From ece0a63689ee54498bb0d372cfb568e431ada5d5 Mon Sep 17 00:00:00 2001
From: Ruidong Tian <tianruidong@linux.alibaba.com>
Date: Mon, 9 Dec 2024 16:28:54 +0800
Subject: [PATCH 77/85] rasdaemon: add rasdaemon json exporter

Signed-off-by: Ruidong Tian <tianruidong@linux.alibaba.com>
---
 Makefile.am                  |   5 +-
 configure.ac                 |  16 +++
 misc/rasdaemon.env           |   2 +
 ras-aer-handler.c            |   9 ++
 ras-arm-handler.c            |   4 +
 ras-mc-handler.c             |   9 ++
 ras-mce-handler.c            |   5 +
 ras-mce-handler.h            |   1 +
 ras-memory-failure-handler.c |   4 +
 ras-record.h                 |   9 ++
 ras-report-json.c            | 240 +++++++++++++++++++++++++++++++++++
 ras-report.h                 |  14 ++
 rasdaemon.c                  |   9 +-
 13 files changed, 325 insertions(+), 2 deletions(-)
 create mode 100644 ras-report-json.c

diff --git a/Makefile.am b/Makefile.am
index fb0248e..2582454 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -54,6 +54,9 @@ endif
 if WITH_ABRT_REPORT
    rasdaemon_SOURCES += ras-report.c
 endif
+if WITH_JSON_REPORT
+   rasdaemon_SOURCES += ras-report-json.c
+endif
 if WITH_HISI_NS_DECODE
    rasdaemon_SOURCES += non-standard-hisi_hip08.c non-standard-hisilicon.c
 endif
@@ -70,7 +73,7 @@ if WITH_YITIAN_NS_DECODE
    rasdaemon_SOURCES += non-standard-yitian.c
 endif
 
-rasdaemon_LDADD = -lpthread $(SQLITE3_LIBS) libtrace/libtrace.a
+rasdaemon_LDADD = -lpthread $(SQLITE3_LIBS) libtrace/libtrace.a $(PCI_LIBS)
 
 include_HEADERS = config.h types.h ras-events.h ras-logger.h ras-mc-handler.h \
 		  ras-aer-handler.h ras-mce-handler.h ras-record.h bitfield.h ras-report.h \
diff --git a/configure.ac b/configure.ac
index 135af9c..c8b4ab6 100644
--- a/configure.ac
+++ b/configure.ac
@@ -131,6 +131,21 @@ AS_IF([test "x$enable_abrt_report" = "xyes" || test "x$enable_all" == "xyes"], [
 AM_CONDITIONAL([WITH_ABRT_REPORT], [test x$enable_abrt_report = xyes || test x$enable_all == xyes])
 AM_COND_IF([WITH_ABRT_REPORT], [USE_ABRT_REPORT="yes"], [USE_ABRT_REPORT="no"])
 
+AC_ARG_ENABLE([json_report],
+    AS_HELP_STRING([--enable-json-report], [enable storing data at SQL lite database (currently experimental)]))
+
+AS_IF([test "x$enable_json_report" = "xyes" || test "x$enable_all" == "xyes"], [
+  AC_CHECK_LIB(pci, pci_lookup_name,[echo "found pci"] , AC_MSG_ERROR([*** Unable to find pci library]), )
+  PCI_LIBS="-lpci"
+  AC_DEFINE(HAVE_JSON_REPORT,1,"have libpci")
+  AC_SUBST([WITH_JSON_REPORT])
+])
+
+AM_CONDITIONAL([WITH_JSON_REPORT], [test x$enable_json_report = xyes || test x$enable_all == xyes])
+AM_COND_IF([WITH_JSON_REPORT], [USE_JSON_REPORT="yes"], [USE_JSON_REPORT="no"])
+
+AC_SUBST([PCI_LIBS])
+
 AC_ARG_ENABLE([hisi_ns_decode],
     AS_HELP_STRING([--enable-hisi-ns-decode], [enable HISI_NS_DECODE events (currently experimental)]))
 
