perf tools: Add wallclock time conversion support
Jiri Olsa
reviewercover perf tools: Add wallclock time conversion support
hi,
this patchset is based on changes made by David Ahern long time ago. The perf code moved a lot since then, but the idea is the same.
The patchset is adding the ability to display TOD/wallclock timestamp in 'perf script' output and in 'perf data convert --to-ctf' subcommand, so the converted CTF data contain TOD/wallclock timestamps.
It's done by adding new header FEATURE 'CLOCK_DATA' to perf.data, that stores reference times for both TOD/wallclock time and for the clock that perf record is configured to use. These reference times are then used to convert sample's timestamps to TOD/wallclock timestamps.
The feature is available only for recording with clockid specified, because it's the only case where we can get reference time to TOD/wallclock time. We can't do that with perf clock yet.
Also available in here:
git://git.kernel.org/pub/scm/linux/kernel/git/jolsa/perf.git
perf/abs_time
thanks,
jirka
---
Jiri Olsa (6):
perf tools: Add clockid_name function
perf tools: Store clock references for -k/--clockid option perf tools: Move clockid_res_ns under clock struct
perf tools: Add support to store time of day in CTF data conversion perf script: Change enum perf_output_field values to be 64 bits perf script: Add tod field to display time of day
tools/perf/Documentation/perf-data.txt | 3 ++ tools/perf/Documentation/perf.data-file-format.txt | 13 +++++++ tools/perf/builtin-data.c | 1 + tools/perf/builtin-record.c | 58 ++++++++++++++++++++++++++++-- tools/perf/builtin-script.c | 195 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++---------------------------------- tools/perf/util/data-convert-bt.c | 56 ++++++++++++++++++----------- tools/perf/util/data-convert.h | 1 + tools/perf/util/env.h | 14 +++++++- tools/perf/util/header.c | 120 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++--- tools/perf/util/header.h | 1 + tools/perf/util/util.h | 2 ++ 11 files changed, 371 insertions(+), 93 deletions(-)
Jiri Olsa
reviewerPATCH 1/6 perf tools: Add clockid_name function
Adding clockid_name function to get the clock name based on its clockid. It will be used in following changes.
--- tools/perf/builtin-record.c | 11 +++++++++++ tools/perf/util/util.h | 2 ++ 2 files changed, 13 insertions(+)
tools/perf/builtin-record.c +11 -0
@@ -2121,6 +2121,17 @@ static const struct clockid_map clockids[] = { CLOCKID_END, }; +const char *clockid_name (clockid_t clk_id) +{ +const struct clockid_map *cm; + +for (cm = clockids; cm->name; cm++) { +if (cm->clockid == clk_id) +return cm->name; + } +return "(not found)" ; +} +static int get_clockid_res (clockid_t clk_id,u64 *res_ns) {struct timespec res;
tools/perf/util/util.h +2 -0
@@ -62,4 +62,6 @@ char *perf_exe(char *buf, int len);#endif #endif +const char *clockid_name (clockid_t clk_id); +#endif /* GIT_COMPAT_UTIL_H */
Signature · 1 line
2.25.4
Re: [PATCH 1/6] perf tools: Add clockid_name function
tools/perf/util/util.h char *perf_exe(char *buf, int len); · line 65
Quoted 9 lines
tools/perf/util/util.h +1
-0@@ -62,4 +62,6 @@ char *perf_exe(char *buf, int len);#endif #endif +const char *clockid_name (clockid_t clk_id);
If it's a generic "util" it probably shouldn't live in builtin-record.c
-Andi
Re: [PATCH 1/6] perf tools: Add clockid_name function
Quoted 11 lines · Andi Kleen
tools/perf/util/util.h char *perf_exe(char *buf, int len); · line 65
Quoted 9 lines
tools/perf/util/util.h +1
-0@@ -62,4 +62,6 @@ char *perf_exe(char *buf, int len);#endif #endif +const char *clockid_name (clockid_t clk_id);If it's a generic "util" it probably shouldn't live in
builtin-record.c
true, but it's where all the clockid data is, I'd need to move parse_clockid as well, I'll check
thanks,
jirka
Jiri Olsa
reviewerPATCH 2/6 perf tools: Store clock references for -k/--clockid option
In reply to Jiri OlsaAdding new CLOCK_DATA feature that stores reference times when -k/--clockid option is specified.
It contains clock id and its reference time together with wall clock time taken at the 'same time', both values are in nanoseconds.
The format of data is as below:
struct {
u32 version; /* version = 1 */
u32 clockid;
u64 clockid_time_ns;
u64 wall_clock_ns;
};This clock reference times will be used in following changes to display wall clock for perf events.
It's available only for recording with clockid specified, because it's the only case where we can get reference time to wallclock time. It's can't do that with perf clock yet.
--- .../Documentation/perf.data-file-format.txt | 13 ++ tools/perf/builtin-record.c | 41 +++++++ tools/perf/util/env.h | 12 ++ tools/perf/util/header.c | 112 ++++++++++++++++++ tools/perf/util/header.h | 1 + 5 files changed, 179 insertions(+)
tools/perf/Documentation/perf.data-file-format.txt +13 -0
@@ -389,6 +389,19 @@ struct { Example: cpu pmu capabilities: branches=32, max_precise=3, pmu_name=icelake + HEADER_CLOCK_DATA = 29, + + Contains clock id and its reference time together with wall clock + time taken at the 'same time', both values are in nanoseconds. + The format of data is as below. + +struct { + u32 version; /* version = 1 */ + u32 clockid; + u64 clockid_time_ns; + u64 wall_clock_ns; +}; + other bits are reserved and should ignored for now HEADER_FEAT_BITS = 256,
tools/perf/builtin-record.c +41 -0
65 lines not on this page · raw message
tools/perf/util/env.h +12 -0
@@ -100,6 +100,18 @@ struct perf_env {/* For fast cpu to numa node lookup via perf_env__numa_node */ int *numa_map;int nr_numa_map; + +/* For real clock time refference. */ +struct { +u64 tod_ns; +u64 clockid_ns; +int clockid; +/* +* enabled is valid for report mode, and is true if above +* values are set, it's set in process_clock_data +*/ +bool enabled; + } clock; };enum perf_compress_type {
tools/perf/util/header.c +112 -0
136 lines not on this page · raw message
tools/perf/util/header.h +1 -0
@@ -44,6 +44,7 @@ enum {HEADER_BPF_BTF ,HEADER_COMPRESSED ,HEADER_CPU_PMU_CAPS , +HEADER_CLOCK_DATA ,HEADER_LAST_FEATURE , HEADER_FEAT_BITS = 256, };
Signature · 1 line
2.25.4
Re: [PATCH 2/6] perf tools: Store clock references for -k/--clockid option
Quoted 94 lines · Jiri Olsa
Adding new
CLOCK_DATAfeature that stores reference times when -k/--clockid option is specified.It contains clock id and its reference time together with wall clock time taken at the 'same time', both values are in nanoseconds.
