Skip to contentSkip to conversation

Freq/CPU%/CORE_BUSY% support

PATCH Superseded [email protected] 2 participants 22 messages Jul 24, 2015 – Jul 28, 2015 perf events

v2 of 10 Superseded Latest is v10 →

lore
Share & tools
lore
#0 lore raw

v2 cover Freq/CPU%/CORE_BUSY% support

From: Kan Liang <[email protected]>

This patch set supports per-sample freq/CPU%/CORE_BUSY% print in perf report -D and --stdio.
For printing these information, the perf.data file must have been obtained by group read and using special events cycles, ref-cycles, msr/tsc/, msr/aperf/ or msr/mperf/.

- Freq (MHz): The frequency during the sample interval. Needs cycles ref-cycles event.
- CPU%: CPU utilization during the sample interval. Needs ref-cycles and msr/tsc/ events.
- CORE_BUSY%: actual percent performance (APERF/MPERF%) during the sample interval. Needs msr/aperf/ and msr/mperf/ events.

For printing CPU% and CORE_BUSY%, please also apply the kernel patch.
http://marc.info/?l=linux-kernel&m=143747254926369&w=2

Here is an example:

$ perf record -e
'{cycles,ref-cycles,msr/tsc/,msr/mperf/,msr/aperf/}:S' ~/tchain_edit
$ perf report --stdio --group --show-freq-perf

Overhead FREQ MHz CPU% CORE_BUSY%
Command Shared Object Symbol
........................................ ......... ..... ..........
........... ................ ......................

    99.54%  99.54%  99.53%  99.53%  99.53%       2301     96         99
tchain_edit  tchain_edit       [.] f3
     0.20%   0.20%   0.20%   0.20%   0.20%       2301     98         99
tchain_edit  tchain_edit       [.] f2
     0.05%   0.05%   0.05%   0.05%   0.05%       2300     98         99
tchain_edit  [kernel.vmlinux]  [k] read_tsc

Changes since V1:
- Save cpu max freq to header when recording
- Read cpu max freq and msr type from header when reporting

Kan Liang (6):
perf,tools: save cpu max freq in perf header perf,tools: read cpu max freq and msr type from header perf,tools: Dump per-sample freq/CPU%/CORE_BUSY% in report -D perf,tools: save misc sample read value in struct perf_sample perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat perf,tools: Show freq/CPU%/CORE_BUSY% in perf report --stdio

 tools/perf/Documentation/perf-report.txt | 12 ++++++
 tools/perf/builtin-annotate.c            |  2 +-
 tools/perf/builtin-diff.c                |  2 +-
 tools/perf/builtin-report.c              | 22 ++++++++++
 tools/perf/perf.h                        |  1 +
 tools/perf/tests/hists_link.c            |  4 +-
 tools/perf/ui/hist.c                     | 71 +++++++++++++++++++++++++++++---
 tools/perf/util/cpumap.c                 | 32 ++++++++++++++
 tools/perf/util/cpumap.h                 |  2 +
 tools/perf/util/event.h                  | 11 +++++
 tools/perf/util/header.c                 | 56 +++++++++++++++++++++++++
 tools/perf/util/header.h                 |  4 ++
 tools/perf/util/hist.c                   | 51 ++++++++++++++++++++---
 tools/perf/util/hist.h                   |  5 +++
 tools/perf/util/session.c                | 49 +++++++++++++++++++---
 tools/perf/util/session.h                | 29 +++++++++++++
 tools/perf/util/sort.c                   |  3 ++
 tools/perf/util/sort.h                   |  3 ++
 tools/perf/util/symbol.h                 |  9 +++-
 tools/perf/util/util.c                   |  2 +
 20 files changed, 349 insertions(+), 21 deletions(-)
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH v2 1/6 perf,tools: save cpu max freq in perf header

From: Kan Liang <[email protected]>

Get cpu max frequency from the first online cpu, and save the MHz value in perf.data header.

---
 tools/perf/util/cpumap.c | 32 ++++++++++++++++++++++++++++++++
 tools/perf/util/cpumap.h |  1 +
 tools/perf/util/header.c | 35 +++++++++++++++++++++++++++++++++++
 tools/perf/util/header.h |  2 ++
 4 files changed, 70 insertions(+)
tools/perf/util/cpumap.c +32 -0
index 3667e21..548ef13 100644
@@ -499,3 +499,35 @@ int cpu__setup_cpunode_map(void)
 	closedir(dir1);
 	return 0;
 }
+
+unsigned int get_cpu_max_freq(void)
+{
+	const char *mnt;
+	char path[PATH_MAX], tmp;
+	FILE *fp;
+	unsigned int freq;
+	int cpu = 0;
+	int ret;
+
+	mnt = sysfs__mountpoint();
+	if (!mnt)
+		return 0;
+
+	snprintf(path, PATH_MAX, "%s/devices/system/cpu/online", mnt);
+	fp = fopen(path, "r");
+	if (fp) {
+		ret = fscanf(fp, "%u%c", &cpu, &tmp);
+		fclose(fp);
+		if (ret < 1)
+			return 0;
+	}
+
+	snprintf(path, PATH_MAX, "%s/devices/system/cpu/cpu%d/cpufreq/cpuinfo_max_freq", mnt, cpu);
+	fp = fopen(path, "r");
+	if (!fp)
+		return 0;
+	ret = fscanf(fp, "%u", &freq);
+	fclose(fp);
+
+	return (ret == 1) ? freq : 0;
+}
tools/perf/util/cpumap.h +1 -0
index 0af9cec..6784677 100644
@@ -58,6 +58,7 @@ int max_node_num;
 int *cpunode_map;
 
 int cpu__setup_cpunode_map(void);
+unsigned int get_cpu_max_freq(void);
 
 static inline int cpu__max_node(void)
 {
tools/perf/util/header.c +35 -0
index 03ace57..287a488 100644
@@ -862,6 +862,16 @@ write_it:
 	return do_write_string(fd, buffer);
 }
 
+static int write_cpu_max_freq(int fd, struct perf_header *h __maybe_unused,
+			      struct perf_evlist *evlist __maybe_unused)
+{
+	u32 freq;
+
+	freq = get_cpu_max_freq() / 1000;
+
+	return do_write(fd, &freq, sizeof(freq));
+}
+
 static int write_branch_stack(int fd __maybe_unused,
 			      struct perf_header *h __maybe_unused,
 		       struct perf_evlist *evlist __maybe_unused)
@@ -1158,6 +1168,11 @@ static void print_cpuid(struct perf_header *ph, int fd __maybe_unused, FILE *fp)
 	fprintf(fp, "# cpuid : %s\n", ph->env.cpuid);
 }
 
+static void print_cpu_max_freq(struct perf_header *ph, int fd __maybe_unused, FILE *fp)
+{
+	fprintf(fp, "# CPU max frequency : %u MHz\n", ph->env.cpu_max_freq);
+}
+
 static void print_branch_stack(struct perf_header *ph __maybe_unused,
 			       int fd __maybe_unused, FILE *fp)
 {
@@ -1471,6 +1486,25 @@ static int process_cpuid(struct perf_file_section *section __maybe_unused,
 	return ph->env.cpuid ? 0 : -ENOMEM;
 }
 
