V2: not to add extra read-back in vcn_v4_0_5_start as there is a read-back already. New comment for better understanding.
On VCN v4.0.5 there is a race condition where the WPTR is not updated after starting from idle when doorbell is used. The read-back of regVCN_RB1_DB_CTRL register after written is to ensure the doorbell_index is updated before it can work properly.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/12528 Cc: stable@vger.kernel.org
Signed-off-by: David (Ming Qiang) Wu David.Wu3@amd.com Reviewed-by: Mario Limonciello mario.limonciello@amd.com Tested-by: Mario Limonciello mario.limonciello@amd.com Reviewed-by: Alex Deucher alexander.deucher@amd.com --- drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c | 4 ++++ 1 file changed, 4 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c index ed00d35039c1..e55b76d71367 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c @@ -1034,6 +1034,10 @@ static int vcn_v4_0_5_start_dpg_mode(struct amdgpu_vcn_inst *vinst, ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT | VCN_RB1_DB_CTRL__EN_MASK);
+ /* Keeping one read-back to ensure all register writes are done, otherwise + * it may introduce race conditions */ + RREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL); + return 0; }