Freq/CPU%/CORE_BUSY% support
v9 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.
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.
Here is an example:
$ perf record --freq-perf ~/tchain_edit
$ perf report --stdio --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_tscChanges since V1:
- Save cpu max freq to header when recording
- Read cpu max freq and msr type from header when reporting
Changes since V2:
- Introduce generic FEAT for CPU related data stored
- Make cpu max freq and msr type part of perf_session_env
- rename cpu_u to cpu_util
- Don't save sample value in perf_sample and discards new iterator. Calculating the freq_perf_info in add_entry_cb callback
- Introduce symbol_conf.freq_perf_type for related hpp column visibility
Changes since V3:
- add a identifier 'tag' for CPU attributes, max frequency.
- add backpointers to evlist for env, and evsel for evlist.
- Use bitmask for freq_perf_type
- Replace macros by functions to caculate freq, cpu_util and core_busy
- Move all caculation codes under symbol_conf.show_freq_perf condition.
Changes since V4:
- Store cpu attributes id as tag and more readable cpu_attr
Changes since V5:
- Rename freq to max_freq and use it
- Add a loop in process_cpu_attributes to facility future extension
Changes since V6:
- Split rename perf_session_env and add backpointer to evlist patches
Changes since V7:
- Introduce --freq-perf option for perf record
- Factor out fix for dump_sample
Changes since V8:
- Rename --show-freq-perf to --freq-perf
- --freq-perf option imply --group
Kan Liang (6):
perf,tools: introduce generic FEAT for CPU attributes perf,tools: read msr pmu type from header.
perf, record: introduce --freq-perf option
perf,tools: Dump per-sample freq/CPU%/CORE_BUSY% in report -D perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat perf,tools: Show freq/CPU%/CORE_BUSY% in perf report by --freq-perf
tools/perf/Documentation/perf-record.txt | 4 ++ tools/perf/Documentation/perf-report.txt | 12 ++++++ tools/perf/builtin-record.c | 39 +++++++++++++++++- tools/perf/builtin-report.c | 59 ++++++++++++++++++++++++++ tools/perf/ui/hist.c | 71 +++++++++++++++++++++++++++++--- tools/perf/util/cpumap.c | 32 ++++++++++++++ tools/perf/util/cpumap.h | 1 + tools/perf/util/header.c | 62 ++++++++++++++++++++++++++++ tools/perf/util/header.h | 13 ++++++ tools/perf/util/hist.h | 3 ++ tools/perf/util/session.c | 31 ++++++++++++-- tools/perf/util/session.h | 64 ++++++++++++++++++++++++++++ tools/perf/util/sort.c | 3 ++ tools/perf/util/sort.h | 3 ++ tools/perf/util/symbol.h | 12 +++++- 15 files changed, 399 insertions(+), 10 deletions(-)
Signature · 1 line
1.8.3.1
PATCH v9 1/6 perf,tools: introduce generic FEAT for CPU attributes
From: Kan Liang <[email protected]>
This patch introduces generic FEAT for CPU attributes. For the patch set, we only need cpu max frequency. But it can be easily extented to support more other CPU attributes.
The cpu max frequency is from the first online cpu.
--- tools/perf/util/cpumap.c | 32 ++++++++++++++++++++++++++ tools/perf/util/cpumap.h | 1 + tools/perf/util/header.c | 59 ++++++++++++++++++++++++++++++++++++++++++++++++ tools/perf/util/header.h | 12 ++++++++++ 4 files changed, 104 insertions(+)
tools/perf/util/cpumap.c +32 -0
@@ -514,3 +514,35 @@ int cpu__setup_cpunode_map(void)closedir (dir1);return 0 ; } + +u64 get_cpu_max_freq (void ) +{ +const char *mnt; +char path[PATH_MAX ], tmp; +FILE *fp; +u64 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,"%lu" , &freq); +fclose (fp); + +return (ret ==1 ) ? freq :0 ; +}
tools/perf/util/cpumap.h +1 -0
@@ -60,6 +60,7 @@ int max_node_num;int *cpunode_map;int cpu__setup_cpunode_map (void ); +u64 get_cpu_max_freq (void );static inline int cpu__max_node (void ) {
tools/perf/util/header.c +59 -0
83 lines not on this page · raw message
tools/perf/util/header.h +12 -0
@@ -31,6 +31,7 @@ enum {HEADER_PMU_MAPPINGS ,HEADER_GROUP_DESC ,HEADER_AUXTRACE , +HEADER_CPU_ATTR ,HEADER_LAST_FEATURE , HEADER_FEAT_BITS = 256, };
@@ -71,6 +72,11 @@ struct cpu_topology_map {int core_id; }; +enum perf_header_cpu_attr { +PERF_HEADER_CPU_MAX_FREQ =0 , +PERF_HEADER_CPU_ATTR_MAX , +}; +struct perf_env {char *hostname;char *os_release;
@@ -95,6 +101,12 @@ struct perf_env {char *numa_nodes;char *pmu_mappings;struct cpu_topology_map *cpu; +union { +u64 cpu_attr[PERF_HEADER_CPU_ATTR_MAX ]; +struct { +u64 max_freq; + } cpuattr; + }; };struct perf_header {
Signature · 1 line
1.8.3.1
Arnaldo Carvalho de Melo
maintainerRe: [PATCH V9 1/6] perf,tools: introduce generic FEAT for CPU attributes
Em Tue, Sep 08, 2015 at 03:32:44PM -0400, [email protected] escreveu:
Quoted 6 lines
From: Kan Liang <[email protected]>
This patch introduces generic FEAT for CPU attributes. For the patch set, we only need cpu max frequency. But it can be easily extented to support more other CPU attributes.
The cpu max frequency is from the first online cpu.
Ok, but don't we have to do error handling? i.e. you are returning 0 for any error in trying to read the cpu max freq, shouldn't we bail out somewhere?
And please move this get_cpu_max_freq() thing out of the cpumap.[ch] files, it is not even a need completely specific to perf tooling, there must be somewhere in tools/lib/api/ (kernel APIs) where this fits, no?
More comments below.
