rpms/rasdaemon
7
0
Fork 0
Code Issues Pull Requests Releases Activity
f8702da1a4
rasdaemon/738bafafdcb2e8b0ced32fff31b13754d571090b.patch
Aristeu Rozanski f8702da1a4 Update rasdaemon package to latest upstream 
- Bumped to 0.6.7
- Backported patches that sit on top of 0.6.7 without being released

Fixes: #2052190

Signed-off-by: Aristeu Rozanski <arozansk@redhat.com>
2022-02-09 10:31:03 -05:00

611 lines
20 KiB
Diff
Raw Blame History

commit 738bafafdcb2e8b0ced32fff31b13754d571090b
Author: Jason Tian <jason@os.amperecomputing.com>
Date: Fri May 28 11:35:43 2021 +0800
Add error handling for Ampere-specific errors.
Save Ampere-specific errors' decode into sqlite3 data
base and log PCIe segment, bus/device/function number
into BMC SEL.
Signed-off-by: Jason Tian <jason@os.amperecomputing.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
diff --git a/non-standard-ampere.c b/non-standard-ampere.c
index 8cceb26..05b5252 100644
--- a/non-standard-ampere.c
+++ b/non-standard-ampere.c
@@ -216,6 +216,13 @@ static const char * const err_bert_sub_type[] = {
"PMPRO Fatal",
};
+static char *sqlite3_table_list[] = {
+ "amp_payload0_event_tab",
+ "amp_payload1_event_tab",
+ "amp_payload2_event_tab",
+ "amp_payload3_event_tab",
+};
+
struct amp_ras_type_info {
int id;
const char *name;
@@ -352,6 +359,359 @@ static const char *oem_subtype_name(const struct amp_ras_type_info *info,
return "unknown";
}
+#ifdef HAVE_SQLITE3
+/*key pair definition for ampere specific error payload type 0*/
+static const struct db_fields amp_payload0_event_fields[] = {
+ { .name = "id", .type = "INTEGER PRIMARY KEY" },
+ { .name = "timestamp", .type = "TEXT" },
+ { .name = "type", .type = "TEXT" },
+ { .name = "subtype", .type = "TEXT" },
+ { .name = "instance", .type = "INTEGER" },
+ { .name = "socket_num", .type = "INTEGER" },
+ { .name = "status_reg", .type = "INTEGER" },
+ { .name = "addr_reg", .type = "INTEGER" },
+ { .name = "misc0", .type = "INTEGER" },
+ { .name = "misc1", .type = "INTEGER" },
+ { .name = "misc2", .type = "INTEGER" },
+ { .name = "misc3", .type = "INTEGER" },
+};
+
+static const struct db_table_descriptor amp_payload0_event_tab = {
+ .name = "amp_payload0_event",
+ .fields = amp_payload0_event_fields,
+ .num_fields = ARRAY_SIZE(amp_payload0_event_fields),
+};
+
+/*key pair definition for ampere specific error payload type 1*/
+static const struct db_fields amp_payload1_event_fields[] = {
+ { .name = "id", .type = "INTEGER PRIMARY KEY" },
+ { .name = "timestamp", .type = "TEXT" },
+ { .name = "type", .type = "TEXT" },
+ { .name = "subtype", .type = "TEXT" },
+ { .name = "instance", .type = "INTEGER" },
+ { .name = "socket_num", .type = "INTEGER" },
+ { .name = "uncore_err_status", .type = "INTEGER" },
+ { .name = "uncore_err_mask", .type = "INTEGER" },
+ { .name = "uncore_err_sev", .type = "INTEGER" },
+ { .name = "core_err_status", .type = "INTEGER" },
+ { .name = "core_err_mask", .type = "INTEGER" },
+ { .name = "root_err_cmd", .type = "INTEGER" },
+ { .name = "root_err_status", .type = "INTEGER" },
+ { .name = "src_id", .type = "INTEGER" },
+ { .name = "reserved1", .type = "INTEGER" },
+ { .name = "reserverd2", .type = "INTEGER" },
+};
+
+static const struct db_table_descriptor amp_payload1_event_tab = {
+ .name = "amp_payload1_event",
+ .fields = amp_payload1_event_fields,
