Re: [PATCH v2] perf record: encode -k clockid frequency into Perf trace
Hi,
On 03.10.2018 16:39, Jiri Olsa wrote:
> On Wed, Oct 03, 2018 at 10:01:56AM +0300, Alexey Budankov wrote:
>> +if (rec->opts.use_clockid && rec->opts.clockid_res_ns)
>> +session->header.env.clockid_res_ns = rec->opts.clockid_res_ns;
>> +else
>> +perf_header__clear_feat(>header, HEADER_CLOCKID);
>> +
>
> could you please keep only bits setting in record__init_features
> and move the header.env.clockid_res_ns assignment out of it?
Ok. Accepted.
>
>> +static int get_clockid_res(clockid_t clk_id, size_t *res_ns)
>> +{
>> +struct timespec res;
>> +
>> +*res_ns = 0;
>> +if (!clock_getres(clk_id, ))
>> +*res_ns = res.tv_nsec + res.tv_sec * NSEC_PER_SEC;
>
> hum, if this one fails (which I guess is unlikely) we should tell
> and probably even quit
Makes sense. Please stay tuned for v3.
Thanks,
Alexey
Re: [PATCH v2] perf record: encode -k clockid frequency into Perf trace
Hi,
On 03.10.2018 16:39, Jiri Olsa wrote:
> On Wed, Oct 03, 2018 at 10:01:56AM +0300, Alexey Budankov wrote:
>> +if (rec->opts.use_clockid && rec->opts.clockid_res_ns)
>> +session->header.env.clockid_res_ns = rec->opts.clockid_res_ns;
>> +else
>> +perf_header__clear_feat(>header, HEADER_CLOCKID);
>> +
>
> could you please keep only bits setting in record__init_features
> and move the header.env.clockid_res_ns assignment out of it?
Ok. Accepted.
>
>> +static int get_clockid_res(clockid_t clk_id, size_t *res_ns)
>> +{
>> +struct timespec res;
>> +
>> +*res_ns = 0;
>> +if (!clock_getres(clk_id, ))
>> +*res_ns = res.tv_nsec + res.tv_sec * NSEC_PER_SEC;
>
> hum, if this one fails (which I guess is unlikely) we should tell
> and probably even quit
Makes sense. Please stay tuned for v3.
Thanks,
Alexey
Re: [PATCH v2] perf record: encode -k clockid frequency into Perf trace
On Wed, Oct 03, 2018 at 10:01:56AM +0300, Alexey Budankov wrote:
>
> Store -k clockid frequency into Perf trace to enable timestamps
> derived metrics conversion into wall clock time on reporting stage.
>
> Below is the example of perf report output:
>
> tools/perf/perf record -k raw -- ../../matrix/linux/matrix.gcc
> ...
> [ perf record: Captured and wrote 31.222 MB perf.data (818054 samples) ]
>
> tools/perf/perf report --header
> #
> ...
> # event : name = cycles:ppp, , size = 112, { sample_period, sample_freq } =
> 4000, sample_type = IP|TID|TIME|PERIOD, disabled = 1, inherit = 1, mmap = 1,
> comm = 1, freq = 1, enable_on_exec = 1, task = 1, precise_ip = 3,
> sample_id_all = 1, exclude_guest = 1, mmap2 = 1, comm_exec = 1, use_clockid =
> 1, clockid = 4
> ...
> # clockid frequency: 1000 MHz
> ...