- Arnaldo
Quoted 20 lines
Signed-off-by Kan Liang <[email protected]>Acked-by Jiri Olsa <[email protected]>--- tools/perf/util/cpumap.c | 32 ++++++++++++++++++++++++++ tools/perf/util/cpumap.h | 1 + tools/perf/util/header.c | 59 ++++++++++++++++++++++++++++++++++++++++++++++++ tools/perf/util/header.h | 12 ++++++++++ 4 files changed, 104 insertions(+)tools/perf/util/cpumap.c +3
-0@@ -514,3 +514,35 @@ int cpu__setup_cpunode_map(void)closedir (dir1);return 0 ; } + +u64 get_cpu_max_freq (void ) +{
tools/perf/util/cpumap.c int cpu__setup_cpunode_map(void) · line 529
Quoted 10 lines
tools/perf/util/cpumap.c +10
-0+const char *mnt; +char path[PATH_MAX ], tmp; +FILE *fp; +u64 freq; +int cpu =0 ; +int ret; + + mnt =sysfs__mountpoint (); +if (!mnt) +return 0 ;
See? Can't find the sysfs mount point? No problem, return 0 max freq.
tools/perf/util/cpumap.c int cpu__setup_cpunode_map(void) · line 537
Quoted 8 lines
tools/perf/util/cpumap.c +8
-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 ;
Can't parse it? 0
tools/perf/util/cpumap.c int cpu__setup_cpunode_map(void) · line 543
Quoted 6 lines
tools/perf/util/cpumap.c +6
-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 ;
Ditto. Return soem error here please.
Quoted 29 lines
tools/perf/util/cpumap.c +5
-0+ ret =fscanf (fp,"%lu" , &freq); +fclose (fp); + +return (ret ==1 ) ? freq :0 ; +}tools/perf/util/cpumap.h +1
-0@@ -60,6 +60,7 @@ int max_node_num;int *cpunode_map;int cpu__setup_cpunode_map (void ); +u64 get_cpu_max_freq (void );static inline int cpu__max_node (void ){
tools/perf/util/header.c +3
-0@@ -885,6 +885,23 @@ write_it:return do_write_string (fd, buffer); } +static int write_cpu_attributes (int fd,struct perf_header *h__maybe_unused , +struct perf_evlist *evlist__maybe_unused ) +{
tools/perf/util/header.c write_it: · line 900
Quoted 10 lines
tools/perf/util/header.c +10
-0+u32 tag_id; +u64 max_freq; +int ret; + + tag_id =PERF_HEADER_CPU_MAX_FREQ ; + ret =do_write (fd, &tag_id,sizeof (tag_id)); +if (ret <0 ) +return ret; + + max_freq =get_cpu_max_freq ();
Here, do the error handling, return whatever errno code this function returned, probably the feature code will use it to warn the user somehow.
Quoted 26 lines
tools/perf/util/header.c +12
-0+ +return do_write (fd, &max_freq,sizeof (max_freq)); +} +static int write_branch_stack (int fd__maybe_unused ,struct perf_header *h__maybe_unused ,struct perf_evlist *evlist__maybe_unused )@@ -1185,6 +1202,11 @@ 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_attributes (struct perf_header *ph,int fd__maybe_unused ,FILE *fp) +{ +fprintf (fp,"# CPU attributes: max frequency = %lu KHz\n" , ph->env.cpuattr.max_freq); +} +static void print_branch_stack (struct perf_header *ph__maybe_unused ,int fd__maybe_unused ,FILE *fp) {@@ -1498,6 +1520,42 @@ static int process_cpuid(struct perf_file_section *section __maybe_unused,return ph->env.cpuid ?0 : -ENOMEM ; } +static int process_cpu_attributes (struct perf_file_section *section__maybe_unused , +struct perf_header *ph,int fd, +void *data__maybe_unused )
tools/perf/util/header.c static int process_cpuid(struct perf_file_section *section __maybe_unused, · line 1535
Quoted 10 lines
tools/perf/util/header.c +10
-0+{ +ssize_t ret; +u32 i, tag_id; +u64 nr; + +for (i =0 ; i <PERF_HEADER_CPU_ATTR_MAX ; i++) { + + ret =readn (fd, &tag_id,sizeof (tag_id)); +if (ret !=sizeof (tag_id)) +return -1 ;
Return some sensible errno... Its not because perf_header__process_sections() doesn't check them that we shouldn't return ;-\ I'll fix perf_header__process_sections() to stop the process and warn the user if some error happens...
tools/perf/util/header.c static int process_cpuid(struct perf_file_section *section __maybe_unused, · line 1543
Quoted 8 lines
tools/perf/util/header.c +8
-0+ +if (ph->needs_swap) + nr =bswap_32 (tag_id); + +if (tag_id >=PERF_HEADER_CPU_ATTR_MAX ) { +pr_debug ("The number of cpu attributes is not expected. " +"You may need to upgrade the perf tool.\n" ); +return -1 ;
ditto
tools/perf/util/header.c static int process_cpuid(struct perf_file_section *section __maybe_unused, · line 1548
Quoted 5 lines
tools/perf/util/header.c +5
-0+ } + + ret =readn (fd, &nr,sizeof (nr)); +if (ret !=sizeof (nr)) +return -1 ;
ditto
Quoted 46 lines
tools/perf/util/header.c +11
-0+ +if (ph->needs_swap) + nr =bswap_64 (nr); + + ph->env.cpu_attr[tag_id] = 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 )@@ -1983,6 +2041,7 @@ static const struct feature_ops feat_ops[HEADER_LAST_FEATURE] = {FEAT_OPP (HEADER_PMU_MAPPINGS , pmu_mappings),FEAT_OPP (HEADER_GROUP_DESC , group_desc),FEAT_OPP (HEADER_AUXTRACE , auxtrace), + FEAT_OPP(HEADER_CPU_ATTR, cpu_attributes), }; struct header_print_data {tools/perf/util/header.h +6
-0@@ -31,6 +31,7 @@ enum {HEADER_PMU_MAPPINGS ,HEADER_GROUP_DESC ,HEADER_AUXTRACE , +HEADER_CPU_ATTR ,HEADER_LAST_FEATURE , HEADER_FEAT_BITS = 256, };@@ -71,6 +72,11 @@ struct cpu_topology_map {int core_id; }; +enum perf_header_cpu_attr { +PERF_HEADER_CPU_MAX_FREQ =0 , +PERF_HEADER_CPU_ATTR_MAX , +}; +struct perf_env {char *hostname;char *os_release;
tools/perf/util/header.h struct perf_env { · line 109
Quoted 9 lines
tools/perf/util/header.h +6
-0@@ -95,6 +101,12 @@ struct perf_env {char *numa_nodes;char *pmu_mappings;struct cpu_topology_map *cpu; +union { +u64 cpu_attr[PERF_HEADER_CPU_ATTR_MAX ]; +struct { +u64 max_freq; + } cpuattr; + };
Ok, these need moving to env.h, but lets first get that patchkit merged... Looking at your other patches
Quoted 5 lines
tools/perf/util/header.h +0
-0};struct perf_header {Signature · 1 line
1.8.3.1
Arnaldo Carvalho de Melo
maintainerRe: [PATCH V9 1/6] perf,tools: introduce generic FEAT for CPU attributes
Em Thu, Sep 10, 2015 at 10:58:38AM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 6 lines
Em Tue, Sep 08, 2015 at 03:32:44PM -0400, [email protected] escreveu:
Quoted 4 lines
This patch introduces generic FEAT for CPU attributes. For the patch set, we only need cpu max frequency. But it can be easily extented to support more other CPU attributes.