+static int process_cpu_max_freq(struct perf_file_section *section __maybe_unused,
+				struct perf_header *ph, int fd,
+				void *data __maybe_unused)
+{
+	ssize_t ret;
+	u32 nr;
+
+	ret = readn(fd, &nr, sizeof(nr));
+	if (ret != sizeof(nr))
+		return -1;
+
+	if (ph->needs_swap)
+		nr = bswap_32(nr);
+
+	ph->env.cpu_max_freq = nr;
+
+	return 0;
+}
+
 static int process_total_mem(struct perf_file_section *section __maybe_unused,
 			     struct perf_header *ph, int fd,
 			     void *data __maybe_unused)
@@ -1885,6 +1919,7 @@ static const struct feature_ops feat_ops[HEADER_LAST_FEATURE] = {
 	FEAT_OPP(HEADER_NRCPUS,		nrcpus),
 	FEAT_OPP(HEADER_CPUDESC,	cpudesc),
 	FEAT_OPP(HEADER_CPUID,		cpuid),
+	FEAT_OPP(HEADER_CPU_MAX_FREQ,	cpu_max_freq),
 	FEAT_OPP(HEADER_TOTAL_MEM,	total_mem),
 	FEAT_OPP(HEADER_EVENT_DESC,	event_desc),
 	FEAT_OPP(HEADER_CMDLINE,	cmdline),
tools/perf/util/header.h +2 -0
index d4d5796..a646025 100644
@@ -22,6 +22,7 @@ enum {
 	HEADER_NRCPUS,
 	HEADER_CPUDESC,
 	HEADER_CPUID,
+	HEADER_CPU_MAX_FREQ,
 	HEADER_TOTAL_MEM,
 	HEADER_CMDLINE,
 	HEADER_EVENT_DESC,
@@ -75,6 +76,7 @@ struct perf_session_env {
 	int			nr_cpus_avail;
 	char			*cpu_desc;
 	char			*cpuid;
+	int			cpu_max_freq;
 	unsigned long long	total_mem;
 
 	int			nr_cmdline;
Signature · 1 line
1.8.3.1
#2 lore raw

Re: [PATCH V2 1/6] perf,tools: save cpu max freq in perf header

On Fri, Jul 24, 2015 at 09:48:27AM -0400, [email protected] wrote:

SNIP

Quoted 41 lines
tools/perf/util/header.c +24 -0
 			      struct perf_header *h __maybe_unused,
 		       struct perf_evlist *evlist __maybe_unused)
@@ -1158,6 +1168,11 @@ static void print_cpuid(struct perf_header *ph, int fd __maybe_unused, FILE *fp)
 	fprintf(fp, "# cpuid : %s\n", ph->env.cpuid);
 }
 
+static void print_cpu_max_freq(struct perf_header *ph, int fd __maybe_unused, FILE *fp)
+{
+	fprintf(fp, "# CPU max frequency : %u MHz\n", ph->env.cpu_max_freq);
+}
+
 static void print_branch_stack(struct perf_header *ph __maybe_unused,
 			       int fd __maybe_unused, FILE *fp)
 {
@@ -1471,6 +1486,25 @@ static int process_cpuid(struct perf_file_section *section __maybe_unused,
 	return ph->env.cpuid ? 0 : -ENOMEM;
 }
 
+static int process_cpu_max_freq(struct perf_file_section *section __maybe_unused,
+				struct perf_header *ph, int fd,
+				void *data __maybe_unused)
+{
+	ssize_t ret;
+	u32 nr;
+
+	ret = readn(fd, &nr, sizeof(nr));
+	if (ret != sizeof(nr))
+		return -1;
+
+	if (ph->needs_swap)
+		nr = bswap_32(nr);
+
+	ph->env.cpu_max_freq = nr;
+
+	return 0;
+}
+
 static int process_total_mem(struct perf_file_section *section __maybe_unused,
 			     struct perf_header *ph, int fd,
 			     void *data __maybe_unused)

tools/perf/util/header.c static const struct feature_ops feat_ops[HEADER_LAST_FEATURE] = { · line 1922

Quoted 4 lines
tools/perf/util/header.c +1 -0
@@ -1885,6 +1919,7 @@ static const struct feature_ops feat_ops[HEADER_LAST_FEATURE] = {
 	FEAT_OPP(HEADER_NRCPUS,		nrcpus),
 	FEAT_OPP(HEADER_CPUDESC,	cpudesc),
 	FEAT_OPP(HEADER_CPUID,		cpuid),
+	FEAT_OPP(HEADER_CPU_MAX_FREQ,	cpu_max_freq),

I wonder we should start generic FEAT for CPU related data stored in some expandable way tag-value or such

jirka

#3 lore raw

Re: [PATCH V2 1/6] perf,tools: save cpu max freq in perf header

In reply to [email protected]

On Fri, Jul 24, 2015 at 09:48:27AM -0400, [email protected] wrote:

SNIP

Quoted 17 lines
tools/perf/util/cpumap.h +1 -0
index 0af9cec..6784677 100644
@@ -58,6 +58,7 @@ int max_node_num;
 int *cpunode_map;
 
 int cpu__setup_cpunode_map(void);
+unsigned int get_cpu_max_freq(void);
 
 static inline int cpu__max_node(void)
 {
tools/perf/util/header.c +0 -0
index 03ace57..287a488 100644

tools/perf/util/header.c write_it: · line 870

Quoted 9 lines
tools/perf/util/header.c +6 -0
@@ -862,6 +862,16 @@ write_it:
 	return do_write_string(fd, buffer);
 }
 
+static int write_cpu_max_freq(int fd, struct perf_header *h __maybe_unused,
+			      struct perf_evlist *evlist __maybe_unused)
+{
+	u32 freq;
+
+	freq = get_cpu_max_freq() / 1000;

why dont you store it as u64 ?

jirka

#4 lore raw

PATCH v2 2/6 perf,tools: read cpu max freq and msr type from header

In reply to [email protected]

From: Kan Liang <[email protected]>

In perf report, read cpu max freq and msr type from header.

---
 tools/perf/builtin-report.c |  3 +++
 tools/perf/util/cpumap.h    |  1 +
 tools/perf/util/header.c    | 21 +++++++++++++++++++++
 tools/perf/util/header.h    |  2 ++
 tools/perf/util/session.h   |  1 +
 5 files changed, 28 insertions(+)
tools/perf/builtin-report.c +3 -0
index 95a4771..2a32e9a 100644
@@ -818,6 +818,9 @@ repeat:
 		symbol_conf.cumulate_callchain = false;
 	}
 
+	cpu_max_freq = session->header.env.cpu_max_freq;
+	msr_pmu_type = perf_header_find_pmu_type(session, "msr");
+
 	if (setup_sorting() < 0) {
 		if (sort_order)
 			parse_options_usage(report_usage, options, "s", 1);
tools/perf/util/cpumap.h +1 -0
index 6784677..70ac686 100644
@@ -56,6 +56,7 @@ static inline bool cpu_map__empty(const struct cpu_map *map)
 int max_cpu_num;
 int max_node_num;
 int *cpunode_map;
+unsigned int cpu_max_freq;
 
 int cpu__setup_cpunode_map(void);
 unsigned int get_cpu_max_freq(void);
tools/perf/util/header.c +21 -0
index 287a488..e83681b 100644
@@ -2847,3 +2847,24 @@ int perf_event__process_build_id(struct perf_tool *tool __maybe_unused,
 				 session);
 	return 0;
 }
+
+int perf_header_find_pmu_type(struct perf_session *session, const char *name)
+{
+	u32 pmu_num = session->header.env.nr_pmu_mappings;
+	u32 type;
+	char *str, *tmp;
+
+	str = session->header.env.pmu_mappings;
+	while (pmu_num--) {
+		type = strtoul(str, &tmp, 0);
+		if (*tmp != ':') {
+			pr_err("unable to read pmu mappings from head\n");
+			return -1;
+		}
+		str = tmp + 1;
+		if (!strcmp(str, name))
+			return type;
+		str += strlen(str) + 1;
+	}
+	return -1;
+}
tools/perf/util/header.h +2 -0
index a646025..233644b 100644
@@ -150,6 +150,8 @@ int perf_event__process_build_id(struct perf_tool *tool,
 				 struct perf_session *session);
 bool is_perf_magic(u64 magic);
 
