This patch series unifies the counters that can be obtained from
perf_evsel__read() to "no scaling".
The counter scaling will be done using a function moved from
tools/perf/util.
The first patch move perf_counts_values__scale from tools/perf/util
to tools/lib/perf so that it can be used with libperf.
The second patch removes the scaling process from
perf_mmap__read_self().
The third patch adds a verification test to make sure that it scales
correctly when multiplexed.
---
Previous version at:
https://lore.kernel.org/linux-perf-users/[email protected]/
Changes in v4:
- Modify type s8 to type __s8
Changes in v3:
- Move scaling process from tools/perf/util to tools/lib/perf
- Remove scaling process from perf_mmap__read_self()
- Remove scaling process for perf_mmap__read_self
- Remove test to verify that no division by zero occurs
Changes in v2:
- Fix not to divide by zero when counter scaling
- Add test to verify that no division by zero occurs
[1] https://github.com/deater/perf_event_tests/blob/master/tests/rdpmc/rdpmc_multiplexing.c
Shunsuke Nakamura (3):
libperf: Move perf_counts_values__scale to tools/lib/perf
libperf: Remove scaling process from perf_mmap__read_self()
libperf tests: Add test_stat_multiplexing test
tools/lib/perf/evsel.c | 19 ++++
tools/lib/perf/include/perf/evsel.h | 4 +
tools/lib/perf/libperf.map | 1 +
tools/lib/perf/mmap.c | 2 -
tools/lib/perf/tests/test-evlist.c | 157 ++++++++++++++++++++++++++++
tools/perf/util/evsel.c | 19 ----
tools/perf/util/evsel.h | 3 -
7 files changed, 181 insertions(+), 24 deletions(-)
--
2.25.1
From: Shunsuke Nakamura <[email protected]>
Move perf_counts_values__scale from tools/perf/util to tools/lib/perf
so that it can be used with libperf.
Signed-off-by: Shunsuke Nakamura <[email protected]>
---
tools/lib/perf/evsel.c | 19 +++++++++++++++++++
tools/lib/perf/include/perf/evsel.h | 4 ++++
tools/lib/perf/libperf.map | 1 +
tools/perf/util/evsel.c | 19 -------------------
tools/perf/util/evsel.h | 3 ---
5 files changed, 24 insertions(+), 22 deletions(-)
diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
index 8441e3e1aaac..782d1466df1f 100644
--- a/tools/lib/perf/evsel.c
+++ b/tools/lib/perf/evsel.c
@@ -431,3 +431,22 @@ void perf_evsel__free_id(struct perf_evsel *evsel)
zfree(&evsel->id);
evsel->ids = 0;
}
+
+void perf_counts_values__scale(struct perf_counts_values *count,
+ bool scale, __s8 *pscaled)
+{
+ __s8 scaled = 0;
+
+ if (scale) {
+ if (count->run == 0) {
+ scaled = -1;
+ count->val = 0;
+ } else if (count->run < count->ena) {
+ scaled = 1;
+ count->val = (u64)((double)count->val * count->ena / count->run);
+ }
+ }
+
+ if (pscaled)
+ *pscaled = scaled;
+}
diff --git a/tools/lib/perf/include/perf/evsel.h b/tools/lib/perf/include/perf/evsel.h
index 60eae25076d3..f401c7484bec 100644
--- a/tools/lib/perf/include/perf/evsel.h
+++ b/tools/lib/perf/include/perf/evsel.h
@@ -4,6 +4,8 @@
#include <stdint.h>
#include <perf/core.h>
+#include <stdbool.h>
+#include <linux/types.h>
struct perf_evsel;
struct perf_event_attr;
@@ -39,5 +41,7 @@ LIBPERF_API int perf_evsel__disable_cpu(struct perf_evsel *evsel, int cpu);
LIBPERF_API struct perf_cpu_map *perf_evsel__cpus(struct perf_evsel *evsel);
LIBPERF_API struct perf_thread_map *perf_evsel__threads(struct perf_evsel *evsel);
