Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kerne…

…l/git/tip/linux-2.6-tip * 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: perf tools: Fix sample size bit operations perf tools: Fix ommitted mmap data update on remap watchdog: Change the default timeout and configure nmi watchdog period based on watchdog_thresh watchdog: Disable watchdog when thresh is zero watchdog: Only disable/enable watchdog if neccessary watchdog: Fix rounding bug in get_sample_period() perf tools: Propagate event parse error handling perf tools: Robustify dynamic sample content fetch perf tools: Pre-check sample size before parsing perf tools: Move evlist sample helpers to evlist area perf tools: Remove junk code in mmap size handling perf tools: Check we are able to read the event size on mmap

Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kerne…
…l/git/tip/linux-2.6-tip * 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: perf tools: Fix sample size bit operations perf tools: Fix ommitted mmap data update on remap watchdog: Change the default timeout and configure nmi watchdog period based on watchdog_thresh watchdog: Disable watchdog when thresh is zero watchdog: Only disable/enable watchdog if neccessary watchdog: Fix rounding bug in get_sample_period() perf tools: Propagate event parse error handling perf tools: Robustify dynamic sample content fetch perf tools: Pre-check sample size before parsing perf tools: Move evlist sample helpers to evlist area perf tools: Remove junk code in mmap size handling perf tools: Check we are able to read the event size on mmap
Linus Torvalds
2 parents 57d19e80f4 3cb6d15408
Showing 18 changed files Side-by-side Diff
arch/x86/kernel/apic/hw_nmi.c
include/linux/nmi.h
include/linux/sched.h
kernel/sysctl.c
kernel/watchdog.c
tools/perf/builtin-test.c
tools/perf/builtin-top.c
tools/perf/util/event.c
tools/perf/util/event.h
tools/perf/util/evlist.c
tools/perf/util/evlist.h
tools/perf/util/evsel.c
tools/perf/util/header.c
tools/perf/util/header.h
tools/perf/util/include/linux/list.h
tools/perf/util/python.c
tools/perf/util/session.c
tools/perf/util/session.h
@@ -19,9 +19,9 @@
 #include <linux/delay.h>
  
 #ifdef CONFIG_HARDLOCKUP_DETECTOR
-u64 hw_nmi_get_sample_period(void)
+u64 hw_nmi_get_sample_period(int watchdog_thresh)
 {
-	return (u64)(cpu_khz) * 1000 * 60;
+	return (u64)(cpu_khz) * 1000 * watchdog_thresh;
 }
 #endif
  
@@ -45,11 +45,12 @@
  
 #ifdef CONFIG_LOCKUP_DETECTOR
 int hw_nmi_is_cpu_stuck(struct pt_regs *);
-u64 hw_nmi_get_sample_period(void);
+u64 hw_nmi_get_sample_period(int watchdog_thresh);
 extern int watchdog_enabled;
+extern int watchdog_thresh;
 struct ctl_table;
-extern int proc_dowatchdog_enabled(struct ctl_table *, int ,
-			void __user *, size_t *, loff_t *);
+extern int proc_dowatchdog(struct ctl_table *, int ,
+			   void __user *, size_t *, loff_t *);
 #endif
  
 #endif
@@ -315,7 +315,6 @@
 				  void __user *buffer,
 				  size_t *lenp, loff_t *ppos);
 extern unsigned int  softlockup_panic;
-extern int softlockup_thresh;
 void lockup_detector_init(void);
 #else
 static inline void touch_softlockup_watchdog(void)
@@ -730,14 +730,16 @@
 		.data           = &watchdog_enabled,
 		.maxlen         = sizeof (int),
 		.mode           = 0644,
-		.proc_handler   = proc_dowatchdog_enabled,
+		.proc_handler   = proc_dowatchdog,
+		.extra1		= &zero,
+		.extra2		= &one,
 	},
 	{
 		.procname	= "watchdog_thresh",
-		.data		= &softlockup_thresh,
+		.data		= &watchdog_thresh,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
-		.proc_handler	= proc_dowatchdog_thresh,
+		.proc_handler	= proc_dowatchdog,
 		.extra1		= &neg_one,
 		.extra2		= &sixty,
 	},
@@ -755,7 +757,9 @@
 		.data           = &watchdog_enabled,
 		.maxlen         = sizeof (int),
 		.mode           = 0644,
-		.proc_handler   = proc_dowatchdog_enabled,
+		.proc_handler   = proc_dowatchdog,
+		.extra1		= &zero,
+		.extra2		= &one,
 	},
 #endif
 #if defined(CONFIG_X86_LOCAL_APIC) && defined(CONFIG_X86)
@@ -28,7 +28,7 @@
 #include <linux/perf_event.h>
  
 int watchdog_enabled = 1;
-int __read_mostly softlockup_thresh = 60;
+int __read_mostly watchdog_thresh = 10;
  
 static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
 static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog);
@@ -91,6 +91,17 @@
 __setup("nosoftlockup", nosoftlockup_setup);
 /*  */
  
