perf record: Support sample-read topdown metric group
A message that is not in the archive (sent off-list or not yet received): [email protected]
PATCH v2 1/4 perf tools: Rename group to topdown
From: Kan Liang <[email protected]>
The group.h/c only include TopDown group related functions. The name "group" is too generic and inaccurate. Use the name "topdown" to replace it.
Move topdown related functions to a dedicated file, topdown.c.
---
tools/perf/arch/x86/util/Build | 2 +-
.../perf/arch/x86/util/{group.c => topdown.c} | 2 +-
tools/perf/builtin-stat.c | 51 +-----------------
tools/perf/util/Build | 1 +
tools/perf/util/topdown.c | 53 +++++++++++++++++++
tools/perf/util/{group.h => topdown.h} | 6 ++-
6 files changed, 61 insertions(+), 54 deletions(-)
rename tools/perf/arch/x86/util/{group.c => topdown.c} (95%)
create mode 100644 tools/perf/util/topdown.c
rename tools/perf/util/{group.h => topdown.h} (52%)tools/perf/arch/x86/util/Build +1 -1
@@ -3,7 +3,7 @@ perf-y += tsc.o perf-y += pmu.o perf-y += kvm-stat.o perf-y += perf_regs.o -perf-y += group.o +perf-y += topdown.o perf-y += machine.o perf-y += event.o
tools/perf/arch/x86/util/topdown.c renamed from tools/perf/arch/x86/util/group.c +1 -1
@@ -1,7 +1,7 @@// SPDX-License-Identifier: GPL-2.0 #include <stdio.h> #include "api/fs/fs.h" -#include "util/group.h" +#include "util/topdown.h" /* * Check whether we can use a group for top down.
tools/perf/builtin-stat.c +1 -50
63 lines not on this page · raw message
tools/perf/util/Build +1 -0
@@ -101,6 +101,7 @@ perf-y += call-path.o perf-y += rwsem.o perf-y += thread-stack.o perf-y += spark.o +perf-y += topdown.o perf-$(CONFIG_AUXTRACE) += auxtrace.o perf-$(CONFIG_AUXTRACE) += intel-pt-decoder/ perf-$(CONFIG_AUXTRACE) += intel-pt.o
tools/perf/util/topdown.c new file +53 -0
53 lines not on this page · raw message
tools/perf/util/topdown.h renamed from tools/perf/util/group.h +4 -2
@@ -1,8 +1,10 @@/* SPDX-License-Identifier: GPL-2.0 */ -#ifndef GROUP_H -#define GROUP_H 1 +#ifndef TOPDOWN_H +#define TOPDOWN_H 1bool arch_topdown_check_group (bool *warn);void arch_topdown_group_warn (void ); +int topdown_filter_events (const char **attr,char **str,bool use_group); +#endif
Signature · 1 line
2.17.1
PATCH v2 2/4 perf record: Support sample-read topdown metric group
In reply to a message not in the archiveFrom: Kan Liang <[email protected]>
With the hardware TopDown metrics feature, sample-read feature should be supported for a topdown group, e.g., sample a non-topdown event and read a topdown metric group. But the current perf record code errors out.
For a topdown metric group, the slots event must be the leader of the group, but the leader slots event doesn't support sampling.
To support sample-read the topdown metric group, use the 2nd event of the group as the "leader" for the purposes of sampling.
Only the platform with Topdown metic feature supports sample-read the topdown group. Add arch_topdown_sample_read() to indicate whether the topdown group supports sample-read.
