We're not using IMemoryHeap as a transport anymore,
instead we're providing a CpuConsumer and use the
IGraphicBufferProducer version of the screenshot API.
However, some GPU drivers don't support properly
a GPU to CPU path, to work around this, we use a
temporary BufferQueue on the server side for the
GL rendering, and we use glReadPixels into the
CpuConsumer (we're now using a CPU to CPU path
which is always supported).
Currently this "wrapping" is always performed,
but it can be bypassed on devices that support
the GPU to CPU path.
This also addresses a DoS attack vector on
SurfaceFlinger, where an application could
consume all of SF's filedescriptors by creating
a lot of screenshots in a row.
Bug: 8390553
Change-Id: I9e81514c2a7711b9bb393f74305be7d2abe08f1c
background:
we have some code to fix-up the IDs of references when
using RefBase's DEBUG_REFS when those refs are managed by
arrays wp<> or sp<> (this is because wp<> / sp<> don't have
a trivial ctor when DEBUG_REFS is enabled, and Vector
treats them as trivial for obvious performance reasons)
this is complicated by the fact that we don't want to have
to recompile everything when enabling DEBUG_REFs (i.e.: the
Vector code cannot know wheter it's enabled or not for its
template stuff).
problem:
there was a bug in the fix-up code for wp<> which was trying
to access the weakref_impl from the RefBase* however, this was
moronic since RefBase could have been destroyed if there wasn't
any more strong refs -- and this happned. Instead we need to get
the weakref_impl directly from the wp<>
Change-Id: Ie16e334204205fdbff142acb9faff8479a78450b
* commit '88f70a2167528ec849845e1203cfd9c6ff35a67d':
Add BufferQueueInterposer and use it for virtual displays
Add DisplaySurface abstraction
Fix argument types in IGraphicBufferProducer methods
Minor cleanups/fixes before virtual display refactoring
- stacks are now saved in /data/debug which must be
created and writable by the user.
- removed "always fatal" DEBUG_REFS option, it wasn't
really needed.
- DEBUG_REFS_ENABLED_BY_DEFAULT is not the default anymore
(usually people want to target which refs they're tracking)
Change-Id: I37fae72e9dacde6ce1fa8f7dbe2bc01b1a1b95e5
Third-party libraries are currently trying to use the
MemoryBase constructor but failing because we fixed the
definition of ssize_t. This is a stop-gap for users of
this private API until we can get them fixed.
Bug: 8253769
Change-Id: Ie7c86f45fa39cb53539cab0ffe8585a585656714
Third-party libraries are currently trying to use the
MemoryBase constructor but failing because we fixed the
definition of ssize_t. This is a stop-gap for users of
this private API until we can get them fixed.
Bug: 8253769
Change-Id: I8a19770f3252d88ee87023fde625cc6289924b0d
ISurface was only used to get the IGraphicBufferProducer from
a Layer. It's now replaced by a BBinder subclass / IBinder and
is only used as a handle to the surface, to both refer to it
and manage its life-time.
Also cleaned-up a bit the ISurfaceComposer interface and
"create layer" code path.
Change-Id: I68d0e02d57b862cffb31d5168c3bc10cea0906eb
When disconnecting from BufferQueue, we now drain the queue
except the head (which means in the screenshot case we won't
have to block, but we might not have a buffer to show, this
will appear as an error in the log).
Bug: 8362363
Change-Id: If80989aac3c917beea2ebddf3cbb502849d394da
we check that the order in which we destroy GLConsumer wrt.
releasing the corresponding EGLSurface via eglMake(Un)Current
doesn't leak a buffer.
On at least 2 devices this test doesn't pass.
Change-Id: I63ab83951b4b0a977f38571158f948cbd9dc7cec
When a binder service's main thread joins the thread pool
it retains its name (whatever the exec name was), which is
very confusing in systrace.
we now rename that thread just like its friends in the
thread pool.
Change-Id: Ibb3b6ff07304b247cfc6fb1694e72350c579513e
CpuConsumer cannot simply assume a slot's buffer is the same buffer
between acquire and release, and therefore it could be possible for
the same slot to get used for a second acquired buffer, if there's a
producer disconnect in between. This would cause a problem when the
first buffer is released by the consumer.
