[Intel-gfx] ✗ Fi.CI.BAT: failure for series starting with [01/15] drm/i915/execlists: Refactor out complete_preempt_context()

[Chris Wilson]

Quoting Patchwork (2018-03-29 00:01:23)
> == Series Details ==
> 
> Series: series starting with [01/15] drm/i915/execlists: Refactor out complete_preempt_context()
> URL   : https://patchwork.freedesktop.org/series/40834/
> State : failure
> 
> == Summary ==
> 
> Series 40834v1 series starting with [01/15] drm/i915/execlists: Refactor out complete_preempt_context()
> https://patchwork.freedesktop.org/api/1.0/series/40834/revisions/1/mbox/
> 
> ---- Possible new issues:
> 
> Test kms_busy:
>         Subgroup basic-flip-a:
>                 pass       -> DMESG-WARN (fi-bdw-5557u)

This looks to be a spin_lock_irq() in the reset path.

> Test kms_cursor_legacy:
>         Subgroup basic-busy-flip-before-cursor-atomic:
>                 pass       -> SKIP       (fi-bdw-5557u)

This is probably the result of using execlists_set_active()
outside of the tasklet.

Honestly, not as scary as I expected.

The question of whether allowing preempt-timeout for bdw was a
deliberate error or not remains open. Otoh, it does give the same
guaranteed QoS to the preempter, but at the same time we don't give much
opportunity for the active context to vacate.
-Chris