rasdaemon: ras-aer-handlder: Add support for AER triggers

Triggers aren't supported on AER events.

Add support for triggers on AER events. It relies on the AER_CE_TRIGGER
and AER_UE_TRIGGER environment variables

Signed-off-by: Bastien Curutchet <[email protected]>

Author: bastien-curutchet

misc/rasdaemon.env

@@ -88,6 +88,8 @@ TRIGGER_DIR=
 #   MC_UE_TRIGGER=mc_event_trigger
 MC_CE_TRIGGER=
 MC_UE_TRIGGER=
+AER_CE_TRIGGER=
+AER_UE_TRIGGER=
 
 # CE Statistic Threshold
 #

ras-aer-handler.c

@@ -4,9 +4,11 @@
  * Copyright (C) 2013 Mauro Carvalho Chehab <[email protected]>
  */
 
+#define _GNU_SOURCE
 #include <pci/pci.h>
 #include <stdio.h>
 #include <stdlib.h>
+#include <stdio.h>
 #include <string.h>
 #include <traceevent/kbuffer.h>
 #include <unistd.h>
@@ -16,6 +18,7 @@
 #include "ras-logger.h"
 #include "ras-report.h"
 #include "unified-sel.h"
+#include "trigger.h"
 #include "types.h"
 
 /* bit field meaning for correctable error */
@@ -53,7 +56,46 @@ static const char *aer_uncor_errors[32] = {
 	[26] = "Poisoned TLP Egrees Blocked",
 };
 
-static bool use_ipmitool = false;
+#define MAX_ENV 30
+static const char *aer_ce_trigger = NULL;
+static const char *aer_ue_trigger = NULL;
+
+void aer_event_trigger_setup(void)
+{
+	const char *trigger;
+
+	trigger = getenv("AER_CE_TRIGGER");
+	if (trigger && strcmp(trigger, "")) {
+		aer_ce_trigger = trigger_check(trigger);
+
+		if (!aer_ce_trigger) {
+			log(ALL, LOG_ERR,
+			    "Cannot access aer_event ce trigger `%s`\n",
+			    trigger);
+		} else {
+			log(ALL, LOG_INFO,
+			    "Setup aer_event ce trigger `%s`\n",
+			    trigger);
+		}
+	} else {
+		log(TERM, LOG_ERR, "\t no AER_CE_TRIGGER (%p)\n", trigger);
+	}
+
+	trigger = getenv("AER_UE_TRIGGER");
+	if (trigger && strcmp(trigger, "")) {
+		aer_ue_trigger = trigger_check(trigger);
+
+		if (!aer_ue_trigger) {
+			log(ALL, LOG_ERR,
+			    "Cannot access aer_event ue trigger `%s`\n",
+			    trigger);
+		} else {
+			log(ALL, LOG_INFO,
+			    "Setup aer_event ue trigger `%s`\n",
+			    trigger);
+		}
+	}
+}
 
 void ras_aer_handler_init(int enable_ipmitool)
 {
@@ -103,6 +145,32 @@ static void get_pci_dev_name(char *bdf, char *pci_name, ssize_t len, u16 *vendor
 	pci_cleanup(pacc);
 }
 
+static void run_aer_trigger(struct ras_aer_event *ev, const char *aer_trigger)
+{
+	char *env[MAX_ENV];
+	int ei = 0;
+	int i;
+
+	if (asprintf(&env[ei++], "PATH=%s", getenv("PATH") ?: "/sbin:/usr/sbin:/bin:/usr/bin") < 0)
+		goto free;
+	if (asprintf(&env[ei++], "TIMESTAMP=%s", ev->timestamp) < 0)
+		goto free;
+	if (asprintf(&env[ei++], "TYPE=%s", ev->error_type) < 0)
+		goto free;
+	if (asprintf(&env[ei++], "MESSAGE=%s", ev->msg) < 0)
+		goto free;
+	if (asprintf(&env[ei++], "NAME=%s", ev->dev_name) < 0)
+		goto free;
+	env[ei] = NULL;
+	assert(ei < MAX_ENV);
+
+	run_trigger(aer_trigger, NULL, env, "aer_event");
+
+free:
+	for (i = 0; i < ei; i++)
+		free(env[i]);
+}
+
 int ras_aer_event_handler(struct trace_seq *s,
 			  struct tep_record *record,
 			  struct tep_event *event, void *context)
@@ -270,5 +338,11 @@ int ras_aer_event_handler(struct trace_seq *s,
 			return -1;
 #endif
 
+	if (aer_ce_trigger && !strcmp(ev.error_type, "Corrected"))
+		run_aer_trigger(&ev, aer_ce_trigger);
+
+	if (aer_ue_trigger && !strncmp(ev.error_type, "Uncorrected", 11))
+		run_aer_trigger(&ev, aer_ue_trigger);
+
 	return 0;
 }

ras-aer-handler.h

@@ -11,6 +11,7 @@
 
 #include "ras-events.h"
 
+void aer_event_trigger_setup(void);
 int ras_aer_event_handler(struct trace_seq *s,
 			  struct tep_record *record,
 			  struct tep_event *event, void *context);

ras-events.c

@@ -56,6 +56,7 @@ char *choices_disable;
 
 static const struct event_trigger event_triggers[] = {
 	{ "mc_event", &mc_event_trigger_setup },
+	{ "aer_event", &aer_event_trigger_setup },
 #ifdef HAVE_MEMORY_FAILURE
 	{ "memory_failure_event", &mem_fail_event_trigger_setup },
 #endif