Age | Commit message (Collapse) | Author |
|
|
|
|
|
Bug: 191603039
Test: N/A
Merged-In: I6320e9b20b1cceb37c0a27f311bbf6a45d3a57bb
Change-Id: If5b28b4d7c4a96cffe96ff52be0887f6545bca30
|
|
|
|
|
|
To prevent deadlocks.
Bug: 188393100
Test: N/A
Change-Id: Icf7a04c8f9d244b7bdd94b2111f41ce489d39302
|
|
|
|
Java Throwable getMessage can (and will!) return null, and when this
happens, we don't get any error information in the native world. So, for
instance, Java ServiceException won't be reported to native code.
This is technically a backwards incompatible change, and we'll have to
watch for breakages, but losing service specific exceptions in such a
clearly egregious way is not acceptable. :)
Fixes: 178861468
Test: boot & aidl_integration_test
Change-Id: If70d8fb445078c2989c4288c01055bb7191a012f
|
|
|
|
|
|
The #ifdef should be #if because RPC_FLAKE_PRONE is defined to false.
[[clang::no_destroy]] is needed for the static variables in
rpcMaybeWaitToFlake otherwise it can crash for taking a mutex that was
destroyed in the onLastStrongRef->sendDecStrong path.
Test: atest binderRpcTest
Fixes: 190828148
Change-Id: Iac6302d737de70f5fde2656dd2e140a6dcd13f7e
|
|
|
|
|
|
|
|
It is possible that development platforms don't have any input devices,
hence /dev/input directory won't be created. Hence check if /dev/input exists
before adding inotify watch and scanDirLocked for /dev/input.
inotify_add_watch is also added for /dev/input when it is created. for e.g.
during EventHubTest.
Signed-off-by: Usama Arif <usama.arif@arm.com>
Change-Id: I8675921280e351dd8885becce79a7cb4a9936c10
Merged-In: I8675921280e351dd8885becce79a7cb4a9936c10
|
|
Increases the meminfo dumpsys timeout from 30s to 60s, since
30s is sometimes not enough.
Test: Check that longer timeout is used in writeDump call
Bug: 191615258
Change-Id: Ic270a81f3dde1329881023d533ffbff43a82127c
|
|
Bug: b/192315789
Test: cts -m CtsNativeHardwareTestCases -t
SingleLayer_ColorTest_CpuWriteColorGpuRead_R16G16B16A16_FLOAT
Change-Id: Iaf3c07fa2856c8e7ddaa5f3d57eaf686fc136fa1
|
|
Use std::filesystem directory iterator to scan event devices in EventHub.
Bug: 160010896
Test: atest inputflinger_tests, atest EventHubTest
Change-Id: Ib5658e96783b8a1099fe0a2cb6dda51703045c84
Merged-In: Ib5658e96783b8a1099fe0a2cb6dda51703045c84
|
|
|
|
Revert "Allow otapreopt_chroot to call otadexopt binder service"
Revert submission 1750144-otapreopt_chroot_speed_up
Reason for revert: Breaks postinstall step on some devices
Reverted Changes:
I56dec13c8:Add ota_dexopt_aidl_interface
Ib32cbbbf8:Allow otapreopt_chroot to call otadexopt binder se...
I02e80a8e6:Move otadexopt-related logic to otapreopt_chroot b...
Bug: 190817237
Change-Id: If090e0cbf8d086b41c0b22a42b5daf4561954156
|
|
|
|
Bug: 192871419
Test: atest installd_dexopt_test
Change-Id: I7aeef7465f3617393ede43b24b142d7149942b3d
|
|
|
|
Bug: N/A
Change-Id: Ia6ca92c424029744ed378c6028bb726f3bf78fc9
|
|
|
|
Without this resulting .o is incorrectly formatted
and fails to load with newer bpfloader.
Bug: 190519702
Test: atest, TreeHugger - existing bpf programs load
examination of bpfloader logs
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ie1a3ed0cdaebc533e4533ed5bfefd369ec2a87b3
|
|
|
|
|
|
If javaAttachThread / javaDetachThread exists, run them before
and after the while loop in join. This ensures the
Java thread is attached / detached appropriately for
binder threads that handles RPC calls.
