[PATCH 2/7] perf trace: add support for pagefault tracing

[Stanislav Fomichev]

This patch adds optional pagefault tracing support to 'perf trace'.
Using -F/--pf option user can specify whether he wants minor, major or
all pagefault events to be traced. This patch adds only live mode,
record and replace will come in a separate patch.

Example output:
1756272.905 ( 0.000 ms): curl/5937 majfault [0x7fa7261978b6] => /usr/lib/x86_64-linux-gnu/libkrb5.so.26.0.0+0x85288 (d.)
1862866.036 ( 0.000 ms): wget/8460 majfault [__clear_user+0x3f] => 0x659cb4 (?k)

Signed-off-by: Stanislav Fomichev <stfomichev@xxxxxxxxxxxxxx>
---
 tools/perf/Documentation/perf-trace.txt |  12 ++++
 tools/perf/builtin-trace.c              | 116 +++++++++++++++++++++++++++++++-
 2 files changed, 127 insertions(+), 1 deletion(-)

diff --git a/tools/perf/Documentation/perf-trace.txt b/tools/perf/Documentation/perf-trace.txt
index fae38d9a44a4..8e5f710aa45d 100644
--- a/tools/perf/Documentation/perf-trace.txt
+++ b/tools/perf/Documentation/perf-trace.txt
@@ -107,6 +107,18 @@ the thread executes on the designated CPUs. Default is to monitor all CPUs.
 	Show tool stats such as number of times fd->pathname was discovered thru
 	hooking the open syscall return + vfs_getname or via reading /proc/pid/fd, etc.
 
+-F=[all|min|maj]::
+--pf=[all|min|maj]::
+	Trace pagefaults. Optionally, you can specify whether you want minor,
+	major or all pagefaults. Default value is maj.
+
+EXAMPLES
+--------
+
+Trace syscalls, major and minor pagefaults:
+
+ $ perf trace -F all
+
 SEE ALSO
 --------
 linkperf:perf-record[1], linkperf:perf-script[1]
diff --git a/tools/perf/builtin-trace.c b/tools/perf/builtin-trace.c
index a9b542918da0..a80aae2bba40 100644
--- a/tools/perf/builtin-trace.c
+++ b/tools/perf/builtin-trace.c
@@ -1177,6 +1177,9 @@ fail:
 	return NULL;
 }
 
+#define TRACE_PFMAJ		(1 << 0)
+#define TRACE_PFMIN		(1 << 1)
+
 struct trace {
 	struct perf_tool	tool;
 	struct {
@@ -1211,6 +1214,7 @@ struct trace {
 	bool			summary_only;
 	bool			show_comm;
 	bool			show_tool_stats;
+	int			trace_pgfaults;
 };
 
 static int trace__set_fd_pathname(struct thread *thread, int fd, const char *pathname)
@@ -1773,6 +1777,59 @@ out_dump:
 	return 0;
 }
 
+static int trace__pgfault(struct trace *trace,
+			  struct perf_evsel *evsel,
+			  union perf_event *event,
+			  struct perf_sample *sample)
+{
+	struct thread *thread;
+	u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
+	struct addr_location al;
+	char map_type = 'd';
+
+	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
+
+	thread__find_addr_location(thread, trace->host, cpumode, MAP__FUNCTION,
+			      sample->ip, &al);
+
+	trace__fprintf_entry_head(trace, thread, 0, sample->time, trace->output);
+
+	fprintf(trace->output, "%sfault ",
+		evsel->attr.config == PERF_COUNT_SW_PAGE_FAULTS_MAJ ?
+		"maj" : "min");
+
+	if (al.sym)
+		fprintf(trace->output, "[%s+0x%lx]",
+			al.sym->name, al.addr - al.sym->start);
+	else
+		fprintf(trace->output, "[0x%lx]", sample->ip);
+
+	fprintf(trace->output, " => ");
+
+	thread__find_addr_map(thread, trace->host, cpumode, MAP__VARIABLE,
+			      sample->addr, &al);
+
+	if (!al.map) {
+		thread__find_addr_map(thread, trace->host, cpumode,
+				      MAP__FUNCTION, sample->addr, &al);
+
+		if (al.map)
+			map_type = 'x';
+	}
+
+	if (al.map) {
+		fprintf(trace->output, "%s+0x%lx",
+			al.map->dso->long_name, al.addr);
+	} else {
+		map_type = '?';
+		fprintf(trace->output, "0x%lx", sample->addr);
+	}
+
+	fprintf(trace->output, " (%c%c)\n", map_type, al.level);
+
+	return 0;
+}
+
 static bool skip_sample(struct trace *trace, struct perf_sample *sample)
 {
 	if ((trace->pid_list && intlist__find(trace->pid_list, sample->pid)) ||
@@ -1887,6 +1944,30 @@ static void perf_evlist__add_vfs_getname(struct perf_evlist *evlist)
 	perf_evlist__add(evlist, evsel);
 }
 
