On one device there is a bug, not yet root-caused, that causes fence
fds to not make it across binder from producer to consumer in the
IGraphicBufferProducer::queueBuffer call. Rather than returning an
error, which the producer typically treats as a fatal error, this
change allows the buffer to be queued with no fence. This avoids an
application crash at the risk of (likely single-frame) visible
corruption.
Bug: 17946343
Change-Id: I9ca89f94098c455e1e90f5f58d5336c936b04a9c
Previously it was possible to have the driver's eglTerminate called beofre
eglDestroyImageKHR in GLConsumer. This was because we didn't increment the
refcount for the lifetime of the image. This could lead to a crash or a deadlock
when multiple threads called terminate and destroy simultaneously.
Bug: 17700483
Change-Id: I7010d0f1b3db875332e95630b5e098a5564ba755
mmap returns MAP_FAILED (which is -1) and not NULL on
failure.
Diagnosed by cferris.
bug: 17909809
Change-Id: I609788ebf94742ef88af002d2d3f3bc9b9e520ac
Temporary extra debug validation for b/17477219: a Parcel recipient is
getting a positive but invalid fd unexpectedly. Trying to track down
where it's coming from.
Debug code for bug: 17477219
Change-Id: Idb1e71621025a3928c7adc88fd44790e1abd2a01
Throttling was previously controlled by a combination of the
driver and the number of buffers in the queue. This patch makes
a more consistent trade-off, which allows two GPU frames pending
but not three. More buffering could improve throughput in the
case of varying frame times, but this also increases latency.
Bug: 17502897
Change-Id: I4ee68019ca94c635294c5959931a555a6c4ef2df
i) Call removeFd() only if the fd in the BitTube has been
previously added to the Looper. Use a flag to determine whether the fd
has been previously added or not.
ii) Increment mPendingFlushEventsToSend after holding a connectionLock.
iii) Store the number of acks that are pending in SensorEventQueue
and send them all at once.
Bug: 17472228
Change-Id: I1ec834fea1112a9cfbd9cddd2198438793698502
Try to clean up the code paths coming in and out of binder IPCs to
plug any places where we could disrupt the gather flag of a thread,
causing it to keep gathering stack crawls (which is the thing that
is causing our strict mode data to become so large).
We now take care of saving and restoring this state in the core
IPC code path, not at the Java layer.
Change-Id: I73d564778da127bdce00f304225930e7f2318293
This is used by media service to schedule video frames at the
proper time, based on precise vsync timings.
Bug: 14659809
Change-Id: I1a90603f3dc09dca9aa4f90a3aa845fab56e0a5e
i) Significant Motion multiple clients fix. Make a copy of
mActiveConnections vector before cleaning up SensorEventConnections
when one-shot sensors trigger.
ii) Maintain a mapping between flush_complete_events and
SensorEventConnections to accurately map flush() API calls and
corresponding flush_complete_events
iii) Remove all references to 1_1 and 1_2 HALs.
iv) Dynamically allocate sensor_event buffers in SensorService main
threadLoop.
Bug: 17412359
Change-Id: If3c3986197660cafef2d2e0b4dc7582e229cf1c4
+ This is needed so that activity manager does not
have to do cpu side rotations when capturing recents
thumbnails.
Change-Id: If998008e675ad01305db8399fd643cf4608b7025
After fixing b/16874785.
This reverts commit f010a05c7e.
Original comment, which actually describes the effect of this:
Change the mExtras field in Binder.h to be a stdatomic.h atomic
value, and replace references to it with proper stdatomic.h calls.
This removes one of a small number of remaining 64 bit
android_atomic references. It also replaces the erroneously
non-atomic read accesses to mExtras.
It would be better if this used the C++11 <atomic> facility,
but we don't quite have that yet.
Fixes
Bug:16513433
Change-Id: I1645ca5d6f60595bf5d388913665ce4b8780b26d
(cherry picked from commit 3effababf2)
If a display is terminated and then initialized, we can't detect
this using the display itself (it has the same value), but all
EglImages still become invalid for the display. This patch detects
this during image binding and forces creation of a new EglImage.
Bug: 10430249
Change-Id: I75101c50962f21263dca3ec6e241a2e5a3c23dad
Modify SurfaceFlinger to use VirtualDisplaySurface in all cases when a virtual
display is used. Add functionality in VirtualDisplaySurface to resize the
buffers aquired in the QueueBufferOutput. Add transaction support in
SurfaceFlinger for resize. Add the modification of the size in DisplayDevice.
Change-Id: Iae7e3556dc06fd18d470adbbd76f7255f6e6dd6b
Tested: None
Commit 78014f32da introduced a bug that
made us pre-allocate buffers into the last available free slots instead
of the first available ones. This in turn caused more re-allocations,
and possibly triggered driver bugs.
Change-Id: Ic4a70e676b4f2bbb054bc873be62ced26e3099a0
i) Send ack for wake_up sensors on the socket connection instead of using Binder RPC.
ii) Cache events per connection in case there are write failures. Compute cache size
from FIFO counts of sensors.
iii) Send FlushCompleteEvent only for apps that explicitly called flush().
Change-Id: I018969736b7794b1b930529586f2294a03ee8667
If getNativeBuffer() is called on a NULL GraphicBuffer the
static_cast of this from GraphicBuffer* to ANativeWindowBuffer*
will return a small pointer like (ANativeWindowBuffer*)0x10.
This value can propagate past NULL checks until it causes a crash
far away from the original NULL pointer. Crash immediately
instead.
Change-Id: Id614b9eb1484108b3c3c733545309844c4b87532
BufferQueueProducer::allocateBuffers used to keep the BufferQueueCore
mutex while doing the buffer allocation, which would cause the consumer
(which also needs the mutex) to block if the allocation takes a long
time.
Instead, release the mutex while doing the allocation, and grab it again
before filling the slots. Keep a bool state and a condvar to prevent
other producers from trying to allocate the slots while the mutex is
released.
Bug: 11792166
Change-Id: I4ab1319995ef892be2beba892f1fdbf50ce0416d
(cherry picked from commit ea96044470)
In most cases, EGLImages can be created one-to-one with graphic
buffers in slots, but that was difficult due to some special
cases:
- ReleaseTexImage binds a custom 'unslotted' debug image.
- When all slots are freed, we still need to hang on to one.
These cases were handled by keeping an additional reference to
the 'current' buffer (mCurrentTextureBuf), but we would create
new images since we can't reference count them in the same way.
This patch uses the same semantics, except that it reference
counts the image (an EglImage wrapper class) rather than just
buffer. The wrapper class also detects the cases when we need
a new EGLImage, and only creates them in those rare cases.
Change-Id: I2915761dbe49d2a9bda1f59e60f857543634636b
