#include "amdgpu_xgmi.h"
#include <linux/pci.h>
#include "amdgpu_ras.h"
+#include "amdgpu_mca.h"
#include "smu_cmn.h"
#include "mp/mp_13_0_6_offset.h"
#include "mp/mp_13_0_6_sh_mask.h"
#define SMU_13_0_6_DSCLK_THRESHOLD 100
+#define MCA_BANK_IPID(_ip, _hwid, _type) \
+ [AMDGPU_MCA_IP_##_ip] = { .hwid = _hwid, .mcatype = _type, }
+
+enum mca_reg_idx {
+ MCA_REG_IDX_CONTROL = 0,
+ MCA_REG_IDX_STATUS = 1,
+ MCA_REG_IDX_ADDR = 2,
+ MCA_REG_IDX_MISC0 = 3,
+ MCA_REG_IDX_CONFIG = 4,
+ MCA_REG_IDX_IPID = 5,
+ MCA_REG_IDX_SYND = 6,
+ MCA_REG_IDX_COUNT = 16,
+};
+
+struct mca_bank_ipid {
+ enum amdgpu_mca_ip ip;
+ uint16_t hwid;
+ uint16_t mcatype;
+};
+
+struct mca_ras_info {
+ enum amdgpu_ras_block blkid;
+ enum amdgpu_mca_ip ip;
+ int *err_code_array;
+ int err_code_count;
+ int (*get_err_count)(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
+ enum amdgpu_mca_error_type type, int idx, uint32_t *count);
+};
+
static const struct cmn2asic_msg_mapping smu_v13_0_6_message_map[SMU_MSG_MAX_COUNT] = {
MSG_MAP(TestMessage, PPSMC_MSG_TestMessage, 0),
MSG_MAP(GetSmuVersion, PPSMC_MSG_GetSmuVersion, 1),
return 0;
}
+static int mca_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_mca_count(struct smu_context *smu, enum amdgpu_mca_error_type type, uint32_t *count)
+{
+ uint32_t msg;
+ int ret;
+
+ if (!count)
+ return -EINVAL;
+
+ switch (type) {
+ case AMDGPU_MCA_ERROR_TYPE_UE:
+ msg = SMU_MSG_QueryValidMcaCount;
+ break;
+ case AMDGPU_MCA_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 __smu_v13_0_6_mca_dump_bank(struct smu_context *smu, enum amdgpu_mca_error_type type,
+ int idx, int offset, uint32_t *val)
+{
+ uint32_t msg, param;
+
+ switch (type) {
+ case AMDGPU_MCA_ERROR_TYPE_UE:
+ msg = SMU_MSG_McaBankDumpDW;
+ break;
+ case AMDGPU_MCA_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, val);
+}
+
+static int smu_v13_0_6_mca_dump_bank(struct smu_context *smu, enum amdgpu_mca_error_type type,
+ int idx, int offset, uint32_t *val, int count)
+{
+ int ret, i;
+
+ if (!val)
+ return -EINVAL;
+
+ for (i = 0; i < count; i++) {
+ ret = __smu_v13_0_6_mca_dump_bank(smu, type, idx, offset + (i << 2), &val[i]);
+ if (ret)
+ return ret;
+ }
+
+ return 0;
+}
+
+static const struct mca_bank_ipid smu_v13_0_6_mca_ipid_table[AMDGPU_MCA_IP_COUNT] = {
+ MCA_BANK_IPID(UMC, 0x96, 0x0),
+ MCA_BANK_IPID(SMU, 0x01, 0x1),
+ MCA_BANK_IPID(MP5, 0x01, 0x2),
+};
+
+static void mca_bank_entry_info_decode(struct mca_bank_entry *entry, struct mca_bank_info *info)
+{
+ uint64_t ipid = entry->regs[MCA_REG_IDX_IPID];
+ uint32_t insthi;
+
+ /* NOTE: All MCA IPID register share the same format,
+ * so the driver can share the MCMP1 register header file.