Instead, use an independent list of acquired buffers to properly track
their state.
Bug: 8291751
Change-Id: I0241ad8704e53d47318c7179b13daed8181b1fab
Writing a NULL Surface was being read as a non-NULL Surface with NULL
mGraphicBufferProducer. Before the SurfaceTextureClient -> Surface
refactoring, you'd get a NULL Surface, and some code relies on that.
Bug: 8291161
Change-Id: I477bfe8882693e53a5f604a3d2c9e3cfe24473b4
- SurfaceFlinger now supports to take a screenshot
directly into an IGraphicBufferProducer
- reimplement the IMemoryHeap screenshot on top
of the above
- reimplement LayerScreenshot such that its
BufferQueue is directly used as the destination
of the screenshot. LayerScreenshot is now a thin
wrapper around Layer
Bug: 6940974
Change-Id: I69a2096b44b91acbb99eba16f83a9c78d94e0d10
Temporary, to fix weekend build, until we get Nvidia code drop.
This reverts commit 9a867a8798
DO NOT MERGE
Change-Id: I7b5dbc4db46ef3d97dc8598057d5487d6971178b
T-junction free regions are useful for rendering regions with various
geometric transformations, and the Region's span-ordered, sorted rect
list supports T-junction free storage without modification.
This approach creates a T-junction free region by splitting each
rectangle that is part of a vertical T-junction. This approach is two
pass (up and down) so that divisions can trickle up/down to other
adjacent spans.
Change-Id: Ifcf5e6fe0034c96b00ef09a4433b2b0fce8f4300
We check that calling eglSwapBuffers() on an abandonned BufferQueue
return EGL_BAD_SURFACE -- this is to ensure consistancy between
drivers.
Change-Id: Ibb548e0cf767ceee69f2fc4a85811d15a6522277
This change eliminates the uses of a NULL sp<Fence> indicating that no waiting
is required. Instead we use a non-NULL but invalid Fence object for which the
wait methods will return immediately.
Bug: 7892871
Change-Id: I5360aebe3090422ef6920d56c99fc4eedc642e48
This is to help implementation of bug #8181262 and maybe
bug #8181261
Note the current code has not yet been tested; it is only
known to compile at this point.
Change-Id: I489674c96d0d3fc0ddacc92611931a19a9ee5230
The C++ class names don't match what the classes do, so rename
ISurfaceTexture to IGraphicBufferProducer, and SurfaceTexture to
GLConsumer.
Bug 7736700
Change-Id: Ia03e468888025b5cae3c0ee1995434515dbea387
CallStack::toString() has a 0 default argument, which ends up getting
passed to strlen(), resulting in a crash.
Change-Id: If706aff8c400983670f49cdbb66e11191ac76e0e
In SurfaceFlingerConsumer, check to see if native fence sync is
enabled. If so, defer the texture binding step to Layer::onDraw.
Change-Id: I7d4034a31c0143207eea2509dfa13ef3820f9b8c
Rearranges updateTexImage() so that the SurfaceFlinger-specific
behavior is in a new SurfaceFlingerConsumer subclass.
SurfaceTexture behavior should not be altered. Instead of
acquire-bind-release we now do acquire-release-bind, but since
it's all done with the lock held there shouldn't be any
externally-visible change.
Change-Id: Ia566e4727945e2cfb9359fc6d2a8f8af64d7b7b7
This change makes GraphicBufferAllocator::alloc wait for pending async frees to
complete before attempting to allocate a gralloc buffer if there are more than
8 pending async frees.
Bug: 7696861
Change-Id: I1fae86e13edefcaa153b8ce9fd057f335716059e
This change makes GraphicBufferAllocator::free queue a job to another thread to
perform the actual free operation. This prevents potentially slow free
operations from blocking rendering.
Bug: 7675940
Change-Id: Id61099d66bb4c3949d04184e0d7f192ac18076b4
This change makes ConsumerBase::onBuffersReleased hold a reference to all its
gralloc buffers until after the mutex is unlocked. This prevents slow
gralloc::free calls from causing lock contention with rendering threads.
