Re: [PATCH v7 2/7] drm/xe: Forcefully tear down exec queues in GuC submit fini

[Michal Wajdeczko <michal.wajdeczko@intel.com>]

On 2/19/2026 7:06 PM, Zhanjun Dong wrote:
> In GuC submit fini, forcefully tear down any exec queues by disabling
> CTs, stopping the scheduler (which cleans up lost G2H), killing all
> remaining queues, and resuming scheduling to allow any remaining cleanup
> actions to complete and signal any remaining fences.
> 
> Split guc_submit_fini into device related and software only part. Using
> device-managed and drm-managed action guarantees the correct ordering of
> cleanup.
> 
> Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
> Cc: stable@vger.kernel.org
> Signed-off-by: Zhanjun Dong <zhanjun.dong@intel.com>
> Signed-off-by: Matthew Brost <matthew.brost@intel.com>
> 
> ---
> history:
> v4:
>  - Split guc_submit_fini into 2 parts, device related by devm and software
>    only part by drmm
> v3:
>  - Add page fault fix
> v2:
>  - Fix VF failure (CI)
> ---
>  drivers/gpu/drm/xe/xe_guc.c        | 18 +++++++++--
>  drivers/gpu/drm/xe/xe_guc.h        |  1 +
>  drivers/gpu/drm/xe/xe_guc_submit.c | 48 +++++++++++++++++++++++-------
>  3 files changed, 55 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_guc.c b/drivers/gpu/drm/xe/xe_guc.c
> index cbbb4d665b8f..f59ae602b8ed 100644
> --- a/drivers/gpu/drm/xe/xe_guc.c
> +++ b/drivers/gpu/drm/xe/xe_guc.c
> @@ -1402,15 +1402,29 @@ int xe_guc_enable_communication(struct xe_guc *guc)
>  	return 0;
>  }
>  
> -int xe_guc_suspend(struct xe_guc *guc)

can you add kernel-doc for this new function?

> +int xe_guc_softreset(struct xe_guc *guc)
>  {
> -	struct xe_gt *gt = guc_to_gt(guc);
>  	u32 action[] = {
>  		XE_GUC_ACTION_CLIENT_SOFT_RESET,
>  	};
>  	int ret;
>  
> +	if (!xe_uc_fw_is_running(&guc->fw))
> +		return 0;
> +
>  	ret = xe_guc_mmio_send(guc, action, ARRAY_SIZE(action));
> +	if (ret)
> +		return ret;
> +
> +	return 0;
> +}
> +
> +int xe_guc_suspend(struct xe_guc *guc)
> +{
> +	struct xe_gt *gt = guc_to_gt(guc);
> +	int ret;
> +
> +	ret = xe_guc_softreset(guc);
>  	if (ret) {
>  		xe_gt_err(gt, "GuC suspend failed: %pe\n", ERR_PTR(ret));
>  		return ret;
> diff --git a/drivers/gpu/drm/xe/xe_guc.h b/drivers/gpu/drm/xe/xe_guc.h
> index 66e7edc70ed9..02514914f404 100644
> --- a/drivers/gpu/drm/xe/xe_guc.h
> +++ b/drivers/gpu/drm/xe/xe_guc.h
> @@ -44,6 +44,7 @@ int xe_guc_opt_in_features_enable(struct xe_guc *guc);
>  void xe_guc_runtime_suspend(struct xe_guc *guc);
>  void xe_guc_runtime_resume(struct xe_guc *guc);
>  int xe_guc_suspend(struct xe_guc *guc);
> +int xe_guc_softreset(struct xe_guc *guc);
>  void xe_guc_notify(struct xe_guc *guc);
>  int xe_guc_auth_huc(struct xe_guc *guc, u32 rsa_addr);
>  int xe_guc_mmio_send(struct xe_guc *guc, const u32 *request, u32 len);
> diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
> index 42712acf2ec2..bfb176c201c7 100644
> --- a/drivers/gpu/drm/xe/xe_guc_submit.c
> +++ b/drivers/gpu/drm/xe/xe_guc_submit.c
> @@ -47,6 +47,8 @@
>  
>  #define XE_GUC_EXEC_QUEUE_CGP_CONTEXT_ERROR_LEN		6
>  
> +static int __xe_guc_submit_reset_prepare(struct xe_guc *guc);
> +
>  static struct xe_guc *
>  exec_queue_to_guc(struct xe_exec_queue *q)
>  {
> @@ -238,7 +240,7 @@ static bool exec_queue_killed_or_banned_or_wedged(struct xe_exec_queue *q)
>  		 EXEC_QUEUE_STATE_BANNED));
>  }
>  
> -static void guc_submit_fini(struct drm_device *drm, void *arg)
> +static void guc_submit_sw_fini(struct drm_device *drm, void *arg)
>  {
>  	struct xe_guc *guc = arg;
>  	struct xe_device *xe = guc_to_xe(guc);
> @@ -256,6 +258,19 @@ static void guc_submit_fini(struct drm_device *drm, void *arg)
>  	xa_destroy(&guc->submission_state.exec_queue_lookup);
>  }
>  
> +static void guc_submit_fini(void *arg)
> +{
> +	struct xe_guc *guc = arg;
> +
> +	/* Forcefully kill any remaining exec queues */
> +	xe_guc_ct_stop(&guc->ct);

