[igt-dev] [PATCH i-g-t] blacklist: Don't run DRRS test on Intel CI system

[Arkadiusz Hiler]

On Tue, Mar 27, 2018 at 11:21:49AM +0200, Maarten Lankhorst wrote:
> Op 27-03-18 om 11:15 schreef Arkadiusz Hiler:
> > On Mon, Mar 26, 2018 at 03:22:56PM +0200, Maarten Lankhorst wrote:
> >> Op 26-03-18 om 08:29 schreef Marta Lofstedt:
> >>> Skipping takes time, specifically for the big amount of drrs
> >>> related kms_frontbuffer_tracking tests. Since we currently
> >>> don't have any system set up with DRRS panels, blacklisting
> >>> all those test will save time, and we can avoid the need to
> >>> increase the Jenkins timeout in order to solve the Bugzilla
> >>> below.
> >>>
> >>> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105617
> >>> Signed-off-by: Marta Lofstedt <marta.lofstedt at intel.com>
> >>> ---
> >>>  tests/intel-ci/blacklist.txt | 1 +
> >>>  1 file changed, 1 insertion(+)
> >>>
> >>> diff --git a/tests/intel-ci/blacklist.txt b/tests/intel-ci/blacklist.txt
> >>> index 7ca313ac..0a8506ca 100644
> >>> --- a/tests/intel-ci/blacklist.txt
> >>> +++ b/tests/intel-ci/blacklist.txt
> >>> @@ -57,6 +57,7 @@ igt at gem_tiled_swapping@(?!non-threaded).*
> >>>  igt at gem_userptr_blits@(major|minor|forked|mlocked|swapping).*
> >>>  igt at gem_wait@.*hang.*
> >>>  igt at gem_write_read_ring_switch(@.*)?
> >>> +igt at kms_frontbuffer_tracking@.*drrs*
> > I think we should start annotating those, so the reevaluation would be
> > easier than 'git blame'-ing the file and going through each commit.
> >
> >>>  ###############################################
> >>>  # Broadcom
> >>>  ###############################################
> >> Hm a single skipped test should take about 100 ms, but I do notice we
> >> probe all outputs in igt_display_init, so could this also be fixed by
> >> https://patchwork.freedesktop.org/patch/212983/ ?
> >>
> >> This should reduce probing to the minimum.
> > Note the discussion around this patch:
> > http://patchwork.freedesktop.org/patch/204944/
> >
> > The idea was to start async probe during the boot, so we won't depend on
> > the fbdev to do the initial probing.
> >
> > But it seems that, after all, UNKNOWN is still a thing. I should have
> > checked myself...
> >
> > I've pulled the numbers for your change:
> >
> >              CI_DRM_3978      IGTPW_1192
> > shard-apl         12982s          12783s
> > shard-hsw         11735s          11560s
> > shard-snb          7033s           6906s
> > shard-kbl          9843s           9048s
> >
> > Looks good.
> >
> > Note: those are the combined runtimes, so the actual wall-clock time
> > saving for our CI is less than that (because shading).
> >
> > -Arek
> 
> https://patchwork.freedesktop.org/patch/212983/ is a better version. :)

I am not claiming the opposite. The initial patch was just a hax to see
what are the savings and check wheter the test results get affected.

Then the discussion about UNKNOWN has started.

> But it's good to see it changing probing reduces runtimes on kbl. Now
> lets see what we can do to reduce runtime even further.

Yep, especially that the number above are for your, more clever, version.

Somewhere in the process I've also tested a patch that goes through
connectors and reprobes only if all of them are != DRM_MODE_CONNECTED.

I'll review your patch later today, let's continue the discussion there.

Cheers,
Arek