perf print-events: Fix "perf list" can not display the PMU prefix for some hybrid cache events
PATCH 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]
--- 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 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--- tools/perf/util/parse-events-hybrid.c | 8 +++++++- tools/perf/util/print-events.c | 2 +- tools/perf/util/print-events.h | 3 ++- 3 files changed, 10 insertions(+), 3 deletions(-)
tools/perf/util/parse-events-hybrid.c +7 -1
@@ -14,6 +14,7 @@#include "pmu.h" #include "pmu-hybrid.h" #include "perf.h" +#include "print-events.h" static void config_hybrid_attr (struct perf_event_attr *attr,int type,int pmu_type)
@@ -48,13 +49,18 @@ static int create_event_hybrid(__u32 config_type, int *idx,__u64 config = attr->config;config_hybrid_attr (attr, config_type, pmu->type); + +if (attr->type ==PERF_TYPE_HW_CACHE + && !is_event_supported (attr->type, attr->config)) + goto out; + evsel =parse_events__add_event_hybrid (list, idx, attr, name, metric_id, pmu, config_terms);if (evsel) evsel->pmu_name =strdup (pmu->name);else return -ENOMEM ; - +out: attr->type = type; attr->config = config;return 0 ;
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,u64 config) +bool is_event_supported (u8 type,u64 config) {bool ret = true;int open_return;
tools/perf/util/print-events.h +2 -1
@@ -1,14 +1,15 @@/* SPDX-License-Identifier: GPL-2.0 */ #ifndef __PERF_PRINT_EVENTS_H#define __PERF_PRINT_EVENTS_H -#include <stdbool.h> +#include <linux/types.h> struct event_symbol ;void print_events (const char *event_glob,bool name_only,bool quiet_flag,bool long_desc,bool details_flag,bool deprecated,const char *pmu_name); +bool is_event_supported (u8 type,u64 config);int print_hwcache_events (const char *event_glob,bool name_only);void print_sdt_events (const char *subsys_glob,const char *event_glob,bool name_only);
Signature · 1 line
2.25.1
Ian Rogers
reviewerRe: [PATCH 2/2] perf parse-events: Remove "not supported" hybrid cache events
Quoted 58 lines · On Wed, Sep 21, 2022 at 6:47 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 elapsedFixes 30def61f64ba ("perf parse-events: Create two hybrid cache events")Signed-off-by Zhengjun Xing <[email protected]>Reviewed-by Kan Liang <[email protected]>--- tools/perf/util/parse-events-hybrid.c | 8 +++++++- tools/perf/util/print-events.c | 2 +- tools/perf/util/print-events.h | 3 ++- 3 files changed, 10 insertions(+), 3 deletions(-)tools/perf/util/parse-events-hybrid.c +1
-0@@ -14,6 +14,7 @@#include "pmu.h" #include "pmu-hybrid.h" #include "perf.h" +#include "print-events.h" static void config_hybrid_attr (struct perf_event_attr *attr,int type,int pmu_type)
tools/perf/util/parse-events-hybrid.c static int create_event_hybrid(__u32 config_type, int *idx, · line 55
Quoted 7 lines
tools/perf/util/parse-events-hybrid.c +4
-0@@ -48,13 +49,18 @@ static int create_event_hybrid(__u32 config_type, int *idx,__u64 config = attr->config;config_hybrid_attr (attr, config_type, pmu->type); + +if (attr->type ==PERF_TYPE_HW_CACHE + && !is_event_supported (attr->type, attr->config)) + goto out;
A comment to explain this would be useful.
tools/perf/util/parse-events-hybrid.c static int create_event_hybrid(__u32 config_type, int *idx, · line 62
Quoted 7 lines
tools/perf/util/parse-events-hybrid.c +1
-0+ evsel =parse_events__add_event_hybrid (list, idx, attr, name, metric_id, pmu, config_terms);if (evsel) evsel->pmu_name =strdup (pmu->name);else return -ENOMEM ;
For consistency should this use the "goto" pattern now? You can also handle the ENOMEM case for strdup.
Quoted 10 lines
tools/perf/util/parse-events-hybrid.c +1
-1- +out: attr->type = type; attr->config = config;return 0 ;tools/perf/util/print-events.c +0
-0
tools/perf/util/print-events.c void print_sdt_events(const char *subsys_glob, const char *event_glob, · line 242
Quoted 5 lines
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,u64 config) +bool is_event_supported (u8 type,u64 config)
This makes me tempted to say this function should be in parse-events.c.