it still looks weird that this GuC sub-component touches other GuC sub-components in its own unwind

maybe all this should be done in devm action added in the xe_guc_init_post_hwconfig() ?

> +	__xe_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;
> @@ -325,7 +340,11 @@ int xe_guc_submit_init(struct xe_guc *guc, unsigned int num_ids)
>  
>  	guc->submission_state.initialized = true;
>  
> -	return drmm_add_action_or_reset(&xe->drm, guc_submit_fini, guc);
> +	err = drmm_add_action_or_reset(&xe->drm, guc_submit_sw_fini, guc);
> +	if (err)
> +		return err;
> +
> +	return devm_add_action_or_reset(xe->drm.dev, guc_submit_fini, guc);
>  }
>  
>  /*
> @@ -2298,6 +2317,7 @@ static const struct xe_exec_queue_ops guc_exec_queue_ops = {
>  static void guc_exec_queue_stop(struct xe_guc *guc, struct xe_exec_queue *q)
>  {
>  	struct xe_gpu_scheduler *sched = &q->guc->sched;
> +	bool do_destroy = false;
>  
>  	/* Stop scheduling + flush any DRM scheduler operations */
>  	xe_sched_submission_stop(sched);
> @@ -2305,7 +2325,7 @@ static void guc_exec_queue_stop(struct xe_guc *guc, struct xe_exec_queue *q)
>  	/* Clean up lost G2H + reset engine state */
>  	if (exec_queue_registered(q)) {
>  		if (exec_queue_destroyed(q))
> -			__guc_exec_queue_destroy(guc, q);
> +			do_destroy = true;
>  	}
>  	if (q->guc->suspend_pending) {
>  		set_exec_queue_suspended(q);
> @@ -2341,18 +2361,15 @@ static void guc_exec_queue_stop(struct xe_guc *guc, struct xe_exec_queue *q)
>  			xe_guc_exec_queue_trigger_cleanup(q);
>  		}
>  	}
> +
> +	if (do_destroy)
> +		__guc_exec_queue_destroy(guc, q);
>  }
>  
> -int xe_guc_submit_reset_prepare(struct xe_guc *guc)
> +static int __xe_guc_submit_reset_prepare(struct xe_guc *guc)

as this function is now static, it doesn't need to have xe_ prefix, let it be:

   static int guc_submit_reset_prepare(struct xe_guc *guc)

>  {
>  	int ret;
>  
> -	if (xe_gt_WARN_ON(guc_to_gt(guc), vf_recovery(guc)))
> -		return 0;
> -
> -	if (!guc->submission_state.initialized)
> -		return 0;
> -
>  	/*
>  	 * Using an atomic here rather than submission_state.lock as this
>  	 * function can be called while holding the CT lock (engine reset
> @@ -2367,6 +2384,17 @@ int xe_guc_submit_reset_prepare(struct xe_guc *guc)
>  	return ret;
>  }
>  
> +int xe_guc_submit_reset_prepare(struct xe_guc *guc)
> +{
> +	if (xe_gt_WARN_ON(guc_to_gt(guc), vf_recovery(guc)))
> +		return 0;
> +
> +	if (!guc->submission_state.initialized)
> +		return 0;
> +
> +	return __xe_guc_submit_reset_prepare(guc);
> +}
> +
>  void xe_guc_submit_reset_wait(struct xe_guc *guc)
>  {
>  	wait_event(guc->ct.wq, xe_device_wedged(guc_to_xe(guc)) ||