diff options
| author | Victor Skvortsov <victor.skvortsov@amd.com> | 2024-10-30 10:18:00 -0400 |
|---|---|---|
| committer | Alex Deucher <alexander.deucher@amd.com> | 2024-11-11 11:55:42 -0500 |
| commit | 84a2947ecc85c67f433f2cc2186e54cdb9047b61 (patch) | |
| tree | 4671794bdd9310925a70b1a967cdf0047ec9d4fa /drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | |
| parent | 907fec2dfd061ca422d8b121f4af1b6062e098ba (diff) | |
| download | lwn-84a2947ecc85c67f433f2cc2186e54cdb9047b61.tar.gz lwn-84a2947ecc85c67f433f2cc2186e54cdb9047b61.zip | |
drm/amdgpu: Implement virt req_ras_err_count
Enable RAS late init if VF RAS Telemetry is supported.
When enabled, the VF can use this interface to query total
RAS error counts from the host.
The VF FB access may abruptly end due to a fatal error,
therefore the VF must cache and sanitize the input.
The Host allows 15 Telemetry messages every 60 seconds, afterwhich
the host will ignore any more in-coming telemetry messages. The VF will
rate limit its msg calling to once every 5 seconds (12 times in 60 seconds).
While the VF is rate limited, it will continue to report the last
good cached data.
v2: Flip generate report & update statistics order for VF
Signed-off-by: Victor Skvortsov <victor.skvortsov@amd.com>
Acked-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Zhigang Luo <zhigang.luo@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 136 |
1 files changed, 136 insertions, 0 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c index 53297c40f09c..c704e9803e11 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c @@ -524,6 +524,8 @@ static int amdgpu_virt_read_pf2vf_data(struct amdgpu_device *adev) adev->unique_id = ((struct amd_sriov_msg_pf2vf_info *)pf2vf_info)->uuid; adev->virt.ras_en_caps.all = ((struct amd_sriov_msg_pf2vf_info *)pf2vf_info)->ras_en_caps.all; + adev->virt.ras_telemetry_en_caps.all = + ((struct amd_sriov_msg_pf2vf_info *)pf2vf_info)->ras_telemetry_en_caps.all; break; default: dev_err(adev->dev, "invalid pf2vf version: 0x%x\n", pf2vf_info->version); @@ -704,6 +706,8 @@ void amdgpu_virt_exchange_data(struct amdgpu_device *adev) adev->virt.fw_reserve.p_vf2pf = (struct amd_sriov_msg_vf2pf_info_header *) (adev->mman.fw_vram_usage_va + (AMD_SRIOV_MSG_VF2PF_OFFSET_KB << 10)); + adev->virt.fw_reserve.ras_telemetry = + (adev->mman.fw_vram_usage_va + (AMD_SRIOV_MSG_RAS_TELEMETRY_OFFSET_KB << 10)); } else if (adev->mman.drv_vram_usage_va) { adev->virt.fw_reserve.p_pf2vf = (struct amd_sriov_msg_pf2vf_info_header *) @@ -711,6 +715,8 @@ void amdgpu_virt_exchange_data(struct amdgpu_device *adev) adev->virt.fw_reserve.p_vf2pf = (struct amd_sriov_msg_vf2pf_info_header *) (adev->mman.drv_vram_usage_va + (AMD_SRIOV_MSG_VF2PF_OFFSET_KB << 10)); + adev->virt.fw_reserve.ras_telemetry = + (adev->mman.drv_vram_usage_va + (AMD_SRIOV_MSG_RAS_TELEMETRY_OFFSET_KB << 10)); } amdgpu_virt_read_pf2vf_data(adev); @@ -1197,3 +1203,133 @@ bool amdgpu_virt_get_ras_capability(struct amdgpu_device *adev) return true; } + +static inline enum amd_sriov_ras_telemetry_gpu_block +amdgpu_ras_block_to_sriov(struct amdgpu_device *adev, enum amdgpu_ras_block block) { + switch (block) { + case AMDGPU_RAS_BLOCK__UMC: + return RAS_TELEMETRY_GPU_BLOCK_UMC; + case AMDGPU_RAS_BLOCK__SDMA: + return RAS_TELEMETRY_GPU_BLOCK_SDMA; + case AMDGPU_RAS_BLOCK__GFX: + return RAS_TELEMETRY_GPU_BLOCK_GFX; + case AMDGPU_RAS_BLOCK__MMHUB: + return RAS_TELEMETRY_GPU_BLOCK_MMHUB; + case AMDGPU_RAS_BLOCK__ATHUB: + return RAS_TELEMETRY_GPU_BLOCK_ATHUB; + case AMDGPU_RAS_BLOCK__PCIE_BIF: + return