--- tools/perf/arch/x86/util/topdown.c | 35 ++++++++++++++++++++++++++++++ tools/perf/util/record.c | 3 ++- tools/perf/util/topdown.c | 5 +++++ tools/perf/util/topdown.h | 2 ++ 4 files changed, 44 insertions(+), 1 deletion(-)
tools/perf/arch/x86/util/topdown.c +35 -0
@@ -1,6 +1,7 @@// SPDX-License-Identifier: GPL-2.0 #include <stdio.h> #include "api/fs/fs.h" +#include "util/pmu.h" #include "util/topdown.h" /*
@@ -26,3 +27,37 @@ void arch_topdown_group_warn(void)"nmi_watchdog enabled with topdown. May give wrong results.\n" "Disable with echo 0 > /proc/sys/kernel/nmi_watchdog\n"); } + +#define TOPDOWN_SLOTS 0x0400 + +static bool is_topdown_slots_event (struct evsel *counter) +{ +if (!counter->pmu_name) +return false; + +if (strcmp (counter->pmu_name,"cpu" )) +return false; + +if (counter->core.attr.config ==TOPDOWN_SLOTS ) +return true; + +return false; +} + +/* +* Check whether a topdown group supports sample-read. +* +* Only Topdown metic supports sample-read. The slots +* event must be the leader of the topdown group. +*/ + +bool arch_topdown_sample_read (struct evsel *leader) +{ +if (!pmu_have_event ("cpu" ,"slots" )) +return false; + +if (is_topdown_slots_event (leader)) +return true; + +return false; +}
tools/perf/util/record.c +2 -1
@@ -13,6 +13,7 @@#include "util/perf_api_probe.h" #include "record.h" #include "../perf-sys.h" +#include "topdown.h" /* * evsel__config_leader_sampling() uses special rules for leader sampling.
@@ -23,7 +24,7 @@ static struct evsel *evsel__read_sampler(struct evsel *evsel, struct evlist *evl {struct evsel *leader = evsel->leader; -if (evsel__is_aux_event (leader)) { +if (evsel__is_aux_event (leader) ||arch_topdown_sample_read (leader)) {evlist__for_each_entry (evlist, evsel) {if (evsel->leader == leader && evsel != evsel->leader)return evsel;
tools/perf/util/topdown.c +5 -0
@@ -51,3 +51,8 @@ __weak bool arch_topdown_check_group(bool *warn)__weak void arch_topdown_group_warn (void ) { } + +__weak bool arch_topdown_sample_read (struct evsel *leader__maybe_unused ) +{ +return false; +}
tools/perf/util/topdown.h +2 -0
@@ -1,9 +1,11 @@/* SPDX-License-Identifier: GPL-2.0 */ #ifndef TOPDOWN_H #define TOPDOWN_H 1 +#include "evsel.h" bool arch_topdown_check_group (bool *warn);void arch_topdown_group_warn (void ); +bool arch_topdown_sample_read (struct evsel *leader);int topdown_filter_events (const char **attr,char **str,bool use_group);
Signature · 1 line
2.17.1
PATCH v2 3/4 perf stat: Support new per thread TopDown metrics
In reply to a message not in the archiveFrom: Andi Kleen <[email protected]>
Icelake has support for reporting per thread TopDown metrics. These are reported differently than the previous TopDown support, each metric is standalone, but scaled to pipeline "slots". We don't need to do anything special for HyperThreading anymore. Teach perf stat --topdown to handle these new metrics and print them in the same way as the previous TopDown metrics. The restrictions of only being able to report information per core is gone.
--- tools/perf/Documentation/perf-stat.txt | 7 +- tools/perf/builtin-stat.c | 30 ++++++++- tools/perf/util/stat-shadow.c | 89 ++++++++++++++++++++++++++ tools/perf/util/stat.c | 4 ++ tools/perf/util/stat.h | 8 +++ 5 files changed, 134 insertions(+), 4 deletions(-)
tools/perf/Documentation/perf-stat.txt +5 -2
@@ -357,6 +357,11 @@ if the workload is actually bound by the CPU and not by something else. For best results it is usually a good idea to use it with interval mode like -I 1000, as the bottleneck of workloads can change often. +This enables --metric-only, unless overridden with --no-metric-only. + +The following restrictions only apply to older Intel CPUs and Atom, +on newer CPUs (IceLake and later) TopDown can be collected for any thread: + The top down metrics are collected per core instead of per CPU thread. Per core mode is automatically enabled and -a (global monitoring) is needed, requiring root rights or
@@ -368,8 +373,6 @@ echo 0 > /proc/sys/kernel/nmi_watchdog for best results. Otherwise the bottlenecks may be inconsistent on workload with changing phases. -This enables --metric-only, unless overridden with --no-metric-only. - To interpret the results it is usually needed to know on which CPUs the workload runs on. If needed the CPUs can be forced using taskset.
