Re: [PATCH v1 7/7] drm/xe/vf: Post migration, repopulate ring area for pending request

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

On 14.05.2025 00:49, Tomasz Lis wrote:
> The commands within ring area allocated for a request may contain
> references to GGTT. These references require update after VF
> migration, in order to continue any preempted LRCs, or jobs which
> were emitted to the ring but not sent to GuC yet.
> 
> This change calls the emit function again for all such jobs,
> as part of post-migration recovery.
> 
> Signed-off-by: Tomasz Lis <tomasz.lis@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_guc_submit.c | 20 ++++++++++++++++++++
>  drivers/gpu/drm/xe/xe_guc_submit.h |  2 ++
>  drivers/gpu/drm/xe/xe_sriov_vf.c   | 23 +++++++++++++++++++++++
>  3 files changed, 45 insertions(+)
> 
> diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
> index c485272829a6..238b6691d575 100644
> --- a/drivers/gpu/drm/xe/xe_guc_submit.c
> +++ b/drivers/gpu/drm/xe/xe_guc_submit.c
> @@ -766,6 +766,26 @@ guc_exec_queue_run_job(struct drm_sched_job *drm_job)
>  	return fence;
>  }
>  
> +/**
> + * xe_exec_queue_jobs_ring_restore - Re-emit ring commands of requests pending on given queue.
> + * @eq: the &xe_exec_queue struct instance
> + */
> +void xe_exec_queue_jobs_ring_restore(struct xe_exec_queue *eq)

are you sure this function shouldn't be placed in xe_exec_queue.c ?

> +{
> +	struct xe_gpu_scheduler *sched = &eq->guc->sched;
> +	struct xe_sched_job *job;
> +
> +	if (exec_queue_killed_or_banned_or_wedged(eq))

this condition likely can be checked by the caller (in xe_guc_submit.c)

> +		return;
> +
> +	list_for_each_entry(job, &sched->base.pending_list, drm.list) {
> +		if (xe_sched_job_is_error(job))
> +			continue;
> +
> +		eq->ring_ops->emit_job(job);
> +	}
> +}
> +
>  static void guc_exec_queue_free_job(struct drm_sched_job *drm_job)
>  {
>  	struct xe_sched_job *job = to_xe_sched_job(drm_job);
> diff --git a/drivers/gpu/drm/xe/xe_guc_submit.h b/drivers/gpu/drm/xe/xe_guc_submit.h
> index 2c2d2936440d..55398e292b79 100644
> --- a/drivers/gpu/drm/xe/xe_guc_submit.h
> +++ b/drivers/gpu/drm/xe/xe_guc_submit.h
> @@ -33,6 +33,8 @@ int xe_guc_exec_queue_memory_cat_error_handler(struct xe_guc *guc, u32 *msg,
>  int xe_guc_exec_queue_reset_failure_handler(struct xe_guc *guc, u32 *msg, u32 len);
>  int xe_guc_error_capture_handler(struct xe_guc *guc, u32 *msg, u32 len);
>  
> +void xe_exec_queue_jobs_ring_restore(struct xe_exec_queue *eq);
> +
>  struct xe_guc_submit_exec_queue_snapshot *
>  xe_guc_exec_queue_snapshot_capture(struct xe_exec_queue *q);
>  void
> diff --git a/drivers/gpu/drm/xe/xe_sriov_vf.c b/drivers/gpu/drm/xe/xe_sriov_vf.c
> index c08c44dbd383..2ff1383f0b1a 100644
> --- a/drivers/gpu/drm/xe/xe_sriov_vf.c
> +++ b/drivers/gpu/drm/xe/xe_sriov_vf.c
> @@ -8,6 +8,7 @@
>  #include "xe_assert.h"
>  #include "xe_device.h"
>  #include "xe_exec_queue_types.h"
> +#include "xe_guc_exec_queue_types.h"
>  #include "xe_gt.h"
>  #include "xe_gt_sriov_printk.h"
>  #include "xe_gt_sriov_vf.h"
> @@ -16,6 +17,7 @@
>  #include "xe_irq.h"
>  #include "xe_lrc.h"
>  #include "xe_pm.h"
> +#include "xe_sched_job_types.h"
>  #include "xe_sriov.h"
>  #include "xe_sriov_printk.h"
>  #include "xe_sriov_vf.h"
> @@ -266,6 +268,26 @@ static void vf_post_migration_fixup_contexts(struct xe_device *xe)
>  	}
>  }
>  
> +static void xe_guc_jobs_ring_rebase(struct xe_guc *guc)

and this one in xe_guc_submit.c ?

> +{
> +	struct xe_exec_queue *eq;
> +	unsigned long index;
> +
> +	mutex_lock(&guc->submission_state.lock);
> +	xa_for_each(&guc->submission_state.exec_queue_lookup, index, eq)
> +		xe_exec_queue_jobs_ring_restore(eq);
> +	mutex_unlock(&guc->submission_state.lock);
> +}
> +
> +static void vf_post_migration_fixup_jobs(struct xe_device *xe)
> +{
> +	struct xe_gt *gt;
> +	unsigned int id;
> +
> +	for_each_gt(gt, xe, id)
> +		xe_guc_jobs_ring_rebase(&gt->uc.guc);
> +}
> +
>  static void vf_post_migration_fixup_ctb(struct xe_device *xe)
>  {
>  	struct xe_gt *gt;
> @@ -348,6 +370,7 @@ static void vf_post_migration_recovery(struct xe_device *xe)
>  	need_fixups = vf_post_migration_fixup_ggtt_nodes(xe);
>  	if (need_fixups) {
>  		vf_post_migration_fixup_contexts(xe);
> +		vf_post_migration_fixup_jobs(xe);

in patch 5/7 you've dropped FIXME so I'm surprised by this step ;)

>  		vf_post_migration_fixup_ctb(xe);
>  	}
>