Re: [PATCH] drm/amd/amdgpu: Avoid writing GMC registers under sriov in gmc9

[Christian König]

Am 02.11.21 um 11:25 schrieb YuBiao Wang:
[Why]
For Vega10, disabling gart of gfxhub and mmhub could mess up KIQ and PSP under 
sriov mode, and lead to DMAR on host side.
[How]
Do not call gmc_gart_disable under sriov but keep vram_unpin to avoid
pin_count leak.

NAK, if you want to do this you should probably avoid the write in the 
low level hardware callbacks and not here.

Christian.


Signed-off-by: YuBiao Wang <yubiao.w...@amd.com>
---
  drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c 
b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index cb82404df534..365059a20ae8 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -1793,14 +1793,13 @@ static void gmc_v9_0_gart_disable(struct amdgpu_device 
*adev)
  {
        adev->gfxhub.funcs->gart_disable(adev);
        adev->mmhub.funcs->gart_disable(adev);
-       amdgpu_gart_table_vram_unpin(adev);
  }
  
  static int gmc_v9_0_hw_fini(void *handle)
  {
        struct amdgpu_device *adev = (struct amdgpu_device *)handle;
  
-	gmc_v9_0_gart_disable(adev);
+       amdgpu_gart_table_vram_unpin(adev);
  
  	if (amdgpu_sriov_vf(adev)) {
                /* full access mode, so don't touch any GMC register */
@@ -1808,6 +1807,7 @@ static int gmc_v9_0_hw_fini(void *handle)
                return 0;
        }
  
+	gmc_v9_0_gart_disable(adev);
        amdgpu_irq_put(adev, &adev->gmc.ecc_irq, 0);
        amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);