The format of data is as below:
struct { u32 version; /* version = 1 */ u32 clockid; u64 clockid_time_ns; u64 wall_clock_ns; };This clock reference times will be used in following changes to display wall clock for perf events.
It's available only for recording with clockid specified, because it's the only case where we can get reference time to wallclock time. It's can't do that with perf clock yet.
Original-patch-by David Ahern <[email protected]>Signed-off-by Jiri Olsa <[email protected]>--- .../Documentation/perf.data-file-format.txt | 13 ++ tools/perf/builtin-record.c | 41 +++++++ tools/perf/util/env.h | 12 ++ tools/perf/util/header.c | 112 ++++++++++++++++++ tools/perf/util/header.h | 1 + 5 files changed, 179 insertions(+)tools/perf/Documentation/perf.data-file-format.txt +13
-0@@ -389,6 +389,19 @@ struct { Example: cpu pmu capabilities: branches=32, max_precise=3, pmu_name=icelake + HEADER_CLOCK_DATA = 29, + + Contains clock id and its reference time together with wall clock + time taken at the 'same time', both values are in nanoseconds. + The format of data is as below. + +struct { + u32 version; /* version = 1 */ + u32 clockid; + u64 clockid_time_ns; + u64 wall_clock_ns; +}; + other bits are reserved and should ignored for now HEADER_FEAT_BITS = 256,tools/perf/builtin-record.c +15
-0@@ -70,6 +70,7 @@#include <linux/time64.h> #include <linux/zalloc.h> #include <linux/bitmap.h> +#include <sys/time.h> struct switch_output {bool enabled;@@ -1203,6 +1204,9 @@ static void record__init_features(struct record *rec)if (!(rec->opts.use_clockid && rec->opts.clockid_res_ns))perf_header__clear_feat (&session->header,HEADER_CLOCKID ); +if (!rec->opts.use_clockid) +perf_header__clear_feat (&session->header,HEADER_CLOCK_DATA ); +perf_header__clear_feat (&session->header,HEADER_DIR_FORMAT );if (!record__comp_enabled (rec))perf_header__clear_feat (&session->header,HEADER_COMPRESSED );@@ -1551,6 +1555,40 @@ static int record__setup_sb_evlist(struct record *rec)return 0 ; } +static int record__init_clock (struct record *rec) +{ +struct perf_session *session = rec->session; +struct timespec ref_clockid; +struct timeval ref_tod; +u64 ref; + +if (!rec->opts.use_clockid) +return 0 ; + + session->header.env.clock.clockid = rec->opts.clockid;
tools/perf/builtin-record.c static int record__setup_sb_evlist(struct record *rec) · line 1578
Quoted 10 lines
tools/perf/builtin-record.c +10
-0+ +if (gettimeofday (&ref_tod,NULL ) !=0 ) { +pr_err ("gettimeofday failed, cannot set reference time.\n" ); +return -1 ; + } + +if (clock_gettime (rec->opts.clockid, &ref_clockid)) { +pr_err ("clock_gettime failed, cannot set reference time.\n" ); +return -1 ; + }
It might also want to be implemented in a loop and iteration with minimal time delta is chosen to improve synchronization accuracy and also mitigate possible context switches between gettimeofday() and clock_gettime() calls.
Alexei
Re: [PATCH 2/6] perf tools: Store clock references for -k/--clockid option
Quoted 102 lines · Alexey Budankov
Quoted 100 lines · Jiri Olsa
Adding new
CLOCK_DATAfeature that stores reference times when -k/--clockid option is specified.It contains clock id and its reference time together with wall clock time taken at the 'same time', both values are in nanoseconds.
The format of data is as below:
struct { u32 version; /* version = 1 */ u32 clockid; u64 clockid_time_ns; u64 wall_clock_ns; };This clock reference times will be used in following changes to display wall clock for perf events.
It's available only for recording with clockid specified, because it's the only case where we can get reference time to wallclock time. It's can't do that with perf clock yet.
Original-patch-by David Ahern <[email protected]>Signed-off-by Jiri Olsa <[email protected]>--- .../Documentation/perf.data-file-format.txt | 13 ++ tools/perf/builtin-record.c | 41 +++++++ tools/perf/util/env.h | 12 ++ tools/perf/util/header.c | 112 ++++++++++++++++++ tools/perf/util/header.h | 1 + 5 files changed, 179 insertions(+)tools/perf/Documentation/perf.data-file-format.txt +13
-0@@ -389,6 +389,19 @@ struct { Example: cpu pmu capabilities: branches=32, max_precise=3, pmu_name=icelake + HEADER_CLOCK_DATA = 29, + + Contains clock id and its reference time together with wall clock + time taken at the 'same time', both values are in nanoseconds. + The format of data is as below. + +struct { + u32 version; /* version = 1 */ + u32 clockid; + u64 clockid_time_ns; + u64 wall_clock_ns; +}; + other bits are reserved and should ignored for now HEADER_FEAT_BITS = 256,tools/perf/builtin-record.c +21
-0@@ -70,6 +70,7 @@#include <linux/time64.h> #include <linux/zalloc.h> #include <linux/bitmap.h> +#include <sys/time.h> struct switch_output {bool enabled;@@ -1203,6 +1204,9 @@ static void record__init_features(struct record *rec)if (!(rec->opts.use_clockid && rec->opts.clockid_res_ns))perf_header__clear_feat (&session->header,HEADER_CLOCKID ); +if (!rec->opts.use_clockid) +perf_header__clear_feat (&session->header,HEADER_CLOCK_DATA ); +perf_header__clear_feat (&session->header,HEADER_DIR_FORMAT );if (!record__comp_enabled (rec))perf_header__clear_feat (&session->header,HEADER_COMPRESSED );@@ -1551,6 +1555,40 @@ static int record__setup_sb_evlist(struct record *rec)return 0 ; } +static int record__init_clock (struct record *rec) +{ +struct perf_session *session = rec->session; +struct timespec ref_clockid; +struct timeval ref_tod; +u64 ref; + +if (!rec->opts.use_clockid) +return 0 ; + + session->header.env.clock.clockid = rec->opts.clockid; + +if (gettimeofday (&ref_tod,NULL ) !=0 ) { +pr_err ("gettimeofday failed, cannot set reference time.\n" ); +return -1 ; + } +
Quoted 8 lines
tools/perf/builtin-record.c static int record__setup_sb_evlist(struct record *rec) · line 1578
Quoted 4 lines
tools/perf/builtin-record.c +4
-0+if (clock_gettime (rec->opts.clockid, &ref_clockid)) { +pr_err ("clock_gettime failed, cannot set reference time.\n" ); +return -1 ; + }It might also want to be implemented in a loop and iteration with minimal time delta is chosen to improve synchronization accuracy and also mitigate possible context switches between
gettimeofday()andclock_gettime()calls.
right, we could make this more accurate.. I'll post some follow up change with that
thanks,
jirka
Re: [PATCH 2/6] perf tools: Store clock references for -k/--clockid option
Quoted 5 lines · Jiri Olsa
Quoted 3 lines
It might also want to be implemented in a loop and iteration with minimal time delta is chosen to improve synchronization accuracy and also mitigate possible context switches between
gettimeofday()andclock_gettime()calls.right, we could make this more accurate.. I'll post some follow up change with that
If it is handled through vdso (e.g., x86) both calls should be very fast with no context switches and the margin of error is well less than the usec resolution in the output.