+int perf_header_find_pmu_type(struct perf_session *s, const char *name);
+
 #define NAME_ALIGN 64
 
 int write_padded(int fd, const void *bf, size_t count, size_t count_aligned);
tools/perf/util/session.h +1 -0
index b44afc7..a339338 100644
@@ -17,6 +17,7 @@ struct thread;
 
 struct auxtrace;
 struct itrace_synth_opts;
+unsigned int msr_pmu_type;
 
 struct perf_session {
 	struct perf_header	header;
Signature · 1 line
1.8.3.1
#5 lore raw

PATCH v2 3/6 perf,tools: Dump per-sample freq/CPU%/CORE_BUSY% in report -D

In reply to [email protected]

From: Kan Liang <[email protected]>

The group read results from cycles/ref-cycles/TSC/ASTATE/MSTATE event can be used to calculate the frequency, CPU Utilization and percent performance during each sampling period.
This patch shows them in report -D.

Here is an example:

$ perf record -e
'{cycles,ref-cycles,msr/tsc/,msr/mperf/,msr/aperf/}:S' ~/tchain_edit

Here is one sample from perf report -D

1972044565107 0x3498 [0x88]: PERF_RECORD_SAMPLE(IP, 0x2): 10608/10608:
0x4005fd period: 564686 addr: 0
... sample_read:
.... group nr 5
..... id 0000000000000012, value 0000000002143901
..... id 0000000000000052, value 0000000002143896
..... id 0000000000000094, value 00000000021e443d
..... id 00000000000000d4, value 00000000021db984
..... id 0000000000000114, value 00000000021db964
..... Freq 2301 MHz
..... CPU% 98%
..... CORE_BUSY% 99%

---
 tools/perf/util/session.c | 33 ++++++++++++++++++++++++++++-----
 tools/perf/util/session.h | 38 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 66 insertions(+), 5 deletions(-)
tools/perf/util/session.c +28 -5
index ed9dc25..7f628d9 100644
@@ -851,8 +851,14 @@ static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
 		printf("%" PRIu64 " ", sample->time);
 }
 
-static void sample_read__printf(struct perf_sample *sample, u64 read_format)
+static void sample_read__printf(struct perf_evlist *evlist,
+				struct perf_sample *sample,
+				u64 read_format)
 {
+	struct perf_evsel *evsel;
+	struct perf_sample_id *sid;
+	u64 data[FREQ_PERF_MAX] = { 0 };
+
 	printf("... sample_read:\n");
 
 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
@@ -875,10 +881,26 @@ static void sample_read__printf(struct perf_sample *sample, u64 read_format)
 			printf("..... id %016" PRIx64
 			       ", value %016" PRIx64 "\n",
 			       value->id, value->value);
+
+			sid = perf_evlist__id2sid(evlist, value->id);
+			evsel = sid->evsel;
+			if (evsel != NULL)
+				SET_FREQ_PERF_VALUE(evsel, data,
+						    value->value);
 		}
 	} else
 		printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
 			sample->read.one.id, sample->read.one.value);
+
+	if (HAS_FREQ(data))
+		printf("..... Freq %lu MHz\n",
+		       (data[FREQ_PERF_CYCLES] * cpu_max_freq) / data[FREQ_PERF_REF_CYCLES]);
+	if (HAS_CPU_U(data))
+		printf("..... CPU%% %lu%%\n",
+		       (100 * data[FREQ_PERF_REF_CYCLES]) / data[FREQ_PERF_TSC]);
+	if (HAS_CORE_BUSY(data))
+		printf("..... CORE_BUSY%% %lu%%\n",
+		       (100 * data[FREQ_PERF_APERF]) / data[FREQ_PERF_MPERF]);
 }
 
 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
@@ -899,8 +921,8 @@ static void dump_event(struct perf_evlist *evlist, union perf_event *event,
 	       event->header.size, perf_event__name(event->header.type));
 }
 
-static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
-			struct perf_sample *sample)
+static void dump_sample(struct perf_evlist *evlist, struct perf_evsel *evsel,
+			union perf_event *event, struct perf_sample *sample)
 {
 	u64 sample_type;
 
@@ -938,7 +960,7 @@ static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
 		printf("... transaction: %" PRIx64 "\n", sample->transaction);
 
 	if (sample_type & PERF_SAMPLE_READ)
-		sample_read__printf(sample, evsel->attr.read_format);
+		sample_read__printf(evlist, sample, evsel->attr.read_format);
 }
 
 static struct machine *machines__find_for_cpumode(struct machines *machines,
@@ -1053,11 +1075,12 @@ static int machines__deliver_event(struct machines *machines,
 
 	switch (event->header.type) {
 	case PERF_RECORD_SAMPLE:
-		dump_sample(evsel, event, sample);
 		if (evsel == NULL) {
 			++evlist->stats.nr_unknown_id;
 			return 0;
 		}
+		dump_sample(evlist, evsel, event, sample);
+
 		if (machine == NULL) {
 			++evlist->stats.nr_unprocessable_samples;
 			return 0;
tools/perf/util/session.h +38 -0
index a339338..db218ba 100644
@@ -43,6 +43,44 @@ struct perf_session {
 #define PRINT_IP_OPT_ONELINE	(1<<4)
 #define PRINT_IP_OPT_SRCLINE	(1<<5)
 
+#define PERF_MSR_TSC		0
+#define PERF_MSR_APERF		1
+#define PERF_MSR_MPERF		2
+
+enum perf_freq_perf_index {
+	FREQ_PERF_TSC		= 0,
+	FREQ_PERF_APERF		= 1,
+	FREQ_PERF_MPERF		= 2,
+	FREQ_PERF_CYCLES	= 3,
+	FREQ_PERF_REF_CYCLES	= 4,
+
+	FREQ_PERF_MAX
+};
+
+#define SET_FREQ_PERF_VALUE(event, array, value)			\
+{									\
+	if (event->attr.type == msr_pmu_type) {				\
+		if (event->attr.config == PERF_MSR_TSC)			\
+			array[FREQ_PERF_TSC] = value;			\
+		if (event->attr.config == PERF_MSR_APERF)		\
+			array[FREQ_PERF_APERF] = value;			\
+		if (event->attr.config == PERF_MSR_MPERF)		\
+			array[FREQ_PERF_MPERF] = value;			\
+	}								\
+	if (event->attr.type == PERF_TYPE_HARDWARE) {			\
+		if (event->attr.config == PERF_COUNT_HW_CPU_CYCLES)	\
+			array[FREQ_PERF_CYCLES] = value;		\
+		if (event->attr.config == PERF_COUNT_HW_REF_CPU_CYCLES)	\
+			array[FREQ_PERF_REF_CYCLES] = value;		\
+	}								\
+}
+
+#define HAS_FREQ(array)							\
+	((array[FREQ_PERF_CYCLES] > 0) && (array[FREQ_PERF_REF_CYCLES] > 0))
+#define HAS_CPU_U(array)						\
+	((array[FREQ_PERF_TSC] > 0) && (array[FREQ_PERF_REF_CYCLES] > 0))
+#define HAS_CORE_BUSY(array)						\
+	((array[FREQ_PERF_APERF] > 0) && (array[FREQ_PERF_MPERF] > 0))
 struct perf_tool;
 
 struct perf_session *perf_session__new(struct perf_data_file *file,
Signature · 1 line
1.8.3.1
#6 lore raw

