Reland (2) of Enable allocator shim for Android (crrev.com/1875043003)

Reland (2) of Enable allocator shim for Android (crrev.com/1875043003)

Reason for reland:
The CL was re-reverted by crrev.com/1884223002 because it broke
the internal orderfile bot.
The reason of the breakage was the following:
 - cygprofile.cc instruments all function calls with a
   __cyg_profile_func_enter preamble.
 - __cyg_profile_func_enter uses __thread. __thread under the hoods
   invokes calloc(), on every thread, to initialize the TLS.
 - The shim layer provides its own implementation of calloc().
 - At this point, calloc() gets instrumented as well and it re-enters
   __cyg_profile_func_enter causing an infinite loop.
The key of the problem here is that __thread silently causes calls to
calloc() in a way that is out of control of cygprofile.cc.

The solution proposed by this CL is the following:
 - Don't use __thread, use explicit POSIX functions for TLS (also there
   doesn't seem to be any precendent of using __thread in the codebase).
 - Use a global variable to prevent re-entrancy of the
   __cyg_profile_func_enter in the global (once per process) TLS slot
   initializer.
 - Re-entrancy is gone.

Original issue's description:
> Enable allocator shim for Android
>
> This is a follow-up to crrev.com/1719433002, which introduced the
> shim for Android, and enables it by default by setting
> use_experimental_allocator_shim=true for Android.
>
> Build/Perf sheriffs heads up
> ----------------------------
> If you see any build error or crash related with __wrap_malloc,
> __wrap_free, __real_malloc, __real_free, etc this CL is to blame.
>
> Performance considerations
> ------------------------
> Binary size diff (GN, arm, static, official build): 24k
>
> I did a mixture of local and trybots run to estimate the perf impact
> of this change. Didn't get any conclusive data, everything I tried
> seems in the same ballpark, below noise levels. More in details:
>
> cc_perftests.PrepareTiles on a Nexus 4.
> Rationale of the choice: in a previous CL (crbug.com/593344), this
> benchmark revealed the presence of two mfences in the malloc path.
> Results: https://goo.gl/8VC3Jp in the same ballpark.
>
> page-cycler on Nexus 9 via trybots:
> Results: http://goo.gl/J3i50a seems to suggest that this CL improves
> both warm and cold times in most cases. I doubt it, more likely it's
> noise.
>
> All the other perf trybots failed. The perf waterfall seems to be in a
> bad state in these days.
>
> BUG=550886, 598075
> TEST=base_unittests --gtest_filter=AllocatorShimTest.*
> TBR=thakis@chromium.org
>
> Committed: https://crrev.com/ebb95496c73dc0d5ce83968ac619921f154305f7
> Cr-Commit-Position: refs/heads/master@{#386386}

BUG=550886, 598075, 602744
TBR=thakis@chromium.org
TEST=gn gen out/Debug --args='is_debug=true target_os="android" use_order_profiling=true target_cpu="arm" is_clang=false';
     ninja -C out/Debug/ cygprofile_unittests;
     adb push out/Debug/cygprofile_unittests /data/local/tmp/cygprofile_unittests_debug;
     adb shell /data/local/tmp/cygprofile_unittests_debug

Committed: https://crrev.com/f7a321facfdabd763ecdbc9536c890fe91c8c079
Cr-Commit-Position: refs/heads/master@{#387594}

[Primiano Tucci (use gerrit), 2016-04-15 09:25:25]

primiano@chromium.org changed reviewers:
+ lizeb@chromium.org, pasko@chromium.org

[Primiano Tucci (use gerrit), 2016-04-15 09:25:26]

PTAL

[Primiano Tucci (use gerrit), 2016-04-15 09:27:40]

Description was changed from

==========
Reland (2) of Enable allocator shim for Android (crrev.com/1875043003)

Reason for reland:
The CL was re-reverted by crrev.com/1884223002 because it broke
the internal orderfile bot.
The reason of the breakage was the following:
 - cygprofile.cc instruments all function calls with a
   __cyg_profile_func_enter preamble.
 - __cyg_profile_func_enter uses __thread. __thread under the hoods
   invokes calloc(), on every thread, to initialize the TLS.
 - The shim layer provides its own implementation of calloc().
 - At this point, calloc() gets instrumented as well and it re-enters
   __cyg_profile_func_enter causing an infinite loop.
The key of the problem here is that __thread silently causes calls to
calloc() in a way that is out of control of cygprofile.cc.