> #
>
> Signed-off-by: Alexey Budankov
> ---
> Changes in v2:
> - renamed clockid_freq to clockid_res_ns and get_clockid_freq() to
> get_clockid_res()
> - avoided redundant define of NSEC_IN_SEC, reused linux/time64.h:NSEC_PER_SEC
> - moved MHz conversion into print_clockid() and shortened write_clockid()
> ---
> tools/perf/builtin-record.c | 21 +++--
> tools/perf/perf.h | 1 +
> tools/perf/util/env.h | 1 +
> tools/perf/util/header.c| 27 +++
> tools/perf/util/header.h| 1 +
> 5 files changed, 49 insertions(+), 2 deletions(-)
>
> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> index 0980dfe3396b..127dc7432394 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -592,6 +592,11 @@ static void record__init_features(struct record *rec)
> if (!rec->opts.full_auxtrace)
> perf_header__clear_feat(>header, HEADER_AUXTRACE);
>
> + if (rec->opts.use_clockid && rec->opts.clockid_res_ns)
> + session->header.env.clockid_res_ns = rec->opts.clockid_res_ns;
> + else
> + perf_header__clear_feat(>header, HEADER_CLOCKID);
> +
could you please keep only bits setting in record__init_features
and move the header.env.clockid_res_ns assignment out of it?
sry for not catching this in the first version
> perf_header__clear_feat(>header, HEADER_STAT);
> }
>
> @@ -1337,6 +1342,17 @@ static const struct clockid_map clockids[] = {
> CLOCKID_END,
> };
>
> +static int get_clockid_res(clockid_t clk_id, size_t *res_ns)
> +{
> + struct timespec res;
> +
> + *res_ns = 0;
> + if (!clock_getres(clk_id, ))
> + *res_ns = res.tv_nsec + res.tv_sec * NSEC_PER_SEC;
hum, if this one fails (which I guess is unlikely) we should tell
and probably even quit
thanks,
jirka
Re: [PATCH v2] perf record: encode -k clockid frequency into Perf trace
On Wed, Oct 03, 2018 at 10:01:56AM +0300, Alexey Budankov wrote:
>
> Store -k clockid frequency into Perf trace to enable timestamps
> derived metrics conversion into wall clock time on reporting stage.
>
> Below is the example of perf report output:
>
> tools/perf/perf record -k raw -- ../../matrix/linux/matrix.gcc
> ...
> [ perf record: Captured and wrote 31.222 MB perf.data (818054 samples) ]
>
> tools/perf/perf report --header
> #
> ...
> # event : name = cycles:ppp, , size = 112, { sample_period, sample_freq } =
> 4000, sample_type = IP|TID|TIME|PERIOD, disabled = 1, inherit = 1, mmap = 1,
> comm = 1, freq = 1, enable_on_exec = 1, task = 1, precise_ip = 3,
> sample_id_all = 1, exclude_guest = 1, mmap2 = 1, comm_exec = 1, use_clockid =
> 1, clockid = 4
> ...
> # clockid frequency: 1000 MHz
> ...
> #
>
> Signed-off-by: Alexey Budankov
> ---
> Changes in v2:
> - renamed clockid_freq to clockid_res_ns and get_clockid_freq() to
> get_clockid_res()
> - avoided redundant define of NSEC_IN_SEC, reused linux/time64.h:NSEC_PER_SEC
> - moved MHz conversion into print_clockid() and shortened write_clockid()
> ---
> tools/perf/builtin-record.c | 21 +++--
> tools/perf/perf.h | 1 +
> tools/perf/util/env.h | 1 +
> tools/perf/util/header.c| 27 +++
> tools/perf/util/header.h| 1 +
> 5 files changed, 49 insertions(+), 2 deletions(-)
>
> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> index 0980dfe3396b..127dc7432394 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -592,6 +592,11 @@ static void record__init_features(struct record *rec)
> if (!rec->opts.full_auxtrace)
> perf_header__clear_feat(>header, HEADER_AUXTRACE);
>
> + if (rec->opts.use_clockid && rec->opts.clockid_res_ns)
> + session->header.env.clockid_res_ns = rec->opts.clockid_res_ns;
> + else
> + perf_header__clear_feat(>header, HEADER_CLOCKID);
> +
could you please keep only bits setting in record__init_features
and move the header.env.clockid_res_ns assignment out of it?
sry for not catching this in the first version
> perf_header__clear_feat(>header, HEADER_STAT);
> }
>
> @@ -1337,6 +1342,17 @@ static const struct clockid_map clockids[] = {
> CLOCKID_END,
> };
>
> +static int get_clockid_res(clockid_t clk_id, size_t *res_ns)
> +{
> + struct timespec res;
> +
> + *res_ns = 0;
> + if (!clock_getres(clk_id, ))
> + *res_ns = res.tv_nsec + res.tv_sec * NSEC_PER_SEC;
hum, if this one fails (which I guess is unlikely) we should tell
and probably even quit
thanks,
jirka
[PATCH v2] perf record: encode -k clockid frequency into Perf trace
Store -k clockid frequency into Perf trace to enable timestamps
derived metrics conversion into wall clock time on reporting stage.