+/*
+ * Hard-lockup warnings should be triggered after just a few seconds. Soft-
+ * lockups can have false positives under extreme conditions. So we generally
+ * want a higher threshold for soft lockups than for hard lockups. So we couple
+ * the thresholds with a factor: we make the soft threshold twice the amount of
+ * time the hard threshold is.
+ */
+static int get_softlockup_thresh()
+{
+	return watchdog_thresh * 2;
+}
  
 /*
  * Returns seconds, approximately.  We don't need nanosecond
  
@@ -105,12 +116,12 @@
 static unsigned long get_sample_period(void)
 {
 	/*
-	 * convert softlockup_thresh from seconds to ns
+	 * convert watchdog_thresh from seconds to ns
 	 * the divide by 5 is to give hrtimer 5 chances to
 	 * increment before the hardlockup detector generates
 	 * a warning
 	 */
-	return softlockup_thresh / 5 * NSEC_PER_SEC;
+	return get_softlockup_thresh() * (NSEC_PER_SEC / 5);
 }
  
 /* Commands for resetting the watchdog */
@@ -182,7 +193,7 @@
 	unsigned long now = get_timestamp(smp_processor_id());
  
 	/* Warn about unreasonable delays: */
-	if (time_after(now, touch_ts + softlockup_thresh))
+	if (time_after(now, touch_ts + get_softlockup_thresh()))
 		return now - touch_ts;
  
 	return 0;
@@ -359,7 +370,7 @@
  
 	/* Try to register using hardware perf events */
 	wd_attr = &wd_hw_attr;
-	wd_attr->sample_period = hw_nmi_get_sample_period();
+	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
 	event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback);
 	if (!IS_ERR(event)) {
 		printk(KERN_INFO "NMI watchdog enabled, takes one hw-pmu counter.\n");
  
  
  
  
@@ -501,28 +512,25 @@
 /* sysctl functions */
 #ifdef CONFIG_SYSCTL
 /*
- * proc handler for /proc/sys/kernel/nmi_watchdog
+ * proc handler for /proc/sys/kernel/nmi_watchdog,watchdog_thresh
  */
  
-int proc_dowatchdog_enabled(struct ctl_table *table, int write,
-		     void __user *buffer, size_t *length, loff_t *ppos)
+int proc_dowatchdog(struct ctl_table *table, int write,
+		    void __user *buffer, size_t *lenp, loff_t *ppos)
 {
-	proc_dointvec(table, write, buffer, length, ppos);
+	int ret;
  
-	if (write) {
-		if (watchdog_enabled)
-			watchdog_enable_all_cpus();
-		else
-			watchdog_disable_all_cpus();
-	}
-	return 0;
-}
+	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
+	if (ret || !write)
+		goto out;
  
-int proc_dowatchdog_thresh(struct ctl_table *table, int write,
-			     void __user *buffer,
-			     size_t *lenp, loff_t *ppos)
-{
-	return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
+	if (watchdog_enabled && watchdog_thresh)
+		watchdog_enable_all_cpus();
+	else
+		watchdog_disable_all_cpus();
+
+out:
+	return ret;
 }
 #endif /* CONFIG_SYSCTL */
  
@@ -474,6 +474,7 @@
 	unsigned int nr_events[nsyscalls],
 		     expected_nr_events[nsyscalls], i, j;
 	struct perf_evsel *evsels[nsyscalls], *evsel;
+	int sample_size = perf_sample_size(attr.sample_type);
  
 	for (i = 0; i < nsyscalls; ++i) {
 		char name[64];
@@ -558,7 +559,13 @@
 			goto out_munmap;
 		}
  
-		perf_event__parse_sample(event, attr.sample_type, false, &sample);
+		err = perf_event__parse_sample(event, attr.sample_type, sample_size,
+					       false, &sample);
+		if (err) {
+			pr_err("Can't parse sample, err = %d\n", err);
+			goto out_munmap;
+		}
+
 		evsel = perf_evlist__id2evsel(evlist, sample.id);
 		if (evsel == NULL) {
 			pr_debug("event with id %" PRIu64
@@ -805,9 +805,14 @@
 {
 	struct perf_sample sample;
 	union perf_event *event;
+	int ret;
  
 	while ((event = perf_evlist__mmap_read(top.evlist, idx)) != NULL) {
-		perf_session__parse_sample(self, event, &sample);
+		ret = perf_session__parse_sample(self, event, &sample);
+		if (ret) {
+			pr_err("Can't parse sample, err = %d\n", ret);
+			continue;
+		}
  