LIBPERF_API struct perf_event_attr *perf_evsel__attr(struct perf_evsel *evsel);
+LIBPERF_API void perf_counts_values__scale(struct perf_counts_values *count,
+ bool scale, __s8 *pscaled);
#endif /* __LIBPERF_EVSEL_H */
diff --git a/tools/lib/perf/libperf.map b/tools/lib/perf/libperf.map
index 71468606e8a7..5979bf92d98f 100644
--- a/tools/lib/perf/libperf.map
+++ b/tools/lib/perf/libperf.map
@@ -50,6 +50,7 @@ LIBPERF_0.0.1 {
perf_mmap__read_init;
perf_mmap__read_done;
perf_mmap__read_event;
+ perf_counts_values__scale;
local:
*;
};
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index ac0127be0459..656c30b988ce 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -1476,25 +1476,6 @@ void evsel__compute_deltas(struct evsel *evsel, int cpu, int thread,
count->run = count->run - tmp.run;
}
-void perf_counts_values__scale(struct perf_counts_values *count,
- bool scale, s8 *pscaled)
-{
- s8 scaled = 0;
-
- if (scale) {
- if (count->run == 0) {
- scaled = -1;
- count->val = 0;
- } else if (count->run < count->ena) {
- scaled = 1;
- count->val = (u64)((double) count->val * count->ena / count->run);
- }
- }
-
- if (pscaled)
- *pscaled = scaled;
-}
-
static int evsel__read_one(struct evsel *evsel, int cpu, int thread)
{
struct perf_counts_values *count = perf_counts(evsel->counts, cpu, thread);
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 29d49a8c1e92..99aa3363def7 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -195,9 +195,6 @@ static inline int evsel__nr_cpus(struct evsel *evsel)
return evsel__cpus(evsel)->nr;
}
-void perf_counts_values__scale(struct perf_counts_values *count,
- bool scale, s8 *pscaled);
-
void evsel__compute_deltas(struct evsel *evsel, int cpu, int thread,
struct perf_counts_values *count);
--
2.25.1
From: Shunsuke Nakamura <[email protected]>
Remove the scaling process from perf_mmap__read_self(), and unify the
counters that can be obtained from perf_evsel__read() to "no scaling".
Acked-by: Jiri Olsa <[email protected]>
Signed-off-by: Shunsuke Nakamura <[email protected]>
---
tools/lib/perf/mmap.c | 2 --
1 file changed, 2 deletions(-)
diff --git a/tools/lib/perf/mmap.c b/tools/lib/perf/mmap.c
index c89dfa5f67b3..aaa457904008 100644
--- a/tools/lib/perf/mmap.c
+++ b/tools/lib/perf/mmap.c
@@ -353,8 +353,6 @@ int perf_mmap__read_self(struct perf_mmap *map, struct perf_counts_values *count
count->ena += delta;
if (idx)
count->run += delta;
-
- cnt = mul_u64_u64_div64(cnt, count->ena, count->run);
}
count->val = cnt;
--
2.25.1
From: Shunsuke Nakamura <[email protected]>
Adds a test for a counter obtained using read() system call during
multiplexing.
Committer testing:
$ sudo make tests -C ./tools/lib/perf V=1
make[1]: Entering directory '/home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/lib/perf'
make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=. obj=libperf
make -C /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/lib/api/ O= libapi.a
make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=./fd obj=libapi
make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=./fs obj=libapi
make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=. obj=tests
make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=./tests obj=tests
running static:
- running tests/test-cpumap.c...OK
- running tests/test-threadmap.c...OK
- running tests/test-evlist.c...
