Re: [Intel-gfx] ✗ Fi.CI.IGT: failure for drm/i915: stop using swiotlb (rev3)

2022-07-25 Thread Tvrtko Ursulin 



On 25/07/2022 15:13, Robert Beckett wrote:

On 25/07/2022 09:23, Tvrtko Ursulin wrote:


On 22/07/2022 20:21, Robert Beckett wrote:

On 22/07/2022 19:05, Patchwork wrote:

*Patch Details*
*Series:*    drm/i915: stop using swiotlb (rev3)
*URL:*    https://patchwork.freedesktop.org/series/106589/ 


*State:*    failure
*Details:* 
https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_106589v3/index.html 
 




  CI Bug Log - changes from CI_DRM_11935_full -> 
Patchwork_106589v3_full



    Summary

*FAILURE*

Serious unknown changes coming with Patchwork_106589v3_full 
absolutely need to be

verified manually.

If you think the reported changes have nothing to do with the changes
introduced in Patchwork_106589v3_full, please notify your bug team 
to allow them
to document this new failure mode, which will reduce false positives 
in CI.



    Participating hosts (10 -> 12)

Additional (2): shard-rkl shard-tglu


    Possible new issues

Here are the unknown changes that may have been introduced in 
Patchwork_106589v3_full:



  IGT changes


    Possible regressions

  *

    igt@gem_mmap_offset@clear:

  o

    shard-tglb: PASS
 


    -> INCOMPLETE
 





I've no idea why CI is seeing a NULL deref here.
Testing locally on my TGL board, it passes fine.
I think I would need some help testing this on other systems to track 
this down.


Is it possible that dma_direct_max_mapping_size can return a whole 
range of "unexpected" values akin to swiotlb_max_segment? Anything 
below PAGE_SIZE, including zero, would confuse the driver.


oh really? scatterlist happily supports non page aligned segments. If 
this is an i915 restriction, it sounds like there could do with being 
some GEM_BUG_ON somewhere to catch this more obviously.


Well zero or one as values returned from swiotlb_max_segment did not 
really make much sense. Hence special handling was put in 
i915_sg_segment_size to correctly interpret them.


So yes a GEM_BUG_ON to ensure value returned from dma_max_mapping_size() 
is at least PAGE_SIZE would be safest. Unless it manages to trigger, in 
which special handling needs to be put back in.


Alternatively we'd need to try to figure out possible return values from 
dma_max_mapping_size(). There doesn't seem to be kernel doc for it 
currently.



Oh..

   return min_t(size_t, UINT_MAX, dma_max_mapping_size(dev)); >
Wrap the min_t value in rounddown(..., PAGE_SIZE).


I'll issue a v5


Or you meant if this one was a i915 limitation? I don't really remember 
and did not do a deep dive to remind myself. Just finger in the air "if 
it is a null sg entry, why would the iterator ever encounter it". So I 
was thinking some sort of a mismatch between found and expected number 
of entries, and hence suspected the lack of rounddown after the 
conversion as the other remaining difference. But as said, did not 
really think it through and checked what in the code would go wrong if 
it wasn't page aligned.


Regards,

Tvrtko



Regards,

Tvrtko




  o

    shard-iclb: PASS
 


    -> INCOMPLETE
 




    Suppressed

The following results come from untrusted machines, tests, or statuses.
They do not affect the overall result.