The solution proposed by this CL is the following:
 - Don't use __thread, use explicit POSIX functions for TLS (also there
   doesn't seem to be any precendent of using __thread in the codebase).
 - Use a global variable to prevent re-entrancy of the
   __cyg_profile_func_enter in the global (once per process) TLS slot
   initializer.
 - Re-entrancy is gone.

Original issue's description:
> Enable allocator shim for Android
>
> This is a follow-up to crrev.com/1719433002, which introduced the
> shim for Android, and enables it by default by setting
> use_experimental_allocator_shim=true for Android.
>
> Build/Perf sheriffs heads up
> ----------------------------
> If you see any build error or crash related with __wrap_malloc,
> __wrap_free, __real_malloc, __real_free, etc this CL is to blame.
>
> Performance considerations
> ------------------------
> Binary size diff (GN, arm, static, official build): 24k
>
> I did a mixture of local and trybots run to estimate the perf impact
> of this change. Didn't get any conclusive data, everything I tried
> seems in the same ballpark, below noise levels. More in details:
>
> cc_perftests.PrepareTiles on a Nexus 4.
> Rationale of the choice: in a previous CL (crbug.com/593344), this
> benchmark revealed the presence of two mfences in the malloc path.
> Results: https://goo.gl/8VC3Jp in the same ballpark.
>
> page-cycler on Nexus 9 via trybots:
> Results: http://goo.gl/J3i50a seems to suggest that this CL improves
> both warm and cold times in most cases. I doubt it, more likely it's
> noise.
>
> All the other perf trybots failed. The perf waterfall seems to be in a
> bad state in these days.
>
> BUG=550886,598075
> TEST=base_unittests --gtest_filter=AllocatorShimTest.*
> TBR=thakis@chromium.org
>
> Committed: https://crrev.com/ebb95496c73dc0d5ce83968ac619921f154305f7
> Cr-Commit-Position: refs/heads/master@{#386386}

BUG=550886,598075, 602744
==========

to

==========
Reland (2) of Enable allocator shim for Android (crrev.com/1875043003)

Reason for reland:
The CL was re-reverted by crrev.com/1884223002 because it broke
the internal orderfile bot.
The reason of the breakage was the following:
 - cygprofile.cc instruments all function calls with a
   __cyg_profile_func_enter preamble.
 - __cyg_profile_func_enter uses __thread. __thread under the hoods
   invokes calloc(), on every thread, to initialize the TLS.
 - The shim layer provides its own implementation of calloc().
 - At this point, calloc() gets instrumented as well and it re-enters
   __cyg_profile_func_enter causing an infinite loop.
The key of the problem here is that __thread silently causes calls to
calloc() in a way that is out of control of cygprofile.cc.

The solution proposed by this CL is the following:
 - Don't use __thread, use explicit POSIX functions for TLS (also there
   doesn't seem to be any precendent of using __thread in the codebase).
 - Use a global variable to prevent re-entrancy of the
   __cyg_profile_func_enter in the global (once per process) TLS slot
   initializer.
 - Re-entrancy is gone.

Original issue's description:
> Enable allocator shim for Android
>
> This is a follow-up to crrev.com/1719433002, which introduced the
> shim for Android, and enables it by default by setting
> use_experimental_allocator_shim=true for Android.
>
> Build/Perf sheriffs heads up
> ----------------------------
> If you see any build error or crash related with __wrap_malloc,
> __wrap_free, __real_malloc, __real_free, etc this CL is to blame.
>
> Performance considerations
> ------------------------
> Binary size diff (GN, arm, static, official build): 24k
>
> I did a mixture of local and trybots run to estimate the perf impact
> of this change. Didn't get any conclusive data, everything I tried
> seems in the same ballpark, below noise levels. More in details:
>
> cc_perftests.PrepareTiles on a Nexus 4.
> Rationale of the choice: in a previous CL (crbug.com/593344), this
> benchmark revealed the presence of two mfences in the malloc path.
> Results: https://goo.gl/8VC3Jp in the same ballpark.
>
> page-cycler on Nexus 9 via trybots:
> Results: http://goo.gl/J3i50a seems to suggest that this CL improves
> both warm and cold times in most cases. I doubt it, more likely it's
> noise.
>
> All the other perf trybots failed. The perf waterfall seems to be in a
> bad state in these days.
>
> BUG=550886,598075
> TEST=base_unittests --gtest_filter=AllocatorShimTest.*
> TBR=thakis@chromium.org
>
> Committed: https://crrev.com/ebb95496c73dc0d5ce83968ac619921f154305f7
> Cr-Commit-Position: refs/heads/master@{#386386}