Namhyung Kim
maintainerRe: [PATCH 2/6] perf tools: Store clock references for -k/--clockid option
In reply to Jiri OlsaQuoted 44 lines · Jiri Olsa
Adding new
CLOCK_DATAfeature that stores reference times when -k/--clockid option is specified.It contains clock id and its reference time together with wall clock time taken at the 'same time', both values are in nanoseconds.
The format of data is as below:
struct { u32 version; /* version = 1 */ u32 clockid; u64 clockid_time_ns; u64 wall_clock_ns; };This clock reference times will be used in following changes to display wall clock for perf events.
It's available only for recording with clockid specified, because it's the only case where we can get reference time to wallclock time. It's can't do that with perf clock yet.
Original-patch-by David Ahern <[email protected]>Signed-off-by Jiri Olsa <[email protected]>--- .../Documentation/perf.data-file-format.txt | 13 ++ tools/perf/builtin-record.c | 41 +++++++ tools/perf/util/env.h | 12 ++ tools/perf/util/header.c | 112 ++++++++++++++++++ tools/perf/util/header.h | 1 + 5 files changed, 179 insertions(+)tools/perf/Documentation/perf.data-file-format.txt +3
-0@@ -389,6 +389,19 @@ struct { Example: cpu pmu capabilities: branches=32, max_precise=3, pmu_name=icelake + HEADER_CLOCK_DATA = 29, + + Contains clock id and its reference time together with wall clock
tools/perf/Documentation/perf.data-file-format.txt struct { · line 404
Quoted 10 lines
tools/perf/Documentation/perf.data-file-format.txt +10
-0+ time taken at the 'same time', both values are in nanoseconds. + The format of data is as below. + +struct { + u32 version; /* version = 1 */ + u32 clockid; + u64 clockid_time_ns; + u64 wall_clock_ns; +}; +
It seems that it's slightly different than what it actually write to a file. Specifically the last two fields should be reversed IMHO.
Quoted 3 lines
tools/perf/Documentation/perf.data-file-format.txt +0
-0other bits are reserved and should ignored for now HEADER_FEAT_BITS = 256,
[SNIP]
tools/perf/util/env.h struct perf_env { · line 104
Quoted 10 lines
tools/perf/util/env.h +2
-0@@ -100,6 +100,18 @@ struct perf_env {/* For fast cpu to numa node lookup via perf_env__numa_node */ int *numa_map;int nr_numa_map; + +/* For real clock time refference. */
typo: reference
Quoted 13 lines
tools/perf/util/env.h +10
-0+struct { +u64 tod_ns; +u64 clockid_ns; +int clockid; +/* +* enabled is valid for report mode, and is true if above +* values are set, it's set in process_clock_data +*/ +bool enabled; + } clock; };enum perf_compress_type {
[SNIP]
Quoted 28 lines
tools/perf/util/env.h +28
-0+static void print_clock_data (struct feat_fd *ff,FILE *fp) +{ +struct timespec clockid_ns; +char tstr[64 ], date[64 ]; +struct timeval tod_ns; +clockid_t clockid; +struct tm ltime; +u64 ref; + +if (!ff->ph->env.clock.enabled) { +fprintf (fp,"# reference time disabled\n" ); +return ; + } + +/* Compute TOD time. */ + ref = ff->ph->env.clock.tod_ns; + tod_ns.tv_sec = ref /NSEC_PER_SEC ; + ref -= tod_ns.tv_sec *NSEC_PER_SEC ; + tod_ns.tv_usec = ref /NSEC_PER_USEC ; + +/* Compute clockid time. */ + ref = ff->ph->env.clock.clockid_ns; + clockid_ns.tv_sec = ref /NSEC_PER_SEC ; + ref -= clockid_ns.tv_sec *NSEC_PER_SEC ; + clockid_ns.tv_nsec = ref; + + clockid = ff->ph->env.clock.clockid; +
tools/perf/util/env.h
Quoted 10 lines
tools/perf/util/env.h +10
-0+if (localtime_r (&tod_ns.tv_sec, <ime) ==NULL ) +snprintf (tstr,sizeof (tstr),"<error>" ); +else { +strftime (date,sizeof (date),"%F %T" , <ime); +scnprintf (tstr,sizeof (tstr),"%s.%06d" , + date, (int ) tod_ns.tv_usec); + } + +fprintf (fp,"# clockid: %s (%u)\n" ,clockid_name (clockid), clockid); + fprintf(fp, "# reference time: %s = %ld.%06d (TOD) = %ld.%ld (%s)\n",
Shouldn't the last one be %ld.%09ld?
Thanks
Namhyung
Quoted 71 lines
tools/perf/util/env.h +40
-0+ tstr, tod_ns.tv_sec, (int ) tod_ns.tv_usec, + clockid_ns.tv_sec, clockid_ns.tv_nsec, + clockid_name(clockid)); +} +static void print_dir_format (struct feat_fd *ff,FILE *fp) {struct perf_session *session;@@ -2738,6 +2815,40 @@ static int process_clockid(struct feat_fd *ff,return 0 ; } +static int process_clock_data (struct feat_fd *ff, +void *_data__maybe_unused ) +{ +u32 data32; +u64 data64; + +/* version */ +if (do_read_u32 (ff, &data32)) +return -1 ; + +if (data32 !=1 ) +return -1 ; + +/* clockid */ +if (do_read_u32 (ff, &data32)) +return -1 ; + + ff->ph->env.clock.clockid = data32; + +/* TOD ref time */ +if (do_read_u64 (ff, &data64)) +return -1 ; + + ff->ph->env.clock.tod_ns = data64; + +/* clockid ref time */ +if (do_read_u64 (ff, &data64)) +return -1 ; + + ff->ph->env.clock.clockid_ns = data64; + ff->ph->env.clock.enabled = true; +return 0 ; +} +static int process_dir_format (struct feat_fd *ff,void *_data__maybe_unused ) {@@ -3008,6 +3119,7 @@ const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE] = {FEAT_OPR (BPF_BTF , bpf_btf, false),FEAT_OPR (COMPRESSED , compressed, false),FEAT_OPR (CPU_PMU_CAPS , cpu_pmu_caps, false), + FEAT_OPR(CLOCK_DATA, clock_data, false), }; struct header_print_data {tools/perf/util/header.h +1
-0@@ -44,6 +44,7 @@ enum {HEADER_BPF_BTF ,HEADER_COMPRESSED ,HEADER_CPU_PMU_CAPS , +HEADER_CLOCK_DATA ,HEADER_LAST_FEATURE , HEADER_FEAT_BITS = 256, };Signature · 1 line
2.25.4
Re: [PATCH 2/6] perf tools: Store clock references for -k/--clockid option
On Mon, Aug 03, 2020 at 12:55:50PM +0900, Namhyung Kim wrote:
SNIP
Quoted 8 lines
Quoted 8 lines
tools/perf/Documentation/perf.data-file-format.txt +0
-0@@ -389,6 +389,19 @@ struct { Example: cpu pmu capabilities: branches=32, max_precise=3, pmu_name=icelake
Quoted 16 lines
tools/perf/Documentation/perf.data-file-format.txt struct { · line 404
Quoted 13 lines
tools/perf/Documentation/perf.data-file-format.txt +13
-0+ HEADER_CLOCK_DATA = 29, + + Contains clock id and its reference time together with wall clock + time taken at the 'same time', both values are in nanoseconds. + The format of data is as below. + +struct { + u32 version; /* version = 1 */ + u32 clockid; + u64 clockid_time_ns; + u64 wall_clock_ns; +}; +It seems that it's slightly different than what it actually write to a file. Specifically the last two fields should be reversed IMHO.