Bug: 7675940
Change-Id: I0ec805d1b612afeeecfffec03f982371d27d93be
ScopedTrace objects were being used in place of ATRACE_NAME because of a
misunderstanding of it's function. Cleared up documentation for usage.
Also explicitly use global namespace for sysprop callback.
Change-Id: I7c248b486b614ccdb841659ca0dcfc644fda670a
This change moves the frame time history tracking code out of Layer and into a
new class called FrameTracker. It also changes the tracking to use signal
timestamps from fences when available for more accurate results.
Change-Id: I323c5f075c58bf86ce363b52af885ad0f6365f2b
in this particular case, this OOB is always harmless
(and that's why it didn't get fixed from MR1), however,
it interfers with valgrind debugging.
Change-Id: Ic977e03287e59c4b124a89146c9023bd0cb540a8
The emulator doesn't support systrace, but we should point that out
at most once per process.
Bug 7436352
Change-Id: I06b2c1ea0df6c02c11cd2496423c337f8d7c62a1
This prevents strong reference cycles when the listener implementation also
holds a strong pointer to the ConsumerBase
Bug: 7425644
Change-Id: I1514b13a32b18d421c902dddebec0765a989c55c
This change fixes a number of small glitches that can occur when
multiple components in the same process are updating surfaces.
One would expect that updates to disjoint sets of surfaces would
not collide but this is not the case. The first component to
close the global transaction causes all pending updates to
be applied, including those that another component might not
have finished setting up if it also had an open transaction
at the same time.
Change-Id: I99345958581abbe0e1e325a5bcba37e8941a313a
This patch adds a hashtable-based LRU cache. This should be
significantly higher performance than the GenerationCache it is intended
to replace. It is a large part of the fix for bug 7271109
TextLayoutCache low-level performance issues.
We added a new method to BasicHashtable to detect when rehashing is
needed, because the internal linked list pointers would get invalidated
by that rehashing.
Also, the hash_type specialized to pointers had a small flaw.
Change-Id: I950c2083f96519777b851dbe157100e0a334caec
This change adds support for displays that are not allowed to display surfaces
with the eSecure flag set. All non-virtual displays are considered secure,
while virtual displays have their secure-ness specified at creation time.
Bug: 7368436
Change-Id: I81ad535d2d1e5a7ff78269017e85b111f0098500
This change adds a transaction flag for WindowManager to indicate that a
transaction is being used to animate windows around the screen. SurfaceFlinger
will not allow more than one of these transactions to be outstanding at a time
to prevent the animation "frames" from being dropped.
Bug: 7353840
Change-Id: I6488a6e0e1ed13d27356d2203c9dc766dc6b1759
This change makes BufferQueue::dequeueBuffer release its mutex before
allocating new buffers. This should alleviate lock contention in
SurfaceFlinger where SF's main thread can get blocked waiting for an allocation
operation to complete.
Bug: 7335075
Change-Id: I1b000539cc616a695afab2e9c68507db69e57b13
error codes are returned in errno, this caused ::waitForwever()
to only wait for 1 second and return improper error code (-1).
needed to help debugging 7316632
Bug: 7316632
Change-Id: Ie144f614a88393393972a3a770c6b4b0581f961a
This needs the ConsumerBase mutex locked, but wasn't locking it. Two
of the four places that called it already held the lock so were fine.
Now addReleaseFence() takes the lock itself, and I added
addReleaseFenceLocked() for the two already-locked callers, since in
one of them dropping the lock would be inconvenient.
Bug: 7289269
Change-Id: I7a5628adb516f8eec782aa6c14128202f96d7b0a
The CleanSpec is missing parens, which causes $P to be interpreted as a
single variable, and the following printed to the log:
Clean step: rm -rf RODUCT_OUT/obj/SHARED_LIBRARIES/libgui_intermediates
This patch adds parens as needed.
Change-Id: I587998fa67a4884418c286360a577cdbb6ea9a21
This change adds debug info to SurfaceFlinger's dumpsys to indicate that the
USE_WAIT_SYNC compile option was enabled, and it removes the
ALLOW_DEQUEUE_CURRENT_BUFFER option.
