Lock ordering bug. The fix is to avoid having to hold both locks
simultaneously in one prong of the deadlock. (Further strengthening:
the code path for potentially acquiring both simultaneously now
begins at a common method, ensuring that lock ordering is always
consistent.)
Change-Id: If9433aff8326fec2ba0e97e96ad2d835b01e2cda
Re-add change 103769 (Change-Id: I9e7c94718a5b1f08840b219b304ba3904259a65f) from gingerbread
after the bad merge was reverted by master change 104487
(Change-Id: I8fad7ee11f4b26630f3cc4b31cc9cafbe66b3bf4).
bug:4157610
Change-Id: I96a75d71220cc7886a5b186b415d8dacf6951ca6
The idea is to assist with debugging by identifying cases in which
the input event stream is corrupted.
Change-Id: I0a00e52bbe2716be1b3dfc7c02a754492d8e7f1f
Revert "am 047f9d12: Merge "resolved conflicts for merge of ce6766ea to honeycomb-plus-aosp" into honeycomb-plus-aosp"
This reverts commit 352f2c33d3b8a0b73dcdab98c7a77f698578b048, reversing
changes made to ce8a55db6feae11c1db45a868a7d8a0e7d105471.
This patch adds a mechanism for capturing, filtering, transforming
and injecting input events at a very low level before the input
dispatcher attempts to deliver them to applications. At this time,
the mechanism is only intended to be used by the accessibility
system to implement built-in system-level accessibility affordances.
The accessibility input filter is currently just a stub.
It logs the input events receives and reinjects them unchanged,
except that it transforms KEYCODE_Q into KEYCODE_Z.
Currently, the accessibility input filter is installed whenever
accessibility is enabled. We'll probably want to change that
so it only enables the input filter when a screen reader is
installed and we want touch exploration.
Change-Id: I35764fdf75522b69d09ebd78c9766eb7593c1afe
Handle the case where the kernel driver is in accessory mode but we failed
to initialize it at the framework level. On disconnnect, check to see if the
accessory kernel driver is enabled rather than checking mCurrentAccessory.
That way we will restore the USB state in the kernel even if mCurrentAccessory
is null.
Change-Id: I35d458f21a8b21611946da523d0f53723cab0540
Signed-off-by: Mike Lockwood <lockwood@android.com>
Two issues.
1) remove default routes for non-default networks.
2) don't report mobile is the active default network just because
it is active.
bug:4157610
Change-Id: I9e7c94718a5b1f08840b219b304ba3904259a65f
When creating new vibration pattern, a callback is registered
in case binding to caller goes away, by linkToDeath().
Need to unregister this callback when we throw away the vibration.
Change-Id: Ibdf0bd415a539054ac7a66f49b33a864f729c546
Split the dump() command between PMS and Settings.
Try to annotate all users of the mPackages lock in preparation for
switch to reader/writer lock.
Also mark some locals final as I was reading through the usage of the
synchronization on mPackages.
Change-Id: Ia7add63e835e67156edf886f98acebe50cc41f70
If you deleted the host routes (started a secondary network like mms, supl
of hipri and then ended it) you would lose the host route to the default
gateway. Then if you needed to re-add the default gateway route (lost
the connection and removed the default route and then re-established)
you couldn't - can't add a gateway that isn't routable apparently.
This happens if you are in a video chat and lose your connection without
losing the interface (PPP keeps it up for a bit).
Fixed it by having addDefaultRoute first add a hsot route for the gateway
before adding the default route. This allows the default add to succeed.
bug:3490353
Change-Id: I415e7319832e6456f8757b14c4f79f098a08839b
Split PackageManagerService from one monolithic class with several inner
classes to several classes. This will help determining how its data
structures can be reshuffled to provide better separation of concerns.
Change-Id: Ic7571daebdcf13ce08e08f34204c5bbf4140139b
Removing some unused methods, adding generics where possible to avoid
future mistakes, and just more cleanups of warnings.
Change-Id: Ie62206f87dc19544b9dc039faa0159fc46ebc259