@@ -223,4 +238,5 @@ compile time options summary
     AMP RAS errors      : $USE_AMP_NS_DECODE
     CPU fault isolation : $USE_CPU_FAULT_ISOLATION
     YITIAN RAS errors   : $USE_YITIAN_NS_DECODE
+    Json exporter       : $USE_JSON_REPORT
 EOF
diff --git a/misc/rasdaemon.env b/misc/rasdaemon.env
index 67f488f..760a42d 100644
--- a/misc/rasdaemon.env
+++ b/misc/rasdaemon.env
@@ -45,6 +45,8 @@ CPU_ISOLATION_CYCLE="24h"
 # Prevent excessive isolation from causing an avalanche effect
 CPU_ISOLATION_LIMIT="10"
 
+
+DISABLE=""
 # Event Trigger
 
 # Event trigger will be executed when the specified event occurs.
diff --git a/ras-aer-handler.c b/ras-aer-handler.c
index b00703e..d0eb4df 100644
--- a/ras-aer-handler.c
+++ b/ras-aer-handler.c
@@ -22,6 +22,8 @@
 #include "libtrace/kbuffer.h"
 #include "ras-aer-handler.h"
 #include "types.h"
+#include "ras-events.h"
+#include "ras-record.h"
 #include "ras-logger.h"
 #include "bitfield.h"
 #include "ras-report.h"
@@ -135,18 +137,22 @@ int ras_aer_event_handler(struct trace_seq *s,
 	case HW_EVENT_AER_UNCORRECTED_NON_FATAL:
 		ev.error_type = "Uncorrected (Non-Fatal)";
 		sel_data[0] = 0xca;
+		ev.severity = GHES_SEV_RECOVERABLE;
 		break;
 	case HW_EVENT_AER_UNCORRECTED_FATAL:
 		ev.error_type = "Uncorrected (Fatal)";
 		sel_data[0] = 0xca;
+		ev.severity = GHES_SEV_PANIC;
 		break;
 	case HW_EVENT_AER_CORRECTED:
 		ev.error_type = "Corrected";
 		sel_data[0] = 0xbf;
+		ev.severity = GHES_SEV_CORRECTED;
 		break;
 	default:
 		ev.error_type = "Unknown severity";
 		sel_data[0] = 0xbf;
+		ev.severity = GHES_SEV_NO;
 	}
 	trace_seq_puts(s, ev.error_type);
 
@@ -184,6 +190,9 @@ int ras_aer_event_handler(struct trace_seq *s,
 	system(ipmi_add_sel);
 #endif
 
+#ifdef HAVE_JSON_REPORT
+	report_aer_event_json(s, &ev);
+#endif
 	run_aer_event_trigger(&ev);
 
 	return 0;
diff --git a/ras-arm-handler.c b/ras-arm-handler.c
index 19150cb..97ebe21 100644
--- a/ras-arm-handler.c
+++ b/ras-arm-handler.c
@@ -264,5 +264,9 @@ int ras_arm_event_handler(struct trace_seq *s,
 	ras_report_arm_event(ras, &ev);
 #endif
 
+#ifdef HAVE_JSON_REPORT
+	report_arm_event_json(s, &ev);
+#endif
+
 	return 0;
 }
diff --git a/ras-mc-handler.c b/ras-mc-handler.c
index a270637..bb93c9d 100644
--- a/ras-mc-handler.c
+++ b/ras-mc-handler.c
@@ -29,6 +29,7 @@
 #include "ras-page-isolation.h"
 #include "types.h"
 #include "ras-report.h"
+#include "ras-events.h"
 #include "trigger.h"
 