oops, you're right.. wil fix in new version
Quoted 7 lines
Quoted 5 lines
tools/perf/Documentation/perf.data-file-format.txt +0
-0other bits are reserved and should ignored for now HEADER_FEAT_BITS = 256,[SNIP]
Quoted 12 lines
tools/perf/util/env.h struct perf_env { · line 104
Quoted 10 lines
tools/perf/util/env.h +2
-0@@ -100,6 +100,18 @@ struct perf_env {/* For fast cpu to numa node lookup via perf_env__numa_node */ int *numa_map;int nr_numa_map; + +/* For real clock time refference. */ typo: reference
ok
SNIP
Quoted 10 lines
tools/perf/util/env.h struct perf_env { · line 112
Quoted 8 lines
tools/perf/util/env.h +8
-0+else { +strftime (date,sizeof (date),"%F %T" , <ime); +scnprintf (tstr,sizeof (tstr),"%s.%06d" , + date, (int ) tod_ns.tv_usec); + } + +fprintf (fp,"# clockid: %s (%u)\n" ,clockid_name (clockid), clockid); + fprintf(fp, "# reference time: %s = %ld.%06d (TOD) = %ld.%ld (%s)\n",Shouldn't the last one be %ld.%09ld?
sure, why not
thanks,
jirka
Jiri Olsa
reviewerPATCH 3/6 perf tools: Move clockid_res_ns under clock struct
In reply to Jiri OlsaMoving clockid_res_ns under clock struct, so we have clock related stuff in one place.
--- tools/perf/builtin-record.c | 6 +++--- tools/perf/util/env.h | 2 +- tools/perf/util/header.c | 8 ++++---- 3 files changed, 8 insertions(+), 8 deletions(-)
tools/perf/builtin-record.c +3 -3
@@ -1565,6 +1565,9 @@ static int record__init_clock(struct record *rec)if (!rec->opts.use_clockid)return 0 ; +if (rec->opts.use_clockid && rec->opts.clockid_res_ns) + session->header.env.clock.clockid_res_ns = rec->opts.clockid_res_ns; + session->header.env.clock.clockid = rec->opts.clockid;if (gettimeofday (&ref_tod,NULL ) !=0 ) {
@@ -1673,9 +1676,6 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)record__init_features (rec); -if (rec->opts.use_clockid && rec->opts.clockid_res_ns) - session->header.env.clockid_res_ns = rec->opts.clockid_res_ns; -if (forks) { err = perf_evlist__prepare_workload(rec->evlist, &opts->target, argv, data->is_pipe,
tools/perf/util/env.h +1 -1
@@ -77,7 +77,6 @@ struct perf_env {struct numa_node *numa_nodes;struct memory_node *memory_nodes;unsigned long long memory_bsize; -u64 clockid_res_ns;/* * bpf_info_lock protects bpf rbtrees. This is needed because the
@@ -105,6 +104,7 @@ struct perf_env {struct {u64 tod_ns;u64 clockid_ns; +u64 clockid_res_ns;int clockid;/* * enabled is valid for report mode, and is true if above
tools/perf/util/header.c +4 -4
@@ -891,8 +891,8 @@ static int write_auxtrace(struct feat_fd *ff,static int write_clockid (struct feat_fd *ff,struct evlist *evlist__maybe_unused ) { -return do_write (ff, &ff->ph->env.clockid_res_ns, -sizeof (ff->ph->env.clockid_res_ns)); +return do_write (ff, &ff->ph->env.clock.clockid_res_ns, +sizeof (ff->ph->env.clock.clockid_res_ns)); } static int write_clock_data(struct feat_fd *ff,
@@ -1580,7 +1580,7 @@ static void print_cpu_topology(struct feat_fd *ff, FILE *fp)static void print_clockid (struct feat_fd *ff,FILE *fp) {fprintf (fp,"# clockid frequency: %" PRIu64" MHz\n" , - ff->ph->env.clockid_res_ns *1000 ); + ff->ph->env.clock.clockid_res_ns *1000 ); }static void print_clock_data (struct feat_fd *ff,FILE *fp)
@@ -2809,7 +2809,7 @@ static int process_mem_topology(struct feat_fd *ff,static int process_clockid (struct feat_fd *ff,void *data__maybe_unused ) { -if (do_read_u64 (ff, &ff->ph->env.clockid_res_ns)) +if (do_read_u64 (ff, &ff->ph->env.clock.clockid_res_ns))return -1 ;return 0 ;
Signature · 1 line
2.25.4
Jiri Olsa
reviewerPATCH 4/6 perf tools: Add support to store time of day in CTF data conversion
In reply to Jiri OlsaAdding support to convert and store time of day in CTF data conversion for 'perf data convert' subcommand.
The perf.data used for conversion needs to have clock data information - must be recorded with -k/--clockid option).
New --tod option is added to 'perf data convert' subcommand to convert data with timestamps converted to wall clock time.
Record data with clockid set:
# perf record -k CLOCK_MONOTONIC kill kill: not enough arguments [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.033 MB perf.data (8 samples) ]
Convert data with TOD timestamps:
# perf data convert --tod --to-ctf ./ctf [ perf data convert: Converted 'perf.data' into CTF data './ctf' ] [ perf data convert: Converted and wrote 0.000 MB (8 samples) ]
Display data in perf script:
# perf script -F+tod --ns
perf 262150 2020-07-13 18:38:50.097678523 153633.958246159: 1 cycles: ...
perf 262150 2020-07-13 18:38:50.097682941 153633.958250577: 1 cycles: ...
perf 262150 2020-07-13 18:38:50.097684997 153633.958252633: 7 cycles: ...
...Display data in babeltrace:
# babeltrace --clock-date ./ctf
[2020-07-13 18:38:50.097678523] (+?.?????????) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ...
