perf print-events: Fix "perf list" can not display the PMU prefix for some hybrid cache events
PATCH v2 1/2 perf print-events: Fix "perf list" can not display the PMU prefix for some hybrid cache events
From: Zhengjun Xing <[email protected]>
Some hybrid hardware cache events are only available on one CPU PMU. For example, 'L1-dcache-load-misses' is only available on cpu_core. We have supported in the perf list clearly reporting this info, the function works fine before but recently the argument "config" in API is_event_supported() is changed from "u64" to "unsigned int" which caused a regression, the "perf list" then can not display the PMU prefix for some hybrid cache events. For the hybrid systems, the PMU type ID is stored at config[63:32], define config to "unsigned int" will miss the PMU type ID information, then the regression happened, the config should be defined as "u64".
Before:
# ./perf list |grep "Hardware cache event" L1-dcache-load-misses [Hardware cache event] L1-dcache-loads [Hardware cache event] L1-dcache-stores [Hardware cache event] L1-icache-load-misses [Hardware cache event] L1-icache-loads [Hardware cache event] LLC-load-misses [Hardware cache event] LLC-loads [Hardware cache event] LLC-store-misses [Hardware cache event] LLC-stores [Hardware cache event] branch-load-misses [Hardware cache event] branch-loads [Hardware cache event] dTLB-load-misses [Hardware cache event] dTLB-loads [Hardware cache event] dTLB-store-misses [Hardware cache event] dTLB-stores [Hardware cache event] iTLB-load-misses [Hardware cache event] node-load-misses [Hardware cache event] node-loads [Hardware cache event]
After:
# ./perf list |grep "Hardware cache event" L1-dcache-loads [Hardware cache event] L1-dcache-stores [Hardware cache event] L1-icache-load-misses [Hardware cache event] LLC-load-misses [Hardware cache event] LLC-loads [Hardware cache event] LLC-store-misses [Hardware cache event] LLC-stores [Hardware cache event] branch-load-misses [Hardware cache event] branch-loads [Hardware cache event] cpu_atom/L1-icache-loads/ [Hardware cache event] cpu_core/L1-dcache-load-misses/ [Hardware cache event] cpu_core/node-load-misses/ [Hardware cache event] cpu_core/node-loads/ [Hardware cache event] dTLB-load-misses [Hardware cache event] dTLB-loads [Hardware cache event] dTLB-store-misses [Hardware cache event] dTLB-stores [Hardware cache event] iTLB-load-misses [Hardware cache event]
---
Change log:
v2:
* Adds Acked-by from Ian Rogers <[email protected]>
* Adds Reported-by from Yi Ammy <[email protected]>
tools/perf/util/print-events.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/print-events.c +1 -1
@@ -239,7 +239,7 @@ void print_sdt_events(const char *subsys_glob, const char *event_glob,strlist__delete (sdtlist); } -static bool is_event_supported (u8 type,unsigned int config) +static bool is_event_supported (u8 type,u64 config) {bool ret = true;int open_return;
Signature · 1 line
2.25.1
PATCH v2 2/2 perf parse-events: Remove "not supported" hybrid cache events
From: Zhengjun Xing <[email protected]>
By default, we create two hybrid cache events, one is for cpu_core, and another is for cpu_atom. But Some hybrid hardware cache events are only available on one CPU PMU. For example, the 'L1-dcache-load-misses' is only available on cpu_core, while the 'L1-icache-loads' is only available on cpu_atom. We need to remove "not supported" hybrid cache events. By extending is_event_supported() to global API and using it to check if the hybrid cache events are supported before being created, we can remove the "not supported" hybrid cache events.
Before:
# ./perf stat -e L1-dcache-load-misses,L1-icache-loads -a sleep 1
Performance counter stats for 'system wide':
52,570 cpu_core/L1-dcache-load-misses/
<not supported> cpu_atom/L1-dcache-load-misses/
<not supported> cpu_core/L1-icache-loads/
1,471,817 cpu_atom/L1-icache-loads/1.004915229 seconds time elapsed
After:
# ./perf stat -e L1-dcache-load-misses,L1-icache-loads -a sleep 1
Performance counter stats for 'system wide':
54,510 cpu_core/L1-dcache-load-misses/
1,441,286 cpu_atom/L1-icache-loads/
1.005114281 seconds time elapsed---
Change log:
v2:
* Adds a comment for removing "not supported" hybrid cache events.