Bug: 7238122
Change-Id: I70e08e34c2ef58aa6d2f88229e781a119f84b5a9
This allows us to blank and unblank displays other than the built-in
display (e.g. HDMI).
Bug: 7240511
Change-Id: I89ea13f9e497be74c3e1231d0c62fb558e93e0f8
This adds a line to the "dumpsys SurfaceFlinger" output that shows
build-time configuration values.
Example:
Build configuration: [sf HAS_CONTEXT_PRIORITY] [libui] \
[libgui USE_FENCE_SYNC]
Bug 7206633
Change-Id: Ibe1856b459d34a4be6ee83a4ebfd2807e6cc68a0
This change fixes an issue causing the mEglContext member of a SurfaceTexture
to get incorrectly zeroed out. This would happen when a call to
ConsumerBase::releaseBufferLocked resulted in the current buffer being freed.
Freeing the current buffer would set SurfaceTexture::mCurrentTexture to -1,
which would then be used by SurfaceTexture::releaseBufferLocked to reset the
current slot's EGLSyncKHR to EGL_NO_SYNC_KHR (= 0). This would overwrite the
mEglContext field, resulting in context mismatch errors in
SurfaceTexture::doGLFenceWaitLocked.
The fix is to simply use the buffer slot that's passed in to
SurfaceTexture::releaseBufferLocked rather than mCurrentTexture.
Change-Id: I0e5e2bd88fcbb354c35a3744f317716fff3e0e41
This change makes updateTexImage default to performing the necessary
synchronization and adds an argument for SurfaceFlinger to disable that
synchronization so that it can be performed lazily.
Change-Id: I7c20923cc786634126fbf7021c9d2541aa77be5d
Bug: 6991805
The Surface createDisplay() call takes a display name for debugging.
This change carries it through SurfaceFlinger and displays it in
the "dumpsys SurfaceFlinger" output.
Bug 7058158
Change-Id: I79f3474a8656ff1beb7b478e0dbf2c5de666118a
The hints were being set a little too late, so the pre-rotation stuff
wasn't quite working.
Bug 7054997
Change-Id: Id8d5c626db7a76f768ba762a145b315878ee08e6
This change adds a call to eglDestroySync after we've dup'd the fd for the
Android fence that the EGLSyncKHR object wraps.
Change-Id: I4fa6ece863260793630d70bb9a69d6284d05d99e
This change adds a compile-option to use eglWaitSyncANDROID to ensure that
texturing operations that access the current buffer of a SurfaceTexture do not
occur until the buffer is completely written. It also moves this
synchronization into a new SurfaceTexture method called doGLFenceWait and
changes SurfaceFlinger's Layer class to use that method rather than performing
its own wait on the fence.
Change-Id: I70afa88086ca7ff49a80e3cd03d423767db7cb88
This change adds support for using Android fences that come from EGLSyncKHR
objects as the release fence for a buffer.
Change-Id: Ice192ce2ec001020f909a2018afdf0f17b24dec9
The ComposerService object wasn't watching for SurfaceFlinger
restarts, which doesn't usually matter because the app framework
restarts when SurfaceFlinger dies. However, mediaserver continues
to run, which means its ComposerService object was trying to use
a dead handle, and playback of DRM movies was failing.
This adds a DeathRecipient listener and some logic to re-establish
the SurfaceFlinger connection.
Bug 6645813
Change-Id: I07581b881d3835601aa57d5358c8259d93bc4515
This change moves some common fence handling code into the base class for
BufferQueue consumer classes. It also makes the ConsumerBase class initialize
a buffer slot's fence with the acquire fence every time a buffer is acquired.
Change-Id: I0bd88bc269e919653b659bfb3ebfb04dd61692a0
We used to keep the bounds of the region as a
separate rectangle. Instead we now store it as the last
element of the Vector<> of Rects.
This has the benefit of being slightly more efficient when
copying regions and reduces the overhead of small regions,
but more importantly will allow us to export the underlaying
SharedBuffer (eventually).