[2020-07-13 18:38:50.097682941] (+0.000004418) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ...
[2020-07-13 18:38:50.097684997] (+0.000002056) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ...
...It's available only for recording with clockid specified, because it's the only case where we can get reference time to wallclock time. It's can't do that with perf clock yet.
Error is display if you want to use --tod on data without clockid specified:
# perf data convert --tod --to-ctf ./ctf Can't provide --tod time, missing clock data. Please record with -k/--clockid option. Failed to setup CTF writer. Error during conversion setup.
--- tools/perf/Documentation/perf-data.txt | 3 ++ tools/perf/builtin-data.c | 1 + tools/perf/util/data-convert-bt.c | 56 +++++++++++++++++--------- tools/perf/util/data-convert.h | 1 + 4 files changed, 41 insertions(+), 20 deletions(-)
tools/perf/Documentation/perf-data.txt +3 -0
@@ -27,6 +27,9 @@ OPTIONS for 'convert' --to-ctf:: Triggers the CTF conversion, specify the path of CTF data directory. +--tod:: + Convert time to wall clock time. + -i:: Specify input perf data file path.
tools/perf/builtin-data.c +1 -0
@@ -65,6 +65,7 @@ static int cmd_data_convert(int argc, const char **argv) OPT_STRING('i', "input", &input_name, "file", "input file name"),#ifdef HAVE_LIBBABELTRACE_SUPPORT OPT_STRING (0 ,"to-ctf" , &to_ctf,NULL ,"Convert to CTF format" ), +OPT_BOOLEAN (0 ,"tod" , &opts.tod,"Convert time to wall clock time" ), #endifOPT_BOOLEAN ('f' ,"force" , &opts.force,"don't complain, do it" ), OPT_BOOLEAN(0, "all", &opts.all, "Convert all events"),
tools/perf/util/data-convert-bt.c +36 -20
119 lines not on this page · raw message
tools/perf/util/data-convert.h +1 -0
@@ -5,6 +5,7 @@struct perf_data_convert_opts {bool force;bool all; +bool tod; };#endif /* __DATA_CONVERT_H */
Signature · 1 line
2.25.4
Namhyung Kim
maintainerRe: [PATCH 4/6] perf tools: Add support to store time of day in CTF data conversion
Quoted 20 lines · Jiri Olsa
Adding support to convert and store time of day in CTF data conversion for 'perf data convert' subcommand.
The perf.data used for conversion needs to have clock data information - must be recorded with -k/--clockid option).
New --tod option is added to 'perf data convert' subcommand to convert data with timestamps converted to wall clock time.
Record data with clockid set:
# perf record -k CLOCK_MONOTONIC kill kill: not enough arguments [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.033 MB perf.data (8 samples) ]Convert data with TOD timestamps:
# perf data convert --tod --to-ctf ./ctf [ perf data convert: Converted 'perf.data' into CTF data './ctf' ] [ perf data convert: Converted and wrote 0.000 MB (8 samples) ]
Quoted 6 lines
Display data in perf script:
# perf script -F+tod --ns perf 262150 2020-07-13 18:38:50.097678523 153633.958246159: 1 cycles: ... perf 262150 2020-07-13 18:38:50.097682941 153633.958250577: 1 cycles: ... perf 262150 2020-07-13 18:38:50.097684997 153633.958252633: 7 cycles: ... ...
I believe this belongs to a later patch.
Thanks
Namhyung
Quoted 27 lines
Display data in babeltrace:
# babeltrace --clock-date ./ctf [2020-07-13 18:38:50.097678523] (+?.?????????) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ... [2020-07-13 18:38:50.097682941] (+0.000004418) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ... [2020-07-13 18:38:50.097684997] (+0.000002056) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ... ...It's available only for recording with clockid specified, because it's the only case where we can get reference time to wallclock time. It's can't do that with perf clock yet.
Error is display if you want to use --tod on data without clockid specified:
# perf data convert --tod --to-ctf ./ctf Can't provide --tod time, missing clock data. Please record with -k/--clockid option. Failed to setup CTF writer. Error during conversion setup.Signed-off-by Jiri Olsa <[email protected]>--- tools/perf/Documentation/perf-data.txt | 3 ++ tools/perf/builtin-data.c | 1 + tools/perf/util/data-convert-bt.c | 56 +++++++++++++++++--------- tools/perf/util/data-convert.h | 1 + 4 files changed, 41 insertions(+), 20 deletions(-)
Re: [PATCH 4/6] perf tools: Add support to store time of day in CTF data conversion
Quoted 21 lines · Namhyung Kim
Quoted 20 lines · Jiri Olsa
Adding support to convert and store time of day in CTF data conversion for 'perf data convert' subcommand.
The perf.data used for conversion needs to have clock data information - must be recorded with -k/--clockid option).
New --tod option is added to 'perf data convert' subcommand to convert data with timestamps converted to wall clock time.
Record data with clockid set:
# perf record -k CLOCK_MONOTONIC kill kill: not enough arguments [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.033 MB perf.data (8 samples) ]Convert data with TOD timestamps:
# perf data convert --tod --to-ctf ./ctf [ perf data convert: Converted 'perf.data' into CTF data './ctf' ] [ perf data convert: Converted and wrote 0.000 MB (8 samples) ]
Quoted 8 lines
Quoted 6 lines
Display data in perf script:
# perf script -F+tod --ns perf 262150 2020-07-13 18:38:50.097678523 153633.958246159: 1 cycles: ... perf 262150 2020-07-13 18:38:50.097682941 153633.958250577: 1 cycles: ... perf 262150 2020-07-13 18:38:50.097684997 153633.958252633: 7 cycles: ... ...I believe this belongs to a later patch.
I wanted to show in changelog that the timestamps match for both script and data convert tools
jirka
Quoted 32 lines
Thanks
NamhyungQuoted 27 lines
Display data in babeltrace:
# babeltrace --clock-date ./ctf [2020-07-13 18:38:50.097678523] (+?.?????????) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ... [2020-07-13 18:38:50.097682941] (+0.000004418) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ... [2020-07-13 18:38:50.097684997] (+0.000002056) cycles: { cpu_id = 0 }, { perf_ip = 0xFFF ... ...It's available only for recording with clockid specified, because it's the only case where we can get reference time to wallclock time. It's can't do that with perf clock yet.
Error is display if you want to use --tod on data without clockid specified:
# perf data convert --tod --to-ctf ./ctf Can't provide --tod time, missing clock data. Please record with -k/--clockid option. Failed to setup CTF writer. Error during conversion setup.Signed-off-by Jiri Olsa <[email protected]>--- tools/perf/Documentation/perf-data.txt | 3 ++ tools/perf/builtin-data.c | 1 + tools/perf/util/data-convert-bt.c | 56 +++++++++++++++++--------- tools/perf/util/data-convert.h | 1 + 4 files changed, 41 insertions(+), 20 deletions(-)
Jiri Olsa
reviewerPATCH 5/6 perf script: Change enum perf_output_field values to be 64 bits
In reply to Jiri OlsaSo it's possible to add new values. I did not find any place where the enum values are passed through some number type, so it's safe to make this change.
