Re: ✗ Fi.CI.IGT: failure for series starting with [1/9] drm/i915: Expose 10:10:10 XRGB formats on SNB-BDW sprites (rev2)

From: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
To: "Ville Syrjälä" <ville.syrjala@linux.intel.com>,
	"Tomi Sarvela" <tomi.p.sarvela@intel.com>,
	intel-gfx@lists.freedesktop.org
Subject: Re: ✗ Fi.CI.IGT: failure for series starting with [1/9] drm/i915:   Expose 10:10:10 XRGB formats on SNB-BDW sprites (rev2)
Date: Tue, 15 Oct 2019 15:08:47 +0300	[thread overview]
Message-ID: <20191015120847.dtzbfatkw7xezi4y@ahiler-desk1.fi.intel.com> (raw)
In-Reply-To: <20191015092559.GL4019@platvala-desk.ger.corp.intel.com>

On Tue, Oct 15, 2019 at 12:25:59PM +0300, Petri Latvala wrote:
> On Tue, Oct 15, 2019 at 09:41:20AM +0300, Arkadiusz Hiler wrote:
> > On Mon, Oct 14, 2019 at 10:23:42PM +0300, Ville Syrjälä wrote:
> > > On Wed, Oct 09, 2019 at 09:12:23PM -0000, Patchwork wrote:
> > > > == Series Details ==
> > > > 
> > > > Series: series starting with [1/9] drm/i915: Expose 10:10:10 XRGB formats on SNB-BDW sprites (rev2)
> > > > URL   : https://patchwork.freedesktop.org/series/67741/
> > > > State : failure
> > > > 
> > > > == Summary ==
> > > > 
> > > > CI Bug Log - changes from CI_DRM_7042_full -> Patchwork_14725_full
> > > > ====================================================
> > > > 
> > > > Summary
> > > > -------
> > > > 
> > > >   **FAILURE**
> > > > 
> > > >   Serious unknown changes coming with Patchwork_14725_full absolutely need to be
> > > >   verified manually.
> > > >   
> > > >   If you think the reported changes have nothing to do with the changes
> > > >   introduced in Patchwork_14725_full, please notify your bug team to allow them
> > > >   to document this new failure mode, which will reduce false positives in CI.
> > > > 
> > > >   
> > > > 
> > > > Possible new issues
> > > > -------------------
> > > > 
> > > >   Here are the unknown changes that may have been introduced in Patchwork_14725_full:
> > > > 
> > > > ### IGT changes ###
> > > > 
> > > > #### Possible regressions ####
> > > > 
> > > >   * igt@gem_eio@in-flight-1us:
> > > >     - shard-snb:          [PASS][1] -> [FAIL][2]
> > > >    [1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_7042/shard-snb7/igt@gem_eio@in-flight-1us.html
> > > >    [2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_14725/shard-snb7/igt@gem_eio@in-flight-1us.html
> > > > 
> > > >   * igt@kms_plane@pixel-format-pipe-a-planes:
> > > >     - shard-iclb:         [PASS][3] -> [FAIL][4] +13 similar issues
> > > >    [3]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_7042/shard-iclb7/igt@kms_plane@pixel-format-pipe-a-planes.html
> > > >    [4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_14725/shard-iclb8/igt@kms_plane@pixel-format-pipe-a-planes.html
> > > 
> > > IGT-Version: 1.24-ge501741f
> > > ...
> > > Testing format AR30(0x30335241) / modifier 0x100000000000003 on A.0
> > > (kms_plane:1411) igt_fb-CRITICAL: Conversion not implemented (from format 0x30335241 to 0x78464749)
> > > 
> > > DRM_FORMAT_ARGB2101010 =  0x30335241
> > > IGT_FORMAT_FLOAT = 0x78464749
> > > 
> > > { .name = "ARGB2101010", .depth = 30, .drm_id = DRM_FORMAT_ARGB2101010,
> > >   .pixman_id = PIXMAN_a2r10g10b10,
> > > 
> > > { .name = "IGT-FLOAT", .depth = -1, .drm_id = IGT_FORMAT_FLOAT,
> > >   .pixman_id = PIXMAN_rgba_float,
> > > 
> > > if ((drm_format_to_pixman(cvt->src.fb->drm_format) != PIXMAN_invalid) &&
> > >     (drm_format_to_pixman(cvt->dst.fb->drm_format) != PIXMAN_invalid)) {
> > > 	cnvert_pixman(cvt);
> > > 	return;
> > > ...