 int ras_mc_event_handler(struct trace_seq *s,
@@ -77,16 +78,20 @@ int ras_mc_event_handler(struct trace_seq *s,
 	switch (val) {
 	case HW_EVENT_ERR_CORRECTED:
 		ev.error_type = "Corrected";
+		ev.severity = GHES_SEV_CORRECTED;
 		break;
 	case HW_EVENT_ERR_UNCORRECTED:
 		ev.error_type = "Uncorrected";
+		ev.severity = GHES_SEV_RECOVERABLE;
 		break;
 	case HW_EVENT_ERR_FATAL:
 		ev.error_type = "Fatal";
+		ev.severity = GHES_SEV_PANIC;
 		break;
 	case HW_EVENT_ERR_INFO:
 	default:
 		ev.error_type = "Info";
+		ev.severity = GHES_SEV_NO;
 	}
 
 	trace_seq_puts(s, ev.error_type);
@@ -202,6 +207,10 @@ int ras_mc_event_handler(struct trace_seq *s,
 
 	run_mc_event_trigger(&ev);
 
+#ifdef HAVE_JSON_REPORT
+	report_mc_event_json(s, &ev);
+#endif
+
 	return 0;
 
 parse_error:
diff --git a/ras-mce-handler.c b/ras-mce-handler.c
index 9601704..ecc6468 100644
--- a/ras-mce-handler.c
+++ b/ras-mce-handler.c
@@ -30,6 +30,7 @@
 #include "ras-logger.h"
 #include "ras-report.h"
 #include "trigger.h"
+#include "ras-events.h"
 
 /*
  * The code below were adapted from Andi Kleen/Intel/SuSe mcelog code,
@@ -581,6 +582,10 @@ int ras_mce_event_handler(struct trace_seq *s,
 	ras_report_mce_event(ras, &e);
 #endif
 
+#ifdef HAVE_JSON_REPORT
+	report_mce_event_json(s, &e);
+#endif
+
 	run_mce_record_trigger(&e);
 
 	return 0;
diff --git a/ras-mce-handler.h b/ras-mce-handler.h
index e1064f6..f0dbdab 100644
--- a/ras-mce-handler.h
+++ b/ras-mce-handler.h
@@ -76,6 +76,7 @@ struct mce_event {
 	uint64_t        synd;   /* MCA_SYND MSR: only valid on SMCA systems */
 	uint64_t        ipid;   /* MCA_IPID MSR: only valid on SMCA systems */
 
+	int		severity;
 	/* Parsed data */
 	char		timestamp[64];
 	char		bank_name[64];
diff --git a/ras-memory-failure-handler.c b/ras-memory-failure-handler.c
index 8bc7a9d..9cd56b4 100644
--- a/ras-memory-failure-handler.c
+++ b/ras-memory-failure-handler.c
@@ -174,5 +174,9 @@ int ras_memory_failure_event_handler(struct trace_seq *s,
 #endif
 	run_mf_event_trigger(&ev);
 
+#ifdef HAVE_JSON_REPORT
+	report_mf_event_json(s, &ev);
+#endif
+
 	return 0;
 }
diff --git a/ras-record.h b/ras-record.h
index f1edcc0..f48fe37 100644
--- a/ras-record.h
+++ b/ras-record.h
@@ -25,6 +25,13 @@
 #include "config.h"
 #include "types.h"
 
+static const char * const severity_strs[] = {
+	"info",
+	"corrected",
+	"recoverable",
+	"fatal",
+};
+
 extern long user_hz;
 
