ANativeWindow::queueBuffer takes ownership of the fence fd passed to
it, and will close it before returning. SurfaceFlinger's screenshot
code was also closing the syncFd it passed to queueBuffer. Most of the
time this meant the second close() silently failed, but in a rare race
condition the file descriptor could be reused between the two
close()s.
Bug: 17946343
Change-Id: Ib74fcb1dce52cc21328059c99b7c4c76f41aa3a5
Print more details about the exact reason that an ANR has occurred.
Also start checking that the window actually has a registered
input connection that is not in a broken state. These windows
are supposed to be cleaned up by the window manager promptly
as if the app had crashed but the pattern of ANRs we are observing
suggests that broken windows might be sticking around longer than
they should.
Bug: 17721767
Change-Id: Ie2803a3fa9642381ecadc198fec15e1b70d93c20
Blobcache is not yet enabled for surfaceflinger (as it should be).
As a temporary workaround, generate all needed shaders during
surfaceflinger initialization instead of doing the compilation
on-demand during ui transitions.
Change-Id: I14455b20a3f85f177d85c9c8b76d8ccc35379b39
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
We normally recompute layer visibility when a layer gets its first
buffer; before then it's treated as invisible. Sideband layers never
get a buffer (as far as SurfaceFlinger knows), so never became
visible. Now we also recompute visibility when a layer gets a new
sideband stream.
Bug: 17752511
Change-Id: I84e150f196eb2eb7bcd2616248e5e3fa73624809
When HWC doesn't provide DPI values for a display, we pick a default
DPI based on resolution. The intent was that 1080p and higher displays
would get XHIGH density, and lower resolutions would get TV density.
In KK (and possibly forever) we had a bug that we'd always use TV
density. That was fixed in L, but that fix exposed a pre-existing bug
that we always used the display's height in its native orientation,
rather than in landscape orientation. So an 800x1280 tablet like N7v1
started getting XHIGH density instead of the intended TV density.
Bug: 17461633
Change-Id: Ia57fa49e61f36bdda63ce283ef62c9953297222c
If the destructor of SensorEventConnection gets called when the main
sendEvents loop of SensorService is executing it may result in a deadlock.
The loop promotes each connection to a strong_pointer, calls sendEvents
and cleans up the connection if necessary. It is possible that the sp's
destructor may delete SensorEventConnection which will call the dtor
~SensorEventConnection(). This dtor again needs SensorService mLock to
execute which may result in a deadlock.
Bug: 17617897
Change-Id: I76c244dbe85fadb591c0bd1a9a5eb01d93f56505
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) Initialize sensor_event.flags to zero or wake_up flag before sending dropped flush_complete_events. This will avoid unnecessary acks.
Currently it is not set to anything which may result in garbage values for flags field.
ii) Change the dump format of SensorEventConnection.
iii) Add const to method signatures wherever applicable.
Bug id: https://b2.corp.google.com/issues/17498626
Change-Id: Ia736ffb01c366540aab73cdcf596ce852c052593
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
But provide a temporary escape hatch for AudioFlinger.
This oneway option will be removed as soon as possible.
Bug: 16408906
Change-Id: I20d6da1969ae05b96e72795463470eb4c1f8fbdc
make all binder call to power manager service one way so that
they don't block calling thread and are processed in the
incoming order.
Bug: 16408906.
Change-Id: I94034df7f0a36d967e085b53fdc37fff7b44757b
i) Fix a possible race condition when adding a pending flush connection
to mActiveSensors data structure.
ii) Invalidate flush_complete_events as soon as they are consumed by a
connection so that they are not used by other connections which are
registered for the same sensor.
Change-Id: I1491bea8c2081c4aab1feacc60add13b7aa26bb7
This makes it possible to tell whether two HW vsync signals are too
far apart because one was late, or because we turned HW vsync off
briefly between them.
Bug: 17259382
Change-Id: If4fba2a8a6013568349949ce87c5c36f0468d2a2
1) Limit maxEventsToWrite depending on the size of the socket
buffer allocated (Fix for write failure on manta).
2) Do not call flush() if the sensor is not activated (OEM issue).
Change-Id: I81ea78dbdef3a3773af505ca1c9a5e39492c0b48
Display configs for external displays are stored in the framework
everytime we receive a hot plug to connect. However, since the
configs are not cleared on disconnect, framework will just assume
that the configs are valid. This does not work for use cases when
you connect/disconnect external displays with different resolutions.
e.g. 1080p to 4K and vice-versa
With this change we clear the display configs and repopulate when
we receive a hot plug to connect.
Change-Id: I2eeab186a8d8668a53390a2413b2ce5e044a1845
Acked-by: Tatenda Chipeperekwa <tatendac@codeaurora.org>
By not committing the results of composition for empty frames, we
avoid spitting out series of black frames for virtual displays that
don't have visible layers. We still draw one black frame when going
from having layers to not having any. In particular, this avoids
having a series of empty frames due to re-compositing the primary
display in the period between creating the virtual display and adding
layers to it.
Bug: 16786752
Change-Id: I7e9b2ed2e407d8d49c7af736b447d4c6181b0ad8
Do not wait for the screen capture to complete within surface flinger,
instead pass a sync point back with the captured gralloc buffer.
Change-Id: I7137c0e0fc710688d1d61f189159418fb27ea263