Below is the example of perf report output:
tools/perf/perf record -k raw -- ../../matrix/linux/matrix.gcc
...
[ perf record: Captured and wrote 31.222 MB perf.data (818054 samples) ]
tools/perf/perf report --header
#
...
# event : name = cycles:ppp, , size = 112, { sample_period, sample_freq } =
4000, sample_type = IP|TID|TIME|PERIOD, disabled = 1, inherit = 1, mmap = 1,
comm = 1, freq = 1, enable_on_exec = 1, task = 1, precise_ip = 3, sample_id_all
= 1, exclude_guest = 1, mmap2 = 1, comm_exec = 1, use_clockid = 1, clockid = 4
...
# clockid frequency: 1000 MHz
...
#
Signed-off-by: Alexey Budankov
---
Changes in v2:
- renamed clockid_freq to clockid_res_ns and get_clockid_freq() to
get_clockid_res()
- avoided redundant define of NSEC_IN_SEC, reused linux/time64.h:NSEC_PER_SEC
- moved MHz conversion into print_clockid() and shortened write_clockid()
---
tools/perf/builtin-record.c | 21 +++--
tools/perf/perf.h | 1 +
tools/perf/util/env.h | 1 +
tools/perf/util/header.c| 27 +++
tools/perf/util/header.h| 1 +
5 files changed, 49 insertions(+), 2 deletions(-)
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 0980dfe3396b..127dc7432394 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -592,6 +592,11 @@ static void record__init_features(struct record *rec)
if (!rec->opts.full_auxtrace)
perf_header__clear_feat(>header, HEADER_AUXTRACE);
+ if (rec->opts.use_clockid && rec->opts.clockid_res_ns)
+ session->header.env.clockid_res_ns = rec->opts.clockid_res_ns;
+ else
+ perf_header__clear_feat(>header, HEADER_CLOCKID);
+
perf_header__clear_feat(>header, HEADER_STAT);
}
@@ -1337,6 +1342,17 @@ static const struct clockid_map clockids[] = {
CLOCKID_END,
};
+static int get_clockid_res(clockid_t clk_id, size_t *res_ns)
+{
+ struct timespec res;
+
+ *res_ns = 0;
+ if (!clock_getres(clk_id, ))
+ *res_ns = res.tv_nsec + res.tv_sec * NSEC_PER_SEC;
+
+ return 0;
+}
+
static int parse_clockid(const struct option *opt, const char *str, int unset)
{
struct record_opts *opts = (struct record_opts *)opt->value;
@@ -1360,7 +1376,7 @@ static int parse_clockid(const struct option *opt, const
char *str, int unset)
/* if its a number, we're done */
if (sscanf(str, "%d", >clockid) == 1)
- return 0;
+ return get_clockid_res(opts->clockid, >clockid_res_ns);
/* allow a "CLOCK_" prefix to the name */
if (!strncasecmp(str, "CLOCK_", 6))
@@ -1369,7 +1385,8 @@ static int parse_clockid(const struct option *opt, const
char *str, int unset)
for (cm = clockids; cm->name; cm++) {
if (!strcasecmp(str, cm->name)) {
opts->clockid = cm->clockid;
- return 0;
+ return get_clockid_res(opts->clockid,
+ >clockid_res_ns);
}
}
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index 21bf7f5a3cf5..981db3c2ed60 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -81,6 +81,7 @@ struct record_opts {
unsigned initial_delay;
bool use_clockid;
clockid_tclockid;
+ size_t clockid_res_ns;
unsigned int proc_map_timeout;
};
diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
index 1f3ccc368530..b167a54d635f 100644
--- a/tools/perf/util/env.h
