Re: [PATCH] drm/xe: Invalidate userptr VMA on page pin fault

["Thomas Hellström" <thomas.hellstrom@linux.intel.com>]

Hi, Matthew

On Fri, 2024-03-08 at 13:37 -0800, Matthew Brost wrote:
> Rather than return an error to the user or ban the VM when userptr
> VMA
> page pin fails with -EFAULT, invalidate VMA mappings. This supports
> the
> UMD use case of freeing userptr while still having bindings.
> 
> Signed-off-by: Matthew Brost <matthew.brost@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_gt_pagefault.c |  4 ++--
>  drivers/gpu/drm/xe/xe_trace.h        |  2 +-
>  drivers/gpu/drm/xe/xe_vm.c           | 20 +++++++++++++-------
>  drivers/gpu/drm/xe/xe_vm_types.h     |  7 ++-----
>  4 files changed, 18 insertions(+), 15 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_gt_pagefault.c
> b/drivers/gpu/drm/xe/xe_gt_pagefault.c
> index 73c535193a98..241c294270d9 100644
> --- a/drivers/gpu/drm/xe/xe_gt_pagefault.c
> +++ b/drivers/gpu/drm/xe/xe_gt_pagefault.c
> @@ -69,7 +69,7 @@ static bool access_is_atomic(enum access_type
> access_type)
>  static bool vma_is_valid(struct xe_tile *tile, struct xe_vma *vma)
>  {
>  	return BIT(tile->id) & vma->tile_present &&
> -		!(BIT(tile->id) & vma->usm.tile_invalidated);
> +		!(BIT(tile->id) & vma->tile_invalidated);
>  }
>  
>  static bool vma_matches(struct xe_vma *vma, u64 page_addr)
> @@ -226,7 +226,7 @@ static int handle_pagefault(struct xe_gt *gt,
> struct pagefault *pf)
>  
>  	if (xe_vma_is_userptr(vma))
>  		ret =
> xe_vma_userptr_check_repin(to_userptr_vma(vma));
> -	vma->usm.tile_invalidated &= ~BIT(tile->id);
> +	vma->tile_invalidated &= ~BIT(tile->id);
>  
>  unlock_dma_resv:
>  	drm_exec_fini(&exec);
> diff --git a/drivers/gpu/drm/xe/xe_trace.h
> b/drivers/gpu/drm/xe/xe_trace.h
> index 4ddc55527f9a..846f14507d5f 100644
> --- a/drivers/gpu/drm/xe/xe_trace.h
> +++ b/drivers/gpu/drm/xe/xe_trace.h
> @@ -468,7 +468,7 @@ DEFINE_EVENT(xe_vma, xe_vma_userptr_invalidate,
>  	     TP_ARGS(vma)
>  );
>  
> -DEFINE_EVENT(xe_vma, xe_vma_usm_invalidate,
> +DEFINE_EVENT(xe_vma, xe_vma_invalidate,
>  	     TP_PROTO(struct xe_vma *vma),
>  	     TP_ARGS(vma)
>  );
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> index 643b3701a738..9a19044f7ef6 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -724,11 +724,18 @@ int xe_vm_userptr_pin(struct xe_vm *vm)
>  	list_for_each_entry_safe(uvma, next, &vm-
> >userptr.repin_list,
>  				 userptr.repin_link) {
>  		err = xe_vma_userptr_pin_pages(uvma);
> -		if (err < 0)
> -			return err;
> -
>  		list_del_init(&uvma->userptr.repin_link);
> -		list_move_tail(&uvma->vma.combined_links.rebind,
> &vm->rebind_list);
> +		if (err == -EFAULT) {
> +			err = xe_vm_invalidate_vma(&uvma->vma);

I think we need to check for FAULT_MODE here. If we hit this path in
FAULT_MODE, we already have an invalid gpu access and can kill the VM.

In preempt-fence mode, we should probably be calling
xe_vm_unbind_vma(), because xe_vm_invalidate_vma() isn't safe to call
outside of the mmu_notifier, and if there are still BOOKKEEP fences
pending- see the asserts in that function.

> +			if (err)
> +				return err;
> +		} else {
> +			if (err < 0)
> +				return err;
> +
> +			list_move_tail(&uvma-
> >vma.combined_links.rebind,
> +				       &vm->rebind_list);
> +		}
>  	}
>  
>  	return 0;
> @@ -3214,9 +3221,8 @@ int xe_vm_invalidate_vma(struct xe_vma *vma)
>  	u8 id;
>  	int ret;
>  
> -	xe_assert(xe, xe_vm_in_fault_mode(xe_vma_vm(vma)));
>  	xe_assert(xe, !xe_vma_is_null(vma));
> -	trace_xe_vma_usm_invalidate(vma);
> +	trace_xe_vma_invalidate(vma);
>  
>  	/* Check that we don't race with page-table updates */
>  	if (IS_ENABLED(CONFIG_PROVE_LOCKING)) {
> @@ -3254,7 +3260,7 @@ int xe_vm_invalidate_vma(struct xe_vma *vma)
>  		}
>  	}
>  
> -	vma->usm.tile_invalidated = vma->tile_mask;
> +	vma->tile_invalidated = vma->tile_mask;
>  
>  	return 0;
>  }
> diff --git a/drivers/gpu/drm/xe/xe_vm_types.h
> b/drivers/gpu/drm/xe/xe_vm_types.h
> index 79b5cab57711..ae5fb565f6bf 100644
> --- a/drivers/gpu/drm/xe/xe_vm_types.h
> +++ b/drivers/gpu/drm/xe/xe_vm_types.h
> @@ -84,11 +84,8 @@ struct xe_vma {
>  		struct work_struct destroy_work;
>  	};
>  
> -	/** @usm: unified shared memory state */
> -	struct {
> -		/** @tile_invalidated: VMA has been invalidated */
> -		u8 tile_invalidated;
> -	} usm;
> +	/** @tile_invalidated: VMA has been invalidated */
> +	u8 tile_invalidated;

Add a comment in the commit message about removing the usm struct?
/Thomas


>  
>  	/** @tile_mask: Tile mask of where to create binding for
> this VMA */
>  	u8 tile_mask;