AlderLake uses a hybrid architecture utilizing Golden Cove cores
(core cpu) and Gracemont cores (atom cpu). Each cpu has dedicated
event list. Some events are available on core cpu, some events
are available on atom cpu and some events can be available on both.
Kernel exports new pmus "cpu_core" and "cpu_atom" through sysfs:
/sys/devices/cpu_core
/sys/devices/cpu_atom
cat /sys/devices/cpu_core/cpus
0-15
cat /sys/devices/cpu_atom/cpus
16-23
In this example, core cpus are 0-15 and atom cpus are 16-23.
To enable a core only event or atom only event:
cpu_core/<event name>/
or
cpu_atom/<event name>/
Count the 'cycles' event on core cpus.
# perf stat -e cpu_core/cycles/ -a -- sleep 1
Performance counter stats for 'system wide':
12,853,951,349 cpu_core/cycles/
1.002581249 seconds time elapsed
If one event is available on both atom cpu and core cpu, two events
are created automatically.
# perf stat -e cycles -a -- sleep 1
Performance counter stats for 'system wide':
12,856,467,438 cpu_core/cycles/
6,404,634,785 cpu_atom/cycles/
1.002453013 seconds time elapsed
Group is supported if the events are from same pmu, otherwise a warning
is displayed.
# perf stat -e '{cpu_core/cycles/,cpu_core/instructions/}' -a -- sleep 1
Performance counter stats for 'system wide':
12,863,866,968 cpu_core/cycles/
554,795,017 cpu_core/instructions/
1.002616117 seconds time elapsed
# perf stat -e '{cpu_core/cycles/,cpu_atom/instructions/}' -a -- sleep 1
WARNING: Group has events from different hybrid PMUs
Performance counter stats for 'system wide':
12,863,344,830 cpu_core/cycles/
<not supported> cpu_atom/instructions/
1.002568830 seconds time elapsed
Note that, since the whole patchset for AlderLake hybrid support is very
large (40+ patches), for simplicity, it's splitted into several patch
series.
The patch series 1 only supports the basic functionality. The other
supports for perf-c2c/perf-mem/topdown/metrics/topology header and etc
will be added in follow-up series.
v2:
---
- Drop kernel patches (Kan posted the series "Add Alder Lake support for perf (kernel)" separately).
- Drop the patches for perf-c2c/perf-mem/topdown/metrics/topology header supports,
which will be added in series 2 or series 3.
- Simplify the arguments of __perf_pmu__new_alias() by passing
the 'struct pme_event' pointer.
- Check sysfs validity before access.
- Use pmu style event name, such as "cpu_core/cycles/".
- Move command output two chars to the right.
- Move pmu hybrid functions to new created pmu-hybrid.c/pmu-hybrid.h.
This is to pass the perf test python case.
Jin Yao (27):
tools headers uapi: Update tools's copy of linux/perf_event.h
perf jevents: Support unit value "cpu_core" and "cpu_atom"
perf pmu: Simplify arguments of __perf_pmu__new_alias
perf pmu: Save pmu name
perf pmu: Save detected hybrid pmus to a global pmu list
perf pmu: Add hybrid helper functions
perf evlist: Hybrid event uses its own cpus
perf stat: Uniquify hybrid event name
perf parse-events: Create two hybrid hardware events
perf parse-events: Create two hybrid cache events
perf parse-events: Support hardware events inside PMU
perf parse-events: Support hybrid raw events
perf evlist: Create two hybrid 'cycles' events by default
perf stat: Add default hybrid events
perf stat: Filter out unmatched aggregation for hybrid event
perf evlist: Warn as events from different hybrid PMUs in a group
perf evsel: Adjust hybrid event and global event mixed group
perf script: Support PERF_TYPE_HARDWARE_PMU and PERF_TYPE_HW_CACHE_PMU
perf tests: Add hybrid cases for 'Parse event definition strings' test
perf tests: Add hybrid cases for 'Roundtrip evsel->name' test
perf tests: Skip 'Setup struct perf_event_attr' test for hybrid
perf tests: Support 'Track with sched_switch' test for hybrid
perf tests: Support 'Parse and process metrics' test for hybrid
perf tests: Support 'Session topology' test for hybrid
perf tests: Support 'Convert perf time to TSC' test for hybrid
perf tests: Skip 'perf stat metrics (shadow stat) test' for hybrid
perf Documentation: Document intel-hybrid support
tools/include/uapi/linux/perf_event.h | 26 ++
tools/perf/Documentation/intel-hybrid.txt | 228 +++++++++++++++++
tools/perf/Documentation/perf-record.txt | 1 +
tools/perf/Documentation/perf-stat.txt | 2 +
tools/perf/builtin-record.c | 13 +-
tools/perf/builtin-script.c | 24 ++
tools/perf/builtin-stat.c | 69 ++++-
tools/perf/pmu-events/jevents.c | 2 +
tools/perf/tests/attr.c | 5 +
tools/perf/tests/evsel-roundtrip-name.c | 19 +-
tools/perf/tests/parse-events.c | 171 +++++++++++++
tools/perf/tests/parse-metric.c | 11 +-
tools/perf/tests/perf-time-to-tsc.c | 16 ++
tools/perf/tests/shell/stat+shadow_stat.sh | 3 +
tools/perf/tests/switch-tracking.c | 10 +-
tools/perf/tests/topology.c | 10 +-
tools/perf/util/Build | 1 +
tools/perf/util/evlist.c | 148 ++++++++++-
tools/perf/util/evlist.h | 6 +
tools/perf/util/evsel.c | 111 ++++++++-
tools/perf/util/evsel.h | 10 +-
tools/perf/util/parse-events.c | 277 ++++++++++++++++++++-
tools/perf/util/parse-events.h | 6 +-
tools/perf/util/parse-events.y | 21 +-
tools/perf/util/pmu-hybrid.c | 75 ++++++
tools/perf/util/pmu-hybrid.h | 29 +++
tools/perf/util/pmu.c | 53 ++--
tools/perf/util/pmu.h | 5 +
tools/perf/util/python-ext-sources | 2 +
tools/perf/util/stat-display.c | 32 ++-
tools/perf/util/stat.h | 1 +
31 files changed, 1311 insertions(+), 76 deletions(-)
create mode 100644 tools/perf/Documentation/intel-hybrid.txt
create mode 100644 tools/perf/util/pmu-hybrid.c
create mode 100644 tools/perf/util/pmu-hybrid.h
--
2.17.1
For some Intel platforms, such as Alderlake, which is a hybrid platform
and it consists of atom cpu and core cpu. Each cpu has dedicated event
list. Part of events are available on core cpu, part of events are
available on atom cpu.
The kernel exports new cpu pmus: cpu_core and cpu_atom. The event in
json is added with a new field "Unit" to indicate which pmu the event
is available on.
For example, one event in cache.json,
{
"BriefDescription": "Counts the number of load ops retired that",
"CollectPEBSRecord": "2",
"Counter": "0,1,2,3",
"EventCode": "0xd2",
"EventName": "MEM_LOAD_UOPS_RETIRED_MISC.MMIO",
"PEBScounters": "0,1,2,3",
"SampleAfterValue": "1000003",
"UMask": "0x80",
"Unit": "cpu_atom"
},
The unit "cpu_atom" indicates this event is only availabe on "cpu_atom".
In generated pmu-events.c, we can see:
{
.name = "mem_load_uops_retired_misc.mmio",
.event = "period=1000003,umask=0x80,event=0xd2",
.desc = "Counts the number of load ops retired that. Unit: cpu_atom ",
.topic = "cache",
.pmu = "cpu_atom",
},
But if without this patch, the "uncore_" prefix is added before "cpu_atom",
such as:
.pmu = "uncore_cpu_atom"
That would be a wrong pmu.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/pmu-events/jevents.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/tools/perf/pmu-events/jevents.c b/tools/perf/pmu-events/jevents.c
index e1f3f5c8c550..b1a15f57c9ad 100644
--- a/tools/perf/pmu-events/jevents.c
+++ b/tools/perf/pmu-events/jevents.c
@@ -285,6 +285,8 @@ static struct map {
{ "imx8_ddr", "imx8_ddr" },
{ "L3PMC", "amd_l3" },
{ "DFPMC", "amd_df" },
+ { "cpu_core", "cpu_core" },
+ { "cpu_atom", "cpu_atom" },
{}
};
--
2.17.1
We identify the cpu_core pmu and cpu_atom pmu by explicitly
checking following files:
For cpu_core, checks:
"/sys/bus/event_source/devices/cpu_core/cpus"
For cpu_atom, checks:
"/sys/bus/event_source/devices/cpu_atom/cpus"
If the 'cpus' file exists, the pmu exists.
But in order not to hardcode the "cpu_core" and "cpu_atom",
and make the code in a generic way. So if the path
"/sys/bus/event_source/devices/cpu_xxx/cpus" exists, the hybrid
pmu exists. All the detected hybrid pmus are linked to a
global list 'perf_pmu__hybrid_pmus' and then next we just need
to iterate the list to get all hybrid pmu by using
perf_pmu__for_each_hybrid_pmu.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/util/Build | 1 +
tools/perf/util/pmu-hybrid.c | 35 +++++++++++++++++++++++++++++++++++
tools/perf/util/pmu-hybrid.h | 18 ++++++++++++++++++
tools/perf/util/pmu.c | 9 ++++++++-
tools/perf/util/pmu.h | 4 ++++
5 files changed, 66 insertions(+), 1 deletion(-)
create mode 100644 tools/perf/util/pmu-hybrid.c
create mode 100644 tools/perf/util/pmu-hybrid.h
diff --git a/tools/perf/util/Build b/tools/perf/util/Build
index e3e12f9d4733..37a8a63c7195 100644
--- a/tools/perf/util/Build
+++ b/tools/perf/util/Build
@@ -69,6 +69,7 @@ perf-y += parse-events-bison.o
perf-y += pmu.o
perf-y += pmu-flex.o
perf-y += pmu-bison.o
+perf-y += pmu-hybrid.o
perf-y += trace-event-read.o
perf-y += trace-event-info.o
perf-y += trace-event-scripting.o
diff --git a/tools/perf/util/pmu-hybrid.c b/tools/perf/util/pmu-hybrid.c
new file mode 100644
index 000000000000..7316bf46e54b
--- /dev/null
+++ b/tools/perf/util/pmu-hybrid.c
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/list.h>
+#include <linux/compiler.h>
+#include <linux/string.h>
+#include <linux/zalloc.h>
+#include <sys/types.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <sys/stat.h>
+#include <unistd.h>
+#include <stdio.h>
+#include <stdbool.h>
+#include <stdarg.h>
+#include <locale.h>
+#include <api/fs/fs.h>
+#include "fncache.h"
+#include "pmu-hybrid.h"
+
+LIST_HEAD(perf_pmu__hybrid_pmus);
+
+bool perf_pmu__hybrid_mounted(const char *name)
+{
+ char path[PATH_MAX];
+ const char *sysfs;
+
+ if (strncmp(name, "cpu_", 4))
+ return false;
+
+ sysfs = sysfs__mountpoint();
+ if (!sysfs)
+ return false;
+
+ snprintf(path, PATH_MAX, CPUS_TEMPLATE_CPU, sysfs, name);
+ return file_available(path);
+}
diff --git a/tools/perf/util/pmu-hybrid.h b/tools/perf/util/pmu-hybrid.h
new file mode 100644
index 000000000000..35bed3714438
--- /dev/null
+++ b/tools/perf/util/pmu-hybrid.h
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __PMU_HYBRID_H
+#define __PMU_HYBRID_H
+
+#include <linux/perf_event.h>
+#include <linux/compiler.h>
+#include <linux/list.h>
+#include <stdbool.h>
+#include "pmu.h"
+
+extern struct list_head perf_pmu__hybrid_pmus;
+
+#define perf_pmu__for_each_hybrid_pmu(pmu) \
+ list_for_each_entry(pmu, &perf_pmu__hybrid_pmus, hybrid_list)
+
+bool perf_pmu__hybrid_mounted(const char *name);
+
+#endif /* __PMU_HYBRID_H */
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 45d8db1af8d2..08280a3e45a8 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -25,6 +25,7 @@
#include "string2.h"
#include "strbuf.h"
#include "fncache.h"
+#include "pmu-hybrid.h"
struct perf_pmu perf_pmu__fake;
@@ -613,7 +614,6 @@ static struct perf_cpu_map *__pmu_cpumask(const char *path)
*/
#define SYS_TEMPLATE_ID "./bus/event_source/devices/%s/identifier"
#define CPUS_TEMPLATE_UNCORE "%s/bus/event_source/devices/%s/cpumask"
-#define CPUS_TEMPLATE_CPU "%s/bus/event_source/devices/%s/cpus"
static struct perf_cpu_map *pmu_cpumask(const char *name)
{
@@ -645,6 +645,9 @@ static bool pmu_is_uncore(const char *name)
char path[PATH_MAX];
const char *sysfs;
+ if (perf_pmu__hybrid_mounted(name))
+ return false;
+
sysfs = sysfs__mountpoint();
snprintf(path, PATH_MAX, CPUS_TEMPLATE_UNCORE, sysfs, name);
return file_available(path);
@@ -946,6 +949,7 @@ static struct perf_pmu *pmu_lookup(const char *name)
pmu->is_uncore = pmu_is_uncore(name);
if (pmu->is_uncore)
pmu->id = pmu_id(name);
+ pmu->is_hybrid = perf_pmu__hybrid_mounted(name);
pmu->max_precise = pmu_max_precise(name);
pmu_add_cpu_aliases(&aliases, pmu);
pmu_add_sys_aliases(&aliases, pmu);
@@ -957,6 +961,9 @@ static struct perf_pmu *pmu_lookup(const char *name)
list_splice(&aliases, &pmu->aliases);
list_add_tail(&pmu->list, &pmus);
+ if (pmu->is_hybrid)
+ list_add_tail(&pmu->hybrid_list, &perf_pmu__hybrid_pmus);
+
pmu->default_config = perf_pmu__get_default_config(pmu);
return pmu;
diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
index 0e724d5b84c6..3b9b4def6032 100644
--- a/tools/perf/util/pmu.h
+++ b/tools/perf/util/pmu.h
@@ -5,6 +5,7 @@
#include <linux/bitmap.h>
#include <linux/compiler.h>
#include <linux/perf_event.h>
+#include <linux/list.h>
#include <stdbool.h>
#include "parse-events.h"
#include "pmu-events/pmu-events.h"
@@ -19,6 +20,7 @@ enum {
#define PERF_PMU_FORMAT_BITS 64
#define EVENT_SOURCE_DEVICE_PATH "/bus/event_source/devices/"
+#define CPUS_TEMPLATE_CPU "%s/bus/event_source/devices/%s/cpus"
struct perf_event_attr;
@@ -34,6 +36,7 @@ struct perf_pmu {
__u32 type;
bool selectable;
bool is_uncore;
+ bool is_hybrid;
bool auxtrace;
int max_precise;
struct perf_event_attr *default_config;
@@ -42,6 +45,7 @@ struct perf_pmu {
struct list_head aliases; /* HEAD struct perf_pmu_alias -> list */
struct list_head caps; /* HEAD struct perf_pmu_caps -> list */
struct list_head list; /* ELEM */
+ struct list_head hybrid_list;
};
extern struct perf_pmu perf_pmu__fake;
--
2.17.1
The functions perf_pmu__is_hybrid and perf_pmu__find_hybrid_pmu
can be used to identify the hybrid platform and return the found
hybrid cpu pmu. All the detected hybrid pmus have been saved in
'perf_pmu__hybrid_pmus' list. So we just need to search this list.
perf_pmu__hybrid_type_to_pmu converts the user specified string
to hybrid pmu name. This is used to support the '--cputype' option
in next patches.
perf_pmu__hybrid_exist checks the existing of hybrid pmu.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/util/pmu-hybrid.c | 40 ++++++++++++++++++++++++++++++++++++
tools/perf/util/pmu-hybrid.h | 11 ++++++++++
2 files changed, 51 insertions(+)
diff --git a/tools/perf/util/pmu-hybrid.c b/tools/perf/util/pmu-hybrid.c
index 7316bf46e54b..86ba84d9469c 100644
--- a/tools/perf/util/pmu-hybrid.c
+++ b/tools/perf/util/pmu-hybrid.c
@@ -33,3 +33,43 @@ bool perf_pmu__hybrid_mounted(const char *name)
snprintf(path, PATH_MAX, CPUS_TEMPLATE_CPU, sysfs, name);
return file_available(path);
}
+
+struct perf_pmu *perf_pmu__find_hybrid_pmu(const char *name)
+{
+ struct perf_pmu *pmu;
+
+ if (!name)
+ return NULL;
+
+ perf_pmu__for_each_hybrid_pmu(pmu) {
+ if (!strcmp(name, pmu->name))
+ return pmu;
+ }
+
+ return NULL;
+}
+
+bool perf_pmu__is_hybrid(const char *name)
+{
+ return perf_pmu__find_hybrid_pmu(name) != NULL;
+}
+
+char *perf_pmu__hybrid_type_to_pmu(const char *type)
+{
+ char *pmu_name = NULL;
+
+ if (asprintf(&pmu_name, "cpu_%s", type) < 0)
+ return NULL;
+
+ if (perf_pmu__is_hybrid(pmu_name))
+ return pmu_name;
+
+ /*
+ * pmu may be not scanned, check the sysfs.
+ */
+ if (perf_pmu__hybrid_mounted(pmu_name))
+ return pmu_name;
+
+ free(pmu_name);
+ return NULL;
+}
diff --git a/tools/perf/util/pmu-hybrid.h b/tools/perf/util/pmu-hybrid.h
index 35bed3714438..7fb2246e939a 100644
--- a/tools/perf/util/pmu-hybrid.h
+++ b/tools/perf/util/pmu-hybrid.h
@@ -15,4 +15,15 @@ extern struct list_head perf_pmu__hybrid_pmus;
bool perf_pmu__hybrid_mounted(const char *name);
+struct perf_pmu *perf_pmu__find_hybrid_pmu(const char *name);
+
+bool perf_pmu__is_hybrid(const char *name);
+
+char *perf_pmu__hybrid_type_to_pmu(const char *type);
+
+static inline bool perf_pmu__hybrid_exist(void)
+{
+ return !list_empty(&perf_pmu__hybrid_pmus);
+}
+
#endif /* __PMU_HYBRID_H */
--
2.17.1
Simplify the arguments of __perf_pmu__new_alias() by passing
the whole 'struct pme_event' pointer.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/util/pmu.c | 36 ++++++++++++++++--------------------
1 file changed, 16 insertions(+), 20 deletions(-)
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 44ef28302fc7..54e586bf19a5 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -306,18 +306,25 @@ static bool perf_pmu_merge_alias(struct perf_pmu_alias *newalias,
}
static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
- char *desc, char *val,
- char *long_desc, char *topic,
- char *unit, char *perpkg,
- char *metric_expr,
- char *metric_name,
- char *deprecated)
+ char *desc, char *val, struct pmu_event *pe)
{
struct parse_events_term *term;
struct perf_pmu_alias *alias;
int ret;
int num;
char newval[256];
+ char *long_desc = NULL, *topic = NULL, *unit = NULL, *perpkg = NULL,
+ *metric_expr = NULL, *metric_name = NULL, *deprecated = NULL;
+
+ if (pe) {
+ long_desc = (char *)pe->long_desc;
+ topic = (char *)pe->topic;
+ unit = (char *)pe->unit;
+ perpkg = (char *)pe->perpkg;
+ metric_expr = (char *)pe->metric_expr;
+ metric_name = (char *)pe->metric_name;
+ deprecated = (char *)pe->deprecated;
+ }
alias = malloc(sizeof(*alias));
if (!alias)
@@ -406,8 +413,7 @@ static int perf_pmu__new_alias(struct list_head *list, char *dir, char *name, FI
/* Remove trailing newline from sysfs file */
strim(buf);
- return __perf_pmu__new_alias(list, dir, name, NULL, buf, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL);
+ return __perf_pmu__new_alias(list, dir, name, NULL, buf, NULL);
}
static inline bool pmu_alias_info_file(char *name)
@@ -793,11 +799,7 @@ void pmu_add_cpu_aliases_map(struct list_head *head, struct perf_pmu *pmu,
/* need type casts to override 'const' */
__perf_pmu__new_alias(head, NULL, (char *)pe->name,
(char *)pe->desc, (char *)pe->event,
- (char *)pe->long_desc, (char *)pe->topic,
- (char *)pe->unit, (char *)pe->perpkg,
- (char *)pe->metric_expr,
- (char *)pe->metric_name,
- (char *)pe->deprecated);
+ pe);
}
}
@@ -864,13 +866,7 @@ static int pmu_add_sys_aliases_iter_fn(struct pmu_event *pe, void *data)
(char *)pe->name,
(char *)pe->desc,
(char *)pe->event,
- (char *)pe->long_desc,
- (char *)pe->topic,
- (char *)pe->unit,
- (char *)pe->perpkg,
- (char *)pe->metric_expr,
- (char *)pe->metric_name,
- (char *)pe->deprecated);
+ pe);
}
return 0;
--
2.17.1
On hybrid platform, one event is available on one pmu
(such as, available on cpu_core or on cpu_atom).
This patch saves the pmu name to the pmu field of struct perf_pmu_alias.
Then next we can know the pmu which the event can be available on.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/util/pmu.c | 10 +++++++++-
tools/perf/util/pmu.h | 1 +
2 files changed, 10 insertions(+), 1 deletion(-)
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 54e586bf19a5..45d8db1af8d2 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -283,6 +283,7 @@ void perf_pmu_free_alias(struct perf_pmu_alias *newalias)
zfree(&newalias->str);
zfree(&newalias->metric_expr);
zfree(&newalias->metric_name);
+ zfree(&newalias->pmu);
parse_events_terms__purge(&newalias->terms);
free(newalias);
}
@@ -297,6 +298,10 @@ static bool perf_pmu_merge_alias(struct perf_pmu_alias *newalias,
list_for_each_entry(a, alist, list) {
if (!strcasecmp(newalias->name, a->name)) {
+ if (newalias->pmu && a->pmu &&
+ !strcasecmp(newalias->pmu, a->pmu)) {
+ continue;
+ }
perf_pmu_update_alias(a, newalias);
perf_pmu_free_alias(newalias);
return true;
@@ -314,7 +319,8 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
int num;
char newval[256];
char *long_desc = NULL, *topic = NULL, *unit = NULL, *perpkg = NULL,
- *metric_expr = NULL, *metric_name = NULL, *deprecated = NULL;
+ *metric_expr = NULL, *metric_name = NULL, *deprecated = NULL,
+ *pmu = NULL;
if (pe) {
long_desc = (char *)pe->long_desc;
@@ -324,6 +330,7 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
metric_expr = (char *)pe->metric_expr;
metric_name = (char *)pe->metric_name;
deprecated = (char *)pe->deprecated;
+ pmu = (char *)pe->pmu;
}
alias = malloc(sizeof(*alias));
@@ -389,6 +396,7 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
}
alias->per_pkg = perpkg && sscanf(perpkg, "%d", &num) == 1 && num == 1;
alias->str = strdup(newval);
+ alias->pmu = pmu ? strdup(pmu) : NULL;
if (deprecated)
alias->deprecated = true;
diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
index 8164388478c6..0e724d5b84c6 100644
--- a/tools/perf/util/pmu.h
+++ b/tools/perf/util/pmu.h
@@ -72,6 +72,7 @@ struct perf_pmu_alias {
bool deprecated;
char *metric_expr;
char *metric_name;
+ char *pmu;
};
struct perf_pmu *perf_pmu__find(const char *name);
--
2.17.1
It would be useful to tell user the pmu which the event belongs to.
perf-stat has supported '--no-merge' option and it can print the pmu
name after the event name, such as:
"cycles [cpu_core]"
Now this option is enabled by default for hybrid platform but change
the format to:
"cpu_core/cycles/"
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/builtin-stat.c | 3 +++
tools/perf/util/stat-display.c | 12 ++++++++++--
2 files changed, 13 insertions(+), 2 deletions(-)
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 68ecf68699a9..6c0a21323814 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -2390,6 +2390,9 @@ int cmd_stat(int argc, const char **argv)
evlist__check_cpu_maps(evsel_list);
+ if (perf_pmu__hybrid_exist())
+ stat_config.no_merge = true;
+
/*
* Initialize thread_map with comm names,
* so we could print it out on output.
diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c
index 7f09cdaf5b60..ed37d8e7ea1a 100644
--- a/tools/perf/util/stat-display.c
+++ b/tools/perf/util/stat-display.c
@@ -526,6 +526,7 @@ static void uniquify_event_name(struct evsel *counter)
{
char *new_name;
char *config;
+ int ret;
if (counter->uniquified_name ||
!counter->pmu_name || !strncmp(counter->name, counter->pmu_name,
@@ -540,8 +541,15 @@ static void uniquify_event_name(struct evsel *counter)
counter->name = new_name;
}
} else {
- if (asprintf(&new_name,
- "%s [%s]", counter->name, counter->pmu_name) > 0) {
+ if (perf_pmu__hybrid_exist()) {
+ ret = asprintf(&new_name, "%s/%s/",
+ counter->pmu_name, counter->name);
+ } else {
+ ret = asprintf(&new_name, "%s [%s]",
+ counter->name, counter->pmu_name);
+ }
+
+ if (ret) {
free(counter->name);
counter->name = new_name;
}
--
2.17.1
On hybrid platform, atom events can be only enabled on atom CPUs. Core
events can be only enabled on core CPUs. So for a hybrid event, it can
be only enabled on it's own CPUs.
But the problem for current perf is, the cpus for evsel (via PMU sysfs)
have been merged to evsel_list->core.all_cpus. It might be all CPUs.
So we need to figure out one way to let the hybrid event only use it's
own CPUs.
The idea is to create a new evlist__invalidate_all_cpus to invalidate
the evsel_list->core.all_cpus then evlist__for_each_cpu returns cpu -1
for hybrid evsel. If cpu is -1, hybrid evsel will use it's own cpus.
We will see following code piece in patch.
if (cpu == -1 && !evlist->thread_mode)
evsel__enable_cpus(pos);
It lets the event be enabled on event's own cpus.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/builtin-stat.c | 37 ++++++++++++++-
tools/perf/util/evlist.c | 72 ++++++++++++++++++++++++++++--
tools/perf/util/evlist.h | 4 ++
tools/perf/util/evsel.h | 8 ++++
tools/perf/util/python-ext-sources | 2 +
5 files changed, 117 insertions(+), 6 deletions(-)
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 2e2e4a8345ea..68ecf68699a9 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -393,6 +393,18 @@ static int read_counter_cpu(struct evsel *counter, struct timespec *rs, int cpu)
return 0;
}
+static int read_counter_cpus(struct evsel *counter, struct timespec *rs)
+{
+ int cpu, nr_cpus, err = 0;
+ struct perf_cpu_map *cpus = evsel__cpus(counter);
+
+ nr_cpus = cpus ? cpus->nr : 1;
+ for (cpu = 0; cpu < nr_cpus; cpu++)
+ err = read_counter_cpu(counter, rs, cpu);
+
+ return err;
+}
+
static int read_affinity_counters(struct timespec *rs)
{
struct evsel *counter;
@@ -414,8 +426,14 @@ static int read_affinity_counters(struct timespec *rs)
if (evsel__cpu_iter_skip(counter, cpu))
continue;
if (!counter->err) {
- counter->err = read_counter_cpu(counter, rs,
- counter->cpu_iter - 1);
+ if (cpu == -1 && !evsel_list->thread_mode) {
+ counter->err = read_counter_cpus(counter, rs);
+ } else if (evsel_list->thread_mode) {
+ counter->err = read_counter_cpu(counter, rs, 0);
+ } else {
+ counter->err = read_counter_cpu(counter, rs,
+ counter->cpu_iter - 1);
+ }
}
}
}
@@ -781,6 +799,21 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
if (group)
evlist__set_leader(evsel_list);
+ /*
+ * On hybrid platform, the cpus for evsel (via PMU sysfs) have been
+ * merged to evsel_list->core.all_cpus. We use evlist__invalidate_all_cpus
+ * to invalidate the evsel_list->core.all_cpus then evlist__for_each_cpu
+ * returns cpu -1 for hybrid evsel. If cpu is -1, hybrid evsel will
+ * use it's own cpus.