Event 0 -- Raw count = 299302036, run = 288430205, enable = 486235490
Scaled count = 504563217 (59.32%, 288430205/486235490)
Event 1 -- Raw count = 298545418, run = 289425078, enable = 486232769
Scaled count = 501554897 (59.52%, 289425078/486232769)
Event 2 -- Raw count = 298838327, run = 290411278, enable = 486229222
Scaled count = 500338444 (59.73%, 290411278/486229222)
Event 3 -- Raw count = 299382984, run = 291406006, enable = 486225338
Scaled count = 499535320 (59.93%, 291406006/486225338)
Event 4 -- Raw count = 300119283, run = 292411181, enable = 486220615
Scaled count = 499037628 (60.14%, 292411181/486220615)
Event 5 -- Raw count = 300255891, run = 293404372, enable = 486215261
Scaled count = 497569260 (60.34%, 293404372/486215261)
Event 6 -- Raw count = 302390604, run = 294222082, enable = 486209147
Scaled count = 499707828 (60.51%, 294222082/486209147)
Event 7 -- Raw count = 303524473, run = 294214550, enable = 486202374
Scaled count = 501587427 (60.51%, 294214550/486202374)
Event 8 -- Raw count = 304282474, run = 294206637, enable = 486194647
Scaled count = 502845590 (60.51%, 294206637/486194647)
Event 9 -- Raw count = 303969062, run = 293818554, enable = 486185895
Scaled count = 502982090 (60.43%, 293818554/486185895)
Event 10 -- Raw count = 303784907, run = 292808731, enable = 486174252
Scaled count = 504398893 (60.23%, 292808731/486174252)
Event 11 -- Raw count = 304129077, run = 291818946, enable = 486164475
Scaled count = 506672904 (60.02%, 291818946/486164475)
Event 12 -- Raw count = 302677227, run = 290819976, enable = 486155254
Scaled count = 505976673 (59.82%, 290819976/486155254)
Event 13 -- Raw count = 301748186, run = 289821793, enable = 486146028
Scaled count = 506151316 (59.62%, 289821793/486146028)
Event 14 -- Raw count = 300216852, run = 288823624, enable = 486136825
Scaled count = 505313468 (59.41%, 288823624/486136825)
Expected: 501623296
High: 506672904 Low: 299302036 Average: 502548997
Average Error = 0.18%
OK
- running tests/test-evsel.c...
loop = 65536, count = 354844
loop = 131072, count = 693386
loop = 262144, count = 1339140
loop = 524288, count = 2726813
loop = 1048576, count = 5263081
loop = 65536, count = 472935
loop = 131072, count = 1018641
loop = 262144, count = 1840393
loop = 524288, count = 3349520
loop = 1048576, count = 6759085
OK
running dynamic:
- running tests/test-cpumap.c...OK
- running tests/test-threadmap.c...OK
- running tests/test-evlist.c...
Event 0 -- Raw count = 297934481, run = 297856898, enable = 501847789
Scaled count = 501978505 (59.35%, 297856898/501847789)
Event 1 -- Raw count = 298651420, run = 297970032, enable = 501845063
Scaled count = 502992665 (59.37%, 297970032/501845063)
Event 2 -- Raw count = 299731258, run = 298965119, enable = 501841526
Scaled count = 503127563 (59.57%, 298965119/501841526)
Event 3 -- Raw count = 300644680, run = 299960013, enable = 501837664
Scaled count = 502983122 (59.77%, 299960013/501837664)
Event 4 -- Raw count = 301854645, run = 300947092, enable = 501833170
Scaled count = 503346526 (59.97%, 300947092/501833170)
Event 5 -- Raw count = 303228020, run = 302371294, enable = 501827577
Scaled count = 503249433 (60.25%, 302371294/501827577)
Event 6 -- Raw count = 304162593, run = 303363734, enable = 501820925
Scaled count = 503142388 (60.45%, 303363734/501820925)
Event 7 -- Raw count = 304802876, run = 304254160, enable = 501814005
Scaled count = 502719016 (60.63%, 304254160/501814005)
Event 8 -- Raw count = 304797825, run = 304246193, enable = 501805969
Scaled count = 502715798 (60.63%, 304246193/501805969)
Event 9 -- Raw count = 304678040, run = 303884090, enable = 501797023
Scaled count = 503108054 (60.56%, 303884090/501797023)
Event 10 -- Raw count = 303209569, run = 302886553, enable = 501785047
Scaled count = 502320180 (60.36%, 302886553/501785047)
Event 11 -- Raw count = 301801701, run = 301456719, enable = 501775302
Scaled count = 502349525 (60.08%, 301456719/501775302)
Event 12 -- Raw count = 300744781, run = 300457620, enable = 501765832
Scaled count = 502245392 (59.88%, 300457620/501765832)
Event 13 -- Raw count = 299656316, run = 299458825, enable = 501756687
Scaled count = 502087592 (59.68%, 299458825/501756687)
Event 14 -- Raw count = 298616379, run = 298460087, enable = 501747345
Scaled count = 502010090 (59.48%, 298460087/501747345)
Expected: 501618012
High: 503346526 Low: 297934481 Average: 502691723
Average Error = 0.21%
OK
- running tests/test-evsel.c...