+ .num_fields = ARRAY_SIZE(amp_payload1_event_fields),
+};
+
+/*key pair definition for ampere specific error payload type 2*/
+static const struct db_fields amp_payload2_event_fields[] = {
+ { .name = "id", .type = "INTEGER PRIMARY KEY" },
+ { .name = "timestamp", .type = "TEXT" },
+ { .name = "type", .type = "TEXT" },
+ { .name = "subtype", .type = "TEXT" },
+ { .name = "instance", .type = "INTEGER" },
+ { .name = "socket_num", .type = "INTEGER" },
+ { .name = "ce_report_reg", .type = "INTEGER" },
+ { .name = "ce_location", .type = "INTEGER" },
+ { .name = "ce_addr", .type = "INTEGER" },
+ { .name = "ue_report_reg", .type = "INTEGER" },
+ { .name = "ue_location", .type = "INTEGER" },
+ { .name = "ue_addr", .type = "INTEGER" },
+ { .name = "reserved1", .type = "INTEGER" },
+ { .name = "reserved2", .type = "INTEGER" },
+ { .name = "reserved2", .type = "INTEGER" },
+};
+
+static const struct db_table_descriptor amp_payload2_event_tab = {
+ .name = "amp_payload2_event",
+ .fields = amp_payload2_event_fields,
+ .num_fields = ARRAY_SIZE(amp_payload2_event_fields),
+};
+
+/*key pair definition for ampere specific error payload type 3*/
+static const struct db_fields amp_payload3_event_fields[] = {
+ { .name = "id", .type = "INTEGER PRIMARY KEY" },
+ { .name = "timestamp", .type = "TEXT" },
+ { .name = "type", .type = "TEXT" },
+ { .name = "subtype", .type = "TEXT" },
+ { .name = "instance", .type = "INTEGER" },
+ { .name = "socket_num", .type = "INTEGER" },
+ { .name = "fw_spec_data0", .type = "INTEGER" },
+ { .name = "fw_spec_data1", .type = "INTEGER" },
+ { .name = "fw_spec_data2", .type = "INTEGER" },
+ { .name = "fw_spec_data3", .type = "INTEGER" },
+ { .name = "fw_spec_data4", .type = "INTEGER" },
+ { .name = "fw_spec_data5", .type = "INTEGER" },
+};
+
+static const struct db_table_descriptor amp_payload3_event_tab = {
+ .name = "amp_payload3_event",
+ .fields = amp_payload3_event_fields,
+ .num_fields = ARRAY_SIZE(amp_payload3_event_fields),
+};
+
+/*Save data with different type into sqlite3 db*/
+static void record_amp_data(struct ras_ns_ev_decoder *ev_decoder,
+ enum amp_oem_data_type data_type,
+ int id, int64_t data, const char *text)
+{
+ switch (data_type) {
+ case AMP_OEM_DATA_TYPE_INT:
+ sqlite3_bind_int(ev_decoder->stmt_dec_record, id, data);
+ break;
+ case AMP_OEM_DATA_TYPE_INT64:
+ sqlite3_bind_int64(ev_decoder->stmt_dec_record, id, data);
+ break;
+ case AMP_OEM_DATA_TYPE_TEXT:
+ sqlite3_bind_text(ev_decoder->stmt_dec_record, id,
+ text, -1, NULL);
+ break;
+ default:
+ break;
+ }
+}
+
+static int store_amp_err_data(struct ras_ns_ev_decoder *ev_decoder,
+ const char *name)
+{
+ int rc;
+
+ rc = sqlite3_step(ev_decoder->stmt_dec_record);
+ if (rc != SQLITE_OK && rc != SQLITE_DONE)
+ log(TERM, LOG_ERR,
+ "Failed to do %s step on sqlite: error = %d\n", name, rc);
+
+ rc = sqlite3_reset(ev_decoder->stmt_dec_record);
+ if (rc != SQLITE_OK && rc != SQLITE_DONE)
+ log(TERM, LOG_ERR,
+ "Failed to reset %s on sqlite: error = %d\n", name, rc);
+
+ rc = sqlite3_clear_bindings(ev_decoder->stmt_dec_record);
+ if (rc != SQLITE_OK && rc != SQLITE_DONE)
+ log(TERM, LOG_ERR,
+ "Failed to clear bindings %s on sqlite: error = %d\n",
+ name, rc);
+
+ return rc;
+}
+
+/*save all Ampere Specific Error Payload type 0 to sqlite3 database*/