BUG=550886,598075, 602744
TBR=thakis@chromium.org
TEST=gn gen out/Debug --args='is_debug=true target_os="android"
use_order_profiling=true target_cpu="arm" is_clang=false';
     ninja -C out/Debug/ cygprofile_unittests;
     adb push out/Debug/cygprofile_unittests
/data/local/tmp/cygprofile_unittests_debug;
     adb shell /data/local/tmp/cygprofile_unittests_debug
==========

[Primiano Tucci (use gerrit), 2016-04-15 09:30:21]

The CQ bit was checked by primiano@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2016-04-15 09:30:28]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1883093005/20001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1883093005/20001

[commit-bot: I haz the power, 2016-04-15 10:26:54]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2016-04-15 10:26:54]

Dry run: This issue passed the CQ dry run.

[pasko, 2016-04-15 11:42:14]

thank you, Primiano

overall looking good, 2 comments below

https://codereview.chromium.org/1883093005/diff/20001/tools/cygprofile/cygpro...
File tools/cygprofile/cygprofile.cc (right):

https://codereview.chromium.org/1883093005/diff/20001/tools/cygprofile/cygpro...
tools/cygprofile/cygprofile.cc:62: // This variable is NOT accessed from
multiple threads (pthread_once guarantees
as per offline discussion, this comment needs to be updated to explain the once
write multiple reads.

https://codereview.chromium.org/1883093005/diff/20001/tools/cygprofile/cygpro...
tools/cygprofile/cygprofile.cc:70: (void) pthread_key_create(&g_tls_slot, NULL);
Everywhere in this file please check return value for:
pthread_key_create()
pthread_setspecific()

.. and do perror(3) on failure. Not many reasons of failure exist, but I think
it is worth the effort to help debugging in future.

[Primiano Tucci (use gerrit), 2016-04-15 12:50:34]

Patchset #3 (id:40001) has been deleted

[Primiano Tucci (use gerrit), 2016-04-15 12:51:07]

Done thanks

https://codereview.chromium.org/1883093005/diff/20001/tools/cygprofile/cygpro...
File tools/cygprofile/cygprofile.cc (right):

https://codereview.chromium.org/1883093005/diff/20001/tools/cygprofile/cygpro...
tools/cygprofile/cygprofile.cc:62: // This variable is NOT accessed from
multiple threads (pthread_once guarantees
On 2016/04/15 11:42:14, pasko wrote:
> as per offline discussion, this comment needs to be updated to explain the
once
> write multiple reads.

Done.

https://codereview.chromium.org/1883093005/diff/20001/tools/cygprofile/cygpro...
tools/cygprofile/cygprofile.cc:70: (void) pthread_key_create(&g_tls_slot, NULL);
On 2016/04/15 11:42:14, pasko wrote:
> Everywhere in this file please check return value for:
> pthread_key_create()
> pthread_setspecific()
> 
> .. and do perror(3) on failure. Not many reasons of failure exist, but I think
> it is worth the effort to help debugging in future.

