[Intel-gfx] [PATCH] drm/i915: Reserve enough fence slot for i915_vma_unbind_async

[Matthew Auld]

On Fri, 23 Dec 2022 at 09:20, Nirmoy Das <nirmoy.das at intel.com> wrote:
>
> A nested dma_resv_reserve_fences(1) will not reserve slot from the
> 2nd call onwards and folowing dma_resv_add_fence() might hit the
> "BUG_ON(fobj->num_fences >= fobj->max_fences)" check.
>
> I915 hit above nested dma_resv case in ttm_bo_handle_move_mem() with
> async unbind:
>
> dma_resv_reserve_fences() from --> ttm_bo_handle_move_mem()
>         dma_resv_reserve_fences() from --> i915_vma_unbind_async()
>         dma_resv_add_fence() from --> i915_vma_unbind_async()
> dma_resv_add_fence() from -->ttm_bo_move_accel_cleanup()
>
> Resolve this by adding an extra fence in i915_vma_unbind_async().
>
> Suggested-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>
> Signed-off-by: Nirmoy Das <nirmoy.das at intel.com>
Reviewed-by: Matthew Auld <matthew.auld at intel.com>

Does this need a fixes tag?

Fixes: 2f6b90da9192 ("drm/i915: Use vma resources for async unbinding")
Cc: <stable at vger.kernel.org> # v5.18+

I can add when pushing.

> ---
>  drivers/gpu/drm/i915/i915_vma.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
> index 7d044888ac33..5ac4c1c2403c 100644
> --- a/drivers/gpu/drm/i915/i915_vma.c
> +++ b/drivers/gpu/drm/i915/i915_vma.c
> @@ -2149,7 +2149,7 @@ int i915_vma_unbind_async(struct i915_vma *vma, bool trylock_vm)
>         if (!obj->mm.rsgt)
>                 return -EBUSY;
>
> -       err = dma_resv_reserve_fences(obj->base.resv, 1);
> +       err = dma_resv_reserve_fences(obj->base.resv, 2);
>         if (err)
>                 return -EBUSY;
>
> --
> 2.38.0
>