+static void record_amp_payload0_err(struct ras_ns_ev_decoder *ev_decoder,
+ const char *type_str, const char *subtype_str,
+ const struct amp_payload0_type_sec *err)
+{
+ if (ev_decoder != NULL) {
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_TEXT,
+ AMP_PAYLOAD0_FIELD_TYPE, 0, type_str);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_TEXT,
+ AMP_PAYLOAD0_FIELD_SUB_TYPE, 0, subtype_str);
+
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD0_FIELD_INS, INSTANCE(err->instance), NULL);
+
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD0_FIELD_SOCKET_NUM,
+ SOCKET_NUM(err->instance), NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD0_FIELD_STATUS_REG, err->err_status, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD0_FIELD_ADDR_REG,
+ err->err_addr, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD0_FIELD_MISC0,
+ err->err_misc_0, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD0_FIELD_MISC1,
+ err->err_misc_1, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD0_FIELD_MISC2,
+ err->err_misc_2, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD0_FIELD_MISC3,
+ err->err_misc_3, NULL);
+ store_amp_err_data(ev_decoder, "amp_payload0_event_tab");
+ }
+}
+
+/*save all Ampere Specific Error Payload type 1 to sqlite3 database*/
+static void record_amp_payload1_err(struct ras_ns_ev_decoder *ev_decoder,
+ const char *type_str, const char *subtype_str,
+ const struct amp_payload1_type_sec *err)
+{
+ if (ev_decoder != NULL) {
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_TEXT,
+ AMP_PAYLOAD1_FIELD_TYPE, 0, type_str);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_TEXT,
+ AMP_PAYLOAD1_FIELD_SUB_TYPE, 0, subtype_str);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD1_FIELD_INS,
+ INSTANCE(err->instance), NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD1_FIELD_SOCKET_NUM,
+ SOCKET_NUM(err->instance), NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD1_FIELD_UNCORE_ERR_STATUS,
+ err->uncore_status, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD1_FIELD_UNCORE_ERR_MASK,
+ err->uncore_mask, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD1_FIELD_UNCORE_ERR_SEV,
+ err->uncore_sev, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD1_FIELD_CORE_ERR_STATUS,
+ err->core_status, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD1_FIELD_CORE_ERR_MASK,
+ err->core_mask, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD1_FIELD_ROOT_ERR_CMD,
+ err->root_err_cmd, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD1_FIELD_ROOT_ERR_STATUS,
+ err->root_status, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD1_FIELD_SRC_ID,
+ err->src_id, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD1_FIELD_RESERVED1,
+ err->reserved1, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD1_FIELD_RESERVED2,
+ err->reserved2, NULL);
+ store_amp_err_data(ev_decoder, "amp_payload1_event_tab");
+ }
+}
+
+/*save all Ampere Specific Error Payload type 2 to sqlite3 database*/
+static void record_amp_payload2_err(struct ras_ns_ev_decoder *ev_decoder,
+ const char *type_str, const char *subtype_str,
+ const struct amp_payload2_type_sec *err)
+{
+ if (ev_decoder != NULL) {
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_TEXT,
+ AMP_PAYLOAD2_FIELD_TYPE, 0, type_str);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_TEXT,