+ */
+ if (evlist__has_hybrid_events(evsel_list)) {
+ evlist__invalidate_all_cpus(evsel_list);
+ if (!target__has_cpu(&target) ||
+ target__has_per_thread(&target)) {
+ evsel_list->thread_mode = true;
+ }
+ }
+
if (affinity__setup(&affinity) < 0)
return -1;
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 882cd1f721d9..3ee12fcd0c9f 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -381,7 +381,8 @@ bool evsel__cpu_iter_skip_no_inc(struct evsel *ev, int cpu)
bool evsel__cpu_iter_skip(struct evsel *ev, int cpu)
{
if (!evsel__cpu_iter_skip_no_inc(ev, cpu)) {
- ev->cpu_iter++;
+ if (cpu != -1)
+ ev->cpu_iter++;
return false;
}
return true;
@@ -410,6 +411,16 @@ static int evlist__is_enabled(struct evlist *evlist)
return false;
}
+static void evsel__disable_cpus(struct evsel *evsel)
+{
+ int cpu, nr_cpus;
+ struct perf_cpu_map *cpus = evsel__cpus(evsel);
+
+ nr_cpus = cpus ? cpus->nr : 1;
+ for (cpu = 0; cpu < nr_cpus; cpu++)
+ evsel__disable_cpu(evsel, cpu);
+}
+
static void __evlist__disable(struct evlist *evlist, char *evsel_name)
{
struct evsel *pos;
@@ -436,7 +447,12 @@ static void __evlist__disable(struct evlist *evlist, char *evsel_name)
has_imm = true;
if (pos->immediate != imm)
continue;
- evsel__disable_cpu(pos, pos->cpu_iter - 1);
+ if (cpu == -1 && !evlist->thread_mode)
+ evsel__disable_cpus(pos);
+ else if (evlist->thread_mode)
+ evsel__disable_cpu(pos, 0);
+ else
+ evsel__disable_cpu(pos, pos->cpu_iter - 1);
}
}
if (!has_imm)
@@ -472,6 +488,15 @@ void evlist__disable_evsel(struct evlist *evlist, char *evsel_name)
__evlist__disable(evlist, evsel_name);
}
+static void evsel__enable_cpus(struct evsel *evsel)
+{
+ int cpu, nr_cpus;
+ struct perf_cpu_map *cpus = evsel__cpus(evsel);
+
+ nr_cpus = cpus ? cpus->nr : 1;
+ for (cpu = 0; cpu < nr_cpus; cpu++)
+ evsel__enable_cpu(evsel, cpu);
+}
static void __evlist__enable(struct evlist *evlist, char *evsel_name)
{
struct evsel *pos;
@@ -491,7 +516,12 @@ static void __evlist__enable(struct evlist *evlist, char *evsel_name)
continue;
if (!evsel__is_group_leader(pos) || !pos->core.fd)
continue;
- evsel__enable_cpu(pos, pos->cpu_iter - 1);
+ if (cpu == -1 && !evlist->thread_mode)
+ evsel__enable_cpus(pos);
+ else if (evlist->thread_mode)
+ evsel__enable_cpu(pos, 0);
+ else
+ evsel__enable_cpu(pos, pos->cpu_iter - 1);
}
}
affinity__cleanup(&affinity);
@@ -1274,6 +1304,16 @@ void evlist__set_selected(struct evlist *evlist, struct evsel *evsel)
evlist->selected = evsel;
}
+static void evsel__close_cpus(struct evsel *evsel)
+{
+ int cpu, nr_cpus;
+ struct perf_cpu_map *cpus = evsel__cpus(evsel);
+
+ nr_cpus = cpus ? cpus->nr : 1;
+ for (cpu = 0; cpu < nr_cpus; cpu++)
+ perf_evsel__close_cpu(&evsel->core, cpu);
+}
+
void evlist__close(struct evlist *evlist)
{
struct evsel *evsel;
@@ -1298,7 +1338,13 @@ void evlist__close(struct evlist *evlist)
evlist__for_each_entry_reverse(evlist, evsel) {
if (evsel__cpu_iter_skip(evsel, cpu))
continue;
- perf_evsel__close_cpu(&evsel->core, evsel->cpu_iter - 1);
+
+ if (cpu == -1 && !evlist->thread_mode)
+ evsel__close_cpus(evsel);
+ else if (evlist->thread_mode)
+ perf_evsel__close_cpu(&evsel->core, 0);
+ else
+ perf_evsel__close_cpu(&evsel->core, evsel->cpu_iter - 1);
}
}
affinity__cleanup(&affinity);
@@ -2130,3 +2176,21 @@ struct evsel *evlist__find_evsel(struct evlist *evlist, int idx)
}
return NULL;
}
+
+bool evlist__has_hybrid_events(struct evlist *evlist)
+{
+ struct evsel *evsel;
+
+ evlist__for_each_entry(evlist, evsel) {
+ if (evsel__is_hybrid_event(evsel))
+ return true;
+ }
+
+ return false;
+}
+
+void evlist__invalidate_all_cpus(struct evlist *evlist)
+{
+ perf_cpu_map__put(evlist->core.all_cpus);
+ evlist->core.all_cpus = perf_cpu_map__empty_new(1);
+}
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index b695ffaae519..0da683511d98 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -52,6 +52,7 @@ struct evlist {
struct perf_evlist core;
int nr_groups;
bool enabled;
+ bool thread_mode;
int id_pos;
int is_pos;
u64 combined_sample_type;
@@ -365,4 +366,7 @@ int evlist__ctlfd_ack(struct evlist *evlist);
#define EVLIST_DISABLED_MSG "Events disabled\n"
struct evsel *evlist__find_evsel(struct evlist *evlist, int idx);
+void evlist__invalidate_all_cpus(struct evlist *evlist);
+
+bool evlist__has_hybrid_events(struct evlist *evlist);
#endif /* __PERF_EVLIST_H */
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 6026487353dd..69aadc52c1bd 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -7,9 +7,11 @@
#include <sys/types.h>
#include <linux/perf_event.h>
#include <linux/types.h>
+#include <string.h>
#include <internal/evsel.h>
#include <perf/evsel.h>
#include "symbol_conf.h"
+#include "pmu-hybrid.h"
#include <internal/cpumap.h>
struct bpf_object;
@@ -435,4 +437,10 @@ struct perf_env *evsel__env(struct evsel *evsel);
int evsel__store_ids(struct evsel *evsel, struct evlist *evlist);
void evsel__zero_per_pkg(struct evsel *evsel);
+
+static inline bool evsel__is_hybrid_event(struct evsel *evsel)
+{
+ return evsel->pmu_name && perf_pmu__is_hybrid(evsel->pmu_name);
+}
+
#endif /* __PERF_EVSEL_H */
diff --git a/tools/perf/util/python-ext-sources b/tools/perf/util/python-ext-sources
index 845dd46e3c61..d7c976671e3a 100644
--- a/tools/perf/util/python-ext-sources
+++ b/tools/perf/util/python-ext-sources
@@ -37,3 +37,5 @@ util/units.c
util/affinity.c
util/rwsem.c
util/hashmap.c
+util/pmu-hybrid.c
+util/fncache.c
--
2.17.1
On hybrid platform, some hardware events are only available
on a specific pmu. For example, 'L1-dcache-load-misses' is only
available on 'cpu_core' pmu. And even for the event which can be
available on both pmus, the user also may want to just enable
one event. So now following syntax is supported:
cpu_core/<hardware event>/
cpu_core/<hardware cache event>/
cpu_core/<pmu event>/
cpu_atom/<hardware event>/
cpu_atom/<hardware cache event>/
cpu_atom/<pmu event>/
It limits the event to be enabled only on a specified pmu.
The patch uses this idea, for example, if we use "cpu_core/LLC-loads/",
in parse_events_add_pmu(), term->config is "LLC-loads".
We create a new "parse_events_state" with the pmu_name and use
parse_events__scanner to scan the term->config (the string "LLC-loads"
in this example). The parse_events_add_cache() will be called during
parsing. The parse_state->pmu_name is used to identify the pmu
where the event is enabled.
Let's see examples:
root@ssp-pwrt-002:~# ./perf stat -e cpu_core/cycles/,cpu_core/LLC-loads/ -vv -- ./triad_loop
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
type 6
size 120
config 0x400000000
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
enable_on_exec 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid 7267 cpu -1 group_fd -1 flags 0x8 = 3
------------------------------------------------------------
perf_event_attr:
type 7
size 120
config 0x400000002
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
enable_on_exec 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid 7267 cpu -1 group_fd -1 flags 0x8 = 4
cycles: 0: 449252097 297999924 297999924
LLC-loads: 0: 1857 297999924 297999924
cycles: 449252097 297999924 297999924
LLC-loads: 1857 297999924 297999924
Performance counter stats for './triad_loop':
449,252,097 cpu_core/cycles/
1,857 cpu_core/LLC-loads/
0.298898415 seconds time elapsed
root@ssp-pwrt-002:~# ./perf stat -e cpu_atom/cycles/,cpu_atom/LLC-loads/ -vv -- taskset -c 16 ./triad_loop
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
type 6
size 120
config 0xa00000000
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
enable_on_exec 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid 7339 cpu -1 group_fd -1 flags 0x8 = 3
------------------------------------------------------------
perf_event_attr:
type 7
size 120
config 0xa00000002
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
enable_on_exec 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid 7339 cpu -1 group_fd -1 flags 0x8 = 4
cycles: 0: 602020010 343657939 342553275
LLC-loads: 0: 3537 343657939 342553275
cycles: 603961400 343657939 342553275
LLC-loads: 3548 343657939 342553275
Performance counter stats for 'taskset -c 16 ./triad_loop':
603,961,400 cpu_atom/cycles/ (99.68%)
3,548 cpu_atom/LLC-loads/ (99.68%)
0.344904585 seconds time elapsed
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/util/parse-events.c | 100 +++++++++++++++++++++++++++++++--
tools/perf/util/parse-events.h | 6 +-
tools/perf/util/parse-events.y | 21 ++-----
3 files changed, 105 insertions(+), 22 deletions(-)
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 09e42245f71a..30435adc7a7b 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -489,7 +489,8 @@ static int create_hybrid_cache_event(struct list_head *list, int *idx,
static int add_hybrid_cache(struct list_head *list, int *idx,
struct perf_event_attr *attr, char *name,
struct list_head *config_terms,
- bool *hybrid)
+ bool *hybrid,
+ struct parse_events_state *parse_state)
{
struct perf_pmu *pmu;
int ret;
@@ -497,6 +498,11 @@ static int add_hybrid_cache(struct list_head *list, int *idx,
*hybrid = false;
perf_pmu__for_each_hybrid_pmu(pmu) {
*hybrid = true;
+ if (parse_state->pmu_name &&
+ strcmp(parse_state->pmu_name, pmu->name)) {
+ continue;
+ }
+
ret = create_hybrid_cache_event(list, idx, attr, name,
config_terms, pmu);
if (ret)
@@ -509,7 +515,8 @@ static int add_hybrid_cache(struct list_head *list, int *idx,
int parse_events_add_cache(struct list_head *list, int *idx,
char *type, char *op_result1, char *op_result2,
struct parse_events_error *err,
- struct list_head *head_config)
+ struct list_head *head_config,
+ struct parse_events_state *parse_state)
{
struct perf_event_attr attr;
LIST_HEAD(config_terms);
@@ -582,7 +589,7 @@ int parse_events_add_cache(struct list_head *list, int *idx,
perf_pmu__scan(NULL);
ret = add_hybrid_cache(list, idx, &attr, config_name ? : name,
- &config_terms, &hybrid);
+ &config_terms, &hybrid, parse_state);
if (hybrid)
return ret;
@@ -1512,6 +1519,11 @@ static int add_hybrid_numeric(struct parse_events_state *parse_state,
*hybrid = false;
perf_pmu__for_each_hybrid_pmu(pmu) {
*hybrid = true;
+ if (parse_state->pmu_name &&
+ strcmp(parse_state->pmu_name, pmu->name)) {
+ continue;
+ }
+
ret = create_hybrid_hw_event(parse_state, list, attr, pmu);
if (ret)
return ret;
@@ -1578,6 +1590,10 @@ static bool config_term_percore(struct list_head *config_terms)
return false;
}
+static int parse_events_with_hybrid_pmu(struct parse_events_state *parse_state,
+ const char *str, char *name, bool *found,
+ struct list_head *list);
+
int parse_events_add_pmu(struct parse_events_state *parse_state,
struct list_head *list, char *name,
struct list_head *head_config,
@@ -1589,7 +1605,7 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
struct perf_pmu *pmu;
struct evsel *evsel;
struct parse_events_error *err = parse_state->error;
- bool use_uncore_alias;
+ bool use_uncore_alias, found;
LIST_HEAD(config_terms);
if (verbose > 1) {
@@ -1605,6 +1621,22 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
fprintf(stderr, "' that may result in non-fatal errors\n");
}
+ if (head_config && perf_pmu__is_hybrid(name)) {
+ struct parse_events_term *term;
+ int ret;
+
+ list_for_each_entry(term, head_config, list) {
+ if (!term->config)
+ continue;
+ ret = parse_events_with_hybrid_pmu(parse_state,
+ term->config,
+ name, &found,
+ list);
+ if (found)
+ return ret;
+ }
+ }
+
pmu = parse_state->fake_pmu ?: perf_pmu__find(name);
if (!pmu) {
char *err_str;
@@ -1713,12 +1745,19 @@ int parse_events_multi_pmu_add(struct parse_events_state *parse_state,
struct perf_pmu *pmu = NULL;
int ok = 0;
+ if (parse_state->pmu_name) {
+ list = alloc_list();
+ if (!list)
+ return -1;
+ *listp = list;
+ return 0;
+ }
+
*listp = NULL;
/* Add it for all PMUs that support the alias */
- list = malloc(sizeof(struct list_head));
+ list = alloc_list();
if (!list)
return -1;
- INIT_LIST_HEAD(list);
while ((pmu = perf_pmu__scan(pmu)) != NULL) {
struct perf_pmu_alias *alias;
@@ -2284,6 +2323,44 @@ int parse_events_terms(struct list_head *terms, const char *str)
return ret;
}
+static int list_num(struct list_head *list)
+{
+ struct list_head *pos;
+ int n = 0;
+
+ list_for_each(pos, list)
+ n++;
+
+ return n;
+}
+
+static int parse_events_with_hybrid_pmu(struct parse_events_state *parse_state,
+ const char *str, char *pmu_name,
+ bool *found, struct list_head *list)
+{
+ struct parse_events_state ps = {
+ .list = LIST_HEAD_INIT(ps.list),
+ .stoken = PE_START_EVENTS,
+ .pmu_name = pmu_name,
+ .idx = parse_state->idx,
+ };
+ int ret;
+
+ *found = false;
+ ret = parse_events__scanner(str, &ps);
+ perf_pmu__parse_cleanup();
+
+ if (!ret) {
+ if (!list_empty(&ps.list)) {
+ *found = true;
+ list_splice(&ps.list, list);
+ parse_state->idx = list_num(list);
+ }
+ }
+
+ return ret;
+}
+
int __parse_events(struct evlist *evlist, const char *str,
struct parse_events_error *err, struct perf_pmu *fake_pmu)
{
@@ -3307,3 +3384,14 @@ char *parse_events_formats_error_string(char *additional_terms)
fail:
return NULL;
}
+
+struct list_head *alloc_list(void)
+{
+ struct list_head *list = malloc(sizeof(*list));
+
+ if (!list)
+ return NULL;
+
+ INIT_LIST_HEAD(list);
+ return list;
+}
diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
index e80c9b74f2f2..39c7121a4659 100644
--- a/tools/perf/util/parse-events.h
+++ b/tools/perf/util/parse-events.h
@@ -138,6 +138,7 @@ struct parse_events_state {
struct list_head *terms;
int stoken;
struct perf_pmu *fake_pmu;
+ char *pmu_name;
};
void parse_events__handle_error(struct parse_events_error *err, int idx,
@@ -188,7 +189,8 @@ int parse_events_add_tool(struct parse_events_state *parse_state,
int parse_events_add_cache(struct list_head *list, int *idx,
char *type, char *op_result1, char *op_result2,
struct parse_events_error *error,
- struct list_head *head_config);
+ struct list_head *head_config,
+ struct parse_events_state *parse_state);
int parse_events_add_breakpoint(struct list_head *list, int *idx,
u64 addr, char *type, u64 len);
int parse_events_add_pmu(struct parse_events_state *parse_state,
@@ -242,6 +244,8 @@ char *parse_events_formats_error_string(char *additional_terms);
void parse_events_print_error(struct parse_events_error *err,
const char *event);
+struct list_head *alloc_list(void);
+
#ifdef HAVE_LIBELF_SUPPORT
/*
* If the probe point starts with '%',
diff --git a/tools/perf/util/parse-events.y b/tools/perf/util/parse-events.y
index d57ac86ce7ca..e0e68c3da9e4 100644
--- a/tools/perf/util/parse-events.y
+++ b/tools/perf/util/parse-events.y
@@ -26,18 +26,6 @@ do { \
YYABORT; \
} while (0)
-static struct list_head* alloc_list(void)
-{
- struct list_head *list;
-
- list = malloc(sizeof(*list));
- if (!list)
- return NULL;
-
- INIT_LIST_HEAD(list);
- return list;
-}
-
static void free_list_evsel(struct list_head* list_evsel)
{
struct evsel *evsel, *tmp;
@@ -454,7 +442,8 @@ PE_NAME_CACHE_TYPE '-' PE_NAME_CACHE_OP_RESULT '-' PE_NAME_CACHE_OP_RESULT opt_e
list = alloc_list();
ABORT_ON(!list);
- err = parse_events_add_cache(list, &parse_state->idx, $1, $3, $5, error, $6);
+ err = parse_events_add_cache(list, &parse_state->idx, $1, $3, $5, error, $6,
+ parse_state);
parse_events_terms__delete($6);
free($1);
free($3);
@@ -475,7 +464,8 @@ PE_NAME_CACHE_TYPE '-' PE_NAME_CACHE_OP_RESULT opt_event_config
list = alloc_list();
ABORT_ON(!list);
- err = parse_events_add_cache(list, &parse_state->idx, $1, $3, NULL, error, $4);
+ err = parse_events_add_cache(list, &parse_state->idx, $1, $3, NULL, error, $4,
+ parse_state);
parse_events_terms__delete($4);
free($1);
free($3);
@@ -495,7 +485,8 @@ PE_NAME_CACHE_TYPE opt_event_config
list = alloc_list();
ABORT_ON(!list);
- err = parse_events_add_cache(list, &parse_state->idx, $1, NULL, NULL, error, $2);
+ err = parse_events_add_cache(list, &parse_state->idx, $1, NULL, NULL, error, $2,
+ parse_state);
parse_events_terms__delete($2);
free($1);
if (err) {
--
2.17.1
On hybrid platform, same raw event is possible to be available on
both cpu_core pmu and cpu_atom pmu. So it's supported to create
two raw events for one event encoding.
root@ssp-pwrt-002:~# ./perf stat -e r3c -a -vv -- sleep 1
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
type 4
size 120
config 0x3c
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 3
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 4
sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 5
sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 7
sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 8
sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 9
sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 10
sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 11
sys_perf_event_open: pid -1 cpu 8 group_fd -1 flags 0x8 = 12
sys_perf_event_open: pid -1 cpu 9 group_fd -1 flags 0x8 = 13
sys_perf_event_open: pid -1 cpu 10 group_fd -1 flags 0x8 = 14
sys_perf_event_open: pid -1 cpu 11 group_fd -1 flags 0x8 = 15
sys_perf_event_open: pid -1 cpu 12 group_fd -1 flags 0x8 = 16
sys_perf_event_open: pid -1 cpu 13 group_fd -1 flags 0x8 = 17
sys_perf_event_open: pid -1 cpu 14 group_fd -1 flags 0x8 = 18
sys_perf_event_open: pid -1 cpu 15 group_fd -1 flags 0x8 = 19
------------------------------------------------------------
perf_event_attr:
type 10
size 120
config 0x3c
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 16 group_fd -1 flags 0x8 = 20
sys_perf_event_open: pid -1 cpu 17 group_fd -1 flags 0x8 = 21
sys_perf_event_open: pid -1 cpu 18 group_fd -1 flags 0x8 = 22
sys_perf_event_open: pid -1 cpu 19 group_fd -1 flags 0x8 = 23
sys_perf_event_open: pid -1 cpu 20 group_fd -1 flags 0x8 = 24
sys_perf_event_open: pid -1 cpu 21 group_fd -1 flags 0x8 = 25
sys_perf_event_open: pid -1 cpu 22 group_fd -1 flags 0x8 = 26
sys_perf_event_open: pid -1 cpu 23 group_fd -1 flags 0x8 = 27
r3c: 0: 807321251 1002093589 1002093589
r3c: 1: 807321699 1002088203 1002088203
r3c: 2: 802163010 1002086701 1002086701
r3c: 3: 802162967 1002080660 1002080660
r3c: 4: 801769096 1002077047 1002077047
r3c: 5: 801766174 1002071197 1002071197
r3c: 6: 804147338 1002065696 1002065696
r3c: 7: 804141152 1002055345 1002055345
r3c: 8: 801743651 1002043364 1002043364
r3c: 9: 801742285 1002036921 1002036921
r3c: 10: 804083297 1002032502 1002032502
r3c: 11: 804084735 1002027992 1002027992
r3c: 12: 804504507 1002026371 1002026371
r3c: 13: 804504679 1002022466 1002022466
r3c: 14: 811424953 1002021767 1002021767
r3c: 15: 811423320 1002021594 1002021594
r3c: 0: 810883154 1002021654 1002021654
r3c: 1: 810881069 1002017334 1002017334
r3c: 2: 810878689 1002014010 1002014010
r3c: 3: 810876654 1002011516 1002011516
r3c: 4: 800488244 1002007858 1002007858
r3c: 5: 800486260 1002003635 1002003635
r3c: 6: 800483374 1002000384 1002000384
r3c: 7: 800481011 1001997122 1001997122
r3c: 12874304114 16032851415 16032851415
r3c: 6445458455 8016073513 8016073513
Performance counter stats for 'system wide':
12,874,304,114 cpu_core/r3c/
6,445,458,455 cpu_atom/r3c/
1.002310991 seconds time elapsed
It also supports the raw event inside pmu. Syntax is similar:
cpu_core/<raw event>/
cpu_atom/<raw event>/
root@ssp-pwrt-002:~# ./perf stat -e cpu_core/r3c/ -vv -- ./triad_loop
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
type 4
size 120
config 0x3c
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
enable_on_exec 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid 12340 cpu -1 group_fd -1 flags 0x8 = 3
cpu_core/r3c/: 0: 449000613 293915211 293915211
cpu_core/r3c/: 449000613 293915211 293915211
Performance counter stats for './triad_loop':
449,000,613 cpu_core/r3c/
0.294859229 seconds time elapsed
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/util/parse-events.c | 56 +++++++++++++++++++++++++++++++++-
1 file changed, 55 insertions(+), 1 deletion(-)
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 30435adc7a7b..9b2a33103a57 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -1532,6 +1532,55 @@ static int add_hybrid_numeric(struct parse_events_state *parse_state,
return 0;
}
+static int create_hybrid_raw_event(struct parse_events_state *parse_state,
+ struct list_head *list,
+ struct perf_event_attr *attr,
+ struct list_head *head_config,
+ struct list_head *config_terms,
+ struct perf_pmu *pmu)
+{
+ struct evsel *evsel;
+
+ attr->type = pmu->type;
+ evsel = __add_event(list, &parse_state->idx, attr, true,
+ get_config_name(head_config),
+ pmu, config_terms, false, NULL);
+ if (evsel)
+ evsel->pmu_name = strdup(pmu->name);
+ else
+ return -ENOMEM;
+
+ return 0;
+}
+
+static int add_hybrid_raw(struct parse_events_state *parse_state,
+ struct list_head *list,
+ struct perf_event_attr *attr,
+ struct list_head *head_config,
+ struct list_head *config_terms,
+ bool *hybrid)
+{
+ struct perf_pmu *pmu;
+ int ret;
+
+ *hybrid = false;
+ perf_pmu__for_each_hybrid_pmu(pmu) {
+ *hybrid = true;
+ if (parse_state->pmu_name &&
+ strcmp(parse_state->pmu_name, pmu->name)) {
+ continue;
+ }
+
+ ret = create_hybrid_raw_event(parse_state, list, attr,
+ head_config, config_terms,
+ pmu);
+ if (ret)
+ return ret;
+ }
+
+ return 0;
+}
+
int parse_events_add_numeric(struct parse_events_state *parse_state,
struct list_head *list,
u32 type, u64 config,
@@ -1558,7 +1607,12 @@ int parse_events_add_numeric(struct parse_events_state *parse_state,
/*
* Skip the software dummy event.
*/
- if (type != PERF_TYPE_SOFTWARE) {
+ if (type == PERF_TYPE_RAW) {
+ ret = add_hybrid_raw(parse_state, list, &attr, head_config,
+ &config_terms, &hybrid);
+ if (hybrid)
+ return ret;
+ } else if (type != PERF_TYPE_SOFTWARE) {
if (!perf_pmu__hybrid_exist())
perf_pmu__scan(NULL);
--
2.17.1
When evlist is empty, for example no '-e' specified in perf record,
one default 'cycles' event is added to evlist.
While on hybrid platform, it needs to create two default 'cycles'
events. One is for core, the other is for atom.
This patch actually calls evsel__new_cycles() two times to create
two 'cycles' events.
root@ssp-pwrt-002:~# ./perf record -vv -- sleep 1
...
------------------------------------------------------------
perf_event_attr:
type 6
size 120
config 0x400000000
{ sample_period, sample_freq } 4000
sample_type IP|TID|TIME|ID|PERIOD
read_format ID
disabled 1
inherit 1
mmap 1
comm 1
freq 1
enable_on_exec 1
task 1
precise_ip 3
sample_id_all 1
exclude_guest 1
mmap2 1
comm_exec 1
ksymbol 1
bpf_event 1
------------------------------------------------------------
sys_perf_event_open: pid 22300 cpu 0 group_fd -1 flags 0x8 = 5
sys_perf_event_open: pid 22300 cpu 1 group_fd -1 flags 0x8 = 6
sys_perf_event_open: pid 22300 cpu 2 group_fd -1 flags 0x8 = 7
sys_perf_event_open: pid 22300 cpu 3 group_fd -1 flags 0x8 = 9
sys_perf_event_open: pid 22300 cpu 4 group_fd -1 flags 0x8 = 10
sys_perf_event_open: pid 22300 cpu 5 group_fd -1 flags 0x8 = 11
sys_perf_event_open: pid 22300 cpu 6 group_fd -1 flags 0x8 = 12
sys_perf_event_open: pid 22300 cpu 7 group_fd -1 flags 0x8 = 13
sys_perf_event_open: pid 22300 cpu 8 group_fd -1 flags 0x8 = 14
sys_perf_event_open: pid 22300 cpu 9 group_fd -1 flags 0x8 = 15
sys_perf_event_open: pid 22300 cpu 10 group_fd -1 flags 0x8 = 16
sys_perf_event_open: pid 22300 cpu 11 group_fd -1 flags 0x8 = 17
sys_perf_event_open: pid 22300 cpu 12 group_fd -1 flags 0x8 = 18
sys_perf_event_open: pid 22300 cpu 13 group_fd -1 flags 0x8 = 19
sys_perf_event_open: pid 22300 cpu 14 group_fd -1 flags 0x8 = 20
sys_perf_event_open: pid 22300 cpu 15 group_fd -1 flags 0x8 = 21
------------------------------------------------------------
perf_event_attr:
type 6
size 120
config 0xa00000000
{ sample_period, sample_freq } 4000
sample_type IP|TID|TIME|ID|PERIOD
read_format ID
disabled 1
inherit 1
freq 1
enable_on_exec 1
precise_ip 3
sample_id_all 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid 22300 cpu 16 group_fd -1 flags 0x8 = 22
sys_perf_event_open: pid 22300 cpu 17 group_fd -1 flags 0x8 = 23
sys_perf_event_open: pid 22300 cpu 18 group_fd -1 flags 0x8 = 24
sys_perf_event_open: pid 22300 cpu 19 group_fd -1 flags 0x8 = 25
sys_perf_event_open: pid 22300 cpu 20 group_fd -1 flags 0x8 = 26
sys_perf_event_open: pid 22300 cpu 21 group_fd -1 flags 0x8 = 27
sys_perf_event_open: pid 22300 cpu 22 group_fd -1 flags 0x8 = 28
sys_perf_event_open: pid 22300 cpu 23 group_fd -1 flags 0x8 = 29
...
We can see one core 'cycles' (0x400000000) is enabled on cpu0-cpu15
and atom 'cycles' (0xa00000000) is enabled on cpu16-cpu23.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/builtin-record.c | 10 ++++++----
tools/perf/util/evlist.c | 32 +++++++++++++++++++++++++++++++-
tools/perf/util/evsel.c | 6 +++---
tools/perf/util/evsel.h | 2 +-
4 files changed, 41 insertions(+), 9 deletions(-)
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 35465d1db6dd..363ea1047148 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -2786,10 +2786,12 @@ int cmd_record(int argc, const char **argv)
if (record.opts.overwrite)
record.opts.tail_synthesize = true;
- if (rec->evlist->core.nr_entries == 0 &&
- __evlist__add_default(rec->evlist, !record.opts.no_samples) < 0) {
- pr_err("Not enough memory for event selector list\n");
- goto out;
+ if (rec->evlist->core.nr_entries == 0) {
+ perf_pmu__scan(NULL);
+ if (__evlist__add_default(rec->evlist, !record.opts.no_samples) < 0) {
+ pr_err("Not enough memory for event selector list\n");
+ goto out;
+ }
}
if (rec->opts.target.tid && !rec->opts.no_inherit_set)
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 3ee12fcd0c9f..f139151b9433 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -244,10 +244,40 @@ void evlist__set_leader(struct evlist *evlist)
}
}
+static int __evlist__add_hybrid_default(struct evlist *evlist, bool precise)
+{
+ struct evsel *evsel;
+ struct perf_pmu *pmu;
+ __u64 config;
+ struct perf_cpu_map *cpus;
+
+ perf_pmu__for_each_hybrid_pmu(pmu) {
+ config = PERF_COUNT_HW_CPU_CYCLES |
+ ((__u64)pmu->type << PERF_PMU_TYPE_SHIFT);
+ evsel = evsel__new_cycles(precise, PERF_TYPE_HARDWARE_PMU,
+ config);
+ if (!evsel)
+ return -ENOMEM;
+
+ cpus = perf_cpu_map__get(pmu->cpus);
+ evsel->core.cpus = cpus;
+ evsel->core.own_cpus = perf_cpu_map__get(cpus);
+ evsel->pmu_name = strdup(pmu->name);
+ evlist__add(evlist, evsel);
+ }
+
+ return 0;
+}
+
int __evlist__add_default(struct evlist *evlist, bool precise)
{
- struct evsel *evsel = evsel__new_cycles(precise);
+ struct evsel *evsel;
+
+ if (perf_pmu__hybrid_exist())
+ return __evlist__add_hybrid_default(evlist, precise);
+ evsel = evsel__new_cycles(precise, PERF_TYPE_HARDWARE,
+ PERF_COUNT_HW_CPU_CYCLES);
if (evsel == NULL)
return -ENOMEM;
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 7ecbc8e2fbfa..e0b6227d263f 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -295,11 +295,11 @@ static bool perf_event_can_profile_kernel(void)
return perf_event_paranoid_check(1);
}
-struct evsel *evsel__new_cycles(bool precise)
+struct evsel *evsel__new_cycles(bool precise, __u32 type, __u64 config)
{
struct perf_event_attr attr = {
- .type = PERF_TYPE_HARDWARE,
- .config = PERF_COUNT_HW_CPU_CYCLES,
+ .type = type,
+ .config = config,
.exclude_kernel = !perf_event_can_profile_kernel(),
};
struct evsel *evsel;
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 69aadc52c1bd..8e9079505e96 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -204,7 +204,7 @@ static inline struct evsel *evsel__newtp(const char *sys, const char *name)
return evsel__newtp_idx(sys, name, 0);
}
-struct evsel *evsel__new_cycles(bool precise);
+struct evsel *evsel__new_cycles(bool precise, __u32 type, __u64 config);
struct tep_event *event_format__new(const char *sys, const char *name);
--
2.17.1
For a hybrid system, the perf subsystem doesn't know which PMU the
events belong to. So the PMU aware version PERF_TYPE_HARDWARE_PMU and
PERF_TYPE_HW_CACHE_PMU are introduced.
Now define the new output[] entries for these two types.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/builtin-script.c | 24 ++++++++++++++++++++++++
1 file changed, 24 insertions(+)
diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
index 5915f19cee55..d0e889e636d5 100644
--- a/tools/perf/builtin-script.c
+++ b/tools/perf/builtin-script.c
@@ -275,6 +275,30 @@ static struct {
.invalid_fields = PERF_OUTPUT_TRACE | PERF_OUTPUT_BPF_OUTPUT,
},
+ [PERF_TYPE_HARDWARE_PMU] = {
+ .user_set = false,
+
+ .fields = PERF_OUTPUT_COMM | PERF_OUTPUT_TID |
+ PERF_OUTPUT_CPU | PERF_OUTPUT_TIME |
+ PERF_OUTPUT_EVNAME | PERF_OUTPUT_IP |
+ PERF_OUTPUT_SYM | PERF_OUTPUT_SYMOFFSET |
+ PERF_OUTPUT_DSO | PERF_OUTPUT_PERIOD,
+
+ .invalid_fields = PERF_OUTPUT_TRACE | PERF_OUTPUT_BPF_OUTPUT,
+ },
+
+ [PERF_TYPE_HW_CACHE_PMU] = {
+ .user_set = false,
+
+ .fields = PERF_OUTPUT_COMM | PERF_OUTPUT_TID |
+ PERF_OUTPUT_CPU | PERF_OUTPUT_TIME |
+ PERF_OUTPUT_EVNAME | PERF_OUTPUT_IP |
+ PERF_OUTPUT_SYM | PERF_OUTPUT_SYMOFFSET |
+ PERF_OUTPUT_DSO | PERF_OUTPUT_PERIOD,
+
+ .invalid_fields = PERF_OUTPUT_TRACE | PERF_OUTPUT_BPF_OUTPUT,
+ },
+
[OUTPUT_TYPE_SYNTH] = {
.user_set = false,
--
2.17.1
A group mixed with hybrid event and global event is allowed. For example,
group leader is 'cpu-clock' and the group member is 'cpu_atom/cycles/'.
e.g.
perf stat -e '{cpu-clock,cpu_atom/cycles/}' -a
The challenge is their available cpus are not fully matched.