--- tools/perf/builtin-script.c | 64 ++++++++++++++++++------------------- 1 file changed, 32 insertions(+), 32 deletions(-)
tools/perf/builtin-script.c +32 -32
@@ -82,38 +82,38 @@ static bool native_arch;unsigned int scripting_max_stack =PERF_MAX_STACK_DEPTH ;enum perf_output_field { -PERF_OUTPUT_COMM =1U <<0 , -PERF_OUTPUT_TID =1U <<1 , -PERF_OUTPUT_PID =1U <<2 , -PERF_OUTPUT_TIME =1U <<3 , -PERF_OUTPUT_CPU =1U <<4 , -PERF_OUTPUT_EVNAME =1U <<5 , -PERF_OUTPUT_TRACE =1U <<6 , -PERF_OUTPUT_IP =1U <<7 , -PERF_OUTPUT_SYM =1U <<8 , -PERF_OUTPUT_DSO =1U <<9 , -PERF_OUTPUT_ADDR =1U <<10 , -PERF_OUTPUT_SYMOFFSET =1U <<11 , -PERF_OUTPUT_SRCLINE =1U <<12 , -PERF_OUTPUT_PERIOD =1U <<13 , -PERF_OUTPUT_IREGS =1U <<14 , -PERF_OUTPUT_BRSTACK =1U <<15 , -PERF_OUTPUT_BRSTACKSYM =1U <<16 , -PERF_OUTPUT_DATA_SRC =1U <<17 , -PERF_OUTPUT_WEIGHT =1U <<18 , -PERF_OUTPUT_BPF_OUTPUT =1U <<19 , -PERF_OUTPUT_CALLINDENT =1U <<20 , -PERF_OUTPUT_INSN =1U <<21 , -PERF_OUTPUT_INSNLEN =1U <<22 , -PERF_OUTPUT_BRSTACKINSN =1U <<23 , -PERF_OUTPUT_BRSTACKOFF =1U <<24 , -PERF_OUTPUT_SYNTH =1U <<25 , -PERF_OUTPUT_PHYS_ADDR =1U <<26 , -PERF_OUTPUT_UREGS =1U <<27 , -PERF_OUTPUT_METRIC =1U <<28 , -PERF_OUTPUT_MISC =1U <<29 , -PERF_OUTPUT_SRCCODE =1U <<30 , -PERF_OUTPUT_IPC =1U <<31 , +PERF_OUTPUT_COMM =1ULL <<0 , +PERF_OUTPUT_TID =1ULL <<1 , +PERF_OUTPUT_PID =1ULL <<2 , +PERF_OUTPUT_TIME =1ULL <<3 , +PERF_OUTPUT_CPU =1ULL <<4 , +PERF_OUTPUT_EVNAME =1ULL <<5 , +PERF_OUTPUT_TRACE =1ULL <<6 , +PERF_OUTPUT_IP =1ULL <<7 , +PERF_OUTPUT_SYM =1ULL <<8 , +PERF_OUTPUT_DSO =1ULL <<9 , +PERF_OUTPUT_ADDR =1ULL <<10 , +PERF_OUTPUT_SYMOFFSET =1ULL <<11 , +PERF_OUTPUT_SRCLINE =1ULL <<12 , +PERF_OUTPUT_PERIOD =1ULL <<13 , +PERF_OUTPUT_IREGS =1ULL <<14 , +PERF_OUTPUT_BRSTACK =1ULL <<15 , +PERF_OUTPUT_BRSTACKSYM =1ULL <<16 , +PERF_OUTPUT_DATA_SRC =1ULL <<17 , +PERF_OUTPUT_WEIGHT =1ULL <<18 , +PERF_OUTPUT_BPF_OUTPUT =1ULL <<19 , +PERF_OUTPUT_CALLINDENT =1ULL <<20 , +PERF_OUTPUT_INSN =1ULL <<21 , +PERF_OUTPUT_INSNLEN =1ULL <<22 , +PERF_OUTPUT_BRSTACKINSN =1ULL <<23 , +PERF_OUTPUT_BRSTACKOFF =1ULL <<24 , +PERF_OUTPUT_SYNTH =1ULL <<25 , +PERF_OUTPUT_PHYS_ADDR =1ULL <<26 , +PERF_OUTPUT_UREGS =1ULL <<27 , +PERF_OUTPUT_METRIC =1ULL <<28 , +PERF_OUTPUT_MISC =1ULL <<29 , +PERF_OUTPUT_SRCCODE =1ULL <<30 , +PERF_OUTPUT_IPC =1ULL <<31 , };struct output_option {
Signature · 1 line
2.25.4
Jiri Olsa
reviewerPATCH 6/6 perf script: Add tod field to display time of day
In reply to Jiri OlsaAdding tod field to display time of day column with time of date (wallclock) time.
# perf record -k CLOCK_MONOTONIC kill kill: not enough arguments [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.033 MB perf.data (8 samples) ]
# perf script
perf 261340 152919.481538: 1 cycles: ffffffff8106d104 ...
perf 261340 152919.481543: 1 cycles: ffffffff8106d104 ...
perf 261340 152919.481545: 7 cycles: ffffffff8106d104 ...
... # perf script --ns
perf 261340 152919.481538922: 1 cycles: ffffffff8106d ...
perf 261340 152919.481543286: 1 cycles: ffffffff8106d ...
perf 261340 152919.481545397: 7 cycles: ffffffff8106d ...
... # perf script -F+tod
perf 261340 2020-07-13 18:26:55.620971 152919.481538: ...
perf 261340 2020-07-13 18:26:55.620975 152919.481543: ...
perf 261340 2020-07-13 18:26:55.620978 152919.481545: ...
... # perf script -F+tod --ns
perf 261340 2020-07-13 18:26:55.620971621 152919.481538922: ...
perf 261340 2020-07-13 18:26:55.620975985 152919.481543286: ...
perf 261340 2020-07-13 18:26:55.620978096 152919.481545397: ...
...It's available only for recording with clockid specified, because it's the only case where we can get reference time to wallclock time. It's can't do that with perf clock yet.
