Merge tag 'perf-core-for-mingo-4.20-20181025' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/urgent

Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

- Introduce 'perf trace --max-events' for stopping 'perf trace' when
  that many syscalls (enter+exit), tracepoints or other events such as
  page faults take place.

  Support that as well on a per-event basis, e.g.:

   perf trace -e sched:*switch/nr=2/,block:*_plug/nr=4/,block:*_unplug/nr=1/,net:*dev_queue/nr=3,max-stack=16/

  Will stop when 2 context switches, 4 block plugs, 1 block unplug and
  3 net_dev_queue tracepoints take place. (Arnaldo Carvalho de Melo)

- Poll for monitored tasks being alive in 'perf stat -p/-t', exiting when
  those tasks all terminate (Jiri Olsa)

- Encode -k clockid frequency into perf.data to enable timestamps derived
  metrics conversion into wall clock time on reporting stage. (Alexey Budankov)

- Improve Intel PT call graph from SQL database and GUI python scripts,
  including adopting the Qt MDI interface to allow for multiple subwindows
  for all the tables, helping in better visualizing the data in the SQL
  tables, also uses, when available, the Intel XED disassembler libraries
  to present the Intel PT data as x86 asm mnemonics. This last feature
  is not currently working in some cases, fix is being discussed (Adrian Hunter)

- Implement a ftrace function_graph view in 'perf script' when processing
  hardware trace data such as Intel PT (Andi Kleen)

- Better integration with the Intel XED disassembler, when available, in
  'perf script' (Andi Kleen)

- Some 'perf trace' drop refcount fixes (Arnaldo Carvalho de Melo)

- Add Sparc support to 'perf annotate', jitdump (David Miller)

- Fix PLT symbols entry/header sizes properly on Sparc (David Miller)

- Fix generation of system call table failure with /tmp mounted with 'noexec'
  in arm64 (Hongxu Jia)

- Allow extended console debug output in 'perf script' (Milian Wolff)

- Flush output stream after events in 'perf script' verbose mode (Milian Wolff)

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

1 files changed, 44 insertions, 6 deletions

diff --git a/tools/perf/builtin-trace.c b/tools/perf/builtin-trace.c
index 90289f31dd87..7081d7ea12e5 100644
--- a/tools/perf/builtin-trace.c
+++ b/tools/perf/builtin-trace.c
@@ -89,6 +89,8 @@ struct trace {
 	u64			base_time;
 	FILE			*output;
 	unsigned long		nr_events;
+	unsigned long		nr_events_printed;
+	unsigned long		max_events;
 	struct strlist		*ev_qualifier;
 	struct {
 		size_t		nr;
@@ -1664,6 +1666,8 @@ static int trace__printf_interrupted_entry(struct trace *trace)
 	printed += fprintf(trace->output, "%-70s) ...\n", ttrace->entry_str);
 	ttrace->entry_pending = false;
 
+	++trace->nr_events_printed;
+
 	return printed;
 }
 
@@ -1810,12 +1814,14 @@ static int trace__resolve_callchain(struct trace *trace, struct perf_evsel *evse
 	int max_stack = evsel->attr.sample_max_stack ?
 			evsel->attr.sample_max_stack :
 			trace->max_stack;
+	int err;
 
-	if (machine__resolve(trace->host, &al, sample) < 0 ||
-	    thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack))
+	if (machine__resolve(trace->host, &al, sample) < 0)
 		return -1;
 
-	return 0;
+	err = thread__resolve_callchain(al.thread, cursor, evsel, sample, NULL, NULL, max_stack);
+	addr_location__put(&al);
+	return err;
 }
 
 static int trace__fprintf_callchain(struct trace *trace, struct perf_sample *sample)
@@ -1940,6 +1946,13 @@ errno_print: {
 
 	fputc('\n', trace->output);
 