tools/perf/builtin-stat.c +28 -2
@@ -128,6 +128,15 @@ static const char * topdown_attrs[] = { NULL, }; +static const char *topdown_metric_attrs[] = { +"slots" , +"topdown-retiring" , +"topdown-bad-spec" , +"topdown-fe-bound" , +"topdown-be-bound" , +NULL , +}; +static const char *smi_cost_attrs = {"{" "msr/aperf/,"
@@ -1691,6 +1700,24 @@ static int add_default_attributes(void)char *str =NULL ;bool warn = false; +if (!force_metric_only) + stat_config.metric_only = true; + +if (topdown_filter_events (topdown_metric_attrs, &str,1 ) <0 ) { +pr_err ("Out of memory\n" ); +return -1 ; + } +if (topdown_metric_attrs[0 ] && str) { +if (!stat_config.interval && !stat_config.metric_only) { +fprintf (stat_config.output, +"Topdown accuracy may decrease when measuring long periods.\n" +"Please print the result regularly, e.g. -I1000\n" ); + } + goto setup_metrics; + } + + str =NULL ; +if (stat_config.aggr_mode !=AGGR_GLOBAL && stat_config.aggr_mode !=AGGR_CORE ) {pr_err ("top down event configuration requires --per-core mode\n" );
@@ -1702,8 +1729,6 @@ static int add_default_attributes(void)return -1 ; } -if (!force_metric_only) - stat_config.metric_only = true;if (topdown_filter_events (topdown_attrs, &str,arch_topdown_check_group (&warn)) <0 ) {pr_err ("Out of memory\n" );
@@ -1712,6 +1737,7 @@ static int add_default_attributes(void)if (topdown_attrs[0 ] && str) {if (warn)arch_topdown_group_warn (); +setup_metrics: err =parse_events (evsel_list, str, &errinfo);if (err) { fprintf(stderr,
tools/perf/util/stat-shadow.c +89 -0
107 lines not on this page · raw message
tools/perf/util/stat.c +4 -0
@@ -95,6 +95,10 @@ static const char *id_str[PERF_STAT_EVSEL_ID__MAX] = {ID (TOPDOWN_SLOTS_RETIRED , topdown-slots-retired),ID (TOPDOWN_FETCH_BUBBLES , topdown-fetch-bubbles),ID (TOPDOWN_RECOVERY_BUBBLES , topdown-recovery-bubbles), +ID (TOPDOWN_RETIRING , topdown-retiring), +ID (TOPDOWN_BAD_SPEC , topdown-bad-spec), +ID (TOPDOWN_FE_BOUND , topdown-fe-bound), +ID (TOPDOWN_BE_BOUND , topdown-be-bound), ID(SMI_NUM, msr/smi/), ID(APERF, msr/aperf/), };
tools/perf/util/stat.h +8 -0
@@ -28,6 +28,10 @@ enum perf_stat_evsel_id {PERF_STAT_EVSEL_ID__TOPDOWN_SLOTS_RETIRED ,PERF_STAT_EVSEL_ID__TOPDOWN_FETCH_BUBBLES ,PERF_STAT_EVSEL_ID__TOPDOWN_RECOVERY_BUBBLES , +PERF_STAT_EVSEL_ID__TOPDOWN_RETIRING , +PERF_STAT_EVSEL_ID__TOPDOWN_BAD_SPEC , +PERF_STAT_EVSEL_ID__TOPDOWN_FE_BOUND , +PERF_STAT_EVSEL_ID__TOPDOWN_BE_BOUND ,PERF_STAT_EVSEL_ID__SMI_NUM ,PERF_STAT_EVSEL_ID__APERF , PERF_STAT_EVSEL_ID__MAX,
@@ -82,6 +86,10 @@ enum stat_type {STAT_TOPDOWN_SLOTS_RETIRED ,STAT_TOPDOWN_FETCH_BUBBLES ,STAT_TOPDOWN_RECOVERY_BUBBLES , +STAT_TOPDOWN_RETIRING , +STAT_TOPDOWN_BAD_SPEC , +STAT_TOPDOWN_FE_BOUND , +STAT_TOPDOWN_BE_BOUND ,STAT_SMI_NUM ,STAT_APERF ,STAT_MAX
Signature · 1 line
2.17.1
Namhyung Kim
maintainerRe: [PATCH V2 3/4] perf stat: Support new per thread TopDown metrics
Hello,
Quoted 80 lines · On Thu, Sep 10, 2020 at 10:48 PM <[email protected]> wrote:
From: Andi Kleen <[email protected]>
Icelake has support for reporting per thread TopDown metrics. These are reported differently than the previous TopDown support, each metric is standalone, but scaled to pipeline "slots". We don't need to do anything special for HyperThreading anymore. Teach perf stat --topdown to handle these new metrics and print them in the same way as the previous TopDown metrics. The restrictions of only being able to report information per core is gone.