 struct ras_events;
@@ -32,6 +39,7 @@ struct ras_events;
 struct ras_mc_event {
 	char timestamp[64];
 	int error_count;
+	int severity;
 	const char *error_type, *msg, *label;
 	unsigned char mc_index;
 	signed char top_layer, middle_layer, lower_layer;
@@ -51,6 +59,7 @@ struct ras_mc_offline_event {
 struct ras_aer_event {
 	char timestamp[64];
 	const char *error_type;
+	int severity;
 	const char *dev_name;
 	uint8_t tlp_header_valid;
 	uint32_t *tlp_header;
diff --git a/ras-report-json.c b/ras-report-json.c
new file mode 100644
index 0000000..f59ca32
--- /dev/null
+++ b/ras-report-json.c
@@ -0,0 +1,240 @@
+/*
+ * Copyright (c) 2016, The Linux Foundation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 and
+ * only version 2 as published by the Free Software Foundation.
+
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <unistd.h>
+#include <sys/types.h>
+#include <sys/utsname.h>
+#include <sys/socket.h>
+#include <sys/un.h>
+#include <pci/pci.h>
+
+#include "libtrace/event-parse.h"
+#include "ras-report.h"
+
+#define NONE ""
+int json_report = 1;
+
+void report_mc_event_json(struct trace_seq *s, struct ras_mc_event *ev)
+{
+	if (!s || !ev || !json_report)
+		return;
+
+	trace_seq_printf(s,
+			"\n{ \"%s\": \"mc_event\", "	\
+			 "\"timestamp\": \"%s\", "	\
+			 "\"severity\": \"%s\", "	\
+			 "\"error_count\": %d, "	\
+			 "\"error_type\": \"%s\", "	\
+			 "\"msg\": \"%s\", "		\
+			 "\"label\": \"%s\", "		\
+			 "\"location\": \"%d:%d:%d:%d\", "	\
+			 "\"address\": \"%#llx\", "	\
+			 "\"grain\": \"%#llx\", "	\
+			 "\"syndrome\": \"%#llx\", "	\
+			 "\"driver_detail\": \"%s\" }",
+			 JSON_REPORT_KEY,
+			 (*ev->timestamp) ? ev->timestamp : NONE,
+			 severity_strs[ev->severity],
+			 ev->error_count,
+			 (ev->error_type) ? ev->error_type : NONE,
+			 (ev->msg) ? ev->msg : NONE,
+			 (ev->label) ? ev->label : NONE,
+			 ev->mc_index, ev->top_layer, ev->middle_layer, ev->lower_layer,
+			 ev->address,
+			 ev->grain,
+			 ev->syndrome,
+			 (ev->driver_detail) ? ev->driver_detail : NONE);
+}
+
+static void get_pci_dev_name(const char *bdf, char *pci_name, ssize_t len, u16 *vendor_id, u16 *device_id)
+{
+	struct pci_access *pacc;
+	struct pci_dev *dev;
+	struct pci_filter filter = {0};
+	int domain, bus, device, function;
+
+	pacc = pci_alloc();
+	if (!pacc)
+		return;
+	pci_init(pacc);
+	pci_scan_bus(pacc);
+
+	if (!pci_name)
+		goto free;
+
+
+	if (sscanf(bdf, "%x:%x.%x", &bus, &device, &function) == 3)
+		domain = 0;
+	else if (sscanf(bdf, "%x:%x:%x", &domain, &bus, &device) == 3)
+		function = 0;
+	else if (sscanf(bdf, "%x:%x:%x.%x", &domain, &bus, &device, &function) != 4)
+		goto free;
+
+	pci_filter_init(pacc, &filter);
+	filter.bus = bus;
+	filter.slot = device;
+	filter.func = function;
+	filter.domain = domain;
+
+	for (dev = pacc->devices; dev; dev = dev->next) {
+		if (pci_filter_match(&filter, dev)) {
+			pci_fill_info(dev, PCI_FILL_IDENT);
+			*vendor_id = dev->vendor_id;
+			*device_id = dev->device_id;
+			pci_lookup_name(pacc, pci_name, len,
+					PCI_LOOKUP_VENDOR | PCI_LOOKUP_DEVICE,
+					dev->vendor_id, dev->device_id);
+			break;
+		}
+	}
+
+free:
+	pci_cleanup(pacc);
+	return;
+}
+
+void report_aer_event_json(struct trace_seq *s, struct ras_aer_event *ev)
+{
+	char pci_name[128];
+	u16 vendor, device;
+
+	if (!s || !ev || !json_report)
+		return;
+
+	get_pci_dev_name(ev->dev_name, pci_name, 128, &vendor, &device);
+
+	trace_seq_printf(s,
+			 "\n{ \"%s\": \"aer_event\", "	\
+			 "\"timestamp\": \"%s\", "	\
+			 "\"severity\": \"%s\", "	\
