[PATCH i-g-t v1] tests/xe_eudebug: use proper address when waiting for fence

Manszewski, Christoph christoph.manszewski at intel.com
Tue Sep 24 14:31:12 UTC 2024


Hi Jan,

On 24.09.2024 10:49, Jan Sokolowski wrote:
> An incorrect address was used as an argument passed to xe_wait_ufence,
> which caused pagefaults as they referenced unmapped addresses.

Yes, for the exec ioctl this is supposed to be a GPU address in the VM. 
Nice fix!

> 
> Use proper address in xe_wait_ufence.
> 
> Signed-off-by: Jan Sokolowski <jan.sokolowski at intel.com>
> Co-developed-by: Dominik Grzegorzek <dominik.grzegorzek at intel.com>
> Signed-off-by: Dominik Grzegorzek <dominik.grzegorzek at intel.com>

Reviewed-by: Christoph Manszewski <christoph.manszewski at intel.com>

Thanks,
Christoph

> ---
>   tests/intel/xe_eudebug.c | 6 +++---
>   1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/tests/intel/xe_eudebug.c b/tests/intel/xe_eudebug.c
> index 40e07ddf6..e28c9ab67 100644
> --- a/tests/intel/xe_eudebug.c
> +++ b/tests/intel/xe_eudebug.c
> @@ -2121,7 +2121,6 @@ static void *vm_bind_clear_thread(void *data)
>   	fence_data = aligned_alloc(xe_get_default_alignment(fd), sizeof(*fence_data));
>   	igt_assert(fence_data);
>   	uf_sync.timeline_value = 1337;
> -	uf_sync.addr = to_user_pointer(fence_data);
>   
>   	igt_debug("Run on: %s%u\n", xe_engine_class_string(priv->hwe->engine_class),
>   		  priv->hwe->engine_instance);
> @@ -2157,6 +2156,7 @@ static void *vm_bind_clear_thread(void *data)
>   
>   		delta = (random() % bo_size) & -4;
>   
> +		uf_sync.addr = to_user_pointer(fence_data);
>   		/* prepare clean bo */
>   		clean_bo = xe_bo_create(fd, vm, bo_size, priv->region,
>   					DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM);
> @@ -2197,9 +2197,9 @@ static void *vm_bind_clear_thread(void *data)
>   		eq_create.extensions = to_user_pointer(&eq_ext);
>   		exec_queue = xe_eudebug_client_exec_queue_create(priv->c, fd, &eq_create);
>   
> -		memset(fence_data, 0, sizeof(*fence_data));
> +		uf_sync.addr = (cs - map) * 4 + batch_offset;
>   		xe_exec_sync(fd, exec_queue, batch_offset, &uf_sync, 1);
> -		xe_wait_ufence(fd, fence_data, uf_sync.timeline_value, 0,
> +		xe_wait_ufence(fd, (uint64_t *)cs, uf_sync.timeline_value, exec_queue,
>   			       XE_EUDEBUG_DEFAULT_TIMEOUT_SEC * NSEC_PER_SEC);
>   
>   		igt_assert_eq(*map, 0);


More information about the igt-dev mailing list