+++ b/tools/perf/util/env.h
@@ -63,6 +63,7 @@ struct perf_env {
struct numa_node*numa_nodes;
struct memory_node *memory_nodes;
unsigned long long memory_bsize;
+ size_t clockid_res_ns;
};
extern struct perf_env perf_env;
diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index 1ec1d9bc2d63..4ce5339158f7 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -1034,6 +1034,13 @@ static int write_auxtrace(struct feat_fd *ff,
return err;
}
+static int write_clockid(struct feat_fd *ff,
+struct perf_evlist *evlist __maybe_unused)
+{
+ return do_write(ff, >ph->env.clockid_res_ns,
+ sizeof(ff->ph->env.clockid_res_ns));
+}
+
static int cpu_cache_level__sort(const void *a, const void *b)
{
struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
@@ -1508,6 +1515,12 @@ static void print_cpu_topology(struct feat_fd *ff, FILE
*fp)
fprintf(fp, "# Core ID and Socket ID information is not
available\n");
}
+static void print_clockid(struct feat_fd *ff, FILE *fp)
+{
+ fprintf(fp, "# clockid frequency: %ld MHz\n",
+
[PATCH v2] perf record: encode -k clockid frequency into Perf trace
Store -k clockid frequency into Perf trace to enable timestamps
derived metrics conversion into wall clock time on reporting stage.
Below is the example of perf report output:
tools/perf/perf record -k raw -- ../../matrix/linux/matrix.gcc
...
[ perf record: Captured and wrote 31.222 MB perf.data (818054 samples) ]
tools/perf/perf report --header
#
...
# event : name = cycles:ppp, , size = 112, { sample_period, sample_freq } =
4000, sample_type = IP|TID|TIME|PERIOD, disabled = 1, inherit = 1, mmap = 1,
comm = 1, freq = 1, enable_on_exec = 1, task = 1, precise_ip = 3, sample_id_all
= 1, exclude_guest = 1, mmap2 = 1, comm_exec = 1, use_clockid = 1, clockid = 4
...
# clockid frequency: 1000 MHz
...
#
Signed-off-by: Alexey Budankov
---
Changes in v2:
- renamed clockid_freq to clockid_res_ns and get_clockid_freq() to
get_clockid_res()
- avoided redundant define of NSEC_IN_SEC, reused linux/time64.h:NSEC_PER_SEC
- moved MHz conversion into print_clockid() and shortened write_clockid()
---
tools/perf/builtin-record.c | 21 +++--
tools/perf/perf.h | 1 +
tools/perf/util/env.h | 1 +
tools/perf/util/header.c| 27 +++
tools/perf/util/header.h| 1 +
5 files changed, 49 insertions(+), 2 deletions(-)
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 0980dfe3396b..127dc7432394 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -592,6 +592,11 @@ static void record__init_features(struct record *rec)
if (!rec->opts.full_auxtrace)
perf_header__clear_feat(>header, HEADER_AUXTRACE);
+ if (rec->opts.use_clockid && rec->opts.clockid_res_ns)
+ session->header.env.clockid_res_ns = rec->opts.clockid_res_ns;
+ else
+ perf_header__clear_feat(>header, HEADER_CLOCKID);
+
perf_header__clear_feat(>header, HEADER_STAT);
}
@@ -1337,6 +1342,17 @@ static const struct clockid_map clockids[] = {
CLOCKID_END,
};
+static int get_clockid_res(clockid_t clk_id, size_t *res_ns)
+{
+ struct timespec res;
+
+ *res_ns = 0;
+ if (!clock_getres(clk_id, ))