Went for CHECK. perror doesn't really help on android due to the lack of stdio
(unless running an executable)

[Primiano Tucci (use gerrit), 2016-04-15 12:53:52]

The CQ bit was checked by primiano@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2016-04-15 12:54:03]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1883093005/60001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1883093005/60001

[pasko, 2016-04-15 13:00:47]

https://codereview.chromium.org/1883093005/diff/60001/tools/cygprofile/cygpro...
File tools/cygprofile/cygprofile.cc (right):

https://codereview.chromium.org/1883093005/diff/60001/tools/cygprofile/cygpro...
tools/cygprofile/cygprofile.cc:61: // to avoid compiler otpimizations, as this
need to be read re-entratively.
nit: s/read re-entratively/read in a re-entrant way/

0.1% easier to read for, at least for me

https://codereview.chromium.org/1883093005/diff/60001/tools/cygprofile/cygpro...
tools/cygprofile/cygprofile.cc:72: CHECK_EQ(0, pthread_key_create(&g_tls_slot,
NULL));
let's do a PCHECK(pthread_key_create(&g_tls_slot, NULL))

[pasko, 2016-04-15 13:03:59]

https://codereview.chromium.org/1883093005/diff/60001/tools/cygprofile/cygpro...
File tools/cygprofile/cygprofile.cc (right):

https://codereview.chromium.org/1883093005/diff/60001/tools/cygprofile/cygpro...
tools/cygprofile/cygprofile.cc:72: CHECK_EQ(0, pthread_key_create(&g_tls_slot,
NULL));
On 2016/04/15 13:00:47, pasko wrote:
> let's do a PCHECK(pthread_key_create(&g_tls_slot, NULL))

plz disregard this (it returns zero on success, and I thought it's vice versa)

[Primiano Tucci (use gerrit), 2016-04-15 13:06:09]

https://codereview.chromium.org/1883093005/diff/60001/tools/cygprofile/cygpro...
File tools/cygprofile/cygprofile.cc (right):

https://codereview.chromium.org/1883093005/diff/60001/tools/cygprofile/cygpro...
tools/cygprofile/cygprofile.cc:61: // to avoid compiler otpimizations, as this
need to be read re-entratively.
On 2016/04/15 13:00:47, pasko wrote:
> nit: s/read re-entratively/read in a re-entrant way/
> 
> 0.1% easier to read for, at least for me

Done.

https://codereview.chromium.org/1883093005/diff/60001/tools/cygprofile/cygpro...
tools/cygprofile/cygprofile.cc:72: CHECK_EQ(0, pthread_key_create(&g_tls_slot,
NULL));
On 2016/04/15 13:00:47, pasko wrote:
> let's do a PCHECK(pthread_key_create(&g_tls_slot, NULL))

Done.

https://codereview.chromium.org/1883093005/diff/60001/tools/cygprofile/cygpro...
tools/cygprofile/cygprofile.cc:72: CHECK_EQ(0, pthread_key_create(&g_tls_slot,
NULL));
On 2016/04/15 13:03:59, pasko wrote:
> On 2016/04/15 13:00:47, pasko wrote:
> > let's do a PCHECK(pthread_key_create(&g_tls_slot, NULL))
> 
> plz disregard this (it returns zero on success, and I thought it's vice versa)

Done.

[pasko, 2016-04-15 13:09:05]

lgtm, thank you

[Primiano Tucci (use gerrit), 2016-04-15 13:12:59]

primiano@chromium.org changed reviewers:
+ thakis@chromium.org

[Primiano Tucci (use gerrit), 2016-04-15 13:12:59]

TBR thakis. Nothing has changed from the previous reland, I just fixed the
cygprofile code.

[Primiano Tucci (use gerrit), 2016-04-15 13:13:04]

The CQ bit was checked by primiano@chromium.org

[commit-bot: I haz the power, 2016-04-15 13:13:21]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1883093005/100001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1883093005/100001

[commit-bot: I haz the power, 2016-04-15 14:14:37]

Description was changed from

==========
Reland (2) of Enable allocator shim for Android (crrev.com/1875043003)

Reason for reland:
The CL was re-reverted by crrev.com/1884223002 because it broke
the internal orderfile bot.
The reason of the breakage was the following:
 - cygprofile.cc instruments all function calls with a
   __cyg_profile_func_enter preamble.
 - __cyg_profile_func_enter uses __thread. __thread under the hoods
   invokes calloc(), on every thread, to initialize the TLS.
 - The shim layer provides its own implementation of calloc().
 - At this point, calloc() gets instrumented as well and it re-enters
   __cyg_profile_func_enter causing an infinite loop.
The key of the problem here is that __thread silently causes calls to
calloc() in a way that is out of control of cygprofile.cc.

