This change fixes several issues where events would be dropped in the
input dispatch pipeline in such a way that the dispatcher could not
accurately track the state of the input device.
Given more robust tracking, we can now also provide robust cancelation
of input events in cases where an application might otherwise become
out of sync with the event stream due to ANR, app switch, policy decisions,
or forced focus transitions.
Pruned some of the input dispatcher log output.
Moved the responsibility for calling intercept*BeforeQueueing into
the input dispatcher instead of the input reader and added support for
early interception of injected events for events coming from trusted
sources. This enables behaviors like injection of media keys while
the screen is off, haptic feedback of injected virtual keys, so injected
events become more "first class" in a way.
Change-Id: Iec6ff1dd21e5f3c7feb80ea4feb5382bd090dbd9
Added a couple of micro-optimizations to avoid calling wake() unnecessarily
and reduce JNI overhead slightly.
Fixed a minor issue where we were not clearing the "next" field of Messages
returned by the MessageQueue so the Message would hold on to its successor
and potentially prevent the GC from collecting it if the message were leaked
somehow.
Change-Id: I488d29417ce0cdd7d0e447cda76ec978ef7f811c
There is a new ANativeWindow::cancelBuffer() API that can be used to
cancel any dequeued buffer, BEFORE it's been enqueued. The buffer is
returned to the list of availlable buffers. dequeue and cancel are not
mutually thread safe, they must be called from the same thread or
external synchronization must be used.
Change-Id: I86cc7985bace8b6a93ad2c75d2bef5c3c2cb4d61
the core screenshot function now can capture the screen at any lower resolution
performing bilinear filtering.
we also now have some client code to interface with the screenshot service.
it's now possible to request a screenshot at a lower resolution.
Change-Id: I33689bba98507ab928d0898b21596d0d2fe4b953
AssetManager instances are created by zygote and passed to all its
children so that they don't have to individually open
frameworks-res.apk. This creates a problem for determining the current
file offset when using lseek() on those files, because you can't
guarantee the cross-process locking of a mutex. Luckily, Linux
implements pread() to get around this suckiness.
The problem is that only Linux implements this, so we have to keep the
old locking for use on host builds with aapt and friends. aapt doesn't
have this same problem of sharing file descriptors across forked
processes, so we can keep the local AutoMutex to protect accesses of
those files.
Change-Id: Ibe9f11499a53fe345f50fbaea438815ec0fd363e
This change narrows the opportunity for a race condition setting the
resource Configuration while devices are being updated.
Change-Id: I58efa563f4129ab0fce7108511d16a99dff7e451
There is apparently still a race upon reading the entry Local File
Header that can't be tracked down, so move the LFH check inside the
mutex-protected block so we can call lseek again to see where we are
when we log an error.
Also, close() can fail so use TEMP_FAILURE_RETRY on it so we don't
unwittingly leak file descriptors when Mean Mr. EINTR comes a-knocking.
Change-Id: I753abad0bd882fe28f7281c406fa76f64393ef4c
Added dumpsys reporting to EventHub.
Made the formatting a bit clearer.
Added 'Locked' suffix to some internal methods of EventHub.
Change-Id: Ic449560bcce378f6361895d27c66854e9724abb0
This feature is currently used to enable dragging the start and end
selection handles of a TextView at the same time. Could be used for
other things later.
Deleted some dead code in ArrowKeyMovementMethod and CursorControllers.
Change-Id: I930accd97ca1ca1917aab8a807db2c950fc7b409
screenshots are taken using ISurfaceComposer::captureScreen() which returns
the size of the screenshot and an IMemoryHeap containing the data.
screenshots have limitations:
- they will always fail if a secure window is up on screen
- require GL_OES_framebuffer_object extension
- in some situation, video planes won't been captured
Change-Id: I741c68a2d2984fb139039301c3349e6780e2cd58
Since we switched to seeking to the LFH to verify its existence instead
of a huge mmap of the file, we have to guarantee that another seek
doesn't happen before we finish our read on the LFH.
Change-Id: If8135d9cb6f2f5cc4db734eafa4f6b5f6269c62a
Also fix a Valgrind complaint by zeroing out the entire epoll event
struct since otherwise the data field union would be partly
uninitialized (but not in a harmful way).