+ AMP_PAYLOAD2_FIELD_SUB_TYPE, 0, subtype_str);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD2_FIELD_INS, INSTANCE(err->instance), NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD2_FIELD_SOCKET_NUM,
+ SOCKET_NUM(err->instance), NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD2_FIELD_CE_REPORT_REG,
+ err->ce_register, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD2_FIELD_CE_LOACATION,
+ err->ce_location, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD2_FIELD_CE_ADDR,
+ err->ce_addr, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD2_FIELD_UE_REPORT_REG,
+ err->ue_register, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD2_FIELD_UE_LOCATION,
+ err->ue_location, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD2_FIELD_UE_ADDR,
+ err->ue_addr, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD2_FIELD_RESERVED1,
+ err->reserved1, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD2_FIELD_RESERVED2,
+ err->reserved2, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD2_FIELD_RESERVED3,
+ err->reserved3, NULL);
+ store_amp_err_data(ev_decoder, "amp_payload2_event_tab");
+ }
+}
+
+/*save all Ampere Specific Error Payload type 3 to sqlite3 database*/
+static void record_amp_payload3_err(struct ras_ns_ev_decoder *ev_decoder,
+ const char *type_str, const char *subtype_str,
+ const struct amp_payload3_type_sec *err)
+{
+ if (ev_decoder != NULL) {
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_TEXT,
+ AMP_PAYLOAD3_FIELD_TYPE, 0, type_str);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_TEXT,
+ AMP_PAYLOAD3_FIELD_SUB_TYPE, 0, subtype_str);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD3_FIELD_INS, INSTANCE(err->instance), NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD3_FIELD_SOCKET_NUM,
+ SOCKET_NUM(err->instance), NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA0,
+ err->fw_speci_data0, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA1,
+ err->fw_speci_data1, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA2,
+ err->fw_speci_data2, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA3,
+ err->fw_speci_data3, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA4,
+ err->fw_speci_data4, NULL);
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_INT64,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA5,
+ err->fw_speci_data5, NULL);
+ store_amp_err_data(ev_decoder, "amp_payload3_event_tab");
+ }
+}
+
+#else
+static void record_amp_data(struct ras_ns_ev_decoder *ev_decoder,
+ enum amp_oem_data_type data_type,
+ int id, int64_t data, const char *text)
+{
+ return 0;
+}
+
+static void record_amp_payload0_err(struct ras_ns_ev_decoder *ev_decoder,
+ const char *type_str, const char *subtype_str,
+ const struct amp_payload0_type_sec *err)
+{
+ return 0;
+}
+
+static void record_amp_payload1_err(struct ras_ns_ev_decoder *ev_decoder,
+ const char *type_str, const char *subtype_str,
+ const struct amp_payload1_type_sec *err)
+{
+ return 0;
+}
+
+static void record_amp_payload2_err(struct ras_ns_ev_decoder *ev_decoder,
+ const char *type_str, const char *subtype_str,
+ const struct amp_payload2_type_sec *err)
+{
+ return 0;
+}
+
+static void record_amp_payload3_err(struct ras_ns_ev_decoder *ev_decoder,
+ const char *type_str, const char *subtype_str,