+ * */
+
+ info->hwid = REG_GET_FIELD(ipid, MCMP1_IPIDT0, HardwareID);
+ info->mcatype = REG_GET_FIELD(ipid, MCMP1_IPIDT0, McaType);
+
+ insthi = REG_GET_FIELD(ipid, MCMP1_IPIDT0, InstanceIdHi);
+ info->aid = ((insthi >> 2) & 0x03);
+ info->socket_id = insthi & 0x03;
+}
+
+static int mca_bank_read_reg(struct amdgpu_device *adev, enum amdgpu_mca_error_type type,
+ int idx, int reg_idx, uint64_t *val)
+{
+ struct smu_context *smu = adev->powerplay.pp_handle;
+ uint32_t data[2] = {0, 0};
+ int ret;
+
+ if (!val || reg_idx >= MCA_REG_IDX_COUNT)
+ return -EINVAL;
+
+ ret = smu_v13_0_6_mca_dump_bank(smu, type, idx, reg_idx * 8, data, ARRAY_SIZE(data));
+ if (ret)
+ return ret;
+
+ *val = (uint64_t)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 == AMDGPU_MCA_ERROR_TYPE_UE ? "UE" : "CE", idx, reg_idx, *val);
+
+ return 0;
+}
+
+static int mca_get_mca_entry(struct amdgpu_device *adev, enum amdgpu_mca_error_type type,
+ int idx, struct mca_bank_entry *entry)
+{
+ int i, ret;
+
+ /* NOTE: populated all mca register by default */
+ for (i = 0; i < ARRAY_SIZE(entry->regs); i++) {
+ ret = mca_bank_read_reg(adev, type, idx, i, &entry->regs[i]);
+ if (ret)
+ return ret;
+ }
+
+ if (ret)
+ return ret;
+
+ entry->idx = idx;
+ entry->type = type;
+
+ mca_bank_entry_info_decode(entry, &entry->info);
+
+ return 0;
+}
+
+static int mca_decode_mca_ipid(struct amdgpu_device *adev, enum amdgpu_mca_error_type type, int idx, int *ip)
+{
+ const struct mca_bank_ipid *ipid;
+ uint64_t val;
+ uint16_t hwid, mcatype;
+ int i, ret;
+
+ ret = mca_bank_read_reg(adev, type, idx, MCA_REG_IDX_IPID, &val);
+ if (ret)
+ return ret;
+
+ hwid = REG_GET_FIELD(val, MCMP1_IPIDT0, HardwareID);
+ mcatype = REG_GET_FIELD(val, MCMP1_IPIDT0, McaType);
+
+ if (hwid) {
+ for (i = 0; i < ARRAY_SIZE(smu_v13_0_6_mca_ipid_table); i++) {
+ ipid = &smu_v13_0_6_mca_ipid_table[i];
+
+ if (!ipid->hwid)
+ continue;
+
+ if (ipid->hwid == hwid && ipid->mcatype == mcatype) {
+ *ip = i;
+ return 0;
+ }
+ }
+ }
+
+ *ip = AMDGPU_MCA_IP_UNKNOW;
+
+ return 0;
+}
+
+static int mca_normal_mca_get_err_count(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
+ enum amdgpu_mca_error_type type, int idx, uint32_t *count)
+{
+ uint64_t status0;
+ int ret;
+
+ ret = mca_bank_read_reg(adev, type, idx, MCA_REG_IDX_STATUS, &status0);
+ if (ret)
+ return ret;
+
+ if (REG_GET_FIELD(status0, MCMP1_STATUST0, Val))
+ *count = 1;
+ else
+ *count = 0;
+
+ return 0;
+}
+
+static bool mca_smu_check_error_code(struct amdgpu_device *adev, const struct mca_ras_info *mca_ras,
+ uint32_t errcode)
+{
+ int i;
+
+ if (!mca_ras->err_code_count || !mca_ras->err_code_array)