* Remove goto and add a strdup check
* "is_event_supported" move to parse-events.c per Ian's suggestion.
* Adds Reported-by from Yi Ammy <[email protected]>
tools/perf/util/parse-events-hybrid.c | 21 ++++++++++++--- tools/perf/util/parse-events.c | 39 +++++++++++++++++++++++++++ tools/perf/util/parse-events.h | 1 + tools/perf/util/print-events.c | 39 --------------------------- 4 files changed, 57 insertions(+), 43 deletions(-)
tools/perf/util/parse-events-hybrid.c +17 -4
@@ -33,7 +33,8 @@ static void config_hybrid_attr(struct perf_event_attr *attr,* If the PMU type ID is 0, the PERF_TYPE_RAW will be applied. */ attr->type = type; - attr->config = attr->config | ((__u64 )pmu_type <<PERF_PMU_TYPE_SHIFT ); + attr->config = (attr->config &PERF_HW_EVENT_MASK ) | + ((__u64 )pmu_type <<PERF_PMU_TYPE_SHIFT ); } static int create_event_hybrid(__u32 config_type, int *idx,
@@ -48,13 +49,25 @@ static int create_event_hybrid(__u32 config_type, int *idx,__u64 config = attr->config;config_hybrid_attr (attr, config_type, pmu->type); + +/* +* Some hybrid hardware cache events are only available on one CPU +* PMU. For example, the 'L1-dcache-load-misses' is only available +* on cpu_core, while the 'L1-icache-loads' is only available on +* cpu_atom. We need to remove "not supported" hybrid cache events. +*/ +if (attr->type ==PERF_TYPE_HW_CACHE + && !is_event_supported (attr->type, attr->config)) +return 0 ; + evsel =parse_events__add_event_hybrid (list, idx, attr, name, metric_id, pmu, config_terms); -if (evsel) +if (evsel) { evsel->pmu_name =strdup (pmu->name); -else +if (!evsel->pmu_name) +return -ENOMEM ; + }else return -ENOMEM ; - attr->type = type; attr->config = config;return 0 ;
tools/perf/util/parse-events.c +39 -0
@@ -28,6 +28,7 @@#include "util/parse-events-hybrid.h" #include "util/pmu-hybrid.h" #include "tracepoint.h" +#include "thread_map.h" #define MAX_NAME_LEN 100
@@ -157,6 +158,44 @@ struct event_symbol event_symbols_sw[PERF_COUNT_SW_MAX] = {#define PERF_EVENT_TYPE (config) __PERF_EVENT_FIELD(config, TYPE)#define PERF_EVENT_ID (config) __PERF_EVENT_FIELD(config, EVENT) +bool is_event_supported (u8 type,u64 config) +{ +bool ret = true; +int open_return; +struct evsel *evsel; +struct perf_event_attr attr = { + .type = type, + .config = config, + .disabled =1 , + }; +struct perf_thread_map *tmap =thread_map__new_by_tid (0 ); + +if (tmap ==NULL ) +return false; + + evsel =evsel__new (&attr); +if (evsel) { + open_return =evsel__open (evsel,NULL , tmap); + ret = open_return >=0 ; + +if (open_return == -EACCES ) { +/* +* This happens if the paranoid value +* /proc/sys/kernel/perf_event_paranoid is set to 2 +* Re-run with exclude_kernel set; we don't do that +* by default as some ARM machines do not support it. +* +*/ + evsel->core.attr.exclude_kernel =1 ; + ret =evsel__open (evsel,NULL , tmap) >=0 ; + } +evsel__delete (evsel); + } + +perf_thread_map__put (tmap); +return ret; +} +const char *event_type (int type) {switch (type) {
tools/perf/util/parse-events.h +1 -0
@@ -19,6 +19,7 @@ struct option;struct perf_pmu ;bool have_tracepoints (struct list_head *evlist); +bool is_event_supported (u8 type,u64 config);const char *event_type (int type);
