Re: [PATCH 2/2] drm/xe: Drop registration of guc_submit_wedged_fini from xe_guc_submit_wedge()

[Rodrigo Vivi <rodrigo.vivi@intel.com>]

On Thu, Mar 26, 2026 at 02:01:16PM -0700, Matthew Brost wrote:
> xe_guc_submit_wedge() runs in the DMA-fence signaling path, where
> GFP_KERNEL memory allocations are not permitted. However, registering
> guc_submit_wedged_fini via drmm_add_action_or_reset() triggers such an
> allocation.
> 
> Avoid this by moving the logic from guc_submit_wedged_fini() into
> guc_submit_fini(), where wedged exec queue references are dropped during
> normal teardown.

interesting and easier than I had imagined.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

> 
> Fixes: 8ed9aaae39f3 ("drm/xe: Force wedged state and block GT reset upon any GPU hang")
> Signed-off-by: Matthew Brost <matthew.brost@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_guc_submit.c | 33 ++++++++----------------------
>  1 file changed, 9 insertions(+), 24 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
> index a145234f662b..10556156eaad 100644
> --- a/drivers/gpu/drm/xe/xe_guc_submit.c
> +++ b/drivers/gpu/drm/xe/xe_guc_submit.c
> @@ -259,24 +259,12 @@ static void guc_submit_sw_fini(struct drm_device *drm, void *arg)
>  }
>  
>  static void guc_submit_fini(void *arg)
> -{
> -	struct xe_guc *guc = arg;
> -
> -	/* Forcefully kill any remaining exec queues */
> -	xe_guc_ct_stop(&guc->ct);
> -	guc_submit_reset_prepare(guc);
> -	xe_guc_softreset(guc);
> -	xe_guc_submit_stop(guc);
> -	xe_uc_fw_sanitize(&guc->fw);
> -	xe_guc_submit_pause_abort(guc);
> -}
> -
> -static void guc_submit_wedged_fini(void *arg)
>  {
>  	struct xe_guc *guc = arg;
>  	struct xe_exec_queue *q;
>  	unsigned long index;
>  
> +	/* Drop any wedged queue refs */
>  	mutex_lock(&guc->submission_state.lock);
>  	xa_for_each(&guc->submission_state.exec_queue_lookup, index, q) {
>  		if (exec_queue_wedged(q)) {
> @@ -286,6 +274,14 @@ static void guc_submit_wedged_fini(void *arg)
>  		}
>  	}
>  	mutex_unlock(&guc->submission_state.lock);
> +
> +	/* Forcefully kill any remaining exec queues */
> +	xe_guc_ct_stop(&guc->ct);
> +	guc_submit_reset_prepare(guc);
> +	xe_guc_softreset(guc);
> +	xe_guc_submit_stop(guc);
> +	xe_uc_fw_sanitize(&guc->fw);
> +	xe_guc_submit_pause_abort(guc);
>  }
>  
>  static const struct xe_exec_queue_ops guc_exec_queue_ops;
> @@ -1320,10 +1316,8 @@ static void disable_scheduling_deregister(struct xe_guc *guc,
>  void xe_guc_submit_wedge(struct xe_guc *guc)
>  {
>  	struct xe_device *xe = guc_to_xe(guc);
> -	struct xe_gt *gt = guc_to_gt(guc);
>  	struct xe_exec_queue *q;
>  	unsigned long index;
> -	int err;
>  
>  	xe_gt_assert(guc_to_gt(guc), guc_to_xe(guc)->wedged.mode);
>  
> @@ -1335,15 +1329,6 @@ void xe_guc_submit_wedge(struct xe_guc *guc)
>  		return;
>  
>  	if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET) {
> -		err = devm_add_action_or_reset(guc_to_xe(guc)->drm.dev,
> -					       guc_submit_wedged_fini, guc);
> -		if (err) {
> -			xe_gt_err(gt, "Failed to register clean-up on wedged.mode=%s; "
> -				  "Although device is wedged.\n",
> -				  xe_wedged_mode_to_string(XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET));
> -			return;
> -		}
> -
>  		mutex_lock(&guc->submission_state.lock);
>  		xa_for_each(&guc->submission_state.exec_queue_lookup, index, q)
>  			if (xe_exec_queue_get_unless_zero(q))
> -- 
> 2.34.1
>