[PATCH] Fix CAT error by adjusting prefetch size for Xe2
Matthew Auld
matthew.auld at intel.com
Thu Jan 18 08:45:12 UTC 2024
On 17/01/2024 20:46, Jagmeet Randhawa wrote:
> Due to batch pre-fetch being quite aggressive on Xe2, we need
> an extra 4K page at the end of the batch to ensure HW doesn't
> pre-fetch beyond the boundaries of the batch itself, hitting
> an invalid ppGTT page, resulting in CAT errors. Fix this by
> accounting for the xe_cs_prefetch_size() when calculating
> the size of the batch.
>
> Suggested-by: Matthew Auld <matthew.auld at intel.com>
> Signed-off-by: Jagmeet Randhawa <jagmeet.randhawa at intel.com>
Commit title needs to have some kind of prefix like:
"tests/intel/xe_exec_store".
If the entire thing is too long then shorten to something like:
"tests/intel/xe_exec_store: account for prefetch".
Otherwise,
Reviewed-by: Matthew Auld <matthew.auld at intel.com>
> ---
> tests/intel/xe_exec_store.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/tests/intel/xe_exec_store.c b/tests/intel/xe_exec_store.c
> index 2e4f57b47..bed118688 100644
> --- a/tests/intel/xe_exec_store.c
> +++ b/tests/intel/xe_exec_store.c
> @@ -201,7 +201,7 @@ static void store_cachelines(int fd, struct drm_xe_engine_class_instance *eci,
> uint32_t *batch_map;
> size_t bo_size = 4096;
>
> - bo_size = ALIGN(bo_size, xe_get_default_alignment(fd));
> + bo_size = ALIGN(bo_size + xe_cs_prefetch_size(fd), xe_get_default_alignment(fd));
> vm = xe_vm_create(fd, 0, 0);
> ahnd = intel_allocator_open(fd, 0, INTEL_ALLOCATOR_SIMPLE);
> exec_queues = xe_exec_queue_create(fd, vm, eci, 0);
More information about the igt-dev
mailing list