tools/perf/util/print-events.c +0 -39
@@ -22,7 +22,6 @@#include "probe-file.h" #include "string2.h" #include "strlist.h" -#include "thread_map.h" #include "tracepoint.h" #include "pfm.h" #include "pmu-hybrid.h"
@@ -239,44 +238,6 @@ void print_sdt_events(const char *subsys_glob, const char *event_glob,strlist__delete (sdtlist); } -static bool is_event_supported (u8 type,u64 config) -{ -bool ret = true; -int open_return; -struct evsel *evsel; -struct perf_event_attr attr = { - .type = type, - .config = config, - .disabled =1 , - }; -struct perf_thread_map *tmap =thread_map__new_by_tid (0 ); - -if (tmap ==NULL ) -return false; - - evsel =evsel__new (&attr); -if (evsel) { - open_return =evsel__open (evsel,NULL , tmap); - ret = open_return >=0 ; - -if (open_return == -EACCES ) { -/* -* This happens if the paranoid value -* /proc/sys/kernel/perf_event_paranoid is set to 2 -* Re-run with exclude_kernel set; we don't do that -* by default as some ARM machines do not support it. -* -*/ - evsel->core.attr.exclude_kernel =1 ; - ret =evsel__open (evsel,NULL , tmap) >=0 ; - } -evsel__delete (evsel); - } - -perf_thread_map__put (tmap); -return ret; -} -int print_hwcache_events (const char *event_glob,bool name_only) {unsigned int type, op, i, evt_i =0 , evt_num =0 , npmus =0 ;
Signature · 1 line
2.25.1
Re: [PATCH v2 2/2] perf parse-events: Remove "not supported" hybrid cache events
Quoted 36 lines · On Thu, Sep 22, 2022 at 7:58 PM <[email protected]> wrote:
From: Zhengjun Xing <[email protected]>
By default, we create two hybrid cache events, one is for
cpu_core, and another is forcpu_atom. But Some hybrid hardware cache events are only available on one CPU PMU. For example, the 'L1-dcache-load-misses' is only available oncpu_core, while the 'L1-icache-loads' is only available oncpu_atom. We need to remove "not supported" hybrid cache events. By extendingis_event_supported()to global API and using it to check if the hybrid cache events are supported before being created, we can remove the "not supported" hybrid cache events.Before:
# ./perf stat -e L1-dcache-load-misses,L1-icache-loads -a sleep 1Performance counter stats for 'system wide':
52,570 cpu_core/L1-dcache-load-misses/ <not supported> cpu_atom/L1-dcache-load-misses/ <not supported> cpu_core/L1-icache-loads/ 1,471,817 cpu_atom/L1-icache-loads/1.004915229 seconds time elapsedAfter:
# ./perf stat -e L1-dcache-load-misses,L1-icache-loads -a sleep 1Performance counter stats for 'system wide':
54,510 cpu_core/L1-dcache-load-misses/ 1,441,286 cpu_atom/L1-icache-loads/ 1.005114281 seconds time elapsed
Quoted 4 lines
Fixes 30def61f64ba ("perf parse-events: Create two hybrid cache events")Signed-off-by Zhengjun Xing <[email protected]>Reviewed-by Kan Liang <[email protected]>Reported-by Yi Ammy <[email protected]>
Thanks,
Ian
Quoted 186 lines
---
Change log:
v2:
* Adds a comment for removing "not supported" hybrid cache events.
* Remove goto and add a strdup check
* "is_event_supported" move toparse-events.cper Ian's suggestion.
