = regs->ARM_lr;
+	frame.pc = regs->ARM_pc;
+#else
+	frame.fp = regs->regs[29];
+	frame.sp = regs->sp;
+	frame.pc = regs->pc;
+#endif
+	walk_stackframe(&frame, report_trace, &depth);
+#else
+	marshal_backtrace(PC_REG & ~1, NO_COOKIE, 1);
+#endif
+}
+
+static void gator_add_sample(int cpu, struct pt_regs *const regs, u64 time)
+{
+	bool in_kernel;
+	unsigned long exec_cookie;
+
+	if (!regs)
+		return;
+
+	in_kernel = !user_mode(regs);
+	exec_cookie = get_exec_cookie(cpu, current);
+
+	if (!marshal_backtrace_header(exec_cookie, current->tgid, current->pid, time))
+		return;
+
+	if (in_kernel) {
+		kernel_backtrace(cpu, regs);
+	} else {
+		/* Cookie+PC */
+		gator_add_trace(cpu, PC_REG);
+
+		/* Backtrace */
+		if (gator_backtrace_depth)
+			arm_backtrace_eabi(cpu, regs, gator_backtrace_depth);
+	}
+
+	marshal_backtrace_footer(time);
+}
diff --git a/drivers/gator/gator_buffer.c b/drivers/gator/gator_buffer.c
new file mode 100644
index 000000000000..910d5aa15066
--- /dev/null
+++ b/drivers/gator/gator_buffer.c
@@ -0,0 +1,171 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+static void marshal_frame(int cpu, int buftype)
+{
+	int frame;
+	bool write_cpu;
+
+	if (!per_cpu(gator_buffer, cpu)[buftype])
+		return;
+
+	switch (buftype) {
+	case SUMMARY_BUF:
+		write_cpu = false;
+		frame = FRAME_SUMMARY;
+		break;
+	case BACKTRACE_BUF:
+		write_cpu = true;
+		frame = FRAME_BACKTRACE;
+		break;
+	case NAME_BUF:
+		write_cpu = true;
+		frame = FRAME_NAME;
+		break;
+	case COUNTER_BUF:
+		write_cpu = false;
+		frame = FRAME_COUNTER;
+		break;
+	case BLOCK_COUNTER_BUF:
+		write_cpu = true;
+		frame = FRAME_BLOCK_COUNTER;
+		break;
+	case ANNOTATE_BUF:
+		write_cpu = false;
+		frame = FRAME_ANNOTATE;
+		break;
+	case SCHED_TRACE_BUF:
+		write_cpu = true;
+		frame = FRAME_SCHED_TRACE;
+		break;
+	case IDLE_BUF:
+		write_cpu = false;
+		frame = FRAME_IDLE;
+		break;
+	case ACTIVITY_BUF:
+		write_cpu = false;
+		frame = FRAME_ACTIVITY;
+		break;
+	default:
+		write_cpu = false;
+		frame = -1;
+		break;
+	}
+
+	/* add response type */
+	if (gator_response_type > 0)
+		gator_buffer_write_packed_int(cpu, buftype, gator_response_type);
+
+	/* leave space for 4-byte unpacked length */
+	per_cpu(gator_buffer_write, cpu)[buftype] = (per_cpu(gator_buffer_write, cpu)[buftype] + sizeof(s32)) & gator_buffer_mask[buftype];
+
+	/* add frame type and core number */
+	gator_buffer_write_packed_int(cpu, buftype, frame);
+	if (write_cpu)
+		gator_buffer_write_packed_int(cpu, buftype, cpu);
+}
+
+static int buffer_bytes_available(int cpu, int buftype)
+{
+	int remaining, filled;
+
+	filled = per_cpu(gator_buffer_write, cpu)[buftype] - per_cpu(gator_buffer_read, cpu)[buftype];
+	if (filled < 0)
+		filled += gator_buffer_size[buftype];
+
+	remaining = gator_buffer_size[buftype] - filled;
+
+	if (per_cpu(buffer_space_available, cpu)[buftype])
+		/* Give some extra room; also allows space to insert the overflow error packet */
+		remaining -= 200;
+	else
+		/* Hysteresis, prevents multiple overflow messages */
+		remaining -= 2000;
+
+	return remaining;
+}
+
+static bool buffer_check_space(int cpu, int buftype, int bytes)
+{
+	int remaining = buffer_bytes_available(cpu, buftype);
+
+	if (remaining < bytes)
+		per_cpu(buffer_space_available, cpu)[buftype] = false;
+	else
+		per_cpu(buffer_space_available, cpu)[buftype] = true;
+
+	return per_cpu(buffer_space_available, cpu)[buftype];
+}
+
+static int contiguous_space_available(int cpu, int buftype)
+{
+	int remaining = buffer_bytes_available(cpu, buftype);
+	int contiguous = gator_buffer_size[buftype] - per_cpu(gator_buffer_write, cpu)[buftype];
+
+	if (remaining < contiguous)
+		return remaining;
+	return contiguous;
+}
+
+static void gator_commit_buffer(int cpu, int buftype, u64 time)
+{
+	int type_length, commit, length, byte;
+	unsigned long flags;
+
+	if (!per_cpu(gator_buffer, cpu)[buftype])
+		return;
+
+	/* post-populate the length, which does not include the response type length nor the length itself, i.e. only the length of the payload */
+	local_irq_save(flags);
+	type_length = gator_response_type ? 1 : 0;
+	commit = per_cpu(gator_buffer_commit, cpu)[buftype];
+	length = per_cpu(gator_buffer_write, cpu)[buftype] - commit;
+	if (length < 0)
+		length += gator_buffer_size[buftype];
+	length = length - type_length - sizeof(s32);
+
+	if (length <= FRAME_HEADER_SIZE) {
+		/* Nothing to write, only the frame header is present */
+		local_irq_restore(flags);
+		return;
+	}
+
+	for (byte = 0; byte < sizeof(s32); byte++)
+		per_cpu(gator_buffer, cpu)[buftype][(commit + type_length + byte) & gator_buffer_mask[buftype]] = (length >> byte * 8) & 0xFF;
+
+	per_cpu(gator_buffer_commit, cpu)[buftype] = per_cpu(gator_buffer_write, cpu)[buftype];
+
+	if (gator_live_rate > 0) {
+		while (time > per_cpu(gator_buffer_commit_time, cpu))
+			per_cpu(gator_buffer_commit_time, cpu) += gator_live_rate;
+	}
+
+	marshal_frame(cpu, buftype);
+	local_irq_restore(flags);
+
+	/* had to delay scheduling work as attempting to schedule work during the context switch is illegal in kernel versions 3.5 and greater */
+	if (per_cpu(in_scheduler_context, cpu)) {
+#ifndef CONFIG_PREEMPT_RT_FULL
+		/* mod_timer can not be used in interrupt context in RT-Preempt full */
+		mod_timer(&gator_buffer_wake_up_timer, jiffies + 1);
+#endif
+	} else {
+		up(&gator_buffer_wake_sem);
+	}
+}
+
+static void buffer_check(int cpu, int buftype, u64 time)
+{
+	int filled = per_cpu(gator_buffer_write, cpu)[buftype] - per_cpu(gator_buffer_commit, cpu)[buftype];
+
+	if (filled < 0)
+		filled += gator_buffer_size[buftype];
+	if (filled >= ((gator_buffer_size[buftype] * 3) / 4))
+		gator_commit_buffer(cpu, buftype, time);
+}
diff --git a/drivers/gator/gator_buffer_write.c b/drivers/gator/gator_buffer_write.c
new file mode 100644
index 000000000000..654ec606cfad
--- /dev/null
+++ b/drivers/gator/gator_buffer_write.c
@@ -0,0 +1,83 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+static void gator_buffer_write_packed_int(int cpu, int buftype, int x)
+{
+	uint32_t write = per_cpu(gator_buffer_write, cpu)[buftype];
+	uint32_t mask = gator_buffer_mask[buftype];
+	char *buffer = per_cpu(gator_buffer, cpu)[buftype];
+	int packedBytes = 0;
+	int more = true;
+
+	while (more) {
+		/* low order 7 bits of x */
+		char b = x & 0x7f;
+
+		x >>= 7;
+
+		if ((x == 0 && (b & 0x40) == 0) || (x == -1 && (b & 0x40) != 0))
+			more = false;
+		else
+			b |= 0x80;
+
+		buffer[(write + packedBytes) & mask] = b;
+		packedBytes++;
+	}
+
+	per_cpu(gator_buffer_write, cpu)[buftype] = (write + packedBytes) & mask;
+}
+
+static void gator_buffer_write_packed_int64(int cpu, int buftype, long long x)
+{
+	uint32_t write = per_cpu(gator_buffer_write, cpu)[buftype];
+	uint32_t mask = gator_buffer_mask[buftype];
+	char *buffer = per_cpu(gator_buffer, cpu)[buftype];
+	int packedBytes = 0;
+	int more = true;
+
+	while (more) {
+		/* low order 7 bits of x */
+		char b = x & 0x7f;
+
+		x >>= 7;
+
+		if ((x == 0 && (b & 0x40) == 0) || (x == -1 && (b & 0x40) != 0))
+			more = false;
+		else
+			b |= 0x80;
+
+		buffer[(write + packedBytes) & mask] = b;
+		packedBytes++;
+	}
+
+	per_cpu(gator_buffer_write, cpu)[buftype] = (write + packedBytes) & mask;
+}
+
+static void gator_buffer_write_bytes(int cpu, int buftype, const char *x, int len)
+{
+	int i;
+	u32 write = per_cpu(gator_buffer_write, cpu)[buftype];
+	u32 mask = gator_buffer_mask[buftype];
+	char *buffer = per_cpu(gator_buffer, cpu)[buftype];
+
+	for (i = 0; i < len; i++) {
+		buffer[write] = x[i];
+		write = (write + 1) & mask;
+	}
+
+	per_cpu(gator_buffer_write, cpu)[buftype] = write;
+}
+
+static void gator_buffer_write_string(int cpu, int buftype, const char *x)
+{
+	int len = strlen(x);
+
+	gator_buffer_write_packed_int(cpu, buftype, len);
+	gator_buffer_write_bytes(cpu, buftype, x, len);
+}
diff --git a/drivers/gator/gator_cookies.c b/drivers/gator/gator_cookies.c
new file mode 100644
index 000000000000..c43cce815226
--- /dev/null
+++ b/drivers/gator/gator_cookies.c
@@ -0,0 +1,446 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+/* must be power of 2 */
+#define COOKIEMAP_ENTRIES	1024
+/* must be a power of 2 - 512/4 = 128 entries */
+#define TRANSLATE_BUFFER_SIZE 512
+#define TRANSLATE_TEXT_SIZE		256
+#define MAX_COLLISIONS		2
+
+static uint32_t *gator_crc32_table;
+static unsigned int translate_buffer_mask;
+
+struct cookie_args {
+	struct task_struct *task;
+	const char *text;
+};
+
+static DEFINE_PER_CPU(char *, translate_text);
+static DEFINE_PER_CPU(uint32_t, cookie_next_key);
+static DEFINE_PER_CPU(uint64_t *, cookie_keys);
+static DEFINE_PER_CPU(uint32_t *, cookie_values);
+static DEFINE_PER_CPU(int, translate_buffer_read);
+static DEFINE_PER_CPU(int, translate_buffer_write);
+static DEFINE_PER_CPU(struct cookie_args *, translate_buffer);
+
+static uint32_t get_cookie(int cpu, struct task_struct *task, const char *text, bool from_wq);
+static void wq_cookie_handler(struct work_struct *unused);
+static DECLARE_WORK(cookie_work, wq_cookie_handler);
+static struct timer_list app_process_wake_up_timer;
+static void app_process_wake_up_handler(unsigned long unused_data);
+
+static uint32_t cookiemap_code(uint64_t value64)
+{
+	uint32_t value = (uint32_t)((value64 >> 32) + value64);
+	uint32_t cookiecode = (value >> 24) & 0xff;
+
+	cookiecode = cookiecode * 31 + ((value >> 16) & 0xff);
+	cookiecode = cookiecode * 31 + ((value >> 8) & 0xff);
+	cookiecode = cookiecode * 31 + ((value >> 0) & 0xff);
+	cookiecode &= (COOKIEMAP_ENTRIES - 1);
+	return cookiecode * MAX_COLLISIONS;
+}
+
+static uint32_t gator_chksum_crc32(const char *data)
+{
+	register unsigned long crc;
+	const unsigned char *block = data;
+	int i, length = strlen(data);
+
+	crc = 0xFFFFFFFF;
+	for (i = 0; i < length; i++)
+		crc = ((crc >> 8) & 0x00FFFFFF) ^ gator_crc32_table[(crc ^ *block++) & 0xFF];
+
+	return (crc ^ 0xFFFFFFFF);
+}
+
+/*
+ * Exists
+ *  Pre:  [0][1][v][3]..[n-1]
+ *  Post: [v][0][1][3]..[n-1]
+ */
+static uint32_t cookiemap_exists(uint64_t key)
+{
+	unsigned long x, flags, retval = 0;
+	int cpu = get_physical_cpu();
+	uint32_t cookiecode = cookiemap_code(key);
+	uint64_t *keys = &(per_cpu(cookie_keys, cpu)[cookiecode]);
+	uint32_t *values = &(per_cpu(cookie_values, cpu)[cookiecode]);
+
+	/* Can be called from interrupt handler or from work queue */
+	local_irq_save(flags);
+	for (x = 0; x < MAX_COLLISIONS; x++) {
+		if (keys[x] == key) {
+			uint32_t value = values[x];
+
+			for (; x > 0; x--) {
+				keys[x] = keys[x - 1];
+				values[x] = values[x - 1];
+			}
+			keys[0] = key;
+			values[0] = value;
+			retval = value;
+			break;
+		}
+	}
+	local_irq_restore(flags);
+
+	return retval;
+}
+
+/*
+ * Add
+ *  Pre:  [0][1][2][3]..[n-1]
+ *  Post: [v][0][1][2]..[n-2]
+ */
+static void cookiemap_add(uint64_t key, uint32_t value)
+{
+	int cpu = get_physical_cpu();
+	int cookiecode = cookiemap_code(key);
+	uint64_t *keys = &(per_cpu(cookie_keys, cpu)[cookiecode]);
+	uint32_t *values = &(per_cpu(cookie_values, cpu)[cookiecode]);
+	int x;
+
+	for (x = MAX_COLLISIONS - 1; x > 0; x--) {
+		keys[x] = keys[x - 1];
+		values[x] = values[x - 1];
+	}
+	keys[0] = key;
+	values[0] = value;
+}
+
+#ifndef CONFIG_PREEMPT_RT_FULL
+static void translate_buffer_write_args(int cpu, struct task_struct *task, const char *text)
+{
+	unsigned long flags;
+	int write;
+	int next_write;
+	struct cookie_args *args;
+
+	local_irq_save(flags);
+
+	write = per_cpu(translate_buffer_write, cpu);
+	next_write = (write + 1) & translate_buffer_mask;
+
+	/* At least one entry must always remain available as when read == write, the queue is empty not full */
+	if (next_write != per_cpu(translate_buffer_read, cpu)) {
+		args = &per_cpu(translate_buffer, cpu)[write];
+		args->task = task;
+		args->text = text;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 39)
+		get_task_struct(task);
+#endif
+		per_cpu(translate_buffer_write, cpu) = next_write;
+	}
+
+	local_irq_restore(flags);
+}
+#endif
+
+static void translate_buffer_read_args(int cpu, struct cookie_args *args)
+{
+	unsigned long flags;
+	int read;
+
+	local_irq_save(flags);
+
+	read = per_cpu(translate_buffer_read, cpu);
+	*args = per_cpu(translate_buffer, cpu)[read];
+	per_cpu(translate_buffer_read, cpu) = (read + 1) & translate_buffer_mask;
+
+	local_irq_restore(flags);
+}
+
+static void wq_cookie_handler(struct work_struct *unused)
+{
+	struct cookie_args args;
+	int cpu = get_physical_cpu(), cookie;
+
+	mutex_lock(&start_mutex);
+
+	if (gator_started != 0) {
+		while (per_cpu(translate_buffer_read, cpu) != per_cpu(translate_buffer_write, cpu)) {
+			translate_buffer_read_args(cpu, &args);
+			cookie = get_cookie(cpu, args.task, args.text, true);
+			marshal_link(cookie, args.task->tgid, args.task->pid);
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 39)
+			put_task_struct(args.task);
+#endif
+		}
+	}
+
+	mutex_unlock(&start_mutex);
+}
+
+static void app_process_wake_up_handler(unsigned long unused_data)
+{
+	/* had to delay scheduling work as attempting to schedule work during the context switch is illegal in kernel versions 3.5 and greater */
+	schedule_work(&cookie_work);
+}
+
+/* Retrieve full name from proc/pid/cmdline for java processes on Android */
+static int translate_app_process(const char **text, int cpu, struct task_struct *task, bool from_wq)
+{
+	void *maddr;
+	unsigned int len;
+	unsigned long addr;
+	struct mm_struct *mm;
+	struct page *page = NULL;
+	struct vm_area_struct *page_vma;
+	int bytes, offset, retval = 0;
+	char *buf = per_cpu(translate_text, cpu);
+
+#ifndef CONFIG_PREEMPT_RT_FULL
+	/* Push work into a work queue if in atomic context as the kernel
+	 * functions below might sleep. Rely on the in_interrupt variable
+	 * rather than in_irq() or in_interrupt() kernel functions, as the
+	 * value of these functions seems inconsistent during a context
+	 * switch between android/linux versions
+	 */
+	if (!from_wq) {
+		/* Check if already in buffer */
+		int pos = per_cpu(translate_buffer_read, cpu);
+
+		while (pos != per_cpu(translate_buffer_write, cpu)) {
+			if (per_cpu(translate_buffer, cpu)[pos].task == task)
+				goto out;
+			pos = (pos + 1) & translate_buffer_mask;
+		}
+
+		translate_buffer_write_args(cpu, task, *text);
+
+		/* Not safe to call in RT-Preempt full in schedule switch context */
+		mod_timer(&app_process_wake_up_timer, jiffies + 1);
+		goto out;
+	}
+#endif
+
+	mm = get_task_mm(task);
+	if (!mm)
+		goto out;
+	if (!mm->arg_end)
+		goto outmm;
+	addr = mm->arg_start;
+	len = mm->arg_end - mm->arg_start;
+
+	if (len > TRANSLATE_TEXT_SIZE)
+		len = TRANSLATE_TEXT_SIZE;
+
+	down_read(&mm->mmap_sem);
+	while (len) {
+		if (get_user_pages(task, mm, addr, 1, 0, 1, &page, &page_vma) <= 0)
+			goto outsem;
+
+		maddr = kmap(page);
+		offset = addr & (PAGE_SIZE - 1);
+		bytes = len;
+		if (bytes > PAGE_SIZE - offset)
+			bytes = PAGE_SIZE - offset;
+
+		copy_from_user_page(page_vma, page, addr, buf, maddr + offset, bytes);
+
+		/* release page allocated by get_user_pages() */
+		kunmap(page);
+		page_cache_release(page);
+
+		len -= bytes;
+		buf += bytes;
+		addr += bytes;
+
+		*text = per_cpu(translate_text, cpu);
+		retval = 1;
+	}
+
+	/* On app_process startup, /proc/pid/cmdline is initially "zygote" then "<pre-initialized>" but changes after an initial startup period */
+	if (strcmp(*text, "zygote") == 0 || strcmp(*text, "<pre-initialized>") == 0)
+		retval = 0;
+
+outsem:
+	up_read(&mm->mmap_sem);
+outmm:
+	mmput(mm);
+out:
+	return retval;
+}
+
+static const char APP_PROCESS[] = "app_process";
+
+static uint32_t get_cookie(int cpu, struct task_struct *task, const char *text, bool from_wq)
+{
+	unsigned long flags, cookie;
+	uint64_t key;
+
+	key = gator_chksum_crc32(text);
+	key = (key << 32) | (uint32_t)task->tgid;
+
+	cookie = cookiemap_exists(key);
+	if (cookie)
+		return cookie;
+
+	/* On 64-bit android app_process can be app_process32 or app_process64 */
+	if (strncmp(text, APP_PROCESS, sizeof(APP_PROCESS) - 1) == 0) {
+		if (!translate_app_process(&text, cpu, task, from_wq))
+			return UNRESOLVED_COOKIE;
+	}
+
+	/* Can be called from interrupt handler or from work queue or from scheduler trace */
+	local_irq_save(flags);
+
+	cookie = UNRESOLVED_COOKIE;
+	if (marshal_cookie_header(text)) {
+		cookie = per_cpu(cookie_next_key, cpu) += nr_cpu_ids;
+		cookiemap_add(key, cookie);
+		marshal_cookie(cookie, text);
+	}
+
+	local_irq_restore(flags);
+
+	return cookie;
+}
+
+static int get_exec_cookie(int cpu, struct task_struct *task)
+{
+	struct mm_struct *mm = task->mm;
+	const char *text;
+
+	/* kernel threads have no address space */
+	if (!mm)
+		return NO_COOKIE;
+
+	if (task && task->mm && task->mm->exe_file) {
+		text = task->mm->exe_file->f_path.dentry->d_name.name;
+		return get_cookie(cpu, task, text, false);
+	}
+
+	return UNRESOLVED_COOKIE;
+}
+
+static unsigned long get_address_cookie(int cpu, struct task_struct *task, unsigned long addr, off_t *offset)
+{
+	unsigned long cookie = NO_COOKIE;
+	struct mm_struct *mm = task->mm;
+	struct vm_area_struct *vma;
+	const char *text;
+
+	if (!mm)
+		return cookie;
+
+	for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
+		if (addr < vma->vm_start || addr >= vma->vm_end)
+			continue;
+
+		if (vma->vm_file) {
+			text = vma->vm_file->f_path.dentry->d_name.name;
+			cookie = get_cookie(cpu, task, text, false);
+			*offset = (vma->vm_pgoff << PAGE_SHIFT) + addr - vma->vm_start;
+		} else {
+			/* must be an anonymous map */
+			*offset = addr;
+		}
+
+		break;
+	}
+
+	if (!vma)
+		cookie = UNRESOLVED_COOKIE;
+
+	return cookie;
+}
+
+static int cookies_initialize(void)
+{
+	uint32_t crc, poly;
+	int i, j, cpu, size, err = 0;
+
+	translate_buffer_mask = TRANSLATE_BUFFER_SIZE / sizeof(per_cpu(translate_buffer, 0)[0]) - 1;
+
+	for_each_present_cpu(cpu) {
+		per_cpu(cookie_next_key, cpu) = nr_cpu_ids + cpu;
+
+		size = COOKIEMAP_ENTRIES * MAX_COLLISIONS * sizeof(uint64_t);
+		per_cpu(cookie_keys, cpu) = kmalloc(size, GFP_KERNEL);
+		if (!per_cpu(cookie_keys, cpu)) {
+			err = -ENOMEM;
+			goto cookie_setup_error;
+		}
+		memset(per_cpu(cookie_keys, cpu), 0, size);
+
+		size = COOKIEMAP_ENTRIES * MAX_COLLISIONS * sizeof(uint32_t);
+		per_cpu(cookie_values, cpu) = kmalloc(size, GFP_KERNEL);
+		if (!per_cpu(cookie_values, cpu)) {
+			err = -ENOMEM;
+			goto cookie_setup_error;
+		}
+		memset(per_cpu(cookie_values, cpu), 0, size);
+
+		per_cpu(translate_buffer, cpu) = kmalloc(TRANSLATE_BUFFER_SIZE, GFP_KERNEL);
+		if (!per_cpu(translate_buffer, cpu)) {
+			err = -ENOMEM;
+			goto cookie_setup_error;
+		}
+
+		per_cpu(translate_buffer_write, cpu) = 0;
+		per_cpu(translate_buffer_read, cpu) = 0;
+
+		per_cpu(translate_text, cpu) = kmalloc(TRANSLATE_TEXT_SIZE, GFP_KERNEL);
+		if (!per_cpu(translate_text, cpu)) {
+			err = -ENOMEM;
+			goto cookie_setup_error;
+		}
+	}
+
+	/* build CRC32 table */
+	poly = 0x04c11db7;
+	gator_crc32_table = kmalloc(256 * sizeof(*gator_crc32_table), GFP_KERNEL);
+	if (!gator_crc32_table) {
+		err = -ENOMEM;
+		goto cookie_setup_error;
+	}
+	for (i = 0; i < 256; i++) {
+		crc = i;
+		for (j = 8; j > 0; j--) {
+			if (crc & 1)
+				crc = (crc >> 1) ^ poly;
+			else
+				crc >>= 1;
+		}
+		gator_crc32_table[i] = crc;
+	}
+
+	setup_timer(&app_process_wake_up_timer, app_process_wake_up_handler, 0);
+
+cookie_setup_error:
+	return err;
+}
+
+static void cookies_release(void)
+{
+	int cpu;
+
+	for_each_present_cpu(cpu) {
+		kfree(per_cpu(cookie_keys, cpu));
+		per_cpu(cookie_keys, cpu) = NULL;
+
+		kfree(per_cpu(cookie_values, cpu));
+		per_cpu(cookie_values, cpu) = NULL;
+
+		kfree(per_cpu(translate_buffer, cpu));
+		per_cpu(translate_buffer, cpu) = NULL;
+		per_cpu(translate_buffer_read, cpu) = 0;
+		per_cpu(translate_buffer_write, cpu) = 0;
+
+		kfree(per_cpu(translate_text, cpu));
+		per_cpu(translate_text, cpu) = NULL;
+	}
+
+	del_timer_sync(&app_process_wake_up_timer);
+	kfree(gator_crc32_table);
+	gator_crc32_table = NULL;
+}
diff --git a/drivers/gator/gator_events_armv6.c b/drivers/gator/gator_events_armv6.c
new file mode 100644
index 000000000000..a157a0013302
--- /dev/null
+++ b/drivers/gator/gator_events_armv6.c
@@ -0,0 +1,234 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "gator.h"
+
+/* gator_events_perf_pmu.c is used if perf is supported */
+#if GATOR_NO_PERF_SUPPORT
+
+static const char *pmnc_name;
+
+/*
+ * Per-CPU PMCR
+ */
+#define PMCR_E			(1 << 0)	/* Enable */
+#define PMCR_P			(1 << 1)	/* Count reset */
+#define PMCR_C			(1 << 2)	/* Cycle counter reset */
+#define PMCR_OFL_PMN0	(1 << 8)	/* Count reg 0 overflow */
+#define PMCR_OFL_PMN1	(1 << 9)	/* Count reg 1 overflow */
+#define PMCR_OFL_CCNT	(1 << 10)	/* Cycle counter overflow */
+
+#define PMN0 0
+#define PMN1 1
+#define CCNT 2
+#define CNTMAX	(CCNT+1)
+
+static int pmnc_counters;
+static unsigned long pmnc_enabled[CNTMAX];
+static unsigned long pmnc_event[CNTMAX];
+static unsigned long pmnc_key[CNTMAX];
+
+static DEFINE_PER_CPU(int[CNTMAX * 2], perfCnt);
+
+static inline void armv6_pmnc_write(u32 val)
+{
+	/* upper 4bits and 7, 11 are write-as-0 */
+	val &= 0x0ffff77f;
+	asm volatile("mcr p15, 0, %0, c15, c12, 0" : : "r" (val));
+}
+
+static inline u32 armv6_pmnc_read(void)
+{
+	u32 val;
+
+	asm volatile("mrc p15, 0, %0, c15, c12, 0" : "=r" (val));
+	return val;
+}
+
+static void armv6_pmnc_reset_counter(unsigned int cnt)
+{
+	u32 val = 0;
+
+	switch (cnt) {
+	case CCNT:
+		asm volatile("mcr p15, 0, %0, c15, c12, 1" : : "r" (val));
+		break;
+	case PMN0:
+		asm volatile("mcr p15, 0, %0, c15, c12, 2" : : "r" (val));
+		break;
+	case PMN1:
+		asm volatile("mcr p15, 0, %0, c15, c12, 3" : : "r" (val));
+		break;
+	}
+}
+
+int gator_events_armv6_create_files(struct super_block *sb, struct dentry *root)
+{
+	struct dentry *dir;
+	int i;
+
+	pmnc_counters = 3;
+
+	for (i = PMN0; i <= CCNT; i++) {
+		char buf[40];
+
+		if (i == CCNT)
+			snprintf(buf, sizeof(buf), "ARM_%s_ccnt", pmnc_name);
+		else
+			snprintf(buf, sizeof(buf), "ARM_%s_cnt%d", pmnc_name, i);
+		dir = gatorfs_mkdir(sb, root, buf);
+		if (!dir)
+			return -1;
+		gatorfs_create_ulong(sb, dir, "enabled", &pmnc_enabled[i]);
+		gatorfs_create_ro_ulong(sb, dir, "key", &pmnc_key[i]);
+		if (i != CCNT)
+			gatorfs_create_ulong(sb, dir, "event", &pmnc_event[i]);
+	}
+
+	return 0;
+}
+
+static int gator_events_armv6_online(int **buffer, bool migrate)
+{
+	unsigned int cnt, len = 0, cpu = smp_processor_id();
+	u32 pmnc;
+
+	if (armv6_pmnc_read() & PMCR_E)
+		armv6_pmnc_write(armv6_pmnc_read() & ~PMCR_E);
+
+	/* initialize PMNC, reset overflow, D bit, C bit and P bit. */
+	armv6_pmnc_write(PMCR_OFL_PMN0 | PMCR_OFL_PMN1 | PMCR_OFL_CCNT |
+			 PMCR_C | PMCR_P);
+
+	/* configure control register */
+	for (pmnc = 0, cnt = PMN0; cnt <= CCNT; cnt++) {
+		unsigned long event;
+
+		if (!pmnc_enabled[cnt])
+			continue;
+
+		event = pmnc_event[cnt] & 255;
+
+		/* Set event (if destined for PMNx counters) */
+		if (cnt == PMN0)
+			pmnc |= event << 20;
+		else if (cnt == PMN1)
+			pmnc |= event << 12;
+
+		/* Reset counter */
+		armv6_pmnc_reset_counter(cnt);
+	}
+	armv6_pmnc_write(pmnc | PMCR_E);
+
+	/* return zero values, no need to read as the counters were just reset */
+	for (cnt = PMN0; cnt <= CCNT; cnt++) {
+		if (pmnc_enabled[cnt]) {
+			per_cpu(perfCnt, cpu)[len++] = pmnc_key[cnt];
+			per_cpu(perfCnt, cpu)[len++] = 0;
+		}
+	}
+
+	if (buffer)
+		*buffer = per_cpu(perfCnt, cpu);
+
+	return len;
+}
+
+static int gator_events_armv6_offline(int **buffer, bool migrate)
+{
+	unsigned int cnt;
+
+	armv6_pmnc_write(armv6_pmnc_read() & ~PMCR_E);
+	for (cnt = PMN0; cnt <= CCNT; cnt++)
+		armv6_pmnc_reset_counter(cnt);
+
+	return 0;
+}
+
+static void gator_events_armv6_stop(void)
+{
+	unsigned int cnt;
+
+	for (cnt = PMN0; cnt <= CCNT; cnt++) {
+		pmnc_enabled[cnt] = 0;
+		pmnc_event[cnt] = 0;
+	}
+}
+
+static int gator_events_armv6_read(int **buffer, bool sched_switch)
+{
+	int cnt, len = 0;
+	int cpu = smp_processor_id();
+
+	/* a context switch may occur before the online hotplug event, thus need to check that the pmu is enabled */
+	if (!(armv6_pmnc_read() & PMCR_E))
+		return 0;
+
+	for (cnt = PMN0; cnt <= CCNT; cnt++) {
+		if (pmnc_enabled[cnt]) {
+			u32 value = 0;
+
+			switch (cnt) {
+			case CCNT:
+				asm volatile("mrc p15, 0, %0, c15, c12, 1" : "=r" (value));
+				break;
+			case PMN0:
+				asm volatile("mrc p15, 0, %0, c15, c12, 2" : "=r" (value));
+				break;
+			case PMN1:
+				asm volatile("mrc p15, 0, %0, c15, c12, 3" : "=r" (value));
+				break;
+			}
+			armv6_pmnc_reset_counter(cnt);
+
+			per_cpu(perfCnt, cpu)[len++] = pmnc_key[cnt];
+			per_cpu(perfCnt, cpu)[len++] = value;
+		}
+	}
+
+	if (buffer)
+		*buffer = per_cpu(perfCnt, cpu);
+
+	return len;
+}
+
+static struct gator_interface gator_events_armv6_interface = {
+	.create_files = gator_events_armv6_create_files,
+	.stop = gator_events_armv6_stop,
+	.online = gator_events_armv6_online,
+	.offline = gator_events_armv6_offline,
+	.read = gator_events_armv6_read,
+};
+
+int gator_events_armv6_init(void)
+{
+	unsigned int cnt;
+
+	switch (gator_cpuid()) {
+	case ARM1136:
+	case ARM1156:
+	case ARM1176:
+		pmnc_name = "ARM11";
+		break;
+	case ARM11MPCORE:
+		pmnc_name = "ARM11MPCore";
+		break;
+	default:
+		return -1;
+	}
+
+	for (cnt = PMN0; cnt <= CCNT; cnt++) {
+		pmnc_enabled[cnt] = 0;
+		pmnc_event[cnt] = 0;
+		pmnc_key[cnt] = gator_events_get_key();
+	}
+
+	return gator_events_install(&gator_events_armv6_interface);
+}
+
+#endif
diff --git a/drivers/gator/gator_events_armv7.c b/drivers/gator/gator_events_armv7.c
new file mode 100644
index 000000000000..09c94220114c
--- /dev/null
+++ b/drivers/gator/gator_events_armv7.c
@@ -0,0 +1,314 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+/*  Disabling interrupts
+ *    Many of the functions below disable interrupts via local_irq_save(). This disabling of interrupts is done to prevent any race conditions
+ *    between multiple entities (e.g. hrtimer interrupts and event based interrupts) calling the same functions. As accessing the pmu involves
+ *    several steps (disable, select, read, enable), these steps must be performed atomically. Normal synchronization routines cannot be used
+ *    as these functions are being called from interrupt context.
+ */
+
+#include "gator.h"
+
+/* gator_events_perf_pmu.c is used if perf is supported */
+#if GATOR_NO_PERF_SUPPORT
+
+/* Per-CPU PMNC: config reg */
+#define PMNC_E		(1 << 0)	/* Enable all counters */
+#define PMNC_P		(1 << 1)	/* Reset all counters */
+#define PMNC_C		(1 << 2)	/* Cycle counter reset */
+#define	PMNC_MASK	0x3f	/* Mask for writable bits */
+
+/* ccnt reg */
+#define CCNT_REG	(1 << 31)
+
+#define CCNT		0
+#define CNT0		1
+#define CNTMAX		(6+1)
+
+static const char *pmnc_name;
+static int pmnc_counters;
+
+static unsigned long pmnc_enabled[CNTMAX];
+static unsigned long pmnc_event[CNTMAX];
+static unsigned long pmnc_key[CNTMAX];
+
+static DEFINE_PER_CPU(int[CNTMAX * 2], perfCnt);
+
+inline void armv7_pmnc_write(u32 val)
+{
+	val &= PMNC_MASK;
+	asm volatile("mcr p15, 0, %0, c9, c12, 0" : : "r" (val));
+}
+
+inline u32 armv7_pmnc_read(void)
+{
+	u32 val;
+
+	asm volatile("mrc p15, 0, %0, c9, c12, 0" : "=r" (val));
+	return val;
+}
+
+inline u32 armv7_ccnt_read(u32 reset_value)
+{
+	unsigned long flags;
+	u32 newval = -reset_value;
+	u32 den = CCNT_REG;
+	u32 val;
+
+	local_irq_save(flags);
+	asm volatile("mcr p15, 0, %0, c9, c12, 2" : : "r" (den));	/* disable */
+	asm volatile("mrc p15, 0, %0, c9, c13, 0" : "=r" (val));	/* read */
+	asm volatile("mcr p15, 0, %0, c9, c13, 0" : : "r" (newval));	/* new value */
+	asm volatile("mcr p15, 0, %0, c9, c12, 1" : : "r" (den));	/* enable */
+	local_irq_restore(flags);
+
+	return val;
+}
+
+inline u32 armv7_cntn_read(unsigned int cnt, u32 reset_value)
+{
+	unsigned long flags;
+	u32 newval = -reset_value;
+	u32 sel = (cnt - CNT0);
+	u32 den = 1 << sel;
+	u32 oldval;
+
+	local_irq_save(flags);
+	asm volatile("mcr p15, 0, %0, c9, c12, 2" : : "r" (den));	/* disable */
+	asm volatile("mcr p15, 0, %0, c9, c12, 5" : : "r" (sel));	/* select */
+	asm volatile("mrc p15, 0, %0, c9, c13, 2" : "=r" (oldval));	/* read */
+	asm volatile("mcr p15, 0, %0, c9, c13, 2" : : "r" (newval));	/* new value */
+	asm volatile("mcr p15, 0, %0, c9, c12, 1" : : "r" (den));	/* enable */
+	local_irq_restore(flags);
+
+	return oldval;
+}
+
+static inline void armv7_pmnc_disable_interrupt(unsigned int cnt)
+{
+	u32 val = cnt ? (1 << (cnt - CNT0)) : (1 << 31);
+
+	asm volatile("mcr p15, 0, %0, c9, c14, 2" : : "r" (val));
+}
+
+inline u32 armv7_pmnc_reset_interrupt(void)
+{
+	/* Get and reset overflow status flags */
+	u32 flags;
+
+	asm volatile("mrc p15, 0, %0, c9, c12, 3" : "=r" (flags));
+	flags &= 0x8000003f;
+	asm volatile("mcr p15, 0, %0, c9, c12, 3" : : "r" (flags));
+	return flags;
+}
+
+static inline u32 armv7_pmnc_enable_counter(unsigned int cnt)
+{
+	u32 val = cnt ? (1 << (cnt - CNT0)) : CCNT_REG;
+
+	asm volatile("mcr p15, 0, %0, c9, c12, 1" : : "r" (val));
+	return cnt;
+}
+
+static inline u32 armv7_pmnc_disable_counter(unsigned int cnt)
+{
+	u32 val = cnt ? (1 << (cnt - CNT0)) : CCNT_REG;
+
+	asm volatile("mcr p15, 0, %0, c9, c12, 2" : : "r" (val));
+	return cnt;
+}
+
+static inline int armv7_pmnc_select_counter(unsigned int cnt)
+{
+	u32 val = (cnt - CNT0);
+
+	asm volatile("mcr p15, 0, %0, c9, c12, 5" : : "r" (val));
+	return cnt;
+}
+
+static inline void armv7_pmnc_write_evtsel(unsigned int cnt, u32 val)
+{
+	if (armv7_pmnc_select_counter(cnt) == cnt)
+		asm volatile("mcr p15, 0, %0, c9, c13, 1" : : "r" (val));
+}
+
+static int gator_events_armv7_create_files(struct super_block *sb, struct dentry *root)
+{
+	struct dentry *dir;
+	int i;
+
+	for (i = 0; i < pmnc_counters; i++) {
+		char buf[40];
+
+		if (i == 0)
+			snprintf(buf, sizeof(buf), "%s_ccnt", pmnc_name);
+		else
+			snprintf(buf, sizeof(buf), "%s_cnt%d", pmnc_name, i - 1);
+		dir = gatorfs_mkdir(sb, root, buf);
+		if (!dir)
+			return -1;
+		gatorfs_create_ulong(sb, dir, "enabled", &pmnc_enabled[i]);
+		gatorfs_create_ro_ulong(sb, dir, "key", &pmnc_key[i]);
+		if (i > 0)
+			gatorfs_create_ulong(sb, dir, "event", &pmnc_event[i]);
+	}
+
+	return 0;
+}
+
+static int gator_events_armv7_online(int **buffer, bool migrate)
+{
+	unsigned int cnt, len = 0, cpu = smp_processor_id();
+
+	if (armv7_pmnc_read() & PMNC_E)
+		armv7_pmnc_write(armv7_pmnc_read() & ~PMNC_E);
+
+	/* Initialize & Reset PMNC: C bit and P bit */
+	armv7_pmnc_write(PMNC_P | PMNC_C);
+
+	/* Reset overflow flags */
+	armv7_pmnc_reset_interrupt();
+
+	for (cnt = CCNT; cnt < CNTMAX; cnt++) {
+		unsigned long event;
+
+		if (!pmnc_enabled[cnt])
+			continue;
+
+		/* Disable counter */
+		armv7_pmnc_disable_counter(cnt);
+
+		event = pmnc_event[cnt] & 255;
+
+		/* Set event (if destined for PMNx counters), we don't need to set the event if it's a cycle count */
+		if (cnt != CCNT)
+			armv7_pmnc_write_evtsel(cnt, event);
+
+		armv7_pmnc_disable_interrupt(cnt);
+
+		/* Reset counter */
+		cnt ? armv7_cntn_read(cnt, 0) : armv7_ccnt_read(0);
+
+		/* Enable counter */
+		armv7_pmnc_enable_counter(cnt);
+	}
+
+	/* enable */
+	armv7_pmnc_write(armv7_pmnc_read() | PMNC_E);
+
+	/* return zero values, no need to read as the counters were just reset */
+	for (cnt = 0; cnt < pmnc_counters; cnt++) {
+		if (pmnc_enabled[cnt]) {
+			per_cpu(perfCnt, cpu)[len++] = pmnc_key[cnt];
+			per_cpu(perfCnt, cpu)[len++] = 0;
+		}
+	}
+
+	if (buffer)
+		*buffer = per_cpu(perfCnt, cpu);
+
+	return len;
+}
+
+static int gator_events_armv7_offline(int **buffer, bool migrate)
+{
+	/* disable all counters, including PMCCNTR; overflow IRQs will not be signaled */
+	armv7_pmnc_write(armv7_pmnc_read() & ~PMNC_E);
+
+	return 0;
+}
+
+static void gator_events_armv7_stop(void)
+{
+	unsigned int cnt;
+
+	for (cnt = CCNT; cnt < CNTMAX; cnt++) {
+		pmnc_enabled[cnt] = 0;
+		pmnc_event[cnt] = 0;
+	}
+}
+
+static int gator_events_armv7_read(int **buffer, bool sched_switch)
+{
+	int cnt, len = 0;
+	int cpu = smp_processor_id();
+
+	/* a context switch may occur before the online hotplug event, thus need to check that the pmu is enabled */
+	if (!(armv7_pmnc_read() & PMNC_E))
+		return 0;
+
+	for (cnt = 0; cnt < pmnc_counters; cnt++) {
+		if (pmnc_enabled[cnt]) {
+			int value;
+
+			if (cnt == CCNT)
+				value = armv7_ccnt_read(0);
+			else
+				value = armv7_cntn_read(cnt, 0);
+			per_cpu(perfCnt, cpu)[len++] = pmnc_key[cnt];
+			per_cpu(perfCnt, cpu)[len++] = value;
+		}
+	}
+
+	if (buffer)
+		*buffer = per_cpu(perfCnt, cpu);
+
+	return len;
+}
+
+static struct gator_interface gator_events_armv7_interface = {
+	.create_files = gator_events_armv7_create_files,
+	.stop = gator_events_armv7_stop,
+	.online = gator_events_armv7_online,
+	.offline = gator_events_armv7_offline,
+	.read = gator_events_armv7_read,
+};
+
+int gator_events_armv7_init(void)
+{
+	unsigned int cnt;
+
+	switch (gator_cpuid()) {
+	case CORTEX_A5:
+		pmnc_name = "ARMv7_Cortex_A5";
+		pmnc_counters = 2;
+		break;
+	case CORTEX_A7:
+		pmnc_name = "ARMv7_Cortex_A7";
+		pmnc_counters = 4;
+		break;
+	case CORTEX_A8:
+		pmnc_name = "ARMv7_Cortex_A8";
+		pmnc_counters = 4;
+		break;
+	case CORTEX_A9:
+		pmnc_name = "ARMv7_Cortex_A9";
+		pmnc_counters = 6;
+		break;
+	case CORTEX_A15:
+		pmnc_name = "ARMv7_Cortex_A15";
+		pmnc_counters = 6;
+		break;
+	/* ARM Cortex A17 is not supported by version of Linux before 3.0 */
+	default:
+		return -1;
+	}
+
+	pmnc_counters++;	/* CNT[n] + CCNT */
+
+	for (cnt = CCNT; cnt < CNTMAX; cnt++) {
+		pmnc_enabled[cnt] = 0;
+		pmnc_event[cnt] = 0;
+		pmnc_key[cnt] = gator_events_get_key();
+	}
+
+	return gator_events_install(&gator_events_armv7_interface);
+}
+
+#endif
diff --git a/drivers/gator/gator_events_block.c b/drivers/gator/gator_events_block.c
new file mode 100644
index 000000000000..a352a54afa02
--- /dev/null
+++ b/drivers/gator/gator_events_block.c
@@ -0,0 +1,160 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#include "gator.h"
+#include <trace/events/block.h>
+
+#define BLOCK_RQ_WR		0
+#define BLOCK_RQ_RD		1
+
+#define BLOCK_TOTAL		(BLOCK_RQ_RD+1)
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
+#define EVENTWRITE REQ_RW
+#else
+#define EVENTWRITE REQ_WRITE
+#endif
+
+static ulong block_rq_wr_enabled;
+static ulong block_rq_rd_enabled;
+static ulong block_rq_wr_key;
+static ulong block_rq_rd_key;
+static atomic_t blockCnt[BLOCK_TOTAL];
+static int blockGet[BLOCK_TOTAL * 4];
+
+/* Tracepoint changed in 3.15 backported to older kernels. The Makefile tries to autodetect the correct value, but if it fails change the #if below */
+#if OLD_BLOCK_RQ_COMPLETE
+GATOR_DEFINE_PROBE(block_rq_complete, TP_PROTO(struct request_queue *q, struct request *rq))
+#else
+GATOR_DEFINE_PROBE(block_rq_complete, TP_PROTO(struct request_queue *q, struct request *rq, unsigned int nr_bytes))
+#endif
+{
+	int write;
+	unsigned int size;
+
+	if (!rq)
+		return;
+
+	write = rq->cmd_flags & EVENTWRITE;
+#if OLD_BLOCK_RQ_COMPLETE
+	size = rq->resid_len;
+#else
+	size = nr_bytes;
+#endif
+
+	if (!size)
+		return;
+
+	if (write) {
+		if (block_rq_wr_enabled)
+			atomic_add(size, &blockCnt[BLOCK_RQ_WR]);
+	} else {
+		if (block_rq_rd_enabled)
+			atomic_add(size, &blockCnt[BLOCK_RQ_RD]);
+	}
+}
+
+static int gator_events_block_create_files(struct super_block *sb, struct dentry *root)
+{
+	struct dentry *dir;
+
+	/* block_complete_wr */
+	dir = gatorfs_mkdir(sb, root, "Linux_block_rq_wr");
+	if (!dir)
+		return -1;
+	gatorfs_create_ulong(sb, dir, "enabled", &block_rq_wr_enabled);
+	gatorfs_create_ro_ulong(sb, dir, "key", &block_rq_wr_key);
+
+	/* block_complete_rd */
+	dir = gatorfs_mkdir(sb, root, "Linux_block_rq_rd");
+	if (!dir)
+		return -1;
+	gatorfs_create_ulong(sb, dir, "enabled", &block_rq_rd_enabled);
+	gatorfs_create_ro_ulong(sb, dir, "key", &block_rq_rd_key);
+
+	return 0;
+}
+
+static int gator_events_block_start(void)
+{
+	/* register tracepoints */
+	if (block_rq_wr_enabled || block_rq_rd_enabled)
+		if (GATOR_REGISTER_TRACE(block_rq_complete))
+			goto fail_block_rq_exit;
+	pr_debug("gator: registered block event tracepoints\n");
+
+	return 0;
+
+	/* unregister tracepoints on error */
+fail_block_rq_exit:
+	pr_err("gator: block event tracepoints failed to activate, please verify that tracepoints are enabled in the linux kernel\n");
+
+	return -1;
+}
+
+static void gator_events_block_stop(void)
+{
+	if (block_rq_wr_enabled || block_rq_rd_enabled)
+		GATOR_UNREGISTER_TRACE(block_rq_complete);
+	pr_debug("gator: unregistered block event tracepoints\n");
+
+	block_rq_wr_enabled = 0;
+	block_rq_rd_enabled = 0;
+}
+
+static int gator_events_block_read(int **buffer, bool sched_switch)
+{
+	int len, value, data = 0;
+
+	if (!on_primary_core())
+		return 0;
+
+	len = 0;
+	if (block_rq_wr_enabled && (value = atomic_read(&blockCnt[BLOCK_RQ_WR])) > 0) {
+		atomic_sub(value, &blockCnt[BLOCK_RQ_WR]);
+		blockGet[len++] = block_rq_wr_key;
+		/* Indicates to Streamline that value bytes were written now, not since the last message */
+		blockGet[len++] = 0;
+		blockGet[len++] = block_rq_wr_key;
+		blockGet[len++] = value;
+		data += value;
+	}
+	if (block_rq_rd_enabled && (value = atomic_read(&blockCnt[BLOCK_RQ_RD])) > 0) {
+		atomic_sub(value, &blockCnt[BLOCK_RQ_RD]);
+		blockGet[len++] = block_rq_rd_key;
+		/* Indicates to Streamline that value bytes were read now, not since the last message */
+		blockGet[len++] = 0;
+		blockGet[len++] = block_rq_rd_key;
+		blockGet[len++] = value;
+		data += value;
+	}
+
+	if (buffer)
+		*buffer = blockGet;
+
+	return len;
+}
+
+static struct gator_interface gator_events_block_interface = {
+	.create_files = gator_events_block_create_files,
+	.start = gator_events_block_start,
+	.stop = gator_events_block_stop,
+	.read = gator_events_block_read,
+};
+
+int gator_events_block_init(void)
+{
+	block_rq_wr_enabled = 0;
+	block_rq_rd_enabled = 0;
+
+	block_rq_wr_key = gator_events_get_key();
+	block_rq_rd_key = gator_events_get_key();
+
+	return gator_events_install(&gator_events_block_interface);
+}
diff --git a/drivers/gator/gator_events_irq.c b/drivers/gator/gator_events_irq.c
new file mode 100644
index 000000000000..5221aac581b3
--- /dev/null
+++ b/drivers/gator/gator_events_irq.c
@@ -0,0 +1,163 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#include "gator.h"
+#include <trace/events/irq.h>
+
+#define HARDIRQ		0
+#define SOFTIRQ		1
+#define TOTALIRQ	(SOFTIRQ+1)
+
+static ulong hardirq_enabled;
+static ulong softirq_enabled;
+static ulong hardirq_key;
+static ulong softirq_key;
+static DEFINE_PER_CPU(atomic_t[TOTALIRQ], irqCnt);
+static DEFINE_PER_CPU(int[TOTALIRQ * 2], irqGet);
+
+GATOR_DEFINE_PROBE(irq_handler_exit,
+		   TP_PROTO(int irq, struct irqaction *action, int ret))
+{
+	atomic_inc(&per_cpu(irqCnt, get_physical_cpu())[HARDIRQ]);
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 37)
+GATOR_DEFINE_PROBE(softirq_exit, TP_PROTO(struct softirq_action *h, struct softirq_action *vec))
+#else
+GATOR_DEFINE_PROBE(softirq_exit, TP_PROTO(unsigned int vec_nr))
+#endif
+{
+	atomic_inc(&per_cpu(irqCnt, get_physical_cpu())[SOFTIRQ]);
+}
+
+static int gator_events_irq_create_files(struct super_block *sb, struct dentry *root)
+{
+	struct dentry *dir;
+
+	/* irq */
+	dir = gatorfs_mkdir(sb, root, "Linux_irq_irq");
+	if (!dir)
+		return -1;
+	gatorfs_create_ulong(sb, dir, "enabled", &hardirq_enabled);
+	gatorfs_create_ro_ulong(sb, dir, "key", &hardirq_key);
+
+	/* soft irq */
+	dir = gatorfs_mkdir(sb, root, "Linux_irq_softirq");
+	if (!dir)
+		return -1;
+	gatorfs_create_ulong(sb, dir, "enabled", &softirq_enabled);
+	gatorfs_create_ro_ulong(sb, dir, "key", &softirq_key);
+
+	return 0;
+}
+
+static int gator_events_irq_online(int **buffer, bool migrate)
+{
+	int len = 0, cpu = get_physical_cpu();
+
+	/* synchronization with the irq_exit functions is not necessary as the values are being reset */
+	if (hardirq_enabled) {
+		atomic_set(&per_cpu(irqCnt, cpu)[HARDIRQ], 0);
+		per_cpu(irqGet, cpu)[len++] = hardirq_key;
+		per_cpu(irqGet, cpu)[len++] = 0;
+	}
+
+	if (softirq_enabled) {
+		atomic_set(&per_cpu(irqCnt, cpu)[SOFTIRQ], 0);
+		per_cpu(irqGet, cpu)[len++] = softirq_key;
+		per_cpu(irqGet, cpu)[len++] = 0;
+	}
+
+	if (buffer)
+		*buffer = per_cpu(irqGet, cpu);
+
+	return len;
+}
+
+static int gator_events_irq_start(void)
+{
+	/* register tracepoints */
+	if (hardirq_enabled)
+		if (GATOR_REGISTER_TRACE(irq_handler_exit))
+			goto fail_hardirq_exit;
+	if (softirq_enabled)
+		if (GATOR_REGISTER_TRACE(softirq_exit))
+			goto fail_softirq_exit;
+	pr_debug("gator: registered irq tracepoints\n");
+
+	return 0;
+
+	/* unregister tracepoints on error */
+fail_softirq_exit:
+	if (hardirq_enabled)
+		GATOR_UNREGISTER_TRACE(irq_handler_exit);
+fail_hardirq_exit:
+	pr_err("gator: irq tracepoints failed to activate, please verify that tracepoints are enabled in the linux kernel\n");
+
+	return -1;
+}
+
+static void gator_events_irq_stop(void)
+{
+	if (hardirq_enabled)
+		GATOR_UNREGISTER_TRACE(irq_handler_exit);
+	if (softirq_enabled)
+		GATOR_UNREGISTER_TRACE(softirq_exit);
+	pr_debug("gator: unregistered irq tracepoints\n");
+
+	hardirq_enabled = 0;
+	softirq_enabled = 0;
+}
+
+static int gator_events_irq_read(int **buffer, bool sched_switch)
+{
+	int len, value;
+	int cpu = get_physical_cpu();
+
+	len = 0;
+	if (hardirq_enabled) {
+		value = atomic_read(&per_cpu(irqCnt, cpu)[HARDIRQ]);
+		atomic_sub(value, &per_cpu(irqCnt, cpu)[HARDIRQ]);
+
+		per_cpu(irqGet, cpu)[len++] = hardirq_key;
+		per_cpu(irqGet, cpu)[len++] = value;
+	}
+
+	if (softirq_enabled) {
+		value = atomic_read(&per_cpu(irqCnt, cpu)[SOFTIRQ]);
+		atomic_sub(value, &per_cpu(irqCnt, cpu)[SOFTIRQ]);
+
+		per_cpu(irqGet, cpu)[len++] = softirq_key;
+		per_cpu(irqGet, cpu)[len++] = value;
+	}
+
+	if (buffer)
+		*buffer = per_cpu(irqGet, cpu);
+
+	return len;
+}
+
+static struct gator_interface gator_events_irq_interface = {
+	.create_files = gator_events_irq_create_files,
+	.online = gator_events_irq_online,
+	.start = gator_events_irq_start,
+	.stop = gator_events_irq_stop,
+	.read = gator_events_irq_read,
+};
+
+int gator_events_irq_init(void)
+{
+	hardirq_key = gator_events_get_key();
+	softirq_key = gator_events_get_key();
+
+	hardirq_enabled = 0;
+	softirq_enabled = 0;
+
+	return gator_events_install(&gator_events_irq_interface);
+}
diff --git a/drivers/gator/gator_events_l2c-310.c b/drivers/gator/gator_events_l2c-310.c
new file mode 100644
index 000000000000..73aaac32327e
--- /dev/null
+++ b/drivers/gator/gator_events_l2c-310.c
@@ -0,0 +1,208 @@
+/**
+ * l2c310 (L2 Cache Controller) event counters for gator
+ *
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#if defined(CONFIG_OF)
+#include <linux/of.h>
+#include <linux/of_address.h>
+#endif
+#include <asm/hardware/cache-l2x0.h>
+
+#include "gator.h"
+
+#define L2C310_COUNTERS_NUM 2
+
+static struct {
+	unsigned long enabled;
+	unsigned long event;
+	unsigned long key;
+} l2c310_counters[L2C310_COUNTERS_NUM];
+
+static int l2c310_buffer[L2C310_COUNTERS_NUM * 2];
+
+static void __iomem *l2c310_base;
+
+static void gator_events_l2c310_reset_counters(void)
+{
+	u32 val = readl(l2c310_base + L2X0_EVENT_CNT_CTRL);
+
+	val |= ((1 << L2C310_COUNTERS_NUM) - 1) << 1;
+
+	writel(val, l2c310_base + L2X0_EVENT_CNT_CTRL);
+}
+
+static int gator_events_l2c310_create_files(struct super_block *sb,
+					    struct dentry *root)
+{
+	int i;
+
+	for (i = 0; i < L2C310_COUNTERS_NUM; i++) {
+		char buf[16];
+		struct dentry *dir;
+
+		snprintf(buf, sizeof(buf), "L2C-310_cnt%d", i);
+		dir = gatorfs_mkdir(sb, root, buf);
+		if (WARN_ON(!dir))
+			return -1;
+		gatorfs_create_ulong(sb, dir, "enabled",
+				     &l2c310_counters[i].enabled);
+		gatorfs_create_ulong(sb, dir, "event",
+				     &l2c310_counters[i].event);
+		gatorfs_create_ro_ulong(sb, dir, "key",
+					&l2c310_counters[i].key);
+	}
+
+	return 0;
+}
+
+static int gator_events_l2c310_start(void)
+{
+	static const unsigned long l2x0_event_cntx_cfg[L2C310_COUNTERS_NUM] = {
+		L2X0_EVENT_CNT0_CFG,
+		L2X0_EVENT_CNT1_CFG,
+	};
+	int i;
+
+	/* Counter event sources */
+	for (i = 0; i < L2C310_COUNTERS_NUM; i++)
+		writel((l2c310_counters[i].event & 0xf) << 2,
+		       l2c310_base + l2x0_event_cntx_cfg[i]);
+
+	gator_events_l2c310_reset_counters();
+
+	/* Event counter enable */
+	writel(1, l2c310_base + L2X0_EVENT_CNT_CTRL);
+
+	return 0;
+}
+
+static void gator_events_l2c310_stop(void)
+{
+	/* Event counter disable */
+	writel(0, l2c310_base + L2X0_EVENT_CNT_CTRL);
+}
+
+static int gator_events_l2c310_read(int **buffer, bool sched_switch)
+{
+	static const unsigned long l2x0_event_cntx_val[L2C310_COUNTERS_NUM] = {
+		L2X0_EVENT_CNT0_VAL,
+		L2X0_EVENT_CNT1_VAL,
+	};
+	int i;
+	int len = 0;
+
+	if (!on_primary_core())
+		return 0;
+
+	for (i = 0; i < L2C310_COUNTERS_NUM; i++) {
+		if (l2c310_counters[i].enabled) {
+			l2c310_buffer[len++] = l2c310_counters[i].key;
+			l2c310_buffer[len++] = readl(l2c310_base +
+						     l2x0_event_cntx_val[i]);
+		}
+	}
+
+	/* l2c310 counters are saturating, not wrapping in case of overflow */
+	gator_events_l2c310_reset_counters();
+
+	if (buffer)
+		*buffer = l2c310_buffer;
+
+	return len;
+}
+
+static struct gator_interface gator_events_l2c310_interface = {
+	.create_files = gator_events_l2c310_create_files,
+	.start = gator_events_l2c310_start,
+	.stop = gator_events_l2c310_stop,
+	.read = gator_events_l2c310_read,
+};
+
+#define L2C310_ADDR_PROBE (~0)
+
+MODULE_PARM_DESC(l2c310_addr, "L2C310 physical base address (0 to disable)");
+static unsigned long l2c310_addr = L2C310_ADDR_PROBE;
+module_param(l2c310_addr, ulong, 0444);
+
+static void __iomem *gator_events_l2c310_probe(void)
+{
+	phys_addr_t variants[] = {
+#if defined(CONFIG_ARCH_EXYNOS4) || defined(CONFIG_ARCH_S5PV310)
+		0x10502000,
+#endif
+#if defined(CONFIG_ARCH_OMAP4)
+		0x48242000,
+#endif
+#if defined(CONFIG_ARCH_TEGRA)
+		0x50043000,
+#endif
+#if defined(CONFIG_ARCH_U8500)
+		0xa0412000,
+#endif
+#if defined(CONFIG_ARCH_VEXPRESS)
+		0x1e00a000, /* A9x4 core tile (HBI-0191) */
+		0x2c0f0000, /* New memory map tiles */
+#endif
+	};
+	int i;
+	void __iomem *base;
+#if defined(CONFIG_OF)
+	struct device_node *node = of_find_all_nodes(NULL);
+
+	if (node) {
+		of_node_put(node);
+
+		node = of_find_compatible_node(NULL, NULL, "arm,pl310-cache");
+		base = of_iomap(node, 0);
+		of_node_put(node);
+
+		return base;
+	}
+#endif
+
+	for (i = 0; i < ARRAY_SIZE(variants); i++) {
+		base = ioremap(variants[i], SZ_4K);
+		if (base) {
+			u32 cache_id = readl(base + L2X0_CACHE_ID);
+
+			if ((cache_id & 0xff0003c0) == 0x410000c0)
+				return base;
+
+			iounmap(base);
+		}
+	}
+
+	return NULL;
+}
+
+int gator_events_l2c310_init(void)
+{
+	int i;
+
+	if (gator_cpuid() != CORTEX_A5 && gator_cpuid() != CORTEX_A9)
+		return -1;
+
+	if (l2c310_addr == L2C310_ADDR_PROBE)
+		l2c310_base = gator_events_l2c310_probe();
+	else if (l2c310_addr)
+		l2c310_base = ioremap(l2c310_addr, SZ_4K);
+
+	if (!l2c310_base)
+		return -1;
+
+	for (i = 0; i < L2C310_COUNTERS_NUM; i++) {
+		l2c310_counters[i].enabled = 0;
+		l2c310_counters[i].key = gator_events_get_key();
+	}
+
+	return gator_events_install(&gator_events_l2c310_interface);
+}
diff --git a/drivers/gator/gator_events_mali_4xx.c b/drivers/gator/gator_events_mali_4xx.c
new file mode 100644
index 000000000000..9cf43fe2c29b
--- /dev/null
+++ b/drivers/gator/gator_events_mali_4xx.c
@@ -0,0 +1,622 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "gator.h"
+
+#include <linux/module.h>
+#include <linux/time.h>
+#include <linux/math64.h>
+
+#include "linux/mali_linux_trace.h"
+
+#include "gator_events_mali_common.h"
+#include "gator_events_mali_4xx.h"
+
+/*
+* There have been four different variants of the comms between gator and Mali depending on driver version:
+* # | DDK vsn range             | Support                                                             | Notes
+*
+* 1 | (obsolete)                | No software counter support                                         | Obsolete patches
+* 2 | (obsolete)                | Tracepoint called for each separate s/w counter value as it appears | Obsolete patches
+* 3 | r3p0-04rel0 - r3p2-01rel2 | Single tracepoint for all s/w counters in a bundle.                 |
+* 4 | r3p2-01rel3 - date        | As above but with extensions for MP devices (Mali-450)              | At least r4p0-00rel1
+*/
+
+#if !defined(GATOR_MALI_INTERFACE_STYLE)
+#define GATOR_MALI_INTERFACE_STYLE (4)
+#endif
+
+#if GATOR_MALI_INTERFACE_STYLE == 1
+#error GATOR_MALI_INTERFACE_STYLE 1 is obsolete
+#elif GATOR_MALI_INTERFACE_STYLE == 2
+#error GATOR_MALI_INTERFACE_STYLE 2 is obsolete
+#elif GATOR_MALI_INTERFACE_STYLE >= 3
+/* Valid GATOR_MALI_INTERFACE_STYLE */
+#else
+#error Unknown GATOR_MALI_INTERFACE_STYLE option.
+#endif
+
+#if GATOR_MALI_INTERFACE_STYLE < 4
+#include "mali/mali_mjollnir_profiling_gator_api.h"
+#else
+#include "mali/mali_utgard_profiling_gator_api.h"
+#endif
+
+/*
+ * Check that the MALI_SUPPORT define is set to one of the allowable device codes.
+ */
+#if (MALI_SUPPORT != MALI_4xx)
+#error MALI_SUPPORT set to an invalid device code: expecting MALI_4xx
+#endif
+
+static const char mali_name[] = "4xx";
+
+/* gatorfs variables for counter enable state,
+ * the event the counter should count and the
+ * 'key' (a unique id set by gatord and returned
+ * by gator.ko)
+ */
+static unsigned long counter_enabled[NUMBER_OF_EVENTS];
+static unsigned long counter_event[NUMBER_OF_EVENTS];
+static unsigned long counter_key[NUMBER_OF_EVENTS];
+
+/* The data we have recorded */
+static u32 counter_data[NUMBER_OF_EVENTS];
+/* The address to sample (or 0 if samples are sent to us) */
+static u32 *counter_address[NUMBER_OF_EVENTS];
+
+/* An array used to return the data we recorded
+ * as key,value pairs hence the *2
+ */
+static int counter_dump[NUMBER_OF_EVENTS * 2];
+static int counter_prev[NUMBER_OF_EVENTS];
+static bool prev_set[NUMBER_OF_EVENTS];
+
+/* Note whether tracepoints have been registered */
+static int trace_registered;
+
+/*
+ * These numbers define the actual numbers of each block type that exist in the system. Initially
+ * these are set to the maxima defined above; if the driver is capable of being queried (newer
+ * drivers only) then the values may be revised.
+ */
+static unsigned int n_vp_cores = MAX_NUM_VP_CORES;
+static unsigned int n_l2_cores = MAX_NUM_L2_CACHE_CORES;
+static unsigned int n_fp_cores = MAX_NUM_FP_CORES;
+
+extern struct mali_counter mali_activity[2];
+static const char *const mali_activity_names[] = {
+	"fragment",
+	"vertex",
+};
+
+/**
+ * Returns non-zero if the given counter ID is an activity counter.
+ */
+static inline int is_activity_counter(unsigned int event_id)
+{
+	return (event_id >= FIRST_ACTIVITY_EVENT &&
+		event_id <= LAST_ACTIVITY_EVENT);
+}
+
+/**
+ * Returns non-zero if the given counter ID is a hardware counter.
+ */
+static inline int is_hw_counter(unsigned int event_id)
+{
+	return (event_id >= FIRST_HW_COUNTER && event_id <= LAST_HW_COUNTER);
+}
+
+/* Probe for hardware counter events */
+GATOR_DEFINE_PROBE(mali_hw_counter, TP_PROTO(unsigned int event_id, unsigned int value))
+{
+	if (is_hw_counter(event_id))
+		counter_data[event_id] = value;
+}
+
+GATOR_DEFINE_PROBE(mali_sw_counters, TP_PROTO(pid_t pid, pid_t tid, void *surface_id, unsigned int *counters))
+{
+	u32 i;
+
+	/* Copy over the values for those counters which are enabled. */
+	for (i = FIRST_SW_COUNTER; i <= LAST_SW_COUNTER; i++) {
+		if (counter_enabled[i])
+			counter_data[i] = (u32)(counters[i - FIRST_SW_COUNTER]);
+	}
+}
+
+/**
+ * Create a single filesystem entry for a specified event.
+ * @param sb the superblock
+ * @param root Filesystem root
+ * @param name The name of the entry to create
+ * @param event The ID of the event
+ * @param create_event_item boolean indicating whether to create an 'event' filesystem entry. True to create.
+ *
+ * @return 0 if ok, non-zero if the create failed.
+ */
+static int create_fs_entry(struct super_block *sb, struct dentry *root, const char *name, int event, int create_event_item)
+{
+	struct dentry *dir;
+
+	dir = gatorfs_mkdir(sb, root, name);
+
+	if (!dir)
+		return -1;
+
+	if (create_event_item)
+		gatorfs_create_ulong(sb, dir, "event", &counter_event[event]);
+
+	gatorfs_create_ulong(sb, dir, "enabled", &counter_enabled[event]);
+	gatorfs_create_ro_ulong(sb, dir, "key", &counter_key[event]);
+
+	return 0;
+}
+
+#if GATOR_MALI_INTERFACE_STYLE > 3
+/*
+ * Read the version info structure if available
+ */
+static void initialise_version_info(void)
+{
+	void (*mali_profiling_get_mali_version_symbol)(struct _mali_profiling_mali_version *values);
+
+	mali_profiling_get_mali_version_symbol = symbol_get(_mali_profiling_get_mali_version);
+
+	if (mali_profiling_get_mali_version_symbol) {
+		struct _mali_profiling_mali_version version_info;
+
+		pr_debug("gator: mali online _mali_profiling_get_mali_version symbol @ %p\n",
+				mali_profiling_get_mali_version_symbol);
+
+		/*
+		 * Revise the number of each different core type using information derived from the DDK.
+		 */
+		mali_profiling_get_mali_version_symbol(&version_info);
+
+		n_fp_cores = version_info.num_of_fp_cores;
+		n_vp_cores = version_info.num_of_vp_cores;
+		n_l2_cores = version_info.num_of_l2_cores;
+
+		/* Release the function - we're done with it. */
+		symbol_put(_mali_profiling_get_mali_version);
+	} else {
+		pr_err("gator: mali online _mali_profiling_get_mali_version symbol not found\n");
+		pr_err("gator:  check your Mali DDK version versus the GATOR_MALI_INTERFACE_STYLE setting\n");
+	}
+}
+#endif
+
+static int create_files(struct super_block *sb, struct dentry *root)
+{
+	int event;
+
+	char buf[40];
+	int core_id;
+	int counter_number;
+
+	pr_debug("gator: Initialising counters with style = %d\n", GATOR_MALI_INTERFACE_STYLE);
+
+#if GATOR_MALI_INTERFACE_STYLE > 3
+	/*
+	 * Initialise first: this sets up the number of cores available (on compatible DDK versions).
+	 * Ideally this would not need guarding but other parts of the code depend on the interface style being set
+	 * correctly; if it is not then the system can enter an inconsistent state.
+	 */
+	initialise_version_info();
+#endif
+
+	mali_activity[0].cores = n_fp_cores;
+	mali_activity[1].cores = n_vp_cores;
+	for (event = 0; event < ARRAY_SIZE(mali_activity); event++) {
+		if (gator_mali_create_file_system(mali_name, mali_activity_names[event], sb, root, &mali_activity[event], NULL) != 0)
+			return -1;
+	}
+
+	/* Vertex processor counters */
+	for (core_id = 0; core_id < n_vp_cores; core_id++) {
+		int activity_counter_id = ACTIVITY_VP_0;
+
+		snprintf(buf, sizeof(buf), "ARM_Mali-%s_VP_%d_active", mali_name, core_id);
+		if (create_fs_entry(sb, root, buf, activity_counter_id, 0) != 0)
+			return -1;
+
+		for (counter_number = 0; counter_number < 2; counter_number++) {
+			int counter_id = COUNTER_VP_0_C0 + (2 * core_id) + counter_number;
+
+			snprintf(buf, sizeof(buf), "ARM_Mali-%s_VP_%d_cnt%d", mali_name, core_id, counter_number);
+			if (create_fs_entry(sb, root, buf, counter_id, 1) != 0)
+				return -1;
+		}
+	}
+
+	/* Fragment processors' counters */
+	for (core_id = 0; core_id < n_fp_cores; core_id++) {
+		int activity_counter_id = ACTIVITY_FP_0 + core_id;
+
+		snprintf(buf, sizeof(buf), "ARM_Mali-%s_FP_%d_active", mali_name, core_id);
+		if (create_fs_entry(sb, root, buf, activity_counter_id, 0) != 0)
+			return -1;
+
+		for (counter_number = 0; counter_number < 2; counter_number++) {
+			int counter_id = COUNTER_FP_0_C0 + (2 * core_id) + counter_number;
+
+			snprintf(buf, sizeof(buf), "ARM_Mali-%s_FP_%d_cnt%d", mali_name, core_id, counter_number);
+			if (create_fs_entry(sb, root, buf, counter_id, 1) != 0)
+				return -1;
+		}
+	}
+
+	/* L2 Cache counters */
+	for (core_id = 0; core_id < n_l2_cores; core_id++) {
+		for (counter_number = 0; counter_number < 2; counter_number++) {
+			int counter_id = COUNTER_L2_0_C0 + (2 * core_id) + counter_number;
+
+			snprintf(buf, sizeof(buf), "ARM_Mali-%s_L2_%d_cnt%d", mali_name, core_id, counter_number);
+			if (create_fs_entry(sb, root, buf, counter_id, 1) != 0)
+				return -1;
+		}
+	}
+
+	/* Now set up the software counter entries */
+	for (event = FIRST_SW_COUNTER; event <= LAST_SW_COUNTER; event++) {
+		snprintf(buf, sizeof(buf), "ARM_Mali-%s_SW_%d", mali_name, event - FIRST_SW_COUNTER);
+
+		if (create_fs_entry(sb, root, buf, event, 0) != 0)
+			return -1;
+	}
+
+	/* Now set up the special counter entries */
+	snprintf(buf, sizeof(buf), "ARM_Mali-%s_Filmstrip_cnt0", mali_name);
+	if (create_fs_entry(sb, root, buf, COUNTER_FILMSTRIP, 1) != 0)
+		return -1;
+
+#ifdef DVFS_REPORTED_BY_DDK
+	snprintf(buf, sizeof(buf), "ARM_Mali-%s_Frequency", mali_name);
+	if (create_fs_entry(sb, root, buf, COUNTER_FREQUENCY, 1) != 0)
+		return -1;
+
+	snprintf(buf, sizeof(buf), "ARM_Mali-%s_Voltage", mali_name);
+	if (create_fs_entry(sb, root, buf, COUNTER_VOLTAGE, 1) != 0)
+		return -1;
+#endif
+
+	return 0;
+}
+
+/*
+ * Local store for the get_counters entry point into the DDK.