+ return true;
+
+ for (i = 0; i < mca_ras->err_code_count; i++) {
+ if (errcode == mca_ras->err_code_array[i])
+ return true;
+ }
+
+ return false;
+}
+
+static int mca_mp5_mca_get_err_count(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
+ enum amdgpu_mca_error_type type, int idx, uint32_t *count)
+{
+ uint64_t status0 = 0, misc0 = 0;
+ uint32_t errcode;
+ int ret;
+
+ if (mca_ras->ip != AMDGPU_MCA_IP_MP5)
+ return -EINVAL;
+
+ ret = mca_bank_read_reg(adev, type, idx, MCA_REG_IDX_STATUS, &status0);
+ if (ret)
+ return ret;
+
+ if (!REG_GET_FIELD(status0, MCMP1_STATUST0, Val)) {
+ *count = 0;
+ return 0;
+ }
+
+ errcode = REG_GET_FIELD(status0, MCMP1_STATUST0, ErrorCode);
+ if (!mca_smu_check_error_code(adev, mca_ras, errcode))
+ return 0;
+
+ if (type == AMDGPU_MCA_ERROR_TYPE_UE &&
+ REG_GET_FIELD(status0, MCMP1_STATUST0, UC) == 1 &&
+ REG_GET_FIELD(status0, MCMP1_STATUST0, PCC) == 1) {
+ if (count)
+ *count = 1;
+ return 0;
+ }
+
+ ret = mca_bank_read_reg(adev, type, idx, MCA_REG_IDX_MISC0, &misc0);
+ if (ret)
+ return ret;
+
+ if (count)
+ *count = REG_GET_FIELD(misc0, MCMP1_MISC0T0, ErrCnt);
+
+ return 0;
+}
+
+static int mca_smu_mca_get_err_count(const struct mca_ras_info *mca_ras, struct amdgpu_device *adev,
+ enum amdgpu_mca_error_type type, int idx, uint32_t *count)
+{
+ uint64_t status0 = 0, misc0 = 0;
+ uint32_t errcode;
+ int ret;
+
+ if (mca_ras->ip != AMDGPU_MCA_IP_SMU)
+ return -EINVAL;
+
+ ret = mca_bank_read_reg(adev, type, idx, MCA_REG_IDX_STATUS, &status0);
+ if (ret)
+ return ret;
+
+ if (!REG_GET_FIELD(status0, MCMP1_STATUST0, Val)) {
+ *count = 0;
+ return 0;
+ }
+
+ errcode = REG_GET_FIELD(status0, MCMP1_STATUST0, ErrorCode);
+ if (!mca_smu_check_error_code(adev, mca_ras, errcode))
+ return 0;
+
+ if (type == AMDGPU_MCA_ERROR_TYPE_UE &&
+ REG_GET_FIELD(status0, MCMP1_STATUST0, UC) == 1 &&
+ REG_GET_FIELD(status0, MCMP1_STATUST0, PCC) == 1) {
+ if (count)
+ *count = 1;
+ return 0;
+ }
+
+ ret = mca_bank_read_reg(adev, type, idx, MCA_REG_IDX_MISC0, &misc0);
+ if (ret)
+ return ret;
+
+ if (count)
+ *count = REG_GET_FIELD(misc0, MCMP1_MISC0T0, ErrCnt);
+
+ return 0;
+}
+
+static int sdma_err_codes[] = { CODE_SDMA0, CODE_SDMA1, CODE_SDMA2, CODE_SDMA3 };
+static int mmhub_err_codes[] = {
+ CODE_DAGB0, CODE_DAGB0 + 1, CODE_DAGB0 + 2, CODE_DAGB0 + 3, CODE_DAGB0 + 4, /* DAGB0-4 */
+ CODE_EA0, CODE_EA0 + 1, CODE_EA0 + 2, CODE_EA0 + 3, CODE_EA0 + 4, /* MMEA0-4*/
+ CODE_VML2, CODE_VML2_WALKER, CODE_MMCANE,
+};
+
+static const struct mca_ras_info mca_ras_table[] = {
+ {
+ .blkid = AMDGPU_RAS_BLOCK__UMC,
+ .ip = AMDGPU_MCA_IP_UMC,
