[igt-dev] [PATCH i-g-t] lib/xe: on xe driver use blitter for x-tile

Rodrigo Vivi rodrigo.vivi at intel.com
Mon Sep 25 20:52:49 UTC 2023 

On Mon, Sep 25, 2023 at 01:42:41PM -0400, Rodrigo Vivi wrote:
> On Mon, Sep 25, 2023 at 06:18:35PM +0300, Juha-Pekka Heikkila wrote:
> > On 25.9.2023 16.37, Rodrigo Vivi wrote:
> > > On Mon, Sep 25, 2023 at 11:02:43AM +0300, Juha-Pekka Heikkila wrote:
> > > > Hi Rodrigo,
> > > > 
> > > > On 23.9.2023 0.47, Rodrigo Vivi wrote:
> > > > > On Thu, Sep 21, 2023 at 07:35:30PM +0300, Juha-Pekka Heikkila wrote:
> > > > > > on xe driver need to use blitter for x-tile
> > > > > 
> > > > > While testing the xe-uapi alignment series I noticed the pass rate had sunk.
> > > > > And then noticed it was all related to display and blit.
> > > > > 
> > > > > A bisect in kernel and in igt brought me to this patch as the culprit for
> > > > > the craches I was seeing on DG2.
> > > > > 
> > > > > And it looks CI is also not very happy:
> > > > > https://intel-gfx-ci.01.org/tree/intel-xe/index.html?
> > > > > 
> > > > > and my bet is on this patch here.
> > > > > 
> > > > > kms_cursor_legacy
> > > > 
> > > > I don't see these crashes, currently this show all green for me:
> > > > https://intel-gfx-ci.01.org/tree/intel-xe/bat-all.html?testfilter=kms_cursor_legacy%7Ckms_flip&hosts=dg2
> > > > 
> > > >  From above link looking at any of those test histories there's no crashes in
> > > > history, this patch went in on IGT_7497 which is still visible on the
> > > > history while writing this.
> > > > 
> > > > I tried to run these tests also on our local dg2 against xe driver but I was
> > > > not able to reproduce any crash here either running tests as per subtest or
> > > > as a blob.
> > > 
> > > So, this is regressing my DG2. I'm trending more to send a revert than a
> > > bug report.
> > > 
> > 
> > No idea what you are trying here.
> 
> The in-denial posture here is what is astonishing me and making me to
> send a revert.
> 
> I'm reporting a real bisected regression that points to this patch, and instead
> of trying to understand the issue you are just pushing back and denying that
> issues my exist,
> 
> > CI all is green, no crashes, as well as
> > our lab 'off the shelf' dg2 everything work.
> 
> Indeed. It was not last Friday, but that was probably coincidence when I looked
> at it. But CI doesn't test all the machines and configurations out there,
> unfortunately. So, if some user reports a regression in special with a bisect,
> we try to at least try to understand what's going on.
> 
> > 
> > Anyway I'll put Lucas and Vandita on CC because reverting this will need to
> > be known by those who have stake in this, reverting this will break lnl and
> > _all_ related upcoming platforms.
> 
> Okay, I will hold the revert for a while.
> 
> I'm going to report a gitlab issue. Let's just please get out of the denial
> mode here and really help to understand what is going on.
> 
> Again, drm-xe-next head kernel and IGT master are crashing all the tests
> that I showed in my first reply all with that crash signature. All of them
> by simply running the BAT/fast-feedback without any special parameter.
> 
> The git bisect on both kernel and IGT led me to this patch here. So, it
> might be a difference on the config or the SKU, but we can *never* *ever*
> deny a regression only because our CI is not catching it or only because
> we cannot reproduce locally.
> 
> Regressions are regressions and specially if they got bisected already.
> 
> In general we should threat the person that reported the issue well, so
> they can provide information that can help the issue to be understood,
> reproduced on other machines, or even get fixes verified and experiments.
> 
> We shouldn't push reporters away by denying the existence of the bug
> only because CI works.

I had opened the issue:
https://gitlab.freedesktop.org/drm/igt-gpu-tools/-/issues/145

However, when trying to build a new kernel with CI .config, I noticed
that I cannot hit the issue anymore.

The issue was with some garbage that my installed previous xe.ko
was leaving and the module reload was not cleaning up. Well, this
should be an issue by itself, but this I cannot reproduce anymore.

I couldn't find the garbage configuration any longer.

Please accept my apologies here.

Sorry for the noise,
Closing the issue now.
Rodrigo.