+ *res_ns = res.tv_nsec + res.tv_sec * NSEC_PER_SEC;
+
+ return 0;
+}
+
static int parse_clockid(const struct option *opt, const char *str, int unset)
{
struct record_opts *opts = (struct record_opts *)opt->value;
@@ -1360,7 +1376,7 @@ static int parse_clockid(const struct option *opt, const
char *str, int unset)
/* if its a number, we're done */
if (sscanf(str, "%d", >clockid) == 1)
- return 0;
+ return get_clockid_res(opts->clockid, >clockid_res_ns);
/* allow a "CLOCK_" prefix to the name */
if (!strncasecmp(str, "CLOCK_", 6))
@@ -1369,7 +1385,8 @@ static int parse_clockid(const struct option *opt, const
char *str, int unset)
for (cm = clockids; cm->name; cm++) {
if (!strcasecmp(str, cm->name)) {
opts->clockid = cm->clockid;
- return 0;
+ return get_clockid_res(opts->clockid,
+ >clockid_res_ns);
}
}
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index 21bf7f5a3cf5..981db3c2ed60 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -81,6 +81,7 @@ struct record_opts {
unsigned initial_delay;
bool use_clockid;
clockid_tclockid;
+ size_t clockid_res_ns;
unsigned int proc_map_timeout;
};
diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
index 1f3ccc368530..b167a54d635f 100644
--- a/tools/perf/util/env.h
+++ b/tools/perf/util/env.h
@@ -63,6 +63,7 @@ struct perf_env {
struct numa_node*numa_nodes;
struct memory_node *memory_nodes;
unsigned long long memory_bsize;
+ size_t clockid_res_ns;
};
extern struct perf_env perf_env;
diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index 1ec1d9bc2d63..4ce5339158f7 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -1034,6 +1034,13 @@ static int write_auxtrace(struct feat_fd *ff,
return err;
}
+static int write_clockid(struct feat_fd *ff,
+struct perf_evlist *evlist __maybe_unused)
+{
+ return do_write(ff, >ph->env.clockid_res_ns,
+ sizeof(ff->ph->env.clockid_res_ns));
+}
+
static int cpu_cache_level__sort(const void *a, const void *b)
{
struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
@@ -1508,6 +1515,12 @@ static void print_cpu_topology(struct feat_fd *ff, FILE
*fp)
fprintf(fp, "# Core ID and Socket ID information is not
available\n");
}
+static void print_clockid(struct feat_fd *ff, FILE *fp)
+{
+ fprintf(fp, "# clockid frequency: %ld MHz\n",
+
[PATCH v2] perf record: encode -k clockid frequency into Perf trace
Store -k clockid frequency into Perf trace to enable timestamps
derived metrics conversion into wall clock time on reporting stage.
Below is the example of perf report output:
tools/perf/perf record -k raw -- ../../matrix/linux/matrix.gcc
...
[ perf record: Captured and wrote 31.222 MB perf.data (818054 samples) ]
tools/perf/perf report --header
#
...
# event : name = cycles:ppp, , size = 112, { sample_period, sample_freq } =
4000, sample_type = IP|TID|TIME|PERIOD, disabled = 1, inherit = 1, mmap = 1,
comm = 1, freq = 1, enable_on_exec = 1, task = 1, precise_ip = 3, sample_id_all
= 1, exclude_guest = 1, mmap2 = 1, comm_exec = 1, use_clockid = 1, clockid = 4
...
# clockid frequency: 1000 MHz
...