> > > igt_assert_f(false, "Conversion not implemented ...);
> > > 
> > > So wtf?
> > > 
> > > Are we somehow compiling igt with an old pixman causing
> > >  #if PIXMAN_VERSION < PIXMAN_VERSION_ENCODE(0, 36, 0)
> > >  #define PIXMAN_rgba_float PIXMAN_invalid
> > >  #endif
> > > to happen?
> > 
> > oof, seems like the building machine got downgraded somehow
> > 
> > ci-worker1:~$ dpkg -l '*pixman*'
> > Desired=Unknown/Install/Remove/Purge/Hold
> > | Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend
> > |/ Err?=(none)/Reinst-required (Status,Err: uppercase=bad)
> > ||/ Name                                            Version                      Architecture                 Description
> > +++-===============================================-============================-============================-===================================================================================================
> > ii  libpixman-1-0:amd64                             0.34.0-2                     amd64                        pixel-manipulation library for X and cairo
> > ii  libpixman-1-dev:amd64                           0.34.0-2                     amd64                        pixel-manipulation library for X and cairo (development files)
> > 
> > that's bad...
> > 
> > > But the reference run shows it testing all the fancy YUV formats so
> > > I don't think that can be the case.
> > 
> > That's the weird bit...
> > 
> > Anyway the building machine needs updating and apt-mark hold.
> > This can cause fallout and we need to file bugs to limit the noise.
> > 
> > There is quite some queue right now, but hopefully by tomorrow it will
> > be drained. I'll do the necessary updates and force IGT run to see what
> > is going to happen in the morning. Then I'll rerun this series.
> > 
> 
> 
> I don't think the builder ever had a higher version.
> 
> The fancy YUV formats work because the runtime lib is new enough, and
> the build-time checks for those are as such:
> 
> #if CAIRO_VERSION < CAIRO_VERSION_ENCODE(1, 17, 2)
> /*
>  * We need cairo 1.17.2 to use HDR formats, but the only thing added is a value
>  * to cairo_format_t.
>  *
>  * To prevent going outside the enum, make cairo_format_t an int and define
>  * ourselves.
> */
> 
> #define	CAIRO_FORMAT_RGB96F (6)
> #define	CAIRO_FORMAT_RGBA128F (7)
> #define	cairo_format_t int
> #endif
> 
> 
>   and
> 
> 
> igt_skip_on_f(status == CAIRO_STATUS_INVALID_FORMAT &&
> 	      cairo_version() < CAIRO_VERSION_ENCODE(1, 17, 2),
> 	      "Cairo version too old, need 1.17.2, have %s\n",
> 	      cairo_version_string());
> 
> igt_skip_on_f(status == CAIRO_STATUS_NO_MEMORY &&
> 	      pixman_version() < PIXMAN_VERSION_ENCODE(0, 36, 0),
> 	      "Pixman version too old, need 0.36.0, have %s\n",
> 	      pixman_version_string());

I am not a fan of this approach as we nest skips pretty deep within
functions that have neither 'require' nor 'skip' in the name. The
issuing longjmp() can cause surprise short-circuiting of tests iterating
through multiple formats...

Some of that would be solved by the "dynamic subsubtests", but the
semantics still feels off.

> In other words, the backwards compatibility for the fancy YUV formats
> at build-time was easy to sneak in by defining some values, is that
> possible to do with the 10bpc stuff? PIXMAN_rgba_float seems to be
> just PIXMAN_FORMAT_BYTE(128, PIXMAN_TYPE_RGBA_FLOAT,32,32,32,32),
> where PIXMAN_TYPE_RGBA_FLOAT is 11.

Yeah, let it be for now unless someone finds time to do the necessary
reworks :-P

-- 
Cheers,
Arek
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx