On Mon, Nov 09, 2015 at 11:52:20AM -0200, Paulo Zanoni wrote: > 2015-11-05 18:40 GMT-02:00 Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>: > > On Thu, Nov 05, 2015 at 06:34:07PM -0200, Paulo Zanoni wrote: > >> 2015-11-05 16:53 GMT-02:00 Rodrigo Vivi <rodrigo.vivi@xxxxxxxxx>: > >> > There are few platforms with other suspend resume bugs that breaks > >> > the full execution. So let's provide a way to skip suspend resume case. > >> > >> Well, I carry a local patch that completely disables suspend subtests > >> for the tests that I usually run, so I really understand your pain. > >> Suspend subtests take a long time to run, and they usually don't work > >> on some of the preproduction machines I still use. > >> > >> But since this problem is not specific to kms_frontbuffer_tracking, > >> maybe we could adopt an igt-wide solution here? Thomas, any idea here? > > > > -x suspend is what I tell piglit on one hsw I have here which hangs on s3. > > The problem with piglit is that it runs every single subtest as a > separate test program invocation. For KMS tests this is a huge problem > since it requires generating the reference CRCs every time and it also > requires a modeset to restore fbcon every single time. For non-eDP the > cost is not big, but for eDP it adds up: running a subset of only 70 > subtests on kms_frontbuffer_tracking, we go from 3:05 to 6:21 in total > execution time. We need to fix this. Generating the reference crc is hard, but getting rid of fbcon and other setup overhead should be doable. Improving piglit runtime means CI can run more tests, which means the test you're writing might actually be used. -Daniel -- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
