The main problem here was in the error recovery when we are waiting
for a process to start but it has failed for some reason. The code
was just setting mPendingBroadcast to null, but this would cause
an eventual ANR because the state was not set back to IDLE so we
would continue waiting for the broadcast without trying to restart
its process.
Now we set it to idle. We also need to reset the "nextReceiver"
index, so there is a new mPendingBroadcastRecvIndex variable holding
what it should be set back to.
While digging into this, I found a number of other lesser problems:
- There is a race when booting the system where we set mSystemReady
to true before restarting the upgrade processes. This could allow
a broadcast to happen between those two and its process to immediately
be removed. To fix this, there is a new mProcessesReady that is set
once we are truly ready to start launching processes.
- There were various places where we were calling sendBroadcastLocked()
without the flag to send only to receivers... if this is called before
mProcessesReady is set, then we would end up sticking any process for
the broadcast on the holding list to not get launched until later
(and hang up all broadcasts as they want for it). Now we always make
sure to set this appropriately.
- sendBroadcastInPackage() was not doing all of the validation that
sendBroadcast() does.
And of course a bunch of new debugging logs that were done in the
course of tracking this down.
Change-Id: I6134bbd94fdb73db8b693507b29499eae012d543
- New API for iterating over history that will allow a better implementation
in the future.
- Now do writes asynchronously.
Also improve the documentation for Activity.onRetainNonInstanceState().
Change-Id: Idf67f2796a8868eb62f288bcbb2bad29876c8554
We now distribute "wifi started" time across all apps that are
holding WIFI locks that cause it to be started. But only when
WIFI would not normally be running. Also have a mechanism to
distribute other WIFI work that has happened across those processes
based on their use.
Also fixed a bug where we were not retaining the CPU speed step
stats across boots...!
Change-Id: I00e3153b98429166273750512cc37e7975211ab9
...that Market took over the screen after signing into a Google account
Don't warn the user about a new activity coming up on an activity that
is finishing.
Change-Id: I573073139d42a485473d0c8a7df450c1a23c35c3
This fixes a problem where applications could ask the location
manager to do very heavy-weight things (like... say... update
location every minute), which would get accounted against the
system instead of the application because ultimately it is the
system making the heavy calls (wake locks, etc).
To solve this, we introduce a new class WorkSource representing
the source of some work. Wake locks and Wifi locks allow you
to set the source to use (but only if you are system code and thus
can get the permission to do so), which is what will be reported
to the battery stats until the actual caller.
For the initial implementation, the location manager keeps track
of all clients requesting periodic updates, and tells its providers
about them as a WorkSource param when setting their min update time.
The network location provider uses this to set the source on the
wake and wifi locks it acquires, when doing work because of the
update period.
This should also be used elsewhere, such as in the GPS provider,
but this is a good start.
Change-Id: I2b6ffafad9e90ecf15d7c502e2db675fd52ae3cf
We weren't logging strictmode violation in the system_server process
in non-user builds (only system apps), even though the rest of the
strictmode logging supports it.
Also add a missing lock in ActivityManagerService.
Change-Id: If2af96a7e4fdde604a647b836097f0029ef1334b
The activity manager looks for cases where one app launches immediately
after another. If this happens, a brief toast is shown telling the user
when app is actually running and what was originally starting.
Change-Id: If94cf5bd393dd0bc0f09789dae044fde1386c481
For the duration of the wake lock, 50% of all CPU usage is now
accounted against the app(s) holding partial wake locks, evenly
distributed between them. This is only while the device is on
battery and screen off.
Change-Id: I3e5c978b792b6ef17bf8540705bfe8343dadd464
- Collect data at better times.
- Collect per-thread CPU usage as soon as possible after the ANR, and print
in log.
- Based on new per-thread CPU usage, limit the number of processes we
collect stacks from to not include inactive not interesting procs.
- Improve the way ProcessStats compute and reports its data.
Change-Id: I12b17fb47d593d175be69bb792c1f57179bf4fdf
There was a flaw in the service management, when the same activity
is doing a bindService() for the same service IBinder. In this case
the activity would correctly keep a list of all generated connections,
however some other data structures would assume there is only one
connection per IBinder, and thus only remember the last.
When that last connection was unbound, the service would be destroyed
since it thought there were no more connections. Then when the
activity was finished, it would try to destroy the service again and
end up with an ANR because the service was already gone and would
not respond.
Change-Id: I59bde38bc24e78147b90b0a7cd525c2a1d20489f
It would grant the permission to the temporary ActivityRecord,
not the real one, so it never got cleaned up.
Also allow granting of permissions to services because... well,
it would be really really useful. And it introduces some
refactoring that we'll need to support cut/paste.
