The GEM CMA helpers uses a custom mmap implementation based on
remap_pfn_range(). While this works when the buffer DMA and physical
addresses are identical, it fails to take IOMMU into account and tries
to mmap the buffer to userspace using the DMA virtual address instead of
the physical address. This results in mapping random physical pages when
the device is behind an IOMMU.
Use the DMA mapping dma_mmap_writecombine() function instead.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@xxxxxxxxxxxxxxxx>
---
drivers/gpu/drm/drm_gem_cma_helper.c | 13 +++++++++++--
1 file changed, 11 insertions(+), 2 deletions(-)
This patches fixes a problem noticed when enabling IOMMU usage with the Renesas
R-Car DU driver. I've tested it both with the IOMMU enabled and disabled.
diff --git a/drivers/gpu/drm/drm_gem_cma_helper.c b/drivers/gpu/drm/drm_gem_cma_helper.c
index 6b51bf9..1d4fd34 100644
--- a/drivers/gpu/drm/drm_gem_cma_helper.c
+++ b/drivers/gpu/drm/drm_gem_cma_helper.c
@@ -253,8 +253,17 @@ static int drm_gem_cma_mmap_obj(struct drm_gem_cma_object *cma_obj,
{
int ret;
- ret = remap_pfn_range(vma, vma->vm_start, cma_obj->paddr >> PAGE_SHIFT,
- vma->vm_end - vma->vm_start, vma->vm_page_prot);
+ /*
+ * Clear the VM_PFNMAP flag that was set by drm_gem_mmap(), and set the
+ * vm_pgoff (used as a fake buffer offset by DRM) to 0 as we want to map
+ * the whole buffer.
+ */
+ vma->vm_flags &= ~VM_PFNMAP;
+ vma->vm_pgoff = 0;
+
+ ret = dma_mmap_writecombine(cma_obj->base.dev->dev, vma,
+ cma_obj->vaddr, cma_obj->paddr,
+ vma->vm_end - vma->vm_start);
if (ret)
drm_gem_vm_close(vma);
--
Regards,
Laurent Pinchart
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/dri-devel
