2009-09-11 10:12:54 +00:00
|
|
|
#include "builtin.h"
|
2009-09-11 10:12:54 +00:00
|
|
|
#include "perf.h"
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
#include "util/util.h"
|
perf tools: Save some loops using perf_evlist__id2evsel
Since we already ask for PERF_SAMPLE_ID and use it to quickly find the
associated evsel, add handler func + data to struct perf_evsel to avoid
using chains of if(strcmp(event_name)) and also to avoid all the linear
list searches via trace_event_find.
To demonstrate the technique convert 'perf sched' to it:
# perf sched record sleep 5m
And then:
Performance counter stats for '/tmp/oldperf sched lat':
646.929438 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,901 page-faults # 0.032 M/sec
1,290,144,450 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,606,158,439 instructions # 1.24 insns per cycle
339,088,395 branches # 524.151 M/sec
4,550,735 branch-misses # 1.34% of all branches
0.647524759 seconds time elapsed
Versus:
Performance counter stats for 'perf sched lat':
473.564691 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,903 page-faults # 0.044 M/sec
944,367,984 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,442,385,571 instructions # 1.53 insns per cycle
308,383,106 branches # 651.195 M/sec
4,481,784 branch-misses # 1.45% of all branches
0.474215751 seconds time elapsed
[root@emilia ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-1kbzpl74lwi6lavpqke2u2p3@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-11-28 19:57:40 +00:00
|
|
|
#include "util/evlist.h"
|
2009-09-11 10:12:54 +00:00
|
|
|
#include "util/cache.h"
|
2011-11-16 19:02:54 +00:00
|
|
|
#include "util/evsel.h"
|
2009-09-11 10:12:54 +00:00
|
|
|
#include "util/symbol.h"
|
|
|
|
#include "util/thread.h"
|
|
|
|
#include "util/header.h"
|
2009-12-11 23:24:02 +00:00
|
|
|
#include "util/session.h"
|
2011-11-28 10:30:20 +00:00
|
|
|
#include "util/tool.h"
|
2014-06-30 20:28:47 +00:00
|
|
|
#include "util/cloexec.h"
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
#include "util/parse-options.h"
|
2009-09-11 10:12:54 +00:00
|
|
|
#include "util/trace-event.h"
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
#include "util/debug.h"
|
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
#include <sys/prctl.h>
|
2012-04-04 08:45:27 +00:00
|
|
|
#include <sys/resource.h>
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
#include <semaphore.h>
|
|
|
|
#include <pthread.h>
|
|
|
|
#include <math.h>
|
perf sched replay: Alloc the memory of pid_to_task dynamically to adapt to the unexpected change of pid_max
The current memory allocation of struct task_desc *pid_to_task[MAX_PID]
is in a permanent and preset way, and it has two problems:
Problem 1: If the pid_max, which is the max number of pids in the
system, is much smaller than MAX_PID (1024*1000), then it causes a waste
of stack memory. This may happen in the case where the number of cpu
cores is much smaller than 1000.
Problem 2: If the pid_max is changed from the default value to a value
larger than MAX_PID, then it will cause assertion failure problem. The
maximum value of pid_max can be set to pid_max_max (see pidmap_init
defined in kernel/pid.c), which equals to PID_MAX_LIMIT. In x86_64,
PID_MAX_LIMIT is 4*1024*1024 (defined in include/linux/threads.h). This
value is much larger than MAX_PID, and will take up 32768 Kbytes
(4*1024*1024*8/1024) for memory allocation of pid_to_task, which is much
larger than the default 8192 Kbytes of the stack size of calling
process.
Due to these two problems, we use calloc to allocate the memory of
pid_to_task dynamically.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ echo 1025000 > /proc/sys/kernel/pid_max
$ cat /proc/sys/kernel/pid_max
1025000
Run some applications until the pid of some process is greater than
the value of MAX_PID (1024*1000).
Before this patch:
$ perf sched replay
run measurement overhead: 221 nsecs
sleep measurement overhead: 55480 nsecs
the run test took 1000008 nsecs
the sleep test took 1063151 nsecs
perf: builtin-sched.c:330: register_pid: Assertion `!(pid >= 1024000)'
failed.
Aborted
After this patch:
$ perf sched replay
run measurement overhead: 221 nsecs
sleep measurement overhead: 55435 nsecs
the run test took 1000004 nsecs
the sleep test took 1059312 nsecs
nr_run_events: 10
nr_sleep_events: 1562
nr_wakeup_events: 5
task 0 ( :1: 1), nr_events: 1
task 1 ( :2: 2), nr_events: 1
task 2 ( :3: 3), nr_events: 1
task 3 ( :5: 5), nr_events: 1
...
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-4-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:30 +00:00
|
|
|
#include <api/fs/fs.h>
|
2009-09-12 01:59:01 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
#define PR_SET_NAME 15 /* Set process name */
|
|
|
|
#define MAX_CPUS 4096
|
|
|
|
#define COMM_LEN 20
|
|
|
|
#define SYM_LEN 129
|
perf sched replay: Increase the MAX_PID value to fix assertion failure problem
Current MAX_PID is only 65536, which will cause assertion failure problem
when CPU cores are more than 64 in x86_64.
This is because the pid_max value in x86_64 is at least
PIDS_PER_CPU_DEFAULT * num_possible_cpus() (see function pidmap_init
defined in kernel/pid.c), where PIDS_PER_CPU_DEFAULT is 1024 (defined in
include/linux/threads.h).
Thus for MAX_PID = 65536, the correspoinding CPU cores are
65536/1024=64. This is obviously not enough at all for x86_64, and will
cause an assertion failure problem due to BUG_ON(pid >= MAX_PID) in the
codes.
We increase MAX_PID value from 65536 to 1024*1000, which can be used in
x86_64 with 1000 cores.
This number is finally decided according to the limitation of stack size
of calling process.
Use 'ulimit -a', the result shows the stack size of any process is 8192
Kbytes, which is defined in include/uapi/linux/resource.h (#define
_STK_LIM (8*1024*1024)).
Thus we choose a large enough value for MAX_PID, and make it satisfy to
the limitation of the stack size, i.e., making the perf process take up
a memory space just smaller than 8192 Kbytes.
We have calculated and tested that 1024*1000 is OK for MAX_PID.
This means perf sched replay can now be used with at most 1000 cores in
x86_64 without any assertion failure problem.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
Before this patch:
$ perf sched replay
run measurement overhead: 240 nsecs
sleep measurement overhead: 55379 nsecs
the run test took 1000004 nsecs
the sleep test took 1059424 nsecs
perf: builtin-sched.c:330: register_pid: Assertion `!(pid >= 65536)'
failed.
Aborted
After this patch:
$ perf sched replay
run measurement overhead: 221 nsecs
sleep measurement overhead: 55397 nsecs
the run test took 999920 nsecs
the sleep test took 1053313 nsecs
nr_run_events: 10
nr_sleep_events: 1562
nr_wakeup_events: 5
task 0 ( :1: 1), nr_events: 1
task 1 ( :2: 2), nr_events: 1
task 2 ( :3: 3), nr_events: 1
task 3 ( :5: 5), nr_events: 1
...
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-3-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:29 +00:00
|
|
|
#define MAX_PID 1024000
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
struct sched_atom;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
struct task_desc {
|
|
|
|
unsigned long nr;
|
|
|
|
unsigned long pid;
|
|
|
|
char comm[COMM_LEN];
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
unsigned long nr_events;
|
|
|
|
unsigned long curr_event;
|
2009-09-14 18:04:48 +00:00
|
|
|
struct sched_atom **atoms;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
pthread_t thread;
|
|
|
|
sem_t sleep_sem;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
sem_t ready_for_work;
|
|
|
|
sem_t work_done_sem;
|
|
|
|
|
|
|
|
u64 cpu_usage;
|
|
|
|
};
|
|
|
|
|
|
|
|
enum sched_event_type {
|
|
|
|
SCHED_EVENT_RUN,
|
|
|
|
SCHED_EVENT_SLEEP,
|
|
|
|
SCHED_EVENT_WAKEUP,
|
2009-10-10 12:46:04 +00:00
|
|
|
SCHED_EVENT_MIGRATION,
|
2009-09-11 10:12:54 +00:00
|
|
|
};
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
struct sched_atom {
|
2009-09-11 10:12:54 +00:00
|
|
|
enum sched_event_type type;
|
perf tools: Reorganize some structs to save space
Using 'pahole --packable' I found some structs that could be reorganized
to eliminate alignment holes, in some cases getting them to be cacheline
multiples.
[acme@doppio linux-2.6-tip]$ codiff perf.old ~/bin/perf
builtin-annotate.c:
struct perf_session | -8
struct perf_header | -8
2 structs changed
builtin-diff.c:
struct sample_data | -8
1 struct changed
diff__process_sample_event | -8
1 function changed, 8 bytes removed, diff: -8
builtin-sched.c:
struct sched_atom | -8
1 struct changed
builtin-timechart.c:
struct per_pid | -8
1 struct changed
cmd_timechart | -16
1 function changed, 16 bytes removed, diff: -16
builtin-probe.c:
struct perf_probe_point | -8
struct perf_probe_event | -8
2 structs changed
opt_add_probe_event | -3
1 function changed, 3 bytes removed, diff: -3
util/probe-finder.c:
struct probe_finder | -8
1 struct changed
find_kprobe_trace_events | -16
1 function changed, 16 bytes removed, diff: -16
/home/acme/bin/perf:
4 functions changed, 43 bytes removed, diff: -43
[acme@doppio linux-2.6-tip]$
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2010-04-05 15:53:45 +00:00
|
|
|
int specific_wait;
|
2009-09-11 10:12:54 +00:00
|
|
|
u64 timestamp;
|
|
|
|
u64 duration;
|
|
|
|
unsigned long nr;
|
|
|
|
sem_t *wait_sem;
|
|
|
|
struct task_desc *wakee;
|
|
|
|
};
|
|
|
|
|
2014-05-05 07:05:54 +00:00
|
|
|
#define TASK_STATE_TO_CHAR_STR "RSDTtZXxKWP"
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
enum thread_state {
|
|
|
|
THREAD_SLEEPING = 0,
|
|
|
|
THREAD_WAIT_CPU,
|
|
|
|
THREAD_SCHED_IN,
|
|
|
|
THREAD_IGNORE
|
|
|
|
};
|
|
|
|
|
|
|
|
struct work_atom {
|
|
|
|
struct list_head list;
|
|
|
|
enum thread_state state;
|
perf tools: Fix processing of randomly serialized sched traces
Currently it's possible to meet such too high latency results
with 'perf sched latency'.
-----------------------------------------------------------------------------------
Task | Runtime ms | Switches | Average delay ms | Maximum delay ms |
-----------------------------------------------------------------------------------
xfce4-panel | 0.222 ms | 2 | avg: 4718.345 ms | max: 9436.493 ms |
scsi_eh_3 | 3.962 ms | 36 | avg: 55.957 ms | max: 1977.829 ms |
The origin is on traces that are sometimes badly serialized across cpus.
For example the raw traces that raised such results for xfce4-panel:
(1) [init]-0 [000] 1494.663899990: sched_switch: task swapper:0 [140] (R) ==> xfce4-panel:4569 [120]
(2) xfce4-panel-4569 [000] 1494.663928373: sched_switch: task xfce4-panel:4569 [120] (S) ==> swapper:0 [140]
(3) Xorg-4276 [001] 1494.663860125: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(4) Xorg-4276 [001] 1504.098252756: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(5) perf-5219 [000] 1504.100353302: sched_switch: task perf:5219 [120] (S) ==> xfce4-panel:4569 [120]
The traces are processed in the order they arrive. Then in (2),
xfce4-panel sleeps, it is first waken up in (3) and eventually
scheduled in (5).
The latency reported is then 1504 - 1495 = 9 secs, as reported by perf
sched. But this is wrong, we are confident in the fact the traces are
nicely serialized while we should actually more trust the timestamps.
If we reorder by timestamps we get:
(1) Xorg-4276 [001] 1494.663860125: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(2) [init]-0 [000] 1494.663899990: sched_switch: task swapper:0 [140] (R) ==> xfce4-panel:4569 [120]
(3) xfce4-panel-4569 [000] 1494.663928373: sched_switch: task xfce4-panel:4569 [120] (S) ==> swapper:0 [140]
(4) Xorg-4276 [001] 1504.098252756: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(5) perf-5219 [000] 1504.100353302: sched_switch: task perf:5219 [120] (S) ==> xfce4-panel:4569 [120]
Now the trace make more sense, xfce4-panel is sleeping. Then it is
woken up in (1), scheduled in (2)
It goes to sleep in (3), woken up in (4) and scheduled in (5).
Now, latency captured between (1) and (2) is of 39 us.
And between (4) and (5) it is 2.1 ms.
Such pattern of bad serializing is the origin of the high latencies
reported by perf sched.
Basically, we need to check whether wake up time is higher than
schedule out time. If it's not the case, we need to tag the current
work atom as invalid.
Beside that, we may need to work later on a better ordering of the
traces given by the kernel.
After this patch:
xfce4-session | 0.221 ms | 1 | avg: 0.538 ms | max: 0.538 ms |
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-14 01:01:12 +00:00
|
|
|
u64 sched_out_time;
|
2009-09-11 10:12:54 +00:00
|
|
|
u64 wake_up_time;
|
|
|
|
u64 sched_in_time;
|
|
|
|
u64 runtime;
|
|
|
|
};
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
struct work_atoms {
|
|
|
|
struct list_head work_list;
|
2009-09-11 10:12:54 +00:00
|
|
|
struct thread *thread;
|
|
|
|
struct rb_node node;
|
|
|
|
u64 max_lat;
|
2009-12-09 20:40:08 +00:00
|
|
|
u64 max_lat_at;
|
2009-09-11 10:12:54 +00:00
|
|
|
u64 total_lat;
|
|
|
|
u64 nb_atoms;
|
|
|
|
u64 total_runtime;
|
|
|
|
};
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
struct perf_sched;
|
2012-09-11 20:29:27 +00:00
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
struct trace_sched_handler {
|
|
|
|
int (*switch_event)(struct perf_sched *sched, struct perf_evsel *evsel,
|
|
|
|
struct perf_sample *sample, struct machine *machine);
|
2012-09-11 20:29:27 +00:00
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
int (*runtime_event)(struct perf_sched *sched, struct perf_evsel *evsel,
|
|
|
|
struct perf_sample *sample, struct machine *machine);
|
2012-09-11 20:29:27 +00:00
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
int (*wakeup_event)(struct perf_sched *sched, struct perf_evsel *evsel,
|
|
|
|
struct perf_sample *sample, struct machine *machine);
|
2012-09-11 20:29:27 +00:00
|
|
|
|
2013-08-08 02:50:47 +00:00
|
|
|
/* PERF_RECORD_FORK event, not sched_process_fork tracepoint */
|
|
|
|
int (*fork_event)(struct perf_sched *sched, union perf_event *event,
|
|
|
|
struct machine *machine);
|
2012-09-11 20:29:27 +00:00
|
|
|
|
|
|
|
int (*migrate_task_event)(struct perf_sched *sched,
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
struct perf_evsel *evsel,
|
|
|
|
struct perf_sample *sample,
|
|
|
|
struct machine *machine);
|
2012-09-11 20:29:27 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
struct perf_sched {
|
|
|
|
struct perf_tool tool;
|
|
|
|
const char *sort_order;
|
|
|
|
unsigned long nr_tasks;
|
perf sched replay: Alloc the memory of pid_to_task dynamically to adapt to the unexpected change of pid_max
The current memory allocation of struct task_desc *pid_to_task[MAX_PID]
is in a permanent and preset way, and it has two problems:
Problem 1: If the pid_max, which is the max number of pids in the
system, is much smaller than MAX_PID (1024*1000), then it causes a waste
of stack memory. This may happen in the case where the number of cpu
cores is much smaller than 1000.
Problem 2: If the pid_max is changed from the default value to a value
larger than MAX_PID, then it will cause assertion failure problem. The
maximum value of pid_max can be set to pid_max_max (see pidmap_init
defined in kernel/pid.c), which equals to PID_MAX_LIMIT. In x86_64,
PID_MAX_LIMIT is 4*1024*1024 (defined in include/linux/threads.h). This
value is much larger than MAX_PID, and will take up 32768 Kbytes
(4*1024*1024*8/1024) for memory allocation of pid_to_task, which is much
larger than the default 8192 Kbytes of the stack size of calling
process.
Due to these two problems, we use calloc to allocate the memory of
pid_to_task dynamically.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ echo 1025000 > /proc/sys/kernel/pid_max
$ cat /proc/sys/kernel/pid_max
1025000
Run some applications until the pid of some process is greater than
the value of MAX_PID (1024*1000).
Before this patch:
$ perf sched replay
run measurement overhead: 221 nsecs
sleep measurement overhead: 55480 nsecs
the run test took 1000008 nsecs
the sleep test took 1063151 nsecs
perf: builtin-sched.c:330: register_pid: Assertion `!(pid >= 1024000)'
failed.
Aborted
After this patch:
$ perf sched replay
run measurement overhead: 221 nsecs
sleep measurement overhead: 55435 nsecs
the run test took 1000004 nsecs
the sleep test took 1059312 nsecs
nr_run_events: 10
nr_sleep_events: 1562
nr_wakeup_events: 5
task 0 ( :1: 1), nr_events: 1
task 1 ( :2: 2), nr_events: 1
task 2 ( :3: 3), nr_events: 1
task 3 ( :5: 5), nr_events: 1
...