Change-Id: If521f509042e7baad7f5dc9bec84b6ba0d90ba09
- Now track wake locks in battery history.
- Now track sensors in battery history.
- Some filtering of sensory data.
- Fixes to some data that wasn't cleared when resetting battery stats.
- Print amount discharged since last charge.
And the big part -- keep track of wake locks held per process,
and kill processes that hold wake locks too much while they are in
the background. This includes information in the battery stats
about the process being killed, which will be available to the
developer if the app is reported.
Change-Id: I97202e94d00aafe0526ba2db74a03212e7539c54
The NotificationManager tries to crash the calling app, but
in the case of a service calling startForeground, the caller
is the ActivityManager, so system_server goes down.
NotificationManagerService#enqueueNotificationInternal is a
new internal-only method that accepts a UID/PID to use when
punishing bogus notifications (such as the one in
http://b/2869787).
Change-Id: I84a9854bae630bc90288cebb94f174809d5dac8c
I was seeing lots of stack traces of people hung for noticeable
amounts of time when switching between activities.
e.g. On of the common gmail stacks showing this pause was:
android.os.StrictMode$StrictModeDiskWriteViolation: policy=391 violation=1
at android.os.StrictMode$AndroidBlockGuardPolicy.startHandlingViolationException(StrictMode.java:272)
at android.os.StrictMode$AndroidBlockGuardPolicy.onWriteToDisk(StrictMode.java:243)
at dalvik.system.BlockGuard$WrappedFileSystem.open(BlockGuard.java:238)
at java.io.FileOutputStream.<init>(FileOutputStream.java:97)
at java.io.FileOutputStream.<init>(FileOutputStream.java:69)
at com.android.server.am.UsageStatsService.writeStatsFLOCK(UsageStatsService.java:424)
at com.android.server.am.UsageStatsService.writeStatsToFile(UsageStatsService.java:398)
at com.android.server.am.UsageStatsService.notePauseComponent(UsageStatsService.java:539)
at com.android.server.am.ActivityManagerService.updateUsageStats(ActivityManagerService.java:1856)
at com.android.server.am.ActivityStack.startPausingLocked(ActivityStack.java:667)
at com.android.server.am.ActivityStack.finishActivityLocked(ActivityStack.java:2925)
at com.android.server.am.ActivityStack.requestFinishActivityLocked(ActivityStack.java:2836)
at com.android.server.am.ActivityManagerService.finishActivity(ActivityManagerService.java:2276)
at android.app.ActivityManagerNative.onTransact(ActivityManagerNative.java:237)
at com.android.server.am.ActivityManagerService.onTransact(ActivityManagerService.java:1415)
at android.os.Binder.execTransact(Binder.java:320)
at dalvik.system.NativeStart.run(Native Method)
at android.app.ActivityManagerProxy.finishActivity(ActivityManagerNative.java:1454)
at android.app.Activity.finish(Activity.java:3260)
at android.app.Activity.onBackPressed(Activity.java:1929)
at android.app.Activity.onKeyUp(Activity.java:1907)
at android.view.KeyEvent.dispatch(KeyEvent.java:1088)
at android.app.Activity.dispatchKeyEvent(Activity.java:2087)
at com.android.internal.policy.impl.PhoneWindow$DecorView.dispatchKeyEvent(PhoneWindow.java:1661)
at android.view.ViewRoot.deliverKeyEventToViewHierarchy(ViewRoot.java:2543)
at android.view.ViewRoot.handleFinishedEvent(ViewRoot.java:2516)
at android.view.ViewRoot.handleMessage(ViewRoot.java:1866)
at android.os.Handler.dispatchMessage(Handler.java:99)
at android.os.Looper.loop(Looper.java:123)
at android.app.ActivityThread.main(ActivityThread.java:3609)
at java.lang.reflect.Method.invokeNative(Native Method)
at java.lang.reflect.Method.invoke(Method.java:521)
at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run(ZygoteInit.java:868)
at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:626)
at dalvik.system.NativeStart.main(Native Method)
Change-Id: Id49157bc635017292eaefddc5e22d73f5f4ab05e
The guard is compiled out by default because it adds overhead to
android.os.Process.setPriority().
Change-Id: Ibb2a648c6349b381abb7ae62a358888b04fba871
This was originally written as an in-case-we-need-it facility, but was
never actually used in production. It also soaked up a surprising amount
of cpu on occasion, as well as doing sketchy things like demoting the
system_server's primary looper thread to the background cgroup at times.
Change-Id: I9a81a8d1e9caea9e0a1277d97785fe96add438d7
Introdude a new ActivityStack class that holds all of the
state and management of a stack of activities. Paves the way
for having multiple activity stacks, though at this point
there should be no change in functionality and the activity
manager is still assuming there is only one stack.
