[Intel-gfx] ✗ Fi.CI.BAT: failure for add more probe failures (rev6)

[Michal Wajdeczko]

On Fri, 02 Aug 2019 20:07:22 +0200, Daniele Ceraolo Spurio  
<daniele.ceraolospurio at intel.com> wrote:

>
>
> On 8/2/19 10:02 AM, Michal Wajdeczko wrote:
>> On Fri, 02 Aug 2019 15:30:19 +0200, Chris Wilson  
>> <chris at chris-wilson.co.uk> wrote:
>>
>>> Quoting Patchwork (2019-08-02 14:12:38)
>>>> == Series Details ==
>>>>
>>>> Series: add more probe failures (rev6)
>>>> URL   : https://patchwork.freedesktop.org/series/64390/
>>>> State : failure
>>>>
>>>> == Summary ==
>>>>
>>>> CI Bug Log - changes from CI_DRM_6614 -> Patchwork_13848
>>>> ====================================================
>>>>
>>>> Summary
>>>> -------
>>>>
>>>>   **FAILURE**
>>>>
>>>>   Serious unknown changes coming with Patchwork_13848 absolutely need  
>>>> to be
>>>>   verified manually.
>>>>
>>>>   If you think the reported changes have nothing to do with the  
>>>> changes
>>>>   introduced in Patchwork_13848, please notify your bug team to allow  
>>>> them
>>>>   to document this new failure mode, which will reduce false  
>>>> positives in CI.
>>>>
>>>>   External URL:  
>>>> https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13848/
>>>>
>>>> Possible new issues
>>>> -------------------
>>>>
>>>>   Here are the unknown changes that may have been introduced in  
>>>> Patchwork_13848:
>>>>
>>>> ### IGT changes ###
>>>>
>>>> #### Possible regressions ####
>>>>
>>>>   * igt at i915_module_load@reload-with-fault-injection:
>>>>     - fi-cfl-guc:         [PASS][1] -> [DMESG-WARN][2]
>>>>    [1]:  
>>>> https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_6614/fi-cfl-guc/igt@i915_module_load@reload-with-fault-injection.html  
>>>>    [2]:  
>>>> https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_13848/fi-cfl-guc/igt@i915_module_load@reload-with-fault-injection.html
>>>
>>> Looks like we didn't flush the cleanup we started before reporting the
>>> error. Maybe,
>>  it's "leaked" guc err log, that we now try to free after gem, see:
>>   
>> https://cgit.freedesktop.org/drm/drm-tip/commit/?id=6f76098fe0f3f0b519b2ad528b4319195d6d0f73
>
> Not sure why but I was convinced that the log release was in uc_fini  
> (but I should probably have checked). late_release isn't the place to  
> release the logs IMO, but not sure where to put it since we don't really  
> do anything with the log at the moment (we stopped wedging on uc failure  
> a long time ago and the log is therefore never accessible). What's the  
> plan with https://patchwork.freedesktop.org/patch/298002/? that would  
> make the log useful again and skip uc_fini on load failure, making that  
> the best place to put the release in.

restoring -EIO is next on my list
for now I want to make probe error path more robust
updated series with fix for last issue will be sent shortly

~Michal