Re: [PATCH v3 5/6] drm/gpusvm: keep a single DMA mapping inline for THP

[Matthew Brost <matthew.brost@intel.com>]

On Fri, Sep 04, 2026 at 04:58:04PM +0000, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> - [Low] Incompatible pointer type in `xe_pt_stage_bind` when `CONFIG_DRM_XE_GPUSVM` is disabled, leading to a build failure.
> --
> 
> commit ba41c31a99e11c5fd960b8101a85807f5290cd76
> Author: Honglei Huang <honghuan@amd.com>
> 
> drm/gpusvm: keep a single DMA mapping inline for THP
> 
> This commit optimizes drm_gpusvm_get_pages() by inlining a single DMA
> mapping entry for Transparent Huge Pages (THP), reducing memory overhead.
> It updates the accessors, including xe_pt_stage_bind() paths, to safely
> handle the new union containing the inline address.
> 
> > diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
> > index 5d990c1c3740a..96ea9735f34ce 100644
> > --- a/drivers/gpu/drm/xe/xe_pt.c
> > +++ b/drivers/gpu/drm/xe/xe_pt.c
> > @@ -831,7 +831,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
> >  			return -EAGAIN;
> >  		}
> >  		if (xe_svm_range_has_dma_mapping(range)) {
> > -			xe_res_first_dma(range->pages.dma_addr, 0,
> > +			xe_res_first_dma(drm_gpusvm_pages_first_dma(&range->pages), 0,
> 
> [Severity: Low]
> Will this cause a build failure when compiling the driver with
> CONFIG_DRM_XE_GPUSVM disabled?
> 
> When CONFIG_DRM_XE_GPUSVM is disabled, struct xe_svm_range uses an anonymous
> struct for its pages field instead of struct drm_gpusvm_pages. Since
> drm_gpusvm_pages_first_dma() expects a const struct drm_gpusvm_pages *,
> passing &range->pages here appears to result in a compiler type mismatch
> error.
> 

Yep. This is related to failure in Intel's CI - kunit builds without
CONFIG_DRM_XE_GPUSVM [1].

I believe Xe side we'd some more inline wrapper:

xe_svm_range_first_dma(struct xe_svm *range)

Then use ifdef around CONFIG_DRM_XE_GPUSVM to either call
drm_gpusvm_pages_first_dma or return NULL. xe_svm.h is structured with
this ifdef in mind. 

If you ask AI how build and run Xe's kunit, it should give you
instructions and should work without any Intel hardware.

Matt

[1] https://patchwork.freedesktop.org/series/173405/

> >  					 xe_svm_range_size(range),
> >  					 &curs);
> >  			xe_svm_range_debug(range, "BIND PREPARE - MIXED");
> 
> [ ... ]
> 
> -- 
> Sashiko AI review · https://sashiko.dev/#/patchset/20260904163956.3433293-1-honghuan@amd.com?part=5