Re: [PATCH V2 3/6] perf,tools: Dump per-sample freq/CPU%/CORE_BUSY% in report -D

On Fri, Jul 24, 2015 at 09:48:29AM -0400, [email protected] wrote:

SNIP

Quoted 28 lines
tools/perf/util/session.c +11 -1
index ed9dc25..7f628d9 100644
@@ -851,8 +851,14 @@ static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
 		printf("%" PRIu64 " ", sample->time);
 }
 
-static void sample_read__printf(struct perf_sample *sample, u64 read_format)
+static void sample_read__printf(struct perf_evlist *evlist,
+				struct perf_sample *sample,
+				u64 read_format)
 {
+	struct perf_evsel *evsel;
+	struct perf_sample_id *sid;
+	u64 data[FREQ_PERF_MAX] = { 0 };
+
 	printf("... sample_read:\n");
 
 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
@@ -875,10 +881,26 @@ static void sample_read__printf(struct perf_sample *sample, u64 read_format)
 			printf("..... id %016" PRIx64
 			       ", value %016" PRIx64 "\n",
 			       value->id, value->value);
+
+			sid = perf_evlist__id2sid(evlist, value->id);
+			evsel = sid->evsel;
+			if (evsel != NULL)

tools/perf/util/session.c static void sample_read__printf(struct perf_sample *sample, u64 read_format) · line 897

Quoted 10 lines
tools/perf/util/session.c +6 -0
+				SET_FREQ_PERF_VALUE(evsel, data,
+						    value->value);
 		}
 	} else
 		printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
 			sample->read.one.id, sample->read.one.value);
+
+	if (HAS_FREQ(data))
+		printf("..... Freq %lu MHz\n",
+		       (data[FREQ_PERF_CYCLES] * cpu_max_freq) / data[FREQ_PERF_REF_CYCLES]);

I dont like having those numbers (cpu_max_freq, msr_pmu_type) pushed into global variables that would be used

I think you should somehow push those session's values through the whole dump_sample code

maybe make them both part of session::perf_session_env and push this pointer along the dump_sample call stack

jirka

#7 lore raw

PATCH v2 4/6 perf,tools: save misc sample read value in struct perf_sample

In reply to [email protected]

From: Kan Liang <[email protected]>

Save group read results from cycles/ref-cycles/TSC/ASTATE/MSTATE in struct perf_sample. The following sample process function can easily use them to caculate freq/CPU%/CORE_BUSY% and add them in hists.

---
 tools/perf/util/event.h   | 11 +++++++++++
 tools/perf/util/session.c | 16 ++++++++++++++++
 tools/perf/util/session.h | 10 ----------
 3 files changed, 27 insertions(+), 10 deletions(-)
tools/perf/util/event.h +11 -0
index c53f363..f7aabe3 100644
@@ -176,6 +176,16 @@ enum {
 	PERF_IP_FLAG_TRACE_BEGIN	|\
 	PERF_IP_FLAG_TRACE_END)
 
+enum perf_freq_perf_index {
+	FREQ_PERF_TSC		= 0,
+	FREQ_PERF_APERF		= 1,
+	FREQ_PERF_MPERF		= 2,
+	FREQ_PERF_CYCLES	= 3,
+	FREQ_PERF_REF_CYCLES	= 4,
+
+	FREQ_PERF_MAX
+};
+
 struct perf_sample {
 	u64 ip;
 	u32 pid, tid;
@@ -191,6 +201,7 @@ struct perf_sample {
 	u64 data_src;
 	u32 flags;
 	u16 insn_len;
+	u64 freq_perf_data[FREQ_PERF_MAX];
 	void *raw_data;
 	struct ip_callchain *callchain;
 	struct branch_stack *branch_stack;
tools/perf/util/session.c +16 -0
index 7f628d9..7da36b1 100644
@@ -998,6 +998,8 @@ static int deliver_sample_value(struct perf_evlist *evlist,
 				struct machine *machine)
 {
 	struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
+	struct perf_evsel *evsel;
+	u64 nr = 0;
 
 	if (sid) {
 		sample->id     = v->id;
@@ -1010,6 +1012,20 @@ static int deliver_sample_value(struct perf_evlist *evlist,
 		return 0;
 	}
 
+	if (perf_evsel__is_group_leader(sid->evsel)) {
+		evsel = sid->evsel;
+		SET_FREQ_PERF_VALUE(evsel, sample->freq_perf_data,
+				    sample->read.group.values[nr].value);
+		evlist__for_each_continue(evlist, evsel) {
+			if ((evsel->leader != sid->evsel) ||
+			    (++nr >= sample->read.group.nr))
+				break;
+
+			SET_FREQ_PERF_VALUE(evsel, sample->freq_perf_data,
+					    sample->read.group.values[nr].value);
+		}
+	}
+
 	return tool->sample(tool, event, sample, sid->evsel, machine);
 }
 
tools/perf/util/session.h +0 -10
index db218ba..ee23872 100644
@@ -47,16 +47,6 @@ struct perf_session {
 #define PERF_MSR_APERF		1
 #define PERF_MSR_MPERF		2
 
-enum perf_freq_perf_index {
-	FREQ_PERF_TSC		= 0,
-	FREQ_PERF_APERF		= 1,
-	FREQ_PERF_MPERF		= 2,
-	FREQ_PERF_CYCLES	= 3,
-	FREQ_PERF_REF_CYCLES	= 4,
-
-	FREQ_PERF_MAX
-};
-
 #define SET_FREQ_PERF_VALUE(event, array, value)			\
 {									\
 	if (event->attr.type == msr_pmu_type) {				\
Signature · 1 line
1.8.3.1
#8 lore raw

Re: [PATCH V2 4/6] perf,tools: save misc sample read value in struct perf_sample

On Fri, Jul 24, 2015 at 09:48:30AM -0400, [email protected] wrote:

SNIP

Quoted 9 lines
tools/perf/util/session.c +2 -0
+	u64 nr = 0;
 
 	if (sid) {
 		sample->id     = v->id;
@@ -1010,6 +1012,20 @@ static int deliver_sample_value(struct perf_evlist *evlist,
 		return 0;
 	}
 
+	if (perf_evsel__is_group_leader(sid->evsel)) {

tools/perf/util/session.c static int deliver_sample_value(struct perf_evlist *evlist, · line 1025

Quoted 10 lines
tools/perf/util/session.c +10 -0
+		evsel = sid->evsel;
+		SET_FREQ_PERF_VALUE(evsel, sample->freq_perf_data,
+				    sample->read.group.values[nr].value);
+		evlist__for_each_continue(evlist, evsel) {
+			if ((evsel->leader != sid->evsel) ||
+			    (++nr >= sample->read.group.nr))
+				break;
+
+			SET_FREQ_PERF_VALUE(evsel, sample->freq_perf_data,
+					    sample->read.group.values[nr].value);

I think this should be in upper layer.. why not do this also within iter_prepare_normal_entry as for the rest of the calculations I suggested in my other email

jirka

author
#9 lore raw

RE: [PATCH V2 4/6] perf,tools: save misc sample read value in struct perf_sample

Quoted 26 lines

-----Original Message-----
From: Jiri Olsa [mailto:[email protected]]
Sent: Sunday, July 26, 2015 12:32 PM
To: Liang, Kan
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Hunter, Adrian; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH V2 4/6] perf,tools: save misc sample read value in struct perf_sample

On Fri, Jul 24, 2015 at 09:48:30AM -0400, [email protected] wrote:

SNIP

tools/perf/util/session.c static int deliver_sample_value(struct

Quoted 5 lines
tools/perf/util/session.c +1 -0
+	u64 nr = 0;
 
 	if (sid) {
 		sample->id     = v->id;
@@ -1010,6 +1012,20 @@ static int deliver_sample_value(struct

perf_evlist *evlist,

Quoted 3 lines
tools/perf/util/session.c +0 -0
 		return 0;
 	}
 
Quoted 15 lines

tools/perf/util/session.c static int deliver_sample_value(struct perf_evlist *evlist,

Quoted 13 lines
tools/perf/util/session.c +11 -0
+	if (perf_evsel__is_group_leader(sid->evsel)) {
+		evsel = sid->evsel;
+		SET_FREQ_PERF_VALUE(evsel, sample->freq_perf_data,
+				    sample->read.group.values[nr].value);
+		evlist__for_each_continue(evlist, evsel) {
+			if ((evsel->leader != sid->evsel) ||
+			    (++nr >= sample->read.group.nr))
+				break;
+
+			SET_FREQ_PERF_VALUE(evsel, sample-
 freq_perf_data,
+					    sample-
 read.group.values[nr].value);

I think this should be in upper layer..