> 
> Thanks,
> Rodrigo.
> 
> > 
> > /Juha-Pekka
> > 
> > > > 
> > > > > 6/6	6/6	6/6	6/6	6/6	6/6	6/6	0/6
> > > > > basic-flip-after-cursor-atomic
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > basic-flip-after-cursor-legacy
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > basic-flip-after-cursor-varying-size
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > basic-flip-before-cursor-atomic
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > basic-flip-before-cursor-legacy
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > basic-flip-before-cursor-varying-size
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > kms_dsc
> > > > > 0/0	0/0	0/0	0/0	0/0	0/0	0/0	0/0
> > > > > dsc-basic
> > > > > skip	skip	skip	skip	notrun	skip	skip	skip
> > > > > kms_flip
> > > > > 16/16	16/16	16/16	16/16	16/16	16/16	16/16	0/16
> > > > > basic-flip-vs-dpms
> > > > > 0/0	0/0	0/0	0/0	0/0	0/0	0/0	0/0
> > > > > a-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > b-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > c-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > d-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > basic-flip-vs-modeset
> > > > > 0/0	0/0	0/0	0/0	0/0	0/0	0/0	0/0
> > > > > a-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > b-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > c-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > d-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > basic-flip-vs-wf_vblank
> > > > > 0/0	0/0	0/0	0/0	0/0	0/0	0/0	0/0
> > > > > a-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > b-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > c-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > d-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > basic-plain-flip
> > > > > 0/0	0/0	0/0	0/0	0/0	0/0	0/0	0/0
> > > > > a-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > b-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > c-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > d-dp2
> > > > > pass	pass	pass	pass	pass	pass	pass	crash
> > > > > 
> > > > > Results for igt at kms_cursor_legacy@basic-flip-before-cursor-varying-size
> > > > > Overview
> > > > > Result: crash
> > > > > 
> > > > > Back to summary
> > > > > 
> > > > > Details
> > > > > Detail	Value
> > > > > Returncode	None
> > > > > Time	0:00:00
> > > > > Stdout	
> > > > > Opened device: /dev/dri/card0
> > > > > Starting subtest: basic-flip-before-cursor-varying-size
> > > > > Using pipe A & DP-2
> > > > > Subtest basic-flip-before-cursor-varying-size: CRASH (0.705s)
> > > > > Stderr	
> > > > > Starting subtest: basic-flip-before-cursor-varying-size
> > > > > Received signal SIGBUS.
> > > > > Stack trace:
> > > > >    #0 [fatal_sig_handler+0x17b]
> > > > >    #1 [__sigaction+0x50]
> > > > >    #2 [emit_blt_block_copy+0x47e]
> > > > >    #3 [blt_block_copy+0xeb]
> > > > >    #4 [blitcopy+0x263]
> > > > >    #5 [setup_linear_mapping+0x163]
> > > > >    #6 [create_cairo_surface__gpu+0x32]
> > > > >    #7 [igt_get_cairo_surface+0x88]
> > > > >    #8 [igt_get_cairo_ctx+0xb]
> > > > >    #9 [igt_create_pattern_fb+0x29]
> > > > >    #10 [set_fb_on_crtc+0x39]
> > > > >    #11 [basic_flip_cursor+0xcf]
> > > > >    #12 [__igt_unique____real_main1887+0xa41]
> > > > >    #13 [main+0x23]
> > > > >    #14 [__libc_start_call_main+0x80]
> > > > >    #15 [__libc_start_main+0x89]
> > > > >    #16 [_start+0x25]
> > > > > Subtest basic-flip-before-cursor-varying-size: CRASH (0.705s)
> > > > > Command	
> > > > > dmesg	
> > > > > <6> [29102.951496] [IGT] kms_cursor_legacy: executing
> > > > > <6> [29103.289314] [IGT] kms_cursor_legacy: starting subtest basic-flip-before-cursor-varying-size
> > > > > <6> [29104.003123] [IGT] kms_cursor_legacy: finished subtest basic-flip-before-cursor-varying-size, CRASH
> > > > > <6> [29104.012383] [IGT] kms_cursor_legacy: exiting, ret=135
> > > > > 
> > > > > 
> > > > > > 
> > > > > > Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila at gmail.com>
> > > > > > ---
> > > > > >    lib/igt_fb.c | 3 ++-
> > > > > >    1 file changed, 2 insertions(+), 1 deletion(-)
> > > > > > 
> > > > > > diff --git a/lib/igt_fb.c b/lib/igt_fb.c
> > > > > > index 4b592825d..c8a7b31a8 100644
> > > > > > --- a/lib/igt_fb.c
> > > > > > +++ b/lib/igt_fb.c
> > > > > > @@ -2557,7 +2557,8 @@ static bool use_blitter(const struct igt_fb *fb)
> > > > > >    	return fb->modifier == I915_FORMAT_MOD_4_TILED ||
> > > > > >    	       fb->modifier == I915_FORMAT_MOD_Y_TILED ||
> > > > > >    	       fb->modifier == I915_FORMAT_MOD_Yf_TILED ||
> > > > > > -	       (is_i915_device(fb->fd) && !gem_has_mappable_ggtt(fb->fd));
> > > > > > +	       (is_i915_device(fb->fd) && !gem_has_mappable_ggtt(fb->fd)) ||
> > > > > > +	       is_xe_device(fb->fd);
> > > > > >    }
> > > > > >    static void init_buf_ccs(struct intel_buf *buf, int ccs_idx,
> > > > > > -- 
> > > > > > 2.25.1
> > > > > > 
> > > > 
> >

More information about the igt-dev mailing list