+ * This is stored here since it is used very regularly.
+ */
+static void (*mali_get_counters)(unsigned int *, unsigned int *, unsigned int *, unsigned int *);
+static u32 (*mali_get_l2_counters)(struct _mali_profiling_l2_counter_values *values);
+
+/*
+ * Examine list of counters between two index limits and determine if any one is enabled.
+ * Returns 1 if any counter is enabled, 0 if none is.
+ */
+static int is_any_counter_enabled(unsigned int first_counter, unsigned int last_counter)
+{
+	unsigned int i;
+
+	for (i = first_counter; i <= last_counter; i++) {
+		if (counter_enabled[i])
+			return 1;	/* At least one counter is enabled */
+	}
+
+	return 0;		/* No s/w counters enabled */
+}
+
+static void init_counters(unsigned int from_counter, unsigned int to_counter)
+{
+	unsigned int counter_id;
+
+	/* If a Mali driver is present and exporting the appropriate symbol
+	 * then we can request the HW counters (of which there are only 2)
+	 * be configured to count the desired events
+	 */
+	mali_profiling_set_event_type *mali_set_hw_event;
+
+	mali_set_hw_event = symbol_get(_mali_profiling_set_event);
+
+	if (mali_set_hw_event) {
+		pr_debug("gator: mali online _mali_profiling_set_event symbol @ %p\n", mali_set_hw_event);
+
+		for (counter_id = from_counter; counter_id <= to_counter; counter_id++) {
+			if (counter_enabled[counter_id])
+				mali_set_hw_event(counter_id, counter_event[counter_id]);
+			else
+				mali_set_hw_event(counter_id, 0xFFFFFFFF);
+		}
+
+		symbol_put(_mali_profiling_set_event);
+	} else {
+		pr_err("gator: mali online _mali_profiling_set_event symbol not found\n");
+	}
+}
+
+static void mali_counter_initialize(void)
+{
+	int i;
+
+	mali_profiling_control_type *mali_control;
+
+	init_counters(COUNTER_L2_0_C0, COUNTER_L2_0_C0 + (2 * n_l2_cores) - 1);
+	init_counters(COUNTER_VP_0_C0, COUNTER_VP_0_C0 + (2 * n_vp_cores) - 1);
+	init_counters(COUNTER_FP_0_C0, COUNTER_FP_0_C0 + (2 * n_fp_cores) - 1);
+
+	/* Generic control interface for Mali DDK. */
+	mali_control = symbol_get(_mali_profiling_control);
+	if (mali_control) {
+		/* The event attribute in the XML file keeps the actual frame rate. */
+		unsigned int rate = counter_event[COUNTER_FILMSTRIP] & 0xff;
+		unsigned int resize_factor = (counter_event[COUNTER_FILMSTRIP] >> 8) & 0xff;
+
+		pr_debug("gator: mali online _mali_profiling_control symbol @ %p\n", mali_control);
+
+		mali_control(SW_COUNTER_ENABLE, (is_any_counter_enabled(FIRST_SW_COUNTER, LAST_SW_COUNTER) ? 1 : 0));
+		mali_control(FBDUMP_CONTROL_ENABLE, (counter_enabled[COUNTER_FILMSTRIP] ? 1 : 0));
+		mali_control(FBDUMP_CONTROL_RATE, rate);
+		mali_control(FBDUMP_CONTROL_RESIZE_FACTOR, resize_factor);
+
+		pr_debug("gator: sent mali_control enabled=%d, rate=%d\n", (counter_enabled[COUNTER_FILMSTRIP] ? 1 : 0), rate);
+
+		symbol_put(_mali_profiling_control);
+	} else {
+		pr_err("gator: mali online _mali_profiling_control symbol not found\n");
+	}
+
+	mali_get_counters = symbol_get(_mali_profiling_get_counters);
+	if (mali_get_counters)
+		pr_debug("gator: mali online _mali_profiling_get_counters symbol @ %p\n", mali_get_counters);
+	else
+		pr_debug("gator WARNING: mali _mali_profiling_get_counters symbol not defined\n");
+
+	mali_get_l2_counters = symbol_get(_mali_profiling_get_l2_counters);
+	if (mali_get_l2_counters)
+		pr_debug("gator: mali online _mali_profiling_get_l2_counters symbol @ %p\n", mali_get_l2_counters);
+	else
+		pr_debug("gator WARNING: mali _mali_profiling_get_l2_counters symbol not defined\n");
+
+	if (!mali_get_counters && !mali_get_l2_counters) {
+		pr_debug("gator: WARNING: no L2 counters available\n");
+		n_l2_cores = 0;
+	}
+
+	/* Clear counters in the start */
+	for (i = 0; i < NUMBER_OF_EVENTS; i++) {
+		counter_data[i] = 0;
+		prev_set[i] = false;
+	}
+}
+
+static void mali_counter_deinitialize(void)
+{
+	mali_profiling_set_event_type *mali_set_hw_event;
+	mali_profiling_control_type *mali_control;
+
+	mali_set_hw_event = symbol_get(_mali_profiling_set_event);
+
+	if (mali_set_hw_event) {
+		int i;
+
+		pr_debug("gator: mali offline _mali_profiling_set_event symbol @ %p\n", mali_set_hw_event);
+		for (i = FIRST_HW_COUNTER; i <= LAST_HW_COUNTER; i++)
+			mali_set_hw_event(i, 0xFFFFFFFF);
+
+		symbol_put(_mali_profiling_set_event);
+	} else {
+		pr_err("gator: mali offline _mali_profiling_set_event symbol not found\n");
+	}
+
+	/* Generic control interface for Mali DDK. */
+	mali_control = symbol_get(_mali_profiling_control);
+
+	if (mali_control) {
+		pr_debug("gator: mali offline _mali_profiling_control symbol @ %p\n", mali_control);
+
+		/* Reset the DDK state - disable counter collection */
+		mali_control(SW_COUNTER_ENABLE, 0);
+
+		mali_control(FBDUMP_CONTROL_ENABLE, 0);
+
+		symbol_put(_mali_profiling_control);
+	} else {
+		pr_err("gator: mali offline _mali_profiling_control symbol not found\n");
+	}
+
+	if (mali_get_counters)
+		symbol_put(_mali_profiling_get_counters);
+
+	if (mali_get_l2_counters)
+		symbol_put(_mali_profiling_get_l2_counters);
+}
+
+static int start(void)
+{
+	/* register tracepoints */
+	if (GATOR_REGISTER_TRACE(mali_hw_counter)) {
+		pr_err("gator: mali_hw_counter tracepoint failed to activate\n");
+		return -1;
+	}
+
+	/* For Mali drivers with built-in support. */
+	if (GATOR_REGISTER_TRACE(mali_sw_counters)) {
+		pr_err("gator: mali_sw_counters tracepoint failed to activate\n");
+		return -1;
+	}
+
+	trace_registered = 1;
+
+	mali_counter_initialize();
+	return 0;
+}
+
+static void stop(void)
+{
+	unsigned int cnt;
+
+	pr_debug("gator: mali stop\n");
+
+	if (trace_registered) {
+		GATOR_UNREGISTER_TRACE(mali_hw_counter);
+
+		/* For Mali drivers with built-in support. */
+		GATOR_UNREGISTER_TRACE(mali_sw_counters);
+
+		pr_debug("gator: mali timeline tracepoint deactivated\n");
+
+		trace_registered = 0;
+	}
+
+	for (cnt = 0; cnt < NUMBER_OF_EVENTS; cnt++) {
+		counter_enabled[cnt] = 0;
+		counter_event[cnt] = 0;
+		counter_address[cnt] = NULL;
+	}
+
+	mali_counter_deinitialize();
+}
+
+static void dump_counters(unsigned int from_counter, unsigned int to_counter, unsigned int *len)
+{
+	unsigned int counter_id;
+
+	for (counter_id = from_counter; counter_id <= to_counter; counter_id++) {
+		if (counter_enabled[counter_id]) {
+			counter_dump[(*len)++] = counter_key[counter_id];
+			counter_dump[(*len)++] = counter_data[counter_id];
+
+			counter_data[counter_id] = 0;
+		}
+	}
+}
+
+static int read(int **buffer, bool sched_switch)
+{
+	int len = 0;
+
+	if (!on_primary_core())
+		return 0;
+
+	/* Read the L2 C0 and C1 here. */
+	if (n_l2_cores > 0 && is_any_counter_enabled(COUNTER_L2_0_C0, COUNTER_L2_0_C0 + (2 * n_l2_cores))) {
+		unsigned int unavailable_l2_caches = 0;
+		struct _mali_profiling_l2_counter_values cache_values;
+		unsigned int cache_id;
+		struct _mali_profiling_core_counters *per_core;
+
+		/* Poke the driver to get the counter values - older style; only one L2 cache */
+		if (mali_get_l2_counters) {
+			unavailable_l2_caches = mali_get_l2_counters(&cache_values);
+		} else if (mali_get_counters) {
+			per_core = &cache_values.cores[0];
+			mali_get_counters(&per_core->source0, &per_core->value0, &per_core->source1, &per_core->value1);
+		} else {
+			/* This should never happen, as n_l2_caches is only set > 0 if one of the above functions is found. */
+		}
+
+		/* Fill in the two cache counter values for each cache block. */
+		for (cache_id = 0; cache_id < n_l2_cores; cache_id++) {
+			unsigned int counter_id_0 = COUNTER_L2_0_C0 + (2 * cache_id);
+			unsigned int counter_id_1 = counter_id_0 + 1;
+
+			if ((1 << cache_id) & unavailable_l2_caches)
+				continue; /* This cache is unavailable (powered-off, possibly). */
+
+			per_core = &cache_values.cores[cache_id];
+
+			if (counter_enabled[counter_id_0] && prev_set[counter_id_0]) {
+				/* Calculate and save src0's counter val0 */
+				counter_dump[len++] = counter_key[counter_id_0];
+				counter_dump[len++] = per_core->value0 - counter_prev[counter_id_0];
+			}
+
+			if (counter_enabled[counter_id_1] && prev_set[counter_id_1]) {
+				/* Calculate and save src1's counter val1 */
+				counter_dump[len++] = counter_key[counter_id_1];
+				counter_dump[len++] = per_core->value1 - counter_prev[counter_id_1];
+			}
+
+			/* Save the previous values for the counters. */
+			counter_prev[counter_id_0] = per_core->value0;
+			prev_set[counter_id_0] = true;
+			counter_prev[counter_id_1] = per_core->value1;
+			prev_set[counter_id_1] = true;
+		}
+	}
+
+	/* Process other (non-timeline) counters. */
+	dump_counters(COUNTER_VP_0_C0, COUNTER_VP_0_C0 + (2 * n_vp_cores) - 1, &len);
+	dump_counters(COUNTER_FP_0_C0, COUNTER_FP_0_C0 + (2 * n_fp_cores) - 1, &len);
+
+	dump_counters(FIRST_SW_COUNTER, LAST_SW_COUNTER, &len);
+
+#ifdef DVFS_REPORTED_BY_DDK
+	{
+		int cnt;
+		/*
+		 * Add in the voltage and frequency counters if enabled. Note
+		 * that, since these are actually passed as events, the counter
+		 * value should not be cleared.
+		 */
+		cnt = COUNTER_FREQUENCY;
+		if (counter_enabled[cnt]) {
+			counter_dump[len++] = counter_key[cnt];
+			counter_dump[len++] = counter_data[cnt];
+		}
+
+		cnt = COUNTER_VOLTAGE;
+		if (counter_enabled[cnt]) {
+			counter_dump[len++] = counter_key[cnt];
+			counter_dump[len++] = counter_data[cnt];
+		}
+	}
+#endif
+
+	if (buffer)
+		*buffer = counter_dump;
+
+	return len;
+}
+
+static struct gator_interface gator_events_mali_interface = {
+	.create_files = create_files,
+	.start = start,
+	.stop = stop,
+	.read = read,
+};
+
+extern void gator_events_mali_log_dvfs_event(unsigned int frequency_mhz, unsigned int voltage_mv)
+{
+#ifdef DVFS_REPORTED_BY_DDK
+	counter_data[COUNTER_FREQUENCY] = frequency_mhz;
+	counter_data[COUNTER_VOLTAGE] = voltage_mv;
+#endif
+}
+
+int gator_events_mali_init(void)
+{
+	unsigned int cnt;
+
+	pr_debug("gator: mali init\n");
+
+	gator_mali_initialise_counters(mali_activity, ARRAY_SIZE(mali_activity));
+
+	for (cnt = 0; cnt < NUMBER_OF_EVENTS; cnt++) {
+		counter_enabled[cnt] = 0;
+		counter_event[cnt] = 0;
+		counter_key[cnt] = gator_events_get_key();
+		counter_address[cnt] = NULL;
+		counter_data[cnt] = 0;
+	}
+
+	trace_registered = 0;
+
+	return gator_events_install(&gator_events_mali_interface);
+}
diff --git a/drivers/gator/gator_events_mali_4xx.h b/drivers/gator/gator_events_mali_4xx.h
new file mode 100644
index 000000000000..976ca8c4cfa1
--- /dev/null
+++ b/drivers/gator/gator_events_mali_4xx.h
@@ -0,0 +1,18 @@
+/**
+ * Copyright (C) ARM Limited 2011-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+/*
+ * Header contains common definitions for the Mali-4xx processors.
+ */
+#if !defined(GATOR_EVENTS_MALI_4xx_H)
+#define GATOR_EVENTS_MALI_4xx_H
+
+extern void gator_events_mali_log_dvfs_event(unsigned int d0, unsigned int d1);
+
+#endif /* GATOR_EVENTS_MALI_4xx_H */
diff --git a/drivers/gator/gator_events_mali_common.c b/drivers/gator/gator_events_mali_common.c
new file mode 100644
index 000000000000..1af87d649afe
--- /dev/null
+++ b/drivers/gator/gator_events_mali_common.c
@@ -0,0 +1,72 @@
+/**
+ * Copyright (C) ARM Limited 2012-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+#include "gator_events_mali_common.h"
+
+extern int gator_mali_create_file_system(const char *mali_name, const char *event_name, struct super_block *sb, struct dentry *root, struct mali_counter *counter, unsigned long *event)
+{
+	int err;
+	char buf[255];
+	struct dentry *dir;
+
+	/* If the counter name is empty ignore it */
+	if (strlen(event_name) != 0) {
+		/* Set up the filesystem entry for this event. */
+		if (mali_name == NULL)
+			snprintf(buf, sizeof(buf), "ARM_Mali-%s", event_name);
+		else
+			snprintf(buf, sizeof(buf), "ARM_Mali-%s_%s", mali_name, event_name);
+
+		dir = gatorfs_mkdir(sb, root, buf);
+
+		if (dir == NULL) {
+			pr_debug("gator: %s: error creating file system for: %s (%s)\n", mali_name, event_name, buf);
+			return -1;
+		}
+
+		err = gatorfs_create_ulong(sb, dir, "enabled", &counter->enabled);
+		if (err != 0) {
+			pr_debug("gator: %s: error calling gatorfs_create_ulong for: %s (%s)\n", mali_name, event_name, buf);
+			return -1;
+		}
+		err = gatorfs_create_ro_ulong(sb, dir, "key", &counter->key);
+		if (err != 0) {
+			pr_debug("gator: %s: error calling gatorfs_create_ro_ulong for: %s (%s)\n", mali_name, event_name, buf);
+			return -1;
+		}
+		if (counter->cores != -1) {
+			err = gatorfs_create_ro_ulong(sb, dir, "cores", &counter->cores);
+			if (err != 0) {
+				pr_debug("gator: %s: error calling gatorfs_create_ro_ulong for: %s (%s)\n", mali_name, event_name, buf);
+				return -1;
+			}
+		}
+		if (event != NULL) {
+			err = gatorfs_create_ulong(sb, dir, "event", event);
+			if (err != 0) {
+				pr_debug("gator: %s: error calling gatorfs_create_ro_ulong for: %s (%s)\n", mali_name, event_name, buf);
+				return -1;
+			}
+		}
+	}
+
+	return 0;
+}
+
+extern void gator_mali_initialise_counters(struct mali_counter counters[], unsigned int n_counters)
+{
+	unsigned int cnt;
+
+	for (cnt = 0; cnt < n_counters; cnt++) {
+		struct mali_counter *counter = &counters[cnt];
+
+		counter->key = gator_events_get_key();
+		counter->enabled = 0;
+		counter->cores = -1;
+	}
+}
diff --git a/drivers/gator/gator_events_mali_common.h b/drivers/gator/gator_events_mali_common.h
new file mode 100644
index 000000000000..e7082e62fe88
--- /dev/null
+++ b/drivers/gator/gator_events_mali_common.h
@@ -0,0 +1,77 @@
+/**
+ * Copyright (C) ARM Limited 2012-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#if !defined(GATOR_EVENTS_MALI_COMMON_H)
+#define GATOR_EVENTS_MALI_COMMON_H
+
+#include "gator.h"
+
+#include <linux/module.h>
+#include <linux/time.h>
+#include <linux/math64.h>
+#include <linux/slab.h>
+#include <linux/io.h>
+
+/* Ensure that MALI_SUPPORT has been defined to something. */
+#ifndef MALI_SUPPORT
+#error MALI_SUPPORT not defined!
+#endif
+
+/* Values for the supported activity event types */
+#define ACTIVITY_START  (1)
+#define ACTIVITY_STOP   (2)
+
+/*
+ * Runtime state information for a counter.
+ */
+struct mali_counter {
+	/* 'key' (a unique id set by gatord and returned by gator.ko) */
+	unsigned long key;
+	/* counter enable state */
+	unsigned long enabled;
+	/* for activity counters, the number of cores, otherwise -1 */
+	unsigned long cores;
+};
+
+/*
+ * Mali-4xx
+ */
+typedef int mali_profiling_set_event_type(unsigned int, int);
+typedef void mali_profiling_control_type(unsigned int, unsigned int);
+
+/*
+ * Driver entry points for functions called directly by gator.
+ */
+extern int _mali_profiling_set_event(unsigned int, int);
+extern void _mali_profiling_control(unsigned int, unsigned int);
+extern void _mali_profiling_get_counters(unsigned int *, unsigned int *, unsigned int *, unsigned int *);
+
+/**
+ * Creates a filesystem entry under /dev/gator relating to the specified event name and key, and
+ * associate the key/enable values with this entry point.
+ *
+ * @param event_name The name of the event.
+ * @param sb Linux super block
+ * @param root Directory under which the entry will be created.
+ * @param counter_key Ptr to location which will be associated with the counter key.
+ * @param counter_enabled Ptr to location which will be associated with the counter enable state.
+ *
+ * @return 0 if entry point was created, non-zero if not.
+ */
+extern int gator_mali_create_file_system(const char *mali_name, const char *event_name, struct super_block *sb, struct dentry *root, struct mali_counter *counter, unsigned long *event);
+
+/**
+ * Initializes the counter array.
+ *
+ * @param keys The array of counters
+ * @param n_counters The number of entries in each of the arrays.
+ */
+extern void gator_mali_initialise_counters(struct mali_counter counters[], unsigned int n_counters);
+
+#endif /* GATOR_EVENTS_MALI_COMMON_H  */
diff --git a/drivers/gator/gator_events_mali_midgard.c b/drivers/gator/gator_events_mali_midgard.c
new file mode 100644
index 000000000000..0aec906d7ae5
--- /dev/null
+++ b/drivers/gator/gator_events_mali_midgard.c
@@ -0,0 +1,562 @@
+/**
+ * Copyright (C) ARM Limited 2011-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#include "gator.h"
+
+#include <linux/module.h>
+#include <linux/time.h>
+#include <linux/math64.h>
+#include <linux/slab.h>
+#include <linux/io.h>
+
+#ifdef MALI_DIR_MIDGARD
+/* New DDK Directory structure with kernel/drivers/gpu/arm/midgard*/
+#include "mali_linux_trace.h"
+#else
+/* Old DDK Directory structure with kernel/drivers/gpu/arm/t6xx*/
+#include "linux/mali_linux_trace.h"
+#endif
+
+#include "gator_events_mali_common.h"
+
+/*
+ * Check that the MALI_SUPPORT define is set to one of the allowable device codes.
+ */
+#if (MALI_SUPPORT != MALI_MIDGARD)
+#error MALI_SUPPORT set to an invalid device code: expecting MALI_MIDGARD
+#endif
+
+static const char mali_name[] = "Midgard";
+
+/* Counters for Mali-Midgard:
+ *
+ *  - Timeline events
+ *    They are tracepoints, but instead of reporting a number they report a START/STOP event.
+ *    They are reported in Streamline as number of microseconds while that particular counter was active.
+ *
+ *  - SW counters
+ *    They are tracepoints reporting a particular number.
+ *    They are accumulated in sw_counter_data array until they are passed to Streamline, then they are zeroed.
+ *
+ *  - Accumulators
+ *    They are the same as software counters but their value is not zeroed.
+ */
+
+/* Timeline (start/stop) activity */
+static const char *const timeline_event_names[] = {
+	"PM_SHADER_0",
+	"PM_SHADER_1",
+	"PM_SHADER_2",
+	"PM_SHADER_3",
+	"PM_SHADER_4",
+	"PM_SHADER_5",
+	"PM_SHADER_6",
+	"PM_SHADER_7",
+	"PM_TILER_0",
+	"PM_L2_0",
+	"PM_L2_1",
+	"MMU_AS_0",
+	"MMU_AS_1",
+	"MMU_AS_2",
+	"MMU_AS_3"
+};
+
+enum {
+	PM_SHADER_0 = 0,
+	PM_SHADER_1,
+	PM_SHADER_2,
+	PM_SHADER_3,
+	PM_SHADER_4,
+	PM_SHADER_5,
+	PM_SHADER_6,
+	PM_SHADER_7,
+	PM_TILER_0,
+	PM_L2_0,
+	PM_L2_1,
+	MMU_AS_0,
+	MMU_AS_1,
+	MMU_AS_2,
+	MMU_AS_3
+};
+/* The number of shader blocks in the enum above */
+#define NUM_PM_SHADER (8)
+
+/* Software Counters */
+static const char *const software_counter_names[] = {
+	"MMU_PAGE_FAULT_0",
+	"MMU_PAGE_FAULT_1",
+	"MMU_PAGE_FAULT_2",
+	"MMU_PAGE_FAULT_3"
+};
+
+enum {
+	MMU_PAGE_FAULT_0 = 0,
+	MMU_PAGE_FAULT_1,
+	MMU_PAGE_FAULT_2,
+	MMU_PAGE_FAULT_3
+};
+
+/* Software Counters */
+static const char *const accumulators_names[] = {
+	"TOTAL_ALLOC_PAGES"
+};
+
+enum {
+	TOTAL_ALLOC_PAGES = 0
+};
+
+#define FIRST_TIMELINE_EVENT (0)
+#define NUMBER_OF_TIMELINE_EVENTS (sizeof(timeline_event_names) / sizeof(timeline_event_names[0]))
+#define FIRST_SOFTWARE_COUNTER (FIRST_TIMELINE_EVENT + NUMBER_OF_TIMELINE_EVENTS)
+#define NUMBER_OF_SOFTWARE_COUNTERS (sizeof(software_counter_names) / sizeof(software_counter_names[0]))
+#define FIRST_ACCUMULATOR (FIRST_SOFTWARE_COUNTER + NUMBER_OF_SOFTWARE_COUNTERS)
+#define NUMBER_OF_ACCUMULATORS (sizeof(accumulators_names) / sizeof(accumulators_names[0]))
+#define FILMSTRIP (NUMBER_OF_TIMELINE_EVENTS + NUMBER_OF_SOFTWARE_COUNTERS + NUMBER_OF_ACCUMULATORS)
+#define NUMBER_OF_EVENTS (NUMBER_OF_TIMELINE_EVENTS + NUMBER_OF_SOFTWARE_COUNTERS + NUMBER_OF_ACCUMULATORS + 1)
+
+/*
+ * gatorfs variables for counter enable state
+ */
+static struct mali_counter counters[NUMBER_OF_EVENTS];
+static unsigned long filmstrip_event;
+
+/* An array used to return the data we recorded
+ * as key,value pairs hence the *2
+ */
+static int counter_dump[NUMBER_OF_EVENTS * 2];
+
+/*
+ * Array holding counter start times (in ns) for each counter. A zero
+ * here indicates that the activity monitored by this counter is not
+ * running.
+ */
+static struct timespec timeline_event_starttime[NUMBER_OF_TIMELINE_EVENTS];
+
+/* The data we have recorded */
+static unsigned int timeline_data[NUMBER_OF_TIMELINE_EVENTS];
+static unsigned int sw_counter_data[NUMBER_OF_SOFTWARE_COUNTERS];
+static unsigned int accumulators_data[NUMBER_OF_ACCUMULATORS];
+
+/* Hold the previous timestamp, used to calculate the sample interval. */
+static struct timespec prev_timestamp;
+
+/**
+ * Returns the timespan (in microseconds) between the two specified timestamps.
+ *
+ * @param start Ptr to the start timestamp
+ * @param end Ptr to the end timestamp
+ *
+ * @return Number of microseconds between the two timestamps (can be negative if start follows end).
+ */
+static inline long get_duration_us(const struct timespec *start, const struct timespec *end)
+{
+	long event_duration_us = (end->tv_nsec - start->tv_nsec) / 1000;
+
+	event_duration_us += (end->tv_sec - start->tv_sec) * 1000000;
+
+	return event_duration_us;
+}
+
+static void record_timeline_event(unsigned int timeline_index, unsigned int type)
+{
+	struct timespec event_timestamp;
+	struct timespec *event_start = &timeline_event_starttime[timeline_index];
+
+	switch (type) {
+	case ACTIVITY_START:
+		/* Get the event time... */
+		getnstimeofday(&event_timestamp);
+
+		/* Remember the start time if the activity is not already started */
+		if (event_start->tv_sec == 0)
+			*event_start = event_timestamp;	/* Structure copy */
+		break;
+
+	case ACTIVITY_STOP:
+		/* if the counter was started... */
+		if (event_start->tv_sec != 0) {
+			/* Get the event time... */
+			getnstimeofday(&event_timestamp);
+
+			/* Accumulate the duration in us */
+			timeline_data[timeline_index] += get_duration_us(event_start, &event_timestamp);
+
+			/* Reset the start time to indicate the activity is stopped. */
+			event_start->tv_sec = 0;
+		}
+		break;
+
+	default:
+		/* Other activity events are ignored. */
+		break;
+	}
+}
+
+/*
+ * Documentation about the following tracepoints is in mali_linux_trace.h
+ */
+
+GATOR_DEFINE_PROBE(mali_pm_status, TP_PROTO(unsigned int event_id, unsigned long long value))
+{
+#define SHADER_PRESENT_LO       0x100	/* (RO) Shader core present bitmap, low word */
+#define TILER_PRESENT_LO        0x110	/* (RO) Tiler core present bitmap, low word */
+#define L2_PRESENT_LO           0x120	/* (RO) Level 2 cache present bitmap, low word */
+#define BIT_AT(value, pos) ((value >> pos) & 1)
+
+	static unsigned long long previous_shader_bitmask;
+	static unsigned long long previous_tiler_bitmask;
+	static unsigned long long previous_l2_bitmask;
+
+	switch (event_id) {
+	case SHADER_PRESENT_LO:
+		{
+			unsigned long long changed_bitmask = previous_shader_bitmask ^ value;
+			int pos;
+
+			for (pos = 0; pos < NUM_PM_SHADER; ++pos) {
+				if (BIT_AT(changed_bitmask, pos))
+					record_timeline_event(PM_SHADER_0 + pos, BIT_AT(value, pos) ? ACTIVITY_START : ACTIVITY_STOP);
+			}
+
+			previous_shader_bitmask = value;
+			break;
+		}
+
+	case TILER_PRESENT_LO:
+		{
+			unsigned long long changed = previous_tiler_bitmask ^ value;
+
+			if (BIT_AT(changed, 0))
+				record_timeline_event(PM_TILER_0, BIT_AT(value, 0) ? ACTIVITY_START : ACTIVITY_STOP);
+
+			previous_tiler_bitmask = value;
+			break;
+		}
+
+	case L2_PRESENT_LO:
+		{
+			unsigned long long changed = previous_l2_bitmask ^ value;
+
+			if (BIT_AT(changed, 0))
+				record_timeline_event(PM_L2_0, BIT_AT(value, 0) ? ACTIVITY_START : ACTIVITY_STOP);
+			if (BIT_AT(changed, 4))
+				record_timeline_event(PM_L2_1, BIT_AT(value, 4) ? ACTIVITY_START : ACTIVITY_STOP);
+
+			previous_l2_bitmask = value;
+			break;
+		}
+
+	default:
+		/* No other blocks are supported at present */
+		break;
+	}
+
+#undef SHADER_PRESENT_LO
+#undef TILER_PRESENT_LO
+#undef L2_PRESENT_LO
+#undef BIT_AT
+}
+
+GATOR_DEFINE_PROBE(mali_page_fault_insert_pages, TP_PROTO(int event_id, unsigned long value))
+{
+	/* We add to the previous since we may receive many tracepoints in one sample period */
+	sw_counter_data[MMU_PAGE_FAULT_0 + event_id] += value;
+}
+
+GATOR_DEFINE_PROBE(mali_mmu_as_in_use, TP_PROTO(int event_id))
+{
+	record_timeline_event(MMU_AS_0 + event_id, ACTIVITY_START);
+}
+
+GATOR_DEFINE_PROBE(mali_mmu_as_released, TP_PROTO(int event_id))
+{
+	record_timeline_event(MMU_AS_0 + event_id, ACTIVITY_STOP);
+}
+
+GATOR_DEFINE_PROBE(mali_total_alloc_pages_change, TP_PROTO(long long int event_id))
+{
+	accumulators_data[TOTAL_ALLOC_PAGES] = event_id;
+}
+
+static int create_files(struct super_block *sb, struct dentry *root)
+{
+	int event;
+	/*
+	 * Create the filesystem for all events
+	 */
+	int counter_index = 0;
+	mali_profiling_control_type *mali_control;
+
+	for (event = FIRST_TIMELINE_EVENT; event < FIRST_TIMELINE_EVENT + NUMBER_OF_TIMELINE_EVENTS; event++) {
+		if (gator_mali_create_file_system(mali_name, timeline_event_names[counter_index], sb, root, &counters[event], NULL) != 0)
+			return -1;
+		counter_index++;
+	}
+	counter_index = 0;
+	for (event = FIRST_SOFTWARE_COUNTER; event < FIRST_SOFTWARE_COUNTER + NUMBER_OF_SOFTWARE_COUNTERS; event++) {
+		if (gator_mali_create_file_system(mali_name, software_counter_names[counter_index], sb, root, &counters[event], NULL) != 0)
+			return -1;
+		counter_index++;
+	}
+	counter_index = 0;
+	for (event = FIRST_ACCUMULATOR; event < FIRST_ACCUMULATOR + NUMBER_OF_ACCUMULATORS; event++) {
+		if (gator_mali_create_file_system(mali_name, accumulators_names[counter_index], sb, root, &counters[event], NULL) != 0)
+			return -1;
+		counter_index++;
+	}
+
+	mali_control = symbol_get(_mali_profiling_control);
+	if (mali_control) {
+		if (gator_mali_create_file_system(mali_name, "Filmstrip_cnt0", sb, root, &counters[FILMSTRIP], &filmstrip_event) != 0)
+			return -1;
+		symbol_put(_mali_profiling_control);
+	}
+
+	return 0;
+}
+
+static int register_tracepoints(void)
+{
+	if (GATOR_REGISTER_TRACE(mali_pm_status)) {
+		pr_debug("gator: Mali-Midgard: mali_pm_status tracepoint failed to activate\n");
+		return 0;
+	}
+
+	if (GATOR_REGISTER_TRACE(mali_page_fault_insert_pages)) {
+		pr_debug("gator: Mali-Midgard: mali_page_fault_insert_pages tracepoint failed to activate\n");
+		return 0;
+	}
+
+	if (GATOR_REGISTER_TRACE(mali_mmu_as_in_use)) {
+		pr_debug("gator: Mali-Midgard: mali_mmu_as_in_use tracepoint failed to activate\n");
+		return 0;
+	}
+
+	if (GATOR_REGISTER_TRACE(mali_mmu_as_released)) {
+		pr_debug("gator: Mali-Midgard: mali_mmu_as_released tracepoint failed to activate\n");
+		return 0;
+	}
+
+	if (GATOR_REGISTER_TRACE(mali_total_alloc_pages_change)) {
+		pr_debug("gator: Mali-Midgard: mali_total_alloc_pages_change tracepoint failed to activate\n");
+		return 0;
+	}
+
+	pr_debug("gator: Mali-Midgard: start\n");
+	pr_debug("gator: Mali-Midgard: mali_pm_status probe is at %p\n", &probe_mali_pm_status);
+	pr_debug("gator: Mali-Midgard: mali_page_fault_insert_pages probe is at %p\n", &probe_mali_page_fault_insert_pages);
+	pr_debug("gator: Mali-Midgard: mali_mmu_as_in_use probe is at %p\n", &probe_mali_mmu_as_in_use);
+	pr_debug("gator: Mali-Midgard: mali_mmu_as_released probe is at %p\n", &probe_mali_mmu_as_released);
+	pr_debug("gator: Mali-Midgard: mali_total_alloc_pages_change probe is at %p\n", &probe_mali_total_alloc_pages_change);
+
+	return 1;
+}
+
+static int start(void)
+{
+	unsigned int cnt;
+	mali_profiling_control_type *mali_control;
+
+	/* Clean all data for the next capture */
+	for (cnt = 0; cnt < NUMBER_OF_TIMELINE_EVENTS; cnt++) {
+		timeline_event_starttime[cnt].tv_sec = timeline_event_starttime[cnt].tv_nsec = 0;
+		timeline_data[cnt] = 0;
+	}
+
+	for (cnt = 0; cnt < NUMBER_OF_SOFTWARE_COUNTERS; cnt++)
+		sw_counter_data[cnt] = 0;
+
+	for (cnt = 0; cnt < NUMBER_OF_ACCUMULATORS; cnt++)
+		accumulators_data[cnt] = 0;
+
+	/* Register tracepoints */
+	if (register_tracepoints() == 0)
+		return -1;
+
+	/* Generic control interface for Mali DDK. */
+	mali_control = symbol_get(_mali_profiling_control);
+	if (mali_control) {
+		/* The event attribute in the XML file keeps the actual frame rate. */
+		unsigned int enabled = counters[FILMSTRIP].enabled ? 1 : 0;
+		unsigned int rate = filmstrip_event & 0xff;
+		unsigned int resize_factor = (filmstrip_event >> 8) & 0xff;
+
+		pr_debug("gator: mali online _mali_profiling_control symbol @ %p\n", mali_control);
+
+#define FBDUMP_CONTROL_ENABLE (1)
+#define FBDUMP_CONTROL_RATE (2)
+#define FBDUMP_CONTROL_RESIZE_FACTOR (4)
+		mali_control(FBDUMP_CONTROL_ENABLE, enabled);
+		mali_control(FBDUMP_CONTROL_RATE, rate);
+		mali_control(FBDUMP_CONTROL_RESIZE_FACTOR, resize_factor);
+
+		pr_debug("gator: sent mali_control enabled=%d, rate=%d, resize_factor=%d\n", enabled, rate, resize_factor);
+
+		symbol_put(_mali_profiling_control);
+	} else {
+		pr_err("gator: mali online _mali_profiling_control symbol not found\n");
+	}
+
+	/*
+	 * Set the first timestamp for calculating the sample interval. The first interval could be quite long,
+	 * since it will be the time between 'start' and the first 'read'.
+	 * This means that timeline values will be divided by a big number for the first sample.
+	 */
+	getnstimeofday(&prev_timestamp);
+
+	return 0;
+}
+
+static void stop(void)
+{
+	mali_profiling_control_type *mali_control;
+
+	pr_debug("gator: Mali-Midgard: stop\n");
+
+	/*
+	 * It is safe to unregister traces even if they were not successfully
+	 * registered, so no need to check.
+	 */
+	GATOR_UNREGISTER_TRACE(mali_pm_status);
+	pr_debug("gator: Mali-Midgard: mali_pm_status tracepoint deactivated\n");
+
+	GATOR_UNREGISTER_TRACE(mali_page_fault_insert_pages);
+	pr_debug("gator: Mali-Midgard: mali_page_fault_insert_pages tracepoint deactivated\n");
+
+	GATOR_UNREGISTER_TRACE(mali_mmu_as_in_use);
+	pr_debug("gator: Mali-Midgard: mali_mmu_as_in_use tracepoint deactivated\n");
+
+	GATOR_UNREGISTER_TRACE(mali_mmu_as_released);
+	pr_debug("gator: Mali-Midgard: mali_mmu_as_released tracepoint deactivated\n");
+
+	GATOR_UNREGISTER_TRACE(mali_total_alloc_pages_change);
+	pr_debug("gator: Mali-Midgard: mali_total_alloc_pages_change tracepoint deactivated\n");
+
+	/* Generic control interface for Mali DDK. */
+	mali_control = symbol_get(_mali_profiling_control);
+	if (mali_control) {
+		pr_debug("gator: mali offline _mali_profiling_control symbol @ %p\n", mali_control);
+
+		mali_control(FBDUMP_CONTROL_ENABLE, 0);
+
+		symbol_put(_mali_profiling_control);
+	} else {
+		pr_err("gator: mali offline _mali_profiling_control symbol not found\n");
+	}
+}
+
+static int read(int **buffer, bool sched_switch)
+{
+	int cnt;
+	int len = 0;
+	long sample_interval_us = 0;
+	struct timespec read_timestamp;
+
+	if (!on_primary_core())
+		return 0;
+
+	/* Get the start of this sample period. */
+	getnstimeofday(&read_timestamp);
+
+	/*
+	 * Calculate the sample interval if the previous sample time is valid.
+	 * We use tv_sec since it will not be 0.
+	 */
+	if (prev_timestamp.tv_sec != 0)
+		sample_interval_us = get_duration_us(&prev_timestamp, &read_timestamp);
+
+	/* Structure copy. Update the previous timestamp. */
+	prev_timestamp = read_timestamp;
+
+	/*
+	 * Report the timeline counters (ACTIVITY_START/STOP)
+	 */
+	for (cnt = FIRST_TIMELINE_EVENT; cnt < (FIRST_TIMELINE_EVENT + NUMBER_OF_TIMELINE_EVENTS); cnt++) {
+		struct mali_counter *counter = &counters[cnt];
+
+		if (counter->enabled) {
+			const int index = cnt - FIRST_TIMELINE_EVENT;
+			unsigned int value;
+
+			/* If the activity is still running, reset its start time to the
+			 * start of this sample period to correct the count. Add the
+			 * time up to the end of the sample onto the count.
+			 */
+			if (timeline_event_starttime[index].tv_sec != 0) {
+				const long event_duration = get_duration_us(&timeline_event_starttime[index], &read_timestamp);
+
+				timeline_data[index] += event_duration;
+				timeline_event_starttime[index] = read_timestamp;	/* Activity is still running. */
+			}
+
+			if (sample_interval_us != 0) {
+				/* Convert the counter to a percent-of-sample value */
+				value = (timeline_data[index] * 100) / sample_interval_us;
+			} else {
+				pr_debug("gator: Mali-Midgard: setting value to zero\n");
+				value = 0;
+			}
+
+			/* Clear the counter value ready for the next sample. */
+			timeline_data[index] = 0;
+
+			counter_dump[len++] = counter->key;
+			counter_dump[len++] = value;
+		}
+	}
+
+	/* Report the software counters */
+	for (cnt = FIRST_SOFTWARE_COUNTER; cnt < (FIRST_SOFTWARE_COUNTER + NUMBER_OF_SOFTWARE_COUNTERS); cnt++) {
+		const struct mali_counter *counter = &counters[cnt];
+
+		if (counter->enabled) {
+			const int index = cnt - FIRST_SOFTWARE_COUNTER;
+
+			counter_dump[len++] = counter->key;
+			counter_dump[len++] = sw_counter_data[index];
+			/* Set the value to zero for the next time */
+			sw_counter_data[index] = 0;
+		}
+	}
+
+	/* Report the accumulators */
+	for (cnt = FIRST_ACCUMULATOR; cnt < (FIRST_ACCUMULATOR + NUMBER_OF_ACCUMULATORS); cnt++) {
+		const struct mali_counter *counter = &counters[cnt];
+
+		if (counter->enabled) {
+			const int index = cnt - FIRST_ACCUMULATOR;
+
+			counter_dump[len++] = counter->key;
+			counter_dump[len++] = accumulators_data[index];
+			/* Do not zero the accumulator */
+		}
+	}
+
+	/* Update the buffer */
+	if (buffer)
+		*buffer = counter_dump;
+
+	return len;
+}
+
+static struct gator_interface gator_events_mali_midgard_interface = {
+	.create_files = create_files,
+	.start = start,
+	.stop = stop,
+	.read = read
+};
+
+extern int gator_events_mali_midgard_init(void)
+{
+	pr_debug("gator: Mali-Midgard: sw_counters init\n");
+
+	gator_mali_initialise_counters(counters, NUMBER_OF_EVENTS);
+
+	return gator_events_install(&gator_events_mali_midgard_interface);
+}
diff --git a/drivers/gator/gator_events_mali_midgard_hw.c b/drivers/gator/gator_events_mali_midgard_hw.c
new file mode 100644
index 000000000000..c8065da56815
--- /dev/null
+++ b/drivers/gator/gator_events_mali_midgard_hw.c
@@ -0,0 +1,977 @@
+/**
+ * Copyright (C) ARM Limited 2012-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#include "gator.h"
+
+#include <linux/module.h>
+#include <linux/time.h>
+#include <linux/math64.h>
+#include <linux/slab.h>
+#include <linux/io.h>
+
+/* Mali Midgard DDK includes */
+#if defined(MALI_SIMPLE_API)
+/* Header with wrapper functions to kbase structures and functions */
+#include "mali/mali_kbase_gator_api.h"
+#elif defined(MALI_DIR_MIDGARD)
+/* New DDK Directory structure with kernel/drivers/gpu/arm/midgard */
+#include "mali_linux_trace.h"
+#include "mali_kbase.h"
+#include "mali_kbase_mem_linux.h"
+#else
+/* Old DDK Directory structure with kernel/drivers/gpu/arm/t6xx */
+#include "linux/mali_linux_trace.h"
+#include "kbase/src/common/mali_kbase.h"
+#include "kbase/src/linux/mali_kbase_mem_linux.h"
+#endif
+
+/* If API version is not specified then assume API version 1. */
+#ifndef MALI_DDK_GATOR_API_VERSION
+#define MALI_DDK_GATOR_API_VERSION 1
+#endif
+
+#if (MALI_DDK_GATOR_API_VERSION != 1) && (MALI_DDK_GATOR_API_VERSION != 2) && (MALI_DDK_GATOR_API_VERSION != 3)
+#error MALI_DDK_GATOR_API_VERSION is invalid (must be 1 for r1/r2 DDK, or 2 for r3/r4 DDK, or 3 for r5 and later DDK).
+#endif
+
+#include "gator_events_mali_common.h"
+
+/*
+ * Mali-Midgard
+ */
+#if MALI_DDK_GATOR_API_VERSION == 3
+static uint32_t (*kbase_gator_instr_hwcnt_dump_irq_symbol)(struct kbase_gator_hwcnt_handles *);
+static uint32_t (*kbase_gator_instr_hwcnt_dump_complete_symbol)(struct kbase_gator_hwcnt_handles *, uint32_t *const);
+static struct kbase_gator_hwcnt_handles *(*kbase_gator_hwcnt_init_symbol)(struct kbase_gator_hwcnt_info *);
+static void (*kbase_gator_hwcnt_term_symbol)(struct kbase_gator_hwcnt_info *, struct kbase_gator_hwcnt_handles *);
+
+#else
+static struct kbase_device *(*kbase_find_device_symbol)(int);
+static struct kbase_context *(*kbase_create_context_symbol)(struct kbase_device *);
+static void (*kbase_destroy_context_symbol)(struct kbase_context *);
+
+#if MALI_DDK_GATOR_API_VERSION == 1
+static void *(*kbase_va_alloc_symbol)(struct kbase_context *, u32);
+static void (*kbase_va_free_symbol)(struct kbase_context *, void *);
+#elif MALI_DDK_GATOR_API_VERSION == 2
+static void *(*kbase_va_alloc_symbol)(struct kbase_context *, u32, struct kbase_hwc_dma_mapping *);
+static void (*kbase_va_free_symbol)(struct kbase_context *, struct kbase_hwc_dma_mapping *);
+#endif
+
+static mali_error (*kbase_instr_hwcnt_enable_symbol)(struct kbase_context *, struct kbase_uk_hwcnt_setup *);
+static mali_error (*kbase_instr_hwcnt_disable_symbol)(struct kbase_context *);
+static mali_error (*kbase_instr_hwcnt_clear_symbol)(struct kbase_context *);
+static mali_error (*kbase_instr_hwcnt_dump_irq_symbol)(struct kbase_context *);
+static mali_bool (*kbase_instr_hwcnt_dump_complete_symbol)(struct kbase_context *, mali_bool *);
+
+static long shader_present_low;
+#endif
+
+/** The interval between reads, in ns.
+ *
+ * Earlier we introduced a 'hold off for 1ms after last read' to
+ * resolve MIDBASE-2178 and MALINE-724. However, the 1ms hold off is
+ * too long if no context switches occur as there is a race between
+ * this value and the tick of the read clock in gator which is also
+ * 1ms. If we 'miss' the current read, the counter values are
+ * effectively 'spread' over 2ms and the values seen are half what
+ * they should be (since Streamline averages over sample time). In the
+ * presence of context switches this spread can vary and markedly
+ * affect the counters. Currently there is no 'proper' solution to
+ * this, but empirically we have found that reducing the minimum read
+ * interval to 950us causes the counts to be much more stable.
+ */
+static const int READ_INTERVAL_NSEC = 950000;
+
+#if GATOR_TEST
+#include "gator_events_mali_midgard_hw_test.c"
+#endif
+
+#if MALI_DDK_GATOR_API_VERSION != 3
+/* Blocks for HW counters */
+enum {
+	JM_BLOCK = 0,
+	TILER_BLOCK,
+	SHADER_BLOCK,
+	MMU_BLOCK
+};
+#endif
+
+static const char *mali_name;
+
+/* Counters for Mali-Midgard:
+ *
+ *    For HW counters we need strings to create /dev/gator/events files.
+ *    Enums are not needed because the position of the HW name in the array is the same
+ *    of the corresponding value in the received block of memory.
+ *    HW counters are requested by calculating a bitmask, passed then to the driver.
+ *    Every millisecond a HW counters dump is requested, and if the previous has been completed they are read.
+ */
+
+/* Hardware Counters */
+#if MALI_DDK_GATOR_API_VERSION == 3
+
+static const char *const *hardware_counter_names;
+static int number_of_hardware_counters;
+
+#else
+
+static const char *const hardware_counter_names[] = {
+	/* Job Manager */
+	"",
+	"",
+	"",
+	"",
+	"MESSAGES_SENT",
+	"MESSAGES_RECEIVED",
+	"GPU_ACTIVE",		/* 6 */
+	"IRQ_ACTIVE",
+	"JS0_JOBS",
+	"JS0_TASKS",
+	"JS0_ACTIVE",
+	"",
+	"JS0_WAIT_READ",
+	"JS0_WAIT_ISSUE",
+	"JS0_WAIT_DEPEND",
+	"JS0_WAIT_FINISH",
+	"JS1_JOBS",
+	"JS1_TASKS",
+	"JS1_ACTIVE",
+	"",
+	"JS1_WAIT_READ",
+	"JS1_WAIT_ISSUE",
+	"JS1_WAIT_DEPEND",
+	"JS1_WAIT_FINISH",
+	"JS2_JOBS",
+	"JS2_TASKS",
+	"JS2_ACTIVE",
+	"",
+	"JS2_WAIT_READ",
+	"JS2_WAIT_ISSUE",
+	"JS2_WAIT_DEPEND",
+	"JS2_WAIT_FINISH",
+	"JS3_JOBS",
+	"JS3_TASKS",
+	"JS3_ACTIVE",
+	"",
+	"JS3_WAIT_READ",
+	"JS3_WAIT_ISSUE",
+	"JS3_WAIT_DEPEND",
+	"JS3_WAIT_FINISH",
+	"JS4_JOBS",
+	"JS4_TASKS",
+	"JS4_ACTIVE",
+	"",
+	"JS4_WAIT_READ",
+	"JS4_WAIT_ISSUE",
+	"JS4_WAIT_DEPEND",
+	"JS4_WAIT_FINISH",
+	"JS5_JOBS",
+	"JS5_TASKS",
+	"JS5_ACTIVE",
+	"",
+	"JS5_WAIT_READ",
+	"JS5_WAIT_ISSUE",
+	"JS5_WAIT_DEPEND",
+	"JS5_WAIT_FINISH",
+	"JS6_JOBS",
+	"JS6_TASKS",
+	"JS6_ACTIVE",
+	"",
+	"JS6_WAIT_READ",
+	"JS6_WAIT_ISSUE",
+	"JS6_WAIT_DEPEND",
+	"JS6_WAIT_FINISH",
+
+	/*Tiler */
+	"",
+	"",
+	"",
+	"JOBS_PROCESSED",
+	"TRIANGLES",
+	"QUADS",
+	"POLYGONS",
+	"POINTS",
+	"LINES",
+	"VCACHE_HIT",
+	"VCACHE_MISS",
+	"FRONT_FACING",
+	"BACK_FACING",
+	"PRIM_VISIBLE",
+	"PRIM_CULLED",
+	"PRIM_CLIPPED",
+	"LEVEL0",
+	"LEVEL1",
+	"LEVEL2",
+	"LEVEL3",
+	"LEVEL4",
+	"LEVEL5",
+	"LEVEL6",
+	"LEVEL7",
+	"COMMAND_1",
+	"COMMAND_2",
+	"COMMAND_3",
+	"COMMAND_4",
+	"COMMAND_4_7",
+	"COMMAND_8_15",
+	"COMMAND_16_63",
+	"COMMAND_64",
+	"COMPRESS_IN",
+	"COMPRESS_OUT",
+	"COMPRESS_FLUSH",
+	"TIMESTAMPS",
+	"PCACHE_HIT",
+	"PCACHE_MISS",
+	"PCACHE_LINE",
+	"PCACHE_STALL",
+	"WRBUF_HIT",
+	"WRBUF_MISS",
+	"WRBUF_LINE",
+	"WRBUF_PARTIAL",
+	"WRBUF_STALL",
+	"ACTIVE",
+	"LOADING_DESC",
+	"INDEX_WAIT",
+	"INDEX_RANGE_WAIT",
+	"VERTEX_WAIT",
+	"PCACHE_WAIT",
+	"WRBUF_WAIT",
+	"BUS_READ",
+	"BUS_WRITE",
+	"",
+	"",
+	"",
+	"",
+	"",
+	"UTLB_STALL",
+	"UTLB_REPLAY_MISS",
+	"UTLB_REPLAY_FULL",
+	"UTLB_NEW_MISS",
+	"UTLB_HIT",
+
+	/* Shader Core */
+	"",
+	"",
+	"",
+	"SHADER_CORE_ACTIVE",
+	"FRAG_ACTIVE",
+	"FRAG_PRIMATIVES",
+	"FRAG_PRIMATIVES_DROPPED",
+	"FRAG_CYCLE_DESC",
+	"FRAG_CYCLES_PLR",
+	"FRAG_CYCLES_VERT",
+	"FRAG_CYCLES_TRISETUP",
+	"FRAG_CYCLES_RAST",
+	"FRAG_THREADS",
+	"FRAG_DUMMY_THREADS",
+	"FRAG_QUADS_RAST",
+	"FRAG_QUADS_EZS_TEST",
+	"FRAG_QUADS_EZS_KILLED",
+	"FRAG_QUADS_LZS_TEST",
+	"FRAG_QUADS_LZS_KILLED",
+	"FRAG_CYCLE_NO_TILE",
+	"FRAG_NUM_TILES",
+	"FRAG_TRANS_ELIM",
+	"COMPUTE_ACTIVE",
+	"COMPUTE_TASKS",
+	"COMPUTE_THREADS",
+	"COMPUTE_CYCLES_DESC",
+	"TRIPIPE_ACTIVE",
+	"ARITH_WORDS",
+	"ARITH_CYCLES_REG",
+	"ARITH_CYCLES_L0",
+	"ARITH_FRAG_DEPEND",
+	"LS_WORDS",
+	"LS_ISSUES",
+	"LS_RESTARTS",
+	"LS_REISSUES_MISS",
+	"LS_REISSUES_VD",
+	"LS_REISSUE_ATTRIB_MISS",
+	"LS_NO_WB",
+	"TEX_WORDS",
+	"TEX_BUBBLES",
+	"TEX_WORDS_L0",
+	"TEX_WORDS_DESC",
+	"TEX_THREADS",
+	"TEX_RECIRC_FMISS",
+	"TEX_RECIRC_DESC",
+	"TEX_RECIRC_MULTI",
+	"TEX_RECIRC_PMISS",
+	"TEX_RECIRC_CONF",
+	"LSC_READ_HITS",
+	"LSC_READ_MISSES",
+	"LSC_WRITE_HITS",
+	"LSC_WRITE_MISSES",
+	"LSC_ATOMIC_HITS",
+	"LSC_ATOMIC_MISSES",
+	"LSC_LINE_FETCHES",
+	"LSC_DIRTY_LINE",
+	"LSC_SNOOPS",
+	"AXI_TLB_STALL",
+	"AXI_TLB_MIESS",
+	"AXI_TLB_TRANSACTION",
+	"LS_TLB_MISS",
+	"LS_TLB_HIT",
+	"AXI_BEATS_READ",
+	"AXI_BEATS_WRITTEN",
+
+	/*L2 and MMU */
+	"",
+	"",
+	"",
+	"",
+	"MMU_HIT",
+	"MMU_NEW_MISS",
+	"MMU_REPLAY_FULL",
+	"MMU_REPLAY_MISS",
+	"MMU_TABLE_WALK",
+	"",
+	"",
+	"",
+	"",
+	"",
+	"",
+	"",
+	"UTLB_HIT",
+	"UTLB_NEW_MISS",
+	"UTLB_REPLAY_FULL",
+	"UTLB_REPLAY_MISS",
+	"UTLB_STALL",
+	"",
+	"",
+	"",
+	"",
+	"",
+	"",
+	"",
+	"",
+	"",
+	"L2_WRITE_BEATS",
+	"L2_READ_BEATS",
+	"L2_ANY_LOOKUP",
+	"L2_READ_LOOKUP",
+	"L2_SREAD_LOOKUP",
+	"L2_READ_REPLAY",
+	"L2_READ_SNOOP",
+	"L2_READ_HIT",
+	"L2_CLEAN_MISS",
+	"L2_WRITE_LOOKUP",
+	"L2_SWRITE_LOOKUP",
+	"L2_WRITE_REPLAY",
+	"L2_WRITE_SNOOP",
+	"L2_WRITE_HIT",
+	"L2_EXT_READ_FULL",
+	"L2_EXT_READ_HALF",
+	"L2_EXT_WRITE_FULL",
+	"L2_EXT_WRITE_HALF",
+	"L2_EXT_READ",
+	"L2_EXT_READ_LINE",
+	"L2_EXT_WRITE",
+	"L2_EXT_WRITE_LINE",
+	"L2_EXT_WRITE_SMALL",
+	"L2_EXT_BARRIER",
+	"L2_EXT_AR_STALL",
+	"L2_EXT_R_BUF_FULL",
+	"L2_EXT_RD_BUF_FULL",
+	"L2_EXT_R_RAW",
+	"L2_EXT_W_STALL",
+	"L2_EXT_W_BUF_FULL",
+	"L2_EXT_R_W_HAZARD",
+	"L2_TAG_HAZARD",
+	"L2_SNOOP_FULL",
+	"L2_REPLAY_FULL"
+};
+
+static const int number_of_hardware_counters = ARRAY_SIZE(hardware_counter_names);
+
+#endif
+
+#define GET_HW_BLOCK(c) (((c) >> 6) & 0x3)
+#define GET_COUNTER_OFFSET(c) ((c) & 0x3f)
+
+#if MALI_DDK_GATOR_API_VERSION == 3
+/* Opaque handles for kbase_context and kbase_hwc_dma_mapping */
+static struct kbase_gator_hwcnt_handles *handles;
+
+/* Information about hardware counters */
+static struct kbase_gator_hwcnt_info *in_out_info;
+
+#else
+/* Memory to dump hardware counters into */
+static void *kernel_dump_buffer;
+
+#if MALI_DDK_GATOR_API_VERSION == 2
+/* DMA state used to manage lifetime of the buffer */
+struct kbase_hwc_dma_mapping kernel_dump_buffer_handle;
+#endif
+
+/* kbase context and device */
+static struct kbase_context *kbcontext;
+static struct kbase_device *kbdevice;
+
+/*
+ * The following function has no external prototype in older DDK
+ * revisions. When the DDK is updated then this should be removed.
+ */
+struct kbase_device *kbase_find_device(int minor);
+#endif
+
+static volatile bool kbase_device_busy;
+static unsigned int num_hardware_counters_enabled;
+
+/* gatorfs variables for counter enable state */
+static struct mali_counter *counters;
+
+/* An array used to return the data we recorded as key,value pairs */
+static int *counter_dump;
+
+extern struct mali_counter mali_activity[3];
+
+static const char *const mali_activity_names[] = {
+	"fragment",
+	"vertex",
+	"opencl",
+};
+
+#define SYMBOL_GET(FUNCTION, ERROR_COUNT) \
+	do { \
+		if (FUNCTION ## _symbol) { \
+			pr_err("gator: mali " #FUNCTION " symbol was already registered\n"); \
+			(ERROR_COUNT)++; \
+		} else { \
+			FUNCTION ## _symbol = symbol_get(FUNCTION); \
+			if (!FUNCTION ## _symbol) { \
+				pr_err("gator: mali online " #FUNCTION " symbol not found\n"); \
+				(ERROR_COUNT)++; \
+			} \
+		} \
+	} while (0)
+
+#define SYMBOL_CLEANUP(FUNCTION) \
+	do { \
+		if (FUNCTION ## _symbol) { \
+			symbol_put(FUNCTION); \
+			FUNCTION ## _symbol = NULL; \
+		} \
+	} while (0)
+
+/**
+ * Execute symbol_get for all the Mali symbols and check for success.
+ * @return the number of symbols not loaded.
+ */
+static int init_symbols(void)
+{
+	int error_count = 0;
+#if MALI_DDK_GATOR_API_VERSION == 3
+	SYMBOL_GET(kbase_gator_instr_hwcnt_dump_irq, error_count);
+	SYMBOL_GET(kbase_gator_instr_hwcnt_dump_complete, error_count);
+	SYMBOL_GET(kbase_gator_hwcnt_init, error_count);
+	SYMBOL_GET(kbase_gator_hwcnt_term, error_count);
+#else
+	SYMBOL_GET(kbase_find_device, error_count);
+	SYMBOL_GET(kbase_create_context, error_count);
+	SYMBOL_GET(kbase_va_alloc, error_count);
+	SYMBOL_GET(kbase_instr_hwcnt_enable, error_count);
+	SYMBOL_GET(kbase_instr_hwcnt_clear, error_count);
+	SYMBOL_GET(kbase_instr_hwcnt_dump_irq, error_count);
+	SYMBOL_GET(kbase_instr_hwcnt_dump_complete, error_count);
+	SYMBOL_GET(kbase_instr_hwcnt_disable, error_count);
+	SYMBOL_GET(kbase_va_free, error_count);
+	SYMBOL_GET(kbase_destroy_context, error_count);
+#endif
+
+	return error_count;
+}
+
+/**
+ * Execute symbol_put for all the registered Mali symbols.
+ */
+static void clean_symbols(void)
+{
+#if MALI_DDK_GATOR_API_VERSION == 3
+	SYMBOL_CLEANUP(kbase_gator_instr_hwcnt_dump_irq);
+	SYMBOL_CLEANUP(kbase_gator_instr_hwcnt_dump_complete);
+	SYMBOL_CLEANUP(kbase_gator_hwcnt_init);
+	SYMBOL_CLEANUP(kbase_gator_hwcnt_term);
+#else
+	SYMBOL_CLEANUP(kbase_find_device);
+	SYMBOL_CLEANUP(kbase_create_context);
+	SYMBOL_CLEANUP(kbase_va_alloc);
+	SYMBOL_CLEANUP(kbase_instr_hwcnt_enable);
+	SYMBOL_CLEANUP(kbase_instr_hwcnt_clear);
+	SYMBOL_CLEANUP(kbase_instr_hwcnt_dump_irq);
+	SYMBOL_CLEANUP(kbase_instr_hwcnt_dump_complete);
+	SYMBOL_CLEANUP(kbase_instr_hwcnt_disable);
+	SYMBOL_CLEANUP(kbase_va_free);
+	SYMBOL_CLEANUP(kbase_destroy_context);
+#endif
+}
+
+/**
+ * Determines whether a read should take place
+ * @param current_time The current time, obtained from getnstimeofday()
+ * @param prev_time_s The number of seconds at the previous read attempt.
+ * @param next_read_time_ns The time (in ns) when the next read should be allowed.
+ *
+ * Note that this function has been separated out here to allow it to be tested.
+ */
+static int is_read_scheduled(const struct timespec *current_time, u32 *prev_time_s, s32 *next_read_time_ns)
+{
+	/* If the current ns count rolls over a second, roll the next read time too. */
+	if (current_time->tv_sec != *prev_time_s)
+		*next_read_time_ns = *next_read_time_ns - NSEC_PER_SEC;
+
+	/* Abort the read if the next read time has not arrived. */
+	if (current_time->tv_nsec < *next_read_time_ns)
+		return 0;
+
+	/* Set the next read some fixed time after this one, and update the read timestamp. */
+	*next_read_time_ns = current_time->tv_nsec + READ_INTERVAL_NSEC;
+
+	*prev_time_s = current_time->tv_sec;
+	return 1;
+}
+
+static int start(void)
+{
+#if MALI_DDK_GATOR_API_VERSION != 3
+	struct kbase_uk_hwcnt_setup setup;
+	unsigned long long shadersPresent = 0;
+	u16 bitmask[] = { 0, 0, 0, 0 };
+	mali_error err;
+#endif
+	int cnt;
+
+#if MALI_DDK_GATOR_API_VERSION == 3
+	/* Setup HW counters */
+	num_hardware_counters_enabled = 0;
+
+	/* Declare and initialise kbase_gator_hwcnt_info structure */
+	in_out_info = kmalloc(sizeof(*in_out_info), GFP_KERNEL);
+	for (cnt = 0; cnt < ARRAY_SIZE(in_out_info->bitmask); cnt++)
+		in_out_info->bitmask[cnt] = 0;
+
+	/* Calculate enable bitmasks based on counters_enabled array */
+	for (cnt = 0; cnt < number_of_hardware_counters; cnt++) {
+		if (counters[cnt].enabled) {
+			int block = GET_HW_BLOCK(cnt);
+			int enable_bit = GET_COUNTER_OFFSET(cnt) / 4;
+
+			in_out_info->bitmask[block] |= (1 << enable_bit);
+			pr_debug("gator: Mali-Midgard: hardware counter %s selected [%d]\n", hardware_counter_names[cnt], cnt);
+			num_hardware_counters_enabled++;
+		}
+	}
+
+	/* Create a kbase context for HW counters */
+	if (num_hardware_counters_enabled > 0) {
+		if (init_symbols() > 0) {
+			clean_symbols();
+			/* No Mali driver code entrypoints found - not a fault. */
+			return 0;
+		}
+
+		handles = kbase_gator_hwcnt_init_symbol(in_out_info);
+
+		if (handles == NULL)
+			goto out;
+
+		kbase_device_busy = false;
+	}
+
+	return 0;
+#else
+	/* Setup HW counters */
+	num_hardware_counters_enabled = 0;
+
+	/* Calculate enable bitmasks based on counters_enabled array */
+	for (cnt = 0; cnt < number_of_hardware_counters; cnt++) {
+		const struct mali_counter *counter = &counters[cnt];
+
+		if (counter->enabled) {
+			int block = GET_HW_BLOCK(cnt);
+			int enable_bit = GET_COUNTER_OFFSET(cnt) / 4;
+
+			bitmask[block] |= (1 << enable_bit);
+			pr_debug("gator: Mali-Midgard: hardware counter %s selected [%d]\n", hardware_counter_names[cnt], cnt);
+			num_hardware_counters_enabled++;
+		}
+	}
+
+	/* Create a kbase context for HW counters */
+	if (num_hardware_counters_enabled > 0) {
+		if (init_symbols() > 0) {
+			clean_symbols();
+			/* No Mali driver code entrypoints found - not a fault. */
+			return 0;
+		}
+
+		kbdevice = kbase_find_device_symbol(-1);
+
+		/* If we already got a context, fail */
+		if (kbcontext) {
+			pr_debug("gator: Mali-Midgard: error context already present\n");
+			goto out;
+		}
+
+		/* kbcontext will only be valid after all the Mali symbols are loaded successfully */
+		kbcontext = kbase_create_context_symbol(kbdevice);
+		if (!kbcontext) {
+			pr_debug("gator: Mali-Midgard: error creating kbase context\n");
+			goto out;
+		}
+
+		/* See if we can get the number of shader cores */
+		shadersPresent = kbdevice->shader_present_bitmap;
+		shader_present_low = (unsigned long)shadersPresent;
+
+		/*
+		 * The amount of memory needed to store the dump (bytes)
+		 * DUMP_SIZE = number of core groups
+		 *             * number of blocks (always 8 for midgard)
+		 *             * number of counters per block (always 64 for midgard)
+		 *             * number of bytes per counter (always 4 in midgard)
+		 * For a Mali-Midgard with a single core group = 1 * 8 * 64 * 4 = 2048
+		 * For a Mali-Midgard with a dual core group   = 2 * 8 * 64 * 4 = 4096
+		 */
+#if MALI_DDK_GATOR_API_VERSION == 1
+		kernel_dump_buffer = kbase_va_alloc_symbol(kbcontext, 4096);
+#elif MALI_DDK_GATOR_API_VERSION == 2
+		kernel_dump_buffer = kbase_va_alloc_symbol(kbcontext, 4096, &kernel_dump_buffer_handle);
+#endif
+		if (!kernel_dump_buffer) {
+			pr_debug("gator: Mali-Midgard: error trying to allocate va\n");
+			goto destroy_context;
+		}
+
+		setup.dump_buffer = (uintptr_t)kernel_dump_buffer;
+		setup.jm_bm = bitmask[JM_BLOCK];
+		setup.tiler_bm = bitmask[TILER_BLOCK];
+		setup.shader_bm = bitmask[SHADER_BLOCK];
+		setup.mmu_l2_bm = bitmask[MMU_BLOCK];
+		/* These counters do not exist on Mali-T60x */
+		setup.l3_cache_bm = 0;
+
+		/* Use kbase API to enable hardware counters and provide dump buffer */
+		err = kbase_instr_hwcnt_enable_symbol(kbcontext, &setup);
+		if (err != MALI_ERROR_NONE) {
+			pr_debug("gator: Mali-Midgard: can't setup hardware counters\n");
+			goto free_buffer;
+		}
+		pr_debug("gator: Mali-Midgard: hardware counters enabled\n");
+		kbase_instr_hwcnt_clear_symbol(kbcontext);
+		pr_debug("gator: Mali-Midgard: hardware counters cleared\n");
+
+		kbase_device_busy = false;
+	}
+
+	return 0;
+
+free_buffer:
+#if MALI_DDK_GATOR_API_VERSION == 1
+	kbase_va_free_symbol(kbcontext, kernel_dump_buffer);
+#elif MALI_DDK_GATOR_API_VERSION == 2
+	kbase_va_free_symbol(kbcontext, &kernel_dump_buffer_handle);
+#endif
+
+destroy_context:
+	kbase_destroy_context_symbol(kbcontext);
+#endif
+
+out:
+	clean_symbols();
+	return -1;
+}
+
+static void stop(void)
+{
+	unsigned int cnt;
+#if MALI_DDK_GATOR_API_VERSION == 3
+	struct kbase_gator_hwcnt_handles *temp_hand;
+#else
+	struct kbase_context *temp_kbcontext;
+#endif
+
+	pr_debug("gator: Mali-Midgard: stop\n");
+
+	/* Set all counters as disabled */
+	for (cnt = 0; cnt < number_of_hardware_counters; cnt++)
+		counters[cnt].enabled = 0;
+
+	/* Destroy the context for HW counters */
+#if MALI_DDK_GATOR_API_VERSION == 3
+	if (num_hardware_counters_enabled > 0 && handles != NULL) {
+		/*
+		 * Set the global variable to NULL before destroying it, because
+		 * other function will check this before using it.
+		 */
+		temp_hand = handles;
+		handles = NULL;
+
+		kbase_gator_hwcnt_term_symbol(in_out_info, temp_hand);
+
+		kfree(in_out_info);
+
+#else
+	if (num_hardware_counters_enabled > 0 && kbcontext != NULL) {
+		/*
+		 * Set the global variable to NULL before destroying it, because
+		 * other function will check this before using it.
+		 */
+		temp_kbcontext = kbcontext;
+		kbcontext = NULL;
+
+		kbase_instr_hwcnt_disable_symbol(temp_kbcontext);
+
+#if MALI_DDK_GATOR_API_VERSION == 1
+		kbase_va_free_symbol(temp_kbcontext, kernel_dump_buffer);
+#elif MALI_DDK_GATOR_API_VERSION == 2
+		kbase_va_free_symbol(temp_kbcontext, &kernel_dump_buffer_handle);
+#endif
+
+		kbase_destroy_context_symbol(temp_kbcontext);
+#endif
+
+		pr_debug("gator: Mali-Midgard: hardware counters stopped\n");
+
+		clean_symbols();
+	}
+}
+
+static int read_counter(const int cnt, const int len, const struct mali_counter *counter)
+{
+	const int block = GET_HW_BLOCK(cnt);
+	const int counter_offset = GET_COUNTER_OFFSET(cnt);
+
+#if MALI_DDK_GATOR_API_VERSION == 3
+	const char *block_base_address = (char *)in_out_info->kernel_dump_buffer;
+	int i;
+	int shader_core_count = 0;
+	u32 value = 0;
+
+	for (i = 0; i < in_out_info->nr_hwc_blocks; i++) {
+		if (block == in_out_info->hwc_layout[i]) {
+			value += *((u32 *)(block_base_address + (0x100 * i)) + counter_offset);
+			if (block == SHADER_BLOCK)
+				++shader_core_count;
+		}
+	}
+
+	if (shader_core_count > 1)
+		value /= shader_core_count;
+#else
+	const char *block_base_address = (char *)kernel_dump_buffer + vithar_blocks[block];
+
+	/* If counter belongs to shader block need to take into account all cores */
+	if (block == SHADER_BLOCK) {
+		int i = 0;
+		int shader_core_count = 0;
+
+		value = 0;
+
+		for (i = 0; i < 4; i++) {
+			if ((shader_present_low >> i) & 1) {
+				value += *((u32 *)(block_base_address + (0x100 * i)) + counter_offset);
+				shader_core_count++;
+			}
+		}
+
+		for (i = 0; i < 4; i++) {
+			if ((shader_present_low >> (i+4)) & 1) {
+				value += *((u32 *)(block_base_address + (0x100 * i) + 0x800) + counter_offset);
+				shader_core_count++;
+			}
+		}
+
+		/* Need to total by number of cores to produce an average */
+		if (shader_core_count != 0)
+			value /= shader_core_count;
+	} else {
+		value = *((u32 *)block_base_address + counter_offset);
+	}
+#endif
+
+	counter_dump[len + 0] = counter->key;
+	counter_dump[len + 1] = value;
+
+	return 2;
+}
+
+static int read(int **buffer, bool sched_switch)
+{
+	int cnt;
+	int len = 0;
+	uint32_t success;
+
+	struct timespec current_time;
+	static u32 prev_time_s;
+	static s32 next_read_time_ns;
+
+	if (!on_primary_core() || sched_switch)
+		return 0;
+
+	getnstimeofday(&current_time);
+
+	/*
+	 * Discard reads unless a respectable time has passed. This
+	 * reduces the load on the GPU without sacrificing accuracy on
+	 * the Streamline display.
