diff options
Diffstat (limited to 'drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c')
| -rw-r--r-- | drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c | 164 |
1 files changed, 151 insertions, 13 deletions
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c index 7e1941cf1796..1b96158b63bf 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c +++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c @@ -45,6 +45,7 @@ #include <linux/pci.h> #include "amdgpu_ras.h" #include "amdgpu_mca.h" +#include "amdgpu_aca.h" #include "smu_cmn.h" #include "mp/mp_13_0_6_offset.h" #include "mp/mp_13_0_6_sh_mask.h" @@ -1438,7 +1439,10 @@ static int smu_v13_0_6_irq_process(struct amdgpu_device *adev, entry->src_data[1]); schedule_work(&smu->throttling_logging_work); } - + break; + default: + dev_dbg(adev->dev, "Unhandled context id %d from client:%d!\n", + ctxid, client_id); break; } } @@ -2547,18 +2551,22 @@ static int mca_umc_mca_get_err_count(const struct mca_ras_info *mca_ras, struct enum amdgpu_mca_error_type type, struct mca_bank_entry *entry, uint32_t *count) { uint64_t status0; + uint32_t ext_error_code; + uint32_t odecc_err_cnt; status0 = entry->regs[MCA_REG_IDX_STATUS]; + ext_error_code = MCA_REG__STATUS__ERRORCODEEXT(status0); + odecc_err_cnt = MCA_REG__MISC0__ERRCNT(entry->regs[MCA_REG_IDX_MISC0]); if (!REG_GET_FIELD(status0, MCMP1_STATUST0, Val)) { *count = 0; return 0; } - if (type == AMDGPU_MCA_ERROR_TYPE_UE && umc_v12_0_is_uncorrectable_error(adev, status0)) - *count = 1; - else if (type == AMDGPU_MCA_ERROR_TYPE_CE && umc_v12_0_is_correctable_error(adev, status0)) - *count = 1; + if (umc_v12_0_is_deferred_error(adev, status0) || + umc_v12_0_is_uncorrectable_error(adev, status0) || + umc_v12_0_is_correctable_error(adev, status0)) + *count = (ext_error_code == 0) ? odecc_err_cnt : 1; return 0; } @@ -2857,6 +2865,143 @@ static const struct amdgpu_mca_smu_funcs smu_v13_0_6_mca_smu_funcs = { .mca_get_valid_mca_count = mca_smu_get_valid_mca_count, }; +static int aca_smu_set_debug_mode(struct amdgpu_device *adev, bool enable) +{ + struct smu_context *smu = adev->powerplay.pp_handle; + + return smu_v13_0_6_mca_set_debug_mode(smu, enable); +} + +static int smu_v13_0_6_get_valid_aca_count(struct smu_context *smu, enum aca_error_type type, u32 *count) +{ + uint32_t msg; + int ret; + + if (!count) + return -EINVAL; + + switch (type) { + case ACA_ERROR_TYPE_UE: + msg = SMU_MSG_QueryValidMcaCount; + break; + case ACA_ERROR_TYPE_CE: + msg = SMU_MSG_QueryValidMcaCeCount; + break; + default: + return -EINVAL; + } + + ret = smu_cmn_send_smc_msg(smu, msg, count); + if (ret) { + *count = 0; + return ret; + } + + return 0; +} + +static int aca_smu_get_valid_aca_count(struct amdgpu_device *adev, + enum aca_error_type type, u32 *count) +{ + struct smu_context *smu = adev->powerplay.pp_handle; + int ret; + + switch (type) { + case ACA_ERROR_TYPE_UE: + case ACA_ERROR_TYPE_CE: + ret = smu_v13_0_6_get_valid_aca_count(smu, type, count); + break; + default: + ret = -EINVAL; + break; + } + + return ret; +} + +static int __smu_v13_0_6_aca_bank_dump(struct smu_context *smu, enum