+			 "\"error_type\": \"%s\", "	\
+			 "\"dev_name\": \"%s\", "	\
+			 "\"pci_dev_name\": \"%s\", "	\
+			 "\"vendor_id\": \"%#x\", "	\
+			 "\"device_id\": \"%#x\", "	\
+			 "\"msg\": \"%s\" }",
+			 JSON_REPORT_KEY,
+			 (*ev->timestamp) ? ev->timestamp : NONE,
+			 severity_strs[ev->severity],
+			 (ev->error_type) ? ev->error_type : NONE,
+			 (ev->dev_name) ? ev->dev_name : NONE,
+			 (*pci_name) ? pci_name : NONE,
+			 vendor, device,
+			 (ev->msg) ? ev->msg : NONE);
+}
+
+void report_arm_event_json(struct trace_seq *s, struct ras_arm_event *ev)
+{
+	if (!s || !ev || !json_report)
+		return;
+
+	trace_seq_printf(s,
+			 "\n{ \"%s\": \"arm_event\", "	\
+			 "\"timestamp\": \"%s\", "	\
+			 "\"error_count\": %d, "	\
+			 "\"affinity\": %d, "		\
+			 "\"mpidr\": \"%#lx\", "	\
+			 "\"midr\": \"%#lx\", "		\
+			 "\"running_state\": %d, "	\
+			 "\"psci_state\": %d }",
+			 JSON_REPORT_KEY,
+			 (*ev->timestamp) ? ev->timestamp : NONE,
+			 ev->error_count,
+			 ev->affinity,
+			 ev->mpidr,
+			 ev->midr,
+			 ev->running_state,
+			 ev->psci_state);
+}
+
+void report_mf_event_json(struct trace_seq *s, struct ras_mf_event *ev)
+{
+	if (!s || !ev || !json_report)
+		return;
+
+	trace_seq_printf(s,
+			 "\n{ \"%s\": \"mf_event\", \"timestamp\": \"%s\", " \
+			 "\"pfn\": %s, \"page_type\": \"%s\", " \
+			 "\"action_result\": \"%s\" }",
+			 JSON_REPORT_KEY,
+			 (*ev->timestamp) ? ev->timestamp : NONE,
+			 (*ev->pfn) ? ev->pfn : NONE,
+			 (ev->page_type) ? ev->page_type : NONE,
+			 (ev->action_result) ? ev->action_result : NONE);
+}
+
+void report_mce_event_json(struct trace_seq *s, struct mce_event *ev)
+{
+	if (!s || !ev || !json_report)
+		return;
+
+	if (ev->status & MCI_STATUS_UC)
+		ev->severity = GHES_SEV_RECOVERABLE;
+	else if (ev->status & MCI_STATUS_DEFERRED)
+		ev->severity = GHES_SEV_RECOVERABLE;
+	else
+		ev->severity = GHES_SEV_CORRECTED;
+
+	trace_seq_printf(s,
+			 "\n{ \"%s\": \"mce_record\", "	\
+			 "\"timestamp\": \"%s\", "	\
+			 "\"severity\": \"%s\", "	\
+			 "\"bank\": %d, "		\
+			 "\"bank_name\": \"%s\", "	\
+			 "\"status\": \"%#lx\", "	\
+			 "\"error_msg\": \"%s\", "	\
+			 "\"mcistatus_msg\": \"%s\", "	\
+			 "\"mcastatus_msg\": \"%s\", "	\
+			 "\"user_action\": \"%s\", "	\
+			 "\"mc_location\": \"%s\", "	\
+			 "\"cpuid\": \"%#x\", "		\
+			 "\"cpu\": %d, "		\
+			 "\"socketid\": %d, "		\
+			 "\"ip\": \"%#lx\", "		\
+			 "\"cs\": \"%#x\", "		\
+			 "\"misc\": \"%#lx\", "		\
+			 "\"addr\": \"%#lx\", "		\
+			 "\"synd\": \"%#lx\", "		\
+			 "\"ipid\": \"%#lx\", "		\
+			 "\"mcgstatus_msg\": \"%s\", "	\
+			 "\"mcgstatus\": \"%#lx\", "	\
+			 "\"mcgcap\": \"%#lx\", "	\
+			 "\"apicid\": \"%#x\" }",
+			 JSON_REPORT_KEY,
+			 (*ev->timestamp) ? ev->timestamp : NONE,
+			 severity_strs[ev->severity],
+			 ev->bank,
+			 (*ev->bank_name) ? ev->bank_name : NONE,
+			 ev->status,
+			 (*ev->error_msg) ? ev->error_msg : NONE,
+			 (*ev->mcistatus_msg) ? ev->mcistatus_msg : NONE,
+			 (*ev->mcastatus_msg) ? ev->mcastatus_msg : NONE,
+			 (*ev->user_action) ? ev->user_action : NONE,
+			 (*ev->mc_location) ? ev->mc_location : NONE,
+			 ev->cpuid,
+			 ev->cpu,
+			 ev->socketid,
+			 ev->ip,
+			 ev->cs,
+			 ev->misc,
+			 ev->addr,
+			 ev->synd,
+			 ev->ipid,
+			 (*ev->mcgstatus_msg) ? ev->mcgstatus_msg : NONE,
+			 ev->mcgstatus,
+			 ev->mcgcap,
+			 ev->apicid);
+}
+
diff --git a/ras-report.h b/ras-report.h
index a2edf3c..fb15dc3 100644
--- a/ras-report.h
+++ b/ras-report.h
@@ -30,6 +30,12 @@
 /* ABRT socket file */
 #define ABRT_SOCKET "/var/run/abrt/abrt.socket"
 