Test: run aservice on Java services
Bug: 190450693
Change-Id: I23f171dd1e08dfacfbbb46ecc9564cbf42d6f353
|
|
Bug: 192049377
Test: manual
Change-Id: I10094499b6141e134cb873d6c6590a1731e5910b
|
|
This auto-generates an enum_values function for all
Rust AIDL enums that works very similarly to enum_range()
in C++.
Bug: 179438017
Test: m
Change-Id: Ice74682dc9e1119aca2a932c3d5644665963dad8
|
|
|
|
|
|
|
|
The logic is a bit complicated.
Fixes: 192321823
Test: fails every time w/o fix, works with fix
Change-Id: I6b4bb126a9fe1fa6d2b39a05805b3a14f53f1a7b
|
|
In a recent CL, I incorrectly stated in a comment (which is now
corrected here) that ABpBinder::onLastStrongRef can assume the
underlying binder also has no reference count. This is simply wrong (and
such a comment should have came with an assertion).
When this was not true (you receive a binder at the same time that
another thread is dropping the last NDK-layer reference to that binder),
the detachObject call was deleting the ABpBinderTag::Value beneath the
code in lookupOrCreateFromBinder that is looking up this binder.
Bug: 192321823
Test: CtsNdkBinderTestCases (more to come)
Change-Id: I71b1081b6d6ca179d17af6bb1e4a1bb3dd4b109b
|
|
|
|
|
|
This allows us to only mount APEXes once during postinstall, instead of
mounting them for each APK being optimized.
Ultimately this CL just moves the while loop from otapreopt_script.sh
into otapreopt_chroot binary, and makes otapreopt_chroot communicate
with otadexopt service via binder.
Without this patch a no-op OTA takes 669.617 seconds
With this patch: 361.991 seconds
Bug: 190817237
Test: manual OTA
Change-Id: I02e80a8e68b6d274a007599371e43cd15330d351
|
|
The default impl is enough but previously move ctor was hidden because
of copy ctor.
Without move constructor, ParcelableHolder makes AIDL-generated C++ code
invalid (compilation error) when used with ParcelFileDescriptor (which
is moveable but not copyable intentionally).
Bug: 192197147
Test: m test-piece-6-cpp test-piece-6-ndk
Change-Id: I7154b4960595ee046167fffdd9da2f485b97da0a
|
|
This is an integration test for servicedispatcher. It tests
servicedispatcher and adb forwarding that the host service
manager, `createRpcDelegateServiceManager()`, provides.
Test: atest
Bug: 182914638
Change-Id: Ia9ff8306dddb1f48422a73594017680930175461
|
|
Make sure we're catching the recently exposed error conditions for these
(turns out these APIs are almost used nowhere!).
Fixes: 192023359
Test: boot, CtsNdkBinderTestCases
Change-Id: If6cdcf18520e874b28b33b80c5b4294ed8d4a302
|
|
|
|
|
|
|
|
* changes:
libbinder_ndk: eradicate global lock
libbinder: add IBinder::withLock
libbinder_ndk: remove global lock on proxy delete
libbinder: attachObject APIs work with threads
rename binderParcelTest binderUnitTest
|
|
This API doesn't really work in an OOM situation (even before the return
type was changed to a void*). Since we don't actually need or want to be
handling these situations gracefully in userspace Android, changing it
to an abort.
Bug: 192023359
Test: N/A
Change-Id: I8b7405bceb528e11a194c0d659d365975107b53f
|
|
Let the nightmares cease!
Without contention, peace.
Note, in the case that an ABpBinder is backed by a BBinder (this happens
when you talk in-process to something between the NDK backend and either
the C++, Rust, or Java backends of AIDL), this introduces an additional
allocation b/c the BBinder's mExtra needs to be allocated for
BBinder::withLock to work. Since this is generally frowned upon, and the
thrashing in this case is already pretty egregious, I don't mind it.
Now, binder proxies can be allocated simultaneously in a process.
Bug: 192023359
Test: CtsNdkBinderTestCases
Change-Id: Ib1c28c9488f7a03579ea9d18347a8fc08cc3f48a
|
|
BpBinder (and BBinder, once its mExtra allocation is made) have very
nice and shiny locks which they keep all for themselves! Stop it,
IBinder! Share that lock!
This provides convenient access to IBinder's lock, in order to avoid
needing additional locks elsewhere.
Bug: 192023359
Test: N/A
Change-Id: Id3485a2ac66d19379dcad2f0b41d6cb7a8a96725
|
|
|