Am 11.04.2018 um 08:26 schrieb Huang Rui:
On Tue, Apr 10, 2018 at 04:59:55PM -0400, Sinan Kaya wrote:
Code is expecing to observe the same number of buffers returned from
dma_map_sg() function compared to sg_alloc_table_from_pages(). This
doesn't hold true universally especially for systems with IOMMU.

IOMMU driver tries to combine buffers into a single DMA address as much
as it can. The right thing is to tell the DMA layer how much combining
IOMMU can do.

Signed-off-by: Sinan Kaya <>
Sinan, I guess Christian's suggestion is to add amdgpu_device_init function
like here:

Yes, exactly. Looks like Sinan just tried to place the call next to pci_set_dma_mask()/pci_set_consistent_dma_mask().

Not necessary a bad idea, but in this case not optimal.

Sinan please refine your patch as Rui suggested and resend.


diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
index 0e798b3..9b96771 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -2339,6 +2339,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
         /* init the mode config */

+       dma_set_max_seg_size(adev->dev, PAGE_SIZE);
         r = amdgpu_device_ip_init(adev);
         if (r) {
                 /* failed in exclusive mode due to timeout */

After that, we don't do it in each generation.

dri-devel mailing list

amd-gfx mailing list

Reply via email to