+#ifdef HAVE_JSON_REPORT
+#define JSON_REPORT_KEY "rasdaemon_event_name"
+
+extern int json_report;
+#endif
+
 #ifdef HAVE_ABRT_REPORT
 
 int ras_report_mc_event(struct ras_events *ras, struct ras_mc_event *ev);
@@ -54,4 +60,12 @@ static inline int ras_report_mf_event(struct ras_events *ras, struct ras_mf_even
 
 #endif
 
+#ifdef HAVE_JSON_REPORT
+void report_mc_event_json(struct trace_seq *s, struct ras_mc_event *ev);
+void report_aer_event_json(struct trace_seq *s, struct ras_aer_event *ev);
+void report_arm_event_json(struct trace_seq *s, struct ras_arm_event *ev);
+void report_mf_event_json(struct trace_seq *s, struct ras_mf_event *ev);
+void report_mce_event_json(struct trace_seq *s, struct mce_event *ev);
+#endif
+
 #endif
diff --git a/rasdaemon.c b/rasdaemon.c
index 88ba1ca..e0e85c1 100644
--- a/rasdaemon.c
+++ b/rasdaemon.c
@@ -26,6 +26,7 @@
 #include "ras-logger.h"
 #include "ras-events.h"
 #include "ras-record.h"
+#include "ras-report.h"
 
 /*
  * Arguments(argp) handling logic and main
@@ -130,10 +131,16 @@ int main(int argc, char *argv[])
 {
 	struct arguments args;
 	int idx = -1;
-	choices_disable = getenv(DISABLE);
 
 	choices_disable = getenv(DISABLE);
 
+#ifdef HAVE_JSON_REPORT
+	if (choices_disable &&
+	    strlen(choices_disable) != 0 &&
+	    strstr(choices_disable, "json_report"))
+		json_report = 0;
+#endif
+
 #ifdef HAVE_MCE
 	const struct argp_option offline_options[] = {
 		{"smca", SMCA, 0, 0, "AMD SMCA Error Decoding"},
-- 
2.33.1