Merge branch 'master' into mc_trigger

Signed-off-by: winterddd <[email protected]>

mce-amd-smca.c

@@ -965,6 +965,18 @@ void decode_smca_error(struct mce_event *e, struct mce_priv *m)
 			     channel, csrow);
 	}
 
+
+	if (e->vdata_len) {
+		uint64_t smca_config = e->vdata[2];
+
+		/*
+		 * BIT 9 of the CONFIG register of a few SMCA Bank types indicates
+		 * presence of FRU Text in SYND 1 / 2 registers
+		 */
+		if (smca_config & BIT(9))
+			memcpy(e->frutext, e->vdata, 16);
+	}
+
 }
 
 int parse_amd_smca_event(struct ras_events *ras, struct mce_event *e)

non-standard-ampere.c

@@ -676,35 +676,30 @@ 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)

non-standard-yitian.c

@@ -164,16 +164,6 @@ void decode_yitian_ddr_payload_err_regs(struct ras_ns_ev_decoder *ev_decoder,
 	const char *subtype_str  = oem_subtype_name(yitian_payload_error_type,
 					header->type, header->subtype);
 
-#ifdef HAVE_SQLITE3
-	if (ras->record_events && !ev_decoder->stmt_dec_record) {
-		if (ras_mc_add_vendor_table(ras, &ev_decoder->stmt_dec_record,
-				&yitian_ddr_payload_section_tab) != SQLITE_OK) {
-			trace_seq_printf(s, "create sql fail\n");
-			return;
-		}
-	}
-#endif
-
 	now = time(NULL);
 	tm = localtime(&now);
 	if (tm)
@@ -217,6 +207,22 @@ void decode_yitian_ddr_payload_err_regs(struct ras_ns_ev_decoder *ev_decoder,
 
 }
 
+static int add_yitian_common_table(struct ras_events *ras,
+				 struct ras_ns_ev_decoder *ev_decoder)
+{
+#ifdef HAVE_SQLITE3
+	if (ras->record_events && !ev_decoder->stmt_dec_record) {
+		if (ras_mc_add_vendor_table(ras, &ev_decoder->stmt_dec_record,
+				&yitian_ddr_payload_section_tab) != SQLITE_OK) {
+			log(TERM, LOG_WARNING,
+				"Failed to create sql yitian_ddr_payload_section_tab\n");
+			return -1;
+		}
+	}
+#endif
+	return 0;
+}
+
 /* error data decoding functions */
 static int decode_yitian710_ns_error(struct ras_events *ras,
 				     struct ras_ns_ev_decoder *ev_decoder,
@@ -239,6 +245,7 @@ static int decode_yitian710_ns_error(struct ras_events *ras,
 struct ras_ns_ev_decoder yitian_ns_oem_decoder[] = {
 	{
 		.sec_type = "a6980811-16ea-4e4d-b936-fb00a23ff29c",
+		.add_table = add_yitian_common_table,
 		.decode = decode_yitian710_ns_error,
 	},
 };

ras-events.c

@@ -62,6 +62,7 @@
 #endif
 
 
+extern char* choices_disable;
 const static struct event_trigger event_triggers[] = {
 	{ "mc_event", &mc_event_trigger_setup },
 };
@@ -156,6 +157,18 @@ static int get_tracing_dir(struct ras_events *ras)
 	return 0;
 }
 
+static int is_disabled_event(char *group, char *event) {
+	char ras_event_name[MAX_PATH + 1];
+
+	snprintf(ras_event_name, sizeof(ras_event_name), "%s:%s",
+			group, event);
+
+	if (choices_disable != NULL && strlen(choices_disable) != 0 && strstr(choices_disable, ras_event_name)) {
+		return 1;
+	}
+	return 0;
+}
+
 /*
  * Tracing enable/disable code
  */
@@ -164,6 +177,7 @@ static int __toggle_ras_mc_event(struct ras_events *ras,
 {
 	int fd, rc;
 	char fname[MAX_PATH + 1];
+	enable = is_disabled_event(group, event) ? 0 : 1;
 
 	snprintf(fname, sizeof(fname), "%s%s:%s\n",
 		 enable ? "" : "!",
@@ -855,6 +869,12 @@ static int add_event_handler(struct ras_events *ras, struct tep_handle *pevent,
 
 	ras->filters[id] = filter;
 