aca_error_type type, + int idx, int offset, u32 *val) +{ + uint32_t msg, param; + + switch (type) { + case ACA_ERROR_TYPE_UE: + msg = SMU_MSG_McaBankDumpDW; + break; + case ACA_ERROR_TYPE_CE: + msg = SMU_MSG_McaBankCeDumpDW; + break; + default: + return -EINVAL; + } + + param = ((idx & 0xffff) << 16) | (offset & 0xfffc); + + return smu_cmn_send_smc_msg_with_param(smu, msg, param, (uint32_t *)val); +} + +static int smu_v13_0_6_aca_bank_dump(struct smu_context *smu, enum aca_error_type type, + int idx, int offset, u32 *val, int count) +{ + int ret, i; + + if (!val) + return -EINVAL; + + for (i = 0; i < count; i++) { + ret = __smu_v13_0_6_aca_bank_dump(smu, type, idx, offset + (i << 2), &val[i]); + if (ret) + return ret; + } + + return 0; +} + +static int aca_bank_read_reg(struct amdgpu_device *adev, enum aca_error_type type, + int idx, int reg_idx, u64 *val) +{ + struct smu_context *smu = adev->powerplay.pp_handle; + u32 data[2] = {0, 0}; + int ret; + + if (!val || reg_idx >= ACA_REG_IDX_COUNT) + return -EINVAL; + + ret = smu_v13_0_6_aca_bank_dump(smu, type, idx, reg_idx * 8, data, ARRAY_SIZE(data)); + if (ret) + return ret; + + *val = (u64)data[1] << 32 | data[0]; + + dev_dbg(adev->dev, "mca read bank reg: type:%s, index: %d, reg_idx: %d, val: 0x%016llx\n", + type == ACA_ERROR_TYPE_UE ? "UE" : "CE", idx, reg_idx, *val); + + return 0; +} + +static int aca_smu_get_valid_aca_bank(struct amdgpu_device *adev, + enum aca_error_type type, int idx, struct aca_bank *bank) +{ + int i, ret, count; + + count = min_t(int, 16, ARRAY_SIZE(bank->regs)); + for (i = 0; i < count; i++) { + ret = aca_bank_read_reg(adev, type, idx, i, &bank->regs[i]); + if (ret) + return ret; + } + + return 0; +} + +static const struct aca_smu_funcs smu_v13_0_6_aca_smu_funcs = { + .max_ue_bank_count = 12, + .max_ce_bank_count = 12, + .set_debug_mode = aca_smu_set_debug_mode, + .get_valid_aca_count = aca_smu_get_valid_aca_count, + .get_valid_aca_bank = aca_smu_get_valid_aca_bank, +}; + static int smu_v13_0_6_select_xgmi_plpd_policy(struct smu_context *smu, enum pp_xgmi_plpd_mode mode) { @@ -2895,13 +3040,6 @@ static int smu_v13_0_6_select_xgmi_plpd_policy(struct smu_context *smu, return ret; } -static ssize_t smu_v13_0_6_get_ecc_info(struct smu_context *smu, - void *table) -{ - /* Support ecc info by default */ - return 0; -} - static const struct pptable_funcs smu_v13_0_6_ppt_funcs = { /* init dpm */ .get_allowed_feature_mask = smu_v13_0_6_get_allowed_feature_mask, @@ -2956,7 +3094,6 @@ static const struct pptable_funcs smu_v13_0_6_ppt_funcs = { .i2c_init = smu_v13_0_6_i2c_control_init, .i2c_fini = smu_v13_0_6_i2c_control_fini, .send_hbm_bad_pages_num = smu_v13_0_6_smu_send_hbm_bad_page_num, - .get_ecc_info = smu_v13_0_6_get_ecc_info, }; void smu_v13_0_6_set_ppt_funcs(struct smu_context *smu) @@ -2969,4 +3106,5 @@ void smu_v13_0_6_set_ppt_funcs(struct smu_context *smu) smu->smc_driver_if_version = SMU13_0_6_DRIVER_IF_VERSION; smu_v13_0_set_smu_mailbox_registers(smu); amdgpu_mca_smu_init_funcs(smu->adev, &smu_v13_0_6_mca_smu_funcs); + amdgpu_aca_set_smu_funcs(smu->adev, &smu_v13_0_6_aca_smu_funcs); } |