As we move away from Android wake locks in JB the legacy /sys/power/wait_for_fb_{sleep,wake} have been removed [1]. Luckly it seems that we don't need this on JB at all as the hw_composer's blank method [2] does not return until the operation is complete. This hopefully will work even for JB ports that have yet to remove this legacy kernel code. [1] https://www.codeaurora.org/cgit/quic/la/kernel/msm/commit/?id=f85607a715a74c65db812cd3901022888257f966 [2] https://www.codeaurora.org/cgit/quic/la/platform/hardware/libhardware/tree/include/hardware/hwcomposer.h?h=b2g/jb_mr1_rb2.17#n510

This seems to work for both my ICS and JB devices, with a hack to widget/gonk/libdisplay/GonkDisplay.cpp not supplied here (see the #error in the patch). It would be convenient to continue to use mozilla::ReadSysFile [1], however this is not exported by libxul ATM it seems. But we could also potentially revert back to the b2g18 method as well [2]. [1] http://mxr.mozilla.org/mozilla-central/source/xpcom/glue/FileUtils.h#178 [2] http://mxr.mozilla.org/mozilla-b2g18/source/widget/gonk/nsWindow.cpp#121

Comment on attachment 768155 [details] [diff] [review] WIP Review of attachment 768155 [details] [diff] [review]: ----------------------------------------------------------------- Overall approach here looks good to me. ::: widget/gonk/libdisplay/GonkDisplayICS.cpp @@ +47,5 @@ > +#error Can't link with ReadSysFile() from here... > + while (true) { > + // Cannot use epoll here because kSleepFile and kWakeFile are > + // always ready to read and blocking. > + ret = mozilla::ReadSysFile(kSleepFile, &buf, sizeof(buf)); IIRC we did this manually before instead of using ReadSysFile. I wouldn't mind switching back to checking this file manually.

BTW, is wait_for_fb_* removed on your port, or still there?

It's gone in our code base entirely. Does this patch work ok on your Nexus 4?

Appears to work. The screen seems to turn on faster too.

Comment on attachment 768635 [details] [diff] [review] v1 Review of attachment 768635 [details] [diff] [review]: ----------------------------------------------------------------- Mostly nits. Looks good overall - thanks. ::: widget/gonk/libdisplay/GonkDisplayICS.cpp @@ +34,5 @@ > + > +namespace { > +static const char* kSleepFile = "/sys/power/wait_for_fb_sleep"; > +static const char* kWakeFile = "/sys/power/wait_for_fb_wake"; > +static mozilla::GonkDisplay::OnEnabledCallbackType sEnabledCallback = nullptr; Static globals are zero/null by default. I know sGonkDisplay is being explicitly nulled elsewhere in this file.. that is a mistake. @@ +65,5 @@ > + } > + sEnabledCallback(true); > + } > + > + return NULL; nullptr @@ +141,5 @@ > > +void > +GonkDisplayICS::OnEnabled(OnEnabledCallbackType callback) > +{ > + if (sEnabledCallback) return; return on the next line @@ +146,5 @@ > + sEnabledCallback = callback; > + > + // Watching screen on/off state by using a pthread > + // which implicitly calls exit() when the main thread ends > + if (pthread_create(&sFramebufferWatchThread, NULL, frameBufferWatcher, NULL)) { nullptr ::: widget/gonk/libdisplay/GonkDisplayJB.cpp @@ +131,5 @@ > > +void > +GonkDisplayJB::OnEnabled(OnEnabledCallbackType callback) > +{ > + mEnabledCallback = callback; This file uses four space indents. ::: widget/gonk/nsWindow.cpp @@ +115,5 @@ > bool mIsOn; > }; > > +static nsRefPtr<ScreenOnOffEvent> sScreenOnEvent = new ScreenOnOffEvent(true); > +static nsRefPtr<ScreenOnOffEvent> sScreenOffEvent = new ScreenOnOffEvent(false); StaticRefPtr is more appropriate in this case. It'll need to be combined with a call to ClearOnShutdown to avoid being counted as a shutdown leak.

review comments addressed, carrying over r+

Try build looks pretty green: https://tbpl.mozilla.org/?tree=Try&rev=3d9aba4acba7

(Apologies, I should have landed this in birch. Please feel free to back me out and abuse me in any way you see fit)

https://hg.mozilla.org/mozilla-central/rev/40a51c5dca9e BURN THE WITCH!!!