* Adds Reported-by from Yi Ammy <[email protected]>tools/perf/util/parse-events-hybrid.c | 21 ++++++++++++--- tools/perf/util/parse-events.c | 39 +++++++++++++++++++++++++++ tools/perf/util/parse-events.h | 1 + tools/perf/util/print-events.c | 39 --------------------------- 4 files changed, 57 insertions(+), 43 deletions(-)tools/perf/util/parse-events-hybrid.c +17
-4@@ -33,7 +33,8 @@ static void config_hybrid_attr(struct perf_event_attr *attr,* If the PMU type ID is 0, the PERF_TYPE_RAW will be applied. */ attr->type = type; - attr->config = attr->config | ((__u64 )pmu_type <<PERF_PMU_TYPE_SHIFT ); + attr->config = (attr->config &PERF_HW_EVENT_MASK ) | + ((__u64 )pmu_type <<PERF_PMU_TYPE_SHIFT ); } static int create_event_hybrid(__u32 config_type, int *idx,@@ -48,13 +49,25 @@ static int create_event_hybrid(__u32 config_type, int *idx,__u64 config = attr->config;config_hybrid_attr (attr, config_type, pmu->type); + +/* +* Some hybrid hardware cache events are only available on one CPU +* PMU. For example, the 'L1-dcache-load-misses' is only available +* on cpu_core, while the 'L1-icache-loads' is only available on +* cpu_atom. We need to remove "not supported" hybrid cache events. +*/ +if (attr->type ==PERF_TYPE_HW_CACHE + && !is_event_supported (attr->type, attr->config)) +return 0 ; + evsel =parse_events__add_event_hybrid (list, idx, attr, name, metric_id, pmu, config_terms); -if (evsel) +if (evsel) { evsel->pmu_name =strdup (pmu->name); -else +if (!evsel->pmu_name) +return -ENOMEM ; + }else return -ENOMEM ; - attr->type = type; attr->config = config;return 0 ;tools/perf/util/parse-events.c +39
-0@@ -28,6 +28,7 @@#include "util/parse-events-hybrid.h" #include "util/pmu-hybrid.h" #include "tracepoint.h" +#include "thread_map.h" #define MAX_NAME_LEN 100@@ -157,6 +158,44 @@ struct event_symbol event_symbols_sw[PERF_COUNT_SW_MAX] = {#define PERF_EVENT_TYPE (config) __PERF_EVENT_FIELD(config, TYPE)#define PERF_EVENT_ID (config) __PERF_EVENT_FIELD(config, EVENT) +bool is_event_supported (u8 type,u64 config) +{ +bool ret = true; +int open_return; +struct evsel *evsel; +struct perf_event_attr attr = { + .type = type, + .config = config, + .disabled =1 , + }; +struct perf_thread_map *tmap =thread_map__new_by_tid (0 ); + +if (tmap ==NULL ) +return false; + + evsel =evsel__new (&attr); +if (evsel) { + open_return =evsel__open (evsel,NULL , tmap); + ret = open_return >=0 ; + +if (open_return == -EACCES ) { +/* +* This happens if the paranoid value +* /proc/sys/kernel/perf_event_paranoid is set to 2 +* Re-run with exclude_kernel set; we don't do that +* by default as some ARM machines do not support it. +* +*/ + evsel->core.attr.exclude_kernel =1 ; + ret =evsel__open (evsel,NULL , tmap) >=0 ; + } +evsel__delete (evsel); + } + +perf_thread_map__put (tmap); +return ret; +} +const char *event_type (int type) {switch (type) {tools/perf/util/parse-events.h +1
-0@@ -19,6 +19,7 @@ struct option;struct perf_pmu ;bool have_tracepoints (struct list_head *evlist); +bool is_event_supported (u8 type,u64 config);const char *event_type (int type);tools/perf/util/print-events.c +0