The cpu max frequency is from the first online cpu.
Quoted 7 lines
Ok, but don't we have to do error handling? i.e. you are returning 0 for any error in trying to read the cpu max freq, shouldn't we bail out somewhere?
And please move this
get_cpu_max_freq()thing out of the cpumap.[ch] files, it is not even a need completely specific to perf tooling, there must be somewhere intools/lib/api/(kernel APIs) where this fits, no?
So, I've updated my perf/env branch with routines to do that, that uses infrastructure to read files from virtual dirs that was there plus a few I just introduced, in the same vein, I will update your patches to use them and put there as well, for your consideration.
The HEAD there is the one below:
- Arnaldo
commit 2bc1fae4ed8a842f52dc374449d37c3ec1fa1986
Author: Arnaldo Carvalho de Melo <[email protected]> Date: Thu Sep 10 12:20:14 2015 -0300
tools lib cpu: Introduce cpu.[ch] to read sysfs cpu related information
E.g.:
$ ./cpu__get_max_freq
3200000
It does that, as Kan's patch does, by looking at these files:
$ cat /sys/devices/system/cpu/online
0-3
$ ./sysfs__read_ull devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq=3200000 $
I.e. find out the first online CPU, then read its cpufreq info.
But do it in tools/lib/api/, so that other tools living code can use it, not just perf.
tools/lib/api/Build +1 -0
@@ -1,2 +1,3 @@ libapi-y += fd/ libapi-y += fs/ +libapi-y += cpu.o
tools/lib/api/cpu.c new file +18 -0
@@ -0,0 +1,18 @@ +#include <stdio.h> + +#include "cpu.h" +#include "fs/fs.h" + +int cpu__get_max_freq (unsigned long long *freq) +{ +char entry[PATH_MAX ]; +int cpu; + +if (sysfs__read_int ("devices/system/cpu/online" , &cpu) <0 ) +return -1 ; + +snprintf (entry,sizeof (entry), +"devices/system/cpu/cpu%d/cpufreq/cpuinfo_max_freq" , cpu); + +return sysfs__read_ull (entry, freq); +}
tools/lib/api/cpu.h new file +6 -0
@@ -0,0 +1,6 @@ +#ifndef __API_CPU__ +#define __API_CPU__ + +int cpu__get_max_freq (unsigned long long *freq); + +#endif /* __API_CPU__ */
Liang, Kan
Acked-byRE: [PATCH V9 1/6] perf,tools: introduce generic FEAT for CPU attributes
Quoted 13 lines
Em Thu, Sep 10, 2015 at 10:58:38AM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 9 lines
Em Tue, Sep 08, 2015 at 03:32:44PM -0400, [email protected] escreveu:
Quoted 4 lines
This patch introduces generic FEAT for CPU attributes. For the patch set, we only need cpu max frequency. But it can be easily extented to support more other CPU attributes.
The cpu max frequency is from the first online cpu.Ok, but don't we have to do error handling? i.e. you are returning 0 for any error in trying to read the cpu max freq, shouldn't we bail out somewhere?
Quoted 8 lines
Quoted 3 lines
And please move this
get_cpu_max_freq()thing out of the cpumap.[ch] files, it is not even a need completely specific to perf tooling, there must be somewhere intools/lib/api/(kernel APIs) where this fits,no?
So, I've updated my perf/env branch with routines to do that, that uses infrastructure to read files from virtual dirs that was there plus a few I just introduced, in the same vein
The two new patches in perf/env are good to me.
Commit 4ee5cc5708d89f380ab5371181b65dd74935352d
Commit 57d54be1e9074049c8695c522a499f8a7d62ef2d
Quoted 3 lines
I will update your patches to use them and
put there as well, for your consideration.
Thanks. Please let me know, when you finished.
I will update the rest of the per-sample freq patches, and repost them for review then.
Thanks,
Kan
Quoted 83 lines
The HEAD there is the one below:
- Arnaldo
commit
2bc1fae4ed8a842f52dc374449d37c3ec1fa1986Author: Arnaldo Carvalho de Melo <[email protected]> Date: Thu Sep 10 12:20:14 2015 -0300tools lib cpu: Introduce cpu.[ch] to read sysfs cpu related information
E.g.:
$ ./cpu__get_max_freq
3200000It does that, as Kan's patch does, by looking at these files:
$ cat
/sys/devices/system/cpu/online
0-3
$ ./sysfs__read_ull devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq=3200000 $I.e. find out the first online CPU, then read its cpufreq info.
But do it in
tools/lib/api/, so that other tools living code can use it, not just perf.Based-on-a-patch-by Kan Liang <[email protected]>Cc Adrian Hunter <[email protected]>Cc Borislav Petkov <[email protected]>Cc David Ahern <[email protected]>Cc Frederic Weisbecker <[email protected]>Cc Jiri Olsa <[email protected]>Cc Namhyung Kim <[email protected]>Cc Stephane Eranian <[email protected]>Signed-off-by Arnaldo Carvalho de Melo <[email protected]>tools/lib/api/Build index renamed from tools/lib/api/Build +0
-0
3653965cf481..e8b8a23b9bf4100644tools/lib/api/Build +1
-0@@ -1,2 +1,3 @@ libapi-y += fd/ libapi-y += fs/ +libapi-y += cpu.otools/lib/api/cpu.c new file +14
-0@@ -0,0 +1,18 @@ +#include <stdio.h> + +#include "cpu.h" +#include "fs/fs.h" + +int cpu__get_max_freq (unsigned long long *freq) { +char entry[PATH_MAX ]; +int cpu; + +if (sysfs__read_int ("devices/system/cpu/online" , &cpu) <0 ) +return -1 ; + + snprintf(entry, sizeof(entry), + "devices/system/cpu/cpu%d/cpufreq/cpuinfo_max_freq",cpu);
tools/lib/api/cpu.c new file +3
-0+ +return sysfs__read_ull (entry, freq); +}tools/lib/api/cpu.h new file +6
-0@@ -0,0 +1,6 @@ +#ifndef __API_CPU__ +#define __API_CPU__ + +int cpu__get_max_freq (unsigned long long *freq); + +#endif /* __API_CPU__ */
PATCH v9 2/6 perf,tools: read msr pmu type from header.
