Report all DIMMs LastShutdownStatus when ADR failure is detected

Signed-off-by: Tomasz Gromadzki <tomasz.gromadzki@hpe.com>

Author: grom72

src/common/set.c

@@ -2776,6 +2776,103 @@ util_unmap_all_hdrs(struct pool_set *set)
 	}
 }
 
+/*
+ * util_ipmctl_log_dimm_shutdown_status -- query ipmctl for every DIMM's last
+ * shutdown time and shutdown status (both latched and unlatched) and report
+ * the output of each command via CORE_LOG_ERROR.
+ *
+ * Steps:
+ *   1. Run "ipmctl show -display DimmID -dimm" and collect all DimmID values
+ *      from lines of the form ---DimmID=0x0001---.
+ *   2. For every discovered DimmID run three ipmctl queries and emit one
+ *      CORE_LOG_ERROR entry per query containing the raw command output.
+ */
+static void
+util_ipmctl_log_dimm_shutdown_status(void)
+{
+	/*
+	 * Collect DimmID strings from "ipmctl show -display DimmID -dimm".
+	 * A DimmID is always 6 characters (e.g. 0x0001) + NUL = 7 bytes.
+	 */
+#define IPMCTL_MAX_DIMMS	64
+#define IPMCTL_DIMMID_LEN	7  /* "0x0001" + NUL */
+#define IPMCTL_CMD_LEN		128
+#define IPMCTL_OUTPUT_LEN	1024
+
+	char dimm_ids[IPMCTL_MAX_DIMMS][IPMCTL_DIMMID_LEN];
+	int ndimms = 0;
+
+	FILE *fp = popen("ipmctl show -display DimmID -dimm", "r");
+	if (fp == NULL) {
+		CORE_LOG_ERROR_W_ERRNO(
+			"popen failed: ipmctl show -display DimmID -dimm");
+		return;
+	}
+
+	char line[256];
+	while (fgets(line, sizeof(line), fp) != NULL) {
+		char id[IPMCTL_DIMMID_LEN];
+		/* match lines of the form ---DimmID=0x0001--- */
+		if (sscanf(line, "---DimmID=%6[^-]---", id) == 1) {
+			if (ndimms < IPMCTL_MAX_DIMMS) {
+				memcpy(dimm_ids[ndimms], id, IPMCTL_DIMMID_LEN);
+				ndimms++;
+			} else {
+				CORE_LOG_ERROR(
+					"ipmctl DIMM count exceeds limit (%d); "
+					"remaining DIMMs skipped",
+					IPMCTL_MAX_DIMMS);
+				break;
+			}
+		}
+	}
+	pclose(fp);
+
+	/* Three queries to run per DIMM, in order */
+	static const char * const queries[] = {
+		"LastShutdownTime",
+		"UnlatchedLastShutdownStatus",
+		"LatchedLastShutdownStatus",
+	};
+
+	for (int i = 0; i < ndimms; i++) {
+		for (size_t q = 0; q < ARRAY_SIZE(queries); q++) {
+			char cmd[IPMCTL_CMD_LEN];
+			snprintf(cmd, sizeof(cmd),
+				"ipmctl show -display %s -dimm %s",
+				queries[q], dimm_ids[i]);
+
+			FILE *qfp = popen(cmd, "r");
+			if (qfp == NULL) {
+				CORE_LOG_ERROR_W_ERRNO(
+					"popen failed: %s", cmd);
+				continue;
+			}
+
+			char output[IPMCTL_OUTPUT_LEN];
+			if (fgets(output, sizeof(output), qfp) == NULL)
+				output[0] = '\0';
+
+			/* strip trailing newline */
+			size_t off = strlen(output);
+			while (off > 0 &&
+				(output[off - 1] == '\n' ||
+				output[off - 1] == '\r')) {
+				output[--off] = '\0';
+			}
+
+			pclose(qfp);
+
+			CORE_LOG_ERROR("DimmID=%s: %s", dimm_ids[i], output);
+		}
+	}
+
+#undef IPMCTL_MAX_DIMMS
+#undef IPMCTL_DIMMID_LEN
+#undef IPMCTL_CMD_LEN
+#undef IPMCTL_OUTPUT_LEN
+}
+
 /*
  * util_replica_check -- check headers, check UUID's, check replicas linkage
  */
@@ -2833,6 +2930,7 @@ util_replica_check(struct pool_set *set, const struct pool_attr *attr)
 			if (shutdown_state_check(&sds, &HDR(rep, 0)->sds,
 					rep)) {
 				CORE_LOG_ERROR("ADR failure detected");
+				util_ipmctl_log_dimm_shutdown_status();
 				errno = EINVAL;
 				return -1;
 			}