-40@@ -22,7 +22,6 @@#include "probe-file.h" #include "string2.h" #include "strlist.h" -#include "thread_map.h" #include "tracepoint.h" #include "pfm.h" #include "pmu-hybrid.h" @@ -239,44 +238,6 @@ void print_sdt_events(const char *subsys_glob, const char *event_glob,strlist__delete (sdtlist); } -static bool is_event_supported (u8 type,u64 config) -{ -bool ret = true; -int open_return; -struct evsel *evsel; -struct perf_event_attr attr = { - .type = type, - .config = config, - .disabled =1 , - }; -struct perf_thread_map *tmap =thread_map__new_by_tid (0 ); - -if (tmap ==NULL ) -return false; - - evsel =evsel__new (&attr); -if (evsel) { - open_return =evsel__open (evsel,NULL , tmap); - ret = open_return >=0 ; - -if (open_return == -EACCES ) { -/* -* This happens if the paranoid value -* /proc/sys/kernel/perf_event_paranoid is set to 2 -* Re-run with exclude_kernel set; we don't do that -* by default as some ARM machines do not support it. -* -*/ - evsel->core.attr.exclude_kernel =1 ; - ret =evsel__open (evsel,NULL , tmap) >=0 ; - } -evsel__delete (evsel); - } - -perf_thread_map__put (tmap); -return ret; -} -int print_hwcache_events (const char *event_glob,bool name_only) {unsigned int type, op, i, evt_i =0 , evt_num =0 , npmus =0 ; --2.25.1
Re: [PATCH v2 2/2] perf parse-events: Remove "not supported" hybrid cache events
Em Fri, Sep 23, 2022 at 09:55:16AM -0700, Ian Rogers escreveu:
Quoted 35 lines · On Thu, Sep 22, 2022 at 7:58 PM <[email protected]> wrote:
Quoted 34 lines
From: Zhengjun Xing <[email protected]>
By default, we create two hybrid cache events, one is for
cpu_core, and another is forcpu_atom. But Some hybrid hardware cache events are only available on one CPU PMU. For example, the 'L1-dcache-load-misses' is only available oncpu_core, while the 'L1-icache-loads' is only available oncpu_atom. We need to remove "not supported" hybrid cache events. By extendingis_event_supported()to global API and using it to check if the hybrid cache events are supported before being created, we can remove the "not supported" hybrid cache events.Before:
# ./perf stat -e L1-dcache-load-misses,L1-icache-loads -a sleep 1Performance counter stats for 'system wide':
52,570 cpu_core/L1-dcache-load-misses/ <not supported> cpu_atom/L1-dcache-load-misses/ <not supported> cpu_core/L1-icache-loads/ 1,471,817 cpu_atom/L1-icache-loads/1.004915229 seconds time elapsedAfter:
# ./perf stat -e L1-dcache-load-misses,L1-icache-loads -a sleep 1Performance counter stats for 'system wide':
54,510 cpu_core/L1-dcache-load-misses/ 1,441,286 cpu_atom/L1-icache-loads/
Quoted 8 lines
Quoted 6 lines
1.005114281 seconds time elapsedFixes 30def61f64ba ("perf parse-events: Create two hybrid cache events")Signed-off-by Zhengjun Xing <[email protected]>Reviewed-by Kan Liang <[email protected]>Reported-by Yi Ammy <[email protected]>Acked-by Ian Rogers <[email protected]>
Thanks, applied.
- Arnaldo
Quoted 189 lines
Thanks,
IanQuoted 186 lines
---
Change log:
v2:
* Adds a comment for removing "not supported" hybrid cache events.
* Remove goto and add a strdup check
* "is_event_supported" move toparse-events.cper Ian's suggestion.
* Adds Reported-by from Yi Ammy <[email protected]>tools/perf/util/parse-events-hybrid.c | 21 ++++++++++++--- tools/perf/util/parse-events.c | 39 +++++++++++++++++++++++++++ tools/perf/util/parse-events.h | 1 + tools/perf/util/print-events.c | 39 --------------------------- 4 files changed, 57 insertions(+), 43 deletions(-)tools/perf/util/parse-events-hybrid.c +17