In reply to [email protected]From: Kan Liang <[email protected]>
Get msr pmu type when processing pmu_mappings
--- tools/perf/util/header.c | 3 +++ tools/perf/util/header.h | 1 + 2 files changed, 4 insertions(+)
tools/perf/util/header.c +3 -0
@@ -1879,6 +1879,9 @@ static int process_pmu_mappings(struct perf_file_section *section __maybe_unused/* include a NULL character at the end */ strbuf_add (&sb,"" ,1 ); +if (!strcmp (name,"msr" )) + ph->env.msr_pmu_type = type; +free (name); pmu_num--; }
tools/perf/util/header.h +1 -0
@@ -107,6 +107,7 @@ struct perf_env {u64 max_freq; } cpuattr; }; +unsigned int msr_pmu_type; };struct perf_header {
Signature · 1 line
1.8.3.1
Arnaldo Carvalho de Melo
maintainerRe: [PATCH V9 2/6] perf,tools: read msr pmu type from header.
Em Tue, Sep 08, 2015 at 03:32:45PM -0400, [email protected] escreveu:
Quoted 3 lines
From: Kan Liang <[email protected]>
Get msr pmu type when processing
pmu_mappings
This one looks ok, I am porting it to my current perf/core, that includes that perf/env branch, will fix up things as we discuss it.
- Arnaldo
Quoted 35 lines
Signed-off-by Kan Liang <[email protected]>Acked-by Jiri Olsa <[email protected]>--- tools/perf/util/header.c | 3 +++ tools/perf/util/header.h | 1 + 2 files changed, 4 insertions(+)tools/perf/util/header.c +3
-0@@ -1879,6 +1879,9 @@ static int process_pmu_mappings(struct perf_file_section *section __maybe_unused/* include a NULL character at the end */ strbuf_add (&sb,"" ,1 ); +if (!strcmp (name,"msr" )) + ph->env.msr_pmu_type = type; +free (name); pmu_num--; }tools/perf/util/header.h +1
-0@@ -107,6 +107,7 @@ struct perf_env {u64 max_freq; } cpuattr; }; +unsigned int msr_pmu_type; };struct perf_header {Signature · 1 line
1.8.3.1
Arnaldo Carvalho de Melo
maintainerRe: [PATCH V9 2/6] perf,tools: read msr pmu type from header.
In reply to [email protected]Em Tue, Sep 08, 2015 at 03:32:45PM -0400, [email protected] escreveu:
Quoted 6 lines
From: Kan Liang <[email protected]>
Get msr pmu type when processing
pmu_mappingsSigned-off-by Kan Liang <[email protected]>Acked-by Jiri Olsa <[email protected]>
This one is simple enough, merged already.
- Arnaldo
Quoted 33 lines
--- tools/perf/util/header.c | 3 +++ tools/perf/util/header.h | 1 + 2 files changed, 4 insertions(+)tools/perf/util/header.c +3
-0@@ -1879,6 +1879,9 @@ static int process_pmu_mappings(struct perf_file_section *section __maybe_unused/* include a NULL character at the end */ strbuf_add (&sb,"" ,1 ); +if (!strcmp (name,"msr" )) + ph->env.msr_pmu_type = type; +free (name); pmu_num--; }tools/perf/util/header.h +1
-0@@ -107,6 +107,7 @@ struct perf_env {u64 max_freq; } cpuattr; }; +unsigned int msr_pmu_type; };struct perf_header {Signature · 1 line
1.8.3.1
PATCH v9 3/6 perf, record: introduce --freq-perf option
In reply to [email protected]From: Kan Liang <[email protected]>
To generate the frequency and performance output, perf must sample read special events like cycles, ref-cycles, msr/tsc/, msr/aperf/ or msr/mperf/.
With the --freq-perf option, perf record can automatically check and add those event into evlist for sampling read.
--- tools/perf/Documentation/perf-record.txt | 4 ++++ tools/perf/builtin-record.c | 39 +++++++++++++++++++++++++++++++- tools/perf/util/session.h | 10 ++++++++ 3 files changed, 52 insertions(+), 1 deletion(-)
tools/perf/Documentation/perf-record.txt +4 -0
@@ -308,6 +308,10 @@ This option sets the time out limit. The default value is 500 ms. Record context switch events i.e. events of type PERF_RECORD_SWITCH or PERF_RECORD_SWITCH_CPU_WIDE. +--freq-perf:: +Add frequency and performance related events to do sample read. +These events include cycles, ref-cycles, msr/tsc/, msr/aperf/ and msr/mperf/. + SEE ALSO -------- linkperf:perf-stat[1], linkperf:perf-list[1]
tools/perf/builtin-record.c +38 -1
@@ -13,7 +13,7 @@#include "util/util.h" #include "util/parse-options.h" #include "util/parse-events.h" - +#include "util/pmu.h" #include "util/callchain.h" #include "util/cgroup.h" #include "util/header.h"
@@ -50,6 +50,7 @@ struct record {bool no_buildid;bool no_buildid_cache;long samples; +bool freq_perf; };static int record__write (struct record *rec,void *bf,size_t size)
@@ -948,6 +949,35 @@ out_free:return ret; } +const char *freq_perf_events[FREQ_PERF_MAX ][3 ] = { + {"msr" ,"tsc" ,"msr/tsc/" }, + {"msr" ,"aperf" ,"msr/aperf/" }, + {"msr" ,"mperf" ,"msr/mperf/" }, + {NULL ,"cycles" ,"cycles" }, + {NULL ,"ref-cycles" ,"ref-cycles" }, +}; + +static int +record_add_freq_perf_events (struct perf_evlist *evlist) +{ +int i; +char freq_perf_attrs[100 ]; + +strcpy (freq_perf_attrs,"{cycles,ref-cycles" ); +for (i =0 ; i <FREQ_PERF_MAX ; i++) { +if ((i ==FREQ_PERF_CYCLES ) || + (i ==FREQ_PERF_REF_CYCLES )) +continue ; +if (pmu_have_event (freq_perf_events[i][0 ], freq_perf_events[i][1 ])) { +strcat (freq_perf_attrs,"," ); +strcat (freq_perf_attrs, freq_perf_events[i][2 ]); + } + } +strcat (freq_perf_attrs,"}:S" ); + +return parse_events (evlist, freq_perf_attrs,NULL ); +} +static const char *const __record_usage[] = {"perf record [<options>] [<command>]" ,"perf record [<options>] -- <command> [<options>]" ,
@@ -1096,6 +1126,8 @@ struct option __record_options[] = { "per thread proc mmap processing timeout in ms"),OPT_BOOLEAN (0 ,"switch-events" , &record.opts.record_switch_events,"Record context switch events" ), +OPT_BOOLEAN (0 ,"freq-perf" , &record.freq_perf, +"Add frequency and performance related events to do sample read." ), OPT_END() };
@@ -1157,6 +1189,11 @@ int cmd_record(int argc, const char **argv, const char *prefix __maybe_unused)if (rec->no_buildid_cache || rec->no_buildid)disable_buildid_cache (); +if (rec->freq_perf &&record_add_freq_perf_events (rec->evlist)) { +pr_err ("Cannot set up freq and performance events\n" ); + goto out_symbol_exit; + } +if (rec->evlist->nr_entries ==0 &&perf_evlist__add_default (rec->evlist) <0 ) {pr_err ("Not enough memory for event selector list\n" );
tools/perf/util/session.h +10 -0
@@ -42,6 +42,16 @@ struct perf_session {#define PRINT_IP_OPT_ONELINE (1<<4)#define PRINT_IP_OPT_SRCLINE (1<<5) +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_tool ; struct perf_session *perf_session__new(struct perf_data_file *file,
Signature · 1 line
1.8.3.1
Re: [PATCH V9 3/6] perf, record: introduce --freq-perf option
Quoted 9 lines · On Tue, Sep 08, 2015 at 03:32:46PM -0400, [email protected] wrote:
From: Kan Liang <[email protected]>
To generate the frequency and performance output, perf must sample read special events like cycles, ref-cycles, msr/tsc/, msr/aperf/ or msr/mperf/.