 		if (event->header.type == PERF_RECORD_SAMPLE)
 			perf_event__process_sample(event, &sample, self);
@@ -9,21 +9,21 @@
 #include "thread_map.h"
  
 static const char *perf_event__names[] = {
-	[0]			 = "TOTAL",
-	[PERF_RECORD_MMAP]	 = "MMAP",
-	[PERF_RECORD_LOST]	 = "LOST",
-	[PERF_RECORD_COMM]	 = "COMM",
-	[PERF_RECORD_EXIT]	 = "EXIT",
-	[PERF_RECORD_THROTTLE]	 = "THROTTLE",
-	[PERF_RECORD_UNTHROTTLE] = "UNTHROTTLE",
-	[PERF_RECORD_FORK]	 = "FORK",
-	[PERF_RECORD_READ]	 = "READ",
-	[PERF_RECORD_SAMPLE]	 = "SAMPLE",
-	[PERF_RECORD_HEADER_ATTR]	 = "ATTR",
-	[PERF_RECORD_HEADER_EVENT_TYPE]	 = "EVENT_TYPE",
-	[PERF_RECORD_HEADER_TRACING_DATA]	 = "TRACING_DATA",
-	[PERF_RECORD_HEADER_BUILD_ID]	 = "BUILD_ID",
-	[PERF_RECORD_FINISHED_ROUND]	 = "FINISHED_ROUND",
+	[0]					= "TOTAL",
+	[PERF_RECORD_MMAP]			= "MMAP",
+	[PERF_RECORD_LOST]			= "LOST",
+	[PERF_RECORD_COMM]			= "COMM",
+	[PERF_RECORD_EXIT]			= "EXIT",
+	[PERF_RECORD_THROTTLE]			= "THROTTLE",
+	[PERF_RECORD_UNTHROTTLE]		= "UNTHROTTLE",
+	[PERF_RECORD_FORK]			= "FORK",
+	[PERF_RECORD_READ]			= "READ",
+	[PERF_RECORD_SAMPLE]			= "SAMPLE",
+	[PERF_RECORD_HEADER_ATTR]		= "ATTR",
+	[PERF_RECORD_HEADER_EVENT_TYPE]		= "EVENT_TYPE",
+	[PERF_RECORD_HEADER_TRACING_DATA]	= "TRACING_DATA",
+	[PERF_RECORD_HEADER_BUILD_ID]		= "BUILD_ID",
+	[PERF_RECORD_FINISHED_ROUND]		= "FINISHED_ROUND",
 };
  
 const char *perf_event__name(unsigned int id)
@@ -33,6 +33,22 @@
 	if (!perf_event__names[id])
 		return "UNKNOWN";
 	return perf_event__names[id];
+}
+
+int perf_sample_size(u64 sample_type)
+{
+	u64 mask = sample_type & PERF_SAMPLE_MASK;
+	int size = 0;
+	int i;
+
+	for (i = 0; i < 64; i++) {
+		if (mask & (1UL << i))
+			size++;
+	}
+
+	size *= sizeof(u64);
+
+	return size;
 }
  
 static struct perf_sample synth_sample = {
@@ -56,6 +56,13 @@
 	u64 id;
 };
  
+
+#define PERF_SAMPLE_MASK				\
+	(PERF_SAMPLE_IP | PERF_SAMPLE_TID |		\
+	 PERF_SAMPLE_TIME | PERF_SAMPLE_ADDR |		\
+	PERF_SAMPLE_ID | PERF_SAMPLE_STREAM_ID |	\
+	 PERF_SAMPLE_CPU | PERF_SAMPLE_PERIOD)
+
 struct sample_event {
 	struct perf_event_header        header;
 	u64 array[];
@@ -75,6 +82,8 @@
 	struct ip_callchain *callchain;
 };
  
+int perf_sample_size(u64 sample_type);
+
 #define BUILD_ID_SIZE 20
  
 struct build_id_event {
@@ -178,7 +187,8 @@
 const char *perf_event__name(unsigned int id);
  
 int perf_event__parse_sample(const union perf_event *event, u64 type,
-			     bool sample_id_all, struct perf_sample *sample);
+			     int sample_size, bool sample_id_all,
+			     struct perf_sample *sample);
  
