Re: [RFC PATCH 0/4] Add cold reset recovery method for critical errors

["Mallesh, Koujalagi" <mallesh.koujalagi@intel.com>]

Hi Christian,

On 11-02-2026 05:57 pm, Christian König wrote:
> On 2/11/26 12:59, Mallesh Koujalagi wrote:
>> This RFC patch series introduces a new DRM wedge recovery method
>> 'DRM_WEDGE_RECOVERY_COLD_RESET' for handling critical errors
>> that cannot be recovered through existing software-based mechanisms.
>>
>> Background
>> ----------
>> Current recovery methods (driver rebind, bus reset, FLR) are effective
>> for most error scenarios. However, certain critical errors
>> affect device-level persistent state that survives warm resets and
>> software recovery attempts. These errors require complete device power
>> cycling to restore functionality.
> I don't think that this is a sufficient justification for making those changes.
>
> Especially since the patch set doesn't seem to add any detection for those cases, but rather just exposes a debugfs file to trigger them.
>
> So what is the actual technical background? In other words when is that necessary?
>
> Regards,
> Christian.

Thanks for the feedback. Sorry I missed to add reference of actual usecase.
This method is for handling errors from power management unit, which 
requires
a complete power cycle (cold reset) to recover.

I'll add actual implementation in next revision. It will be an extension of
our WIP RAS infrastructure which is being developed in parallel.
(see https://patchwork.freedesktop.org/series/160482/)

Current RFC series is to get community insight on proposed recovery method
as part of wedging uapi.

Thanks,
-/Mallesh


>> Proposed Solution
>> -----------------
>> This series adds DRM_WEDGE_RECOVERY_COLD_RESET (BIT(4)) as a new
>> recovery method to the DRM wedging framework. When this method is set,
>> it signals to userspace that only a complete device cold reset (power
>> cycle) can restore normal operation.
>>
>> Example uevent received:
>>    SUBSYSTEM=drm
>>    WEDGED=cold-reset
>>    DEVPATH=/devices/.../drm/card0
>>
>> Testing
>> -------
>> The debugfs interface allows testing the cold reset recovery path:
>>
>>    echo 1 > /sys/kernel/debug/dri/N/trigger_critical_error
>>
>> This triggers the critical error handler, wedges the device with
>> cold reset method, and sends the appropriate uevent to userspace.
>>
>> Cc: André Almeida <andrealmeid@igalia.com>
>> Cc: Christian König <christian.koenig@amd.com>
>> Cc: David Airlie <airlied@gmail.com>
>> Cc: Simona Vetter <simona.vetter@ffwll.ch>
>> Cc: Maxime Ripard <mripard@kernel.org>
>>
>> Mallesh Koujalagi (4):
>>    drm: Add DRM_WEDGE_RECOVERY_COLD_RESET for critical error
>>    drm/doc: Document DRM_WEDGE_RECOVERY_COLD_RESET recovery method
>>    drm/xe: Add handler for critical errors which require cold-reset
>>    drm/xe/debugfs: Add interface to trigger critical error handler
>>
>>   Documentation/gpu/drm-uapi.rst   | 73 +++++++++++++++++++++++++++++++-
>>   drivers/gpu/drm/drm_drv.c        |  2 +
>>   drivers/gpu/drm/xe/xe_debugfs.c  | 38 +++++++++++++++++
>>   drivers/gpu/drm/xe/xe_hw_error.c | 28 ++++++++++++
>>   drivers/gpu/drm/xe/xe_hw_error.h |  1 +
>>   include/drm/drm_device.h         |  1 +
>>   6 files changed, 142 insertions(+), 1 deletion(-)
>>