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-4-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:30 +00:00
|
|
|
struct task_desc **pid_to_task;
|
2012-09-11 20:29:27 +00:00
|
|
|
struct task_desc **tasks;
|
|
|
|
const struct trace_sched_handler *tp_handler;
|
|
|
|
pthread_mutex_t start_work_mutex;
|
|
|
|
pthread_mutex_t work_done_wait_mutex;
|
|
|
|
int profile_cpu;
|
|
|
|
/*
|
|
|
|
* Track the current task - that way we can know whether there's any
|
|
|
|
* weird events, such as a task being switched away that is not current.
|
|
|
|
*/
|
|
|
|
int max_cpu;
|
|
|
|
u32 curr_pid[MAX_CPUS];
|
|
|
|
struct thread *curr_thread[MAX_CPUS];
|
|
|
|
char next_shortname1;
|
|
|
|
char next_shortname2;
|
|
|
|
unsigned int replay_repeat;
|
|
|
|
unsigned long nr_run_events;
|
|
|
|
unsigned long nr_sleep_events;
|
|
|
|
unsigned long nr_wakeup_events;
|
|
|
|
unsigned long nr_sleep_corrections;
|
|
|
|
unsigned long nr_run_events_optimized;
|
|
|
|
unsigned long targetless_wakeups;
|
|
|
|
unsigned long multitarget_wakeups;
|
|
|
|
unsigned long nr_runs;
|
|
|
|
unsigned long nr_timestamps;
|
|
|
|
unsigned long nr_unordered_timestamps;
|
|
|
|
unsigned long nr_context_switch_bugs;
|
|
|
|
unsigned long nr_events;
|
|
|
|
unsigned long nr_lost_chunks;
|
|
|
|
unsigned long nr_lost_events;
|
|
|
|
u64 run_measurement_overhead;
|
|
|
|
u64 sleep_measurement_overhead;
|
|
|
|
u64 start_time;
|
|
|
|
u64 cpu_usage;
|
|
|
|
u64 runavg_cpu_usage;
|
|
|
|
u64 parent_cpu_usage;
|
|
|
|
u64 runavg_parent_cpu_usage;
|
|
|
|
u64 sum_runtime;
|
|
|
|
u64 sum_fluct;
|
|
|
|
u64 run_avg;
|
|
|
|
u64 all_runtime;
|
|
|
|
u64 all_count;
|
|
|
|
u64 cpu_last_switched[MAX_CPUS];
|
|
|
|
struct rb_root atom_root, sorted_atom_root;
|
|
|
|
struct list_head sort_list, cmp_pid;
|
perf sched replay: Fix the EMFILE error caused by the limitation of the maximum open files
The soft maximum number of open files for a calling process is 1024,
which is defined as INR_OPEN_CUR in include/uapi/linux/fs.h, and the
hard maximum number of open files for a calling process is 4096, which
is defined as INR_OPEN_MAX in include/uapi/linux/fs.h.
Both INR_OPEN_CUR and INR_OPEN_MAX are used to limit the value of
RLIMIT_NOFILE in include/asm-generic/resource.h.
And the soft maximum number finally decides the limitation of the
maximum files which are allowed to be opened.
That is to say a process can use at most 1024 file descriptors for its
o pened files, or an EMFILE error will happen.
This error can be fixed by increasing the soft maximum number, under the
constraint that the soft maximum number can not exceed the hard maximum
number, or both soft and hard maximum number should be increased
simultaneously with privilege.
For perf sched replay, it uses sys_perf_event_open to create the file
descriptor for each of the tasks in order to handle information of perf
events.
That is to say each task needs a unique file descriptor. In x86_64,
there may be over 1024 or 4096 tasks correspoinding to the record in
perf.data, which causes that no enough file descriptors can be used.
As a result, EMFILE error happens and stops the replay process. To solve
this problem, we adaptively increase the soft and hard maximum number of
open files with a '-f' option.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ cat /proc/sys/fs/file-max
6815744
$ ulimit -Sn
1024
$ ulimit -Hn
4096
Before this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
After this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
Have a try with -f option
$ perf sched replay -f
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
------------------------------------------------------------
#1 : 54.401, ravg: 54.40, cpu: 3285.21 / 3285.21
#2 : 199.548, ravg: 68.92, cpu: 4999.65 / 3456.66
#3 : 170.483, ravg: 79.07, cpu: 1349.94 / 3245.99
#4 : 192.034, ravg: 90.37, cpu: 1322.88 / 3053.67
#5 : 182.929, ravg: 99.62, cpu: 1406.51 / 2888.96
#6 : 152.974, ravg: 104.96, cpu: 1167.54 / 2716.82
#7 : 155.579, ravg: 110.02, cpu: 2992.53 / 2744.39
#8 : 130.557, ravg: 112.08, cpu: 1126.43 / 2582.59
#9 : 138.520, ravg: 114.72, cpu: 1253.22 / 2449.65
#10 : 134.328, ravg: 116.68, cpu: 1587.95 / 2363.48
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-8-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:34 +00:00
|
|
|
bool force;
|
2012-09-11 20:29:27 +00:00
|
|
|
};
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
static u64 get_nsecs(void)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
|
|
|
struct timespec ts;
|
|
|
|
|
|
|
|
clock_gettime(CLOCK_MONOTONIC, &ts);
|
|
|
|
|
|
|
|
return ts.tv_sec * 1000000000ULL + ts.tv_nsec;
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void burn_nsecs(struct perf_sched *sched, u64 nsecs)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2009-09-11 10:12:54 +00:00
|
|
|
u64 T0 = get_nsecs(), T1;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
do {
|
|
|
|
T1 = get_nsecs();
|
2012-09-11 20:29:27 +00:00
|
|
|
} while (T1 + sched->run_measurement_overhead < T0 + nsecs);
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
static void sleep_nsecs(u64 nsecs)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
|
|
|
struct timespec ts;
|
|
|
|
|
|
|
|
ts.tv_nsec = nsecs % 999999999;
|
|
|
|
ts.tv_sec = nsecs / 999999999;
|
|
|
|
|
|
|
|
nanosleep(&ts, NULL);
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void calibrate_run_measurement_overhead(struct perf_sched *sched)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2009-09-11 10:12:54 +00:00
|
|
|
u64 T0, T1, delta, min_delta = 1000000000ULL;
|
2009-09-11 10:12:54 +00:00
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < 10; i++) {
|
|
|
|
T0 = get_nsecs();
|
2012-09-11 20:29:27 +00:00
|
|
|
burn_nsecs(sched, 0);
|
2009-09-11 10:12:54 +00:00
|
|
|
T1 = get_nsecs();
|
|
|
|
delta = T1-T0;
|
|
|
|
min_delta = min(min_delta, delta);
|
|
|
|
}
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->run_measurement_overhead = min_delta;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2011-01-22 22:37:02 +00:00
|
|
|
printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta);
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void calibrate_sleep_measurement_overhead(struct perf_sched *sched)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2009-09-11 10:12:54 +00:00
|
|
|
u64 T0, T1, delta, min_delta = 1000000000ULL;
|
2009-09-11 10:12:54 +00:00
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < 10; i++) {
|
|
|
|
T0 = get_nsecs();
|
|
|
|
sleep_nsecs(10000);
|
|
|
|
T1 = get_nsecs();
|
|
|
|
delta = T1-T0;
|
|
|
|
min_delta = min(min_delta, delta);
|
|
|
|
}
|
|
|
|
min_delta -= 10000;
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->sleep_measurement_overhead = min_delta;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2011-01-22 22:37:02 +00:00
|
|
|
printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta);
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
static struct sched_atom *
|
2009-09-11 10:12:54 +00:00
|
|
|
get_new_event(struct task_desc *task, u64 timestamp)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2009-11-24 14:05:16 +00:00
|
|
|
struct sched_atom *event = zalloc(sizeof(*event));
|
2009-09-11 10:12:54 +00:00
|
|
|
unsigned long idx = task->nr_events;
|
|
|
|
size_t size;
|
|
|
|
|
|
|
|
event->timestamp = timestamp;
|
|
|
|
event->nr = idx;
|
|
|
|
|
|
|
|
task->nr_events++;
|
2009-09-14 18:04:48 +00:00
|
|
|
size = sizeof(struct sched_atom *) * task->nr_events;
|
|
|
|
task->atoms = realloc(task->atoms, size);
|
|
|
|
BUG_ON(!task->atoms);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
task->atoms[idx] = event;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
return event;
|
|
|
|
}
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
static struct sched_atom *last_event(struct task_desc *task)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
|
|
|
if (!task->nr_events)
|
|
|
|
return NULL;
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
return task->atoms[task->nr_events - 1];
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task,
|
|
|
|
u64 timestamp, u64 duration)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2009-09-14 18:04:48 +00:00
|
|
|
struct sched_atom *event, *curr_event = last_event(task);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
/*
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
* optimize an existing RUN event by merging this one
|
|
|
|
* to it:
|
|
|
|
*/
|
2009-09-11 10:12:54 +00:00
|
|
|
if (curr_event && curr_event->type == SCHED_EVENT_RUN) {
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->nr_run_events_optimized++;
|
2009-09-11 10:12:54 +00:00
|
|
|
curr_event->duration += duration;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
event = get_new_event(task, timestamp);
|
|
|
|
|
|
|
|
event->type = SCHED_EVENT_RUN;
|
|
|
|
event->duration = duration;
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->nr_run_events++;
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task,
|
|
|
|
u64 timestamp, struct task_desc *wakee)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2009-09-14 18:04:48 +00:00
|
|
|
struct sched_atom *event, *wakee_event;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
event = get_new_event(task, timestamp);
|
|
|
|
event->type = SCHED_EVENT_WAKEUP;
|
|
|
|
event->wakee = wakee;
|
|
|
|
|
|
|
|
wakee_event = last_event(wakee);
|
|
|
|
if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) {
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->targetless_wakeups++;
|
2009-09-11 10:12:54 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
if (wakee_event->wait_sem) {
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->multitarget_wakeups++;
|
2009-09-11 10:12:54 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2009-11-24 14:05:16 +00:00
|
|
|
wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem));
|
2009-09-11 10:12:54 +00:00
|
|
|
sem_init(wakee_event->wait_sem, 0, 0);
|
|
|
|
wakee_event->specific_wait = 1;
|
|
|
|
event->wait_sem = wakee_event->wait_sem;
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->nr_wakeup_events++;
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task,
|
|
|
|
u64 timestamp, u64 task_state __maybe_unused)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2009-09-14 18:04:48 +00:00
|
|
|
struct sched_atom *event = get_new_event(task, timestamp);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
event->type = SCHED_EVENT_SLEEP;
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->nr_sleep_events++;
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static struct task_desc *register_pid(struct perf_sched *sched,
|
|
|
|
unsigned long pid, const char *comm)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
|
|
|
struct task_desc *task;
|
perf sched replay: Alloc the memory of pid_to_task dynamically to adapt to the unexpected change of pid_max
The current memory allocation of struct task_desc *pid_to_task[MAX_PID]
is in a permanent and preset way, and it has two problems:
Problem 1: If the pid_max, which is the max number of pids in the
system, is much smaller than MAX_PID (1024*1000), then it causes a waste
of stack memory. This may happen in the case where the number of cpu
cores is much smaller than 1000.
Problem 2: If the pid_max is changed from the default value to a value
larger than MAX_PID, then it will cause assertion failure problem. The
maximum value of pid_max can be set to pid_max_max (see pidmap_init
defined in kernel/pid.c), which equals to PID_MAX_LIMIT. In x86_64,
PID_MAX_LIMIT is 4*1024*1024 (defined in include/linux/threads.h). This
value is much larger than MAX_PID, and will take up 32768 Kbytes
(4*1024*1024*8/1024) for memory allocation of pid_to_task, which is much
larger than the default 8192 Kbytes of the stack size of calling
process.
Due to these two problems, we use calloc to allocate the memory of
pid_to_task dynamically.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ echo 1025000 > /proc/sys/kernel/pid_max
$ cat /proc/sys/kernel/pid_max
1025000
Run some applications until the pid of some process is greater than
the value of MAX_PID (1024*1000).
Before this patch:
$ perf sched replay
run measurement overhead: 221 nsecs
sleep measurement overhead: 55480 nsecs
the run test took 1000008 nsecs
the sleep test took 1063151 nsecs
perf: builtin-sched.c:330: register_pid: Assertion `!(pid >= 1024000)'
failed.
Aborted
After this patch:
$ perf sched replay
run measurement overhead: 221 nsecs
sleep measurement overhead: 55435 nsecs
the run test took 1000004 nsecs
the sleep test took 1059312 nsecs
nr_run_events: 10
nr_sleep_events: 1562
nr_wakeup_events: 5
task 0 ( :1: 1), nr_events: 1
task 1 ( :2: 2), nr_events: 1
task 2 ( :3: 3), nr_events: 1
task 3 ( :5: 5), nr_events: 1
...
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-4-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:30 +00:00
|
|
|
static int pid_max;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
perf sched replay: Alloc the memory of pid_to_task dynamically to adapt to the unexpected change of pid_max
The current memory allocation of struct task_desc *pid_to_task[MAX_PID]
is in a permanent and preset way, and it has two problems:
Problem 1: If the pid_max, which is the max number of pids in the
system, is much smaller than MAX_PID (1024*1000), then it causes a waste
of stack memory. This may happen in the case where the number of cpu
cores is much smaller than 1000.
Problem 2: If the pid_max is changed from the default value to a value
larger than MAX_PID, then it will cause assertion failure problem. The
maximum value of pid_max can be set to pid_max_max (see pidmap_init
defined in kernel/pid.c), which equals to PID_MAX_LIMIT. In x86_64,
PID_MAX_LIMIT is 4*1024*1024 (defined in include/linux/threads.h). This
value is much larger than MAX_PID, and will take up 32768 Kbytes
(4*1024*1024*8/1024) for memory allocation of pid_to_task, which is much
larger than the default 8192 Kbytes of the stack size of calling
process.
Due to these two problems, we use calloc to allocate the memory of
pid_to_task dynamically.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ echo 1025000 > /proc/sys/kernel/pid_max
$ cat /proc/sys/kernel/pid_max
1025000
Run some applications until the pid of some process is greater than
the value of MAX_PID (1024*1000).
Before this patch:
$ perf sched replay
run measurement overhead: 221 nsecs
sleep measurement overhead: 55480 nsecs
the run test took 1000008 nsecs
the sleep test took 1063151 nsecs
perf: builtin-sched.c:330: register_pid: Assertion `!(pid >= 1024000)'
failed.
Aborted
After this patch:
$ perf sched replay
run measurement overhead: 221 nsecs
sleep measurement overhead: 55435 nsecs
the run test took 1000004 nsecs
the sleep test took 1059312 nsecs
nr_run_events: 10
nr_sleep_events: 1562
nr_wakeup_events: 5
task 0 ( :1: 1), nr_events: 1
task 1 ( :2: 2), nr_events: 1
task 2 ( :3: 3), nr_events: 1
task 3 ( :5: 5), nr_events: 1
...
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-4-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:30 +00:00
|
|
|
if (sched->pid_to_task == NULL) {
|
|
|
|
if (sysctl__read_int("kernel/pid_max", &pid_max) < 0)
|
|
|
|
pid_max = MAX_PID;
|
|
|
|
BUG_ON((sched->pid_to_task = calloc(pid_max, sizeof(struct task_desc *))) == NULL);
|
|
|
|
}
|
2015-03-31 13:46:31 +00:00
|
|
|
if (pid >= (unsigned long)pid_max) {
|
|
|
|
BUG_ON((sched->pid_to_task = realloc(sched->pid_to_task, (pid + 1) *
|
|
|
|
sizeof(struct task_desc *))) == NULL);
|
|
|
|
while (pid >= (unsigned long)pid_max)
|
|
|
|
sched->pid_to_task[pid_max++] = NULL;
|
|
|
|
}
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
task = sched->pid_to_task[pid];
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
if (task)
|
|
|
|
return task;
|
|
|
|
|
2009-11-24 14:05:16 +00:00
|
|
|
task = zalloc(sizeof(*task));
|
2009-09-11 10:12:54 +00:00
|
|
|
task->pid = pid;
|
2012-09-11 20:29:27 +00:00
|
|
|
task->nr = sched->nr_tasks;
|
2009-09-11 10:12:54 +00:00
|
|
|
strcpy(task->comm, comm);
|
|
|
|
/*
|
|
|
|
* every task starts in sleeping state - this gets ignored
|
|
|
|
* if there's no wakeup pointing to this sleep state:
|
|
|
|
*/
|
2012-09-11 20:29:27 +00:00
|
|
|
add_sched_event_sleep(sched, task, 0, 0);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->pid_to_task[pid] = task;
|
|
|
|
sched->nr_tasks++;
|
2015-03-31 13:46:28 +00:00
|
|
|
sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_desc *));
|
2012-09-11 20:29:27 +00:00
|
|
|
BUG_ON(!sched->tasks);
|
|
|
|
sched->tasks[task->nr] = task;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
if (verbose)
|
2012-09-11 20:29:27 +00:00
|
|
|
printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
return task;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void print_task_traces(struct perf_sched *sched)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
|
|
|
struct task_desc *task;
|
|
|
|
unsigned long i;
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
for (i = 0; i < sched->nr_tasks; i++) {
|
|
|
|
task = sched->tasks[i];
|
2009-09-11 10:12:54 +00:00
|
|
|
printf("task %6ld (%20s:%10ld), nr_events: %ld\n",
|
2009-09-11 10:12:54 +00:00
|
|
|
task->nr, task->comm, task->pid, task->nr_events);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void add_cross_task_wakeups(struct perf_sched *sched)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
|
|
|
struct task_desc *task1, *task2;
|
|
|
|
unsigned long i, j;
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
for (i = 0; i < sched->nr_tasks; i++) {
|
|
|
|
task1 = sched->tasks[i];
|
2009-09-11 10:12:54 +00:00
|
|
|
j = i + 1;
|
2012-09-11 20:29:27 +00:00
|
|
|
if (j == sched->nr_tasks)
|
2009-09-11 10:12:54 +00:00
|
|
|
j = 0;
|
2012-09-11 20:29:27 +00:00
|
|
|
task2 = sched->tasks[j];
|
|
|
|
add_sched_event_wakeup(sched, task1, 0, task2);
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void perf_sched__process_event(struct perf_sched *sched,
|
|
|
|
struct sched_atom *atom)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
|
|
|
int ret = 0;
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
switch (atom->type) {
|
2009-09-11 10:12:54 +00:00
|
|
|
case SCHED_EVENT_RUN:
|
2012-09-11 20:29:27 +00:00
|
|
|
burn_nsecs(sched, atom->duration);
|
2009-09-11 10:12:54 +00:00
|
|
|
break;
|
|
|
|
case SCHED_EVENT_SLEEP:
|
2009-09-14 18:04:48 +00:00
|
|
|
if (atom->wait_sem)
|
|
|
|
ret = sem_wait(atom->wait_sem);
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(ret);
|
|
|
|
break;
|
|
|
|
case SCHED_EVENT_WAKEUP:
|
2009-09-14 18:04:48 +00:00
|
|
|
if (atom->wait_sem)
|
|
|
|
ret = sem_post(atom->wait_sem);
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(ret);
|
|
|
|
break;
|
2009-10-10 12:46:04 +00:00
|
|
|
case SCHED_EVENT_MIGRATION:
|
|
|
|
break;
|
2009-09-11 10:12:54 +00:00
|
|
|
default:
|
|
|
|
BUG_ON(1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
static u64 get_cpu_usage_nsec_parent(void)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
|
|
|
struct rusage ru;
|
2009-09-11 10:12:54 +00:00
|
|
|
u64 sum;
|
2009-09-11 10:12:54 +00:00
|
|
|
int err;
|
|
|
|
|
|
|
|
err = getrusage(RUSAGE_SELF, &ru);
|
|
|
|
BUG_ON(err);
|
|
|
|
|
|
|
|
sum = ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3;
|
|
|
|
sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3;
|
|
|
|
|
|
|
|
return sum;
|
|
|
|
}
|
|
|
|
|
perf sched replay: Fix the EMFILE error caused by the limitation of the maximum open files
The soft maximum number of open files for a calling process is 1024,
which is defined as INR_OPEN_CUR in include/uapi/linux/fs.h, and the
hard maximum number of open files for a calling process is 4096, which
is defined as INR_OPEN_MAX in include/uapi/linux/fs.h.