 #endif /* __PERF_RECORD_H */
@@ -459,4 +459,35 @@
  
 	return 0;
 }
+
+u64 perf_evlist__sample_type(struct perf_evlist *evlist)
+{
+	struct perf_evsel *pos;
+	u64 type = 0;
+
+	list_for_each_entry(pos, &evlist->entries, node) {
+		if (!type)
+			type = pos->attr.sample_type;
+		else if (type != pos->attr.sample_type)
+			die("non matching sample_type");
+	}
+
+	return type;
+}
+
+bool perf_evlist__sample_id_all(const struct perf_evlist *evlist)
+{
+	bool value = false, first = true;
+	struct perf_evsel *pos;
+
+	list_for_each_entry(pos, &evlist->entries, node) {
+		if (first) {
+			value = pos->attr.sample_id_all;
+			first = false;
+		} else if (value != pos->attr.sample_id_all)
+			die("non matching sample_id_all");
+	}
+
+	return value;
+}
@@ -66,5 +66,8 @@
 void perf_evlist__delete_maps(struct perf_evlist *evlist);
 int perf_evlist__set_filters(struct perf_evlist *evlist);
  
+u64 perf_evlist__sample_type(struct perf_evlist *evlist);
+bool perf_evlist__sample_id_all(const struct perf_evlist *evlist);
+
 #endif /* __PERF_EVLIST_H */
@@ -303,8 +303,20 @@
 	return 0;
 }
  
+static bool sample_overlap(const union perf_event *event,
+			   const void *offset, u64 size)
+{
+	const void *base = event;
+
+	if (offset + size > base + event->header.size)
+		return true;
+
+	return false;
+}
+
 int perf_event__parse_sample(const union perf_event *event, u64 type,
-			     bool sample_id_all, struct perf_sample *data)
+			     int sample_size, bool sample_id_all,
+			     struct perf_sample *data)
 {
 	const u64 *array;
  
@@ -319,6 +331,9 @@
  
 	array = event->sample.array;
  
+	if (sample_size + sizeof(event->header) > event->header.size)
+		return -EFAULT;
+
 	if (type & PERF_SAMPLE_IP) {
 		data->ip = event->ip.ip;
 		array++;
  
  
  
@@ -369,14 +384,29 @@
 	}
  
 	if (type & PERF_SAMPLE_CALLCHAIN) {
+		if (sample_overlap(event, array, sizeof(data->callchain->nr)))
+			return -EFAULT;
+
 		data->callchain = (struct ip_callchain *)array;
+
+		if (sample_overlap(event, array, data->callchain->nr))
+			return -EFAULT;
+
 		array += 1 + data->callchain->nr;
 	}
  
 	if (type & PERF_SAMPLE_RAW) {
 		u32 *p = (u32 *)array;
+
+		if (sample_overlap(event, array, sizeof(u32)))
+			return -EFAULT;
+
 		data->raw_size = *p;
 		p++;
+
+		if (sample_overlap(event, p, data->raw_size))
+			return -EFAULT;
+
 		data->raw_data = p;
 	}
  
@@ -934,37 +934,6 @@
 	return -ENOMEM;
 }
  
-u64 perf_evlist__sample_type(struct perf_evlist *evlist)
-{
-	struct perf_evsel *pos;
-	u64 type = 0;
-
-	list_for_each_entry(pos, &evlist->entries, node) {
-		if (!type)
-			type = pos->attr.sample_type;
-		else if (type != pos->attr.sample_type)
-			die("non matching sample_type");
-	}
-
-	return type;
-}
-
-bool perf_evlist__sample_id_all(const struct perf_evlist *evlist)
-{
-	bool value = false, first = true;
-	struct perf_evsel *pos;
-
-	list_for_each_entry(pos, &evlist->entries, node) {
-		if (first) {
-			value = pos->attr.sample_id_all;
-			first = false;
-		} else if (value != pos->attr.sample_id_all)
-			die("non matching sample_id_all");
-	}
-
-	return value;
-}
-
 int perf_event__synthesize_attr(struct perf_event_attr *attr, u16 ids, u64 *id,
 				perf_event__handler_t process,
 				struct perf_session *session)
@@ -64,8 +64,6 @@
 int perf_header__push_event(u64 id, const char *name);
 char *perf_header__find_event(u64 id);
  
-u64 perf_evlist__sample_type(struct perf_evlist *evlist);
-bool perf_evlist__sample_id_all(const struct perf_evlist *evlist);
 void perf_header__set_feat(struct perf_header *header, int feat);
 void perf_header__clear_feat(struct perf_header *header, int feat);
 bool perf_header__has_feat(const struct perf_header *header, int feat);
 #include <linux/kernel.h>
+#include <linux/prefetch.h>
+
 #include "../../../../include/linux/list.h"
  
 #ifndef PERF_LIST_H
@@ -675,6 +675,7 @@
 	union perf_event *event;
 	int sample_id_all = 1, cpu;
 	static char *kwlist[] = {"sample_id_all", NULL, NULL};
+	int err;
  
 	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "i|i", kwlist,
 					 &cpu, &sample_id_all))
  
