[virtio-dev] [PATCH 2/2] drm/virtio: add iommu support.

[Jiandi An]

On 09/12/2018 02:25 AM, Gerd Hoffmann wrote:
>   Hi,
> 
>> I attempted to fix it in the ttm layer and here is the discussion
>> https://lore.kernel.org/lkml/b44280d7-eb13-0996-71f5-3fbdeb466801@amd.com/
>>
>> The ttm maintainer Christian is suggesting to map and set ttm->pages as decrypted
>> right after ttm->pages are allocated.
>>
>> Just checking with you guys maybe there is a better way to handle this in
>> the virtio gpu layer instead of the common ttm layer.  Could you guys shine some
>> light on this?  Thanks.
> 
> I think the tty layer is the right place to fix this.  virtio just calls
> down to ttm for mappings.  I think virtio should just hint to ttm using
> a flag in some struct, probably ttm_buffer_object or
> ttm_mem_type_manager, that the objects need decrypted mappings.
> 
> cheers,
>   Gerd
> 

I tested this patch with non SEV guest.  It gives a blank black screen if booting
with swiotlb=force.  dma_sync is missing if dma op uses swiotlb as bounce
buffer.  I tried to put a dma_sync_sg_for_device() on virtio_gpu_object obj->pages-sgl
before VIRTIO_GPU_CMD_TRANSFER_TO_HOST_2D is sent.  This fixes the kernel console path.
Once display manger is kicked off for example (sudo systemctl start lightdm.service) and
resource id 3 gets created from user space down, it still gives a blank black screen.

In addition, I added dma_sync_sg_for_device() before sending VIRTIO_GPU_CMD_RESOURCE_FLUSH
and VIRTIO_GPU_CMD_SET_SCANOUT, still blank black screen after display manger is kicked off.

Do you know which path I'm still missing as far as VIRTIO_GPU_CMD goes?

Thanks
- Jiandi