  *

    igt@gem_mmap_offset@clear:

  o

    {shard-rkl}: NOTRUN -> INCOMPLETE
 



  o

    {shard-tglu}: NOTRUN -> INCOMPLETE
 


    +1 similar issue


    New tests

New tests have been introduced between CI_DRM_11935_full and 
Patchwork_106589v3_full:



  New IGT tests (27)

  *

    igt@kms_flip@basic-flip-vs-dpms@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.66] s
  *

    igt@kms_flip@basic-flip-vs-modeset@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.62] s
  *

    igt@kms_flip@blocking-absolute-wf_vblank-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.59] s
  *

    igt@kms_flip@blocking-absolute-wf_vblank@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.62] s
  *

    igt@kms_flip@bo-too-big-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.39] s
  *

    igt@kms_flip@bo-too-big@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.41] s
  *

    igt@kms_flip@busy-flip@d-hdmi-a1:

  o Stat

Re: [Intel-gfx] ✗ Fi.CI.IGT: failure for drm/i915: stop using swiotlb (rev3)

2022-07-25 Thread Robert Beckett 




On 25/07/2022 09:23, Tvrtko Ursulin wrote:


On 22/07/2022 20:21, Robert Beckett wrote:

On 22/07/2022 19:05, Patchwork wrote:

*Patch Details*
*Series:*    drm/i915: stop using swiotlb (rev3)
*URL:*    https://patchwork.freedesktop.org/series/106589/ 


*State:*    failure
*Details:* 
https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_106589v3/index.html  




  CI Bug Log - changes from CI_DRM_11935_full -> Patchwork_106589v3_full


    Summary

*FAILURE*

Serious unknown changes coming with Patchwork_106589v3_full 
absolutely need to be

verified manually.

If you think the reported changes have nothing to do with the changes
introduced in Patchwork_106589v3_full, please notify your bug team to 
allow them
to document this new failure mode, which will reduce false positives 
in CI.



    Participating hosts (10 -> 12)

Additional (2): shard-rkl shard-tglu


    Possible new issues

Here are the unknown changes that may have been introduced in 
Patchwork_106589v3_full:



  IGT changes


    Possible regressions

  *

    igt@gem_mmap_offset@clear:

  o

    shard-tglb: PASS
 


    -> INCOMPLETE
 





I've no idea why CI is seeing a NULL deref here.
Testing locally on my TGL board, it passes fine.
I think I would need some help testing this on other systems to track 
this down.


Is it possible that dma_direct_max_mapping_size can return a whole range 
of "unexpected" values akin to swiotlb_max_segment? Anything below 
PAGE_SIZE, including zero, would confuse the driver.


oh really? scatterlist happily supports non page aligned segments. If 
this is an i915 restriction, it sounds like there could do with being 
some GEM_BUG_ON somewhere to catch this more obviously.




Oh..

   return min_t(size_t, UINT_MAX, dma_max_mapping_size(dev)); >
Wrap the min_t value in rounddown(..., PAGE_SIZE).


I'll issue a v5



Regards,

Tvrtko




  o

    shard-iclb: PASS
 


    -> INCOMPLETE
 




    Suppressed

The following results come from untrusted machines, tests, or statuses.
They do not affect the overall result.

  *

    igt@gem_mmap_offset@clear:

  o

    {shard-rkl}: NOTRUN -> INCOMPLETE
 



  o

    {shard-tglu}: NOTRUN -> INCOMPLETE
 


    +1 similar issue


    New tests

New tests have been introduced between CI_DRM_11935_full and 
Patchwork_106589v3_full:



  New IGT tests (27)

  *

    igt@kms_flip@basic-flip-vs-dpms@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.66] s
  *

    igt@kms_flip@basic-flip-vs-modeset@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.62] s
  *

    igt@kms_flip@blocking-absolute-wf_vblank-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.59] s
  *

    igt@kms_flip@blocking-absolute-wf_vblank@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.62] s
  *

    igt@kms_flip@bo-too-big-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.39] s
  *

    igt@kms_flip@bo-too-big@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.41] s
  *

    igt@kms_flip@busy-flip@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.62] s
  *

    igt@kms_flip@dpms-off-confusion-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.60] s
  *

    igt@kms_flip@dpms-vs-vblank-race-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [2.72] s
  *

    igt@kms_flip@flip-vs-absolute-wf_vblank-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [8.03] s
  *

    igt@kms_flip@flip-vs-dpms-off-vs-modeset-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.60] s
  *

    igt@kms_flip@flip-vs-dpms-off-vs-modeset@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.57] s
  *

    igt@kms_flip@flip-vs-fences-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.62] s
  *

    igt@kms_flip@flip-vs-fences@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.66] s
  *

    igt@kms_flip@flip-vs-modeset-vs-hang@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [30.02] s
  *

    igt@kms_flip@flip-vs-panning-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
 