With the --freq-perf option, perf record can automatically check and add those event into evlist for sampling read.Signed-off-by Kan Liang <[email protected]>
thanks,
jirka
Arnaldo Carvalho de Melo
maintainerRe: [PATCH V9 3/6] perf, record: introduce --freq-perf option
In reply to [email protected]Em Tue, Sep 08, 2015 at 03:32:46PM -0400, [email protected] escreveu:
Quoted 88 lines
From: Kan Liang <[email protected]>
To generate the frequency and performance output, perf must sample read special events like cycles, ref-cycles, msr/tsc/, msr/aperf/ or msr/mperf/.
With the --freq-perf option, perf record can automatically check and add those event into evlist for sampling read.Signed-off-by Kan Liang <[email protected]>--- tools/perf/Documentation/perf-record.txt | 4 ++++ tools/perf/builtin-record.c | 39 +++++++++++++++++++++++++++++++- tools/perf/util/session.h | 10 ++++++++ 3 files changed, 52 insertions(+), 1 deletion(-)tools/perf/Documentation/perf-record.txt +4
-0@@ -308,6 +308,10 @@ This option sets the time out limit. The default value is 500 ms. Record context switch events i.e. events of type PERF_RECORD_SWITCH or PERF_RECORD_SWITCH_CPU_WIDE. +--freq-perf:: +Add frequency and performance related events to do sample read. +These events include cycles, ref-cycles, msr/tsc/, msr/aperf/ and msr/mperf/. + SEE ALSO -------- linkperf:perf-stat[1], linkperf:perf-list[1]tools/perf/builtin-record.c +31
-1@@ -13,7 +13,7 @@#include "util/util.h" #include "util/parse-options.h" #include "util/parse-events.h" - +#include "util/pmu.h" #include "util/callchain.h" #include "util/cgroup.h" #include "util/header.h" @@ -50,6 +50,7 @@ struct record {bool no_buildid;bool no_buildid_cache;long samples; +bool freq_perf; };static int record__write (struct record *rec,void *bf,size_t size)@@ -948,6 +949,35 @@ out_free:return ret; } +const char *freq_perf_events[FREQ_PERF_MAX ][3 ] = { + {"msr" ,"tsc" ,"msr/tsc/" }, + {"msr" ,"aperf" ,"msr/aperf/" }, + {"msr" ,"mperf" ,"msr/mperf/" }, + {NULL ,"cycles" ,"cycles" }, + {NULL ,"ref-cycles" ,"ref-cycles" }, +}; + +static int +record_add_freq_perf_events (struct perf_evlist *evlist) +{ +int i; +char freq_perf_attrs[100 ]; + +strcpy (freq_perf_attrs,"{cycles,ref-cycles" ); +for (i =0 ; i <FREQ_PERF_MAX ; i++) { +if ((i ==FREQ_PERF_CYCLES ) || + (i ==FREQ_PERF_REF_CYCLES )) +continue ; +if (pmu_have_event (freq_perf_events[i][0 ], freq_perf_events[i][1 ])) { +strcat (freq_perf_attrs,"," ); +strcat (freq_perf_attrs, freq_perf_events[i][2 ]); + } + } +strcat (freq_perf_attrs,"}:S" ); + +return parse_events (evlist, freq_perf_attrs,NULL ); +} +static const char *const __record_usage[] = {"perf record [<options>] [<command>]" ,"perf record [<options>] -- <command> [<options>]" ,
tools/perf/builtin-record.c struct option __record_options[] = { · line 1130
Quoted 5 lines
tools/perf/builtin-record.c +2
-0@@ -1096,6 +1126,8 @@ struct option __record_options[] = { "per thread proc mmap processing timeout in ms"),OPT_BOOLEAN (0 ,"switch-events" , &record.opts.record_switch_events,"Record context switch events" ), +OPT_BOOLEAN (0 ,"freq-perf" , &record.freq_perf, + "Add frequency and performance related events to do sample read."),
Isn't this too vague? Can you try rewriting it? what do you mean
with "performance related events"?
You forgot to add the entry totools/perf/Documentation/perf-record.txt, where a less terse explanation could have helped me understand the part above, oops, sorry, you added it, but it was really terse! Can you elaborate a bit? Why would one want to use this? To achieve what? Don't assume everybody knows :-)
Quoted 20 lines
tools/perf/builtin-record.c +5
-0OPT_END() };@@ -1157,6 +1189,11 @@ int cmd_record(int argc, const char **argv, const char *prefix __maybe_unused)if (rec->no_buildid_cache || rec->no_buildid)disable_buildid_cache (); +if (rec->freq_perf &&record_add_freq_perf_events (rec->evlist)) { +pr_err ("Cannot set up freq and performance events\n" ); + goto out_symbol_exit; + } +if (rec->evlist->nr_entries ==0 &&perf_evlist__add_default (rec->evlist) <0 ) {pr_err ("Not enough memory for event selector list\n" );tools/perf/util/session.h +0
-0
tools/perf/util/session.h struct perf_session { · line 45
Quoted 4 lines
tools/perf/util/session.h +1
-0@@ -42,6 +42,16 @@ struct perf_session {#define PRINT_IP_OPT_ONELINE (1<<4)#define PRINT_IP_OPT_SRCLINE (1<<5) +enum perf_freq_perf_index {
That is really an overly long enum name! What about:
enum perf_freqs {
PERF_FREQ_TSC,
...