+	 */
+	if (!is_read_scheduled(&current_time, &prev_time_s, &next_read_time_ns))
+		return 0;
+
+	/*
+	 * Report the HW counters
+	 * Only process hardware counters if at least one of the hardware counters is enabled.
+	 */
+	if (num_hardware_counters_enabled > 0) {
+#if MALI_DDK_GATOR_API_VERSION != 3
+		const unsigned int vithar_blocks[] = {
+			0x700,	/* VITHAR_JOB_MANAGER,     Block 0 */
+			0x400,	/* VITHAR_TILER,           Block 1 */
+			0x000,	/* VITHAR_SHADER_CORE,     Block 2 */
+			0x500	/* VITHAR_MEMORY_SYSTEM,   Block 3 */
+		};
+#endif
+
+#if MALI_DDK_GATOR_API_VERSION == 3
+		if (!handles)
+			return -1;
+
+		/* Mali symbols can be called safely since a kbcontext is valid */
+		if (kbase_gator_instr_hwcnt_dump_complete_symbol(handles, &success) == MALI_TRUE) {
+#else
+		if (!kbcontext)
+			return -1;
+
+		/* Mali symbols can be called safely since a kbcontext is valid */
+		if (kbase_instr_hwcnt_dump_complete_symbol(kbcontext, &success) == MALI_TRUE) {
+#endif
+			kbase_device_busy = false;
+
+			if (success == MALI_TRUE) {
+				/* Cycle through hardware counters and accumulate totals */
+				for (cnt = 0; cnt < number_of_hardware_counters; cnt++) {
+					const struct mali_counter *counter = &counters[cnt];
+
+					if (counter->enabled)
+						len += read_counter(cnt, len, counter);
+				}
+			}
+		}
+
+		if (!kbase_device_busy) {
+			kbase_device_busy = true;
+#if MALI_DDK_GATOR_API_VERSION == 3
+			kbase_gator_instr_hwcnt_dump_irq_symbol(handles);
+#else
+			kbase_instr_hwcnt_dump_irq_symbol(kbcontext);
+#endif
+		}
+	}
+
+	/* Update the buffer */
+	if (buffer)
+		*buffer = counter_dump;
+
+	return len;
+}
+
+static int create_files(struct super_block *sb, struct dentry *root)
+{
+	unsigned int event;
+	/*
+	 * Create the filesystem for all events
+	 */
+	for (event = 0; event < ARRAY_SIZE(mali_activity); event++) {
+		if (gator_mali_create_file_system("Midgard", mali_activity_names[event], sb, root, &mali_activity[event], NULL) != 0)
+			return -1;
+	}
+
+	for (event = 0; event < number_of_hardware_counters; event++) {
+		if (gator_mali_create_file_system(mali_name, hardware_counter_names[event], sb, root, &counters[event], NULL) != 0)
+			return -1;
+	}
+
+	return 0;
+}
+
+static void shutdown(void)
+{
+#if MALI_DDK_GATOR_API_VERSION == 3
+	void (*kbase_gator_hwcnt_term_names_symbol)(void) = NULL;
+	int error_count = 0;
+#endif
+
+	kfree(counters);
+	kfree(counter_dump);
+
+#if MALI_DDK_GATOR_API_VERSION == 3
+	SYMBOL_GET(kbase_gator_hwcnt_term_names, error_count);
+
+	number_of_hardware_counters = -1;
+	hardware_counter_names = NULL;
+	if (kbase_gator_hwcnt_term_names_symbol != NULL) {
+		kbase_gator_hwcnt_term_names_symbol();
+		pr_err("Released symbols\n");
+	}
+
+	SYMBOL_CLEANUP(kbase_gator_hwcnt_term_names);
+#endif
+}
+
+static struct gator_interface gator_events_mali_midgard_interface = {
+	.shutdown = shutdown,
+	.create_files = create_files,
+	.start = start,
+	.stop = stop,
+	.read = read
+};
+
+int gator_events_mali_midgard_hw_init(void)
+{
+#if MALI_DDK_GATOR_API_VERSION == 3
+	const char *const *(*kbase_gator_hwcnt_init_names_symbol)(uint32_t *) = NULL;
+	int error_count = 0;
+#endif
+
+	pr_debug("gator: Mali-Midgard: sw_counters init\n");
+
+#if GATOR_TEST
+	test_all_is_read_scheduled();
+#endif
+
+#if MALI_DDK_GATOR_API_VERSION == 3
+	SYMBOL_GET(kbase_gator_hwcnt_init_names, error_count);
+	if (error_count > 0) {
+		SYMBOL_CLEANUP(kbase_gator_hwcnt_init_names);
+		return 1;
+	}
+
+	number_of_hardware_counters = -1;
+	hardware_counter_names = kbase_gator_hwcnt_init_names_symbol(&number_of_hardware_counters);
+
+	SYMBOL_CLEANUP(kbase_gator_hwcnt_init_names);
+
+	if ((hardware_counter_names == NULL) || (number_of_hardware_counters <= 0)) {
+		pr_err("gator: Error reading hardware counters names: got %d names\n", number_of_hardware_counters);
+		return -1;
+	}
+#else
+	mali_name = "Midgard";
+#endif
+
+	counters = kmalloc(sizeof(*counters)*number_of_hardware_counters, GFP_KERNEL);
+	counter_dump = kmalloc(sizeof(*counter_dump)*number_of_hardware_counters*2, GFP_KERNEL);
+
+	gator_mali_initialise_counters(mali_activity, ARRAY_SIZE(mali_activity));
+	gator_mali_initialise_counters(counters, number_of_hardware_counters);
+
+	return gator_events_install(&gator_events_mali_midgard_interface);
+}
diff --git a/drivers/gator/gator_events_mali_midgard_hw_test.c b/drivers/gator/gator_events_mali_midgard_hw_test.c
new file mode 100644
index 000000000000..31a91e1c72b2
--- /dev/null
+++ b/drivers/gator/gator_events_mali_midgard_hw_test.c
@@ -0,0 +1,55 @@
+/**
+ * Copyright (C) ARM Limited 2012-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+/**
+ * Test functions for mali_t600_hw code.
+ */
+
+static int is_read_scheduled(const struct timespec *current_time, u32 *prev_time_s, s32 *next_read_time_ns);
+
+static int test_is_read_scheduled(u32 s, u32 ns, u32 prev_s, s32 next_ns, int expected_result, s32 expected_next_ns)
+{
+	struct timespec current_time;
+	u32 prev_time_s = prev_s;
+	s32 next_read_time_ns = next_ns;
+
+	current_time.tv_sec = s;
+	current_time.tv_nsec = ns;
+
+	if (is_read_scheduled(&current_time, &prev_time_s, &next_read_time_ns) != expected_result) {
+		pr_err("Failed do_read(%u, %u, %u, %d): expected %d\n", s, ns, prev_s, next_ns, expected_result);
+		return 0;
+	}
+
+	if (next_read_time_ns != expected_next_ns) {
+		pr_err("Failed: next_read_ns expected=%d, actual=%d\n", expected_next_ns, next_read_time_ns);
+		return 0;
+	}
+
+	return 1;
+}
+
+static void test_all_is_read_scheduled(void)
+{
+	const int HIGHEST_NS = 999999999;
+	int n_tests_passed = 0;
+
+	pr_err("gator: running tests on %s\n", __FILE__);
+
+	n_tests_passed += test_is_read_scheduled(0, 0, 0, 0, 1, READ_INTERVAL_NSEC);	/* Null time */
+	n_tests_passed += test_is_read_scheduled(100, 1000, 0, 0, 1, READ_INTERVAL_NSEC + 1000);	/* Initial values */
+
+	n_tests_passed += test_is_read_scheduled(100, HIGHEST_NS, 100, HIGHEST_NS + 500, 0, HIGHEST_NS + 500);
+	n_tests_passed += test_is_read_scheduled(101, 0001, 100, HIGHEST_NS + 500, 0, HIGHEST_NS + 500 - NSEC_PER_SEC);
+	n_tests_passed += test_is_read_scheduled(101, 600, 100, HIGHEST_NS + 500 - NSEC_PER_SEC, 1, 600 + READ_INTERVAL_NSEC);
+
+	n_tests_passed += test_is_read_scheduled(101, 600, 100, HIGHEST_NS + 500, 1, 600 + READ_INTERVAL_NSEC);
+
+	pr_err("gator: %d tests passed\n", n_tests_passed);
+}
diff --git a/drivers/gator/gator_events_meminfo.c b/drivers/gator/gator_events_meminfo.c
new file mode 100644
index 000000000000..c625ac5af9cd
--- /dev/null
+++ b/drivers/gator/gator_events_meminfo.c
@@ -0,0 +1,470 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#include "gator.h"
+
+#include <linux/hardirq.h>
+#include <linux/kthread.h>
+#include <linux/sched.h>
+#include <linux/semaphore.h>
+#include <linux/workqueue.h>
+#include <trace/events/kmem.h>
+
+#define USE_THREAD defined(CONFIG_PREEMPT_RT_FULL)
+
+enum {
+	MEMINFO_MEMFREE,
+	MEMINFO_MEMUSED,
+	MEMINFO_BUFFERRAM,
+	MEMINFO_TOTAL,
+};
+
+enum {
+	PROC_SIZE,
+	PROC_SHARE,
+	PROC_TEXT,
+	PROC_DATA,
+	PROC_COUNT,
+};
+
+static const char * const meminfo_names[] = {
+	"Linux_meminfo_memfree",
+	"Linux_meminfo_memused",
+	"Linux_meminfo_bufferram",
+};
+
+static const char * const proc_names[] = {
+	"Linux_proc_statm_size",
+	"Linux_proc_statm_share",
+	"Linux_proc_statm_text",
+	"Linux_proc_statm_data",
+};
+
+static bool meminfo_global_enabled;
+static ulong meminfo_enabled[MEMINFO_TOTAL];
+static ulong meminfo_keys[MEMINFO_TOTAL];
+static long long meminfo_buffer[2 * (MEMINFO_TOTAL + 2)];
+static int meminfo_length;
+static bool new_data_avail;
+
+static bool proc_global_enabled;
+static ulong proc_enabled[PROC_COUNT];
+static ulong proc_keys[PROC_COUNT];
+static DEFINE_PER_CPU(long long, proc_buffer[2 * (PROC_COUNT + 3)]);
+
+#if USE_THREAD
+
+static int gator_meminfo_func(void *data);
+static bool gator_meminfo_run;
+/* Initialize semaphore unlocked to initialize memory values */
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
+static DECLARE_MUTEX(gator_meminfo_sem);
+#else
+static DEFINE_SEMAPHORE(gator_meminfo_sem);
+#endif
+
+static void notify(void)
+{
+	up(&gator_meminfo_sem);
+}
+
+#else
+
+static unsigned int mem_event;
+static void wq_sched_handler(struct work_struct *wsptr);
+DECLARE_WORK(work, wq_sched_handler);
+static struct timer_list meminfo_wake_up_timer;
+static void meminfo_wake_up_handler(unsigned long unused_data);
+
+static void notify(void)
+{
+	mem_event++;
+}
+
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
+GATOR_DEFINE_PROBE(mm_page_free_direct, TP_PROTO(struct page *page, unsigned int order))
+#else
+GATOR_DEFINE_PROBE(mm_page_free, TP_PROTO(struct page *page, unsigned int order))
+#endif
+{
+	notify();
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
+GATOR_DEFINE_PROBE(mm_pagevec_free, TP_PROTO(struct page *page, int cold))
+#else
+GATOR_DEFINE_PROBE(mm_page_free_batched, TP_PROTO(struct page *page, int cold))
+#endif
+{
+	notify();
+}
+
+GATOR_DEFINE_PROBE(mm_page_alloc, TP_PROTO(struct page *page, unsigned int order, gfp_t gfp_flags, int migratetype))
+{
+	notify();
+}
+
+static int gator_events_meminfo_create_files(struct super_block *sb, struct dentry *root)
+{
+	struct dentry *dir;
+	int i;
+
+	for (i = 0; i < MEMINFO_TOTAL; i++) {
+		dir = gatorfs_mkdir(sb, root, meminfo_names[i]);
+		if (!dir)
+			return -1;
+		gatorfs_create_ulong(sb, dir, "enabled", &meminfo_enabled[i]);
+		gatorfs_create_ro_ulong(sb, dir, "key", &meminfo_keys[i]);
+	}
+
+	for (i = 0; i < PROC_COUNT; ++i) {
+		dir = gatorfs_mkdir(sb, root, proc_names[i]);
+		if (!dir)
+			return -1;
+		gatorfs_create_ulong(sb, dir, "enabled", &proc_enabled[i]);
+		gatorfs_create_ro_ulong(sb, dir, "key", &proc_keys[i]);
+	}
+
+	return 0;
+}
+
+static int gator_events_meminfo_start(void)
+{
+	int i;
+
+	new_data_avail = false;
+	meminfo_global_enabled = 0;
+	for (i = 0; i < MEMINFO_TOTAL; i++) {
+		if (meminfo_enabled[i]) {
+			meminfo_global_enabled = 1;
+			break;
+		}
+	}
+
+	proc_global_enabled = 0;
+	for (i = 0; i < PROC_COUNT; ++i) {
+		if (proc_enabled[i]) {
+			proc_global_enabled = 1;
+			break;
+		}
+	}
+	if (meminfo_enabled[MEMINFO_MEMUSED])
+		proc_global_enabled = 1;
+
+	if (meminfo_global_enabled == 0)
+		return 0;
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
+	if (GATOR_REGISTER_TRACE(mm_page_free_direct))
+#else
+	if (GATOR_REGISTER_TRACE(mm_page_free))
+#endif
+		goto mm_page_free_exit;
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
+	if (GATOR_REGISTER_TRACE(mm_pagevec_free))
+#else
+	if (GATOR_REGISTER_TRACE(mm_page_free_batched))
+#endif
+		goto mm_page_free_batched_exit;
+	if (GATOR_REGISTER_TRACE(mm_page_alloc))
+		goto mm_page_alloc_exit;
+
+#if USE_THREAD
+	/* Start worker thread */
+	gator_meminfo_run = true;
+	/* Since the mutex starts unlocked, memory values will be initialized */
+	if (IS_ERR(kthread_run(gator_meminfo_func, NULL, "gator_meminfo")))
+		goto kthread_run_exit;
+#else
+	setup_timer(&meminfo_wake_up_timer, meminfo_wake_up_handler, 0);
+#endif
+
+	return 0;
+
+#if USE_THREAD
+kthread_run_exit:
+	GATOR_UNREGISTER_TRACE(mm_page_alloc);
+#endif
+mm_page_alloc_exit:
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
+	GATOR_UNREGISTER_TRACE(mm_pagevec_free);
+#else
+	GATOR_UNREGISTER_TRACE(mm_page_free_batched);
+#endif
+mm_page_free_batched_exit:
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
+	GATOR_UNREGISTER_TRACE(mm_page_free_direct);
+#else
+	GATOR_UNREGISTER_TRACE(mm_page_free);
+#endif
+mm_page_free_exit:
+	return -1;
+}
+
+static void gator_events_meminfo_stop(void)
+{
+	if (meminfo_global_enabled) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
+		GATOR_UNREGISTER_TRACE(mm_page_free_direct);
+		GATOR_UNREGISTER_TRACE(mm_pagevec_free);
+#else
+		GATOR_UNREGISTER_TRACE(mm_page_free);
+		GATOR_UNREGISTER_TRACE(mm_page_free_batched);
+#endif
+		GATOR_UNREGISTER_TRACE(mm_page_alloc);
+
+#if USE_THREAD
+		/* Stop worker thread */
+		gator_meminfo_run = false;
+		up(&gator_meminfo_sem);
+#else
+		del_timer_sync(&meminfo_wake_up_timer);
+#endif
+	}
+}
+
+static void do_read(void)
+{
+	struct sysinfo info;
+	int i, len;
+	unsigned long long value;
+
+	meminfo_length = len = 0;
+
+	si_meminfo(&info);
+	for (i = 0; i < MEMINFO_TOTAL; i++) {
+		if (meminfo_enabled[i]) {
+			switch (i) {
+			case MEMINFO_MEMFREE:
+				value = info.freeram * PAGE_SIZE;
+				break;
+			case MEMINFO_MEMUSED:
+				/* pid -1 means system wide */
+				meminfo_buffer[len++] = 1;
+				meminfo_buffer[len++] = -1;
+				/* Emit value */
+				meminfo_buffer[len++] = meminfo_keys[MEMINFO_MEMUSED];
+				meminfo_buffer[len++] = (info.totalram - info.freeram) * PAGE_SIZE;
+				/* Clear pid */
+				meminfo_buffer[len++] = 1;
+				meminfo_buffer[len++] = 0;
+				continue;
+			case MEMINFO_BUFFERRAM:
+				value = info.bufferram * PAGE_SIZE;
+				break;
+			default:
+				value = 0;
+				break;
+			}
+			meminfo_buffer[len++] = meminfo_keys[i];
+			meminfo_buffer[len++] = value;
+		}
+	}
+
+	meminfo_length = len;
+	new_data_avail = true;
+}
+
+#if USE_THREAD
+
+static int gator_meminfo_func(void *data)
+{
+	for (;;) {
+		if (down_killable(&gator_meminfo_sem))
+			break;
+
+		/* Eat up any pending events */
+		while (!down_trylock(&gator_meminfo_sem))
+			;
+
+		if (!gator_meminfo_run)
+			break;
+
+		do_read();
+	}
+
+	return 0;
+}
+
+#else
+
+/* Must be run in process context as the kernel function si_meminfo() can sleep */
+static void wq_sched_handler(struct work_struct *wsptr)
+{
+	do_read();
+}
+
+static void meminfo_wake_up_handler(unsigned long unused_data)
+{
+	/* had to delay scheduling work as attempting to schedule work during the context switch is illegal in kernel versions 3.5 and greater */
+	schedule_work(&work);
+}
+
+#endif
+
+static int gator_events_meminfo_read(long long **buffer)
+{
+#if !USE_THREAD
+	static unsigned int last_mem_event;
+#endif
+
+	if (!on_primary_core() || !meminfo_global_enabled)
+		return 0;
+
+#if !USE_THREAD
+	if (last_mem_event != mem_event) {
+		last_mem_event = mem_event;
+		mod_timer(&meminfo_wake_up_timer, jiffies + 1);
+	}
+#endif
+
+	if (!new_data_avail)
+		return 0;
+
+	new_data_avail = false;
+
+	if (buffer)
+		*buffer = meminfo_buffer;
+
+	return meminfo_length;
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 34) && LINUX_VERSION_CODE < KERNEL_VERSION(3, 4, 0)
+
+static inline unsigned long gator_get_mm_counter(struct mm_struct *mm, int member)
+{
+#ifdef SPLIT_RSS_COUNTING
+	long val = atomic_long_read(&mm->rss_stat.count[member]);
+
+	if (val < 0)
+		val = 0;
+	return (unsigned long)val;
+#else
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 0, 0)
+	return mm->rss_stat.count[member];
+#else
+	return atomic_long_read(&mm->rss_stat.count[member]);
+#endif
+#endif
+}
+
+#define get_mm_counter(mm, member) gator_get_mm_counter(mm, member)
+
+#endif
+
+static int gator_events_meminfo_read_proc(long long **buffer, struct task_struct *task)
+{
+	struct mm_struct *mm;
+	u64 share = 0;
+	int i;
+	long long value;
+	int len = 0;
+	int cpu = get_physical_cpu();
+	long long *buf = per_cpu(proc_buffer, cpu);
+
+	if (!proc_global_enabled)
+		return 0;
+
+	/* Collect the memory stats of the process instead of the thread */
+	if (task->group_leader != NULL)
+		task = task->group_leader;
+
+	/* get_task_mm/mmput is not needed in this context because the task and it's mm are required as part of the sched_switch */
+	mm = task->mm;
+	if (mm == NULL)
+		return 0;
+
+	/* Derived from task_statm in fs/proc/task_mmu.c */
+	if (meminfo_enabled[MEMINFO_MEMUSED] || proc_enabled[PROC_SHARE]) {
+		share = get_mm_counter(mm,
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 34)
+							   file_rss
+#else
+							   MM_FILEPAGES
+#endif
+							   );
+	}
+
+	/* key of 1 indicates a pid */
+	buf[len++] = 1;
+	buf[len++] = task->pid;
+
+	for (i = 0; i < PROC_COUNT; ++i) {
+		if (proc_enabled[i]) {
+			switch (i) {
+			case PROC_SIZE:
+				value = mm->total_vm;
+				break;
+			case PROC_SHARE:
+				value = share;
+				break;
+			case PROC_TEXT:
+				value = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> PAGE_SHIFT;
+				break;
+			case PROC_DATA:
+				value = mm->total_vm - mm->shared_vm;
+				break;
+			}
+
+			buf[len++] = proc_keys[i];
+			buf[len++] = value * PAGE_SIZE;
+		}
+	}
+
+	if (meminfo_enabled[MEMINFO_MEMUSED]) {
+		value = share + get_mm_counter(mm,
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 32) && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 34)
+									   anon_rss
+#else
+									   MM_ANONPAGES
+#endif
+									   );
+		/* Send resident for this pid */
+		buf[len++] = meminfo_keys[MEMINFO_MEMUSED];
+		buf[len++] = value * PAGE_SIZE;
+	}
+
+	/* Clear pid */
+	buf[len++] = 1;
+	buf[len++] = 0;
+
+	if (buffer)
+		*buffer = buf;
+
+	return len;
+}
+
+static struct gator_interface gator_events_meminfo_interface = {
+	.create_files = gator_events_meminfo_create_files,
+	.start = gator_events_meminfo_start,
+	.stop = gator_events_meminfo_stop,
+	.read64 = gator_events_meminfo_read,
+	.read_proc = gator_events_meminfo_read_proc,
+};
+
+int gator_events_meminfo_init(void)
+{
+	int i;
+
+	meminfo_global_enabled = 0;
+	for (i = 0; i < MEMINFO_TOTAL; i++) {
+		meminfo_enabled[i] = 0;
+		meminfo_keys[i] = gator_events_get_key();
+	}
+
+	proc_global_enabled = 0;
+	for (i = 0; i < PROC_COUNT; ++i) {
+		proc_enabled[i] = 0;
+		proc_keys[i] = gator_events_get_key();
+	}
+
+	return gator_events_install(&gator_events_meminfo_interface);
+}
diff --git a/drivers/gator/gator_events_mmapped.c b/drivers/gator/gator_events_mmapped.c
new file mode 100644
index 000000000000..6b2af995ed41
--- /dev/null
+++ b/drivers/gator/gator_events_mmapped.c
@@ -0,0 +1,209 @@
+/*
+ * Example events provider
+ *
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Similar entries to those below must be present in the events.xml file.
+ * To add them to the events.xml, create an events-mmap.xml with the
+ * following contents and rebuild gatord:
+ *
+ * <category name="mmapped">
+ *   <event counter="mmapped_cnt0" title="Simulated1" name="Sine" display="maximum" class="absolute" description="Sort-of-sine"/>
+ *   <event counter="mmapped_cnt1" title="Simulated2" name="Triangle" display="maximum" class="absolute" description="Triangular wave"/>
+ *   <event counter="mmapped_cnt2" title="Simulated3" name="PWM" display="maximum" class="absolute" description="PWM Signal"/>
+ * </category>
+ *
+ * When adding custom events, be sure to do the following:
+ * - add any needed .c files to the gator driver Makefile
+ * - call gator_events_install in the events init function
+ * - add the init function to GATOR_EVENTS_LIST in gator_main.c
+ * - add a new events-*.xml file to the gator daemon and rebuild
+ *
+ * Troubleshooting:
+ * - verify the new events are part of events.xml, which is created when building the daemon
+ * - verify the new events exist at /dev/gator/events/ once gatord is launched
+ * - verify the counter name in the XML matches the name at /dev/gator/events
+ */
+
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/ratelimit.h>
+
+#include "gator.h"
+
+#define MMAPPED_COUNTERS_NUM 3
+
+static int mmapped_global_enabled;
+
+static struct {
+	unsigned long enabled;
+	unsigned long key;
+} mmapped_counters[MMAPPED_COUNTERS_NUM];
+
+static int mmapped_buffer[MMAPPED_COUNTERS_NUM * 2];
+
+static s64 prev_time;
+
+/* Adds mmapped_cntX directories and enabled, event, and key files to /dev/gator/events */
+static int gator_events_mmapped_create_files(struct super_block *sb,
+					     struct dentry *root)
+{
+	int i;
+
+	for (i = 0; i < MMAPPED_COUNTERS_NUM; i++) {
+		char buf[16];
+		struct dentry *dir;
+
+		snprintf(buf, sizeof(buf), "mmapped_cnt%d", i);
+		dir = gatorfs_mkdir(sb, root, buf);
+		if (WARN_ON(!dir))
+			return -1;
+		gatorfs_create_ulong(sb, dir, "enabled",
+				     &mmapped_counters[i].enabled);
+		gatorfs_create_ro_ulong(sb, dir, "key",
+					&mmapped_counters[i].key);
+	}
+
+	return 0;
+}
+
+static int gator_events_mmapped_start(void)
+{
+	int i;
+	struct timespec ts;
+
+	getnstimeofday(&ts);
+	prev_time = timespec_to_ns(&ts);
+
+	mmapped_global_enabled = 0;
+	for (i = 0; i < MMAPPED_COUNTERS_NUM; i++) {
+		if (mmapped_counters[i].enabled) {
+			mmapped_global_enabled = 1;
+			break;
+		}
+	}
+
+	return 0;
+}
+
+static void gator_events_mmapped_stop(void)
+{
+}
+
+/* This function "simulates" counters, generating values of fancy
+ * functions like sine or triangle... */
+static int mmapped_simulate(int counter, int delta_in_us)
+{
+	int result = 0;
+
+	switch (counter) {
+	case 0:		/* sort-of-sine */
+		{
+			static int t;
+			int x;
+
+			t += delta_in_us;
+			if (t > 2048000)
+				t = 0;
+
+			if (t % 1024000 < 512000)
+				x = 512000 - (t % 512000);
+			else
+				x = t % 512000;
+
+			result = 32 * x / 512000;
+			result = result * result;
+
+			if (t < 1024000)
+				result = 1922 - result;
+		}
+		break;
+	case 1:		/* triangle */
+		{
+			static int v, d = 1;
+
+			v = v + d * delta_in_us;
+			if (v < 0) {
+				v = 0;
+				d = 1;
+			} else if (v > 1000000) {
+				v = 1000000;
+				d = -1;
+			}
+
+			result = v;
+		}
+		break;
+	case 2:		/* PWM signal */
+		{
+			static int dc, x, t;
+
+			t += delta_in_us;
+			if (t > 1000000)
+				t = 0;
+			if (x / 1000000 != (x + delta_in_us) / 1000000)
+				dc = (dc + 100000) % 1000000;
+			x += delta_in_us;
+
+			result = t < dc ? 0 : 10;
+		}
+		break;
+	}
+
+	return result;
+}
+
+static int gator_events_mmapped_read(int **buffer, bool sched_switch)
+{
+	int i;
+	int len = 0;
+	int delta_in_us;
+	struct timespec ts;
+	s64 time;
+
+	/* System wide counters - read from one core only */
+	if (!on_primary_core() || !mmapped_global_enabled)
+		return 0;
+
+	getnstimeofday(&ts);
+	time = timespec_to_ns(&ts);
+	delta_in_us = (int)(time - prev_time) / 1000;
+	prev_time = time;
+
+	for (i = 0; i < MMAPPED_COUNTERS_NUM; i++) {
+		if (mmapped_counters[i].enabled) {
+			mmapped_buffer[len++] = mmapped_counters[i].key;
+			mmapped_buffer[len++] =
+			    mmapped_simulate(i, delta_in_us);
+		}
+	}
+
+	if (buffer)
+		*buffer = mmapped_buffer;
+
+	return len;
+}
+
+static struct gator_interface gator_events_mmapped_interface = {
+	.create_files = gator_events_mmapped_create_files,
+	.start = gator_events_mmapped_start,
+	.stop = gator_events_mmapped_stop,
+	.read = gator_events_mmapped_read,
+};
+
+/* Must not be static! */
+int __init gator_events_mmapped_init(void)
+{
+	int i;
+
+	for (i = 0; i < MMAPPED_COUNTERS_NUM; i++) {
+		mmapped_counters[i].enabled = 0;
+		mmapped_counters[i].key = gator_events_get_key();
+	}
+
+	return gator_events_install(&gator_events_mmapped_interface);
+}
diff --git a/drivers/gator/gator_events_net.c b/drivers/gator/gator_events_net.c
new file mode 100644
index 000000000000..d21b4db7b77c
--- /dev/null
+++ b/drivers/gator/gator_events_net.c
@@ -0,0 +1,172 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#include "gator.h"
+#include <linux/netdevice.h>
+#include <linux/hardirq.h>
+
+#define NETRX		0
+#define NETTX		1
+#define TOTALNET	2
+
+static ulong netrx_enabled;
+static ulong nettx_enabled;
+static ulong netrx_key;
+static ulong nettx_key;
+static int rx_total, tx_total;
+static ulong netPrev[TOTALNET];
+static int netGet[TOTALNET * 4];
+
+static struct timer_list net_wake_up_timer;
+
+/* Must be run in process context as the kernel function dev_get_stats() can sleep */
+static void get_network_stats(struct work_struct *wsptr)
+{
+	int rx = 0, tx = 0;
+	struct net_device *dev;
+
+	for_each_netdev(&init_net, dev) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
+		const struct net_device_stats *stats = dev_get_stats(dev);
+#else
+		struct rtnl_link_stats64 temp;
+		const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
+#endif
+		rx += stats->rx_bytes;
+		tx += stats->tx_bytes;
+	}
+	rx_total = rx;
+	tx_total = tx;
+}
+
+DECLARE_WORK(wq_get_stats, get_network_stats);
+
+static void net_wake_up_handler(unsigned long unused_data)
+{
+	/* had to delay scheduling work as attempting to schedule work during the context switch is illegal in kernel versions 3.5 and greater */
+	schedule_work(&wq_get_stats);
+}
+
+static void calculate_delta(int *rx, int *tx)
+{
+	int rx_calc, tx_calc;
+
+	rx_calc = (int)(rx_total - netPrev[NETRX]);
+	if (rx_calc < 0)
+		rx_calc = 0;
+	netPrev[NETRX] += rx_calc;
+
+	tx_calc = (int)(tx_total - netPrev[NETTX]);
+	if (tx_calc < 0)
+		tx_calc = 0;
+	netPrev[NETTX] += tx_calc;
+
+	*rx = rx_calc;
+	*tx = tx_calc;
+}
+
+static int gator_events_net_create_files(struct super_block *sb, struct dentry *root)
+{
+	/* Network counters are not currently supported in RT-Preempt full because mod_timer is used */
+#ifndef CONFIG_PREEMPT_RT_FULL
+	struct dentry *dir;
+
+	dir = gatorfs_mkdir(sb, root, "Linux_net_rx");
+	if (!dir)
+		return -1;
+	gatorfs_create_ulong(sb, dir, "enabled", &netrx_enabled);
+	gatorfs_create_ro_ulong(sb, dir, "key", &netrx_key);
+
+	dir = gatorfs_mkdir(sb, root, "Linux_net_tx");
+	if (!dir)
+		return -1;
+	gatorfs_create_ulong(sb, dir, "enabled", &nettx_enabled);
+	gatorfs_create_ro_ulong(sb, dir, "key", &nettx_key);
+#endif
+
+	return 0;
+}
+
+static int gator_events_net_start(void)
+{
+	get_network_stats(0);
+	netPrev[NETRX] = rx_total;
+	netPrev[NETTX] = tx_total;
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
+	setup_timer(&net_wake_up_timer, net_wake_up_handler, 0);
+#else
+	setup_deferrable_timer_on_stack(&net_wake_up_timer, net_wake_up_handler, 0);
+#endif
+	return 0;
+}
+
+static void gator_events_net_stop(void)
+{
+	del_timer_sync(&net_wake_up_timer);
+	netrx_enabled = 0;
+	nettx_enabled = 0;
+}
+
+static int gator_events_net_read(int **buffer, bool sched_switch)
+{
+	int len, rx_delta, tx_delta;
+	static int last_rx_delta, last_tx_delta;
+
+	if (!on_primary_core())
+		return 0;
+
+	if (!netrx_enabled && !nettx_enabled)
+		return 0;
+
+	mod_timer(&net_wake_up_timer, jiffies + 1);
+
+	calculate_delta(&rx_delta, &tx_delta);
+
+	len = 0;
+	if (netrx_enabled && last_rx_delta != rx_delta) {
+		last_rx_delta = rx_delta;
+		netGet[len++] = netrx_key;
+		/* indicates to Streamline that rx_delta bytes were transmitted now, not since the last message */
+		netGet[len++] = 0;
+		netGet[len++] = netrx_key;
+		netGet[len++] = rx_delta;
+	}
+
+	if (nettx_enabled && last_tx_delta != tx_delta) {
+		last_tx_delta = tx_delta;
+		netGet[len++] = nettx_key;
+		/* indicates to Streamline that tx_delta bytes were transmitted now, not since the last message */
+		netGet[len++] = 0;
+		netGet[len++] = nettx_key;
+		netGet[len++] = tx_delta;
+	}
+
+	if (buffer)
+		*buffer = netGet;
+
+	return len;
+}
+
+static struct gator_interface gator_events_net_interface = {
+	.create_files = gator_events_net_create_files,
+	.start = gator_events_net_start,
+	.stop = gator_events_net_stop,
+	.read = gator_events_net_read,
+};
+
+int gator_events_net_init(void)
+{
+	netrx_key = gator_events_get_key();
+	nettx_key = gator_events_get_key();
+
+	netrx_enabled = 0;
+	nettx_enabled = 0;
+
+	return gator_events_install(&gator_events_net_interface);
+}
diff --git a/drivers/gator/gator_events_perf_pmu.c b/drivers/gator/gator_events_perf_pmu.c
new file mode 100644
index 000000000000..47cf278e508b
--- /dev/null
+++ b/drivers/gator/gator_events_perf_pmu.c
@@ -0,0 +1,574 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "gator.h"
+
+/* gator_events_armvX.c is used for Linux 2.6.x */
+#if GATOR_PERF_PMU_SUPPORT
+
+#include <linux/io.h>
+#ifdef CONFIG_OF
+#include <linux/of_address.h>
+#endif
+#include <linux/perf_event.h>
+#include <linux/slab.h>
+
+extern bool event_based_sampling;
+
+/* Maximum number of per-core counters - currently reserves enough space for two full hardware PMUs for big.LITTLE */
+#define CNTMAX 16
+#define CCI_400 4
+#define CCN_5XX 8
+/* Maximum number of uncore counters */
+/* + 1 for the cci-400 cycles counter */
+/* + 1 for the CCN-5xx cycles counter */
+#define UCCNT (CCI_400 + 1 + CCN_5XX + 1)
+
+/* Default to 0 if unable to probe the revision which was the previous behavior */
+#define DEFAULT_CCI_REVISION 0
+
+/* A gator_attr is needed for every counter */
+struct gator_attr {
+	/* Set once in gator_events_perf_pmu_*_init - the name of the event in the gatorfs */
+	char name[40];
+	/* Exposed in gatorfs - set by gatord to enable this counter */
+	unsigned long enabled;
+	/* Set once in gator_events_perf_pmu_*_init - the perf type to use, see perf_type_id in the perf_event.h header file. */
+	unsigned long type;
+	/* Exposed in gatorfs - set by gatord to select the event to collect */
+	unsigned long event;
+	/* Exposed in gatorfs - set by gatord with the sample period to use and enable EBS for this counter */
+	unsigned long count;
+	/* Exposed as read only in gatorfs - set once in __attr_init as the key to use in the APC data */
+	unsigned long key;
+};
+
+/* Per-core counter attributes */
+static struct gator_attr attrs[CNTMAX];
+/* Number of initialized per-core counters */
+static int attr_count;
+/* Uncore counter attributes */
+static struct gator_attr uc_attrs[UCCNT];
+/* Number of initialized uncore counters */
+static int uc_attr_count;
+
+struct gator_event {
+	int curr;
+	int prev;
+	int prev_delta;
+	bool zero;
+	struct perf_event *pevent;
+	struct perf_event_attr *pevent_attr;
+};
+
+static DEFINE_PER_CPU(struct gator_event[CNTMAX], events);
+static struct gator_event uc_events[UCCNT];
+static DEFINE_PER_CPU(int[(CNTMAX + UCCNT)*2], perf_cnt);
+
+static void gator_events_perf_pmu_stop(void);
+
+static int __create_files(struct super_block *sb, struct dentry *root, struct gator_attr *const attr)
+{
+	struct dentry *dir;
+
+	if (attr->name[0] == '\0')
+		return 0;
+	dir = gatorfs_mkdir(sb, root, attr->name);
+	if (!dir)
+		return -1;
+	gatorfs_create_ulong(sb, dir, "enabled", &attr->enabled);
+	gatorfs_create_ulong(sb, dir, "count", &attr->count);
+	gatorfs_create_ro_ulong(sb, dir, "key", &attr->key);
+	gatorfs_create_ulong(sb, dir, "event", &attr->event);
+
+	return 0;
+}
+
+static int gator_events_perf_pmu_create_files(struct super_block *sb, struct dentry *root)
+{
+	int cnt;
+
+	for (cnt = 0; cnt < attr_count; cnt++) {
+		if (__create_files(sb, root, &attrs[cnt]) != 0)
+			return -1;
+	}
+
+	for (cnt = 0; cnt < uc_attr_count; cnt++) {
+		if (__create_files(sb, root, &uc_attrs[cnt]) != 0)
+			return -1;
+	}
+
+	return 0;
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 1, 0)
+static void ebs_overflow_handler(struct perf_event *event, int unused, struct perf_sample_data *data, struct pt_regs *regs)
+#else
+static void ebs_overflow_handler(struct perf_event *event, struct perf_sample_data *data, struct pt_regs *regs)
+#endif
+{
+	gator_backtrace_handler(regs);
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 1, 0)
+static void dummy_handler(struct perf_event *event, int unused, struct perf_sample_data *data, struct pt_regs *regs)
+#else
+static void dummy_handler(struct perf_event *event, struct perf_sample_data *data, struct pt_regs *regs)
+#endif
+{
+	/* Required as perf_event_create_kernel_counter() requires an overflow handler, even though all we do is poll */
+}
+
+static int gator_events_perf_pmu_read(int **buffer, bool sched_switch);
+
+static int gator_events_perf_pmu_online(int **buffer, bool migrate)
+{
+	return gator_events_perf_pmu_read(buffer, false);
+}
+
+static void __online_dispatch(int cpu, bool migrate, struct gator_attr *const attr, struct gator_event *const event)
+{
+	perf_overflow_handler_t handler;
+
+	event->zero = true;
+
+	if (event->pevent != NULL || event->pevent_attr == 0 || migrate)
+		return;
+
+	if (attr->count > 0)
+		handler = ebs_overflow_handler;
+	else
+		handler = dummy_handler;
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 1, 0)
+	event->pevent = perf_event_create_kernel_counter(event->pevent_attr, cpu, 0, handler);
+#else
+	event->pevent = perf_event_create_kernel_counter(event->pevent_attr, cpu, 0, handler, 0);
+#endif
+	if (IS_ERR(event->pevent)) {
+		pr_debug("gator: unable to online a counter on cpu %d\n", cpu);
+		event->pevent = NULL;
+		return;
+	}
+
+	if (event->pevent->state != PERF_EVENT_STATE_ACTIVE) {
+		pr_debug("gator: inactive counter on cpu %d\n", cpu);
+		perf_event_release_kernel(event->pevent);
+		event->pevent = NULL;
+		return;
+	}
+}
+
+static void gator_events_perf_pmu_online_dispatch(int cpu, bool migrate)
+{
+	int cnt;
+
+	cpu = pcpu_to_lcpu(cpu);
+
+	for (cnt = 0; cnt < attr_count; cnt++)
+		__online_dispatch(cpu, migrate, &attrs[cnt], &per_cpu(events, cpu)[cnt]);
+
+	if (cpu == 0) {
+		for (cnt = 0; cnt < uc_attr_count; cnt++)
+			__online_dispatch(cpu, migrate, &uc_attrs[cnt], &uc_events[cnt]);
+	}
+}
+
+static void __offline_dispatch(int cpu, struct gator_event *const event)
+{
+	struct perf_event *pe = NULL;
+
+	if (event->pevent) {
+		pe = event->pevent;
+		event->pevent = NULL;
+	}
+
+	if (pe)
+		perf_event_release_kernel(pe);
+}
+
+static void gator_events_perf_pmu_offline_dispatch(int cpu, bool migrate)
+{
+	int cnt;
+
+	if (migrate)
+		return;
+	cpu = pcpu_to_lcpu(cpu);
+
+	for (cnt = 0; cnt < attr_count; cnt++)
+		__offline_dispatch(cpu, &per_cpu(events, cpu)[cnt]);
+
+	if (cpu == 0) {
+		for (cnt = 0; cnt < uc_attr_count; cnt++)
+			__offline_dispatch(cpu, &uc_events[cnt]);
+	}
+}
+
+static int __check_ebs(struct gator_attr *const attr)
+{
+	if (attr->count > 0) {
+		if (!event_based_sampling) {
+			event_based_sampling = true;
+		} else {
+			pr_warning("gator: Only one ebs counter is allowed\n");
+			return -1;
+		}
+	}
+
+	return 0;
+}
+
+static int __start(struct gator_attr *const attr, struct gator_event *const event)
+{
+	u32 size = sizeof(struct perf_event_attr);
+
+	event->pevent = NULL;
+	/* Skip disabled counters */
+	if (!attr->enabled)
+		return 0;
+
+	event->prev = 0;
+	event->curr = 0;
+	event->prev_delta = 0;
+	event->pevent_attr = kmalloc(size, GFP_KERNEL);
+	if (!event->pevent_attr) {
+		gator_events_perf_pmu_stop();
+		return -1;
+	}
+
+	memset(event->pevent_attr, 0, size);
+	event->pevent_attr->type = attr->type;
+	event->pevent_attr->size = size;
+	event->pevent_attr->config = attr->event;
+	event->pevent_attr->sample_period = attr->count;
+	event->pevent_attr->pinned = 1;
+
+	return 0;
+}
+
+static int gator_events_perf_pmu_start(void)
+{
+	int cnt, cpu;
+
+	event_based_sampling = false;
+	for (cnt = 0; cnt < attr_count; cnt++) {
+		if (__check_ebs(&attrs[cnt]) != 0)
+			return -1;
+	}
+
+	for (cnt = 0; cnt < uc_attr_count; cnt++) {
+		if (__check_ebs(&uc_attrs[cnt]) != 0)
+			return -1;
+	}
+
+	for_each_present_cpu(cpu) {
+		for (cnt = 0; cnt < attr_count; cnt++) {
+			if (__start(&attrs[cnt], &per_cpu(events, cpu)[cnt]) != 0)
+				return -1;
+		}
+	}
+
+	for (cnt = 0; cnt < uc_attr_count; cnt++) {
+		if (__start(&uc_attrs[cnt], &uc_events[cnt]) != 0)
+			return -1;
+	}
+
+	return 0;
+}
+
+static void __event_stop(struct gator_event *const event)
+{
+	kfree(event->pevent_attr);
+	event->pevent_attr = NULL;
+}
+
+static void __attr_stop(struct gator_attr *const attr)
+{
+	attr->enabled = 0;
+	attr->event = 0;
+	attr->count = 0;
+}
+
+static void gator_events_perf_pmu_stop(void)
+{
+	unsigned int cnt, cpu;
+
+	for_each_present_cpu(cpu) {
+		for (cnt = 0; cnt < attr_count; cnt++)
+			__event_stop(&per_cpu(events, cpu)[cnt]);
+	}
+
+	for (cnt = 0; cnt < uc_attr_count; cnt++)
+		__event_stop(&uc_events[cnt]);
+
+	for (cnt = 0; cnt < attr_count; cnt++)
+		__attr_stop(&attrs[cnt]);
+
+	for (cnt = 0; cnt < uc_attr_count; cnt++)
+		__attr_stop(&uc_attrs[cnt]);
+}
+
+static void __read(int *const len, int cpu, struct gator_attr *const attr, struct gator_event *const event)
+{
+	int delta;
+	struct perf_event *const ev = event->pevent;
+
+	if (ev != NULL && ev->state == PERF_EVENT_STATE_ACTIVE) {
+		/* After creating the perf counter in __online_dispatch, there
+		 * is a race condition between gator_events_perf_pmu_online and
+		 * gator_events_perf_pmu_read. So have
+		 * gator_events_perf_pmu_online call gator_events_perf_pmu_read
+		 * and in __read check to see if it's the first call after
+		 * __online_dispatch and if so, run the online code.
+		 */
+		if (event->zero) {
+			ev->pmu->read(ev);
+			event->prev = event->curr = local64_read(&ev->count);
+			event->prev_delta = 0;
+			per_cpu(perf_cnt, cpu)[(*len)++] = attr->key;
+			per_cpu(perf_cnt, cpu)[(*len)++] = 0;
+			event->zero = false;
+		} else {
+			ev->pmu->read(ev);
+			event->curr = local64_read(&ev->count);
+			delta = event->curr - event->prev;
+			if (delta != 0 || delta != event->prev_delta) {
+				event->prev_delta = delta;
+				event->prev = event->curr;
+				per_cpu(perf_cnt, cpu)[(*len)++] = attr->key;
+				if (delta < 0)
+					delta *= -1;
+				per_cpu(perf_cnt, cpu)[(*len)++] = delta;
+			}
+		}
+	}
+}
+
+static int gator_events_perf_pmu_read(int **buffer, bool sched_switch)
+{
+	int cnt, len = 0;
+	const int cpu = get_logical_cpu();
+
+	for (cnt = 0; cnt < attr_count; cnt++)
+		__read(&len, cpu, &attrs[cnt], &per_cpu(events, cpu)[cnt]);
+
+	if (cpu == 0) {
+		for (cnt = 0; cnt < uc_attr_count; cnt++)
+			__read(&len, cpu, &uc_attrs[cnt], &uc_events[cnt]);
+	}
+
+	if (buffer)
+		*buffer = per_cpu(perf_cnt, cpu);
+
+	return len;
+}
+
+static struct gator_interface gator_events_perf_pmu_interface = {
+	.create_files = gator_events_perf_pmu_create_files,
+	.start = gator_events_perf_pmu_start,
+	.stop = gator_events_perf_pmu_stop,
+	.online = gator_events_perf_pmu_online,
+	.online_dispatch = gator_events_perf_pmu_online_dispatch,
+	.offline_dispatch = gator_events_perf_pmu_offline_dispatch,
+	.read = gator_events_perf_pmu_read,
+};
+
+static void __attr_init(struct gator_attr *const attr)
+{
+	attr->name[0] = '\0';
+	attr->enabled = 0;
+	attr->type = 0;
+	attr->event = 0;
+	attr->count = 0;
+	attr->key = gator_events_get_key();
+}
+
+#ifdef CONFIG_OF
+
+static const struct of_device_id arm_cci_matches[] = {
+	{.compatible = "arm,cci-400" },
+	{},
+};
+
+static int probe_cci_revision(void)
+{
+	struct device_node *np;
+	struct resource res;
+	void __iomem *cci_ctrl_base;
+	int rev;
+	int ret = DEFAULT_CCI_REVISION;
+
+	np = of_find_matching_node(NULL, arm_cci_matches);
+	if (!np)
+		return ret;
+
+	if (of_address_to_resource(np, 0, &res))
+		goto node_put;
+
+	cci_ctrl_base = ioremap(res.start, resource_size(&res));
+
+	rev = (readl_relaxed(cci_ctrl_base + 0xfe8) >> 4) & 0xf;
+
+	if (rev <= 4)
+		ret = 0;
+	else if (rev <= 6)
+		ret = 1;
+
+	iounmap(cci_ctrl_base);
+
+ node_put:
+	of_node_put(np);
+
+	return ret;
+}
+
+#else
+
+static int probe_cci_revision(void)
+{
+	return DEFAULT_CCI_REVISION;
+}
+
+#endif
+
+static void gator_events_perf_pmu_uncore_init(const char *const name, const int type, const int count)
+{
+	int cnt;
+
+	snprintf(uc_attrs[uc_attr_count].name, sizeof(uc_attrs[uc_attr_count].name), "%s_ccnt", name);
+	uc_attrs[uc_attr_count].type = type;
+	++uc_attr_count;
+
+	for (cnt = 0; cnt < count; ++cnt, ++uc_attr_count) {
+		struct gator_attr *const attr = &uc_attrs[uc_attr_count];
+
+		snprintf(attr->name, sizeof(attr->name), "%s_cnt%d", name, cnt);
+		attr->type = type;
+	}
+}
+
+static void gator_events_perf_pmu_cci_init(const int type)
+{
+	const char *cci_name;
+
+	switch (probe_cci_revision()) {
+	case 0:
+		cci_name = "CCI_400";
+		break;
+	case 1:
+		cci_name = "CCI_400-r1";
+		break;
+	default:
+		pr_debug("gator: unrecognized cci-400 revision\n");
+		return;
+	}
+
+	gator_events_perf_pmu_uncore_init(cci_name, type, CCI_400);
+}
+
+static void gator_events_perf_pmu_cpu_init(const struct gator_cpu *const gator_cpu, const int type)
+{
+	int cnt;
+
+	snprintf(attrs[attr_count].name, sizeof(attrs[attr_count].name), "%s_ccnt", gator_cpu->pmnc_name);
+	attrs[attr_count].type = type;
+	++attr_count;
+
+	for (cnt = 0; cnt < gator_cpu->pmnc_counters; ++cnt, ++attr_count) {
+		struct gator_attr *const attr = &attrs[attr_count];
+
+		snprintf(attr->name, sizeof(attr->name), "%s_cnt%d", gator_cpu->pmnc_name, cnt);
+		attr->type = type;
+	}
+}
+
+int gator_events_perf_pmu_init(void)
+{
+	struct perf_event_attr pea;
+	struct perf_event *pe;
+	const struct gator_cpu *gator_cpu;
+	int type;
+	int cpu;
+	int cnt;
+	bool found_cpu = false;
+
+	for (cnt = 0; cnt < CNTMAX; cnt++)
+		__attr_init(&attrs[cnt]);
+	for (cnt = 0; cnt < UCCNT; cnt++)
+		__attr_init(&uc_attrs[cnt]);
+
+	memset(&pea, 0, sizeof(pea));
+	pea.size = sizeof(pea);
+	pea.config = 0xFF;
+	attr_count = 0;
+	uc_attr_count = 0;
+	for (type = PERF_TYPE_MAX; type < 0x20; ++type) {
+		pea.type = type;
+
+		/* A particular PMU may work on some but not all cores, so try on each core */
+		pe = NULL;
+		for_each_present_cpu(cpu) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 1, 0)
+			pe = perf_event_create_kernel_counter(&pea, cpu, 0, dummy_handler);
+#else
+			pe = perf_event_create_kernel_counter(&pea, cpu, 0, dummy_handler, 0);
+#endif
+			if (!IS_ERR(pe))
+				break;
+		}
+		/* Assume that valid PMUs are contiguous */
+		if (IS_ERR(pe)) {
+			pea.config = 0xff00;
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 1, 0)
+			pe = perf_event_create_kernel_counter(&pea, 0, 0, dummy_handler);
+#else
+			pe = perf_event_create_kernel_counter(&pea, 0, 0, dummy_handler, 0);
+#endif
+			if (IS_ERR(pe))
+				break;
+		}
+
+		if (pe->pmu != NULL && type == pe->pmu->type) {
+			if (strcmp("CCI", pe->pmu->name) == 0 || strcmp("CCI_400", pe->pmu->name) == 0 || strcmp("CCI_400-r1", pe->pmu->name) == 0) {
+				gator_events_perf_pmu_cci_init(type);
+			} else if (strcmp("ccn", pe->pmu->name) == 0) {
+				gator_events_perf_pmu_uncore_init("ARM_CCN_5XX", type, CCN_5XX);
+			} else if ((gator_cpu = gator_find_cpu_by_pmu_name(pe->pmu->name)) != NULL) {
+				found_cpu = true;
+				gator_events_perf_pmu_cpu_init(gator_cpu, type);
+			}
+			/* Initialize gator_attrs for dynamic PMUs here */
+		}
+
+		perf_event_release_kernel(pe);
+	}
+
+	if (!found_cpu) {
+		const struct gator_cpu *const gator_cpu = gator_find_cpu_by_cpuid(gator_cpuid());
+
+		if (gator_cpu == NULL)
+			return -1;
+		gator_events_perf_pmu_cpu_init(gator_cpu, PERF_TYPE_RAW);
+	}
+
+	/* Initialize gator_attrs for non-dynamic PMUs here */
+
+	if (attr_count > CNTMAX) {
+		pr_err("gator: Too many perf counters\n");
+		return -1;
+	}
+
+	if (uc_attr_count > UCCNT) {
+		pr_err("gator: Too many perf uncore counters\n");
+		return -1;
+	}
+
+	return gator_events_install(&gator_events_perf_pmu_interface);
+}
+
+#endif
diff --git a/drivers/gator/gator_events_sched.c b/drivers/gator/gator_events_sched.c
new file mode 100644
index 000000000000..637107d6af1d
--- /dev/null
+++ b/drivers/gator/gator_events_sched.c
@@ -0,0 +1,113 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#include "gator.h"
+#include <trace/events/sched.h>
+
+#define SCHED_SWITCH	0
+#define SCHED_TOTAL		(SCHED_SWITCH+1)
+
+static ulong sched_switch_enabled;
+static ulong sched_switch_key;
+static DEFINE_PER_CPU(int[SCHED_TOTAL], schedCnt);
+static DEFINE_PER_CPU(int[SCHED_TOTAL * 2], schedGet);
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 35)
+GATOR_DEFINE_PROBE(sched_switch, TP_PROTO(struct rq *rq, struct task_struct *prev, struct task_struct *next))
+#else
+GATOR_DEFINE_PROBE(sched_switch, TP_PROTO(struct task_struct *prev, struct task_struct *next))
+#endif
+{
+	unsigned long flags;
+
+	/* disable interrupts to synchronize with gator_events_sched_read()
+	 * spinlocks not needed since percpu buffers are used
+	 */
+	local_irq_save(flags);
+	per_cpu(schedCnt, get_physical_cpu())[SCHED_SWITCH]++;
+	local_irq_restore(flags);
+}
+
+static int gator_events_sched_create_files(struct super_block *sb, struct dentry *root)
+{
+	struct dentry *dir;
+
+	/* switch */
+	dir = gatorfs_mkdir(sb, root, "Linux_sched_switch");
+	if (!dir)
+		return -1;
+	gatorfs_create_ulong(sb, dir, "enabled", &sched_switch_enabled);
+	gatorfs_create_ro_ulong(sb, dir, "key", &sched_switch_key);
+
+	return 0;
+}
+
+static int gator_events_sched_start(void)
+{
+	/* register tracepoints */
+	if (sched_switch_enabled)
+		if (GATOR_REGISTER_TRACE(sched_switch))
+			goto sched_switch_exit;
+	pr_debug("gator: registered scheduler event tracepoints\n");
+
+	return 0;
+
+	/* unregister tracepoints on error */
+sched_switch_exit:
+	pr_err("gator: scheduler event tracepoints failed to activate, please verify that tracepoints are enabled in the linux kernel\n");
+
+	return -1;
+}
+
+static void gator_events_sched_stop(void)
+{
+	if (sched_switch_enabled)
+		GATOR_UNREGISTER_TRACE(sched_switch);
+	pr_debug("gator: unregistered scheduler event tracepoints\n");
+
+	sched_switch_enabled = 0;
+}
+
+static int gator_events_sched_read(int **buffer, bool sched_switch)
+{
+	unsigned long flags;
+	int len, value;
+	int cpu = get_physical_cpu();
+
+	len = 0;
+	if (sched_switch_enabled) {
+		local_irq_save(flags);
+		value = per_cpu(schedCnt, cpu)[SCHED_SWITCH];
+		per_cpu(schedCnt, cpu)[SCHED_SWITCH] = 0;
+		local_irq_restore(flags);
+		per_cpu(schedGet, cpu)[len++] = sched_switch_key;
+		per_cpu(schedGet, cpu)[len++] = value;
+	}
+
+	if (buffer)
+		*buffer = per_cpu(schedGet, cpu);
+
+	return len;
+}
+
+static struct gator_interface gator_events_sched_interface = {
+	.create_files = gator_events_sched_create_files,
+	.start = gator_events_sched_start,
+	.stop = gator_events_sched_stop,
+	.read = gator_events_sched_read,
+};
+
+int gator_events_sched_init(void)
+{
+	sched_switch_enabled = 0;
+
+	sched_switch_key = gator_events_get_key();
+
+	return gator_events_install(&gator_events_sched_interface);
+}
diff --git a/drivers/gator/gator_events_scorpion.c b/drivers/gator/gator_events_scorpion.c
new file mode 100644
index 000000000000..49219362db09
--- /dev/null
+++ b/drivers/gator/gator_events_scorpion.c
@@ -0,0 +1,674 @@
+/**
+ * Copyright (C) ARM Limited 2011-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "gator.h"
+
+/* gator_events_perf_pmu.c is used if perf is supported */
+#if GATOR_NO_PERF_SUPPORT
+
+static const char *pmnc_name;
+static int pmnc_counters;
+
+/* Per-CPU PMNC: config reg */
+#define PMNC_E		(1 << 0)	/* Enable all counters */
+#define PMNC_P		(1 << 1)	/* Reset all counters */
+#define PMNC_C		(1 << 2)	/* Cycle counter reset */
+#define PMNC_D		(1 << 3)	/* CCNT counts every 64th cpu cycle */
+#define PMNC_X		(1 << 4)	/* Export to ETM */
+#define PMNC_DP		(1 << 5)	/* Disable CCNT if non-invasive debug */
+#define	PMNC_MASK	0x3f	/* Mask for writable bits */
+
+/* ccnt reg */
+#define CCNT_REG	(1 << 31)
+
+#define CCNT		0
+#define CNT0		1
+#define CNTMAX		(4+1)
+
+static unsigned long pmnc_enabled[CNTMAX];
+static unsigned long pmnc_event[CNTMAX];
+static unsigned long pmnc_key[CNTMAX];
+
+static DEFINE_PER_CPU(int[CNTMAX * 2], perfCnt);
+
+enum scorpion_perf_types {
+	SCORPION_ICACHE_EXPL_INV = 0x4c,
+	SCORPION_ICACHE_MISS = 0x4d,
+	SCORPION_ICACHE_ACCESS = 0x4e,
+	SCORPION_ICACHE_CACHEREQ_L2 = 0x4f,
+	SCORPION_ICACHE_NOCACHE_L2 = 0x50,
+	SCORPION_HIQUP_NOPED = 0x51,
+	SCORPION_DATA_ABORT = 0x52,
+	SCORPION_IRQ = 0x53,
+	SCORPION_FIQ = 0x54,
+	SCORPION_ALL_EXCPT = 0x55,
+	SCORPION_UNDEF = 0x56,
+	SCORPION_SVC = 0x57,
+	SCORPION_SMC = 0x58,
+	SCORPION_PREFETCH_ABORT = 0x59,
+	SCORPION_INDEX_CHECK = 0x5a,
+	SCORPION_NULL_CHECK = 0x5b,
+	SCORPION_EXPL_ICIALLU = 0x5c,
+	SCORPION_IMPL_ICIALLU = 0x5d,
+	SCORPION_NONICIALLU_BTAC_INV = 0x5e,
+	SCORPION_ICIMVAU_IMPL_ICIALLU = 0x5f,
+	SCORPION_SPIPE_ONLY_CYCLES = 0x60,
+	SCORPION_XPIPE_ONLY_CYCLES = 0x61,
+	SCORPION_DUAL_CYCLES = 0x62,
+	SCORPION_DISPATCH_ANY_CYCLES = 0x63,
+	SCORPION_FIFO_FULLBLK_CMT = 0x64,
+	SCORPION_FAIL_COND_INST = 0x65,
+	SCORPION_PASS_COND_INST = 0x66,
+	SCORPION_ALLOW_VU_CLK = 0x67,
+	SCORPION_VU_IDLE = 0x68,
+	SCORPION_ALLOW_L2_CLK = 0x69,
+	SCORPION_L2_IDLE = 0x6a,
+	SCORPION_DTLB_IMPL_INV_SCTLR_DACR = 0x6b,
+	SCORPION_DTLB_EXPL_INV = 0x6c,
+	SCORPION_DTLB_MISS = 0x6d,
+	SCORPION_DTLB_ACCESS = 0x6e,
+	SCORPION_ITLB_MISS = 0x6f,
+	SCORPION_ITLB_IMPL_INV = 0x70,
+	SCORPION_ITLB_EXPL_INV = 0x71,
+	SCORPION_UTLB_D_MISS = 0x72,
+	SCORPION_UTLB_D_ACCESS = 0x73,
+	SCORPION_UTLB_I_MISS = 0x74,
+	SCORPION_UTLB_I_ACCESS = 0x75,
+	SCORPION_UTLB_INV_ASID = 0x76,
+	SCORPION_UTLB_INV_MVA = 0x77,
+	SCORPION_UTLB_INV_ALL = 0x78,
+	SCORPION_S2_HOLD_RDQ_UNAVAIL = 0x79,
+	SCORPION_S2_HOLD = 0x7a,
+	SCORPION_S2_HOLD_DEV_OP = 0x7b,
+	SCORPION_S2_HOLD_ORDER = 0x7c,
+	SCORPION_S2_HOLD_BARRIER = 0x7d,
+	SCORPION_VIU_DUAL_CYCLE = 0x7e,
+	SCORPION_VIU_SINGLE_CYCLE = 0x7f,
+	SCORPION_VX_PIPE_WAR_STALL_CYCLES = 0x80,
+	SCORPION_VX_PIPE_WAW_STALL_CYCLES = 0x81,
+	SCORPION_VX_PIPE_RAW_STALL_CYCLES = 0x82,
+	SCORPION_VX_PIPE_LOAD_USE_STALL = 0x83,
+	SCORPION_VS_PIPE_WAR_STALL_CYCLES = 0x84,
+	SCORPION_VS_PIPE_WAW_STALL_CYCLES = 0x85,
+	SCORPION_VS_PIPE_RAW_STALL_CYCLES = 0x86,
+	SCORPION_EXCEPTIONS_INV_OPERATION = 0x87,
+	SCORPION_EXCEPTIONS_DIV_BY_ZERO = 0x88,
+	SCORPION_COND_INST_FAIL_VX_PIPE = 0x89,
+	SCORPION_COND_INST_FAIL_VS_PIPE = 0x8a,
+	SCORPION_EXCEPTIONS_OVERFLOW = 0x8b,
+	SCORPION_EXCEPTIONS_UNDERFLOW = 0x8c,
+	SCORPION_EXCEPTIONS_DENORM = 0x8d,
+#ifdef CONFIG_ARCH_MSM_SCORPIONMP
+	SCORPIONMP_NUM_BARRIERS = 0x8e,
+	SCORPIONMP_BARRIER_CYCLES = 0x8f,
+#else
+	SCORPION_BANK_AB_HIT = 0x8e,
+	SCORPION_BANK_AB_ACCESS = 0x8f,
+	SCORPION_BANK_CD_HIT = 0x90,
+	SCORPION_BANK_CD_ACCESS = 0x91,
+	SCORPION_BANK_AB_DSIDE_HIT = 0x92,
+	SCORPION_BANK_AB_DSIDE_ACCESS = 0x93,
+	SCORPION_BANK_CD_DSIDE_HIT = 0x94,
+	SCORPION_BANK_CD_DSIDE_ACCESS = 0x95,
+	SCORPION_BANK_AB_ISIDE_HIT = 0x96,
+	SCORPION_BANK_AB_ISIDE_ACCESS = 0x97,
+	SCORPION_BANK_CD_ISIDE_HIT = 0x98,
+	SCORPION_BANK_CD_ISIDE_ACCESS = 0x99,
+	SCORPION_ISIDE_RD_WAIT = 0x9a,
+	SCORPION_DSIDE_RD_WAIT = 0x9b,
+	SCORPION_BANK_BYPASS_WRITE = 0x9c,
+	SCORPION_BANK_AB_NON_CASTOUT = 0x9d,
+	SCORPION_BANK_AB_L2_CASTOUT = 0x9e,
+	SCORPION_BANK_CD_NON_CASTOUT = 0x9f,
+	SCORPION_BANK_CD_L2_CASTOUT = 0xa0,
+#endif
+	MSM_MAX_EVT
+};
+
+struct scorp_evt {
+	u32 evt_type;
+	u32 val;
+	u8 grp;
+	u32 evt_type_act;
+};
+
+static const struct scorp_evt sc_evt[] = {
+	{SCORPION_ICACHE_EXPL_INV, 0x80000500, 0, 0x4d},
+	{SCORPION_ICACHE_MISS, 0x80050000, 0, 0x4e},
+	{SCORPION_ICACHE_ACCESS, 0x85000000, 0, 0x4f},
+	{SCORPION_ICACHE_CACHEREQ_L2, 0x86000000, 0, 0x4f},
+	{SCORPION_ICACHE_NOCACHE_L2, 0x87000000, 0, 0x4f},
+	{SCORPION_HIQUP_NOPED, 0x80080000, 0, 0x4e},
+	{SCORPION_DATA_ABORT, 0x8000000a, 0, 0x4c},
+	{SCORPION_IRQ, 0x80000a00, 0, 0x4d},
+	{SCORPION_FIQ, 0x800a0000, 0, 0x4e},
+	{SCORPION_ALL_EXCPT, 0x8a000000, 0, 0x4f},
+	{SCORPION_UNDEF, 0x8000000b, 0, 0x4c},
+	{SCORPION_SVC, 0x80000b00, 0, 0x4d},
+	{SCORPION_SMC, 0x800b0000, 0, 0x4e},
+	{SCORPION_PREFETCH_ABORT, 0x8b000000, 0, 0x4f},
+	{SCORPION_INDEX_CHECK, 0x8000000c, 0, 0x4c},
+	{SCORPION_NULL_CHECK, 0x80000c00, 0, 0x4d},
+	{SCORPION_EXPL_ICIALLU, 0x8000000d, 0, 0x4c},
+	{SCORPION_IMPL_ICIALLU, 0x80000d00, 0, 0x4d},
+	{SCORPION_NONICIALLU_BTAC_INV, 0x800d0000, 0, 0x4e},
+	{SCORPION_ICIMVAU_IMPL_ICIALLU, 0x8d000000, 0, 0x4f},
+
+	{SCORPION_SPIPE_ONLY_CYCLES, 0x80000600, 1, 0x51},
+	{SCORPION_XPIPE_ONLY_CYCLES, 0x80060000, 1, 0x52},
+	{SCORPION_DUAL_CYCLES, 0x86000000, 1, 0x53},
+	{SCORPION_DISPATCH_ANY_CYCLES, 0x89000000, 1, 0x53},
+	{SCORPION_FIFO_FULLBLK_CMT, 0x8000000d, 1, 0x50},
+	{SCORPION_FAIL_COND_INST, 0x800d0000, 1, 0x52},
+	{SCORPION_PASS_COND_INST, 0x8d000000, 1, 0x53},
+	{SCORPION_ALLOW_VU_CLK, 0x8000000e, 1, 0x50},
+	{SCORPION_VU_IDLE, 0x80000e00, 1, 0x51},
+	{SCORPION_ALLOW_L2_CLK, 0x800e0000, 1, 0x52},
+	{SCORPION_L2_IDLE, 0x8e000000, 1, 0x53},
+
+	{SCORPION_DTLB_IMPL_INV_SCTLR_DACR, 0x80000001, 2, 0x54},
+	{SCORPION_DTLB_EXPL_INV, 0x80000100, 2, 0x55},
+	{SCORPION_DTLB_MISS, 0x80010000, 2, 0x56},
+	{SCORPION_DTLB_ACCESS, 0x81000000, 2, 0x57},
+	{SCORPION_ITLB_MISS, 0x80000200, 2, 0x55},
+	{SCORPION_ITLB_IMPL_INV, 0x80020000, 2, 0x56},
+	{SCORPION_ITLB_EXPL_INV, 0x82000000, 2, 0x57},
+	{SCORPION_UTLB_D_MISS, 0x80000003, 2, 0x54},
+	{SCORPION_UTLB_D_ACCESS, 0x80000300, 2, 0x55},
+	{SCORPION_UTLB_I_MISS, 0x80030000, 2, 0x56},
+	{SCORPION_UTLB_I_ACCESS, 0x83000000, 2, 0x57},
+	{SCORPION_UTLB_INV_ASID, 0x80000400, 2, 0x55},
+	{SCORPION_UTLB_INV_MVA, 0x80040000, 2, 0x56},
+	{SCORPION_UTLB_INV_ALL, 0x84000000, 2, 0x57},
+	{SCORPION_S2_HOLD_RDQ_UNAVAIL, 0x80000800, 2, 0x55},
+	{SCORPION_S2_HOLD, 0x88000000, 2, 0x57},
+	{SCORPION_S2_HOLD_DEV_OP, 0x80000900, 2, 0x55},
+	{SCORPION_S2_HOLD_ORDER, 0x80090000, 2, 0x56},
+	{SCORPION_S2_HOLD_BARRIER, 0x89000000, 2, 0x57},
+
+	{SCORPION_VIU_DUAL_CYCLE, 0x80000001, 4, 0x5c},
+	{SCORPION_VIU_SINGLE_CYCLE, 0x80000100, 4, 0x5d},
+	{SCORPION_VX_PIPE_WAR_STALL_CYCLES, 0x80000005, 4, 0x5c},
+	{SCORPION_VX_PIPE_WAW_STALL_CYCLES, 0x80000500, 4, 0x5d},
+	{SCORPION_VX_PIPE_RAW_STALL_CYCLES, 0x80050000, 4, 0x5e},
+	{SCORPION_VX_PIPE_LOAD_USE_STALL, 0x80000007, 4, 0x5c},
+	{SCORPION_VS_PIPE_WAR_STALL_CYCLES, 0x80000008, 4, 0x5c},
+	{SCORPION_VS_PIPE_WAW_STALL_CYCLES, 0x80000800, 4, 0x5d},
+	{SCORPION_VS_PIPE_RAW_STALL_CYCLES, 0x80080000, 4, 0x5e},
+	{SCORPION_EXCEPTIONS_INV_OPERATION, 0x8000000b, 4, 0x5c},
+	{SCORPION_EXCEPTIONS_DIV_BY_ZERO, 0x80000b00, 4, 0x5d},
+	{SCORPION_COND_INST_FAIL_VX_PIPE, 0x800b0000, 4, 0x5e},
+	{SCORPION_COND_INST_FAIL_VS_PIPE, 0x8b000000, 4, 0x5f},
+	{SCORPION_EXCEPTIONS_OVERFLOW, 0x8000000c, 4, 0x5c},
+	{SCORPION_EXCEPTIONS_UNDERFLOW, 0x80000c00, 4, 0x5d},
+	{SCORPION_EXCEPTIONS_DENORM, 0x8c000000, 4, 0x5f},
+
+#ifdef CONFIG_ARCH_MSM_SCORPIONMP
+	{SCORPIONMP_NUM_BARRIERS, 0x80000e00, 3, 0x59},
+	{SCORPIONMP_BARRIER_CYCLES, 0x800e0000, 3, 0x5a},
+#else
+	{SCORPION_BANK_AB_HIT, 0x80000001, 3, 0x58},
+	{SCORPION_BANK_AB_ACCESS, 0x80000100, 3, 0x59},
+	{SCORPION_BANK_CD_HIT, 0x80010000, 3, 0x5a},
+	{SCORPION_BANK_CD_ACCESS, 0x81000000, 3, 0x5b},
+	{SCORPION_BANK_AB_DSIDE_HIT, 0x80000002, 3, 0x58},
+	{SCORPION_BANK_AB_DSIDE_ACCESS, 0x80000200, 3, 0x59},
+	{SCORPION_BANK_CD_DSIDE_HIT, 0x80020000, 3, 0x5a},
+	{SCORPION_BANK_CD_DSIDE_ACCESS, 0x82000000, 3, 0x5b},
+	{SCORPION_BANK_AB_ISIDE_HIT, 0x80000003, 3, 0x58},
+	{SCORPION_BANK_AB_ISIDE_ACCESS, 0x80000300, 3, 0x59},
+	{SCORPION_BANK_CD_ISIDE_HIT, 0x80030000, 3, 0x5a},
+	{SCORPION_BANK_CD_ISIDE_ACCESS, 0x83000000, 3, 0x5b},
+	{SCORPION_ISIDE_RD_WAIT, 0x80000009, 3, 0x58},
+	{SCORPION_DSIDE_RD_WAIT, 0x80090000, 3, 0x5a},
+	{SCORPION_BANK_BYPASS_WRITE, 0x8000000a, 3, 0x58},
+	{SCORPION_BANK_AB_NON_CASTOUT, 0x8000000c, 3, 0x58},
+	{SCORPION_BANK_AB_L2_CASTOUT, 0x80000c00, 3, 0x59},
+	{SCORPION_BANK_CD_NON_CASTOUT, 0x800c0000, 3, 0x5a},
+	{SCORPION_BANK_CD_L2_CASTOUT, 0x8c000000, 3, 0x5b},
+#endif
+};
+
+static inline void scorpion_pmnc_write(u32 val)
+{
+	val &= PMNC_MASK;
+	asm volatile("mcr p15, 0, %0, c9, c12, 0" : : "r" (val));
+}
+
+static inline u32 scorpion_pmnc_read(void)
+{
+	u32 val;
+
+	asm volatile("mrc p15, 0, %0, c9, c12, 0" : "=r" (val));
+	return val;
+}
+
+static inline u32 scorpion_ccnt_read(void)
+{
+	u32 val;
+
+	asm volatile("mrc p15, 0, %0, c9, c13, 0" : "=r" (val));
+	return val;
+}
+
+static inline u32 scorpion_cntn_read(void)
+{
+	u32 val;
+
+	asm volatile("mrc p15, 0, %0, c9, c13, 2" : "=r" (val));
+	return val;
+}
+
+static inline u32 scorpion_pmnc_enable_counter(unsigned int cnt)
+{
+	u32 val;
+
+	if (cnt >= CNTMAX) {
+		pr_err("gator: CPU%u enabling wrong PMNC counter %d\n", smp_processor_id(), cnt);
+		return -1;
+	}
+
+	if (cnt == CCNT)
+		val = CCNT_REG;
+	else
+		val = (1 << (cnt - CNT0));
+
+	asm volatile("mcr p15, 0, %0, c9, c12, 1" : : "r" (val));
+
+	return cnt;
+}
+
+static inline u32 scorpion_pmnc_disable_counter(unsigned int cnt)
+{
+	u32 val;
+
+	if (cnt >= CNTMAX) {
+		pr_err("gator: CPU%u disabling wrong PMNC counter %d\n", smp_processor_id(), cnt);
+		return -1;
+	}
+
+	if (cnt == CCNT)
+		val = CCNT_REG;
+	else
+		val = (1 << (cnt - CNT0));
+
+	asm volatile("mcr p15, 0, %0, c9, c12, 2" : : "r" (val));
+
+	return cnt;
+}
+
+static inline int scorpion_pmnc_select_counter(unsigned int cnt)
+{
+	u32 val;
+
+	if ((cnt == CCNT) || (cnt >= CNTMAX)) {
+		pr_err("gator: CPU%u selecting wrong PMNC counter %d\n", smp_processor_id(), cnt);
+		return -1;
+	}
+
+	val = (cnt - CNT0);
+	asm volatile("mcr p15, 0, %0, c9, c12, 5" : : "r" (val));
+
+	return cnt;
+}
+
+static u32 scorpion_read_lpm0(void)
+{
+	u32 val;
+
+	asm volatile("mrc p15, 0, %0, c15, c0, 0" : "=r" (val));
+	return val;
+}
+
+static void scorpion_write_lpm0(u32 val)
+{
+	asm volatile("mcr p15, 0, %0, c15, c0, 0" : : "r" (val));
+}
+
+static u32 scorpion_read_lpm1(void)
+{
+	u32 val;
+
+	asm volatile("mrc p15, 1, %0, c15, c0, 0" : "=r" (val));
+	return val;
+}
+
+static void scorpion_write_lpm1(u32 val)
+{
+	asm volatile("mcr p15, 1, %0, c15, c0, 0" : : "r" (val));
+}
+
+static u32 scorpion_read_lpm2(void)
+{
+	u32 val;
+
+	asm volatile("mrc p15, 2, %0, c15, c0, 0" : "=r" (val));
+	return val;
+}
+
+static void scorpion_write_lpm2(u32 val)
+{
+	asm volatile("mcr p15, 2, %0, c15, c0, 0" : : "r" (val));
+}
+
+static u32 scorpion_read_l2lpm(void)
+{
+	u32 val;
+
+	asm volatile("mrc p15, 3, %0, c15, c2, 0" : "=r" (val));
+	return val;
+}
+
+static void scorpion_write_l2lpm(u32 val)
+{
+	asm volatile("mcr p15, 3, %0, c15, c2, 0" : : "r" (val));
+}
+
+static u32 scorpion_read_vlpm(void)
+{
+	u32 val;
+
+	asm volatile("mrc p10, 7, %0, c11, c0, 0" : "=r" (val));
+	return val;
+}
+
+static void scorpion_write_vlpm(u32 val)
+{
+	asm volatile("mcr p10, 7, %0, c11, c0, 0" : : "r" (val));
+}
+
+struct scorpion_access_funcs {
+	u32 (*read)(void);
+	void (*write)(u32);
+};
+
+struct scorpion_access_funcs scor_func[] = {
+	{scorpion_read_lpm0, scorpion_write_lpm0},
+	{scorpion_read_lpm1, scorpion_write_lpm1},
+	{scorpion_read_lpm2, scorpion_write_lpm2},
+	{scorpion_read_l2lpm, scorpion_write_l2lpm},
+	{scorpion_read_vlpm, scorpion_write_vlpm},
+};
+
+u32 venum_orig_val;
+u32 fp_orig_val;
+
+static void scorpion_pre_vlpm(void)
+{
+	u32 venum_new_val;
+	u32 fp_new_val;
+
+	/* CPACR Enable CP10 access */
+	asm volatile("mrc p15, 0, %0, c1, c0, 2" : "=r" (venum_orig_val));
+	venum_new_val = venum_orig_val | 0x00300000;
+	asm volatile("mcr p15, 0, %0, c1, c0, 2" : : "r" (venum_new_val));
+	/* Enable FPEXC */
+	asm volatile("mrc p10, 7, %0, c8, c0, 0" : "=r" (fp_orig_val));
+	fp_new_val = fp_orig_val | 0x40000000;
+	asm volatile("mcr p10, 7, %0, c8, c0, 0" : : "r" (fp_new_val));
+}
+
+static void scorpion_post_vlpm(void)
+{
+	/* Restore FPEXC */
+	asm volatile("mcr p10, 7, %0, c8, c0, 0" : : "r" (fp_orig_val));
+	/* Restore CPACR */
+	asm volatile("mcr p15, 0, %0, c1, c0, 2" : : "r" (venum_orig_val));
+}
+
+#define COLMN0MASK 0x000000ff
+#define COLMN1MASK 0x0000ff00
+#define COLMN2MASK 0x00ff0000
+static u32 scorpion_get_columnmask(u32 setval)
+{
+	if (setval & COLMN0MASK)
+		return 0xffffff00;
+	if (setval & COLMN1MASK)
+		return 0xffff00ff;
+	if (setval & COLMN2MASK)
+		return 0xff00ffff;
+	return 0x80ffffff;
+}
+
+static void scorpion_evt_setup(u32 gr, u32 setval)
+{
+	u32 val;
+
+	if (gr == 4)
+		scorpion_pre_vlpm();
+	val = scorpion_get_columnmask(setval) & scor_func[gr].read();
+	val = val | setval;
+	scor_func[gr].write(val);
+	if (gr == 4)
+		scorpion_post_vlpm();
+}
+
+static int get_scorpion_evtinfo(unsigned int evt_type, struct scorp_evt *evtinfo)
+{
+	u32 idx;
+
+	if ((evt_type < 0x4c) || (evt_type >= MSM_MAX_EVT))
+		return 0;
+	idx = evt_type - 0x4c;
+	if (sc_evt[idx].evt_type == evt_type) {
+		evtinfo->val = sc_evt[idx].val;
+		evtinfo->grp = sc_evt[idx].grp;
+		evtinfo->evt_type_act = sc_evt[idx].evt_type_act;
+		return 1;
+	}
+	return 0;
+}
+
+static inline void scorpion_pmnc_write_evtsel(unsigned int cnt, u32 val)
+{
+	if (scorpion_pmnc_select_counter(cnt) == cnt) {
+		if (val < 0x40) {
+			asm volatile("mcr p15, 0, %0, c9, c13, 1" : : "r" (val));
+		} else {
+			u32 zero = 0;
+			struct scorp_evt evtinfo;
+			/* extract evtinfo.grp and evtinfo.tevt_type_act from val */
+			if (get_scorpion_evtinfo(val, &evtinfo) == 0)
+				return;
+			asm volatile("mcr p15, 0, %0, c9, c13, 1" : : "r" (evtinfo.evt_type_act));
+			asm volatile("mcr p15, 0, %0, c9, c15, 0" : : "r" (zero));
+			scorpion_evt_setup(evtinfo.grp, val);
+		}
+	}
+}
+
+static void scorpion_pmnc_reset_counter(unsigned int cnt)
+{
+	u32 val = 0;
+
+	if (cnt == CCNT) {
+		scorpion_pmnc_disable_counter(cnt);
+
+		asm volatile("mcr p15, 0, %0, c9, c13, 0" : : "r" (val));
+
+		if (pmnc_enabled[cnt] != 0)
+			scorpion_pmnc_enable_counter(cnt);
+
+	} else if (cnt >= CNTMAX) {
+		pr_err("gator: CPU%u resetting wrong PMNC counter %d\n", smp_processor_id(), cnt);
+	} else {
+		scorpion_pmnc_disable_counter(cnt);
+
+		if (scorpion_pmnc_select_counter(cnt) == cnt)
+			asm volatile("mcr p15, 0, %0, c9, c13, 2" : : "r" (val));
+
+		if (pmnc_enabled[cnt] != 0)
+			scorpion_pmnc_enable_counter(cnt);
+	}
+}
+
+static int gator_events_scorpion_create_files(struct super_block *sb, struct dentry *root)
+{
+	struct dentry *dir;
+	int i;
+
+	for (i = 0; i < pmnc_counters; i++) {
+		char buf[40];
+
+		if (i == 0)
+			snprintf(buf, sizeof(buf), "%s_ccnt", pmnc_name);
+		else
+			snprintf(buf, sizeof(buf), "%s_cnt%d", pmnc_name, i - 1);
+		dir = gatorfs_mkdir(sb, root, buf);
+		if (!dir)
+			return -1;
+		gatorfs_create_ulong(sb, dir, "enabled", &pmnc_enabled[i]);
+		gatorfs_create_ro_ulong(sb, dir, "key", &pmnc_key[i]);
+		if (i > 0)
+			gatorfs_create_ulong(sb, dir, "event", &pmnc_event[i]);
+	}
+
+	return 0;
+}
+
+static int gator_events_scorpion_online(int **buffer, bool migrate)
+{
+	unsigned int cnt, len = 0, cpu = smp_processor_id();
+
+	if (scorpion_pmnc_read() & PMNC_E)
+		scorpion_pmnc_write(scorpion_pmnc_read() & ~PMNC_E);
+
+	/* Initialize & Reset PMNC: C bit and P bit */
+	scorpion_pmnc_write(PMNC_P | PMNC_C);
+
+	for (cnt = CCNT; cnt < CNTMAX; cnt++) {
+		unsigned long event;
+
+		if (!pmnc_enabled[cnt])
+			continue;
+
+		/* disable counter */
+		scorpion_pmnc_disable_counter(cnt);
+
+		event = pmnc_event[cnt] & 255;
+
+		/* Set event (if destined for PMNx counters), We don't need to set the event if it's a cycle count */
+		if (cnt != CCNT)
+			scorpion_pmnc_write_evtsel(cnt, event);
+
+		/* reset counter */
+		scorpion_pmnc_reset_counter(cnt);
+
+		/* Enable counter, do not enable interrupt for this counter */
+		scorpion_pmnc_enable_counter(cnt);
+	}
+
+	/* enable */
+	scorpion_pmnc_write(scorpion_pmnc_read() | PMNC_E);
+
+	/* read the counters and toss the invalid data, return zero instead */
+	for (cnt = 0; cnt < pmnc_counters; cnt++) {
+		if (pmnc_enabled[cnt]) {
+			if (cnt == CCNT)
+				scorpion_ccnt_read();
+			else if (scorpion_pmnc_select_counter(cnt) == cnt)
+				scorpion_cntn_read();
+			scorpion_pmnc_reset_counter(cnt);
+
+			per_cpu(perfCnt, cpu)[len++] = pmnc_key[cnt];
+			per_cpu(perfCnt, cpu)[len++] = 0;
+		}
+	}
+
+	if (buffer)
+		*buffer = per_cpu(perfCnt, cpu);
+
+	return len;
+}
+
+static int gator_events_scorpion_offline(int **buffer, bool migrate)
+{
+	scorpion_pmnc_write(scorpion_pmnc_read() & ~PMNC_E);
+	return 0;
+}
+
+static void gator_events_scorpion_stop(void)
+{
+	unsigned int cnt;
+
+	for (cnt = CCNT; cnt < CNTMAX; cnt++) {
+		pmnc_enabled[cnt] = 0;
+		pmnc_event[cnt] = 0;
+	}
+}
+
+static int gator_events_scorpion_read(int **buffer, bool sched_switch)
+{
+	int cnt, len = 0;
+	int cpu = smp_processor_id();
+
+	/* a context switch may occur before the online hotplug event, thus need to check that the pmu is enabled */
+	if (!(scorpion_pmnc_read() & PMNC_E))
+		return 0;
+
+	for (cnt = 0; cnt < pmnc_counters; cnt++) {
+		if (pmnc_enabled[cnt]) {
+			int value;
+
+			if (cnt == CCNT)
+				value = scorpion_ccnt_read();
+			else if (scorpion_pmnc_select_counter(cnt) == cnt)
+				value = scorpion_cntn_read();
+			else
+				value = 0;
+			scorpion_pmnc_reset_counter(cnt);
+
+			per_cpu(perfCnt, cpu)[len++] = pmnc_key[cnt];
+			per_cpu(perfCnt, cpu)[len++] = value;
+		}
+	}
+
+	if (buffer)
+		*buffer = per_cpu(perfCnt, cpu);
+
+	return len;
+}
+
+static struct gator_interface gator_events_scorpion_interface = {
+	.create_files = gator_events_scorpion_create_files,
+	.stop = gator_events_scorpion_stop,
+	.online = gator_events_scorpion_online,
+	.offline = gator_events_scorpion_offline,
+	.read = gator_events_scorpion_read,
+};
+
+int gator_events_scorpion_init(void)
+{
+	unsigned int cnt;
+
+	switch (gator_cpuid()) {
+	case SCORPION:
+		pmnc_name = "Scorpion";
+		pmnc_counters = 4;
+		break;
+	case SCORPIONMP:
+		pmnc_name = "ScorpionMP";
+		pmnc_counters = 4;
+		break;
+	default:
+		return -1;
+	}
+
+	/* CNT[n] + CCNT */
+	pmnc_counters++;
+
+	for (cnt = CCNT; cnt < CNTMAX; cnt++) {
+		pmnc_enabled[cnt] = 0;
+		pmnc_event[cnt] = 0;
+		pmnc_key[cnt] = gator_events_get_key();
+	}
+
+	return gator_events_install(&gator_events_scorpion_interface);
+}
+
+#endif
diff --git a/drivers/gator/gator_fs.c b/drivers/gator/gator_fs.c
new file mode 100644
index 000000000000..d8fb357b9eda
--- /dev/null
+++ b/drivers/gator/gator_fs.c
@@ -0,0 +1,370 @@
+/**
+ * @file gatorfs.c
+ *
+ * @remark Copyright 2002 OProfile authors
+ * @remark Read the file COPYING
+ *
+ * @author John Levon
+ *
+ * A simple filesystem for configuration and
+ * access of oprofile.