Re: [Intel-gfx] ✗ Fi.CI.IGT: failure for drm/i915: stop using swiotlb (rev3)

2022-07-25 Thread Tvrtko Ursulin 



On 22/07/2022 20:21, Robert Beckett wrote:

On 22/07/2022 19:05, Patchwork wrote:

*Patch Details*
*Series:*    drm/i915: stop using swiotlb (rev3)
*URL:*    https://patchwork.freedesktop.org/series/106589/ 


*State:*    failure
*Details:* 
https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_106589v3/index.html 




  CI Bug Log - changes from CI_DRM_11935_full -> Patchwork_106589v3_full


    Summary

*FAILURE*

Serious unknown changes coming with Patchwork_106589v3_full absolutely 
need to be

verified manually.

If you think the reported changes have nothing to do with the changes
introduced in Patchwork_106589v3_full, please notify your bug team to 
allow them
to document this new failure mode, which will reduce false positives 
in CI.



    Participating hosts (10 -> 12)

Additional (2): shard-rkl shard-tglu


    Possible new issues

Here are the unknown changes that may have been introduced in 
Patchwork_106589v3_full:



  IGT changes


    Possible regressions

  *

    igt@gem_mmap_offset@clear:

  o

    shard-tglb: PASS

 


    -> INCOMPLETE

 





I've no idea why CI is seeing a NULL deref here.
Testing locally on my TGL board, it passes fine.
I think I would need some help testing this on other systems to track 
this down.


Is it possible that dma_direct_max_mapping_size can return a whole range 
of "unexpected" values akin to swiotlb_max_segment? Anything below 
PAGE_SIZE, including zero, would confuse the driver.


Oh..

  return min_t(size_t, UINT_MAX, dma_max_mapping_size(dev));

Wrap the min_t value in rounddown(..., PAGE_SIZE).

Regards,

Tvrtko




  o

    shard-iclb: PASS

 


    -> INCOMPLETE

 




    Suppressed

The following results come from untrusted machines, tests, or statuses.
They do not affect the overall result.

  *

    igt@gem_mmap_offset@clear:

  o

    {shard-rkl}: NOTRUN -> INCOMPLETE

 



  o

    {shard-tglu}: NOTRUN -> INCOMPLETE

 


    +1 similar issue


    New tests

New tests have been introduced between CI_DRM_11935_full and 
Patchwork_106589v3_full:



  New IGT tests (27)

  *

    igt@kms_flip@basic-flip-vs-dpms@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.66] s
  *

    igt@kms_flip@basic-flip-vs-modeset@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.62] s
  *

    igt@kms_flip@blocking-absolute-wf_vblank-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.59] s
  *

    igt@kms_flip@blocking-absolute-wf_vblank@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.62] s
  *

    igt@kms_flip@bo-too-big-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.39] s
  *

    igt@kms_flip@bo-too-big@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.41] s
  *

    igt@kms_flip@busy-flip@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.62] s
  *

    igt@kms_flip@dpms-off-confusion-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.60] s
  *

    igt@kms_flip@dpms-vs-vblank-race-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [2.72] s
  *

    igt@kms_flip@flip-vs-absolute-wf_vblank-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [8.03] s
  *

    igt@kms_flip@flip-vs-dpms-off-vs-modeset-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.60] s
  *

    igt@kms_flip@flip-vs-dpms-off-vs-modeset@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.57] s
  *

    igt@kms_flip@flip-vs-fences-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.62] s
  *

    igt@kms_flip@flip-vs-fences@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.66] s
  *

    igt@kms_flip@flip-vs-modeset-vs-hang@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [30.02] s
  *

    igt@kms_flip@flip-vs-panning-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.60] s
  *

    igt@kms_flip@flip-vs-panning@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.62] s
  *

    igt@kms_flip@flip-vs-rmfb-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(

Re: [Intel-gfx] ✗ Fi.CI.IGT: failure for drm/i915: stop using swiotlb (rev3)