loop = 65536, count = 328183
loop = 131072, count = 681853
loop = 262144, count = 1334936
loop = 524288, count = 2702197
loop = 1048576, count = 5294233
loop = 65536, count = 486498
loop = 131072, count = 928179
loop = 262144, count = 1921944
loop = 524288, count = 3687007
loop = 1048576, count = 6721087
OK
make[1]: Leaving directory '/home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/lib/perf'
Acked-by: Jiri Olsa <[email protected]>
Signed-off-by: Shunsuke Nakamura <[email protected]>
---
tools/lib/perf/tests/test-evlist.c | 157 +++++++++++++++++++++++++++++
1 file changed, 157 insertions(+)
diff --git a/tools/lib/perf/tests/test-evlist.c b/tools/lib/perf/tests/test-evlist.c
index ce91a582f0e4..064edd0e995c 100644
--- a/tools/lib/perf/tests/test-evlist.c
+++ b/tools/lib/perf/tests/test-evlist.c
@@ -21,6 +21,9 @@
#include "tests.h"
#include <internal/evsel.h>
+#define EVENT_NUM 15
+#define WAIT_COUNT 100000000UL
+
static int libperf_print(enum libperf_print_level level,
const char *fmt, va_list ap)
{
@@ -413,6 +416,159 @@ static int test_mmap_cpus(void)
return 0;
}
+static double display_error(long long average,
+ long long high,
+ long long low,
+ long long expected)
+{
+ double error;
+
+ error = (((double)average - expected) / expected) * 100.0;
+
+ __T_VERBOSE(" Expected: %lld\n", expected);
+ __T_VERBOSE(" High: %lld Low: %lld Average: %lld\n",
+ high, low, average);
+
+ __T_VERBOSE(" Average Error = %.2f%%\n", error);
+
+ return error;
+}
+
+static int test_stat_multiplexing(void)
+{
+ struct perf_counts_values expected_counts = { .val = 0 };
+ struct perf_counts_values counts[EVENT_NUM] = {{ .val = 0 },};
+ struct perf_thread_map *threads;
+ struct perf_evlist *evlist;
+ struct perf_evsel *evsel;
+ struct perf_event_attr attr = {
+ .type = PERF_TYPE_HARDWARE,
+ .config = PERF_COUNT_HW_INSTRUCTIONS,
+ .read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
+ PERF_FORMAT_TOTAL_TIME_RUNNING,
+ .disabled = 1,
+ };
+ int err, i, nonzero = 0;
+ unsigned long count;
+ long long max = 0, min = 0, avg = 0;
+ double error = 0.0;
+ __s8 scaled = 0;
+
+ /* read for non-multiplexing event count */
+ threads = perf_thread_map__new_dummy();
+ __T("failed to create threads", threads);
+
+ perf_thread_map__set_pid(threads, 0, 0);
+
+ evsel = perf_evsel__new(&attr);
+ __T("failed to create evsel", evsel);
+
+ err = perf_evsel__open(evsel, NULL, threads);
+ __T("failed to open evsel", err == 0);
+
+ err = perf_evsel__enable(evsel);
+ __T("failed to enable evsel", err == 0);
+
+ /* wait loop */
+ count = WAIT_COUNT;
+ while (count--)
+ ;
+
+ perf_evsel__read(evsel, 0, 0, &expected_counts);
+ __T("failed to read value for evsel", expected_counts.val != 0);
+ __T("failed to read non-multiplexing event count",
+ expected_counts.ena == expected_counts.run);
+
