Skip to contentSkip to conversation

perf intel-pt: Fix occasional decoding errors when tracing system-wide

PATCH New Adrian Hunter 2 participants 2 messages Aug 15, 2016 – Aug 16, 2016 perf events

lore
Share & tools
lore

Adrian Hunter

author reviewer
#0 lore raw

PATCH 1/1 perf intel-pt: Fix occasional decoding errors when tracing system-wide

In order to successfully decode Intel PT traces, context switch events are needed from the moment the trace starts. Currently that is ensured by using the 'immediate' flag which enables the switch event when it is opened. However, since commit 86c2786994bd ("perf intel-pt: Add support for PERF_RECORD_SWITCH") that might not always happen. When tracing system-wide the context switch event is added to the tracking event which was not set as 'immediate'. Change that so it is.

Fixes 86c2786994bd ("perf intel-pt: Add support for PERF_RECORD_SWITCH")
---
 tools/perf/arch/x86/util/intel-pt.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
tools/perf/arch/x86/util/intel-pt.c +5 -1
index fb51457ba338..a2412e9d883b 100644
@@ -501,7 +501,7 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
 	struct intel_pt_recording *ptr =
 			container_of(itr, struct intel_pt_recording, itr);
 	struct perf_pmu *intel_pt_pmu = ptr->intel_pt_pmu;
-	bool have_timing_info;
+	bool have_timing_info, need_immediate = false;
 	struct perf_evsel *evsel, *intel_pt_evsel = NULL;
 	const struct cpu_map *cpus = evlist->cpus;
 	bool privileged = geteuid() == 0 || perf_event_paranoid() < 0;
@@ -655,6 +655,7 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
 				ptr->have_sched_switch = 3;
 			} else {
 				opts->record_switch_events = true;
+				need_immediate = true;
 				if (cpu_wide)
 					ptr->have_sched_switch = 3;
 				else
@@ -700,6 +701,9 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
 		tracking_evsel->attr.freq = 0;
 		tracking_evsel->attr.sample_period = 1;
 
+		if (need_immediate)
+			tracking_evsel->immediate = true;
+
 		/* In per-cpu case, always need the time of mmap events etc */
 		if (!cpu_map__empty(cpus)) {
 			perf_evsel__set_sample_bit(tracking_evsel, TIME);
Signature · 1 line
1.9.1
#1 lore raw
perf intel-pt: Fix occasional decoding errors when tracing system-wide
Commit-ID:  3d918fb13abdbeca7947578f5d7e426eafad7f5e
Gitweb:     http://git.kernel.org/tip/3d918fb13abdbeca7947578f5d7e426eafad7f5e
Author:     Adrian Hunter <[email protected]>
AuthorDate: Mon, 15 Aug 2016 10:23:04 +0300
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Mon, 15 Aug 2016 18:11:18 -0300

perf intel-pt: Fix occasional decoding errors when tracing system-wide

In order to successfully decode Intel PT traces, context switch events are needed from the moment the trace starts. Currently that is ensured by using the 'immediate' flag which enables the switch event when it is opened.

However, since commit 86c2786994bd ("perf intel-pt: Add support for PERF_RECORD_SWITCH") that might not always happen. When tracing system-wide the context switch event is added to the tracking event which was not set as 'immediate'. Change that so it is.

Cc Jiri Olsa <[email protected]>
Fixes 86c2786994bd ("perf intel-pt: Add support for PERF_RECORD_SWITCH")
---
 tools/perf/arch/x86/util/intel-pt.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
tools/perf/arch/x86/util/intel-pt.c +5 -1
index fb51457..a2412e9 100644
@@ -501,7 +501,7 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
 	struct intel_pt_recording *ptr =
 			container_of(itr, struct intel_pt_recording, itr);
 	struct perf_pmu *intel_pt_pmu = ptr->intel_pt_pmu;
-	bool have_timing_info;
+	bool have_timing_info, need_immediate = false;
 	struct perf_evsel *evsel, *intel_pt_evsel = NULL;
 	const struct cpu_map *cpus = evlist->cpus;
 	bool privileged = geteuid() == 0 || perf_event_paranoid() < 0;
@@ -655,6 +655,7 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
 				ptr->have_sched_switch = 3;
 			} else {
 				opts->record_switch_events = true;
+				need_immediate = true;
 				if (cpu_wide)
 					ptr->have_sched_switch = 3;
 				else
@@ -700,6 +701,9 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
 		tracking_evsel->attr.freq = 0;
 		tracking_evsel->attr.sample_period = 1;
 
+		if (need_immediate)
+			tracking_evsel->immediate = true;
+
 		/* In per-cpu case, always need the time of mmap events etc */
 		if (!cpu_map__empty(cpus)) {
 			perf_evsel__set_sample_bit(tracking_evsel, TIME);