Re: [PATCH v11 3/3] drm/xe: Handle PUNIT errors by requesting cold-reset recovery

On 21-07-2026 11:07 pm, Rodrigo Vivi wrote:
> On Mon, Jul 20, 2026 at 10:48:15AM +0000, [email protected] wrote:
>> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
>> - [High] Calling `xe_device_declare_wedged()` during early driver probe fails to notify userspace because the DRM device is not yet registered in sysfs.
>> --
>>
>> --- Patch [3]: [PATCH v11 3/3] drm/xe: Handle PUNIT errors by requesting cold-reset recovery ---
>> commit e4801339581daca0be0db10f90d94f25e72cfa52
>> Author: Mallesh Koujalagi <[email protected]>
>>
>> drm/xe: Handle PUNIT errors by requesting cold-reset recovery
>>
>> When PUNIT (power management unit) errors are detected that persist across
>> warm resets, mark the device as wedged with DRM_WEDGE_RECOVERY_COLD_RESET
>> and notify userspace that a complete device power cycle is required to
>> restore normal operation.
>>
>>> diff --git a/drivers/gpu/drm/xe/xe_ras.c b/drivers/gpu/drm/xe/xe_ras.c
>>> index a31e06b8aa67a..92b4181026cbe 100644
>>> --- a/drivers/gpu/drm/xe/xe_ras.c
>>> +++ b/drivers/gpu/drm/xe/xe_ras.c
>> [ ... ]
>>> @@ -236,6 +236,12 @@ static u8 handle_core_compute_errors(struct xe_ras_error_array *arr)
>>>   	return XE_RAS_RECOVERY_ACTION_RECOVERED;
>>>   }
>>>   
>>> +static void punit_error_handler(struct xe_device *xe)
>>> +{
>>> +	xe_device_set_wedged_method(xe, DRM_WEDGE_RECOVERY_COLD_RESET);
>>> +	xe_device_declare_wedged(xe);
>>          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>
>> [Severity: High]
>> Does calling xe_device_declare_wedged() here fail to notify userspace when
>> triggered during early driver probe?
>>
>> If a PUNIT error is present at driver load, the call chain is:
>>
>> xe_device_probe()
>>    xe_ras_init()
>>      xe_ras_process_errors()
>>        handle_soc_internal_errors()
>>          punit_error_handler()
>>            xe_device_declare_wedged()
>>              drm_dev_wedged_event()
>>
>> Since drm_dev_register() has not yet been called at this stage of the probe
>> path, the sysfs directory for the DRM device does not exist, and the KOBJ_ADD
>> event has not been emitted.
>>
>> Consequently, userspace (e.g., udev) will likely drop or ignore the KOBJ_CHANGE
>> uevent emitted by drm_dev_wedged_event().
>>
>> Will this cause the device to finish probe successfully but leave userspace
>> unaware of the wedged state, breaking the automated cold reset recovery
>> mechanism?
> Mallesh, could you please add a fault-inject test to this flow here and
> ensure that we are indeed getting the notification we are willing to get?
>
> It looks that right now we are really never getting the uevent that we
> are willing to get.
>
> Thanks,
> Rodrigo.

In boot-time punit error, the device goes into boot survivability mode

and never reaches the punit_error_handler() → xe_device_declare_wedged() 
path before drm_dev_register().

Because of this, the dropped uevent scenario does not actually occur in 
practice.

Thanks,

-/Mallesh


>>> +}
>>> +
>>>   static u8 handle_soc_internal_errors(struct xe_device *xe, struct xe_ras_error_array *arr)
>>>   {
>>>   	struct xe_ras_soc_error *info = (void *)arr->details;
>> -- 
>> Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=3