+ err = perf_evsel__disable(evsel);
+ __T("failed to enable evsel", err == 0);
+
+ perf_evsel__close(evsel);
+ perf_evsel__delete(evsel);
+
+ perf_thread_map__put(threads);
+
+ /* read for multiplexing event count */
+ threads = perf_thread_map__new_dummy();
+ __T("failed to create threads", threads);
+
+ perf_thread_map__set_pid(threads, 0, 0);
+
+ evlist = perf_evlist__new();
+ __T("failed to create evlist", evlist);
+
+ for (i = 0; i < EVENT_NUM; i++) {
+ evsel = perf_evsel__new(&attr);
+ __T("failed to create evsel", evsel);
+
+ perf_evlist__add(evlist, evsel);
+ }
+ perf_evlist__set_maps(evlist, NULL, threads);
+
+ err = perf_evlist__open(evlist);
+ __T("failed to open evsel", err == 0);
+
+ perf_evlist__enable(evlist);
+
+ /* wait loop */
+ count = WAIT_COUNT;
+ while (count--)
+ ;
+
+ i = 0;
+ perf_evlist__for_each_evsel(evlist, evsel) {
+ perf_evsel__read(evsel, 0, 0, &counts[i]);
+ __T("failed to read value for evsel", counts[i].val != 0);
+ i++;
+ }
+
+ perf_evlist__disable(evlist);
+
+ min = counts[0].val;
+ for (i = 0; i < EVENT_NUM; i++) {
+ __T_VERBOSE("Event %2d -- Raw count = %lu, run = %lu, enable = %lu\n",
+ i, counts[i].val, counts[i].run, counts[i].ena);
+
+ perf_counts_values__scale(&counts[i], true, &scaled);
+ if (scaled == 1) {
+ __T_VERBOSE("\t Scaled count = %lu (%.2lf%%, %lu/%lu)\n",
+ counts[i].val,
+ (double)counts[i].run / (double)counts[i].ena * 100.0,
+ counts[i].run, counts[i].ena);
+ } else if (scaled == -1) {
+ __T_VERBOSE("\t Not Runnnig\n");
+ } else {
+ __T_VERBOSE("\t Not Scaling\n");
+ }
+
+ if (counts[i].val > max)
+ max = counts[i].val;
+
+ if (counts[i].val < min)
+ min = counts[i].val;
+
+ avg += counts[i].val;
+
+ if (counts[i].val != 0)
+ nonzero++;
+ }
+
+ if (nonzero != 0)
+ avg = avg / nonzero;
+ else
+ avg = 0;
+
+ error = display_error(avg, max, min, expected_counts.val);
+
+ __T("Error out of range!", ((error <= 1.0) && (error >= -1.0)));
+
+ perf_evlist__close(evlist);
+ perf_evlist__delete(evlist);
+
+ perf_thread_map__put(threads);
+
+ return 0;
+}
+
int test_evlist(int argc, char **argv)
{
__T_START;
@@ -424,6 +580,7 @@ int test_evlist(int argc, char **argv)
test_stat_thread_enable();
test_mmap_thread();
test_mmap_cpus();
+ test_stat_multiplexing();
__T_END;
return tests_failed == 0 ? 0 : -1;
--
2.25.1
> Move perf_counts_values__scale from tools/perf/util to tools/lib/perf
> so that it can be used with libperf.
>
> Signed-off-by: Shunsuke Nakamura <[email protected]>
> ---
> tools/lib/perf/evsel.c | 19 +++++++++++++++++++
> tools/lib/perf/include/perf/evsel.h | 4 ++++
> tools/lib/perf/libperf.map | 1 +
> tools/perf/util/evsel.c | 19 -------------------
> tools/perf/util/evsel.h | 3 ---
> 5 files changed, 24 insertions(+), 22 deletions(-)
I missed to update tools/lib/perf/Documentation/libperf.txt, so I will fix it.