+	if (is_disabled_event(group, event)) {
+		log(ALL, LOG_INFO, "Disabled %s:%s tracing from config\n",
+                    group, event);
+		return -EINVAL;
+	}
+
 	/* Enable RAS events */
 	rc = __toggle_ras_mc_event(ras, group, event, 1);
 	free(page);
@@ -924,7 +944,7 @@ int handle_ras_events(int record_events)
 			       ras_mc_event_handler, NULL, MC_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "ras", "mc_event");
 
@@ -933,7 +953,7 @@ int handle_ras_events(int record_events)
 			       ras_aer_event_handler, NULL, AER_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "ras", "aer_event");
 #endif
@@ -943,7 +963,7 @@ int handle_ras_events(int record_events)
 			       ras_non_standard_event_handler, NULL, NON_STANDARD_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "ras", "non_standard_event");
 #endif
@@ -953,7 +973,7 @@ int handle_ras_events(int record_events)
 			       ras_arm_event_handler, NULL, ARM_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "ras", "arm_event");
 #endif
@@ -987,7 +1007,7 @@ int handle_ras_events(int record_events)
 		/* tell kernel we are listening, so don't printk to console */
 		(void)open("/sys/kernel/debug/ras/daemon_active", 0);
 		num_events++;
-	} else
+	} else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "ras", "extlog_mem_event");
 #endif
@@ -1004,7 +1024,7 @@ int handle_ras_events(int record_events)
 			       ras_devlink_event_handler, filter_str, DEVLINK_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "devlink", "devlink_health_report");
 #endif
@@ -1016,7 +1036,7 @@ int handle_ras_events(int record_events)
 				NULL, DISKERROR_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "block", "block_rq_error");
 #else
@@ -1027,7 +1047,7 @@ int handle_ras_events(int record_events)
 					NULL, DISKERROR_EVENT);
 		if (!rc)
 			num_events++;
-		else
+		else if (rc != -EINVAL)
 			log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 			    "block", "block_rq_complete");
 	}
@@ -1039,7 +1059,7 @@ int handle_ras_events(int record_events)
 			       ras_memory_failure_event_handler, NULL, MF_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "ras", "memory_failure_event");
 #endif
@@ -1049,63 +1069,63 @@ int handle_ras_events(int record_events)
 			       ras_cxl_poison_event_handler, NULL, CXL_POISON_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "cxl", "cxl_poison");
 
 	rc = add_event_handler(ras, pevent, page_size, "cxl", "cxl_aer_uncorrectable_error",
 			       ras_cxl_aer_ue_event_handler, NULL, CXL_AER_UE_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "cxl", "cxl_aer_uncorrectable_error");
 
 	rc = add_event_handler(ras, pevent, page_size, "cxl", "cxl_aer_correctable_error",
 			       ras_cxl_aer_ce_event_handler, NULL, CXL_AER_CE_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "cxl", "cxl_aer_correctable_error");
 
 	rc = add_event_handler(ras, pevent, page_size, "cxl", "cxl_overflow",
 			       ras_cxl_overflow_event_handler, NULL, CXL_OVERFLOW_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "cxl", "cxl_overflow");
 
 	rc = add_event_handler(ras, pevent, page_size, "cxl", "cxl_generic_event",
 			       ras_cxl_generic_event_handler, NULL, CXL_GENERIC_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "cxl", "cxl_generic_event");
 
 	rc = add_event_handler(ras, pevent, page_size, "cxl", "cxl_general_media",
 			       ras_cxl_general_media_event_handler, NULL, CXL_GENERAL_MEDIA_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "cxl", "cxl_general_media");
 
 	rc = add_event_handler(ras, pevent, page_size, "cxl", "cxl_dram",
 			       ras_cxl_dram_event_handler, NULL, CXL_DRAM_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "cxl", "cxl_dram");
 