#
Signed-off-by: Alexey Budankov
---
Changes in v2:
- renamed clockid_freq to clockid_res_ns and get_clockid_freq() to
get_clockid_res()
- avoided redundant define of NSEC_IN_SEC, reused linux/time64.h:NSEC_PER_SEC
- moved MHz conversion into print_clockid() and shortened write_clockid()
---
tools/perf/builtin-record.c | 21 +++--
tools/perf/perf.h | 1 +
tools/perf/util/env.h | 1 +
tools/perf/util/header.c| 27 +++
tools/perf/util/header.h| 1 +
5 files changed, 49 insertions(+), 2 deletions(-)
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 0980dfe3396b..127dc7432394 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -592,6 +592,11 @@ static void record__init_features(struct record *rec)
if (!rec->opts.full_auxtrace)
perf_header__clear_feat(>header, HEADER_AUXTRACE);
+ if (rec->opts.use_clockid && rec->opts.clockid_res_ns)
+ session->header.env.clockid_res_ns = rec->opts.clockid_res_ns;
+ else
+ perf_header__clear_feat(>header, HEADER_CLOCKID);
+
perf_header__clear_feat(>header, HEADER_STAT);
}
@@ -1337,6 +1342,17 @@ static const struct clockid_map clockids[] = {
CLOCKID_END,
};
+static int get_clockid_res(clockid_t clk_id, size_t *res_ns)
+{
+ struct timespec res;
+
+ *res_ns = 0;
+ if (!clock_getres(clk_id, ))
+ *res_ns = res.tv_nsec + res.tv_sec * NSEC_PER_SEC;
+
+ return 0;
+}
+
static int parse_clockid(const struct option *opt, const char *str, int unset)
{
struct record_opts *opts = (struct record_opts *)opt->value;
@@ -1360,7 +1376,7 @@ static int parse_clockid(const struct option *opt, const
char *str, int unset)
/* if its a number, we're done */
if (sscanf(str, "%d", >clockid) == 1)
- return 0;
+ return get_clockid_res(opts->clockid, >clockid_res_ns);
/* allow a "CLOCK_" prefix to the name */
if (!strncasecmp(str, "CLOCK_", 6))
@@ -1369,7 +1385,8 @@ static int parse_clockid(const struct option *opt, const
char *str, int unset)
for (cm = clockids; cm->name; cm++) {
if (!strcasecmp(str, cm->name)) {
opts->clockid = cm->clockid;
- return 0;
+ return get_clockid_res(opts->clockid,
+ >clockid_res_ns);
}
}
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index 21bf7f5a3cf5..981db3c2ed60 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -81,6 +81,7 @@ struct record_opts {
unsigned initial_delay;
bool use_clockid;
clockid_tclockid;
+ size_t clockid_res_ns;
unsigned int proc_map_timeout;
};
diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
index 1f3ccc368530..b167a54d635f 100644
--- a/tools/perf/util/env.h
+++ b/tools/perf/util/env.h
@@ -63,6 +63,7 @@ struct perf_env {
struct numa_node*numa_nodes;
struct memory_node *memory_nodes;
unsigned long long memory_bsize;
+ size_t clockid_res_ns;
};
extern struct perf_env perf_env;
diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index 1ec1d9bc2d63..4ce5339158f7 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -1034,6 +1034,13 @@ static int write_auxtrace(struct feat_fd *ff,
return err;
}
+static int write_clockid(struct feat_fd *ff,
+struct perf_evlist *evlist __maybe_unused)
+{
+ return do_write(ff, >ph->env.clockid_res_ns,
+ sizeof(ff->ph->env.clockid_res_ns));
+}
+
static int cpu_cache_level__sort(const void *a, const void *b)
{
struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
@@ -1508,6 +1515,12 @@ static void print_cpu_topology(struct feat_fd *ff, FILE
*fp)
fprintf(fp, "# Core ID and Socket ID information is not
available\n");
}
+static void print_clockid(struct feat_fd *ff, FILE *fp)
+{
+ fprintf(fp, "# clockid frequency: %ld MHz\n",
+
[PATCH v2] perf record: encode -k clockid frequency into Perf trace
Store -k clockid frequency into Perf trace to enable timestamps
derived metrics conversion into wall clock time on reporting stage.
Below is the example of perf report output:
tools/perf/perf record -k raw -- ../../matrix/linux/matrix.gcc
...
[ perf record: Captured and wrote 31.222 MB perf.data (818054 samples) ]
tools/perf/perf report --header
#
...
# event : name = cycles:ppp, , size = 112, { sample_period, sample_freq } =
4000, sample_type = IP|TID|TIME|PERIOD, disabled = 1, inherit = 1, mmap = 1,
comm = 1, freq = 1, enable_on_exec = 1, task = 1, precise_ip = 3, sample_id_all
= 1, exclude_guest = 1, mmap2 = 1, comm_exec = 1, use_clockid = 1, clockid = 4
...
# clockid frequency: 1000 MHz
...
