Re: [PATCH] drm/amdgpu/vcn: reset fw_shared when VCPU buffers corrupted on vcn v4.0.3

[Christian König <christian.koenig@xxxxxxx>]

Am 20.11.24 um 13:34 schrieb Xiang Liu:
> It is not necessarily corrupted. When there is RAS fatal error, device
> memory access is blocked. Hence vcpu bo cannot be saved to system memory
> as in a regular suspend sequence before going for reset. In other full
> device reset cases, that gets saved and restored during resume.
>
> v2: Remove redundant code like vcn_v4_0 did
> v2: Refine commit message
> v3: Drop the volatile
> v3: Refine commit message
>
> Signed-off-by: Xiang Liu <xiang.liu@xxxxxxx>

Acked-by: Christian König <christian.koenig@xxxxxxx>

> ---
>   drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c | 30 ++++++++++++++++++-------
>   1 file changed, 22 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c
> index d011e4678ca1..c678631c6887 100644
> --- a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c
> +++ b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c
> @@ -123,6 +123,20 @@ static int vcn_v4_0_3_early_init(struct amdgpu_ip_block *ip_block)
>   	return amdgpu_vcn_early_init(adev);
>   }
>   
> +static int vcn_v4_0_3_fw_shared_init(struct amdgpu_device *adev, int inst_idx)
> +{
> +	struct amdgpu_vcn4_fw_shared *fw_shared;
> +
> +	fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
> +	fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
> +	fw_shared->sq.is_enabled = 1;
> +
> +	if (amdgpu_vcnfw_log)
> +		amdgpu_vcn_fwlog_init(&adev->vcn.inst[inst_idx]);
> +
> +	return 0;
> +}
> +
>   /**
>    * vcn_v4_0_3_sw_init - sw init for VCN block
>    *
> @@ -155,8 +169,6 @@ static int vcn_v4_0_3_sw_init(struct amdgpu_ip_block *ip_block)
>   		return r;
>   
>   	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
> -		volatile struct amdgpu_vcn4_fw_shared *fw_shared;
> -
>   		vcn_inst = GET_INST(VCN, i);
>   
>   		ring = &adev->vcn.inst[i].ring_enc[0];
> @@ -179,12 +191,7 @@ static int vcn_v4_0_3_sw_init(struct amdgpu_ip_block *ip_block)
>   		if (r)
>   			return r;
>   
> -		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
> -		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
> -		fw_shared->sq.is_enabled = true;
> -
> -		if (amdgpu_vcnfw_log)
> -			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
> +		vcn_v4_0_3_fw_shared_init(adev, i);
>   	}
>   
>   	if (amdgpu_sriov_vf(adev)) {
> @@ -280,6 +287,8 @@ static int vcn_v4_0_3_hw_init(struct amdgpu_ip_block *ip_block)
>   		}
>   	} else {
>   		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
> +			struct amdgpu_vcn4_fw_shared *fw_shared;
> +
>   			vcn_inst = GET_INST(VCN, i);
>   			ring = &adev->vcn.inst[i].ring_enc[0];
>   
> @@ -303,6 +312,11 @@ static int vcn_v4_0_3_hw_init(struct amdgpu_ip_block *ip_block)
>   					regVCN_RB1_DB_CTRL);
>   			}
>   
> +			/* Re-init fw_shared when RAS fatal error occurred */
> +			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
> +			if (!fw_shared->sq.is_enabled)
> +				vcn_v4_0_3_fw_shared_init(adev, i);
> +
>   			r = amdgpu_ring_test_helper(ring);
>   			if (r)
>   				return r;