The problem comes from a deadlock with AudioPolicyService mutex: When the second ringtone starts,
this mutex is locked by AudioPolicyService::startOutput() which in turn calls setParameters() to change the output device.
Audioflinger::ThreadBase::setParameters() signals the parameter change to the AudioFlinger mixer thread and waits for a condition
indicating that the parameter change has been processed.
At the same time, the mixer thread detects that the audio track corresponding to the first ring tone has been killed and calls its destructor.
This calls AudioPolicyService::releaseOutput() which tries to lock the AudioPolicyService mutex.
If this happens before the mixer thread can process the setParameters() command we are deadlocked.
The deadlock ends because setParameters() uses a timeout when waiting for the condition.
This regression was introduced by change 33736 fixing issue 2265163.
The fix consists in calling AudioPolicyService::releaseOutput() from Track::destroy() instead of from Track destructor: as detroy() is never called from the mixer thread loop (as opposed to the destructor) the deadlock described above cannot occur.
Binary XML file line #37: Error inflating class <unknown> after adding a secondary account
Now that I have these debug logs, I want to keep them since they will make
debugging these kinds of issues a lot easier in the future. (Note in this
case there was no problem in the framework.)
Change-Id: If2b0bbeda4706b7c5dc1ba4a5db04b74f40e1543
This is a second attempt to fix the audio routed to earpiece syndrom.
The root cause identified this time is the crash of an application having an active AudioTrack playing on the VOICE_CALL stream type.
When this happens, the AudioTrack destructor is not called and the audio policy manager is not notified of the track stop.
Results a situation where the VOICE_CALL stream is considered as always in use by audio policy manager which makes that audio is routed to earpiece.
The fix consists in moving the track start/stop/close notification to audio policiy manager from AudioTrack to AudioFlinger Track objet.
The net result is that in the case of a client application crash, the AudioFlinger TrackHandle object (which implements the remote side of the IAudioTrack binder interface) destructor is called which in turn destroys the Track object and we can notify the audio policy manager of the track stop and removal.
The same modification is made for AudioRecord although no bug related to record has been reported yet.
Also fixed a potential problem if record stop is called while the record thread is exiting.
since we're using the GPU for composition, don't use a texture for dimming,
instead simply use an alpha-blended quad.
also workaround what looks like a GL driver bug by calling glFinish() before
glReadPixels().
2206097: Broken suggestions while composing message
2166583: Color artifacts with MDP dithering
2261119: Passion transition animations are rough
2216759: Screen flicker when dropdown list in background window shows or hides
This is part of enabling GPU composition instead of using the MDP. This change
is dependent on another change in the vendor project.
Specifically this change disables the use of EGLImageKHR for s/w buffers
for cache coherency reasons. memcpy is used instead.
Surface::validate() could sometimes dereference a null pointer before checking it wasn't null.
This will prevent the application to crash when given bad parameters or used incorrectly.
However, the bug above probably has another cause.
in the kernel requires a guard page, so 1M allocations fragment memory very
badly. Subtracting a couple of pages so that they fit in a power of
two allows the kernel to make more efficient use of its virtual address space.
Signed-off-by: Rebecca Schultz Zavin <rebecca@android.com>
This builds on the EGLImage solution. We simply use copybit to convert from the
YUV frame into an EGLImage created for that purpose and proceed with the
regular EGLImage code.
We need to do this because "regular" GL doesn't support YUV textures.
We could improve upon this by detecting exacly what the GL supports and bypass
this extra step if not required, but we'll do this later if needed.
This change goes with a kernel driver change that reduces the audio buffer size from 4800 bytes (~27ms) to 3072 bytes (~17ms).
- The AudioFlinger modifcations in change 0bca68cfff161abbc992fec82dc7c88079dd1a36 have been removed: the short sleep period was counter productive when the AudioTrack is using the call back thread as it causes to many preemptions.
- AudioFlinger mixer thread now detects long standby exit time and in this case anticipates start by writing 0s as soon as a track is enabled even if not ready for mixing.
- AudioTrack::start() is modified to start call back thread before starting the IAudioTrack so that thread startup time is masked by IAudioTrack start and mixer thread wakeup time.
To prevent buggy command implementations from poisoning binder threads'
scheduling class & priority for future command execution, we now reset the
cgroup and thread priority to foreground/normal when a binder service thread
finishes executing the designated command.
Change-Id: Ibc0ab2485751453f6dc96fdb4eb877fd02796e3f
we lost the concept of vertical stride when moving video playback to EGLImage.