}Quoted 14 lines
tools/perf/util/session.h +9
-0+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_tool ; struct perf_session *perf_session__new(struct perf_data_file *file,Signature · 1 line
1.8.3.1
PATCH v9 4/6 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 --freq-perf ~/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 | 28 +++++++++++++++++++++--- tools/perf/util/session.h | 54 +++++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 79 insertions(+), 3 deletions(-)
tools/perf/util/session.c +25 -3
@@ -7,7 +7,6 @@#include <sys/mman.h> #include "evlist.h" -#include "evsel.h" #include "session.h" #include "tool.h" #include "sort.h"
@@ -878,8 +877,15 @@ 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_sample *sample, +struct perf_evsel *evsel) { +u64 read_format = evsel->attr.read_format; +struct perf_evlist *evlist = evsel->evlist; +struct perf_sample_id *sid; +perf_freq_t data = {0 }; +u64 cpu_max_freq = evlist->env->cpuattr.max_freq; +printf ("... sample_read:\n" );if (read_format &PERF_FORMAT_TOTAL_TIME_ENABLED )
@@ -902,10 +908,26 @@ 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 ) +perf_freq__init (evlist->env->msr_pmu_type, + evsel, data, value->value); } }else printf ("..... id %016" PRIx64", value %016" PRIx64"\n" , sample->read.one.id, sample->read.one.value); + +if (perf_freq__has_freq (data)) +printf ("..... Freq %lu MHz\n" , +perf_freq__get_freq (data, cpu_max_freq/1000 )); +if (perf_freq__has_cpu_util (data)) +printf ("..... CPU%% %lu%%\n" , +perf_freq__get_cpu_util (data)); +if (perf_freq__has_core_busy (data)) +printf ("..... CORE_BUSY%% %lu%%\n" , +perf_freq__get_core_busy (data)); } static void dump_event(struct perf_evlist *evlist, union perf_event *event,
@@ -965,7 +987,7 @@ 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 (sample, evsel); } static struct machine *machines__find_for_cpumode(struct machines *machines,
tools/perf/util/session.h +54 -0
72 lines not on this page · raw message
Signature · 1 line
1.8.3.1
Arnaldo Carvalho de Melo
maintainerRe: [PATCH V9 4/6] perf,tools: Dump per-sample freq/CPU%/CORE_BUSY% in report -D
Em Tue, Sep 08, 2015 at 03:32:47PM -0400, [email protected] escreveu:
Quoted 46 lines
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 --freq-perf ~/tchain_editHere is one sample from perf report -D
1972044565107
0x3498[0x88]:PERF_RECORD_SAMPLE(IP, 0x2): 10608/10608:0x4005fdperiod: 564686 addr: 0
...sample_read:
.... group nr 5
..... id 0000000000000012, value 0000000002143901
..... id 0000000000000052, value 0000000002143896
..... id 0000000000000094, value00000000021e443d
..... id00000000000000d4, value00000000021db984
..... id 0000000000000114, value00000000021db964
..... Freq 2301 MHz
..... CPU% 98%
..... CORE_BUSY% 99%Signed-off-by Kan Liang <[email protected]>Acked-by Jiri Olsa <[email protected]>--- tools/perf/util/session.c | 28 +++++++++++++++++++++--- tools/perf/util/session.h | 54 +++++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 79 insertions(+), 3 deletions(-)tools/perf/util/session.c +0
-1@@ -7,7 +7,6 @@#include <sys/mman.h> #include "evlist.h" -#include "evsel.h" #include "session.h" #include "tool.h" #include "sort.h"
tools/perf/util/session.c static void perf_evlist__print_tstamp(struct perf_evlist *evlist, · line 881
Quoted 6 lines
tools/perf/util/session.c +2
-1@@ -878,8 +877,15 @@ 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_sample *sample, +struct perf_evsel *evsel)
Hey, this is a _printf_ function, i.e. one thinks about functions with such name it doing just that, printing whatever is _already_ in the data structures, at maximum messing with _output_ related internal state.
This probably needs setting up at some other place, so that when it comes here it is already all worked out.
- Arnaldo
Quoted 127 lines
tools/perf/util/session.c +23
-1{ +u64 read_format = evsel->attr.read_format; +struct perf_evlist *evlist = evsel->evlist; +struct perf_sample_id *sid; +perf_freq_t data = {0 }; +u64 cpu_max_freq = evlist->env->cpuattr.max_freq; +printf ("... sample_read:\n" );if (read_format &PERF_FORMAT_TOTAL_TIME_ENABLED )@@ -902,10 +908,26 @@ 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 ) +perf_freq__init (evlist->env->msr_pmu_type, + evsel, data, value->value); } }else printf ("..... id %016" PRIx64", value %016" PRIx64"\n" , sample->read.one.id, sample->read.one.value); + +if (perf_freq__has_freq (data)) +printf ("..... Freq %lu MHz\n" , +perf_freq__get_freq (data, cpu_max_freq/1000 )); +if (perf_freq__has_cpu_util (data)) +printf ("..... CPU%% %lu%%\n" , +perf_freq__get_cpu_util (data)); +if (perf_freq__has_core_busy (data)) +printf ("..... CORE_BUSY%% %lu%%\n" , +perf_freq__get_core_busy (data)); } static void dump_event(struct perf_evlist *evlist, union perf_event *event,@@ -965,7 +987,7 @@ 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 (sample, evsel); } static struct machine *machines__find_for_cpumode(struct machines *machines,tools/perf/util/session.h +54
