Re: [PATCH v2 4/4] drm/xe/multi_queue: replay dynamic CGP updates lost during VF migration

On Fri, Jul 31, 2026 at 04:26:27PM -0700, Niranjana Vishwanathapura wrote:
> When guc_exec_queue_revert_pending_state_change() sets needs_cgp_sync for
> a queue, replay that update during unpause by injecting a CGP_SYNC_MSG
> static message at the head of the scheduler queue.
> 
> The message handler __guc_exec_queue_process_msg_cgp_sync() calls
> guc_exec_queue_send_cgp_sync() which re-reads the current priority from
> q->multi_queue.priority and re-issues the CGP write + CGP_SYNC H2G, so
> the queue's CGP entry is restored to the correct value after migration.
> 
> Assisted-by: Github-Copilot:Claude-opus-4.8
> Signed-off-by: Niranjana Vishwanathapura <[email protected]>

Reviewed-by: Matthew Brost <[email protected]>

> ---
>  drivers/gpu/drm/xe/xe_guc_exec_queue_types.h |  2 +-
>  drivers/gpu/drm/xe/xe_guc_submit.c           | 28 ++++++++++++++++++++
>  2 files changed, 29 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h b/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h
> index 573b920edb41..d27826b36649 100644
> --- a/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h
> +++ b/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h
> @@ -36,7 +36,7 @@ struct xe_guc_exec_queue {
>  	 * a message needs to sent through the GPU scheduler but memory
>  	 * allocations are not allowed.
>  	 */
> -#define MAX_STATIC_MSG_TYPE	3
> +#define MAX_STATIC_MSG_TYPE	4
>  	struct xe_sched_msg static_msgs[MAX_STATIC_MSG_TYPE];
>  	/** @destroy_async: do final destroy async from this worker */
>  	struct work_struct destroy_async;
> diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
> index c018bc0d8d6f..9036f89dff7d 100644
> --- a/drivers/gpu/drm/xe/xe_guc_submit.c
> +++ b/drivers/gpu/drm/xe/xe_guc_submit.c
> @@ -1960,11 +1960,25 @@ static void __guc_exec_queue_process_msg_set_multi_queue_priority(struct xe_sche
>  	kfree(msg);
>  }
>  
> +static void __guc_exec_queue_process_msg_cgp_sync(struct xe_sched_msg *msg)
> +{
> +	struct xe_exec_queue *q = msg->private_data;
> +
> +	/*
> +	 * Replay a dynamic CGP update lost across VF migration by re-issuing the
> +	 * CGP update + CGP_SYNC (re-applies the current priority from
> +	 * q->multi_queue.priority).
> +	 */
> +	if (guc_exec_queue_allowed_to_change_state(q))
> +		guc_exec_queue_send_cgp_sync(q, 0);
> +}
> +
>  #define CLEANUP				1	/* Non-zero values to catch uninitialized msg */
>  #define SET_SCHED_PROPS			2
>  #define SUSPEND				3
>  #define RESUME				4
>  #define SET_MULTI_QUEUE_PRIORITY	5
> +#define CGP_SYNC_MSG			6
>  #define OPCODE_MASK	0xf
>  #define MSG_LOCKED	BIT(8)
>  #define MSG_HEAD	BIT(9)
> @@ -1991,6 +2005,9 @@ static void guc_exec_queue_process_msg(struct xe_sched_msg *msg)
>  	case SET_MULTI_QUEUE_PRIORITY:
>  		__guc_exec_queue_process_msg_set_multi_queue_priority(msg);
>  		break;
> +	case CGP_SYNC_MSG:
> +		__guc_exec_queue_process_msg_cgp_sync(msg);
> +		break;
>  	default:
>  		XE_WARN_ON("Unknown message type");
>  	}
> @@ -2160,6 +2177,7 @@ static bool guc_exec_queue_try_add_msg(struct xe_exec_queue *q,
>  #define STATIC_MSG_CLEANUP	0
>  #define STATIC_MSG_SUSPEND	1
>  #define STATIC_MSG_RESUME	2
> +#define STATIC_MSG_CGP_SYNC	3
>  static void guc_exec_queue_destroy(struct xe_exec_queue *q)
>  {
>  	struct xe_sched_msg *msg = q->guc->static_msgs + STATIC_MSG_CLEANUP;
> @@ -3024,6 +3042,16 @@ static void guc_exec_queue_replay_pending_state_change(struct xe_exec_queue *q)
>  	struct xe_gpu_scheduler *sched = &q->guc->sched;
>  	struct xe_sched_msg *msg;
>  
> +	if (q->guc->multi_queue.needs_cgp_sync) {
> +		msg = q->guc->static_msgs + STATIC_MSG_CGP_SYNC;
> +
> +		xe_sched_msg_lock(sched);
> +		guc_exec_queue_try_add_msg_head(q, msg, CGP_SYNC_MSG);
> +		xe_sched_msg_unlock(sched);
> +
> +		q->guc->multi_queue.needs_cgp_sync = false;
> +	}
> +
>  	if (q->guc->needs_cleanup) {
>  		msg = q->guc->static_msgs + STATIC_MSG_CLEANUP;
>  
> -- 
> 2.43.0
>