Error is display if you want to use --tod on data without clockid specified:
# perf script -F+tod Can't provide 'tod' time, missing clock data. Please record with -k/--clockid option.
--- tools/perf/builtin-script.c | 131 +++++++++++++++++++++++++++--------- 1 file changed, 98 insertions(+), 33 deletions(-)
tools/perf/builtin-script.c +98 -33
@@ -114,6 +114,32 @@ enum perf_output_field {PERF_OUTPUT_MISC =1ULL <<29 ,PERF_OUTPUT_SRCCODE =1ULL <<30 ,PERF_OUTPUT_IPC =1ULL <<31 , + PERF_OUTPUT_TOD = 1ULL << 32, +}; + +struct perf_script { +struct perf_tool tool; +struct perf_session *session; +bool show_task_events; +bool show_mmap_events; +bool show_switch_events; +bool show_namespace_events; +bool show_lost_events; +bool show_round_events; +bool show_bpf_events; +bool show_cgroup_events; +bool show_text_poke_events; +bool allocated; +bool per_event_dump; +bool stitch_lbr; +struct evswitch evswitch; +struct perf_cpu_map *cpus; +struct perf_thread_map *threads; +int name_width; +const char *time_str; +struct perf_time_interval *ptime_range; +int range_size; +int range_num; }; struct output_option {
@@ -152,6 +178,7 @@ struct output_option { {.str = "misc", .field = PERF_OUTPUT_MISC}, {.str = "srccode", .field = PERF_OUTPUT_SRCCODE}, {.str = "ipc", .field = PERF_OUTPUT_IPC}, + {.str = "tod", .field = PERF_OUTPUT_TOD}, }; enum {
@@ -502,6 +529,7 @@ static void set_print_ip_opts(struct perf_event_attr *attr)*/ static int perf_session__check_output_opt (struct perf_session *session) { +bool tod = false;unsigned int j;struct evsel *evsel;
@@ -528,6 +556,7 @@ static int perf_session__check_output_opt(struct perf_session *session)continue ;set_print_ip_opts (&evsel->core.attr); + tod |= output[j].fields &PERF_OUTPUT_TOD ; }if (!no_callchain) {
@@ -568,13 +597,17 @@ static int perf_session__check_output_opt(struct perf_session *session) } } +if (tod && !session->header.env.clock.enabled) { +pr_err ("Can't provide 'tod' time, missing clock data. " +"Please record with -k/--clockid option.\n" ); +return -1 ; + } out:return 0 ; }static int perf_sample__fprintf_regs (struct regs_dump *regs,uint64_t mask, -FILE *fp -) +FILE *fp) {unsigned i =0 , r;int printed =0 ;
@@ -592,6 +625,56 @@ static int perf_sample__fprintf_regs(struct regs_dump *regs, uint64_t mask,return printed; } +#define DEFAULT_TOD_FMT "%F %H:%M:%S" + +static char * +tod_scnprintf (struct perf_script *script,char *buf,int buflen, +u64 timestamp) +{ +u64 tod_ns, clockid_ns; +struct perf_env *env; +unsigned long nsec; +struct tm ltime; +char date[64 ]; +time_t sec; + + buf[0 ] ='\0' ; +if (buflen <64 || !script) +return buf; + + env = &script->session->header.env; +if (!env->clock.enabled) { +scnprintf (buf, buflen,"disabled" ); +return buf; + } + + clockid_ns = env->clock.clockid_ns; + tod_ns = env->clock.tod_ns; + +if (timestamp > clockid_ns) + tod_ns += timestamp - clockid_ns; +else + tod_ns -= clockid_ns - timestamp; + + sec = (time_t) (tod_ns /NSEC_PER_SEC ); + nsec = tod_ns - sec *NSEC_PER_SEC ; + +if (localtime_r (&sec, <ime) ==NULL ) { +scnprintf (buf, buflen,"failed" ); + }else { +strftime (date,sizeof (date),DEFAULT_TOD_FMT , <ime); + +if (symbol_conf.nanosecs) { +snprintf (buf, buflen,"%s.%09lu" , date, nsec); + }else { +snprintf (buf, buflen,"%s.%06lu" , + date, nsec /NSEC_PER_USEC ); + } + } + +return buf; +} +static int perf_sample__fprintf_iregs (struct perf_sample *sample,struct perf_event_attr *attr,FILE *fp) {
@@ -606,7 +689,8 @@ static int perf_sample__fprintf_uregs(struct perf_sample *sample, attr->sample_regs_user, fp); } -static int perf_sample__fprintf_start (struct perf_sample *sample, +static int perf_sample__fprintf_start (struct perf_script *script, +struct perf_sample *sample,struct thread *thread,struct evsel *evsel,u32 type,FILE *fp)
@@ -615,6 +699,7 @@ static int perf_sample__fprintf_start(struct perf_sample *sample,unsigned long secs;unsigned long long nsecs;int printed =0 ; +char tstr[128 ];if (PRINT_FIELD (COMM )) {if (latency_format)
@@ -683,6 +768,11 @@ static int perf_sample__fprintf_start(struct perf_sample *sample, printed += ret; } +if (PRINT_FIELD (TOD )) { +tod_scnprintf (script, tstr,sizeof (tstr), sample->time); + printed +=fprintf (fp,"%s " , tstr); + } +if (PRINT_FIELD (TIME )) {u64 t = sample->time;if (reltime) {
@@ -1667,31 +1757,6 @@ static int perf_sample__fprintf_synth(struct perf_sample *sample,return 0 ; } -struct perf_script { -struct perf_tool tool; -struct perf_session *session; -bool show_task_events; -bool show_mmap_events; -bool show_switch_events; -bool show_namespace_events; -bool show_lost_events; -bool show_round_events; -bool show_bpf_events; -bool show_cgroup_events; -bool show_text_poke_events; -bool allocated; -bool per_event_dump; -bool stitch_lbr; -struct evswitch evswitch; -struct perf_cpu_map *cpus; -struct perf_thread_map *threads; -int name_width; -const char *time_str; -struct perf_time_interval *ptime_range; -int range_size; -int range_num; -}; -static int evlist__max_name_len (struct evlist *evlist) {struct evsel *evsel;
@@ -1739,7 +1804,7 @@ static void script_print_metric(struct perf_stat_config *config __maybe_unused,if (!fmt)return ; -perf_sample__fprintf_start (mctx->sample, mctx->thread, mctx->evsel, +perf_sample__fprintf_start (NULL , mctx->sample, mctx->thread, mctx->evsel,PERF_RECORD_SAMPLE , mctx->fp);fputs ("\tmetric: " , mctx->fp);if (color)
@@ -1754,7 +1819,7 @@ static void script_new_line(struct perf_stat_config *config __maybe_unused, {struct metric_ctx *mctx = ctx; -perf_sample__fprintf_start (mctx->sample, mctx->thread, mctx->evsel, +perf_sample__fprintf_start (NULL , mctx->sample, mctx->thread, mctx->evsel,PERF_RECORD_SAMPLE , mctx->fp);fputs ("\tmetric: " , mctx->fp); }
@@ -1865,7 +1930,7 @@ static void process_event(struct perf_script *script, ++es->samples; -perf_sample__fprintf_start (sample, thread, evsel, +perf_sample__fprintf_start (script, sample, thread, evsel,PERF_RECORD_SAMPLE , fp);if (PRINT_FIELD (PERIOD ))
@@ -2174,7 +2239,7 @@ static int print_event_with_time(struct perf_tool *tool, thread =machine__findnew_thread (machine, pid, tid);if (thread && evsel) { -perf_sample__fprintf_start (sample, thread, evsel, +perf_sample__fprintf_start (script, sample, thread, evsel, event->header.type, stdout); }
@@ -3439,7 +3504,7 @@ int cmd_script(int argc, const char **argv) "Fields: comm,tid,pid,time,cpu,event,trace,ip,sym,dso," "addr,symoff,srcline,period,iregs,uregs,brstack," "brstacksym,flags,bpf-output,brstackinsn,brstackoff," - "callindent,insn,insnlen,synth,phys_addr,metric,misc,ipc", + "callindent,insn,insnlen,synth,phys_addr,metric,misc,ipc,tod", parse_output_fields),OPT_BOOLEAN ('a' ,"all-cpus" , &system_wide, "system-wide collection from all CPUs"),
Signature · 1 line
2.25.4
[email protected]
maintainerQuoted 3 lines · Jiri Olsa
The patchset is adding the ability to display TOD/wallclock timestamp in 'perf script' output and in 'perf data convert --to-ctf' subcommand, so the converted CTF data contain TOD/wallclock timestamps.