The solution proposed by this CL is the following:
 - Don't use __thread, use explicit POSIX functions for TLS (also there
   doesn't seem to be any precendent of using __thread in the codebase).
 - Use a global variable to prevent re-entrancy of the
   __cyg_profile_func_enter in the global (once per process) TLS slot
   initializer.
 - Re-entrancy is gone.

Original issue's description:
> Enable allocator shim for Android
>
> This is a follow-up to crrev.com/1719433002, which introduced the
> shim for Android, and enables it by default by setting
> use_experimental_allocator_shim=true for Android.
>
> Build/Perf sheriffs heads up
> ----------------------------
> If you see any build error or crash related with __wrap_malloc,
> __wrap_free, __real_malloc, __real_free, etc this CL is to blame.
>
> Performance considerations
> ------------------------
> Binary size diff (GN, arm, static, official build): 24k
>
> I did a mixture of local and trybots run to estimate the perf impact
> of this change. Didn't get any conclusive data, everything I tried
> seems in the same ballpark, below noise levels. More in details:
>
> cc_perftests.PrepareTiles on a Nexus 4.
> Rationale of the choice: in a previous CL (crbug.com/593344), this
> benchmark revealed the presence of two mfences in the malloc path.
> Results: https://goo.gl/8VC3Jp in the same ballpark.
>
> page-cycler on Nexus 9 via trybots:
> Results: http://goo.gl/J3i50a seems to suggest that this CL improves
> both warm and cold times in most cases. I doubt it, more likely it's
> noise.
>
> All the other perf trybots failed. The perf waterfall seems to be in a
> bad state in these days.
>
> BUG=550886,598075
> TEST=base_unittests --gtest_filter=AllocatorShimTest.*
> TBR=thakis@chromium.org
>
> Committed: https://crrev.com/ebb95496c73dc0d5ce83968ac619921f154305f7
> Cr-Commit-Position: refs/heads/master@{#386386}

BUG=550886,598075, 602744
TBR=thakis@chromium.org
TEST=gn gen out/Debug --args='is_debug=true target_os="android"
use_order_profiling=true target_cpu="arm" is_clang=false';
     ninja -C out/Debug/ cygprofile_unittests;
     adb push out/Debug/cygprofile_unittests
/data/local/tmp/cygprofile_unittests_debug;
     adb shell /data/local/tmp/cygprofile_unittests_debug
==========

to

==========
Reland (2) of Enable allocator shim for Android (crrev.com/1875043003)

Reason for reland:
The CL was re-reverted by crrev.com/1884223002 because it broke
the internal orderfile bot.
The reason of the breakage was the following:
 - cygprofile.cc instruments all function calls with a
   __cyg_profile_func_enter preamble.
 - __cyg_profile_func_enter uses __thread. __thread under the hoods
   invokes calloc(), on every thread, to initialize the TLS.
 - The shim layer provides its own implementation of calloc().
 - At this point, calloc() gets instrumented as well and it re-enters
   __cyg_profile_func_enter causing an infinite loop.
The key of the problem here is that __thread silently causes calls to
calloc() in a way that is out of control of cygprofile.cc.

The solution proposed by this CL is the following:
 - Don't use __thread, use explicit POSIX functions for TLS (also there
   doesn't seem to be any precendent of using __thread in the codebase).
 - Use a global variable to prevent re-entrancy of the
   __cyg_profile_func_enter in the global (once per process) TLS slot
   initializer.
 - Re-entrancy is gone.