Acked-by Jiri Olsa <[email protected]>Co-developed-by Kan Liang <[email protected]>Signed-off-by Kan Liang <[email protected]>Signed-off-by Andi Kleen <[email protected]>--- tools/perf/Documentation/perf-stat.txt | 7 +- tools/perf/builtin-stat.c | 30 ++++++++- tools/perf/util/stat-shadow.c | 89 ++++++++++++++++++++++++++ tools/perf/util/stat.c | 4 ++ tools/perf/util/stat.h | 8 +++ 5 files changed, 134 insertions(+), 4 deletions(-)tools/perf/Documentation/perf-stat.txt +5
-2@@ -357,6 +357,11 @@ if the workload is actually bound by the CPU and not by something else. For best results it is usually a good idea to use it with interval mode like -I 1000, as the bottleneck of workloads can change often. +This enables --metric-only, unless overridden with --no-metric-only. + +The following restrictions only apply to older Intel CPUs and Atom, +on newer CPUs (IceLake and later) TopDown can be collected for any thread: + The top down metrics are collected per core instead of per CPU thread. Per core mode is automatically enabled and -a (global monitoring) is needed, requiring root rights or@@ -368,8 +373,6 @@ echo 0 > /proc/sys/kernel/nmi_watchdog for best results. Otherwise the bottlenecks may be inconsistent on workload with changing phases. -This enables --metric-only, unless overridden with --no-metric-only. - To interpret the results it is usually needed to know on which CPUs the workload runs on. If needed the CPUs can be forced using taskset.tools/perf/builtin-stat.c +16
-0@@ -128,6 +128,15 @@ static const char * topdown_attrs[] = { NULL, }; +static const char *topdown_metric_attrs[] = { +"slots" , +"topdown-retiring" , +"topdown-bad-spec" , +"topdown-fe-bound" , +"topdown-be-bound" , +NULL , +}; +static const char *smi_cost_attrs = {"{" "msr/aperf/," @@ -1691,6 +1700,24 @@ static int add_default_attributes(void)char *str =NULL ;bool warn = false; +if (!force_metric_only) + stat_config.metric_only = true; + +if (topdown_filter_events (topdown_metric_attrs, &str,1 ) <0 ) { +pr_err ("Out of memory\n" ); +return -1 ; + }
tools/perf/builtin-stat.c static int add_default_attributes(void) · line 1719
Quoted 10 lines
tools/perf/builtin-stat.c +10
-0+if (topdown_metric_attrs[0 ] && str) { +if (!stat_config.interval && !stat_config.metric_only) { +fprintf (stat_config.output, +"Topdown accuracy may decrease when measuring long periods.\n" +"Please print the result regularly, e.g. -I1000\n" ); + } + goto setup_metrics; + } + + str =NULL ;
zfree(&str) ?
Thanks
Namhyung
Quoted 21 lines
tools/perf/builtin-stat.c +2
-2+if (stat_config.aggr_mode !=AGGR_GLOBAL && stat_config.aggr_mode !=AGGR_CORE ) {pr_err ("top down event configuration requires --per-core mode\n" );@@ -1702,8 +1729,6 @@ static int add_default_attributes(void)return -1 ; } -if (!force_metric_only) - stat_config.metric_only = true;if (topdown_filter_events (topdown_attrs, &str,arch_topdown_check_group (&warn)) <0 ) {pr_err ("Out of memory\n" );@@ -1712,6 +1737,7 @@ static int add_default_attributes(void)if (topdown_attrs[0 ] && str) {if (warn)arch_topdown_group_warn (); +setup_metrics: err =parse_events (evsel_list, str, &errinfo);if (err) { fprintf(stderr,
Re: [PATCH V2 3/4] perf stat: Support new per thread TopDown metrics
Quoted 87 lines · Namhyung Kim
Hello,
Quoted 84 lines · On Thu, Sep 10, 2020 at 10:48 PM <[email protected]> wrote:
From: Andi Kleen <[email protected]>
Icelake has support for reporting per thread TopDown metrics. These are reported differently than the previous TopDown support, each metric is standalone, but scaled to pipeline "slots". We don't need to do anything special for HyperThreading anymore. Teach perf stat --topdown to handle these new metrics and print them in the same way as the previous TopDown metrics. The restrictions of only being able to report information per core is gone.