+ const struct amp_payload3_type_sec *err)
+{
+ return 0;
+}
+
+static int store_amp_err_data(struct ras_ns_ev_decoder *ev_decoder, char *name)
+{
+ return 0;
+}
+#endif
/*decode ampere specific error payload type 0, the CPU's data is save*/
/*to sqlite by ras-arm-handler, others are saved by this function.*/
@@ -434,6 +794,7 @@ void decode_amp_payload0_err_regs(struct ras_ns_ev_decoder *ev_decoder,
*p = '\0';
}
+ record_amp_payload0_err(ev_decoder, type_str, subtype_str, err);
i = 0;
p = NULL;
end = NULL;
@@ -517,6 +878,7 @@ static void decode_amp_payload1_err_regs(struct ras_ns_ev_decoder *ev_decoder,
*p = '\0';
}
+ record_amp_payload1_err(ev_decoder, type_str, subtype_str, err);
i = 0;
p = NULL;
end = NULL;
@@ -601,6 +963,7 @@ static void decode_amp_payload2_err_regs(struct ras_ns_ev_decoder *ev_decoder,
*p = '\0';
}
+ record_amp_payload2_err(ev_decoder, type_str, subtype_str, err);
i = 0;
p = NULL;
end = NULL;
@@ -673,6 +1036,7 @@ static void decode_amp_payload3_err_regs(struct ras_ns_ev_decoder *ev_decoder,
*p = '\0';
}
+ record_amp_payload3_err(ev_decoder, type_str, subtype_str, err);
i = 0;
p = NULL;
end = NULL;
@@ -687,6 +1051,38 @@ static int decode_amp_oem_type_error(struct ras_events *ras,
{
int payload_type = PAYLOAD_TYPE(event->error[0]);
+#ifdef HAVE_SQLITE3
+ struct db_table_descriptor db_tab;
+ int id = 0;
+
+ if (payload_type == PAYLOAD_TYPE_0) {
+ db_tab = amp_payload0_event_tab;
+ id = AMP_PAYLOAD0_FIELD_TIMESTAMP;
+ } else if (payload_type == PAYLOAD_TYPE_1) {
+ db_tab = amp_payload1_event_tab;
+ id = AMP_PAYLOAD1_FIELD_TIMESTAMP;
+ } else if (payload_type == PAYLOAD_TYPE_2) {
+ db_tab = amp_payload2_event_tab;
+ id = AMP_PAYLOAD2_FIELD_TIMESTAMP;
+ } else if (payload_type == PAYLOAD_TYPE_3) {
+ db_tab = amp_payload3_event_tab;
+ id = AMP_PAYLOAD3_FIELD_TIMESTAMP;
+ } else
+ return -1;
+
+ if (!ev_decoder->stmt_dec_record) {
+ if (ras_mc_add_vendor_table(ras, &ev_decoder->stmt_dec_record,
+ &db_tab) != SQLITE_OK) {
+ trace_seq_printf(s,
+ "create sql %s fail\n",
+ sqlite3_table_list[payload_type]);
+ return -1;
+ }
+ }
+ record_amp_data(ev_decoder, AMP_OEM_DATA_TYPE_TEXT,
+ id, 0, event->timestamp);
+#endif
+
if (payload_type == PAYLOAD_TYPE_0) {
const struct amp_payload0_type_sec *err =
(struct amp_payload0_type_sec *)event->error;
diff --git a/non-standard-ampere.h b/non-standard-ampere.h
index aacf3a8..f463c53 100644
--- a/non-standard-ampere.h
+++ b/non-standard-ampere.h
@@ -102,6 +102,79 @@ struct amp_payload3_type_sec {
uint64_t fw_speci_data5;
};
+enum amp_oem_data_type {
+ AMP_OEM_DATA_TYPE_INT,
+ AMP_OEM_DATA_TYPE_INT64,
+ AMP_OEM_DATA_TYPE_TEXT,
+};
+
+enum {
+ AMP_PAYLOAD0_FIELD_ID,
+ AMP_PAYLOAD0_FIELD_TIMESTAMP,
+ AMP_PAYLOAD0_FIELD_TYPE,
+ AMP_PAYLOAD0_FIELD_SUB_TYPE,
+ AMP_PAYLOAD0_FIELD_INS,
+ AMP_PAYLOAD0_FIELD_SOCKET_NUM,
+ AMP_PAYLOAD0_FIELD_STATUS_REG,
+ AMP_PAYLOAD0_FIELD_ADDR_REG,
+ AMP_PAYLOAD0_FIELD_MISC0,
+ AMP_PAYLOAD0_FIELD_MISC1,
+ AMP_PAYLOAD0_FIELD_MISC2,
+ AMP_PAYLOAD0_FIELD_MISC3,
+};
+
+enum {
+ AMP_PAYLOAD1_FIELD_ID,
+ AMP_PAYLOAD1_FIELD_TIMESTAMP,
+ AMP_PAYLOAD1_FIELD_TYPE,
+ AMP_PAYLOAD1_FIELD_SUB_TYPE,
+ AMP_PAYLOAD1_FIELD_INS,
+ AMP_PAYLOAD1_FIELD_SOCKET_NUM,
+ AMP_PAYLOAD1_FIELD_UNCORE_ERR_STATUS,
+ AMP_PAYLOAD1_FIELD_UNCORE_ERR_MASK,
+ AMP_PAYLOAD1_FIELD_UNCORE_ERR_SEV,