Thanks,
Ian
Quoted 27 lines
tools/perf/util/print-events.c +0
-0{bool ret = true;int open_return;tools/perf/util/print-events.h +2
-2@@ -1,14 +1,15 @@/* SPDX-License-Identifier: GPL-2.0 */ #ifndef __PERF_PRINT_EVENTS_H#define __PERF_PRINT_EVENTS_H -#include <stdbool.h> +#include <linux/types.h> struct event_symbol ;void print_events (const char *event_glob,bool name_only,bool quiet_flag,bool long_desc,bool details_flag,bool deprecated,const char *pmu_name); +bool is_event_supported (u8 type,u64 config);int print_hwcache_events (const char *event_glob,bool name_only);void print_sdt_events (const char *subsys_glob,const char *event_glob,bool name_only); --2.25.1
Re: [PATCH 2/2] perf parse-events: Remove "not supported" hybrid cache events
Quoted 61 lines · Ian Rogers
Quoted 60 lines · On Wed, Sep 21, 2022 at 6:47 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 elapsedFixes 30def61f64ba ("perf parse-events: Create two hybrid cache events")Signed-off-by Zhengjun Xing <[email protected]>Reviewed-by Kan Liang <[email protected]>--- tools/perf/util/parse-events-hybrid.c | 8 +++++++- tools/perf/util/print-events.c | 2 +- tools/perf/util/print-events.h | 3 ++- 3 files changed, 10 insertions(+), 3 deletions(-)tools/perf/util/parse-events-hybrid.c +1
-0@@ -14,6 +14,7 @@#include "pmu.h" #include "pmu-hybrid.h" #include "perf.h" +#include "print-events.h" static void config_hybrid_attr (struct perf_event_attr *attr,int type,int pmu_type)@@ -48,13 +49,18 @@ static int create_event_hybrid(__u32 config_type, int *idx,__u64 config = attr->config;
Quoted 7 lines
tools/perf/util/parse-events-hybrid.c static int create_event_hybrid(__u32 config_type, int *idx, · line 55
Quoted 5 lines
tools/perf/util/parse-events-hybrid.c +4
-0config_hybrid_attr (attr, config_type, pmu->type); + +if (attr->type ==PERF_TYPE_HW_CACHE + && !is_event_supported (attr->type, attr->config)) + goto out;A comment to explain this would be useful.
Thanks, I will add a comment in the next version.
Quoted 9 lines
tools/perf/util/parse-events-hybrid.c static int create_event_hybrid(__u32 config_type, int *idx, · line 62
Quoted 8 lines
tools/perf/util/parse-events-hybrid.c +1
-0+ evsel =parse_events__add_event_hybrid (list, idx, attr, name, metric_id, pmu, config_terms);if (evsel) evsel->pmu_name =strdup (pmu->name);else return -ENOMEM ;
Quoted 3 lines
For consistency should this use the "goto" pattern now? You can also handle the ENOMEM case for strdup.
Yes, I will add a strdup check in the next version.
Quoted 13 lines
Quoted 13 lines
tools/perf/util/parse-events-hybrid.c +1
-1- +out: attr->type = type; attr->config = config;return 0 ;tools/perf/util/print-events.c +0
-0@@ -239,7 +239,7 @@ void print_sdt_events(const char *subsys_glob, const char *event_glob,strlist__delete (sdtlist); }
Quoted 5 lines
tools/perf/util/print-events.c void print_sdt_events(const char *subsys_glob, const char *event_glob, · line 242
Quoted 2 lines
tools/perf/util/print-events.c +1
-1-static bool is_event_supported (u8 type,u64 config) +bool is_event_supported (u8 type,u64 config)This makes me tempted to say this function should be in
parse-events.c.
"is_event_supported" move to parse-events.c should be better, I will do it in the next version.
Quoted 30 lines
Thanks,
IanQuoted 27 lines
tools/perf/util/print-events.c +0
-0{bool ret = true;int open_return;tools/perf/util/print-events.h +2
-2@@ -1,14 +1,15 @@/* SPDX-License-Identifier: GPL-2.0 */ #ifndef __PERF_PRINT_EVENTS_H#define __PERF_PRINT_EVENTS_H -#include <stdbool.h> +#include <linux/types.h> struct event_symbol ;void print_events (const char *event_glob,bool name_only,bool quiet_flag,bool long_desc,bool details_flag,bool deprecated,const char *pmu_name); +bool is_event_supported (u8 type,u64 config);int print_hwcache_events (const char *event_glob,bool name_only);void print_sdt_events (const char *subsys_glob,const char *event_glob,bool name_only); --2.25.1
Signature · 1 line
Zhengjun Xing
Quoted 55 lines · On Wed, Sep 21, 2022 at 6:47 PM <[email protected]> wrote:
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 APIis_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]
Quoted 3 lines
Fixes 9b7c7728f4e4 ("perf parse-events: Break out tracepoint and printing")Signed-off-by Zhengjun Xing <[email protected]>Reviewed-by Kan Liang <[email protected]>
Sorry for this breakage, I suspect that a long review on the refactoring CL meant that I missed the intervening fix. Can we add a test on this? It would need to be hybrid specific and skip otherwise.
Thanks,
Ian
Quoted 20 lines
--- tools/perf/util/print-events.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)tools/perf/util/print-events.c +1
-2@@ -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; --2.25.1
Quoted 60 lines · Ian Rogers
Quoted 55 lines · On Wed, Sep 21, 2022 at 6:47 PM <[email protected]> wrote:
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 APIis_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]Quoted 3 lines
Fixes 9b7c7728f4e4 ("perf parse-events: Break out tracepoint and printing")Signed-off-by Zhengjun Xing <[email protected]>Reviewed-by Kan Liang <[email protected]>
Quoted 5 lines
Acked-by Ian Rogers <[email protected]>Sorry for this breakage, I suspect that a long review on the refactoring CL meant that I missed the intervening fix. Can we add a test on this? It would need to be hybrid specific and skip otherwise.
Thanks. I will add a test case for hybrid-specific later.
Quoted 24 lines
Thanks,
IanQuoted 20 lines
--- tools/perf/util/print-events.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)tools/perf/util/print-events.c +1
-2@@ -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; --2.25.1
Signature · 1 line
Zhengjun Xing