+static int perf_evlist__add_pgfault(struct perf_evlist *evlist,
+				    u64 config)
+{
+	struct perf_evsel *evsel;
+	struct perf_event_attr attr = {
+		.type = PERF_TYPE_SOFTWARE,
+		.mmap_data = 1,
+		.sample_period = 1,
+	};
+
+	attr.config = config;
+
+	event_attr_init(&attr);
+
+	evsel = perf_evsel__new(&attr);
+	if (!evsel)
+		return -ENOMEM;
+
+	evsel->handler = trace__pgfault;
+	perf_evlist__add(evlist, evsel);
+
+	return 0;
+}
+
 static int trace__run(struct trace *trace, int argc, const char **argv)
 {
 	struct perf_evlist *evlist = perf_evlist__new();
@@ -1907,6 +1988,14 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
 
 	perf_evlist__add_vfs_getname(evlist);
 
+	if ((trace->trace_pgfaults & TRACE_PFMAJ) &&
+	    perf_evlist__add_pgfault(evlist, PERF_COUNT_SW_PAGE_FAULTS_MAJ))
+		goto out_error_tp;
+
+	if ((trace->trace_pgfaults & TRACE_PFMIN) &&
+	    perf_evlist__add_pgfault(evlist, PERF_COUNT_SW_PAGE_FAULTS_MIN))
+		goto out_error_tp;
+
 	if (trace->sched &&
 		perf_evlist__add_newtp(evlist, "sched", "sched_stat_runtime",
 				trace__sched_stat_runtime))
@@ -1987,7 +2076,8 @@ again:
 				goto next_event;
 			}
 
-			if (sample.raw_data == NULL) {
+			if (evsel->attr.type == PERF_TYPE_TRACEPOINT &&
+			    sample.raw_data == NULL) {
 				fprintf(trace->output, "%s sample with no payload for tid: %d, cpu %d, raw_size=%d, skipping...\n",
 				       perf_evsel__name(evsel), sample.tid,
 				       sample.cpu, sample.raw_size);
@@ -2269,6 +2359,23 @@ static int trace__open_output(struct trace *trace, const char *filename)
 	return trace->output == NULL ? -errno : 0;
 }
 
+static int parse_pagefaults(const struct option *opt, const char *str,
+			    int unset __maybe_unused)
+{
+	int *trace_pgfaults = opt->value;
+
+	if (strcmp(str, "all") == 0)
+		*trace_pgfaults |= TRACE_PFMAJ | TRACE_PFMIN;
+	else if (strcmp(str, "maj") == 0)
+		*trace_pgfaults |= TRACE_PFMAJ;
+	else if (strcmp(str, "min") == 0)
+		*trace_pgfaults |= TRACE_PFMIN;
+	else
+		return -1;
+
+	return 0;
+}
+
 int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
 {
 	const char * const trace_usage[] = {
@@ -2335,6 +2442,8 @@ int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
 		    "Show only syscall summary with statistics"),
 	OPT_BOOLEAN('S', "with-summary", &trace.summary,
 		    "Show all syscalls and summary with statistics"),
+	OPT_CALLBACK_DEFAULT('F', "pf", &trace.trace_pgfaults, "all|maj|min",
+		     "Trace pagefaults", parse_pagefaults, "maj"),
 	OPT_END()
 	};
 	int err;
@@ -2349,6 +2458,11 @@ int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
 	if (trace.summary_only)
 		trace.summary = trace.summary_only;
 
+	if (trace.trace_pgfaults) {
+		trace.opts.sample_address = true;
+		trace.opts.sample_time = true;
+	}
+
 	if (output_name != NULL) {
 		err = trace__open_output(&trace, output_name);
 		if (err < 0) {
-- 
1.8.3.2

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/