Change-Id: I2091ce517e87fcad7c9caf90e2c5e4854a7ca465
The idea is that if you're writing code that wants fd/events/data on
return from pollOnce() / pollAll() you should really pass in all of those
arguments. When I changed the Looper API earlier, it was difficult
to ensure that all callers were passing the right parameters since they
were relying on default parameters to some degree so usage mistakes
would not have been caught by the compiler.
Change-Id: I1f2812894270aaf1515017ac1616b6b312d9b565
Redesigned the input dispatcher's ANR timeout mechanism so it is much
closer to Froyo's policy. ANR is only ever signalled if the dispatcher
is waiting on a window to finish processing its previous event(s) and
there is new pending input.
In the old code, we tracked the dispatch timeout separately for each
input channel. This was somewhat complicated and also resulted in the
situation where applications could ANR long after the user had pushed
them into the background.
Change-Id: I666ecada0952d4b95f1d67b9f733842b745c7f4b
As part of this change, consolidated and cleaned up the Looper API so
that there are fewer distinctions between the NDK and non-NDK declarations
(no need for two callback types, etc.).
Removed the dependence on specific constants from sys/poll.h such as
POLLIN. Instead looper.h defines events like LOOPER_EVENT_INPUT for
the events that it supports. That should help make any future
under-the-hood implementation changes easier.
Fixed a couple of compiler warnings along the way.
Change-Id: I449a7ec780bf061bdd325452f823673e2b39b6ae
There are 16 events logged in the event log:
SF_APP_DEQUEUE_BEFORE
SF_APP_DEQUEUE_AFTER
SF_APP_LOCK_BEFORE
SF_APP_LOCK_AFTER
SF_APP_QUEUE
SF_REPAINT
SF_COMPOSITION_COMPLETE
SF_UNLOCK_CLIENTS
SF_SWAP_BUFFERS
SF_REPAINT_DONE
SF_FB_POST_BEFORE
SF_FB_POST_AFTER
SF_FB_DEQUEUE_BEFORE
SF_FB_DEQUEUE_AFTER
SF_FB_LOCK_BEFORE
SF_FB_LOCK_AFTER
all events log the buffer conserned and a timestamp in microseconds.
by default the logging is not enabled, to turn it on:
adb shell service call SurfaceFlinger 1006 i31 1
adb shell setprop debug.graphic_log 1
The effect is immediate in SurfaceFlinger, but applications need to be
restarted.
Change-Id: Ifc2e31f7aed072d9a7dede20ff2ce59231edbec1
This change is essentially a rewrite of the main input dispatcher loop
with the target identification folded in. Since the input dispatcher now
has all of the window state, it can make better decisions about
when to ANR.
Added a .5 second deadline for processing app switch keys. This behavior
predates Gingerbread but had not previously been ported.
Fixed some timing inaccuracies in the ANR accounting that could cause
applications to ANR sooner than they should have.
Added a mechanism for tracking key and motion events that have been
dispatched to a window so that appropriate cancelation events can be
synthesized when recovering from ANR. This change helps to keep
applications in sync so they don't end up with stuck buttons upon
recovery from ANRs.
Added more comments to describe the tricky parts of PollLoop.
Change-Id: I13dffca27acb436fc383980db536abc4d8b9e6f1
Added the MotionEvent.FLAG_WINDOW_IS_OBSCURED flag which is set by the
input manager whenever another visible window is partly or wholly obscured
the target of a touch event so that applications can filter touches
accordingly.
Added a "filterTouchesWhenObscured" attribute to View which can be used to
enable filtering of touches when the view's window is obscured.
Change-Id: I936d9c85013fd2d77fb296a600528d30a29027d2
The ALooper API now uses an explicit "identifier" for the integer
that is returned rather than implicitly using the fd. This allows
the APIs that had the fd to be a little more sane.
Change-Id: I8507f535ad484c0bdc4a1bd016d87bb09acd7ff0
This was causing stack stitching problems where a one-way call with
violations followed by a two-way call without violations was getting
the previous one-way call's violation stack stitched on to the second
caller's stack.
The solution is a little more indirect than I would've liked
(preserving the binder's onTransact flags until enforceInterface) but
was seemingly necessary to work without changing the AIDL compiler.
