Re: [PATCH v4 3/3] drm/xe/guc_ads: use uncached mapping for UM queue BO

[Michal Wajdeczko <michal.wajdeczko@intel.com>]

On 7/30/2026 2:47 AM, Jia Yao wrote:
> The HW unit that writes fault descriptors into the UM queue ring buffer
> may not be coherent with the CPU cache.  If a descriptor write is still
> sitting in a GPU L3 cache line when GuC reads the slot, GuC sees the
> stale (all-zero) content from DRAM and forwards an invalid descriptor to
> the driver (Faulted Address = 0, ASID = 0).
> 
> Mark the UM queue BO as uncached (XE_BO_FLAG_NEEDS_UC) so that writes
> bypass the cache and are immediately visible to GuC.
> 
> Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")

note that XE_BO_FLAG_NEEDS_UC flag is not available at this point

it was initially added as XE_BO_NEEDS_UC flag in a commit f15de1936f8d
and then renamed to XE_BO_FLAG_NEEDS_UC in commit 62742d126631451

> Cc: Gwan-gyeong Mun <gwan-gyeong.mun@intel.com>
> Cc: Matthew Auld <matthew.auld@intel.com>
> Cc: stable@vger.kernel.org
> Signed-off-by: Jia Yao <jia.yao@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_guc_ads.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_guc_ads.c b/drivers/gpu/drm/xe/xe_guc_ads.c
> index 53636333fb45..76029575aa45 100644
> --- a/drivers/gpu/drm/xe/xe_guc_ads.c
> +++ b/drivers/gpu/drm/xe/xe_guc_ads.c
> @@ -412,7 +412,8 @@ int xe_guc_ads_init(struct xe_guc_ads *ads)
>  		u32 um_flags = XE_BO_FLAG_VRAM_IF_DGFX(tile) |
>  			       XE_BO_FLAG_GGTT |
>  			       XE_BO_FLAG_GGTT_INVALIDATE |
> -			       XE_BO_FLAG_PINNED_NORESTORE;
> +			       XE_BO_FLAG_PINNED_NORESTORE |
> +			       XE_BO_FLAG_NEEDS_UC;
>  
>  		bo = xe_managed_bo_create_pin_map(xe, tile, um_size, um_flags);
>  		if (IS_ERR(bo))