OK, I'll move it to deliver_sample_group.

Quoted 3 lines

why not do this also within
iter_prepare_normal_entry as for the rest of the calculations I suggested in my other email

Because we cannot get evlist in iter_add/prepare function. We cannot go through all group members' value.
It's too complex (need to change many interfaces) to pass the evlist to iter_add/prepare function

Thanks,
Kan

#10 lore raw

Re: [PATCH V2 4/6] perf,tools: save misc sample read value in struct perf_sample

On Mon, Jul 27, 2015 at 10:24:49PM +0000, Liang, Kan wrote:

SNIP

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

freq_perf_data,

(quoted diff) +1 -0
+					    sample-

read.group.values[nr].value);

I think this should be in upper layer..

OK, I'll move it to deliver_sample_group.

Quoted 8 lines
Quoted 3 lines

why not do this also within
iter_prepare_normal_entry as for the rest of the calculations I suggested in my other email

Because we cannot get evlist in iter_add/prepare function. We cannot go through all group members' value.
It's too complex (need to change many interfaces) to pass the evlist to iter_add/prepare function

hum, I remember Namhyung added iter->add_entry_cb, but seems it's called after the entry is added.. how about adding pre-addition callback?

it can reach the report structure with perf_evlist

jirka

#11 lore raw

PATCH v2 5/6 perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat

In reply to [email protected]

From: Kan Liang <[email protected]>

Introduce a new hist_iter ops (hist_iter_freq_perf) to caculate the freq/CPU%/CORE_BUSY% freq when processing samples, and save them in hist_entry.

---
 tools/perf/builtin-annotate.c |  2 +-
 tools/perf/builtin-diff.c     |  2 +-
 tools/perf/tests/hists_link.c |  4 ++--
 tools/perf/util/hist.c        | 51 ++++++++++++++++++++++++++++++++++++++-----
 tools/perf/util/hist.h        |  2 ++
 tools/perf/util/sort.h        |  3 +++
 tools/perf/util/symbol.h      |  6 +++++
 7 files changed, 60 insertions(+), 10 deletions(-)
tools/perf/builtin-annotate.c +1 -1
index 2c1bec3..06e2f87 100644
@@ -71,7 +71,7 @@ static int perf_evsel__add_sample(struct perf_evsel *evsel,
 		return 0;
 	}
 
-	he = __hists__add_entry(hists, al, NULL, NULL, NULL, 1, 1, 0, true);
+	he = __hists__add_entry(hists, al, NULL, NULL, NULL, 1, 1, 0, NULL, true);
 	if (he == NULL)
 		return -ENOMEM;
 
tools/perf/builtin-diff.c +1 -1
index daaa7dc..2fffcc4 100644
@@ -315,7 +315,7 @@ static int hists__add_entry(struct hists *hists,
 			    u64 weight, u64 transaction)
 {
 	if (__hists__add_entry(hists, al, NULL, NULL, NULL, period, weight,
-			       transaction, true) != NULL)
+			       transaction, NULL, true) != NULL)
 		return 0;
 	return -ENOMEM;
 }
tools/perf/tests/hists_link.c +2 -2
index 8c102b0..5d9f9e3 100644
@@ -90,7 +90,7 @@ static int add_hist_entries(struct perf_evlist *evlist, struct machine *machine)
 				goto out;
 
 			he = __hists__add_entry(hists, &al, NULL,
-						NULL, NULL, 1, 1, 0, true);
+						NULL, NULL, 1, 1, 0, NULL, true);
 			if (he == NULL) {
 				addr_location__put(&al);
 				goto out;
@@ -116,7 +116,7 @@ static int add_hist_entries(struct perf_evlist *evlist, struct machine *machine)
 				goto out;
 
 			he = __hists__add_entry(hists, &al, NULL,
-						NULL, NULL, 1, 1, 0, true);
+						NULL, NULL, 1, 1, 0, NULL, true);
 			if (he == NULL) {
 				addr_location__put(&al);
 				goto out;
tools/perf/util/hist.c +45 -6

105 lines not on this page · raw message

tools/perf/util/hist.h +2 -0
index 5ed8d9c..70bd557 100644
@@ -102,6 +102,7 @@ extern const struct hist_iter_ops hist_iter_normal;
 extern const struct hist_iter_ops hist_iter_branch;
 extern const struct hist_iter_ops hist_iter_mem;
 extern const struct hist_iter_ops hist_iter_cumulative;
+extern const struct hist_iter_ops hist_iter_freq_perf;
 
 struct hist_entry *__hists__add_entry(struct hists *hists,
 				      struct addr_location *al,
@@ -109,6 +110,7 @@ struct hist_entry *__hists__add_entry(struct hists *hists,
 				      struct branch_info *bi,
 				      struct mem_info *mi, u64 period,
 				      u64 weight, u64 transaction,
+				      struct freq_perf_info *freq_perf,
 				      bool sample_self);
 int hist_entry_iter__add(struct hist_entry_iter *iter, struct addr_location *al,
 			 int max_stack_depth, void *arg);
tools/perf/util/sort.h +3 -0
index e97cd47..90422ed 100644
@@ -54,6 +54,9 @@ struct he_stat {
 	u64			period_guest_us;
 	u64			weight;
 	u32			nr_events;
+	u64			freq;
+	u64			cpu_u;
+	u64			core_busy;
 };
 
 struct hist_entry_diff {
tools/perf/util/symbol.h +6 -0
index b98ce51..fa0ccf3 100644
@@ -180,6 +180,12 @@ struct mem_info {
 	union perf_mem_data_src data_src;
 };
 
+struct freq_perf_info {
+	u64	freq;
+	u64	cpu_u;
+	u64	core_busy;
+};
+
 struct addr_location {
 	struct machine *machine;
 	struct thread *thread;
Signature · 1 line
1.8.3.1
#12 lore raw

Re: [PATCH V2 5/6] perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat

On Fri, Jul 24, 2015 at 09:48:31AM -0400, [email protected] wrote:

SNIP

Quoted 11 lines
tools/perf/util/hist.c +4 -0
 		},
 		.parent = sym_parent,
 		.filtered = symbol__parent_filter(sym_parent) | al->filtered,
@@ -481,6 +486,32 @@ iter_add_next_nop_entry(struct hist_entry_iter *iter __maybe_unused,
 }
 
 static int
+iter_add_single_freq_perf_entry(struct hist_entry_iter *iter, struct addr_location *al)
+{
+	struct perf_evsel *evsel = iter->evsel;
+	struct perf_sample *sample = iter->sample;

tools/perf/util/hist.c iter_add_next_nop_entry(struct hist_entry_iter *iter __maybe_unused, · line 502