-0@@ -11,6 +11,7 @@#include "ordered-events.h" #include <linux/rbtree.h> #include <linux/perf_event.h> +#include "evsel.h" struct ip_callchain ;struct thread ;@@ -42,6 +43,10 @@ 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 ,@@ -52,6 +57,55 @@ enum perf_freq_perf_index {FREQ_PERF_MAX }; +typedef u64 perf_freq_t [FREQ_PERF_MAX ]; + +static inline void perf_freq__init (unsigned int msr_pmu_type, +struct perf_evsel *evsel, +perf_freq_t array, +u64 value) +{ +if (evsel->attr.type == msr_pmu_type) { +if (evsel->attr.config ==PERF_MSR_TSC ) + array[FREQ_PERF_TSC ] = value; +if (evsel->attr.config ==PERF_MSR_APERF ) + array[FREQ_PERF_APERF ] = value; +if (evsel->attr.config ==PERF_MSR_MPERF ) + array[FREQ_PERF_MPERF ] = value; + } +if (evsel->attr.type ==PERF_TYPE_HARDWARE ) { +if (evsel->attr.config ==PERF_COUNT_HW_CPU_CYCLES ) + array[FREQ_PERF_CYCLES ] = value; +if (evsel->attr.config ==PERF_COUNT_HW_REF_CPU_CYCLES ) + array[FREQ_PERF_REF_CYCLES ] = value; + } +} + +static inline bool perf_freq__has_freq (perf_freq_t array) +{ +return ((array[FREQ_PERF_CYCLES ] >0 ) && (array[FREQ_PERF_REF_CYCLES ] >0 )); +} +static inline u64 perf_freq__get_freq (perf_freq_t array,u64 cpu_max_freq) +{ +return ((array[FREQ_PERF_CYCLES ] * cpu_max_freq) / array[FREQ_PERF_REF_CYCLES ]); +} +static inline bool perf_freq__has_cpu_util (perf_freq_t array) +{ +return ((array[FREQ_PERF_TSC ] >0 ) && (array[FREQ_PERF_REF_CYCLES ] >0 )); +} +static inline u64 perf_freq__get_cpu_util (perf_freq_t array) +{ +return ((100 * array[FREQ_PERF_REF_CYCLES ]) / array[FREQ_PERF_TSC ]); +} + +static inline bool perf_freq__has_core_busy (perf_freq_t array) +{ +return ((array[FREQ_PERF_APERF ] >0 ) && (array[FREQ_PERF_MPERF ] >0 )); +} +static inline u64 perf_freq__get_core_busy (perf_freq_t array) +{ +return ((100 * array[FREQ_PERF_APERF ]) / array[FREQ_PERF_MPERF ]); +} +struct perf_tool ; struct perf_session *perf_session__new(struct perf_data_file *file,Signature · 1 line
1.8.3.1
PATCH v9 5/6 perf,tools: caculate and save freq/CPU%/CORE_BUSY% in he_stat
In reply to [email protected]From: Kan Liang <[email protected]>
Caculate freq/CPU%/CORE_BUSY% in add_entry_cb, and update the value in he_stat.
--- tools/perf/builtin-report.c | 36 ++++++++++++++++++++++++++++++++++++ tools/perf/util/sort.h | 3 +++ 2 files changed, 39 insertions(+)
tools/perf/builtin-report.c +36 -0
@@ -90,6 +90,38 @@ static int report__config(const char *var, const char *value, void *cb)return perf_default_config (var, value, cb); } +static void set_he_freq_perf (struct perf_session *session, +struct hist_entry_iter *iter) +{ +struct hist_entry *he = iter->he; +struct perf_evsel *evsel = iter->evsel; +struct perf_evsel *leader = evsel; +struct perf_sample *sample = iter->sample; +struct perf_evlist *evlist = session->evlist; +u64 cpu_max_freq = session->header.env.cpuattr.max_freq; +perf_freq_t data = {0 }; +u64 nr =0 ; + +perf_freq__init (session->header.env.msr_pmu_type, + evsel, data, + sample->read.group.values[nr].value); +evlist__for_each_continue (evlist, evsel) { +if ((evsel->leader != leader) || + (++nr >= sample->read.group.nr)) +break ; +perf_freq__init (session->header.env.msr_pmu_type, + evsel, data, + sample->read.group.values[nr].value); + } + +if (perf_freq__has_freq (data)) + he->stat.freq =perf_freq__get_freq (data, cpu_max_freq/1000 ); +if (perf_freq__has_cpu_util (data)) + he->stat.cpu_util =perf_freq__get_cpu_util (data); +if (perf_freq__has_core_busy (data)) + he->stat.core_busy =perf_freq__get_core_busy (data); +} +static int hist_iter__report_callback (struct hist_entry_iter *iter,struct addr_location *al,bool single,void *arg)
@@ -101,6 +133,10 @@ static int hist_iter__report_callback(struct hist_entry_iter *iter,struct mem_info *mi;struct branch_info *bi; +if ((iter->ops == &hist_iter_normal) && +perf_evsel__is_group_leader (evsel)) +set_he_freq_perf (rep->session, iter); +if (!ui__has_annotation ())return 0 ;
tools/perf/util/sort.h +3 -0
@@ -54,6 +54,9 @@ struct he_stat {u64 period_guest_us;u64 weight;u32 nr_events; +u64 freq; +u64 cpu_util; +u64 core_busy; };struct hist_entry_diff {
Signature · 1 line
1.8.3.1
PATCH v9 6/6 perf,tools: Show freq/CPU%/CORE_BUSY% in perf report by --freq-perf
In reply to [email protected]From: Kan Liang <[email protected]>
Show frequency, CPU Utilization and percent performance for each symbol in perf report by --freq-perf
In sampling group, only group leader do sampling. So only need to print group leader's freq in --group.
--freq-perf option also implies --group.
Here is an example.
