From e17768691dd8d0664413de3123621daa0504054c Mon Sep 17 00:00:00 2001 From: Bokun Zhang Date: Wed, 29 Nov 2023 19:11:22 -0500 Subject: [PATCH] drm/amd/amdgpu: SRIOV full reset issue with VCN - After a full reset, VF's FB will be cleaned. This includes the VCN's fw_shared memory. However, there is no suspend-resume routine for SRIOV VF. Therefore, the data in the fw_shared memory will be lost forever and it causes engine hang later on. We must repopulate the data in fw_shared during SRIOV hw_init Signed-off-by: Bokun Zhang Reviewed-by: Emily Deng Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c index 28b7f96b42e59..169ed400ee7b7 100644 --- a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c @@ -1280,6 +1280,9 @@ static int vcn_v4_0_start_sriov(struct amdgpu_device *adev) if (adev->vcn.harvest_config & (1 << i)) continue; + // Must re/init fw_shared at beginning + vcn_v4_0_fw_shared_init(adev, i); + table_size = 0; MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i, -- 2.30.2