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/pagemap.h>
+#include <linux/uaccess.h>
+
+#define gatorfs_MAGIC 0x24051020
+#define TMPBUFSIZE 50
+DEFINE_SPINLOCK(gatorfs_lock);
+
+static struct inode *gatorfs_get_inode(struct super_block *sb, int mode)
+{
+	struct inode *inode = new_inode(sb);
+
+	if (inode) {
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 37)
+		inode->i_ino = get_next_ino();
+#endif
+		inode->i_mode = mode;
+		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+	}
+	return inode;
+}
+
+static const struct super_operations s_ops = {
+	.statfs = simple_statfs,
+	.drop_inode = generic_delete_inode,
+};
+
+static ssize_t gatorfs_ulong_to_user(unsigned long val, char __user *buf, size_t count, loff_t *offset)
+{
+	char tmpbuf[TMPBUFSIZE];
+	size_t maxlen = snprintf(tmpbuf, TMPBUFSIZE, "%lu\n", val);
+
+	if (maxlen > TMPBUFSIZE)
+		maxlen = TMPBUFSIZE;
+	return simple_read_from_buffer(buf, count, offset, tmpbuf, maxlen);
+}
+
+static ssize_t gatorfs_u64_to_user(u64 val, char __user *buf, size_t count, loff_t *offset)
+{
+	char tmpbuf[TMPBUFSIZE];
+	size_t maxlen = snprintf(tmpbuf, TMPBUFSIZE, "%llu\n", val);
+
+	if (maxlen > TMPBUFSIZE)
+		maxlen = TMPBUFSIZE;
+	return simple_read_from_buffer(buf, count, offset, tmpbuf, maxlen);
+}
+
+static int gatorfs_ulong_from_user(unsigned long *val, char const __user *buf, size_t count)
+{
+	char tmpbuf[TMPBUFSIZE];
+	unsigned long flags;
+
+	if (!count)
+		return 0;
+
+	if (count > TMPBUFSIZE - 1)
+		return -EINVAL;
+
+	memset(tmpbuf, 0x0, TMPBUFSIZE);
+
+	if (copy_from_user(tmpbuf, buf, count))
+		return -EFAULT;
+
+	spin_lock_irqsave(&gatorfs_lock, flags);
+	*val = simple_strtoul(tmpbuf, NULL, 0);
+	spin_unlock_irqrestore(&gatorfs_lock, flags);
+	return 0;
+}
+
+static int gatorfs_u64_from_user(u64 *val, char const __user *buf, size_t count)
+{
+	char tmpbuf[TMPBUFSIZE];
+	unsigned long flags;
+
+	if (!count)
+		return 0;
+
+	if (count > TMPBUFSIZE - 1)
+		return -EINVAL;
+
+	memset(tmpbuf, 0x0, TMPBUFSIZE);
+
+	if (copy_from_user(tmpbuf, buf, count))
+		return -EFAULT;
+
+	spin_lock_irqsave(&gatorfs_lock, flags);
+	*val = simple_strtoull(tmpbuf, NULL, 0);
+	spin_unlock_irqrestore(&gatorfs_lock, flags);
+	return 0;
+}
+
+static ssize_t ulong_read_file(struct file *file, char __user *buf, size_t count, loff_t *offset)
+{
+	unsigned long *val = file->private_data;
+
+	return gatorfs_ulong_to_user(*val, buf, count, offset);
+}
+
+static ssize_t u64_read_file(struct file *file, char __user *buf, size_t count, loff_t *offset)
+{
+	u64 *val = file->private_data;
+
+	return gatorfs_u64_to_user(*val, buf, count, offset);
+}
+
+static ssize_t ulong_write_file(struct file *file, char const __user *buf, size_t count, loff_t *offset)
+{
+	unsigned long *value = file->private_data;
+	int retval;
+
+	if (*offset)
+		return -EINVAL;
+
+	retval = gatorfs_ulong_from_user(value, buf, count);
+
+	if (retval)
+		return retval;
+	return count;
+}
+
+static ssize_t u64_write_file(struct file *file, char const __user *buf, size_t count, loff_t *offset)
+{
+	u64 *value = file->private_data;
+	int retval;
+
+	if (*offset)
+		return -EINVAL;
+
+	retval = gatorfs_u64_from_user(value, buf, count);
+
+	if (retval)
+		return retval;
+	return count;
+}
+
+static int default_open(struct inode *inode, struct file *filp)
+{
+	if (inode->i_private)
+		filp->private_data = inode->i_private;
+	return 0;
+}
+
+static const struct file_operations ulong_fops = {
+	.read = ulong_read_file,
+	.write = ulong_write_file,
+	.open = default_open,
+};
+
+static const struct file_operations u64_fops = {
+	.read = u64_read_file,
+	.write = u64_write_file,
+	.open = default_open,
+};
+
+static const struct file_operations ulong_ro_fops = {
+	.read = ulong_read_file,
+	.open = default_open,
+};
+
+static const struct file_operations u64_ro_fops = {
+	.read = u64_read_file,
+	.open = default_open,
+};
+
+static struct dentry *__gatorfs_create_file(struct super_block *sb,
+					    struct dentry *root,
+					    char const *name,
+					    const struct file_operations *fops,
+					    int perm)
+{
+	struct dentry *dentry;
+	struct inode *inode;
+
+	dentry = d_alloc_name(root, name);
+	if (!dentry)
+		return NULL;
+	inode = gatorfs_get_inode(sb, S_IFREG | perm);
+	if (!inode) {
+		dput(dentry);
+		return NULL;
+	}
+	inode->i_fop = fops;
+	d_add(dentry, inode);
+	return dentry;
+}
+
+int gatorfs_create_ulong(struct super_block *sb, struct dentry *root,
+			 char const *name, unsigned long *val)
+{
+	struct dentry *d = __gatorfs_create_file(sb, root, name,
+						 &ulong_fops, 0644);
+	if (!d)
+		return -EFAULT;
+
+	d->d_inode->i_private = val;
+	return 0;
+}
+
+static int gatorfs_create_u64(struct super_block *sb, struct dentry *root,
+			      char const *name, u64 *val)
+{
+	struct dentry *d = __gatorfs_create_file(sb, root, name,
+						 &u64_fops, 0644);
+	if (!d)
+		return -EFAULT;
+
+	d->d_inode->i_private = val;
+	return 0;
+}
+
+int gatorfs_create_ro_ulong(struct super_block *sb, struct dentry *root,
+			    char const *name, unsigned long *val)
+{
+	struct dentry *d = __gatorfs_create_file(sb, root, name,
+						 &ulong_ro_fops, 0444);
+	if (!d)
+		return -EFAULT;
+
+	d->d_inode->i_private = val;
+	return 0;
+}
+
+static int gatorfs_create_ro_u64(struct super_block *sb, struct dentry *root,
+				 char const *name, u64 *val)
+{
+	struct dentry *d =
+	    __gatorfs_create_file(sb, root, name, &u64_ro_fops, 0444);
+	if (!d)
+		return -EFAULT;
+
+	d->d_inode->i_private = val;
+	return 0;
+}
+
+static ssize_t atomic_read_file(struct file *file, char __user *buf, size_t count, loff_t *offset)
+{
+	atomic_t *val = file->private_data;
+
+	return gatorfs_ulong_to_user(atomic_read(val), buf, count, offset);
+}
+
+static const struct file_operations atomic_ro_fops = {
+	.read = atomic_read_file,
+	.open = default_open,
+};
+
+static int gatorfs_create_file(struct super_block *sb, struct dentry *root,
+			       char const *name, const struct file_operations *fops)
+{
+	if (!__gatorfs_create_file(sb, root, name, fops, 0644))
+		return -EFAULT;
+	return 0;
+}
+
+static int gatorfs_create_file_perm(struct super_block *sb, struct dentry *root,
+				    char const *name,
+				    const struct file_operations *fops, int perm)
+{
+	if (!__gatorfs_create_file(sb, root, name, fops, perm))
+		return -EFAULT;
+	return 0;
+}
+
+struct dentry *gatorfs_mkdir(struct super_block *sb,
+			     struct dentry *root, char const *name)
+{
+	struct dentry *dentry;
+	struct inode *inode;
+
+	dentry = d_alloc_name(root, name);
+	if (!dentry)
+		return NULL;
+	inode = gatorfs_get_inode(sb, S_IFDIR | 0755);
+	if (!inode) {
+		dput(dentry);
+		return NULL;
+	}
+	inode->i_op = &simple_dir_inode_operations;
+	inode->i_fop = &simple_dir_operations;
+	d_add(dentry, inode);
+	return dentry;
+}
+
+static int gatorfs_fill_super(struct super_block *sb, void *data, int silent)
+{
+	struct inode *root_inode;
+	struct dentry *root_dentry;
+
+	sb->s_blocksize = PAGE_CACHE_SIZE;
+	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
+	sb->s_magic = gatorfs_MAGIC;
+	sb->s_op = &s_ops;
+	sb->s_time_gran = 1;
+
+	root_inode = gatorfs_get_inode(sb, S_IFDIR | 0755);
+	if (!root_inode)
+		return -ENOMEM;
+	root_inode->i_op = &simple_dir_inode_operations;
+	root_inode->i_fop = &simple_dir_operations;
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 4, 0)
+	root_dentry = d_alloc_root(root_inode);
+#else
+	root_dentry = d_make_root(root_inode);
+#endif
+
+	if (!root_dentry) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 4, 0)
+		iput(root_inode);
+#endif
+		return -ENOMEM;
+	}
+
+	sb->s_root = root_dentry;
+
+	gator_op_create_files(sb, root_dentry);
+
+	return 0;
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 39)
+static int gatorfs_get_sb(struct file_system_type *fs_type,
+			  int flags, const char *dev_name, void *data,
+			  struct vfsmount *mnt)
+{
+	return get_sb_single(fs_type, flags, data, gatorfs_fill_super, mnt);
+}
+#else
+static struct dentry *gatorfs_mount(struct file_system_type *fs_type,
+				    int flags, const char *dev_name, void *data)
+{
+	return mount_nodev(fs_type, flags, data, gatorfs_fill_super);
+}
+#endif
+
+static struct file_system_type gatorfs_type = {
+	.owner = THIS_MODULE,
+	.name = "gatorfs",
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 39)
+	.get_sb = gatorfs_get_sb,
+#else
+	.mount = gatorfs_mount,
+#endif
+
+	.kill_sb = kill_litter_super,
+};
+
+static int __init gatorfs_register(void)
+{
+	return register_filesystem(&gatorfs_type);
+}
+
+static void gatorfs_unregister(void)
+{
+	unregister_filesystem(&gatorfs_type);
+}
diff --git a/drivers/gator/gator_hrtimer_gator.c b/drivers/gator/gator_hrtimer_gator.c
new file mode 100644
index 000000000000..c1525e10a8da
--- /dev/null
+++ b/drivers/gator/gator_hrtimer_gator.c
@@ -0,0 +1,80 @@
+/**
+ * Copyright (C) ARM Limited 2011-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+void (*callback)(void);
+DEFINE_PER_CPU(struct hrtimer, percpu_hrtimer);
+DEFINE_PER_CPU(ktime_t, hrtimer_expire);
+DEFINE_PER_CPU(int, hrtimer_is_active);
+static ktime_t profiling_interval;
+static void gator_hrtimer_online(void);
+static void gator_hrtimer_offline(void);
+
+static enum hrtimer_restart gator_hrtimer_notify(struct hrtimer *hrtimer)
+{
+	int cpu = get_logical_cpu();
+
+	hrtimer_forward(hrtimer, per_cpu(hrtimer_expire, cpu), profiling_interval);
+	per_cpu(hrtimer_expire, cpu) = ktime_add(per_cpu(hrtimer_expire, cpu), profiling_interval);
+	(*callback)();
+	return HRTIMER_RESTART;
+}
+
+static void gator_hrtimer_online(void)
+{
+	int cpu = get_logical_cpu();
+	struct hrtimer *hrtimer = &per_cpu(percpu_hrtimer, cpu);
+
+	if (per_cpu(hrtimer_is_active, cpu) || profiling_interval.tv64 == 0)
+		return;
+
+	per_cpu(hrtimer_is_active, cpu) = 1;
+	hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
+	hrtimer->function = gator_hrtimer_notify;
+#ifdef CONFIG_PREEMPT_RT_BASE
+	hrtimer->irqsafe = 1;
+#endif
+	per_cpu(hrtimer_expire, cpu) = ktime_add(hrtimer->base->get_time(), profiling_interval);
+	hrtimer_start(hrtimer, per_cpu(hrtimer_expire, cpu), HRTIMER_MODE_ABS_PINNED);
+}
+
+static void gator_hrtimer_offline(void)
+{
+	int cpu = get_logical_cpu();
+	struct hrtimer *hrtimer = &per_cpu(percpu_hrtimer, cpu);
+
+	if (!per_cpu(hrtimer_is_active, cpu))
+		return;
+
+	per_cpu(hrtimer_is_active, cpu) = 0;
+	hrtimer_cancel(hrtimer);
+}
+
+static int gator_hrtimer_init(int interval, void (*func)(void))
+{
+	int cpu;
+
+	(callback) = (func);
+
+	for_each_present_cpu(cpu) {
+		per_cpu(hrtimer_is_active, cpu) = 0;
+	}
+
+	/* calculate profiling interval */
+	if (interval > 0)
+		profiling_interval = ns_to_ktime(1000000000UL / interval);
+	else
+		profiling_interval.tv64 = 0;
+
+	return 0;
+}
+
+static void gator_hrtimer_shutdown(void)
+{
+	/* empty */
+}
diff --git a/drivers/gator/gator_iks.c b/drivers/gator/gator_iks.c
new file mode 100644
index 000000000000..fb78c10fd987
--- /dev/null
+++ b/drivers/gator/gator_iks.c
@@ -0,0 +1,197 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#if GATOR_IKS_SUPPORT
+
+#include <linux/of.h>
+#include <asm/bL_switcher.h>
+#include <asm/smp_plat.h>
+#include <trace/events/power_cpu_migrate.h>
+
+static bool map_cpuids;
+static int mpidr_cpuids[NR_CPUS];
+static const struct gator_cpu *mpidr_cpus[NR_CPUS];
+static int __lcpu_to_pcpu[NR_CPUS];
+
+static const struct gator_cpu *gator_find_cpu_by_dt_name(const char *const name)
+{
+	int i;
+
+	for (i = 0; gator_cpus[i].cpuid != 0; ++i) {
+		const struct gator_cpu *const gator_cpu = &gator_cpus[i];
+
+		if (gator_cpu->dt_name != NULL && strcmp(gator_cpu->dt_name, name) == 0)
+			return gator_cpu;
+	}
+
+	return NULL;
+}
+
+static void calc_first_cluster_size(void)
+{
+	int len;
+	const u32 *val;
+	const char *compatible;
+	struct device_node *cn = NULL;
+	int mpidr_cpuids_count = 0;
+
+	/* Zero is a valid cpuid, so initialize the array to 0xff's */
+	memset(&mpidr_cpuids, 0xff, sizeof(mpidr_cpuids));
+	memset(&mpidr_cpus, 0, sizeof(mpidr_cpus));
+
+	while ((cn = of_find_node_by_type(cn, "cpu"))) {
+		BUG_ON(mpidr_cpuids_count >= NR_CPUS);
+
+		val = of_get_property(cn, "reg", &len);
+		if (!val || len != 4) {
+			pr_err("%s missing reg property\n", cn->full_name);
+			continue;
+		}
+		compatible = of_get_property(cn, "compatible", NULL);
+		if (compatible == NULL) {
+			pr_err("%s missing compatible property\n", cn->full_name);
+			continue;
+		}
+
+		mpidr_cpuids[mpidr_cpuids_count] = be32_to_cpup(val);
+		mpidr_cpus[mpidr_cpuids_count] = gator_find_cpu_by_dt_name(compatible);
+		++mpidr_cpuids_count;
+	}
+
+	map_cpuids = (mpidr_cpuids_count == nr_cpu_ids);
+}
+
+static int linearize_mpidr(int mpidr)
+{
+	int i;
+
+	for (i = 0; i < nr_cpu_ids; ++i) {
+		if (mpidr_cpuids[i] == mpidr)
+			return i;
+	}
+
+	BUG();
+}
+
+int lcpu_to_pcpu(const int lcpu)
+{
+	int pcpu;
+
+	if (!map_cpuids)
+		return lcpu;
+
+	BUG_ON(lcpu >= nr_cpu_ids || lcpu < 0);
+	pcpu = __lcpu_to_pcpu[lcpu];
+	BUG_ON(pcpu >= nr_cpu_ids || pcpu < 0);
+	return pcpu;
+}
+
+int pcpu_to_lcpu(const int pcpu)
+{
+	int lcpu;
+
+	if (!map_cpuids)
+		return pcpu;
+
+	BUG_ON(pcpu >= nr_cpu_ids || pcpu < 0);
+	for (lcpu = 0; lcpu < nr_cpu_ids; ++lcpu) {
+		if (__lcpu_to_pcpu[lcpu] == pcpu) {
+			BUG_ON(lcpu >= nr_cpu_ids || lcpu < 0);
+			return lcpu;
+		}
+	}
+	BUG();
+}
+
+static void gator_update_cpu_mapping(u32 cpu_hwid)
+{
+	int lcpu = smp_processor_id();
+	int pcpu = linearize_mpidr(cpu_hwid & MPIDR_HWID_BITMASK);
+
+	BUG_ON(lcpu >= nr_cpu_ids || lcpu < 0);
+	BUG_ON(pcpu >= nr_cpu_ids || pcpu < 0);
+	__lcpu_to_pcpu[lcpu] = pcpu;
+}
+
+GATOR_DEFINE_PROBE(cpu_migrate_begin, TP_PROTO(u64 timestamp, u32 cpu_hwid))
+{
+	const int cpu = get_physical_cpu();
+
+	gator_timer_offline((void *)1);
+	gator_timer_offline_dispatch(cpu, true);
+}
+
+GATOR_DEFINE_PROBE(cpu_migrate_finish, TP_PROTO(u64 timestamp, u32 cpu_hwid))
+{
+	int cpu;
+
+	gator_update_cpu_mapping(cpu_hwid);
+
+	/* get_physical_cpu must be called after gator_update_cpu_mapping */
+	cpu = get_physical_cpu();
+	gator_timer_online_dispatch(cpu, true);
+	gator_timer_online((void *)1);
+}
+
+GATOR_DEFINE_PROBE(cpu_migrate_current, TP_PROTO(u64 timestamp, u32 cpu_hwid))
+{
+	gator_update_cpu_mapping(cpu_hwid);
+}
+
+static void gator_send_iks_core_names(void)
+{
+	int cpu;
+	/* Send the cpu names */
+	preempt_disable();
+	for (cpu = 0; cpu < nr_cpu_ids; ++cpu) {
+		if (mpidr_cpus[cpu] != NULL)
+			gator_send_core_name(cpu, mpidr_cpus[cpu]->cpuid);
+	}
+	preempt_enable();
+}
+
+static int gator_migrate_start(void)
+{
+	int retval = 0;
+
+	if (!map_cpuids)
+		return retval;
+
+	if (retval == 0)
+		retval = GATOR_REGISTER_TRACE(cpu_migrate_begin);
+	if (retval == 0)
+		retval = GATOR_REGISTER_TRACE(cpu_migrate_finish);
+	if (retval == 0)
+		retval = GATOR_REGISTER_TRACE(cpu_migrate_current);
+	if (retval == 0) {
+		/* Initialize the logical to physical cpu mapping */
+		memset(&__lcpu_to_pcpu, 0xff, sizeof(__lcpu_to_pcpu));
+		bL_switcher_trace_trigger();
+	}
+	return retval;
+}
+
+static void gator_migrate_stop(void)
+{
+	if (!map_cpuids)
+		return;
+
+	GATOR_UNREGISTER_TRACE(cpu_migrate_current);
+	GATOR_UNREGISTER_TRACE(cpu_migrate_finish);
+	GATOR_UNREGISTER_TRACE(cpu_migrate_begin);
+}
+
+#else
+
+#define calc_first_cluster_size()
+#define gator_send_iks_core_names()
+#define gator_migrate_start() 0
+#define gator_migrate_stop()
+
+#endif
diff --git a/drivers/gator/gator_main.c b/drivers/gator/gator_main.c
new file mode 100644
index 000000000000..30bf60d95286
--- /dev/null
+++ b/drivers/gator/gator_main.c
@@ -0,0 +1,1491 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+/* This version must match the gator daemon version */
+#define PROTOCOL_VERSION 20
+static unsigned long gator_protocol_version = PROTOCOL_VERSION;
+
+#include <linux/slab.h>
+#include <linux/cpu.h>
+#include <linux/sched.h>
+#include <linux/irq.h>
+#include <linux/vmalloc.h>
+#include <linux/hardirq.h>
+#include <linux/highmem.h>
+#include <linux/pagemap.h>
+#include <linux/suspend.h>
+#include <linux/module.h>
+#include <linux/perf_event.h>
+#include <linux/utsname.h>
+#include <linux/kthread.h>
+#include <asm/stacktrace.h>
+#include <linux/uaccess.h>
+
+#include "gator.h"
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 32)
+#error kernels prior to 2.6.32 are not supported
+#endif
+
+#if defined(MODULE) && !defined(CONFIG_MODULES)
+#error Cannot build a module against a kernel that does not support modules. To resolve, either rebuild the kernel to support modules or build gator as part of the kernel.
+#endif
+
+#if !defined(CONFIG_GENERIC_TRACER) && !defined(CONFIG_TRACING)
+#error gator requires the kernel to have CONFIG_GENERIC_TRACER or CONFIG_TRACING defined
+#endif
+
+#ifndef CONFIG_PROFILING
+#error gator requires the kernel to have CONFIG_PROFILING defined
+#endif
+
+#ifndef CONFIG_HIGH_RES_TIMERS
+#error gator requires the kernel to have CONFIG_HIGH_RES_TIMERS defined to support PC sampling
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 12, 0) && defined(__arm__) && defined(CONFIG_SMP) && !defined(CONFIG_LOCAL_TIMERS)
+#error gator requires the kernel to have CONFIG_LOCAL_TIMERS defined on SMP systems
+#endif
+
+#if (GATOR_PERF_SUPPORT) && (!(GATOR_PERF_PMU_SUPPORT))
+#ifndef CONFIG_PERF_EVENTS
+#error gator requires the kernel to have CONFIG_PERF_EVENTS defined to support pmu hardware counters
+#elif !defined CONFIG_HW_PERF_EVENTS
+#error gator requires the kernel to have CONFIG_HW_PERF_EVENTS defined to support pmu hardware counters
+#endif
+#endif
+
+/******************************************************************************
+ * DEFINES
+ ******************************************************************************/
+#define SUMMARY_BUFFER_SIZE       (1*1024)
+#define BACKTRACE_BUFFER_SIZE     (128*1024)
+#define NAME_BUFFER_SIZE          (64*1024)
+#define COUNTER_BUFFER_SIZE       (64*1024)	/* counters have the core as part of the data and the core value in the frame header may be discarded */
+#define BLOCK_COUNTER_BUFFER_SIZE (128*1024)
+#define ANNOTATE_BUFFER_SIZE      (128*1024)	/* annotate counters have the core as part of the data and the core value in the frame header may be discarded */
+#define SCHED_TRACE_BUFFER_SIZE   (128*1024)
+#define IDLE_BUFFER_SIZE          (32*1024)	/* idle counters have the core as part of the data and the core value in the frame header may be discarded */
+#define ACTIVITY_BUFFER_SIZE      (128*1024)
+
+#define NO_COOKIE      0U
+#define UNRESOLVED_COOKIE ~0U
+
+#define FRAME_SUMMARY       1
+#define FRAME_BACKTRACE     2
+#define FRAME_NAME          3
+#define FRAME_COUNTER       4
+#define FRAME_BLOCK_COUNTER 5
+#define FRAME_ANNOTATE      6
+#define FRAME_SCHED_TRACE   7
+#define FRAME_IDLE          9
+#define FRAME_ACTIVITY     13
+
+#define MESSAGE_END_BACKTRACE 1
+
+/* Name Frame Messages */
+#define MESSAGE_COOKIE      1
+#define MESSAGE_THREAD_NAME 2
+#define MESSAGE_LINK        4
+
+/* Scheduler Trace Frame Messages */
+#define MESSAGE_SCHED_SWITCH 1
+#define MESSAGE_SCHED_EXIT   2
+
+/* Idle Frame Messages */
+#define MESSAGE_IDLE_ENTER 1
+#define MESSAGE_IDLE_EXIT  2
+
+/* Summary Frame Messages */
+#define MESSAGE_SUMMARY   1
+#define MESSAGE_CORE_NAME 3
+
+/* Activity Frame Messages */
+#define MESSAGE_SWITCH 2
+#define MESSAGE_EXIT   3
+
+#define MAXSIZE_PACK32     5
+#define MAXSIZE_PACK64    10
+
+#define FRAME_HEADER_SIZE 3
+
+#if defined(__arm__)
+#define PC_REG regs->ARM_pc
+#elif defined(__aarch64__)
+#define PC_REG regs->pc
+#else
+#define PC_REG regs->ip
+#endif
+
+enum {
+	SUMMARY_BUF,
+	BACKTRACE_BUF,
+	NAME_BUF,
+	COUNTER_BUF,
+	BLOCK_COUNTER_BUF,
+	ANNOTATE_BUF,
+	SCHED_TRACE_BUF,
+	IDLE_BUF,
+	ACTIVITY_BUF,
+	NUM_GATOR_BUFS
+};
+
+/******************************************************************************
+ * Globals
+ ******************************************************************************/
+static unsigned long gator_cpu_cores;
+/* Size of the largest buffer. Effectively constant, set in gator_op_create_files */
+static unsigned long userspace_buffer_size;
+static unsigned long gator_backtrace_depth;
+/* How often to commit the buffers for live in nanoseconds */
+static u64 gator_live_rate;
+
+static unsigned long gator_started;
+static u64 gator_monotonic_started;
+static u64 gator_sync_time;
+static u64 gator_hibernate_time;
+static unsigned long gator_buffer_opened;
+static unsigned long gator_timer_count;
+static unsigned long gator_response_type;
+static DEFINE_MUTEX(start_mutex);
+static DEFINE_MUTEX(gator_buffer_mutex);
+
+bool event_based_sampling;
+
+static DECLARE_WAIT_QUEUE_HEAD(gator_buffer_wait);
+static DECLARE_WAIT_QUEUE_HEAD(gator_annotate_wait);
+static struct timer_list gator_buffer_wake_up_timer;
+static bool gator_buffer_wake_run;
+/* Initialize semaphore unlocked to initialize memory values */
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
+static DECLARE_MUTEX(gator_buffer_wake_sem);
+#else
+static DEFINE_SEMAPHORE(gator_buffer_wake_sem);
+#endif
+static struct task_struct *gator_buffer_wake_thread;
+static LIST_HEAD(gator_events);
+
+static DEFINE_PER_CPU(u64, last_timestamp);
+
+static bool printed_monotonic_warning;
+
+static u32 gator_cpuids[NR_CPUS];
+static bool sent_core_name[NR_CPUS];
+
+static DEFINE_PER_CPU(bool, in_scheduler_context);
+
+/******************************************************************************
+ * Prototypes
+ ******************************************************************************/
+static u64 gator_get_time(void);
+static void gator_emit_perf_time(u64 time);
+static void gator_op_create_files(struct super_block *sb, struct dentry *root);
+
+/* gator_buffer is protected by being per_cpu and by having IRQs
+ * disabled when writing to it. Most marshal_* calls take care of this
+ * except for marshal_cookie*, marshal_backtrace* and marshal_frame
+ * where the caller is responsible for doing so. No synchronization is
+ * needed with the backtrace buffer as it is per cpu and is only used
+ * from the hrtimer. The annotate_lock must be held when using the
+ * annotation buffer as it is not per cpu. collect_counters which is
+ * the sole writer to the block counter frame is additionally
+ * protected by the per cpu collecting flag.
+ */
+
+/* Size of the buffer, must be a power of 2. Effectively constant, set in gator_op_setup. */
+static uint32_t gator_buffer_size[NUM_GATOR_BUFS];
+/* gator_buffer_size - 1, bitwise and with pos to get offset into the array. Effectively constant, set in gator_op_setup. */
+static uint32_t gator_buffer_mask[NUM_GATOR_BUFS];
+/* Read position in the buffer. Initialized to zero in gator_op_setup and incremented after bytes are read by userspace in userspace_buffer_read */
+static DEFINE_PER_CPU(int[NUM_GATOR_BUFS], gator_buffer_read);
+/* Write position in the buffer. Initialized to zero in gator_op_setup and incremented after bytes are written to the buffer */
+static DEFINE_PER_CPU(int[NUM_GATOR_BUFS], gator_buffer_write);
+/* Commit position in the buffer. Initialized to zero in gator_op_setup and incremented after a frame is ready to be read by userspace */
+static DEFINE_PER_CPU(int[NUM_GATOR_BUFS], gator_buffer_commit);
+/* If set to false, decreases the number of bytes returned by
+ * buffer_bytes_available. Set in buffer_check_space if no space is
+ * remaining. Initialized to true in gator_op_setup. This means that
+ * if we run out of space, continue to report that no space is
+ * available until bytes are read by userspace
+ */
+static DEFINE_PER_CPU(int[NUM_GATOR_BUFS], buffer_space_available);
+/* The buffer. Allocated in gator_op_setup */
+static DEFINE_PER_CPU(char *[NUM_GATOR_BUFS], gator_buffer);
+/* The time after which the buffer should be committed for live display */
+static DEFINE_PER_CPU(u64, gator_buffer_commit_time);
+
+/* List of all gator events - new events must be added to this list */
+#define GATOR_EVENTS_LIST \
+	GATOR_EVENT(gator_events_armv6_init) \
+	GATOR_EVENT(gator_events_armv7_init) \
+	GATOR_EVENT(gator_events_block_init) \
+	GATOR_EVENT(gator_events_ccn504_init) \
+	GATOR_EVENT(gator_events_irq_init) \
+	GATOR_EVENT(gator_events_l2c310_init) \
+	GATOR_EVENT(gator_events_mali_init) \
+	GATOR_EVENT(gator_events_mali_midgard_hw_init) \
+	GATOR_EVENT(gator_events_mali_midgard_init) \
+	GATOR_EVENT(gator_events_meminfo_init) \
+	GATOR_EVENT(gator_events_mmapped_init) \
+	GATOR_EVENT(gator_events_net_init) \
+	GATOR_EVENT(gator_events_perf_pmu_init) \
+	GATOR_EVENT(gator_events_sched_init) \
+	GATOR_EVENT(gator_events_scorpion_init) \
+
+#define GATOR_EVENT(EVENT_INIT) __weak int EVENT_INIT(void);
+GATOR_EVENTS_LIST
+#undef GATOR_EVENT
+
+static int (*gator_events_list[])(void) = {
+#define GATOR_EVENT(EVENT_INIT) EVENT_INIT,
+GATOR_EVENTS_LIST
+#undef GATOR_EVENT
+};
+
+/******************************************************************************
+ * Application Includes
+ ******************************************************************************/
+#include "gator_fs.c"
+#include "gator_buffer_write.c"
+#include "gator_buffer.c"
+#include "gator_marshaling.c"
+#include "gator_hrtimer_gator.c"
+#include "gator_cookies.c"
+#include "gator_annotate.c"
+#include "gator_trace_sched.c"
+#include "gator_trace_power.c"
+#include "gator_trace_gpu.c"
+#include "gator_backtrace.c"
+
+/******************************************************************************
+ * Misc
+ ******************************************************************************/
+
+static const struct gator_cpu gator_cpus[] = {
+	{
+		.cpuid = ARM1136,
+		.core_name = "ARM1136",
+		.pmnc_name = "ARM_ARM11",
+		.dt_name = "arm,arm1136",
+		.pmnc_counters = 3,
+	},
+	{
+		.cpuid = ARM1156,
+		.core_name = "ARM1156",
+		.pmnc_name = "ARM_ARM11",
+		.dt_name = "arm,arm1156",
+		.pmnc_counters = 3,
+	},
+	{
+		.cpuid = ARM1176,
+		.core_name = "ARM1176",
+		.pmnc_name = "ARM_ARM11",
+		.dt_name = "arm,arm1176",
+		.pmnc_counters = 3,
+	},
+	{
+		.cpuid = ARM11MPCORE,
+		.core_name = "ARM11MPCore",
+		.pmnc_name = "ARM_ARM11MPCore",
+		.dt_name = "arm,arm11mpcore",
+		.pmnc_counters = 3,
+	},
+	{
+		.cpuid = CORTEX_A5,
+		.core_name = "Cortex-A5",
+		.pmnc_name = "ARMv7_Cortex_A5",
+		.dt_name = "arm,cortex-a5",
+		.pmnc_counters = 2,
+	},
+	{
+		.cpuid = CORTEX_A7,
+		.core_name = "Cortex-A7",
+		.pmnc_name = "ARMv7_Cortex_A7",
+		.dt_name = "arm,cortex-a7",
+		.pmnc_counters = 4,
+	},
+	{
+		.cpuid = CORTEX_A8,
+		.core_name = "Cortex-A8",
+		.pmnc_name = "ARMv7_Cortex_A8",
+		.dt_name = "arm,cortex-a8",
+		.pmnc_counters = 4,
+	},
+	{
+		.cpuid = CORTEX_A9,
+		.core_name = "Cortex-A9",
+		.pmnc_name = "ARMv7_Cortex_A9",
+		.dt_name = "arm,cortex-a9",
+		.pmnc_counters = 6,
+	},
+	{
+		.cpuid = CORTEX_A15,
+		.core_name = "Cortex-A15",
+		.pmnc_name = "ARMv7_Cortex_A15",
+		.dt_name = "arm,cortex-a15",
+		.pmnc_counters = 6,
+	},
+	{
+		.cpuid = CORTEX_A17,
+		.core_name = "Cortex-A17",
+		.pmnc_name = "ARMv7_Cortex_A17",
+		.dt_name = "arm,cortex-a17",
+		.pmnc_counters = 6,
+	},
+	{
+		.cpuid = SCORPION,
+		.core_name = "Scorpion",
+		.pmnc_name = "Scorpion",
+		.pmnc_counters = 4,
+	},
+	{
+		.cpuid = SCORPIONMP,
+		.core_name = "ScorpionMP",
+		.pmnc_name = "ScorpionMP",
+		.pmnc_counters = 4,
+	},
+	{
+		.cpuid = KRAITSIM,
+		.core_name = "KraitSIM",
+		.pmnc_name = "Krait",
+		.pmnc_counters = 4,
+	},
+	{
+		.cpuid = KRAIT,
+		.core_name = "Krait",
+		.pmnc_name = "Krait",
+		.pmnc_counters = 4,
+	},
+	{
+		.cpuid = KRAIT_S4_PRO,
+		.core_name = "Krait S4 Pro",
+		.pmnc_name = "Krait",
+		.pmnc_counters = 4,
+	},
+	{
+		.cpuid = CORTEX_A53,
+		.core_name = "Cortex-A53",
+		.pmnc_name = "ARM_Cortex-A53",
+		.dt_name = "arm,cortex-a53",
+		.pmnc_counters = 6,
+	},
+	{
+		.cpuid = CORTEX_A57,
+		.core_name = "Cortex-A57",
+		.pmnc_name = "ARM_Cortex-A57",
+		.dt_name = "arm,cortex-a57",
+		.pmnc_counters = 6,
+	},
+	{
+		.cpuid = AARCH64,
+		.core_name = "AArch64",
+		.pmnc_name = "ARM_AArch64",
+		.pmnc_counters = 6,
+	},
+	{
+		.cpuid = OTHER,
+		.core_name = "Other",
+		.pmnc_name = "Other",
+		.pmnc_counters = 6,
+	},
+	{}
+};
+
+const struct gator_cpu *gator_find_cpu_by_cpuid(const u32 cpuid)
+{
+	int i;
+
+	for (i = 0; gator_cpus[i].cpuid != 0; ++i) {
+		const struct gator_cpu *const gator_cpu = &gator_cpus[i];
+
+		if (gator_cpu->cpuid == cpuid)
+			return gator_cpu;
+	}
+
+	return NULL;
+}
+
+static const char OLD_PMU_PREFIX[] = "ARMv7 Cortex-";
+static const char NEW_PMU_PREFIX[] = "ARMv7_Cortex_";
+
+const struct gator_cpu *gator_find_cpu_by_pmu_name(const char *const name)
+{
+	int i;
+
+	for (i = 0; gator_cpus[i].cpuid != 0; ++i) {
+		const struct gator_cpu *const gator_cpu = &gator_cpus[i];
+
+		if (gator_cpu->pmnc_name != NULL &&
+		    /* Do the names match exactly? */
+		    (strcasecmp(gator_cpu->pmnc_name, name) == 0 ||
+		     /* Do these names match but have the old vs new prefix? */
+		     ((strncasecmp(name, OLD_PMU_PREFIX, sizeof(OLD_PMU_PREFIX) - 1) == 0 &&
+		       strncasecmp(gator_cpu->pmnc_name, NEW_PMU_PREFIX, sizeof(NEW_PMU_PREFIX) - 1) == 0 &&
+		       strcasecmp(name + sizeof(OLD_PMU_PREFIX) - 1, gator_cpu->pmnc_name + sizeof(NEW_PMU_PREFIX) - 1) == 0))))
+			return gator_cpu;
+	}
+
+	return NULL;
+}
+
+u32 gator_cpuid(void)
+{
+#if defined(__arm__) || defined(__aarch64__)
+	u32 val;
+#if !defined(__aarch64__)
+	asm volatile("mrc p15, 0, %0, c0, c0, 0" : "=r" (val));
+#else
+	asm volatile("mrs %0, midr_el1" : "=r" (val));
+#endif
+	return (val >> 4) & 0xfff;
+#else
+	return OTHER;
+#endif
+}
+
+static void gator_buffer_wake_up(unsigned long data)
+{
+	wake_up(&gator_buffer_wait);
+}
+
+static int gator_buffer_wake_func(void *data)
+{
+	for (;;) {
+		if (down_killable(&gator_buffer_wake_sem))
+			break;
+
+		/* Eat up any pending events */
+		while (!down_trylock(&gator_buffer_wake_sem))
+			;
+
+		if (!gator_buffer_wake_run)
+			break;
+
+		gator_buffer_wake_up(0);
+	}
+
+	return 0;
+}
+
+/******************************************************************************
+ * Commit interface
+ ******************************************************************************/
+static bool buffer_commit_ready(int *cpu, int *buftype)
+{
+	int cpu_x, x;
+
+	for_each_present_cpu(cpu_x) {
+		for (x = 0; x < NUM_GATOR_BUFS; x++)
+			if (per_cpu(gator_buffer_commit, cpu_x)[x] != per_cpu(gator_buffer_read, cpu_x)[x]) {
+				*cpu = cpu_x;
+				*buftype = x;
+				return true;
+			}
+	}
+	*cpu = -1;
+	*buftype = -1;
+	return false;
+}
+
+/******************************************************************************
+ * hrtimer interrupt processing
+ ******************************************************************************/
+static void gator_timer_interrupt(void)
+{
+	struct pt_regs *const regs = get_irq_regs();
+
+	gator_backtrace_handler(regs);
+}
+
+void gator_backtrace_handler(struct pt_regs *const regs)
+{
+	u64 time = gator_get_time();
+	int cpu = get_physical_cpu();
+
+	/* Output backtrace */
+	gator_add_sample(cpu, regs, time);
+
+	/* Collect counters */
+	if (!per_cpu(collecting, cpu))
+		collect_counters(time, current, false);
+
+	/* No buffer flushing occurs during sched switch for RT-Preempt full. The block counter frame will be flushed by collect_counters, but the sched buffer needs to be explicitly flushed */
+#ifdef CONFIG_PREEMPT_RT_FULL
+	buffer_check(cpu, SCHED_TRACE_BUF, time);
+#endif
+}
+
+static int gator_running;
+
+/* This function runs in interrupt context and on the appropriate core */
+static void gator_timer_offline(void *migrate)
+{
+	struct gator_interface *gi;
+	int i, len, cpu = get_physical_cpu();
+	int *buffer;
+	u64 time;
+
+	gator_trace_sched_offline();
+	gator_trace_power_offline();
+
+	if (!migrate)
+		gator_hrtimer_offline();
+
+	/* Offline any events and output counters */
+	time = gator_get_time();
+	if (marshal_event_header(time)) {
+		list_for_each_entry(gi, &gator_events, list) {
+			if (gi->offline) {
+				len = gi->offline(&buffer, migrate);
+				marshal_event(len, buffer);
+			}
+		}
+		/* Only check after writing all counters so that time and corresponding counters appear in the same frame */
+		buffer_check(cpu, BLOCK_COUNTER_BUF, time);
+	}
+
+	/* Flush all buffers on this core */
+	for (i = 0; i < NUM_GATOR_BUFS; i++)
+		gator_commit_buffer(cpu, i, time);
+}
+
+/* This function runs in interrupt context and may be running on a core other than core 'cpu' */
+static void gator_timer_offline_dispatch(int cpu, bool migrate)
+{
+	struct gator_interface *gi;
+
+	list_for_each_entry(gi, &gator_events, list) {
+		if (gi->offline_dispatch)
+			gi->offline_dispatch(cpu, migrate);
+	}
+}
+
+static void gator_timer_stop(void)
+{
+	int cpu;
+
+	if (gator_running) {
+		on_each_cpu(gator_timer_offline, NULL, 1);
+		for_each_online_cpu(cpu) {
+			gator_timer_offline_dispatch(lcpu_to_pcpu(cpu), false);
+		}
+
+		gator_running = 0;
+		gator_hrtimer_shutdown();
+	}
+}
+
+static void gator_send_core_name(const int cpu, const u32 cpuid)
+{
+#if defined(__arm__) || defined(__aarch64__)
+	if (!sent_core_name[cpu] || (cpuid != gator_cpuids[cpu])) {
+		const struct gator_cpu *const gator_cpu = gator_find_cpu_by_cpuid(cpuid);
+		const char *core_name = NULL;
+		char core_name_buf[32];
+
+		/* Save off this cpuid */
+		gator_cpuids[cpu] = cpuid;
+		if (gator_cpu != NULL) {
+			core_name = gator_cpu->core_name;
+		} else {
+			if (cpuid == -1)
+				snprintf(core_name_buf, sizeof(core_name_buf), "Unknown");
+			else
+				snprintf(core_name_buf, sizeof(core_name_buf), "Unknown (0x%.3x)", cpuid);
+			core_name = core_name_buf;
+		}
+
+		marshal_core_name(cpu, cpuid, core_name);
+		sent_core_name[cpu] = true;
+	}
+#endif
+}
+
+static void gator_read_cpuid(void *arg)
+{
+	gator_cpuids[get_physical_cpu()] = gator_cpuid();
+}
+
+/* This function runs in interrupt context and on the appropriate core */
+static void gator_timer_online(void *migrate)
+{
+	struct gator_interface *gi;
+	int len, cpu = get_physical_cpu();
+	int *buffer;
+	u64 time;
+
+	/* Send what is currently running on this core */
+	marshal_sched_trace_switch(current->pid, 0);
+
+	gator_trace_power_online();
+
+	/* online any events and output counters */
+	time = gator_get_time();
+	if (marshal_event_header(time)) {
+		list_for_each_entry(gi, &gator_events, list) {
+			if (gi->online) {
+				len = gi->online(&buffer, migrate);
+				marshal_event(len, buffer);
+			}
+		}
+		/* Only check after writing all counters so that time and corresponding counters appear in the same frame */
+		buffer_check(cpu, BLOCK_COUNTER_BUF, time);
+	}
+
+	if (!migrate)
+		gator_hrtimer_online();
+
+	gator_send_core_name(cpu, gator_cpuid());
+}
+
+/* This function runs in interrupt context and may be running on a core other than core 'cpu' */
+static void gator_timer_online_dispatch(int cpu, bool migrate)
+{
+	struct gator_interface *gi;
+
+	list_for_each_entry(gi, &gator_events, list) {
+		if (gi->online_dispatch)
+			gi->online_dispatch(cpu, migrate);
+	}
+}
+
+#include "gator_iks.c"
+
+static int gator_timer_start(unsigned long sample_rate)
+{
+	int cpu;
+
+	if (gator_running) {
+		pr_notice("gator: already running\n");
+		return 0;
+	}
+
+	gator_running = 1;
+
+	/* event based sampling trumps hr timer based sampling */
+	if (event_based_sampling)
+		sample_rate = 0;
+
+	if (gator_hrtimer_init(sample_rate, gator_timer_interrupt) == -1)
+		return -1;
+
+	/* Send off the previously saved cpuids */
+	for_each_present_cpu(cpu) {
+		preempt_disable();
+		gator_send_core_name(cpu, gator_cpuids[cpu]);
+		preempt_enable();
+	}
+
+	gator_send_iks_core_names();
+	for_each_online_cpu(cpu) {
+		gator_timer_online_dispatch(lcpu_to_pcpu(cpu), false);
+	}
+	on_each_cpu(gator_timer_online, NULL, 1);
+
+	return 0;
+}
+
+static u64 gator_get_time(void)
+{
+	struct timespec ts;
+	u64 timestamp;
+	u64 prev_timestamp;
+	u64 delta;
+	int cpu = smp_processor_id();
+
+	/* Match clock_gettime(CLOCK_MONOTONIC_RAW, &ts) from userspace */
+	getrawmonotonic(&ts);
+	timestamp = timespec_to_ns(&ts);
+
+	/* getrawmonotonic is not monotonic on all systems. Detect and
+	 * attempt to correct these cases. up to 0.5ms delta has been seen
+	 * on some systems, which can skew Streamline data when viewing at
+	 * high resolution. This doesn't work well with interrupts, but that
+	 * it's OK - the real concern is to catch big jumps in time
+	 */
+	prev_timestamp = per_cpu(last_timestamp, cpu);
+	if (prev_timestamp <= timestamp) {
+		per_cpu(last_timestamp, cpu) = timestamp;
+	} else {
+		delta = prev_timestamp - timestamp;
+		/* Log the error once */
+		if (!printed_monotonic_warning && delta > 500000) {
+			pr_err("%s: getrawmonotonic is not monotonic  cpu: %i  delta: %lli\nSkew in Streamline data may be present at the fine zoom levels\n", __func__, cpu, delta);
+			printed_monotonic_warning = true;
+		}
+		timestamp = prev_timestamp;
+	}
+
+	return timestamp - gator_monotonic_started;
+}
+
+static void gator_emit_perf_time(u64 time)
+{
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 10, 0)
+	if (time >= gator_sync_time) {
+		int cpu = get_physical_cpu();
+
+		marshal_event_single64(0, -1, local_clock());
+		gator_sync_time += NSEC_PER_SEC;
+		gator_commit_buffer(cpu, COUNTER_BUF, time);
+	}
+#endif
+}
+
+/******************************************************************************
+ * cpu hotplug and pm notifiers
+ ******************************************************************************/
+static int __cpuinit gator_hotcpu_notify(struct notifier_block *self, unsigned long action, void *hcpu)
+{
+	int cpu = lcpu_to_pcpu((long)hcpu);
+
+	switch (action) {
+	case CPU_DOWN_PREPARE:
+	case CPU_DOWN_PREPARE_FROZEN:
+		smp_call_function_single(cpu, gator_timer_offline, NULL, 1);
+		gator_timer_offline_dispatch(cpu, false);
+		break;
+	case CPU_ONLINE:
+	case CPU_ONLINE_FROZEN:
+		gator_timer_online_dispatch(cpu, false);
+		smp_call_function_single(cpu, gator_timer_online, NULL, 1);
+		break;
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block __refdata gator_hotcpu_notifier = {
+	.notifier_call = gator_hotcpu_notify,
+};
+
+/* n.b. calling "on_each_cpu" only runs on those that are online.
+ * Registered linux events are not disabled, so their counters will
+ * continue to collect
+ */
+static int gator_pm_notify(struct notifier_block *nb, unsigned long event, void *dummy)
+{
+	int cpu;
+	struct timespec ts;
+
+	switch (event) {
+	case PM_HIBERNATION_PREPARE:
+	case PM_SUSPEND_PREPARE:
+		unregister_hotcpu_notifier(&gator_hotcpu_notifier);
+		unregister_scheduler_tracepoints();
+		on_each_cpu(gator_timer_offline, NULL, 1);
+		for_each_online_cpu(cpu) {
+			gator_timer_offline_dispatch(lcpu_to_pcpu(cpu), false);
+		}
+
+		/* Record the wallclock hibernate time */
+		getnstimeofday(&ts);
+		gator_hibernate_time = timespec_to_ns(&ts) - gator_get_time();
+		break;
+	case PM_POST_HIBERNATION:
+	case PM_POST_SUSPEND:
+		/* Adjust gator_monotonic_started for the time spent sleeping, as gator_get_time does not account for it */
+		if (gator_hibernate_time > 0) {
+			getnstimeofday(&ts);
+			gator_monotonic_started += gator_hibernate_time + gator_get_time() - timespec_to_ns(&ts);
+			gator_hibernate_time = 0;
+		}
+
+		for_each_online_cpu(cpu) {
+			gator_timer_online_dispatch(lcpu_to_pcpu(cpu), false);
+		}
+		on_each_cpu(gator_timer_online, NULL, 1);
+		register_scheduler_tracepoints();
+		register_hotcpu_notifier(&gator_hotcpu_notifier);
+		break;
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block gator_pm_notifier = {
+	.notifier_call = gator_pm_notify,
+};
+
+static int gator_notifier_start(void)
+{
+	int retval;
+
+	retval = register_hotcpu_notifier(&gator_hotcpu_notifier);
+	if (retval == 0)
+		retval = register_pm_notifier(&gator_pm_notifier);
+	return retval;
+}
+
+static void gator_notifier_stop(void)
+{
+	unregister_pm_notifier(&gator_pm_notifier);
+	unregister_hotcpu_notifier(&gator_hotcpu_notifier);
+}
+
+/******************************************************************************
+ * Main
+ ******************************************************************************/
+static void gator_summary(void)
+{
+	u64 timestamp, uptime;
+	struct timespec ts;
+	char uname_buf[512];
+
+	snprintf(uname_buf, sizeof(uname_buf), "%s %s %s %s %s GNU/Linux", utsname()->sysname, utsname()->nodename, utsname()->release, utsname()->version, utsname()->machine);
+
+	getnstimeofday(&ts);
+	timestamp = timespec_to_ns(&ts);
+
+	/* Similar to reading /proc/uptime from fs/proc/uptime.c, calculate uptime */
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 11, 0)
+	{
+		void (*m2b)(struct timespec *ts);
+
+		do_posix_clock_monotonic_gettime(&ts);
+		/* monotonic_to_bootbased is not defined for some versions of Android */
+		m2b = symbol_get(monotonic_to_bootbased);
+		if (m2b)
+			m2b(&ts);
+	}
+#else
+	get_monotonic_boottime(&ts);
+#endif
+	uptime = timespec_to_ns(&ts);
+
+	/* Disable preemption as gator_get_time calls smp_processor_id to verify time is monotonic */
+	preempt_disable();
+	/* Set monotonic_started to zero as gator_get_time is uptime minus monotonic_started */
+	gator_monotonic_started = 0;
+	gator_monotonic_started = gator_get_time();
+
+	marshal_summary(timestamp, uptime, gator_monotonic_started, uname_buf);
+	gator_sync_time = 0;
+	gator_emit_perf_time(gator_monotonic_started);	
+	preempt_enable();
+}
+
+int gator_events_install(struct gator_interface *interface)
+{
+	list_add_tail(&interface->list, &gator_events);
+
+	return 0;
+}
+
+int gator_events_get_key(void)
+{
+	/* key 0 is reserved as a timestamp. key 1 is reserved as the marker
+	 * for thread specific counters. key 2 is reserved as the marker for
+	 * core. Odd keys are assigned by the driver, even keys by the
+	 * daemon.