For example, 'cpu-clock' is available on CPU0-CPU23, but 'cpu_core/cycles/'
is available on CPU16-CPU23.
When getting the group id for group member, we must be very careful
because the cpu for 'cpu-clock' is not equal to the cpu for 'cpu_atom/cycles/'.
Actually the cpu here is the index of evsel->core.cpus, not the real CPU ID.
e.g. cpu0 for 'cpu-clock' is CPU0, but cpu0 for 'cpu_atom/cycles/' is CPU16.
Another challenge is for group read. The events in group may be not
available on all cpus. For example the leader is a software event and
it's available on CPU0-CPU1, but the group member is a hybrid event and
it's only available on CPU1. For CPU0, we have only one event, but for CPU1
we have two events. So we need to change the read size according to
the real number of events on that cpu.
Let's see examples,
root@ssp-pwrt-002:~# ./perf stat -e '{cpu-clock,cpu_atom/cycles/}' -a -vvv -- sleep 1
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
type 1
size 120
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING|ID|GROUP
disabled 1
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 3
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 4
sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 5
sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 7
sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 8
sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 9
sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 10
sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 11
sys_perf_event_open: pid -1 cpu 8 group_fd -1 flags 0x8 = 12
sys_perf_event_open: pid -1 cpu 9 group_fd -1 flags 0x8 = 13
sys_perf_event_open: pid -1 cpu 10 group_fd -1 flags 0x8 = 14
sys_perf_event_open: pid -1 cpu 11 group_fd -1 flags 0x8 = 15
sys_perf_event_open: pid -1 cpu 12 group_fd -1 flags 0x8 = 16
sys_perf_event_open: pid -1 cpu 13 group_fd -1 flags 0x8 = 17
sys_perf_event_open: pid -1 cpu 14 group_fd -1 flags 0x8 = 18
sys_perf_event_open: pid -1 cpu 15 group_fd -1 flags 0x8 = 19
sys_perf_event_open: pid -1 cpu 16 group_fd -1 flags 0x8 = 20
sys_perf_event_open: pid -1 cpu 17 group_fd -1 flags 0x8 = 21
sys_perf_event_open: pid -1 cpu 18 group_fd -1 flags 0x8 = 22
sys_perf_event_open: pid -1 cpu 19 group_fd -1 flags 0x8 = 23
sys_perf_event_open: pid -1 cpu 20 group_fd -1 flags 0x8 = 24
sys_perf_event_open: pid -1 cpu 21 group_fd -1 flags 0x8 = 25
sys_perf_event_open: pid -1 cpu 22 group_fd -1 flags 0x8 = 26
sys_perf_event_open: pid -1 cpu 23 group_fd -1 flags 0x8 = 27
------------------------------------------------------------
perf_event_attr:
type 6
size 120
config 0xa00000000
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING|ID|GROUP
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 16 group_fd 20 flags 0x8 = 28
sys_perf_event_open: pid -1 cpu 17 group_fd 21 flags 0x8 = 29
sys_perf_event_open: pid -1 cpu 18 group_fd 22 flags 0x8 = 30
sys_perf_event_open: pid -1 cpu 19 group_fd 23 flags 0x8 = 31
sys_perf_event_open: pid -1 cpu 20 group_fd 24 flags 0x8 = 32
sys_perf_event_open: pid -1 cpu 21 group_fd 25 flags 0x8 = 33
sys_perf_event_open: pid -1 cpu 22 group_fd 26 flags 0x8 = 34
sys_perf_event_open: pid -1 cpu 23 group_fd 27 flags 0x8 = 35
cpu-clock: 0: 1002495141 1002496165 1002496165
cpu-clock: 1: 1002494114 1002494741 1002494741
cpu-clock: 2: 1002489755 1002491096 1002491096
cpu-clock: 3: 1002486160 1002487660 1002487660
cpu-clock: 4: 1002482342 1002483461 1002483461
cpu-clock: 5: 1002480405 1002480756 1002480756
cpu-clock: 6: 1002478135 1002478698 1002478698
cpu-clock: 7: 1002476940 1002477251 1002477251
cpu-clock: 8: 1002474281 1002475616 1002475616
cpu-clock: 9: 1002471033 1002471983 1002471983
cpu-clock: 10: 1002468437 1002469553 1002469553
cpu-clock: 11: 1002467503 1002467892 1002467892
cpu-clock: 12: 1002465167 1002466190 1002466190
cpu-clock: 13: 1002463794 1002464109 1002464109
cpu-clock: 14: 1002460746 1002461897 1002461897
cpu-clock: 15: 1002460006 1002460371 1002460371
cpu-clock: 16: 1002457619 1002458056 1002458056
cpu-clock: 17: 1002451809 1002452414 1002452414
cpu-clock: 18: 1002446385 1002446927 1002446927
cpu-clock: 19: 1002442633 1002443203 1002443203
cpu-clock: 20: 1002438330 1002438939 1002438939
cpu-clock: 21: 1002432839 1002433483 1002433483
cpu-clock: 22: 1002428951 1002429416 1002429416
cpu-clock: 23: 1002423932 1002424604 1002424604
cycles: 0: 800847113 1002458056 1002458056
cycles: 1: 800843983 1002452414 1002452414
cycles: 2: 800840233 1002446927 1002446927
cycles: 3: 800837260 1002443203 1002443203
cycles: 4: 800832030 1002438939 1002438939
cycles: 5: 800829207 1002433483 1002433483
cycles: 6: 800825621 1002429416 1002429416
cycles: 7: 800822445 1002424604 1002424604
cpu-clock: 24059136457 24059154481 24059154481
cycles: 6406677892 8019527042 8019527042
Performance counter stats for 'system wide':
24,059.14 msec cpu-clock # 23.994 CPUs utilized
6,406,677,892 cpu_atom/cycles/ # 266.289 M/sec
1.002699058 seconds time elapsed
For cpu_atom/cycles/, cpu16-cpu23 are set with valid group fd (cpu-clock's fd
on that cpu). For counting results, cpu-clock has 24 cpus aggregation and
cpu_atom/cycles/ has 8 cpus aggregation. That's expected.
But if the event order is changed, e.g. '{cpu_atom/cycles/,cpu-clock}',
there leaves more works to do.
root@ssp-pwrt-002:~# ./perf stat -e '{cpu_atom/cycles/,cpu-clock}' -a -vvv -- sleep 1
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
type 6
size 120
config 0xa00000000
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING|ID|GROUP
disabled 1
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 16 group_fd -1 flags 0x8 = 3
sys_perf_event_open: pid -1 cpu 17 group_fd -1 flags 0x8 = 4
sys_perf_event_open: pid -1 cpu 18 group_fd -1 flags 0x8 = 5
sys_perf_event_open: pid -1 cpu 19 group_fd -1 flags 0x8 = 7
sys_perf_event_open: pid -1 cpu 20 group_fd -1 flags 0x8 = 8
sys_perf_event_open: pid -1 cpu 21 group_fd -1 flags 0x8 = 9
sys_perf_event_open: pid -1 cpu 22 group_fd -1 flags 0x8 = 10
sys_perf_event_open: pid -1 cpu 23 group_fd -1 flags 0x8 = 11
------------------------------------------------------------
perf_event_attr:
type 1
size 120
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING|ID|GROUP
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 12
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 13
sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 14
sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 15
sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 16
sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 17
sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 18
sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 19
sys_perf_event_open: pid -1 cpu 8 group_fd -1 flags 0x8 = 20
sys_perf_event_open: pid -1 cpu 9 group_fd -1 flags 0x8 = 21
sys_perf_event_open: pid -1 cpu 10 group_fd -1 flags 0x8 = 22
sys_perf_event_open: pid -1 cpu 11 group_fd -1 flags 0x8 = 23
sys_perf_event_open: pid -1 cpu 12 group_fd -1 flags 0x8 = 24
sys_perf_event_open: pid -1 cpu 13 group_fd -1 flags 0x8 = 25
sys_perf_event_open: pid -1 cpu 14 group_fd -1 flags 0x8 = 26
sys_perf_event_open: pid -1 cpu 15 group_fd -1 flags 0x8 = 27
sys_perf_event_open: pid -1 cpu 16 group_fd 3 flags 0x8 = 28
sys_perf_event_open: pid -1 cpu 17 group_fd 4 flags 0x8 = 29
sys_perf_event_open: pid -1 cpu 18 group_fd 5 flags 0x8 = 30
sys_perf_event_open: pid -1 cpu 19 group_fd 7 flags 0x8 = 31
sys_perf_event_open: pid -1 cpu 20 group_fd 8 flags 0x8 = 32
sys_perf_event_open: pid -1 cpu 21 group_fd 9 flags 0x8 = 33
sys_perf_event_open: pid -1 cpu 22 group_fd 10 flags 0x8 = 34
sys_perf_event_open: pid -1 cpu 23 group_fd 11 flags 0x8 = 35
cycles: 0: 810965983 1002124999 1002124999
cycles: 1: 810962706 1002118442 1002118442
cycles: 2: 810959729 1002114853 1002114853
cycles: 3: 810958079 1002111730 1002111730
cycles: 4: 800570097 1002108582 1002108582
cycles: 5: 800569278 1002106441 1002106441
cycles: 6: 800568167 1002104339 1002104339
cycles: 7: 800566760 1002102953 1002102953
WARNING: for cpu-clock, some CPU counts not read
cpu-clock: 0: 0 0 0
cpu-clock: 1: 0 0 0
cpu-clock: 2: 0 0 0
cpu-clock: 3: 0 0 0
cpu-clock: 4: 0 0 0
cpu-clock: 5: 0 0 0
cpu-clock: 6: 0 0 0
cpu-clock: 7: 0 0 0
cpu-clock: 8: 0 0 0
cpu-clock: 9: 0 0 0
cpu-clock: 10: 0 0 0
cpu-clock: 11: 0 0 0
cpu-clock: 12: 0 0 0
cpu-clock: 13: 0 0 0
cpu-clock: 14: 0 0 0
cpu-clock: 15: 0 0 0
cpu-clock: 16: 1002125111 1002124999 1002124999
cpu-clock: 17: 1002118626 1002118442 1002118442
cpu-clock: 18: 1002115058 1002114853 1002114853
cpu-clock: 19: 1002111740 1002111730 1002111730
cpu-clock: 20: 1002109031 1002108582 1002108582
cpu-clock: 21: 1002105927 1002106441 1002106441
cpu-clock: 22: 1002104010 1002104339 1002104339
cpu-clock: 23: 1002102730 1002102953 1002102953
cycles: 6446120799 8016892339 8016892339
cpu-clock: 8016892233 8016892339 8016892339
Performance counter stats for 'system wide':
6,446,120,799 cpu_atom/cycles/ # 804.067 M/sec
8,016.89 msec cpu-clock # 7.999 CPUs utilized
1.002212870 seconds time elapsed
For cpu-clock, cpu16-cpu23 are set with valid group fd (cpu_atom/cycles/'s
fd on that cpu). For counting results, cpu_atom/cycles/ has 8 cpus aggregation
, that's correct. But for cpu-clock, it also has 8 cpus aggregation
(cpu16-cpu23, but not all cpus), the code should be improved. Now one warning
is displayed: "WARNING: for cpu-clock, some CPU counts not read".
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/util/evsel.c | 105 ++++++++++++++++++++++++++++++++++++++--
tools/perf/util/stat.h | 1 +
2 files changed, 101 insertions(+), 5 deletions(-)
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index e0b6227d263f..862fdc145f05 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -1464,15 +1464,26 @@ static void evsel__set_count(struct evsel *counter, int cpu, int thread, u64 val
perf_counts__set_loaded(counter->counts, cpu, thread, true);
}
-static int evsel__process_group_data(struct evsel *leader, int cpu, int thread, u64 *data)
+static int evsel_cpuid_match(struct evsel *evsel1, struct evsel *evsel2,
+ int cpu)
+{
+ int cpuid;
+
+ cpuid = perf_cpu_map__cpu(evsel1->core.cpus, cpu);
+ return perf_cpu_map__idx(evsel2->core.cpus, cpuid);
+}
+
+static int evsel__process_group_data(struct evsel *leader, int cpu, int thread,
+ u64 *data, int nr_members)
{
u64 read_format = leader->core.attr.read_format;
struct sample_read_value *v;
u64 nr, ena = 0, run = 0, i;
+ int idx;
nr = *data++;
- if (nr != (u64) leader->core.nr_members)
+ if (nr != (u64) nr_members)
return -EINVAL;
if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
@@ -1492,24 +1503,85 @@ static int evsel__process_group_data(struct evsel *leader, int cpu, int thread,
if (!counter)
return -EINVAL;
- evsel__set_count(counter, cpu, thread, v[i].value, ena, run);
+ if (evsel__is_hybrid_event(counter) ||
+ evsel__is_hybrid_event(leader)) {
+ idx = evsel_cpuid_match(leader, counter, cpu);
+ if (idx == -1)
+ return -EINVAL;
+ } else
+ idx = cpu;
+
+ evsel__set_count(counter, idx, thread, v[i].value, ena, run);
}
return 0;
}
+static int hybrid_read_size(struct evsel *leader, int cpu, int *nr_members)
+{
+ struct evsel *pos;
+ int nr = 1, back, new_size = 0, idx;
+
+ for_each_group_member(pos, leader) {
+ idx = evsel_cpuid_match(leader, pos, cpu);
+ if (idx != -1)
+ nr++;
+ }
+
+ if (nr != leader->core.nr_members) {
+ back = leader->core.nr_members;
+ leader->core.nr_members = nr;
+ new_size = perf_evsel__read_size(&leader->core);
+ leader->core.nr_members = back;
+ }
+
+ *nr_members = nr;
+ return new_size;
+}
+
static int evsel__read_group(struct evsel *leader, int cpu, int thread)
{
struct perf_stat_evsel *ps = leader->stats;
u64 read_format = leader->core.attr.read_format;
int size = perf_evsel__read_size(&leader->core);
+ int new_size, nr_members;
u64 *data = ps->group_data;
if (!(read_format & PERF_FORMAT_ID))
return -EINVAL;
- if (!evsel__is_group_leader(leader))
+ if (!evsel__is_group_leader(leader)) {
+ if (evsel__is_hybrid_event(leader->leader) &&
+ !evsel__is_hybrid_event(leader)) {
+ /*
+ * The group leader is hybrid event and it's
+ * only available on part of cpus. But the group
+ * member are available on all cpus. TODO:
+ * read the counts on the rest of cpus for group
+ * member.
+ */
+ WARN_ONCE(1, "WARNING: for %s, some CPU counts "
+ "not read\n", leader->name);
+ return 0;
+ }
return -EINVAL;
+ }
+
+ /*
+ * For example the leader is a software event and it's available on
+ * cpu0-cpu1, but the group member is a hybrid event and it's only
+ * available on cpu1. For cpu0, we have only one event, but for cpu1
+ * we have two events. So we need to change the read size according to
+ * the real number of events on a given cpu.
+ */
+ new_size = hybrid_read_size(leader, cpu, &nr_members);
+ if (new_size)
+ size = new_size;
+
+ if (ps->group_data && ps->group_data_size < size) {
+ zfree(&ps->group_data);
+ data = NULL;
+ }
if (!data) {
data = zalloc(size);
@@ -1517,6 +1589,7 @@ static int evsel__read_group(struct evsel *leader, int cpu, int thread)
return -ENOMEM;
ps->group_data = data;
+ ps->group_data_size = size;
}
if (FD(leader, cpu, thread) < 0)
@@ -1525,7 +1598,7 @@ static int evsel__read_group(struct evsel *leader, int cpu, int thread)
if (readn(FD(leader, cpu, thread), data, size) <= 0)
return -errno;
- return evsel__process_group_data(leader, cpu, thread, data);
+ return evsel__process_group_data(leader, cpu, thread, data, nr_members);
}
int evsel__read_counter(struct evsel *evsel, int cpu, int thread)
@@ -1572,6 +1645,28 @@ static int get_group_fd(struct evsel *evsel, int cpu, int thread)
*/
BUG_ON(!leader->core.fd);
+ /*
+ * If leader is not hybrid event, it's available on
+ * all cpus (e.g. software event). But hybrid evsel
+ * member is only available on part of cpus. So need
+ * to get the leader's fd from correct cpu.
+ */
+ if (evsel__is_hybrid_event(evsel) &&
+ !evsel__is_hybrid_event(leader)) {
+ cpu = evsel_cpuid_match(evsel, leader, cpu);
+ BUG_ON(cpu == -1);
+ }
+
+ /*
+ * Leader is hybrid event but member is global event.
+ */
+ if (!evsel__is_hybrid_event(evsel) &&
+ evsel__is_hybrid_event(leader)) {
+ cpu = evsel_cpuid_match(evsel, leader, cpu);
+ if (cpu == -1)
+ return -1;
+ }
+
fd = FD(leader, cpu, thread);
BUG_ON(fd == -1);
diff --git a/tools/perf/util/stat.h b/tools/perf/util/stat.h
index d85c292148bb..4aec97d32e69 100644
--- a/tools/perf/util/stat.h
+++ b/tools/perf/util/stat.h
@@ -46,6 +46,7 @@ struct perf_stat_evsel {
struct stats res_stats[3];
enum perf_stat_evsel_id id;
u64 *group_data;
+ int group_data_size;
};
enum aggr_mode {
--
2.17.1
Add basic hybrid test cases for 'Parse event definition strings' test.
root@otcpl-adl-s-2:~# ./perf test 6
6: Parse event definition strings : Ok
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/tests/parse-events.c | 171 ++++++++++++++++++++++++++++++++
1 file changed, 171 insertions(+)
diff --git a/tools/perf/tests/parse-events.c b/tools/perf/tests/parse-events.c
index a7f6661e6112..aec929867020 100644
--- a/tools/perf/tests/parse-events.c
+++ b/tools/perf/tests/parse-events.c
@@ -1512,6 +1512,123 @@ static int test__all_tracepoints(struct evlist *evlist)
return test__checkevent_tracepoint_multi(evlist);
}
+static int test__hybrid_hw_event_with_pmu(struct evlist *evlist)
+{
+ struct evsel *evsel = evlist__first(evlist);
+
+ TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->core.nr_entries);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE_PMU == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong config", 0x400000000 == evsel->core.attr.config);
+ return 0;
+}
+
+static int test__hybrid_hw_event(struct evlist *evlist)
+{
+ struct evsel *evsel1 = evlist__first(evlist);
+ struct evsel *evsel2 = evlist__last(evlist);
+
+ TEST_ASSERT_VAL("wrong number of entries", 2 == evlist->core.nr_entries);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE_PMU == evsel1->core.attr.type);
+ TEST_ASSERT_VAL("wrong config", 0x400000000 == evsel1->core.attr.config);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE_PMU == evsel2->core.attr.type);
+ TEST_ASSERT_VAL("wrong config", 0xa00000000 == evsel2->core.attr.config);
+ return 0;
+}
+
+static int test__hybrid_hw_group_event(struct evlist *evlist)
+{
+ struct evsel *evsel, *leader;
+
+ evsel = leader = evlist__first(evlist);
+ TEST_ASSERT_VAL("wrong number of entries", 2 == evlist->core.nr_entries);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE_PMU == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong config", 0x400000000 == evsel->core.attr.config);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+
+ evsel = evsel__next(evsel);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE_PMU == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong config", 0x400000001 == evsel->core.attr.config);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+ return 0;
+}
+
+static int test__hybrid_sw_hw_group_event(struct evlist *evlist)
+{
+ struct evsel *evsel, *leader;
+
+ evsel = leader = evlist__first(evlist);
+ TEST_ASSERT_VAL("wrong number of entries", 2 == evlist->core.nr_entries);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_SOFTWARE == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+
+ evsel = evsel__next(evsel);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE_PMU == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong config", 0x400000000 == evsel->core.attr.config);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+ return 0;
+}
+
+static int test__hybrid_hw_sw_group_event(struct evlist *evlist)
+{
+ struct evsel *evsel, *leader;
+
+ evsel = leader = evlist__first(evlist);
+ TEST_ASSERT_VAL("wrong number of entries", 2 == evlist->core.nr_entries);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE_PMU == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong config", 0x400000000 == evsel->core.attr.config);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+
+ evsel = evsel__next(evsel);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_SOFTWARE == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+ return 0;
+}
+
+static int test__hybrid_group_modifier1(struct evlist *evlist)
+{
+ struct evsel *evsel, *leader;
+
+ evsel = leader = evlist__first(evlist);
+ TEST_ASSERT_VAL("wrong number of entries", 2 == evlist->core.nr_entries);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE_PMU == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong config", 0x400000000 == evsel->core.attr.config);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+ TEST_ASSERT_VAL("wrong exclude_user", evsel->core.attr.exclude_user);
+ TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->core.attr.exclude_kernel);
+
+ evsel = evsel__next(evsel);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE_PMU == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong config", 0x400000001 == evsel->core.attr.config);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+ TEST_ASSERT_VAL("wrong exclude_user", !evsel->core.attr.exclude_user);
+ TEST_ASSERT_VAL("wrong exclude_kernel", evsel->core.attr.exclude_kernel);
+ return 0;
+}
+
+static int test__hybrid_raw1(struct evlist *evlist)
+{
+ struct evsel *evsel = evlist__first(evlist);
+
+ TEST_ASSERT_VAL("wrong number of entries", 2 == evlist->core.nr_entries);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_RAW == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong config", 0x1a == evsel->core.attr.config);
+
+ /* The type of second event is randome value */
+ evsel = evsel__next(evsel);
+ TEST_ASSERT_VAL("wrong config", 0x1a == evsel->core.attr.config);
+ return 0;
+}
+
+static int test__hybrid_raw2(struct evlist *evlist)
+{
+ struct evsel *evsel = evlist__first(evlist);
+
+ TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->core.nr_entries);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_RAW == evsel->core.attr.type);
+ TEST_ASSERT_VAL("wrong config", 0x1a == evsel->core.attr.config);
+ return 0;
+}
+
struct evlist_test {
const char *name;
__u32 type;
@@ -1868,6 +1985,54 @@ static struct terms_test test__terms[] = {
},
};
+static struct evlist_test test__hybrid_events[] = {
+ {
+ .name = "cpu_core/cycles/",
+ .check = test__hybrid_hw_event_with_pmu,
+ .id = 0,
+ },
+ {
+ .name = "cycles",
+ .check = test__hybrid_hw_event,
+ .id = 1,
+ },
+ {
+ .name = "{cpu_core/cycles/,cpu_core/instructions/}",
+ .check = test__hybrid_hw_group_event,
+ .id = 2,
+ },
+ {
+ .name = "{cpu-clock,cpu_core/cycles/}",
+ .check = test__hybrid_sw_hw_group_event,
+ .id = 3,
+ },
+ {
+ .name = "{cpu_core/cycles/,cpu-clock}",
+ .check = test__hybrid_hw_sw_group_event,
+ .id = 4,
+ },
+ {
+ .name = "{cpu_core/cycles:k/,cpu_core/instructions:u/}",
+ .check = test__hybrid_group_modifier1,
+ .id = 5,
+ },
+ {
+ .name = "r1a",
+ .check = test__hybrid_raw1,
+ .id = 6,
+ },
+ {
+ .name = "cpu_core/r1a/",
+ .check = test__hybrid_raw2,
+ .id = 7,
+ },
+ {
+ .name = "cpu_core/config=10,config1,config2=3,period=1000/u",
+ .check = test__checkevent_pmu,
+ .id = 0,
+ },
+};
+
static int test_event(struct evlist_test *e)
{
struct parse_events_error err;
@@ -2035,6 +2200,12 @@ do { \
ret2 = ret1; \
} while (0)
+ perf_pmu__scan(NULL);
+ if (perf_pmu__hybrid_exist()) {
+ TEST_EVENTS(test__hybrid_events);
+ return ret2;
+ }
+
TEST_EVENTS(test__events);
if (test_pmu())
--
2.17.1
Since for one hw event, two hybrid events are created.
For example,
evsel->idx evsel__name(evsel)
0 cycles
1 cycles
2 instructions
3 instructions
...
So for comparing the evsel name on hybrid, the evsel->idx
needs to be divided by 2.
root@otcpl-adl-s-2:~# ./perf test 14
14: Roundtrip evsel->name : Ok
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/tests/evsel-roundtrip-name.c | 19 ++++++++++++-------
1 file changed, 12 insertions(+), 7 deletions(-)
diff --git a/tools/perf/tests/evsel-roundtrip-name.c b/tools/perf/tests/evsel-roundtrip-name.c
index f7f3e5b4c180..2b938a15901e 100644
--- a/tools/perf/tests/evsel-roundtrip-name.c
+++ b/tools/perf/tests/evsel-roundtrip-name.c
@@ -62,7 +62,8 @@ static int perf_evsel__roundtrip_cache_name_test(void)
return ret;
}
-static int __perf_evsel__name_array_test(const char *names[], int nr_names)
+static int __perf_evsel__name_array_test(const char *names[], int nr_names,
+ int distance)
{
int i, err;
struct evsel *evsel;
@@ -82,9 +83,9 @@ static int __perf_evsel__name_array_test(const char *names[], int nr_names)
err = 0;
evlist__for_each_entry(evlist, evsel) {
- if (strcmp(evsel__name(evsel), names[evsel->idx])) {
+ if (strcmp(evsel__name(evsel), names[evsel->idx / distance])) {
--err;
- pr_debug("%s != %s\n", evsel__name(evsel), names[evsel->idx]);
+ pr_debug("%s != %s\n", evsel__name(evsel), names[evsel->idx / distance]);
}
}
@@ -93,18 +94,22 @@ static int __perf_evsel__name_array_test(const char *names[], int nr_names)
return err;
}
-#define perf_evsel__name_array_test(names) \
- __perf_evsel__name_array_test(names, ARRAY_SIZE(names))
+#define perf_evsel__name_array_test(names, distance) \
+ __perf_evsel__name_array_test(names, ARRAY_SIZE(names), distance)
int test__perf_evsel__roundtrip_name_test(struct test *test __maybe_unused, int subtest __maybe_unused)
{
int err = 0, ret = 0;
- err = perf_evsel__name_array_test(evsel__hw_names);
+ perf_pmu__scan(NULL);
+ if (perf_pmu__hybrid_exist())
+ return perf_evsel__name_array_test(evsel__hw_names, 2);
+
+ err = perf_evsel__name_array_test(evsel__hw_names, 1);
if (err)
ret = err;
- err = __perf_evsel__name_array_test(evsel__sw_names, PERF_COUNT_SW_DUMMY + 1);
+ err = __perf_evsel__name_array_test(evsel__sw_names, PERF_COUNT_SW_DUMMY + 1, 1);
if (err)
ret = err;
--
2.17.1
Since for "cycles:u' on hybrid platform, it creates two "cycles".
So the number of events in evlist is not expected in next test
steps. Now we just use one event "cpu_core/cycles:u/" for hybrid.
root@otcpl-adl-s-2:~# ./perf test 35
35: Track with sched_switch : Ok
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/tests/switch-tracking.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/tools/perf/tests/switch-tracking.c b/tools/perf/tests/switch-tracking.c
index 3ebaa758df77..13a11ce51a1a 100644
--- a/tools/perf/tests/switch-tracking.c
+++ b/tools/perf/tests/switch-tracking.c
@@ -340,6 +340,11 @@ int test__switch_tracking(struct test *test __maybe_unused, int subtest __maybe_
struct evsel *switch_evsel, *tracking_evsel;
const char *comm;
int err = -1;
+ bool hybrid = false;
+
+ perf_pmu__scan(NULL);
+ if (perf_pmu__hybrid_exist())
+ hybrid = true;
threads = thread_map__new(-1, getpid(), UINT_MAX);
if (!threads) {
@@ -371,7 +376,10 @@ int test__switch_tracking(struct test *test __maybe_unused, int subtest __maybe_
cpu_clocks_evsel = evlist__last(evlist);
/* Second event */
- err = parse_events(evlist, "cycles:u", NULL);
+ if (!hybrid)
+ err = parse_events(evlist, "cycles:u", NULL);
+ else
+ err = parse_events(evlist, "cpu_core/cycles:u/", NULL);
if (err) {
pr_debug("Failed to parse event cycles:u\n");
goto out_err;
--
2.17.1
Since for "cycles:u' on hybrid platform, it creates two "cycles".
So the second evsel in evlist also needs initialization.
With this patch,
root@otcpl-adl-s-2:~# ./perf test 71
71: Convert perf time to TSC : Ok
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/tests/perf-time-to-tsc.c | 16 ++++++++++++++++
1 file changed, 16 insertions(+)
diff --git a/tools/perf/tests/perf-time-to-tsc.c b/tools/perf/tests/perf-time-to-tsc.c
index 680c3cffb128..b472205ec8e3 100644
--- a/tools/perf/tests/perf-time-to-tsc.c
+++ b/tools/perf/tests/perf-time-to-tsc.c
@@ -66,6 +66,11 @@ int test__perf_time_to_tsc(struct test *test __maybe_unused, int subtest __maybe
u64 test_tsc, comm1_tsc, comm2_tsc;
u64 test_time, comm1_time = 0, comm2_time = 0;
struct mmap *md;
+ bool hybrid = false;
+
+ perf_pmu__scan(NULL);
+ if (perf_pmu__hybrid_exist())
+ hybrid = true;
threads = thread_map__new(-1, getpid(), UINT_MAX);
CHECK_NOT_NULL__(threads);
@@ -88,6 +93,17 @@ int test__perf_time_to_tsc(struct test *test __maybe_unused, int subtest __maybe
evsel->core.attr.disabled = 1;
evsel->core.attr.enable_on_exec = 0;
+ /*
+ * For hybrid "cycles:u", it creates two events.
+ * Init the second evsel here.
+ */
+ if (hybrid) {
+ evsel = evsel__next(evsel);
+ evsel->core.attr.comm = 1;
+ evsel->core.attr.disabled = 1;
+ evsel->core.attr.enable_on_exec = 0;
+ }
+
CHECK__(evlist__open(evlist));
CHECK__(evlist__mmap(evlist, UINT_MAX));
--
2.17.1
Currently we don't support shadow stat for hybrid.
root@ssp-pwrt-002:~# ./perf stat -e cycles,instructions -a -- sleep 1
Performance counter stats for 'system wide':
12,883,109,591 cpu_core/cycles/
6,405,163,221 cpu_atom/cycles/
555,553,778 cpu_core/instructions/
841,158,734 cpu_atom/instructions/
1.002644773 seconds time elapsed
Now there is no shadow stat 'insn per cycle' reported. We will support
it later and now just skip the 'perf stat metrics (shadow stat) test'.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/tests/shell/stat+shadow_stat.sh | 3 +++
1 file changed, 3 insertions(+)
diff --git a/tools/perf/tests/shell/stat+shadow_stat.sh b/tools/perf/tests/shell/stat+shadow_stat.sh
index ebebd3596cf9..e6e35fc6c882 100755
--- a/tools/perf/tests/shell/stat+shadow_stat.sh
+++ b/tools/perf/tests/shell/stat+shadow_stat.sh
@@ -7,6 +7,9 @@ set -e
# skip if system-wide mode is forbidden
perf stat -a true > /dev/null 2>&1 || exit 2
+# skip if on hybrid platform
+perf stat -a -e cycles sleep 1 2>&1 | grep -e cpu_core && exit 2
+
test_global_aggr()
{
perf stat -a --no-big-num -e cycles,instructions sleep 1 2>&1 | \
--
2.17.1
For hardware events, they have pre-defined configs. The kernel
needs to know where the event comes from (e.g. from cpu_core pmu
or from cpu_atom pmu). But the perf type 'PERF_TYPE_HARDWARE'
can't carry pmu information.