Both INR_OPEN_CUR and INR_OPEN_MAX are used to limit the value of
RLIMIT_NOFILE in include/asm-generic/resource.h.
And the soft maximum number finally decides the limitation of the
maximum files which are allowed to be opened.
That is to say a process can use at most 1024 file descriptors for its
o pened files, or an EMFILE error will happen.
This error can be fixed by increasing the soft maximum number, under the
constraint that the soft maximum number can not exceed the hard maximum
number, or both soft and hard maximum number should be increased
simultaneously with privilege.
For perf sched replay, it uses sys_perf_event_open to create the file
descriptor for each of the tasks in order to handle information of perf
events.
That is to say each task needs a unique file descriptor. In x86_64,
there may be over 1024 or 4096 tasks correspoinding to the record in
perf.data, which causes that no enough file descriptors can be used.
As a result, EMFILE error happens and stops the replay process. To solve
this problem, we adaptively increase the soft and hard maximum number of
open files with a '-f' option.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ cat /proc/sys/fs/file-max
6815744
$ ulimit -Sn
1024
$ ulimit -Hn
4096
Before this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
After this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
Have a try with -f option
$ perf sched replay -f
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
------------------------------------------------------------
#1 : 54.401, ravg: 54.40, cpu: 3285.21 / 3285.21
#2 : 199.548, ravg: 68.92, cpu: 4999.65 / 3456.66
#3 : 170.483, ravg: 79.07, cpu: 1349.94 / 3245.99
#4 : 192.034, ravg: 90.37, cpu: 1322.88 / 3053.67
#5 : 182.929, ravg: 99.62, cpu: 1406.51 / 2888.96
#6 : 152.974, ravg: 104.96, cpu: 1167.54 / 2716.82
#7 : 155.579, ravg: 110.02, cpu: 2992.53 / 2744.39
#8 : 130.557, ravg: 112.08, cpu: 1126.43 / 2582.59
#9 : 138.520, ravg: 114.72, cpu: 1253.22 / 2449.65
#10 : 134.328, ravg: 116.68, cpu: 1587.95 / 2363.48
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-8-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:34 +00:00
|
|
|
static int self_open_counters(struct perf_sched *sched, unsigned long cur_task)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2009-12-09 09:51:30 +00:00
|
|
|
struct perf_event_attr attr;
|
perf sched replay: Fix the EMFILE error caused by the limitation of the maximum open files
The soft maximum number of open files for a calling process is 1024,
which is defined as INR_OPEN_CUR in include/uapi/linux/fs.h, and the
hard maximum number of open files for a calling process is 4096, which
is defined as INR_OPEN_MAX in include/uapi/linux/fs.h.
Both INR_OPEN_CUR and INR_OPEN_MAX are used to limit the value of
RLIMIT_NOFILE in include/asm-generic/resource.h.
And the soft maximum number finally decides the limitation of the
maximum files which are allowed to be opened.
That is to say a process can use at most 1024 file descriptors for its
o pened files, or an EMFILE error will happen.
This error can be fixed by increasing the soft maximum number, under the
constraint that the soft maximum number can not exceed the hard maximum
number, or both soft and hard maximum number should be increased
simultaneously with privilege.
For perf sched replay, it uses sys_perf_event_open to create the file
descriptor for each of the tasks in order to handle information of perf
events.
That is to say each task needs a unique file descriptor. In x86_64,
there may be over 1024 or 4096 tasks correspoinding to the record in
perf.data, which causes that no enough file descriptors can be used.
As a result, EMFILE error happens and stops the replay process. To solve
this problem, we adaptively increase the soft and hard maximum number of
open files with a '-f' option.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ cat /proc/sys/fs/file-max
6815744
$ ulimit -Sn
1024
$ ulimit -Hn
4096
Before this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
After this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
Have a try with -f option
$ perf sched replay -f
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
------------------------------------------------------------
#1 : 54.401, ravg: 54.40, cpu: 3285.21 / 3285.21
#2 : 199.548, ravg: 68.92, cpu: 4999.65 / 3456.66
#3 : 170.483, ravg: 79.07, cpu: 1349.94 / 3245.99
#4 : 192.034, ravg: 90.37, cpu: 1322.88 / 3053.67
#5 : 182.929, ravg: 99.62, cpu: 1406.51 / 2888.96
#6 : 152.974, ravg: 104.96, cpu: 1167.54 / 2716.82
#7 : 155.579, ravg: 110.02, cpu: 2992.53 / 2744.39
#8 : 130.557, ravg: 112.08, cpu: 1126.43 / 2582.59
#9 : 138.520, ravg: 114.72, cpu: 1253.22 / 2449.65
#10 : 134.328, ravg: 116.68, cpu: 1587.95 / 2363.48
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-8-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:34 +00:00
|
|
|
char sbuf[STRERR_BUFSIZE], info[STRERR_BUFSIZE];
|
2009-12-09 09:51:30 +00:00
|
|
|
int fd;
|
perf sched replay: Fix the EMFILE error caused by the limitation of the maximum open files
The soft maximum number of open files for a calling process is 1024,
which is defined as INR_OPEN_CUR in include/uapi/linux/fs.h, and the
hard maximum number of open files for a calling process is 4096, which
is defined as INR_OPEN_MAX in include/uapi/linux/fs.h.
Both INR_OPEN_CUR and INR_OPEN_MAX are used to limit the value of
RLIMIT_NOFILE in include/asm-generic/resource.h.
And the soft maximum number finally decides the limitation of the
maximum files which are allowed to be opened.
That is to say a process can use at most 1024 file descriptors for its
o pened files, or an EMFILE error will happen.
This error can be fixed by increasing the soft maximum number, under the
constraint that the soft maximum number can not exceed the hard maximum
number, or both soft and hard maximum number should be increased
simultaneously with privilege.
For perf sched replay, it uses sys_perf_event_open to create the file
descriptor for each of the tasks in order to handle information of perf
events.
That is to say each task needs a unique file descriptor. In x86_64,
there may be over 1024 or 4096 tasks correspoinding to the record in
perf.data, which causes that no enough file descriptors can be used.
As a result, EMFILE error happens and stops the replay process. To solve
this problem, we adaptively increase the soft and hard maximum number of
open files with a '-f' option.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ cat /proc/sys/fs/file-max
6815744
$ ulimit -Sn
1024
$ ulimit -Hn
4096
Before this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
After this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
Have a try with -f option
$ perf sched replay -f
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
------------------------------------------------------------
#1 : 54.401, ravg: 54.40, cpu: 3285.21 / 3285.21
#2 : 199.548, ravg: 68.92, cpu: 4999.65 / 3456.66
#3 : 170.483, ravg: 79.07, cpu: 1349.94 / 3245.99
#4 : 192.034, ravg: 90.37, cpu: 1322.88 / 3053.67
#5 : 182.929, ravg: 99.62, cpu: 1406.51 / 2888.96
#6 : 152.974, ravg: 104.96, cpu: 1167.54 / 2716.82
#7 : 155.579, ravg: 110.02, cpu: 2992.53 / 2744.39
#8 : 130.557, ravg: 112.08, cpu: 1126.43 / 2582.59
#9 : 138.520, ravg: 114.72, cpu: 1253.22 / 2449.65
#10 : 134.328, ravg: 116.68, cpu: 1587.95 / 2363.48
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-8-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:34 +00:00
|
|
|
struct rlimit limit;
|
|
|
|
bool need_privilege = false;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-12-09 09:51:30 +00:00
|
|
|
memset(&attr, 0, sizeof(attr));
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-12-09 09:51:30 +00:00
|
|
|
attr.type = PERF_TYPE_SOFTWARE;
|
|
|
|
attr.config = PERF_COUNT_SW_TASK_CLOCK;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
perf sched replay: Fix the EMFILE error caused by the limitation of the maximum open files
The soft maximum number of open files for a calling process is 1024,
which is defined as INR_OPEN_CUR in include/uapi/linux/fs.h, and the
hard maximum number of open files for a calling process is 4096, which
is defined as INR_OPEN_MAX in include/uapi/linux/fs.h.
Both INR_OPEN_CUR and INR_OPEN_MAX are used to limit the value of
RLIMIT_NOFILE in include/asm-generic/resource.h.
And the soft maximum number finally decides the limitation of the
maximum files which are allowed to be opened.
That is to say a process can use at most 1024 file descriptors for its
o pened files, or an EMFILE error will happen.
This error can be fixed by increasing the soft maximum number, under the
constraint that the soft maximum number can not exceed the hard maximum
number, or both soft and hard maximum number should be increased
simultaneously with privilege.
For perf sched replay, it uses sys_perf_event_open to create the file
descriptor for each of the tasks in order to handle information of perf
events.
That is to say each task needs a unique file descriptor. In x86_64,
there may be over 1024 or 4096 tasks correspoinding to the record in
perf.data, which causes that no enough file descriptors can be used.
As a result, EMFILE error happens and stops the replay process. To solve
this problem, we adaptively increase the soft and hard maximum number of
open files with a '-f' option.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ cat /proc/sys/fs/file-max
6815744
$ ulimit -Sn
1024
$ ulimit -Hn
4096
Before this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
After this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
Have a try with -f option
$ perf sched replay -f
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
------------------------------------------------------------
#1 : 54.401, ravg: 54.40, cpu: 3285.21 / 3285.21
#2 : 199.548, ravg: 68.92, cpu: 4999.65 / 3456.66
#3 : 170.483, ravg: 79.07, cpu: 1349.94 / 3245.99
#4 : 192.034, ravg: 90.37, cpu: 1322.88 / 3053.67
#5 : 182.929, ravg: 99.62, cpu: 1406.51 / 2888.96
#6 : 152.974, ravg: 104.96, cpu: 1167.54 / 2716.82
#7 : 155.579, ravg: 110.02, cpu: 2992.53 / 2744.39
#8 : 130.557, ravg: 112.08, cpu: 1126.43 / 2582.59
#9 : 138.520, ravg: 114.72, cpu: 1253.22 / 2449.65
#10 : 134.328, ravg: 116.68, cpu: 1587.95 / 2363.48
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-8-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:34 +00:00
|
|
|
force_again:
|
2014-06-30 20:28:47 +00:00
|
|
|
fd = sys_perf_event_open(&attr, 0, -1, -1,
|
|
|
|
perf_event_open_cloexec_flag());
|
2009-12-09 09:51:30 +00:00
|
|
|
|
perf sched replay: Handle the dead halt of sem_wait when create_tasks() fails for any task
Since there is sem_wait for each task in the wait_for_tasks(), e.g.
sem_wait(&task->work_done_sem).
The sem_wait can continue only when work_done_sem is greater than 0, or
it will be blocked.
For perf sched replay, one task may sem_post the work_done_sem of
another task, which causes the work_done_sem of that task processed in a
reasonable sequence, e.g. sem_post, sem_wait, sem_wait, sem_post...
This sequence simulates the sched process of the running tasks at the
time when perf sched record runs.
As a result, all the tasks are required and their threads must be
successfully created.
If any one (task A) of the tasks fails to create its thread, then
another task (task B), whose work_done_sem needs sem_post from that
failed task A, may likely block itself due to seg_wait.
And this is a dead halt, since task B's thread_func cannot continue at
all.
To solve this problem, perf sched replay should exit once any task fails
to create its thread.
Example:
Test environment: x86_64 with 160 cores
Before this patch:
$ perf sched replay
...
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
------------------------------------------------------------ <- dead halt
After this patch:
$ perf sched replay
...
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
$
As shown above, perf sched replay finishes the process after printing an
error message and does not block itself.
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-7-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:33 +00:00
|
|
|
if (fd < 0) {
|
perf sched replay: Fix the EMFILE error caused by the limitation of the maximum open files
The soft maximum number of open files for a calling process is 1024,
which is defined as INR_OPEN_CUR in include/uapi/linux/fs.h, and the
hard maximum number of open files for a calling process is 4096, which
is defined as INR_OPEN_MAX in include/uapi/linux/fs.h.
Both INR_OPEN_CUR and INR_OPEN_MAX are used to limit the value of
RLIMIT_NOFILE in include/asm-generic/resource.h.
And the soft maximum number finally decides the limitation of the
maximum files which are allowed to be opened.
That is to say a process can use at most 1024 file descriptors for its
o pened files, or an EMFILE error will happen.
This error can be fixed by increasing the soft maximum number, under the
constraint that the soft maximum number can not exceed the hard maximum
number, or both soft and hard maximum number should be increased
simultaneously with privilege.
For perf sched replay, it uses sys_perf_event_open to create the file
descriptor for each of the tasks in order to handle information of perf
events.
That is to say each task needs a unique file descriptor. In x86_64,
there may be over 1024 or 4096 tasks correspoinding to the record in
perf.data, which causes that no enough file descriptors can be used.
As a result, EMFILE error happens and stops the replay process. To solve
this problem, we adaptively increase the soft and hard maximum number of
open files with a '-f' option.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ cat /proc/sys/fs/file-max
6815744
$ ulimit -Sn
1024
$ ulimit -Hn
4096
Before this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
After this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
Have a try with -f option
$ perf sched replay -f
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
------------------------------------------------------------
#1 : 54.401, ravg: 54.40, cpu: 3285.21 / 3285.21
#2 : 199.548, ravg: 68.92, cpu: 4999.65 / 3456.66
#3 : 170.483, ravg: 79.07, cpu: 1349.94 / 3245.99
#4 : 192.034, ravg: 90.37, cpu: 1322.88 / 3053.67
#5 : 182.929, ravg: 99.62, cpu: 1406.51 / 2888.96
#6 : 152.974, ravg: 104.96, cpu: 1167.54 / 2716.82
#7 : 155.579, ravg: 110.02, cpu: 2992.53 / 2744.39
#8 : 130.557, ravg: 112.08, cpu: 1126.43 / 2582.59
#9 : 138.520, ravg: 114.72, cpu: 1253.22 / 2449.65
#10 : 134.328, ravg: 116.68, cpu: 1587.95 / 2363.48
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-8-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:34 +00:00
|
|
|
if (errno == EMFILE) {
|
|
|
|
if (sched->force) {
|
|
|
|
BUG_ON(getrlimit(RLIMIT_NOFILE, &limit) == -1);
|
|
|
|
limit.rlim_cur += sched->nr_tasks - cur_task;
|
|
|
|
if (limit.rlim_cur > limit.rlim_max) {
|
|
|
|
limit.rlim_max = limit.rlim_cur;
|
|
|
|
need_privilege = true;
|
|
|
|
}
|
|
|
|
if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
|
|
|
|
if (need_privilege && errno == EPERM)
|
|
|
|
strcpy(info, "Need privilege\n");
|
|
|
|
} else
|
|
|
|
goto force_again;
|
|
|
|
} else
|
|
|
|
strcpy(info, "Have a try with -f option\n");
|
|
|
|
}
|
2012-09-12 02:11:06 +00:00
|
|
|
pr_err("Error: sys_perf_event_open() syscall returned "
|
perf sched replay: Fix the EMFILE error caused by the limitation of the maximum open files
The soft maximum number of open files for a calling process is 1024,
which is defined as INR_OPEN_CUR in include/uapi/linux/fs.h, and the
hard maximum number of open files for a calling process is 4096, which
is defined as INR_OPEN_MAX in include/uapi/linux/fs.h.
Both INR_OPEN_CUR and INR_OPEN_MAX are used to limit the value of
RLIMIT_NOFILE in include/asm-generic/resource.h.
And the soft maximum number finally decides the limitation of the
maximum files which are allowed to be opened.
That is to say a process can use at most 1024 file descriptors for its
o pened files, or an EMFILE error will happen.
This error can be fixed by increasing the soft maximum number, under the
constraint that the soft maximum number can not exceed the hard maximum
number, or both soft and hard maximum number should be increased
simultaneously with privilege.
For perf sched replay, it uses sys_perf_event_open to create the file
descriptor for each of the tasks in order to handle information of perf
events.
That is to say each task needs a unique file descriptor. In x86_64,
there may be over 1024 or 4096 tasks correspoinding to the record in
perf.data, which causes that no enough file descriptors can be used.
As a result, EMFILE error happens and stops the replay process. To solve
this problem, we adaptively increase the soft and hard maximum number of
open files with a '-f' option.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ cat /proc/sys/fs/file-max
6815744
$ ulimit -Sn
1024
$ ulimit -Hn
4096
Before this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
After this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
Have a try with -f option
$ perf sched replay -f
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
------------------------------------------------------------
#1 : 54.401, ravg: 54.40, cpu: 3285.21 / 3285.21
#2 : 199.548, ravg: 68.92, cpu: 4999.65 / 3456.66
#3 : 170.483, ravg: 79.07, cpu: 1349.94 / 3245.99
#4 : 192.034, ravg: 90.37, cpu: 1322.88 / 3053.67
#5 : 182.929, ravg: 99.62, cpu: 1406.51 / 2888.96
#6 : 152.974, ravg: 104.96, cpu: 1167.54 / 2716.82
#7 : 155.579, ravg: 110.02, cpu: 2992.53 / 2744.39
#8 : 130.557, ravg: 112.08, cpu: 1126.43 / 2582.59
#9 : 138.520, ravg: 114.72, cpu: 1253.22 / 2449.65
#10 : 134.328, ravg: 116.68, cpu: 1587.95 / 2363.48
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-8-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:34 +00:00
|
|
|
"with %d (%s)\n%s", fd,
|
|
|
|
strerror_r(errno, sbuf, sizeof(sbuf)), info);
|
perf sched replay: Handle the dead halt of sem_wait when create_tasks() fails for any task
Since there is sem_wait for each task in the wait_for_tasks(), e.g.
sem_wait(&task->work_done_sem).