Original issue's description:
> Enable allocator shim for Android
>
> This is a follow-up to crrev.com/1719433002, which introduced the
> shim for Android, and enables it by default by setting
> use_experimental_allocator_shim=true for Android.
>
> Build/Perf sheriffs heads up
> ----------------------------
> If you see any build error or crash related with __wrap_malloc,
> __wrap_free, __real_malloc, __real_free, etc this CL is to blame.
>
> Performance considerations
> ------------------------
> Binary size diff (GN, arm, static, official build): 24k
>
> I did a mixture of local and trybots run to estimate the perf impact
> of this change. Didn't get any conclusive data, everything I tried
> seems in the same ballpark, below noise levels. More in details:
>
> cc_perftests.PrepareTiles on a Nexus 4.
> Rationale of the choice: in a previous CL (crbug.com/593344), this
> benchmark revealed the presence of two mfences in the malloc path.
> Results: https://goo.gl/8VC3Jp in the same ballpark.
>
> page-cycler on Nexus 9 via trybots:
> Results: http://goo.gl/J3i50a seems to suggest that this CL improves
> both warm and cold times in most cases. I doubt it, more likely it's
> noise.
>
> All the other perf trybots failed. The perf waterfall seems to be in a
> bad state in these days.
>
> BUG=550886,598075
> TEST=base_unittests --gtest_filter=AllocatorShimTest.*
> TBR=thakis@chromium.org
>
> Committed: https://crrev.com/ebb95496c73dc0d5ce83968ac619921f154305f7
> Cr-Commit-Position: refs/heads/master@{#386386}

BUG=550886,598075, 602744
TBR=thakis@chromium.org
TEST=gn gen out/Debug --args='is_debug=true target_os="android"
use_order_profiling=true target_cpu="arm" is_clang=false';
     ninja -C out/Debug/ cygprofile_unittests;
     adb push out/Debug/cygprofile_unittests
/data/local/tmp/cygprofile_unittests_debug;
     adb shell /data/local/tmp/cygprofile_unittests_debug
==========

[commit-bot: I haz the power, 2016-04-15 14:14:39]

Committed patchset #5 (id:100001)

[commit-bot: I haz the power, 2016-04-15 14:16:12]

Description was changed from

==========
Reland (2) of Enable allocator shim for Android (crrev.com/1875043003)

Reason for reland:
The CL was re-reverted by crrev.com/1884223002 because it broke
the internal orderfile bot.
The reason of the breakage was the following:
 - cygprofile.cc instruments all function calls with a
   __cyg_profile_func_enter preamble.
 - __cyg_profile_func_enter uses __thread. __thread under the hoods
   invokes calloc(), on every thread, to initialize the TLS.
 - The shim layer provides its own implementation of calloc().
 - At this point, calloc() gets instrumented as well and it re-enters
   __cyg_profile_func_enter causing an infinite loop.
The key of the problem here is that __thread silently causes calls to
calloc() in a way that is out of control of cygprofile.cc.

The solution proposed by this CL is the following:
 - Don't use __thread, use explicit POSIX functions for TLS (also there
   doesn't seem to be any precendent of using __thread in the codebase).
 - Use a global variable to prevent re-entrancy of the
   __cyg_profile_func_enter in the global (once per process) TLS slot
   initializer.
 - Re-entrancy is gone.

Original issue's description:
> Enable allocator shim for Android
>
> This is a follow-up to crrev.com/1719433002, which introduced the
> shim for Android, and enables it by default by setting
> use_experimental_allocator_shim=true for Android.
>
> Build/Perf sheriffs heads up
> ----------------------------
> If you see any build error or crash related with __wrap_malloc,
> __wrap_free, __real_malloc, __real_free, etc this CL is to blame.
>
> Performance considerations
> ------------------------
> Binary size diff (GN, arm, static, official build): 24k
>
> I did a mixture of local and trybots run to estimate the perf impact
> of this change. Didn't get any conclusive data, everything I tried
> seems in the same ballpark, below noise levels. More in details:
>
> cc_perftests.PrepareTiles on a Nexus 4.
> Rationale of the choice: in a previous CL (crbug.com/593344), this
> benchmark revealed the presence of two mfences in the malloc path.
> Results: https://goo.gl/8VC3Jp in the same ballpark.
>
> page-cycler on Nexus 9 via trybots:
> Results: http://goo.gl/J3i50a seems to suggest that this CL improves
> both warm and cold times in most cases. I doubt it, more likely it's
> noise.
>
> All the other perf trybots failed. The perf waterfall seems to be in a
> bad state in these days.
>
> BUG=550886,598075
> TEST=base_unittests --gtest_filter=AllocatorShimTest.*
> TBR=thakis@chromium.org
>
> Committed: https://crrev.com/ebb95496c73dc0d5ce83968ac619921f154305f7
> Cr-Commit-Position: refs/heads/master@{#386386}