Change-Id: Iea4859a24c9269061043755ec58a615028d4183b
Implement notification manager handling of bad notifications, to
call a new activity manager to have the owner's process crashed
(if there is one).
Change-Id: Ib15e8d0c598756f3b39c99cc2045c18e054daf6b
- Move PackageInfo out of ActivityThread, renaming to LoadedApk.
- Rename some of the other PacakgeInfo inner classes to better
represent what they are.
- Rename HistoryRecord to ActivityRecord.
- Introduce AppGlobals, to eventually let ActivityThread become
package scoped.
Change-Id: Ib714c54ceb3cdbb525dce3db9505f31042e88cf0
An Activity can declare itself to be "immersive" either by
setting android:immersive="true" in AndroidManifest or by
calling setImmersive(true).
Immersive activities "should" not be interrupted, for
example by Notifications with an associated
fullScreenIntent. (In the future we may even prevent any
non-system application from successfully calling
startActivity() if the foreground activity is immersive.)
Notifications with FLAG_HIGH_PRIORITY set will be shown to
the user in some less-obtrusive way if the frontmost
activity is immersive.
Change-Id: I8d0c25cc4e22371c27cbf2bb6372d2c95d57b2d7
We now clear the battery stats when unplugging after the
battery is full. This allows us to use the "total" stats as
a new "since last charged" stat. Total is gone. I never used
it, it was worthless. Since last charged is a lot more
interesting.
The battery history now collects a lot more stats, and keeps
control over how much it can collect. Printing is now more
descriptive.
The kinds of stats have been renamed to SINCE_UNPLUGGED and
SINCE_DISCHARGED. The other two stats are still there, but
no longer printed; a future change will eliminate them
completely along with all of their state.
Change-Id: I4e9fcfcf8c30510092c76a8594f6021e9502fbc1
Modify OOM adj classes a bit, to take into account the new
heavy weight app type, and give "foreground services" their
own category to have a bettery chance to manager them when
things go wrong.
Also add some new code to battery stats to keep a history
of changes to the battery level.
Change-Id: I29f5ab6938777e1a7eafd7d8c38b5e564cc9f96a
This is a new public API for developers to opt-in to strict rules
about what they're allowed to do on certain threads. (this is the
public face of the @hide dalvik.system.BlockGuard, added recently...)
In practice this will be used for developers to opt-in to declaring
that they don't want to be allowed to do various operations (such as
disk I/O or network operations) on their main UI threads. (these
operations are often accidental, or even when they are fast come with
a good chance of being slow or very slow in some cases....)
Implementation wise, this is just a thread-local integer that has a
bitmask of the things that aren't allowed, and more bits for saying
what the violation penalty is. The penalties, of which multiple can
be chosen, include:
* logging
* dropbox uploading for analysis/reporting
* annoying dialog
* full-on crashing
These are all only very roughly implemented at this point, but all
parts now minimally work end-to-end now, so this is a good checkpoint
commit before this gets too large.
Future CLs will polish all the above 4 penalties, including
checksumming of stacktraces and minimizing penalties for duplicate
violations.
Change-Id: Icbe61a2e950119519e7364030b10c3c28d243abe
Merge commit 'ac24d23cd4a96f38b4e9cb0318a7c298794b9b6a' into kraken
* commit 'ac24d23cd4a96f38b4e9cb0318a7c298794b9b6a':
Don't bring up Launcher until after boot complete
The preboot upgrade handling was bringing up the acore process with a default
application object, then the normal "start the HOME app" code was bringing up
Launcher2 [hosted in acore] in anticipation of boot completion... but then it
saw that the host process was alive and continued with Launcher2's init.
Launcher2 depends on a custom application object, however, so it crashed
immediately.
This change ensures that the HOME app is not actually initted at that level
until after boot has completed, at which point its proper application class
can be instantiated.
Fixes bug #2732250
Change-Id: I1a15384e2c0d50e14300df0c0db236bd7b1a187c
Only one can be running at a time, their process can not be killed,
and a notification is posted while it is running.
Change-Id: I843015723947e0c934ae63a1aeee139327c0bc01
This gives us a snapshot of what the system process was doing after 30 seconds
of apparent inactivity as well as after 1 minute, to help distinguishing actual
deadlocks from too-slow progress, livelock, etc.
Change-Id: I19758861d1b25f298e88788e8f1c7ec7bf828823
They're designed to handle one time upgrade code, which may
take a significant amount of time. Third party code cannot
receive this broadcast so the timing should be able to be
kept under control.
Bug: 2713849
Change-Id: I3733b6530198bebd9b8c2ba6f9f27a7999336b8d