#
Signed-off-by: Alexey Budankov
---
Changes in v2:
- renamed clockid_freq to clockid_res_ns and get_clockid_freq() to
get_clockid_res()
- avoided redundant define of NSEC_IN_SEC, reused linux/time64.h:NSEC_PER_SEC
- moved MHz conversion into print_clockid() and shortened write_clockid()
---
tools/perf/builtin-record.c | 21 +++--
tools/perf/perf.h | 1 +
tools/perf/util/env.h | 1 +
tools/perf/util/header.c| 27 +++
tools/perf/util/header.h| 1 +
5 files changed, 49 insertions(+), 2 deletions(-)
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 0980dfe3396b..127dc7432394 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -592,6 +592,11 @@ static void record__init_features(struct record *rec)
if (!rec->opts.full_auxtrace)
perf_header__clear_feat(>header, HEADER_AUXTRACE);
+ if (rec->opts.use_clockid && rec->opts.clockid_res_ns)
+ session->header.env.clockid_res_ns = rec->opts.clockid_res_ns;
+ else
+ perf_header__clear_feat(>header, HEADER_CLOCKID);
+
perf_header__clear_feat(>header, HEADER_STAT);
}
@@ -1337,6 +1342,17 @@ static const struct clockid_map clockids[] = {
CLOCKID_END,
};
+static int get_clockid_res(clockid_t clk_id, size_t *res_ns)
+{
+ struct timespec res;
+
+ *res_ns = 0;
+ if (!clock_getres(clk_id, ))
+ *res_ns = res.tv_nsec + res.tv_sec * NSEC_PER_SEC;
+
+ return 0;
+}
+
static int parse_clockid(const struct option *opt, const char *str, int unset)
{
struct record_opts *opts = (struct record_opts *)opt->value;
@@ -1360,7 +1376,7 @@ static int parse_clockid(const struct option *opt, const
char *str, int unset)
/* if its a number, we're done */
if (sscanf(str, "%d", >clockid) == 1)
- return 0;
+ return get_clockid_res(opts->clockid, >clockid_res_ns);
/* allow a "CLOCK_" prefix to the name */
if (!strncasecmp(str, "CLOCK_", 6))
@@ -1369,7 +1385,8 @@ static int parse_clockid(const struct option *opt, const
char *str, int unset)
for (cm = clockids; cm->name; cm++) {
if (!strcasecmp(str, cm->name)) {
opts->clockid = cm->clockid;
- return 0;
+ return get_clockid_res(opts->clockid,
+ >clockid_res_ns);
}
}
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index 21bf7f5a3cf5..981db3c2ed60 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -81,6 +81,7 @@ struct record_opts {
unsigned initial_delay;
bool use_clockid;
clockid_tclockid;
+ size_t clockid_res_ns;
unsigned int proc_map_timeout;
};
diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
index 1f3ccc368530..b167a54d635f 100644
--- a/tools/perf/util/env.h
+++ b/tools/perf/util/env.h
@@ -63,6 +63,7 @@ struct perf_env {
struct numa_node*numa_nodes;
struct memory_node *memory_nodes;
unsigned long long memory_bsize;
+ size_t clockid_res_ns;
};
extern struct perf_env perf_env;
diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index 1ec1d9bc2d63..4ce5339158f7 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -1034,6 +1034,13 @@ static int write_auxtrace(struct feat_fd *ff,
return err;
}
+static int write_clockid(struct feat_fd *ff,
+struct perf_evlist *evlist __maybe_unused)
+{
+ return do_write(ff, >ph->env.clockid_res_ns,
+ sizeof(ff->ph->env.clockid_res_ns));
+}
+
static int cpu_cache_level__sort(const void *a, const void *b)
{
struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
@@ -1508,6 +1515,12 @@ static void print_cpu_topology(struct feat_fd *ff, FILE
*fp)
fprintf(fp, "# Core ID and Socket ID information is not
available\n");
}
+static void print_clockid(struct feat_fd *ff, FILE *fp)
+{
+ fprintf(fp, "# clockid frequency: %ld MHz\n",
+
