[PATCH] drm/xe/vf: Fix VM crash during VF driver release

[Michal Wajdeczko]

On 7/26/2025 3:39 AM, Satyanarayana K V P wrote:
> The VF CCS save/restore series (patchwork #149108) has a dependency
> on the migration framework. A recent migration update in commit
> d65ff1ec8535 ("drm/xe: Split xe_migrate allocation from initialization")
> caused a VM crash during XE driver release for iGPU devices.

you may want to include crash log here for reference

> 
> Update the VF CCS migration initialization sequence to align with the new
> migration framework changes, resolving the release-time crash.
> 
> Fixes: f3009272ff2e ("drm/xe/vf: Create contexts for CCS read write")
> 

no empty lines here

> Signed-off-by: Satyanarayana K V P <satyanarayana.k.v.p at intel.com>
> Cc: Michal Wajdeczko <michal.wajdeczko at intel.com>
> Cc: Matthew Brost <matthew.brost at intel.com>
> Cc: Matthew Auld <matthew.auld at intel.com>
> Cc: Piotr Piórkowski <piotr.piorkowski at intel.com>
> ---
>  drivers/gpu/drm/xe/xe_gt.c           | 2 +-
>  drivers/gpu/drm/xe/xe_migrate.c      | 5 ++---
>  drivers/gpu/drm/xe/xe_migrate.h      | 2 +-
>  drivers/gpu/drm/xe/xe_sriov_vf_ccs.c | 9 ++++++++-
>  4 files changed, 12 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
> index c8eda36546d3..1ef6eea7df19 100644
> --- a/drivers/gpu/drm/xe/xe_gt.c
> +++ b/drivers/gpu/drm/xe/xe_gt.c
> @@ -564,7 +564,7 @@ static int gt_init_with_all_forcewake(struct xe_gt *gt)
>  	if (xe_gt_is_main_type(gt)) {
>  		struct xe_tile *tile = gt_to_tile(gt);
>  
> -		tile->migrate = xe_migrate_init(tile);
> +		tile->migrate = xe_migrate_init(tile, tile->migrate);
>  		if (IS_ERR(tile->migrate)) {
>  			err = PTR_ERR(tile->migrate);
>  			goto err_force_wake;
> diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c
> index 90065d7d29ff..ae03440a276c 100644
> --- a/drivers/gpu/drm/xe/xe_migrate.c
> +++ b/drivers/gpu/drm/xe/xe_migrate.c
> @@ -400,11 +400,10 @@ struct xe_migrate *xe_migrate_alloc(struct xe_tile *tile)
>   *
>   * Return: Pointer to a migrate context on success. Error pointer on error.
>   */
> -struct xe_migrate *xe_migrate_init(struct xe_tile *tile)
> +struct xe_migrate *xe_migrate_init(struct xe_tile *tile, struct xe_migrate *m)

hmm, why are we passing 'tile' here? it should be just 'm' as 'tile' can be
accessed from m->tile since it was initialized in xe_migrate_alloc()

and then we can simply return err code, not play with ERR_PTR

>  {
> -	struct xe_device *xe = tile_to_xe(tile);
>  	struct xe_gt *primary_gt = tile->primary_gt;
> -	struct xe_migrate *m = tile->migrate;
> +	struct xe_device *xe = tile_to_xe(tile);
>  	struct xe_vm *vm;
>  	int err;
>  
> diff --git a/drivers/gpu/drm/xe/xe_migrate.h b/drivers/gpu/drm/xe/xe_migrate.h
> index 3758f9615484..a5dfb34fc072 100644
> --- a/drivers/gpu/drm/xe/xe_migrate.h
> +++ b/drivers/gpu/drm/xe/xe_migrate.h
> @@ -105,7 +105,7 @@ struct xe_migrate_pt_update {
>  };
>  
>  struct xe_migrate *xe_migrate_alloc(struct xe_tile *tile);
> -struct xe_migrate *xe_migrate_init(struct xe_tile *tile);
> +struct xe_migrate *xe_migrate_init(struct xe_tile *tile, struct xe_migrate *m);
>  
>  struct dma_fence *xe_migrate_to_vram(struct xe_migrate *m,
>  				     unsigned long npages,
> diff --git a/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c b/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c
> index af43e04179aa..1752864f7ffe 100644
> --- a/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c
> +++ b/drivers/gpu/drm/xe/xe_sriov_vf_ccs.c
> @@ -270,11 +270,18 @@ int xe_sriov_vf_ccs_init(struct xe_device *xe)
>  		ctx = &tile->sriov.vf.ccs[ctx_id];
>  		ctx->ctx_id = ctx_id;
>  
> -		migrate = xe_migrate_init(tile);
> +		migrate = xe_migrate_alloc(tile);
>  		if (IS_ERR(migrate)) {
>  			err = PTR_ERR(migrate);
>  			goto err_ret;
>  		}
> +
> +		migrate = xe_migrate_init(tile, migrate);
> +		if (IS_ERR(migrate)) {
> +			err = PTR_ERR(migrate);
> +			goto err_ret;
> +		}
> +
>  		ctx->migrate = migrate;
>  
>  		err = alloc_bb_pool(tile, ctx);