RAS_TELEMETRY_GPU_BLOCK_PCIE_BIF; + case AMDGPU_RAS_BLOCK__HDP: + return RAS_TELEMETRY_GPU_BLOCK_HDP; + case AMDGPU_RAS_BLOCK__XGMI_WAFL: + return RAS_TELEMETRY_GPU_BLOCK_XGMI_WAFL; + case AMDGPU_RAS_BLOCK__DF: + return RAS_TELEMETRY_GPU_BLOCK_DF; + case AMDGPU_RAS_BLOCK__SMN: + return RAS_TELEMETRY_GPU_BLOCK_SMN; + case AMDGPU_RAS_BLOCK__SEM: + return RAS_TELEMETRY_GPU_BLOCK_SEM; + case AMDGPU_RAS_BLOCK__MP0: + return RAS_TELEMETRY_GPU_BLOCK_MP0; + case AMDGPU_RAS_BLOCK__MP1: + return RAS_TELEMETRY_GPU_BLOCK_MP1; + case AMDGPU_RAS_BLOCK__FUSE: + return RAS_TELEMETRY_GPU_BLOCK_FUSE; + case AMDGPU_RAS_BLOCK__MCA: + return RAS_TELEMETRY_GPU_BLOCK_MCA; + case AMDGPU_RAS_BLOCK__VCN: + return RAS_TELEMETRY_GPU_BLOCK_VCN; + case AMDGPU_RAS_BLOCK__JPEG: + return RAS_TELEMETRY_GPU_BLOCK_JPEG; + case AMDGPU_RAS_BLOCK__IH: + return RAS_TELEMETRY_GPU_BLOCK_IH; + case AMDGPU_RAS_BLOCK__MPIO: + return RAS_TELEMETRY_GPU_BLOCK_MPIO; + default: + dev_err(adev->dev, "Unsupported SRIOV RAS telemetry block 0x%x\n", block); + return RAS_TELEMETRY_GPU_BLOCK_COUNT; + } +} + +static int amdgpu_virt_cache_host_error_counts(struct amdgpu_device *adev, + struct amdsriov_ras_telemetry *host_telemetry) +{ + struct amd_sriov_ras_telemetry_error_count *tmp = NULL; + uint32_t checksum, used_size; + + checksum = host_telemetry->header.checksum; + used_size = host_telemetry->header.used_size; + + if (used_size > (AMD_SRIOV_RAS_TELEMETRY_SIZE_KB << 10)) + return 0; + + tmp = kmalloc(used_size, GFP_KERNEL); + if (!tmp) + return -ENOMEM; + + memcpy(tmp, &host_telemetry->body.error_count, used_size); + + if (checksum != amd_sriov_msg_checksum(tmp, used_size, 0, 0)) + goto out; + + memcpy(&adev->virt.count_cache, tmp, + min(used_size, sizeof(adev->virt.count_cache))); +out: + kfree(tmp); + + return 0; +} + +static int amdgpu_virt_req_ras_err_count_internal(struct amdgpu_device *adev, bool force_update) +{ + struct amdgpu_virt *virt = &adev->virt; + + /* Host allows 15 ras telemetry requests per 60 seconds. Afterwhich, the Host + * will ignore incoming guest messages. Ratelimit the guest messages to + * prevent guest self DOS. + */ + if (__ratelimit(&adev->virt.ras_telemetry_rs) || force_update) { + if (!virt->ops->req_ras_err_count(adev)) + amdgpu_virt_cache_host_error_counts(adev, + adev->virt.fw_reserve.ras_telemetry); + } + + return 0; +} + +/* Bypass ACA interface and query ECC counts directly from host */ +int amdgpu_virt_req_ras_err_count(struct amdgpu_device *adev, enum amdgpu_ras_block block, + struct ras_err_data *err_data) +{ + enum amd_sriov_ras_telemetry_gpu_block sriov_block; + + sriov_block = amdgpu_ras_block_to_sriov(adev, block); + + if (sriov_block >= RAS_TELEMETRY_GPU_BLOCK_COUNT || + !amdgpu_sriov_ras_telemetry_block_en(adev, sriov_block)) + return -EOPNOTSUPP; + + /* Host Access may be lost during reset, just return last cached data. */ + if (down_read_trylock(&adev->reset_domain->sem)) { + amdgpu_virt_req_ras_err_count_internal(adev, false); + up_read(&adev->reset_domain->sem); + } + + err_data->ue_count = adev->virt.count_cache.block[sriov_block].ue_count; + err_data->ce_count = adev->virt.count_cache.block[sriov_block].ce_count; + err_data->de_count = adev->virt.count_cache.block[sriov_block].de_count; + + return 0; +} + +int amdgpu_virt_ras_telemetry_post_reset(struct amdgpu_device *adev) +{ + unsigned long ue_count, ce_count; + + if (amdgpu_sriov_ras_telemetry_en(adev)) { + amdgpu_virt_req_ras_err_count_internal(adev, true); + amdgpu_ras_query_error_count(adev, &ce_count, &ue_count, NULL); + } + + return 0; +} |