Change-Id: I80790e4fb1a09a747a5616000cfef852ac4ce9e9
since regions are copy-on-write, this prevents to
duplicate the region entirely and offseting by 0,0 is
fairly common.
Change-Id: I9b8c286315a2e00dda01c2456397d72b5e12006b
This change adds some infrastructure for testing the BufferQueue class. It
also includes a test that tests the new check in BufferQueue::acquireBuffer
that prevents the consumer from acquiring more than one buffer beyond the max
acquired buffer count that was set.
Change-Id: I38554ad3f9a53d2ddeba7ef0deee35ec2e2f9775
This change disables the CpuConsumer tests because they require a Gralloc
format that is not supported on all devices.
Change-Id: Ifaa618062c1dae53d9fcb9e16ba92c480d3dbd0c
This change adds an error check to ensure that consumers don't acquire more
buffers than the maximum that they set.
Change-Id: I026643564bde52732e4ee6146972b207ddbbba77
This change makes BufferQueue derive the min undequeued buffer count from a max
acquired buffer count that is set by the consumer. This value may be set at
any time that a producer is not connected to the BufferQueue rather than at
BufferQueue construction time.
Change-Id: Icf9f1d91ec612a079968ba0a4621deffe48f4e22
This change is a clean up of some of the handling of the maximum number of
buffers that are allowed at once. It mostly renames a few member variables and
methods, but it includes a couple small refactorings.
Change-Id: I9959310f563d09583548d4291e1050a7bbc7d87d
This is a compatibility shim for one product whose drivers
are depending on SurfaceComposerClient::getDisplayInfo(
int, DisplayInfo*) when it really shouldn't.
Revert this patch when the problem has been resolved.
Bug: 7065398
Change-Id: I6542691b81fd1b1e1d79500a62e82d40a3d51db7
Use only display tokens in the API to refer to new displays.
Don't require the caller to specify the display when creating
a surface (since in general a surface could be shown on
any display).
This is intended to be a minimum change just to update the API.
Note that SurfaceFlinger still uses DisplayID in a few places
internally that might cause some features not to work properly
when there are multiple displays (LayerScreenshot, for example).
Change-Id: I3d91eec2da406eefd97bcd53655d403ad865a7e6
This change makes SurfaceTexture inherit from ConsumerBase. It removes all of
the functionality from SurfaceTexture that is now provided by the base class.
This includes fixes for two bugs that were found after checking this change in
the first time and then reverting it.
Change-Id: Ie2d9f4f27cfef26fdac341de3152e842b01a58d2
BufferItemConsumer allows for acquiring BufferQueue's BufferItems,
which contain all the data and metadata the BufferQueue has for a
given graphics buffer.
This consumer is useful when direct access to the native buffer_handles
is needed by the client.
Also includes a minor cleanup of CpuConsumer's use of 'virtual'.
Bug: 6243944
Change-Id: If7dc4192b15ac499555f1eda42a85140f2434795
some binaries are using these private APIs and broke
(as they should!) with this change. Temporarily restore the
virtuals to work around this.
Bug: 6977550
Change-Id: I7c37f24b16e4d586b89205c493db5169cf87e024
This change fixes SurfaceTexture::freeBufferLocked so that it calls the base
class implementation.
Change-Id: I45d76fb2eb02c1fa6e4e917823ead83e2086bd15
This change makes SurfaceTexture inherit from ConsumerBase. It removes all of
the functionality from SurfaceTexture that is now provided by the base class.
Change-Id: I4a881df42810a14ee32d4ef7c8772a8f2510f4c7
this change introduces a new class LightFlattenable<> which is
a protocol to flatten simple objects that don't require
binders or file descriptors; the benefit of this protocol is that
it doesn't require the objects to have a virtual table and give us
a consitant way of doing this.
we also introduce an implementation of this protocol for
POD structures, LightFlattenablePod<>.
Parcel has been update to handle this protocol automatically.
Sensor, Rect, Point and Region now use this new protocol.