So the kernel introduces a new type 'PERF_TYPE_HARDWARE_PMU'.
The new attr.config layout for PERF_TYPE_HARDWARE_PMU is:
0xDD000000AA
AA: original hardware event ID
DD: PMU type ID
PMU type ID is retrieved from sysfs. For example,
cat /sys/devices/cpu_atom/type
10
cat /sys/devices/cpu_core/type
4
When enabling a hybrid hardware event without specified pmu, such as,
'perf stat -e cycles -a', two events are created automatically. One
is for atom, the other is for core.
root@ssp-pwrt-002:~# ./perf stat -e cycles -vv -a -- sleep 1
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
type 6
size 120
config 0x400000000
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 3
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 4
sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 5
sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 7
sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 8
sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 9
sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 10
sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 11
sys_perf_event_open: pid -1 cpu 8 group_fd -1 flags 0x8 = 12
sys_perf_event_open: pid -1 cpu 9 group_fd -1 flags 0x8 = 13
sys_perf_event_open: pid -1 cpu 10 group_fd -1 flags 0x8 = 14
sys_perf_event_open: pid -1 cpu 11 group_fd -1 flags 0x8 = 15
sys_perf_event_open: pid -1 cpu 12 group_fd -1 flags 0x8 = 16
sys_perf_event_open: pid -1 cpu 13 group_fd -1 flags 0x8 = 17
sys_perf_event_open: pid -1 cpu 14 group_fd -1 flags 0x8 = 18
sys_perf_event_open: pid -1 cpu 15 group_fd -1 flags 0x8 = 19
------------------------------------------------------------
perf_event_attr:
type 6
size 120
config 0xa00000000
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 16 group_fd -1 flags 0x8 = 20
sys_perf_event_open: pid -1 cpu 17 group_fd -1 flags 0x8 = 21
sys_perf_event_open: pid -1 cpu 18 group_fd -1 flags 0x8 = 22
sys_perf_event_open: pid -1 cpu 19 group_fd -1 flags 0x8 = 23
sys_perf_event_open: pid -1 cpu 20 group_fd -1 flags 0x8 = 24
sys_perf_event_open: pid -1 cpu 21 group_fd -1 flags 0x8 = 25
sys_perf_event_open: pid -1 cpu 22 group_fd -1 flags 0x8 = 26
sys_perf_event_open: pid -1 cpu 23 group_fd -1 flags 0x8 = 27
cycles: 0: 810754998 1002563650 1002563650
cycles: 1: 810749852 1002559947 1002559947
cycles: 2: 808096005 1002555036 1002555036
cycles: 3: 808090246 1002543496 1002543496
cycles: 4: 800933425 1002536659 1002536659
cycles: 5: 800928573 1002528386 1002528386
cycles: 6: 800924347 1002520527 1002520527
cycles: 7: 800922009 1002513176 1002513176
cycles: 8: 800919624 1002507326 1002507326
cycles: 9: 800917204 1002500663 1002500663
cycles: 10: 802096579 1002494280 1002494280
cycles: 11: 802093770 1002486404 1002486404
cycles: 12: 803284338 1002479491 1002479491
cycles: 13: 803277609 1002469777 1002469777
cycles: 14: 800875902 1002458861 1002458861
cycles: 15: 800873241 1002451350 1002451350
cycles: 0: 800837379 1002444645 1002444645
cycles: 1: 800833400 1002438505 1002438505
cycles: 2: 800829291 1002433698 1002433698
cycles: 3: 800824390 1002427584 1002427584
cycles: 4: 800819360 1002422099 1002422099
cycles: 5: 800814787 1002415845 1002415845
cycles: 6: 800810125 1002410301 1002410301
cycles: 7: 800791893 1002386845 1002386845
cycles: 12855737722 16040169029 16040169029
cycles: 6406560625 8019379522 8019379522
Performance counter stats for 'system wide':
12,855,737,722 cpu_core/cycles/
6,406,560,625 cpu_atom/cycles/
1.002774658 seconds time elapsed
type 6 is PERF_TYPE_HARDWARE_PMU.
0x4 in 0x400000000 indicates the cpu_core pmu.
0xa in 0xa00000000 indicates the cpu_atom pmu.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/util/parse-events.c | 73 ++++++++++++++++++++++++++++++++++
1 file changed, 73 insertions(+)
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 42c84adeb2fb..c6c76fc810a3 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -446,6 +446,24 @@ static int config_attr(struct perf_event_attr *attr,
struct parse_events_error *err,
config_term_func_t config_term);
+static void config_hybrid_attr(struct perf_event_attr *attr,
+ int type, int pmu_type)
+{
+ /*
+ * attr.config layout:
+ * PERF_TYPE_HARDWARE_PMU: 0xDD000000AA
+ * AA: hardware event ID
+ * DD: PMU type ID
+ * PERF_TYPE_HW_CACHE_PMU: 0xDD00CCBBAA
+ * AA: hardware cache ID
+ * BB: hardware cache op ID
+ * CC: hardware cache op result ID
+ * DD: PMU type ID
+ */
+ attr->type = type;
+ attr->config = attr->config | ((__u64)pmu_type << PERF_PMU_TYPE_SHIFT);
+}
+
int parse_events_add_cache(struct list_head *list, int *idx,
char *type, char *op_result1, char *op_result2,
struct parse_events_error *err,
@@ -1409,6 +1427,47 @@ int parse_events_add_tracepoint(struct list_head *list, int *idx,
err, head_config);
}
+static int create_hybrid_hw_event(struct parse_events_state *parse_state,
+ struct list_head *list,
+ struct perf_event_attr *attr,
+ struct perf_pmu *pmu)
+{
+ struct evsel *evsel;
+ __u32 type = attr->type;
+ __u64 config = attr->config;
+
+ config_hybrid_attr(attr, PERF_TYPE_HARDWARE_PMU, pmu->type);
+ evsel = __add_event(list, &parse_state->idx, attr, true, NULL,
+ pmu, NULL, false, NULL);
+ if (evsel)
+ evsel->pmu_name = strdup(pmu->name);
+ else
+ return -ENOMEM;
+
+ attr->type = type;
+ attr->config = config;
+ return 0;
+}
+
+static int add_hybrid_numeric(struct parse_events_state *parse_state,
+ struct list_head *list,
+ struct perf_event_attr *attr,
+ bool *hybrid)
+{
+ struct perf_pmu *pmu;
+ int ret;
+
+ *hybrid = false;
+ perf_pmu__for_each_hybrid_pmu(pmu) {
+ *hybrid = true;
+ ret = create_hybrid_hw_event(parse_state, list, attr, pmu);
+ if (ret)
+ return ret;
+ }
+
+ return 0;
+}
+
int parse_events_add_numeric(struct parse_events_state *parse_state,
struct list_head *list,
u32 type, u64 config,
@@ -1416,6 +1475,8 @@ int parse_events_add_numeric(struct parse_events_state *parse_state,
{
struct perf_event_attr attr;
LIST_HEAD(config_terms);
+ bool hybrid;
+ int ret;
memset(&attr, 0, sizeof(attr));
attr.type = type;
@@ -1430,6 +1491,18 @@ int parse_events_add_numeric(struct parse_events_state *parse_state,
return -ENOMEM;
}
+ /*
+ * Skip the software dummy event.
+ */
+ if (type != PERF_TYPE_SOFTWARE) {
+ if (!perf_pmu__hybrid_exist())
+ perf_pmu__scan(NULL);
+
+ ret = add_hybrid_numeric(parse_state, list, &attr, &hybrid);
+ if (hybrid)
+ return ret;
+ }
+
return add_event(list, &parse_state->idx, &attr,
get_config_name(head_config), &config_terms);
}
--
2.17.1
For cache events, they have pre-defined configs. The kernel needs
to know where the cache event comes from (e.g. from cpu_core pmu
or from cpu_atom pmu). But the perf type 'PERF_TYPE_HW_CACHE'
can't carry pmu information.
So the kernel introduces a new type 'PERF_TYPE_HW_CACHE_PMU'.
The new attr.config layout for PERF_TYPE_HW_CACHE_PMU is
0xDD00CCBBAA
AA: hardware cache ID
BB: hardware cache op ID
CC: hardware cache op result ID
DD: PMU type ID
Similar as hardware event, PMU type ID is retrieved from sysfs.
When enabling a hybrid cache event without specified pmu, such as,
'perf stat -e L1-dcache-loads -a', two events are created
automatically. One is for atom, the other is for core.
root@ssp-pwrt-002:~# ./perf stat -e L1-dcache-loads -vv -a -- sleep 1
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
type 7
size 120
config 0x400000000
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 3
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 4
sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 5
sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 7
sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 8
sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 9
sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 10
sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 11
sys_perf_event_open: pid -1 cpu 8 group_fd -1 flags 0x8 = 12
sys_perf_event_open: pid -1 cpu 9 group_fd -1 flags 0x8 = 13
sys_perf_event_open: pid -1 cpu 10 group_fd -1 flags 0x8 = 14
sys_perf_event_open: pid -1 cpu 11 group_fd -1 flags 0x8 = 15
sys_perf_event_open: pid -1 cpu 12 group_fd -1 flags 0x8 = 16
sys_perf_event_open: pid -1 cpu 13 group_fd -1 flags 0x8 = 17
sys_perf_event_open: pid -1 cpu 14 group_fd -1 flags 0x8 = 18
sys_perf_event_open: pid -1 cpu 15 group_fd -1 flags 0x8 = 19
------------------------------------------------------------
perf_event_attr:
type 7
size 120
config 0xa00000000
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
disabled 1
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 16 group_fd -1 flags 0x8 = 20
sys_perf_event_open: pid -1 cpu 17 group_fd -1 flags 0x8 = 21
sys_perf_event_open: pid -1 cpu 18 group_fd -1 flags 0x8 = 22
sys_perf_event_open: pid -1 cpu 19 group_fd -1 flags 0x8 = 23
sys_perf_event_open: pid -1 cpu 20 group_fd -1 flags 0x8 = 24
sys_perf_event_open: pid -1 cpu 21 group_fd -1 flags 0x8 = 25
sys_perf_event_open: pid -1 cpu 22 group_fd -1 flags 0x8 = 26
sys_perf_event_open: pid -1 cpu 23 group_fd -1 flags 0x8 = 27
L1-dcache-loads: 0: 13103284 1002535421 1002535421
L1-dcache-loads: 1: 12995797 1002532807 1002532807
L1-dcache-loads: 2: 13428186 1002528572 1002528572
L1-dcache-loads: 3: 12913469 1002517437 1002517437
L1-dcache-loads: 4: 12857843 1002507079 1002507079
L1-dcache-loads: 5: 12812079 1002498279 1002498279
L1-dcache-loads: 6: 12829938 1002490010 1002490010
L1-dcache-loads: 7: 12807085 1002481860 1002481860
L1-dcache-loads: 8: 12907189 1002473181 1002473181
L1-dcache-loads: 9: 12823095 1002465895 1002465895
L1-dcache-loads: 10: 12892770 1002459322 1002459322
L1-dcache-loads: 11: 12789718 1002451607 1002451607
L1-dcache-loads: 12: 12838931 1002442632 1002442632
L1-dcache-loads: 13: 12803756 1002434133 1002434133
L1-dcache-loads: 14: 12840574 1002426060 1002426060
L1-dcache-loads: 15: 12799075 1002415964 1002415964
L1-dcache-loads: 0: 39394457 1002406287 1002406287
L1-dcache-loads: 1: 39372632 1002400502 1002400502
L1-dcache-loads: 2: 39405247 1002394865 1002394865
L1-dcache-loads: 3: 39400547 1002389099 1002389099
L1-dcache-loads: 4: 39410752 1002383106 1002383106
L1-dcache-loads: 5: 39402983 1002375365 1002375365
L1-dcache-loads: 6: 39388775 1002369374 1002369374
L1-dcache-loads: 7: 39408527 1002363344 1002363344
L1-dcache-loads: 206442789 16039660259 16039660259
L1-dcache-loads: 315183920 8019081942 8019081942
Performance counter stats for 'system wide':
206,442,789 cpu_core/L1-dcache-loads/
315,183,920 cpu_atom/L1-dcache-loads/
1.002751663 seconds time elapsed
type 7 is PERF_TYPE_HW_CACHE_PMU.
0x4 in 0x400000000 indicates the cpu_core pmu.
0xa in 0xa00000000 indicates the cpu_atom pmu.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/util/parse-events.c | 54 +++++++++++++++++++++++++++++++++-
1 file changed, 53 insertions(+), 1 deletion(-)
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index c6c76fc810a3..09e42245f71a 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -464,6 +464,48 @@ static void config_hybrid_attr(struct perf_event_attr *attr,
attr->config = attr->config | ((__u64)pmu_type << PERF_PMU_TYPE_SHIFT);
}
+static int create_hybrid_cache_event(struct list_head *list, int *idx,
+ struct perf_event_attr *attr, char *name,
+ struct list_head *config_terms,
+ struct perf_pmu *pmu)
+{
+ struct evsel *evsel;
+ __u32 type = attr->type;
+ __u64 config = attr->config;
+
+ config_hybrid_attr(attr, PERF_TYPE_HW_CACHE_PMU, pmu->type);
+ evsel = __add_event(list, idx, attr, true, name,
+ pmu, config_terms, false, NULL);
+ if (evsel)
+ evsel->pmu_name = strdup(pmu->name);
+ else
+ return -ENOMEM;
+
+ attr->type = type;
+ attr->config = config;
+ return 0;
+}
+
+static int add_hybrid_cache(struct list_head *list, int *idx,
+ struct perf_event_attr *attr, char *name,
+ struct list_head *config_terms,
+ bool *hybrid)
+{
+ struct perf_pmu *pmu;
+ int ret;
+
+ *hybrid = false;
+ perf_pmu__for_each_hybrid_pmu(pmu) {
+ *hybrid = true;
+ ret = create_hybrid_cache_event(list, idx, attr, name,
+ config_terms, pmu);
+ if (ret)
+ return ret;
+ }
+
+ return 0;
+}
+
int parse_events_add_cache(struct list_head *list, int *idx,
char *type, char *op_result1, char *op_result2,
struct parse_events_error *err,
@@ -474,7 +516,8 @@ int parse_events_add_cache(struct list_head *list, int *idx,
char name[MAX_NAME_LEN], *config_name;
int cache_type = -1, cache_op = -1, cache_result = -1;
char *op_result[2] = { op_result1, op_result2 };
- int i, n;
+ int i, n, ret;
+ bool hybrid;
/*
* No fallback - if we cannot get a clear cache type
@@ -534,6 +577,15 @@ int parse_events_add_cache(struct list_head *list, int *idx,
if (get_config_terms(head_config, &config_terms))
return -ENOMEM;
}
+
+ if (!perf_pmu__hybrid_exist())
+ perf_pmu__scan(NULL);
+
+ ret = add_hybrid_cache(list, idx, &attr, config_name ? : name,
+ &config_terms, &hybrid);
+ if (hybrid)
+ return ret;
+
return add_event(list, idx, &attr, config_name ? : name, &config_terms);
}
--
2.17.1
Add some words and examples to help understanding of
Intel hybrid perf support.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/Documentation/intel-hybrid.txt | 228 ++++++++++++++++++++++
tools/perf/Documentation/perf-record.txt | 1 +
tools/perf/Documentation/perf-stat.txt | 2 +
3 files changed, 231 insertions(+)
create mode 100644 tools/perf/Documentation/intel-hybrid.txt
diff --git a/tools/perf/Documentation/intel-hybrid.txt b/tools/perf/Documentation/intel-hybrid.txt
new file mode 100644
index 000000000000..ff641d9ac81b
--- /dev/null
+++ b/tools/perf/Documentation/intel-hybrid.txt
@@ -0,0 +1,228 @@
+Intel hybrid support
+--------------------
+Support for Intel hybrid events within perf tools.
+
+For some Intel platforms, such as AlderLake, which is hybrid platform and
+it consists of atom cpu and core cpu. Each cpu has dedicated event list.
+Part of events are available on core cpu, part of events are available
+on atom cpu and even part of events are available on both.
+
+Kernel exports two new cpu pmus via sysfs:
+/sys/devices/cpu_core
+/sys/devices/cpu_atom
+
+The 'cpus' files are created under the directories. For example,
+
+cat /sys/devices/cpu_core/cpus
+0-15
+
+cat /sys/devices/cpu_atom/cpus
+16-23
+
+It indicates cpu0-cpu15 are core cpus and cpu16-cpu23 are atom cpus.
+
+Quickstart
+
+List hybrid event
+-----------------
+
+As before, use perf-list to list the symbolic event.
+
+perf list
+
+inst_retired.any
+ [Fixed Counter: Counts the number of instructions retired. Unit: cpu_atom]
+inst_retired.any
+ [Number of instructions retired. Fixed Counter - architectural event. Unit: cpu_core]
+
+The 'Unit: xxx' is added to brief description to indicate which pmu
+the event is belong to. Same event name but with different pmu can
+be supported.
+
+Enable hybrid event with a specific pmu
+---------------------------------------
+
+To enable a core only event or atom only event, following syntax is supported:
+
+ cpu_core/<event name>/
+or
+ cpu_atom/<event name>/
+
+For example, count the 'cycles' event on core cpus.
+
+ perf stat -e cpu_core/cycles/
+
+Create two events for one hardware event automatically
+------------------------------------------------------
+
+When creating one event and the event is available on both atom and core,
+two events are created automatically. One is for atom, the other is for
+core. Most of hardware events and cache events are available on both
+cpu_core and cpu_atom.
+
+For hardware events, they have pre-defined configs (e.g. 0 for cycles).
+But on hybrid platform, kernel needs to know where the event comes from
+(from atom or from core). The original perf event type PERF_TYPE_HARDWARE
+can't carry pmu information. So a new type PERF_TYPE_HARDWARE_PMU is
+introduced.
+
+The new attr.config layout for PERF_TYPE_HARDWARE_PMU:
+
+0xDD000000AA
+AA: original hardware event ID
+DD: PMU type ID
+
+Cache event is similar. A new type PERF_TYPE_HW_CACHE_PMU is introduced.
+
+The new attr.config layout for PERF_TYPE_HW_CACHE_PMU:
+
+0xDD00CCBBAA
+AA: original hardware cache ID
+BB: original hardware cache op ID
+CC: original hardware cache op result ID
+DD: PMU type ID
+
+PMU type ID is retrieved from sysfs
+
+cat /sys/devices/cpu_atom/type
+10
+
+cat /sys/devices/cpu_core/type
+4
+
+When enabling a hardware event without specified pmu, such as,
+perf stat -e cycles -a (use system-wide in this example), two events
+are created automatically.
+
+ ------------------------------------------------------------
+ perf_event_attr:
+ type 6
+ size 120
+ config 0x400000000
+ sample_type IDENTIFIER
+ read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
+ disabled 1
+ inherit 1
+ exclude_guest 1
+ ------------------------------------------------------------
+
+and
+
+ ------------------------------------------------------------
+ perf_event_attr:
+ type 6
+ size 120
+ config 0xa00000000
+ sample_type IDENTIFIER
+ read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
+ disabled 1
+ inherit 1
+ exclude_guest 1
+ ------------------------------------------------------------
+
+type 6 is PERF_TYPE_HARDWARE_PMU.
+0x4 in 0x400000000 indicates it's cpu_core pmu.
+0xa in 0xa00000000 indicates it's cpu_atom pmu (atom pmu type id is random).
+
+The kernel creates 'cycles' (0x400000000) on cpu0-cpu15 (core cpus),
+and create 'cycles' (0xa00000000) on cpu16-cpu23 (atom cpus).
+
+For perf-stat result, it displays two events:
+
+ Performance counter stats for 'system wide':
+
+ 12,869,720,529 cpu_core/cycles/
+ 6,405,459,328 cpu_atom/cycles/
+
+The first 'cycles' is core event, the second 'cycles' is atom event.
+
+Thread mode example:
+--------------------
+
+perf-stat reports the scaled counts for hybrid event and with a percentage
+displayed. The percentage is the event's running time/enabling time.
+
+One example, 'triad_loop' runs on cpu16 (atom core), while we can see the
+scaled value for core cycles is 160,444,092 and the percentage is 0.47%.
+
+perf stat -e cycles -- taskset -c 16 ./triad_loop
+
+As previous, two events are created.
+
+------------------------------------------------------------
+perf_event_attr:
+ type 6
+ size 120
+ config 0x400000000
+ sample_type IDENTIFIER
+ read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
+ disabled 1
+ inherit 1
+ enable_on_exec 1
+ exclude_guest 1
+------------------------------------------------------------
+
+and
+
+------------------------------------------------------------
+perf_event_attr:
+ type 6
+ size 120
+ config 0xa00000000
+ sample_type IDENTIFIER
+ read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
+ disabled 1
+ inherit 1
+ enable_on_exec 1
+ exclude_guest 1
+------------------------------------------------------------
+
+cycles: 0: 784136 339210144 1013444
+cycles: 0: 601988862 339210144 338196700
+cycles: 262458394 339210144 1013444
+cycles: 603792788 339210144 338196700
+
+ Performance counter stats for 'taskset -c 16 ./triad_loop':
+
+ 262,458,394 cpu_core/cycles/ (0.30%)
+ 603,792,788 cpu_atom/cycles/ (99.70%)
+
+ 0.340467913 seconds time elapsed
+
+perf-record:
+------------
+
+If there is no '-e' specified in perf record, on hybrid platform,
+it creates two default 'cycles' and adds them to event list. One
+is for core, the other is for atom.
+
+perf-stat:
+----------
+
+If there is no '-e' specified in perf stat, on hybrid platform,
+besides of software events, following events are created and
+added to event list in order.
+
+core 'cycles',
+atom 'cycles',
+core 'instructions',
+atom 'instructions',
+core 'branches',
+atom 'branches',
+core 'branch-misses',
+atom 'branch-misses'
+
+Of course, both perf-stat and perf-record support to enable
+hybrid event with a specific pmu.
+
+e.g.
+perf stat -e cpu_core/cycles/
+perf stat -e cpu_atom/cycles/
+perf stat -e cpu_core/r1a/
+perf stat -e cpu_atom/L1-icache-loads/
+perf stat -e cpu_core/cycles/,cpu_atom/instructions/
+perf stat -e '{cpu_core/cycles/,cpu_core/instructions/}'
+
+But '{cpu_core/cycles/,cpu_atom/instructions/}' will return
+"<not supported>" for 'instructions', because the pmus in
+group are not matched (cpu_core vs. cpu_atom).
\ No newline at end of file
diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
index f3161c9673e9..d71bac847936 100644
--- a/tools/perf/Documentation/perf-record.txt
+++ b/tools/perf/Documentation/perf-record.txt
@@ -695,6 +695,7 @@ measurements:
wait -n ${perf_pid}
exit $?
+include::intel-hybrid.txt[]
SEE ALSO
--------
diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
index 08a1714494f8..d0def5c1715a 100644
--- a/tools/perf/Documentation/perf-stat.txt
+++ b/tools/perf/Documentation/perf-stat.txt
@@ -527,6 +527,8 @@ The fields are in this order:
Additional metrics may be printed with all earlier fields being empty.
+include::intel-hybrid.txt[]
+
SEE ALSO
--------
linkperf:perf-top[1], linkperf:perf-list[1]
--
2.17.1
Previously if '-e' is not specified in perf stat, some software events
and hardware events are added to evlist by default.
root@otcpl-adl-s-2:~# ./perf stat -- ./triad_loop
Performance counter stats for './triad_loop':
109.43 msec task-clock # 0.993 CPUs utilized
1 context-switches # 0.009 K/sec
0 cpu-migrations # 0.000 K/sec
105 page-faults # 0.960 K/sec
401,161,982 cycles # 3.666 GHz
1,601,216,357 instructions # 3.99 insn per cycle
200,217,751 branches # 1829.686 M/sec
14,555 branch-misses # 0.01% of all branches
0.110176860 seconds time elapsed
Among the events, cycles, instructions, branches and branch-misses
are hardware events.
One hybrid platform, two events are created for one hardware event.
core cycles,
atom cycles,
core instructions,
atom instructions,
core branches,
atom branches,
core branch-misses,
atom branch-misses
These events will be added to evlist in order on hybrid platform
if '-e' is not set.
Since parse_events() has been supported to create two hardware events
for one event on hybrid platform, so we just use parse_events(evlist,
"cycles,instructions,branches,branch-misses") to create the default
events and add them to evlist.
After:
root@ssp-pwrt-002:~# ./perf stat -- ./triad_loop
Performance counter stats for './triad_loop':
290.77 msec task-clock # 0.996 CPUs utilized
25 context-switches # 0.086 K/sec
13 cpu-migrations # 0.045 K/sec
107 page-faults # 0.368 K/sec
449,620,957 cpu_core/cycles/ # 1546.334 M/sec
<not counted> cpu_atom/cycles/ (0.00%)
1,601,499,820 cpu_core/instructions/ # 5507.870 M/sec
<not counted> cpu_atom/instructions/ (0.00%)
200,272,310 cpu_core/branches/ # 688.776 M/sec
<not counted> cpu_atom/branches/ (0.00%)
15,255 cpu_core/branch-misses/ # 0.052 M/sec
<not counted> cpu_atom/branch-misses/ (0.00%)
0.291897676 seconds time elapsed
We can see two events are created for one hardware event.
First one is core event the second one is atom event.
One thing is, the shadow stats looks a bit different, now it's just
'M/sec'.
The perf_stat__update_shadow_stats and perf_stat__print_shadow_stats
need to be improved in future if we want to get the original shadow
stats.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/builtin-stat.c | 22 ++++++++++++++++++++++
1 file changed, 22 insertions(+)
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 6c0a21323814..7a732508b2b4 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -1162,6 +1162,13 @@ static int parse_stat_cgroups(const struct option *opt,
return parse_cgroups(opt, str, unset);
}
+static int add_default_hybrid_events(struct evlist *evlist)
+{
+ struct parse_events_error err;
+
+ return parse_events(evlist, "cycles,instructions,branches,branch-misses", &err);
+}
+
static struct option stat_options[] = {
OPT_BOOLEAN('T', "transaction", &transaction_run,
"hardware transaction statistics"),
@@ -1637,6 +1644,12 @@ static int add_default_attributes(void)
{ .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS },
{ .type = PERF_TYPE_HARDWARE, .config = PERF_COUNT_HW_BRANCH_MISSES },
+};
+ struct perf_event_attr default_sw_attrs[] = {
+ { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_TASK_CLOCK },
+ { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_CONTEXT_SWITCHES },
+ { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_CPU_MIGRATIONS },
+ { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_PAGE_FAULTS },
};
/*
@@ -1874,6 +1887,15 @@ static int add_default_attributes(void)
}
if (!evsel_list->core.nr_entries) {
+ perf_pmu__scan(NULL);
+ if (perf_pmu__hybrid_exist()) {
+ if (evlist__add_default_attrs(evsel_list,
+ default_sw_attrs) < 0) {
+ return -1;
+ }
+ return add_default_hybrid_events(evsel_list);
+ }
+
if (target__has_cpu(&target))
default_attrs0[0].config = PERF_COUNT_SW_CPU_CLOCK;
--
2.17.1
perf-stat has supported some aggregation modes, such as --per-core,
--per-socket and etc. While for hybrid event, it may only available
on part of cpus. So for --per-core, we need to filter out the
unavailable cores, for --per-socket, filter out the unavailable
sockets, and so on.
Before:
root@ssp-pwrt-002:~# ./perf stat --per-core -e cpu_core/cycles/ -a -- sleep 1
Performance counter stats for 'system wide':
S0-D0-C0 2 1,604,426,524 cpu_core/cycles/
S0-D0-C4 2 1,604,408,224 cpu_core/cycles/
S0-D0-C8 2 1,605,995,644 cpu_core/cycles/
S0-D0-C12 2 1,628,056,554 cpu_core/cycles/
S0-D0-C16 2 1,611,488,734 cpu_core/cycles/
S0-D0-C20 2 1,616,314,761 cpu_core/cycles/
S0-D0-C24 2 1,603,558,295 cpu_core/cycles/
S0-D0-C28 2 1,603,541,128 cpu_core/cycles/
S0-D0-C32 0 <not counted> cpu_core/cycles/
S0-D0-C33 0 <not counted> cpu_core/cycles/
S0-D0-C34 0 <not counted> cpu_core/cycles/
S0-D0-C35 0 <not counted> cpu_core/cycles/
S0-D0-C36 0 <not counted> cpu_core/cycles/
S0-D0-C37 0 <not counted> cpu_core/cycles/
S0-D0-C38 0 <not counted> cpu_core/cycles/
S0-D0-C39 0 <not counted> cpu_core/cycles/
After:
root@ssp-pwrt-002:~# ./perf stat --per-core -e cpu_core/cycles/ -a -- sleep 1
Performance counter stats for 'system wide':
S0-D0-C0 2 1,621,781,943 cpu_core/cycles/
S0-D0-C4 2 1,621,755,088 cpu_core/cycles/
S0-D0-C8 2 1,604,276,920 cpu_core/cycles/
S0-D0-C12 2 1,603,446,963 cpu_core/cycles/
S0-D0-C16 2 1,604,231,725 cpu_core/cycles/
S0-D0-C20 2 1,603,435,286 cpu_core/cycles/
S0-D0-C24 2 1,603,387,250 cpu_core/cycles/
S0-D0-C28 2 1,604,173,183 cpu_core/cycles/
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/util/stat-display.c | 20 ++++++++++++++++++++
1 file changed, 20 insertions(+)
diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c
index ed37d8e7ea1a..2db7c36a03ad 100644
--- a/tools/perf/util/stat-display.c
+++ b/tools/perf/util/stat-display.c
@@ -634,6 +634,20 @@ static void aggr_cb(struct perf_stat_config *config,
}
}
+static bool aggr_id_hybrid_matched(struct perf_stat_config *config,
+ struct evsel *counter, struct aggr_cpu_id id)
+{
+ struct aggr_cpu_id s;
+
+ for (int i = 0; i < evsel__nr_cpus(counter); i++) {
+ s = config->aggr_get_id(config, evsel__cpus(counter), i);
+ if (cpu_map__compare_aggr_cpu_id(s, id))
+ return true;
+ }
+
+ return false;
+}
+
static void print_counter_aggrdata(struct perf_stat_config *config,
struct evsel *counter, int s,
char *prefix, bool metric_only,
@@ -647,6 +661,12 @@ static void print_counter_aggrdata(struct perf_stat_config *config,
double uval;
ad.id = id = config->aggr_map->map[s];
+
+ if (perf_pmu__hybrid_exist() &&
+ !aggr_id_hybrid_matched(config, counter, id)) {
+ return;
+ }
+
ad.val = ad.ena = ad.run = 0;
ad.nr = 0;
if (!collect_data(config, counter, aggr_cb, &ad))
--
2.17.1
If a group has events which are from different hybrid PMUs,
shows a warning.