-4@@ -33,7 +33,8 @@ static void config_hybrid_attr(struct perf_event_attr *attr,* If the PMU type ID is 0, the PERF_TYPE_RAW will be applied. */ attr->type = type; - attr->config = attr->config | ((__u64 )pmu_type <<PERF_PMU_TYPE_SHIFT ); + attr->config = (attr->config &PERF_HW_EVENT_MASK ) | + ((__u64 )pmu_type <<PERF_PMU_TYPE_SHIFT ); } static int create_event_hybrid(__u32 config_type, int *idx,@@ -48,13 +49,25 @@ static int create_event_hybrid(__u32 config_type, int *idx,__u64 config = attr->config;config_hybrid_attr (attr, config_type, pmu->type); + +/* +* Some hybrid hardware cache events are only available on one CPU +* PMU. For example, the 'L1-dcache-load-misses' is only available +* on cpu_core, while the 'L1-icache-loads' is only available on +* cpu_atom. We need to remove "not supported" hybrid cache events. +*/ +if (attr->type ==PERF_TYPE_HW_CACHE + && !is_event_supported (attr->type, attr->config)) +return 0 ; + evsel =parse_events__add_event_hybrid (list, idx, attr, name, metric_id, pmu, config_terms); -if (evsel) +if (evsel) { evsel->pmu_name =strdup (pmu->name); -else +if (!evsel->pmu_name) +return -ENOMEM ; + }else return -ENOMEM ; - attr->type = type; attr->config = config;return 0 ;tools/perf/util/parse-events.c +39
-0@@ -28,6 +28,7 @@#include "util/parse-events-hybrid.h" #include "util/pmu-hybrid.h" #include "tracepoint.h" +#include "thread_map.h" #define MAX_NAME_LEN 100@@ -157,6 +158,44 @@ struct event_symbol event_symbols_sw[PERF_COUNT_SW_MAX] = {#define PERF_EVENT_TYPE (config) __PERF_EVENT_FIELD(config, TYPE)#define PERF_EVENT_ID (config) __PERF_EVENT_FIELD(config, EVENT) +bool is_event_supported (u8 type,u64 config) +{ +bool ret = true; +int open_return; +struct evsel *evsel; +struct perf_event_attr attr = { + .type = type, + .config = config, + .disabled =1 , + }; +struct perf_thread_map *tmap =thread_map__new_by_tid (0 ); + +if (tmap ==NULL ) +return false; + + evsel =evsel__new (&attr); +if (evsel) { + open_return =evsel__open (evsel,NULL , tmap); + ret = open_return >=0 ; + +if (open_return == -EACCES ) { +/* +* This happens if the paranoid value +* /proc/sys/kernel/perf_event_paranoid is set to 2 +* Re-run with exclude_kernel set; we don't do that +* by default as some ARM machines do not support it. +* +*/ + evsel->core.attr.exclude_kernel =1 ; + ret =evsel__open (evsel,NULL , tmap) >=0 ; + } +evsel__delete (evsel); + } + +perf_thread_map__put (tmap); +return ret; +} +const char *event_type (int type) {switch (type) {tools/perf/util/parse-events.h +1
-0@@ -19,6 +19,7 @@ struct option;struct perf_pmu ;bool have_tracepoints (struct list_head *evlist); +bool is_event_supported (u8 type,u64 config);const char *event_type (int type);tools/perf/util/print-events.c +0
-40@@ -22,7 +22,6 @@#include "probe-file.h" #include "string2.h" #include "strlist.h" -#include "thread_map.h" #include "tracepoint.h" #include "pfm.h" #include "pmu-hybrid.h" @@ -239,44 +238,6 @@ void print_sdt_events(const char *subsys_glob, const char *event_glob,strlist__delete (sdtlist); } -static bool is_event_supported (u8 type,u64 config) -{ -bool ret = true; -int open_return; -struct evsel *evsel; -struct perf_event_attr attr = { - .type = type, - .config = config, - .disabled =1 , - }; -struct perf_thread_map *tmap =thread_map__new_by_tid (0 ); - -if (tmap ==NULL ) -return false; - - evsel =evsel__new (&attr); -if (evsel) { - open_return =evsel__open (evsel,NULL , tmap); - ret = open_return >=0 ; - -if (open_return == -EACCES ) { -/* -* This happens if the paranoid value -* /proc/sys/kernel/perf_event_paranoid is set to 2 -* Re-run with exclude_kernel set; we don't do that -* by default as some ARM machines do not support it. -* -*/ - evsel->core.attr.exclude_kernel =1 ; - ret =evsel__open (evsel,NULL , tmap) >=0 ; - } -evsel__delete (evsel); - } - -perf_thread_map__put (tmap); -return ret; -} -int print_hwcache_events (const char *event_glob,bool name_only) {unsigned int type, op, i, evt_i =0 , evt_num =0 , npmus =0 ; --2.25.1
Signature · 1 line
- Arnaldo