2022-07-22 Thread Robert Beckett 




On 22/07/2022 19:05, Patchwork wrote:

*Patch Details*
*Series:*   drm/i915: stop using swiotlb (rev3)
*URL:*	https://patchwork.freedesktop.org/series/106589/ 


*State:*failure
*Details:* 
https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_106589v3/index.html 




  CI Bug Log - changes from CI_DRM_11935_full -> Patchwork_106589v3_full


Summary

*FAILURE*

Serious unknown changes coming with Patchwork_106589v3_full absolutely 
need to be

verified manually.

If you think the reported changes have nothing to do with the changes
introduced in Patchwork_106589v3_full, please notify your bug team to 
allow them

to document this new failure mode, which will reduce false positives in CI.


Participating hosts (10 -> 12)

Additional (2): shard-rkl shard-tglu


Possible new issues

Here are the unknown changes that may have been introduced in 
Patchwork_106589v3_full:



  IGT changes


Possible regressions

  *

igt@gem_mmap_offset@clear:

  o

shard-tglb: PASS


-> INCOMPLETE





I've no idea why CI is seeing a NULL deref here.
Testing locally on my TGL board, it passes fine.
I think I would need some help testing this on other systems to track 
this down.



  o

shard-iclb: PASS


-> INCOMPLETE




Suppressed

The following results come from untrusted machines, tests, or statuses.
They do not affect the overall result.

  *

igt@gem_mmap_offset@clear:

  o

{shard-rkl}: NOTRUN -> INCOMPLETE



  o

{shard-tglu}: NOTRUN -> INCOMPLETE


+1 similar issue


New tests

New tests have been introduced between CI_DRM_11935_full and 
Patchwork_106589v3_full:



  New IGT tests (27)

  *

igt@kms_flip@basic-flip-vs-dpms@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.66] s
  *

igt@kms_flip@basic-flip-vs-modeset@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.62] s
  *

igt@kms_flip@blocking-absolute-wf_vblank-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.59] s
  *

igt@kms_flip@blocking-absolute-wf_vblank@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.62] s
  *

igt@kms_flip@bo-too-big-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.39] s
  *

igt@kms_flip@bo-too-big@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.41] s
  *

igt@kms_flip@busy-flip@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.62] s
  *

igt@kms_flip@dpms-off-confusion-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.60] s
  *

igt@kms_flip@dpms-vs-vblank-race-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [2.72] s
  *

igt@kms_flip@flip-vs-absolute-wf_vblank-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [8.03] s
  *

igt@kms_flip@flip-vs-dpms-off-vs-modeset-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.60] s
  *

igt@kms_flip@flip-vs-dpms-off-vs-modeset@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [0.57] s
  *

igt@kms_flip@flip-vs-fences-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.62] s
  *

igt@kms_flip@flip-vs-fences@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.66] s
  *

igt@kms_flip@flip-vs-modeset-vs-hang@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [30.02] s
  *

igt@kms_flip@flip-vs-panning-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.60] s
  *

igt@kms_flip@flip-vs-panning@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.62] s
  *

igt@kms_flip@flip-vs-rmfb-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.59] s
  *

igt@kms_flip@flip-vs-rmfb@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.60] s
  *

igt@kms_flip@modeset-vs-vblank-race@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [2.81] s
  *

igt@kms_flip@plain-flip-fb-recreate-interruptible@d-hdmi-a1:

  o Statuses : 1 pass(s)
  o Exec time: [7.72] s
  *

igt@kms_f