This is to remind the user not to put the core event and atom
event into one group.
root@ssp-pwrt-002:~# ./perf stat -e "{cpu_core/cycles/,cpu_atom/cycles/}" -- sleep 1
WARNING: Group has events from different hybrid PMUs
Performance counter stats for 'sleep 1':
<not counted> cpu_core/cycles/
<not supported> cpu_atom/cycles/
1.002585908 seconds time elapsed
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/builtin-record.c | 3 +++
tools/perf/builtin-stat.c | 7 ++++++
tools/perf/util/evlist.c | 44 +++++++++++++++++++++++++++++++++++++
tools/perf/util/evlist.h | 2 ++
4 files changed, 56 insertions(+)
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 363ea1047148..188a1198cd4b 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -929,6 +929,9 @@ static int record__open(struct record *rec)
pos = evlist__reset_weak_group(evlist, pos, true);
goto try_again;
}
+
+ if (errno == EINVAL && perf_pmu__hybrid_exist())
+ evlist__warn_hybrid_group(evlist);
rc = -errno;
evsel__open_strerror(pos, &opts->target, errno, msg, sizeof(msg));
ui__error("%s\n", msg);
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 7a732508b2b4..6f780a039db0 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -239,6 +239,9 @@ static void evlist__check_cpu_maps(struct evlist *evlist)
struct evsel *evsel, *pos, *leader;
char buf[1024];
+ if (evlist__hybrid_exist(evlist))
+ return;
+
evlist__for_each_entry(evlist, evsel) {
leader = evsel->leader;
@@ -726,6 +729,10 @@ enum counter_recovery {
static enum counter_recovery stat_handle_error(struct evsel *counter)
{
char msg[BUFSIZ];
+
+ if (perf_pmu__hybrid_exist() && errno == EINVAL)
+ evlist__warn_hybrid_group(evsel_list);
+
/*
* PPC returns ENXIO for HW counters until 2.6.37
* (behavior changed with commit b0a873e).
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index f139151b9433..5ec891418cdd 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -2224,3 +2224,47 @@ void evlist__invalidate_all_cpus(struct evlist *evlist)
perf_cpu_map__put(evlist->core.all_cpus);
evlist->core.all_cpus = perf_cpu_map__empty_new(1);
}
+
+static bool group_hybrid_conflict(struct evsel *leader)
+{
+ struct evsel *pos, *prev = NULL;
+
+ for_each_group_evsel(pos, leader) {
+ if (!pos->pmu_name || !perf_pmu__is_hybrid(pos->pmu_name))
+ continue;
+
+ if (prev && strcmp(prev->pmu_name, pos->pmu_name))
+ return true;
+
+ prev = pos;
+ }
+
+ return false;
+}
+
+void evlist__warn_hybrid_group(struct evlist *evlist)
+{
+ struct evsel *evsel;
+
+ evlist__for_each_entry(evlist, evsel) {
+ if (evsel__is_group_leader(evsel) &&
+ evsel->core.nr_members > 1 &&
+ group_hybrid_conflict(evsel)) {
+ WARN_ONCE(1, "WARNING: Group has events from "
+ "different hybrid PMUs\n");
+ return;
+ }
+ }
+}
+
+bool evlist__hybrid_exist(struct evlist *evlist)
+{
+ struct evsel *evsel;
+
+ evlist__for_each_entry(evlist, evsel) {
+ if (evsel__is_hybrid_event(evsel))
+ return true;
+ }
+
+ return false;
+}
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 0da683511d98..33dec3bb5739 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -369,4 +369,6 @@ struct evsel *evlist__find_evsel(struct evlist *evlist, int idx);
void evlist__invalidate_all_cpus(struct evlist *evlist);
bool evlist__has_hybrid_events(struct evlist *evlist);
+void evlist__warn_hybrid_group(struct evlist *evlist);
+bool evlist__hybrid_exist(struct evlist *evlist);
#endif /* __PERF_EVLIST_H */
--
2.17.1
Force to create one event "cpu_core/cycles/" by default,
otherwise in evlist__valid_sample_type, the checking of
'if (evlist->core.nr_entries == 1)' would be failed.
root@otcpl-adl-s-2:~# ./perf test 41
41: Session topology : Ok
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/tests/topology.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/tools/perf/tests/topology.c b/tools/perf/tests/topology.c
index 74748ed75b2c..0f6e73baab2d 100644
--- a/tools/perf/tests/topology.c
+++ b/tools/perf/tests/topology.c
@@ -40,7 +40,15 @@ static int session_write_header(char *path)
session = perf_session__new(&data, false, NULL);
TEST_ASSERT_VAL("can't get session", !IS_ERR(session));
- session->evlist = evlist__new_default();
+ perf_pmu__scan(NULL);
+ if (!perf_pmu__hybrid_exist()) {
+ session->evlist = evlist__new_default();
+ } else {
+ struct parse_events_error err;
+
+ session->evlist = evlist__new();
+ parse_events(session->evlist, "cpu_core/cycles/", &err);
+ }
TEST_ASSERT_VAL("can't get evlist", session->evlist);
perf_header__set_feat(&session->header, HEADER_CPU_TOPOLOGY);
--
2.17.1
For hybrid, kernel introduces new perf type PERF_TYPE_HARDWARE_PMU (6)
and it's assigned to hybrid hardware events.
root@otcpl-adl-s-2:~# ./perf test 17 -vvv
...
compare
matching [event:base-stat]
to [event-6-17179869184-4]
[cpu] * 0
[flags] 0|8 8
[type] 0 6
->FAIL
match: [event:base-stat] matches []
event:base-stat does not match, but is optional
matched
compare
matching [event-6-17179869184-4]
to [event:base-stat]
[cpu] 0 *
[flags] 8 0|8
[type] 6 0
->FAIL
match: [event-6-17179869184-4] matches []
expected type=6, got 0
expected config=17179869184, got 0
FAILED './tests/attr/test-stat-C0' - match failure
The type matching is failed because one type is 0 but the
type of hybrid hardware event is 6. We temporarily skip this
test case and TODO in future.
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/tests/attr.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/tools/perf/tests/attr.c b/tools/perf/tests/attr.c
index dd39ce9b0277..fc7f74159764 100644
--- a/tools/perf/tests/attr.c
+++ b/tools/perf/tests/attr.c
@@ -34,6 +34,7 @@
#include "event.h"
#include "util.h"
#include "tests.h"
+#include "pmu-hybrid.h"
#define ENV "PERF_TEST_ATTR"
@@ -184,6 +185,10 @@ int test__attr(struct test *test __maybe_unused, int subtest __maybe_unused)
char path_dir[PATH_MAX];
char *exec_path;
+ perf_pmu__scan(NULL);
+ if (perf_pmu__hybrid_exist())
+ return 0;
+
/* First try development tree tests. */
if (!lstat("./tests", &st))
return run_dir("./tests", "./perf");
--
2.17.1
Some events are not supported. Only pick up some cases for hybrid.
root@otcpl-adl-s-2:~# ./perf test 67
67: Parse and process metrics : Ok
Signed-off-by: Jin Yao <[email protected]>
---
tools/perf/tests/parse-metric.c | 11 ++++++++---
1 file changed, 8 insertions(+), 3 deletions(-)
diff --git a/tools/perf/tests/parse-metric.c b/tools/perf/tests/parse-metric.c
index 55bf52e588be..149b18f1f96a 100644
--- a/tools/perf/tests/parse-metric.c
+++ b/tools/perf/tests/parse-metric.c
@@ -370,12 +370,17 @@ static int test_metric_group(void)
int test__parse_metric(struct test *test __maybe_unused, int subtest __maybe_unused)
{
+ perf_pmu__scan(NULL);
+
TEST_ASSERT_VAL("IPC failed", test_ipc() == 0);
TEST_ASSERT_VAL("frontend failed", test_frontend() == 0);
- TEST_ASSERT_VAL("cache_miss_cycles failed", test_cache_miss_cycles() == 0);
TEST_ASSERT_VAL("DCache_L2 failed", test_dcache_l2() == 0);
TEST_ASSERT_VAL("recursion fail failed", test_recursion_fail() == 0);
- TEST_ASSERT_VAL("test metric group", test_metric_group() == 0);
TEST_ASSERT_VAL("Memory bandwidth", test_memory_bandwidth() == 0);
- return 0;
+
+ if (!perf_pmu__hybrid_exist()) {
+ TEST_ASSERT_VAL("cache_miss_cycles failed", test_cache_miss_cycles() == 0);
+ TEST_ASSERT_VAL("test metric group", test_metric_group() == 0);
+ }
+ return 0;
}
--
2.17.1
On Thu, Mar 11, 2021 at 03:07:26PM +0800, Jin Yao wrote:
> On hybrid platform, some hardware events are only available
> on a specific pmu. For example, 'L1-dcache-load-misses' is only
> available on 'cpu_core' pmu. And even for the event which can be
> available on both pmus, the user also may want to just enable
> one event. So now following syntax is supported:
>
> cpu_core/<hardware event>/
> cpu_core/<hardware cache event>/
> cpu_core/<pmu event>/
>
> cpu_atom/<hardware event>/
> cpu_atom/<hardware cache event>/
> cpu_atom/<pmu event>/
>
> It limits the event to be enabled only on a specified pmu.
>
> The patch uses this idea, for example, if we use "cpu_core/LLC-loads/",
> in parse_events_add_pmu(), term->config is "LLC-loads".
hum, I don't understand how this doest not work even now,
I assume both cpu_core and cpu_atom have sysfs device directory
with events/ directory right?
and whatever is defined in events we allow in parsing syntax..
why can't we treat them like 2 separated pmus?
thanks,
jirka
>
> We create a new "parse_events_state" with the pmu_name and use
> parse_events__scanner to scan the term->config (the string "LLC-loads"
> in this example). The parse_events_add_cache() will be called during
> parsing. The parse_state->pmu_name is used to identify the pmu
> where the event is enabled.
>
> Let's see examples:
>
> root@ssp-pwrt-002:~# ./perf stat -e cpu_core/cycles/,cpu_core/LLC-loads/ -vv -- ./triad_loop
> Control descriptor is not initialized
> ------------------------------------------------------------
> perf_event_attr:
> type 6
> size 120
> config 0x400000000
> sample_type IDENTIFIER
> read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
> disabled 1
> inherit 1
> enable_on_exec 1
> exclude_guest 1
> ------------------------------------------------------------
> sys_perf_event_open: pid 7267 cpu -1 group_fd -1 flags 0x8 = 3
> ------------------------------------------------------------
> perf_event_attr:
> type 7
> size 120
> config 0x400000002
> sample_type IDENTIFIER
> read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
> disabled 1
> inherit 1
> enable_on_exec 1
> exclude_guest 1
> ------------------------------------------------------------
> sys_perf_event_open: pid 7267 cpu -1 group_fd -1 flags 0x8 = 4
> cycles: 0: 449252097 297999924 297999924
> LLC-loads: 0: 1857 297999924 297999924
> cycles: 449252097 297999924 297999924
> LLC-loads: 1857 297999924 297999924
>
> Performance counter stats for './triad_loop':
>
> 449,252,097 cpu_core/cycles/
> 1,857 cpu_core/LLC-loads/
>
> 0.298898415 seconds time elapsed
>
> root@ssp-pwrt-002:~# ./perf stat -e cpu_atom/cycles/,cpu_atom/LLC-loads/ -vv -- taskset -c 16 ./triad_loop
> Control descriptor is not initialized
> ------------------------------------------------------------
> perf_event_attr:
> type 6
> size 120
> config 0xa00000000
> sample_type IDENTIFIER
> read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
> disabled 1
> inherit 1
> enable_on_exec 1
> exclude_guest 1
> ------------------------------------------------------------
> sys_perf_event_open: pid 7339 cpu -1 group_fd -1 flags 0x8 = 3
> ------------------------------------------------------------
> perf_event_attr:
> type 7
> size 120
> config 0xa00000002
> sample_type IDENTIFIER
> read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
> disabled 1
> inherit 1
> enable_on_exec 1
> exclude_guest 1
> ------------------------------------------------------------
> sys_perf_event_open: pid 7339 cpu -1 group_fd -1 flags 0x8 = 4
> cycles: 0: 602020010 343657939 342553275
> LLC-loads: 0: 3537 343657939 342553275
> cycles: 603961400 343657939 342553275
> LLC-loads: 3548 343657939 342553275
>
> Performance counter stats for 'taskset -c 16 ./triad_loop':
>
> 603,961,400 cpu_atom/cycles/ (99.68%)
> 3,548 cpu_atom/LLC-loads/ (99.68%)
>
> 0.344904585 seconds time elapsed
>
> Signed-off-by: Jin Yao <[email protected]>
> ---
> tools/perf/util/parse-events.c | 100 +++++++++++++++++++++++++++++++--
> tools/perf/util/parse-events.h | 6 +-
> tools/perf/util/parse-events.y | 21 ++-----
> 3 files changed, 105 insertions(+), 22 deletions(-)
>
> diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
> index 09e42245f71a..30435adc7a7b 100644
> --- a/tools/perf/util/parse-events.c
> +++ b/tools/perf/util/parse-events.c
> @@ -489,7 +489,8 @@ static int create_hybrid_cache_event(struct list_head *list, int *idx,
> static int add_hybrid_cache(struct list_head *list, int *idx,
> struct perf_event_attr *attr, char *name,
> struct list_head *config_terms,
> - bool *hybrid)
> + bool *hybrid,
> + struct parse_events_state *parse_state)
> {
> struct perf_pmu *pmu;
> int ret;
> @@ -497,6 +498,11 @@ static int add_hybrid_cache(struct list_head *list, int *idx,
> *hybrid = false;
> perf_pmu__for_each_hybrid_pmu(pmu) {
> *hybrid = true;
> + if (parse_state->pmu_name &&
> + strcmp(parse_state->pmu_name, pmu->name)) {
> + continue;
> + }
> +
> ret = create_hybrid_cache_event(list, idx, attr, name,
> config_terms, pmu);
> if (ret)
> @@ -509,7 +515,8 @@ static int add_hybrid_cache(struct list_head *list, int *idx,
> int parse_events_add_cache(struct list_head *list, int *idx,
> char *type, char *op_result1, char *op_result2,
> struct parse_events_error *err,
> - struct list_head *head_config)
> + struct list_head *head_config,
> + struct parse_events_state *parse_state)
> {
> struct perf_event_attr attr;
> LIST_HEAD(config_terms);
> @@ -582,7 +589,7 @@ int parse_events_add_cache(struct list_head *list, int *idx,
> perf_pmu__scan(NULL);
>
> ret = add_hybrid_cache(list, idx, &attr, config_name ? : name,
> - &config_terms, &hybrid);
> + &config_terms, &hybrid, parse_state);
> if (hybrid)
> return ret;
>
> @@ -1512,6 +1519,11 @@ static int add_hybrid_numeric(struct parse_events_state *parse_state,
> *hybrid = false;
> perf_pmu__for_each_hybrid_pmu(pmu) {
> *hybrid = true;
> + if (parse_state->pmu_name &&
> + strcmp(parse_state->pmu_name, pmu->name)) {
> + continue;
> + }
> +
> ret = create_hybrid_hw_event(parse_state, list, attr, pmu);
> if (ret)
> return ret;
> @@ -1578,6 +1590,10 @@ static bool config_term_percore(struct list_head *config_terms)
> return false;
> }
>
> +static int parse_events_with_hybrid_pmu(struct parse_events_state *parse_state,
> + const char *str, char *name, bool *found,
> + struct list_head *list);
> +
> int parse_events_add_pmu(struct parse_events_state *parse_state,
> struct list_head *list, char *name,
> struct list_head *head_config,
> @@ -1589,7 +1605,7 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
> struct perf_pmu *pmu;
> struct evsel *evsel;
> struct parse_events_error *err = parse_state->error;
> - bool use_uncore_alias;
> + bool use_uncore_alias, found;
> LIST_HEAD(config_terms);
>
> if (verbose > 1) {
> @@ -1605,6 +1621,22 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
> fprintf(stderr, "' that may result in non-fatal errors\n");
> }
>
> + if (head_config && perf_pmu__is_hybrid(name)) {
> + struct parse_events_term *term;
> + int ret;
> +
> + list_for_each_entry(term, head_config, list) {
> + if (!term->config)
> + continue;
> + ret = parse_events_with_hybrid_pmu(parse_state,
> + term->config,
> + name, &found,
> + list);
> + if (found)
> + return ret;
> + }
> + }
> +
> pmu = parse_state->fake_pmu ?: perf_pmu__find(name);
> if (!pmu) {
> char *err_str;
> @@ -1713,12 +1745,19 @@ int parse_events_multi_pmu_add(struct parse_events_state *parse_state,
> struct perf_pmu *pmu = NULL;
> int ok = 0;
>
> + if (parse_state->pmu_name) {
> + list = alloc_list();
> + if (!list)
> + return -1;
> + *listp = list;
> + return 0;
> + }
> +
> *listp = NULL;
> /* Add it for all PMUs that support the alias */
> - list = malloc(sizeof(struct list_head));
> + list = alloc_list();
> if (!list)
> return -1;
> - INIT_LIST_HEAD(list);
> while ((pmu = perf_pmu__scan(pmu)) != NULL) {
> struct perf_pmu_alias *alias;
>
> @@ -2284,6 +2323,44 @@ int parse_events_terms(struct list_head *terms, const char *str)
> return ret;
> }
>
> +static int list_num(struct list_head *list)
> +{
> + struct list_head *pos;
> + int n = 0;
> +
> + list_for_each(pos, list)
> + n++;
> +
> + return n;
> +}
> +
> +static int parse_events_with_hybrid_pmu(struct parse_events_state *parse_state,
> + const char *str, char *pmu_name,
> + bool *found, struct list_head *list)
> +{
> + struct parse_events_state ps = {
> + .list = LIST_HEAD_INIT(ps.list),
> + .stoken = PE_START_EVENTS,
> + .pmu_name = pmu_name,
> + .idx = parse_state->idx,
> + };
> + int ret;
> +
> + *found = false;
> + ret = parse_events__scanner(str, &ps);
> + perf_pmu__parse_cleanup();
> +
> + if (!ret) {
> + if (!list_empty(&ps.list)) {
> + *found = true;
> + list_splice(&ps.list, list);
> + parse_state->idx = list_num(list);
> + }
> + }
> +
> + return ret;
> +}
> +
> int __parse_events(struct evlist *evlist, const char *str,
> struct parse_events_error *err, struct perf_pmu *fake_pmu)
> {
> @@ -3307,3 +3384,14 @@ char *parse_events_formats_error_string(char *additional_terms)
> fail:
> return NULL;
> }
> +
> +struct list_head *alloc_list(void)
> +{
> + struct list_head *list = malloc(sizeof(*list));
> +
> + if (!list)
> + return NULL;
> +
> + INIT_LIST_HEAD(list);
> + return list;
> +}
> diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
> index e80c9b74f2f2..39c7121a4659 100644
> --- a/tools/perf/util/parse-events.h
> +++ b/tools/perf/util/parse-events.h
> @@ -138,6 +138,7 @@ struct parse_events_state {
> struct list_head *terms;
> int stoken;
> struct perf_pmu *fake_pmu;
> + char *pmu_name;
> };
>
> void parse_events__handle_error(struct parse_events_error *err, int idx,
> @@ -188,7 +189,8 @@ int parse_events_add_tool(struct parse_events_state *parse_state,
> int parse_events_add_cache(struct list_head *list, int *idx,
> char *type, char *op_result1, char *op_result2,
> struct parse_events_error *error,
> - struct list_head *head_config);
> + struct list_head *head_config,
> + struct parse_events_state *parse_state);
> int parse_events_add_breakpoint(struct list_head *list, int *idx,
> u64 addr, char *type, u64 len);
> int parse_events_add_pmu(struct parse_events_state *parse_state,
> @@ -242,6 +244,8 @@ char *parse_events_formats_error_string(char *additional_terms);
> void parse_events_print_error(struct parse_events_error *err,
> const char *event);
>
> +struct list_head *alloc_list(void);
> +
> #ifdef HAVE_LIBELF_SUPPORT
> /*
> * If the probe point starts with '%',
> diff --git a/tools/perf/util/parse-events.y b/tools/perf/util/parse-events.y
> index d57ac86ce7ca..e0e68c3da9e4 100644
> --- a/tools/perf/util/parse-events.y
> +++ b/tools/perf/util/parse-events.y
> @@ -26,18 +26,6 @@ do { \
> YYABORT; \
> } while (0)
>
> -static struct list_head* alloc_list(void)
> -{
> - struct list_head *list;
> -
> - list = malloc(sizeof(*list));
> - if (!list)
> - return NULL;
> -
> - INIT_LIST_HEAD(list);
> - return list;
> -}
> -
> static void free_list_evsel(struct list_head* list_evsel)
> {
> struct evsel *evsel, *tmp;
> @@ -454,7 +442,8 @@ PE_NAME_CACHE_TYPE '-' PE_NAME_CACHE_OP_RESULT '-' PE_NAME_CACHE_OP_RESULT opt_e
>
> list = alloc_list();
> ABORT_ON(!list);
> - err = parse_events_add_cache(list, &parse_state->idx, $1, $3, $5, error, $6);
> + err = parse_events_add_cache(list, &parse_state->idx, $1, $3, $5, error, $6,
> + parse_state);
> parse_events_terms__delete($6);
> free($1);
> free($3);
> @@ -475,7 +464,8 @@ PE_NAME_CACHE_TYPE '-' PE_NAME_CACHE_OP_RESULT opt_event_config
>
> list = alloc_list();
> ABORT_ON(!list);
> - err = parse_events_add_cache(list, &parse_state->idx, $1, $3, NULL, error, $4);
> + err = parse_events_add_cache(list, &parse_state->idx, $1, $3, NULL, error, $4,
> + parse_state);
> parse_events_terms__delete($4);
> free($1);
> free($3);
> @@ -495,7 +485,8 @@ PE_NAME_CACHE_TYPE opt_event_config
>
> list = alloc_list();
> ABORT_ON(!list);
> - err = parse_events_add_cache(list, &parse_state->idx, $1, NULL, NULL, error, $2);
> + err = parse_events_add_cache(list, &parse_state->idx, $1, NULL, NULL, error, $2,
> + parse_state);
> parse_events_terms__delete($2);
> free($1);
> if (err) {
> --
> 2.17.1
>
On Thu, Mar 11, 2021 at 03:07:22PM +0800, Jin Yao wrote:
> On hybrid platform, atom events can be only enabled on atom CPUs. Core
> events can be only enabled on core CPUs. So for a hybrid event, it can
> be only enabled on it's own CPUs.
>
> But the problem for current perf is, the cpus for evsel (via PMU sysfs)
> have been merged to evsel_list->core.all_cpus. It might be all CPUs.
>
> So we need to figure out one way to let the hybrid event only use it's
> own CPUs.
>
> The idea is to create a new evlist__invalidate_all_cpus to invalidate
> the evsel_list->core.all_cpus then evlist__for_each_cpu returns cpu -1
> for hybrid evsel. If cpu is -1, hybrid evsel will use it's own cpus.
that's wild.. I don't understand when you say we don't have
cpus for evsel, because they have been merged.. each evsel
has evsel->core.own_cpus coming from pmu->cpus, right?
why can't you just filter out cpus that are in there?
jirka
>
> We will see following code piece in patch.
>
> if (cpu == -1 && !evlist->thread_mode)
> evsel__enable_cpus(pos);
>
> It lets the event be enabled on event's own cpus.
>
> Signed-off-by: Jin Yao <[email protected]>
> ---
> tools/perf/builtin-stat.c | 37 ++++++++++++++-
> tools/perf/util/evlist.c | 72 ++++++++++++++++++++++++++++--
> tools/perf/util/evlist.h | 4 ++
> tools/perf/util/evsel.h | 8 ++++
> tools/perf/util/python-ext-sources | 2 +
> 5 files changed, 117 insertions(+), 6 deletions(-)
>
> diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
> index 2e2e4a8345ea..68ecf68699a9 100644
> --- a/tools/perf/builtin-stat.c
> +++ b/tools/perf/builtin-stat.c
> @@ -393,6 +393,18 @@ static int read_counter_cpu(struct evsel *counter, struct timespec *rs, int cpu)
> return 0;
> }
>
> +static int read_counter_cpus(struct evsel *counter, struct timespec *rs)
> +{
> + int cpu, nr_cpus, err = 0;
> + struct perf_cpu_map *cpus = evsel__cpus(counter);
> +
> + nr_cpus = cpus ? cpus->nr : 1;
> + for (cpu = 0; cpu < nr_cpus; cpu++)
> + err = read_counter_cpu(counter, rs, cpu);
> +
> + return err;
> +}
> +
> static int read_affinity_counters(struct timespec *rs)
> {
> struct evsel *counter;
> @@ -414,8 +426,14 @@ static int read_affinity_counters(struct timespec *rs)
> if (evsel__cpu_iter_skip(counter, cpu))
> continue;
> if (!counter->err) {
> - counter->err = read_counter_cpu(counter, rs,
> - counter->cpu_iter - 1);
> + if (cpu == -1 && !evsel_list->thread_mode) {
> + counter->err = read_counter_cpus(counter, rs);
> + } else if (evsel_list->thread_mode) {
> + counter->err = read_counter_cpu(counter, rs, 0);
> + } else {
> + counter->err = read_counter_cpu(counter, rs,
> + counter->cpu_iter - 1);
> + }
> }
> }
> }
> @@ -781,6 +799,21 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
> if (group)
> evlist__set_leader(evsel_list);
>
> + /*
> + * On hybrid platform, the cpus for evsel (via PMU sysfs) have been
> + * merged to evsel_list->core.all_cpus. We use evlist__invalidate_all_cpus
> + * to invalidate the evsel_list->core.all_cpus then evlist__for_each_cpu
> + * returns cpu -1 for hybrid evsel. If cpu is -1, hybrid evsel will
> + * use it's own cpus.
> + */
> + if (evlist__has_hybrid_events(evsel_list)) {
> + evlist__invalidate_all_cpus(evsel_list);
> + if (!target__has_cpu(&target) ||
> + target__has_per_thread(&target)) {
> + evsel_list->thread_mode = true;
> + }
> + }
> +
> if (affinity__setup(&affinity) < 0)
> return -1;
>
> diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
> index 882cd1f721d9..3ee12fcd0c9f 100644
> --- a/tools/perf/util/evlist.c
> +++ b/tools/perf/util/evlist.c
> @@ -381,7 +381,8 @@ bool evsel__cpu_iter_skip_no_inc(struct evsel *ev, int cpu)
> bool evsel__cpu_iter_skip(struct evsel *ev, int cpu)
> {
> if (!evsel__cpu_iter_skip_no_inc(ev, cpu)) {
> - ev->cpu_iter++;
> + if (cpu != -1)
> + ev->cpu_iter++;
> return false;
> }
> return true;
> @@ -410,6 +411,16 @@ static int evlist__is_enabled(struct evlist *evlist)
> return false;
> }
>
> +static void evsel__disable_cpus(struct evsel *evsel)
> +{
> + int cpu, nr_cpus;
> + struct perf_cpu_map *cpus = evsel__cpus(evsel);
> +
> + nr_cpus = cpus ? cpus->nr : 1;
> + for (cpu = 0; cpu < nr_cpus; cpu++)
> + evsel__disable_cpu(evsel, cpu);
> +}
> +
> static void __evlist__disable(struct evlist *evlist, char *evsel_name)
> {
> struct evsel *pos;
> @@ -436,7 +447,12 @@ static void __evlist__disable(struct evlist *evlist, char *evsel_name)
> has_imm = true;
> if (pos->immediate != imm)
> continue;
> - evsel__disable_cpu(pos, pos->cpu_iter - 1);
> + if (cpu == -1 && !evlist->thread_mode)
> + evsel__disable_cpus(pos);
> + else if (evlist->thread_mode)
> + evsel__disable_cpu(pos, 0);
> + else
> + evsel__disable_cpu(pos, pos->cpu_iter - 1);
> }
> }
> if (!has_imm)
> @@ -472,6 +488,15 @@ void evlist__disable_evsel(struct evlist *evlist, char *evsel_name)
> __evlist__disable(evlist, evsel_name);
> }
>
> +static void evsel__enable_cpus(struct evsel *evsel)
> +{
> + int cpu, nr_cpus;
> + struct perf_cpu_map *cpus = evsel__cpus(evsel);
> +
> + nr_cpus = cpus ? cpus->nr : 1;
> + for (cpu = 0; cpu < nr_cpus; cpu++)
> + evsel__enable_cpu(evsel, cpu);
> +}
> static void __evlist__enable(struct evlist *evlist, char *evsel_name)
> {
> struct evsel *pos;
> @@ -491,7 +516,12 @@ static void __evlist__enable(struct evlist *evlist, char *evsel_name)
> continue;
> if (!evsel__is_group_leader(pos) || !pos->core.fd)
> continue;
> - evsel__enable_cpu(pos, pos->cpu_iter - 1);
> + if (cpu == -1 && !evlist->thread_mode)
> + evsel__enable_cpus(pos);
> + else if (evlist->thread_mode)
> + evsel__enable_cpu(pos, 0);
> + else
> + evsel__enable_cpu(pos, pos->cpu_iter - 1);
> }
> }
> affinity__cleanup(&affinity);
> @@ -1274,6 +1304,16 @@ void evlist__set_selected(struct evlist *evlist, struct evsel *evsel)
> evlist->selected = evsel;
> }
>
> +static void evsel__close_cpus(struct evsel *evsel)
> +{
> + int cpu, nr_cpus;
> + struct perf_cpu_map *cpus = evsel__cpus(evsel);
> +
> + nr_cpus = cpus ? cpus->nr : 1;
> + for (cpu = 0; cpu < nr_cpus; cpu++)
> + perf_evsel__close_cpu(&evsel->core, cpu);
> +}
> +
> void evlist__close(struct evlist *evlist)
> {
> struct evsel *evsel;
> @@ -1298,7 +1338,13 @@ void evlist__close(struct evlist *evlist)
> evlist__for_each_entry_reverse(evlist, evsel) {
> if (evsel__cpu_iter_skip(evsel, cpu))
> continue;
> - perf_evsel__close_cpu(&evsel->core, evsel->cpu_iter - 1);
> +
> + if (cpu == -1 && !evlist->thread_mode)
> + evsel__close_cpus(evsel);
> + else if (evlist->thread_mode)
> + perf_evsel__close_cpu(&evsel->core, 0);
> + else
> + perf_evsel__close_cpu(&evsel->core, evsel->cpu_iter - 1);
> }
> }
> affinity__cleanup(&affinity);
> @@ -2130,3 +2176,21 @@ struct evsel *evlist__find_evsel(struct evlist *evlist, int idx)
> }
> return NULL;
> }
> +
> +bool evlist__has_hybrid_events(struct evlist *evlist)
> +{
> + struct evsel *evsel;
> +
> + evlist__for_each_entry(evlist, evsel) {
> + if (evsel__is_hybrid_event(evsel))
> + return true;
> + }
> +
> + return false;
> +}
> +
> +void evlist__invalidate_all_cpus(struct evlist *evlist)
> +{
> + perf_cpu_map__put(evlist->core.all_cpus);
> + evlist->core.all_cpus = perf_cpu_map__empty_new(1);
> +}
> diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
> index b695ffaae519..0da683511d98 100644
> --- a/tools/perf/util/evlist.h
> +++ b/tools/perf/util/evlist.h
> @@ -52,6 +52,7 @@ struct evlist {
> struct perf_evlist core;
> int nr_groups;
> bool enabled;
> + bool thread_mode;
> int id_pos;
> int is_pos;
> u64 combined_sample_type;
> @@ -365,4 +366,7 @@ int evlist__ctlfd_ack(struct evlist *evlist);
> #define EVLIST_DISABLED_MSG "Events disabled\n"
>
> struct evsel *evlist__find_evsel(struct evlist *evlist, int idx);
> +void evlist__invalidate_all_cpus(struct evlist *evlist);
> +
> +bool evlist__has_hybrid_events(struct evlist *evlist);
> #endif /* __PERF_EVLIST_H */
> diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
> index 6026487353dd..69aadc52c1bd 100644
> --- a/tools/perf/util/evsel.h
> +++ b/tools/perf/util/evsel.h
> @@ -7,9 +7,11 @@
> #include <sys/types.h>
> #include <linux/perf_event.h>
> #include <linux/types.h>
> +#include <string.h>
> #include <internal/evsel.h>
> #include <perf/evsel.h>
> #include "symbol_conf.h"
> +#include "pmu-hybrid.h"
> #include <internal/cpumap.h>
>
> struct bpf_object;
> @@ -435,4 +437,10 @@ struct perf_env *evsel__env(struct evsel *evsel);
> int evsel__store_ids(struct evsel *evsel, struct evlist *evlist);
>
> void evsel__zero_per_pkg(struct evsel *evsel);
> +
> +static inline bool evsel__is_hybrid_event(struct evsel *evsel)
> +{
> + return evsel->pmu_name && perf_pmu__is_hybrid(evsel->pmu_name);
> +}
> +
> #endif /* __PERF_EVSEL_H */
> diff --git a/tools/perf/util/python-ext-sources b/tools/perf/util/python-ext-sources
> index 845dd46e3c61..d7c976671e3a 100644
> --- a/tools/perf/util/python-ext-sources
> +++ b/tools/perf/util/python-ext-sources
> @@ -37,3 +37,5 @@ util/units.c
> util/affinity.c
> util/rwsem.c
> util/hashmap.c
> +util/pmu-hybrid.c
> +util/fncache.c
> --
> 2.17.1
>
On Thu, Mar 11, 2021 at 03:07:24PM +0800, Jin Yao wrote:
SNIP
> cycles: 4: 800933425 1002536659 1002536659
> cycles: 5: 800928573 1002528386 1002528386
> cycles: 6: 800924347 1002520527 1002520527
> cycles: 7: 800922009 1002513176 1002513176
> cycles: 8: 800919624 1002507326 1002507326
> cycles: 9: 800917204 1002500663 1002500663
> cycles: 10: 802096579 1002494280 1002494280
> cycles: 11: 802093770 1002486404 1002486404
> cycles: 12: 803284338 1002479491 1002479491
> cycles: 13: 803277609 1002469777 1002469777
> cycles: 14: 800875902 1002458861 1002458861
> cycles: 15: 800873241 1002451350 1002451350
> cycles: 0: 800837379 1002444645 1002444645
> cycles: 1: 800833400 1002438505 1002438505
> cycles: 2: 800829291 1002433698 1002433698
> cycles: 3: 800824390 1002427584 1002427584
> cycles: 4: 800819360 1002422099 1002422099
> cycles: 5: 800814787 1002415845 1002415845
> cycles: 6: 800810125 1002410301 1002410301
> cycles: 7: 800791893 1002386845 1002386845
> cycles: 12855737722 16040169029 16040169029
> cycles: 6406560625 8019379522 8019379522
>
> Performance counter stats for 'system wide':
>
> 12,855,737,722 cpu_core/cycles/
> 6,406,560,625 cpu_atom/cycles/
so we do that no_merge stuff for uncore pmus, why can't we do
that in here? that'd seems like generic way
jirka
Hi Jiri,
On 3/13/2021 3:15 AM, Jiri Olsa wrote:
> On Thu, Mar 11, 2021 at 03:07:22PM +0800, Jin Yao wrote:
>> On hybrid platform, atom events can be only enabled on atom CPUs. Core
>> events can be only enabled on core CPUs. So for a hybrid event, it can
>> be only enabled on it's own CPUs.