+	 */
+	static int key = 3;
+	const int ret = key;
+
+	key += 2;
+	return ret;
+}
+
+static int gator_init(void)
+{
+	int i;
+
+	calc_first_cluster_size();
+
+	/* events sources */
+	for (i = 0; i < ARRAY_SIZE(gator_events_list); i++)
+		if (gator_events_list[i])
+			gator_events_list[i]();
+
+	gator_trace_sched_init();
+	gator_trace_power_init();
+
+	return 0;
+}
+
+static void gator_exit(void)
+{
+	struct gator_interface *gi;
+
+	list_for_each_entry(gi, &gator_events, list)
+		if (gi->shutdown)
+			gi->shutdown();
+}
+
+static int gator_start(void)
+{
+	unsigned long cpu, i;
+	struct gator_interface *gi;
+
+	gator_buffer_wake_run = true;
+	gator_buffer_wake_thread = kthread_run(gator_buffer_wake_func, NULL, "gator_bwake");
+	if (IS_ERR(gator_buffer_wake_thread))
+		goto bwake_failure;
+
+	if (gator_migrate_start())
+		goto migrate_failure;
+
+	/* Initialize the buffer with the frame type and core */
+	for_each_present_cpu(cpu) {
+		for (i = 0; i < NUM_GATOR_BUFS; i++)
+			marshal_frame(cpu, i);
+		per_cpu(last_timestamp, cpu) = 0;
+	}
+	printed_monotonic_warning = false;
+
+	/* Capture the start time */
+	gator_summary();
+
+	/* start all events */
+	list_for_each_entry(gi, &gator_events, list) {
+		if (gi->start && gi->start() != 0) {
+			struct list_head *ptr = gi->list.prev;
+
+			while (ptr != &gator_events) {
+				gi = list_entry(ptr, struct gator_interface, list);
+
+				if (gi->stop)
+					gi->stop();
+
+				ptr = ptr->prev;
+			}
+			goto events_failure;
+		}
+	}
+
+	/* cookies shall be initialized before trace_sched_start() and gator_timer_start() */
+	if (cookies_initialize())
+		goto cookies_failure;
+	if (gator_annotate_start())
+		goto annotate_failure;
+	if (gator_trace_sched_start())
+		goto sched_failure;
+	if (gator_trace_power_start())
+		goto power_failure;
+	if (gator_trace_gpu_start())
+		goto gpu_failure;
+	if (gator_timer_start(gator_timer_count))
+		goto timer_failure;
+	if (gator_notifier_start())
+		goto notifier_failure;
+
+	return 0;
+
+notifier_failure:
+	gator_timer_stop();
+timer_failure:
+	gator_trace_gpu_stop();
+gpu_failure:
+	gator_trace_power_stop();
+power_failure:
+	gator_trace_sched_stop();
+sched_failure:
+	gator_annotate_stop();
+annotate_failure:
+	cookies_release();
+cookies_failure:
+	/* stop all events */
+	list_for_each_entry(gi, &gator_events, list)
+		if (gi->stop)
+			gi->stop();
+events_failure:
+	gator_migrate_stop();
+migrate_failure:
+	gator_buffer_wake_run = false;
+	up(&gator_buffer_wake_sem);
+	gator_buffer_wake_thread = NULL;
+bwake_failure:
+
+	return -1;
+}
+
+static void gator_stop(void)
+{
+	struct gator_interface *gi;
+
+	gator_annotate_stop();
+	gator_trace_sched_stop();
+	gator_trace_power_stop();
+	gator_trace_gpu_stop();
+
+	/* stop all interrupt callback reads before tearing down other interfaces */
+	gator_notifier_stop();	/* should be called before gator_timer_stop to avoid re-enabling the hrtimer after it has been offlined */
+	gator_timer_stop();
+
+	/* stop all events */
+	list_for_each_entry(gi, &gator_events, list)
+		if (gi->stop)
+			gi->stop();
+
+	gator_migrate_stop();
+
+	gator_buffer_wake_run = false;
+	up(&gator_buffer_wake_sem);
+	gator_buffer_wake_thread = NULL;
+}
+
+/******************************************************************************
+ * Filesystem
+ ******************************************************************************/
+/* fopen("buffer") */
+static int gator_op_setup(void)
+{
+	int err = 0;
+	int cpu, i;
+
+	mutex_lock(&start_mutex);
+
+	gator_buffer_size[SUMMARY_BUF] = SUMMARY_BUFFER_SIZE;
+	gator_buffer_mask[SUMMARY_BUF] = SUMMARY_BUFFER_SIZE - 1;
+
+	gator_buffer_size[BACKTRACE_BUF] = BACKTRACE_BUFFER_SIZE;
+	gator_buffer_mask[BACKTRACE_BUF] = BACKTRACE_BUFFER_SIZE - 1;
+
+	gator_buffer_size[NAME_BUF] = NAME_BUFFER_SIZE;
+	gator_buffer_mask[NAME_BUF] = NAME_BUFFER_SIZE - 1;
+
+	gator_buffer_size[COUNTER_BUF] = COUNTER_BUFFER_SIZE;
+	gator_buffer_mask[COUNTER_BUF] = COUNTER_BUFFER_SIZE - 1;
+
+	gator_buffer_size[BLOCK_COUNTER_BUF] = BLOCK_COUNTER_BUFFER_SIZE;
+	gator_buffer_mask[BLOCK_COUNTER_BUF] = BLOCK_COUNTER_BUFFER_SIZE - 1;
+
+	gator_buffer_size[ANNOTATE_BUF] = ANNOTATE_BUFFER_SIZE;
+	gator_buffer_mask[ANNOTATE_BUF] = ANNOTATE_BUFFER_SIZE - 1;
+
+	gator_buffer_size[SCHED_TRACE_BUF] = SCHED_TRACE_BUFFER_SIZE;
+	gator_buffer_mask[SCHED_TRACE_BUF] = SCHED_TRACE_BUFFER_SIZE - 1;
+
+	gator_buffer_size[IDLE_BUF] = IDLE_BUFFER_SIZE;
+	gator_buffer_mask[IDLE_BUF] = IDLE_BUFFER_SIZE - 1;
+
+	gator_buffer_size[ACTIVITY_BUF] = ACTIVITY_BUFFER_SIZE;
+	gator_buffer_mask[ACTIVITY_BUF] = ACTIVITY_BUFFER_SIZE - 1;
+
+	/* Initialize percpu per buffer variables */
+	for (i = 0; i < NUM_GATOR_BUFS; i++) {
+		/* Verify buffers are a power of 2 */
+		if (gator_buffer_size[i] & (gator_buffer_size[i] - 1)) {
+			err = -ENOEXEC;
+			goto setup_error;
+		}
+
+		for_each_present_cpu(cpu) {
+			per_cpu(gator_buffer_read, cpu)[i] = 0;
+			per_cpu(gator_buffer_write, cpu)[i] = 0;
+			per_cpu(gator_buffer_commit, cpu)[i] = 0;
+			per_cpu(buffer_space_available, cpu)[i] = true;
+			per_cpu(gator_buffer_commit_time, cpu) = gator_live_rate;
+
+			/* Annotation is a special case that only uses a single buffer */
+			if (cpu > 0 && i == ANNOTATE_BUF) {
+				per_cpu(gator_buffer, cpu)[i] = NULL;
+				continue;
+			}
+
+			per_cpu(gator_buffer, cpu)[i] = vmalloc(gator_buffer_size[i]);
+			if (!per_cpu(gator_buffer, cpu)[i]) {
+				err = -ENOMEM;
+				goto setup_error;
+			}
+		}
+	}
+
+setup_error:
+	mutex_unlock(&start_mutex);
+	return err;
+}
+
+/* Actually start profiling (echo 1>/dev/gator/enable) */
+static int gator_op_start(void)
+{
+	int err = 0;
+
+	mutex_lock(&start_mutex);
+
+	if (gator_started || gator_start())
+		err = -EINVAL;
+	else
+		gator_started = 1;
+
+	mutex_unlock(&start_mutex);
+
+	return err;
+}
+
+/* echo 0>/dev/gator/enable */
+static void gator_op_stop(void)
+{
+	mutex_lock(&start_mutex);
+
+	if (gator_started) {
+		gator_stop();
+
+		mutex_lock(&gator_buffer_mutex);
+
+		gator_started = 0;
+		gator_monotonic_started = 0;
+		cookies_release();
+		wake_up(&gator_buffer_wait);
+
+		mutex_unlock(&gator_buffer_mutex);
+	}
+
+	mutex_unlock(&start_mutex);
+}
+
+static void gator_shutdown(void)
+{
+	int cpu, i;
+
+	mutex_lock(&start_mutex);
+
+	for_each_present_cpu(cpu) {
+		mutex_lock(&gator_buffer_mutex);
+		for (i = 0; i < NUM_GATOR_BUFS; i++) {
+			vfree(per_cpu(gator_buffer, cpu)[i]);
+			per_cpu(gator_buffer, cpu)[i] = NULL;
+			per_cpu(gator_buffer_read, cpu)[i] = 0;
+			per_cpu(gator_buffer_write, cpu)[i] = 0;
+			per_cpu(gator_buffer_commit, cpu)[i] = 0;
+			per_cpu(buffer_space_available, cpu)[i] = true;
+			per_cpu(gator_buffer_commit_time, cpu) = 0;
+		}
+		mutex_unlock(&gator_buffer_mutex);
+	}
+
+	memset(&sent_core_name, 0, sizeof(sent_core_name));
+
+	mutex_unlock(&start_mutex);
+}
+
+static int gator_set_backtrace(unsigned long val)
+{
+	int err = 0;
+
+	mutex_lock(&start_mutex);
+
+	if (gator_started)
+		err = -EBUSY;
+	else
+		gator_backtrace_depth = val;
+
+	mutex_unlock(&start_mutex);
+
+	return err;
+}
+
+static ssize_t enable_read(struct file *file, char __user *buf, size_t count, loff_t *offset)
+{
+	return gatorfs_ulong_to_user(gator_started, buf, count, offset);
+}
+
+static ssize_t enable_write(struct file *file, char const __user *buf, size_t count, loff_t *offset)
+{
+	unsigned long val;
+	int retval;
+
+	if (*offset)
+		return -EINVAL;
+
+	retval = gatorfs_ulong_from_user(&val, buf, count);
+	if (retval)
+		return retval;
+
+	if (val)
+		retval = gator_op_start();
+	else
+		gator_op_stop();
+
+	if (retval)
+		return retval;
+	return count;
+}
+
+static const struct file_operations enable_fops = {
+	.read = enable_read,
+	.write = enable_write,
+};
+
+static int userspace_buffer_open(struct inode *inode, struct file *file)
+{
+	int err = -EPERM;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (test_and_set_bit_lock(0, &gator_buffer_opened))
+		return -EBUSY;
+
+	err = gator_op_setup();
+	if (err)
+		goto fail;
+
+	/* NB: the actual start happens from userspace
+	 * echo 1 >/dev/gator/enable
+	 */
+
+	return 0;
+
+fail:
+	__clear_bit_unlock(0, &gator_buffer_opened);
+	return err;
+}
+
+static int userspace_buffer_release(struct inode *inode, struct file *file)
+{
+	gator_op_stop();
+	gator_shutdown();
+	__clear_bit_unlock(0, &gator_buffer_opened);
+	return 0;
+}
+
+static ssize_t userspace_buffer_read(struct file *file, char __user *buf, size_t count, loff_t *offset)
+{
+	int commit, length1, length2, read;
+	char *buffer1;
+	char *buffer2;
+	int cpu, buftype;
+	int written = 0;
+
+	/* ensure there is enough space for a whole frame */
+	if (count < userspace_buffer_size || *offset)
+		return -EINVAL;
+
+	/* sleep until the condition is true or a signal is received the
+	 * condition is checked each time gator_buffer_wait is woken up
+	 */
+	wait_event_interruptible(gator_buffer_wait, buffer_commit_ready(&cpu, &buftype) || !gator_started);
+
+	if (signal_pending(current))
+		return -EINTR;
+
+	if (buftype == -1 || cpu == -1)
+		return 0;
+
+	mutex_lock(&gator_buffer_mutex);
+
+	do {
+		read = per_cpu(gator_buffer_read, cpu)[buftype];
+		commit = per_cpu(gator_buffer_commit, cpu)[buftype];
+
+		/* May happen if the buffer is freed during pending reads. */
+		if (!per_cpu(gator_buffer, cpu)[buftype])
+			break;
+
+		/* determine the size of two halves */
+		length1 = commit - read;
+		length2 = 0;
+		buffer1 = &(per_cpu(gator_buffer, cpu)[buftype][read]);
+		buffer2 = &(per_cpu(gator_buffer, cpu)[buftype][0]);
+		if (length1 < 0) {
+			length1 = gator_buffer_size[buftype] - read;
+			length2 = commit;
+		}
+
+		if (length1 + length2 > count - written)
+			break;
+
+		/* start, middle or end */
+		if (length1 > 0 && copy_to_user(&buf[written], buffer1, length1))
+			break;
+
+		/* possible wrap around */
+		if (length2 > 0 && copy_to_user(&buf[written + length1], buffer2, length2))
+			break;
+
+		per_cpu(gator_buffer_read, cpu)[buftype] = commit;
+		written += length1 + length2;
+
+		/* Wake up annotate_write if more space is available */
+		if (buftype == ANNOTATE_BUF)
+			wake_up(&gator_annotate_wait);
+	} while (buffer_commit_ready(&cpu, &buftype));
+
+	mutex_unlock(&gator_buffer_mutex);
+
+	/* kick just in case we've lost an SMP event */
+	wake_up(&gator_buffer_wait);
+
+	return written > 0 ? written : -EFAULT;
+}
+
+static const struct file_operations gator_event_buffer_fops = {
+	.open = userspace_buffer_open,
+	.release = userspace_buffer_release,
+	.read = userspace_buffer_read,
+};
+
+static ssize_t depth_read(struct file *file, char __user *buf, size_t count, loff_t *offset)
+{
+	return gatorfs_ulong_to_user(gator_backtrace_depth, buf, count, offset);
+}
+
+static ssize_t depth_write(struct file *file, char const __user *buf, size_t count, loff_t *offset)
+{
+	unsigned long val;
+	int retval;
+
+	if (*offset)
+		return -EINVAL;
+
+	retval = gatorfs_ulong_from_user(&val, buf, count);
+	if (retval)
+		return retval;
+
+	retval = gator_set_backtrace(val);
+
+	if (retval)
+		return retval;
+	return count;
+}
+
+static const struct file_operations depth_fops = {
+	.read = depth_read,
+	.write = depth_write
+};
+
+static void gator_op_create_files(struct super_block *sb, struct dentry *root)
+{
+	struct dentry *dir;
+	struct gator_interface *gi;
+	int cpu;
+
+	/* reinitialize default values */
+	gator_cpu_cores = 0;
+	for_each_present_cpu(cpu) {
+		gator_cpu_cores++;
+	}
+	userspace_buffer_size = BACKTRACE_BUFFER_SIZE;
+	gator_response_type = 1;
+	gator_live_rate = 0;
+
+	gatorfs_create_file(sb, root, "enable", &enable_fops);
+	gatorfs_create_file(sb, root, "buffer", &gator_event_buffer_fops);
+	gatorfs_create_file(sb, root, "backtrace_depth", &depth_fops);
+	gatorfs_create_ro_ulong(sb, root, "cpu_cores", &gator_cpu_cores);
+	gatorfs_create_ro_ulong(sb, root, "buffer_size", &userspace_buffer_size);
+	gatorfs_create_ulong(sb, root, "tick", &gator_timer_count);
+	gatorfs_create_ulong(sb, root, "response_type", &gator_response_type);
+	gatorfs_create_ro_ulong(sb, root, "version", &gator_protocol_version);
+	gatorfs_create_ro_u64(sb, root, "started", &gator_monotonic_started);
+	gatorfs_create_u64(sb, root, "live_rate", &gator_live_rate);
+
+	/* Annotate interface */
+	gator_annotate_create_files(sb, root);
+
+	/* Linux Events */
+	dir = gatorfs_mkdir(sb, root, "events");
+	list_for_each_entry(gi, &gator_events, list)
+		if (gi->create_files)
+			gi->create_files(sb, dir);
+
+	/* Sched Events */
+	sched_trace_create_files(sb, dir);
+
+	/* Power interface */
+	gator_trace_power_create_files(sb, dir);
+}
+
+/******************************************************************************
+ * Module
+ ******************************************************************************/
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 15, 0)
+
+#define GATOR_TRACEPOINTS \
+	GATOR_HANDLE_TRACEPOINT(block_rq_complete); \
+	GATOR_HANDLE_TRACEPOINT(cpu_frequency); \
+	GATOR_HANDLE_TRACEPOINT(cpu_idle); \
+	GATOR_HANDLE_TRACEPOINT(cpu_migrate_begin); \
+	GATOR_HANDLE_TRACEPOINT(cpu_migrate_current); \
+	GATOR_HANDLE_TRACEPOINT(cpu_migrate_finish); \
+	GATOR_HANDLE_TRACEPOINT(irq_handler_exit); \
+	GATOR_HANDLE_TRACEPOINT(mali_hw_counter); \
+	GATOR_HANDLE_TRACEPOINT(mali_job_slots_event); \
+	GATOR_HANDLE_TRACEPOINT(mali_mmu_as_in_use); \
+	GATOR_HANDLE_TRACEPOINT(mali_mmu_as_released); \
+	GATOR_HANDLE_TRACEPOINT(mali_page_fault_insert_pages); \
+	GATOR_HANDLE_TRACEPOINT(mali_pm_status); \
+	GATOR_HANDLE_TRACEPOINT(mali_sw_counter); \
+	GATOR_HANDLE_TRACEPOINT(mali_sw_counters); \
+	GATOR_HANDLE_TRACEPOINT(mali_timeline_event); \
+	GATOR_HANDLE_TRACEPOINT(mali_total_alloc_pages_change); \
+	GATOR_HANDLE_TRACEPOINT(mm_page_alloc); \
+	GATOR_HANDLE_TRACEPOINT(mm_page_free); \
+	GATOR_HANDLE_TRACEPOINT(mm_page_free_batched); \
+	GATOR_HANDLE_TRACEPOINT(sched_process_exec); \
+	GATOR_HANDLE_TRACEPOINT(sched_process_fork); \
+	GATOR_HANDLE_TRACEPOINT(sched_process_free); \
+	GATOR_HANDLE_TRACEPOINT(sched_switch); \
+	GATOR_HANDLE_TRACEPOINT(softirq_exit); \
+	GATOR_HANDLE_TRACEPOINT(task_rename); \
+
+#define GATOR_HANDLE_TRACEPOINT(probe_name) \
+	struct tracepoint *gator_tracepoint_##probe_name
+GATOR_TRACEPOINTS;
+#undef GATOR_HANDLE_TRACEPOINT
+
+static void gator_save_tracepoint(struct tracepoint *tp, void *priv)
+{
+#define GATOR_HANDLE_TRACEPOINT(probe_name) \
+	do { \
+		if (strcmp(tp->name, #probe_name) == 0) { \
+			gator_tracepoint_##probe_name = tp; \
+			return; \
+		} \
+	} while (0)
+GATOR_TRACEPOINTS;
+#undef GATOR_HANDLE_TRACEPOINT
+}
+
+#else
+
+#define for_each_kernel_tracepoint(fct, priv)
+
+#endif
+
+static int __init gator_module_init(void)
+{
+	for_each_kernel_tracepoint(gator_save_tracepoint, NULL);
+
+	if (gatorfs_register())
+		return -1;
+
+	if (gator_init()) {
+		gatorfs_unregister();
+		return -1;
+	}
+
+	setup_timer(&gator_buffer_wake_up_timer, gator_buffer_wake_up, 0);
+
+	/* Initialize the list of cpuids */
+	memset(gator_cpuids, -1, sizeof(gator_cpuids));
+	on_each_cpu(gator_read_cpuid, NULL, 1);
+
+	return 0;
+}
+
+static void __exit gator_module_exit(void)
+{
+	del_timer_sync(&gator_buffer_wake_up_timer);
+	tracepoint_synchronize_unregister();
+	gator_exit();
+	gatorfs_unregister();
+}
+
+module_init(gator_module_init);
+module_exit(gator_module_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("ARM Ltd");
+MODULE_DESCRIPTION("Gator system profiler");
+#define STRIFY2(ARG) #ARG
+#define STRIFY(ARG) STRIFY2(ARG)
+MODULE_VERSION(STRIFY(PROTOCOL_VERSION));
diff --git a/drivers/gator/gator_marshaling.c b/drivers/gator/gator_marshaling.c
new file mode 100644
index 000000000000..0d1167643642
--- /dev/null
+++ b/drivers/gator/gator_marshaling.c
@@ -0,0 +1,371 @@
+/**
+ * Copyright (C) ARM Limited 2012-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#define NEWLINE_CANARY \
+	/* Unix */ \
+	"1\n" \
+	/* Windows */ \
+	"2\r\n" \
+	/* Mac OS */ \
+	"3\r" \
+	/* RISC OS */ \
+	"4\n\r" \
+	/* Add another character so the length isn't 0x0a bytes */ \
+	"5"
+
+#ifdef MALI_SUPPORT
+#include "gator_events_mali_common.h"
+#endif
+
+static void marshal_summary(long long timestamp, long long uptime, long long monotonic_delta, const char *uname)
+{
+	unsigned long flags;
+	int cpu = 0;
+
+	local_irq_save(flags);
+	gator_buffer_write_packed_int(cpu, SUMMARY_BUF, MESSAGE_SUMMARY);
+	gator_buffer_write_string(cpu, SUMMARY_BUF, NEWLINE_CANARY);
+	gator_buffer_write_packed_int64(cpu, SUMMARY_BUF, timestamp);
+	gator_buffer_write_packed_int64(cpu, SUMMARY_BUF, uptime);
+	gator_buffer_write_packed_int64(cpu, SUMMARY_BUF, monotonic_delta);
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "uname");
+	gator_buffer_write_string(cpu, SUMMARY_BUF, uname);
+#if GATOR_IKS_SUPPORT
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "iks");
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "");
+#endif
+#ifdef CONFIG_PREEMPT_RTB
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "preempt_rtb");
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "");
+#endif
+#ifdef CONFIG_PREEMPT_RT_FULL
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "preempt_rt_full");
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "");
+#endif
+	/* Let Streamline know which GPU is used so that it can label the GPU Activity appropriately. This is a temporary fix, to be improved in a future release. */
+#ifdef MALI_SUPPORT
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "mali_type");
+#if (MALI_SUPPORT == MALI_4xx)
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "4xx");
+#elif (MALI_SUPPORT == MALI_MIDGARD)
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "6xx");
+#else
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "unknown");
+#endif
+#endif
+	gator_buffer_write_string(cpu, SUMMARY_BUF, "");
+	/* Commit the buffer now so it can be one of the first frames read by Streamline */
+	local_irq_restore(flags);
+	gator_commit_buffer(cpu, SUMMARY_BUF, gator_get_time());
+}
+
+static bool marshal_cookie_header(const char *text)
+{
+	int cpu = get_physical_cpu();
+
+	return buffer_check_space(cpu, NAME_BUF, strlen(text) + 3 * MAXSIZE_PACK32);
+}
+
+static void marshal_cookie(int cookie, const char *text)
+{
+	int cpu = get_physical_cpu();
+	/* buffer_check_space already called by marshal_cookie_header */
+	gator_buffer_write_packed_int(cpu, NAME_BUF, MESSAGE_COOKIE);
+	gator_buffer_write_packed_int(cpu, NAME_BUF, cookie);
+	gator_buffer_write_string(cpu, NAME_BUF, text);
+	buffer_check(cpu, NAME_BUF, gator_get_time());
+}
+
+static void marshal_thread_name(int pid, char *name)
+{
+	unsigned long flags, cpu;
+	u64 time;
+
+	local_irq_save(flags);
+	cpu = get_physical_cpu();
+	time = gator_get_time();
+	if (buffer_check_space(cpu, NAME_BUF, TASK_COMM_LEN + 3 * MAXSIZE_PACK32 + MAXSIZE_PACK64)) {
+		gator_buffer_write_packed_int(cpu, NAME_BUF, MESSAGE_THREAD_NAME);
+		gator_buffer_write_packed_int64(cpu, NAME_BUF, time);
+		gator_buffer_write_packed_int(cpu, NAME_BUF, pid);
+		gator_buffer_write_string(cpu, NAME_BUF, name);
+	}
+	local_irq_restore(flags);
+	buffer_check(cpu, NAME_BUF, time);
+}
+
+static void marshal_link(int cookie, int tgid, int pid)
+{
+	unsigned long cpu = get_physical_cpu(), flags;
+	u64 time;
+
+	local_irq_save(flags);
+	time = gator_get_time();
+	if (buffer_check_space(cpu, NAME_BUF, MAXSIZE_PACK64 + 5 * MAXSIZE_PACK32)) {
+		gator_buffer_write_packed_int(cpu, NAME_BUF, MESSAGE_LINK);
+		gator_buffer_write_packed_int64(cpu, NAME_BUF, time);
+		gator_buffer_write_packed_int(cpu, NAME_BUF, cookie);
+		gator_buffer_write_packed_int(cpu, NAME_BUF, tgid);
+		gator_buffer_write_packed_int(cpu, NAME_BUF, pid);
+	}
+	local_irq_restore(flags);
+	/* Check and commit; commit is set to occur once buffer is 3/4 full */
+	buffer_check(cpu, NAME_BUF, time);
+}
+
+static bool marshal_backtrace_header(int exec_cookie, int tgid, int pid, u64 time)
+{
+	int cpu = get_physical_cpu();
+
+	if (!buffer_check_space(cpu, BACKTRACE_BUF, MAXSIZE_PACK64 + 5 * MAXSIZE_PACK32 + gator_backtrace_depth * 2 * MAXSIZE_PACK32)) {
+		/* Check and commit; commit is set to occur once buffer is 3/4 full */
+		buffer_check(cpu, BACKTRACE_BUF, time);
+
+		return false;
+	}
+
+	gator_buffer_write_packed_int64(cpu, BACKTRACE_BUF, time);
+	gator_buffer_write_packed_int(cpu, BACKTRACE_BUF, exec_cookie);
+	gator_buffer_write_packed_int(cpu, BACKTRACE_BUF, tgid);
+	gator_buffer_write_packed_int(cpu, BACKTRACE_BUF, pid);
+
+	return true;
+}
+
+static void marshal_backtrace(unsigned long address, int cookie, int in_kernel)
+{
+	int cpu = get_physical_cpu();
+
+	if (cookie == 0 && !in_kernel)
+		cookie = UNRESOLVED_COOKIE;
+	gator_buffer_write_packed_int(cpu, BACKTRACE_BUF, cookie);
+	gator_buffer_write_packed_int64(cpu, BACKTRACE_BUF, address);
+}
+
+static void marshal_backtrace_footer(u64 time)
+{
+	int cpu = get_physical_cpu();
+
+	gator_buffer_write_packed_int(cpu, BACKTRACE_BUF, MESSAGE_END_BACKTRACE);
+
+	/* Check and commit; commit is set to occur once buffer is 3/4 full */
+	buffer_check(cpu, BACKTRACE_BUF, time);
+}
+
+static bool marshal_event_header(u64 time)
+{
+	unsigned long flags, cpu = get_physical_cpu();
+	bool retval = false;
+
+	local_irq_save(flags);
+	if (buffer_check_space(cpu, BLOCK_COUNTER_BUF, MAXSIZE_PACK32 + MAXSIZE_PACK64)) {
+		gator_buffer_write_packed_int(cpu, BLOCK_COUNTER_BUF, 0);	/* key of zero indicates a timestamp */
+		gator_buffer_write_packed_int64(cpu, BLOCK_COUNTER_BUF, time);
+		retval = true;
+	}
+	local_irq_restore(flags);
+
+	return retval;
+}
+
+static void marshal_event(int len, int *buffer)
+{
+	unsigned long i, flags, cpu = get_physical_cpu();
+
+	if (len <= 0)
+		return;
+
+	/* length must be even since all data is a (key, value) pair */
+	if (len & 0x1) {
+		pr_err("gator: invalid counter data detected and discarded\n");
+		return;
+	}
+
+	/* events must be written in key,value pairs */
+	local_irq_save(flags);
+	for (i = 0; i < len; i += 2) {
+		if (!buffer_check_space(cpu, BLOCK_COUNTER_BUF, 2 * MAXSIZE_PACK32))
+			break;
+		gator_buffer_write_packed_int(cpu, BLOCK_COUNTER_BUF, buffer[i]);
+		gator_buffer_write_packed_int(cpu, BLOCK_COUNTER_BUF, buffer[i + 1]);
+	}
+	local_irq_restore(flags);
+}
+
+static void marshal_event64(int len, long long *buffer64)
+{
+	unsigned long i, flags, cpu = get_physical_cpu();
+
+	if (len <= 0)
+		return;
+
+	/* length must be even since all data is a (key, value) pair */
+	if (len & 0x1) {
+		pr_err("gator: invalid counter data detected and discarded\n");
+		return;
+	}
+
+	/* events must be written in key,value pairs */
+	local_irq_save(flags);
+	for (i = 0; i < len; i += 2) {
+		if (!buffer_check_space(cpu, BLOCK_COUNTER_BUF, 2 * MAXSIZE_PACK64))
+			break;
+		gator_buffer_write_packed_int64(cpu, BLOCK_COUNTER_BUF, buffer64[i]);
+		gator_buffer_write_packed_int64(cpu, BLOCK_COUNTER_BUF, buffer64[i + 1]);
+	}
+	local_irq_restore(flags);
+}
+
+static void __maybe_unused marshal_event_single(int core, int key, int value)
+{
+	unsigned long flags, cpu;
+	u64 time;
+
+	local_irq_save(flags);
+	cpu = get_physical_cpu();
+	time = gator_get_time();
+	if (buffer_check_space(cpu, COUNTER_BUF, MAXSIZE_PACK64 + 3 * MAXSIZE_PACK32)) {
+		gator_buffer_write_packed_int64(cpu, COUNTER_BUF, time);
+		gator_buffer_write_packed_int(cpu, COUNTER_BUF, core);
+		gator_buffer_write_packed_int(cpu, COUNTER_BUF, key);
+		gator_buffer_write_packed_int(cpu, COUNTER_BUF, value);
+	}
+	local_irq_restore(flags);
+	/* Check and commit; commit is set to occur once buffer is 3/4 full */
+	buffer_check(cpu, COUNTER_BUF, time);
+}
+
+static void __maybe_unused marshal_event_single64(int core, int key, long long value)
+{
+	unsigned long flags, cpu;
+	u64 time;
+
+	local_irq_save(flags);
+	cpu = get_physical_cpu();
+	time = gator_get_time();
+	if (buffer_check_space(cpu, COUNTER_BUF, 2 * MAXSIZE_PACK64 + 2 * MAXSIZE_PACK32)) {
+		gator_buffer_write_packed_int64(cpu, COUNTER_BUF, time);
+		gator_buffer_write_packed_int(cpu, COUNTER_BUF, core);
+		gator_buffer_write_packed_int(cpu, COUNTER_BUF, key);
+		gator_buffer_write_packed_int64(cpu, COUNTER_BUF, value);
+	}
+	local_irq_restore(flags);
+	/* Check and commit; commit is set to occur once buffer is 3/4 full */
+	buffer_check(cpu, COUNTER_BUF, time);
+}
+
+static void marshal_sched_trace_switch(int pid, int state)
+{
+	unsigned long cpu = get_physical_cpu(), flags;
+	u64 time;
+
+	if (!per_cpu(gator_buffer, cpu)[SCHED_TRACE_BUF])
+		return;
+
+	local_irq_save(flags);
+	time = gator_get_time();
+	if (buffer_check_space(cpu, SCHED_TRACE_BUF, MAXSIZE_PACK64 + 5 * MAXSIZE_PACK32)) {
+		gator_buffer_write_packed_int(cpu, SCHED_TRACE_BUF, MESSAGE_SCHED_SWITCH);
+		gator_buffer_write_packed_int64(cpu, SCHED_TRACE_BUF, time);
+		gator_buffer_write_packed_int(cpu, SCHED_TRACE_BUF, pid);
+		gator_buffer_write_packed_int(cpu, SCHED_TRACE_BUF, state);
+	}
+	local_irq_restore(flags);
+	/* Check and commit; commit is set to occur once buffer is 3/4 full */
+	buffer_check(cpu, SCHED_TRACE_BUF, time);
+}
+
+static void marshal_sched_trace_exit(int tgid, int pid)
+{
+	unsigned long cpu = get_physical_cpu(), flags;
+	u64 time;
+
+	if (!per_cpu(gator_buffer, cpu)[SCHED_TRACE_BUF])
+		return;
+
+	local_irq_save(flags);
+	time = gator_get_time();
+	if (buffer_check_space(cpu, SCHED_TRACE_BUF, MAXSIZE_PACK64 + 2 * MAXSIZE_PACK32)) {
+		gator_buffer_write_packed_int(cpu, SCHED_TRACE_BUF, MESSAGE_SCHED_EXIT);
+		gator_buffer_write_packed_int64(cpu, SCHED_TRACE_BUF, time);
+		gator_buffer_write_packed_int(cpu, SCHED_TRACE_BUF, pid);
+	}
+	local_irq_restore(flags);
+	/* Check and commit; commit is set to occur once buffer is 3/4 full */
+	buffer_check(cpu, SCHED_TRACE_BUF, time);
+}
+
+#if GATOR_CPU_FREQ_SUPPORT
+static void marshal_idle(int core, int state)
+{
+	unsigned long flags, cpu;
+	u64 time;
+
+	local_irq_save(flags);
+	cpu = get_physical_cpu();
+	time = gator_get_time();
+	if (buffer_check_space(cpu, IDLE_BUF, MAXSIZE_PACK64 + 2 * MAXSIZE_PACK32)) {
+		gator_buffer_write_packed_int(cpu, IDLE_BUF, state);
+		gator_buffer_write_packed_int64(cpu, IDLE_BUF, time);
+		gator_buffer_write_packed_int(cpu, IDLE_BUF, core);
+	}
+	local_irq_restore(flags);
+	/* Check and commit; commit is set to occur once buffer is 3/4 full */
+	buffer_check(cpu, IDLE_BUF, time);
+}
+#endif
+
+#if defined(__arm__) || defined(__aarch64__)
+static void marshal_core_name(const int core, const int cpuid, const char *name)
+{
+	int cpu = get_physical_cpu();
+	unsigned long flags;
+
+	local_irq_save(flags);
+	if (buffer_check_space(cpu, SUMMARY_BUF, MAXSIZE_PACK32 + MAXSIZE_CORE_NAME)) {
+		gator_buffer_write_packed_int(cpu, SUMMARY_BUF, MESSAGE_CORE_NAME);
+		gator_buffer_write_packed_int(cpu, SUMMARY_BUF, core);
+		gator_buffer_write_packed_int(cpu, SUMMARY_BUF, cpuid);
+		gator_buffer_write_string(cpu, SUMMARY_BUF, name);
+	}
+	/* Commit core names now so that they can show up in live */
+	local_irq_restore(flags);
+	gator_commit_buffer(cpu, SUMMARY_BUF, gator_get_time());
+}
+#endif
+
+static void marshal_activity_switch(int core, int key, int activity, int pid, int state)
+{
+	unsigned long cpu = get_physical_cpu(), flags;
+	u64 time;
+
+	if (!per_cpu(gator_buffer, cpu)[ACTIVITY_BUF])
+		return;
+
+	local_irq_save(flags);
+	time = gator_get_time();
+	if (buffer_check_space(cpu, ACTIVITY_BUF, MAXSIZE_PACK64 + 5 * MAXSIZE_PACK32)) {
+		gator_buffer_write_packed_int(cpu, ACTIVITY_BUF, MESSAGE_SWITCH);
+		gator_buffer_write_packed_int64(cpu, ACTIVITY_BUF, time);
+		gator_buffer_write_packed_int(cpu, ACTIVITY_BUF, core);
+		gator_buffer_write_packed_int(cpu, ACTIVITY_BUF, key);
+		gator_buffer_write_packed_int(cpu, ACTIVITY_BUF, activity);
+		gator_buffer_write_packed_int(cpu, ACTIVITY_BUF, pid);
+		gator_buffer_write_packed_int(cpu, ACTIVITY_BUF, state);
+	}
+	local_irq_restore(flags);
+	/* Check and commit; commit is set to occur once buffer is 3/4 full */
+	buffer_check(cpu, ACTIVITY_BUF, time);
+}
+
+void gator_marshal_activity_switch(int core, int key, int activity, int pid)
+{
+	/* state is reserved for cpu use only */
+	marshal_activity_switch(core, key, activity, pid, 0);
+}
diff --git a/drivers/gator/gator_trace_gpu.c b/drivers/gator/gator_trace_gpu.c
new file mode 100644
index 000000000000..5de9152e365a
--- /dev/null
+++ b/drivers/gator/gator_trace_gpu.c
@@ -0,0 +1,321 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "gator.h"
+
+#include <linux/slab.h>
+#include <linux/module.h>
+#include <linux/time.h>
+#include <linux/math64.h>
+
+#ifdef MALI_SUPPORT
+#ifdef MALI_DIR_MIDGARD
+/* New DDK Directory structure with kernel/drivers/gpu/arm/midgard*/
+#include "mali_linux_trace.h"
+#else
+/* Old DDK Directory structure with kernel/drivers/gpu/arm/t6xx*/
+#include "linux/mali_linux_trace.h"
+#endif
+#endif
+
+/*
+ * Taken from MALI_PROFILING_EVENT_TYPE_* items in Mali DDK.
+ */
+#define EVENT_TYPE_SINGLE  0
+#define EVENT_TYPE_START   1
+#define EVENT_TYPE_STOP    2
+#define EVENT_TYPE_SUSPEND 3
+#define EVENT_TYPE_RESUME  4
+
+/* Note whether tracepoints have been registered */
+static int mali_timeline_trace_registered;
+static int mali_job_slots_trace_registered;
+
+enum {
+	GPU_UNIT_NONE = 0,
+	GPU_UNIT_VP,
+	GPU_UNIT_FP,
+	GPU_UNIT_CL,
+	NUMBER_OF_GPU_UNITS
+};
+
+#if defined(MALI_SUPPORT)
+
+struct mali_activity {
+	int core;
+	int key;
+	int count;
+	int last_activity;
+	int last_pid;
+};
+
+#define NUMBER_OF_GPU_CORES 16
+static struct mali_activity mali_activities[NUMBER_OF_GPU_UNITS*NUMBER_OF_GPU_CORES];
+static DEFINE_SPINLOCK(mali_activities_lock);
+
+/* Only one event should be running on a unit and core at a time (ie,
+ * a start event can only be followed by a stop and vice versa), but
+ * because the kernel only knows when a job is enqueued and not
+ * started, it is possible for a start1, start2, stop1, stop2. Change
+ * it back into start1, stop1, start2, stop2 by queueing up start2 and
+ * releasing it when stop1 is received.
+ */
+
+static int mali_activity_index(int core, int key)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(mali_activities); ++i) {
+		if ((mali_activities[i].core == core) && (mali_activities[i].key == key))
+			break;
+		if ((mali_activities[i].core == 0) && (mali_activities[i].key == 0)) {
+			mali_activities[i].core = core;
+			mali_activities[i].key = key;
+			break;
+		}
+	}
+	BUG_ON(i >= ARRAY_SIZE(mali_activities));
+
+	return i;
+}
+
+static void mali_activity_enqueue(int core, int key, int activity, int pid)
+{
+	int i;
+	int count;
+
+	spin_lock(&mali_activities_lock);
+	i = mali_activity_index(core, key);
+
+	count = mali_activities[i].count;
+	BUG_ON(count < 0);
+	++mali_activities[i].count;
+	if (count) {
+		mali_activities[i].last_activity = activity;
+		mali_activities[i].last_pid = pid;
+	}
+	spin_unlock(&mali_activities_lock);
+
+	if (!count)
+		gator_marshal_activity_switch(core, key, activity, pid);
+}
+
+static void mali_activity_stop(int core, int key)
+{
+	int i;
+	int count;
+	int last_activity = 0;
+	int last_pid = 0;
+
+	spin_lock(&mali_activities_lock);
+	i = mali_activity_index(core, key);
+
+	if (mali_activities[i].count == 0) {
+		spin_unlock(&mali_activities_lock);
+		return;
+	}
+	--mali_activities[i].count;
+	count = mali_activities[i].count;
+	if (count) {
+		last_activity = mali_activities[i].last_activity;
+		last_pid = mali_activities[i].last_pid;
+	}
+	spin_unlock(&mali_activities_lock);
+
+	gator_marshal_activity_switch(core, key, 0, 0);
+	if (count)
+		gator_marshal_activity_switch(core, key, last_activity, last_pid);
+}
+
+void mali_activity_clear(struct mali_counter mali_activity[], size_t mali_activity_size)
+{
+	int activity;
+	int cores;
+	int core;
+
+	for (activity = 0; activity < mali_activity_size; ++activity) {
+		cores = mali_activity[activity].cores;
+		if (cores < 0)
+			cores = 1;
+		for (core = 0; core < cores; ++core) {
+			if (mali_activity[activity].enabled) {
+				preempt_disable();
+				gator_marshal_activity_switch(core, mali_activity[activity].key, 0, 0);
+				preempt_enable();
+			}
+		}
+	}
+}
+
+#endif
+
+#if defined(MALI_SUPPORT) && (MALI_SUPPORT != MALI_MIDGARD)
+#include "gator_events_mali_4xx.h"
+
+/*
+ * Taken from MALI_PROFILING_EVENT_CHANNEL_* in Mali DDK.
+ */
+enum {
+	EVENT_CHANNEL_SOFTWARE = 0,
+	EVENT_CHANNEL_VP0 = 1,
+	EVENT_CHANNEL_FP0 = 5,
+	EVENT_CHANNEL_FP1,
+	EVENT_CHANNEL_FP2,
+	EVENT_CHANNEL_FP3,
+	EVENT_CHANNEL_FP4,
+	EVENT_CHANNEL_FP5,
+	EVENT_CHANNEL_FP6,
+	EVENT_CHANNEL_FP7,
+	EVENT_CHANNEL_GPU = 21
+};
+
+/**
+ * These events are applicable when the type MALI_PROFILING_EVENT_TYPE_SINGLE is used from the GPU channel
+ */
+enum {
+	EVENT_REASON_SINGLE_GPU_NONE = 0,
+	EVENT_REASON_SINGLE_GPU_FREQ_VOLT_CHANGE = 1,
+};
+
+struct mali_counter mali_activity[2];
+
+GATOR_DEFINE_PROBE(mali_timeline_event, TP_PROTO(unsigned int event_id, unsigned int d0, unsigned int d1, unsigned int d2, unsigned int d3, unsigned int d4))
+{
+	unsigned int component, state;
+
+	/* do as much work as possible before disabling interrupts */
+	component = (event_id >> 16) & 0xFF;	/* component is an 8-bit field */
+	state = (event_id >> 24) & 0xF;	/* state is a 4-bit field */
+
+	switch (state) {
+	case EVENT_TYPE_START:
+		if (component == EVENT_CHANNEL_VP0) {
+			/* tgid = d0; pid = d1; */
+			if (mali_activity[1].enabled)
+				mali_activity_enqueue(0, mali_activity[1].key, 1, d1);
+		} else if (component >= EVENT_CHANNEL_FP0 && component <= EVENT_CHANNEL_FP7) {
+			/* tgid = d0; pid = d1; */
+			if (mali_activity[0].enabled)
+				mali_activity_enqueue(component - EVENT_CHANNEL_FP0, mali_activity[0].key, 1, d1);
+		}
+		break;
+
+	case EVENT_TYPE_STOP:
+		if (component == EVENT_CHANNEL_VP0) {
+			if (mali_activity[1].enabled)
+				mali_activity_stop(0, mali_activity[1].key);
+		} else if (component >= EVENT_CHANNEL_FP0 && component <= EVENT_CHANNEL_FP7) {
+			if (mali_activity[0].enabled)
+				mali_activity_stop(component - EVENT_CHANNEL_FP0, mali_activity[0].key);
+		}
+		break;
+
+	case EVENT_TYPE_SINGLE:
+		if (component == EVENT_CHANNEL_GPU) {
+			unsigned int reason = (event_id & 0xffff);
+
+			if (reason == EVENT_REASON_SINGLE_GPU_FREQ_VOLT_CHANGE)
+				gator_events_mali_log_dvfs_event(d0, d1);
+		}
+		break;
+
+	default:
+		break;
+	}
+}
+#endif
+
+#if defined(MALI_SUPPORT) && (MALI_SUPPORT == MALI_MIDGARD)
+
+struct mali_counter mali_activity[3];
+
+#if defined(MALI_JOB_SLOTS_EVENT_CHANGED)
+GATOR_DEFINE_PROBE(mali_job_slots_event, TP_PROTO(unsigned int event_id, unsigned int tgid, unsigned int pid, unsigned char job_id))
+#else
+GATOR_DEFINE_PROBE(mali_job_slots_event, TP_PROTO(unsigned int event_id, unsigned int tgid, unsigned int pid))
+#endif
+{
+	unsigned int component, state, unit;
+#if !defined(MALI_JOB_SLOTS_EVENT_CHANGED)
+	unsigned char job_id = 0;
+#endif
+
+	component = (event_id >> 16) & 0xFF;	/* component is an 8-bit field */
+	state = (event_id >> 24) & 0xF;	/* state is a 4-bit field */
+
+	switch (component) {
+	case 0:
+		unit = GPU_UNIT_FP;
+		break;
+	case 1:
+		unit = GPU_UNIT_VP;
+		break;
+	case 2:
+		unit = GPU_UNIT_CL;
+		break;
+	default:
+		unit = GPU_UNIT_NONE;
+	}
+
+	if (unit != GPU_UNIT_NONE) {
+		switch (state) {
+		case EVENT_TYPE_START:
+			if (mali_activity[component].enabled)
+				mali_activity_enqueue(0, mali_activity[component].key, 1, (pid != 0 ? pid : tgid));
+			break;
+		case EVENT_TYPE_STOP:
+		default: /* Some jobs can be soft-stopped, so ensure that this terminates the activity trace. */
+			if (mali_activity[component].enabled)
+				mali_activity_stop(0, mali_activity[component].key);
+			break;
+		}
+	}
+}
+#endif
+
+static int gator_trace_gpu_start(void)
+{
+	/*
+	 * Returns nonzero for installation failed
+	 * Absence of gpu trace points is not an error
+	 */
+
+#if defined(MALI_SUPPORT)
+	memset(&mali_activities, 0, sizeof(mali_activities));
+#endif
+	mali_timeline_trace_registered = mali_job_slots_trace_registered = 0;
+
+#if defined(MALI_SUPPORT) && (MALI_SUPPORT != MALI_MIDGARD)
+	mali_activity_clear(mali_activity, ARRAY_SIZE(mali_activity));
+	if (!GATOR_REGISTER_TRACE(mali_timeline_event))
+		mali_timeline_trace_registered = 1;
+#endif
+
+#if defined(MALI_SUPPORT) && (MALI_SUPPORT == MALI_MIDGARD)
+	mali_activity_clear(mali_activity, ARRAY_SIZE(mali_activity));
+	if (!GATOR_REGISTER_TRACE(mali_job_slots_event))
+		mali_job_slots_trace_registered = 1;
+#endif
+
+	return 0;
+}
+
+static void gator_trace_gpu_stop(void)
+{
+#if defined(MALI_SUPPORT) && (MALI_SUPPORT != MALI_MIDGARD)
+	if (mali_timeline_trace_registered)
+		GATOR_UNREGISTER_TRACE(mali_timeline_event);
+#endif
+
+#if defined(MALI_SUPPORT) && (MALI_SUPPORT == MALI_MIDGARD)
+	if (mali_job_slots_trace_registered)
+		GATOR_UNREGISTER_TRACE(mali_job_slots_event);
+#endif
+
+	mali_timeline_trace_registered = mali_job_slots_trace_registered = 0;
+}
diff --git a/drivers/gator/gator_trace_power.c b/drivers/gator/gator_trace_power.c
new file mode 100644
index 000000000000..46e04b29a187
--- /dev/null
+++ b/drivers/gator/gator_trace_power.c
@@ -0,0 +1,192 @@
+/**
+ * Copyright (C) ARM Limited 2011-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#include <linux/cpufreq.h>
+#include <trace/events/power.h>
+
+#if defined(__arm__)
+
+#include <asm/mach-types.h>
+
+#define implements_wfi() (!machine_is_omap3_beagle())
+
+#else
+
+#define implements_wfi() false
+
+#endif
+
+/* cpu_frequency and cpu_idle trace points were introduced in Linux
+ * kernel v2.6.38 the now deprecated power_frequency trace point was
+ * available prior to 2.6.38, but only for x86
+ */
+#if GATOR_CPU_FREQ_SUPPORT
+enum {
+	POWER_CPU_FREQ,
+	POWER_TOTAL
+};
+
+static DEFINE_PER_CPU(ulong, idle_prev_state);
+static ulong power_cpu_enabled[POWER_TOTAL];
+static ulong power_cpu_key[POWER_TOTAL];
+static ulong power_cpu_cores;
+
+static int gator_trace_power_create_files(struct super_block *sb, struct dentry *root)
+{
+	struct dentry *dir;
+	int cpu;
+	bool found_nonzero_freq = false;
+
+	/* Even if CONFIG_CPU_FREQ is defined, it still may not be
+	 * used. Check for non-zero values from cpufreq_quick_get
+	 */
+	for_each_online_cpu(cpu) {
+		if (cpufreq_quick_get(cpu) > 0) {
+			found_nonzero_freq = true;
+			break;
+		}
+	}
+
+	if (found_nonzero_freq) {
+		/* cpu_frequency */
+		dir = gatorfs_mkdir(sb, root, "Linux_power_cpu_freq");
+		if (!dir)
+			return -1;
+		gatorfs_create_ulong(sb, dir, "enabled", &power_cpu_enabled[POWER_CPU_FREQ]);
+		gatorfs_create_ro_ulong(sb, dir, "key", &power_cpu_key[POWER_CPU_FREQ]);
+	}
+
+	return 0;
+}
+
+/* 'cpu' may not equal smp_processor_id(), i.e. may not be running on the core that is having the freq/idle state change */
+GATOR_DEFINE_PROBE(cpu_frequency, TP_PROTO(unsigned int frequency, unsigned int cpu))
+{
+	cpu = lcpu_to_pcpu(cpu);
+	marshal_event_single64(cpu, power_cpu_key[POWER_CPU_FREQ], frequency * 1000L);
+}
+
+GATOR_DEFINE_PROBE(cpu_idle, TP_PROTO(unsigned int state, unsigned int cpu))
+{
+	cpu = lcpu_to_pcpu(cpu);
+
+	if (state == per_cpu(idle_prev_state, cpu))
+		return;
+
+	if (implements_wfi()) {
+		if (state == PWR_EVENT_EXIT) {
+			/* transition from wfi to non-wfi */
+			marshal_idle(cpu, MESSAGE_IDLE_EXIT);
+		} else {
+			/* transition from non-wfi to wfi */
+			marshal_idle(cpu, MESSAGE_IDLE_ENTER);
+		}
+	}
+
+	per_cpu(idle_prev_state, cpu) = state;
+}
+
+static void gator_trace_power_online(void)
+{
+	int pcpu = get_physical_cpu();
+	int lcpu = get_logical_cpu();
+
+	if (power_cpu_enabled[POWER_CPU_FREQ])
+		marshal_event_single64(pcpu, power_cpu_key[POWER_CPU_FREQ], cpufreq_quick_get(lcpu) * 1000L);
+}
+
+static void gator_trace_power_offline(void)
+{
+	/* Set frequency to zero on an offline */
+	int cpu = get_physical_cpu();
+
+	if (power_cpu_enabled[POWER_CPU_FREQ])
+		marshal_event_single(cpu, power_cpu_key[POWER_CPU_FREQ], 0);
+}
+
+static int gator_trace_power_start(void)
+{
+	int cpu;
+
+	/* register tracepoints */
+	if (power_cpu_enabled[POWER_CPU_FREQ])
+		if (GATOR_REGISTER_TRACE(cpu_frequency))
+			goto fail_cpu_frequency_exit;
+
+	/* Always register for cpu_idle for detecting WFI */
+	if (GATOR_REGISTER_TRACE(cpu_idle))
+		goto fail_cpu_idle_exit;
+	pr_debug("gator: registered power event tracepoints\n");
+
+	for_each_present_cpu(cpu) {
+		per_cpu(idle_prev_state, cpu) = 0;
+	}
+
+	return 0;
+
+	/* unregister tracepoints on error */
+fail_cpu_idle_exit:
+	if (power_cpu_enabled[POWER_CPU_FREQ])
+		GATOR_UNREGISTER_TRACE(cpu_frequency);
+fail_cpu_frequency_exit:
+	pr_err("gator: power event tracepoints failed to activate, please verify that tracepoints are enabled in the linux kernel\n");
+
+	return -1;
+}
+
+static void gator_trace_power_stop(void)
+{
+	int i;
+
+	if (power_cpu_enabled[POWER_CPU_FREQ])
+		GATOR_UNREGISTER_TRACE(cpu_frequency);
+	GATOR_UNREGISTER_TRACE(cpu_idle);
+	pr_debug("gator: unregistered power event tracepoints\n");
+
+	for (i = 0; i < POWER_TOTAL; i++)
+		power_cpu_enabled[i] = 0;
+}
+
+static void gator_trace_power_init(void)
+{
+	int i;
+
+	power_cpu_cores = nr_cpu_ids;
+	for (i = 0; i < POWER_TOTAL; i++) {
+		power_cpu_enabled[i] = 0;
+		power_cpu_key[i] = gator_events_get_key();
+	}
+}
+#else
+static int gator_trace_power_create_files(struct super_block *sb, struct dentry *root)
+{
+	return 0;
+}
+
+static void gator_trace_power_online(void)
+{
+}
+
+static void gator_trace_power_offline(void)
+{
+}
+
+static int gator_trace_power_start(void)
+{
+	return 0;
+}
+
+static void gator_trace_power_stop(void)
+{
+}
+
+static void gator_trace_power_init(void)
+{
+}
+#endif
diff --git a/drivers/gator/gator_trace_sched.c b/drivers/gator/gator_trace_sched.c
new file mode 100644
index 000000000000..6d7cbd7348e1
--- /dev/null
+++ b/drivers/gator/gator_trace_sched.c
@@ -0,0 +1,321 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#include <trace/events/sched.h>
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
+#include <trace/events/task.h>
+#endif
+
+#include "gator.h"
+
+#define TASK_MAP_ENTRIES		1024	/* must be power of 2 */
+#define TASK_MAX_COLLISIONS		2
+
+enum {
+	STATE_WAIT_ON_OTHER = 0,
+	STATE_CONTENTION,
+	STATE_WAIT_ON_IO,
+	CPU_WAIT_TOTAL
+};
+
+static DEFINE_PER_CPU(uint64_t *, taskname_keys);
+static DEFINE_PER_CPU(int, collecting);
+
+/* this array is never read as the cpu wait charts are derived
+ * counters the files are needed, nonetheless, to show that these
+ * counters are available
+ */
+static ulong cpu_wait_enabled[CPU_WAIT_TOTAL];
+static ulong sched_cpu_key[CPU_WAIT_TOTAL];
+
+static int sched_trace_create_files(struct super_block *sb, struct dentry *root)
+{
+	struct dentry *dir;
+
+	/* CPU Wait - Contention */
+	dir = gatorfs_mkdir(sb, root, "Linux_cpu_wait_contention");
+	if (!dir)
+		return -1;
+	gatorfs_create_ulong(sb, dir, "enabled", &cpu_wait_enabled[STATE_CONTENTION]);
+	gatorfs_create_ro_ulong(sb, dir, "key", &sched_cpu_key[STATE_CONTENTION]);
+
+	/* CPU Wait - I/O */
+	dir = gatorfs_mkdir(sb, root, "Linux_cpu_wait_io");
+	if (!dir)
+		return -1;
+	gatorfs_create_ulong(sb, dir, "enabled", &cpu_wait_enabled[STATE_WAIT_ON_IO]);
+	gatorfs_create_ro_ulong(sb, dir, "key", &sched_cpu_key[STATE_WAIT_ON_IO]);
+
+	return 0;
+}
+
+static void emit_pid_name(const char *comm, struct task_struct *task)
+{
+	bool found = false;
+	char taskcomm[TASK_COMM_LEN + 3];
+	unsigned long x, cpu = get_physical_cpu();
+	uint64_t *keys = &(per_cpu(taskname_keys, cpu)[(task->pid & 0xFF) * TASK_MAX_COLLISIONS]);
+	uint64_t value;
+
+	value = gator_chksum_crc32(comm);
+	value = (value << 32) | (uint32_t)task->pid;
+
+	/* determine if the thread name was emitted already */
+	for (x = 0; x < TASK_MAX_COLLISIONS; x++) {
+		if (keys[x] == value) {
+			found = true;
+			break;
+		}
+	}
+
+	if (!found) {
+		/* shift values, new value always in front */
+		uint64_t oldv, newv = value;
+
+		for (x = 0; x < TASK_MAX_COLLISIONS; x++) {
+			oldv = keys[x];
+			keys[x] = newv;
+			newv = oldv;
+		}
+
+		/* emit pid names, cannot use get_task_comm, as it's not exported on all kernel versions */
+		if (strlcpy(taskcomm, comm, TASK_COMM_LEN) == TASK_COMM_LEN - 1) {
+			/* append ellipses if comm has length of TASK_COMM_LEN - 1 */
+			strcat(taskcomm, "...");
+		}
+
+		marshal_thread_name(task->pid, taskcomm);
+	}
+}
+
+static void collect_counters(u64 time, struct task_struct *task, bool sched_switch)
+{
+	int *buffer, len, cpu = get_physical_cpu();
+	long long *buffer64;
+	struct gator_interface *gi;
+
+	if (marshal_event_header(time)) {
+		list_for_each_entry(gi, &gator_events, list) {
+			if (gi->read) {
+				len = gi->read(&buffer, sched_switch);
+				marshal_event(len, buffer);
+			} else if (gi->read64) {
+				len = gi->read64(&buffer64);
+				marshal_event64(len, buffer64);
+			}
+			if (gi->read_proc && task != NULL) {
+				len = gi->read_proc(&buffer64, task);
+				marshal_event64(len, buffer64);
+			}
+		}
+		if (cpu == 0)
+			gator_emit_perf_time(time);
+		/* Only check after writing all counters so that time and corresponding counters appear in the same frame */
+		buffer_check(cpu, BLOCK_COUNTER_BUF, time);
+
+		/* Commit buffers on timeout */
+		if (gator_live_rate > 0 && time >= per_cpu(gator_buffer_commit_time, cpu)) {
+			static const int buftypes[] = { NAME_BUF, COUNTER_BUF, BLOCK_COUNTER_BUF, SCHED_TRACE_BUF, ACTIVITY_BUF };
+			int i;
+
+			for (i = 0; i < ARRAY_SIZE(buftypes); ++i)
+				gator_commit_buffer(cpu, buftypes[i], time);
+
+			/* spinlocks are noops on uniprocessor machines and mutexes do
+			 * not work in sched_switch context in RT-Preempt full, so
+			 * disable proactive flushing of the annotate frame on
+			 * uniprocessor machines.
+			 */
+#ifdef CONFIG_SMP
+			/* Try to preemptively flush the annotate buffer to reduce the chance of the buffer being full */
+			if (on_primary_core() && spin_trylock(&annotate_lock)) {
+				gator_commit_buffer(0, ANNOTATE_BUF, time);
+				spin_unlock(&annotate_lock);
+			}
+#endif
+		}
+	}
+}
+
+/* special case used during a suspend of the system */
+static void trace_sched_insert_idle(void)
+{
+	marshal_sched_trace_switch(0, 0);
+}
+
+static void gator_trace_emit_link(struct task_struct *p)
+{
+	int cookie;
+	int cpu = get_physical_cpu();
+
+	cookie = get_exec_cookie(cpu, p);
+	emit_pid_name(p->comm, p);
+
+	marshal_link(cookie, p->tgid, p->pid);
+}
+
+GATOR_DEFINE_PROBE(sched_process_fork, TP_PROTO(struct task_struct *parent, struct task_struct *child))
+{
+	gator_trace_emit_link(child);
+}
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
+GATOR_DEFINE_PROBE(sched_process_exec, TP_PROTO(struct task_struct *p, pid_t old_pid, struct linux_binprm *bprm))
+{
+	gator_trace_emit_link(p);
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 15, 0)
+GATOR_DEFINE_PROBE(task_rename, TP_PROTO(struct task_struct *task, char *comm))
+#else
+GATOR_DEFINE_PROBE(task_rename, TP_PROTO(struct task_struct *task, const char *comm))
+#endif
+{
+	emit_pid_name(comm, task);
+}
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 35)
+GATOR_DEFINE_PROBE(sched_switch, TP_PROTO(struct rq *rq, struct task_struct *prev, struct task_struct *next))
+#else
+GATOR_DEFINE_PROBE(sched_switch, TP_PROTO(struct task_struct *prev, struct task_struct *next))
+#endif
+{
+	int state;
+	int cpu = get_physical_cpu();
+
+	per_cpu(in_scheduler_context, cpu) = true;
+
+	/* do as much work as possible before disabling interrupts */
+	if (prev->state == TASK_RUNNING)
+		state = STATE_CONTENTION;
+	else if (prev->in_iowait)
+		state = STATE_WAIT_ON_IO;
+	else
+		state = STATE_WAIT_ON_OTHER;
+
+	per_cpu(collecting, cpu) = 1;
+	collect_counters(gator_get_time(), prev, true);
+	per_cpu(collecting, cpu) = 0;
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 4, 0)
+	gator_trace_emit_link(next);
+#endif
+	marshal_sched_trace_switch(next->pid, state);
+
+	per_cpu(in_scheduler_context, cpu) = false;
+}
+
+GATOR_DEFINE_PROBE(sched_process_free, TP_PROTO(struct task_struct *p))
+{
+	marshal_sched_trace_exit(p->tgid, p->pid);
+}
+
+static void do_nothing(void *info)
+{
+	/* Intentionally do nothing */
+	(void)info;
+}
+
+static int register_scheduler_tracepoints(void)
+{
+	/* register tracepoints */
+	if (GATOR_REGISTER_TRACE(sched_process_fork))
+		goto fail_sched_process_fork;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
+	if (GATOR_REGISTER_TRACE(sched_process_exec))
+		goto fail_sched_process_exec;
+	if (GATOR_REGISTER_TRACE(task_rename))
+		goto fail_task_rename;
+#endif
+	if (GATOR_REGISTER_TRACE(sched_switch))
+		goto fail_sched_switch;
+	if (GATOR_REGISTER_TRACE(sched_process_free))
+		goto fail_sched_process_free;
+	pr_debug("gator: registered tracepoints\n");
+
+	/* Now that the scheduler tracepoint is registered, force a context
+	 * switch on all cpus to capture what is currently running.
+	 */
+	on_each_cpu(do_nothing, NULL, 0);
+
+	return 0;
+
+	/* unregister tracepoints on error */
+fail_sched_process_free:
+	GATOR_UNREGISTER_TRACE(sched_switch);
+fail_sched_switch:
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
+	GATOR_UNREGISTER_TRACE(task_rename);
+fail_task_rename:
+	GATOR_UNREGISTER_TRACE(sched_process_exec);
+fail_sched_process_exec:
+#endif
+	GATOR_UNREGISTER_TRACE(sched_process_fork);
+fail_sched_process_fork:
+	pr_err("gator: tracepoints failed to activate, please verify that tracepoints are enabled in the linux kernel\n");
+
+	return -1;
+}
+
+static void unregister_scheduler_tracepoints(void)
+{
+	GATOR_UNREGISTER_TRACE(sched_process_fork);
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
+	GATOR_UNREGISTER_TRACE(sched_process_exec);
+	GATOR_UNREGISTER_TRACE(task_rename);
+#endif
+	GATOR_UNREGISTER_TRACE(sched_switch);
+	GATOR_UNREGISTER_TRACE(sched_process_free);
+	pr_debug("gator: unregistered tracepoints\n");
+}
+
+static void gator_trace_sched_stop(void)
+{
+	int cpu;
+
+	unregister_scheduler_tracepoints();
+
+	for_each_present_cpu(cpu) {
+		kfree(per_cpu(taskname_keys, cpu));
+	}
+}
+
+static int gator_trace_sched_start(void)
+{
+	int cpu, size;
+	int ret;
+
+	for_each_present_cpu(cpu) {
+		size = TASK_MAP_ENTRIES * TASK_MAX_COLLISIONS * sizeof(uint64_t);
+		per_cpu(taskname_keys, cpu) = kmalloc(size, GFP_KERNEL);
+		if (!per_cpu(taskname_keys, cpu))
+			return -1;
+		memset(per_cpu(taskname_keys, cpu), 0, size);
+	}
+
+	ret = register_scheduler_tracepoints();
+
+	return ret;
+}
+
+static void gator_trace_sched_offline(void)
+{
+	trace_sched_insert_idle();
+}
+
+static void gator_trace_sched_init(void)
+{
+	int i;
+
+	for (i = 0; i < CPU_WAIT_TOTAL; i++) {
+		cpu_wait_enabled[i] = 0;
+		sched_cpu_key[i] = gator_events_get_key();
+	}
+}
diff --git a/drivers/gator/mali/mali_kbase_gator_api.h b/drivers/gator/mali/mali_kbase_gator_api.h
new file mode 100644
index 000000000000..5ed069797e36
--- /dev/null
+++ b/drivers/gator/mali/mali_kbase_gator_api.h
@@ -0,0 +1,219 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#ifndef _KBASE_GATOR_API_H_
+#define _KBASE_GATOR_API_H_
+
+/**
+ * @brief This file describes the API used by Gator to collect hardware counters data from a Mali device.
+ */
+
+/* This define is used by the gator kernel module compile to select which DDK
+ * API calling convention to use. If not defined (legacy DDK) gator assumes
+ * version 1. The version to DDK release mapping is:
+ *     Version 1 API: DDK versions r1px, r2px
+ *     Version 2 API: DDK versions r3px, r4px
+ *     Version 3 API: DDK version r5p0 and newer
+ *
+ * API Usage
+ * =========
+ *
+ * 1] Call kbase_gator_hwcnt_init_names() to return the list of short counter
+ * names for the GPU present in this device.
+ *
+ * 2] Create a kbase_gator_hwcnt_info structure and set the counter enables for
+ * the counters you want enabled. The enables can all be set for simplicity in
+ * most use cases, but disabling some will let you minimize bandwidth impact.
+ *
+ * 3] Call kbase_gator_hwcnt_init() using the above structure, to create a
+ * counter context. On successful return the DDK will have populated the
+ * structure with a variety of useful information.
+ *
+ * 4] Call kbase_gator_hwcnt_dump_irq() to queue a non-blocking request for a
+ * counter dump. If this returns a non-zero value the request has been queued,
+ * otherwise the driver has been unable to do so (typically because of another
+ * user of the instrumentation exists concurrently).
+ *
+ * 5] Call kbase_gator_hwcnt_dump_complete() to test whether the  previously
+ * requested dump has been succesful. If this returns non-zero the counter dump
+ * has resolved, but the value of *success must also be tested as the dump
+ * may have not been successful. If it returns zero the counter dump was
+ * abandoned due to the device being busy (typically because of another
+ * user of the instrumentation exists concurrently).
+ *
+ * 6] Process the counters stored in the buffer pointed to by ...
+ *
+ *        kbase_gator_hwcnt_info->kernel_dump_buffer
+ *
+ *    In pseudo code you can find all of the counters via this approach:
+ *
+ *
+ *        hwcnt_info # pointer to kbase_gator_hwcnt_info structure
+ *        hwcnt_name # pointer to name list
+ *
+ *        u32 * hwcnt_data = (u32*)hwcnt_info->kernel_dump_buffer
+ *
+ *        # Iterate over each 64-counter block in this GPU configuration
+ *        for( i = 0; i < hwcnt_info->nr_hwc_blocks; i++) {
+ *            hwc_type type = hwcnt_info->hwc_layout[i];
+ *
+ *            # Skip reserved type blocks - they contain no counters at all
+ *            if( type == RESERVED_BLOCK ) {
+ *                continue;
+ *            }
+ *
+ *            size_t name_offset = type * 64;
+ *            size_t data_offset = i * 64;
+ *
+ *            # Iterate over the names of the counters in this block type
+ *            for( j = 0; j < 64; j++) {
+ *                const char * name = hwcnt_name[name_offset+j];
+ *
+ *                # Skip empty name strings - there is no counter here
+ *                if( name[0] == '\0' ) {
+ *                    continue;
+ *                }
+ *
+ *                u32 data = hwcnt_data[data_offset+j];
+ *
+ *                printk( "COUNTER: %s DATA: %u\n", name, data );
+ *            }
+ *        }
+ *
+ *
+ *     Note that in most implementations you typically want to either SUM or
+ *     AVERAGE multiple instances of the same counter if, for example, you have
+ *     multiple shader cores or multiple L2 caches. The most sensible view for
+ *     analysis is to AVERAGE shader core counters, but SUM L2 cache and MMU
+ *     counters.
+ *
+ * 7] Goto 4, repeating until you want to stop collecting counters.
+ *
+ * 8] Release the dump resources by calling kbase_gator_hwcnt_term().
+ *
+ * 9] Release the name table resources by calling kbase_gator_hwcnt_term_names().
+ *    This function must only be called if init_names() returned a non-NULL value.
+ **/
+
+#define MALI_DDK_GATOR_API_VERSION 3
+
+#if !defined(MALI_TRUE)
+	#define MALI_TRUE                ((uint32_t)1)
+#endif
+
+#if !defined(MALI_FALSE)
+	#define MALI_FALSE               ((uint32_t)0)
+#endif
+
+enum hwc_type {
+	JM_BLOCK = 0,
+	TILER_BLOCK,
+	SHADER_BLOCK,
+	MMU_L2_BLOCK,
+	RESERVED_BLOCK
+};
+
+struct kbase_gator_hwcnt_info {
+
+	/* Passed from Gator to kbase */
+
+	/* the bitmask of enabled hardware counters for each counter block */
+	uint16_t bitmask[4];
+
+	/* Passed from kbase to Gator */
+
+	/* ptr to counter dump memory */
+	void *kernel_dump_buffer;
+
+	/* size of counter dump memory */
+	uint32_t size;
+
+	/* the ID of the Mali device */
+	uint32_t gpu_id;
+
+	/* the number of shader cores in the GPU */
+	uint32_t nr_cores;
+
+	/* the number of core groups */
+	uint32_t nr_core_groups;
+
+	/* the memory layout of the performance counters */
+	enum hwc_type *hwc_layout;
+
+	/* the total number of hardware couter blocks */
+	uint32_t nr_hwc_blocks;
+};
+
+/**
+ * @brief Opaque block of Mali data which Gator needs to return to the API later.
+ */
+struct kbase_gator_hwcnt_handles;
+
+/**
+ * @brief Initialize the resources Gator needs for performance profiling.
+ *
+ * @param in_out_info   A pointer to a structure containing the enabled counters passed from Gator and all the Mali
+ *                      specific information that will be returned to Gator. On entry Gator must have populated the
+ *                      'bitmask' field with the counters it wishes to enable for each class of counter block.
+ *                      Each entry in the array corresponds to a single counter class based on the "hwc_type"
+ *                      enumeration, and each bit corresponds to an enable for 4 sequential counters (LSB enables
+ *                      the first 4 counters in the block, and so on). See the GPU counter array as returned by
+ *                      kbase_gator_hwcnt_get_names() for the index values of each counter for the curernt GPU.
+ *
+ * @return              Pointer to an opaque handle block on success, NULL on error.
+ */
+extern struct kbase_gator_hwcnt_handles *kbase_gator_hwcnt_init(struct kbase_gator_hwcnt_info *in_out_info);
+
+/**
+ * @brief Free all resources once Gator has finished using performance counters.
+ *
+ * @param in_out_info       A pointer to a structure containing the enabled counters passed from Gator and all the
+ *                          Mali specific information that will be returned to Gator.
+ * @param opaque_handles    A wrapper structure for kbase structures.
+ */
+extern void kbase_gator_hwcnt_term(struct kbase_gator_hwcnt_info *in_out_info, struct kbase_gator_hwcnt_handles *opaque_handles);
+
+/**
+ * @brief Poll whether a counter dump is successful.
+ *
+ * @param opaque_handles    A wrapper structure for kbase structures.
+ * @param[out] success      Non-zero on success, zero on failure.
+ *
+ * @return                  Zero if the dump is still pending, non-zero if the dump has completed. Note that a
+ *                          completed dump may not have dumped succesfully, so the caller must test for both
+ *                          a completed and successful dump before processing counters.
+ */
+extern uint32_t kbase_gator_instr_hwcnt_dump_complete(struct kbase_gator_hwcnt_handles *opaque_handles, uint32_t * const success);
+
+/**
+ * @brief Request the generation of a new counter dump.
+ *
+ * @param opaque_handles    A wrapper structure for kbase structures.
+ *
+ * @return                  Zero if the hardware device is busy and cannot handle the request, non-zero otherwise.
+ */
+extern uint32_t kbase_gator_instr_hwcnt_dump_irq(struct kbase_gator_hwcnt_handles *opaque_handles);
+
+/**
+ * @brief This function is used to fetch the names table based on the Mali device in use.
+ *
+ * @param[out] total_number_of_counters The total number of counters short names in the Mali devices' list.
+ *
+ * @return                              Pointer to an array of strings of length *total_number_of_counters.
+ */
+extern const char * const *kbase_gator_hwcnt_init_names(uint32_t *total_number_of_counters);
+
+/**
+ * @brief This function is used to terminate the use of the names table.
+ *
+ * This function must only be called if the initial call to kbase_gator_hwcnt_init_names returned a non-NULL value.
+ */
+extern void kbase_gator_hwcnt_term_names(void);
+
+#endif
diff --git a/drivers/gator/mali/mali_mjollnir_profiling_gator_api.h b/drivers/gator/mali/mali_mjollnir_profiling_gator_api.h
new file mode 100644
index 000000000000..2bc0b037eee6
--- /dev/null
+++ b/drivers/gator/mali/mali_mjollnir_profiling_gator_api.h
@@ -0,0 +1,159 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#ifndef __MALI_MJOLLNIR_PROFILING_GATOR_API_H__
+#define __MALI_MJOLLNIR_PROFILING_GATOR_API_H__
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+
+
+/*
+ * The number of processor cores.  Update to suit your hardware implementation.
+ */
+#define MAX_NUM_FP_CORES            (4)
+#define MAX_NUM_VP_CORES            (1)
+#define MAX_NUM_L2_CACHE_CORES      (1)
+
+enum counters {
+	/* Timeline activity */
+	ACTIVITY_VP_0 = 0,
+	ACTIVITY_FP_0,
+	ACTIVITY_FP_1,
+	ACTIVITY_FP_2,
+	ACTIVITY_FP_3,
+
+	/* L2 cache counters */
+	COUNTER_L2_0_C0,
+	COUNTER_L2_0_C1,
+
+	/* Vertex processor counters */
+	COUNTER_VP_0_C0,
+	COUNTER_VP_0_C1,
+
+	/* Fragment processor counters */
+	COUNTER_FP_0_C0,
+	COUNTER_FP_0_C1,
+	COUNTER_FP_1_C0,
+	COUNTER_FP_1_C1,
+	COUNTER_FP_2_C0,
+	COUNTER_FP_2_C1,
+	COUNTER_FP_3_C0,
+	COUNTER_FP_3_C1,
+
+	/* EGL Software Counters */
+	COUNTER_EGL_BLIT_TIME,
+
+	/* GLES Software Counters */
+	COUNTER_GLES_DRAW_ELEMENTS_CALLS,
+	COUNTER_GLES_DRAW_ELEMENTS_NUM_INDICES,
+	COUNTER_GLES_DRAW_ELEMENTS_NUM_TRANSFORMED,
+	COUNTER_GLES_DRAW_ARRAYS_CALLS,
+	COUNTER_GLES_DRAW_ARRAYS_NUM_TRANSFORMED,
+	COUNTER_GLES_DRAW_POINTS,
+	COUNTER_GLES_DRAW_LINES,
+	COUNTER_GLES_DRAW_LINE_LOOP,
+	COUNTER_GLES_DRAW_LINE_STRIP,
+	COUNTER_GLES_DRAW_TRIANGLES,
+	COUNTER_GLES_DRAW_TRIANGLE_STRIP,
+	COUNTER_GLES_DRAW_TRIANGLE_FAN,
+	COUNTER_GLES_NON_VBO_DATA_COPY_TIME,
+	COUNTER_GLES_UNIFORM_BYTES_COPIED_TO_MALI,
+	COUNTER_GLES_UPLOAD_TEXTURE_TIME,
+	COUNTER_GLES_UPLOAD_VBO_TIME,
+	COUNTER_GLES_NUM_FLUSHES,
+	COUNTER_GLES_NUM_VSHADERS_GENERATED,
+	COUNTER_GLES_NUM_FSHADERS_GENERATED,
+	COUNTER_GLES_VSHADER_GEN_TIME,
+	COUNTER_GLES_FSHADER_GEN_TIME,
+	COUNTER_GLES_INPUT_TRIANGLES,
+	COUNTER_GLES_VXCACHE_HIT,
+	COUNTER_GLES_VXCACHE_MISS,
+	COUNTER_GLES_VXCACHE_COLLISION,
+	COUNTER_GLES_CULLED_TRIANGLES,
+	COUNTER_GLES_CULLED_LINES,
+	COUNTER_GLES_BACKFACE_TRIANGLES,
+	COUNTER_GLES_GBCLIP_TRIANGLES,
+	COUNTER_GLES_GBCLIP_LINES,
+	COUNTER_GLES_TRIANGLES_DRAWN,
+	COUNTER_GLES_DRAWCALL_TIME,
+	COUNTER_GLES_TRIANGLES_COUNT,
+	COUNTER_GLES_INDEPENDENT_TRIANGLES_COUNT,
+	COUNTER_GLES_STRIP_TRIANGLES_COUNT,
+	COUNTER_GLES_FAN_TRIANGLES_COUNT,
+	COUNTER_GLES_LINES_COUNT,
+	COUNTER_GLES_INDEPENDENT_LINES_COUNT,
+	COUNTER_GLES_STRIP_LINES_COUNT,
+	COUNTER_GLES_LOOP_LINES_COUNT,
+
+	COUNTER_FILMSTRIP,
+	COUNTER_FREQUENCY,
+	COUNTER_VOLTAGE,
+
+	NUMBER_OF_EVENTS
+};
+
+#define FIRST_ACTIVITY_EVENT    ACTIVITY_VP_0
+#define LAST_ACTIVITY_EVENT     ACTIVITY_FP_3
+
+#define FIRST_HW_COUNTER        COUNTER_L2_0_C0
+#define LAST_HW_COUNTER         COUNTER_FP_3_C1
+
+#define FIRST_SW_COUNTER        COUNTER_EGL_BLIT_TIME
+#define LAST_SW_COUNTER         COUNTER_GLES_LOOP_LINES_COUNT
+
+/* Signifies that the system is able to report voltage and frequency numbers. */
+#define DVFS_REPORTED_BY_DDK 1
+
+/**
+ * Structure to pass performance counter data of a Mali core
+ */
+struct _mali_profiling_core_counters {
+	u32 source0;
+	u32 value0;
+	u32 source1;
+	u32 value1;
+};
+
+/*
+ * For compatibility with utgard.
