Re: [PATCH 18/18] drm/xe: Finish disconnect HW fences from module

[Philipp Stanner <phasta@mailbox.org>]

On Thu, 2025-11-13 at 15:51 +0100, Christian König wrote:
> From: Matthew Brost <matthew.brost@intel.com>
> 
> Be safe when dereferencing fence->xe.
> 
> Signed-off-by: Matthew Brost <matthew.brost@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_hw_fence.c | 13 +++++++------
>  1 file changed, 7 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_hw_fence.c b/drivers/gpu/drm/xe/xe_hw_fence.c
> index f5fad4426729..8181dfc628e4 100644
> --- a/drivers/gpu/drm/xe/xe_hw_fence.c
> +++ b/drivers/gpu/drm/xe/xe_hw_fence.c
> @@ -159,9 +159,7 @@ static struct xe_hw_fence_irq *xe_hw_fence_irq(struct xe_hw_fence *fence)
>  
>  static const char *xe_hw_fence_get_driver_name(struct dma_fence *dma_fence)
>  {
> -	struct xe_hw_fence *fence = to_xe_hw_fence(dma_fence);
> -
> -	return dev_name(fence->xe->drm.dev);
> +	return "xe";
>  }

May I ask what those name callbacks are used for in Xe?
I had so far only seen those dma_fence callbacks being used in i915.

I'm asking because in Rust DmaFence I discovered that having those
callbacks in the first place creates the life time problems Christian
addresses in this series.

I think we had begun brainstorming whether removing get_driver_name()
and get_timeline_name() in dma_fence's backend_ops would be the cleaner
thing to do, compared to the RCU guards.


P.