Mir's Android Input implementation has wake-locks stubbed out

[Daniel van Vugt]

Correct use of wake locks is very important to keeping responsiveness high, and the kernel awake when you need it (e.g. processing input events). Unfortunately in Mir we seem to have stubbed out this functionality, which will hurt our input latency:

grep -ir 'wake_lock' 3rd_party/
3rd_party/android-input/android/frameworks/base/services/input/EventHub.cpp:#define acquire_wake_lock(lock, id) {}
3rd_party/android-input/android/frameworks/base/services/input/EventHub.cpp:#define release_wake_lock(id) {}

I can understand why -- because wake locks in the form Android has them do not exist outside of Android kernels. Mainline kernels use something different. But using nothing at all is bad for performance.

[Daniel van Vugt (vanvugt)]

The implementation on Android seems pretty simple:

https://android.googlesource.com/platform/hardware/libhardware_legacy/+/master/power/power.c

Exported via libhardware?

[Daniel van Vugt (vanvugt)]

I vaguely recall our client-side input logic needed fixing for this too. We were signalling from the client library that input event handling was completed before it even began. I assume we need to fix that too, to get the duration of the wake lock right in the server.

[Daniel van Vugt (vanvugt)]

Similarly: bug 1388490

[Daniel van Vugt (vanvugt)]

Attached a branch that attempts to solve the problem by re-introducing wake locks. Not seeing any measurable benefit from it yet though.

[Cemil Azizoglu (cemil-azizoglu)]

I'm guessing we stubbed them out because there is another mechanism to keep CPU/display awake. What does tvoss say about this?

[PS Jenkins bot (ps-jenkins)]

Fix committed into lp:mir at revision None, scheduled for release in mir, milestone 0.14.0

[Alan Griffiths (alan-griffiths)]

I don't understand why anyone thinks this valid. We've taken some code from a context in which wakelocks (presumably) made sense and used it in a context where they don't. That's a feature not a bug.

[Alberto Aguirre (albaguirre)]

There is absolutely no need for MIR to acquire wakelocks ever. Wakelocks are a suspend prevention mechanism which in Ubuntu Touch is taken care of by powerd.

The user is already interacting with the screen in any case so the system will not be suspending during interaction.

In android the EventHub is a background service hence the wakelocks.

[Daniel van Vugt (vanvugt)]

Thanks. In future please keep an eye out for bug reports so that you can make comments in the bug before fixes are proposed.