+ AMP_PAYLOAD1_FIELD_CORE_ERR_STATUS,
+ AMP_PAYLOAD1_FIELD_CORE_ERR_MASK,
+ AMP_PAYLOAD1_FIELD_ROOT_ERR_CMD,
+ AMP_PAYLOAD1_FIELD_ROOT_ERR_STATUS,
+ AMP_PAYLOAD1_FIELD_SRC_ID,
+ AMP_PAYLOAD1_FIELD_RESERVED1,
+ AMP_PAYLOAD1_FIELD_RESERVED2,
+};
+
+enum {
+ AMP_PAYLOAD2_FIELD_ID,
+ AMP_PAYLOAD2_FIELD_TIMESTAMP,
+ AMP_PAYLOAD2_FIELD_TYPE,
+ AMP_PAYLOAD2_FIELD_SUB_TYPE,
+ AMP_PAYLOAD2_FIELD_INS,
+ AMP_PAYLOAD2_FIELD_SOCKET_NUM,
+ AMP_PAYLOAD2_FIELD_CE_REPORT_REG,
+ AMP_PAYLOAD2_FIELD_CE_LOACATION,
+ AMP_PAYLOAD2_FIELD_CE_ADDR,
+ AMP_PAYLOAD2_FIELD_UE_REPORT_REG,
+ AMP_PAYLOAD2_FIELD_UE_LOCATION,
+ AMP_PAYLOAD2_FIELD_UE_ADDR,
+ AMP_PAYLOAD2_FIELD_RESERVED1,
+ AMP_PAYLOAD2_FIELD_RESERVED2,
+ AMP_PAYLOAD2_FIELD_RESERVED3,
+};
+
+enum {
+ AMP_PAYLOAD3_FIELD_ID,
+ AMP_PAYLOAD3_FIELD_TIMESTAMP,
+ AMP_PAYLOAD3_FIELD_TYPE,
+ AMP_PAYLOAD3_FIELD_SUB_TYPE,
+ AMP_PAYLOAD3_FIELD_INS,
+ AMP_PAYLOAD3_FIELD_SOCKET_NUM,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA0,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA1,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA2,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA3,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA4,
+ AMP_PAYLOAD3_FIELD_FW_SPEC_DATA5
+};
+
void decode_amp_payload0_err_regs(struct ras_ns_ev_decoder *ev_decoder,
struct trace_seq *s,
const struct amp_payload0_type_sec *err);
diff --git a/ras-aer-handler.c b/ras-aer-handler.c
index 8ddd439..6f4cb2b 100644
--- a/ras-aer-handler.c
+++ b/ras-aer-handler.c
@@ -67,6 +67,9 @@ int ras_aer_event_handler(struct trace_seq *s,
struct tm *tm;
struct ras_aer_event ev;
char buf[BUF_LEN];
+ char ipmi_add_sel[105];
+ uint8_t sel_data[5];
+ int seg, bus, dev, fn;
/*
* Newer kernels (3.10-rc1 or upper) provide an uptime clock.
@@ -129,15 +132,19 @@ int ras_aer_event_handler(struct trace_seq *s,
switch (severity_val) {
case HW_EVENT_AER_UNCORRECTED_NON_FATAL:
ev.error_type = "Uncorrected (Non-Fatal)";
+ sel_data[0] = 0xca;
break;
case HW_EVENT_AER_UNCORRECTED_FATAL:
ev.error_type = "Uncorrected (Fatal)";
+ sel_data[0] = 0xca;
break;
case HW_EVENT_AER_CORRECTED:
ev.error_type = "Corrected";
+ sel_data[0] = 0xbf;
break;
default:
ev.error_type = "Unknown severity";
+ sel_data[0] = 0xbf;
}
trace_seq_puts(s, ev.error_type);
@@ -151,5 +158,29 @@ int ras_aer_event_handler(struct trace_seq *s,
ras_report_aer_event(ras, &ev);
#endif
+#ifdef HAVE_AMP_NS_DECODE
+ /*
+ * Get PCIe AER error source seg/bus/dev/fn and save it into
+ * BMC OEM SEL, ipmitool raw 0x0a 0x44 is IPMI command-Add SEL
+ * entry, please refer IPMI specificaiton chapter 31.6. 0xcd3a
+ * is manufactuer ID(ampere),byte 12 is sensor num(CE is 0xBF,
+ * UE is 0xCA), byte 13~14 is segment number, byte 15 is bus
+ * number, byte 16[7:3] is device number, byte 16[2:0] is
+ * function number
+ */
+ sscanf(ev.dev_name, "%x:%x:%x.%x", &seg, &bus, &dev, &fn);
+
+ sel_data[1] = seg & 0xff;
+ sel_data[2] = (seg & 0xff00) >> 8;
+ sel_data[3] = bus;
+ sel_data[4] = (((dev & 0x1f) << 3) | (fn & 0x7));
+
+ sprintf(ipmi_add_sel,
+ "ipmitool raw 0x0a 0x44 0x00 0x00 0xc0 0x00 0x00 0x00 0x00 0x3a 0xcd 0x00 0xc0 0x%02x 0x%02x 0x%02x 0x%02x 0x%02x",
+ sel_data[0], sel_data[1], sel_data[2], sel_data[3], sel_data[4]);
+
+ system(ipmi_add_sel);
+#endif
+
return 0;
}
Reference in New Issue View Git Blame Copy Permalink