+ */
+struct _mali_profiling_l2_counter_values {
+	struct _mali_profiling_core_counters cores[MAX_NUM_L2_CACHE_CORES];
+};
+
+struct _mali_profiling_mali_version {
+	u32 mali_product_id;
+	u32 mali_version_major;
+	u32 mali_version_minor;
+	u32 num_of_l2_cores;
+	u32 num_of_fp_cores;
+	u32 num_of_vp_cores;
+};
+
+extern void _mali_profiling_get_mali_version(struct _mali_profiling_mali_version *values);
+extern u32 _mali_profiling_get_l2_counters(struct _mali_profiling_l2_counter_values *values);
+
+/*
+ * List of possible actions allowing DDK to be controlled by Streamline.
+ * The following numbers are used by DDK to control the frame buffer dumping.
+ */
+#define FBDUMP_CONTROL_ENABLE (1)
+#define FBDUMP_CONTROL_RATE (2)
+#define SW_COUNTER_ENABLE      (3)
+#define FBDUMP_CONTROL_RESIZE_FACTOR (4)
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* __MALI_MJOLLNIR_PROFILING_GATOR_API_H__ */
diff --git a/drivers/gator/mali/mali_utgard_profiling_gator_api.h b/drivers/gator/mali/mali_utgard_profiling_gator_api.h
new file mode 100644
index 000000000000..d6465312628e
--- /dev/null
+++ b/drivers/gator/mali/mali_utgard_profiling_gator_api.h
@@ -0,0 +1,197 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#ifndef __MALI_UTGARD_PROFILING_GATOR_API_H__
+#define __MALI_UTGARD_PROFILING_GATOR_API_H__
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+
+#define MALI_PROFILING_API_VERSION 4
+
+#define MAX_NUM_L2_CACHE_CORES 3
+#define MAX_NUM_FP_CORES 8
+#define MAX_NUM_VP_CORES 1
+
+/** The list of events supported by the Mali DDK. */
+enum {
+	/* Vertex processor activity */
+	ACTIVITY_VP_0 = 0,
+
+	/* Fragment processor activity */
+	ACTIVITY_FP_0, /* 1 */
+	ACTIVITY_FP_1,
+	ACTIVITY_FP_2,
+	ACTIVITY_FP_3,
+	ACTIVITY_FP_4,
+	ACTIVITY_FP_5,
+	ACTIVITY_FP_6,
+	ACTIVITY_FP_7,
+
+	/* L2 cache counters */
+	COUNTER_L2_0_C0,
+	COUNTER_L2_0_C1,
+	COUNTER_L2_1_C0,
+	COUNTER_L2_1_C1,
+	COUNTER_L2_2_C0,
+	COUNTER_L2_2_C1,
+
+	/* Vertex processor counters */
+	COUNTER_VP_0_C0, /*15*/
+	COUNTER_VP_0_C1,
+
+	/* Fragment processor counters */
+	COUNTER_FP_0_C0,
+	COUNTER_FP_0_C1,
+	COUNTER_FP_1_C0,
+	COUNTER_FP_1_C1,
+	COUNTER_FP_2_C0,
+	COUNTER_FP_2_C1,
+	COUNTER_FP_3_C0,
+	COUNTER_FP_3_C1,
+	COUNTER_FP_4_C0,
+	COUNTER_FP_4_C1,
+	COUNTER_FP_5_C0,
+	COUNTER_FP_5_C1,
+	COUNTER_FP_6_C0,
+	COUNTER_FP_6_C1,
+	COUNTER_FP_7_C0,
+	COUNTER_FP_7_C1, /* 32 */
+
+	/*
+	 * If more hardware counters are added, the _mali_osk_hw_counter_table
+	 * below should also be updated.
+	 */
+
+	/* EGL software counters */
+	COUNTER_EGL_BLIT_TIME,
+
+	/* GLES software counters */
+	COUNTER_GLES_DRAW_ELEMENTS_CALLS,
+	COUNTER_GLES_DRAW_ELEMENTS_NUM_INDICES,
+	COUNTER_GLES_DRAW_ELEMENTS_NUM_TRANSFORMED,
+	COUNTER_GLES_DRAW_ARRAYS_CALLS,
+	COUNTER_GLES_DRAW_ARRAYS_NUM_TRANSFORMED,
+	COUNTER_GLES_DRAW_POINTS,
+	COUNTER_GLES_DRAW_LINES,
+	COUNTER_GLES_DRAW_LINE_LOOP,
+	COUNTER_GLES_DRAW_LINE_STRIP,
+	COUNTER_GLES_DRAW_TRIANGLES,
+	COUNTER_GLES_DRAW_TRIANGLE_STRIP,
+	COUNTER_GLES_DRAW_TRIANGLE_FAN,
+	COUNTER_GLES_NON_VBO_DATA_COPY_TIME,
+	COUNTER_GLES_UNIFORM_BYTES_COPIED_TO_MALI,
+	COUNTER_GLES_UPLOAD_TEXTURE_TIME,
+	COUNTER_GLES_UPLOAD_VBO_TIME,
+	COUNTER_GLES_NUM_FLUSHES,
+	COUNTER_GLES_NUM_VSHADERS_GENERATED,
+	COUNTER_GLES_NUM_FSHADERS_GENERATED,
+	COUNTER_GLES_VSHADER_GEN_TIME,
+	COUNTER_GLES_FSHADER_GEN_TIME,
+	COUNTER_GLES_INPUT_TRIANGLES,
+	COUNTER_GLES_VXCACHE_HIT,
+	COUNTER_GLES_VXCACHE_MISS,
+	COUNTER_GLES_VXCACHE_COLLISION,
+	COUNTER_GLES_CULLED_TRIANGLES,
+	COUNTER_GLES_CULLED_LINES,
+	COUNTER_GLES_BACKFACE_TRIANGLES,
+	COUNTER_GLES_GBCLIP_TRIANGLES,
+	COUNTER_GLES_GBCLIP_LINES,
+	COUNTER_GLES_TRIANGLES_DRAWN,
+	COUNTER_GLES_DRAWCALL_TIME,
+	COUNTER_GLES_TRIANGLES_COUNT,
+	COUNTER_GLES_INDEPENDENT_TRIANGLES_COUNT,
+	COUNTER_GLES_STRIP_TRIANGLES_COUNT,
+	COUNTER_GLES_FAN_TRIANGLES_COUNT,
+	COUNTER_GLES_LINES_COUNT,
+	COUNTER_GLES_INDEPENDENT_LINES_COUNT,
+	COUNTER_GLES_STRIP_LINES_COUNT,
+	COUNTER_GLES_LOOP_LINES_COUNT,
+
+	/* Framebuffer capture pseudo-counter */
+	COUNTER_FILMSTRIP,
+
+	NUMBER_OF_EVENTS
+} _mali_osk_counter_id;
+
+#define FIRST_ACTIVITY_EVENT    ACTIVITY_VP_0
+#define LAST_ACTIVITY_EVENT     ACTIVITY_FP_7
+
+#define FIRST_HW_COUNTER        COUNTER_L2_0_C0
+#define LAST_HW_COUNTER         COUNTER_FP_7_C1
+
+#define FIRST_SW_COUNTER        COUNTER_EGL_BLIT_TIME
+#define LAST_SW_COUNTER         COUNTER_GLES_LOOP_LINES_COUNT
+
+#define FIRST_SPECIAL_COUNTER   COUNTER_FILMSTRIP
+#define LAST_SPECIAL_COUNTER    COUNTER_FILMSTRIP
+
+/**
+ * Structure to pass performance counter data of a Mali core
+ */
+struct _mali_profiling_core_counters {
+	u32 source0;
+	u32 value0;
+	u32 source1;
+	u32 value1;
+};
+
+/**
+ * Structure to pass performance counter data of Mali L2 cache cores
+ */
+struct _mali_profiling_l2_counter_values {
+	struct _mali_profiling_core_counters cores[MAX_NUM_L2_CACHE_CORES];
+};
+
+/**
+ * Structure to pass data defining Mali instance in use:
+ *
+ * mali_product_id - Mali product id
+ * mali_version_major - Mali version major number
+ * mali_version_minor - Mali version minor number
+ * num_of_l2_cores - number of L2 cache cores
+ * num_of_fp_cores - number of fragment processor cores
+ * num_of_vp_cores - number of vertex processor cores
+ */
+struct _mali_profiling_mali_version {
+	u32 mali_product_id;
+	u32 mali_version_major;
+	u32 mali_version_minor;
+	u32 num_of_l2_cores;
+	u32 num_of_fp_cores;
+	u32 num_of_vp_cores;
+};
+
+/*
+ * List of possible actions to be controlled by Streamline.
+ * The following numbers are used by gator to control the frame buffer dumping and s/w counter reporting.
+ * We cannot use the enums in mali_uk_types.h because they are unknown inside gator.
+ */
+#define FBDUMP_CONTROL_ENABLE (1)
+#define FBDUMP_CONTROL_RATE (2)
+#define SW_COUNTER_ENABLE (3)
+#define FBDUMP_CONTROL_RESIZE_FACTOR (4)
+
+void _mali_profiling_control(u32 action, u32 value);
+
+u32 _mali_profiling_get_l2_counters(struct _mali_profiling_l2_counter_values *values);
+
+int _mali_profiling_set_event(u32 counter_id, s32 event_id);
+
+u32 _mali_profiling_get_api_version(void);
+
+void _mali_profiling_get_mali_version(struct _mali_profiling_mali_version *values);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* __MALI_UTGARD_PROFILING_GATOR_API_H__ */
diff --git a/drivers/gator/mali_midgard.mk b/drivers/gator/mali_midgard.mk
new file mode 100644
index 000000000000..1b784d5c3d58
--- /dev/null
+++ b/drivers/gator/mali_midgard.mk
@@ -0,0 +1,39 @@
+# Defines for Mali-Midgard driver
+EXTRA_CFLAGS += -DMALI_USE_UMP=1 \
+                -DMALI_LICENSE_IS_GPL=1 \
+                -DMALI_BASE_TRACK_MEMLEAK=0 \
+                -DMALI_DEBUG=0 \
+                -DMALI_ERROR_INJECT_ON=0 \
+                -DMALI_CUSTOMER_RELEASE=1 \
+                -DMALI_UNIT_TEST=0 \
+                -DMALI_BACKEND_KERNEL=1 \
+                -DMALI_NO_MALI=0
+
+DDK_DIR ?= .
+ifneq ($(wildcard $(DDK_DIR)/drivers/gpu/arm/t6xx),)
+KBASE_DIR = $(DDK_DIR)/drivers/gpu/arm/t6xx/kbase
+OSK_DIR = $(DDK_DIR)/drivers/gpu/arm/t6xx/kbase/osk
+endif
+
+ifneq ($(wildcard $(DDK_DIR)/drivers/gpu/arm/midgard),)
+KBASE_DIR = $(DDK_DIR)/drivers/gpu/arm/midgard
+OSK_DIR = $(DDK_DIR)/drivers/gpu/arm/midgard/osk
+EXTRA_CFLAGS += -DMALI_DIR_MIDGARD=1
+endif
+
+ifneq ($(wildcard $(DDK_DIR)/drivers/gpu/arm/midgard/mali_kbase_gator_api.h),)
+EXTRA_CFLAGS += -DMALI_SIMPLE_API=1
+endif
+
+UMP_DIR = $(DDK_DIR)/include/linux
+
+# Include directories in the DDK
+EXTRA_CFLAGS += -I$(KBASE_DIR)/ \
+                -I$(KBASE_DIR)/.. \
+                -I$(OSK_DIR)/.. \
+                -I$(UMP_DIR)/.. \
+                -I$(DDK_DIR)/include \
+                -I$(KBASE_DIR)/osk/src/linux/include \
+                -I$(KBASE_DIR)/platform_dummy \
+                -I$(KBASE_DIR)/src
+
diff --git a/tools/gator/Makefile b/tools/gator/Makefile
new file mode 100644
index 000000000000..4ab73b0ebedb
--- /dev/null
+++ b/tools/gator/Makefile
@@ -0,0 +1,4 @@
+all:
+	$(MAKE) -C daemon $@
+clean:
+	$(MAKE) -C daemon $@
diff --git a/tools/gator/daemon/Android.mk b/tools/gator/daemon/Android.mk
new file mode 100644
index 000000000000..970ac6946150
--- /dev/null
+++ b/tools/gator/daemon/Android.mk
@@ -0,0 +1,76 @@
+LOCAL_PATH := $(call my-dir)
+include $(CLEAR_VARS)
+
+XML_H := $(shell cd $(LOCAL_PATH) && make events_xml.h defaults_xml.h)
+
+LOCAL_SRC_FILES := \
+	AnnotateListener.cpp \
+	Buffer.cpp \
+	CCNDriver.cpp \
+	CPUFreqDriver.cpp \
+	CapturedXML.cpp \
+	Child.cpp \
+	Command.cpp \
+	ConfigurationXML.cpp \
+	DiskIODriver.cpp \
+	Driver.cpp \
+	DriverSource.cpp \
+	DynBuf.cpp \
+	EventsXML.cpp \
+	ExternalSource.cpp \
+	FSDriver.cpp \
+	Fifo.cpp \
+	FtraceDriver.cpp \
+	FtraceSource.cpp \
+	HwmonDriver.cpp \
+	KMod.cpp \
+	LocalCapture.cpp \
+	Logging.cpp \
+	main.cpp \
+	MaliVideoDriver.cpp \
+	MemInfoDriver.cpp\
+	Monitor.cpp \
+	NetDriver.cpp \
+	OlySocket.cpp \
+	OlyUtility.cpp \
+	PerfBuffer.cpp \
+	PerfDriver.cpp \
+	PerfGroup.cpp \
+	PerfSource.cpp \
+	Proc.cpp \
+	Sender.cpp \
+	SessionData.cpp \
+	SessionXML.cpp \
+	Setup.cpp \
+	Source.cpp \
+	StreamlineSetup.cpp \
+	UEvent.cpp \
+	UserSpaceSource.cpp \
+	libsensors/access.c \
+	libsensors/conf-lex.c \
+	libsensors/conf-parse.c \
+	libsensors/data.c \
+	libsensors/error.c \
+	libsensors/general.c \
+	libsensors/init.c \
+	libsensors/sysfs.c \
+	mxml/mxml-attr.c \
+	mxml/mxml-entity.c \
+	mxml/mxml-file.c \
+	mxml/mxml-get.c \
+	mxml/mxml-index.c \
+	mxml/mxml-node.c \
+	mxml/mxml-private.c \
+	mxml/mxml-search.c \
+	mxml/mxml-set.c \
+	mxml/mxml-string.c
+
+LOCAL_CFLAGS += -Wall -O3 -fno-exceptions -pthread -DETCDIR=\"/etc\" -Ilibsensors -fPIE
+LOCAL_LDFLAGS += -fPIE -pie
+
+LOCAL_C_INCLUDES := $(LOCAL_PATH)
+
+LOCAL_MODULE := gatord
+LOCAL_MODULE_TAGS := optional
+
+include $(BUILD_EXECUTABLE)
diff --git a/tools/gator/daemon/AnnotateListener.cpp b/tools/gator/daemon/AnnotateListener.cpp
new file mode 100644
index 000000000000..50110b4dc84c
--- /dev/null
+++ b/tools/gator/daemon/AnnotateListener.cpp
@@ -0,0 +1,69 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "AnnotateListener.h"
+
+#include <unistd.h>
+
+#include "OlySocket.h"
+
+struct AnnotateClient {
+	AnnotateClient *next;
+	int fd;
+};
+
+AnnotateListener::AnnotateListener() : mClients(NULL), mSock(NULL) {
+}
+
+AnnotateListener::~AnnotateListener() {
+	close();
+	delete mSock;
+}
+
+void AnnotateListener::setup() {
+	mSock = new OlyServerSocket(8082);
+}
+
+int AnnotateListener::getFd() {
+	return mSock->getFd();
+}
+
+void AnnotateListener::handle() {
+	AnnotateClient *const client = new AnnotateClient();
+	client->fd = mSock->acceptConnection();
+	client->next = mClients;
+	mClients = client;
+}
+
+void AnnotateListener::close() {
+	mSock->closeServerSocket();
+	while (mClients != NULL) {
+		::close(mClients->fd);
+		AnnotateClient *next = mClients->next;
+		delete mClients;
+		mClients = next;
+	}
+}
+
+void AnnotateListener::signal() {
+	const char ch = 0;
+	AnnotateClient **ptr = &mClients;
+	AnnotateClient *client = mClients;
+	while (client != NULL) {
+		if (write(client->fd, &ch, sizeof(ch)) != 1) {
+			::close(client->fd);
+			AnnotateClient *next = client->next;
+			delete client;
+			*ptr = next;
+			client = next;
+			continue;
+		}
+		ptr = &client->next;
+		client = client->next;
+	}
+}
diff --git a/tools/gator/daemon/AnnotateListener.h b/tools/gator/daemon/AnnotateListener.h
new file mode 100644
index 000000000000..cdefef12db22
--- /dev/null
+++ b/tools/gator/daemon/AnnotateListener.h
@@ -0,0 +1,31 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+class AnnotateClient;
+class OlyServerSocket;
+
+class AnnotateListener {
+public:
+	AnnotateListener();
+	~AnnotateListener();
+
+	void setup();
+	int getFd();
+
+	void handle();
+	void close();
+	void signal();
+
+private:
+	AnnotateClient *mClients;
+	OlyServerSocket *mSock;
+
+	// Intentionally unimplemented
+	AnnotateListener(const AnnotateListener &);
+	AnnotateListener &operator=(const AnnotateListener &);
+};
diff --git a/tools/gator/daemon/Application.mk b/tools/gator/daemon/Application.mk
new file mode 100644
index 000000000000..3ada471cac19
--- /dev/null
+++ b/tools/gator/daemon/Application.mk
@@ -0,0 +1,3 @@
+APP_PLATFORM := android-8
+# Replace armeabi-v7a with arm64-v8a to build an arm64 gatord or with armeabi to build an ARM11 gatord
+APP_ABI := armeabi-v7a
diff --git a/tools/gator/daemon/Buffer.cpp b/tools/gator/daemon/Buffer.cpp
new file mode 100644
index 000000000000..8fa628015069
--- /dev/null
+++ b/tools/gator/daemon/Buffer.cpp
@@ -0,0 +1,428 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "Buffer.h"
+
+#include "Logging.h"
+#include "Sender.h"
+#include "SessionData.h"
+
+#define mask (mSize - 1)
+
+enum {
+	CODE_PEA         = 1,
+	CODE_KEYS        = 2,
+	CODE_FORMAT      = 3,
+	CODE_MAPS        = 4,
+	CODE_COMM        = 5,
+	CODE_KEYS_OLD    = 6,
+	CODE_ONLINE_CPU  = 7,
+	CODE_OFFLINE_CPU = 8,
+	CODE_KALLSYMS    = 9,
+};
+
+// Summary Frame Messages
+enum {
+	MESSAGE_SUMMARY = 1,
+	MESSAGE_CORE_NAME = 3,
+};
+
+// From gator_marshaling.c
+#define NEWLINE_CANARY \
+	/* Unix */ \
+	"1\n" \
+	/* Windows */ \
+	"2\r\n" \
+	/* Mac OS */ \
+	"3\r" \
+	/* RISC OS */ \
+	"4\n\r" \
+	/* Add another character so the length isn't 0x0a bytes */ \
+	"5"
+
+Buffer::Buffer(const int32_t core, const int32_t buftype, const int size, sem_t *const readerSem) : mBuf(new char[size]), mReaderSem(readerSem), mCommitTime(gSessionData->mLiveRate), mSize(size), mReadPos(0), mWritePos(0), mCommitPos(0), mAvailable(true), mIsDone(false), mCore(core), mBufType(buftype) {
+	if ((mSize & mask) != 0) {
+		logg->logError(__FILE__, __LINE__, "Buffer size is not a power of 2");
+		handleException();
+	}
+	sem_init(&mWriterSem, 0, 0);
+	frame();
+}
+
+Buffer::~Buffer() {
+	delete [] mBuf;
+	sem_destroy(&mWriterSem);
+}
+
+void Buffer::write(Sender *const sender) {
+	if (!commitReady()) {
+		return;
+	}
+
+	// commit and read are updated by the writer, only read them once
+	int commitPos = mCommitPos;
+	int readPos = mReadPos;
+
+	// determine the size of two halves
+	int length1 = commitPos - readPos;
+	char *buffer1 = mBuf + readPos;
+	int length2 = 0;
+	char *buffer2 = mBuf;
+	if (length1 < 0) {
+		length1 = mSize - readPos;
+		length2 = commitPos;
+	}
+
+	logg->logMessage("Sending data length1: %i length2: %i", length1, length2);
+
+	// start, middle or end
+	if (length1 > 0) {
+		sender->writeData(buffer1, length1, RESPONSE_APC_DATA);
+	}
+
+	// possible wrap around
+	if (length2 > 0) {
+		sender->writeData(buffer2, length2, RESPONSE_APC_DATA);
+	}
+
+	mReadPos = commitPos;
+
+	// send a notification that space is available
+	sem_post(&mWriterSem);
+}
+
+bool Buffer::commitReady() const {
+	return mCommitPos != mReadPos;
+}
+
+int Buffer::bytesAvailable() const {
+	int filled = mWritePos - mReadPos;
+	if (filled < 0) {
+		filled += mSize;
+	}
+
+	int remaining = mSize - filled;
+
+	if (mAvailable) {
+		// Give some extra room; also allows space to insert the overflow error packet
+		remaining -= 200;
+	} else {
+		// Hysteresis, prevents multiple overflow messages
+		remaining -= 2000;
+	}
+
+	return remaining;
+}
+
+bool Buffer::checkSpace(const int bytes) {
+	const int remaining = bytesAvailable();
+
+	if (remaining < bytes) {
+		mAvailable = false;
+	} else {
+		mAvailable = true;
+	}
+
+	return mAvailable;
+}
+
+int Buffer::contiguousSpaceAvailable() const {
+	int remaining = bytesAvailable();
+	int contiguous = mSize - mWritePos;
+	if (remaining < contiguous) {
+		return remaining;
+	} else {
+		return contiguous;
+	}
+}
+
+void Buffer::commit(const uint64_t time) {
+	// post-populate the length, which does not include the response type length nor the length itself, i.e. only the length of the payload
+	const int typeLength = gSessionData->mLocalCapture ? 0 : 1;
+	int length = mWritePos - mCommitPos;
+	if (length < 0) {
+		length += mSize;
+	}
+	length = length - typeLength - sizeof(int32_t);
+	for (size_t byte = 0; byte < sizeof(int32_t); byte++) {
+		mBuf[(mCommitPos + typeLength + byte) & mask] = (length >> byte * 8) & 0xFF;
+	}
+
+	logg->logMessage("Committing data mReadPos: %i mWritePos: %i mCommitPos: %i", mReadPos, mWritePos, mCommitPos);
+	mCommitPos = mWritePos;
+
+	if (gSessionData->mLiveRate > 0) {
+		while (time > mCommitTime) {
+			mCommitTime += gSessionData->mLiveRate;
+		}
+	}
+
+	if (!mIsDone) {
+		frame();
+	}
+
+	// send a notification that data is ready
+	sem_post(mReaderSem);
+}
+
+void Buffer::check(const uint64_t time) {
+	int filled = mWritePos - mCommitPos;
+	if (filled < 0) {
+		filled += mSize;
+	}
+	if (filled >= ((mSize * 3) / 4) || (gSessionData->mLiveRate > 0 && time >= mCommitTime)) {
+		commit(time);
+	}
+}
+
+void Buffer::packInt(char *const buf, const int size, int &writePos, int32_t x) {
+	int packedBytes = 0;
+	int more = true;
+	while (more) {
+		// low order 7 bits of x
+		char b = x & 0x7f;
+		x >>= 7;
+
+		if ((x == 0 && (b & 0x40) == 0) || (x == -1 && (b & 0x40) != 0)) {
+			more = false;
+		} else {
+			b |= 0x80;
+		}
+
+		buf[(writePos + packedBytes) & /*mask*/(size - 1)] = b;
+		packedBytes++;
+	}
+
+	writePos = (writePos + packedBytes) & /*mask*/(size - 1);
+}
+
+void Buffer::packInt(int32_t x) {
+	packInt(mBuf, mSize, mWritePos, x);
+}
+
+void Buffer::packInt64(char *const buf, const int size, int &writePos, int64_t x) {
+	int packedBytes = 0;
+	int more = true;
+	while (more) {
+		// low order 7 bits of x
+		char b = x & 0x7f;
+		x >>= 7;
+
+		if ((x == 0 && (b & 0x40) == 0) || (x == -1 && (b & 0x40) != 0)) {
+			more = false;
+		} else {
+			b |= 0x80;
+		}
+
+		buf[(writePos + packedBytes) & /*mask*/(size - 1)] = b;
+		packedBytes++;
+	}
+
+	writePos = (writePos + packedBytes) & /*mask*/(size - 1);
+}
+
+void Buffer::packInt64(int64_t x) {
+	packInt64(mBuf, mSize, mWritePos, x);
+}
+
+void Buffer::writeBytes(const void *const data, size_t count) {
+	size_t i;
+	for (i = 0; i < count; ++i) {
+		mBuf[(mWritePos + i) & mask] = static_cast<const char *>(data)[i];
+	}
+
+	mWritePos = (mWritePos + i) & mask;
+}
+
+void Buffer::writeString(const char *const str) {
+	const int len = strlen(str);
+	packInt(len);
+	writeBytes(str, len);
+}
+
+void Buffer::frame() {
+	if (!gSessionData->mLocalCapture) {
+		packInt(RESPONSE_APC_DATA);
+	}
+	// Reserve space for the length
+	mWritePos += sizeof(int32_t);
+	packInt(mBufType);
+	if ((mBufType == FRAME_BLOCK_COUNTER) || (mBufType == FRAME_PERF_ATTRS) || (mBufType == FRAME_PERF)) {
+		packInt(mCore);
+	}
+}
+
+void Buffer::summary(const uint64_t currTime, const int64_t timestamp, const int64_t uptime, const int64_t monotonicDelta, const char *const uname) {
+	packInt(MESSAGE_SUMMARY);
+	writeString(NEWLINE_CANARY);
+	packInt64(timestamp);
+	packInt64(uptime);
+	packInt64(monotonicDelta);
+	writeString("uname");
+	writeString(uname);
+	writeString("");
+	check(currTime);
+}
+
+void Buffer::coreName(const uint64_t currTime, const int core, const int cpuid, const char *const name) {
+	if (checkSpace(3 * MAXSIZE_PACK32 + 0x100)) {
+		packInt(MESSAGE_CORE_NAME);
+		packInt(core);
+		packInt(cpuid);
+		writeString(name);
+	}
+	check(currTime);
+}
+
+bool Buffer::eventHeader(const uint64_t curr_time) {
+	bool retval = false;
+	if (checkSpace(MAXSIZE_PACK32 + MAXSIZE_PACK64)) {
+		// key of zero indicates a timestamp
+		packInt(0);
+		packInt64(curr_time);
+		retval = true;
+	}
+
+	return retval;
+}
+
+bool Buffer::eventTid(const int tid) {
+	bool retval = false;
+	if (checkSpace(2 * MAXSIZE_PACK32)) {
+		// key of 1 indicates a tid
+		packInt(1);
+		packInt(tid);
+		retval = true;
+	}
+
+	return retval;
+}
+
+void Buffer::event(const int key, const int32_t value) {
+	if (checkSpace(2 * MAXSIZE_PACK32)) {
+		packInt(key);
+		packInt(value);
+	}
+}
+
+void Buffer::event64(const int key, const int64_t value) {
+	if (checkSpace(MAXSIZE_PACK64 + MAXSIZE_PACK32)) {
+		packInt(key);
+		packInt64(value);
+	}
+}
+
+void Buffer::pea(const uint64_t currTime, const struct perf_event_attr *const pea, int key) {
+	while (!checkSpace(2 * MAXSIZE_PACK32 + pea->size)) {
+		sem_wait(&mWriterSem);
+	}
+	packInt(CODE_PEA);
+	writeBytes(pea, pea->size);
+	packInt(key);
+	check(currTime);
+}
+
+void Buffer::keys(const uint64_t currTime, const int count, const __u64 *const ids, const int *const keys) {
+	while (!checkSpace(2 * MAXSIZE_PACK32 + count * (MAXSIZE_PACK32 + MAXSIZE_PACK64))) {
+		sem_wait(&mWriterSem);
+	}
+	packInt(CODE_KEYS);
+	packInt(count);
+	for (int i = 0; i < count; ++i) {
+		packInt64(ids[i]);
+		packInt(keys[i]);
+	}
+	check(currTime);
+}
+
+void Buffer::keysOld(const uint64_t currTime, const int keyCount, const int *const keys, const int bytes, const char *const buf) {
+	while (!checkSpace((2 + keyCount) * MAXSIZE_PACK32 + bytes)) {
+		sem_wait(&mWriterSem);
+	}
+	packInt(CODE_KEYS_OLD);
+	packInt(keyCount);
+	for (int i = 0; i < keyCount; ++i) {
+		packInt(keys[i]);
+	}
+	writeBytes(buf, bytes);
+	check(currTime);
+}
+
+void Buffer::format(const uint64_t currTime, const int length, const char *const format) {
+	while (!checkSpace(MAXSIZE_PACK32 + length + 1)) {
+		sem_wait(&mWriterSem);
+	}
+	packInt(CODE_FORMAT);
+	writeBytes(format, length + 1);
+	check(currTime);
+}
+
+void Buffer::maps(const uint64_t currTime, const int pid, const int tid, const char *const maps) {
+	const int mapsLen = strlen(maps) + 1;
+	while (!checkSpace(3 * MAXSIZE_PACK32 + mapsLen)) {
+		sem_wait(&mWriterSem);
+	}
+	packInt(CODE_MAPS);
+	packInt(pid);
+	packInt(tid);
+	writeBytes(maps, mapsLen);
+	check(currTime);
+}
+
+void Buffer::comm(const uint64_t currTime, const int pid, const int tid, const char *const image, const char *const comm) {
+	const int imageLen = strlen(image) + 1;
+	const int commLen = strlen(comm) + 1;
+	while (!checkSpace(3 * MAXSIZE_PACK32 + imageLen + commLen)) {
+		sem_wait(&mWriterSem);
+	}
+	packInt(CODE_COMM);
+	packInt(pid);
+	packInt(tid);
+	writeBytes(image, imageLen);
+	writeBytes(comm, commLen);
+	check(currTime);
+}
+
+void Buffer::onlineCPU(const uint64_t currTime, const uint64_t time, const int cpu) {
+	while (!checkSpace(MAXSIZE_PACK32 + MAXSIZE_PACK64)) {
+		sem_wait(&mWriterSem);
+	}
+	packInt(CODE_ONLINE_CPU);
+	packInt64(time);
+	packInt(cpu);
+	check(currTime);
+}
+
+void Buffer::offlineCPU(const uint64_t currTime, const uint64_t time, const int cpu) {
+	while (!checkSpace(MAXSIZE_PACK32 + MAXSIZE_PACK64)) {
+		sem_wait(&mWriterSem);
+	}
+	packInt(CODE_OFFLINE_CPU);
+	packInt64(time);
+	packInt(cpu);
+	check(currTime);
+}
+
+void Buffer::kallsyms(const uint64_t currTime, const char *const kallsyms) {
+	const int kallsymsLen = strlen(kallsyms) + 1;
+	while (!checkSpace(3 * MAXSIZE_PACK32 + kallsymsLen)) {
+		sem_wait(&mWriterSem);
+	}
+	packInt(CODE_KALLSYMS);
+	writeBytes(kallsyms, kallsymsLen);
+	check(currTime);
+}
+
+void Buffer::setDone() {
+	mIsDone = true;
+	commit(0);
+}
+
+bool Buffer::isDone() const {
+	return mIsDone && mReadPos == mCommitPos && mCommitPos == mWritePos;
+}
diff --git a/tools/gator/daemon/Buffer.h b/tools/gator/daemon/Buffer.h
new file mode 100644
index 000000000000..6cffd8e39a36
--- /dev/null
+++ b/tools/gator/daemon/Buffer.h
@@ -0,0 +1,106 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef BUFFER_H
+#define BUFFER_H
+
+#include <stdint.h>
+#include <semaphore.h>
+
+#include "k/perf_event.h"
+
+class Sender;
+
+enum {
+	FRAME_SUMMARY       =  1,
+	FRAME_BLOCK_COUNTER =  5,
+	FRAME_EXTERNAL      = 10,
+	FRAME_PERF_ATTRS    = 11,
+	FRAME_PERF          = 12,
+};
+
+class Buffer {
+public:
+	static const size_t MAXSIZE_PACK32 = 5;
+	static const size_t MAXSIZE_PACK64 = 10;
+
+	Buffer(int32_t core, int32_t buftype, const int size, sem_t *const readerSem);
+	~Buffer();
+
+	void write(Sender *sender);
+
+	int bytesAvailable() const;
+	int contiguousSpaceAvailable() const;
+	void commit(const uint64_t time);
+	void check(const uint64_t time);
+
+	// Summary messages
+	void summary(const uint64_t currTime, const int64_t timestamp, const int64_t uptime, const int64_t monotonicDelta, const char *const uname);
+	void coreName(const uint64_t currTime, const int core, const int cpuid, const char *const name);
+
+	// Block Counter messages
+	bool eventHeader(uint64_t curr_time);
+	bool eventTid(int tid);
+	void event(int key, int32_t value);
+	void event64(int key, int64_t value);
+
+	// Perf Attrs messages
+	void pea(const uint64_t currTime, const struct perf_event_attr *const pea, int key);
+	void keys(const uint64_t currTime, const int count, const __u64 *const ids, const int *const keys);
+	void keysOld(const uint64_t currTime, const int keyCount, const int *const keys, const int bytes, const char *const buf);
+	void format(const uint64_t currTime, const int length, const char *const format);
+	void maps(const uint64_t currTime, const int pid, const int tid, const char *const maps);
+	void comm(const uint64_t currTime, const int pid, const int tid, const char *const image, const char *const comm);
+	void onlineCPU(const uint64_t currTime, const uint64_t time, const int cpu);
+	void offlineCPU(const uint64_t currTime, const uint64_t time, const int cpu);
+	void kallsyms(const uint64_t currTime, const char *const kallsyms);
+
+	void setDone();
+	bool isDone() const;
+
+	// Prefer a new member to using these functions if possible
+	char *getWritePos() { return mBuf + mWritePos; }
+	void advanceWrite(int bytes) { mWritePos = (mWritePos + bytes) & /*mask*/(mSize - 1); }
+	static void packInt(char *const buf, const int size, int &writePos, int32_t x);
+	void packInt(int32_t x);
+	static void packInt64(char *const buf, const int size, int &writePos, int64_t x);
+	void packInt64(int64_t x);
+	void writeBytes(const void *const data, size_t count);
+	void writeString(const char *const str);
+
+	static void writeLEInt(unsigned char *buf, int v) {
+		buf[0] = (v >> 0) & 0xFF;
+		buf[1] = (v >> 8) & 0xFF;
+		buf[2] = (v >> 16) & 0xFF;
+		buf[3] = (v >> 24) & 0xFF;
+	}
+
+private:
+	void frame();
+	bool commitReady() const;
+	bool checkSpace(int bytes);
+
+	char *const mBuf;
+	sem_t *const mReaderSem;
+	uint64_t mCommitTime;
+	sem_t mWriterSem;
+	const int mSize;
+	int mReadPos;
+	int mWritePos;
+	int mCommitPos;
+	bool mAvailable;
+	bool mIsDone;
+	const int32_t mCore;
+	const int32_t mBufType;
+
+	// Intentionally unimplemented
+	Buffer(const Buffer &);
+	Buffer &operator=(const Buffer &);
+};
+
+#endif // BUFFER_H
diff --git a/tools/gator/daemon/CCNDriver.cpp b/tools/gator/daemon/CCNDriver.cpp
new file mode 100644
index 000000000000..dd1a2b133842
--- /dev/null
+++ b/tools/gator/daemon/CCNDriver.cpp
@@ -0,0 +1,295 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "CCNDriver.h"
+
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+#include "k/perf_event.h"
+
+#include "Config.h"
+#include "DriverSource.h"
+#include "Logging.h"
+
+static const char TAG_CATEGORY[] = "category";
+static const char TAG_COUNTER_SET[] = "counter_set";
+static const char TAG_EVENT[] = "event";
+static const char TAG_OPTION[] = "option";
+static const char TAG_OPTION_SET[] = "option_set";
+
+static const char ATTR_AVERAGE_SELECTION[] = "average_selection";
+static const char ATTR_COUNTER[] = "counter";
+static const char ATTR_COUNTER_SET[] = "counter_set";
+static const char ATTR_COUNT[] = "count";
+static const char ATTR_DESCRIPTION[] = "description";
+static const char ATTR_DISPLAY[] = "display";
+static const char ATTR_EVENT[] = "event";
+static const char ATTR_EVENT_DELTA[] = "event_delta";
+static const char ATTR_NAME[] = "name";
+static const char ATTR_OPTION_SET[] = "option_set";
+static const char ATTR_TITLE[] = "title";
+static const char ATTR_UNITS[] = "units";
+
+static const char XP_REGION[] = "XP_Region";
+static const char HNF_REGION[] = "HN-F_Region";
+static const char RNI_REGION[] = "RN-I_Region";
+static const char SBAS_REGION[] = "SBAS_Region";
+static const char CCN_5XX[] = "CCN-5xx";
+#define ARM_CCN_5XX "ARM_CCN_5XX_"
+
+static const char *const VC_TYPES[] = { "REQ", "RSP", "SNP", "DAT" };
+static const char *const XP_EVENT_NAMES[] = { NULL, "H-bit", "S-bit", "P-Cnt", "TknV" };
+static const char *const XP_EVENT_DESCRIPTIONS[] = { NULL, "Set H-bit, signaled when this XP sets the H-bit.", "Set S-bit, signaled when this XP sets the S-bit.", "Set P-Cnt, signaled when this XP sets the P-Cnt. This is not applicable for the SNP VC.", "No TknV, signaled when this XP transmits a valid packet." };
+static const char *const HNF_EVENT_NAMES[] = { NULL, "Cache Miss", "L3 SF Cache Access", "Cache Fill", "POCQ Retry", "POCQ Reqs Recvd", "SF Hit", "SF Evictions", "Snoops Sent", "Snoops Broadcast", "L3 Eviction", "L3 Fill Invalid Way", "MC Retries", "MC Reqs", "QOS HH Retry" };
+static const char *const HNF_EVENT_DESCRIPTIONS[] = { NULL, "Counts the total cache misses. This is the first time lookup result, and is high priority.", "Counts the number of cache accesses. This is the first time access, and is high priority.", "Counts the total allocations in the HN L3 cache, and all cache line allocations to the L3 cache.", "Counts the number of requests that have been retried.", "Counts the number of requests received by HN.", "Counts the number of snoop filter hits.", "Counts the number of snoop filter evictions. Cache invalidations are initiated.", "Counts the number of snoops sent. Does not differentiate between broadcast or directed snoops.", "Counts the number of snoop broadcasts sent.", "Counts the number of L3 evictions.", "Counts the number of L3 fills to an invalid way.", "Counts the number of transactions retried by the memory controller.", "Counts the number of requests to the memory controller.", "Counts the number of times a highest-priority QoS class was retried at the HN-F." };
+static const char *const RNI_EVENT_NAMES[] = { NULL, "S0 RDataBeats", "S1 RDataBeats", "S2 RDataBeats", "RXDAT Flits received", "TXDAT Flits sent", "Total TXREQ Flits sent", "Retried TXREQ Flits sent", "RRT full", "WRT full", "Replayed TXREQ Flits" };
+static const char *const RNI_EVENT_DESCRIPTIONS[] = { NULL, "S0 RDataBeats.", "S1 RDataBeats.", "S2 RDataBeats.", "RXDAT Flits received.", "TXDAT Flits sent.", "Total TXREQ Flits sent.", "Retried TXREQ Flits sent.", "RRT full.", "WRT full.", "Replayed TXREQ Flits." };
+static const char *const SBAS_EVENT_NAMES[] = { NULL, "S0 RDataBeats", NULL, NULL, "RXDAT Flits received", "TXDAT Flits sent", "Total TXREQ Flits sent", "Retried TXREQ Flits sent", "RRT full", "WRT full", "Replayed TXREQ Flits" };
+static const char *const SBAS_EVENT_DESCRIPTIONS[] = { NULL, "S0 RDataBeats.", NULL, NULL, "RXDAT Flits received.", "TXDAT Flits sent.", "Total TXREQ Flits sent.", "Retried TXREQ Flits sent.", "RRT full.", "WRT full.", "Replayed TXREQ Flits." };
+
+// This class is used only to poll for CCN-5xx configuration and emit events XML for it. All other operations are handled by PerfDriver
+
+static int sys_perf_event_open(struct perf_event_attr *const attr, const pid_t pid, const int cpu, const int group_fd, const unsigned long flags) {
+	return syscall(__NR_perf_event_open, attr, pid, cpu, group_fd, flags);
+}
+
+static unsigned int getConfig(unsigned int node, unsigned int type, unsigned int event, unsigned int port, unsigned int vc) {
+  return
+    ((node  & 0xff) <<  0) |
+    ((type  & 0xff) <<  8) |
+    ((event & 0xff) << 16) |
+    ((port  & 0x03) << 24) |
+    ((vc    & 0x07) << 26) |
+    0;
+}
+
+static bool perfPoll(struct perf_event_attr *const pea) {
+	int fd = sys_perf_event_open(pea, -1, 0, -1, 0);
+	if (fd < 0) {
+		return false;
+	}
+	close(fd);
+	return true;
+}
+
+CCNDriver::CCNDriver() : mNodeTypes(NULL), mXpCount(0) {
+}
+
+CCNDriver::~CCNDriver() {
+	delete mNodeTypes;
+}
+
+bool CCNDriver::claimCounter(const Counter &) const {
+	// Handled by PerfDriver
+	return false;
+}
+
+void CCNDriver::resetCounters() {
+	// Handled by PerfDriver
+}
+
+void CCNDriver::setupCounter(Counter &) {
+	// Handled by PerfDriver
+}
+
+void CCNDriver::readEvents(mxml_node_t *const) {
+	struct stat st;
+	if (stat("/sys/bus/event_source/devices/ccn", &st) != 0) {
+		// Not found
+		return;
+	}
+
+	int type;
+	if (DriverSource::readIntDriver("/sys/bus/event_source/devices/ccn/type", &type) != 0) {
+		logg->logError(__FILE__, __LINE__, "Unable to read CCN-5xx type");
+		handleException();
+	}
+
+	// Detect number of xps
+	struct perf_event_attr pea;
+	memset(&pea, 0, sizeof(pea));
+	pea.type = type;
+	pea.size = sizeof(pea);
+
+	mXpCount = 1;
+	while (true) {
+		pea.config = getConfig(0, 0x08, 1, 0, 1) | mXpCount;
+		if (!perfPoll(&pea)) {
+			break;
+		}
+		mXpCount *= 2;
+	};
+	{
+		int lower = mXpCount/2 + 1;
+		while (lower < mXpCount) {
+			int mid = (lower + mXpCount)/2;
+			pea.config = getConfig(0, 0x08, 1, 0, 1) | mid;
+			if (perfPoll(&pea)) {
+				lower = mid + 1;
+			} else {
+				mXpCount = mid;
+			}
+		}
+	}
+
+	mNodeTypes = new NodeType[2*mXpCount];
+
+	// Detect node types
+	for (int i = 0; i < 2*mXpCount; ++i) {
+		pea.config = getConfig(0, 0x04, 1, 0, 0) | i;
+		if (perfPoll(&pea)) {
+			mNodeTypes[i] = NT_HNF;
+			continue;
+		}
+
+		pea.config = getConfig(0, 0x16, 1, 0, 0) | i;
+		if (perfPoll(&pea)) {
+			mNodeTypes[i] = NT_RNI;
+			continue;
+		}
+
+		pea.config = getConfig(0, 0x10, 1, 0, 0) | i;
+		if (perfPoll(&pea)) {
+			mNodeTypes[i] = NT_SBAS;
+			continue;
+		}
+
+		mNodeTypes[i] = NT_UNKNOWN;
+	}
+}
+
+int CCNDriver::writeCounters(mxml_node_t *const) const {
+	// Handled by PerfDriver
+	return 0;
+}
+
+void CCNDriver::writeEvents(mxml_node_t *const root) const {
+	mxml_node_t *const counter_set = mxmlNewElement(root, TAG_COUNTER_SET);
+	mxmlElementSetAttr(counter_set, ATTR_NAME, ARM_CCN_5XX "cnt");
+	mxmlElementSetAttr(counter_set, ATTR_COUNT, "8");
+
+	mxml_node_t *const category = mxmlNewElement(root, TAG_CATEGORY);
+	mxmlElementSetAttr(category, ATTR_NAME, CCN_5XX);
+	mxmlElementSetAttr(category, TAG_COUNTER_SET, ARM_CCN_5XX "cnt");
+
+	mxml_node_t *const clock_event = mxmlNewElement(category, TAG_EVENT);
+	mxmlElementSetAttr(clock_event, ATTR_COUNTER, ARM_CCN_5XX "ccnt");
+	mxmlElementSetAttr(clock_event, ATTR_EVENT, "0xff00");
+	mxmlElementSetAttr(clock_event, ATTR_TITLE, "CCN-5xx Clock");
+	mxmlElementSetAttr(clock_event, ATTR_NAME, "Cycles");
+	mxmlElementSetAttr(clock_event, ATTR_DISPLAY, "hertz");
+	mxmlElementSetAttr(clock_event, ATTR_UNITS, "Hz");
+	mxmlElementSetAttr(clock_event, ATTR_AVERAGE_SELECTION, "yes");
+	mxmlElementSetAttr(clock_event, ATTR_DESCRIPTION, "The number of core clock cycles");
+
+	mxml_node_t *const xp_option_set = mxmlNewElement(category, TAG_OPTION_SET);
+	mxmlElementSetAttr(xp_option_set, ATTR_NAME, XP_REGION);
+
+	for (int i = 0; i < mXpCount; ++i) {
+		mxml_node_t *const option = mxmlNewElement(xp_option_set, TAG_OPTION);
+		mxmlElementSetAttrf(option, ATTR_EVENT_DELTA, "0x%x", getConfig(i, 0, 0, 0, 0));
+		mxmlElementSetAttrf(option, ATTR_NAME, "XP %i", i);
+		mxmlElementSetAttrf(option, ATTR_DESCRIPTION, "Crosspoint %i", i);
+	}
+
+	for (int vc = 0; vc < ARRAY_LENGTH(VC_TYPES); ++vc) {
+		if (VC_TYPES[vc] == NULL) {
+			continue;
+		}
+		for (int bus = 0; bus < 2; ++bus) {
+			for (int eventId = 0; eventId < ARRAY_LENGTH(XP_EVENT_NAMES); ++eventId) {
+				if (XP_EVENT_NAMES[eventId] == NULL) {
+					continue;
+				}
+				mxml_node_t *const event = mxmlNewElement(category, TAG_EVENT);
+				mxmlElementSetAttrf(event, ATTR_EVENT, "0x%x", getConfig(0, 0x08, eventId, bus, vc));
+				mxmlElementSetAttr(event, ATTR_OPTION_SET, XP_REGION);
+				mxmlElementSetAttr(event, ATTR_TITLE, CCN_5XX);
+				mxmlElementSetAttrf(event, ATTR_NAME, "Bus %i: %s: %s", bus, VC_TYPES[vc], XP_EVENT_NAMES[eventId]);
+				mxmlElementSetAttrf(event, ATTR_DESCRIPTION, "Bus %i: %s: %s", bus, VC_TYPES[vc], XP_EVENT_DESCRIPTIONS[eventId]);
+			}
+		}
+	}
+
+	mxml_node_t *const hnf_option_set = mxmlNewElement(category, TAG_OPTION_SET);
+	mxmlElementSetAttr(hnf_option_set, ATTR_NAME, HNF_REGION);
+
+	for (int eventId = 0; eventId < ARRAY_LENGTH(HNF_EVENT_NAMES); ++eventId) {
+		if (HNF_EVENT_NAMES[eventId] == NULL) {
+			continue;
+		}
+		mxml_node_t *const event = mxmlNewElement(category, TAG_EVENT);
+		mxmlElementSetAttrf(event, ATTR_EVENT, "0x%x", getConfig(0, 0x04, eventId, 0, 0));
+		mxmlElementSetAttr(event, ATTR_OPTION_SET, HNF_REGION);
+		mxmlElementSetAttr(event, ATTR_TITLE, CCN_5XX);
+		mxmlElementSetAttr(event, ATTR_NAME, HNF_EVENT_NAMES[eventId]);
+		mxmlElementSetAttr(event, ATTR_DESCRIPTION, HNF_EVENT_DESCRIPTIONS[eventId]);
+	}
+
+	mxml_node_t *const rni_option_set = mxmlNewElement(category, TAG_OPTION_SET);
+	mxmlElementSetAttr(rni_option_set, ATTR_NAME, RNI_REGION);
+
+	for (int eventId = 0; eventId < ARRAY_LENGTH(RNI_EVENT_NAMES); ++eventId) {
+		if (RNI_EVENT_NAMES[eventId] == NULL) {
+			continue;
+		}
+		mxml_node_t *const event = mxmlNewElement(category, TAG_EVENT);
+		mxmlElementSetAttrf(event, ATTR_EVENT, "0x%x", getConfig(0, 0x16, eventId, 0, 0));
+		mxmlElementSetAttr(event, ATTR_OPTION_SET, RNI_REGION);
+		mxmlElementSetAttr(event, ATTR_TITLE, CCN_5XX);
+		mxmlElementSetAttr(event, ATTR_NAME, RNI_EVENT_NAMES[eventId]);
+		mxmlElementSetAttr(event, ATTR_DESCRIPTION, RNI_EVENT_DESCRIPTIONS[eventId]);
+	}
+
+	mxml_node_t *const sbas_option_set = mxmlNewElement(category, TAG_OPTION_SET);
+	mxmlElementSetAttr(sbas_option_set, ATTR_NAME, SBAS_REGION);
+
+	for (int eventId = 0; eventId < ARRAY_LENGTH(SBAS_EVENT_NAMES); ++eventId) {
+		if (SBAS_EVENT_NAMES[eventId] == NULL) {
+			continue;
+		}
+		mxml_node_t *const event = mxmlNewElement(category, TAG_EVENT);
+		mxmlElementSetAttrf(event, ATTR_EVENT, "0x%x", getConfig(0, 0x10, eventId, 0, 0));
+		mxmlElementSetAttr(event, ATTR_OPTION_SET, SBAS_REGION);
+		mxmlElementSetAttr(event, ATTR_TITLE, CCN_5XX);
+		mxmlElementSetAttr(event, ATTR_NAME, SBAS_EVENT_NAMES[eventId]);
+		mxmlElementSetAttr(event, ATTR_DESCRIPTION, SBAS_EVENT_DESCRIPTIONS[eventId]);
+	}
+
+	for (int i = 0; i < 2*mXpCount; ++i) {
+		switch (mNodeTypes[i]) {
+		case NT_HNF: {
+			mxml_node_t *const option = mxmlNewElement(hnf_option_set, TAG_OPTION);
+			mxmlElementSetAttrf(option, ATTR_EVENT_DELTA, "0x%x", getConfig(i, 0, 0, 0, 0));
+			mxmlElementSetAttrf(option, ATTR_NAME, "HN-F %i", i);
+			mxmlElementSetAttrf(option, ATTR_DESCRIPTION, "Fully-coherent Home Node %i", i);
+			break;
+		}
+		case NT_RNI: {
+			mxml_node_t *const option = mxmlNewElement(rni_option_set, TAG_OPTION);
+			mxmlElementSetAttrf(option, ATTR_EVENT_DELTA, "0x%x", getConfig(i, 0, 0, 0, 0));
+			mxmlElementSetAttrf(option, ATTR_NAME, "RN-I %i", i);
+			mxmlElementSetAttrf(option, ATTR_DESCRIPTION, "I/O-coherent Requesting Node %i", i);
+			break;
+		}
+		case NT_SBAS: {
+			mxml_node_t *const option = mxmlNewElement(sbas_option_set, TAG_OPTION);
+			mxmlElementSetAttrf(option, ATTR_EVENT_DELTA, "0x%x", getConfig(i, 0, 0, 0, 0));
+			mxmlElementSetAttrf(option, ATTR_NAME, "SBAS %i", i);
+			mxmlElementSetAttrf(option, ATTR_DESCRIPTION, "ACE master to CHI protocol bridge %i", i);
+			break;
+		}
+		default:
+			continue;
+		}
+	}
+}
diff --git a/tools/gator/daemon/CCNDriver.h b/tools/gator/daemon/CCNDriver.h
new file mode 100644
index 000000000000..fb4c717e969a
--- /dev/null
+++ b/tools/gator/daemon/CCNDriver.h
@@ -0,0 +1,43 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef CCNDRIVER_H
+#define CCNDRIVER_H
+
+#include "Driver.h"
+
+class CCNDriver : public Driver {
+public:
+	CCNDriver();
+	~CCNDriver();
+
+	bool claimCounter(const Counter &counter) const;
+	void resetCounters();
+	void setupCounter(Counter &counter);
+
+	void readEvents(mxml_node_t *const);
+	int writeCounters(mxml_node_t *const root) const;
+	void writeEvents(mxml_node_t *const) const;
+
+private:
+	enum NodeType {
+		NT_UNKNOWN,
+		NT_HNF,
+		NT_RNI,
+		NT_SBAS,
+	};
+
+	NodeType *mNodeTypes;
+	int mXpCount;
+
+	// Intentionally unimplemented
+	CCNDriver(const CCNDriver &);
+	CCNDriver &operator=(const CCNDriver &);
+};
+
+#endif // CCNDRIVER_H
diff --git a/tools/gator/daemon/CPUFreqDriver.cpp b/tools/gator/daemon/CPUFreqDriver.cpp
new file mode 100644
index 000000000000..41f9d6f2b3f4
--- /dev/null
+++ b/tools/gator/daemon/CPUFreqDriver.cpp
@@ -0,0 +1,58 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "CPUFreqDriver.h"
+
+#include "Buffer.h"
+#include "DriverSource.h"
+#include "Logging.h"
+#include "SessionData.h"
+
+CPUFreqDriver::CPUFreqDriver() : mPrev() {
+}
+
+CPUFreqDriver::~CPUFreqDriver() {
+}
+
+void CPUFreqDriver::readEvents(mxml_node_t *const) {
+	// Only for use with perf
+	if (!gSessionData->perf.isSetup()) {
+		return;
+	}
+
+	setCounters(new DriverCounter(getCounters(), strdup("Linux_power_cpu_freq")));
+}
+
+void CPUFreqDriver::read(Buffer *const buffer) {
+	char buf[64];
+	const DriverCounter *const counter = getCounters();
+	if ((counter == NULL) || !counter->isEnabled()) {
+		return;
+	}
+
+	const int key = getCounters()->getKey();
+	bool resetCores = false;
+	for (int i = 0; i < gSessionData->mCores; ++i) {
+		snprintf(buf, sizeof(buf), "/sys/devices/system/cpu/cpu%i/cpufreq/cpuinfo_cur_freq", i);
+		int64_t freq;
+		if (DriverSource::readInt64Driver(buf, &freq) != 0) {
+			freq = 0;
+		}
+		if (mPrev[i] != freq) {
+			mPrev[i] = freq;
+			// Change cores
+			buffer->event64(2, i);
+			resetCores = true;
+			buffer->event64(key, 1000*freq);
+		}
+	}
+	if (resetCores) {
+		// Revert cores, UserSpaceSource is all on core 0
+		buffer->event64(2, 0);
+	}
+}
diff --git a/tools/gator/daemon/CPUFreqDriver.h b/tools/gator/daemon/CPUFreqDriver.h
new file mode 100644
index 000000000000..ad8c9aaa9e7d
--- /dev/null
+++ b/tools/gator/daemon/CPUFreqDriver.h
@@ -0,0 +1,34 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef CPUFREQDRIVER_H
+#define CPUFREQDRIVER_H
+
+#include "Config.h"
+#include "Driver.h"
+
+class CPUFreqDriver : public PolledDriver {
+private:
+	typedef PolledDriver super;
+
+public:
+	CPUFreqDriver();
+	~CPUFreqDriver();
+
+	void readEvents(mxml_node_t *const root);
+	void read(Buffer *const buffer);
+
+private:
+	int64_t mPrev[NR_CPUS];
+
+	// Intentionally unimplemented
+	CPUFreqDriver(const CPUFreqDriver &);
+	CPUFreqDriver &operator=(const CPUFreqDriver &);
+};
+
+#endif // CPUFREQDRIVER_H
diff --git a/tools/gator/daemon/CapturedXML.cpp b/tools/gator/daemon/CapturedXML.cpp
new file mode 100644
index 000000000000..0b5802c893bb
--- /dev/null
+++ b/tools/gator/daemon/CapturedXML.cpp
@@ -0,0 +1,145 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "CapturedXML.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <dirent.h>
+
+#include "SessionData.h"
+#include "Logging.h"
+#include "OlyUtility.h"
+
+CapturedXML::CapturedXML() {
+}
+
+CapturedXML::~CapturedXML() {
+}
+
+mxml_node_t* CapturedXML::getTree(bool includeTime) {
+	mxml_node_t *xml;
+	mxml_node_t *captured;
+	mxml_node_t *target;
+	int x;
+
+	xml = mxmlNewXML("1.0");
+
+	captured = mxmlNewElement(xml, "captured");
+	mxmlElementSetAttr(captured, "version", "1");
+	if (gSessionData->perf.isSetup()) {
+		mxmlElementSetAttr(captured, "type", "Perf");
+		mxmlElementSetAttr(captured, "perf_beta", "yes");
+	}
+	mxmlElementSetAttrf(captured, "protocol", "%d", PROTOCOL_VERSION);
+	if (includeTime) { // Send the following only after the capture is complete
+		if (time(NULL) > 1267000000) { // If the time is reasonable (after Feb 23, 2010)
+			mxmlElementSetAttrf(captured, "created", "%lu", time(NULL)); // Valid until the year 2038
+		}
+	}
+
+	target = mxmlNewElement(captured, "target");
+	mxmlElementSetAttr(target, "name", gSessionData->mCoreName);
+	mxmlElementSetAttrf(target, "sample_rate", "%d", gSessionData->mSampleRate);
+	mxmlElementSetAttrf(target, "cores", "%d", gSessionData->mCores);
+	mxmlElementSetAttrf(target, "cpuid", "0x%x", gSessionData->mMaxCpuId);
+
+	if (!gSessionData->mOneShot && (gSessionData->mSampleRate > 0)) {
+		mxmlElementSetAttr(target, "supports_live", "yes");
+	}
+
+	if (gSessionData->mLocalCapture) {
+		mxmlElementSetAttr(target, "local_capture", "yes");
+	}
+
+	mxml_node_t *counters = NULL;
+	for (x = 0; x < MAX_PERFORMANCE_COUNTERS; x++) {
+		const Counter & counter = gSessionData->mCounters[x];
+		if (counter.isEnabled()) {
+			if (counters == NULL) {
+				counters = mxmlNewElement(captured, "counters");
+			}
+			mxml_node_t *const node = mxmlNewElement(counters, "counter");
+			mxmlElementSetAttrf(node, "key", "0x%x", counter.getKey());
+			mxmlElementSetAttr(node, "type", counter.getType());
+			if (counter.getEvent() != -1) {
+				mxmlElementSetAttrf(node, "event", "0x%x", counter.getEvent());
+			}
+			if (counter.getCount() > 0) {
+				mxmlElementSetAttrf(node, "count", "%d", counter.getCount());
+			}
+			if (counter.getCores() > 0) {
+				mxmlElementSetAttrf(node, "cores", "%d", counter.getCores());
+			}
+		}
+	}
+
+	return xml;
+}
+
+char* CapturedXML::getXML(bool includeTime) {
+	char* xml_string;
+	mxml_node_t *xml = getTree(includeTime);
+	xml_string = mxmlSaveAllocString(xml, mxmlWhitespaceCB);
+	mxmlDelete(xml);
+	return xml_string;
+}
+
+void CapturedXML::write(char* path) {
+	char file[PATH_MAX];
+
+	// Set full path
+	snprintf(file, PATH_MAX, "%s/captured.xml", path);
+
+	char* xml = getXML(true);
+	if (util->writeToDisk(file, xml) < 0) {
+		logg->logError(__FILE__, __LINE__, "Error writing %s\nPlease verify the path.", file);
+		handleException();
+	}
+
+	free(xml);
+}
+
+// whitespace callback utility function used with mini-xml
+const char * mxmlWhitespaceCB(mxml_node_t *node, int loc) {
+	const char *name;
+
+	name = mxmlGetElement(node);
+
+	if (loc == MXML_WS_BEFORE_OPEN) {
+		// Single indentation
+		if (!strcmp(name, "target") || !strcmp(name, "counters"))
+			return "\n  ";
+
+		// Double indentation
+		if (!strcmp(name, "counter"))
+			return "\n    ";
+
+		// Avoid a carriage return on the first line of the xml file
+		if (!strncmp(name, "?xml", 4))
+			return NULL;
+
+		// Default - no indentation
+		return "\n";
+	}
+
+	if (loc == MXML_WS_BEFORE_CLOSE) {
+		// No indentation
+		if (!strcmp(name, "captured"))
+			return "\n";
+
+		// Single indentation
+		if (!strcmp(name, "counters"))
+			return "\n  ";
+
+		// Default - no carriage return
+		return NULL;
+	}
+
+	return NULL;
+}
diff --git a/tools/gator/daemon/CapturedXML.h b/tools/gator/daemon/CapturedXML.h
new file mode 100644
index 000000000000..b704f6e53bb5
--- /dev/null
+++ b/tools/gator/daemon/CapturedXML.h
@@ -0,0 +1,26 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef __CAPTURED_XML_H__
+#define __CAPTURED_XML_H__
+
+#include "mxml/mxml.h"
+
+class CapturedXML {
+public:
+	CapturedXML();
+	~CapturedXML();
+	char* getXML(bool includeTime); // the string should be freed by the caller
+	void write(char* path);
+private:
+	mxml_node_t* getTree(bool includeTime);
+};
+
+const char * mxmlWhitespaceCB(mxml_node_t *node, int where);
+
+#endif //__CAPTURED_XML_H__
diff --git a/tools/gator/daemon/Child.cpp b/tools/gator/daemon/Child.cpp
new file mode 100644
index 000000000000..6b5bbb3bf6af
--- /dev/null
+++ b/tools/gator/daemon/Child.cpp
@@ -0,0 +1,392 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "Child.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <signal.h>
+#include <unistd.h>
+#include <sys/prctl.h>
+
+#include "CapturedXML.h"
+#include "Command.h"
+#include "ConfigurationXML.h"
+#include "Driver.h"
+#include "DriverSource.h"
+#include "ExternalSource.h"
+#include "FtraceSource.h"
+#include "LocalCapture.h"
+#include "Logging.h"
+#include "OlySocket.h"
+#include "OlyUtility.h"
+#include "PerfSource.h"
+#include "Sender.h"
+#include "SessionData.h"
+#include "StreamlineSetup.h"
+#include "UserSpaceSource.h"
+
+static sem_t haltPipeline, senderThreadStarted, startProfile, senderSem; // Shared by Child and spawned threads
+static Source *primarySource = NULL;
+static Source *externalSource = NULL;
+static Source *userSpaceSource = NULL;
+static Source *ftraceSource = NULL;
+static Sender* sender = NULL;        // Shared by Child.cpp and spawned threads
+Child* child = NULL;                 // shared by Child.cpp and main.cpp
+
+extern void cleanUp();
+void handleException() {
+	if (child && child->numExceptions++ > 0) {
+		// it is possible one of the below functions itself can cause an exception, thus allow only one exception
+		logg->logMessage("Received multiple exceptions, terminating the child");
+		exit(1);
+	}
+	fprintf(stderr, "%s", logg->getLastError());
+
+	if (child && child->socket) {
+		if (sender) {
+			// send the error, regardless of the command sent by Streamline
+			sender->writeData(logg->getLastError(), strlen(logg->getLastError()), RESPONSE_ERROR);
+
+			// cannot close the socket before Streamline issues the command, so wait for the command before exiting
+			if (gSessionData->mWaitingOnCommand) {
+				char discard;
+				child->socket->receiveNBytes(&discard, 1);
+			}
+
+			// Ensure all data is flushed
+			child->socket->shutdownConnection();
+
+			// this indirectly calls close socket which will ensure the data has been sent
+			delete sender;
+		}
+	}
+
+	if (gSessionData->mLocalCapture)
+		cleanUp();
+
+	exit(1);
+}
+
+// CTRL C Signal Handler for child process
+static void child_handler(int signum) {
+	static bool beenHere = false;
+	if (beenHere == true) {
+		logg->logMessage("Gator is being forced to shut down.");
+		exit(1);
+	}
+	beenHere = true;
+	logg->logMessage("Gator is shutting down.");
+	if (signum == SIGALRM || !primarySource) {
+		exit(1);
+	} else {
+		child->endSession();
+		alarm(5); // Safety net in case endSession does not complete within 5 seconds
+	}
+}
+
+static void *durationThread(void *) {
+	prctl(PR_SET_NAME, (unsigned long)&"gatord-duration", 0, 0, 0);
+	sem_wait(&startProfile);
+	if (gSessionData->mSessionIsActive) {
+		// Time out after duration seconds
+		// Add a second for host-side filtering
+		sleep(gSessionData->mDuration + 1);
+		if (gSessionData->mSessionIsActive) {
+			logg->logMessage("Duration expired.");
+			child->endSession();
+		}
+	}
+	logg->logMessage("Exit duration thread");
+	return 0;
+}
+
+static void *stopThread(void *) {
+	OlySocket* socket = child->socket;
+
+	prctl(PR_SET_NAME, (unsigned long)&"gatord-stopper", 0, 0, 0);
+	while (gSessionData->mSessionIsActive) {
+		// This thread will stall until the APC_STOP or PING command is received over the socket or the socket is disconnected
+		unsigned char header[5];
+		const int result = socket->receiveNBytes((char*)&header, sizeof(header));
+		const char type = header[0];
+		const int length = (header[1] << 0) | (header[2] << 8) | (header[3] << 16) | (header[4] << 24);
+		if (result == -1) {
+			child->endSession();
+		} else if (result > 0) {
+			if ((type != COMMAND_APC_STOP) && (type != COMMAND_PING)) {
+				logg->logMessage("INVESTIGATE: Received unknown command type %d", type);
+			} else {
+				// verify a length of zero
+				if (length == 0) {
+					if (type == COMMAND_APC_STOP) {
+						logg->logMessage("Stop command received.");
+						child->endSession();
+					} else {
+						// Ping is used to make sure gator is alive and requires an ACK as the response
+						logg->logMessage("Ping command received.");
+						sender->writeData(NULL, 0, RESPONSE_ACK);
+					}
+				} else {
+					logg->logMessage("INVESTIGATE: Received stop command but with length = %d", length);
+				}
+			}
+		}
+	}
+
+	logg->logMessage("Exit stop thread");
+	return 0;
+}
+
+static void *senderThread(void *) {
+	char end_sequence[] = {RESPONSE_APC_DATA, 0, 0, 0, 0};
+
+	sem_post(&senderThreadStarted);
+	prctl(PR_SET_NAME, (unsigned long)&"gatord-sender", 0, 0, 0);
+	sem_wait(&haltPipeline);
+
+	while (!primarySource->isDone() ||
+	       !externalSource->isDone() ||
+	       (userSpaceSource != NULL && !userSpaceSource->isDone()) ||
+	       (ftraceSource != NULL && !ftraceSource->isDone())) {
+		sem_wait(&senderSem);
+
+		primarySource->write(sender);
+		externalSource->write(sender);
+		if (userSpaceSource != NULL) {
+			userSpaceSource->write(sender);
+		}
+		if (ftraceSource != NULL) {
+			ftraceSource->write(sender);
+		}
+	}
+
+	// write end-of-capture sequence
+	if (!gSessionData->mLocalCapture) {
+		sender->writeData(end_sequence, sizeof(end_sequence), RESPONSE_APC_DATA);
+	}
+
+	logg->logMessage("Exit sender thread");
+	return 0;
+}
+
+Child::Child() {
+	initialization();
+	gSessionData->mLocalCapture = true;
+}
+
+Child::Child(OlySocket* sock, int conn) {
+	initialization();
+	socket = sock;
+	mNumConnections = conn;
+}
+
+Child::~Child() {
+}
+
+void Child::initialization() {
+	// Set up different handlers for signals
+	gSessionData->mSessionIsActive = true;
+	signal(SIGINT, child_handler);
+	signal(SIGTERM, child_handler);
+	signal(SIGABRT, child_handler);
+	signal(SIGALRM, child_handler);
+	socket = NULL;
+	numExceptions = 0;
+	mNumConnections = 0;
+
+	// Initialize semaphores
+	sem_init(&senderThreadStarted, 0, 0);
+	sem_init(&startProfile, 0, 0);
+	sem_init(&senderSem, 0, 0);
+}
+
+void Child::endSession() {
+	gSessionData->mSessionIsActive = false;
+	primarySource->interrupt();
+	externalSource->interrupt();
+	if (userSpaceSource != NULL) {
+		userSpaceSource->interrupt();
+	}
+	if (ftraceSource != NULL) {
+		ftraceSource->interrupt();
+	}
+	sem_post(&haltPipeline);
+}
+
+void Child::run() {
+	LocalCapture* localCapture = NULL;
+	pthread_t durationThreadID, stopThreadID, senderThreadID;
+
+	prctl(PR_SET_NAME, (unsigned long)&"gatord-child", 0, 0, 0);
+
+	// Disable line wrapping when generating xml files; carriage returns and indentation to be added manually
+	mxmlSetWrapMargin(0);
+
+	// Instantiate the Sender - must be done first, after which error messages can be sent
+	sender = new Sender(socket);
+
+	if (mNumConnections > 1) {