The sem_wait can continue only when work_done_sem is greater than 0, or
it will be blocked.
For perf sched replay, one task may sem_post the work_done_sem of
another task, which causes the work_done_sem of that task processed in a
reasonable sequence, e.g. sem_post, sem_wait, sem_wait, sem_post...
This sequence simulates the sched process of the running tasks at the
time when perf sched record runs.
As a result, all the tasks are required and their threads must be
successfully created.
If any one (task A) of the tasks fails to create its thread, then
another task (task B), whose work_done_sem needs sem_post from that
failed task A, may likely block itself due to seg_wait.
And this is a dead halt, since task B's thread_func cannot continue at
all.
To solve this problem, perf sched replay should exit once any task fails
to create its thread.
Example:
Test environment: x86_64 with 160 cores
Before this patch:
$ perf sched replay
...
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
------------------------------------------------------------ <- dead halt
After this patch:
$ perf sched replay
...
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
$
As shown above, perf sched replay finishes the process after printing an
error message and does not block itself.
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-7-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:33 +00:00
|
|
|
exit(EXIT_FAILURE);
|
|
|
|
}
|
2009-12-09 09:51:30 +00:00
|
|
|
return fd;
|
|
|
|
}
|
|
|
|
|
|
|
|
static u64 get_cpu_usage_nsec_self(int fd)
|
|
|
|
{
|
|
|
|
u64 runtime;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = read(fd, &runtime, sizeof(runtime));
|
|
|
|
BUG_ON(ret != sizeof(runtime));
|
|
|
|
|
|
|
|
return runtime;
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
struct sched_thread_parms {
|
|
|
|
struct task_desc *task;
|
|
|
|
struct perf_sched *sched;
|
perf sched replay: Fix the segmentation fault problem caused by pr_err in threads
The pr_err in self_open_counters() prints error message to stderr.
Unlike stdout, stderr uses memory buffer on the stack of each calling
process.
The pr_err in self_open_counters() works in a thread called thread_func
created in function create_tasks, which concurrently creates
sched->nr_tasks threads.
If the error happens and pr_err prints the error message in each of
these threads, the stack size of the perf process (default is 8192
kbytes) will quickly run out and the segmentation fault will happen
then.
To solve this problem, pr_err with self_open_counters() should be moved
from newly created threads to the old main thread of the perf process.
Then the pr_err can work in a stable situation without the strange
segmentation fault problem.
Example:
Test environment: x86_64 with 160 cores
Before this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Segmentation fault
After this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
...
As shown above, the result continues without any segmentation fault.
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-6-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:32 +00:00
|
|
|
int fd;
|
2012-09-11 20:29:27 +00:00
|
|
|
};
|
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
static void *thread_func(void *ctx)
|
|
|
|
{
|
2012-09-11 20:29:27 +00:00
|
|
|
struct sched_thread_parms *parms = ctx;
|
|
|
|
struct task_desc *this_task = parms->task;
|
|
|
|
struct perf_sched *sched = parms->sched;
|
2009-09-11 10:12:54 +00:00
|
|
|
u64 cpu_usage_0, cpu_usage_1;
|
2009-09-11 10:12:54 +00:00
|
|
|
unsigned long i, ret;
|
|
|
|
char comm2[22];
|
perf sched replay: Fix the segmentation fault problem caused by pr_err in threads
The pr_err in self_open_counters() prints error message to stderr.
Unlike stdout, stderr uses memory buffer on the stack of each calling
process.
The pr_err in self_open_counters() works in a thread called thread_func
created in function create_tasks, which concurrently creates
sched->nr_tasks threads.
If the error happens and pr_err prints the error message in each of
these threads, the stack size of the perf process (default is 8192
kbytes) will quickly run out and the segmentation fault will happen
then.
To solve this problem, pr_err with self_open_counters() should be moved
from newly created threads to the old main thread of the perf process.
Then the pr_err can work in a stable situation without the strange
segmentation fault problem.
Example:
Test environment: x86_64 with 160 cores
Before this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Segmentation fault
After this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
...
As shown above, the result continues without any segmentation fault.
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-6-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:32 +00:00
|
|
|
int fd = parms->fd;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2013-12-27 19:55:14 +00:00
|
|
|
zfree(&parms);
|
2012-09-11 20:29:27 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
sprintf(comm2, ":%s", this_task->comm);
|
|
|
|
prctl(PR_SET_NAME, comm2);
|
2012-09-09 01:53:06 +00:00
|
|
|
if (fd < 0)
|
|
|
|
return NULL;
|
2009-09-11 10:12:54 +00:00
|
|
|
again:
|
|
|
|
ret = sem_post(&this_task->ready_for_work);
|
|
|
|
BUG_ON(ret);
|
2012-09-11 20:29:27 +00:00
|
|
|
ret = pthread_mutex_lock(&sched->start_work_mutex);
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(ret);
|
2012-09-11 20:29:27 +00:00
|
|
|
ret = pthread_mutex_unlock(&sched->start_work_mutex);
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(ret);
|
|
|
|
|
2009-12-09 09:51:30 +00:00
|
|
|
cpu_usage_0 = get_cpu_usage_nsec_self(fd);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
for (i = 0; i < this_task->nr_events; i++) {
|
|
|
|
this_task->curr_event = i;
|
2012-09-11 20:29:27 +00:00
|
|
|
perf_sched__process_event(sched, this_task->atoms[i]);
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2009-12-09 09:51:30 +00:00
|
|
|
cpu_usage_1 = get_cpu_usage_nsec_self(fd);
|
2009-09-11 10:12:54 +00:00
|
|
|
this_task->cpu_usage = cpu_usage_1 - cpu_usage_0;
|
|
|
|
ret = sem_post(&this_task->work_done_sem);
|
|
|
|
BUG_ON(ret);
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
ret = pthread_mutex_lock(&sched->work_done_wait_mutex);
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(ret);
|
2012-09-11 20:29:27 +00:00
|
|
|
ret = pthread_mutex_unlock(&sched->work_done_wait_mutex);
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(ret);
|
|
|
|
|
|
|
|
goto again;
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void create_tasks(struct perf_sched *sched)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
|
|
|
struct task_desc *task;
|
|
|
|
pthread_attr_t attr;
|
|
|
|
unsigned long i;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
err = pthread_attr_init(&attr);
|
|
|
|
BUG_ON(err);
|
2011-01-10 16:14:23 +00:00
|
|
|
err = pthread_attr_setstacksize(&attr,
|
|
|
|
(size_t) max(16 * 1024, PTHREAD_STACK_MIN));
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(err);
|
2012-09-11 20:29:27 +00:00
|
|
|
err = pthread_mutex_lock(&sched->start_work_mutex);
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(err);
|
2012-09-11 20:29:27 +00:00
|
|
|
err = pthread_mutex_lock(&sched->work_done_wait_mutex);
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(err);
|
2012-09-11 20:29:27 +00:00
|
|
|
for (i = 0; i < sched->nr_tasks; i++) {
|
|
|
|
struct sched_thread_parms *parms = malloc(sizeof(*parms));
|
|
|
|
BUG_ON(parms == NULL);
|
|
|
|
parms->task = task = sched->tasks[i];
|
|
|
|
parms->sched = sched;
|
perf sched replay: Fix the EMFILE error caused by the limitation of the maximum open files
The soft maximum number of open files for a calling process is 1024,
which is defined as INR_OPEN_CUR in include/uapi/linux/fs.h, and the
hard maximum number of open files for a calling process is 4096, which
is defined as INR_OPEN_MAX in include/uapi/linux/fs.h.
Both INR_OPEN_CUR and INR_OPEN_MAX are used to limit the value of
RLIMIT_NOFILE in include/asm-generic/resource.h.
And the soft maximum number finally decides the limitation of the
maximum files which are allowed to be opened.
That is to say a process can use at most 1024 file descriptors for its
o pened files, or an EMFILE error will happen.
This error can be fixed by increasing the soft maximum number, under the
constraint that the soft maximum number can not exceed the hard maximum
number, or both soft and hard maximum number should be increased
simultaneously with privilege.
For perf sched replay, it uses sys_perf_event_open to create the file
descriptor for each of the tasks in order to handle information of perf
events.
That is to say each task needs a unique file descriptor. In x86_64,
there may be over 1024 or 4096 tasks correspoinding to the record in
perf.data, which causes that no enough file descriptors can be used.
As a result, EMFILE error happens and stops the replay process. To solve
this problem, we adaptively increase the soft and hard maximum number of
open files with a '-f' option.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ cat /proc/sys/fs/file-max
6815744
$ ulimit -Sn
1024
$ ulimit -Hn
4096
Before this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
After this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
Have a try with -f option
$ perf sched replay -f
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
------------------------------------------------------------
#1 : 54.401, ravg: 54.40, cpu: 3285.21 / 3285.21
#2 : 199.548, ravg: 68.92, cpu: 4999.65 / 3456.66
#3 : 170.483, ravg: 79.07, cpu: 1349.94 / 3245.99
#4 : 192.034, ravg: 90.37, cpu: 1322.88 / 3053.67
#5 : 182.929, ravg: 99.62, cpu: 1406.51 / 2888.96
#6 : 152.974, ravg: 104.96, cpu: 1167.54 / 2716.82
#7 : 155.579, ravg: 110.02, cpu: 2992.53 / 2744.39
#8 : 130.557, ravg: 112.08, cpu: 1126.43 / 2582.59
#9 : 138.520, ravg: 114.72, cpu: 1253.22 / 2449.65
#10 : 134.328, ravg: 116.68, cpu: 1587.95 / 2363.48
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-8-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:34 +00:00
|
|
|
parms->fd = self_open_counters(sched, i);
|
2009-09-11 10:12:54 +00:00
|
|
|
sem_init(&task->sleep_sem, 0, 0);
|
|
|
|
sem_init(&task->ready_for_work, 0, 0);
|
|
|
|
sem_init(&task->work_done_sem, 0, 0);
|
|
|
|
task->curr_event = 0;
|
2012-09-11 20:29:27 +00:00
|
|
|
err = pthread_create(&task->thread, &attr, thread_func, parms);
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(err);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void wait_for_tasks(struct perf_sched *sched)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2009-09-11 10:12:54 +00:00
|
|
|
u64 cpu_usage_0, cpu_usage_1;
|
2009-09-11 10:12:54 +00:00
|
|
|
struct task_desc *task;
|
|
|
|
unsigned long i, ret;
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->start_time = get_nsecs();
|
|
|
|
sched->cpu_usage = 0;
|
|
|
|
pthread_mutex_unlock(&sched->work_done_wait_mutex);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
for (i = 0; i < sched->nr_tasks; i++) {
|
|
|
|
task = sched->tasks[i];
|
2009-09-11 10:12:54 +00:00
|
|
|
ret = sem_wait(&task->ready_for_work);
|
|
|
|
BUG_ON(ret);
|
|
|
|
sem_init(&task->ready_for_work, 0, 0);
|
|
|
|
}
|
2012-09-11 20:29:27 +00:00
|
|
|
ret = pthread_mutex_lock(&sched->work_done_wait_mutex);
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(ret);
|
|
|
|
|
|
|
|
cpu_usage_0 = get_cpu_usage_nsec_parent();
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
pthread_mutex_unlock(&sched->start_work_mutex);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
for (i = 0; i < sched->nr_tasks; i++) {
|
|
|
|
task = sched->tasks[i];
|
2009-09-11 10:12:54 +00:00
|
|
|
ret = sem_wait(&task->work_done_sem);
|
|
|
|
BUG_ON(ret);
|
|
|
|
sem_init(&task->work_done_sem, 0, 0);
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->cpu_usage += task->cpu_usage;
|
2009-09-11 10:12:54 +00:00
|
|
|
task->cpu_usage = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
cpu_usage_1 = get_cpu_usage_nsec_parent();
|
2012-09-11 20:29:27 +00:00
|
|
|
if (!sched->runavg_cpu_usage)
|
|
|
|
sched->runavg_cpu_usage = sched->cpu_usage;
|
2015-03-31 13:46:36 +00:00
|
|
|
sched->runavg_cpu_usage = (sched->runavg_cpu_usage * (sched->replay_repeat - 1) + sched->cpu_usage) / sched->replay_repeat;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0;
|
|
|
|
if (!sched->runavg_parent_cpu_usage)
|
|
|
|
sched->runavg_parent_cpu_usage = sched->parent_cpu_usage;
|
2015-03-31 13:46:36 +00:00
|
|
|
sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * (sched->replay_repeat - 1) +
|
|
|
|
sched->parent_cpu_usage)/sched->replay_repeat;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
ret = pthread_mutex_lock(&sched->start_work_mutex);
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(ret);
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
for (i = 0; i < sched->nr_tasks; i++) {
|
|
|
|
task = sched->tasks[i];
|
2009-09-11 10:12:54 +00:00
|
|
|
sem_init(&task->sleep_sem, 0, 0);
|
|
|
|
task->curr_event = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void run_one_test(struct perf_sched *sched)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2011-01-24 16:13:04 +00:00
|
|
|
u64 T0, T1, delta, avg_delta, fluct;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
T0 = get_nsecs();
|
2012-09-11 20:29:27 +00:00
|
|
|
wait_for_tasks(sched);
|
2009-09-11 10:12:54 +00:00
|
|
|
T1 = get_nsecs();
|
|
|
|
|
|
|
|
delta = T1 - T0;
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->sum_runtime += delta;
|
|
|
|
sched->nr_runs++;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
avg_delta = sched->sum_runtime / sched->nr_runs;
|
2009-09-11 10:12:54 +00:00
|
|
|
if (delta < avg_delta)
|
|
|
|
fluct = avg_delta - delta;
|
|
|
|
else
|
|
|
|
fluct = delta - avg_delta;
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->sum_fluct += fluct;
|
|
|
|
if (!sched->run_avg)
|
|
|
|
sched->run_avg = delta;
|
2015-03-31 13:46:36 +00:00
|
|
|
sched->run_avg = (sched->run_avg * (sched->replay_repeat - 1) + delta) / sched->replay_repeat;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / 1000000.0);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
printf("ravg: %0.2f, ", (double)sched->run_avg / 1e6);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
printf("cpu: %0.2f / %0.2f",
|
2012-09-11 20:29:27 +00:00
|
|
|
(double)sched->cpu_usage / 1e6, (double)sched->runavg_cpu_usage / 1e6);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
#if 0
|
|
|
|
/*
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
* rusage statistics done by the parent, these are less
|
2012-09-11 20:29:27 +00:00
|
|
|
* accurate than the sched->sum_exec_runtime based statistics:
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
*/
|
2009-09-11 10:12:54 +00:00
|
|
|
printf(" [%0.2f / %0.2f]",
|
2012-09-11 20:29:27 +00:00
|
|
|
(double)sched->parent_cpu_usage/1e6,
|
|
|
|
(double)sched->runavg_parent_cpu_usage/1e6);
|
2009-09-11 10:12:54 +00:00
|
|
|
#endif
|
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
printf("\n");
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
if (sched->nr_sleep_corrections)
|
|
|
|
printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections);
|
|