+ .get_err_count = mca_normal_mca_get_err_count,
+ }, {
+ .blkid = AMDGPU_RAS_BLOCK__GFX,
+ .ip = AMDGPU_MCA_IP_MP5,
+ .get_err_count = mca_mp5_mca_get_err_count,
+ }, {
+ .blkid = AMDGPU_RAS_BLOCK__SDMA,
+ .ip = AMDGPU_MCA_IP_SMU,
+ .err_code_array = sdma_err_codes,
+ .err_code_count = ARRAY_SIZE(sdma_err_codes),
+ .get_err_count = mca_smu_mca_get_err_count,
+ }, {
+ .blkid = AMDGPU_RAS_BLOCK__MMHUB,
+ .ip = AMDGPU_MCA_IP_SMU,
+ .err_code_array = mmhub_err_codes,
+ .err_code_count = ARRAY_SIZE(mmhub_err_codes),
+ .get_err_count = mca_smu_mca_get_err_count,
+ },
+};
+
+static const struct mca_ras_info *mca_get_mca_ras_info(struct amdgpu_device *adev, enum amdgpu_ras_block blkid)
+{
+ int i;
+
+ for (i = 0; i < ARRAY_SIZE(mca_ras_table); i++) {
+ if (mca_ras_table[i].blkid == blkid)
+ return &mca_ras_table[i];
+ }
+
+ return NULL;
+}
+
+static int mca_get_valid_mca_count(struct amdgpu_device *adev, enum amdgpu_mca_error_type type, uint32_t *count)
+{
+ struct smu_context *smu = adev->powerplay.pp_handle;
+ int ret;
+
+ switch (type) {
+ case AMDGPU_MCA_ERROR_TYPE_UE:
+ case AMDGPU_MCA_ERROR_TYPE_CE:
+ ret = smu_v13_0_6_get_valid_mca_count(smu, type, count);
+ break;
+ default:
+ ret = -EINVAL;
+ break;
+ }
+
+ return ret;
+}
+
+static bool mca_bank_is_valid(struct amdgpu_device *adev, const struct mca_ras_info *mca_ras,
+ enum amdgpu_mca_error_type type, int idx)
+{
+ int ret, ip = AMDGPU_MCA_IP_UNKNOW;
+
+ ret = mca_decode_mca_ipid(adev, type, idx, &ip);
+ if (ret)
+ return false;
+
+ if (ip == AMDGPU_MCA_IP_UNKNOW)
+ return false;
+
+ return ip == mca_ras->ip;
+}
+
+static int mca_get_valid_mca_idx(struct amdgpu_device *adev, const struct mca_ras_info *mca_ras,
+ enum amdgpu_mca_error_type type,
+ uint32_t mca_cnt, int *idx_array, int idx_array_size)
+{
+ int i, idx_cnt = 0;
+
+ for (i = 0; i < mca_cnt; i++) {
+ if (!mca_bank_is_valid(adev, mca_ras, type, i))
+ continue;
+
+ if (idx_array) {
+ if (idx_cnt < idx_array_size)
+ idx_array[idx_cnt] = i;
+ else
+ return -EINVAL;
+ }
+
+ idx_cnt++;
+ }
+
+ return idx_cnt;
+}
+
+static int __mca_smu_get_error_count(struct amdgpu_device *adev, const struct mca_ras_info *mca_ras, enum amdgpu_mca_error_type type, uint32_t *count)
+{
+ uint32_t result, mca_cnt, total = 0;
+ int idx_array[16];
+ int i, ret, idx_cnt = 0;
+
+ ret = mca_get_valid_mca_count(adev, type, &mca_cnt);
+ if (ret)
+ return ret;
+
+ /* if valid mca bank count is 0, the driver can return 0 directly */
+ if (!mca_cnt) {
+ *count = 0;
+ return 0;
+ }
+
+ if (!mca_ras->get_err_count)
+ return -EINVAL;
+
+ idx_cnt = mca_get_valid_mca_idx(adev, mca_ras, type, mca_cnt, idx_array, ARRAY_SIZE(idx_array));
