Re: [PATCH] drm/amdgpu: skip PF-only programming on the VF

[Lazar, Lijo]

On 29-Sep-26 11:15 AM, Wang, YuBiao (Ethan) wrote:

AMD General

Hi Lijo,

Host driver will program the VF copies for FB_LOCATION and SYSTEM_APERTURE 
registers via PSP interface so amdgpu doesn't need to handle it under sriov.

Thanks for the clarification. The patch looks good.

Reviewed-by: Lijo Lazar <lijo.la...@amd.com>

Thanks,
Lijo

Regards & Thanks,
Yubiao


-----Original Message-----
From: Lazar, Lijo <lijo.la...@amd.com>
Sent: Tuesday, September 29, 2026 1:35 PM
To: Wang, YuBiao (Ethan) <yubiao.w...@amd.com>; amd-gfx@lists.freedesktop.org
Cc: Deucher, Alexander <alexander.deuc...@amd.com>; Zhang, Hawking 
<hawking.zh...@amd.com>
Subject: Re: [PATCH] drm/amdgpu: skip PF-only programming on the VF



On 24-Sep-26 12:31 PM, YuBiao Wang wrote:

Golden register init, the SDMA RB command switch setup, the GFXHUB
L2 protection fault controls and the guest-side FB_LOCATION
programming all touch registers a VF does not own, so leave them to the PF.

Signed-off-by: YuBiao Wang <yubiao.w...@amd.com>
---
   drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c    |  3 ++-
   drivers/gpu/drm/amd/amdgpu/gfxhub_v12_1.c | 24 +++--------------------
   drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c    |  3 ++-
   3 files changed, 7 insertions(+), 23 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c
b/drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c
index db541939555b..4f6eac30b235 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c
@@ -2665,7 +2665,8 @@ static int gfx_v12_1_hw_init(struct amdgpu_ip_block 
*ip_block)
       if (r)
               return r;

-     gfx_v12_1_init_golden_registers(adev);
+     if (!amdgpu_sriov_vf(adev))
+             gfx_v12_1_init_golden_registers(adev);

       gfx_v12_1_constants_init(adev);

diff --git a/drivers/gpu/drm/amd/amdgpu/gfxhub_v12_1.c
b/drivers/gpu/drm/amd/amdgpu/gfxhub_v12_1.c
index 7fa451b373fc..b13967a1c8fb 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfxhub_v12_1.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfxhub_v12_1.c
@@ -482,27 +482,6 @@ static void gfxhub_v12_1_xcc_program_invalidation(struct 
amdgpu_device *adev,
   static int gfxhub_v12_1_xcc_gart_enable(struct amdgpu_device *adev,
                                       uint32_t xcc_mask)
   {
-     uint32_t i;
-
-     if (amdgpu_sriov_vf(adev)) {
-             /* GCMC_VM_FB_LOCATION_BASE/TOP are VF copy registers
-              * VBIO post does not program them at boot up phase
-              * Need driver to program them from guest side */
-             for_each_inst(i, xcc_mask) {
-                     WREG32_SOC15(GC, GET_INST(GC, i),
-                                  regGCMC_VM_FB_LOCATION_BASE_LO32,
-                                  lower_32_bits(adev->gmc.vram_start >> 24));
-                     WREG32_SOC15(GC, GET_INST(GC, i),
-                                  regGCMC_VM_FB_LOCATION_BASE_HI32,
-                                  upper_32_bits(adev->gmc.vram_start >> 24));
-                     WREG32_SOC15(GC, GET_INST(GC, i),
-                                  regGCMC_VM_FB_LOCATION_TOP_LO32,
-                                  lower_32_bits(adev->gmc.vram_end >> 24));
-                     WREG32_SOC15(GC, GET_INST(GC, i),
-                                  regGCMC_VM_FB_LOCATION_TOP_HI32,
-                                  upper_32_bits(adev->gmc.vram_end >> 24));

These have VF copies, and likely needs programming as in 
gfxhub_v12_1_xcc_init_system_aperture_regs. Could you check?

Thanks,
Lijo

-             }
-     }
       /* GART Enable. */
       gfxhub_v12_1_xcc_init_gart_aperture_regs(adev, xcc_mask);
       gfxhub_v12_1_xcc_init_system_aperture_regs(adev, xcc_mask); @@
-576,6 +555,9 @@ static void gfxhub_v12_1_xcc_set_fault_enable_default(struct 
amdgpu_device *adev
       u32 tmp;
       int i;

+     if (amdgpu_sriov_vf(adev))
+             return;
+
       for_each_inst(i, xcc_mask) {
               tmp = RREG32_SOC15(GC, GET_INST(GC, i),
                                  regGCVM_L2_PROTECTION_FAULT_CNTL_LO32);
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c
b/drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c
index 321c67e01443..d1e09acc684a 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c
@@ -1397,7 +1397,8 @@ static int sdma_v7_1_hw_init(struct
amdgpu_ip_block *ip_block)

       inst_mask = GENMASK(adev->sdma.num_instances - 1, 0);

-     sdma_v7_1_rb_cmd_switch(adev, inst_mask);
+     if (!amdgpu_sriov_vf(adev))
+             sdma_v7_1_rb_cmd_switch(adev, inst_mask);

       r = sdma_v7_1_inst_start(adev, inst_mask);
       if (r)