Quoted 10 lines
tools/perf/util/hist.c +10 -0
+	struct hist_entry *he;
+	struct freq_perf_info info = {0};
+	u64 *data = sample->freq_perf_data;
+
+	if (data[FREQ_PERF_REF_CYCLES] > 0)
+		info.freq = (data[FREQ_PERF_CYCLES] * cpu_max_freq) / data[FREQ_PERF_REF_CYCLES];
+	if (data[FREQ_PERF_TSC] > 0)
+		info.cpu_u = (100 * data[FREQ_PERF_REF_CYCLES]) / data[FREQ_PERF_TSC];
+	if (data[FREQ_PERF_MPERF] > 0)
+		info.core_busy = (100 * data[FREQ_PERF_APERF]) / data[FREQ_PERF_MPERF];

seems to me the new iterator is too big gun for this, why not initialize 'struct freq_perf_info' in iter_prepare_normal_entry ?

jirka

author
#13 lore raw

RE: [PATCH V2 5/6] perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat

Quoted 21 lines
Quoted 3 lines

-481,6 +486,32 @@ iter_add_next_nop_entry(struct hist_entry_iter

*iter

tools/perf/util/hist.c iter_add_next_nop_entry(struct hist_entry_iter *iter __maybe_unused,

Quoted 13 lines

__maybe_unused, }

static int

tools/perf/util/hist.c +10 -0
+iter_add_single_freq_perf_entry(struct hist_entry_iter *iter, struct
+addr_location *al) {
+	struct perf_evsel *evsel = iter->evsel;
+	struct perf_sample *sample = iter->sample;
+	struct hist_entry *he;
+	struct freq_perf_info info = {0};
+	u64 *data = sample->freq_perf_data;
+
+	if (data[FREQ_PERF_REF_CYCLES] > 0)
+		info.freq = (data[FREQ_PERF_CYCLES] * cpu_max_freq) /

data[FREQ_PERF_REF_CYCLES];

Review

Quoted 2 lines
(quoted diff) +2 -0
+	if (data[FREQ_PERF_TSC] > 0)
+		info.cpu_u = (100 * data[FREQ_PERF_REF_CYCLES]) /

data[FREQ_PERF_TSC];

Quoted 7 lines

Review

Quoted 3 lines
(quoted diff) +3 -0
+	if (data[FREQ_PERF_MPERF] > 0)
+		info.core_busy = (100 * data[FREQ_PERF_APERF]) /
+data[FREQ_PERF_MPERF];

seems to me the new iterator is too big gun for this, why not initialize 'struct freq_perf_info' in iter_prepare_normal_entry ?

Yes, we can initialize it in normal process.
But we only need the freq_perf_info when "--show-freq-perf" is set. For most of the cases, it wastes time to calculate the unused information for each sample. That could impact the performance for processing.

Thanks,
Kan

#14 lore raw

Re: [PATCH V2 5/6] perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat

Quoted 29 lines · Kan
Quoted 21 lines
Quoted 3 lines

-481,6 +486,32 @@ iter_add_next_nop_entry(struct hist_entry_iter

*iter

tools/perf/util/hist.c iter_add_next_nop_entry(struct hist_entry_iter *iter __maybe_unused,

Quoted 13 lines

__maybe_unused, }

static int

tools/perf/util/hist.c +10 -0
+iter_add_single_freq_perf_entry(struct hist_entry_iter *iter, struct
+addr_location *al) {
+	struct perf_evsel *evsel = iter->evsel;
+	struct perf_sample *sample = iter->sample;
+	struct hist_entry *he;
+	struct freq_perf_info info = {0};
+	u64 *data = sample->freq_perf_data;
+
+	if (data[FREQ_PERF_REF_CYCLES] > 0)
+		info.freq = (data[FREQ_PERF_CYCLES] * cpu_max_freq) /

data[FREQ_PERF_REF_CYCLES];

Review

Quoted 2 lines
(quoted diff) +2 -0
+	if (data[FREQ_PERF_TSC] > 0)
+		info.cpu_u = (100 * data[FREQ_PERF_REF_CYCLES]) /

data[FREQ_PERF_TSC];

Quoted 7 lines

Review

Quoted 3 lines
(quoted diff) +3 -0
+	if (data[FREQ_PERF_MPERF] > 0)
+		info.core_busy = (100 * data[FREQ_PERF_APERF]) /
+data[FREQ_PERF_MPERF];

seems to me the new iterator is too big gun for this, why not initialize 'struct freq_perf_info' in iter_prepare_normal_entry ?

Quoted 4 lines

Yes, we can initialize it in normal process.
But we only need the freq_perf_info when "--show-freq-perf" is set. For most of the cases, it wastes time to calculate the unused information for each sample. That could impact the performance for processing.

hum, it could fill in freq_perf_info only when --show-freq-perf is set, other time it'd pass NULL

jirka

author
#15 lore raw

RE: [PATCH V2 5/6] perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat

In reply to Jiri Olsa
Quoted 29 lines · On Fri, Jul 24, 2015 at 09:48:31AM -0400, [email protected] wrote:

SNIP

Quoted 5 lines
tools/perf/util/hist.c +0 -0
 		},
 		.parent = sym_parent,
 		.filtered = symbol__parent_filter(sym_parent) | al-

filtered, @@
-481,6 +486,32 @@ iter_add_next_nop_entry(struct hist_entry_iter

*iter

tools/perf/util/hist.c iter_add_next_nop_entry(struct hist_entry_iter *iter __maybe_unused,

Quoted 13 lines

__maybe_unused, }

static int

tools/perf/util/hist.c +10 -0
+iter_add_single_freq_perf_entry(struct hist_entry_iter *iter, struct
+addr_location *al) {
+	struct perf_evsel *evsel = iter->evsel;
+	struct perf_sample *sample = iter->sample;
+	struct hist_entry *he;
+	struct freq_perf_info info = {0};
+	u64 *data = sample->freq_perf_data;
+
+	if (data[FREQ_PERF_REF_CYCLES] > 0)
+		info.freq = (data[FREQ_PERF_CYCLES] * cpu_max_freq) /

data[FREQ_PERF_REF_CYCLES];

Review

Quoted 2 lines
(quoted diff) +2 -0
+	if (data[FREQ_PERF_TSC] > 0)
+		info.cpu_u = (100 * data[FREQ_PERF_REF_CYCLES]) /

data[FREQ_PERF_TSC];

Quoted 7 lines

Review

Quoted 3 lines
(quoted diff) +3 -0
+	if (data[FREQ_PERF_MPERF] > 0)
+		info.core_busy = (100 * data[FREQ_PERF_APERF]) /
+data[FREQ_PERF_MPERF];

seems to me the new iterator is too big gun for this, why not initialize 'struct freq_perf_info' in iter_prepare_normal_entry ?

How about initializing in iter_add_single_normal_entry?

We only use freq_perf_info in iter_add_single_normal_entry. If initializing in iter_prepare_normal_entry, we have to save the freq_perf_info in hist_entry_iter.