BUG=550886,598075, 602744
TBR=thakis@chromium.org
TEST=gn gen out/Debug --args='is_debug=true target_os="android"
use_order_profiling=true target_cpu="arm" is_clang=false';
     ninja -C out/Debug/ cygprofile_unittests;
     adb push out/Debug/cygprofile_unittests
/data/local/tmp/cygprofile_unittests_debug;
     adb shell /data/local/tmp/cygprofile_unittests_debug
==========

to

==========
Reland (2) of Enable allocator shim for Android (crrev.com/1875043003)

Reason for reland:
The CL was re-reverted by crrev.com/1884223002 because it broke
the internal orderfile bot.
The reason of the breakage was the following:
 - cygprofile.cc instruments all function calls with a
   __cyg_profile_func_enter preamble.
 - __cyg_profile_func_enter uses __thread. __thread under the hoods
   invokes calloc(), on every thread, to initialize the TLS.
 - The shim layer provides its own implementation of calloc().
 - At this point, calloc() gets instrumented as well and it re-enters
   __cyg_profile_func_enter causing an infinite loop.
The key of the problem here is that __thread silently causes calls to
calloc() in a way that is out of control of cygprofile.cc.

The solution proposed by this CL is the following:
 - Don't use __thread, use explicit POSIX functions for TLS (also there
   doesn't seem to be any precendent of using __thread in the codebase).
 - Use a global variable to prevent re-entrancy of the
   __cyg_profile_func_enter in the global (once per process) TLS slot
   initializer.
 - Re-entrancy is gone.

Original issue's description:
> Enable allocator shim for Android
>
> This is a follow-up to crrev.com/1719433002, which introduced the
> shim for Android, and enables it by default by setting
> use_experimental_allocator_shim=true for Android.
>
> Build/Perf sheriffs heads up
> ----------------------------
> If you see any build error or crash related with __wrap_malloc,
> __wrap_free, __real_malloc, __real_free, etc this CL is to blame.
>
> Performance considerations
> ------------------------
> Binary size diff (GN, arm, static, official build): 24k
>
> I did a mixture of local and trybots run to estimate the perf impact
> of this change. Didn't get any conclusive data, everything I tried
> seems in the same ballpark, below noise levels. More in details:
>
> cc_perftests.PrepareTiles on a Nexus 4.
> Rationale of the choice: in a previous CL (crbug.com/593344), this
> benchmark revealed the presence of two mfences in the malloc path.
> Results: https://goo.gl/8VC3Jp in the same ballpark.
>
> page-cycler on Nexus 9 via trybots:
> Results: http://goo.gl/J3i50a seems to suggest that this CL improves
> both warm and cold times in most cases. I doubt it, more likely it's
> noise.
>
> All the other perf trybots failed. The perf waterfall seems to be in a
> bad state in these days.
>
> BUG=550886,598075
> TEST=base_unittests --gtest_filter=AllocatorShimTest.*
> TBR=thakis@chromium.org
>
> Committed: https://crrev.com/ebb95496c73dc0d5ce83968ac619921f154305f7
> Cr-Commit-Position: refs/heads/master@{#386386}

BUG=550886,598075, 602744
TBR=thakis@chromium.org
TEST=gn gen out/Debug --args='is_debug=true target_os="android"
use_order_profiling=true target_cpu="arm" is_clang=false';
     ninja -C out/Debug/ cygprofile_unittests;
     adb push out/Debug/cygprofile_unittests
/data/local/tmp/cygprofile_unittests_debug;
     adb shell /data/local/tmp/cygprofile_unittests_debug

Committed: https://crrev.com/f7a321facfdabd763ecdbc9536c890fe91c8c079
Cr-Commit-Position: refs/heads/master@{#387594}
==========

[commit-bot: I haz the power, 2016-04-15 14:16:13]

Patchset 5 (id:??) landed as
https://crrev.com/f7a321facfdabd763ecdbc9536c890fe91c8c079
Cr-Commit-Position: refs/heads/master@{#387594}