|
|
sched->nr_sleep_corrections = 0;
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void test_calibrations(struct perf_sched *sched)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2009-09-11 10:12:54 +00:00
|
|
|
u64 T0, T1;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
T0 = get_nsecs();
|
2012-09-11 20:29:27 +00:00
|
|
|
burn_nsecs(sched, 1e6);
|
2009-09-11 10:12:54 +00:00
|
|
|
T1 = get_nsecs();
|
|
|
|
|
2011-01-22 22:37:02 +00:00
|
|
|
printf("the run test took %" PRIu64 " nsecs\n", T1 - T0);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
T0 = get_nsecs();
|
|
|
|
sleep_nsecs(1e6);
|
|
|
|
T1 = get_nsecs();
|
|
|
|
|
2011-01-22 22:37:02 +00:00
|
|
|
printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0);
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2012-09-09 01:53:06 +00:00
|
|
|
static int
|
2012-09-11 20:29:27 +00:00
|
|
|
replay_wakeup_event(struct perf_sched *sched,
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
struct perf_evsel *evsel, struct perf_sample *sample,
|
|
|
|
struct machine *machine __maybe_unused)
|
2009-09-12 01:59:01 +00:00
|
|
|
{
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
const char *comm = perf_evsel__strval(evsel, sample, "comm");
|
|
|
|
const u32 pid = perf_evsel__intval(evsel, sample, "pid");
|
2009-09-12 01:59:01 +00:00
|
|
|
struct task_desc *waker, *wakee;
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
if (verbose) {
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
printf("sched_wakeup event %p\n", evsel);
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid);
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
waker = register_pid(sched, sample->tid, "<unknown>");
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
wakee = register_pid(sched, pid, comm);
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
add_sched_event_wakeup(sched, waker, sample->time, wakee);
|
2012-09-09 01:53:06 +00:00
|
|
|
return 0;
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
static int replay_switch_event(struct perf_sched *sched,
|
|
|
|
struct perf_evsel *evsel,
|
|
|
|
struct perf_sample *sample,
|
|
|
|
struct machine *machine __maybe_unused)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
const char *prev_comm = perf_evsel__strval(evsel, sample, "prev_comm"),
|
|
|
|
*next_comm = perf_evsel__strval(evsel, sample, "next_comm");
|
|
|
|
const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
|
|
|
|
next_pid = perf_evsel__intval(evsel, sample, "next_pid");
|
|
|
|
const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
|
2012-09-10 22:15:03 +00:00
|
|
|
struct task_desc *prev, __maybe_unused *next;
|
2012-08-07 14:33:42 +00:00
|
|
|
u64 timestamp0, timestamp = sample->time;
|
|
|
|
int cpu = sample->cpu;
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
s64 delta;
|
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
if (verbose)
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
printf("sched_switch event %p\n", evsel);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
if (cpu >= MAX_CPUS || cpu < 0)
|
2012-09-09 01:53:06 +00:00
|
|
|
return 0;
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
timestamp0 = sched->cpu_last_switched[cpu];
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
if (timestamp0)
|
|
|
|
delta = timestamp - timestamp0;
|
|
|
|
else
|
|
|
|
delta = 0;
|
|
|
|
|
2012-09-09 01:53:06 +00:00
|
|
|
if (delta < 0) {
|
2012-09-12 02:11:06 +00:00
|
|
|
pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
|
|
|
}
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n",
|
|
|
|
prev_comm, prev_pid, next_comm, next_pid, delta);
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
prev = register_pid(sched, prev_pid, prev_comm);
|
|
|
|
next = register_pid(sched, next_pid, next_comm);
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->cpu_last_switched[cpu] = timestamp;
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
add_sched_event_run(sched, prev, timestamp, delta);
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
add_sched_event_sleep(sched, prev, timestamp, prev_state);
|
2012-09-09 01:53:06 +00:00
|
|
|
|
|
|
|
return 0;
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2013-08-08 02:50:47 +00:00
|
|
|
static int replay_fork_event(struct perf_sched *sched,
|
|
|
|
union perf_event *event,
|
|
|
|
struct machine *machine)
|
2009-09-12 01:59:01 +00:00
|
|
|
{
|
2013-08-08 02:50:47 +00:00
|
|
|
struct thread *child, *parent;
|
|
|
|
|
2013-08-27 08:23:03 +00:00
|
|
|
child = machine__findnew_thread(machine, event->fork.pid,
|
|
|
|
event->fork.tid);
|
|
|
|
parent = machine__findnew_thread(machine, event->fork.ppid,
|
|
|
|
event->fork.ptid);
|
2013-08-08 02:50:47 +00:00
|
|
|
|
|
|
|
if (child == NULL || parent == NULL) {
|
|
|
|
pr_debug("thread does not exist on fork event: child %p, parent %p\n",
|
|
|
|
child, parent);
|
|
|
|
return 0;
|
|
|
|
}
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
|
2009-09-12 01:59:01 +00:00
|
|
|
if (verbose) {
|
2013-08-08 02:50:47 +00:00
|
|
|
printf("fork event\n");
|
2013-09-11 12:46:56 +00:00
|
|
|
printf("... parent: %s/%d\n", thread__comm_str(parent), parent->tid);
|
|
|
|
printf("... child: %s/%d\n", thread__comm_str(child), child->tid);
|
2009-09-12 01:59:01 +00:00
|
|
|
}
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
|
2013-09-11 12:46:56 +00:00
|
|
|
register_pid(sched, parent->tid, thread__comm_str(parent));
|
|
|
|
register_pid(sched, child->tid, thread__comm_str(child));
|
2012-09-09 01:53:06 +00:00
|
|
|
return 0;
|
2009-09-12 01:59:01 +00:00
|
|
|
}
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
struct sort_dimension {
|
|
|
|
const char *name;
|
2009-09-11 10:12:54 +00:00
|
|
|
sort_fn_t cmp;
|
2009-09-11 10:12:54 +00:00
|
|
|
struct list_head list;
|
|
|
|
};
|
|
|
|
|
2009-09-13 01:36:29 +00:00
|
|
|
static int
|
2009-09-14 18:04:48 +00:00
|
|
|
thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r)
|
2009-09-13 01:36:29 +00:00
|
|
|
{
|
|
|
|
struct sort_dimension *sort;
|
|
|
|
int ret = 0;
|
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
BUG_ON(list_empty(list));
|
|
|
|
|
2009-09-13 01:36:29 +00:00
|
|
|
list_for_each_entry(sort, list, list) {
|
|
|
|
ret = sort->cmp(l, r);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
static struct work_atoms *
|
2009-09-11 10:12:54 +00:00
|
|
|
thread_atoms_search(struct rb_root *root, struct thread *thread,
|
|
|
|
struct list_head *sort_list)
|
|
|
|
{
|
|
|
|
struct rb_node *node = root->rb_node;
|
2009-09-14 18:04:48 +00:00
|
|
|
struct work_atoms key = { .thread = thread };
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
while (node) {
|
2009-09-14 18:04:48 +00:00
|
|
|
struct work_atoms *atoms;
|
2009-09-11 10:12:54 +00:00
|
|
|
int cmp;
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
atoms = container_of(node, struct work_atoms, node);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
|
|
|
cmp = thread_lat_cmp(sort_list, &key, atoms);
|
|
|
|
if (cmp > 0)
|
|
|
|
node = node->rb_left;
|
|
|
|
else if (cmp < 0)
|
|
|
|
node = node->rb_right;
|
|
|
|
else {
|
|
|
|
BUG_ON(thread != atoms->thread);
|
|
|
|
return atoms;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2009-09-12 06:06:14 +00:00
|
|
|
static void
|
2009-09-14 18:04:48 +00:00
|
|
|
__thread_latency_insert(struct rb_root *root, struct work_atoms *data,
|
2009-09-13 01:36:29 +00:00
|
|
|
struct list_head *sort_list)
|
2009-09-12 06:06:14 +00:00
|
|
|
{
|
|
|
|
struct rb_node **new = &(root->rb_node), *parent = NULL;
|
|
|
|
|
|
|
|
while (*new) {
|
2009-09-14 18:04:48 +00:00
|
|
|
struct work_atoms *this;
|
2009-09-13 01:36:29 +00:00
|
|
|
int cmp;
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
this = container_of(*new, struct work_atoms, node);
|
2009-09-12 06:06:14 +00:00
|
|
|
parent = *new;
|
2009-09-13 01:36:29 +00:00
|
|
|
|
|
|
|
cmp = thread_lat_cmp(sort_list, data, this);
|
|
|
|
|
|
|
|
if (cmp > 0)
|
2009-09-12 06:06:14 +00:00
|
|
|
new = &((*new)->rb_left);
|
|
|
|
else
|
2009-09-13 01:36:29 +00:00
|
|
|
new = &((*new)->rb_right);
|
2009-09-12 06:06:14 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
rb_link_node(&data->node, parent, new);
|
|
|
|
rb_insert_color(&data->node, root);
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread)
|
2009-09-12 06:06:14 +00:00
|
|
|
{
|
2009-11-24 14:05:16 +00:00
|
|
|
struct work_atoms *atoms = zalloc(sizeof(*atoms));
|
2012-09-09 01:53:06 +00:00
|
|
|
if (!atoms) {
|
|
|
|
pr_err("No memory at %s\n", __func__);
|
|
|
|
return -1;
|
|
|
|
}
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2015-03-03 01:21:35 +00:00
|
|
|
atoms->thread = thread__get(thread);
|
2009-09-14 18:04:48 +00:00
|
|
|
INIT_LIST_HEAD(&atoms->work_list);
|
2012-09-11 20:29:27 +00:00
|
|
|
__thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid);
|
2012-09-09 01:53:06 +00:00
|
|
|
return 0;
|
2009-09-12 06:06:14 +00:00
|
|
|
}
|
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
static char sched_out_state(u64 prev_state)
|
2009-09-12 06:06:14 +00:00
|
|
|
{
|
|
|
|
const char *str = TASK_STATE_TO_CHAR_STR;
|
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
return str[prev_state];
|
2009-09-12 06:06:14 +00:00
|
|
|
}
|
|
|
|
|
2012-09-09 01:53:06 +00:00
|
|
|
static int
|
2009-09-14 18:04:48 +00:00
|
|
|
add_sched_out_event(struct work_atoms *atoms,
|
|
|
|
char run_state,
|
|
|
|
u64 timestamp)
|
2009-09-12 06:06:14 +00:00
|
|
|
{
|
2009-11-24 14:05:16 +00:00
|
|
|
struct work_atom *atom = zalloc(sizeof(*atom));
|
2012-09-09 01:53:06 +00:00
|
|
|
if (!atom) {
|
|
|
|
pr_err("Non memory at %s", __func__);
|
|
|
|
return -1;
|
|
|
|
}
|
2009-09-12 06:06:14 +00:00
|
|
|
|
perf tools: Fix processing of randomly serialized sched traces
Currently it's possible to meet such too high latency results
with 'perf sched latency'.
-----------------------------------------------------------------------------------
Task | Runtime ms | Switches | Average delay ms | Maximum delay ms |
-----------------------------------------------------------------------------------
xfce4-panel | 0.222 ms | 2 | avg: 4718.345 ms | max: 9436.493 ms |
scsi_eh_3 | 3.962 ms | 36 | avg: 55.957 ms | max: 1977.829 ms |
The origin is on traces that are sometimes badly serialized across cpus.
For example the raw traces that raised such results for xfce4-panel:
(1) [init]-0 [000] 1494.663899990: sched_switch: task swapper:0 [140] (R) ==> xfce4-panel:4569 [120]
(2) xfce4-panel-4569 [000] 1494.663928373: sched_switch: task xfce4-panel:4569 [120] (S) ==> swapper:0 [140]
(3) Xorg-4276 [001] 1494.663860125: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(4) Xorg-4276 [001] 1504.098252756: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(5) perf-5219 [000] 1504.100353302: sched_switch: task perf:5219 [120] (S) ==> xfce4-panel:4569 [120]
The traces are processed in the order they arrive. Then in (2),
xfce4-panel sleeps, it is first waken up in (3) and eventually
scheduled in (5).
The latency reported is then 1504 - 1495 = 9 secs, as reported by perf
sched. But this is wrong, we are confident in the fact the traces are
nicely serialized while we should actually more trust the timestamps.
If we reorder by timestamps we get:
(1) Xorg-4276 [001] 1494.663860125: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(2) [init]-0 [000] 1494.663899990: sched_switch: task swapper:0 [140] (R) ==> xfce4-panel:4569 [120]
(3) xfce4-panel-4569 [000] 1494.663928373: sched_switch: task xfce4-panel:4569 [120] (S) ==> swapper:0 [140]
(4) Xorg-4276 [001] 1504.098252756: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(5) perf-5219 [000] 1504.100353302: sched_switch: task perf:5219 [120] (S) ==> xfce4-panel:4569 [120]
Now the trace make more sense, xfce4-panel is sleeping. Then it is
woken up in (1), scheduled in (2)
It goes to sleep in (3), woken up in (4) and scheduled in (5).
Now, latency captured between (1) and (2) is of 39 us.
And between (4) and (5) it is 2.1 ms.
Such pattern of bad serializing is the origin of the high latencies
reported by perf sched.
Basically, we need to check whether wake up time is higher than
schedule out time. If it's not the case, we need to tag the current
work atom as invalid.
Beside that, we may need to work later on a better ordering of the
traces given by the kernel.
After this patch:
xfce4-session | 0.221 ms | 1 | avg: 0.538 ms | max: 0.538 ms |
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-14 01:01:12 +00:00
|
|
|
atom->sched_out_time = timestamp;
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
if (run_state == 'R') {
|
2009-09-11 10:12:54 +00:00
|
|
|
atom->state = THREAD_WAIT_CPU;
|
perf tools: Fix processing of randomly serialized sched traces
Currently it's possible to meet such too high latency results
with 'perf sched latency'.
-----------------------------------------------------------------------------------
Task | Runtime ms | Switches | Average delay ms | Maximum delay ms |
-----------------------------------------------------------------------------------
xfce4-panel | 0.222 ms | 2 | avg: 4718.345 ms | max: 9436.493 ms |
scsi_eh_3 | 3.962 ms | 36 | avg: 55.957 ms | max: 1977.829 ms |
The origin is on traces that are sometimes badly serialized across cpus.
For example the raw traces that raised such results for xfce4-panel:
(1) [init]-0 [000] 1494.663899990: sched_switch: task swapper:0 [140] (R) ==> xfce4-panel:4569 [120]
(2) xfce4-panel-4569 [000] 1494.663928373: sched_switch: task xfce4-panel:4569 [120] (S) ==> swapper:0 [140]
(3) Xorg-4276 [001] 1494.663860125: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(4) Xorg-4276 [001] 1504.098252756: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(5) perf-5219 [000] 1504.100353302: sched_switch: task perf:5219 [120] (S) ==> xfce4-panel:4569 [120]
The traces are processed in the order they arrive. Then in (2),
xfce4-panel sleeps, it is first waken up in (3) and eventually
scheduled in (5).
The latency reported is then 1504 - 1495 = 9 secs, as reported by perf
sched. But this is wrong, we are confident in the fact the traces are
nicely serialized while we should actually more trust the timestamps.
If we reorder by timestamps we get:
(1) Xorg-4276 [001] 1494.663860125: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(2) [init]-0 [000] 1494.663899990: sched_switch: task swapper:0 [140] (R) ==> xfce4-panel:4569 [120]
(3) xfce4-panel-4569 [000] 1494.663928373: sched_switch: task xfce4-panel:4569 [120] (S) ==> swapper:0 [140]
(4) Xorg-4276 [001] 1504.098252756: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(5) perf-5219 [000] 1504.100353302: sched_switch: task perf:5219 [120] (S) ==> xfce4-panel:4569 [120]
Now the trace make more sense, xfce4-panel is sleeping. Then it is
woken up in (1), scheduled in (2)
It goes to sleep in (3), woken up in (4) and scheduled in (5).
Now, latency captured between (1) and (2) is of 39 us.
And between (4) and (5) it is 2.1 ms.
Such pattern of bad serializing is the origin of the high latencies
reported by perf sched.
Basically, we need to check whether wake up time is higher than
schedule out time. If it's not the case, we need to tag the current
work atom as invalid.
Beside that, we may need to work later on a better ordering of the
traces given by the kernel.