Thanks,
Kan

#16 lore raw

Re: [PATCH V2 5/6] perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat

Quoted 36 lines · Kan
Quoted 29 lines · On Fri, Jul 24, 2015 at 09:48:31AM -0400, [email protected] wrote:

SNIP

Quoted 5 lines
tools/perf/util/hist.c +0 -0
 		},
 		.parent = sym_parent,
 		.filtered = symbol__parent_filter(sym_parent) | al-

filtered, @@
-481,6 +486,32 @@ iter_add_next_nop_entry(struct hist_entry_iter

*iter

tools/perf/util/hist.c iter_add_next_nop_entry(struct hist_entry_iter *iter __maybe_unused,

Quoted 13 lines

__maybe_unused, }

static int

tools/perf/util/hist.c +10 -0
+iter_add_single_freq_perf_entry(struct hist_entry_iter *iter, struct
+addr_location *al) {
+	struct perf_evsel *evsel = iter->evsel;
+	struct perf_sample *sample = iter->sample;
+	struct hist_entry *he;
+	struct freq_perf_info info = {0};
+	u64 *data = sample->freq_perf_data;
+
+	if (data[FREQ_PERF_REF_CYCLES] > 0)
+		info.freq = (data[FREQ_PERF_CYCLES] * cpu_max_freq) /

data[FREQ_PERF_REF_CYCLES];

Review

Quoted 2 lines
(quoted diff) +2 -0
+	if (data[FREQ_PERF_TSC] > 0)
+		info.cpu_u = (100 * data[FREQ_PERF_REF_CYCLES]) /

data[FREQ_PERF_TSC];

Quoted 7 lines

Review

Quoted 3 lines
(quoted diff) +3 -0
+	if (data[FREQ_PERF_MPERF] > 0)
+		info.core_busy = (100 * data[FREQ_PERF_APERF]) /
+data[FREQ_PERF_MPERF];

seems to me the new iterator is too big gun for this, why not initialize 'struct freq_perf_info' in iter_prepare_normal_entry ?

Quoted 5 lines

How about initializing in iter_add_single_normal_entry?

We only use freq_perf_info in iter_add_single_normal_entry. If initializing in iter_prepare_normal_entry, we have to save the freq_perf_info in hist_entry_iter.

sounds good

jirka

#17 lore raw

Re: [PATCH V2 5/6] perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat

In reply to [email protected]

On Fri, Jul 24, 2015 at 09:48:31AM -0400, [email protected] wrote:

SNIP

Quoted 14 lines
tools/perf/util/hist.h +1 -0
 				      struct addr_location *al,
@@ -109,6 +110,7 @@ struct hist_entry *__hists__add_entry(struct hists *hists,
 				      struct branch_info *bi,
 				      struct mem_info *mi, u64 period,
 				      u64 weight, u64 transaction,
+				      struct freq_perf_info *freq_perf,
 				      bool sample_self);
 int hist_entry_iter__add(struct hist_entry_iter *iter, struct addr_location *al,
 			 int max_stack_depth, void *arg);
tools/perf/util/sort.h +0 -0
index e97cd47..90422ed 100644

tools/perf/util/sort.h struct he_stat { · line 58

Quoted 5 lines
tools/perf/util/sort.h +2 -0
@@ -54,6 +54,9 @@ struct he_stat {
 	u64			period_guest_us;
 	u64			weight;
 	u32			nr_events;
+	u64			freq;
+	u64			cpu_u;

could the cpu_u have more descriptive name?

thanks,
jirka

Quoted 3 lines
tools/perf/util/sort.h +1 -0
+	u64			core_busy;
 };
 

SNIP

author
#18 lore raw

RE: [PATCH V2 5/6] perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat

Quoted 15 lines · On Fri, Jul 24, 2015 at 09:48:31AM -0400, [email protected] wrote:

SNIP

tools/perf/util/hist.h struct hist_entry *__hists__add_entry(struct

Quoted 2 lines
tools/perf/util/hist.h +0 -0
 				      struct addr_location *al,
@@ -109,6 +110,7 @@ struct hist_entry *__hists__add_entry(struct

hists *hists,

tools/perf/util/hist.h struct hist_entry *__hists__add_entry(struct hists *hists,

Quoted 6 lines
tools/perf/util/hist.h +1 -0
 				      struct branch_info *bi,
 				      struct mem_info *mi, u64 period,
 				      u64 weight, u64 transaction,
+				      struct freq_perf_info *freq_perf,
 				      bool sample_self);
 int hist_entry_iter__add(struct hist_entry_iter *iter, struct

addr_location *al,

Quoted 13 lines

tools/perf/util/sort.h struct he_stat { · line 58

Quoted 11 lines
tools/perf/util/hist.h +0 -0
 			 int max_stack_depth, void *arg);
tools/perf/util/sort.h index renamed from tools/perf/util/sort.h +0 -0

e97cd47..90422ed 100644

tools/perf/util/sort.h +2 -0
@@ -54,6 +54,9 @@ struct he_stat {
 	u64			period_guest_us;
 	u64			weight;
 	u32			nr_events;
+	u64			freq;
+	u64			cpu_u;

could the cpu_u have more descriptive name?

How about "cpu_util" or "cpu_utilization"?

Thanks,
Kan

#19 lore raw

Re: [PATCH V2 5/6] perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat

Quoted 29 lines · Kan
Quoted 15 lines · On Fri, Jul 24, 2015 at 09:48:31AM -0400, [email protected] wrote:

SNIP

tools/perf/util/hist.h struct hist_entry *__hists__add_entry(struct

Quoted 2 lines
tools/perf/util/hist.h +0 -0
 				      struct addr_location *al,
@@ -109,6 +110,7 @@ struct hist_entry *__hists__add_entry(struct

hists *hists,

tools/perf/util/hist.h struct hist_entry *__hists__add_entry(struct hists *hists,

Quoted 6 lines
tools/perf/util/hist.h +1 -0
 				      struct branch_info *bi,
 				      struct mem_info *mi, u64 period,
 				      u64 weight, u64 transaction,
+				      struct freq_perf_info *freq_perf,
 				      bool sample_self);
 int hist_entry_iter__add(struct hist_entry_iter *iter, struct

addr_location *al,

Quoted 13 lines

tools/perf/util/sort.h struct he_stat { · line 58

Quoted 11 lines
tools/perf/util/hist.h +0 -0
 			 int max_stack_depth, void *arg);
tools/perf/util/sort.h index renamed from tools/perf/util/sort.h +0 -0

e97cd47..90422ed 100644

tools/perf/util/sort.h +2 -0
@@ -54,6 +54,9 @@ struct he_stat {
 	u64			period_guest_us;
 	u64			weight;
 	u32			nr_events;
+	u64			freq;
+	u64			cpu_u;

could the cpu_u have more descriptive name?

Quoted 1 line

How about "cpu_util" or "cpu_utilization"?

cpu_util sounds good to me

thanks,
jirka

#20 lore raw

PATCH v2 6/6 perf,tools: Show freq/CPU%/CORE_BUSY% in perf report --stdio

In reply to [email protected]

From: Kan Liang <[email protected]>

Show frequency, CPU Utilization and percent performance for each symbol in perf report by --stdio --show-freq-perf

In sampling group, only group leader do sampling. So only need to print group leader's freq in --group.

Here is an example.

$ perf report --stdio --group --show-freq-perf

Overhead FREQ MHz CPU% CORE_BUSY%
Command Shared Object Symbol
........................................ ......... ..... ..........
........... ................ ......................

    99.54%  99.54%  99.53%  99.53%  99.53%       2301     96         99
tchain_edit  tchain_edit       [.] f3
     0.20%   0.20%   0.20%   0.20%   0.20%       2301     98         99
tchain_edit  tchain_edit       [.] f2
     0.05%   0.05%   0.05%   0.05%   0.05%       2300     98         99
tchain_edit  [kernel.vmlinux]  [k] read_tsc
---
 tools/perf/Documentation/perf-report.txt | 12 ++++++
 tools/perf/builtin-report.c              | 19 +++++++++
 tools/perf/perf.h                        |  1 +
 tools/perf/ui/hist.c                     | 71 +++++++++++++++++++++++++++++---
 tools/perf/util/hist.h                   |  3 ++
 tools/perf/util/session.c                |  2 +-
 tools/perf/util/sort.c                   |  3 ++
 tools/perf/util/symbol.h                 |  3 +-
 tools/perf/util/util.c                   |  2 +
 9 files changed, 109 insertions(+), 7 deletions(-)
tools/perf/Documentation/perf-report.txt +12 -0
index c33b69f..faa8825 100644
@@ -303,6 +303,18 @@ OPTIONS
 	special event -e cpu/mem-loads/ or -e cpu/mem-stores/. See
 	'perf mem' for simpler access.
 
