Am 17.04.25 um 18:37 schrieb Felix Kuehling: > If peer memory is XGMI-accessible, we should never access it through PCIe > P2P DMA mappings. PCIe P2P is slower, has different coherence behaviour, > limited or no support for atomics, or may not work at all. Fail with a > warning if DMABUF mappings of such memory are attempted. > > Signed-off-by: Felix Kuehling <felix.kuehl...@amd.com>
Reviewed-by: Christian König <christian.koe...@amd.com> > --- > drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c | 5 +++++ > 1 file changed, 5 insertions(+) > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c > b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c > index e6913fcf2c7b..44e120f9f764 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c > @@ -199,6 +199,11 @@ static struct sg_table *amdgpu_dma_buf_map(struct > dma_buf_attachment *attach, > break; > > case TTM_PL_VRAM: > + /* XGMI-accessible memory should never be DMA-mapped */ > + if (WARN_ON(amdgpu_dmabuf_is_xgmi_accessible( > + dma_buf_attach_adev(attach), bo))) > + return ERR_PTR(-EINVAL); > + > r = amdgpu_vram_mgr_alloc_sgt(adev, bo->tbo.resource, 0, > bo->tbo.base.size, attach->dev, > dir, &sgt);