@@ -690,11 +691,17 @@
 			return PyErr_NoMemory();
  
 		first = list_entry(evlist->entries.next, struct perf_evsel, node);
-		perf_event__parse_sample(event, first->attr.sample_type, sample_id_all,
-					 &pevent->sample);
+		err = perf_event__parse_sample(event, first->attr.sample_type,
+					       perf_sample_size(first->attr.sample_type),
+					       sample_id_all, &pevent->sample);
+		if (err) {
+			pr_err("Can't parse sample, err = %d\n", err);
+			goto end;
+		}
+
 		return pyevent;
 	}
-
+end:
 	Py_INCREF(Py_None);
 	return Py_None;
 }
@@ -97,6 +97,7 @@
 void perf_session__update_sample_type(struct perf_session *self)
 {
 	self->sample_type = perf_evlist__sample_type(self->evlist);
+	self->sample_size = perf_sample_size(self->sample_type);
 	self->sample_id_all = perf_evlist__sample_id_all(self->evlist);
 	perf_session__id_header_size(self);
 }
@@ -479,6 +480,7 @@
 	struct perf_sample sample;
 	u64 limit = os->next_flush;
 	u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL;
+	int ret;
  
 	if (!ops->ordered_samples || !limit)
 		return;
@@ -487,9 +489,12 @@
 		if (iter->timestamp > limit)
 			break;
  
-		perf_session__parse_sample(s, iter->event, &sample);
-		perf_session_deliver_event(s, iter->event, &sample, ops,
-					   iter->file_offset);
+		ret = perf_session__parse_sample(s, iter->event, &sample);
+		if (ret)
+			pr_err("Can't parse sample, err = %d\n", ret);
+		else
+			perf_session_deliver_event(s, iter->event, &sample, ops,
+						   iter->file_offset);
  
 		os->last_flush = iter->timestamp;
 		list_del(&iter->list);
@@ -805,7 +810,9 @@
 	/*
 	 * For all kernel events we get the sample data
 	 */
-	perf_session__parse_sample(session, event, &sample);
+	ret = perf_session__parse_sample(session, event, &sample);
+	if (ret)
+		return ret;
  
 	/* Preprocess sample records - precheck callchains */
 	if (perf_session__preprocess_sample(session, event, &sample))
@@ -953,6 +960,30 @@
 	return err;
 }
  
+static union perf_event *
+fetch_mmaped_event(struct perf_session *session,
+		   u64 head, size_t mmap_size, char *buf)
+{
+	union perf_event *event;
+
+	/*
+	 * Ensure we have enough space remaining to read
+	 * the size of the event in the headers.
+	 */
+	if (head + sizeof(event->header) > mmap_size)
+		return NULL;
+
+	event = (union perf_event *)(buf + head);
+
+	if (session->header.needs_swap)
+		perf_event_header__bswap(&event->header);
+
+	if (head + event->header.size > mmap_size)
+		return NULL;
+
+	return event;
+}
+
 int __perf_session__process_events(struct perf_session *session,
 				   u64 data_offset, u64 data_size,
 				   u64 file_size, struct perf_event_ops *ops)
@@ -1007,15 +1038,8 @@
 	file_pos = file_offset + head;
  
