LWN: Comments on "Scheduling for the Android display pipeline"

Scheduling for the Android display pipeline

alison — Sun, 26 Jan 2020 22:59:45 +0000

VSYNC "occurs at the refresh rate of the display, and is thus periodic: a display with a 60Hz refresh rate generates a VSYNC every 16.6667ms." With the newish G-Sync/FreeSync HW introduced by Nvidia and AMD respectively, VSYNC is now effectively an IRQ rather than a clock. Do any mobile devices yet make use of such a dynamic scan-out mechanism? Might being able to slightly adjust the scan-out time help?

Scheduling for the Android display pipeline

scientes — Mon, 20 Jan 2020 21:21:51 +0000

Yes, here[1] is Gernot Heiser talking about it, and how it is different from how traditional RT schedulers work.

[1] https://youtu.be/6s5FDX5PkZI?t=801

Scheduling for the Android display pipeline

scientes — Sat, 18 Jan 2020 18:34:26 +0000

Didn't the seL4 scheduler start doing this type of analysis?

Scheduling for the Android display pipeline

pj — Thu, 16 Jan 2020 21:07:47 +0000

It's clear (to me) that a good solution to this is going to have to teach the scheduler about thread dependencies... does the scheduler take into account that thread A is doing an epoll() on a socket whose other end is owned by thread B ? Fencing operations could likewise be instrumented to 'show' when a thread is waiting on a fence.