+--show-freq-perf::
+	Show CPU frequency and performance result from sample read.
+	To generate the frequency and performance output, the perf.data file
+	must have been obtained by group read and using special events cycles,
+	ref-cycles, msr/tsc/, msr/aperf/ or msr/mperf/
+	Freq MHz: The frequency during the sample interval. Needs cycles and
+		  ref-cycles event.
+	CPU%: CPU utilization during the sample interval. Needs ref-cycles and
+	      msr/tsc/ events.
+	CORE_BUSY%: actual percent performance (APERF/MPERF%) during the
+		    sample interval. Needs msr/aperf/ and msr/mperf/ events.
+
 --percent-limit::
 	Do not show entries which have an overhead under that percent.
 	(Default: 0).
tools/perf/builtin-report.c +19 -0
index 2a32e9a..f52c63d 100644
@@ -164,6 +164,8 @@ static int process_sample_event(struct perf_tool *tool,
 		iter.ops = &hist_iter_mem;
 	else if (symbol_conf.cumulate_callchain)
 		iter.ops = &hist_iter_cumulative;
+	else if (symbol_conf.show_freq_perf)
+		iter.ops = &hist_iter_freq_perf;
 	else
 		iter.ops = &hist_iter_normal;
 
@@ -721,6 +723,8 @@ int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
 	OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
 		    "Enable kernel symbol demangling"),
 	OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
+	OPT_BOOLEAN(0, "show-freq-perf", &symbol_conf.show_freq_perf,
+		    "show CPU freqency and performance info"),
 	OPT_CALLBACK(0, "percent-limit", &report, "percent",
 		     "Don't show entries under that percent", parse_percent_limit),
 	OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
@@ -733,7 +737,9 @@ int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
 	struct perf_data_file file = {
 		.mode  = PERF_DATA_MODE_READ,
 	};
+	struct perf_evsel *pos;
 	int ret = hists__init();
+	bool freq_perf_info[FREQ_PERF_MAX] = {0};
 
 	if (ret < 0)
 		return ret;
@@ -821,6 +827,19 @@ repeat:
 	cpu_max_freq = session->header.env.cpu_max_freq;
 	msr_pmu_type = perf_header_find_pmu_type(session, "msr");
 
+	if (symbol_conf.show_freq_perf) {
+		perf_freq = perf_cpu_u = perf_core_busy = false;
+		evlist__for_each(session->evlist, pos) {
+			SET_FREQ_PERF_VALUE(pos, freq_perf_info, true);
+		}
+		if (HAS_FREQ(freq_perf_info))
+			perf_freq = true;
+		if (HAS_CPU_U(freq_perf_info))
+			perf_cpu_u = true;
+		if (HAS_CORE_BUSY(freq_perf_info))
+			perf_core_busy = true;
+	}
+
 	if (setup_sorting() < 0) {
 		if (sort_order)
 			parse_options_usage(report_usage, options, "s", 1);
tools/perf/perf.h +1 -0
index 937b16a..87daab8 100644
@@ -33,6 +33,7 @@ static inline unsigned long long rdclock(void)
 
 extern const char *input_name;
 extern bool perf_host, perf_guest;
+extern bool perf_freq, perf_cpu_u, perf_core_busy;
 extern const char perf_version_string[];
 
 void pthread__unblock_sigwinch(void);
tools/perf/ui/hist.c +66 -5

140 lines not on this page · raw message

tools/perf/util/hist.h +3 -0
index 70bd557..ec64234 100644
@@ -237,6 +237,9 @@ enum {
 	PERF_HPP__OVERHEAD_ACC,
 	PERF_HPP__SAMPLES,
 	PERF_HPP__PERIOD,
+	PERF_HPP__FREQ,
+	PERF_HPP__CPU_U,
+	PERF_HPP__CORE_BUSY,
 
 	PERF_HPP__MAX_INDEX
 };
tools/perf/util/session.c +1 -1
index 7da36b1..c5010f5 100644
@@ -1012,7 +1012,7 @@ static int deliver_sample_value(struct perf_evlist *evlist,
 		return 0;
 	}
 
-	if (perf_evsel__is_group_leader(sid->evsel)) {
+	if (symbol_conf.show_freq_perf && perf_evsel__is_group_leader(sid->evsel)) {
 		evsel = sid->evsel;
 		SET_FREQ_PERF_VALUE(evsel, sample->freq_perf_data,
 				    sample->read.group.values[nr].value);
tools/perf/util/sort.c +3 -0
index 4c65a14..690e173 100644
@@ -1225,6 +1225,9 @@ static struct hpp_dimension hpp_sort_dimensions[] = {
 	DIM(PERF_HPP__OVERHEAD_ACC, "overhead_children"),
 	DIM(PERF_HPP__SAMPLES, "sample"),
 	DIM(PERF_HPP__PERIOD, "period"),
+	DIM(PERF_HPP__FREQ, "freq"),
+	DIM(PERF_HPP__CPU_U, "cpu_u"),
+	DIM(PERF_HPP__CORE_BUSY, "core_busy"),
 };
 
 #undef DIM
tools/perf/util/symbol.h +2 -1
index fa0ccf3..7d70c89 100644
@@ -106,7 +106,8 @@ struct symbol_conf {
 			filter_relative,
 			show_hist_headers,
 			branch_callstack,
-			has_filter;
+			has_filter,
+			show_freq_perf;
 	const char	*vmlinux_name,
 			*kallsyms_name,
 			*source_prefix,
tools/perf/util/util.c +2 -0
index edc2d63..648b307 100644
@@ -34,6 +34,8 @@ bool test_attr__enabled;
 bool perf_host  = true;
 bool perf_guest = false;
 
+bool perf_freq, perf_cpu_u, perf_core_busy;
+
 char tracing_events_path[PATH_MAX + 1] = "/sys/kernel/debug/tracing/events";
 
 void event_attr_init(struct perf_event_attr *attr)
Signature · 1 line
1.8.3.1
#21 lore raw

Re: [PATCH V2 6/6] perf,tools: Show freq/CPU%/CORE_BUSY% in perf report --stdio

On Fri, Jul 24, 2015 at 09:48:32AM -0400, [email protected] wrote:

SNIP

Quoted 19 lines
tools/perf/util/symbol.h +2 -1
index fa0ccf3..7d70c89 100644
@@ -106,7 +106,8 @@ struct symbol_conf {
 			filter_relative,
 			show_hist_headers,
 			branch_callstack,
-			has_filter;
+			has_filter,
+			show_freq_perf;
 	const char	*vmlinux_name,
 			*kallsyms_name,
 			*source_prefix,
tools/perf/util/util.c +0 -0
index edc2d63..648b307 100644

tools/perf/util/util.c bool test_attr__enabled; · line 37

Quoted 4 lines
tools/perf/util/util.c +1 -0
@@ -34,6 +34,8 @@ bool test_attr__enabled;
 bool perf_host  = true;
 bool perf_guest = false;
 
+bool perf_freq, perf_cpu_u, perf_core_busy;

how about 'symbol_conf.freq_perf_type' having display info encoded in?

thanks,
jirka