 more:
-	event = (union perf_event *)(buf + head);
-
-	if (session->header.needs_swap)
-		perf_event_header__bswap(&event->header);
-	size = event->header.size;
-	if (size == 0)
-		size = 8;
-
-	if (head + event->header.size > mmap_size) {
+	event = fetch_mmaped_event(session, head, mmap_size, buf);
+	if (!event) {
 		if (mmaps[map_idx]) {
 			munmap(mmaps[map_idx], mmap_size);
 			mmaps[map_idx] = NULL;
@@ -43,6 +43,7 @@
 	 */
 	struct hists		hists;
 	u64			sample_type;
+	int			sample_size;
 	int			fd;
 	bool			fd_pipe;
 	bool			repipe;
@@ -159,6 +160,7 @@
 					     struct perf_sample *sample)
 {
 	return perf_event__parse_sample(event, session->sample_type,
+					session->sample_size,
 					session->sample_id_all, sample);
 }
...	...	@@ -19,9 +19,9 @@
19	19	#include <linux/delay.h>
20	20
21	21	#ifdef CONFIG_HARDLOCKUP_DETECTOR
22		-u64 hw_nmi_get_sample_period(void)
	22	+u64 hw_nmi_get_sample_period(int watchdog_thresh)
23	23	{
24		- return (u64)(cpu_khz) * 1000 * 60;
	24	+ return (u64)(cpu_khz) * 1000 * watchdog_thresh;
25	25	}
26	26	#endif
27	27
...	...	@@ -45,11 +45,12 @@
45	45
46	46	#ifdef CONFIG_LOCKUP_DETECTOR
47	47	int hw_nmi_is_cpu_stuck(struct pt_regs *);
48		-u64 hw_nmi_get_sample_period(void);
	48	+u64 hw_nmi_get_sample_period(int watchdog_thresh);
49	49	extern int watchdog_enabled;
	50	+extern int watchdog_thresh;
50	51	struct ctl_table;
51		-extern int proc_dowatchdog_enabled(struct ctl_table *, int ,
52		- void __user , size_t , loff_t *);
	52	+extern int proc_dowatchdog(struct ctl_table *, int ,
	53	+ void __user , size_t , loff_t *);
53	54	#endif
54	55
55	56	#endif
...	...	@@ -315,7 +315,6 @@
315	315	void __user *buffer,
316	316	size_t lenp, loff_t ppos);
317	317	extern unsigned int softlockup_panic;
318		-extern int softlockup_thresh;
319	318	void lockup_detector_init(void);
320	319	#else
321	320	static inline void touch_softlockup_watchdog(void)
...	...	@@ -730,14 +730,16 @@
730	730	.data = &watchdog_enabled,
731	731	.maxlen = sizeof (int),
732	732	.mode = 0644,
733		- .proc_handler = proc_dowatchdog_enabled,
	733	+ .proc_handler = proc_dowatchdog,
	734	+ .extra1 = &zero,
	735	+ .extra2 = &one,
734	736	},
735	737	{
736	738	.procname = "watchdog_thresh",
737		- .data = &softlockup_thresh,
	739	+ .data = &watchdog_thresh,
738	740	.maxlen = sizeof(int),
739	741	.mode = 0644,
740		- .proc_handler = proc_dowatchdog_thresh,
	742	+ .proc_handler = proc_dowatchdog,
741	743	.extra1 = &neg_one,
742	744	.extra2 = &sixty,
743	745	},
...	...	@@ -755,7 +757,9 @@
755	757	.data = &watchdog_enabled,
756	758	.maxlen = sizeof (int),
757	759	.mode = 0644,
758		- .proc_handler = proc_dowatchdog_enabled,
	760	+ .proc_handler = proc_dowatchdog,
	761	+ .extra1 = &zero,
	762	+ .extra2 = &one,
759	763	},
760	764	#endif
761	765	#if defined(CONFIG_X86_LOCAL_APIC) && defined(CONFIG_X86)
...	...	@@ -28,7 +28,7 @@
28	28	#include <linux/perf_event.h>
29	29
30	30	int watchdog_enabled = 1;
31		-int __read_mostly softlockup_thresh = 60;
	31	+int __read_mostly watchdog_thresh = 10;
32	32
33	33	static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
34	34	static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog);
...	...	@@ -91,6 +91,17 @@
91	91	__setup("nosoftlockup", nosoftlockup_setup);
92	92	/* */
93	93
	94	+/*
	95	+ * Hard-lockup warnings should be triggered after just a few seconds. Soft-
	96	+ * lockups can have false positives under extreme conditions. So we generally
	97	+ * want a higher threshold for soft lockups than for hard lockups. So we couple
	98	+ * the thresholds with a factor: we make the soft threshold twice the amount of
	99	+ * time the hard threshold is.
	100	+ */
	101	+static int get_softlockup_thresh()
	102	+{
	103	+ return watchdog_thresh * 2;
	104	+}
94	105
95	106	/*
96	107	* Returns seconds, approximately. We don't need nanosecond
97	108