After this patch:
xfce4-session | 0.221 ms | 1 | avg: 0.538 ms | max: 0.538 ms |
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-14 01:01:12 +00:00
|
|
|
atom->wake_up_time = atom->sched_out_time;
|
2009-09-12 22:46:19 +00:00
|
|
|
}
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
list_add_tail(&atom->list, &atoms->work_list);
|
2012-09-09 01:53:06 +00:00
|
|
|
return 0;
|
2009-09-12 06:06:14 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
2012-09-10 22:15:03 +00:00
|
|
|
add_runtime_event(struct work_atoms *atoms, u64 delta,
|
|
|
|
u64 timestamp __maybe_unused)
|
2009-09-14 18:04:48 +00:00
|
|
|
{
|
|
|
|
struct work_atom *atom;
|
|
|
|
|
|
|
|
BUG_ON(list_empty(&atoms->work_list));
|
|
|
|
|
|
|
|
atom = list_entry(atoms->work_list.prev, struct work_atom, list);
|
|
|
|
|
|
|
|
atom->runtime += delta;
|
|
|
|
atoms->total_runtime += delta;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
add_sched_in_event(struct work_atoms *atoms, u64 timestamp)
|
2009-09-12 06:06:14 +00:00
|
|
|
{
|
2009-09-11 10:12:54 +00:00
|
|
|
struct work_atom *atom;
|
2009-09-12 23:56:25 +00:00
|
|
|
u64 delta;
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
if (list_empty(&atoms->work_list))
|
2009-09-12 06:06:14 +00:00
|
|
|
return;
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
atom = list_entry(atoms->work_list.prev, struct work_atom, list);
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
if (atom->state != THREAD_WAIT_CPU)
|
2009-09-12 06:06:14 +00:00
|
|
|
return;
|
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
if (timestamp < atom->wake_up_time) {
|
|
|
|
atom->state = THREAD_IGNORE;
|
2009-09-12 06:06:14 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
atom->state = THREAD_SCHED_IN;
|
|
|
|
atom->sched_in_time = timestamp;
|
2009-09-12 23:56:25 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
delta = atom->sched_in_time - atom->wake_up_time;
|
2009-09-12 23:56:25 +00:00
|
|
|
atoms->total_lat += delta;
|
2009-12-09 20:40:08 +00:00
|
|
|
if (delta > atoms->max_lat) {
|
2009-09-12 23:56:25 +00:00
|
|
|
atoms->max_lat = delta;
|
2009-12-09 20:40:08 +00:00
|
|
|
atoms->max_lat_at = timestamp;
|
|
|
|
}
|
2009-09-12 23:56:25 +00:00
|
|
|
atoms->nb_atoms++;
|
2009-09-12 06:06:14 +00:00
|
|
|
}
|
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
static int latency_switch_event(struct perf_sched *sched,
|
|
|
|
struct perf_evsel *evsel,
|
|
|
|
struct perf_sample *sample,
|
|
|
|
struct machine *machine)
|
2009-09-12 06:06:14 +00:00
|
|
|
{
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
|
|
|
|
next_pid = perf_evsel__intval(evsel, sample, "next_pid");
|
|
|
|
const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
|
2009-09-14 18:04:48 +00:00
|
|
|
struct work_atoms *out_events, *in_events;
|
2009-09-12 06:06:14 +00:00
|
|
|
struct thread *sched_out, *sched_in;
|
2012-08-07 14:33:42 +00:00
|
|
|
u64 timestamp0, timestamp = sample->time;
|
|
|
|
int cpu = sample->cpu;
|
2009-09-12 08:08:34 +00:00
|
|
|
s64 delta;
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
BUG_ON(cpu >= MAX_CPUS || cpu < 0);
|
2009-09-12 08:08:34 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
timestamp0 = sched->cpu_last_switched[cpu];
|
|
|
|
sched->cpu_last_switched[cpu] = timestamp;
|
2009-09-12 08:08:34 +00:00
|
|
|
if (timestamp0)
|
|
|
|
delta = timestamp - timestamp0;
|
|
|
|
else
|
|
|
|
delta = 0;
|
|
|
|
|
2012-09-09 01:53:06 +00:00
|
|
|
if (delta < 0) {
|
|
|
|
pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
|
|
|
|
return -1;
|
|
|
|
}
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2014-07-14 10:02:25 +00:00
|
|
|
sched_out = machine__findnew_thread(machine, -1, prev_pid);
|
|
|
|
sched_in = machine__findnew_thread(machine, -1, next_pid);
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
|
2009-09-14 18:04:48 +00:00
|
|
|
if (!out_events) {
|
2012-09-11 20:29:27 +00:00
|
|
|
if (thread_atoms_insert(sched, sched_out))
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
2012-09-11 20:29:27 +00:00
|
|
|
out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
|
2012-09-09 01:53:06 +00:00
|
|
|
if (!out_events) {
|
|
|
|
pr_err("out-event: Internal tree error");
|
|
|
|
return -1;
|
|
|
|
}
|
2009-09-14 18:04:48 +00:00
|
|
|
}
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
if (add_sched_out_event(out_events, sched_out_state(prev_state), timestamp))
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
2009-09-14 18:04:48 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
|
2009-09-14 18:04:48 +00:00
|
|
|
if (!in_events) {
|
2012-09-11 20:29:27 +00:00
|
|
|
if (thread_atoms_insert(sched, sched_in))
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
2012-09-11 20:29:27 +00:00
|
|
|
in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
|
2012-09-09 01:53:06 +00:00
|
|
|
if (!in_events) {
|
|
|
|
pr_err("in-event: Internal tree error");
|
|
|
|
return -1;
|
|
|
|
}
|
2009-09-14 18:04:48 +00:00
|
|
|
/*
|
|
|
|
* Take came in we have not heard about yet,
|
|
|
|
* add in an initial atom in runnable state:
|
|
|
|
*/
|
2012-09-09 01:53:06 +00:00
|
|
|
if (add_sched_out_event(in_events, 'R', timestamp))
|
|
|
|
return -1;
|
2009-09-12 06:06:14 +00:00
|
|
|
}
|
2009-09-14 18:04:48 +00:00
|
|
|
add_sched_in_event(in_events, timestamp);
|
2012-09-09 01:53:06 +00:00
|
|
|
|
|
|
|
return 0;
|
2009-09-14 18:04:48 +00:00
|
|
|
}
|
2009-09-12 06:06:14 +00:00
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
static int latency_runtime_event(struct perf_sched *sched,
|
|
|
|
struct perf_evsel *evsel,
|
|
|
|
struct perf_sample *sample,
|
|
|
|
struct machine *machine)
|
2009-09-14 18:04:48 +00:00
|
|
|
{
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
const u32 pid = perf_evsel__intval(evsel, sample, "pid");
|
|
|
|
const u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
|
2014-07-14 10:02:25 +00:00
|
|
|
struct thread *thread = machine__findnew_thread(machine, -1, pid);
|
2012-09-11 20:29:27 +00:00
|
|
|
struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
|
2012-08-07 14:33:42 +00:00
|
|
|
u64 timestamp = sample->time;
|
|
|
|
int cpu = sample->cpu;
|
2009-09-14 18:04:48 +00:00
|
|
|
|
|
|
|
BUG_ON(cpu >= MAX_CPUS || cpu < 0);
|
|
|
|
if (!atoms) {
|
2012-09-11 20:29:27 +00:00
|
|
|
if (thread_atoms_insert(sched, thread))
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
2012-09-11 20:29:27 +00:00
|
|
|
atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
|
2012-09-09 01:53:06 +00:00
|
|
|
if (!atoms) {
|
2012-09-12 02:11:06 +00:00
|
|
|
pr_err("in-event: Internal tree error");
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
if (add_sched_out_event(atoms, 'R', timestamp))
|
|
|
|
return -1;
|
2009-09-12 06:06:14 +00:00
|
|
|
}
|
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
add_runtime_event(atoms, runtime, timestamp);
|
2012-09-09 01:53:06 +00:00
|
|
|
return 0;
|
2009-09-12 06:06:14 +00:00
|
|
|
}
|
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
static int latency_wakeup_event(struct perf_sched *sched,
|
|
|
|
struct perf_evsel *evsel,
|
|
|
|
struct perf_sample *sample,
|
|
|
|
struct machine *machine)
|
2009-09-12 06:06:14 +00:00
|
|
|
{
|
2014-05-12 18:19:46 +00:00
|
|
|
const u32 pid = perf_evsel__intval(evsel, sample, "pid");
|
2009-09-14 18:04:48 +00:00
|
|
|
struct work_atoms *atoms;
|
2009-09-11 10:12:54 +00:00
|
|
|
struct work_atom *atom;
|
2009-09-12 06:06:14 +00:00
|
|
|
struct thread *wakee;
|
2012-08-07 14:33:42 +00:00
|
|
|
u64 timestamp = sample->time;
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2014-07-14 10:02:25 +00:00
|
|
|
wakee = machine__findnew_thread(machine, -1, pid);
|
2012-09-11 20:29:27 +00:00
|
|
|
atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
|
2009-09-12 21:11:32 +00:00
|
|
|
if (!atoms) {
|
2012-09-11 20:29:27 +00:00
|
|
|
if (thread_atoms_insert(sched, wakee))
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
2012-09-11 20:29:27 +00:00
|
|
|
atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
|
2012-09-09 01:53:06 +00:00
|
|
|
if (!atoms) {
|
2012-09-12 02:11:06 +00:00
|
|
|
pr_err("wakeup-event: Internal tree error");
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
if (add_sched_out_event(atoms, 'S', timestamp))
|
|
|
|
return -1;
|
2009-09-12 06:06:14 +00:00
|
|
|
}
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
BUG_ON(list_empty(&atoms->work_list));
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
atom = list_entry(atoms->work_list.prev, struct work_atom, list);
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2009-10-10 12:46:04 +00:00
|
|
|
/*
|
2014-05-13 01:38:21 +00:00
|
|
|
* As we do not guarantee the wakeup event happens when
|
|
|
|
* task is out of run queue, also may happen when task is
|
|
|
|
* on run queue and wakeup only change ->state to TASK_RUNNING,
|
|
|
|
* then we should not set the ->wake_up_time when wake up a
|
|
|
|
* task which is on run queue.
|
|
|
|
*
|
2009-10-10 12:46:04 +00:00
|
|
|
* You WILL be missing events if you've recorded only
|
|
|
|
* one CPU, or are only looking at only one, so don't
|
2014-05-13 01:38:21 +00:00
|
|
|
* skip in this case.
|
2009-10-10 12:46:04 +00:00
|
|
|
*/
|
2012-09-11 20:29:27 +00:00
|
|
|
if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING)
|
2014-05-13 01:38:21 +00:00
|
|
|
return 0;
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->nr_timestamps++;
|
2009-09-13 16:15:54 +00:00
|
|
|
if (atom->sched_out_time > timestamp) {
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->nr_unordered_timestamps++;
|
2012-09-09 01:53:06 +00:00
|
|
|
return 0;
|
2009-09-13 16:15:54 +00:00
|
|
|
}
|
perf tools: Fix processing of randomly serialized sched traces
Currently it's possible to meet such too high latency results
with 'perf sched latency'.
-----------------------------------------------------------------------------------
Task | Runtime ms | Switches | Average delay ms | Maximum delay ms |
-----------------------------------------------------------------------------------
xfce4-panel | 0.222 ms | 2 | avg: 4718.345 ms | max: 9436.493 ms |
scsi_eh_3 | 3.962 ms | 36 | avg: 55.957 ms | max: 1977.829 ms |
The origin is on traces that are sometimes badly serialized across cpus.
For example the raw traces that raised such results for xfce4-panel:
(1) [init]-0 [000] 1494.663899990: sched_switch: task swapper:0 [140] (R) ==> xfce4-panel:4569 [120]
(2) xfce4-panel-4569 [000] 1494.663928373: sched_switch: task xfce4-panel:4569 [120] (S) ==> swapper:0 [140]
(3) Xorg-4276 [001] 1494.663860125: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(4) Xorg-4276 [001] 1504.098252756: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(5) perf-5219 [000] 1504.100353302: sched_switch: task perf:5219 [120] (S) ==> xfce4-panel:4569 [120]
The traces are processed in the order they arrive. Then in (2),
xfce4-panel sleeps, it is first waken up in (3) and eventually
scheduled in (5).
The latency reported is then 1504 - 1495 = 9 secs, as reported by perf
sched. But this is wrong, we are confident in the fact the traces are
nicely serialized while we should actually more trust the timestamps.
If we reorder by timestamps we get:
(1) Xorg-4276 [001] 1494.663860125: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(2) [init]-0 [000] 1494.663899990: sched_switch: task swapper:0 [140] (R) ==> xfce4-panel:4569 [120]
(3) xfce4-panel-4569 [000] 1494.663928373: sched_switch: task xfce4-panel:4569 [120] (S) ==> swapper:0 [140]
(4) Xorg-4276 [001] 1504.098252756: sched_wakeup: task xfce4-panel:4569 [120] success=1 [000]
(5) perf-5219 [000] 1504.100353302: sched_switch: task perf:5219 [120] (S) ==> xfce4-panel:4569 [120]
Now the trace make more sense, xfce4-panel is sleeping. Then it is
woken up in (1), scheduled in (2)
It goes to sleep in (3), woken up in (4) and scheduled in (5).
Now, latency captured between (1) and (2) is of 39 us.
And between (4) and (5) it is 2.1 ms.
Such pattern of bad serializing is the origin of the high latencies
reported by perf sched.
Basically, we need to check whether wake up time is higher than
schedule out time. If it's not the case, we need to tag the current
work atom as invalid.
Beside that, we may need to work later on a better ordering of the
traces given by the kernel.
After this patch:
xfce4-session | 0.221 ms | 1 | avg: 0.538 ms | max: 0.538 ms |
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-14 01:01:12 +00:00
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
atom->state = THREAD_WAIT_CPU;
|
|
|
|
atom->wake_up_time = timestamp;
|
2012-09-09 01:53:06 +00:00
|
|
|
return 0;
|
2009-09-12 06:06:14 +00:00
|
|
|
}
|
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
static int latency_migrate_task_event(struct perf_sched *sched,
|
|
|
|
struct perf_evsel *evsel,
|
|
|
|
struct perf_sample *sample,
|
|
|
|
struct machine *machine)
|
2009-10-10 12:46:04 +00:00
|
|
|
{
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
const u32 pid = perf_evsel__intval(evsel, sample, "pid");
|
2012-08-07 14:33:42 +00:00
|
|
|
u64 timestamp = sample->time;
|
2009-10-10 12:46:04 +00:00
|
|
|
struct work_atoms *atoms;
|
|
|
|
struct work_atom *atom;
|
|
|
|
struct thread *migrant;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Only need to worry about migration when profiling one CPU.
|
|
|
|
*/
|
2012-09-11 20:29:27 +00:00
|
|
|
if (sched->profile_cpu == -1)
|
2012-09-09 01:53:06 +00:00
|
|
|
return 0;
|
2009-10-10 12:46:04 +00:00
|
|
|
|
2014-07-14 10:02:25 +00:00
|
|
|
migrant = machine__findnew_thread(machine, -1, pid);
|
2012-09-11 20:29:27 +00:00
|
|
|
atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
|
2009-10-10 12:46:04 +00:00
|
|
|
if (!atoms) {
|
2012-09-11 20:29:27 +00:00
|
|
|
if (thread_atoms_insert(sched, migrant))
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
2013-09-11 12:46:56 +00:00
|
|
|
register_pid(sched, migrant->tid, thread__comm_str(migrant));
|
2012-09-11 20:29:27 +00:00
|
|
|
atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
|
2012-09-09 01:53:06 +00:00
|
|
|
if (!atoms) {
|
2012-09-12 02:11:06 +00:00
|
|
|
pr_err("migration-event: Internal tree error");
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
if (add_sched_out_event(atoms, 'R', timestamp))
|
|
|
|
return -1;
|
2009-10-10 12:46:04 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
BUG_ON(list_empty(&atoms->work_list));
|
|
|
|
|
|
|
|
atom = list_entry(atoms->work_list.prev, struct work_atom, list);
|
|
|
|
atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp;
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->nr_timestamps++;
|
2009-10-10 12:46:04 +00:00
|
|
|
|
|
|
|
if (atom->sched_out_time > timestamp)
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->nr_unordered_timestamps++;
|
2012-09-09 01:53:06 +00:00
|
|
|
|
|
|
|
return 0;
|
2009-10-10 12:46:04 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list)
|
2009-09-12 06:06:14 +00:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
int ret;
|
2009-09-12 23:56:25 +00:00
|
|
|
u64 avg;
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
if (!work_list->nb_atoms)
|
2009-09-12 06:06:14 +00:00
|
|
|
return;
|
2009-09-13 16:15:54 +00:00
|
|
|
/*
|
|
|
|
* Ignore idle threads:
|
|
|
|
*/
|
2013-09-11 12:46:56 +00:00
|
|
|
if (!strcmp(thread__comm_str(work_list->thread), "swapper"))
|
2009-09-13 16:15:54 +00:00
|
|
|
return;
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->all_runtime += work_list->total_runtime;
|
|
|
|
sched->all_count += work_list->nb_atoms;
|
2009-09-12 23:56:25 +00:00
|
|
|
|
2013-09-11 12:46:56 +00:00
|
|
|
ret = printf(" %s:%d ", thread__comm_str(work_list->thread), work_list->thread->tid);
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2009-09-14 16:30:44 +00:00
|
|
|
for (i = 0; i < 24 - ret; i++)
|
2009-09-12 06:06:14 +00:00
|
|
|
printf(" ");
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
avg = work_list->total_lat / work_list->nb_atoms;
|
2009-09-12 06:06:14 +00:00
|
|
|
|
2014-03-17 14:18:21 +00:00
|
|
|
printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %13.6f s\n",
|
2009-09-14 18:04:48 +00:00
|
|
|
(double)work_list->total_runtime / 1e6,
|
|
|
|
work_list->nb_atoms, (double)avg / 1e6,
|
2009-12-09 20:40:08 +00:00
|
|
|
(double)work_list->max_lat / 1e6,
|
|
|
|
(double)work_list->max_lat_at / 1e9);
|
2009-09-12 06:06:14 +00:00
|
|
|
}
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
static int pid_cmp(struct work_atoms *l, struct work_atoms *r)
|
2009-09-13 01:36:29 +00:00
|
|
|
{
|
2013-07-04 13:20:31 +00:00
|
|
|
if (l->thread->tid < r->thread->tid)
|
2009-09-13 01:36:29 +00:00
|
|
|
return -1;
|
2013-07-04 13:20:31 +00:00
|
|
|
if (l->thread->tid > r->thread->tid)
|
2009-09-13 01:36:29 +00:00
|
|
|
return 1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
static int avg_cmp(struct work_atoms *l, struct work_atoms *r)
|
2009-09-13 01:36:29 +00:00
|
|
|
{
|
|
|
|
u64 avgl, avgr;
|
|
|
|
|
|
|
|
if (!l->nb_atoms)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
if (!r->nb_atoms)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
avgl = l->total_lat / l->nb_atoms;
|
|
|
|
avgr = r->total_lat / r->nb_atoms;
|
|
|
|
|
|
|
|
if (avgl < avgr)
|
|
|
|
return -1;
|
|
|
|
if (avgl > avgr)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
static int max_cmp(struct work_atoms *l, struct work_atoms *r)
|
2009-09-13 01:36:29 +00:00
|
|
|
{
|
|
|
|
if (l->max_lat < r->max_lat)
|
|
|
|
return -1;
|
|
|
|
if (l->max_lat > r->max_lat)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
static int switch_cmp(struct work_atoms *l, struct work_atoms *r)
|
2009-09-13 01:36:29 +00:00
|
|
|
{
|
|
|
|
if (l->nb_atoms < r->nb_atoms)
|
|
|
|
return -1;
|
|
|
|
if (l->nb_atoms > r->nb_atoms)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-09-14 18:04:48 +00:00
|
|
|
static int runtime_cmp(struct work_atoms *l, struct work_atoms *r)
|
2009-09-13 01:36:29 +00:00
|
|
|
{
|
|
|
|
if (l->total_runtime < r->total_runtime)
|
|
|
|
return -1;
|
|
|
|
if (l->total_runtime > r->total_runtime)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-10-05 20:17:29 +00:00
|
|
|
static int sort_dimension__add(const char *tok, struct list_head *list)
|
2009-09-13 01:36:29 +00:00
|
|
|
{
|
2012-09-11 20:29:27 +00:00
|
|
|
size_t i;
|
|
|
|
static struct sort_dimension avg_sort_dimension = {
|
|
|
|
.name = "avg",
|
|
|
|
.cmp = avg_cmp,
|
|
|
|
};
|
|
|
|
static struct sort_dimension max_sort_dimension = {
|
|
|
|
.name = "max",
|
|
|
|
.cmp = max_cmp,
|
|
|
|
};
|
|
|
|
static struct sort_dimension pid_sort_dimension = {
|
|
|
|
.name = "pid",
|
|
|
|
.cmp = pid_cmp,
|
|
|
|
};
|
|
|
|
static struct sort_dimension runtime_sort_dimension = {
|
|
|
|
.name = "runtime",
|
|
|
|
.cmp = runtime_cmp,
|
|
|
|
};
|
|
|
|
static struct sort_dimension switch_sort_dimension = {
|
|
|
|
.name = "switch",
|
|
|
|
.cmp = switch_cmp,
|
|
|
|
};
|
|
|
|
struct sort_dimension *available_sorts[] = {
|
|
|
|
&pid_sort_dimension,
|
|
|
|
&avg_sort_dimension,
|
|
|
|
&max_sort_dimension,
|
|
|
|
&switch_sort_dimension,
|
|
|
|
&runtime_sort_dimension,
|
|
|
|
};
|
2009-09-13 01:36:29 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
for (i = 0; i < ARRAY_SIZE(available_sorts); i++) {
|
2009-09-13 01:36:29 +00:00
|
|
|
if (!strcmp(available_sorts[i]->name, tok)) {
|
|
|
|
list_add_tail(&available_sorts[i]->list, list);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void perf_sched__sort_lat(struct perf_sched *sched)
|
2009-09-13 01:36:29 +00:00
|
|
|
{
|
|
|
|
struct rb_node *node;
|
|
|
|
|
|
|
|
for (;;) {
|
2009-09-14 18:04:48 +00:00
|
|
|
struct work_atoms *data;
|
2012-09-11 20:29:27 +00:00
|
|
|
node = rb_first(&sched->atom_root);
|
2009-09-13 01:36:29 +00:00
|
|
|
if (!node)
|
|
|
|
break;
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
rb_erase(node, &sched->atom_root);
|
2009-09-14 18:04:48 +00:00
|
|
|
data = rb_entry(node, struct work_atoms, node);
|
2012-09-11 20:29:27 +00:00
|
|
|
__thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list);
|
2009-09-13 01:36:29 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static int process_sched_wakeup_event(struct perf_tool *tool,
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
struct perf_evsel *evsel,
|
2012-09-10 22:15:03 +00:00
|
|
|
struct perf_sample *sample,
|
2012-09-11 16:18:47 +00:00
|
|
|
struct machine *machine)
|
2009-09-12 01:59:01 +00:00
|
|
|
{
|
2012-09-11 20:29:27 +00:00
|
|
|
struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
|
2009-09-12 01:59:01 +00:00
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
if (sched->tp_handler->wakeup_event)
|
|
|
|
return sched->tp_handler->wakeup_event(sched, evsel, sample, machine);
|
2012-09-09 01:53:06 +00:00
|
|
|
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
return 0;
|
2009-09-12 01:59:01 +00:00
|
|
|
}
|
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
static int map_switch_event(struct perf_sched *sched, struct perf_evsel *evsel,
|
|
|
|
struct perf_sample *sample, struct machine *machine)
|
2009-09-16 15:40:48 +00:00
|
|
|
{
|
2014-05-16 05:37:05 +00:00
|
|
|
const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
|
|
|
|
struct thread *sched_in;
|
2009-09-16 15:40:48 +00:00
|
|
|
int new_shortname;
|
2012-08-07 14:33:42 +00:00
|
|
|
u64 timestamp0, timestamp = sample->time;
|
2009-09-16 15:40:48 +00:00
|
|
|
s64 delta;
|
2012-08-07 14:33:42 +00:00
|
|
|
int cpu, this_cpu = sample->cpu;
|
2009-09-16 15:40:48 +00:00
|
|
|
|
|
|
|
BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0);
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
if (this_cpu > sched->max_cpu)
|
|
|
|
sched->max_cpu = this_cpu;
|
2009-09-16 15:40:48 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
timestamp0 = sched->cpu_last_switched[this_cpu];
|
|
|
|
sched->cpu_last_switched[this_cpu] = timestamp;
|
2009-09-16 15:40:48 +00:00
|
|
|
if (timestamp0)
|
|
|
|
delta = timestamp - timestamp0;
|
|
|
|
else
|
|
|
|
delta = 0;
|
|
|
|
|
2012-09-09 01:53:06 +00:00
|
|
|
if (delta < 0) {
|
2012-09-12 02:11:06 +00:00
|
|
|
pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
|
|
|
}
|
2009-09-16 15:40:48 +00:00
|
|
|
|
2014-07-14 10:02:25 +00:00
|
|
|
sched_in = machine__findnew_thread(machine, -1, next_pid);
|
2009-09-16 15:40:48 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->curr_thread[this_cpu] = sched_in;
|
2009-09-16 15:40:48 +00:00
|
|
|
|
|
|
|
printf(" ");
|
|
|
|
|
|
|
|
new_shortname = 0;
|
|
|
|
if (!sched_in->shortname[0]) {
|
2014-05-06 05:39:01 +00:00
|
|
|
if (!strcmp(thread__comm_str(sched_in), "swapper")) {
|
|
|
|
/*
|
|
|
|
* Don't allocate a letter-number for swapper:0
|
|
|
|
* as a shortname. Instead, we use '.' for it.