+	/*
+	 * We only consider an 'event' for the sake of --max-events a non-filtered
+	 * sys_enter + sys_exit and other tracepoint events.
+	 */
+	if (++trace->nr_events_printed == trace->max_events && trace->max_events != ULONG_MAX)
+		interrupted = true;
+
 	if (callchain_ret > 0)
 		trace__fprintf_callchain(trace, sample);
 	else if (callchain_ret < 0)
@@ -2072,14 +2085,25 @@ static void bpf_output__fprintf(struct trace *trace,
 {
 	binary__fprintf(sample->raw_data, sample->raw_size, 8,
 			bpf_output__printer, NULL, trace->output);
+	++trace->nr_events_printed;
 }
 
 static int trace__event_handler(struct trace *trace, struct perf_evsel *evsel,
 				union perf_event *event __maybe_unused,
 				struct perf_sample *sample)
 {
-	struct thread *thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
+	struct thread *thread;
 	int callchain_ret = 0;
+	/*
+	 * Check if we called perf_evsel__disable(evsel) due to, for instance,
+	 * this event's max_events having been hit and this is an entry coming
+	 * from the ring buffer that we should discard, since the max events
+	 * have already been considered/printed.
+	 */
+	if (evsel->disabled)
+		return 0;
+
+	thread = machine__findnew_thread(trace->host, sample->pid, sample->tid);
 
 	if (sample->callchain) {
 		callchain_ret = trace__resolve_callchain(trace, evsel, sample, &callchain_cursor);
@@ -2127,6 +2151,12 @@ static int trace__event_handler(struct trace *trace, struct perf_evsel *evsel,
 			event_format__fprintf(evsel->tp_format, sample->cpu,
 					      sample->raw_data, sample->raw_size,
 					      trace->output);
+			++trace->nr_events_printed;
+
+			if (evsel->max_events != ULONG_MAX && ++evsel->nr_events_printed == evsel->max_events) {
+				perf_evsel__disable(evsel);
+				perf_evsel__close(evsel);
+			}
 		}
 	}
 
@@ -2137,8 +2167,8 @@ newline:
 		trace__fprintf_callchain(trace, sample);
 	else if (callchain_ret < 0)
 		pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
-	thread__put(thread);
 out:
+	thread__put(thread);
 	return 0;
 }
 
@@ -2225,6 +2255,8 @@ static int trace__pgfault(struct trace *trace,
 		trace__fprintf_callchain(trace, sample);
 	else if (callchain_ret < 0)
 		pr_err("Problem processing %s callchain, skipping...\n", perf_evsel__name(evsel));
+
+	++trace->nr_events_printed;
 out:
 	err = 0;
 out_put:
@@ -2402,6 +2434,9 @@ static void trace__handle_event(struct trace *trace, union perf_event *event, st
 		tracepoint_handler handler = evsel->handler;
 		handler(trace, evsel, event, sample);
 	}
+
+	if (trace->nr_events_printed >= trace->max_events && trace->max_events != ULONG_MAX)
+		interrupted = true;
 }
 
 static int trace__add_syscall_newtp(struct trace *trace)
@@ -2706,7 +2741,7 @@ next_event:
 		int timeout = done ? 100 : -1;
 
 		if (!draining && perf_evlist__poll(evlist, timeout) > 0) {
-			if (perf_evlist__filter_pollfd(evlist, POLLERR | POLLHUP) == 0)
+			if (perf_evlist__filter_pollfd(evlist, POLLERR | POLLHUP | POLLNVAL) == 0)
 				draining = true;
 
 			goto again;
@@ -3249,6 +3284,7 @@ int cmd_trace(int argc, const char **argv)
 		.trace_syscalls = false,
 		.kernel_syscallchains = false,
 		.max_stack = UINT_MAX,
+		.max_events = ULONG_MAX,
 	};
 	const char *output_name = NULL;
 	const struct option trace_options[] = {
@@ -3301,6 +3337,8 @@ int cmd_trace(int argc, const char **argv)
 		     &record_parse_callchain_opt),
 	OPT_BOOLEAN(0, "kernel-syscall-graph", &trace.kernel_syscallchains,
 		    "Show the kernel callchains on the syscall exit path"),
+	OPT_ULONG(0, "max-events", &trace.max_events,
+		"Set the maximum number of events to print, exit after that is reached. "),
 	OPT_UINTEGER(0, "min-stack", &trace.min_stack,
 		     "Set the minimum stack depth when parsing the callchain, "
 		     "anything below the specified depth will be ignored."),