...	...	@@ -105,12 +116,12 @@
105	116	static unsigned long get_sample_period(void)
106	117	{
107	118	/*
108		- * convert softlockup_thresh from seconds to ns
	119	+ * convert watchdog_thresh from seconds to ns
109	120	* the divide by 5 is to give hrtimer 5 chances to
110	121	* increment before the hardlockup detector generates
111	122	* a warning
112	123	*/
113		- return softlockup_thresh / 5 * NSEC_PER_SEC;
	124	+ return get_softlockup_thresh() * (NSEC_PER_SEC / 5);
114	125	}
115	126
116	127	/* Commands for resetting the watchdog */
...	...	@@ -182,7 +193,7 @@
182	193	unsigned long now = get_timestamp(smp_processor_id());
183	194
184	195	/* Warn about unreasonable delays: */
185		- if (time_after(now, touch_ts + softlockup_thresh))
	196	+ if (time_after(now, touch_ts + get_softlockup_thresh()))
186	197	return now - touch_ts;
187	198
188	199	return 0;
...	...	@@ -359,7 +370,7 @@
359	370
360	371	/* Try to register using hardware perf events */
361	372	wd_attr = &wd_hw_attr;
362		- wd_attr->sample_period = hw_nmi_get_sample_period();
	373	+ wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
363	374	event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback);
364	375	if (!IS_ERR(event)) {
365	376	printk(KERN_INFO "NMI watchdog enabled, takes one hw-pmu counter.\n");
366	377
367	378
368	379
369	380
...	...	@@ -501,28 +512,25 @@
501	512	/* sysctl functions */
502	513	#ifdef CONFIG_SYSCTL
503	514	/*
504		- * proc handler for /proc/sys/kernel/nmi_watchdog
	515	+ * proc handler for /proc/sys/kernel/nmi_watchdog,watchdog_thresh
505	516	*/
506	517
507		-int proc_dowatchdog_enabled(struct ctl_table *table, int write,
508		- void __user buffer, size_t length, loff_t *ppos)
	518	+int proc_dowatchdog(struct ctl_table *table, int write,
	519	+ void __user buffer, size_t lenp, loff_t *ppos)
509	520	{
510		- proc_dointvec(table, write, buffer, length, ppos);
	521	+ int ret;
511	522
512		- if (write) {
513		- if (watchdog_enabled)
514		- watchdog_enable_all_cpus();
515		- else
516		- watchdog_disable_all_cpus();
517		- }
518		- return 0;
519		-}
	523	+ ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
	524	+ if (ret \|\| !write)
	525	+ goto out;
520	526
521		-int proc_dowatchdog_thresh(struct ctl_table *table, int write,
522		- void __user *buffer,
523		- size_t lenp, loff_t ppos)
524		-{
525		- return proc_dointvec_minmax(table, write, buffer, lenp, ppos);
	527	+ if (watchdog_enabled && watchdog_thresh)
	528	+ watchdog_enable_all_cpus();
	529	+ else
	530	+ watchdog_disable_all_cpus();
	531	+
	532	+out:
	533	+ return ret;
526	534	}
527	535	#endif /* CONFIG_SYSCTL */
528	536
...	...	@@ -474,6 +474,7 @@
474	474	unsigned int nr_events[nsyscalls],
475	475	expected_nr_events[nsyscalls], i, j;
476	476	struct perf_evsel evsels[nsyscalls], evsel;
	477	+ int sample_size = perf_sample_size(attr.sample_type);
477	478
478	479	for (i = 0; i < nsyscalls; ++i) {
479	480	char name[64];
...	...	@@ -558,7 +559,13 @@
558	559	goto out_munmap;
559	560	}
560	561
561		- perf_event__parse_sample(event, attr.sample_type, false, &sample);
	562	+ err = perf_event__parse_sample(event, attr.sample_type, sample_size,
	563	+ false, &sample);
	564	+ if (err) {
	565	+ pr_err("Can't parse sample, err = %d\n", err);
	566	+ goto out_munmap;
	567	+ }
	568	+
562	569	evsel = perf_evlist__id2evsel(evlist, sample.id);
563	570	if (evsel == NULL) {
564	571	pr_debug("event with id %" PRIu64
...	...	@@ -805,9 +805,14 @@
805	805	{
806	806	struct perf_sample sample;
807	807	union perf_event *event;
	808	+ int ret;
808	809
809	810	while ((event = perf_evlist__mmap_read(top.evlist, idx)) != NULL) {
810		- perf_session__parse_sample(self, event, &sample);
	811	+ ret = perf_session__parse_sample(self, event, &sample);
	812	+ if (ret) {
	813	+ pr_err("Can't parse sample, err = %d\n", ret);
	814	+ continue;
	815	+ }
