GitLab

Simpleperf stat/record command may need 1 second to set security.perf_harden
property. So the monitored child threads should run longer than 1 second.

Bug: 29763505
Change-Id: I9bf6dcba03f3fc5c91130f38a08ce0062e104103

BUG: 29388879
Change-Id: I6ede2273419ba986d08decc0540741c54ed736b1
Signed-off-by: Mohamad Ayyash <mkayyash@google.com>

* changes:
  Disable fragments and duplicates detection for squashfs
  mksquashfsimage.sh: Add support for passing block map files

They don't play nicely with generating block maps. Fragments reduce
performance. A later patch will add workaround for duplicate detection.

BUG: 22322817
Change-Id: I01c731202ef7a2450b5d1beca627a0158c9cc3b4
Signed-off-by: Mohamad Ayyash <mkayyash@google.com>

BUG: 22322817
Change-Id: I7e17f6ad3a503ee97f7a81a1a8382cff47a913c6
Signed-off-by: Mohamad Ayyash <mkayyash@google.com>

Bug: 29264324
Change-Id: Ia78ef924f049630968d222729cf3c1f378105b8a
(cherry picked from commit 5896132e)

The property security.perf_harden is added in https://android-review.googlesource.com/#/c/233736/5. And simpleperf needs to notice that.

Bug: 29054680
Change-Id: I5f1593f5b389d182a56c4bf3bd438a1dc2b66686
(cherry picked from commit ebf79f3a)

Bug: 29071416
Change-Id: Ie39cad3151907aabbf523895d6cde37b82ff29a2

Since it's not possible for libfec to identify whether the signature
field was successfully corrected, return both and allow the caller to
validate either signature.

Bug: 28943429
Change-Id: Ie913c21ba1d07d6df4c6feeb7226b2ec963f4d19

Bug: 19537384
Bug: 27199751
Bug: 28719525
Change-Id: Ie983ec12ac6c550fa76c89cd44343220688a99b4

am: 4b7f0b27

* commit '4b7f0b27':
  ANRdaemon: add README and a bash script to easily get trace.

Change-Id: I0ea20dd173cfaef70df4d07672810ca233fcb5f2

am: ecb5849a

* commit 'ecb5849a':
  ANRdaemon: add a few new trace options and poke all userland services

Change-Id: Id35a4a5d9ccab90c0ab661e835fcd55787efa09a

am: c33bbcd4

* commit 'c33bbcd4':
  ANRdaemon: increase the CPU threshold resolution to 0.01%

Change-Id: Ib6cc29b5ac94344c9c7a9766b6548563da060888

am: 9e90477a

* commit '9e90477a':
  ANRdaemon: add README and a bash script to easily get trace.

Change-Id: Ief3f17421cc970a6b8676921b414d093c43ea880

am: 5b42895d

* commit '5b42895d':
  ANRdaemon: add a few new trace options and poke all userland services

Change-Id: Ic1eac97f0b965cc8f1e2b73f26456975150acb48

am: 6d25c0bd

* commit '6d25c0bd':
  ANRdaemon: increase the CPU threshold resolution to 0.01%

Change-Id: I4b8feac7f8908943ae543f096e97896e50963697

Bug: 28989601
Change-Id: I5c9b819a6b05003e68d26bfacccf4c539ddab7ae

Bug: 28989601
Change-Id: Ic0fdb6e0601d5e0a285d42cd9fe1a3d8d6f86286

Bug: 28989601
Change-Id: If2a305bd14e8bd7c22115cbca668bd7777d0e7d2

The absence of a policy reference in the unencrypted directory now
causes e4crypt_set_directory_policy to fail with an error. Callers
should call e4crypt_is_native (now moved into here) before calling this.

Bug: 28318405
Change-Id: I209292aba3abad3b19105c9afe2b84e8b3dd6874

BUG: 27698960
Change-Id: Ia31b8319e0d1c8644ae1798116007a4b35c39e9b
Signed-off-by: Mohamad Ayyash <mkayyash@google.com>

This happens around the boundary of block groups where a slightly
different block boundary is set based on size of filesystem which means
allocations based on a previous filesystem layout need to verify they're
not crossing that new bounday

BUG: 27698960
Change-Id: I45d444b4477f14f71e8f17144523505a7525b4e2
Signed-off-by: Mohamad Ayyash <mkayyash@google.com>

1) Set EXT4_BG_INODE_UNINIT so fsck will skip scanning unused inodes
2) Use EXT4_FEATURE_RO_COMPAT_GDT_CSUM to enable group descriptor
checksums in filesystem which reduces overhead for fsck

BUG: 24747200
BUG: 28013112
Change-Id: I118ab26935442f452962b3a7c29c61acf76ad4d9
Signed-off-by: Mohamad Ayyash <mkayyash@google.com>

Device may not have libbacktrace_offline shared library, so always
using libbacktrace_offline static library.
Use simpleperf namespace to avoid conflicts with ThreadEntry in
libbacktrace.

Bug: 28152982
Change-Id: Ia2cdbee3b28556c790ccd7b061a801ed96f9b97c
(cherry picked from commit 040f7b4c)

It is now also used in mksquashfs so it was added to libcutils

BUG: 27467028
Change-Id: I81c87938d5b6b200a8ad817f0c64c5ca699f451c
Signed-off-by: Mohamad Ayyash <mkayyash@google.com>

Change-Id: Ic6994988152942e162e0b977afe48144b4d8bb32

Bug: 24272167
Change-Id: I831c9d26495e77e041a9461ddbafc70d7d11f9b7

Change-Id: Ie025a3d70237a7a959203d8cd066d05acf9e94b2

In order to keep all the benchmark binaries in the same
output place.

Change-Id: I188f7933c230d8955905217b915899dd19e614bf

Also add comment for how to generate each perf.data in testdata/.
Add --log <leve> argument in unit test to help debugging.

Bug: 26962895
Change-Id: Iaa5a81cd8da5174d5b5b4e7847811e2432bf0db8
(cherry picked from commit 8f680f60)

Broke this while adding app filtering.

Change-Id: Ie9d4ae3b7f99169a15c04b9e2bc11d1cb2a62a95

Bug: 26449138
Change-Id: Ie91d114bf669c686f618c252e58feff4c853ef91

When running unit tests on arm64 devices:
[OK] ReportCommandTest.dwarf_callgraph
[OK] record_cmd.dwarf_callchain_sampling.
ERROR: can't unwind data recorded on a different architecture.

It is because ReportCommandtest.dwarf_callgraph opens a perf.data
recorded on x86_64, and changes current_arch. It causes a problem when
the test record_cmd.dwarf_callchain_sampling calls libbacktrace built
on aarch64. Athough it doesn't make the test fail, we should fix this.

Bug: 27860749

Change-Id: I2cd70369a769ef2199cab2302b8b824369be0907
(cherry picked from commit 48460892)

simpleperf looks for _text symbol in /proc/kallsyms to get
kernel address space information. But that info is no longer
needed. As some kernels don't have _text symbol, we should
remove the corresponding code.

Bug: 27814103
Change-Id: I4eb9869b58c51846c99b1204f304cd107a0eb354
(cherry picked from commit 7134f388)

Bug: 26512380
Change-Id: Ia65986700f198a0f09db4a45a7d1bde173e557e6