Here we bring it back in a somewhat hacky-way that will work only for the
softgl/mdp backend.
Reduce sleep time in AudioFlinger mixer thread when no data has been written to output to speed up startup time when exiting standby.
The rest of the modifications for this issues is in kernel driver:
commit 0dbb0ee136ed8de757df1ae26d84556c1751deae for buffer size modification from 8192 to 4800 bytes.
Another kernel improvement that is not submitted yes will reduce delay when audio output is exiting standby.
add a way to convert a mapped "pushbuffer" buffer to a gralloc handle
which then can be safely used by surfaceflinger, without including
gralloc_priv.h
Temporarily make a function public that doesn't need to be. When
host gcc-4.0.3 is gone from the build servers we can undo this.
(Cherry-picked from eclair-mr2.)
Use EGLImageKHR instead of copybit directly.
We now have the basis to use streaming YUV textures (well, in fact
we already are). When/if we use the GPU instead of the MDP we'll
need to make sure it supports the appropriate YUV format.
Also make sure we compile if EGL_ANDROID_image_native_buffer is not supported
Instead of using glTex{Sub}Image2D() to refresh the textures, we're using an EGLImageKHR object
backed up by a gralloc buffer. The data is updated using memcpy(). This is faster than
glTex{Sub}Image2D() because the texture is not swizzled. It also uses less memory because
EGLImageKHW is not limited to power-of-two dimensions.
Added setVoiceVolume() method to AudioSystem, AudioFlinger, IAudioFlinger, AudioPolicyService.
Removed call to AudioHardwareInterface::setVoiceVolume() from AudioFlinger::setStreamVolume().
Added a workarouond to request the A2DP output standby directly to audio hardware when the sink is suspended as it seems that the suspend request often fails.
Also take into account resume requests received while a suspend request is pending.
when running out of memory, a null handle is returned but the error code may not be set.
In that case we need to return NO_MEMORY instead of NO_ERROR, so that the calling code
won't try to dereference the null pointer.
When switching rapidily orientation back and forth, surfaces end-up
acquiring the freeze-lock when the first orientation change happens,
but never release it because by the time the 2nd orientation change
comes in, the surface size is back to its original size and
doesn't appear to have resized.
we now always release the freeze-lock when we receive a buffer of the
expected size.
This also fixes [2152536] ANR in browser
When SF is enqueuing buffers faster than SF dequeues them.
The update flag in SF is not counted and under some situations SF will only
dequeue the first buffer. The state at this point is not technically
corrupted, it's valid, but just delayed by one buffer.
In the case of the Browser ANR, because the last enqueued buffer was delayed
the resizing of the current buffer couldn't happen.
The system would always fall back onto its feet if anything -else- in
tried to draw, because the "late" buffer would be picked up then.
A window is created and the browser is about to render into it the
very first time, at that point it does an IPC to SF to request a new
buffer. Meanwhile, the window manager removes that window from the
list and the shared memory block it uses is marked as invalid.
However, at that point, another window is created and is given the
same index (that just go freed), but a different identity and resets
the "invalid" bit in the shared block. When we go back to the buffer
allocation code, we're stuck because the surface we're allocating for
is gone and we don't detect it's invalid because the invalid bit has
been reset.
It is not sufficient to check for the invalid bit, I should
also check that identities match.
This change is a complement to the main fix in kernel driver for the same issue (partner change #1250).
It removes clicks sometimes heard after the end of the tones while audio flinger is sending 0s to the audio output stream.
The problem was that the sleep time between two writes was more than the duration of one audio output stream buffer which could cause some underrun.
Also fixed a recent regression in ToneGenerator that made that the end of previous tone was repeated at the beginning of current one under certain timing circumstances when the maximum tone duration was specified.
When EGLImage extension is not available, SurfaceFlinger will fallback to using
glTexImage2D and glTexSubImage2D instead, which requires 50% more memory and an
extra copy. However this code path has never been exercised and had some bugs
which this patch fix.
Mainly the scale factor wasn't computed right when falling back on glDrawElements.
We also fallback to this mode of operation if a buffer doesn't have the adequate
usage bits for EGLImage usage.
This changes only code that is currently not executed. Some refactoring was needed to
keep the change clean. This doesn't change anything functionaly.
The ANR is caused by SurfaceFlinger waiting for buffers of a removed surface to become availlable.
When it is removed from the current list, a Surface is marked as NO_INIT, which causes SF to return
immediately in the above case. For some reason, the surface here wasn't marked as NO_INIT.
This change makes the code more robust by always (irregadless or errors) setting the NO_INIT status
in all code paths where a surface is removed from the list.
