Sympton:
When application crash in a special timing and system server's InputMethodManager is binding, deadlock may occur.
Root Cause:
Thread(1): When using InputMethodManager in system server, it will lock mH(handler) of InputMethodManager,
and sometimes it will call to InputMethodManagerService::showCurrentInputLocked and will call bindService at some condition,
then it will also lock ActivityManagerService.
Thread(2): When an application crashed, it will lock ActivityManager when showing crash dialog, inside the dialog,
it will call setEnabled of Button and lock mH of InputMethodManager.setEnable of TextView will lock IMM's handler.
So the deadlock happened as the flow: (2) lock AMS -> (1)lock mH -> (2)wait mH -> (1) wait AMS
Solution:
Reduce nested lock of error dialog: post message to let (2) lock mH after release AMS lock.
Change-Id: Id85c29406236db3b5fca9655fde1fcaf0afd1337
We should call ResourceTable::getLocales directly,
and not AssetManager::getLocales. The latter will convert
"tl" to "fil" so we'll end up thinking we have resources
for "fil" when we don't really have any.
bug: 15873165
Change-Id: I9753e4608aaecede328a40ee1f3ee6b016d0dedc
Symptom: SkBitmap::getAddr(int, int) may return NULL due to unrecognized config
(ex: kRLE_Index8_Config). This will cause memcmp method to crash. Since bm0 and
bm1 both have pixel data() (have passed NULL == getPixels() check), those 2
bitmaps should be valid (only unrecognized), we return JNI_FALSE to warn user
those 2 unrecognized config bitmaps may be different.
Change-Id: I6970c27de412110a3035d0a783112c4cd3ebc35b
Symptom: Invisible views are still shown in popup window
Root Cause: glClear isn't called if there is no draw command
Solution: Even if there is no draw command, it still calls glClear
by startFrame().
Reproduce steps: App set one view in a popup window to invisible,
it will be still visible.
Change-Id: I36bc7b752434ad388fbb80ea63341ac778f48d35
Symptom: If app applies animation to enlarge a bitmap, there will be an residual
line on the screen
Root Cause: On platform which has Tile Rendering implementation (ex. Qualcomm
CPU), startTiling() call will restrict the framebuffer region which
GPU can affect. So the expansion of clear region by 1 will not take
effect if startTiling region is not expanded.
Solution: Expand the startTiling region by 1, too.
Reproduce steps: Apply animation to enlarge (and then shrink) a bitmap icon.
Change-Id: I7b4a59e180daa29dbe909d9e11f4093ae1d7396f
AudioTrack and AudioRecord JNI should not cast
jshortArray to jbyteArray. This appeared to work with Dalvik
but causes data corruption with ART.
(cherry picked from commit 9d02848e902d04417df616354db3a18c03e639b7)
Change-Id: Ie36624d3ea06042373c64edced4b5e30b7b1ee86
This was broken by commit b27c137087ee92674 which
removed a badly phrased check. Prior to that commit,
there was a check that excluded locales whose toString()
length was not 5.
This change reinstates that check in a less roundabout
way, by excluding BCP-47 tags that don't have a country
subtag.
bug: 16038949
Change-Id: Ic24b1df87b2e2ce47d9ae04c7759088721f27b4f
Movistar Spain needs to add a specific APN to offer tethering
in their devices.
Change-Id: I5554ff98a5ac4c5f33b9af0083e1fac8305c9e13
Signed-off-by: Albert <alberto.crespell@gmail.com>
* Added a new JNI call to pass arrays of Allocations to the RS runtime.
* Added a new version of ForEach that takes an array of Allocations.
* Added some casts to disambiguate existing calls to forEach.
Change-Id: I46d2834c37075b2a2407fd8b010546818a4540d1
Root Cause:
When there is only home activity existed,
updating home apk will call forceStopPackageLocked to finish the activity.
Then activity history becomes empty, then home will be launched,
but its package is still target to close that results a loop.
Solution:
If home activity has been force-stopped, do not stop the same home activity again.
Change-Id: Icff12028d407873c2e6f50a06bcad231b908ccbd
Root Cause:
Use removeTask with flag ActivityManager.REMOVE_TASK_KILL_PROCESS will set waitingToKill = "remove task"
to the target process when its setSchedGroup is not BG_NONINTERACTIVE.
Later the target process may be killed when applying oom-adj due to setSchedGroup has changed to BG_NONINTERACTIVE.
If the process is needed to restart, the process record will be resued.
Then the restarted process may be killed again because its waitingToKill is not null.
Solution:
Clean waitingToKill when process is dead.
Change-Id: I5ffb5388127f4221da4c700d3f1c224f7ca6e7b2
Symptom: Unable to kill by Am again and will be skipped updating its oom adj.
Root Cause:
A restarted process will reuse original process record.
The flag killedByAm will keep previous state.
Solution:
Reset the flag killedByAm to false when the process is started.
Note: Found another similiar patch If95137d91939cc44882ad2813131bcde0edd0c1b
Change-Id: I59a86648ca8d0aed4c489d92751af120aae5ef90
For JB-MR1, there was a hack that used "tl" where
we really meant "fil" because ICU didn't have localizations
for "fil". This has now been fixed, and we now support 3 letter
language codes for AAPT so we can use "fil" where required.
For the benefit of apps that need to target older platforms,
we fall back to "tl" if the app has assets for "tl" and the
resource locale is "fil".
See bugs 7291355, 7207176 and 8049507 for more context.
Change-Id: I1ac8502525f99b40f9091d5efd2df33518d47a41
Prior support forced all third party apps
to be resolved against the default stanza
of the mac_permissions.xml file when assigning
seinfo labels. This meant that all third party
apps, in effect, were untrusted regardless of
cert and therefore received the same selinux domain.
This also had the unfortunate side effect of forcing
certain third party apps into the wrong domains
because of shared userid requests among apps.
This patch removes that restriction and instead
allows all apps, regardless of location, to be
matched against the full mac_permissions.xml
policy file. This then allows all apps signed
with known good certs to receive the same selinux
domains of other apps with whom they share trust.
Change-Id: Iba569c046135c0e81140faf6296c5da26a243037
Signed-off-by: rpcraig <rpcraig@tycho.ncsc.mil>
Locale settings are persisted to system properties
and not to system settings, so the values read here
are used pretty much everywhere.
bug: 15873165
bug: 10090157
Change-Id: Ie7cda166439112c615afbd38253cf4fbb5791242
This code assumed that the locale is always 5 chars in
length and was of the form xx-YY. This is not necessarily
true. The language can be 3 letters in length and the locale
might have a script and variant.
Also cleans up several nasty pieces of code and eliminates
unnecessary array copies and improves readability by using
idiomatic java.
bug: 15873165
bug: 10090157
Change-Id: Iab1cfd7b78e5772b1245654f2153caf63a96033d
