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
This is (intendend to be) a no-op change.
At this stage, Binder RPCs just have an additional uint32 passed around
in the header, right before the interface name. But nothing is actually
done with them yet. That value should right now always be 0.
This now boots and seems to work.
Change-Id: I135b7c84f07575e6b9717fef2424d301a450df7b
On binder incalls, the handler thread is given the caller's priority by the
driver, but not the caller's cgroup. We have explicit code that sets the
handler's cgroup to match the caller's, *except* that the system process
explicitly disables this behavior. This led to a siuation in which we were
running binder incalls to the system process at nice=10 but cgroup=fg.
That's fine as far as it goes, except that if a GC happened in the handler
thread, it would be promoted to foreground priority and cgroup both, to avoid
having the GC take forever. Then, when GC finished, the original priority
is reset, and the cgroup set *based on that priority*. This would push the
handler thread into nice=10 cgroup=bg_non_interactive -- which matches the
caller, but is supposed to be impossible in the system process.
The end result of this was that we could be running "lengthy" operations in
the system process in the background. Unfortunately, some of the operations
that wound up like this would hold important global system locks for up to
twenty seconds as a result, making the entire device unresponsive to input
for that period.
This CL fixes the binder incall setup to ensure that within the system process,
a binder incall is always begun from the normal foreground priority as well
as cgroup. In practice now the device still becomes laggy/sluggish when the
offending lock-holding time-consuming incall occurs, but since it still runs
as a foreground task it is able to proceed to completion within a short time
rather than taking 20 seconds.
Fixes bug #2403717
Change-Id: Id046aeabd0e80c48eef94accc37842835eab308d
This is a very simply implementation: upon receiving an IPC, if the handling
thread is at a background priority (the driver will have taken care of
propagating this from the calling thread), then stick it in to the background
scheduling group. Plus an API to turn this off for the process, which is
used by the system process.
This also pulls some of the code for managing scheduling classes out of
the Process JNI wrappers and in to some convenience methods in thread.h.
To prevent buggy command implementations from poisoning binder threads'
scheduling class & priority for future command execution, we now reset the
cgroup and thread priority to foreground/normal when a binder service thread
finishes executing the designated command.
Change-Id: Ibc0ab2485751453f6dc96fdb4eb877fd02796e3f