Additionaly added more information in the logs, should this happen again.
The core logging in BackupManagerService and in the Google backup transport are
still enabled at this point.
Change-Id: I10abfa565bbd1097dd3631051b6aca163e4af33a
Wait for the parameter set completed condition with a time out in ThreadBase::setParameters().
Also lock AudioFlinger mutex before accessing thread list in AudioFlinger::setParameters() and keep a strong reference
on the thread being used in case it is exited while processing the request.
* changes:
fix [2152247] Windows sometimes drawn scaled up.
invalidate the surface when the physical changes
introduce the notion of the requested size in the Layer state
remove unused code
We were emitting GL commands, calling composition complete and releasing clients
without ever calling eglSwapBuffers(), which is completely wrong on non-direct
renders. This could cause transient drawing artifacts when unfreezing the
screen (upon orientaion change for instance) and could also block the clients
for ever as they are waiting for their previous buffer to be rendered.
Turning off backup in the Settings UI constitutes an opt-out of the whole
mechanism. For privacy reasons we instruct the backend to wipe all of the data
belonging to this device when the user does this. If the attempt fails it is
rescheduled in the future based on the transport's requestBackupTime()
suggestion. If network connectivity changes prompt the transport to indicate a
backup pass is appropriate "now," any pending init operation is processed before
the backup schedule is resumed.
The broadcasts used internally to the backup manager are now fully protected;
third party apps can neither send nor receive them.
(Also a minor logging change; don't log 'appropriate' EOF encountered during
parsing of a backup data stream.)
There was a regression introduced in AudioFlinger by change 24114 for suspended output:
The suspended output was not reading and mixing audio tracks.
When the phone is ringing, the A2DP output is suspended if the SCO headset and A2DP headset are the same. As the ringtone is played over the duplicated output, the fact that the A2DP output was not reading data was causing the hardware output to be stalled from time to time.
This appears to fix the sim-eng build on the gDapper build machines.
Basic problem is that LayerBuffer::OverlaySource has a constructor that
calls SurfaceFlinger.signalEvent(). SurfaceFlinger lists LayerBuffer
as a friend, but that's not enough to convince gcc that the embedded
OverlaySource class is also a friend. I don't see a way to make them
friendly, so I marked signalEvent() as public.
a new method, compostionComplete() is added to the framebuffer hal, it is used by surfaceflinger to signal the driver that the composition is complete, BEFORE it releases its client. This gives a chance to the driver to
The fix consists in locking AudioFlinger::mLock mutex in the TrackBase destructor before clearing the strong pointer to the shared memory client. The mutex is not locked in removeclient() any more which implies that we must make sure that the Client destructor is always called from the TrackBase destructor or that we hold the mLock mutex before calling deleting the Client.
Take 2. We needed to check that the usage flags are "good enough" as opposed to "the same".
This reverts commit 8f17a762fe9e9f31e4e86cb60ff2bfb6b10fdee6.
The problem comes from the fact that when the duplicated output is closed after BT headset disconnection, the OUTPUT_CLOSED notification is not sent to AudioSystem. Then the mapping between notification stream and duplicated output cached in AudioSystem is not cleared and next time a notification is played, the duplicated output is selected and the createTrack() request is refused by AudioFlinger as the selected output doesn't exist.
The notification is ignored by AudioFlinger because when it is sent by the terminating playback thread, the thread has already been removed from the playback thread list.
The fix consists in sending the notification in closeOutput() and not when exiting the playback thread.
The same fix is applied to record threads.
This is due to a regression introduced by change 24114: when no audio tracks are ready for mixing, 0s are written to audio hardware. However this should only happen if tracks have already been mixed since the audio flinger thread woke up.
Also do not write 0s to audio hardware in direct output threads when audio format is not linear PCM.
Appears to have been broken by:
commit 9779b221e999583ff89e0dfc40e56398737adbb3
Author: Mathias Agopian <mathias@google.com>
Date: Mon Sep 7 16:32:45 2009 -0700
fix [2068105] implement queueBuffer/lockBuffer/dequeueBuffer properly
For some reason we don't like to have "-lpthread" globally -- it's a no-op
on device builds, but required for many host tools and all sim binaries --
so adding the use of pthread calls requires adding the library explicitly.
AudioFlinger: verify that mCblk is not null before using it in Track and RecordTrack contructors.
IAudioFlinger: check result of remote transaction before reading IAudioTrack and IAudioRecord.