>>
>> But the problem for current perf is, the cpus for evsel (via PMU sysfs)
>> have been merged to evsel_list->core.all_cpus. It might be all CPUs.
>>
>> So we need to figure out one way to let the hybrid event only use it's
>> own CPUs.
>>
>> The idea is to create a new evlist__invalidate_all_cpus to invalidate
>> the evsel_list->core.all_cpus then evlist__for_each_cpu returns cpu -1
>> for hybrid evsel. If cpu is -1, hybrid evsel will use it's own cpus.
>
> that's wild.. I don't understand when you say we don't have
> cpus for evsel, because they have been merged.. each evsel
> has evsel->core.own_cpus coming from pmu->cpus, right?
>
> why can't you just filter out cpus that are in there?
>
> jirka
>
Yes, you're right. This patch is wide and actually it's not very necessary.
The current framework has processed the cpus for evsel well even for hybrid evsel. So this patch can
be dropped.
Thanks
Jin Yao
Hi Jiri,
On 3/13/2021 3:15 AM, Jiri Olsa wrote:
> On Thu, Mar 11, 2021 at 03:07:24PM +0800, Jin Yao wrote:
>
> SNIP
>
>> cycles: 4: 800933425 1002536659 1002536659
>> cycles: 5: 800928573 1002528386 1002528386
>> cycles: 6: 800924347 1002520527 1002520527
>> cycles: 7: 800922009 1002513176 1002513176
>> cycles: 8: 800919624 1002507326 1002507326
>> cycles: 9: 800917204 1002500663 1002500663
>> cycles: 10: 802096579 1002494280 1002494280
>> cycles: 11: 802093770 1002486404 1002486404
>> cycles: 12: 803284338 1002479491 1002479491
>> cycles: 13: 803277609 1002469777 1002469777
>> cycles: 14: 800875902 1002458861 1002458861
>> cycles: 15: 800873241 1002451350 1002451350
>> cycles: 0: 800837379 1002444645 1002444645
>> cycles: 1: 800833400 1002438505 1002438505
>> cycles: 2: 800829291 1002433698 1002433698
>> cycles: 3: 800824390 1002427584 1002427584
>> cycles: 4: 800819360 1002422099 1002422099
>> cycles: 5: 800814787 1002415845 1002415845
>> cycles: 6: 800810125 1002410301 1002410301
>> cycles: 7: 800791893 1002386845 1002386845
>> cycles: 12855737722 16040169029 16040169029
>> cycles: 6406560625 8019379522 8019379522
>>
>> Performance counter stats for 'system wide':
>>
>> 12,855,737,722 cpu_core/cycles/
>> 6,406,560,625 cpu_atom/cycles/
>
> so we do that no_merge stuff for uncore pmus, why can't we do
> that in here? that'd seems like generic way
>
> jirka
>
We have set the "stat_config.no_merge = true;" in "[PATCH v2 08/27] perf stat: Uniquify hybrid event
name".
For hybrid hardware events, they have different configs. The config is 0xDD000000AA (0x400000000 for
core vs. 0xa00000000 for atom in this example)
We use perf_pmu__for_each_hybrid_pmu() to iterate all hybrid PMUs, generate the configs and create
the evsels for each hybrid PMU. This logic and the code are not complex and easy to understand.
Uncore looks complicated. It has uncore alias concept which is for different PMUs but with same
prefix. Such as "uncore_cbox" for "uncore_cbox_0" to "uncore_cbox_9". But the uncore alias concept
doesn't apply to hybrid pmu (we just have "cpu_core" and "cpu_atom" here). And actually I also don't
want to mix the core stuff with uncore stuff, that would be hard for understanding.
Perhaps I misunderstand, correct me if I'm wrong.
Thanks
Jin Yao
Hi Jiri,
On 3/13/2021 3:15 AM, Jiri Olsa wrote:
> On Thu, Mar 11, 2021 at 03:07:26PM +0800, Jin Yao wrote:
>> On hybrid platform, some hardware events are only available
>> on a specific pmu. For example, 'L1-dcache-load-misses' is only
>> available on 'cpu_core' pmu. And even for the event which can be
>> available on both pmus, the user also may want to just enable
>> one event. So now following syntax is supported:
>>
>> cpu_core/<hardware event>/
>> cpu_core/<hardware cache event>/
>> cpu_core/<pmu event>/
>>
>> cpu_atom/<hardware event>/
>> cpu_atom/<hardware cache event>/
>> cpu_atom/<pmu event>/
>>
>> It limits the event to be enabled only on a specified pmu.
>>
>> The patch uses this idea, for example, if we use "cpu_core/LLC-loads/",
>> in parse_events_add_pmu(), term->config is "LLC-loads".
>
> hum, I don't understand how this doest not work even now,
> I assume both cpu_core and cpu_atom have sysfs device directory
> with events/ directory right?
>
Yes, we have cpu_core and cpu_atom directories with events.
root@ssp-pwrt-002:/sys/devices/cpu_atom/events# ls
branch-instructions bus-cycles cache-references instructions mem-stores topdown-bad-spec
topdown-fe-bound
branch-misses cache-misses cpu-cycles mem-loads ref-cycles topdown-be-bound
topdown-retiring
root@ssp-pwrt-002:/sys/devices/cpu_core/events# ls
branch-instructions cache-misses instructions mem-stores topdown-bad-spec
topdown-fe-bound topdown-mem-bound
branch-misses cache-references mem-loads ref-cycles topdown-be-bound
topdown-fetch-lat topdown-retiring
bus-cycles cpu-cycles mem-loads-aux slots topdown-br-mispredict
topdown-heavy-ops
> and whatever is defined in events we allow in parsing syntax..
>
> why can't we treat them like 2 separated pmus?
>
But if without this patch, it reports the error,
root@ssp-pwrt-002:~# ./perf stat -e cpu_core/cycles/ -a -vv -- sleep 1
event syntax error: 'cpu_core/cycles/'
\___ unknown term 'cycles' for pmu 'cpu_core'
valid terms:
event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
Initial error:
event syntax error: 'cpu_core/cycles/'
\___ unknown term 'cycles' for pmu 'cpu_core'
valid terms:
event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
Run 'perf list' for a list of valid events
The 'cycles' is treated as a unknown term, then it errors out.
So we have to create another parser to scan the term.
Thanks
Jin Yao
> thanks,
> jirka
>
>>
>> We create a new "parse_events_state" with the pmu_name and use
>> parse_events__scanner to scan the term->config (the string "LLC-loads"
>> in this example). The parse_events_add_cache() will be called during
>> parsing. The parse_state->pmu_name is used to identify the pmu
>> where the event is enabled.
>>
>> Let's see examples:
>>
>> root@ssp-pwrt-002:~# ./perf stat -e cpu_core/cycles/,cpu_core/LLC-loads/ -vv -- ./triad_loop
>> Control descriptor is not initialized
>> ------------------------------------------------------------
>> perf_event_attr:
>> type 6
>> size 120
>> config 0x400000000
>> sample_type IDENTIFIER
>> read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
>> disabled 1
>> inherit 1
>> enable_on_exec 1
>> exclude_guest 1
>> ------------------------------------------------------------
>> sys_perf_event_open: pid 7267 cpu -1 group_fd -1 flags 0x8 = 3
>> ------------------------------------------------------------
>> perf_event_attr:
>> type 7
>> size 120
>> config 0x400000002
>> sample_type IDENTIFIER
>> read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
>> disabled 1
>> inherit 1
>> enable_on_exec 1
>> exclude_guest 1
>> ------------------------------------------------------------
>> sys_perf_event_open: pid 7267 cpu -1 group_fd -1 flags 0x8 = 4
>> cycles: 0: 449252097 297999924 297999924
>> LLC-loads: 0: 1857 297999924 297999924
>> cycles: 449252097 297999924 297999924
>> LLC-loads: 1857 297999924 297999924
>>
>> Performance counter stats for './triad_loop':
>>
>> 449,252,097 cpu_core/cycles/
>> 1,857 cpu_core/LLC-loads/
>>
>> 0.298898415 seconds time elapsed
>>
>> root@ssp-pwrt-002:~# ./perf stat -e cpu_atom/cycles/,cpu_atom/LLC-loads/ -vv -- taskset -c 16 ./triad_loop
>> Control descriptor is not initialized
>> ------------------------------------------------------------
>> perf_event_attr:
>> type 6
>> size 120
>> config 0xa00000000
>> sample_type IDENTIFIER
>> read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
>> disabled 1
>> inherit 1
>> enable_on_exec 1
>> exclude_guest 1
>> ------------------------------------------------------------
>> sys_perf_event_open: pid 7339 cpu -1 group_fd -1 flags 0x8 = 3
>> ------------------------------------------------------------
>> perf_event_attr:
>> type 7
>> size 120
>> config 0xa00000002
>> sample_type IDENTIFIER
>> read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
>> disabled 1
>> inherit 1
>> enable_on_exec 1
>> exclude_guest 1
>> ------------------------------------------------------------
>> sys_perf_event_open: pid 7339 cpu -1 group_fd -1 flags 0x8 = 4
>> cycles: 0: 602020010 343657939 342553275
>> LLC-loads: 0: 3537 343657939 342553275
>> cycles: 603961400 343657939 342553275
>> LLC-loads: 3548 343657939 342553275
>>
>> Performance counter stats for 'taskset -c 16 ./triad_loop':
>>
>> 603,961,400 cpu_atom/cycles/ (99.68%)
>> 3,548 cpu_atom/LLC-loads/ (99.68%)
>>
>> 0.344904585 seconds time elapsed
>>
>> Signed-off-by: Jin Yao <[email protected]>
>> ---
>> tools/perf/util/parse-events.c | 100 +++++++++++++++++++++++++++++++--
>> tools/perf/util/parse-events.h | 6 +-
>> tools/perf/util/parse-events.y | 21 ++-----
>> 3 files changed, 105 insertions(+), 22 deletions(-)
>>
>> diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
>> index 09e42245f71a..30435adc7a7b 100644
>> --- a/tools/perf/util/parse-events.c
>> +++ b/tools/perf/util/parse-events.c
>> @@ -489,7 +489,8 @@ static int create_hybrid_cache_event(struct list_head *list, int *idx,
>> static int add_hybrid_cache(struct list_head *list, int *idx,
>> struct perf_event_attr *attr, char *name,
>> struct list_head *config_terms,
>> - bool *hybrid)
>> + bool *hybrid,
>> + struct parse_events_state *parse_state)
>> {
>> struct perf_pmu *pmu;
>> int ret;
>> @@ -497,6 +498,11 @@ static int add_hybrid_cache(struct list_head *list, int *idx,
>> *hybrid = false;
>> perf_pmu__for_each_hybrid_pmu(pmu) {
>> *hybrid = true;
>> + if (parse_state->pmu_name &&
>> + strcmp(parse_state->pmu_name, pmu->name)) {
>> + continue;
>> + }
>> +
>> ret = create_hybrid_cache_event(list, idx, attr, name,
>> config_terms, pmu);
>> if (ret)
>> @@ -509,7 +515,8 @@ static int add_hybrid_cache(struct list_head *list, int *idx,
>> int parse_events_add_cache(struct list_head *list, int *idx,
>> char *type, char *op_result1, char *op_result2,
>> struct parse_events_error *err,
>> - struct list_head *head_config)
>> + struct list_head *head_config,
>> + struct parse_events_state *parse_state)
>> {
>> struct perf_event_attr attr;
>> LIST_HEAD(config_terms);
>> @@ -582,7 +589,7 @@ int parse_events_add_cache(struct list_head *list, int *idx,
>> perf_pmu__scan(NULL);
>>
>> ret = add_hybrid_cache(list, idx, &attr, config_name ? : name,
>> - &config_terms, &hybrid);
>> + &config_terms, &hybrid, parse_state);
>> if (hybrid)
>> return ret;
>>
>> @@ -1512,6 +1519,11 @@ static int add_hybrid_numeric(struct parse_events_state *parse_state,
>> *hybrid = false;
>> perf_pmu__for_each_hybrid_pmu(pmu) {
>> *hybrid = true;
>> + if (parse_state->pmu_name &&
>> + strcmp(parse_state->pmu_name, pmu->name)) {
>> + continue;
>> + }
>> +
>> ret = create_hybrid_hw_event(parse_state, list, attr, pmu);
>> if (ret)
>> return ret;
>> @@ -1578,6 +1590,10 @@ static bool config_term_percore(struct list_head *config_terms)
>> return false;
>> }
>>
>> +static int parse_events_with_hybrid_pmu(struct parse_events_state *parse_state,
>> + const char *str, char *name, bool *found,
>> + struct list_head *list);
>> +
>> int parse_events_add_pmu(struct parse_events_state *parse_state,
>> struct list_head *list, char *name,
>> struct list_head *head_config,
>> @@ -1589,7 +1605,7 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
>> struct perf_pmu *pmu;
>> struct evsel *evsel;
>> struct parse_events_error *err = parse_state->error;
>> - bool use_uncore_alias;
>> + bool use_uncore_alias, found;
>> LIST_HEAD(config_terms);
>>
>> if (verbose > 1) {
>> @@ -1605,6 +1621,22 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
>> fprintf(stderr, "' that may result in non-fatal errors\n");
>> }
>>
>> + if (head_config && perf_pmu__is_hybrid(name)) {
>> + struct parse_events_term *term;
>> + int ret;
>> +
>> + list_for_each_entry(term, head_config, list) {
>> + if (!term->config)
>> + continue;
>> + ret = parse_events_with_hybrid_pmu(parse_state,
>> + term->config,
>> + name, &found,
>> + list);
>> + if (found)
>> + return ret;
>> + }
>> + }
>> +
>> pmu = parse_state->fake_pmu ?: perf_pmu__find(name);
>> if (!pmu) {
>> char *err_str;
>> @@ -1713,12 +1745,19 @@ int parse_events_multi_pmu_add(struct parse_events_state *parse_state,
>> struct perf_pmu *pmu = NULL;
>> int ok = 0;
>>
>> + if (parse_state->pmu_name) {
>> + list = alloc_list();
>> + if (!list)
>> + return -1;
>> + *listp = list;
>> + return 0;
>> + }
>> +
>> *listp = NULL;
>> /* Add it for all PMUs that support the alias */
>> - list = malloc(sizeof(struct list_head));
>> + list = alloc_list();
>> if (!list)
>> return -1;
>> - INIT_LIST_HEAD(list);
>> while ((pmu = perf_pmu__scan(pmu)) != NULL) {
>> struct perf_pmu_alias *alias;
>>
>> @@ -2284,6 +2323,44 @@ int parse_events_terms(struct list_head *terms, const char *str)
>> return ret;
>> }
>>
>> +static int list_num(struct list_head *list)
>> +{
>> + struct list_head *pos;
>> + int n = 0;
>> +
>> + list_for_each(pos, list)
>> + n++;
>> +
>> + return n;
>> +}
>> +
>> +static int parse_events_with_hybrid_pmu(struct parse_events_state *parse_state,
>> + const char *str, char *pmu_name,
>> + bool *found, struct list_head *list)
>> +{
>> + struct parse_events_state ps = {
>> + .list = LIST_HEAD_INIT(ps.list),
>> + .stoken = PE_START_EVENTS,
>> + .pmu_name = pmu_name,
>> + .idx = parse_state->idx,
>> + };
>> + int ret;
>> +
>> + *found = false;
>> + ret = parse_events__scanner(str, &ps);
>> + perf_pmu__parse_cleanup();
>> +
>> + if (!ret) {
>> + if (!list_empty(&ps.list)) {
>> + *found = true;
>> + list_splice(&ps.list, list);
>> + parse_state->idx = list_num(list);
>> + }
>> + }
>> +
>> + return ret;
>> +}
>> +
>> int __parse_events(struct evlist *evlist, const char *str,
>> struct parse_events_error *err, struct perf_pmu *fake_pmu)
>> {
>> @@ -3307,3 +3384,14 @@ char *parse_events_formats_error_string(char *additional_terms)
>> fail:
>> return NULL;
>> }
>> +
>> +struct list_head *alloc_list(void)
>> +{
>> + struct list_head *list = malloc(sizeof(*list));
>> +
>> + if (!list)
>> + return NULL;
>> +
>> + INIT_LIST_HEAD(list);
>> + return list;
>> +}
>> diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
>> index e80c9b74f2f2..39c7121a4659 100644
>> --- a/tools/perf/util/parse-events.h
>> +++ b/tools/perf/util/parse-events.h
>> @@ -138,6 +138,7 @@ struct parse_events_state {
>> struct list_head *terms;
>> int stoken;
>> struct perf_pmu *fake_pmu;
>> + char *pmu_name;
>> };
>>
>> void parse_events__handle_error(struct parse_events_error *err, int idx,
>> @@ -188,7 +189,8 @@ int parse_events_add_tool(struct parse_events_state *parse_state,
>> int parse_events_add_cache(struct list_head *list, int *idx,
>> char *type, char *op_result1, char *op_result2,
>> struct parse_events_error *error,
>> - struct list_head *head_config);
>> + struct list_head *head_config,
>> + struct parse_events_state *parse_state);
>> int parse_events_add_breakpoint(struct list_head *list, int *idx,
>> u64 addr, char *type, u64 len);
>> int parse_events_add_pmu(struct parse_events_state *parse_state,
>> @@ -242,6 +244,8 @@ char *parse_events_formats_error_string(char *additional_terms);
>> void parse_events_print_error(struct parse_events_error *err,
>> const char *event);
>>
>> +struct list_head *alloc_list(void);
>> +
>> #ifdef HAVE_LIBELF_SUPPORT
>> /*
>> * If the probe point starts with '%',
>> diff --git a/tools/perf/util/parse-events.y b/tools/perf/util/parse-events.y
>> index d57ac86ce7ca..e0e68c3da9e4 100644
>> --- a/tools/perf/util/parse-events.y
>> +++ b/tools/perf/util/parse-events.y
>> @@ -26,18 +26,6 @@ do { \
>> YYABORT; \
>> } while (0)
>>
>> -static struct list_head* alloc_list(void)
>> -{
>> - struct list_head *list;
>> -
>> - list = malloc(sizeof(*list));
>> - if (!list)
>> - return NULL;
>> -
>> - INIT_LIST_HEAD(list);
>> - return list;
>> -}
>> -
>> static void free_list_evsel(struct list_head* list_evsel)
>> {
>> struct evsel *evsel, *tmp;
>> @@ -454,7 +442,8 @@ PE_NAME_CACHE_TYPE '-' PE_NAME_CACHE_OP_RESULT '-' PE_NAME_CACHE_OP_RESULT opt_e
>>
>> list = alloc_list();
>> ABORT_ON(!list);
>> - err = parse_events_add_cache(list, &parse_state->idx, $1, $3, $5, error, $6);
>> + err = parse_events_add_cache(list, &parse_state->idx, $1, $3, $5, error, $6,
>> + parse_state);
>> parse_events_terms__delete($6);
>> free($1);
>> free($3);
>> @@ -475,7 +464,8 @@ PE_NAME_CACHE_TYPE '-' PE_NAME_CACHE_OP_RESULT opt_event_config
>>
>> list = alloc_list();
>> ABORT_ON(!list);
>> - err = parse_events_add_cache(list, &parse_state->idx, $1, $3, NULL, error, $4);
>> + err = parse_events_add_cache(list, &parse_state->idx, $1, $3, NULL, error, $4,
>> + parse_state);
>> parse_events_terms__delete($4);
>> free($1);
>> free($3);
>> @@ -495,7 +485,8 @@ PE_NAME_CACHE_TYPE opt_event_config
>>
>> list = alloc_list();
>> ABORT_ON(!list);
>> - err = parse_events_add_cache(list, &parse_state->idx, $1, NULL, NULL, error, $2);
>> + err = parse_events_add_cache(list, &parse_state->idx, $1, NULL, NULL, error, $2,
>> + parse_state);
>> parse_events_terms__delete($2);
>> free($1);
>> if (err) {
>> --
>> 2.17.1
>>
>
On Mon, Mar 15, 2021 at 10:28:12AM +0800, Jin, Yao wrote:
> Hi Jiri,
>
> On 3/13/2021 3:15 AM, Jiri Olsa wrote:
> > On Thu, Mar 11, 2021 at 03:07:26PM +0800, Jin Yao wrote:
> > > On hybrid platform, some hardware events are only available
> > > on a specific pmu. For example, 'L1-dcache-load-misses' is only
> > > available on 'cpu_core' pmu. And even for the event which can be
> > > available on both pmus, the user also may want to just enable
> > > one event. So now following syntax is supported:
> > >
> > > cpu_core/<hardware event>/
> > > cpu_core/<hardware cache event>/
> > > cpu_core/<pmu event>/
> > >
> > > cpu_atom/<hardware event>/
> > > cpu_atom/<hardware cache event>/
> > > cpu_atom/<pmu event>/
> > >
> > > It limits the event to be enabled only on a specified pmu.
> > >
> > > The patch uses this idea, for example, if we use "cpu_core/LLC-loads/",
> > > in parse_events_add_pmu(), term->config is "LLC-loads".
> >
> > hum, I don't understand how this doest not work even now,
> > I assume both cpu_core and cpu_atom have sysfs device directory
> > with events/ directory right?
> >
>
> Yes, we have cpu_core and cpu_atom directories with events.
>
> root@ssp-pwrt-002:/sys/devices/cpu_atom/events# ls
> branch-instructions bus-cycles cache-references instructions
> mem-stores topdown-bad-spec topdown-fe-bound
> branch-misses cache-misses cpu-cycles mem-loads
> ref-cycles topdown-be-bound topdown-retiring
>
> root@ssp-pwrt-002:/sys/devices/cpu_core/events# ls
> branch-instructions cache-misses instructions mem-stores
> topdown-bad-spec topdown-fe-bound topdown-mem-bound
> branch-misses cache-references mem-loads ref-cycles
> topdown-be-bound topdown-fetch-lat topdown-retiring
> bus-cycles cpu-cycles mem-loads-aux slots
> topdown-br-mispredict topdown-heavy-ops
>
> > and whatever is defined in events we allow in parsing syntax..
> >
> > why can't we treat them like 2 separated pmus?
> >
>
> But if without this patch, it reports the error,
>
> root@ssp-pwrt-002:~# ./perf stat -e cpu_core/cycles/ -a -vv -- sleep 1
> event syntax error: 'cpu_core/cycles/'
> \___ unknown term 'cycles' for pmu 'cpu_core'
yep, because there's special care for 'cycles' unfortunately,
but you should be able to run 'cpu_core/cpu-cycles/' right?
>
> valid terms: event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
>
> Initial error:
> event syntax error: 'cpu_core/cycles/'
> \___ unknown term 'cycles' for pmu 'cpu_core'
>
> valid terms: event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
> Run 'perf list' for a list of valid events
>
> The 'cycles' is treated as a unknown term, then it errors out.
yep, because it's not in events.. we could add special rule to
treat cycles as cpu-cycles inside pmu definition ;-)
jirka
On Mon, Mar 15, 2021 at 10:04:56AM +0800, Jin, Yao wrote:
> Hi Jiri,
>
> On 3/13/2021 3:15 AM, Jiri Olsa wrote:
> > On Thu, Mar 11, 2021 at 03:07:24PM +0800, Jin Yao wrote:
> >
> > SNIP
> >
> > > cycles: 4: 800933425 1002536659 1002536659
> > > cycles: 5: 800928573 1002528386 1002528386
> > > cycles: 6: 800924347 1002520527 1002520527
> > > cycles: 7: 800922009 1002513176 1002513176
> > > cycles: 8: 800919624 1002507326 1002507326
> > > cycles: 9: 800917204 1002500663 1002500663
> > > cycles: 10: 802096579 1002494280 1002494280
> > > cycles: 11: 802093770 1002486404 1002486404
> > > cycles: 12: 803284338 1002479491 1002479491
> > > cycles: 13: 803277609 1002469777 1002469777
> > > cycles: 14: 800875902 1002458861 1002458861
> > > cycles: 15: 800873241 1002451350 1002451350
> > > cycles: 0: 800837379 1002444645 1002444645
> > > cycles: 1: 800833400 1002438505 1002438505
> > > cycles: 2: 800829291 1002433698 1002433698
> > > cycles: 3: 800824390 1002427584 1002427584
> > > cycles: 4: 800819360 1002422099 1002422099
> > > cycles: 5: 800814787 1002415845 1002415845
> > > cycles: 6: 800810125 1002410301 1002410301
> > > cycles: 7: 800791893 1002386845 1002386845
> > > cycles: 12855737722 16040169029 16040169029
> > > cycles: 6406560625 8019379522 8019379522
> > >
> > > Performance counter stats for 'system wide':
> > >
> > > 12,855,737,722 cpu_core/cycles/
> > > 6,406,560,625 cpu_atom/cycles/
> >
> > so we do that no_merge stuff for uncore pmus, why can't we do
> > that in here? that'd seems like generic way
> >
> > jirka
> >
>
> We have set the "stat_config.no_merge = true;" in "[PATCH v2 08/27] perf
> stat: Uniquify hybrid event name".
>
> For hybrid hardware events, they have different configs. The config is
> 0xDD000000AA (0x400000000 for core vs. 0xa00000000 for atom in this example)
>
> We use perf_pmu__for_each_hybrid_pmu() to iterate all hybrid PMUs, generate
> the configs and create the evsels for each hybrid PMU. This logic and the
> code are not complex and easy to understand.
>
> Uncore looks complicated. It has uncore alias concept which is for different
> PMUs but with same prefix. Such as "uncore_cbox" for "uncore_cbox_0" to
> "uncore_cbox_9". But the uncore alias concept doesn't apply to hybrid pmu
> (we just have "cpu_core" and "cpu_atom" here). And actually I also don't
> want to mix the core stuff with uncore stuff, that would be hard for
> understanding.
>
> Perhaps I misunderstand, correct me if I'm wrong.
not sure, I thought the merging stuff was more generic,
because the change looks too specific for me, I'll try
to check on it more deeply
jirka
On Thu, Mar 11, 2021 at 03:07:23PM +0800, Jin Yao wrote:
> It would be useful to tell user the pmu which the event belongs to.
> perf-stat has supported '--no-merge' option and it can print the pmu
> name after the event name, such as:
>
> "cycles [cpu_core]"
>
> Now this option is enabled by default for hybrid platform but change
> the format to:
>
> "cpu_core/cycles/"
>
> Signed-off-by: Jin Yao <[email protected]>
> ---
> tools/perf/builtin-stat.c | 3 +++
> tools/perf/util/stat-display.c | 12 ++++++++++--
> 2 files changed, 13 insertions(+), 2 deletions(-)
>
> diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
> index 68ecf68699a9..6c0a21323814 100644
> --- a/tools/perf/builtin-stat.c
> +++ b/tools/perf/builtin-stat.c
> @@ -2390,6 +2390,9 @@ int cmd_stat(int argc, const char **argv)
>
> evlist__check_cpu_maps(evsel_list);
>
> + if (perf_pmu__hybrid_exist())
> + stat_config.no_merge = true;
> +
> /*
> * Initialize thread_map with comm names,
> * so we could print it out on output.
> diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c
> index 7f09cdaf5b60..ed37d8e7ea1a 100644
> --- a/tools/perf/util/stat-display.c
> +++ b/tools/perf/util/stat-display.c
> @@ -526,6 +526,7 @@ static void uniquify_event_name(struct evsel *counter)
> {
> char *new_name;
> char *config;
> + int ret;
>
> if (counter->uniquified_name ||
> !counter->pmu_name || !strncmp(counter->name, counter->pmu_name,
> @@ -540,8 +541,15 @@ static void uniquify_event_name(struct evsel *counter)
> counter->name = new_name;
> }
> } else {
> - if (asprintf(&new_name,
> - "%s [%s]", counter->name, counter->pmu_name) > 0) {
> + if (perf_pmu__hybrid_exist()) {
I'm still not sure about the whole thing, but should you
check in here jut for count->pmu_name instead for hybrid
globaly?
jirka
> + ret = asprintf(&new_name, "%s/%s/",
> + counter->pmu_name, counter->name);
> + } else {
> + ret = asprintf(&new_name, "%s [%s]",
> + counter->name, counter->pmu_name);
> + }
> +
> + if (ret) {
> free(counter->name);
> counter->name = new_name;
> }
> --
> 2.17.1
>
On Thu, Mar 11, 2021 at 03:07:24PM +0800, Jin Yao wrote:
> For hardware events, they have pre-defined configs. The kernel
> needs to know where the event comes from (e.g. from cpu_core pmu
> or from cpu_atom pmu). But the perf type 'PERF_TYPE_HARDWARE'
> can't carry pmu information.
>
> So the kernel introduces a new type 'PERF_TYPE_HARDWARE_PMU'.
> The new attr.config layout for PERF_TYPE_HARDWARE_PMU is:
>
> 0xDD000000AA
> AA: original hardware event ID
> DD: PMU type ID
>
> PMU type ID is retrieved from sysfs. For example,
>
> cat /sys/devices/cpu_atom/type
> 10
>
> cat /sys/devices/cpu_core/type
> 4
>
> When enabling a hybrid hardware event without specified pmu, such as,
> 'perf stat -e cycles -a', two events are created automatically. One
> is for atom, the other is for core.
ok I think I understand the need for this (and the following) patch
the perf_hw_id counters could be global, so when you specify only
event like:
-e cycles
you want all the cycles, which on hybrid system means cycles from
more than one pmus
SNIP
> @@ -1416,6 +1475,8 @@ int parse_events_add_numeric(struct parse_events_state *parse_state,
> {
> struct perf_event_attr attr;
> LIST_HEAD(config_terms);
> + bool hybrid;
> + int ret;
>
> memset(&attr, 0, sizeof(attr));
> attr.type = type;
> @@ -1430,6 +1491,18 @@ int parse_events_add_numeric(struct parse_events_state *parse_state,
> return -ENOMEM;
> }
>
> + /*
> + * Skip the software dummy event.
> + */
> + if (type != PERF_TYPE_SOFTWARE) {
> + if (!perf_pmu__hybrid_exist())
> + perf_pmu__scan(NULL);
this could be checked in the following add_hybrid_numeric call
> +
> + ret = add_hybrid_numeric(parse_state, list, &attr, &hybrid);
> + if (hybrid)
> + return ret;
> + }
could we add this to separate object.. hybrid.c or maybe parse-events-hybrid.c,
there's already global __add_event wrapper - parse_events__add_event
jirka
> +
> return add_event(list, &parse_state->idx, &attr,
> get_config_name(head_config), &config_terms);
> }
> --
> 2.17.1
>
On Thu, Mar 11, 2021 at 03:07:32PM +0800, Jin Yao wrote:
> A group mixed with hybrid event and global event is allowed. For example,
> group leader is 'cpu-clock' and the group member is 'cpu_atom/cycles/'.
>
> e.g.
> perf stat -e '{cpu-clock,cpu_atom/cycles/}' -a
>
> The challenge is their available cpus are not fully matched.
> For example, 'cpu-clock' is available on CPU0-CPU23, but 'cpu_core/cycles/'
> is available on CPU16-CPU23.
>
> When getting the group id for group member, we must be very careful
> because the cpu for 'cpu-clock' is not equal to the cpu for 'cpu_atom/cycles/'.
> Actually the cpu here is the index of evsel->core.cpus, not the real CPU ID.
> e.g. cpu0 for 'cpu-clock' is CPU0, but cpu0 for 'cpu_atom/cycles/' is CPU16.
>
> Another challenge is for group read. The events in group may be not
> available on all cpus. For example the leader is a software event and
> it's available on CPU0-CPU1, but the group member is a hybrid event and
> it's only available on CPU1. For CPU0, we have only one event, but for CPU1
> we have two events. So we need to change the read size according to
> the real number of events on that cpu.
ugh, this is really bad.. do we really want to support it? ;-)
I guess we need that for metrics..
SNIP
>
> Performance counter stats for 'system wide':
>
> 24,059.14 msec cpu-clock # 23.994 CPUs utilized
> 6,406,677,892 cpu_atom/cycles/ # 266.289 M/sec
>
> 1.002699058 seconds time elapsed
>
> For cpu_atom/cycles/, cpu16-cpu23 are set with valid group fd (cpu-clock's fd
> on that cpu). For counting results, cpu-clock has 24 cpus aggregation and
> cpu_atom/cycles/ has 8 cpus aggregation. That's expected.
>
> But if the event order is changed, e.g. '{cpu_atom/cycles/,cpu-clock}',
> there leaves more works to do.
>
> root@ssp-pwrt-002:~# ./perf stat -e '{cpu_atom/cycles/,cpu-clock}' -a -vvv -- sleep 1
what id you add the other hybrid pmu event? or just cycles?
SNIP
> +static int hybrid_read_size(struct evsel *leader, int cpu, int *nr_members)
> +{
> + struct evsel *pos;
> + int nr = 1, back, new_size = 0, idx;
> +
> + for_each_group_member(pos, leader) {
> + idx = evsel_cpuid_match(leader, pos, cpu);
> + if (idx != -1)
> + nr++;
> + }
> +
> + if (nr != leader->core.nr_members) {
> + back = leader->core.nr_members;
> + leader->core.nr_members = nr;
> + new_size = perf_evsel__read_size(&leader->core);
> + leader->core.nr_members = back;
> + }
> +
> + *nr_members = nr;
> + return new_size;
> +}
> +
> static int evsel__read_group(struct evsel *leader, int cpu, int thread)
> {
> struct perf_stat_evsel *ps = leader->stats;
> u64 read_format = leader->core.attr.read_format;
> int size = perf_evsel__read_size(&leader->core);
> + int new_size, nr_members;
> u64 *data = ps->group_data;
>
> if (!(read_format & PERF_FORMAT_ID))
> return -EINVAL;
I wonder if we do not find some reasonable generic way to process
this, porhaps we should make some early check that this evlist has
hybrid event and the move the implementation in some separated
hybrid-XXX object, so we don't confuse the code
jirka
On Thu, Mar 11, 2021 at 03:07:25PM +0800, Jin Yao wrote:
SNIP
> + config_terms, pmu);
> + if (ret)
> + return ret;
> + }
> +
> + return 0;
> +}
> +
> int parse_events_add_cache(struct list_head *list, int *idx,
> char *type, char *op_result1, char *op_result2,
> struct parse_events_error *err,
> @@ -474,7 +516,8 @@ int parse_events_add_cache(struct list_head *list, int *idx,
> char name[MAX_NAME_LEN], *config_name;
> int cache_type = -1, cache_op = -1, cache_result = -1;
> char *op_result[2] = { op_result1, op_result2 };
> - int i, n;
> + int i, n, ret;
> + bool hybrid;
>
> /*
> * No fallback - if we cannot get a clear cache type
> @@ -534,6 +577,15 @@ int parse_events_add_cache(struct list_head *list, int *idx,
> if (get_config_terms(head_config, &config_terms))
> return -ENOMEM;
> }
> +
> + if (!perf_pmu__hybrid_exist())
> + perf_pmu__scan(NULL);
actualy how about construct like:
perf_pmu_is_hybrid()
return hybrid_add_event_cache(...)
return add_event(...)
with:
perf_pmu_is_hybrid()
{
static bool initialized;
if (!initialized) {
initialized = true;
perf_pmu__scan(NULL)
}
return ...
}
jirka
> +
> + ret = add_hybrid_cache(list, idx, &attr, config_name ? : name,
> + &config_terms, &hybrid);
> + if (hybrid)
> + return ret;
> +
> return add_event(list, idx, &attr, config_name ? : name, &config_terms);
> }
>
> --
> 2.17.1
>
Hi Jiri,
On 3/16/2021 7:03 AM, Jiri Olsa wrote:
> On Thu, Mar 11, 2021 at 03:07:19PM +0800, Jin Yao wrote:
>> On hybrid platform, one event is available on one pmu
>> (such as, available on cpu_core or on cpu_atom).
>>
>> This patch saves the pmu name to the pmu field of struct perf_pmu_alias.
>> Then next we can know the pmu which the event can be available on.
>>
>> Signed-off-by: Jin Yao <[email protected]>
>> ---
>> tools/perf/util/pmu.c | 10 +++++++++-
>> tools/perf/util/pmu.h | 1 +
>> 2 files changed, 10 insertions(+), 1 deletion(-)
>>
>> diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
>> index 54e586bf19a5..45d8db1af8d2 100644
>> --- a/tools/perf/util/pmu.c
>> +++ b/tools/perf/util/pmu.c
>> @@ -283,6 +283,7 @@ void perf_pmu_free_alias(struct perf_pmu_alias *newalias)
>> zfree(&newalias->str);
>> zfree(&newalias->metric_expr);
>> zfree(&newalias->metric_name);
>> + zfree(&newalias->pmu);
>> parse_events_terms__purge(&newalias->terms);
>> free(newalias);
>> }
>> @@ -297,6 +298,10 @@ static bool perf_pmu_merge_alias(struct perf_pmu_alias *newalias,
>>
>> list_for_each_entry(a, alist, list) {
>> if (!strcasecmp(newalias->name, a->name)) {
>> + if (newalias->pmu && a->pmu &&
>> + !strcasecmp(newalias->pmu, a->pmu)) {
>> + continue;
>> + }
>> perf_pmu_update_alias(a, newalias);
>> perf_pmu_free_alias(newalias);
>> return true;
>> @@ -314,7 +319,8 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
>> int num;
>> char newval[256];
>> char *long_desc = NULL, *topic = NULL, *unit = NULL, *perpkg = NULL,
>> - *metric_expr = NULL, *metric_name = NULL, *deprecated = NULL;
>> + *metric_expr = NULL, *metric_name = NULL, *deprecated = NULL,
>> + *pmu = NULL;
>>
>> if (pe) {
>> long_desc = (char *)pe->long_desc;
>> @@ -324,6 +330,7 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
>> metric_expr = (char *)pe->metric_expr;
>> metric_name = (char *)pe->metric_name;
>> deprecated = (char *)pe->deprecated;
>> + pmu = (char *)pe->pmu;
>> }
>>
>> alias = malloc(sizeof(*alias));
>> @@ -389,6 +396,7 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
>> }
>> alias->per_pkg = perpkg && sscanf(perpkg, "%d", &num) == 1 && num == 1;
>> alias->str = strdup(newval);
>> + alias->pmu = pmu ? strdup(pmu) : NULL;
>>
>> if (deprecated)
>> alias->deprecated = true;
>> diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
>> index 8164388478c6..0e724d5b84c6 100644
>> --- a/tools/perf/util/pmu.h
>> +++ b/tools/perf/util/pmu.h
>> @@ -72,6 +72,7 @@ struct perf_pmu_alias {
>> bool deprecated;
>> char *metric_expr;
>> char *metric_name;
>> + char *pmu;
>
> please use pmu_name
>
> thanks,
> jirka
>
OK, I will use pmu_name in next version.
Thanks
Jin Yao
Hi Jiri,
On 3/16/2021 1:37 AM, Jiri Olsa wrote:
> On Mon, Mar 15, 2021 at 10:28:12AM +0800, Jin, Yao wrote:
>> Hi Jiri,
>>
>> On 3/13/2021 3:15 AM, Jiri Olsa wrote:
>>> On Thu, Mar 11, 2021 at 03:07:26PM +0800, Jin Yao wrote:
>>>> On hybrid platform, some hardware events are only available
>>>> on a specific pmu. For example, 'L1-dcache-load-misses' is only
>>>> available on 'cpu_core' pmu. And even for the event which can be
>>>> available on both pmus, the user also may want to just enable
>>>> one event. So now following syntax is supported:
>>>>
>>>> cpu_core/<hardware event>/
>>>> cpu_core/<hardware cache event>/
>>>> cpu_core/<pmu event>/
>>>>
>>>> cpu_atom/<hardware event>/
>>>> cpu_atom/<hardware cache event>/
>>>> cpu_atom/<pmu event>/
>>>>
>>>> It limits the event to be enabled only on a specified pmu.
>>>>
>>>> The patch uses this idea, for example, if we use "cpu_core/LLC-loads/",
>>>> in parse_events_add_pmu(), term->config is "LLC-loads".
>>>
>>> hum, I don't understand how this doest not work even now,
>>> I assume both cpu_core and cpu_atom have sysfs device directory
>>> with events/ directory right?
>>>
>>
>> Yes, we have cpu_core and cpu_atom directories with events.
>>
>> root@ssp-pwrt-002:/sys/devices/cpu_atom/events# ls
>> branch-instructions bus-cycles cache-references instructions
>> mem-stores topdown-bad-spec topdown-fe-bound
>> branch-misses cache-misses cpu-cycles mem-loads
>> ref-cycles topdown-be-bound topdown-retiring
>>
>> root@ssp-pwrt-002:/sys/devices/cpu_core/events# ls
>> branch-instructions cache-misses instructions mem-stores
>> topdown-bad-spec topdown-fe-bound topdown-mem-bound
>> branch-misses cache-references mem-loads ref-cycles
>> topdown-be-bound topdown-fetch-lat topdown-retiring
>> bus-cycles cpu-cycles mem-loads-aux slots
>> topdown-br-mispredict topdown-heavy-ops
>>
>>> and whatever is defined in events we allow in parsing syntax..
>>>
>>> why can't we treat them like 2 separated pmus?
>>>
>>
>> But if without this patch, it reports the error,
>>
>> root@ssp-pwrt-002:~# ./perf stat -e cpu_core/cycles/ -a -vv -- sleep 1
>> event syntax error: 'cpu_core/cycles/'
>> \___ unknown term 'cycles' for pmu 'cpu_core'
>
> yep, because there's special care for 'cycles' unfortunately,
> but you should be able to run 'cpu_core/cpu-cycles/' right?
>
Yes, cpu_core/cpu-cycles/ is OK.
# ./perf stat -e cpu_core/cpu-cycles/ -a -- sleep 1
Performance counter stats for 'system wide':
12,831,980,326 cpu_core/cpu-cycles/
1.003132639 seconds time elapsed
>>
>> valid terms: event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
>>
>> Initial error:
>> event syntax error: 'cpu_core/cycles/'
>> \___ unknown term 'cycles' for pmu 'cpu_core'
>>
>> valid terms: event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
>> Run 'perf list' for a list of valid events
>>
>> The 'cycles' is treated as a unknown term, then it errors out.
>
> yep, because it's not in events.. we could add special rule to
> treat cycles as cpu-cycles inside pmu definition ;-)
>
> jirka
>
But not only the cycles, the branches has error too.
# ./perf stat -e cpu_core/branches/ -a -- sleep 1
event syntax error: 'cpu_core/branches/'
\___ unknown term 'branches' for pmu 'cpu_core'
valid terms:
event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
Initial error:
event syntax error: 'cpu_core/branches/'
\___ unknown term 'branches' for pmu 'cpu_core'
Of course, branch-instructions runs OK.
# ./perf stat -e cpu_core/branch-instructions/ -a -- sleep 1
Performance counter stats for 'system wide':
136,655,302 cpu_core/branch-instructions/
1.003171561 seconds time elapsed
So we need special rules for both cycles and branches.
The worse thing is, we also need to process the hardware cache events.
# ./perf stat -e cpu_core/LLC-loads/
event syntax error: 'cpu_core/LLC-loads/'
\___ unknown term 'LLC-loads' for pmu 'cpu_core'
valid terms:
event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
Initial error:
event syntax error: 'cpu_core/LLC-loads/'
\___ unknown term 'LLC-loads' for pmu 'cpu_core'
If we use special rules for establishing all event mapping, that looks too much. :(
Thanks
Jin Yao
Hi Jiri,
On 3/16/2021 7:03 AM, Jiri Olsa wrote:
> On Thu, Mar 11, 2021 at 03:07:31PM +0800, Jin Yao wrote:
>
> SNIP
>
>> goto try_again;
>> }
>> +
>> + if (errno == EINVAL && perf_pmu__hybrid_exist())
>> + evlist__warn_hybrid_group(evlist);
>> rc = -errno;
>> evsel__open_strerror(pos, &opts->target, errno, msg, sizeof(msg));
>> ui__error("%s\n", msg);
>> diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
>> index 7a732508b2b4..6f780a039db0 100644
>> --- a/tools/perf/builtin-stat.c
>> +++ b/tools/perf/builtin-stat.c
>> @@ -239,6 +239,9 @@ static void evlist__check_cpu_maps(struct evlist *evlist)
>> struct evsel *evsel, *pos, *leader;
>> char buf[1024];
>>
>> + if (evlist__hybrid_exist(evlist))
>> + return;
>
> this should be in separate patch and explained
>
Now I have another idea. If a group consists of atom events and core events, we still follow current
disabling group solution?
I mean removing following code:
if (evlist__hybrid_exist(evlist))
return;
evlist__check_cpu_maps then continues running and disabling the group. But also report with a
warning that says "WARNING: Group has events from different hybrid PMUs".
Do you like this way?
>> +
>> evlist__for_each_entry(evlist, evsel) {
>> leader = evsel->leader;
>>
>> @@ -726,6 +729,10 @@ enum counter_recovery {
>> static enum counter_recovery stat_handle_error(struct evsel *counter)
>> {
>> char msg[BUFSIZ];
>> +
>> + if (perf_pmu__hybrid_exist() && errno == EINVAL)
>> + evlist__warn_hybrid_group(evsel_list);
>> +
>> /*
>> * PPC returns ENXIO for HW counters until 2.6.37
>> * (behavior changed with commit b0a873e).
>> diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
>> index f139151b9433..5ec891418cdd 100644
>> --- a/tools/perf/util/evlist.c
>> +++ b/tools/perf/util/evlist.c
>> @@ -2224,3 +2224,47 @@ void evlist__invalidate_all_cpus(struct evlist *evlist)
>> perf_cpu_map__put(evlist->core.all_cpus);
>> evlist->core.all_cpus = perf_cpu_map__empty_new(1);
>> }
>> +
>> +static bool group_hybrid_conflict(struct evsel *leader)
>> +{
>> + struct evsel *pos, *prev = NULL;
>> +
>> + for_each_group_evsel(pos, leader) {
>> + if (!pos->pmu_name || !perf_pmu__is_hybrid(pos->pmu_name))
>> + continue;
>> +
>> + if (prev && strcmp(prev->pmu_name, pos->pmu_name))
>> + return true;
>> +
>> + prev = pos;
>> + }
>> +
>> + return false;
>> +}
>> +
>> +void evlist__warn_hybrid_group(struct evlist *evlist)
>> +{
>> + struct evsel *evsel;
>> +
>> + evlist__for_each_entry(evlist, evsel) {
>> + if (evsel__is_group_leader(evsel) &&
>> + evsel->core.nr_members > 1 &&
>
> hm, could we just iterate all the members and make sure the first found
> hybrid event's pmu matches the pmu of the rest hybrid events in the list?
>
'{cpu_core/event1/,cpu_core/event2/}','{cpu_atom/event3/,cpu_atom/event4/}'
Two or more groups need to be supported. We get the first hybrid event's pmu (cpu_core in this
example) but it doesn't match the cpu_atom/event3/ and cpu_atom/event4/. But actually this case
should be supported, right?
>> + group_hybrid_conflict(evsel)) {
>> + WARN_ONCE(1, "WARNING: Group has events from "
>> + "different hybrid PMUs\n");
>> + return;
>> + }
>> + }
>> +}
>> +
>> +bool evlist__hybrid_exist(struct evlist *evlist)
>
> evlist__has_hybrid seems better
>
Yes, agree.
Thanks
Jin Yao
>
> jirka
>
On Thu, Mar 11, 2021 at 03:07:31PM +0800, Jin Yao wrote:
SNIP
> goto try_again;
> }
> +
> + if (errno == EINVAL && perf_pmu__hybrid_exist())
> + evlist__warn_hybrid_group(evlist);
> rc = -errno;
> evsel__open_strerror(pos, &opts->target, errno, msg, sizeof(msg));
> ui__error("%s\n", msg);
> diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
> index 7a732508b2b4..6f780a039db0 100644
> --- a/tools/perf/builtin-stat.c
> +++ b/tools/perf/builtin-stat.c
> @@ -239,6 +239,9 @@ static void evlist__check_cpu_maps(struct evlist *evlist)
> struct evsel *evsel, *pos, *leader;
> char buf[1024];
>
> + if (evlist__hybrid_exist(evlist))
> + return;
this should be in separate patch and explained
> +
> evlist__for_each_entry(evlist, evsel) {
> leader = evsel->leader;
>
> @@ -726,6 +729,10 @@ enum counter_recovery {
> static enum counter_recovery stat_handle_error(struct evsel *counter)
> {
> char msg[BUFSIZ];
> +
> + if (perf_pmu__hybrid_exist() && errno == EINVAL)
> + evlist__warn_hybrid_group(evsel_list);
> +
> /*
> * PPC returns ENXIO for HW counters until 2.6.37
> * (behavior changed with commit b0a873e).
> diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
> index f139151b9433..5ec891418cdd 100644
> --- a/tools/perf/util/evlist.c
> +++ b/tools/perf/util/evlist.c
> @@ -2224,3 +2224,47 @@ void evlist__invalidate_all_cpus(struct evlist *evlist)
> perf_cpu_map__put(evlist->core.all_cpus);
> evlist->core.all_cpus = perf_cpu_map__empty_new(1);
> }
> +
> +static bool group_hybrid_conflict(struct evsel *leader)
> +{
> + struct evsel *pos, *prev = NULL;
> +
> + for_each_group_evsel(pos, leader) {
> + if (!pos->pmu_name || !perf_pmu__is_hybrid(pos->pmu_name))
> + continue;
> +
> + if (prev && strcmp(prev->pmu_name, pos->pmu_name))
> + return true;
> +
> + prev = pos;
> + }
> +
> + return false;
> +}
> +
> +void evlist__warn_hybrid_group(struct evlist *evlist)
> +{
> + struct evsel *evsel;
> +
> + evlist__for_each_entry(evlist, evsel) {
> + if (evsel__is_group_leader(evsel) &&
> + evsel->core.nr_members > 1 &&
hm, could we just iterate all the members and make sure the first found
hybrid event's pmu matches the pmu of the rest hybrid events in the list?
> + group_hybrid_conflict(evsel)) {
> + WARN_ONCE(1, "WARNING: Group has events from "
> + "different hybrid PMUs\n");
> + return;
> + }
> + }
> +}
> +
> +bool evlist__hybrid_exist(struct evlist *evlist)
evlist__has_hybrid seems better
jirka
On Thu, Mar 11, 2021 at 03:07:19PM +0800, Jin Yao wrote:
> On hybrid platform, one event is available on one pmu
> (such as, available on cpu_core or on cpu_atom).
>
> This patch saves the pmu name to the pmu field of struct perf_pmu_alias.
> Then next we can know the pmu which the event can be available on.
>
> Signed-off-by: Jin Yao <[email protected]>
> ---
> tools/perf/util/pmu.c | 10 +++++++++-
> tools/perf/util/pmu.h | 1 +
> 2 files changed, 10 insertions(+), 1 deletion(-)
>
> diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
> index 54e586bf19a5..45d8db1af8d2 100644
> --- a/tools/perf/util/pmu.c
> +++ b/tools/perf/util/pmu.c
> @@ -283,6 +283,7 @@ void perf_pmu_free_alias(struct perf_pmu_alias *newalias)
> zfree(&newalias->str);
> zfree(&newalias->metric_expr);
> zfree(&newalias->metric_name);
> + zfree(&newalias->pmu);
> parse_events_terms__purge(&newalias->terms);
> free(newalias);
> }
> @@ -297,6 +298,10 @@ static bool perf_pmu_merge_alias(struct perf_pmu_alias *newalias,
>
> list_for_each_entry(a, alist, list) {
> if (!strcasecmp(newalias->name, a->name)) {
> + if (newalias->pmu && a->pmu &&
> + !strcasecmp(newalias->pmu, a->pmu)) {
> + continue;
> + }
> perf_pmu_update_alias(a, newalias);
> perf_pmu_free_alias(newalias);
> return true;
> @@ -314,7 +319,8 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
> int num;
> char newval[256];
> char *long_desc = NULL, *topic = NULL, *unit = NULL, *perpkg = NULL,
> - *metric_expr = NULL, *metric_name = NULL, *deprecated = NULL;
> + *metric_expr = NULL, *metric_name = NULL, *deprecated = NULL,
> + *pmu = NULL;
>
> if (pe) {
> long_desc = (char *)pe->long_desc;
> @@ -324,6 +330,7 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
> metric_expr = (char *)pe->metric_expr;
> metric_name = (char *)pe->metric_name;
> deprecated = (char *)pe->deprecated;
> + pmu = (char *)pe->pmu;
> }
>
> alias = malloc(sizeof(*alias));
> @@ -389,6 +396,7 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
> }
> alias->per_pkg = perpkg && sscanf(perpkg, "%d", &num) == 1 && num == 1;
> alias->str = strdup(newval);
> + alias->pmu = pmu ? strdup(pmu) : NULL;
>
> if (deprecated)
> alias->deprecated = true;
> diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
> index 8164388478c6..0e724d5b84c6 100644
> --- a/tools/perf/util/pmu.h
> +++ b/tools/perf/util/pmu.h
> @@ -72,6 +72,7 @@ struct perf_pmu_alias {
> bool deprecated;
> char *metric_expr;
> char *metric_name;
> + char *pmu;
please use pmu_name
thanks,
jirka
Hi Jiri,
On 3/16/2021 7:05 AM, Jiri Olsa wrote:
> On Thu, Mar 11, 2021 at 03:07:24PM +0800, Jin Yao wrote:
>> For hardware events, they have pre-defined configs. The kernel
>> needs to know where the event comes from (e.g. from cpu_core pmu
>> or from cpu_atom pmu). But the perf type 'PERF_TYPE_HARDWARE'
>> can't carry pmu information.
>>
>> So the kernel introduces a new type 'PERF_TYPE_HARDWARE_PMU'.
>> The new attr.config layout for PERF_TYPE_HARDWARE_PMU is:
>>
>> 0xDD000000AA
>> AA: original hardware event ID
>> DD: PMU type ID
>>
>> PMU type ID is retrieved from sysfs. For example,
>>
>> cat /sys/devices/cpu_atom/type
>> 10
>>
>> cat /sys/devices/cpu_core/type
>> 4
>>
>> When enabling a hybrid hardware event without specified pmu, such as,
>> 'perf stat -e cycles -a', two events are created automatically. One
>> is for atom, the other is for core.
>
> ok I think I understand the need for this (and the following) patch
> the perf_hw_id counters could be global, so when you specify only
> event like:
>
> -e cycles
>
> you want all the cycles, which on hybrid system means cycles from
> more than one pmus
>
Yes, on hybrid system it means the cycles from two pmus. One cycle is from cpu_core pmu, another
cycles is from cpu_atom pmu.
> SNIP
>
>> @@ -1416,6 +1475,8 @@ int parse_events_add_numeric(struct parse_events_state *parse_state,
>> {
>> struct perf_event_attr attr;
>> LIST_HEAD(config_terms);
>> + bool hybrid;
>> + int ret;
>>
>> memset(&attr, 0, sizeof(attr));
>> attr.type = type;
>> @@ -1430,6 +1491,18 @@ int parse_events_add_numeric(struct parse_events_state *parse_state,
>> return -ENOMEM;
>> }
>>
>> + /*
>> + * Skip the software dummy event.
>> + */
>> + if (type != PERF_TYPE_SOFTWARE) {
>> + if (!perf_pmu__hybrid_exist())
>> + perf_pmu__scan(NULL);
>
> this could be checked in the following add_hybrid_numeric call
>
Yes, that should be OK. I will move the check in the next version.
>> +
>> + ret = add_hybrid_numeric(parse_state, list, &attr, &hybrid);
>> + if (hybrid)
>> + return ret;
>> + }
>
> could we add this to separate object.. hybrid.c or maybe parse-events-hybrid.c,
>
> there's already global __add_event wrapper - parse_events__add_event
>
>
> jirka
>
Use a new parse-events-hybrid.c, hmm, well that's OK.
Thanks
Jin Yao
>> +
>> return add_event(list, &parse_state->idx, &attr,
>> get_config_name(head_config), &config_terms);
>> }
>> --
>> 2.17.1
>>
>
Hi Jiri,
On 3/16/2021 7:05 AM, Jiri Olsa wrote:
> On Thu, Mar 11, 2021 at 03:07:25PM +0800, Jin Yao wrote:
>
> SNIP
>
>> + config_terms, pmu);
>> + if (ret)
>> + return ret;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> int parse_events_add_cache(struct list_head *list, int *idx,
>> char *type, char *op_result1, char *op_result2,
>> struct parse_events_error *err,
>> @@ -474,7 +516,8 @@ int parse_events_add_cache(struct list_head *list, int *idx,
>> char name[MAX_NAME_LEN], *config_name;
>> int cache_type = -1, cache_op = -1, cache_result = -1;
>> char *op_result[2] = { op_result1, op_result2 };
>> - int i, n;
>> + int i, n, ret;
>> + bool hybrid;
>>
>> /*
>> * No fallback - if we cannot get a clear cache type
>> @@ -534,6 +577,15 @@ int parse_events_add_cache(struct list_head *list, int *idx,
>> if (get_config_terms(head_config, &config_terms))
>> return -ENOMEM;
>> }
>> +
>> + if (!perf_pmu__hybrid_exist())
>> + perf_pmu__scan(NULL);
>
> actualy how about construct like:
>
> perf_pmu_is_hybrid()
> return hybrid_add_event_cache(...)
>
> return add_event(...)
>
>
> with:
> perf_pmu_is_hybrid()
> {
> static bool initialized;
>
> if (!initialized) {
> initialized = true;
> perf_pmu__scan(NULL)
> }
>
> return ...
> }
>
> jirka
>
Thanks, that's a good solution. I will do that in v3.
Thanks
Jin Yao
>> +
>> + ret = add_hybrid_cache(list, idx, &attr, config_name ? : name,
>> + &config_terms, &hybrid);
>> + if (hybrid)
>> + return ret;
>> +
>> return add_event(list, idx, &attr, config_name ? : name, &config_terms);
>> }
>>
>> --
>> 2.17.1
>>
>
Hi Jiri,
On 3/16/2021 7:04 AM, Jiri Olsa wrote:
> On Thu, Mar 11, 2021 at 03:07:32PM +0800, Jin Yao wrote:
>> A group mixed with hybrid event and global event is allowed. For example,
>> group leader is 'cpu-clock' and the group member is 'cpu_atom/cycles/'.
>>
>> e.g.
>> perf stat -e '{cpu-clock,cpu_atom/cycles/}' -a
>>
>> The challenge is their available cpus are not fully matched.
>> For example, 'cpu-clock' is available on CPU0-CPU23, but 'cpu_core/cycles/'
>> is available on CPU16-CPU23.
>>
>> When getting the group id for group member, we must be very careful
>> because the cpu for 'cpu-clock' is not equal to the cpu for 'cpu_atom/cycles/'.
>> Actually the cpu here is the index of evsel->core.cpus, not the real CPU ID.
>> e.g. cpu0 for 'cpu-clock' is CPU0, but cpu0 for 'cpu_atom/cycles/' is CPU16.
