Re: [PATCH i-g-t] lib/xe: Retry xe_vm_bind on transient -EBUSY

[Jani Nikula <jani.nikula@intel.com>]

On Tue, 12 May 2026, nishit.sharma@intel.com wrote:
> From: Nishit Sharma <nishit.sharma@intel.com>
>
> __xe_vm_bind_assert() can intermittently fail with -EBUSY when the
> kernel rejects the bind because it detects in-flight work that has
> not yet completed. This is transient condition the call is expected
> to succeed once the competing work settles.
>
> Add a retry loop (up to 5 attempts) that re-issues __xe_vm_bind()
> whenever -EBUSY is returned, before asserting success. All other
> error codes are treated as fatal and propagate immediately to
> igt_assert_eq() as before.
>
> Signed-off-by: Nishit Sharma <nishit.sharma@intel.com>
> ---
>  lib/xe/xe_ioctl.c | 21 ++++++++++++++++++---
>  1 file changed, 18 insertions(+), 3 deletions(-)
>
> diff --git a/lib/xe/xe_ioctl.c b/lib/xe/xe_ioctl.c
> index e13195e16..3a307f95f 100644
> --- a/lib/xe/xe_ioctl.c
> +++ b/lib/xe/xe_ioctl.c
> @@ -242,9 +242,24 @@ void  __xe_vm_bind_assert(int fd, uint32_t vm, uint32_t exec_queue, uint32_t bo,
>  			  uint32_t op, uint32_t flags, struct drm_xe_sync *sync,
>  			  uint32_t num_syncs, uint32_t prefetch_region, uint64_t ext)
>  {
> -	igt_assert_eq(__xe_vm_bind(fd, vm, exec_queue, bo, offset, addr, size,
> -				   op, flags, sync, num_syncs, prefetch_region,
> -				   DEFAULT_PAT_INDEX, ext), 0);
> +	int ret, retries = 0;
> +
> +retry:
> +	ret = __xe_vm_bind(fd, vm, exec_queue, bo, offset, addr, size,
> +			   op, flags, sync, num_syncs, prefetch_region,
> +			   DEFAULT_PAT_INDEX, ext);
> +
> +	if (ret && retries < 5) {
> +		/*
> +		 * Retry on transient -EBUSY, which can occur when the
> +		 * kernel is contended by in-flight background work
> +		 */
> +		retries++;
> +		goto retry;
> +	}

Please use a loop for looping.

BR,
Jani.



> +
> +	igt_assert_eq(ret, 0);
> +	return;
>  }
>  
>  void xe_vm_prefetch_async(int fd, uint32_t vm, uint32_t exec_queue, uint64_t offset,

-- 
Jani Nikula, Intel