+ if (idx_cnt < 0)
+ return -EINVAL;
+
+ for (i = 0; i < idx_cnt; i++) {
+ result = 0;
+ ret = mca_ras->get_err_count(mca_ras, adev, type, idx_array[i], &result);
+ if (ret)
+ return ret;
+
+ total += result;
+ }
+
+ *count = total;
+
+ return 0;
+}
+
+static int mca_smu_get_error_count(struct amdgpu_device *adev, enum amdgpu_ras_block blk,
+ enum amdgpu_mca_error_type type, uint32_t *count)
+{
+ const struct mca_ras_info *mca_ras;
+
+ if (!count)
+ return -EINVAL;
+
+ mca_ras = mca_get_mca_ras_info(adev, blk);
+ if (!mca_ras)
+ return -EOPNOTSUPP;
+
+ return __mca_smu_get_error_count(adev, mca_ras, type, count);
+}
+
+static int __mca_smu_get_ras_mca_idx_array(struct amdgpu_device *adev, const struct mca_ras_info *mca_ras,
+ enum amdgpu_mca_error_type type, int *idx_array, int *idx_array_size)
+{
+ uint32_t mca_cnt = 0;
+ int ret, idx_cnt = 0;
+
+ ret = mca_get_valid_mca_count(adev, type, &mca_cnt);
+ if (ret)
+ return ret;
+
+ /* if valid mca bank count is 0, the driver can return 0 directly */
+ if (!mca_cnt) {
+ *idx_array_size = 0;
+ return 0;
+ }
+
+ idx_cnt = mca_get_valid_mca_idx(adev, mca_ras, type, mca_cnt, idx_array, *idx_array_size);
+ if (idx_cnt < 0)
+ return -EINVAL;
+
+ *idx_array_size = idx_cnt;
+
+ return 0;
+}
+
+static int mca_smu_get_ras_mca_idx_array(struct amdgpu_device *adev, enum amdgpu_ras_block blk,
+ enum amdgpu_mca_error_type type, int *idx_array, int *idx_array_size)
+{
+ const struct mca_ras_info *mca_ras;
+
+ mca_ras = mca_get_mca_ras_info(adev, blk);
+ if (!mca_ras)
+ return -EOPNOTSUPP;
+
+ return __mca_smu_get_ras_mca_idx_array(adev, mca_ras, type, idx_array, idx_array_size);
+}
+
+static int mca_smu_get_mca_entry(struct amdgpu_device *adev,
+ enum amdgpu_mca_error_type type, int idx, struct mca_bank_entry *entry)
+{
+ return mca_get_mca_entry(adev, type, idx, entry);
+}
+
+static int mca_smu_get_valid_mca_count(struct amdgpu_device *adev,
+ enum amdgpu_mca_error_type type, uint32_t *count)
+{
+ return mca_get_valid_mca_count(adev, type, count);
+}
+
+static const struct amdgpu_mca_smu_funcs smu_v13_0_6_mca_smu_funcs = {
+ .max_ue_count = 12,
+ .max_ce_count = 12,
+ .mca_set_debug_mode = mca_smu_set_debug_mode,
+ .mca_get_error_count = mca_smu_get_error_count,
+ .mca_get_mca_entry = mca_smu_get_mca_entry,
+ .mca_get_valid_mca_count = mca_smu_get_valid_mca_count,
+ .mca_get_ras_mca_idx_array = mca_smu_get_ras_mca_idx_array,
+};
+
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,
smu->table_map = smu_v13_0_6_table_map;
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);
}
projects / linux.git / commitdiff