|
|
|
|
*/
|
|
|
|
sched_in->shortname[0] = '.';
|
|
|
|
sched_in->shortname[1] = ' ';
|
2009-09-16 15:40:48 +00:00
|
|
|
} else {
|
2014-05-06 05:39:01 +00:00
|
|
|
sched_in->shortname[0] = sched->next_shortname1;
|
|
|
|
sched_in->shortname[1] = sched->next_shortname2;
|
|
|
|
|
|
|
|
if (sched->next_shortname1 < 'Z') {
|
|
|
|
sched->next_shortname1++;
|
2009-09-16 15:40:48 +00:00
|
|
|
} else {
|
2014-05-06 05:39:01 +00:00
|
|
|
sched->next_shortname1 = 'A';
|
|
|
|
if (sched->next_shortname2 < '9')
|
|
|
|
sched->next_shortname2++;
|
|
|
|
else
|
|
|
|
sched->next_shortname2 = '0';
|
2009-09-16 15:40:48 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
new_shortname = 1;
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
for (cpu = 0; cpu <= sched->max_cpu; cpu++) {
|
2009-09-16 15:40:48 +00:00
|
|
|
if (cpu != this_cpu)
|
|
|
|
printf(" ");
|
|
|
|
else
|
|
|
|
printf("*");
|
|
|
|
|
2014-05-06 05:39:01 +00:00
|
|
|
if (sched->curr_thread[cpu])
|
|
|
|
printf("%2s ", sched->curr_thread[cpu]->shortname);
|
|
|
|
else
|
2009-09-16 15:40:48 +00:00
|
|
|
printf(" ");
|
|
|
|
}
|
|
|
|
|
|
|
|
printf(" %12.6f secs ", (double)timestamp/1e9);
|
|
|
|
if (new_shortname) {
|
|
|
|
printf("%s => %s:%d\n",
|
2013-09-11 12:46:56 +00:00
|
|
|
sched_in->shortname, thread__comm_str(sched_in), sched_in->tid);
|
2009-09-16 15:40:48 +00:00
|
|
|
} else {
|
|
|
|
printf("\n");
|
|
|
|
}
|
2012-09-09 01:53:06 +00:00
|
|
|
|
|
|
|
return 0;
|
2009-09-16 15:40:48 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static int process_sched_switch_event(struct perf_tool *tool,
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
struct perf_evsel *evsel,
|
2012-09-10 22:15:03 +00:00
|
|
|
struct perf_sample *sample,
|
2012-09-11 16:18:47 +00:00
|
|
|
struct machine *machine)
|
2009-09-12 01:59:01 +00:00
|
|
|
{
|
2012-09-11 20:29:27 +00:00
|
|
|
struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
|
2012-09-09 01:53:06 +00:00
|
|
|
int this_cpu = sample->cpu, err = 0;
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
|
|
|
|
next_pid = perf_evsel__intval(evsel, sample, "next_pid");
|
2009-09-12 01:59:01 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
if (sched->curr_pid[this_cpu] != (u32)-1) {
|
2009-09-16 12:07:00 +00:00
|
|
|
/*
|
|
|
|
* Are we trying to switch away a PID that is
|
|
|
|
* not current?
|
|
|
|
*/
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
if (sched->curr_pid[this_cpu] != prev_pid)
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->nr_context_switch_bugs++;
|
2009-09-16 12:07:00 +00:00
|
|
|
}
|
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
if (sched->tp_handler->switch_event)
|
|
|
|
err = sched->tp_handler->switch_event(sched, evsel, sample, machine);
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
|
|
|
|
sched->curr_pid[this_cpu] = next_pid;
|
2012-09-09 01:53:06 +00:00
|
|
|
return err;
|
2009-09-12 01:59:01 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static int process_sched_runtime_event(struct perf_tool *tool,
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
struct perf_evsel *evsel,
|
2012-09-10 22:15:03 +00:00
|
|
|
struct perf_sample *sample,
|
2012-09-11 16:18:47 +00:00
|
|
|
struct machine *machine)
|
2009-09-14 18:04:48 +00:00
|
|
|
{
|
2012-09-11 20:29:27 +00:00
|
|
|
struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
|
2009-09-14 18:04:48 +00:00
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
if (sched->tp_handler->runtime_event)
|
|
|
|
return sched->tp_handler->runtime_event(sched, evsel, sample, machine);
|
2012-09-09 01:53:06 +00:00
|
|
|
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
return 0;
|
2009-09-14 18:04:48 +00:00
|
|
|
}
|
|
|
|
|
2013-08-08 02:50:47 +00:00
|
|
|
static int perf_sched__process_fork_event(struct perf_tool *tool,
|
|
|
|
union perf_event *event,
|
|
|
|
struct perf_sample *sample,
|
|
|
|
struct machine *machine)
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2012-09-11 20:29:27 +00:00
|
|
|
struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
|
2009-09-12 00:43:45 +00:00
|
|
|
|
2013-08-08 02:50:47 +00:00
|
|
|
/* run the fork event through the perf machineruy */
|
|
|
|
perf_event__process_fork(tool, event, sample, machine);
|
|
|
|
|
|
|
|
/* and then run additional processing needed for this command */
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
if (sched->tp_handler->fork_event)
|
2013-08-08 02:50:47 +00:00
|
|
|
return sched->tp_handler->fork_event(sched, event, machine);
|
2012-09-09 01:53:06 +00:00
|
|
|
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
return 0;
|
perf sched: Implement the scheduling workload replay engine
Integrate the schedbench.c bits with the raw trace events
that we get from the perf machinery, and activate the
workload replayer/simulator.
Example of a captured 'make -j' workload:
$ perf sched
run measurement overhead: 90 nsecs
sleep measurement overhead: 2724743 nsecs
the run test took 1000081 nsecs
the sleep test took 2981111 nsecs
version = 0.5
...
nr_run_events: 70
nr_sleep_events: 66
nr_wakeup_events: 9
target-less wakeups: 71
multi-target wakeups: 47
run events optimized: 139
task 0 ( perf: 6607), nr_events: 2
task 1 ( perf: 6608), nr_events: 6
task 2 ( : 0), nr_events: 1
task 3 ( make: 6609), nr_events: 5
task 4 ( sh: 6610), nr_events: 4
task 5 ( make: 6611), nr_events: 6
task 6 ( sh: 6612), nr_events: 4
task 7 ( make: 6613), nr_events: 5
task 8 ( migration/11: 25), nr_events: 1
task 9 ( migration/13: 29), nr_events: 1
task 10 ( migration/15: 33), nr_events: 1
task 11 ( migration/9: 21), nr_events: 1
task 12 ( sh: 6614), nr_events: 4
task 13 ( make: 6615), nr_events: 5
task 14 ( sh: 6616), nr_events: 4
task 15 ( make: 6617), nr_events: 7
task 16 ( migration/3: 9), nr_events: 1
task 17 ( migration/5: 13), nr_events: 1
task 18 ( migration/7: 17), nr_events: 1
task 19 ( migration/1: 5), nr_events: 1
task 20 ( sh: 6618), nr_events: 4
task 21 ( make: 6619), nr_events: 5
task 22 ( sh: 6620), nr_events: 4
task 23 ( make: 6621), nr_events: 10
task 24 ( sh: 6623), nr_events: 3
task 25 ( gcc: 6624), nr_events: 4
task 26 ( gcc: 6625), nr_events: 4
task 27 ( gcc: 6626), nr_events: 5
task 28 ( collect2: 6627), nr_events: 5
task 29 ( sh: 6622), nr_events: 1
task 30 ( make: 6628), nr_events: 7
task 31 ( sh: 6630), nr_events: 4
task 32 ( gcc: 6631), nr_events: 4
task 33 ( sh: 6629), nr_events: 1
task 34 ( gcc: 6632), nr_events: 4
task 35 ( gcc: 6633), nr_events: 4
task 36 ( collect2: 6634), nr_events: 4
task 37 ( make: 6635), nr_events: 8
task 38 ( sh: 6637), nr_events: 4
task 39 ( sh: 6636), nr_events: 1
task 40 ( gcc: 6638), nr_events: 4
task 41 ( gcc: 6639), nr_events: 4
task 42 ( gcc: 6640), nr_events: 4
task 43 ( collect2: 6641), nr_events: 4
task 44 ( make: 6642), nr_events: 6
task 45 ( sh: 6643), nr_events: 5
task 46 ( sh: 6644), nr_events: 3
task 47 ( sh: 6645), nr_events: 4
task 48 ( make: 6646), nr_events: 6
task 49 ( sh: 6647), nr_events: 3
task 50 ( make: 6648), nr_events: 5
task 51 ( sh: 6649), nr_events: 5
task 52 ( sh: 6650), nr_events: 6
task 53 ( make: 6651), nr_events: 4
task 54 ( make: 6652), nr_events: 5
task 55 ( make: 6653), nr_events: 4
task 56 ( make: 6654), nr_events: 4
task 57 ( make: 6655), nr_events: 5
task 58 ( sh: 6656), nr_events: 4
task 59 ( gcc: 6657), nr_events: 9
task 60 ( ksoftirqd/3: 10), nr_events: 1
task 61 ( gcc: 6658), nr_events: 4
task 62 ( make: 6659), nr_events: 5
task 63 ( sh: 6660), nr_events: 3
task 64 ( gcc: 6661), nr_events: 5
task 65 ( collect2: 6662), nr_events: 4
------------------------------------------------------------
#1 : 256.745, ravg: 256.74, cpu: 0.00 / 0.00
#2 : 439.372, ravg: 275.01, cpu: 0.00 / 0.00
#3 : 411.971, ravg: 288.70, cpu: 0.00 / 0.00
#4 : 385.500, ravg: 298.38, cpu: 0.00 / 0.00
#5 : 366.526, ravg: 305.20, cpu: 0.00 / 0.00
#6 : 381.281, ravg: 312.81, cpu: 0.00 / 0.00
#7 : 410.756, ravg: 322.60, cpu: 0.00 / 0.00
#8 : 368.009, ravg: 327.14, cpu: 0.00 / 0.00
#9 : 408.098, ravg: 335.24, cpu: 0.00 / 0.00
#10 : 368.582, ravg: 338.57, cpu: 0.00 / 0.00
I.e. we successfully analyzed the trace, replayed it
via real threads and measured the replayed workload's
scheduling properties.
This is how it looked like in 'top' output:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
7164 mingo 20 0 1434m 8080 888 R 57.0 0.1 0:02.04 :perf
7165 mingo 20 0 1434m 8080 888 R 41.8 0.1 0:01.52 :perf
7228 mingo 20 0 1434m 8080 888 R 39.8 0.1 0:01.44 :gcc
7225 mingo 20 0 1434m 8080 888 R 33.8 0.1 0:01.26 :gcc
7202 mingo 20 0 1434m 8080 888 R 31.2 0.1 0:01.16 :sh
7222 mingo 20 0 1434m 8080 888 R 25.2 0.1 0:00.96 :sh
7211 mingo 20 0 1434m 8080 888 R 21.9 0.1 0:00.82 :sh
7213 mingo 20 0 1434m 8080 888 D 19.2 0.1 0:00.74 :sh
7194 mingo 20 0 1434m 8080 888 D 18.6 0.1 0:00.72 :make
There's still various kinks in it - more patches to come.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static int process_sched_migrate_task_event(struct perf_tool *tool,
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
struct perf_evsel *evsel,
|
2012-09-10 22:15:03 +00:00
|
|
|
struct perf_sample *sample,
|
2012-09-11 16:18:47 +00:00
|
|
|
struct machine *machine)
|
2009-10-10 12:46:04 +00:00
|
|
|
{
|
2012-09-11 20:29:27 +00:00
|
|
|
struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
|
2009-10-10 12:46:04 +00:00
|
|
|
|
perf sched: Don't read all tracepoint variables in advance
Do it just at the actual consumer of these fields, that way we avoid
needless lookups:
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
98.848272 task-clock # 0.993 CPUs utilized ( +- 0.48% )
11 context-switches # 0.112 K/sec ( +- 2.83% )
0 cpu-migrations # 0.003 K/sec ( +- 50.92% )
7,604 page-faults # 0.077 M/sec ( +- 0.00% )
332,216,085 cycles # 3.361 GHz ( +- 0.14% ) [82.87%]
100,623,710 stalled-cycles-frontend # 30.29% frontend cycles idle ( +- 0.53% ) [82.95%]
58,788,692 stalled-cycles-backend # 17.70% backend cycles idle ( +- 0.59% ) [67.15%]
609,402,433 instructions # 1.83 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.76%]
131,277,138 branches # 1328.067 M/sec ( +- 0.06% ) [83.77%]
1,117,871 branch-misses # 0.85% of all branches ( +- 0.32% ) [83.51%]
0.099580430 seconds time elapsed ( +- 0.48% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kracdpw8wqlr0xjh75uk8g11@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
if (sched->tp_handler->migrate_task_event)
|
|
|
|
return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine);
|
2012-09-09 01:53:06 +00:00
|
|
|
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
return 0;
|
2009-10-10 12:46:04 +00:00
|
|
|
}
|
|
|
|
|
2012-09-09 01:53:06 +00:00
|
|
|
typedef int (*tracepoint_handler)(struct perf_tool *tool,
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
struct perf_evsel *evsel,
|
2012-09-09 01:53:06 +00:00
|
|
|
struct perf_sample *sample,
|
2012-09-11 16:18:47 +00:00
|
|
|
struct machine *machine);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-10 22:15:03 +00:00
|
|
|
static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused,
|
|
|
|
union perf_event *event __maybe_unused,
|
perf tools: Save some loops using perf_evlist__id2evsel
Since we already ask for PERF_SAMPLE_ID and use it to quickly find the
associated evsel, add handler func + data to struct perf_evsel to avoid
using chains of if(strcmp(event_name)) and also to avoid all the linear
list searches via trace_event_find.
To demonstrate the technique convert 'perf sched' to it:
# perf sched record sleep 5m
And then:
Performance counter stats for '/tmp/oldperf sched lat':
646.929438 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,901 page-faults # 0.032 M/sec
1,290,144,450 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,606,158,439 instructions # 1.24 insns per cycle
339,088,395 branches # 524.151 M/sec
4,550,735 branch-misses # 1.34% of all branches
0.647524759 seconds time elapsed
Versus:
Performance counter stats for 'perf sched lat':
473.564691 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,903 page-faults # 0.044 M/sec
944,367,984 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,442,385,571 instructions # 1.53 insns per cycle
308,383,106 branches # 651.195 M/sec
4,481,784 branch-misses # 1.45% of all branches
0.474215751 seconds time elapsed
[root@emilia ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-1kbzpl74lwi6lavpqke2u2p3@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-11-28 19:57:40 +00:00
|
|
|
struct perf_sample *sample,
|
|
|
|
struct perf_evsel *evsel,
|
|
|
|
struct machine *machine)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2012-09-09 01:53:06 +00:00
|
|
|
int err = 0;
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2013-11-06 13:17:38 +00:00
|
|
|
if (evsel->handler != NULL) {
|
|
|
|
tracepoint_handler f = evsel->handler;
|
perf sched: Use perf_evsel__{int,str}val
This patch also stops reading the common fields, as they were not being used except
for one ->common_pid case that was replaced by sample->tid, i.e. the info is already
in the perf_sample struct.
Also it only fills the _event structures when there is a handler.
