GitLab

Move 3A notification synthesis for HAL3 devices from
Camera3Device::processCaptureResult to Camera2Client's
FrameProcessor. This will ensure that calls to processCaptureResult
from HAL can never block on Camera2Client internal mutexes.

Bug: 9923891
Change-Id: I5184649bf45c0807babe6b8c0e1239e959cd3480

Preview callback oneshot flag need set when a still picture is being captured
and preview callback oneshot is enabled, otherwise, the callback enabling status
could be lost when preview is restarted after still catpure.

Bug 9742091

Change-Id: If9562369013e53727edfdd16f3d00d6a9392e70d

When starting a recording session, check for and clear out recording
buffers still owned by stagefright from a previous recording session,
if any.  There should never be any, but in case they are due to a bug
elsewhere in the system, clearing out the buffers makes recovery
possible.

Bug: 9591080
Change-Id: I3a2a1256860bd174381cd525948419f985c9b415

bug: 9672606
Change-Id: I98cf65260d5e43d34e3897452f6f26ed2c89294e

Check status of transactions to scheduling policy service
and re-acquire a binder interface in case of DEAD_OBJECT.

Bug: 8875559.
Change-Id: I1e00bd44e2d4723b3ec95d5c31d9652ba08e238a

- Tear down conflicting streams when necessary.

- Shut down callbacks if recording starts

- Do not allow callbacks to start if recording is active

Per the current camera API, recording and preview callbacks cannot be
active simultaneously. However, the framework did not explicitly
disallow this, and in fact left the streams configured once they were
created, even if switching between the two operational modes.

In addition, no guards existed for trying to enable both recording and
callbacks at the same time.

Bug: 9423825

Change-Id: I7d6e6114c2e14fcfb5299b4c72ad557895cbf4b8

EQ and bass boost can improperly interact with other effects
 (e.g. virtualizer) outside of the bundle. Apply them first
 in an effect chain, which is equivalent to using them to affect
 the content of a track, rather than how a track is presented,
 when additional effects are cascaded behind EQ and BB.

Bug 8589316

Change-Id: Ifb34e46bc1969f613cdc90ee2a79af677e2fad92

It's possible, during shutdown, for callback processor's heap to be
destroyed when it's about to send the callback back to the user.
Properly copy the heap reference to a local variable before unlocking
the mutex.

Bug: 9485959
Change-Id: I301347b77145f19c7ac721b9127dc74f122acce2

Bug: 9466221
Change-Id: I9ed0740561b2e04954eebc1b8aec89cfab40a3a0

While this wait should be immediate, it's important to guarantee
that preview is stopped when takePicture completes, and that the
device knows itself to be idle.

Bug: 9413025
Change-Id: I1c151ba920e226107ee048918187364990f444c0

To improve robustness, make sure the HAL device instance is always
closed, even if it cannot be put into the IDLE state.

Bug: 9414167
Change-Id: I8e7035ff0afe7214951f9824e561b340cb37592e

If a HAL-level stream reconfigure is not needed, still need to
transition camera device to ACTIVE state in configureStreams.
Otherwise, shutdown functions will not wait for device to drain
properly.

Bug: 9413025
Change-Id: I0b15792097c7e5771015f2a362735d5f88b8c6ea

Only change the nativewindow crop before queuing frames
with the new crop.

Change-Id: Icd8a84c82e66a8be3d6bc1675c777b1de8dc39b9
Signed-off-by: Lajos Molnar <lajos@google.com>
Bug: 9395753

If configuring the same set of streams more than once, don't actually
call into the HAL for the second and subsequent configure calls, since
they're no-ops.

This can speed up camera operation substantially if the
HAL implementation does not detect no-ops on its own and does a full
shutdown/restart on each configure call.

Bug: 9392513
Change-Id: I23baf4acbae2304735899adcf8e17565fa94d31d

PlaybackThread::addTrack_l() uses the assumption that
effects are attached to a track only if the track accumulation
buffer is different from the mixer thread output buffer.
This is not true for direct output threads where only one track is active
an only one buffer is needed.
This assumption is an optimization to avoid checking for effect chains with
the same session ID each time a track is processed. The optimization
is not key if only one track is attached to the thread which is the case for
direct outputs.

Current code fails to increment the active track count in the effect chain
on direct output threads when a track is started thus making the effect
framework clear the mix buffer and produce silence each time the mixer runs.

The fix consists in removing the optimization described above.

Bug: 9324989.

Change-Id: Id7a6337450ed90d326299c2ce9fc02f4b9e2fa6f

Instead of creating the JPEG output stream only at first use,
create it at time of preview start. This is important for reducing
the first-capture latency on HAL3 devices, and for ZSL on them.

Bug: 9339858
Change-Id: I4296f706cfd151c47ef315149e87d01fe554aaa4

AudioFlinger should not cache the audio device
passed to the audio HAL when it is AUDIO_DEVICE_NONE
but keep previous valid selection instead.

Bug: 9323399.
Change-Id: I6f9480e55a21be4115453e6a5eebc5cf2536c476

Acquiring mInputMutex before SharedParameters mutex in preview callback thread
causes circular locking dependency between callback thread and capture callback
thread.
Enforce the right lock ordering to break the dead lock.

Bug: 9323319
Change-Id: Iea2e025e4d9e29debcf74297c19930075017e179

When camera open fails in camera3device initialize call, setErrorStateLockedV()
tries to access RequestThread object that is not created yet.

Bug: 9319518
Change-Id: Id5c6c0150ecaecb198045d5c6eb1b2ceebf5a727

Callback stream in callbackprocessor could be delleted while process or
discard callback are ongoing, which leads CpuConsumer related call crash.

Bug 9303911

Change-Id: Ib6137e512208e5603468eba166640729b744e8f9

Due to new deadlock opportunities from HAL3 integration, recording
frame processing needs to move out of the new frame handler into its
own thread in StreamingProcessor.

Bug: 9282913
Change-Id: I8d2926f2de9d0d395eef3c328c65caebbdbd2827

While the HAL should ignore it, it's faster to immediately respond
with success.

Also filter out cancel autofocus calls for FIXED/INFINITY focus.

Bug: 9263856
Change-Id: I637ec7c8b459d654ef11cb1142b959f2674224c5

Change-Id: Ie0dd00045aba668d8b49da73224e7a7c9c04f69b
related-to-bug: 8873723
(cherry picked from commit 2704965b8a1ff3b7450ff58ccecf86d8ec688c40)

Remove the invalid assumption that when a port-reconfiguration event
is received, buffers cannot be downstream (waiting to be rendered).

Luckily, these buffers are properly handled (freed) after they are
sent to be rendered.

Change-Id: I1df39c1ffc2bfb96f8b7b4ee5be07cae654f956f
Signed-off-by: Lajos Molnar <lajos@google.com>
Bug: 8736466

MediaMuxer.writeSampleData() uses MediaCodec.BufferInfo
objects to specify the flags, so use directly the MediaCodec
flags.  This is more intuitive and avoids users of the
MediaMuxer API having to create translations between
MediaMuxer's and MediaCodec's flag constants.

Change-Id: I18e12a6839ffb2c63da62516a4dc4c94e8bb3d17
Signed-off-by: Lajos Molnar <lajos@google.com>
Bug: 9169479

This fixes the multiple-disconnect-from-BufferQueue bug.

(In logs)
BufferQueue: [SurfaceView] disconnect: connected to another api (cur=0, req=4)

Bug: 9114681
Change-Id: I21ae1c5bd993701eea13648aa9994979a24eb210

Bug: 9111852
Change-Id: Idad7e0d2f912341bd643d0ad1a0861fb2043aa90