It should also be sufficiently cheap, since no new calls to
thread-local IPCThreadState lookups were required. The additional
work is just same-thread getter/setters on the existing
IPCThreadState.
Change-Id: I4b6db1d445c56e868e6d0d7be3ba6849f4ef23ae
Finished the input device capability API.
Added a mechanism for calibrating touch devices to obtain more
accurate information about the touch contact area.
Improved pointer location to show new coordinates and capabilities.
Optimized pointer location display and formatting to avoid allocating large
numbers of temporary objects. The GC churn was causing the application to
stutter very badly when more than a couple of fingers were down).
Added more diagnostics.
Change-Id: Ie25380278ed6f16c5b04cd9df848015850383498
On single-touch devices, pointer up/down is signalled by a BTN_TOUCH
key event. Previously we handled BTN_TOUCH immediately but some drivers
may produce the sequence BTN_TOUCH, ABS_X, ABS_Y, SYN_REPORT on pointer down
which caused us to emit a bad initial pointer down location.
Now we wait for SYN_REPORT before reporting the up or down.
On multi-touch devices, pointer up can be signalled by as little as
the sequence SYN_MT_REPORT, SYN_REPORT. This change ensures that we
handle this case.
Added support for reading ABS_MT_PRESSURE when available.
Corrected mapping of touchMajor/touchMinor on single touch devices.
Minor code cleanup.
Change-Id: Ic7ec4811241ed85a06e59b8a839ca05180d491d4
Sometimes the wrong fd was accessed when the device was addressed
by device id.
The earlier implementation assumed that two arrays were in sync
but one of them was compacted when devices were removed. Instead
of that dependency the device now keeps track of it's file descriptor.
Change-Id: Ib0f320603aafb07ded354bc3687de9759c9068f2
* Add flags field in OBB footer to support overlays.
* Remove unused 'crypto' and 'filesystem' fields in obbtool (could
later be supported in the "flags" field of the OBB footer).
* Add notes to document OBB classes before shipping.
Change-Id: I386b43c32c5edef55210acb5d3322639c08010ba
Previously, the input dispatcher assumed that the input channel's
receive pipe file descriptor was a sufficiently unique identifier for
looking up input channels in its various tables. However, it can happen
that an input channel is disposed and then a new input channel is
immediately created that reuses the same file descriptor. Ordinarily
this is not a problem, however there is a small opportunity for a race
to arise in InputQueue.
When InputQueue receives an input event from the dispatcher, it
generates a finishedToken that encodes the channel's receive pipe fd,
and a sequence number. The finishedToken is used by the ViewRoot
as a handle for the event so that it can tell the InputQueue when
the event has finished being processed.
Here is the race:
1. InputQueue receives an input event, assigns a new finishedToken.
2. ViewRoot begins processing the input event.
3. During processing, ViewRoot unregisters the InputChannel.
4. A new InputChannel is created and is registered with the Input Queue.
This InputChannel happens to have the same receive pipe fd as
the one previously registered.
5. ViewRoot tells the InputQueue that it has finished processing the
input event, passing along the original finishedToken.
6. InputQueue throws an exception because the finishedToken's receive
pipe fd is registered but the sequence number is incorrect so it
assumes that the client has called finish spuriously.
The fix is to include a unique connection id within the finishedToken so
that the InputQueue can accurately confirm that the token belongs to
the currently registered InputChannel rather than to an old one that
happened to have the same receive pipe fd. When it notices this, it
ignores the spurious finish.
I've also made a couple of other small changes to avoid similar races
elsewhere.
This patch set also includes a fix to synthesize a finished signal
when the input channel is unregistered on the client side to
help keep the server and client in sync.
Bug: 2834068
Change-Id: I1de34a36249ab74c359c2c67a57e333543400f7b
This change involves adding a new method to IWindowManager,
monitorInput() that returns an InputChannel to receive a copy of all
input that is dispatched to applications. The caller must have
the READ_INPUT_STATE permission to make this request (similar to
other window manager methods such as getKeycodeState).
Change-Id: Icd14d810174a5b2928671ef16de73af88302aea0
Add "obbtool" host command for adding, removing, and querying Opaque
Binary Blob (OBB) information from a file.
Change-Id: Id2ac41e687ad2a500c362616d6738a8ae7e8f5c3