Change-Id: Icb3ce7fa1d785249eb666f39c2129f2fc143ea4a
- displays are represented by a binder on the client side
- c++ clients can now create and modify displays
Change-Id: I203ea5b4beae0819d742ec5171c27568f4e8354b
- make errors that will always cause a memory corruption always fatal
(for eg: KeyedVector<>::editValue{For|At}() failure)
- make other errors fatal in debug mode, those that can be caught by
the caller.
- fix typos
Change-Id: I65cc7d81035c37ce2906fc4500c50e5d5b5c49eb
This change refactors the FramebufferSurface class to inherit from the new
ConsumerBase class.
Bug: 6620200
Change-Id: I46ec942ddb019658e3c5e79465548b171b2261f2
this attribute can be set through a regular transaction using
SurfaceComposerClient (just like any other attribute, eg: position or size)
Change-Id: I701a47c677ea6442ca713728a93335328cd2b172
no change of functionality -- the old behavior is implemented
on top of this new protocol.
this new protocol will allow, eventually, to pass informations
about displays and layer stacks.
Change-Id: Ic6c2295e61ec8ecbc8ce01ab7664e35d928202fc
There was an issue in Surface::lock where failure to lock a surface
resulted in two bad things happening:
- success was returned to the caller (it was apparently locked).
- an uninitialised pointer was returned as the buffer.
Change-Id: I8b0df81400e0fa0542a8bb993d76923ac96b686e
When updateTexImage acquires a buffer but then aborts (due to an error
or the buffer being rejected), it releases the newly-acquired buffer.
It was passing the buffer slot's fences to releaseBuffer, even though
they hadn't been created after the acquire yet. This wasn't a bug,
since the fences would be cleared just after the buffer slot was last
released, but explicitly passing null fences makes this clearer.
Change-Id: I087f2ec3fd02c40f57782c1fca24eb9567e2943d
both API and implementation will change, this is just a prototype
intended to show feasability.
SurfaceFlinger is passed an ISurfaceTexture through a new
callback, it is in turn used to create an EGLSurface which
surfaceflinger will draw into in addition to the main screen.
Change-Id: Id0bbb0b854bb7bae44d57246a90b65d4567f9a21
Hand merge from ics-aah
> Utils: Fix a bug in the linear transformation code.
>
> Fix a bug where an incorrect result would be computed if you used the
> linear transformation code to do a reverse transformation (from B's
> domain into A's domain) when the scaler fraction was negative.
>
> Change-Id: I8e5f109314d235a177ab41f65d3c4cd08cff78be
> Signed-off-by: John Grossman <johngro@google.com>
Change-Id: Id90e18f685c61c1a89fd91c32adcf01363b3e8f3
Signed-off-by: John Grossman <johngro@google.com>
When the app_process is shutting down the main thread will close the
binder fd while pool threads are executing an ioctl (in
IPCThreadState::stopProcess called by AppRuntime::onStarted in
app_main.c).
The binder driver will then return all pending calls in ioctl
without any error and with a command. One of the threads gets a
BR_SPAWN_LOOPER which will create a new thread (the other thread
gets a BR_NOOP). This new thread then calls
vm->AttachCurrentThread. Usually this results in a log entry with
"AndroidRuntime: NOTE: attach of thread 'Binder Thread #3' failed",
but sometimes it also causes a SIGSEGV. This depends on the timing
between the new thread an the main thread that calls DestroyJavaVM
(in AndroidRuntime::start).
If IPCThreadState.cpp is compiled with "#define LOG_NDEBUG 0" the
pool thread will loop and hit the
ALOG_ASSERT(mProcess->mDriverFD >= 0) in
IPCThreadState::talkWithDriver.
Crashes like this has been seen when running the am command and
other commands that use the app_process.
This fix makes sure that any command that is received when the driver
fd is closed are ignored and IPCThreadState::talkWithDriver instead
returns an error which will cause the pool thread to exit and detach
itself from the vm. A check to avoid calling ioctl to a fd with -1
was also added in IPCThreadState::threadDestructor.
Another solution might be to change the binder driver so that it
returns an error when the fd is closed (or atleast not a
BR_SPAWN_LOOPER command). It might also be possible to call exit(0)
which is done when System.exit(0) is called from java.
Change-Id: I3d1f0ff64896c44be2a5994b3a90f7a06d27f429