But why? Wallclock is a horrible piece of crap. Why would you want to do this?
Quoted 9 lines · On 7/30/20 4:14 PM, [email protected] wrote:
Quoted 3 lines · Jiri Olsa
The patchset is adding the ability to display TOD/wallclock timestamp in 'perf script' output and in 'perf data convert --to-ctf' subcommand, so the converted CTF data contain TOD/wallclock timestamps.
But why? Wallclock is a horrible piece of crap. Why would you want to do this?
Same reason I brought this up 9+ years ago: userspace lives on time-of-day, and troubleshooting is based on correlating timestamps from multiple sources. To correlate a perf event to syslog or an application log, we need time-of-day.
Quoted 7 lines · David Ahern
Quoted 6 lines · On 7/30/20 4:14 PM, [email protected] wrote:
Quoted 3 lines · Jiri Olsa
The patchset is adding the ability to display TOD/wallclock timestamp in 'perf script' output and in 'perf data convert --to-ctf' subcommand, so the converted CTF data contain TOD/wallclock timestamps.
Quoted 8 lines
Quoted 3 lines
But why? Wallclock is a horrible piece of crap. Why would you want to do this?
Same reason I brought this up 9+ years ago: userspace lives on time-of-day, and troubleshooting is based on correlating timestamps from multiple sources. To correlate a perf event to syslog or an application log, we need time-of-day.
yep, we have a customer that needs to compare data from multiple servers
jirka
Quoted 1 line
yep, we have a customer that needs to compare data from multiple servers
It's also needed to correlate over different guests on the same machine. This is an important use case.
It would be nice if we could find a way to export the per guest TSC offset from KVM, then we could use the more reliable TSC for this, at least as long as no TSC scaling is used.
But failing that wall clock is probably the best you can do.
-Andi
[email protected]
maintainerQuoted 4 lines · Andi Kleen
Quoted 1 line
yep, we have a customer that needs to compare data from multiple servers
It's also needed to correlate over different guests on the same machine. This is an important use case.
Both these cases you want to sync up CLOCK_MONOTONIC, using walltime is just utterly misguided.
What happens if the servers have (per accident or otherwise) different DST settings, or someone does a clock_setttime() for giggles.
All you really want is a clock that runs at the same rate but is not subject to random jumps and user foibles.
Quoted 8 lines · On 7/31/20 12:05 PM, [email protected] wrote:
Quoted 4 lines · Andi Kleen
Quoted 1 line
yep, we have a customer that needs to compare data from multiple servers
It's also needed to correlate over different guests on the same machine. This is an important use case.
Both these cases you want to sync up
CLOCK_MONOTONIC, using walltime is just utterly misguided.
Every userspace component logs in walltime. You can say that is misguided, but that is the way it is. The missing piece is the ability to correlate kernel events to userspace logs.
Quoted 3 lines
What happens if the servers have (per accident or otherwise) different DST settings, or someone does a
clock_setttime()for giggles.
Yes, someone *could* change the time. Someone *could* start ntpd or other time server in the middle of a session. While technically such things can happen, that is not real life in most environments (e.g., Data center servers). ntpd (or other) is started at boot, and it is just the little misc adjustments that happen over time.
We could add tracepoints and detect the changes and invalidate the reference time. We could add tracepoints to track the adjustments and update the reference time. In my experience over 9+ years using this tool (out of tree patches) that has never been the problem.
Quoted 4 lines
All you really want is a clock that runs at the same rate but is not subject to random jumps and user foibles.
All I want is to compare user logs to a kernel event via timestamps.
Arnaldo Carvalho de Melo
maintainerEm Fri, Jul 31, 2020 at 06:46:21PM -0600, David Ahern escreveu:
Quoted 28 lines · On 7/31/20 12:05 PM, [email protected] wrote:
Quoted 8 lines · Andi Kleen
Quoted 4 lines
Quoted 1 line
yep, we have a customer that needs to compare data from multiple servers
It's also needed to correlate over different guests on the same machine. This is an important use case.
Both these cases you want to sync up
CLOCK_MONOTONIC, using walltime is just utterly misguided.Every userspace component logs in walltime. You can say that is misguided, but that is the way it is. The missing piece is the ability to correlate kernel events to userspace logs.
Quoted 2 lines
What happens if the servers have (per accident or otherwise) different DST settings, or someone does a
clock_setttime()for giggles.Yes, someone *could* change the time. Someone *could* start ntpd or other time server in the middle of a session. While technically such things can happen, that is not real life in most environments (e.g., Data center servers). ntpd (or other) is started at boot, and it is just the little misc adjustments that happen over time.
We could add tracepoints and detect the changes and invalidate the reference time. We could add tracepoints to track the adjustments and update the reference time. In my experience over 9+ years using this tool (out of tree patches) that has never been the problem.
Quoted 4 lines
Quoted 2 lines
All you really want is a clock that runs at the same rate but is not subject to random jumps and user foibles.
All I want is to compare user logs to a kernel event via timestamps.
Can we have both possibilities and leave the decision on which one to use in the hands of those who have a gun to shoot wherever they want, maybe in the foot?
- Arnaldo
[email protected]
maintainerQuoted 7 lines · David Ahern
Quoted 6 lines · On 7/30/20 4:14 PM, [email protected] wrote:
Quoted 3 lines · Jiri Olsa
The patchset is adding the ability to display TOD/wallclock timestamp in 'perf script' output and in 'perf data convert --to-ctf' subcommand, so the converted CTF data contain TOD/wallclock timestamps.
Quoted 8 lines
Quoted 3 lines
But why? Wallclock is a horrible piece of crap. Why would you want to do this?
Same reason I brought this up 9+ years ago: userspace lives on time-of-day, and troubleshooting is based on correlating timestamps from multiple sources. To correlate a perf event to syslog or an application log, we need time-of-day.
You need a sync'ed CLOCK_MONOTONIC for that, CLOCK_REALTIME 'aka' wallclock is a trainwreck, you don't ever want that.