+		logg->logError(__FILE__, __LINE__, "Session already in progress");
+		handleException();
+	}
+
+	// Populate gSessionData with the configuration
+	{ ConfigurationXML configuration; }
+
+	// Set up the driver; must be done after gSessionData->mPerfCounterType[] is populated
+	if (!gSessionData->perf.isSetup()) {
+		primarySource = new DriverSource(&senderSem, &startProfile);
+	} else {
+		primarySource = new PerfSource(&senderSem, &startProfile);
+	}
+
+	// Initialize all drivers
+	for (Driver *driver = Driver::getHead(); driver != NULL; driver = driver->getNext()) {
+		driver->resetCounters();
+	}
+
+	// Set up counters using the associated driver's setup function
+	for (int i = 0; i < MAX_PERFORMANCE_COUNTERS; i++) {
+		Counter & counter = gSessionData->mCounters[i];
+		if (counter.isEnabled()) {
+			counter.getDriver()->setupCounter(counter);
+		}
+	}
+
+	// Start up and parse session xml
+	if (socket) {
+		// Respond to Streamline requests
+		StreamlineSetup ss(socket);
+	} else {
+		char* xmlString;
+		xmlString = util->readFromDisk(gSessionData->mSessionXMLPath);
+		if (xmlString == 0) {
+			logg->logError(__FILE__, __LINE__, "Unable to read session xml file: %s", gSessionData->mSessionXMLPath);
+			handleException();
+		}
+		gSessionData->parseSessionXML(xmlString);
+		localCapture = new LocalCapture();
+		localCapture->createAPCDirectory(gSessionData->mTargetPath);
+		localCapture->copyImages(gSessionData->mImages);
+		localCapture->write(xmlString);
+		sender->createDataFile(gSessionData->mAPCDir);
+		free(xmlString);
+	}
+
+	if (gSessionData->kmod.isMaliCapture() && (gSessionData->mSampleRate == 0)) {
+		logg->logError(__FILE__, __LINE__, "Mali counters are not supported with Sample Rate: None.");
+		handleException();
+	}
+
+	// Must be after session XML is parsed
+	if (!primarySource->prepare()) {
+		if (gSessionData->perf.isSetup()) {
+			logg->logError(__FILE__, __LINE__, "Unable to prepare gator driver for capture");
+		} else {
+			logg->logError(__FILE__, __LINE__, "Unable to communicate with the perf API, please ensure that CONFIG_TRACING and CONFIG_CONTEXT_SWITCH_TRACER are enabled. Please refer to README_Streamline.txt for more information.");
+		}
+		handleException();
+	}
+
+	// Sender thread shall be halted until it is signaled for one shot mode
+	sem_init(&haltPipeline, 0, gSessionData->mOneShot ? 0 : 2);
+
+	// Must be initialized before senderThread is started as senderThread checks externalSource
+	externalSource = new ExternalSource(&senderSem);
+	if (!externalSource->prepare()) {
+		logg->logError(__FILE__, __LINE__, "Unable to prepare external source for capture");
+		handleException();
+	}
+	externalSource->start();
+
+	// Create the duration, stop, and sender threads
+	bool thread_creation_success = true;
+	if (gSessionData->mDuration > 0 && pthread_create(&durationThreadID, NULL, durationThread, NULL)) {
+		thread_creation_success = false;
+	} else if (socket && pthread_create(&stopThreadID, NULL, stopThread, NULL)) {
+		thread_creation_success = false;
+	} else if (pthread_create(&senderThreadID, NULL, senderThread, NULL)) {
+		thread_creation_success = false;
+	}
+
+	bool startUSSource = false;
+	for (int i = 0; i < ARRAY_LENGTH(gSessionData->usDrivers); ++i) {
+		if (gSessionData->usDrivers[i]->countersEnabled()) {
+			startUSSource = true;
+		}
+	}
+	if (startUSSource) {
+		userSpaceSource = new UserSpaceSource(&senderSem);
+		if (!userSpaceSource->prepare()) {
+			logg->logError(__FILE__, __LINE__, "Unable to prepare userspace source for capture");
+			handleException();
+		}
+		userSpaceSource->start();
+	}
+
+	if (gSessionData->ftraceDriver.countersEnabled()) {
+		ftraceSource = new FtraceSource(&senderSem);
+		if (!ftraceSource->prepare()) {
+			logg->logError(__FILE__, __LINE__, "Unable to prepare userspace source for capture");
+			handleException();
+		}
+		ftraceSource->start();
+	}
+
+	if (gSessionData->mAllowCommands && (gSessionData->mCaptureCommand != NULL)) {
+		pthread_t thread;
+		if (pthread_create(&thread, NULL, commandThread, NULL)) {
+			thread_creation_success = false;
+		}
+	}
+
+	if (!thread_creation_success) {
+		logg->logError(__FILE__, __LINE__, "Failed to create gator threads");
+		handleException();
+	}
+
+	// Wait until thread has started
+	sem_wait(&senderThreadStarted);
+
+	// Start profiling
+	primarySource->run();
+
+	if (ftraceSource != NULL) {
+		ftraceSource->join();
+	}
+	if (userSpaceSource != NULL) {
+		userSpaceSource->join();
+	}
+	externalSource->join();
+
+	// Wait for the other threads to exit
+	pthread_join(senderThreadID, NULL);
+
+	// Shutting down the connection should break the stop thread which is stalling on the socket recv() function
+	if (socket) {
+		logg->logMessage("Waiting on stop thread");
+		socket->shutdownConnection();
+		pthread_join(stopThreadID, NULL);
+	}
+
+	// Write the captured xml file
+	if (gSessionData->mLocalCapture) {
+		CapturedXML capturedXML;
+		capturedXML.write(gSessionData->mAPCDir);
+	}
+
+	logg->logMessage("Profiling ended.");
+
+	delete ftraceSource;
+	delete userSpaceSource;
+	delete externalSource;
+	delete primarySource;
+	delete sender;
+	delete localCapture;
+}
diff --git a/tools/gator/daemon/Child.h b/tools/gator/daemon/Child.h
new file mode 100644
index 000000000000..cc78202ceb5c
--- /dev/null
+++ b/tools/gator/daemon/Child.h
@@ -0,0 +1,33 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef __CHILD_H__
+#define __CHILD_H__
+
+class OlySocket;
+
+class Child {
+public:
+	Child();
+	Child(OlySocket* sock, int numConnections);
+	~Child();
+	void run();
+	OlySocket *socket;
+	void endSession();
+	int numExceptions;
+private:
+	int mNumConnections;
+
+	void initialization();
+
+	// Intentionally unimplemented
+	Child(const Child &);
+	Child &operator=(const Child &);
+};
+
+#endif //__CHILD_H__
diff --git a/tools/gator/daemon/Command.cpp b/tools/gator/daemon/Command.cpp
new file mode 100644
index 000000000000..28d73cf5a905
--- /dev/null
+++ b/tools/gator/daemon/Command.cpp
@@ -0,0 +1,172 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "Command.h"
+
+#include <fcntl.h>
+#include <pwd.h>
+#include <stdio.h>
+#include <sys/prctl.h>
+#include <sys/resource.h>
+#include <sys/stat.h>
+#include <sys/syscall.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+
+#include "Logging.h"
+#include "SessionData.h"
+
+static int getUid(const char *const name, char *const shPath, const char *const tmpDir) {
+	// Lookups may fail when using a different libc or a statically compiled executable
+	char gatorTemp[32];
+	snprintf(gatorTemp, sizeof(gatorTemp), "%s/gator_temp", tmpDir);
+
+	const int fd = open(gatorTemp, 600, O_CREAT | O_CLOEXEC);
+	if (fd < 0) {
+		return -1;
+	}
+	close(fd);
+
+	char cmd[128];
+	snprintf(cmd, sizeof(cmd), "chown %s %s || rm %s", name, gatorTemp, gatorTemp);
+
+	const int pid = fork();
+	if (pid < 0) {
+		logg->logError(__FILE__, __LINE__, "fork failed");
+		handleException();
+	}
+	if (pid == 0) {
+		char cargv1[] = "-c";
+		char *cargv[] = {
+			shPath,
+			cargv1,
+			cmd,
+			NULL,
+		};
+
+		execv(cargv[0], cargv);
+		exit(-1);
+	}
+	while ((waitpid(pid, NULL, 0) < 0) && (errno == EINTR));
+
+	struct stat st;
+	int result = -1;
+	if (stat(gatorTemp, &st) == 0) {
+		result = st.st_uid;
+	}
+	unlink(gatorTemp);
+	return result;
+}
+
+static int getUid(const char *const name) {
+	// Look up the username
+	struct passwd *const user = getpwnam(name);
+	if (user != NULL) {
+		return user->pw_uid;
+	}
+
+
+	// Are we on Linux
+	char cargv0l[] = "/bin/sh";
+	if ((access(cargv0l, X_OK) == 0) && (access("/tmp", W_OK) == 0)) {
+		return getUid(name, cargv0l, "/tmp");
+	}
+
+	// Are we on android
+	char cargv0a[] = "/system/bin/sh";
+	if ((access(cargv0a, X_OK) == 0) && (access("/data", W_OK) == 0)) {
+		return getUid(name, cargv0a, "/data");
+	}
+
+	return -1;
+}
+
+void *commandThread(void *) {
+	prctl(PR_SET_NAME, (unsigned long)&"gatord-command", 0, 0, 0);
+
+	const char *const name = gSessionData->mCaptureUser == NULL ? "nobody" : gSessionData->mCaptureUser;
+	const int uid = getUid(name);
+	if (uid < 0) {
+		logg->logError(__FILE__, __LINE__, "Unable to lookup the user %s, please double check that the user exists", name);
+		handleException();
+	}
+
+	sleep(3);
+
+	char buf[128];
+	int pipefd[2];
+	if (pipe_cloexec(pipefd) != 0) {
+		logg->logError(__FILE__, __LINE__, "pipe failed");
+		handleException();
+	}
+
+	const int pid = fork();
+	if (pid < 0) {
+		logg->logError(__FILE__, __LINE__, "fork failed");
+		handleException();
+	}
+	if (pid == 0) {
+		char cargv0l[] = "/bin/sh";
+		char cargv0a[] = "/system/bin/sh";
+		char cargv1[] = "-c";
+		char *cargv[] = {
+			cargv0l,
+			cargv1,
+			gSessionData->mCaptureCommand,
+			NULL,
+		};
+
+		buf[0] = '\0';
+		close(pipefd[0]);
+
+		// Gator runs at a high priority, reset the priority to the default
+		if (setpriority(PRIO_PROCESS, syscall(__NR_gettid), 0) == -1) {
+			snprintf(buf, sizeof(buf), "setpriority failed");
+			goto fail_exit;
+		}
+
+		if (setuid(uid) != 0) {
+			snprintf(buf, sizeof(buf), "setuid failed");
+			goto fail_exit;
+		}
+
+		{
+			const char *const path = gSessionData->mCaptureWorkingDir == NULL ? "/" : gSessionData->mCaptureWorkingDir;
+			if (chdir(path) != 0) {
+				snprintf(buf, sizeof(buf), "Unable to cd to %s, please verify the directory exists and is accessable to %s", path, name);
+				goto fail_exit;
+			}
+		}
+
+		execv(cargv[0], cargv);
+		cargv[0] = cargv0a;
+		execv(cargv[0], cargv);
+		snprintf(buf, sizeof(buf), "execv failed");
+
+	fail_exit:
+		if (buf[0] != '\0') {
+			const ssize_t bytes = write(pipefd[1], buf, sizeof(buf));
+			// Can't do anything if this fails
+			(void)bytes;
+		}
+
+		exit(-1);
+	}
+
+	close(pipefd[1]);
+	const ssize_t bytes = read(pipefd[0], buf, sizeof(buf));
+	if (bytes > 0) {
+		logg->logError(__FILE__, __LINE__, buf);
+		handleException();
+	}
+	close(pipefd[0]);
+
+	return NULL;
+}
diff --git a/tools/gator/daemon/Command.h b/tools/gator/daemon/Command.h
new file mode 100644
index 000000000000..17244b7aaebc
--- /dev/null
+++ b/tools/gator/daemon/Command.h
@@ -0,0 +1,14 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef COMMAND_H
+#define COMMAND_H
+
+void *commandThread(void *);
+
+#endif // COMMAND_H
diff --git a/tools/gator/daemon/Config.h b/tools/gator/daemon/Config.h
new file mode 100644
index 000000000000..bee383a1c797
--- /dev/null
+++ b/tools/gator/daemon/Config.h
@@ -0,0 +1,28 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef CONFIG_H
+#define CONFIG_H
+
+#define ARRAY_LENGTH(A) static_cast<int>(sizeof(A)/sizeof((A)[0]))
+#define ACCESS_ONCE(x) (*(volatile typeof(x)*)&(x))
+
+#define MAX_PERFORMANCE_COUNTERS 50
+#define NR_CPUS 32
+
+template<typename T>
+static inline T min(const T a, const T b) {
+	return (a < b ? a : b);
+}
+
+template<typename T>
+static inline T max(const T a, const T b) {
+	return (a > b ? a : b);
+}
+
+#endif // CONFIG_H
diff --git a/tools/gator/daemon/ConfigurationXML.cpp b/tools/gator/daemon/ConfigurationXML.cpp
new file mode 100644
index 000000000000..6590dd389196
--- /dev/null
+++ b/tools/gator/daemon/ConfigurationXML.cpp
@@ -0,0 +1,217 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "ConfigurationXML.h"
+
+#include <string.h>
+#include <stdlib.h>
+#include <dirent.h>
+
+#include "Driver.h"
+#include "Logging.h"
+#include "OlyUtility.h"
+#include "SessionData.h"
+
+static const char* ATTR_COUNTER            = "counter";
+static const char* ATTR_REVISION           = "revision";
+static const char* ATTR_EVENT              = "event";
+static const char* ATTR_COUNT              = "count";
+static const char* ATTR_CORES              = "cores";
+
+ConfigurationXML::ConfigurationXML() {
+	const char * configuration_xml;
+	unsigned int configuration_xml_len;
+	getDefaultConfigurationXml(configuration_xml, configuration_xml_len);
+
+	char path[PATH_MAX];
+
+	getPath(path);
+	mConfigurationXML = util->readFromDisk(path);
+
+	for (int retryCount = 0; retryCount < 2; ++retryCount) {
+		if (mConfigurationXML == NULL) {
+			logg->logMessage("Unable to locate configuration.xml, using default in binary");
+			// null-terminate configuration_xml
+			mConfigurationXML = (char*)malloc(configuration_xml_len + 1);
+			memcpy(mConfigurationXML, (const void*)configuration_xml, configuration_xml_len);
+			mConfigurationXML[configuration_xml_len] = 0;
+		}
+
+		int ret = parse(mConfigurationXML);
+		if (ret == 1) {
+			remove();
+
+			// Free the current configuration and reload
+			free((void*)mConfigurationXML);
+			mConfigurationXML = NULL;
+			continue;
+		}
+
+		break;
+	}
+
+	validate();
+}
+
+ConfigurationXML::~ConfigurationXML() {
+	if (mConfigurationXML) {
+		free((void*)mConfigurationXML);
+	}
+}
+
+int ConfigurationXML::parse(const char* configurationXML) {
+	mxml_node_t *tree, *node;
+	int ret;
+
+	// clear counter overflow
+	gSessionData->mCounterOverflow = 0;
+	gSessionData->mIsEBS = false;
+	mIndex = 0;
+
+	// disable all counters prior to parsing the configuration xml
+	for (int i = 0; i < MAX_PERFORMANCE_COUNTERS; i++) {
+		gSessionData->mCounters[i].setEnabled(false);
+	}
+
+	tree = mxmlLoadString(NULL, configurationXML, MXML_NO_CALLBACK);
+
+	node = mxmlGetFirstChild(tree);
+	while (node && mxmlGetType(node) != MXML_ELEMENT)
+		node = mxmlWalkNext(node, tree, MXML_NO_DESCEND);
+
+	ret = configurationsTag(node);
+
+	node = mxmlGetFirstChild(node);
+	while (node) {
+		if (mxmlGetType(node) != MXML_ELEMENT) {
+			node = mxmlWalkNext(node, tree, MXML_NO_DESCEND);
+			continue;
+		}
+		configurationTag(node);
+		node = mxmlWalkNext(node, tree, MXML_NO_DESCEND);
+	}
+
+	mxmlDelete(tree);
+
+	return ret;
+}
+
+void ConfigurationXML::validate(void) {
+	for (int i = 0; i < MAX_PERFORMANCE_COUNTERS; i++) {
+		const Counter & counter = gSessionData->mCounters[i];
+		if (counter.isEnabled()) {
+			if (strcmp(counter.getType(), "") == 0) {
+				logg->logError(__FILE__, __LINE__, "Invalid required attribute in configuration.xml:\n  counter=\"%s\"\n  event=%d\n", counter.getType(), counter.getEvent());
+				handleException();
+			}
+
+			// iterate through the remaining enabled performance counters
+			for (int j = i + 1; j < MAX_PERFORMANCE_COUNTERS; j++) {
+				const Counter & counter2 = gSessionData->mCounters[j];
+				if (counter2.isEnabled()) {
+					// check if the types are the same
+					if (strcmp(counter.getType(), counter2.getType()) == 0) {
+						logg->logError(__FILE__, __LINE__, "Duplicate performance counter type in configuration.xml: %s", counter.getType());
+						handleException();
+					}
+				}
+			}
+		}
+	}
+}
+
+#define CONFIGURATION_REVISION 3
+int ConfigurationXML::configurationsTag(mxml_node_t *node) {
+	const char* revision_string;
+
+	revision_string = mxmlElementGetAttr(node, ATTR_REVISION);
+	if (!revision_string) {
+		return 1; //revision issue;
+	}
+
+	int revision = strtol(revision_string, NULL, 10);
+	if (revision < CONFIGURATION_REVISION) {
+		return 1; // revision issue
+	}
+
+	// A revision >= CONFIGURATION_REVISION is okay
+	// Greater than can occur when Streamline is newer than gator
+
+	return 0;
+}
+
+void ConfigurationXML::configurationTag(mxml_node_t *node) {
+	// handle all other performance counters
+	if (mIndex >= MAX_PERFORMANCE_COUNTERS) {
+		mIndex++;
+		gSessionData->mCounterOverflow = mIndex;
+		return;
+	}
+
+	// read attributes
+	Counter & counter = gSessionData->mCounters[mIndex];
+	counter.clear();
+	if (mxmlElementGetAttr(node, ATTR_COUNTER)) counter.setType(mxmlElementGetAttr(node, ATTR_COUNTER));
+	if (mxmlElementGetAttr(node, ATTR_EVENT)) counter.setEvent(strtol(mxmlElementGetAttr(node, ATTR_EVENT), NULL, 16));
+	if (mxmlElementGetAttr(node, ATTR_COUNT)) counter.setCount(strtol(mxmlElementGetAttr(node, ATTR_COUNT), NULL, 10));
+	if (mxmlElementGetAttr(node, ATTR_CORES)) counter.setCores(strtol(mxmlElementGetAttr(node, ATTR_CORES), NULL, 10));
+	if (counter.getCount() > 0) {
+		gSessionData->mIsEBS = true;
+	}
+	counter.setEnabled(true);
+
+	// Associate a driver with each counter
+	for (Driver *driver = Driver::getHead(); driver != NULL; driver = driver->getNext()) {
+		if (driver->claimCounter(counter)) {
+			if (counter.getDriver() != NULL) {
+				logg->logError(__FILE__, __LINE__, "More than one driver has claimed %s:%i", counter.getType(), counter.getEvent());
+				handleException();
+			}
+			counter.setDriver(driver);
+		}
+	}
+
+	// If no driver is associated with the counter, disable it
+	if (counter.getDriver() == NULL) {
+		logg->logMessage("No driver has claimed %s:%i", counter.getType(), counter.getEvent());
+		counter.setEnabled(false);
+	}
+
+	if (counter.isEnabled()) {
+		// update counter index
+		mIndex++;
+	}
+}
+
+void ConfigurationXML::getDefaultConfigurationXml(const char * & xml, unsigned int & len) {
+#include "defaults_xml.h" // defines and initializes char defaults_xml[] and int defaults_xml_len
+	xml = (const char *)defaults_xml;
+	len = defaults_xml_len;
+}
+
+void ConfigurationXML::getPath(char* path) {
+	if (gSessionData->mConfigurationXMLPath) {
+		strncpy(path, gSessionData->mConfigurationXMLPath, PATH_MAX);
+	} else {
+		if (util->getApplicationFullPath(path, PATH_MAX) != 0) {
+			logg->logMessage("Unable to determine the full path of gatord, the cwd will be used");
+		}
+		strncat(path, "configuration.xml", PATH_MAX - strlen(path) - 1);
+	}
+}
+
+void ConfigurationXML::remove() {
+	char path[PATH_MAX];
+	getPath(path);
+
+	if (::remove(path) != 0) {
+		logg->logError(__FILE__, __LINE__, "Invalid configuration.xml file detected and unable to delete it. To resolve, delete configuration.xml on disk");
+		handleException();
+	}
+	logg->logMessage("Invalid configuration.xml file detected and removed");
+}
diff --git a/tools/gator/daemon/ConfigurationXML.h b/tools/gator/daemon/ConfigurationXML.h
new file mode 100644
index 000000000000..efa415e508b6
--- /dev/null
+++ b/tools/gator/daemon/ConfigurationXML.h
@@ -0,0 +1,38 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef COUNTERS_H
+#define COUNTERS_H
+
+#include "mxml/mxml.h"
+
+class ConfigurationXML {
+public:
+	static void getDefaultConfigurationXml(const char * & xml, unsigned int & len);
+	static void getPath(char* path);
+	static void remove();
+
+	ConfigurationXML();
+	~ConfigurationXML();
+	const char* getConfigurationXML() {return mConfigurationXML;}
+	void validate(void);
+
+private:
+	char* mConfigurationXML;
+	int mIndex;
+
+	int parse(const char* xmlFile);
+	int configurationsTag(mxml_node_t *node);
+	void configurationTag(mxml_node_t *node);
+
+	// Intentionally unimplemented
+	ConfigurationXML(const ConfigurationXML &);
+	ConfigurationXML &operator=(const ConfigurationXML &);
+};
+
+#endif // COUNTERS_H
diff --git a/tools/gator/daemon/Counter.h b/tools/gator/daemon/Counter.h
new file mode 100644
index 000000000000..5202aa046362
--- /dev/null
+++ b/tools/gator/daemon/Counter.h
@@ -0,0 +1,65 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef COUNTER_H
+#define COUNTER_H
+
+#include <string.h>
+
+class Driver;
+
+class Counter {
+public:
+	static const size_t MAX_STRING_LEN = 80;
+	static const size_t MAX_DESCRIPTION_LEN = 400;
+
+	Counter () {
+		clear();
+	}
+
+	void clear () {
+		mType[0] = '\0';
+		mEnabled = false;
+		mEvent = -1;
+		mCount = 0;
+		mCores = -1;
+		mKey = 0;
+		mDriver = NULL;
+	}
+
+	void setType(const char *const type) { strncpy(mType, type, sizeof(mType)); mType[sizeof(mType) - 1] = '\0'; }
+	void setEnabled(const bool enabled) { mEnabled = enabled; }
+	void setEvent(const int event) { mEvent = event; }
+	void setCount(const int count) { mCount = count; }
+	void setCores(const int cores) { mCores = cores; }
+	void setKey(const int key) { mKey = key; }
+	void setDriver(Driver *const driver) { mDriver = driver; }
+
+	const char *getType() const { return mType;}
+	bool isEnabled() const { return mEnabled; }
+	int getEvent() const { return mEvent; }
+	int getCount() const { return mCount; }
+	int getCores() const { return mCores; }
+	int getKey() const { return mKey; }
+	Driver *getDriver() const { return mDriver; }
+
+private:
+	// Intentionally unimplemented
+	Counter(const Counter &);
+	Counter & operator=(const Counter &);
+
+	char mType[MAX_STRING_LEN];
+	bool mEnabled;
+	int mEvent;
+	int mCount;
+	int mCores;
+	int mKey;
+	Driver *mDriver;
+};
+
+#endif // COUNTER_H
diff --git a/tools/gator/daemon/DiskIODriver.cpp b/tools/gator/daemon/DiskIODriver.cpp
new file mode 100644
index 000000000000..5deb0f375f3a
--- /dev/null
+++ b/tools/gator/daemon/DiskIODriver.cpp
@@ -0,0 +1,125 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+// Define to get format macros from inttypes.h
+#define __STDC_FORMAT_MACROS
+
+#include "DiskIODriver.h"
+
+#include <inttypes.h>
+
+#include "Logging.h"
+#include "SessionData.h"
+
+class DiskIOCounter : public DriverCounter {
+public:
+	DiskIOCounter(DriverCounter *next, char *const name, int64_t *const value);
+	~DiskIOCounter();
+
+	int64_t read();
+
+private:
+	int64_t *const mValue;
+	int64_t mPrev;
+
+	// Intentionally unimplemented
+	DiskIOCounter(const DiskIOCounter &);
+	DiskIOCounter &operator=(const DiskIOCounter &);
+};
+
+DiskIOCounter::DiskIOCounter(DriverCounter *next, char *const name, int64_t *const value) : DriverCounter(next, name), mValue(value), mPrev(0) {
+}
+
+DiskIOCounter::~DiskIOCounter() {
+}
+
+int64_t DiskIOCounter::read() {
+	int64_t result = *mValue - mPrev;
+	mPrev = *mValue;
+	// Kernel assumes a sector is 512 bytes
+	return result << 9;
+}
+
+DiskIODriver::DiskIODriver() : mBuf(), mReadBytes(0), mWriteBytes(0) {
+}
+
+DiskIODriver::~DiskIODriver() {
+}
+
+void DiskIODriver::readEvents(mxml_node_t *const) {
+	// Only for use with perf
+	if (!gSessionData->perf.isSetup()) {
+		return;
+	}
+
+	setCounters(new DiskIOCounter(getCounters(), strdup("Linux_block_rq_rd"), &mReadBytes));
+	setCounters(new DiskIOCounter(getCounters(), strdup("Linux_block_rq_wr"), &mWriteBytes));
+}
+
+void DiskIODriver::doRead() {
+	if (!countersEnabled()) {
+		return;
+	}
+
+	if (!mBuf.read("/proc/diskstats")) {
+		logg->logError(__FILE__, __LINE__, "Unable to read /proc/diskstats");
+		handleException();
+	}
+
+	mReadBytes = 0;
+	mWriteBytes = 0;
+
+	char *lastName = NULL;
+	int lastNameLen = -1;
+	char *start = mBuf.getBuf();
+	while (*start != '\0') {
+		char *end = strchr(start, '\n');
+		if (end != NULL) {
+			*end = '\0';
+		}
+
+		int nameStart = -1;
+		int nameEnd = -1;
+		int64_t readBytes = -1;
+		int64_t writeBytes = -1;
+		const int count = sscanf(start, "%*d %*d %n%*s%n %*u %*u %" SCNu64 " %*u %*u %*u %" SCNu64, &nameStart, &nameEnd, &readBytes, &writeBytes);
+		if (count != 2) {
+			logg->logError(__FILE__, __LINE__, "Unable to parse /proc/diskstats");
+			handleException();
+		}
+
+		// Skip partitions which are identified if the name is a substring of the last non-partition
+		if ((lastName == NULL) || (strncmp(lastName, start + nameStart, lastNameLen) != 0)) {
+			lastName = start + nameStart;
+			lastNameLen = nameEnd - nameStart;
+			mReadBytes += readBytes;
+			mWriteBytes += writeBytes;
+		}
+
+		if (end == NULL) {
+			break;
+		}
+		start = end + 1;
+	}
+}
+
+void DiskIODriver::start() {
+	doRead();
+	// Initialize previous values
+	for (DriverCounter *counter = getCounters(); counter != NULL; counter = counter->getNext()) {
+		if (!counter->isEnabled()) {
+			continue;
+		}
+		counter->read();
+	}
+}
+
+void DiskIODriver::read(Buffer *const buffer) {
+	doRead();
+	super::read(buffer);
+}
diff --git a/tools/gator/daemon/DiskIODriver.h b/tools/gator/daemon/DiskIODriver.h
new file mode 100644
index 000000000000..d0db18c77d04
--- /dev/null
+++ b/tools/gator/daemon/DiskIODriver.h
@@ -0,0 +1,39 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef DISKIODRIVER_H
+#define DISKIODRIVER_H
+
+#include "Driver.h"
+#include "DynBuf.h"
+
+class DiskIODriver : public PolledDriver {
+private:
+	typedef PolledDriver super;
+
+public:
+	DiskIODriver();
+	~DiskIODriver();
+
+	void readEvents(mxml_node_t *const root);
+	void start();
+	void read(Buffer *const buffer);
+
+private:
+	void doRead();
+
+	DynBuf mBuf;
+	int64_t mReadBytes;
+	int64_t mWriteBytes;
+
+	// Intentionally unimplemented
+	DiskIODriver(const DiskIODriver &);
+	DiskIODriver &operator=(const DiskIODriver &);
+};
+
+#endif // DISKIODRIVER_H
diff --git a/tools/gator/daemon/Driver.cpp b/tools/gator/daemon/Driver.cpp
new file mode 100644
index 000000000000..275da31c7a0d
--- /dev/null
+++ b/tools/gator/daemon/Driver.cpp
@@ -0,0 +1,96 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "Driver.h"
+
+#include "Buffer.h"
+#include "SessionData.h"
+
+DriverCounter::DriverCounter(DriverCounter *const next, const char *const name) : mNext(next), mName(name), mKey(getEventKey()), mEnabled(false) {
+}
+
+DriverCounter::~DriverCounter() {
+	delete mName;
+}
+
+Driver *Driver::head = NULL;
+
+Driver::Driver() : next(head) {
+	head = this;
+}
+
+SimpleDriver::~SimpleDriver() {
+	DriverCounter *counters = mCounters;
+	while (counters != NULL) {
+		DriverCounter *counter = counters;
+		counters = counter->getNext();
+		delete counter;
+	}
+}
+
+DriverCounter *SimpleDriver::findCounter(const Counter &counter) const {
+	for (DriverCounter *driverCounter = mCounters; driverCounter != NULL; driverCounter = driverCounter->getNext()) {
+		if (strcmp(driverCounter->getName(), counter.getType()) == 0) {
+			return driverCounter;
+		}
+	}
+
+	return NULL;
+}
+
+bool SimpleDriver::claimCounter(const Counter &counter) const {
+	return findCounter(counter) != NULL;
+}
+
+bool SimpleDriver::countersEnabled() const {
+	for (DriverCounter *counter = mCounters; counter != NULL; counter = counter->getNext()) {
+		if (counter->isEnabled()) {
+			return true;
+		}
+	}
+	return false;
+}
+
+void SimpleDriver::resetCounters() {
+	for (DriverCounter *counter = mCounters; counter != NULL; counter = counter->getNext()) {
+		counter->setEnabled(false);
+	}
+}
+
+void SimpleDriver::setupCounter(Counter &counter) {
+	DriverCounter *const driverCounter = findCounter(counter);
+	if (driverCounter == NULL) {
+		counter.setEnabled(false);
+		return;
+	}
+	driverCounter->setEnabled(true);
+	counter.setKey(driverCounter->getKey());
+}
+
+int SimpleDriver::writeCounters(mxml_node_t *root) const {
+	int count = 0;
+	for (DriverCounter *counter = mCounters; counter != NULL; counter = counter->getNext()) {
+		mxml_node_t *node = mxmlNewElement(root, "counter");
+		mxmlElementSetAttr(node, "name", counter->getName());
+		++count;
+	}
+
+	return count;
+}
+
+PolledDriver::~PolledDriver() {
+}
+
+void PolledDriver::read(Buffer *const buffer) {
+	for (DriverCounter *counter = getCounters(); counter != NULL; counter = counter->getNext()) {
+		if (!counter->isEnabled()) {
+			continue;
+		}
+		buffer->event64(counter->getKey(), counter->read());
+	}
+}
diff --git a/tools/gator/daemon/Driver.h b/tools/gator/daemon/Driver.h
new file mode 100644
index 000000000000..72870e3dbca1
--- /dev/null
+++ b/tools/gator/daemon/Driver.h
@@ -0,0 +1,118 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef DRIVER_H
+#define DRIVER_H
+
+#include <stdint.h>
+
+#include "mxml/mxml.h"
+
+class Buffer;
+class Counter;
+
+class DriverCounter {
+public:
+	DriverCounter(DriverCounter *const next, const char *const name);
+	virtual ~DriverCounter();
+
+	DriverCounter *getNext() const { return mNext; }
+	const char *getName() const { return mName; }
+	int getKey() const { return mKey; }
+	bool isEnabled() const { return mEnabled; }
+	void setEnabled(const bool enabled) { mEnabled = enabled; }
+	virtual int64_t read() { return -1; }
+
+private:
+	DriverCounter *const mNext;
+	const char *const mName;
+	const int mKey;
+	bool mEnabled;
+
+	// Intentionally unimplemented
+	DriverCounter(const DriverCounter &);
+	DriverCounter &operator=(const DriverCounter &);
+};
+
+class Driver {
+public:
+	static Driver *getHead() { return head; }
+
+	virtual ~Driver() {}
+
+	// Returns true if this driver can manage the counter
+	virtual bool claimCounter(const Counter &counter) const = 0;
+	// Clears and disables all counters
+	virtual void resetCounters() = 0;
+	// Enables and prepares the counter for capture
+	virtual void setupCounter(Counter &counter) = 0;
+
+	// Performs any actions needed for setup or based on eventsXML
+	virtual void readEvents(mxml_node_t *const) {}
+	// Emits available counters
+	virtual int writeCounters(mxml_node_t *const root) const = 0;
+	// Emits possible dynamically generated events/counters
+	virtual void writeEvents(mxml_node_t *const) const {}
+
+	Driver *getNext() const { return next; }
+
+protected:
+	Driver();
+
+private:
+	static Driver *head;
+	Driver *next;
+
+	// Intentionally unimplemented
+	Driver(const Driver &);
+	Driver &operator=(const Driver &);
+};
+
+class SimpleDriver : public Driver {
+public:
+	virtual ~SimpleDriver();
+
+	bool claimCounter(const Counter &counter) const;
+	bool countersEnabled() const;
+	void resetCounters();
+	void setupCounter(Counter &counter);
+	int writeCounters(mxml_node_t *root) const;
+
+protected:
+	SimpleDriver() : mCounters(NULL) {}
+
+	DriverCounter *getCounters() const { return mCounters; }
+	void setCounters(DriverCounter *const counter) { mCounters = counter; }
+
+	DriverCounter *findCounter(const Counter &counter) const;
+
+private:
+	DriverCounter *mCounters;
+
+	// Intentionally unimplemented
+	SimpleDriver(const SimpleDriver &);
+	SimpleDriver &operator=(const SimpleDriver &);
+};
+
+class PolledDriver : public SimpleDriver {
+public:
+	virtual ~PolledDriver();
+
+	virtual void start() {}
+	virtual void read(Buffer *const buffer);
+
+protected:
+	PolledDriver() {}
+
+private:
+	// Intentionally unimplemented
+	PolledDriver(const PolledDriver &);
+	PolledDriver &operator=(const PolledDriver &);
+};
+
+#endif // DRIVER_H
diff --git a/tools/gator/daemon/DriverSource.cpp b/tools/gator/daemon/DriverSource.cpp
new file mode 100644
index 000000000000..7f299b646952
--- /dev/null
+++ b/tools/gator/daemon/DriverSource.cpp
@@ -0,0 +1,322 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+// Define to get format macros from inttypes.h
+#define __STDC_FORMAT_MACROS
+
+#include "DriverSource.h"
+
+#include <fcntl.h>
+#include <inttypes.h>
+#include <sys/prctl.h>
+#include <unistd.h>
+
+#include "Buffer.h"
+#include "Child.h"
+#include "DynBuf.h"
+#include "Fifo.h"
+#include "Logging.h"
+#include "Proc.h"
+#include "Sender.h"
+#include "SessionData.h"
+
+extern Child *child;
+
+DriverSource::DriverSource(sem_t *senderSem, sem_t *startProfile) : mBuffer(NULL), mFifo(NULL), mSenderSem(senderSem), mStartProfile(startProfile), mBufferSize(0), mBufferFD(0), mLength(1) {
+	int driver_version = 0;
+
+	mBuffer = new Buffer(0, FRAME_PERF_ATTRS, 4*1024*1024, senderSem);
+	if (readIntDriver("/dev/gator/version", &driver_version) == -1) {
+		logg->logError(__FILE__, __LINE__, "Error reading gator driver version");
+		handleException();
+	}
+
+	// Verify the driver version matches the daemon version
+	if (driver_version != PROTOCOL_VERSION) {
+		if ((driver_version > PROTOCOL_DEV) || (PROTOCOL_VERSION > PROTOCOL_DEV)) {
+			// One of the mismatched versions is development version
+			logg->logError(__FILE__, __LINE__,
+				"DEVELOPMENT BUILD MISMATCH: gator driver version \"%d\" is not in sync with gator daemon version \"%d\".\n"
+				">> The following must be synchronized from engineering repository:\n"
+				">> * gator driver\n"
+				">> * gator daemon\n"
+				">> * Streamline", driver_version, PROTOCOL_VERSION);
+			handleException();
+		} else {
+			// Release version mismatch
+			logg->logError(__FILE__, __LINE__,
+				"gator driver version \"%d\" is different than gator daemon version \"%d\".\n"
+				">> Please upgrade the driver and daemon to the latest versions.", driver_version, PROTOCOL_VERSION);
+			handleException();
+		}
+	}
+
+	int enable = -1;
+	if (readIntDriver("/dev/gator/enable", &enable) != 0 || enable != 0) {
+		logg->logError(__FILE__, __LINE__, "Driver already enabled, possibly a session is already in progress.");
+		handleException();
+	}
+
+	readIntDriver("/dev/gator/cpu_cores", &gSessionData->mCores);
+	if (gSessionData->mCores == 0) {
+		gSessionData->mCores = 1;
+	}
+
+	if (readIntDriver("/dev/gator/buffer_size", &mBufferSize) || mBufferSize <= 0) {
+		logg->logError(__FILE__, __LINE__, "Unable to read the driver buffer size");
+		handleException();
+	}
+}
+
+DriverSource::~DriverSource() {
+	delete mFifo;
+
+	// Write zero for safety, as a zero should have already been written
+	writeDriver("/dev/gator/enable", "0");
+
+	// Calls event_buffer_release in the driver
+	if (mBufferFD) {
+		close(mBufferFD);
+	}
+}
+
+bool DriverSource::prepare() {
+	// Create user-space buffers, add 5 to the size to account for the 1-byte type and 4-byte length
+	logg->logMessage("Created %d MB collector buffer with a %d-byte ragged end", gSessionData->mTotalBufferSize, mBufferSize);
+	mFifo = new Fifo(mBufferSize + 5, gSessionData->mTotalBufferSize*1024*1024, mSenderSem);
+
+	return true;
+}
+
+void DriverSource::bootstrapThread() {
+	prctl(PR_SET_NAME, (unsigned long)&"gatord-proc", 0, 0, 0);
+
+	DynBuf printb;
+	DynBuf b1;
+	DynBuf b2;
+	const uint64_t currTime = getTime();
+
+	if (!readProcComms(currTime, mBuffer, &printb, &b1, &b2)) {
+		logg->logError(__FILE__, __LINE__, "readProcComms failed");
+		handleException();
+	}
+
+	mBuffer->commit(currTime);
+	mBuffer->setDone();
+}
+
+void *DriverSource::bootstrapThreadStatic(void *arg) {
+	static_cast<DriverSource *>(arg)->bootstrapThread();
+	return NULL;
+}
+
+void DriverSource::run() {
+	// Get the initial pointer to the collect buffer
+	char *collectBuffer = mFifo->start();
+	int bytesCollected = 0;
+
+	logg->logMessage("********** Profiling started **********");
+
+	// Set the maximum backtrace depth
+	if (writeReadDriver("/dev/gator/backtrace_depth", &gSessionData->mBacktraceDepth)) {
+		logg->logError(__FILE__, __LINE__, "Unable to set the driver backtrace depth");
+		handleException();
+	}
+
+	// open the buffer which calls userspace_buffer_open() in the driver
+	mBufferFD = open("/dev/gator/buffer", O_RDONLY | O_CLOEXEC);
+	if (mBufferFD < 0) {
+		logg->logError(__FILE__, __LINE__, "The gator driver did not set up properly. Please view the linux console or dmesg log for more information on the failure.");
+		handleException();
+	}
+
+	// set the tick rate of the profiling timer
+	if (writeReadDriver("/dev/gator/tick", &gSessionData->mSampleRate) != 0) {
+		logg->logError(__FILE__, __LINE__, "Unable to set the driver tick");
+		handleException();
+	}
+
+	// notify the kernel of the response type
+	int response_type = gSessionData->mLocalCapture ? 0 : RESPONSE_APC_DATA;
+	if (writeDriver("/dev/gator/response_type", response_type)) {
+		logg->logError(__FILE__, __LINE__, "Unable to write the response type");
+		handleException();
+	}
+
+	// Set the live rate
+	if (writeReadDriver("/dev/gator/live_rate", &gSessionData->mLiveRate)) {
+		logg->logError(__FILE__, __LINE__, "Unable to set the driver live rate");
+		handleException();
+	}
+
+	logg->logMessage("Start the driver");
+
+	// This command makes the driver start profiling by calling gator_op_start() in the driver
+	if (writeDriver("/dev/gator/enable", "1") != 0) {
+		logg->logError(__FILE__, __LINE__, "The gator driver did not start properly. Please view the linux console or dmesg log for more information on the failure.");
+		handleException();
+	}
+
+	lseek(mBufferFD, 0, SEEK_SET);
+
+	sem_post(mStartProfile);
+
+	pthread_t bootstrapThreadID;
+	if (pthread_create(&bootstrapThreadID, NULL, bootstrapThreadStatic, this) != 0) {
+		logg->logError(__FILE__, __LINE__, "Unable to start the gator_bootstrap thread");
+		handleException();
+	}
+
+	// Collect Data
+	do {
+		// This command will stall until data is received from the driver
+		// Calls event_buffer_read in the driver
+		errno = 0;
+		bytesCollected = read(mBufferFD, collectBuffer, mBufferSize);
+
+		// If read() returned due to an interrupt signal, re-read to obtain the last bit of collected data
+		if (bytesCollected == -1 && errno == EINTR) {
+			bytesCollected = read(mBufferFD, collectBuffer, mBufferSize);
+		}
+
+		// return the total bytes written
+		logg->logMessage("Driver read of %d bytes", bytesCollected);
+
+		// In one shot mode, stop collection once all the buffers are filled
+		if (gSessionData->mOneShot && gSessionData->mSessionIsActive) {
+			if (bytesCollected == -1 || mFifo->willFill(bytesCollected)) {
+				logg->logMessage("One shot");
+				child->endSession();
+			}
+		}
+		collectBuffer = mFifo->write(bytesCollected);
+	} while (bytesCollected > 0);
+
+	logg->logMessage("Exit collect data loop");
+
+	pthread_join(bootstrapThreadID, NULL);
+}
+
+void DriverSource::interrupt() {
+	// This command should cause the read() function in collect() to return and stop the driver from profiling
+	if (writeDriver("/dev/gator/enable", "0") != 0) {
+		logg->logMessage("Stopping kernel failed");
+	}
+}
+
+bool DriverSource::isDone() {
+	return mLength <= 0 && (mBuffer == NULL || mBuffer->isDone());
+}
+
+void DriverSource::write(Sender *sender) {
+	char *data = mFifo->read(&mLength);
+	if (data != NULL) {
+		sender->writeData(data, mLength, RESPONSE_APC_DATA);
+		mFifo->release();
+		// Assume the summary packet is in the first block received from the driver
+		gSessionData->mSentSummary = true;
+	}
+	if (mBuffer != NULL && !mBuffer->isDone()) {
+		mBuffer->write(sender);
+		if (mBuffer->isDone()) {
+			Buffer *buf = mBuffer;
+			mBuffer = NULL;
+			delete buf;
+		}
+	}
+}
+
+int DriverSource::readIntDriver(const char *fullpath, int *value) {
+	char data[40]; // Sufficiently large to hold any integer
+	const int fd = open(fullpath, O_RDONLY | O_CLOEXEC);
+	if (fd < 0) {
+		return -1;
+	}
+
+	const ssize_t bytes = read(fd, data, sizeof(data) - 1);
+	close(fd);
+	if (bytes < 0) {
+		return -1;
+	}
+	data[bytes] = '\0';
+
+	char *endptr;
+	errno = 0;
+	*value = strtol(data, &endptr, 10);
+	if (errno != 0 || *endptr != '\n') {
+		logg->logMessage("Invalid value in file %s", fullpath);
+		return -1;
+	}
+
+	return 0;
+}
+
+int DriverSource::readInt64Driver(const char *fullpath, int64_t *value) {
+	char data[40]; // Sufficiently large to hold any integer
+	const int fd = open(fullpath, O_RDONLY | O_CLOEXEC);
+	if (fd < 0) {
+		return -1;
+	}
+
+	const ssize_t bytes = read(fd, data, sizeof(data) - 1);
+	close(fd);
+	if (bytes < 0) {
+		return -1;
+	}
+	data[bytes] = '\0';
+
+	char *endptr;
+	errno = 0;
+	*value = strtoll(data, &endptr, 10);
+	if (errno != 0 || (*endptr != '\n' && *endptr != '\0')) {
+		logg->logMessage("Invalid value in file %s", fullpath);
+		return -1;
+	}
+
+	return 0;
+}
+
+int DriverSource::writeDriver(const char *fullpath, const char *data) {
+	int fd = open(fullpath, O_WRONLY | O_CLOEXEC);
+	if (fd < 0) {
+		return -1;
+	}
+	if (::write(fd, data, strlen(data)) < 0) {
+		close(fd);
+		logg->logMessage("Opened but could not write to %s", fullpath);
+		return -1;
+	}
+	close(fd);
+	return 0;
+}
+
+int DriverSource::writeDriver(const char *path, int value) {
+	char data[40]; // Sufficiently large to hold any integer
+	snprintf(data, sizeof(data), "%d", value);
+	return writeDriver(path, data);
+}
+
+int DriverSource::writeDriver(const char *path, int64_t value) {
+	char data[40]; // Sufficiently large to hold any integer
+	snprintf(data, sizeof(data), "%" PRIi64, value);
+	return writeDriver(path, data);
+}
+
+int DriverSource::writeReadDriver(const char *path, int *value) {
+	if (writeDriver(path, *value) || readIntDriver(path, value)) {
+		return -1;
+	}
+	return 0;
+}
+
+int DriverSource::writeReadDriver(const char *path, int64_t *value) {
+	if (writeDriver(path, *value) || readInt64Driver(path, value)) {
+		return -1;
+	}
+	return 0;
+}
diff --git a/tools/gator/daemon/DriverSource.h b/tools/gator/daemon/DriverSource.h
new file mode 100644
index 000000000000..ec27b0815bbf
--- /dev/null
+++ b/tools/gator/daemon/DriverSource.h
@@ -0,0 +1,57 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef DRIVERSOURCE_H
+#define DRIVERSOURCE_H
+
+#include <semaphore.h>
+#include <stdint.h>
+
+#include "Source.h"
+
+class Buffer;
+class Fifo;
+
+class DriverSource : public Source {
+public:
+	DriverSource(sem_t *senderSem, sem_t *startProfile);
+	~DriverSource();
+
+	bool prepare();
+	void run();
+	void interrupt();
+
+	bool isDone();
+	void write(Sender *sender);
+
+	static int readIntDriver(const char *fullpath, int *value);
+	static int readInt64Driver(const char *fullpath, int64_t *value);
+	static int writeDriver(const char *fullpath, const char *data);
+	static int writeDriver(const char *path, int value);
+	static int writeDriver(const char *path, int64_t value);
+	static int writeReadDriver(const char *path, int *value);
+	static int writeReadDriver(const char *path, int64_t *value);
+
+private:
+	static void *bootstrapThreadStatic(void *arg);
+	void bootstrapThread();
+
+	Buffer *mBuffer;
+	Fifo *mFifo;
+	sem_t *const mSenderSem;
+	sem_t *const mStartProfile;
+	int mBufferSize;
+	int mBufferFD;
+	int mLength;
+
+	// Intentionally unimplemented
+	DriverSource(const DriverSource &);
+	DriverSource &operator=(const DriverSource &);
+};
+
+#endif // DRIVERSOURCE_H
diff --git a/tools/gator/daemon/DynBuf.cpp b/tools/gator/daemon/DynBuf.cpp
new file mode 100644
index 000000000000..df20713ad63c
--- /dev/null
+++ b/tools/gator/daemon/DynBuf.cpp
@@ -0,0 +1,139 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "DynBuf.h"
+
+#include <errno.h>
+#include <fcntl.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <unistd.h>
+
+#include "Logging.h"
+
+// Pick an aggressive size as buffer is primarily used for disk IO
+#define MIN_BUFFER_FREE (1 << 12)
+
+int DynBuf::resize(const size_t minCapacity) {
+	size_t scaledCapacity = 2 * capacity;
+	if (scaledCapacity < minCapacity) {
+		scaledCapacity = minCapacity;
+	}
+	if (scaledCapacity < 2 * MIN_BUFFER_FREE) {
+		scaledCapacity = 2 * MIN_BUFFER_FREE;
+	}
+	capacity = scaledCapacity;
+
+	buf = static_cast<char *>(realloc(buf, capacity));
+	if (buf == NULL) {
+		return -errno;
+	}
+
+	return 0;
+}
+
+bool DynBuf::read(const char *const path) {
+	int result = false;
+
+	const int fd = open(path, O_RDONLY | O_CLOEXEC);
+	if (fd < 0) {
+		logg->logMessage("%s(%s:%i): open failed", __FUNCTION__, __FILE__, __LINE__);
+		return false;
+	}
+
+	length = 0;
+
+	for (;;) {
+		const size_t minCapacity = length + MIN_BUFFER_FREE + 1;
+		if (capacity < minCapacity) {
+			if (resize(minCapacity) != 0) {
+				logg->logMessage("%s(%s:%i): DynBuf::resize failed", __FUNCTION__, __FILE__, __LINE__);
+				goto fail;
+			}
+		}
+
+		const ssize_t bytes = ::read(fd, buf + length, capacity - length - 1);
+		if (bytes < 0) {
+			logg->logMessage("%s(%s:%i): read failed", __FUNCTION__, __FILE__, __LINE__);
+			goto fail;
+		} else if (bytes == 0) {
+			break;
+		}
+		length += bytes;
+	}
+
+	buf[length] = '\0';
+	result = true;
+
+ fail:
+	close(fd);
+
+	return result;
+}
+
+int DynBuf::readlink(const char *const path) {
+	ssize_t bytes = MIN_BUFFER_FREE;
+
+	for (;;) {
+		if (static_cast<size_t>(bytes) >= capacity) {
+			const int err = resize(2 * bytes);
+			if (err != 0) {
+				return err;
+			}
+		}
+		bytes = ::readlink(path, buf, capacity);
+		if (bytes < 0) {
+			return -errno;
+		} else if (static_cast<size_t>(bytes) < capacity) {
+			break;
+		}
+	}
+
+	length = bytes;
+	buf[bytes] = '\0';
+
+	return 0;
+}
+
+bool DynBuf::printf(const char *format, ...) {
+	va_list ap;
+
+	if (capacity <= 0) {
+		if (resize(2 * MIN_BUFFER_FREE) != 0) {
+			logg->logMessage("%s(%s:%i): DynBuf::resize failed", __FUNCTION__, __FILE__, __LINE__);
+			return false;
+		}
+	}
+
+	va_start(ap, format);
+	int bytes = vsnprintf(buf, capacity, format, ap);
+	va_end(ap);
+	if (bytes < 0) {
+		logg->logMessage("%s(%s:%i): fsnprintf failed", __FUNCTION__, __FILE__, __LINE__);
+		return false;
+	}
+
+	if (static_cast<size_t>(bytes) > capacity) {
+		if (resize(bytes + 1) != 0) {
+			logg->logMessage("%s(%s:%i): DynBuf::resize failed", __FUNCTION__, __FILE__, __LINE__);
+			return false;
+		}
+
+		va_start(ap, format);
+		bytes = vsnprintf(buf, capacity, format, ap);
+		va_end(ap);
+		if (bytes < 0) {
+			logg->logMessage("%s(%s:%i): fsnprintf failed", __FUNCTION__, __FILE__, __LINE__);
+			return false;
+		}
+	}
+
+	length = bytes;
+
+	return true;
+}
diff --git a/tools/gator/daemon/DynBuf.h b/tools/gator/daemon/DynBuf.h
new file mode 100644
index 000000000000..2f4554ab2e49
--- /dev/null
+++ b/tools/gator/daemon/DynBuf.h
@@ -0,0 +1,52 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef DYNBUF_H
+#define DYNBUF_H
+
+#include <stdlib.h>
+
+class DynBuf {
+public:
+	DynBuf() : capacity(0), length(0), buf(NULL) {}
+	~DynBuf() {
+		reset();
+	}
+
+	inline void reset() {
+		capacity = 0;
+		length = 0;
+		if (buf != NULL) {
+			free(buf);
+			buf = NULL;
+		}
+	}
+
+	bool read(const char *const path);
+	// On error instead of printing the error and returning false, this returns -errno
+	int readlink(const char *const path);
+	__attribute__ ((format(printf, 2, 3)))
+	bool printf(const char *format, ...);
+
+	size_t getLength() const { return length; }
+	const char *getBuf() const { return buf; }
+	char *getBuf() { return buf; }
+
+private:
+	int resize(const size_t minCapacity);
+
+	size_t capacity;
+	size_t length;
+	char *buf;
+
+	// Intentionally undefined
+	DynBuf(const DynBuf &);
+	DynBuf &operator=(const DynBuf &);
+};
+
+#endif // DYNBUF_H
diff --git a/tools/gator/daemon/EventsXML.cpp b/tools/gator/daemon/EventsXML.cpp
new file mode 100644
index 000000000000..d905bbabe988
--- /dev/null
+++ b/tools/gator/daemon/EventsXML.cpp
@@ -0,0 +1,76 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "EventsXML.h"
+
+#include "CapturedXML.h"
+#include "Logging.h"
+#include "OlyUtility.h"
+#include "SessionData.h"
+
+mxml_node_t *EventsXML::getTree() {
+#include "events_xml.h" // defines and initializes char events_xml[] and int events_xml_len
+	char path[PATH_MAX];
+	mxml_node_t *xml;
+	FILE *fl;
+
+	// Avoid unused variable warning
+	(void)events_xml_len;
+
+	// Load the provided or default events xml
+	if (gSessionData->mEventsXMLPath) {
+		strncpy(path, gSessionData->mEventsXMLPath, PATH_MAX);
+	} else {
+		util->getApplicationFullPath(path, PATH_MAX);
+		strncat(path, "events.xml", PATH_MAX - strlen(path) - 1);
+	}
+	fl = fopen(path, "r");
+	if (fl) {
+		xml = mxmlLoadFile(NULL, fl, MXML_NO_CALLBACK);
+		fclose(fl);
+	} else {
+		logg->logMessage("Unable to locate events.xml, using default");
+		xml = mxmlLoadString(NULL, (const char *)events_xml, MXML_NO_CALLBACK);
+	}
+
+	return xml;
+}
+
+char *EventsXML::getXML() {
+	mxml_node_t *xml = getTree();
+
+	// Add dynamic events from the drivers
+	mxml_node_t *events = mxmlFindElement(xml, xml, "events", NULL, NULL, MXML_DESCEND);
+	if (!events) {
+		logg->logError(__FILE__, __LINE__, "Unable to find <events> node in the events.xml");
+		handleException();
+	}
+	for (Driver *driver = Driver::getHead(); driver != NULL; driver = driver->getNext()) {
+		driver->writeEvents(events);
+	}
+
+	char *string = mxmlSaveAllocString(xml, mxmlWhitespaceCB);
+	mxmlDelete(xml);
+
+	return string;
+}
+
+void EventsXML::write(const char *path) {
+	char file[PATH_MAX];
+
+	// Set full path
+	snprintf(file, PATH_MAX, "%s/events.xml", path);
+
+	char *buf = getXML();
+	if (util->writeToDisk(file, buf) < 0) {
+		logg->logError(__FILE__, __LINE__, "Error writing %s\nPlease verify the path.", file);
+		handleException();
+	}
+
+	free(buf);
+}
diff --git a/tools/gator/daemon/EventsXML.h b/tools/gator/daemon/EventsXML.h
new file mode 100644
index 000000000000..ff7a02fd3c78
--- /dev/null
+++ b/tools/gator/daemon/EventsXML.h
@@ -0,0 +1,21 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef EVENTS_XML
+#define EVENTS_XML
+
+#include "mxml/mxml.h"
+
+class EventsXML {
+public:
+	mxml_node_t *getTree();
+	char *getXML();
+	void write(const char* path);
+};
+
+#endif // EVENTS_XML
diff --git a/tools/gator/daemon/ExternalSource.cpp b/tools/gator/daemon/ExternalSource.cpp
new file mode 100644
index 000000000000..8f5e6b684c53
--- /dev/null
+++ b/tools/gator/daemon/ExternalSource.cpp
@@ -0,0 +1,260 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "ExternalSource.h"
+
+#include <fcntl.h>
+#include <sys/prctl.h>
+#include <unistd.h>
+
+#include "Logging.h"
+#include "OlySocket.h"
+#include "SessionData.h"
+
+static const char MALI_VIDEO[] = "\0mali-video";
+static const char MALI_VIDEO_STARTUP[] = "\0mali-video-startup";
+static const char MALI_VIDEO_V1[] = "MALI_VIDEO 1\n";
+static const char MALI_GRAPHICS[] = "\0mali_thirdparty_server";
+static const char MALI_GRAPHICS_STARTUP[] = "\0mali_thirdparty_client";
+static const char MALI_GRAPHICS_V1[] = "MALI_GRAPHICS 1\n";
+
+static bool setNonblock(const int fd) {
+	int flags;
+
+	flags = fcntl(fd, F_GETFL);
+	if (flags < 0) {
+		logg->logMessage("fcntl getfl failed");
+		return false;
+	}
+
+	if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0) {
+		logg->logMessage("fcntl setfl failed");
+		return false;
+	}
+
+	return true;
+}
+
+ExternalSource::ExternalSource(sem_t *senderSem) : mBuffer(0, FRAME_EXTERNAL, 128*1024, senderSem), mMonitor(), mMveStartupUds(MALI_VIDEO_STARTUP, sizeof(MALI_VIDEO_STARTUP)), mMaliStartupUds(MALI_GRAPHICS_STARTUP, sizeof(MALI_GRAPHICS_STARTUP)), mAnnotate(8083), mInterruptFd(-1), mMaliUds(-1), mMveUds(-1) {
+	sem_init(&mBufferSem, 0, 0);
+}
+
+ExternalSource::~ExternalSource() {
+}
+
+void ExternalSource::waitFor(const int bytes) {
+	while (mBuffer.bytesAvailable() <= bytes) {
+		sem_wait(&mBufferSem);
+	}
+}
+
+void ExternalSource::configureConnection(const int fd, const char *const handshake, size_t size) {
+	if (!setNonblock(fd)) {
+		logg->logError(__FILE__, __LINE__, "Unable to set nonblock on fh");
+		handleException();
+	}
+
+	if (!mMonitor.add(fd)) {
+		logg->logError(__FILE__, __LINE__, "Unable to add fh to monitor");
+		handleException();
+	}
+
+	// Write the handshake to the circular buffer
+	waitFor(Buffer::MAXSIZE_PACK32 + size - 1);
+	mBuffer.packInt(fd);
+	mBuffer.writeBytes(handshake, size - 1);
+	mBuffer.commit(1);
+}
+
+bool ExternalSource::connectMali() {
+	mMaliUds = OlySocket::connect(MALI_GRAPHICS, sizeof(MALI_GRAPHICS));
+	if (mMaliUds < 0) {
+		return false;
+	}
+
+	configureConnection(mMaliUds, MALI_GRAPHICS_V1, sizeof(MALI_GRAPHICS_V1));
+
+	return true;
+}
+
+bool ExternalSource::connectMve() {
+	if (!gSessionData->maliVideo.countersEnabled()) {
+		return true;
+	}
+
+	mMveUds = OlySocket::connect(MALI_VIDEO, sizeof(MALI_VIDEO));
+	if (mMveUds < 0) {
+		return false;
+	}
+
+	if (!gSessionData->maliVideo.start(mMveUds)) {
+		return false;
+	}
+
+	configureConnection(mMveUds, MALI_VIDEO_V1, sizeof(MALI_VIDEO_V1));
+
+	return true;
+}
+
+bool ExternalSource::prepare() {
+	if (!mMonitor.init() ||
+			!setNonblock(mMveStartupUds.getFd()) || !mMonitor.add(mMveStartupUds.getFd()) ||
+			!setNonblock(mMaliStartupUds.getFd()) || !mMonitor.add(mMaliStartupUds.getFd()) ||
+			!setNonblock(mAnnotate.getFd()) || !mMonitor.add(mAnnotate.getFd()) ||
+			false) {
+		return false;
+	}
+
+	connectMali();
+	connectMve();
+
+	return true;
+}
+
+void ExternalSource::run() {
+	int pipefd[2];
+
+	prctl(PR_SET_NAME, (unsigned long)&"gatord-external", 0, 0, 0);
+
+	if (pipe_cloexec(pipefd) != 0) {
+		logg->logError(__FILE__, __LINE__, "pipe failed");
+		handleException();
+	}
+	mInterruptFd = pipefd[1];
+
+	if (!mMonitor.add(pipefd[0])) {
+		logg->logError(__FILE__, __LINE__, "Monitor::add failed");
+		handleException();
+	}
+
+	// Notify annotate clients to retry connecting to gatord
+	gSessionData->annotateListener.signal();
+
+	while (gSessionData->mSessionIsActive) {
+		struct epoll_event events[16];
+		// Clear any pending sem posts
+		while (sem_trywait(&mBufferSem) == 0);
+		int ready = mMonitor.wait(events, ARRAY_LENGTH(events), -1);
+		if (ready < 0) {
+			logg->logError(__FILE__, __LINE__, "Monitor::wait failed");
+			handleException();
+		}
+
+		const uint64_t currTime = getTime();
+
+		for (int i = 0; i < ready; ++i) {
+			const int fd = events[i].data.fd;
+			if (fd == mMveStartupUds.getFd()) {
+				// Mali Video Engine says it's alive
+				int client = mMveStartupUds.acceptConnection();
+				// Don't read from this connection, establish a new connection to Mali-V500
+				close(client);
+				if (!connectMve()) {
+					logg->logError(__FILE__, __LINE__, "Unable to configure incoming Mali video connection");
+					handleException();
+				}
+			} else if (fd == mMaliStartupUds.getFd()) {
+				// Mali Graphics says it's alive
+				int client = mMaliStartupUds.acceptConnection();
+				// Don't read from this connection, establish a new connection to Mali Graphics
+				close(client);
+				if (!connectMali()) {
+					logg->logError(__FILE__, __LINE__, "Unable to configure incoming Mali graphics connection");
+					handleException();
+				}
+			} else if (fd == mAnnotate.getFd()) {
+				int client = mAnnotate.acceptConnection();
+				if (!setNonblock(client) || !mMonitor.add(client)) {
+					logg->logError(__FILE__, __LINE__, "Unable to set socket options on incoming annotation connection");
+					handleException();
+				}
+			} else if (fd == pipefd[0]) {
+				// Means interrupt has been called and mSessionIsActive should be reread
+			} else {
+				/* This can result in some starvation if there are multiple
+				 * threads which are annotating heavily, but it is not
+				 * recommended that threads annotate that much as it can also
+				 * starve out the gator data.