$ perf report --stdio --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 | 25 ++++++++++- tools/perf/ui/hist.c | 71 +++++++++++++++++++++++++++++--- tools/perf/util/hist.h | 3 ++ tools/perf/util/session.c | 33 ++++++++------- tools/perf/util/sort.c | 3 ++ tools/perf/util/symbol.h | 12 +++++- 7 files changed, 137 insertions(+), 22 deletions(-)
tools/perf/Documentation/perf-report.txt +12 -0
@@ -307,6 +307,18 @@ OPTIONS special event -e cpu/mem-loads/ or -e cpu/mem-stores/. See 'perf mem' for simpler access. +--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 +24 -1
@@ -134,7 +134,8 @@ static int hist_iter__report_callback(struct hist_entry_iter *iter,struct branch_info *bi;if ((iter->ops == &hist_iter_normal) && -perf_evsel__is_group_leader (evsel)) +perf_evsel__is_group_leader (evsel) && + symbol_conf.freq_perf)set_he_freq_perf (rep->session, iter);if (!ui__has_annotation ())
@@ -785,6 +786,8 @@ 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 ,"freq-perf" , &symbol_conf.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",
@@ -803,7 +806,9 @@ 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 (); +perf_freq_t freq_data_status = {0 };if (ret <0 )return ret;
@@ -888,6 +893,24 @@ repeat: symbol_conf.cumulate_callchain = false; } + +if (symbol_conf.freq_perf) { + symbol_conf.freq_perf_type =0 ; +evlist__for_each (session->evlist, pos) { +perf_freq__init (session->header.env.msr_pmu_type, + pos, freq_data_status,1 ); + } +if (perf_freq__has_freq (freq_data_status)) + symbol_conf.freq_perf_type |=1U <<DISPLAY_FREQ ; +if (perf_freq__has_cpu_util (freq_data_status)) + symbol_conf.freq_perf_type |=1U <<DISPLAY_CPU_UTIL ; +if (perf_freq__has_core_busy (freq_data_status)) + symbol_conf.freq_perf_type |=1U <<DISPLAY_CORE_BUSY ; + +/* --freq-perf option imply --group */ + symbol_conf.event_group = true; + } +if (setup_sorting () <0 ) {if (sort_order)parse_options_usage (report_usage, options,"s" ,1 );
tools/perf/ui/hist.c +66 -5
140 lines not on this page · raw message
tools/perf/util/hist.h +3 -0
@@ -241,6 +241,9 @@ enum {PERF_HPP__OVERHEAD_ACC ,PERF_HPP__SAMPLES ,PERF_HPP__PERIOD , +PERF_HPP__FREQ , +PERF_HPP__CPU_UTIL , +PERF_HPP__CORE_BUSY , PERF_HPP__MAX_INDEX };
tools/perf/util/session.c +18 -15
@@ -908,26 +908,29 @@ static void sample_read__printf(struct perf_sample *sample,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 ) -perf_freq__init (evlist->env->msr_pmu_type, - evsel, data, value->value); +if (symbol_conf.freq_perf) { + sid =perf_evlist__id2sid (evlist, value->id); + evsel = sid->evsel; +if (evsel !=NULL ) +perf_freq__init (evlist->env->msr_pmu_type, + evsel, data, value->value); + } } }else printf ("..... id %016" PRIx64", value %016" PRIx64"\n" , sample->read.one.id, sample->read.one.value); -if (perf_freq__has_freq (data)) -printf ("..... Freq %lu MHz\n" , -perf_freq__get_freq (data, cpu_max_freq/1000 )); -if (perf_freq__has_cpu_util (data)) -printf ("..... CPU%% %lu%%\n" , -perf_freq__get_cpu_util (data)); -if (perf_freq__has_core_busy (data)) -printf ("..... CORE_BUSY%% %lu%%\n" , -perf_freq__get_core_busy (data)); +if (symbol_conf.freq_perf) { +if (perf_freq__has_freq (data)) +printf ("..... Freq %lu MHz\n" , +perf_freq__get_freq (data, cpu_max_freq/1000 )); +if (perf_freq__has_cpu_util (data)) +printf ("..... CPU%% %lu%%\n" , +perf_freq__get_cpu_util (data)); +if (perf_freq__has_core_busy (data)) +printf ("..... CORE_BUSY%% %lu%%\n" , +perf_freq__get_core_busy (data)); + } } static void dump_event(struct perf_evlist *evlist, union perf_event *event,
tools/perf/util/sort.c +3 -0
@@ -1325,6 +1325,9 @@ 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_UTIL ,"cpu_u" ), + DIM(PERF_HPP__CORE_BUSY, "core_busy"), };#undef DIM
tools/perf/util/symbol.h +11 -1
@@ -80,6 +80,14 @@ static inline size_t symbol__size(const struct symbol *sym)struct strlist ;struct intlist ; +enum freq_perf_type_index { +DISPLAY_FREQ =0 , +DISPLAY_CPU_UTIL , +DISPLAY_CORE_BUSY , + +DISPLAY_MAX +}; +struct symbol_conf {unsigned short priv_size;unsigned short nr_events;
@@ -107,7 +115,8 @@ struct symbol_conf { show_hist_headers, branch_callstack, has_filter, - show_ref_callgraph; + show_ref_callgraph, + freq_perf;const char *vmlinux_name, *kallsyms_name, *source_prefix,
@@ -132,6 +141,7 @@ struct symbol_conf {struct intlist *pid_list, *tid_list;const char *symfs; +u64 freq_perf_type; };extern struct symbol_conf symbol_conf;
Signature · 1 line
1.8.3.1
Re: [PATCH V9 6/6] perf,tools: Show freq/CPU%/CORE_BUSY% in perf report by --freq-perf
Quoted 19 lines · On Tue, Sep 08, 2015 at 03:32:49PM -0400, [email protected] wrote:
From: Kan Liang <[email protected]>
Show frequency, CPU Utilization and percent performance for each symbol in perf report by --freq-perf
In sampling group, only group leader do sampling. So only need to print group leader's freq in --group.
--freq-perf option also implies --group.
Here is an example.
$ perf report --stdio --freq-perfOverhead FREQ MHz CPU% CORE_BUSY%
Command Shared Object Symbol
........................................ ......... ..... ..........
........... ................ ......................
Quoted 8 lines
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_tscSigned-off-by Kan Liang <[email protected]>
it seems ok to me, though I think it could be split into HPP__SINGLE_PRINT_FNS support and the rest..
but I prefer if Arnaldo or Namhyung could review this one
thanks,
jirka
Quoted 62 lines · On Tue, Sep 08, 2015 at 03:32:43PM -0400, [email protected] wrote:
From: Kan Liang <[email protected]>
This patch set supports per-sample freq/CPU%/CORE_BUSY% print in perf report.
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.Here is an example:
$ perf record --freq-perf ~/tchain_edit$ perf report --stdio --freq-perfOverhead 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_tscChanges since V1:
- Save cpu max freq to header when recording
- Read cpu max freq and msr type from header when reportingChanges since V2:
- Introduce generic FEAT for CPU related data stored
- Make cpu max freq and msr type part ofperf_session_env
- renamecpu_utocpu_util
- Don't save sample value inperf_sampleand discards new iterator. Calculating thefreq_perf_infoinadd_entry_cbcallback
- Introduce symbol_conf.freq_perf_type for related hpp column visibilityChanges since V3:
- add a identifier 'tag' for CPU attributes, max frequency.
- add backpointers to evlist for env, and evsel for evlist.
- Use bitmask forfreq_perf_type
- Replace macros by functions to caculate freq,cpu_utilandcore_busy
- Move all caculation codes under symbol_conf.show_freq_perf condition.Changes since V4:
- Store cpu attributes id as tag and more readablecpu_attrChanges since V5:
- Rename freq tomax_freqand use it
- Add a loop inprocess_cpu_attributesto facility future extensionChanges since V6:
- Split renameperf_session_envand add backpointer to evlist patches
Quoted 7 lines
Changes since V7:
- Introduce --freq-perf option for perf record
- Factor out fix fordump_sampleChanges since V8:
- Rename --show-freq-perf to --freq-perf
- --freq-perf option imply --group
I believe I just acked all of it but the last one, otherwise I'm ok with it
thanks,
jirka