 	rc = add_event_handler(ras, pevent, page_size, "cxl", "cxl_memory_module",
 			       ras_cxl_memory_module_event_handler, NULL, CXL_MEMORY_MODULE_EVENT);
 	if (!rc)
 		num_events++;
-	else
+	else if (rc != -EINVAL)
 		log(ALL, LOG_ERR, "Can't get traces from %s:%s\n",
 		    "cxl", "memory_module");
 #endif

ras-mce-handler.c

@@ -372,6 +372,24 @@ static void report_mce_event(struct ras_events *ras,
 
 	trace_seq_printf(s, ", apicid= %x", e->apicid);
 
+	if (!e->vdata_len)
+		return;
+
+	if (strlen(e->frutext)) {
+		trace_seq_printf(s, ", FRU Text= %s", e->frutext);
+		trace_seq_printf(s, ", Vendor Data= ");
+		for (int i = 2; i < e->vdata_len/8; i++) {
+			trace_seq_printf(s, "0x%lx", e->vdata[i]);
+			trace_seq_printf(s, " ");
+		}
+	} else {
+		trace_seq_printf(s, ", Vendor Data= ");
+		for (int i = 0; i < e->vdata_len/8; i ++) {
+			trace_seq_printf(s, "0x%lx", e->vdata[i]);
+			trace_seq_printf(s, " ");
+		}
+	}
+
 	/*
 	 * FIXME: The original mcelog userspace tool uses DMI to map from
 	 * address to DIMM. From the comments there, the code there doesn't
@@ -548,6 +566,9 @@ int ras_mce_event_handler(struct trace_seq *s,
 		return -1;
 	e.ipid = val;
 
+	/* Get Vendor-specfic Data, if any */
+	e.vdata = tep_get_field_raw(s, event, "v_data", record, &e.vdata_len, 1);
+
 	switch (mce->cputype) {
 	case CPU_GENERIC:
 		break;

ras-mce-handler.h

@@ -75,8 +75,11 @@ struct mce_event {
 	uint8_t		cpuvendor;
 	uint64_t        synd;   /* MCA_SYND MSR: only valid on SMCA systems */
 	uint64_t        ipid;   /* MCA_IPID MSR: only valid on SMCA systems */
+	int32_t		vdata_len;
+	const uint64_t	*vdata;
 
 	/* Parsed data */
+	char		frutext[17];
 	char		timestamp[64];
 	char		bank_name[64];
 	char		error_msg[4096];

ras-page-isolation.c

@@ -171,18 +171,18 @@ static void parse_isolation_env(struct isolation *config)
 	config->unit = no_unit ? config->unit : "";
 }
 
-static void parse_env_string(struct isolation *config, char *str)
+static void parse_env_string(struct isolation *config, char *str, unsigned int size)
 {
 	int i;
 
 	if (config->overflow) {
 		/* when overflow, use basic unit */
 		for (i = 0; config->units[i].name; i++) ;
-		sprintf(str, "%lu%s", config->val, config->units[i-1].name);
+		snprintf(str, size, "%lu%s", config->val, config->units[i-1].name);
 		log(TERM, LOG_INFO, "%s is set overflow(%s), truncate it\n",
 				config->name, config->env);
 	} else {
-		sprintf(str, "%s%s", config->env, config->unit);
+		snprintf(str, size, "%s%s", config->env, config->unit);
 	}
 }
 
@@ -199,8 +199,8 @@ static void page_isolation_init(void)
 
 	parse_isolation_env(&threshold);
 	parse_isolation_env(&cycle);
-	parse_env_string(&threshold, threshold_string);
-	parse_env_string(&cycle, cycle_string);
+	parse_env_string(&threshold, threshold_string, sizeof(threshold_string));
+	parse_env_string(&cycle, cycle_string, sizeof(cycle_string));
 	log(TERM, LOG_INFO, "Threshold of memory Corrected Errors is %s / %s\n",
 			threshold_string, cycle_string);
 }