+				 */
+				while (gSessionData->mSessionIsActive) {
+					// Wait until there is enough room for the fd, two headers and two ints
+					waitFor(7*Buffer::MAXSIZE_PACK32 + 2*sizeof(uint32_t));
+					mBuffer.packInt(fd);
+					const int contiguous = mBuffer.contiguousSpaceAvailable();
+					const int bytes = read(fd, mBuffer.getWritePos(), contiguous);
+					if (bytes < 0) {
+						if (errno == EAGAIN) {
+							// Nothing left to read
+							mBuffer.commit(currTime);
+							break;
+						}
+						// Something else failed, close the socket
+						mBuffer.commit(currTime);
+						mBuffer.packInt(-1);
+						mBuffer.packInt(fd);
+						mBuffer.commit(currTime);
+						close(fd);
+						break;
+					} else if (bytes == 0) {
+						// The other side is closed
+						mBuffer.commit(currTime);
+						mBuffer.packInt(-1);
+						mBuffer.packInt(fd);
+						mBuffer.commit(currTime);
+						close(fd);
+						break;
+					}
+
+					mBuffer.advanceWrite(bytes);
+					mBuffer.commit(currTime);
+
+					// Short reads also mean nothing is left to read
+					if (bytes < contiguous) {
+						break;
+					}
+				}
+			}
+		}
+	}
+
+	mBuffer.setDone();
+
+	if (mMveUds >= 0) {
+		gSessionData->maliVideo.stop(mMveUds);
+	}
+
+	mInterruptFd = -1;
+	close(pipefd[0]);
+	close(pipefd[1]);
+}
+
+void ExternalSource::interrupt() {
+	if (mInterruptFd >= 0) {
+		int8_t c = 0;
+		// Write to the pipe to wake the monitor which will cause mSessionIsActive to be reread
+		if (::write(mInterruptFd, &c, sizeof(c)) != sizeof(c)) {
+			logg->logError(__FILE__, __LINE__, "write failed");
+			handleException();
+		}
+	}
+}
+
+bool ExternalSource::isDone() {
+	return mBuffer.isDone();
+}
+
+void ExternalSource::write(Sender *sender) {
+	// Don't send external data until the summary packet is sent so that monotonic delta is available
+	if (!gSessionData->mSentSummary) {
+		return;
+	}
+	if (!mBuffer.isDone()) {
+		mBuffer.write(sender);
+		sem_post(&mBufferSem);
+	}
+}
diff --git a/tools/gator/daemon/ExternalSource.h b/tools/gator/daemon/ExternalSource.h
new file mode 100644
index 000000000000..919e75e8a41a
--- /dev/null
+++ b/tools/gator/daemon/ExternalSource.h
@@ -0,0 +1,53 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef EXTERNALSOURCE_H
+#define EXTERNALSOURCE_H
+
+#include <semaphore.h>
+
+#include "Buffer.h"
+#include "Monitor.h"
+#include "OlySocket.h"
+#include "Source.h"
+
+// Counters from external sources like graphics drivers and annotations
+class ExternalSource : public Source {
+public:
+	ExternalSource(sem_t *senderSem);
+	~ExternalSource();
+
+	bool prepare();
+	void run();
+	void interrupt();
+
+	bool isDone();
+	void write(Sender *sender);
+
+private:
+	void waitFor(const int bytes);
+	void configureConnection(const int fd, const char *const handshake, size_t size);
+	bool connectMali();
+	bool connectMve();
+
+	sem_t mBufferSem;
+	Buffer mBuffer;
+	Monitor mMonitor;
+	OlyServerSocket mMveStartupUds;
+	OlyServerSocket mMaliStartupUds;
+	OlyServerSocket mAnnotate;
+	int mInterruptFd;
+	int mMaliUds;
+	int mMveUds;
+
+	// Intentionally unimplemented
+	ExternalSource(const ExternalSource &);
+	ExternalSource &operator=(const ExternalSource &);
+};
+
+#endif // EXTERNALSOURCE_H
diff --git a/tools/gator/daemon/FSDriver.cpp b/tools/gator/daemon/FSDriver.cpp
new file mode 100644
index 000000000000..dd8eb804dc99
--- /dev/null
+++ b/tools/gator/daemon/FSDriver.cpp
@@ -0,0 +1,158 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "FSDriver.h"
+
+#include <fcntl.h>
+#include <regex.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "DriverSource.h"
+#include "Logging.h"
+
+class FSCounter : public DriverCounter {
+public:
+	FSCounter(DriverCounter *next, char *name, char *path, const char *regex);
+	~FSCounter();
+
+	const char *getPath() const { return mPath; }
+
+	int64_t read();
+
+private:
+	char *const mPath;
+	regex_t mReg;
+	bool mUseRegex;
+
+	// Intentionally unimplemented
+	FSCounter(const FSCounter &);
+	FSCounter &operator=(const FSCounter &);
+};
+
+FSCounter::FSCounter(DriverCounter *next, char *name, char *path, const char *regex) : DriverCounter(next, name), mPath(path), mUseRegex(regex != NULL) {
+	if (mUseRegex) {
+		int result = regcomp(&mReg, regex, REG_EXTENDED);
+		if (result != 0) {
+			char buf[128];
+			regerror(result, &mReg, buf, sizeof(buf));
+			logg->logError(__FILE__, __LINE__, "Invalid regex '%s': %s", regex, buf);
+			handleException();
+		}
+	}
+}
+
+FSCounter::~FSCounter() {
+	free(mPath);
+	if (mUseRegex) {
+		regfree(&mReg);
+	}
+}
+
+int64_t FSCounter::read() {
+	int64_t value;
+	if (mUseRegex) {
+		char buf[4096];
+		size_t pos = 0;
+		const int fd = open(mPath, O_RDONLY | O_CLOEXEC);
+		if (fd < 0) {
+			goto fail;
+		}
+		while (pos < sizeof(buf) - 1) {
+			const ssize_t bytes = ::read(fd, buf + pos, sizeof(buf) - pos - 1);
+			if (bytes < 0) {
+				goto fail;
+			} else if (bytes == 0) {
+				break;
+			}
+			pos += bytes;
+		}
+		close(fd);
+		buf[pos] = '\0';
+
+		regmatch_t match[2];
+		int result = regexec(&mReg, buf, 2, match, 0);
+		if (result != 0) {
+			regerror(result, &mReg, buf, sizeof(buf));
+			logg->logError(__FILE__, __LINE__, "Parsing %s failed: %s", mPath, buf);
+			handleException();
+		}
+
+		if (match[1].rm_so < 0) {
+			logg->logError(__FILE__, __LINE__, "Parsing %s failed", mPath);
+			handleException();
+		}
+
+		errno = 0;
+		value = strtoll(buf + match[1].rm_so, NULL, 0);
+		if (errno != 0) {
+			logg->logError(__FILE__, __LINE__, "Parsing %s failed: %s", mPath, strerror(errno));
+			handleException();
+		}
+	} else {
+		if (DriverSource::readInt64Driver(mPath, &value) != 0) {
+			goto fail;
+		}
+	}
+	return value;
+
+ fail:
+	logg->logError(__FILE__, __LINE__, "Unable to read %s", mPath);
+	handleException();
+}
+
+FSDriver::FSDriver() {
+}
+
+FSDriver::~FSDriver() {
+}
+
+void FSDriver::readEvents(mxml_node_t *const xml) {
+	mxml_node_t *node = xml;
+	while (true) {
+		node = mxmlFindElement(node, xml, "event", NULL, NULL, MXML_DESCEND);
+		if (node == NULL) {
+			break;
+		}
+		const char *counter = mxmlElementGetAttr(node, "counter");
+		if (counter == NULL) {
+			continue;
+		}
+
+		if (counter[0] == '/') {
+			logg->logError(__FILE__, __LINE__, "Old style filesystem counter (%s) detected, please create a new unique counter value and move the filename into the path attribute, see events-Filesystem.xml for examples", counter);
+			handleException();
+		}
+
+		if (strncmp(counter, "filesystem_", 11) != 0) {
+			continue;
+		}
+
+		const char *path = mxmlElementGetAttr(node, "path");
+		if (path == NULL) {
+			logg->logError(__FILE__, __LINE__, "The filesystem counter %s is missing the required path attribute", counter);
+			handleException();
+		}
+		const char *regex = mxmlElementGetAttr(node, "regex");
+		setCounters(new FSCounter(getCounters(), strdup(counter), strdup(path), regex));
+	}
+}
+
+int FSDriver::writeCounters(mxml_node_t *root) const {
+	int count = 0;
+	for (FSCounter *counter = static_cast<FSCounter *>(getCounters()); counter != NULL; counter = static_cast<FSCounter *>(counter->getNext())) {
+		if (access(counter->getPath(), R_OK) == 0) {
+			mxml_node_t *node = mxmlNewElement(root, "counter");
+			mxmlElementSetAttr(node, "name", counter->getName());
+			++count;
+		}
+	}
+
+	return count;
+}
diff --git a/tools/gator/daemon/FSDriver.h b/tools/gator/daemon/FSDriver.h
new file mode 100644
index 000000000000..a7dc8b4df9dd
--- /dev/null
+++ b/tools/gator/daemon/FSDriver.h
@@ -0,0 +1,29 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef FSDRIVER_H
+#define FSDRIVER_H
+
+#include "Driver.h"
+
+class FSDriver : public PolledDriver {
+public:
+	FSDriver();
+	~FSDriver();
+
+	void readEvents(mxml_node_t *const xml);
+
+	int writeCounters(mxml_node_t *root) const;
+
+private:
+	// Intentionally unimplemented
+	FSDriver(const FSDriver &);
+	FSDriver &operator=(const FSDriver &);
+};
+
+#endif // FSDRIVER_H
diff --git a/tools/gator/daemon/Fifo.cpp b/tools/gator/daemon/Fifo.cpp
new file mode 100644
index 000000000000..41275fd287b8
--- /dev/null
+++ b/tools/gator/daemon/Fifo.cpp
@@ -0,0 +1,127 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "Fifo.h"
+
+#include <stdlib.h>
+
+#include "Logging.h"
+
+// bufferSize is the amount of data to be filled
+// singleBufferSize is the maximum size that may be filled during a single write
+// (bufferSize + singleBufferSize) will be allocated
+Fifo::Fifo(int singleBufferSize, int bufferSize, sem_t* readerSem) {
+  mWrite = mRead = mReadCommit = mRaggedEnd = 0;
+  mWrapThreshold = bufferSize;
+  mSingleBufferSize = singleBufferSize;
+  mReaderSem = readerSem;
+  mBuffer = (char*)malloc(bufferSize + singleBufferSize);
+  mEnd = false;
+
+  if (mBuffer == NULL) {
+    logg->logError(__FILE__, __LINE__, "failed to allocate %d bytes", bufferSize + singleBufferSize);
+    handleException();
+  }
+
+  if (sem_init(&mWaitForSpaceSem, 0, 0)) {
+    logg->logError(__FILE__, __LINE__, "sem_init() failed");
+    handleException();
+  }
+}
+
+Fifo::~Fifo() {
+  free(mBuffer);
+  sem_destroy(&mWaitForSpaceSem);
+}
+
+int Fifo::numBytesFilled() const {
+  return mWrite - mRead + mRaggedEnd;
+}
+
+char* Fifo::start() const {
+  return mBuffer;
+}
+
+bool Fifo::isEmpty() const {
+  return mRead == mWrite && mRaggedEnd == 0;
+}
+
+bool Fifo::isFull() const {
+  return willFill(0);
+}
+
+// Determines if the buffer will fill assuming 'additional' bytes will be added to the buffer
+// 'full' means there is less than singleBufferSize bytes available contiguously; it does not mean there are zero bytes available
+bool Fifo::willFill(int additional) const {
+  if (mWrite > mRead) {
+    if (numBytesFilled() + additional < mWrapThreshold) {
+      return false;
+    }
+  } else {
+    if (numBytesFilled() + additional < mWrapThreshold - mSingleBufferSize) {
+      return false;
+    }
+  }
+  return true;
+}
+
+// This function will stall until contiguous singleBufferSize bytes are available
+char* Fifo::write(int length) {
+  if (length <= 0) {
+    length = 0;
+    mEnd = true;
+  }
+
+  // update the write pointer
+  mWrite += length;
+
+  // handle the wrap-around
+  if (mWrite >= mWrapThreshold) {
+    mRaggedEnd = mWrite;
+    mWrite = 0;
+  }
+
+  // send a notification that data is ready
+  sem_post(mReaderSem);
+
+  // wait for space
+  while (isFull()) {
+    sem_wait(&mWaitForSpaceSem);
+  }
+
+  return &mBuffer[mWrite];
+}
+
+void Fifo::release() {
+  // update the read pointer now that the data has been handled
+  mRead = mReadCommit;
+
+  // handle the wrap-around
+  if (mRead >= mWrapThreshold) {
+    mRaggedEnd = mRead = mReadCommit = 0;
+  }
+
+  // send a notification that data is free (space is available)
+  sem_post(&mWaitForSpaceSem);
+}
+
+// This function will return null if no data is available
+char* Fifo::read(int *const length) {
+  // wait for data
+  if (isEmpty() && !mEnd) {
+    return NULL;
+  }
+
+  // obtain the length
+  do {
+    mReadCommit = mRaggedEnd ? mRaggedEnd : mWrite;
+    *length = mReadCommit - mRead;
+  } while (*length < 0); // plugs race condition without using semaphores
+
+  return &mBuffer[mRead];
+}
diff --git a/tools/gator/daemon/Fifo.h b/tools/gator/daemon/Fifo.h
new file mode 100644
index 000000000000..21c8d8580391
--- /dev/null
+++ b/tools/gator/daemon/Fifo.h
@@ -0,0 +1,48 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef __FIFO_H__
+#define __FIFO_H__
+
+#ifdef WIN32
+#include <windows.h>
+#define sem_t HANDLE
+#define sem_init(sem, pshared, value) ((*(sem) = CreateSemaphore(NULL, value, LONG_MAX, NULL)) == NULL)
+#define sem_wait(sem) WaitForSingleObject(*(sem), INFINITE)
+#define sem_post(sem) ReleaseSemaphore(*(sem), 1, NULL)
+#define sem_destroy(sem) CloseHandle(*(sem))
+#else
+#include <semaphore.h>
+#endif
+
+class Fifo {
+public:
+  Fifo(int singleBufferSize, int totalBufferSize, sem_t* readerSem);
+  ~Fifo();
+  int numBytesFilled() const;
+  bool isEmpty() const;
+  bool isFull() const;
+  bool willFill(int additional) const;
+  char* start() const;
+  char* write(int length);
+  void release();
+  char* read(int *const length);
+
+private:
+  int mSingleBufferSize, mWrite, mRead, mReadCommit, mRaggedEnd, mWrapThreshold;
+  sem_t mWaitForSpaceSem;
+  sem_t* mReaderSem;
+  char* mBuffer;
+  bool mEnd;
+
+  // Intentionally unimplemented
+  Fifo(const Fifo &);
+  Fifo &operator=(const Fifo &);
+};
+
+#endif //__FIFO_H__
diff --git a/tools/gator/daemon/FtraceDriver.cpp b/tools/gator/daemon/FtraceDriver.cpp
new file mode 100644
index 000000000000..b156f1c0b8b4
--- /dev/null
+++ b/tools/gator/daemon/FtraceDriver.cpp
@@ -0,0 +1,118 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "FtraceDriver.h"
+
+#include <regex.h>
+
+#include "Logging.h"
+
+class FtraceCounter : public DriverCounter {
+public:
+	FtraceCounter(DriverCounter *next, char *name, const char *regex);
+	~FtraceCounter();
+
+	int read(const char *const line, int64_t *values);
+
+private:
+	regex_t reg;
+
+	// Intentionally unimplemented
+	FtraceCounter(const FtraceCounter &);
+	FtraceCounter &operator=(const FtraceCounter &);
+};
+
+FtraceCounter::FtraceCounter(DriverCounter *next, char *name, const char *regex) : DriverCounter(next, name) {
+	int result = regcomp(&reg, regex, REG_EXTENDED);
+	if (result != 0) {
+		char buf[128];
+		regerror(result, &reg, buf, sizeof(buf));
+		logg->logError(__FILE__, __LINE__, "Invalid regex '%s': %s", regex, buf);
+		handleException();
+	}
+}
+
+FtraceCounter::~FtraceCounter() {
+	regfree(&reg);
+}
+
+int FtraceCounter::read(const char *const line, int64_t *values) {
+	regmatch_t match[2];
+	int result = regexec(&reg, line, 2, match, 0);
+	if (result != 0) {
+		// No match
+		return 0;
+	}
+
+	if (match[1].rm_so < 0) {
+		logg->logError(__FILE__, __LINE__, "Parsing %s failed", getName());
+		handleException();
+	}
+
+	errno = 0;
+	int64_t value = strtoll(line + match[1].rm_so, NULL, 0);
+	if (errno != 0) {
+		logg->logError(__FILE__, __LINE__, "Parsing %s failed: %s", getName(), strerror(errno));
+		handleException();
+	}
+
+	values[0] = getKey();
+	values[1] = value;
+
+	return 1;
+}
+
+FtraceDriver::FtraceDriver() : mValues(NULL) {
+}
+
+FtraceDriver::~FtraceDriver() {
+	delete [] mValues;
+}
+
+void FtraceDriver::readEvents(mxml_node_t *const xml) {
+	mxml_node_t *node = xml;
+	int count = 0;
+	while (true) {
+		node = mxmlFindElement(node, xml, "event", NULL, NULL, MXML_DESCEND);
+		if (node == NULL) {
+			break;
+		}
+		const char *counter = mxmlElementGetAttr(node, "counter");
+		if (counter == NULL) {
+			continue;
+		}
+
+		if (strncmp(counter, "ftrace_", 7) != 0) {
+			continue;
+		}
+
+		const char *regex = mxmlElementGetAttr(node, "regex");
+		if (regex == NULL) {
+			logg->logError(__FILE__, __LINE__, "The regex counter %s is missing the required regex attribute", counter);
+			handleException();
+		}
+		setCounters(new FtraceCounter(getCounters(), strdup(counter), regex));
+		++count;
+	}
+
+	mValues = new int64_t[2*count];
+}
+
+int FtraceDriver::read(const char *line, int64_t **buf) {
+	int count = 0;
+
+	for (FtraceCounter *counter = static_cast<FtraceCounter *>(getCounters()); counter != NULL; counter = static_cast<FtraceCounter *>(counter->getNext())) {
+		if (!counter->isEnabled()) {
+			continue;
+		}
+		count += counter->read(line, mValues + 2*count);
+	}
+
+	*buf = mValues;
+	return count;
+}
diff --git a/tools/gator/daemon/FtraceDriver.h b/tools/gator/daemon/FtraceDriver.h
new file mode 100644
index 000000000000..5f958bec672c
--- /dev/null
+++ b/tools/gator/daemon/FtraceDriver.h
@@ -0,0 +1,31 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef FTRACEDRIVER_H
+#define FTRACEDRIVER_H
+
+#include "Driver.h"
+
+class FtraceDriver : public SimpleDriver {
+public:
+	FtraceDriver();
+	~FtraceDriver();
+
+	void readEvents(mxml_node_t *const xml);
+
+	int read(const char *line, int64_t **buf);
+
+private:
+	int64_t *mValues;
+
+	// Intentionally unimplemented
+	FtraceDriver(const FtraceDriver &);
+	FtraceDriver &operator=(const FtraceDriver &);
+};
+
+#endif // FTRACEDRIVER_H
diff --git a/tools/gator/daemon/FtraceSource.cpp b/tools/gator/daemon/FtraceSource.cpp
new file mode 100644
index 000000000000..521633357417
--- /dev/null
+++ b/tools/gator/daemon/FtraceSource.cpp
@@ -0,0 +1,158 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "FtraceSource.h"
+
+#include <fcntl.h>
+#include <signal.h>
+#include <sys/prctl.h>
+#include <sys/syscall.h>
+#include <unistd.h>
+
+#include "DriverSource.h"
+#include "Logging.h"
+#include "SessionData.h"
+
+static void handler(int signum)
+{
+	(void)signum;
+};
+
+FtraceSource::FtraceSource(sem_t *senderSem) : mFtraceFh(NULL), mBuffer(0, FRAME_BLOCK_COUNTER, 128*1024, senderSem), mTid(-1), mTracingOn(0) {
+}
+
+FtraceSource::~FtraceSource() {
+}
+
+bool FtraceSource::prepare() {
+	{
+		struct sigaction act;
+		act.sa_handler = handler;
+		act.sa_flags = (int)SA_RESETHAND;
+		if (sigaction(SIGUSR1, &act, NULL) != 0) {
+			logg->logError(__FILE__, __LINE__, "sigaction failed: %s\n", strerror(errno));
+			handleException();
+		}
+	}
+
+	if (DriverSource::readIntDriver("/sys/kernel/debug/tracing/tracing_on", &mTracingOn)) {
+		logg->logError(__FILE__, __LINE__, "Unable to read if ftrace is enabled");
+		handleException();
+	}
+
+	if (DriverSource::writeDriver("/sys/kernel/debug/tracing/tracing_on", "0") != 0) {
+		logg->logError(__FILE__, __LINE__, "Unable to turn ftrace off before truncating the buffer");
+		handleException();
+	}
+
+	{
+		int fd;
+		fd = open("/sys/kernel/debug/tracing/trace", O_WRONLY | O_TRUNC | O_CLOEXEC, 0666);
+		if (fd < 0) {
+			logg->logError(__FILE__, __LINE__, "Unable truncate ftrace buffer: %s", strerror(errno));
+			handleException();
+		}
+		close(fd);
+	}
+
+	if (DriverSource::writeDriver("/sys/kernel/debug/tracing/trace_clock", "perf") != 0) {
+		logg->logError(__FILE__, __LINE__, "Unable to switch ftrace to the perf clock, please ensure you are running Linux 3.10 or later");
+		handleException();
+	}
+
+	mFtraceFh = fopen_cloexec("/sys/kernel/debug/tracing/trace_pipe", "rb");
+	if (mFtraceFh == NULL) {
+		logg->logError(__FILE__, __LINE__, "Unable to open trace_pipe");
+		handleException();
+	}
+
+	return true;
+}
+
+void FtraceSource::run() {
+	prctl(PR_SET_NAME, (unsigned long)&"gatord-ftrace", 0, 0, 0);
+	mTid = syscall(__NR_gettid);
+
+	if (DriverSource::writeDriver("/sys/kernel/debug/tracing/tracing_on", "1") != 0) {
+		logg->logError(__FILE__, __LINE__, "Unable to turn ftrace on");
+		handleException();
+	}
+
+	while (gSessionData->mSessionIsActive) {
+		char buf[1<<12];
+
+		if (fgets(buf, sizeof(buf), mFtraceFh) == NULL) {
+			if (errno == EINTR) {
+				// Interrupted by interrupt - likely user request to terminate
+				break;
+			}
+			logg->logError(__FILE__, __LINE__, "Unable read trace data: %s", strerror(errno));
+			handleException();
+		}
+
+		const uint64_t currTime = getTime();
+
+		char *const colon = strstr(buf, ": ");
+		if (colon == NULL) {
+			logg->logError(__FILE__, __LINE__, "Unable find colon: %s", buf);
+			handleException();
+		}
+		*colon = '\0';
+
+		char *const space = strrchr(buf, ' ');
+		if (space == NULL) {
+			logg->logError(__FILE__, __LINE__, "Unable find space: %s", buf);
+			handleException();
+		}
+		*colon = ':';
+
+		int64_t *data = NULL;
+		int count = gSessionData->ftraceDriver.read(colon + 2, &data);
+		if (count > 0) {
+			errno = 0;
+			const long long time = strtod(space, NULL) * 1000000000;
+			if (errno != 0) {
+				logg->logError(__FILE__, __LINE__, "Unable to parse time: %s", strerror(errno));
+				handleException();
+			}
+			mBuffer.event64(-1, time);
+
+			for (int i = 0; i < count; ++i) {
+				mBuffer.event64(data[2*i + 0], data[2*i + 1]);
+			}
+
+			mBuffer.check(currTime);
+		}
+
+	}
+
+	mBuffer.setDone();
+
+	DriverSource::writeDriver("/sys/kernel/debug/tracing/tracing_on", mTracingOn);
+	fclose(mFtraceFh);
+	DriverSource::writeDriver("/sys/kernel/debug/tracing/trace_clock", "local");
+}
+
+void FtraceSource::interrupt() {
+	// Closing the underlying file handle does not result in the read on the ftrace file handle to return, so send a signal to the thread
+	syscall(__NR_tgkill, getpid(), mTid, SIGUSR1);
+}
+
+bool FtraceSource::isDone() {
+	return mBuffer.isDone();
+}
+
+void FtraceSource::write(Sender *sender) {
+	// Don't send ftrace data until the summary packet is sent so that monotonic delta is available
+	if (!gSessionData->mSentSummary) {
+		return;
+	}
+	if (!mBuffer.isDone()) {
+		mBuffer.write(sender);
+	}
+}
diff --git a/tools/gator/daemon/FtraceSource.h b/tools/gator/daemon/FtraceSource.h
new file mode 100644
index 000000000000..2391b881494e
--- /dev/null
+++ b/tools/gator/daemon/FtraceSource.h
@@ -0,0 +1,43 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef FTRACESOURCE_H
+#define FTRACESOURCE_H
+
+#include <semaphore.h>
+#include <stdio.h>
+
+#include "Buffer.h"
+#include "Source.h"
+
+class FtraceSource : public Source {
+public:
+	FtraceSource(sem_t *senderSem);
+	~FtraceSource();
+
+	bool prepare();
+	void run();
+	void interrupt();
+
+	bool isDone();
+	void write(Sender *sender);
+
+private:
+	void waitFor(const int bytes);
+
+	FILE *mFtraceFh;
+	Buffer mBuffer;
+	int mTid;
+	int mTracingOn;
+
+	// Intentionally unimplemented
+	FtraceSource(const FtraceSource &);
+	FtraceSource &operator=(const FtraceSource &);
+};
+
+#endif // FTRACESOURCE_H
diff --git a/tools/gator/daemon/HwmonDriver.cpp b/tools/gator/daemon/HwmonDriver.cpp
new file mode 100644
index 000000000000..9d161ae5ac56
--- /dev/null
+++ b/tools/gator/daemon/HwmonDriver.cpp
@@ -0,0 +1,245 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "HwmonDriver.h"
+
+#include "libsensors/sensors.h"
+
+#include "Logging.h"
+
+// feature->type to input map
+static sensors_subfeature_type getInput(const sensors_feature_type type) {
+	switch (type) {
+	case SENSORS_FEATURE_IN: return SENSORS_SUBFEATURE_IN_INPUT;
+	case SENSORS_FEATURE_FAN: return SENSORS_SUBFEATURE_FAN_INPUT;
+	case SENSORS_FEATURE_TEMP: return SENSORS_SUBFEATURE_TEMP_INPUT;
+	case SENSORS_FEATURE_POWER: return SENSORS_SUBFEATURE_POWER_INPUT;
+	case SENSORS_FEATURE_ENERGY: return SENSORS_SUBFEATURE_ENERGY_INPUT;
+	case SENSORS_FEATURE_CURR: return SENSORS_SUBFEATURE_CURR_INPUT;
+	case SENSORS_FEATURE_HUMIDITY: return SENSORS_SUBFEATURE_HUMIDITY_INPUT;
+	default:
+		logg->logError(__FILE__, __LINE__, "Unsupported hwmon feature %i", type);
+		handleException();
+	}
+};
+
+class HwmonCounter : public DriverCounter {
+public:
+	HwmonCounter(DriverCounter *next, char *const name, const sensors_chip_name *chip, const sensors_feature *feature);
+	~HwmonCounter();
+
+	const char *getLabel() const { return label; }
+	const char *getTitle() const { return title; }
+	bool isDuplicate() const { return duplicate; }
+	const char *getDisplay() const { return display; }
+	const char *getCounterClass() const { return counter_class; }
+	const char *getUnit() const { return unit; }
+	int getModifier() const { return modifier; }
+
+	int64_t read();
+
+private:
+	void init(const sensors_chip_name *chip, const sensors_feature *feature);
+
+	const sensors_chip_name *chip;
+	const sensors_feature *feature;
+	char *label;
+	const char *title;
+	const char *display;
+	const char *counter_class;
+	const char *unit;
+	double previous_value;
+	int modifier;
+	int monotonic: 1,
+		duplicate : 1;
+
+	// Intentionally unimplemented
+	HwmonCounter(const HwmonCounter &);
+	HwmonCounter &operator=(const HwmonCounter &);
+};
+
+HwmonCounter::HwmonCounter(DriverCounter *next, char *const name, const sensors_chip_name *chip, const sensors_feature *feature) : DriverCounter(next, name), chip(chip), feature(feature), duplicate(false) {
+	label = sensors_get_label(chip, feature);
+
+	switch (feature->type) {
+	case SENSORS_FEATURE_IN:
+		title = "Voltage";
+		display = "maximum";
+		counter_class = "absolute";
+		unit = "V";
+		modifier = 1000;
+		monotonic = false;
+		break;
+	case SENSORS_FEATURE_FAN:
+		title = "Fan";
+		display = "average";
+		counter_class = "absolute";
+		unit = "RPM";
+		modifier = 1;
+		monotonic = false;
+		break;
+	case SENSORS_FEATURE_TEMP:
+		title = "Temperature";
+		display = "maximum";
+		counter_class = "absolute";
+		unit = "°C";
+		modifier = 1000;
+		monotonic = false;
+		break;
+	case SENSORS_FEATURE_POWER:
+		title = "Power";
+		display = "maximum";
+		counter_class = "absolute";
+		unit = "W";
+		modifier = 1000000;
+		monotonic = false;
+		break;
+	case SENSORS_FEATURE_ENERGY:
+		title = "Energy";
+		display = "accumulate";
+		counter_class = "delta";
+		unit = "J";
+		modifier = 1000000;
+		monotonic = true;
+		break;
+	case SENSORS_FEATURE_CURR:
+		title = "Current";
+		display = "maximum";
+		counter_class = "absolute";
+		unit = "A";
+		modifier = 1000;
+		monotonic = false;
+		break;
+	case SENSORS_FEATURE_HUMIDITY:
+		title = "Humidity";
+		display = "average";
+		counter_class = "absolute";
+		unit = "%";
+		modifier = 1000;
+		monotonic = false;
+		break;
+	default:
+		logg->logError(__FILE__, __LINE__, "Unsupported hwmon feature %i", feature->type);
+		handleException();
+	}
+
+	for (HwmonCounter * counter = static_cast<HwmonCounter *>(next); counter != NULL; counter = static_cast<HwmonCounter *>(counter->getNext())) {
+		if (strcmp(label, counter->getLabel()) == 0 && strcmp(title, counter->getTitle()) == 0) {
+			duplicate = true;
+			counter->duplicate = true;
+			break;
+		}
+	}
+}
+
+HwmonCounter::~HwmonCounter() {
+	free((void *)label);
+}
+
+int64_t HwmonCounter::read() {
+	double value;
+	double result;
+	const sensors_subfeature *subfeature;
+
+	// Keep in sync with the read check in HwmonDriver::readEvents
+	subfeature = sensors_get_subfeature(chip, feature, getInput(feature->type));
+	if (!subfeature) {
+		logg->logError(__FILE__, __LINE__, "No input value for hwmon sensor %s", label);
+		handleException();
+	}
+
+	if (sensors_get_value(chip, subfeature->number, &value) != 0) {
+		logg->logError(__FILE__, __LINE__, "Can't get input value for hwmon sensor %s", label);
+		handleException();
+	}
+
+	result = (monotonic ? value - previous_value : value);
+	previous_value = value;
+
+	return result;
+}
+
+HwmonDriver::HwmonDriver() {
+}
+
+HwmonDriver::~HwmonDriver() {
+	sensors_cleanup();
+}
+
+void HwmonDriver::readEvents(mxml_node_t *const) {
+	int err = sensors_init(NULL);
+	if (err) {
+		logg->logMessage("Failed to initialize libsensors! (%d)", err);
+		return;
+	}
+	sensors_sysfs_no_scaling = 1;
+
+	int chip_nr = 0;
+	const sensors_chip_name *chip;
+	while ((chip = sensors_get_detected_chips(NULL, &chip_nr))) {
+		int feature_nr = 0;
+		const sensors_feature *feature;
+		while ((feature = sensors_get_features(chip, &feature_nr))) {
+			// Keep in sync with HwmonCounter::read
+			// Can this counter be read?
+			double value;
+			const sensors_subfeature *const subfeature = sensors_get_subfeature(chip, feature, getInput(feature->type));
+			if ((subfeature == NULL) || (sensors_get_value(chip, subfeature->number, &value) != 0)) {
+				continue;
+			}
+
+			// Get the name of the counter
+			int len = sensors_snprintf_chip_name(NULL, 0, chip) + 1;
+			char *chip_name = new char[len];
+			sensors_snprintf_chip_name(chip_name, len, chip);
+			len = snprintf(NULL, 0, "hwmon_%s_%d_%d", chip_name, chip_nr, feature->number) + 1;
+			char *const name = new char[len];
+			snprintf(name, len, "hwmon_%s_%d_%d", chip_name, chip_nr, feature->number);
+			delete [] chip_name;
+
+			setCounters(new HwmonCounter(getCounters(), name, chip, feature));
+		}
+	}
+}
+
+void HwmonDriver::writeEvents(mxml_node_t *root) const {
+	root = mxmlNewElement(root, "category");
+	mxmlElementSetAttr(root, "name", "hwmon");
+
+	char buf[1024];
+	for (HwmonCounter *counter = static_cast<HwmonCounter *>(getCounters()); counter != NULL; counter = static_cast<HwmonCounter *>(counter->getNext())) {
+		mxml_node_t *node = mxmlNewElement(root, "event");
+		mxmlElementSetAttr(node, "counter", counter->getName());
+		mxmlElementSetAttr(node, "title", counter->getTitle());
+		if (counter->isDuplicate()) {
+			mxmlElementSetAttrf(node, "name", "%s (0x%x)", counter->getLabel(), counter->getKey());
+		} else {
+			mxmlElementSetAttr(node, "name", counter->getLabel());
+		}
+		mxmlElementSetAttr(node, "display", counter->getDisplay());
+		mxmlElementSetAttr(node, "class", counter->getCounterClass());
+		mxmlElementSetAttr(node, "units", counter->getUnit());
+		if (counter->getModifier() != 1) {
+			mxmlElementSetAttrf(node, "modifier", "%d", counter->getModifier());
+		}
+		if (strcmp(counter->getDisplay(), "average") == 0 || strcmp(counter->getDisplay(), "maximum") == 0) {
+			mxmlElementSetAttr(node, "average_selection", "yes");
+		}
+		snprintf(buf, sizeof(buf), "libsensors %s sensor %s (%s)", counter->getTitle(), counter->getLabel(), counter->getName());
+		mxmlElementSetAttr(node, "description", buf);
+	}
+}
+
+void HwmonDriver::start() {
+	for (DriverCounter *counter = getCounters(); counter != NULL; counter = counter->getNext()) {
+		if (!counter->isEnabled()) {
+			continue;
+		}
+		counter->read();
+	}
+}
diff --git a/tools/gator/daemon/HwmonDriver.h b/tools/gator/daemon/HwmonDriver.h
new file mode 100644
index 000000000000..f28d825e3b7b
--- /dev/null
+++ b/tools/gator/daemon/HwmonDriver.h
@@ -0,0 +1,31 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef HWMONDRIVER_H
+#define HWMONDRIVER_H
+
+#include "Driver.h"
+
+class HwmonDriver : public PolledDriver {
+public:
+	HwmonDriver();
+	~HwmonDriver();
+
+	void readEvents(mxml_node_t *const root);
+
+	void writeEvents(mxml_node_t *root) const;
+
+	void start();
+
+private:
+	// Intentionally unimplemented
+	HwmonDriver(const HwmonDriver &);
+	HwmonDriver &operator=(const HwmonDriver &);
+};
+
+#endif // HWMONDRIVER_H
diff --git a/tools/gator/daemon/KMod.cpp b/tools/gator/daemon/KMod.cpp
new file mode 100644
index 000000000000..fe9dc6a7e4f7
--- /dev/null
+++ b/tools/gator/daemon/KMod.cpp
@@ -0,0 +1,119 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "KMod.h"
+
+#include <sys/types.h>
+#include <dirent.h>
+#include <unistd.h>
+
+#include "ConfigurationXML.h"
+#include "Counter.h"
+#include "DriverSource.h"
+#include "Logging.h"
+#include "SessionData.h"
+
+// Claim all the counters in /dev/gator/events
+bool KMod::claimCounter(const Counter &counter) const {
+	char text[128];
+	snprintf(text, sizeof(text), "/dev/gator/events/%s", counter.getType());
+	return access(text, F_OK) == 0;
+}
+
+void KMod::resetCounters() {
+	char base[128];
+	char text[128];
+
+	// Initialize all perf counters in the driver, i.e. set enabled to zero
+	struct dirent *ent;
+	DIR* dir = opendir("/dev/gator/events");
+	if (dir) {
+		while ((ent = readdir(dir)) != NULL) {
+			// skip hidden files, current dir, and parent dir
+			if (ent->d_name[0] == '.')
+				continue;
+			snprintf(base, sizeof(base), "/dev/gator/events/%s", ent->d_name);
+			snprintf(text, sizeof(text), "%s/enabled", base);
+			DriverSource::writeDriver(text, 0);
+			snprintf(text, sizeof(text), "%s/count", base);
+			DriverSource::writeDriver(text, 0);
+		}
+		closedir(dir);
+	}
+}
+
+static const char ARM_MALI_MIDGARD[] = "ARM_Mali-Midgard_";
+static const char ARM_MALI_T[] = "ARM_Mali-T";
+
+void KMod::setupCounter(Counter &counter) {
+	char base[128];
+	char text[128];
+	snprintf(base, sizeof(base), "/dev/gator/events/%s", counter.getType());
+
+	if ((strncmp(counter.getType(), ARM_MALI_MIDGARD, sizeof(ARM_MALI_MIDGARD) - 1) == 0 ||
+	     strncmp(counter.getType(), ARM_MALI_T, sizeof(ARM_MALI_T) - 1) == 0)) {
+		mIsMaliCapture = true;
+	}
+
+	snprintf(text, sizeof(text), "%s/enabled", base);
+	int enabled = true;
+	if (DriverSource::writeReadDriver(text, &enabled) || !enabled) {
+		counter.setEnabled(false);
+		return;
+	}
+
+	int value = 0;
+	snprintf(text, sizeof(text), "%s/key", base);
+	DriverSource::readIntDriver(text, &value);
+	counter.setKey(value);
+
+	snprintf(text, sizeof(text), "%s/cores", base);
+	if (DriverSource::readIntDriver(text, &value) == 0) {
+		counter.setCores(value);
+	}
+
+	snprintf(text, sizeof(text), "%s/event", base);
+	DriverSource::writeDriver(text, counter.getEvent());
+	snprintf(text, sizeof(text), "%s/count", base);
+	if (access(text, F_OK) == 0) {
+		int count = counter.getCount();
+		if (DriverSource::writeReadDriver(text, &count) && counter.getCount() > 0) {
+			logg->logError(__FILE__, __LINE__, "Cannot enable EBS for %s:%i with a count of %d\n", counter.getType(), counter.getEvent(), counter.getCount());
+			handleException();
+		}
+		counter.setCount(count);
+	} else if (counter.getCount() > 0) {
+		ConfigurationXML::remove();
+		logg->logError(__FILE__, __LINE__, "Event Based Sampling is only supported with kernel versions 3.0.0 and higher with CONFIG_PERF_EVENTS=y, and CONFIG_HW_PERF_EVENTS=y. The invalid configuration.xml has been removed.\n");
+		handleException();
+	}
+}
+
+int KMod::writeCounters(mxml_node_t *root) const {
+	struct dirent *ent;
+	mxml_node_t *counter;
+
+	// counters.xml is simply a file listing of /dev/gator/events
+	DIR* dir = opendir("/dev/gator/events");
+	if (dir == NULL) {
+		return 0;
+	}
+
+	int count = 0;
+	while ((ent = readdir(dir)) != NULL) {
+		// skip hidden files, current dir, and parent dir
+		if (ent->d_name[0] == '.')
+			continue;
+		counter = mxmlNewElement(root, "counter");
+		mxmlElementSetAttr(counter, "name", ent->d_name);
+		++count;
+	}
+	closedir(dir);
+
+	return count;
+}
diff --git a/tools/gator/daemon/KMod.h b/tools/gator/daemon/KMod.h
new file mode 100644
index 000000000000..900a60e87d24
--- /dev/null
+++ b/tools/gator/daemon/KMod.h
@@ -0,0 +1,32 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef KMOD_H
+#define KMOD_H
+
+#include "Driver.h"
+
+// Driver for the gator kernel module
+class KMod : public Driver {
+public:
+	KMod() : mIsMaliCapture(false) {}
+	~KMod() {}
+
+	bool claimCounter(const Counter &counter) const;
+	void resetCounters();
+	void setupCounter(Counter &counter);
+
+	int writeCounters(mxml_node_t *root) const;
+
+	bool isMaliCapture() const { return mIsMaliCapture; }
+
+private:
+	bool mIsMaliCapture;
+};
+
+#endif // KMOD_H
diff --git a/tools/gator/daemon/LocalCapture.cpp b/tools/gator/daemon/LocalCapture.cpp
new file mode 100644
index 000000000000..d2a4b799d7ac
--- /dev/null
+++ b/tools/gator/daemon/LocalCapture.cpp
@@ -0,0 +1,131 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "LocalCapture.h"
+
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <dirent.h>
+#include <string.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+#include "SessionData.h"
+#include "Logging.h"
+#include "OlyUtility.h"
+#include "EventsXML.h"
+
+LocalCapture::LocalCapture() {}
+
+LocalCapture::~LocalCapture() {}
+
+void LocalCapture::createAPCDirectory(char* target_path) {
+	gSessionData->mAPCDir = createUniqueDirectory(target_path, ".apc");
+	if ((removeDirAndAllContents(gSessionData->mAPCDir) != 0 || mkdir(gSessionData->mAPCDir, S_IRWXU | S_IRWXG | S_IROTH | S_IXOTH) != 0)) {
+		logg->logError(__FILE__, __LINE__, "Unable to create directory %s", gSessionData->mAPCDir);
+		handleException();
+	}
+}
+
+void LocalCapture::write(char* string) {
+	char file[PATH_MAX];
+
+	// Set full path
+	snprintf(file, PATH_MAX, "%s/session.xml", gSessionData->mAPCDir);
+
+	// Write the file
+	if (util->writeToDisk(file, string) < 0) {
+		logg->logError(__FILE__, __LINE__, "Error writing %s\nPlease verify the path.", file);
+		handleException();
+	}
+
+	// Write events XML
+	EventsXML eventsXML;
+	eventsXML.write(gSessionData->mAPCDir);
+}
+
+char* LocalCapture::createUniqueDirectory(const char* initialPath, const char* ending) {
+	char* output;
+	char path[PATH_MAX];
+
+	// Ensure the path is an absolute path, i.e. starts with a slash
+	if (initialPath == 0 || strlen(initialPath) == 0) {
+		logg->logError(__FILE__, __LINE__, "Missing -o command line option required for a local capture.");
+		handleException();
+	} else if (initialPath[0] != '/') {
+		if (getcwd(path, PATH_MAX) == 0) {
+			logg->logMessage("Unable to retrieve the current working directory");
+		}
+		strncat(path, "/", PATH_MAX - strlen(path) - 1);
+		strncat(path, initialPath, PATH_MAX - strlen(path) - 1);
+	} else {
+		strncpy(path, initialPath, PATH_MAX);
+		path[PATH_MAX - 1] = 0; // strncpy does not guarantee a null-terminated string
+	}
+
+	// Add ending if it is not already there
+	if (strcmp(&path[strlen(path) - strlen(ending)], ending) != 0) {
+		strncat(path, ending, PATH_MAX - strlen(path) - 1);
+	}
+
+	output = strdup(path);
+
+	return output;
+}
+
+int LocalCapture::removeDirAndAllContents(char* path) {
+	int error = 0;
+	struct stat mFileInfo;
+	// Does the path exist?
+	if (stat(path, &mFileInfo) == 0) {
+		// Is it a directory?
+		if (mFileInfo.st_mode & S_IFDIR) {
+			DIR * dir = opendir(path);
+			dirent* entry = readdir(dir);
+			while (entry) {
+				if (strcmp(entry->d_name, ".") != 0 && strcmp(entry->d_name, "..") != 0) {
+					char* newpath = (char*)malloc(strlen(path) + strlen(entry->d_name) + 2);
+					sprintf(newpath, "%s/%s", path, entry->d_name);
+					error = removeDirAndAllContents(newpath);
+					free(newpath);
+					if (error) {
+						break;
+					}
+				}
+				entry = readdir(dir);
+			}
+			closedir(dir);
+			if (error == 0) {
+				error = rmdir(path);
+			}
+		} else {
+			error = remove(path);
+		}
+	}
+	return error;
+}
+
+void LocalCapture::copyImages(ImageLinkList* ptr) {
+	char dstfilename[PATH_MAX];
+
+	while (ptr) {
+		strncpy(dstfilename, gSessionData->mAPCDir, PATH_MAX);
+		dstfilename[PATH_MAX - 1] = 0; // strncpy does not guarantee a null-terminated string
+		if (gSessionData->mAPCDir[strlen(gSessionData->mAPCDir) - 1] != '/') {
+			strncat(dstfilename, "/", PATH_MAX - strlen(dstfilename) - 1);
+		}
+		strncat(dstfilename, util->getFilePart(ptr->path), PATH_MAX - strlen(dstfilename) - 1);
+		if (util->copyFile(ptr->path, dstfilename)) {
+			logg->logMessage("copied file %s to %s", ptr->path, dstfilename);
+		} else {
+			logg->logMessage("copy of file %s to %s failed", ptr->path, dstfilename);
+		}
+
+		ptr = ptr->next;
+	}
+}
diff --git a/tools/gator/daemon/LocalCapture.h b/tools/gator/daemon/LocalCapture.h
new file mode 100644
index 000000000000..25d281f8328b
--- /dev/null
+++ b/tools/gator/daemon/LocalCapture.h
@@ -0,0 +1,26 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef __LOCAL_CAPTURE_H__
+#define __LOCAL_CAPTURE_H__
+
+struct ImageLinkList;
+
+class LocalCapture {
+public:
+	LocalCapture();
+	~LocalCapture();
+	void write(char* string);
+	void copyImages(ImageLinkList* ptr);
+	void createAPCDirectory(char* target_path);
+private:
+	char* createUniqueDirectory(const char* path, const char* ending);
+	int removeDirAndAllContents(char* path);
+};
+
+#endif //__LOCAL_CAPTURE_H__
diff --git a/tools/gator/daemon/Logging.cpp b/tools/gator/daemon/Logging.cpp
new file mode 100644
index 000000000000..41ffa1a45151
--- /dev/null
+++ b/tools/gator/daemon/Logging.cpp
@@ -0,0 +1,78 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "Logging.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdarg.h>
+#include <string.h>
+
+#ifdef WIN32
+#define MUTEX_INIT()    mLoggingMutex = CreateMutex(NULL, false, NULL);
+#define MUTEX_LOCK()    WaitForSingleObject(mLoggingMutex, 0xFFFFFFFF);
+#define MUTEX_UNLOCK()  ReleaseMutex(mLoggingMutex);
+#define snprintf _snprintf
+#else
+#include <pthread.h>
+#define MUTEX_INIT()    pthread_mutex_init(&mLoggingMutex, NULL)
+#define MUTEX_LOCK()    pthread_mutex_lock(&mLoggingMutex)
+#define MUTEX_UNLOCK()  pthread_mutex_unlock(&mLoggingMutex)
+#endif
+
+// Global thread-safe logging
+Logging* logg = NULL;
+
+Logging::Logging(bool debug) {
+	mDebug = debug;
+	MUTEX_INIT();
+
+	strcpy(mErrBuf, "Unknown Error");
+	strcpy(mLogBuf, "Unknown Message");
+}
+
+Logging::~Logging() {
+}
+
+void Logging::logError(const char* file, int line, const char* fmt, ...) {
+	va_list args;
+
+	MUTEX_LOCK();
+	if (mDebug) {
+		snprintf(mErrBuf, sizeof(mErrBuf), "ERROR[%s:%d]: ", file, line);
+	} else {
+		mErrBuf[0] = 0;
+	}
+
+	va_start(args, fmt);
+	vsnprintf(mErrBuf + strlen(mErrBuf), sizeof(mErrBuf) - 2 - strlen(mErrBuf), fmt, args); //  subtract 2 for \n and \0
+	va_end(args);
+
+	if (strlen(mErrBuf) > 0) {
+		strcat(mErrBuf, "\n");
+	}
+	MUTEX_UNLOCK();
+}
+
+void Logging::logMessage(const char* fmt, ...) {
+	if (mDebug) {
+		va_list args;
+
+		MUTEX_LOCK();
+		strcpy(mLogBuf, "INFO: ");
+
+		va_start(args, fmt);
+		vsnprintf(mLogBuf + strlen(mLogBuf), sizeof(mLogBuf) - 2 - strlen(mLogBuf), fmt, args); //  subtract 2 for \n and \0
+		va_end(args);
+		strcat(mLogBuf, "\n");
+
+		fprintf(stdout, "%s", mLogBuf);
+		fflush(stdout);
+		MUTEX_UNLOCK();
+	}
+}
diff --git a/tools/gator/daemon/Logging.h b/tools/gator/daemon/Logging.h
new file mode 100644
index 000000000000..09e93ff13f7a
--- /dev/null
+++ b/tools/gator/daemon/Logging.h
@@ -0,0 +1,36 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef __LOGGING_H__
+#define __LOGGING_H__
+
+#include <pthread.h>
+
+#define DRIVER_ERROR "\n Driver issue:\n  >> gator.ko must be built against the current kernel version & configuration\n  >> gator.ko should be co-located with gatord in the same directory\n  >>   OR insmod gator.ko prior to launching gatord"
+
+class Logging {
+public:
+	Logging(bool debug);
+	~Logging();
+	void logError(const char* file, int line, const char* fmt, ...);
+	void logMessage(const char* fmt, ...);
+	char* getLastError() {return mErrBuf;}
+	char* getLastMessage() {return mLogBuf;}
+
+private:
+	char mErrBuf[4096]; // Arbitrarily large buffer to hold a string
+	char mLogBuf[4096]; // Arbitrarily large buffer to hold a string
+	bool mDebug;
+	pthread_mutex_t mLoggingMutex;
+};
+
+extern Logging* logg;
+
+extern void handleException() __attribute__ ((noreturn));
+
+#endif //__LOGGING_H__
diff --git a/tools/gator/daemon/Makefile b/tools/gator/daemon/Makefile
new file mode 100644
index 000000000000..27531b438b63
--- /dev/null
+++ b/tools/gator/daemon/Makefile
@@ -0,0 +1,22 @@
+#
+# Makefile for ARM Streamline - Gator Daemon
+#
+
+# Uncomment and define CROSS_COMPILE if it is not already defined
+# CROSS_COMPILE=/path/to/cross-compiler/arm-linux-gnueabihf-
+# NOTE: This toolchain uses the hardfloat abi by default. For non-hardfloat
+# targets run 'make SOFTFLOAT=1 SYSROOT=/path/to/sysroot', see
+# README_Streamline.txt for more details
+
+CC = $(CROSS_COMPILE)gcc
+CXX = $(CROSS_COMPILE)g++
+
+ifeq ($(SOFTFLOAT),1)
+	CPPFLAGS += -marm -mthumb-interwork -march=armv4t -mfloat-abi=soft
+	LDFLAGS += -marm -march=armv4t -mfloat-abi=soft
+endif
+ifneq ($(SYSROOT),)
+	LDFLAGS += --sysroot=$(SYSROOT)
+endif
+
+include common.mk
diff --git a/tools/gator/daemon/Makefile_aarch64 b/tools/gator/daemon/Makefile_aarch64
new file mode 100644
index 000000000000..efd1fa002182
--- /dev/null
+++ b/tools/gator/daemon/Makefile_aarch64
@@ -0,0 +1,12 @@
+#
+# Makefile for ARM Streamline - Gator Daemon
+# make -f Makefile_aarch64
+#
+
+# Uncomment and define CROSS_COMPILE if it is not already defined
+# CROSS_COMPILE=/path/to/cross-compiler/aarch64-linux-gnu-
+
+CC = $(CROSS_COMPILE)gcc
+CXX = $(CROSS_COMPILE)g++
+
+include common.mk
diff --git a/tools/gator/daemon/MaliVideoDriver.cpp b/tools/gator/daemon/MaliVideoDriver.cpp
new file mode 100644
index 000000000000..5eef2643ab15
--- /dev/null
+++ b/tools/gator/daemon/MaliVideoDriver.cpp
@@ -0,0 +1,191 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "MaliVideoDriver.h"
+
+#include <unistd.h>
+
+#include "Buffer.h"
+#include "Counter.h"
+#include "Logging.h"
+#include "SessionData.h"
+
+// From instr/src/mve_instr_comm_protocol.h
+typedef enum mve_instr_configuration_type {
+	MVE_INSTR_RAW         = 1 << 0,
+	MVE_INSTR_COUNTERS    = 1 << 1,
+	MVE_INSTR_EVENTS      = 1 << 2,
+	MVE_INSTR_ACTIVITIES  = 1 << 3,
+
+	// Raw always pushed regardless
+	MVE_INSTR_PULL        = 1 << 12,
+	// Raw always unpacked regardless
+	MVE_INSTR_PACKED_COMM = 1 << 13,
+	// Don’t send ACKt response
+	MVE_INSTR_NO_AUTO_ACK   = 1 << 14,
+} mve_instr_configuration_type_t;
+
+static const char COUNTER[] = "ARM_Mali-V500_cnt";
+static const char EVENT[] = "ARM_Mali-V500_evn";
+static const char ACTIVITY[] = "ARM_Mali-V500_act";
+
+class MaliVideoCounter : public DriverCounter {
+public:
+	MaliVideoCounter(DriverCounter *next, const char *name, const MaliVideoCounterType type, const int id) : DriverCounter(next, name), mType(type), mId(id) {
+	}
+
+	~MaliVideoCounter() {
+	}
+
+	MaliVideoCounterType getType() const { return mType; }
+	int getId() const { return mId; }
+
+private:
+	const MaliVideoCounterType mType;
+	// Mali Video id
+	const int mId;
+};
+
+MaliVideoDriver::MaliVideoDriver() {
+}
+
+MaliVideoDriver::~MaliVideoDriver() {
+}
+
+void MaliVideoDriver::readEvents(mxml_node_t *const xml) {
+	mxml_node_t *node = xml;
+	while (true) {
+		node = mxmlFindElement(node, xml, "event", NULL, NULL, MXML_DESCEND);
+		if (node == NULL) {
+			break;
+		}
+		const char *counter = mxmlElementGetAttr(node, "counter");
+		if (counter == NULL) {
+			// Ignore
+		} else if (strncmp(counter, COUNTER, sizeof(COUNTER) - 1) == 0) {
+			const int i = strtol(counter + sizeof(COUNTER) - 1, NULL, 10);
+			setCounters(new MaliVideoCounter(getCounters(), strdup(counter), MVCT_COUNTER, i));
+		} else if (strncmp(counter, EVENT, sizeof(EVENT) - 1) == 0) {
+			const int i = strtol(counter + sizeof(EVENT) - 1, NULL, 10);
+			setCounters(new MaliVideoCounter(getCounters(), strdup(counter), MVCT_EVENT, i));
+		} else if (strncmp(counter, ACTIVITY, sizeof(ACTIVITY) - 1) == 0) {
+			const int i = strtol(counter + sizeof(ACTIVITY) - 1, NULL, 10);
+			setCounters(new MaliVideoCounter(getCounters(), strdup(counter), MVCT_ACTIVITY, i));
+		}
+	}
+}
+
+int MaliVideoDriver::writeCounters(mxml_node_t *root) const {
+	if (access("/dev/mv500", F_OK) != 0) {
+		return 0;
+	}
+
+	return super::writeCounters(root);
+}
+
+void MaliVideoDriver::marshalEnable(const MaliVideoCounterType type, char *const buf, const size_t bufsize, int &pos) {
+	// size
+	int numEnabled = 0;
+	for (MaliVideoCounter *counter = static_cast<MaliVideoCounter *>(getCounters()); counter != NULL; counter = static_cast<MaliVideoCounter *>(counter->getNext())) {
+		if (counter->isEnabled() && (counter->getType() == type)) {
+			++numEnabled;
+		}
+	}
+	Buffer::packInt(buf, bufsize, pos, numEnabled*sizeof(uint32_t));
+	for (MaliVideoCounter *counter = static_cast<MaliVideoCounter *>(getCounters()); counter != NULL; counter = static_cast<MaliVideoCounter *>(counter->getNext())) {
+		if (counter->isEnabled() && (counter->getType() == type)) {
+			Buffer::packInt(buf, bufsize, pos, counter->getId());
+		}
+	}
+}
+
+static bool writeAll(const int mveUds, const char *const buf, const int pos) {
+	int written = 0;
+	while (written < pos) {
+		size_t bytes = ::write(mveUds, buf + written, pos - written);
+		if (bytes <= 0) {
+			logg->logMessage("%s(%s:%i): write failed", __FUNCTION__, __FILE__, __LINE__);
+			return false;
+		}
+		written += bytes;
+	}
+
+	return true;
+}
+
+bool MaliVideoDriver::start(const int mveUds) {
+	char buf[256];
+	int pos = 0;
+
+	// code - MVE_INSTR_STARTUP
+	buf[pos++] = 'C';
+	buf[pos++] = 'L';
+	buf[pos++] = 'N';
+	buf[pos++] = 'T';
+	// size
+	Buffer::packInt(buf, sizeof(buf), pos, sizeof(uint32_t));
+	// client_version_number
+	Buffer::packInt(buf, sizeof(buf), pos, 1);
+
+	// code - MVE_INSTR_CONFIGURE
+	buf[pos++] = 'C';
+	buf[pos++] = 'N';
+	buf[pos++] = 'F';
+	buf[pos++] = 'G';
+	// size
+	Buffer::packInt(buf, sizeof(buf), pos, 5*sizeof(uint32_t));
+	// configuration
+	Buffer::packInt(buf, sizeof(buf), pos, MVE_INSTR_COUNTERS | MVE_INSTR_EVENTS | MVE_INSTR_ACTIVITIES | MVE_INSTR_PACKED_COMM);
+	// communication_protocol_version
+	Buffer::packInt(buf, sizeof(buf), pos, 1);
+	// data_protocol_version
+	Buffer::packInt(buf, sizeof(buf), pos, 1);
+	// sample_rate - convert samples/second to ms/sample
+	Buffer::packInt(buf, sizeof(buf), pos, 1000/gSessionData->mSampleRate);
+	// live_rate - convert ns/flush to ms/flush
+	Buffer::packInt(buf, sizeof(buf), pos, gSessionData->mLiveRate/1000000);
+
+	// code - MVE_INSTR_ENABLE_COUNTERS
+	buf[pos++] = 'C';
+	buf[pos++] = 'F';
+	buf[pos++] = 'G';
+	buf[pos++] = 'c';
+	marshalEnable(MVCT_COUNTER, buf, sizeof(buf), pos);
+
+	// code - MVE_INSTR_ENABLE_EVENTS
+	buf[pos++] = 'C';
+	buf[pos++] = 'F';
+	buf[pos++] = 'G';
+	buf[pos++] = 'e';
+	marshalEnable(MVCT_EVENT, buf, sizeof(buf), pos);
+
+	// code - MVE_INSTR_ENABLE_ACTIVITIES
+	buf[pos++] = 'C';
+	buf[pos++] = 'F';
+	buf[pos++] = 'G';
+	buf[pos++] = 'a';
+	marshalEnable(MVCT_ACTIVITY, buf, sizeof(buf), pos);
+
+	return writeAll(mveUds, buf, pos);
+}
+
+void MaliVideoDriver::stop(const int mveUds) {
+	char buf[8];
+	int pos = 0;
+
+	// code - MVE_INSTR_STOP
+	buf[pos++] = 'S';
+	buf[pos++] = 'T';
+	buf[pos++] = 'O';
+	buf[pos++] = 'P';
+	marshalEnable(MVCT_COUNTER, buf, sizeof(buf), pos);
+
+	writeAll(mveUds, buf, pos);
+
+	close(mveUds);
+}
diff --git a/tools/gator/daemon/MaliVideoDriver.h b/tools/gator/daemon/MaliVideoDriver.h
new file mode 100644
index 000000000000..204a57a447ac
--- /dev/null
+++ b/tools/gator/daemon/MaliVideoDriver.h
@@ -0,0 +1,45 @@
+/**
+ * Copyright (C) ARM Limited 2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef MALIVIDEODRIVER_H
+#define MALIVIDEODRIVER_H
+
+#include "Driver.h"
+
+class MaliVideoCounter;
+
+enum MaliVideoCounterType {
+	MVCT_COUNTER,
+	MVCT_EVENT,
+	MVCT_ACTIVITY,
+};
+
+class MaliVideoDriver : public SimpleDriver {
+private:
+	typedef SimpleDriver super;
+
+public:
+	MaliVideoDriver();
+	~MaliVideoDriver();
+
+	void readEvents(mxml_node_t *const root);
+
+	int writeCounters(mxml_node_t *root) const;
+
+	bool start(const int mveUds);
+	void stop(const int mveUds);
+
+private:
+	void marshalEnable(const MaliVideoCounterType type, char *const buf, const size_t bufsize, int &pos);
+
+	// Intentionally unimplemented
+	MaliVideoDriver(const MaliVideoDriver &);
+	MaliVideoDriver &operator=(const MaliVideoDriver &);
+};
+
+#endif // MALIVIDEODRIVER_H
diff --git a/tools/gator/daemon/MemInfoDriver.cpp b/tools/gator/daemon/MemInfoDriver.cpp
new file mode 100644
index 000000000000..cce15c16fcdc
--- /dev/null
+++ b/tools/gator/daemon/MemInfoDriver.cpp
@@ -0,0 +1,93 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "MemInfoDriver.h"
+
+#include "Logging.h"
+#include "SessionData.h"
+
+class MemInfoCounter : public DriverCounter {
+public:
+	MemInfoCounter(DriverCounter *next, char *const name, int64_t *const value);
+	~MemInfoCounter();
+
+	int64_t read();
+
+private:
+	int64_t *const mValue;
+
+	// Intentionally unimplemented
+	MemInfoCounter(const MemInfoCounter &);
+	MemInfoCounter &operator=(const MemInfoCounter &);
+};
+
+MemInfoCounter::MemInfoCounter(DriverCounter *next, char *const name, int64_t *const value) : DriverCounter(next, name), mValue(value) {
+}
+
+MemInfoCounter::~MemInfoCounter() {
+}
+
+int64_t MemInfoCounter::read() {
+	return *mValue;
+}
+
+MemInfoDriver::MemInfoDriver() : mBuf(), mMemUsed(0), mMemFree(0), mBuffers(0) {
+}
+
+MemInfoDriver::~MemInfoDriver() {
+}
+
+void MemInfoDriver::readEvents(mxml_node_t *const) {
+	// Only for use with perf
+	if (!gSessionData->perf.isSetup()) {
+		return;
+	}
+
+	setCounters(new MemInfoCounter(getCounters(), strdup("Linux_meminfo_memused2"), &mMemUsed));
+	setCounters(new MemInfoCounter(getCounters(), strdup("Linux_meminfo_memfree"), &mMemFree));
+	setCounters(new MemInfoCounter(getCounters(), strdup("Linux_meminfo_bufferram"), &mBuffers));
+}
+
+void MemInfoDriver::read(Buffer *const buffer) {
+	if (!countersEnabled()) {
+		return;
+	}
+
+	if (!mBuf.read("/proc/meminfo")) {
+		logg->logError(__FILE__, __LINE__, "Failed to read /proc/meminfo");
+		handleException();
+	}
+
+	char *key = mBuf.getBuf();
+	char *colon;
+	int64_t memTotal = 0;
+	while ((colon = strchr(key, ':')) != NULL) {
+		char *end = strchr(colon + 1, '\n');
+		if (end != NULL) {
+			*end = '\0';
+		}
+		*colon = '\0';
+
+		if (strcmp(key, "MemTotal") == 0) {
+			memTotal = strtoll(colon + 1, NULL, 10) << 10;
+		} else if (strcmp(key, "MemFree") == 0) {
+			mMemFree = strtoll(colon + 1, NULL, 10) << 10;
+		} else if (strcmp(key, "Buffers") == 0) {
+			mBuffers = strtoll(colon + 1, NULL, 10) << 10;
+		}
+
+		if (end == NULL) {
+			break;
+		}
+		key = end + 1;
+	}
+
+	mMemUsed = memTotal - mMemFree;
+
+	super::read(buffer);
+}
diff --git a/tools/gator/daemon/MemInfoDriver.h b/tools/gator/daemon/MemInfoDriver.h
new file mode 100644
index 000000000000..eb1b0417f309
--- /dev/null
+++ b/tools/gator/daemon/MemInfoDriver.h
@@ -0,0 +1,37 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef MEMINFODRIVER_H
+#define MEMINFODRIVER_H
+
+#include "Driver.h"
+#include "DynBuf.h"
+
+class MemInfoDriver : public PolledDriver {
+private:
+	typedef PolledDriver super;
+
+public:
+	MemInfoDriver();
+	~MemInfoDriver();
+
+	void readEvents(mxml_node_t *const root);
+	void read(Buffer *const buffer);
+
+private:
+	DynBuf mBuf;
+	int64_t mMemUsed;
+	int64_t mMemFree;
+	int64_t mBuffers;
+
+	// Intentionally unimplemented
+	MemInfoDriver(const MemInfoDriver &);
+	MemInfoDriver &operator=(const MemInfoDriver &);
+};
+
+#endif // MEMINFODRIVER_H
diff --git a/tools/gator/daemon/Monitor.cpp b/tools/gator/daemon/Monitor.cpp
new file mode 100644
index 000000000000..74f22ee29fec
--- /dev/null
+++ b/tools/gator/daemon/Monitor.cpp
@@ -0,0 +1,82 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "Monitor.h"
+
+#include <errno.h>
+#include <fcntl.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "Logging.h"
+
+Monitor::Monitor() : mFd(-1) {
+}
+
+Monitor::~Monitor() {
+	if (mFd >= 0) {
+		::close(mFd);
+	}
+}
+
+void Monitor::close() {
+	if (mFd >= 0) {
+		::close(mFd);
+		mFd = -1;
+	}
+}
+
+bool Monitor::init() {
+#ifdef EPOLL_CLOEXEC
+	mFd = epoll_create1(EPOLL_CLOEXEC);
+#else
+	mFd = epoll_create(16);
+#endif
+	if (mFd < 0) {
+		logg->logMessage("%s(%s:%i): epoll_create1 failed", __FUNCTION__, __FILE__, __LINE__);
+		return false;
+	}
+
+#ifndef EPOLL_CLOEXEC
+  int fdf = fcntl(mFd, F_GETFD);
+  if ((fdf == -1) || (fcntl(mFd, F_SETFD, fdf | FD_CLOEXEC) != 0)) {
+		logg->logMessage("%s(%s:%i): fcntl failed", __FUNCTION__, __FILE__, __LINE__);
+    ::close(mFd);
+    return -1;
+  }
+#endif
+
+	return true;
+}
+
+bool Monitor::add(const int fd) {
+	struct epoll_event event;
+	memset(&event, 0, sizeof(event));
+	event.data.fd = fd;
+	event.events = EPOLLIN;
+	if (epoll_ctl(mFd, EPOLL_CTL_ADD, fd, &event) != 0) {
+		logg->logMessage("%s(%s:%i): epoll_ctl failed", __FUNCTION__, __FILE__, __LINE__);
+		return false;
+	}
+
+	return true;
+}
+
+int Monitor::wait(struct epoll_event *const events, int maxevents, int timeout) {
+	int result = epoll_wait(mFd, events, maxevents, timeout);
+	if (result < 0) {
+		// Ignore if the call was interrupted as this will happen when SIGINT is received
+		if (errno == EINTR) {
+			result = 0;
+		} else {
+			logg->logMessage("%s(%s:%i): epoll_wait failed", __FUNCTION__, __FILE__, __LINE__);
+		}
+	}
+
+	return result;
+}
diff --git a/tools/gator/daemon/Monitor.h b/tools/gator/daemon/Monitor.h
new file mode 100644
index 000000000000..7194e0e4ca50
--- /dev/null
+++ b/tools/gator/daemon/Monitor.h
@@ -0,0 +1,33 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef MONITOR_H
+#define MONITOR_H
+
+#include <sys/epoll.h>
+
+class Monitor {
+public:
+	Monitor();
+	~Monitor();
+
+	void close();
+	bool init();
+	bool add(const int fd);
+	int wait(struct epoll_event *const events, int maxevents, int timeout);
+
+private:
+
+	int mFd;
+
+	// Intentionally unimplemented
+	Monitor(const Monitor &);
+	Monitor &operator=(const Monitor &);
+};
+
+#endif // MONITOR_H
diff --git a/tools/gator/daemon/NetDriver.cpp b/tools/gator/daemon/NetDriver.cpp
new file mode 100644
index 000000000000..e75c0695779a
--- /dev/null
+++ b/tools/gator/daemon/NetDriver.cpp
@@ -0,0 +1,129 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+// Define to get format macros from inttypes.h
+#define __STDC_FORMAT_MACROS
+
+#include "NetDriver.h"
+
+#include <inttypes.h>
+
+#include "Logging.h"
+#include "SessionData.h"
+
+class NetCounter : public DriverCounter {
+public:
+	NetCounter(DriverCounter *next, char *const name, int64_t *const value);
+	~NetCounter();
+
+	int64_t read();
+
+private:
+	int64_t *const mValue;
+	int64_t mPrev;
+
+	// Intentionally unimplemented
+	NetCounter(const NetCounter &);
+	NetCounter &operator=(const NetCounter &);
+};
+
+NetCounter::NetCounter(DriverCounter *next, char *const name, int64_t *const value) : DriverCounter(next, name), mValue(value), mPrev(0) {
+}
+
+NetCounter::~NetCounter() {
+}
+
+int64_t NetCounter::read() {
+	int64_t result = *mValue - mPrev;
+	mPrev = *mValue;
+	return result;
+}
+
+NetDriver::NetDriver() : mBuf(), mReceiveBytes(0), mTransmitBytes(0) {
+}
+
+NetDriver::~NetDriver() {
+}
+
+void NetDriver::readEvents(mxml_node_t *const) {
+	// Only for use with perf
+	if (!gSessionData->perf.isSetup()) {
+		return;
+	}
+
+	setCounters(new NetCounter(getCounters(), strdup("Linux_net_rx"), &mReceiveBytes));
+	setCounters(new NetCounter(getCounters(), strdup("Linux_net_tx"), &mTransmitBytes));
+}
+
+bool NetDriver::doRead() {
+	if (!countersEnabled()) {
+		return true;
+	}
+
+	if (!mBuf.read("/proc/net/dev")) {
+		return false;
+	}
+
+	// Skip the header
+	char *key;
+	if (((key = strchr(mBuf.getBuf(), '\n')) == NULL) ||
+			((key = strchr(key + 1, '\n')) == NULL)) {
+		return false;
+	}
+	key = key + 1;
+
+	mReceiveBytes = 0;
+	mTransmitBytes = 0;
+
+	char *colon;
+	while ((colon = strchr(key, ':')) != NULL) {
+		char *end = strchr(colon + 1, '\n');
+		if (end != NULL) {
+			*end = '\0';
+		}
+		*colon = '\0';
+
+		int64_t receiveBytes;
+		int64_t transmitBytes;
+		const int count = sscanf(colon + 1, " %" SCNu64 " %*u %*u %*u %*u %*u %*u %*u %" SCNu64, &receiveBytes, &transmitBytes);
+		if (count != 2) {
+			return false;
+		}
+		mReceiveBytes += receiveBytes;
+		mTransmitBytes += transmitBytes;
+
+		if (end == NULL) {
+			break;
+		}
+		key = end + 1;
+	}
+
+	return true;
+}
+
+void NetDriver::start() {
+	if (!doRead()) {
+		logg->logError(__FILE__, __LINE__, "Unable to read network stats");
+		handleException();
+	}
+	// Initialize previous values
+	for (DriverCounter *counter = getCounters(); counter != NULL; counter = counter->getNext()) {
+		if (!counter->isEnabled()) {
+			continue;
+		}
+		counter->read();
+	}
+}
+
+void NetDriver::read(Buffer *const buffer) {
+	if (!doRead()) {
+		logg->logError(__FILE__, __LINE__, "Unable to read network stats");
+		handleException();
+	}
+	super::read(buffer);
+}
diff --git a/tools/gator/daemon/NetDriver.h b/tools/gator/daemon/NetDriver.h
new file mode 100644
index 000000000000..50ff850bfc6d
--- /dev/null
+++ b/tools/gator/daemon/NetDriver.h
@@ -0,0 +1,39 @@
+/**
+ * Copyright (C) ARM Limited 2013-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef NETDRIVER_H
+#define NETDRIVER_H
+
+#include "Driver.h"
+#include "DynBuf.h"
+
+class NetDriver : public PolledDriver {
+private:
+	typedef PolledDriver super;
+
+public:
+	NetDriver();
+	~NetDriver();
+
+	void readEvents(mxml_node_t *const root);
+	void start();
+	void read(Buffer *const buffer);
+
+private:
+	bool doRead();
+
+	DynBuf mBuf;
+	int64_t mReceiveBytes;
+	int64_t mTransmitBytes;
+
+	// Intentionally unimplemented
+	NetDriver(const NetDriver &);
+	NetDriver &operator=(const NetDriver &);
+};
+
+#endif // NETDRIVER_H
diff --git a/tools/gator/daemon/OlySocket.cpp b/tools/gator/daemon/OlySocket.cpp
new file mode 100644
index 000000000000..aa0ce4929916
--- /dev/null
+++ b/tools/gator/daemon/OlySocket.cpp
@@ -0,0 +1,324 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "OlySocket.h"
+
+#include <stdio.h>
+#include <string.h>
+#ifdef WIN32
+#include <Winsock2.h>
+#include <ws2tcpip.h>
+#else
+#include <netinet/in.h>
+#include <sys/un.h>
+#include <unistd.h>
+#include <netdb.h>
+#include <fcntl.h>
+#endif
+
+#include "Logging.h"
+
+#ifdef WIN32
+#define CLOSE_SOCKET(x) closesocket(x)
+#define SHUTDOWN_RX_TX SD_BOTH
+#define snprintf       _snprintf
+#else
+#define CLOSE_SOCKET(x) close(x)
+#define SHUTDOWN_RX_TX SHUT_RDWR
+#endif
+
+int socket_cloexec(int domain, int type, int protocol) {
+#ifdef SOCK_CLOEXEC
+  return socket(domain, type | SOCK_CLOEXEC, protocol);
+#else
+  int sock = socket(domain, type, protocol);
+#ifdef FD_CLOEXEC
+  if (sock < 0) {
+    return -1;
+  }
+  int fdf = fcntl(sock, F_GETFD);
+  if ((fdf == -1) || (fcntl(sock, F_SETFD, fdf | FD_CLOEXEC) != 0)) {
+    close(sock);
+    return -1;
+  }
+#endif
+  return sock;
+#endif
+}
+
+int accept_cloexec(int sockfd, struct sockaddr *addr, socklen_t *addrlen) {
+  int sock;
+#ifdef SOCK_CLOEXEC
+  sock = accept4(sockfd, addr, addrlen, SOCK_CLOEXEC);
+  if (sock >= 0) {
+    return sock;
+  }
+  // accept4 with SOCK_CLOEXEC may not work on all kernels, so fallback
+#endif
+  sock = accept(sockfd, addr, addrlen);
+#ifdef FD_CLOEXEC
+  if (sock < 0) {
+    return -1;
+  }
+  int fdf = fcntl(sock, F_GETFD);
+  if ((fdf == -1) || (fcntl(sock, F_SETFD, fdf | FD_CLOEXEC) != 0)) {
+    close(sock);
+    return -1;
+  }
+#endif
+  return sock;
+}
+
+OlyServerSocket::OlyServerSocket(int port) {
+#ifdef WIN32
+  WSADATA wsaData;
+  if (WSAStartup(0x0202, &wsaData) != 0) {
+    logg->logError(__FILE__, __LINE__, "Windows socket initialization failed");
+    handleException();
+  }
+#endif
+
+  createServerSocket(port);
+}
+
+OlySocket::OlySocket(int socketID) : mSocketID(socketID) {
+}
+
+#ifndef WIN32
+
+#define MIN(A, B) ({ \
+  const __typeof__(A) __a = A; \
+  const __typeof__(B) __b = B; \
+  __a > __b ? __b : __a; \
+})
+
+OlyServerSocket::OlyServerSocket(const char* path, const size_t pathSize) {
+  // Create socket
+  mFDServer = socket_cloexec(PF_UNIX, SOCK_STREAM, 0);
+  if (mFDServer < 0) {
+    logg->logError(__FILE__, __LINE__, "Error creating server socket");
+    handleException();
+  }
+
+  // Create sockaddr_in structure, ensuring non-populated fields are zero
+  struct sockaddr_un sockaddr;
+  memset((void*)&sockaddr, 0, sizeof(sockaddr));
+  sockaddr.sun_family = AF_UNIX;
+  memcpy(sockaddr.sun_path, path, MIN(pathSize, sizeof(sockaddr.sun_path)));
+  sockaddr.sun_path[sizeof(sockaddr.sun_path) - 1] = '\0';
+
+  // Bind the socket to an address
+  if (bind(mFDServer, (const struct sockaddr*)&sockaddr, sizeof(sockaddr)) < 0) {
+    logg->logError(__FILE__, __LINE__, "Binding of server socket failed.");
+    handleException();
+  }
+
+  // Listen for connections on this socket
+  if (listen(mFDServer, 1) < 0) {
+    logg->logError(__FILE__, __LINE__, "Listening of server socket failed");
+    handleException();
+  }
+}
+
+int OlySocket::connect(const char* path, const size_t pathSize) {
+  int fd = socket_cloexec(PF_UNIX, SOCK_STREAM, 0);
+  if (fd < 0) {
+    return -1;
+  }
+
+  // Create sockaddr_in structure, ensuring non-populated fields are zero
+  struct sockaddr_un sockaddr;
+  memset((void*)&sockaddr, 0, sizeof(sockaddr));
+  sockaddr.sun_family = AF_UNIX;
+  memcpy(sockaddr.sun_path, path, MIN(pathSize, sizeof(sockaddr.sun_path)));
+  sockaddr.sun_path[sizeof(sockaddr.sun_path) - 1] = '\0';
+
+  if (::connect(fd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr)) < 0) {
+    close(fd);
+    return -1;
+  }
+
+  return fd;
+}
+
+#endif
+
+OlySocket::~OlySocket() {
+  if (mSocketID > 0) {
+    CLOSE_SOCKET(mSocketID);
+  }
+}
+
+OlyServerSocket::~OlyServerSocket() {
+  if (mFDServer > 0) {
+    CLOSE_SOCKET(mFDServer);
+  }
+}
+
+void OlySocket::shutdownConnection() {
+  // Shutdown is primarily used to unblock other threads that are blocking on send/receive functions
+  shutdown(mSocketID, SHUTDOWN_RX_TX);
+}
+
+void OlySocket::closeSocket() {
+  // Used for closing an accepted socket but keeping the server socket active
+  if (mSocketID > 0) {
+    CLOSE_SOCKET(mSocketID);
+    mSocketID = -1;
+  }
+}
+
+void OlyServerSocket::closeServerSocket() {
+  if (CLOSE_SOCKET(mFDServer) != 0) {
+    logg->logError(__FILE__, __LINE__, "Failed to close server socket.");
+    handleException();
+  }
+  mFDServer = 0;
+}
+
+void OlyServerSocket::createServerSocket(int port) {
+  int family = AF_INET6;
+
+  // Create socket
+  mFDServer = socket_cloexec(PF_INET6, SOCK_STREAM, IPPROTO_TCP);
+  if (mFDServer < 0) {
+    family = AF_INET;
+    mFDServer = socket_cloexec(PF_INET, SOCK_STREAM, IPPROTO_TCP);
+    if (mFDServer < 0) {
+      logg->logError(__FILE__, __LINE__, "Error creating server socket");
+      handleException();
+    }
+  }
+
+  // Enable address reuse, another solution would be to create the server socket once and only close it when the object exits
+  int on = 1;
+  if (setsockopt(mFDServer, SOL_SOCKET, SO_REUSEADDR, (const char*)&on, sizeof(on)) != 0) {
+    logg->logError(__FILE__, __LINE__, "Setting server socket options failed");
+    handleException();
+  }
+
+  // Create sockaddr_in structure, ensuring non-populated fields are zero
+  struct sockaddr_in6 sockaddr;
+  memset((void*)&sockaddr, 0, sizeof(sockaddr));
+  sockaddr.sin6_family = family;
+  sockaddr.sin6_port = htons(port);
+  sockaddr.sin6_addr = in6addr_any;
+
+  // Bind the socket to an address
+  if (bind(mFDServer, (const struct sockaddr*)&sockaddr, sizeof(sockaddr)) < 0) {
+    logg->logError(__FILE__, __LINE__, "Binding of server socket failed.\nIs an instance already running?");
+    handleException();
+  }
+
+  // Listen for connections on this socket
+  if (listen(mFDServer, 1) < 0) {
+    logg->logError(__FILE__, __LINE__, "Listening of server socket failed");
+    handleException();
+  }
+}
+
+// mSocketID is always set to the most recently accepted connection
+// The user of this class should maintain the different socket connections, e.g. by forking the process
+int OlyServerSocket::acceptConnection() {
+  int socketID;
+  if (mFDServer <= 0) {
+    logg->logError(__FILE__, __LINE__, "Attempting multiple connections on a single connection server socket or attempting to accept on a client socket");
+    handleException();
+  }
+
+  // Accept a connection, note that this call blocks until a client connects
+  socketID = accept_cloexec(mFDServer, NULL, NULL);
+  if (socketID < 0) {
+    logg->logError(__FILE__, __LINE__, "Socket acceptance failed");
+    handleException();
+  }
+  return socketID;
+}
+
+void OlySocket::send(const char* buffer, int size) {
+  if (size <= 0 || buffer == NULL) {
+    return;
+  }
+
+  while (size > 0) {
+    int n = ::send(mSocketID, buffer, size, 0);
+    if (n < 0) {
+      logg->logError(__FILE__, __LINE__, "Socket send error");
+      handleException();
+    }
+    size -= n;
+    buffer += n;
+  }
+}
+
+// Returns the number of bytes received
+int OlySocket::receive(char* buffer, int size) {
+  if (size <= 0 || buffer == NULL) {
+    return 0;
+  }
+
+  int bytes = recv(mSocketID, buffer, size, 0);
+  if (bytes < 0) {
+    logg->logError(__FILE__, __LINE__, "Socket receive error");
+    handleException();
+  } else if (bytes == 0) {
+    logg->logMessage("Socket disconnected");
+    return -1;
+  }
+  return bytes;
+}
+
+// Receive exactly size bytes of data. Note, this function will block until all bytes are received
+int OlySocket::receiveNBytes(char* buffer, int size) {
+  int bytes = 0;
+  while (size > 0 && buffer != NULL) {
+    bytes = recv(mSocketID, buffer, size, 0);
+    if (bytes < 0) {
+      logg->logError(__FILE__, __LINE__, "Socket receive error");
+      handleException();
+    } else if (bytes == 0) {
+      logg->logMessage("Socket disconnected");
+      return -1;
+    }
+    buffer += bytes;
+    size -= bytes;
+  }
+  return bytes;
+}
+
+// Receive data until a carriage return, line feed, or null is encountered, or the buffer fills
+int OlySocket::receiveString(char* buffer, int size) {
+  int bytes_received = 0;
+  bool found = false;
+
+  if (buffer == 0) {
+    return 0;
+  }
+
+  while (!found && bytes_received < size) {
+    // Receive a single character
+    int bytes = recv(mSocketID, &buffer[bytes_received], 1, 0);
+    if (bytes < 0) {
+      logg->logError(__FILE__, __LINE__, "Socket receive error");
+      handleException();
+    } else if (bytes == 0) {
+      logg->logMessage("Socket disconnected");
+      return -1;
+    }
+
+    // Replace carriage returns and line feeds with zero
+    if (buffer[bytes_received] == '\n' || buffer[bytes_received] == '\r' || buffer[bytes_received] == '\0') {
+      buffer[bytes_received] = '\0';
+      found = true;
+    }
+
+    bytes_received++;
+  }
+
+  return bytes_received;
+}
diff --git a/tools/gator/daemon/OlySocket.h b/tools/gator/daemon/OlySocket.h
new file mode 100644
index 000000000000..6b53b01fc3ee
--- /dev/null
+++ b/tools/gator/daemon/OlySocket.h
@@ -0,0 +1,64 @@
+/**
+ * Copyright (C) ARM Limited 2010-2014. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef __OLY_SOCKET_H__
+#define __OLY_SOCKET_H__
+
+#include <stddef.h>
+
+#ifdef WIN32
+typedef socklen_t int;
+#else
+#include <sys/socket.h>
+#endif
+
+class OlySocket {
+public:
+#ifndef WIN32
+  static int connect(const char* path, const size_t pathSize);
+#endif
+
+  OlySocket(int socketID);
+  ~OlySocket();
+
+  void closeSocket();
+  void shutdownConnection();
+  void send(const char* buffer, int size);
+  int receive(char* buffer, int size);
+  int receiveNBytes(char* buffer, int size);
+  int receiveString(char* buffer, int size);
+
+  bool isValid() const { return mSocketID >= 0; }