IAudioTrack and IAudioRecord: check result of remote transaction before reading IMemory.
we could have several thread waiting on the condition and they all need to wake-up.
also added a debug "mTid" field in the class, which contains the tid of the thread (as opposed to pthread_t), this
is useful when debugging under gdb for instance.
we ended-up locking a Mutex that had been destroyed.
This happened because we gave an sp<Source> to the outside world,
and were called after LayerBuffer had been destroyed.
Instead we now give a wp<LayerBuffer> to the outside and have it
do the destruction.
Add a parameter to ToneGenerator.startTone() allowing the caller to specify the tone duration. This is used by the phone application to have a precise control on the DTMF tone duration which was not possible with the use of delayed messaged.
Also modified AudioFlinger output threads so that 0s are written to the audio output stream when no more tracks are ready to mix instead of just sleeping. This avoids an issue where the end of a previous DTMF tone could stay in audio hardware buffers and be played just before the beginning of the next DTMF tone.
Rewrote SurfaceFlinger's buffer management from the ground-up.
The design now support an arbitrary number of buffers per surface, however the current implementation is limited to four. Currently only 2 buffers are used in practice.
The main new feature is to be able to dequeue all buffers at once (very important when there are only two).
A client can dequeue all buffers until there are none available, it can lock all buffers except the last one that is used for composition. The client will block then, until a new buffer is enqueued.
The current implementation requires that buffers are locked in the same order they are dequeued and enqueued in the same order they are locked. Only one buffer can be locked at a time.
eg. Allowed sequence: DQ, DQ, LOCK, Q, LOCK, Q
eg. Forbidden sequence: DQ, DQ, LOCK, LOCK, Q, Q
Do not ramp volume if the first frame of a track is processed after the track was stopped.
In the case of very short sounds, the track stop request can be received by AudioFlinger just after the start request before the first frame is mixed by AudioMixer. In this case, the track is already in stopped state and initial volume is applied with a ramp for the first frame processed which should not be the case: initial volume change is always applied immediatelly.
This addresses a few parts of the bug:
- There was a small issue in the window manager where we could show a window
too early before the transition animation starts, which was introduced
by the recent wallpaper work. This was the cause of the flicker when
starting the dialer for the first time.
- There was a much larger problem that has existing forever where moving
an application token to the front or back was not synchronized with the
application animation transaction. This was the cause of the flicker
when hanging up (now that the in-call screen moves to the back instead
of closing and we always have a wallpaper visible). The approach to
solving this is to have the window manager go ahead and move the app
tokens (it must in order to keep in sync with the activity manager), but
to delay the actual window movement: perform the movement to front when
the animation starts, and to back when it ends. Actually, when the
animation ends, we just go and completely rebuild the window list to
ensure it is correct, because there can be ways people can add windows
while in this intermediate state where they could end up at the wrong
place once we do the delayed movement to the front or back. And it is
simply reasuring to know that every time we finish a full app transition,
we re-evaluate the world and put everything in its proper place.
Also included in this change are a few little tweaks to the input system,
to perform better logging, and completely ignore input devices that do not
have any of our input classes. There is also a little cleanup of evaluating
configuration changes to not do more work than needed when an input
devices appears or disappears, and to only log a config change message when
the config is truly changing.
Change-Id: Ifb2db77f8867435121722a6abeb946ec7c3ea9d3
In practice, no one ever writes an apostrophe in an aapt string with the
intent of using it to quote whitespace -- they always mean to include a
literal apostrophe in the string and then are surprised when they find
the apostrophe missing. Make this an error so that it is discovered
right away instead of waiting until late in QA or after the strings have
already been sent for translation. (And fix a recently-introduced string
that has exactly this problem.)
Silence the warning about an empty span in a string, since this seems to
annoy people instead of finding any real problems.
Make the error about having a translated string with no base string into
a warning, since this is a big pain when making changes to an application
that has already had some translations done, and the dead translations
should be removed by a later translation import anyway.
In AudioFlinger::MixerThread::putTracks(), change the mFillingUpStatus flag to FS_FILLING for active tracks so that mute request is executed without ramping volume down when the track is moved from A2DP to hardware output.
Also modified AudioFlinger::setStreamOutput() so that the notification of the change is sent only once to AudioSystem.
(in this case the state is dumped without the proper locks held which could result to a crash)
in addition, the last transaction and swap times are printed to the dump as well as the time spent
*currently* in these function. For instance, if SF is unresponsive because eglSwapBuffers() is stuck,
this will show up here.
what happened is that the efective pixel format is calculated by SF but Surface nevew had access to it directly.
in particular this caused query(FORMAT) to return the requested format instead of the effective format.