Acked-by Jiri Olsa <[email protected]>Co-developed-by Kan Liang <[email protected]>Signed-off-by Kan Liang <[email protected]>Signed-off-by Andi Kleen <[email protected]>--- tools/perf/Documentation/perf-stat.txt | 7 +- tools/perf/builtin-stat.c | 30 ++++++++- tools/perf/util/stat-shadow.c | 89 ++++++++++++++++++++++++++ tools/perf/util/stat.c | 4 ++ tools/perf/util/stat.h | 8 +++ 5 files changed, 134 insertions(+), 4 deletions(-)tools/perf/Documentation/perf-stat.txt +5
-2@@ -357,6 +357,11 @@ if the workload is actually bound by the CPU and not by something else. For best results it is usually a good idea to use it with interval mode like -I 1000, as the bottleneck of workloads can change often. +This enables --metric-only, unless overridden with --no-metric-only. + +The following restrictions only apply to older Intel CPUs and Atom, +on newer CPUs (IceLake and later) TopDown can be collected for any thread: + The top down metrics are collected per core instead of per CPU thread. Per core mode is automatically enabled and -a (global monitoring) is needed, requiring root rights or@@ -368,8 +373,6 @@ echo 0 > /proc/sys/kernel/nmi_watchdog for best results. Otherwise the bottlenecks may be inconsistent on workload with changing phases. -This enables --metric-only, unless overridden with --no-metric-only. - To interpret the results it is usually needed to know on which CPUs the workload runs on. If needed the CPUs can be forced using taskset.tools/perf/builtin-stat.c +20
-0@@ -128,6 +128,15 @@ static const char * topdown_attrs[] = { NULL, }; +static const char *topdown_metric_attrs[] = { +"slots" , +"topdown-retiring" , +"topdown-bad-spec" , +"topdown-fe-bound" , +"topdown-be-bound" , +NULL , +}; +static const char *smi_cost_attrs = {"{" "msr/aperf/," @@ -1691,6 +1700,24 @@ static int add_default_attributes(void)char *str =NULL ;bool warn = false; +if (!force_metric_only) + stat_config.metric_only = true; + +if (topdown_filter_events (topdown_metric_attrs, &str,1 ) <0 ) { +pr_err ("Out of memory\n" ); +return -1 ; + } +if (topdown_metric_attrs[0 ] && str) { +if (!stat_config.interval && !stat_config.metric_only) { +fprintf (stat_config.output, + "Topdown accuracy may decrease when measuring long periods.\n"
Quoted 8 lines
tools/perf/builtin-stat.c static int add_default_attributes(void) · line 1719
Quoted 6 lines
tools/perf/builtin-stat.c +6
-0+ "Please print the result regularly, e.g. -I1000\n"); + } + goto setup_metrics; + } + + str =NULL ;
zfree(&str)?
Yes, even the topdown events don't exist, str is still allocated. The str should be free.
I will send V3 to fix it shortly.
Thanks,
Kan
Quoted 26 lines
Thanks
NamhyungQuoted 21 lines
tools/perf/builtin-stat.c +2
-2+if (stat_config.aggr_mode !=AGGR_GLOBAL && stat_config.aggr_mode !=AGGR_CORE ) {pr_err ("top down event configuration requires --per-core mode\n" );@@ -1702,8 +1729,6 @@ static int add_default_attributes(void)return -1 ; } -if (!force_metric_only) - stat_config.metric_only = true;if (topdown_filter_events (topdown_attrs, &str,arch_topdown_check_group (&warn)) <0 ) {pr_err ("Out of memory\n" );@@ -1712,6 +1737,7 @@ static int add_default_attributes(void)if (topdown_attrs[0 ] && str) {if (warn)arch_topdown_group_warn (); +setup_metrics: err =parse_events (evsel_list, str, &errinfo);if (err) { fprintf(stderr,