>>
>> Another challenge is for group read. The events in group may be not
>> available on all cpus. For example the leader is a software event and
>> it's available on CPU0-CPU1, but the group member is a hybrid event and
>> it's only available on CPU1. For CPU0, we have only one event, but for CPU1
>> we have two events. So we need to change the read size according to
>> the real number of events on that cpu.
>
> ugh, this is really bad.. do we really want to support it? ;-)
> I guess we need that for metrics..
>
Yes, it's a bit of pain but the user case makes sense. Some metrics need the event group which
consists of global event + hybrid event.
For example, CPU_Utilization = 'cpu_clk_unhalted.ref_tsc' / 'msr/tsc/'.
'msr/tsc/' is a global event. It's valid on all CPUs.
But 'cpu_clk_unhalted.ref' is hybrid event.
'cpu_core/cpu_clk_unhalted.ref/' is valid on core CPUs
'cpu_atom/cpu_clk_unhalted.ref/' is valid on atom CPUs.
So we have to support this usage. :)
> SNIP
>
>>
>> Performance counter stats for 'system wide':
>>
>> 24,059.14 msec cpu-clock # 23.994 CPUs utilized
>> 6,406,677,892 cpu_atom/cycles/ # 266.289 M/sec
>>
>> 1.002699058 seconds time elapsed
>>
>> For cpu_atom/cycles/, cpu16-cpu23 are set with valid group fd (cpu-clock's fd
>> on that cpu). For counting results, cpu-clock has 24 cpus aggregation and
>> cpu_atom/cycles/ has 8 cpus aggregation. That's expected.
>>
>> But if the event order is changed, e.g. '{cpu_atom/cycles/,cpu-clock}',
>> there leaves more works to do.
>>
>> root@ssp-pwrt-002:~# ./perf stat -e '{cpu_atom/cycles/,cpu-clock}' -a -vvv -- sleep 1
>
> what id you add the other hybrid pmu event? or just cycles?
>
Do you mean the config for cpu_atom/cycles/? Let's see the log.
root@ssp-pwrt-002:~# perf stat -e '{cpu_atom/cycles/,cpu-clock}' -a -vvv -- sleep 1
Control descriptor is not initialized
------------------------------------------------------------
perf_event_attr:
type 6
size 120
config 0xa00000000
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING|ID|GROUP
disabled 1
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 16 group_fd -1 flags 0x8 = 3
sys_perf_event_open: pid -1 cpu 17 group_fd -1 flags 0x8 = 4
sys_perf_event_open: pid -1 cpu 18 group_fd -1 flags 0x8 = 5
sys_perf_event_open: pid -1 cpu 19 group_fd -1 flags 0x8 = 7
sys_perf_event_open: pid -1 cpu 20 group_fd -1 flags 0x8 = 8
sys_perf_event_open: pid -1 cpu 21 group_fd -1 flags 0x8 = 9
sys_perf_event_open: pid -1 cpu 22 group_fd -1 flags 0x8 = 10
sys_perf_event_open: pid -1 cpu 23 group_fd -1 flags 0x8 = 11
------------------------------------------------------------
perf_event_attr:
type 1
size 120
sample_type IDENTIFIER
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING|ID|GROUP
inherit 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 12
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 13
sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 14
sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 15
sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 16
sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 17
sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 18
sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 19
sys_perf_event_open: pid -1 cpu 8 group_fd -1 flags 0x8 = 20
sys_perf_event_open: pid -1 cpu 9 group_fd -1 flags 0x8 = 21
sys_perf_event_open: pid -1 cpu 10 group_fd -1 flags 0x8 = 22
sys_perf_event_open: pid -1 cpu 11 group_fd -1 flags 0x8 = 23
sys_perf_event_open: pid -1 cpu 12 group_fd -1 flags 0x8 = 24
sys_perf_event_open: pid -1 cpu 13 group_fd -1 flags 0x8 = 25
sys_perf_event_open: pid -1 cpu 14 group_fd -1 flags 0x8 = 26
sys_perf_event_open: pid -1 cpu 15 group_fd -1 flags 0x8 = 27
sys_perf_event_open: pid -1 cpu 16 group_fd 3 flags 0x8 = 28
sys_perf_event_open: pid -1 cpu 17 group_fd 4 flags 0x8 = 29
sys_perf_event_open: pid -1 cpu 18 group_fd 5 flags 0x8 = 30
sys_perf_event_open: pid -1 cpu 19 group_fd 7 flags 0x8 = 31
sys_perf_event_open: pid -1 cpu 20 group_fd 8 flags 0x8 = 32
sys_perf_event_open: pid -1 cpu 21 group_fd 9 flags 0x8 = 33
sys_perf_event_open: pid -1 cpu 22 group_fd 10 flags 0x8 = 34
sys_perf_event_open: pid -1 cpu 23 group_fd 11 flags 0x8 = 35
cycles: 0: 800791792 1002389889 1002389889
cycles: 1: 800788198 1002383611 1002383611
cycles: 2: 800783491 1002377507 1002377507
cycles: 3: 800777752 1002371035 1002371035
cycles: 4: 800771559 1002363669 1002363669
cycles: 5: 800766391 1002356944 1002356944
cycles: 6: 800761593 1002350144 1002350144
cycles: 7: 800756258 1002343203 1002343203
WARNING: for cpu-clock, some CPU counts not read
cpu-clock: 0: 0 0 0
cpu-clock: 1: 0 0 0
cpu-clock: 2: 0 0 0
cpu-clock: 3: 0 0 0
cpu-clock: 4: 0 0 0
cpu-clock: 5: 0 0 0
cpu-clock: 6: 0 0 0
cpu-clock: 7: 0 0 0
cpu-clock: 8: 0 0 0
cpu-clock: 9: 0 0 0
cpu-clock: 10: 0 0 0
cpu-clock: 11: 0 0 0
cpu-clock: 12: 0 0 0
cpu-clock: 13: 0 0 0
cpu-clock: 14: 0 0 0
cpu-clock: 15: 0 0 0
cpu-clock: 16: 1002390566 1002389889 1002389889
cpu-clock: 17: 1002383263 1002383611 1002383611
cpu-clock: 18: 1002377257 1002377507 1002377507
cpu-clock: 19: 1002370895 1002371035 1002371035
cpu-clock: 20: 1002363611 1002363669 1002363669
cpu-clock: 21: 1002356623 1002356944 1002356944
cpu-clock: 22: 1002349562 1002350144 1002350144
cpu-clock: 23: 1002343089 1002343203 1002343203
cycles: 6406197034 8018936002 8018936002
cpu-clock: 8018934866 8018936002 8018936002
Performance counter stats for 'system wide':
6,406,197,034 cpu_atom/cycles/ # 798.884 M/sec
8,018.93 msec cpu-clock # 7.999 CPUs utilized
1.002475994 seconds time elapsed
'0xa00000000' is the config for cpu_atom/cycles/ and we can see this event is only enabled on CPU16-23.
'cpu-clock' is a global event and it's enable don CPU0-23 (but only have valid group fd on CPU16-23).
>
> SNIP
>
>> +static int hybrid_read_size(struct evsel *leader, int cpu, int *nr_members)
>> +{
>> + struct evsel *pos;
>> + int nr = 1, back, new_size = 0, idx;
>> +
>> + for_each_group_member(pos, leader) {
>> + idx = evsel_cpuid_match(leader, pos, cpu);
>> + if (idx != -1)
>> + nr++;
>> + }
>> +
>> + if (nr != leader->core.nr_members) {
>> + back = leader->core.nr_members;
>> + leader->core.nr_members = nr;
>> + new_size = perf_evsel__read_size(&leader->core);
>> + leader->core.nr_members = back;
>> + }
>> +
>> + *nr_members = nr;
>> + return new_size;
>> +}
>> +
>> static int evsel__read_group(struct evsel *leader, int cpu, int thread)
>> {
>> struct perf_stat_evsel *ps = leader->stats;
>> u64 read_format = leader->core.attr.read_format;
>> int size = perf_evsel__read_size(&leader->core);
>> + int new_size, nr_members;
>> u64 *data = ps->group_data;
>>
>> if (!(read_format & PERF_FORMAT_ID))
>> return -EINVAL;
>
> I wonder if we do not find some reasonable generic way to process
> this, porhaps we should make some early check that this evlist has
> hybrid event and the move the implementation in some separated
> hybrid-XXX object, so we don't confuse the code
>
> jirka
>
Agree. The code looks a bit tricky and hard to understand for non-hybrid platform.
Maybe it's a good idea to move this code to hybrid-xxx files.
Thanks
Jin Yao
On Tue, Mar 16, 2021 at 09:49:42AM +0800, Jin, Yao wrote:
SNIP
>
> Performance counter stats for 'system wide':
>
> 136,655,302 cpu_core/branch-instructions/
>
> 1.003171561 seconds time elapsed
>
> So we need special rules for both cycles and branches.
>
> The worse thing is, we also need to process the hardware cache events.
>
> # ./perf stat -e cpu_core/LLC-loads/
> event syntax error: 'cpu_core/LLC-loads/'
> \___ unknown term 'LLC-loads' for pmu 'cpu_core'
>
> valid terms: event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
>
> Initial error:
> event syntax error: 'cpu_core/LLC-loads/'
> \___ unknown term 'LLC-loads' for pmu 'cpu_core'
>
> If we use special rules for establishing all event mapping, that looks too much. :(
hmmm but wait, currently we do not support events like this:
'cpu/cycles/'
'cpu/branches/'
the pmu style accepts only 'events' or 'format' terms within //
we made hw events like 'cycles','instructions','branches' special
to be used without the pmu
so why do we need to support cpu_code/cycles/ ?
jirka
On 3/16/2021 10:04 PM, Jiri Olsa wrote:
> On Tue, Mar 16, 2021 at 09:49:42AM +0800, Jin, Yao wrote:
>
> SNIP
>
>>
>> Performance counter stats for 'system wide':
>>
>> 136,655,302 cpu_core/branch-instructions/
>>
>> 1.003171561 seconds time elapsed
>>
>> So we need special rules for both cycles and branches.
>>
>> The worse thing is, we also need to process the hardware cache events.
>>
>> # ./perf stat -e cpu_core/LLC-loads/
>> event syntax error: 'cpu_core/LLC-loads/'
>> \___ unknown term 'LLC-loads' for pmu 'cpu_core'
>>
>> valid terms: event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
>>
>> Initial error:
>> event syntax error: 'cpu_core/LLC-loads/'
>> \___ unknown term 'LLC-loads' for pmu 'cpu_core'
>>
>> If we use special rules for establishing all event mapping, that looks too much. :(
>
> hmmm but wait, currently we do not support events like this:
>
> 'cpu/cycles/'
> 'cpu/branches/'
>
> the pmu style accepts only 'events' or 'format' terms within //
>
> we made hw events like 'cycles','instructions','branches' special
> to be used without the pmu
>
> so why do we need to support cpu_code/cycles/ ?
>
> jirka
>
Actually we have to support pmu style event for hybrid platform.
User may want to enable the events from specified pmus and also with flexible grouping.
For example,
perf stat -e '{cpu_core/cycles/,cpu_core/instructions/}' -e '{cpu_atom/cycles/,cpu_atom/instructions/}'
This usage is common and reasonable. So I think we may need to support pmu style events.
Thanks
Jin Yao
On Wed, Mar 17, 2021 at 10:12:03AM +0800, Jin, Yao wrote:
>
>
> On 3/16/2021 10:04 PM, Jiri Olsa wrote:
> > On Tue, Mar 16, 2021 at 09:49:42AM +0800, Jin, Yao wrote:
> >
> > SNIP
> >
> > >
> > > Performance counter stats for 'system wide':
> > >
> > > 136,655,302 cpu_core/branch-instructions/
> > >
> > > 1.003171561 seconds time elapsed
> > >
> > > So we need special rules for both cycles and branches.
> > >
> > > The worse thing is, we also need to process the hardware cache events.
> > >
> > > # ./perf stat -e cpu_core/LLC-loads/
> > > event syntax error: 'cpu_core/LLC-loads/'
> > > \___ unknown term 'LLC-loads' for pmu 'cpu_core'
> > >
> > > valid terms: event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
> > >
> > > Initial error:
> > > event syntax error: 'cpu_core/LLC-loads/'
> > > \___ unknown term 'LLC-loads' for pmu 'cpu_core'
> > >
> > > If we use special rules for establishing all event mapping, that looks too much. :(
> >
> > hmmm but wait, currently we do not support events like this:
> >
> > 'cpu/cycles/'
> > 'cpu/branches/'
> >
> > the pmu style accepts only 'events' or 'format' terms within //
> >
> > we made hw events like 'cycles','instructions','branches' special
> > to be used without the pmu
> >
> > so why do we need to support cpu_code/cycles/ ?
> >
> > jirka
> >
>
> Actually we have to support pmu style event for hybrid platform.
>
> User may want to enable the events from specified pmus and also with flexible grouping.
>
> For example,
>
> perf stat -e '{cpu_core/cycles/,cpu_core/instructions/}' -e '{cpu_atom/cycles/,cpu_atom/instructions/}'
>
> This usage is common and reasonable. So I think we may need to support pmu style events.
sure, but we don't support 'cpu/cycles/' but we support 'cpu/cpu-cycles/'
why do you insist on supporting cpu_core/cycles/ ?
jirka
Hi Jiri,
On 3/17/2021 6:06 PM, Jiri Olsa wrote:
> On Wed, Mar 17, 2021 at 10:12:03AM +0800, Jin, Yao wrote:
>>
>>
>> On 3/16/2021 10:04 PM, Jiri Olsa wrote:
>>> On Tue, Mar 16, 2021 at 09:49:42AM +0800, Jin, Yao wrote:
>>>
>>> SNIP
>>>
>>>>
>>>> Performance counter stats for 'system wide':
>>>>
>>>> 136,655,302 cpu_core/branch-instructions/
>>>>
>>>> 1.003171561 seconds time elapsed
>>>>
>>>> So we need special rules for both cycles and branches.
>>>>
>>>> The worse thing is, we also need to process the hardware cache events.
>>>>
>>>> # ./perf stat -e cpu_core/LLC-loads/
>>>> event syntax error: 'cpu_core/LLC-loads/'
>>>> \___ unknown term 'LLC-loads' for pmu 'cpu_core'
>>>>
>>>> valid terms: event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
>>>>
>>>> Initial error:
>>>> event syntax error: 'cpu_core/LLC-loads/'
>>>> \___ unknown term 'LLC-loads' for pmu 'cpu_core'
>>>>
>>>> If we use special rules for establishing all event mapping, that looks too much. :(
>>>
>>> hmmm but wait, currently we do not support events like this:
>>>
>>> 'cpu/cycles/'
>>> 'cpu/branches/'
>>>
>>> the pmu style accepts only 'events' or 'format' terms within //
>>>
>>> we made hw events like 'cycles','instructions','branches' special
>>> to be used without the pmu
>>>
>>> so why do we need to support cpu_code/cycles/ ?
>>>
>>> jirka
>>>
>>
>> Actually we have to support pmu style event for hybrid platform.
>>
>> User may want to enable the events from specified pmus and also with flexible grouping.
>>
>> For example,
>>
>> perf stat -e '{cpu_core/cycles/,cpu_core/instructions/}' -e '{cpu_atom/cycles/,cpu_atom/instructions/}'
>>
>> This usage is common and reasonable. So I think we may need to support pmu style events.
>
> sure, but we don't support 'cpu/cycles/' but we support 'cpu/cpu-cycles/'
> why do you insist on supporting cpu_core/cycles/ ?
>
> jirka
>
I'm OK to only support 'cpu_core/cpu-cycles/' or 'cpu_atom/cpu-cycles/'. But what would we do for
cache event?
'perf stat -e LLC-loads' is OK, but 'perf stat -e cpu/LLC-loads/' is not supported currently.
For hybrid platform, user may only want to enable the LLC-loads on core CPUs or on atom CPUs. That's
reasonable. While if we don't support the pmu style event, how to satisfy this requirement?
If we can support the pmu style event, we can also use the same way for cpu_core/cycles/. At least
it's not a bad thing, right? :)
Thanks
Jin Yao
Em Wed, Mar 17, 2021 at 08:17:52PM +0800, Jin, Yao escreveu:
> Hi Jiri,
>
> On 3/17/2021 6:06 PM, Jiri Olsa wrote:
> > On Wed, Mar 17, 2021 at 10:12:03AM +0800, Jin, Yao wrote:
> > >
> > >
> > > On 3/16/2021 10:04 PM, Jiri Olsa wrote:
> > > > On Tue, Mar 16, 2021 at 09:49:42AM +0800, Jin, Yao wrote:
> > > >
> > > > SNIP
> > > >
> > > > >
> > > > > Performance counter stats for 'system wide':
> > > > >
> > > > > 136,655,302 cpu_core/branch-instructions/
> > > > >
> > > > > 1.003171561 seconds time elapsed
> > > > >
> > > > > So we need special rules for both cycles and branches.
> > > > >
> > > > > The worse thing is, we also need to process the hardware cache events.
> > > > >
> > > > > # ./perf stat -e cpu_core/LLC-loads/
> > > > > event syntax error: 'cpu_core/LLC-loads/'
> > > > > \___ unknown term 'LLC-loads' for pmu 'cpu_core'
> > > > >
> > > > > valid terms: event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
> > > > >
> > > > > Initial error:
> > > > > event syntax error: 'cpu_core/LLC-loads/'
> > > > > \___ unknown term 'LLC-loads' for pmu 'cpu_core'
> > > > >
> > > > > If we use special rules for establishing all event mapping, that looks too much. :(
> > > >
> > > > hmmm but wait, currently we do not support events like this:
> > > >
> > > > 'cpu/cycles/'
> > > > 'cpu/branches/'
> > > >
> > > > the pmu style accepts only 'events' or 'format' terms within //
> > > >
> > > > we made hw events like 'cycles','instructions','branches' special
> > > > to be used without the pmu
> > > >
> > > > so why do we need to support cpu_code/cycles/ ?
> > > Actually we have to support pmu style event for hybrid platform.
> > > User may want to enable the events from specified pmus and also with flexible grouping.
> > > For example,
> > > perf stat -e '{cpu_core/cycles/,cpu_core/instructions/}' -e '{cpu_atom/cycles/,cpu_atom/instructions/}'
> > > This usage is common and reasonable. So I think we may need to support pmu style events.
> > sure, but we don't support 'cpu/cycles/' but we support 'cpu/cpu-cycles/'
> > why do you insist on supporting cpu_core/cycles/ ?
>
> I'm OK to only support 'cpu_core/cpu-cycles/' or 'cpu_atom/cpu-cycles/'. But
> what would we do for cache event?
>
> 'perf stat -e LLC-loads' is OK, but 'perf stat -e cpu/LLC-loads/' is not supported currently.
>
> For hybrid platform, user may only want to enable the LLC-loads on core CPUs
> or on atom CPUs. That's reasonable. While if we don't support the pmu style
> event, how to satisfy this requirement?
>
> If we can support the pmu style event, we can also use the same way for
> cpu_core/cycles/. At least it's not a bad thing, right? :)
While we're discussing, do we really want to use the "core" and "atom"
terms here? I thought cpu/cycles/ would be ok for the main (Big) CPU and
that we should come up with some short name for the "litle" CPUs.
Won't we have the same situation with ARM where we want to know the
number of cycles spent on a BIG core and also on a little one?
Perhaps 'cycles' should mean all cycles, and then we use 'big/cycles/' and
'little/cycles/'?
- Arnaldo
On Wed, Mar 17, 2021 at 08:17:52PM +0800, Jin, Yao wrote:
SNIP
> > >
> > > For example,
> > >
> > > perf stat -e '{cpu_core/cycles/,cpu_core/instructions/}' -e '{cpu_atom/cycles/,cpu_atom/instructions/}'
> > >
> > > This usage is common and reasonable. So I think we may need to support pmu style events.
> >
> > sure, but we don't support 'cpu/cycles/' but we support 'cpu/cpu-cycles/'
> > why do you insist on supporting cpu_core/cycles/ ?
> >
> > jirka
> >
>
> I'm OK to only support 'cpu_core/cpu-cycles/' or 'cpu_atom/cpu-cycles/'. But
> what would we do for cache event?
>
> 'perf stat -e LLC-loads' is OK, but 'perf stat -e cpu/LLC-loads/' is not supported currently.
ugh, I keep forgetting those ;-)
>
> For hybrid platform, user may only want to enable the LLC-loads on core CPUs
> or on atom CPUs. That's reasonable. While if we don't support the pmu style
> event, how to satisfy this requirement?
>
> If we can support the pmu style event, we can also use the same way for
> cpu_core/cycles/. At least it's not a bad thing, right? :)
right, it's probably best to use the pmu/LLC-.../ for this,
I'll check the patch again
jirka
On Tue, Mar 16, 2021 at 01:25:29PM +0800, Jin, Yao wrote:
> Hi Jiri,
>
> On 3/16/2021 7:03 AM, Jiri Olsa wrote:
> > On Thu, Mar 11, 2021 at 03:07:31PM +0800, Jin Yao wrote:
> >
> > SNIP
> >
> > > goto try_again;
> > > }
> > > +
> > > + if (errno == EINVAL && perf_pmu__hybrid_exist())
> > > + evlist__warn_hybrid_group(evlist);
> > > rc = -errno;
> > > evsel__open_strerror(pos, &opts->target, errno, msg, sizeof(msg));
> > > ui__error("%s\n", msg);
> > > diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
> > > index 7a732508b2b4..6f780a039db0 100644
> > > --- a/tools/perf/builtin-stat.c
> > > +++ b/tools/perf/builtin-stat.c
> > > @@ -239,6 +239,9 @@ static void evlist__check_cpu_maps(struct evlist *evlist)
> > > struct evsel *evsel, *pos, *leader;
> > > char buf[1024];
> > > + if (evlist__hybrid_exist(evlist))
> > > + return;
> >
> > this should be in separate patch and explained
> >
>
> Now I have another idea. If a group consists of atom events and core events,
> we still follow current disabling group solution?
>
> I mean removing following code:
>
> if (evlist__hybrid_exist(evlist))
> return;
>
> evlist__check_cpu_maps then continues running and disabling the group. But
> also report with a warning that says "WARNING: Group has events from
> different hybrid PMUs".
>
> Do you like this way?
I'm not sure I follow completely.. would be best over the code
jirka
On Wed, Mar 17, 2021 at 10:42:45AM -0300, Arnaldo Carvalho de Melo wrote:
> Em Wed, Mar 17, 2021 at 08:17:52PM +0800, Jin, Yao escreveu:
> > Hi Jiri,
> >
> > On 3/17/2021 6:06 PM, Jiri Olsa wrote:
> > > On Wed, Mar 17, 2021 at 10:12:03AM +0800, Jin, Yao wrote:
> > > >
> > > >
> > > > On 3/16/2021 10:04 PM, Jiri Olsa wrote:
> > > > > On Tue, Mar 16, 2021 at 09:49:42AM +0800, Jin, Yao wrote:
> > > > >
> > > > > SNIP
> > > > >
> > > > > >
> > > > > > Performance counter stats for 'system wide':
> > > > > >
> > > > > > 136,655,302 cpu_core/branch-instructions/
> > > > > >
> > > > > > 1.003171561 seconds time elapsed
> > > > > >
> > > > > > So we need special rules for both cycles and branches.
> > > > > >
> > > > > > The worse thing is, we also need to process the hardware cache events.
> > > > > >
> > > > > > # ./perf stat -e cpu_core/LLC-loads/
> > > > > > event syntax error: 'cpu_core/LLC-loads/'
> > > > > > \___ unknown term 'LLC-loads' for pmu 'cpu_core'
> > > > > >
> > > > > > valid terms: event,pc,edge,offcore_rsp,ldlat,inv,umask,frontend,cmask,config,config1,config2,name,period,percore
> > > > > >
> > > > > > Initial error:
> > > > > > event syntax error: 'cpu_core/LLC-loads/'
> > > > > > \___ unknown term 'LLC-loads' for pmu 'cpu_core'
> > > > > >
> > > > > > If we use special rules for establishing all event mapping, that looks too much. :(
> > > > >
> > > > > hmmm but wait, currently we do not support events like this:
> > > > >
> > > > > 'cpu/cycles/'
> > > > > 'cpu/branches/'
> > > > >
> > > > > the pmu style accepts only 'events' or 'format' terms within //
> > > > >
> > > > > we made hw events like 'cycles','instructions','branches' special
> > > > > to be used without the pmu
> > > > >
> > > > > so why do we need to support cpu_code/cycles/ ?
>
> > > > Actually we have to support pmu style event for hybrid platform.
>
> > > > User may want to enable the events from specified pmus and also with flexible grouping.
>
> > > > For example,
>
> > > > perf stat -e '{cpu_core/cycles/,cpu_core/instructions/}' -e '{cpu_atom/cycles/,cpu_atom/instructions/}'
>
> > > > This usage is common and reasonable. So I think we may need to support pmu style events.
>
> > > sure, but we don't support 'cpu/cycles/' but we support 'cpu/cpu-cycles/'
> > > why do you insist on supporting cpu_core/cycles/ ?
>
> >
> > I'm OK to only support 'cpu_core/cpu-cycles/' or 'cpu_atom/cpu-cycles/'. But
> > what would we do for cache event?
> >
> > 'perf stat -e LLC-loads' is OK, but 'perf stat -e cpu/LLC-loads/' is not supported currently.
> >
> > For hybrid platform, user may only want to enable the LLC-loads on core CPUs
> > or on atom CPUs. That's reasonable. While if we don't support the pmu style
> > event, how to satisfy this requirement?
> >
> > If we can support the pmu style event, we can also use the same way for
> > cpu_core/cycles/. At least it's not a bad thing, right? :)
>
> While we're discussing, do we really want to use the "core" and "atom"
> terms here? I thought cpu/cycles/ would be ok for the main (Big) CPU and
> that we should come up with some short name for the "litle" CPUs.
>
> Won't we have the same situation with ARM where we want to know the
> number of cycles spent on a BIG core and also on a little one?
>
> Perhaps 'cycles' should mean all cycles, and then we use 'big/cycles/' and
> 'little/cycles/'?
do arm servers already export multiple pmus like this?
I did not notice
it'd be definitely great to have some unite way for this,
so far we have the hybrid pmu detection and support in
hw events like cycles/instructions.. which should be easy
to follow on arm
there's also support to have these events on specific pmu
pmu/cycles/ , which I still need to check on
jirka
Em Thu, Mar 18, 2021 at 01:16:37PM +0100, Jiri Olsa escreveu:
> On Wed, Mar 17, 2021 at 10:42:45AM -0300, Arnaldo Carvalho de Melo wrote:
> > Em Wed, Mar 17, 2021 at 08:17:52PM +0800, Jin, Yao escreveu:
> > > I'm OK to only support 'cpu_core/cpu-cycles/' or 'cpu_atom/cpu-cycles/'. But
> > > what would we do for cache event?
> > > 'perf stat -e LLC-loads' is OK, but 'perf stat -e cpu/LLC-loads/' is not supported currently.
> > > For hybrid platform, user may only want to enable the LLC-loads on core CPUs
> > > or on atom CPUs. That's reasonable. While if we don't support the pmu style
> > > event, how to satisfy this requirement?
> > > If we can support the pmu style event, we can also use the same way for
> > > cpu_core/cycles/. At least it's not a bad thing, right? :)
> > While we're discussing, do we really want to use the "core" and "atom"
> > terms here? I thought cpu/cycles/ would be ok for the main (Big) CPU and
> > that we should come up with some short name for the "litle" CPUs.
> > Won't we have the same situation with ARM where we want to know the
> > number of cycles spent on a BIG core and also on a little one?
> > Perhaps 'cycles' should mean all cycles, and then we use 'big/cycles/' and
> > 'little/cycles/'?
> do arm servers already export multiple pmus like this?
> I did not notice
I haven't checked, but AFAIK this BIG/Little kind of arch started there,
Mark?
- Arnaldo
> it'd be definitely great to have some unite way for this,
> so far we have the hybrid pmu detection and support in
> hw events like cycles/instructions.. which should be easy
> to follow on arm
>
> there's also support to have these events on specific pmu
> pmu/cycles/ , which I still need to check on
On 3/18/2021 9:21 AM, Arnaldo Carvalho de Melo wrote:
> Em Thu, Mar 18, 2021 at 01:16:37PM +0100, Jiri Olsa escreveu:
>> On Wed, Mar 17, 2021 at 10:42:45AM -0300, Arnaldo Carvalho de Melo wrote:
>>> Em Wed, Mar 17, 2021 at 08:17:52PM +0800, Jin, Yao escreveu:
>>>> I'm OK to only support 'cpu_core/cpu-cycles/' or 'cpu_atom/cpu-cycles/'. But
>>>> what would we do for cache event?
>
>>>> 'perf stat -e LLC-loads' is OK, but 'perf stat -e cpu/LLC-loads/' is not supported currently.
>
>>>> For hybrid platform, user may only want to enable the LLC-loads on core CPUs
>>>> or on atom CPUs. That's reasonable. While if we don't support the pmu style
>>>> event, how to satisfy this requirement?
>
>>>> If we can support the pmu style event, we can also use the same way for
>>>> cpu_core/cycles/. At least it's not a bad thing, right? :)
>
>>> While we're discussing, do we really want to use the "core" and "atom"
>>> terms here? I thought cpu/cycles/ would be ok for the main (Big) CPU and
>>> that we should come up with some short name for the "litle" CPUs.
>
>>> Won't we have the same situation with ARM where we want to know the
>>> number of cycles spent on a BIG core and also on a little one?
>
>>> Perhaps 'cycles' should mean all cycles, and then we use 'big/cycles/' and
>>> 'little/cycles/'?
>
>> do arm servers already export multiple pmus like this?
>> I did not notice
>
> I haven't checked, but AFAIK this BIG/Little kind of arch started there,
> Mark?
Here is the cover letter of the ARM big.little patch set. ARM also
exports multiple PMUs, e.g., armv7_cortex_a15 and armv7_cortex_a7.
https://lore.kernel.org/lkml/[email protected]/
We follow a similar way to handle the Intel hybrid PMUs. The naming rule
is also similar, "cpu_" + CPU type.
We don't use the old name "cpu" for the main CPU type, because we want
to make sure every software updated for the hybrid architecture.
Otherwise, the old script with "cpu//" can still run on a hybrid
architecture. Users cannot notice that the monitored scope is already
implicitly changed. The results may be not what they want.
Thanks,
Kan
>
> - Arnaldo
>
>> it'd be definitely great to have some unite way for this,
>> so far we have the hybrid pmu detection and support in
>> hw events like cycles/instructions.. which should be easy
>> to follow on arm
>>
>> there's also support to have these events on specific pmu
>> pmu/cycles/ , which I still need to check on
> While we're discussing, do we really want to use the "core" and "atom"
> terms here? I thought cpu/cycles/ would be ok for the main (Big) CPU and
Yes absolutely.
> that we should come up with some short name for the "litle" CPUs.
There actually isn't a main CPU.
There's nothing "better" about the big cores vs the Atoms
anyways. They're all important CPUs.
And the system might have no "big" CPUs, but we won't know
until we finished onlining all CPUs.
Or on Lakefield there are four Atoms and only a single big core.
So with a non hybrid aware profiler tool you would miss most of the
system if we used cpu// for the big core.
Also I think renaming is a good idea because it forces the software
or configuration to handle hybrid. Otherwise you just get subtle breakage
all the time with some CPUs not getting profiled.
It's a similar strategy as we do in the source code when semantics
change.
ARM did this right.
-Andi