811	816
812	817	if (event->header.type == PERF_RECORD_SAMPLE)
813	818	perf_event__process_sample(event, &sample, self);
...	...	@@ -9,21 +9,21 @@
9	9	#include "thread_map.h"
10	10
11	11	static const char *perf_event__names[] = {
12		- [0] = "TOTAL",
13		- [PERF_RECORD_MMAP] = "MMAP",
14		- [PERF_RECORD_LOST] = "LOST",
15		- [PERF_RECORD_COMM] = "COMM",
16		- [PERF_RECORD_EXIT] = "EXIT",
17		- [PERF_RECORD_THROTTLE] = "THROTTLE",
18		- [PERF_RECORD_UNTHROTTLE] = "UNTHROTTLE",
19		- [PERF_RECORD_FORK] = "FORK",
20		- [PERF_RECORD_READ] = "READ",
21		- [PERF_RECORD_SAMPLE] = "SAMPLE",
22		- [PERF_RECORD_HEADER_ATTR] = "ATTR",
23		- [PERF_RECORD_HEADER_EVENT_TYPE] = "EVENT_TYPE",
24		- [PERF_RECORD_HEADER_TRACING_DATA] = "TRACING_DATA",
25		- [PERF_RECORD_HEADER_BUILD_ID] = "BUILD_ID",
26		- [PERF_RECORD_FINISHED_ROUND] = "FINISHED_ROUND",
	12	+ [0] = "TOTAL",
	13	+ [PERF_RECORD_MMAP] = "MMAP",
	14	+ [PERF_RECORD_LOST] = "LOST",
	15	+ [PERF_RECORD_COMM] = "COMM",
	16	+ [PERF_RECORD_EXIT] = "EXIT",
	17	+ [PERF_RECORD_THROTTLE] = "THROTTLE",
	18	+ [PERF_RECORD_UNTHROTTLE] = "UNTHROTTLE",
	19	+ [PERF_RECORD_FORK] = "FORK",
	20	+ [PERF_RECORD_READ] = "READ",
	21	+ [PERF_RECORD_SAMPLE] = "SAMPLE",
	22	+ [PERF_RECORD_HEADER_ATTR] = "ATTR",
	23	+ [PERF_RECORD_HEADER_EVENT_TYPE] = "EVENT_TYPE",
	24	+ [PERF_RECORD_HEADER_TRACING_DATA] = "TRACING_DATA",
	25	+ [PERF_RECORD_HEADER_BUILD_ID] = "BUILD_ID",
	26	+ [PERF_RECORD_FINISHED_ROUND] = "FINISHED_ROUND",
27	27	};
28	28
29	29	const char *perf_event__name(unsigned int id)
...	...	@@ -33,6 +33,22 @@
33	33	if (!perf_event__names[id])
34	34	return "UNKNOWN";
35	35	return perf_event__names[id];
	36	+}
	37	+
	38	+int perf_sample_size(u64 sample_type)
	39	+{
	40	+ u64 mask = sample_type & PERF_SAMPLE_MASK;
	41	+ int size = 0;
	42	+ int i;
	43	+
	44	+ for (i = 0; i < 64; i++) {
	45	+ if (mask & (1UL << i))
	46	+ size++;
	47	+ }
	48	+
	49	+ size *= sizeof(u64);
	50	+
	51	+ return size;
36	52	}
37	53
38	54	static struct perf_sample synth_sample = {
...	...	@@ -56,6 +56,13 @@
56	56	u64 id;
57	57	};
58	58
	59	+
	60	+#define PERF_SAMPLE_MASK \
	61	+ (PERF_SAMPLE_IP \| PERF_SAMPLE_TID \| \
	62	+ PERF_SAMPLE_TIME \| PERF_SAMPLE_ADDR \| \
	63	+ PERF_SAMPLE_ID \| PERF_SAMPLE_STREAM_ID \| \
	64	+ PERF_SAMPLE_CPU \| PERF_SAMPLE_PERIOD)
	65	+
59	66	struct sample_event {
60	67	struct perf_event_header header;
61	68	u64 array[];
...	...	@@ -75,6 +82,8 @@
75	82	struct ip_callchain *callchain;
76	83	};
77	84
	85	+int perf_sample_size(u64 sample_type);
	86	+
78	87	#define BUILD_ID_SIZE 20
79	88
80	89	struct build_id_event {
...	...	@@ -178,7 +187,8 @@
178	187	const char *perf_event__name(unsigned int id);
179	188
180	189	int perf_event__parse_sample(const union perf_event *event, u64 type,
181		- bool sample_id_all, struct perf_sample *sample);
	190	+ int sample_size, bool sample_id_all,
	191	+ struct perf_sample *sample);
182	192
183	193	#endif /* __PERF_RECORD_H */
...	...	@@ -459,4 +459,35 @@
459	459
460	460	return 0;
461	461	}
	462	+
	463	+u64 perf_evlist__sample_type(struct perf_evlist *evlist)
	464	+{
	465	+ struct perf_evsel *pos;
	466	+ u64 type = 0;
	467	+
	468	+ list_for_each_entry(pos, &evlist->entries, node) {
	469	+ if (!type)
	470	+ type = pos->attr.sample_type;
	471	+ else if (type != pos->attr.sample_type)
	472	+ die("non matching sample_type");
	473	+ }
	474	+
	475	+ return type;
	476	+}
	477	+
	478	+bool perf_evlist__sample_id_all(const struct perf_evlist *evlist)
	479	+{
	480	+ bool value = false, first = true;
	481	+ struct perf_evsel *pos;
	482	+
	483	+ list_for_each_entry(pos, &evlist->entries, node) {
	484	+ if (first) {
	485	+ value = pos->attr.sample_id_all;
	486	+ first = false;
	487	+ } else if (value != pos->attr.sample_id_all)
	488	+ die("non matching sample_id_all");
	489	+ }
	490	+
	491	+ return value;
	492	+}