Re: [PATCH] drm/xe/display: consider DPT when WA 22019338487 is active

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

On Tue, 09 Jun 2026, Matthew Auld <matthew.auld@intel.com> wrote:
> WA 22019338487 (22019338487_display) indicates that stolen memory should
> not be used for display allocations on affected platforms (like Lunar
> Lake).
>
> While the fbdev allocation in xe_display_bo.c properly respected this
> workaround, the Display Page Table (DPT) allocation in xe_fb_pin.c
> continued to unconditionally attempt to allocate from stolen memory on
> all integrated GPUs.
>
> Check XE_DEVICE_WA(xe, 22019338487_display) before attempting to
> allocate the DPT from stolen memory. If the workaround applies, skip the
> stolen allocation attempt and let the driver naturally fall back to
> allocating from system memory.
>
> Without this we will end up hammering stolen when programming the DPT on
> the host side during the normal operation, which seems to be exactly
> what the WA wants us to avoid.
>
> There are a bunch of users all getting some kind of hard hang in the fb
> pin programming sequence on LNL, so wondering if this could help there.
>
> Assisted-by: Gemini:gemini-3.1-pro-preview
> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/7513
> Signed-off-by: Matthew Auld <matthew.auld@intel.com>
> Fixes: 775d0adc01a5 ("drm/xe/fbdev: Limit the usage of stolen for LNL+")
> Cc: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
> Cc: Uma Shankar <uma.shankar@intel.com>
> Cc: Nikolay Mikhaylov <sonny@milton.pro>
> Cc: <stable@vger.kernel.org> # v6.12+
> ---
>  drivers/gpu/drm/xe/display/xe_fb_pin.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/xe/display/xe_fb_pin.c b/drivers/gpu/drm/xe/display/xe_fb_pin.c
> index f93c98bec5b5..46b1fd620527 100644
> --- a/drivers/gpu/drm/xe/display/xe_fb_pin.c
> +++ b/drivers/gpu/drm/xe/display/xe_fb_pin.c
> @@ -20,6 +20,9 @@
>  #include "xe_pat.h"
>  #include "xe_pm.h"
>  #include "xe_vram_types.h"
> +#include "xe_wa.h"
> +
> +#include <generated/xe_device_wa_oob.h>
>  
>  static void
>  write_dpt_rotated(struct xe_bo *bo, struct iosys_map *map, u32 *dpt_ofs, u32 bo_ofs,
> @@ -172,7 +175,7 @@ static int __xe_pin_fb_vma_dpt(struct drm_gem_object *obj,
>  						   XE_BO_FLAG_GGTT |
>  						   XE_BO_FLAG_PAGETABLE,
>  						   pin_params->alignment, false);
> -	else
> +	else if (!XE_DEVICE_WA(xe, 22019338487_display))

Nitpick, please don't make if not-some-obscure-workaround the happy day
scenario path.

I'd rather see

	else if (XE_DEVICE_WA(xe, 22019338487_display))
		dpt = ERR_PTR(-ENODEV);


BR,
Jani.

>  		dpt = xe_bo_create_pin_map_at_novm(xe, tile0,
>  						   dpt_size,  ~0ull,
>  						   ttm_bo_type_kernel,
> @@ -180,6 +183,9 @@ static int __xe_pin_fb_vma_dpt(struct drm_gem_object *obj,
>  						   XE_BO_FLAG_GGTT |
>  						   XE_BO_FLAG_PAGETABLE,
>  						   pin_params->alignment, false);
> +	else
> +		dpt = ERR_PTR(-ENODEV);
> +
>  	if (IS_ERR(dpt))
>  		dpt = xe_bo_create_pin_map_at_novm(xe, tile0,
>  						   dpt_size,  ~0ull,

-- 
Jani Nikula, Intel