Re: [PATCH v4 3/5] drm/xe: Make xe_device_declare_wedged() IRQ safe

[Raag Jadav <raag.jadav@intel.com>]

On Sat, Oct 10, 2026 at 02:36:20PM +0530, Raag Jadav wrote:
> Currently, xe_device_declare_wedged() implementation wedges the GTs and
> sends wedged uevent to userspace. This is problematic for usecases which
> require declaring the device as wedged in IRQ context, as it allocates
> memory for event buffer and acquires mutexes deep into it's call path.
> 
> Fix this by deferring wedge handling into its dedicated worker, so that
> xe_device_declare_wedged() can be called in IRQ context.
> 
> First user of this requirement is introduced in commit def675cf3f10
> ("drm/xe/mert: Improve handling of MERT CAT errors"), which declares the
> device as wedged on catastrophic errors received in form of an IRQ.

I just realized that this is fundamentally flawed. IRQ just means the
device is in D0 state and doesn't necessarily indicate non-zero refcount
which is a prerequisite for xe_device_declare_wedged() call.

I'll be dropping this patch in next revision but the rest of the series
still needs to land for other long standing stable fixes and debt cleanup.

Raag

> Fixes: def675cf3f10 ("drm/xe/mert: Improve handling of MERT CAT errors")
> Signed-off-by: Raag Jadav <raag.jadav@intel.com>
> Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
> ---
> v2: Split fixes into separate patches (Rodrigo, Michal)
> ---
>  drivers/gpu/drm/xe/xe_device.c       | 51 +++++++++++++++++-----------
>  drivers/gpu/drm/xe/xe_device_types.h |  2 ++
>  2 files changed, 34 insertions(+), 19 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
> index 6ae1da578bda..2a7794aa71f9 100644
> --- a/drivers/gpu/drm/xe/xe_device.c
> +++ b/drivers/gpu/drm/xe/xe_device.c
> @@ -697,8 +697,16 @@ static int xe_device_vram_alloc(struct xe_device *xe)
>  	return 0;
>  }
>  
> +static void wedged_work(struct work_struct *work);
>  static void xe_device_wedged_fini(struct drm_device *drm, void *arg);
>  
> +static int xe_device_wedged_init(struct xe_device *xe)
> +{
> +	INIT_WORK(&xe->wedged.work, wedged_work);
> +
> +	return drmm_add_action_or_reset(&xe->drm, xe_device_wedged_fini, xe);
> +}
> +
>  /**
>   * xe_device_probe_early: Device early probe
>   * @xe: xe device instance
> @@ -767,7 +775,7 @@ int xe_device_probe_early(struct xe_device *xe)
>  	if (err)
>  		return err;
>  
> -	err = drmm_add_action_or_reset(&xe->drm, xe_device_wedged_fini, xe);
> +	err = xe_device_wedged_init(xe);
>  	if (err)
>  		return err;
>  
> @@ -1392,6 +1400,26 @@ void xe_device_set_wedged_method(struct xe_device *xe, unsigned long method)
>  	xe->wedged.method = method;
>  }
>  
> +static void wedged_work(struct work_struct *work)
> +{
> +	struct xe_device *xe = container_of(work, struct xe_device, wedged.work);
> +
> +	/*
> +	 * XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET is intended for debugging
> +	 * hangs, so wedge the device with 'none' recovery method and have
> +	 * it available to the user for debugging.
> +	 */
> +	if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET)
> +		xe_device_set_wedged_method(xe, DRM_WEDGE_RECOVERY_NONE);
> +	/* If no wedge recovery method is set, use default */
> +	else if (!xe->wedged.method)
> +		xe_device_set_wedged_method(xe, DRM_WEDGE_RECOVERY_REBIND |
> +					    DRM_WEDGE_RECOVERY_BUS_RESET);
> +
> +	/* Notify userspace of wedged device */
> +	drm_dev_wedged_event(&xe->drm, xe->wedged.method, NULL);
> +}
> +
>  #define WEDGED_URL	"https://docs.kernel.org/gpu/drm-uapi.html#device-wedging"
>  #define XE_BUG_URL	"https://gitlab.freedesktop.org/drm/xe/kernel/issues/new"
>  
> @@ -1432,26 +1460,11 @@ void xe_device_declare_wedged(struct xe_device *xe)
>  			    "For recovery procedure, refer to %s\n"
>  			    "Please file a _new_ bug report at %s\n",
>  			    WEDGED_URL, XE_BUG_URL);
> -	}
>  
> -	for_each_initialized_gt(gt, xe, id)
> -		xe_gt_declare_wedged(gt);
> +		for_each_initialized_gt(gt, xe, id)
> +			xe_gt_declare_wedged(gt);
>  
> -	if (xe_device_wedged(xe)) {
> -		/*
> -		 * XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET is intended for debugging
> -		 * hangs, so wedge the device with 'none' recovery method and have
> -		 * it available to the user for debugging.
> -		 */
> -		if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET)
> -			xe_device_set_wedged_method(xe, DRM_WEDGE_RECOVERY_NONE);
> -		/* If no wedge recovery method is set, use default */
> -		else if (!xe->wedged.method)
> -			xe_device_set_wedged_method(xe, DRM_WEDGE_RECOVERY_REBIND |
> -						    DRM_WEDGE_RECOVERY_BUS_RESET);
> -
> -		/* Notify userspace of wedged device */
> -		drm_dev_wedged_event(&xe->drm, xe->wedged.method, NULL);
> +		schedule_work(&xe->wedged.work);
>  	}
>  }
>  
> diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
> index 4e0e1bc040c1..ec5c1f0b8f00 100644
> --- a/drivers/gpu/drm/xe/xe_device_types.h
> +++ b/drivers/gpu/drm/xe/xe_device_types.h
> @@ -547,6 +547,8 @@ struct xe_device {
>  		unsigned long method;
>  		/** @wedged.inconsistent_reset: Inconsistent reset policy state between GTs */
>  		bool inconsistent_reset;
> +		/** @wedged.work: Worker for wedge handling */
> +		struct work_struct work;
>  	} wedged;
>  
>  	/** @devres_group: devres group */
> -- 
> 2.43.0
>