[root@sandy ~]# perf sched record sleep 30s
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 8.585 MB perf.data (~375063 samples) ]
Before:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
129.117838 task-clock # 0.994 CPUs utilized ( +- 0.28% )
14 context-switches # 0.111 K/sec ( +- 2.10% )
0 cpu-migrations # 0.002 K/sec ( +- 66.67% )
7,654 page-faults # 0.059 M/sec ( +- 0.67% )
438,121,661 cycles # 3.393 GHz ( +- 0.06% ) [83.06%]
150,808,605 stalled-cycles-frontend # 34.42% frontend cycles idle ( +- 0.14% ) [83.10%]
80,748,941 stalled-cycles-backend # 18.43% backend cycles idle ( +- 0.64% ) [66.73%]
758,605,879 instructions # 1.73 insns per cycle
# 0.20 stalled cycles per insn ( +- 0.08% ) [83.54%]
162,164,321 branches # 1255.940 M/sec ( +- 0.10% ) [83.70%]
1,609,903 branch-misses # 0.99% of all branches ( +- 0.08% ) [83.62%]
0.129949153 seconds time elapsed ( +- 0.28% )
After:
[root@sandy ~]# perf stat -r 10 perf sched lat > /dev/null
Performance counter stats for 'perf sched lat' (10 runs):
103.592215 task-clock # 0.993 CPUs utilized ( +- 0.33% )
12 context-switches # 0.114 K/sec ( +- 3.29% )
0 cpu-migrations # 0.000 K/sec
7,605 page-faults # 0.073 M/sec ( +- 0.00% )
345,796,112 cycles # 3.338 GHz ( +- 0.07% ) [82.90%]
106,876,796 stalled-cycles-frontend # 30.91% frontend cycles idle ( +- 0.38% ) [83.23%]
62,060,877 stalled-cycles-backend # 17.95% backend cycles idle ( +- 0.80% ) [67.14%]
628,246,586 instructions # 1.82 insns per cycle
# 0.17 stalled cycles per insn ( +- 0.04% ) [83.64%]
134,962,057 branches # 1302.820 M/sec ( +- 0.10% ) [83.64%]
1,233,037 branch-misses # 0.91% of all branches ( +- 0.29% ) [83.41%]
0.104333272 seconds time elapsed ( +- 0.33% )
[root@sandy ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-weu9t63zkrfrazkn0gxj48xy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-11 22:29:17 +00:00
|
|
|
err = f(tool, evsel, sample, machine);
|
perf tools: Save some loops using perf_evlist__id2evsel
Since we already ask for PERF_SAMPLE_ID and use it to quickly find the
associated evsel, add handler func + data to struct perf_evsel to avoid
using chains of if(strcmp(event_name)) and also to avoid all the linear
list searches via trace_event_find.
To demonstrate the technique convert 'perf sched' to it:
# perf sched record sleep 5m
And then:
Performance counter stats for '/tmp/oldperf sched lat':
646.929438 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,901 page-faults # 0.032 M/sec
1,290,144,450 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,606,158,439 instructions # 1.24 insns per cycle
339,088,395 branches # 524.151 M/sec
4,550,735 branch-misses # 1.34% of all branches
0.647524759 seconds time elapsed
Versus:
Performance counter stats for 'perf sched lat':
473.564691 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,903 page-faults # 0.044 M/sec
944,367,984 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,442,385,571 instructions # 1.53 insns per cycle
308,383,106 branches # 651.195 M/sec
4,481,784 branch-misses # 1.45% of all branches
0.474215751 seconds time elapsed
[root@emilia ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-1kbzpl74lwi6lavpqke2u2p3@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-11-28 19:57:40 +00:00
|
|
|
}
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2012-09-09 01:53:06 +00:00
|
|
|
return err;
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2015-03-03 01:28:41 +00:00
|
|
|
static int perf_sched__read_events(struct perf_sched *sched)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
perf tools: Save some loops using perf_evlist__id2evsel
Since we already ask for PERF_SAMPLE_ID and use it to quickly find the
associated evsel, add handler func + data to struct perf_evsel to avoid
using chains of if(strcmp(event_name)) and also to avoid all the linear
list searches via trace_event_find.
To demonstrate the technique convert 'perf sched' to it:
# perf sched record sleep 5m
And then:
Performance counter stats for '/tmp/oldperf sched lat':
646.929438 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,901 page-faults # 0.032 M/sec
1,290,144,450 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,606,158,439 instructions # 1.24 insns per cycle
339,088,395 branches # 524.151 M/sec
4,550,735 branch-misses # 1.34% of all branches
0.647524759 seconds time elapsed
Versus:
Performance counter stats for 'perf sched lat':
473.564691 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,903 page-faults # 0.044 M/sec
944,367,984 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,442,385,571 instructions # 1.53 insns per cycle
308,383,106 branches # 651.195 M/sec
4,481,784 branch-misses # 1.45% of all branches
0.474215751 seconds time elapsed
[root@emilia ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-1kbzpl74lwi6lavpqke2u2p3@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-11-28 19:57:40 +00:00
|
|
|
const struct perf_evsel_str_handler handlers[] = {
|
|
|
|
{ "sched:sched_switch", process_sched_switch_event, },
|
|
|
|
{ "sched:sched_stat_runtime", process_sched_runtime_event, },
|
|
|
|
{ "sched:sched_wakeup", process_sched_wakeup_event, },
|
|
|
|
{ "sched:sched_wakeup_new", process_sched_wakeup_event, },
|
|
|
|
{ "sched:sched_migrate_task", process_sched_migrate_task_event, },
|
|
|
|
};
|
2012-06-27 16:08:42 +00:00
|
|
|
struct perf_session *session;
|
2013-10-15 14:27:32 +00:00
|
|
|
struct perf_data_file file = {
|
|
|
|
.path = input_name,
|
|
|
|
.mode = PERF_DATA_MODE_READ,
|
perf sched replay: Support using -f to override perf.data file ownership
Enable to use perf.data when it is not owned by current user or root.
Example:
$ ls -al perf.data
-rw------- 1 Yunlong.Song Yunlong.Song 5321918 Mar 25 15:14 perf.data
$ sudo id
uid=0(root) gid=0(root) groups=0(root),64(pkcs11)
Before this patch:
$ sudo perf sched replay -f
run measurement overhead: 98 nsecs
sleep measurement overhead: 52909 nsecs
the run test took 1000015 nsecs
the sleep test took 1054253 nsecs
File perf.data not owned by current user or root (use -f to override)
As shown above, the -f option does not work at all.
After this patch:
$ sudo perf sched replay -f
run measurement overhead: 221 nsecs
sleep measurement overhead: 40514 nsecs
the run test took 1000003 nsecs
the sleep test took 1056098 nsecs
nr_run_events: 10
nr_sleep_events: 1562
nr_wakeup_events: 5
task 0 ( :1: 1), nr_events: 1
task 1 ( :2: 2), nr_events: 1
task 2 ( :3: 3), nr_events: 1
...
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
------------------------------------------------------------
#1 : 50.198, ravg: 50.20, cpu: 2335.18 / 2335.18
#2 : 219.099, ravg: 67.09, cpu: 2835.11 / 2385.17
#3 : 238.626, ravg: 84.24, cpu: 3278.26 / 2474.48
#4 : 200.364, ravg: 95.85, cpu: 2977.41 / 2524.77
#5 : 176.882, ravg: 103.96, cpu: 2801.35 / 2552.43
#6 : 191.093, ravg: 112.67, cpu: 2813.70 / 2578.56
#7 : 189.448, ravg: 120.35, cpu: 2809.21 / 2601.62
#8 : 200.637, ravg: 128.38, cpu: 2849.91 / 2626.45
#9 : 248.338, ravg: 140.37, cpu: 4380.61 / 2801.87
#10 : 511.139, ravg: 177.45, cpu: 3077.73 / 2829.45
As shown above, the -f option really works now.
Besides for replay, -f option can also work for latency and map.
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-9-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:35 +00:00
|
|
|
.force = sched->force,
|
2013-10-15 14:27:32 +00:00
|
|
|
};
|
2015-03-03 01:28:41 +00:00
|
|
|
int rc = -1;
|
2012-06-27 16:08:42 +00:00
|
|
|
|
2013-10-15 14:27:32 +00:00
|
|
|
session = perf_session__new(&file, false, &sched->tool);
|
2012-09-09 01:53:06 +00:00
|
|
|
if (session == NULL) {
|
|
|
|
pr_debug("No Memory for session\n");
|
|
|
|
return -1;
|
|
|
|
}
|
2009-12-11 23:24:02 +00:00
|
|
|
|
2014-08-12 06:40:45 +00:00
|
|
|
symbol__init(&session->header.env);
|
2014-08-12 06:40:41 +00:00
|
|
|
|
2012-09-09 01:53:06 +00:00
|
|
|
if (perf_session__set_tracepoints_handlers(session, handlers))
|
|
|
|
goto out_delete;
|
perf tools: Save some loops using perf_evlist__id2evsel
Since we already ask for PERF_SAMPLE_ID and use it to quickly find the
associated evsel, add handler func + data to struct perf_evsel to avoid
using chains of if(strcmp(event_name)) and also to avoid all the linear
list searches via trace_event_find.
To demonstrate the technique convert 'perf sched' to it:
# perf sched record sleep 5m
And then:
Performance counter stats for '/tmp/oldperf sched lat':
646.929438 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,901 page-faults # 0.032 M/sec
1,290,144,450 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,606,158,439 instructions # 1.24 insns per cycle
339,088,395 branches # 524.151 M/sec
4,550,735 branch-misses # 1.34% of all branches
0.647524759 seconds time elapsed
Versus:
Performance counter stats for 'perf sched lat':
473.564691 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,903 page-faults # 0.044 M/sec
944,367,984 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,442,385,571 instructions # 1.53 insns per cycle
308,383,106 branches # 651.195 M/sec
4,481,784 branch-misses # 1.45% of all branches
0.474215751 seconds time elapsed
[root@emilia ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-1kbzpl74lwi6lavpqke2u2p3@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-11-28 19:57:40 +00:00
|
|
|
|
2010-05-14 16:16:55 +00:00
|
|
|
if (perf_session__has_traces(session, "record -R")) {
|
2015-03-03 14:58:45 +00:00
|
|
|
int err = perf_session__process_events(session);
|
2012-09-09 01:53:06 +00:00
|
|
|
if (err) {
|
|
|
|
pr_err("Failed to process events, error %d", err);
|
|
|
|
goto out_delete;
|
|
|
|
}
|
2011-08-08 21:03:34 +00:00
|
|
|
|
2015-02-14 17:50:11 +00:00
|
|
|
sched->nr_events = session->evlist->stats.nr_events[0];
|
|
|
|
sched->nr_lost_events = session->evlist->stats.total_lost;
|
|
|
|
sched->nr_lost_chunks = session->evlist->stats.nr_events[PERF_RECORD_LOST];
|
2010-05-14 16:16:55 +00:00
|
|
|
}
|
2009-12-27 23:37:02 +00:00
|
|
|
|
2015-03-03 01:28:41 +00:00
|
|
|
rc = 0;
|
2012-09-09 01:53:06 +00:00
|
|
|
out_delete:
|
|
|
|
perf_session__delete(session);
|
2015-03-03 01:28:41 +00:00
|
|
|
return rc;
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void print_bad_events(struct perf_sched *sched)
|
2009-09-16 15:40:48 +00:00
|
|
|
{
|
2012-09-11 20:29:27 +00:00
|
|
|
if (sched->nr_unordered_timestamps && sched->nr_timestamps) {
|
2009-09-16 15:40:48 +00:00
|
|
|
printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n",
|
2012-09-11 20:29:27 +00:00
|
|
|
(double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0,
|
|
|
|
sched->nr_unordered_timestamps, sched->nr_timestamps);
|
2009-09-16 15:40:48 +00:00
|
|
|
}
|
2012-09-11 20:29:27 +00:00
|
|
|
if (sched->nr_lost_events && sched->nr_events) {
|
2009-09-16 15:40:48 +00:00
|
|
|
printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
|
2012-09-11 20:29:27 +00:00
|
|
|
(double)sched->nr_lost_events/(double)sched->nr_events * 100.0,
|
|
|
|
sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks);
|
2009-09-16 15:40:48 +00:00
|
|
|
}
|
2012-09-11 20:29:27 +00:00
|
|
|
if (sched->nr_context_switch_bugs && sched->nr_timestamps) {
|
2009-09-16 15:40:48 +00:00
|
|
|
printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)",
|
2012-09-11 20:29:27 +00:00
|
|
|
(double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0,
|
|
|
|
sched->nr_context_switch_bugs, sched->nr_timestamps);
|
|
|
|
if (sched->nr_lost_events)
|
2009-09-16 15:40:48 +00:00
|
|
|
printf(" (due to lost events?)");
|
|
|
|
printf("\n");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static int perf_sched__lat(struct perf_sched *sched)
|
2009-09-16 15:40:48 +00:00
|
|
|
{
|
|
|
|
struct rb_node *next;
|
|
|
|
|
|
|
|
setup_pager();
|
2013-08-08 02:50:44 +00:00
|
|
|
|
2015-03-03 01:28:41 +00:00
|
|
|
if (perf_sched__read_events(sched))
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
2013-08-08 02:50:44 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
perf_sched__sort_lat(sched);
|
2009-09-16 15:40:48 +00:00
|
|
|
|
2014-03-17 14:18:21 +00:00
|
|
|
printf("\n -----------------------------------------------------------------------------------------------------------------\n");
|
|
|
|
printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n");
|
|
|
|
printf(" -----------------------------------------------------------------------------------------------------------------\n");
|
2009-09-16 15:40:48 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
next = rb_first(&sched->sorted_atom_root);
|
2009-09-16 15:40:48 +00:00
|
|
|
|
|
|
|
while (next) {
|
|
|
|
struct work_atoms *work_list;
|
|
|
|
|
|
|
|
work_list = rb_entry(next, struct work_atoms, node);
|
2012-09-11 20:29:27 +00:00
|
|
|
output_lat_thread(sched, work_list);
|
2009-09-16 15:40:48 +00:00
|
|
|
next = rb_next(next);
|
2015-03-03 01:28:41 +00:00
|
|
|
thread__zput(work_list->thread);
|
2009-09-16 15:40:48 +00:00
|
|
|
}
|
|
|
|
|
2014-03-17 14:18:21 +00:00
|
|
|
printf(" -----------------------------------------------------------------------------------------------------------------\n");
|
2011-01-22 22:37:02 +00:00
|
|
|
printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n",
|
2012-09-11 20:29:27 +00:00
|
|
|
(double)sched->all_runtime / 1e6, sched->all_count);
|
2009-09-16 15:40:48 +00:00
|
|
|
|
|
|
|
printf(" ---------------------------------------------------\n");
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
print_bad_events(sched);
|
2009-09-16 15:40:48 +00:00
|
|
|
printf("\n");
|
|
|
|
|
2012-09-09 01:53:06 +00:00
|
|
|
return 0;
|
2009-09-16 15:40:48 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static int perf_sched__map(struct perf_sched *sched)
|
2009-09-16 15:40:48 +00:00
|
|
|
{
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->max_cpu = sysconf(_SC_NPROCESSORS_CONF);
|
2009-09-17 16:24:55 +00:00
|
|
|
|
2009-09-16 15:40:48 +00:00
|
|
|
setup_pager();
|
2015-03-03 01:28:41 +00:00
|
|
|
if (perf_sched__read_events(sched))
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
2012-09-11 20:29:27 +00:00
|
|
|
print_bad_events(sched);
|
2012-09-09 01:53:06 +00:00
|
|
|
return 0;
|
2009-09-16 15:40:48 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static int perf_sched__replay(struct perf_sched *sched)
|
2009-09-16 15:40:48 +00:00
|
|
|
{
|
|
|
|
unsigned long i;
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
calibrate_run_measurement_overhead(sched);
|
|
|
|
calibrate_sleep_measurement_overhead(sched);
|
2009-09-16 15:40:48 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
test_calibrations(sched);
|
2009-09-16 15:40:48 +00:00
|
|
|
|
2015-03-03 01:28:41 +00:00
|
|
|
if (perf_sched__read_events(sched))
|
2012-09-09 01:53:06 +00:00
|
|
|
return -1;
|
2009-09-16 15:40:48 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
printf("nr_run_events: %ld\n", sched->nr_run_events);
|
|
|
|
printf("nr_sleep_events: %ld\n", sched->nr_sleep_events);
|
|
|
|
printf("nr_wakeup_events: %ld\n", sched->nr_wakeup_events);
|
2009-09-16 15:40:48 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
if (sched->targetless_wakeups)
|
|
|
|
printf("target-less wakeups: %ld\n", sched->targetless_wakeups);
|
|
|
|
if (sched->multitarget_wakeups)
|
|
|
|
printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups);
|
|
|
|
if (sched->nr_run_events_optimized)
|
2009-09-16 15:40:48 +00:00
|
|
|
printf("run atoms optimized: %ld\n",
|
2012-09-11 20:29:27 +00:00
|
|
|
sched->nr_run_events_optimized);
|
2009-09-16 15:40:48 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
print_task_traces(sched);
|
|
|
|
add_cross_task_wakeups(sched);
|
2009-09-16 15:40:48 +00:00
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
create_tasks(sched);
|
2009-09-16 15:40:48 +00:00
|
|
|
printf("------------------------------------------------------------\n");
|
2012-09-11 20:29:27 +00:00
|
|
|
for (i = 0; i < sched->replay_repeat; i++)
|
|
|
|
run_one_test(sched);
|
2012-09-09 01:53:06 +00:00
|
|
|
|
|
|
|
return 0;
|
2009-09-16 15:40:48 +00:00
|
|
|
}
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
static void setup_sorting(struct perf_sched *sched, const struct option *options,
|
|
|
|
const char * const usage_msg[])
|
2009-09-13 01:36:29 +00:00
|
|
|
{
|
2012-09-11 20:29:27 +00:00
|
|
|
char *tmp, *tok, *str = strdup(sched->sort_order);
|
2009-09-13 01:36:29 +00:00
|
|
|
|
|
|
|
for (tok = strtok_r(str, ", ", &tmp);
|
|
|
|
tok; tok = strtok_r(NULL, ", ", &tmp)) {
|
2012-09-11 20:29:27 +00:00
|
|
|
if (sort_dimension__add(tok, &sched->sort_list) < 0) {
|
2009-09-13 01:36:29 +00:00
|
|
|
error("Unknown --sort key: `%s'", tok);
|
2012-09-11 20:29:27 +00:00
|
|
|
usage_with_options(usage_msg, options);
|
2009-09-13 01:36:29 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
free(str);
|
|
|
|
|
2012-09-11 20:29:27 +00:00
|
|
|
sort_dimension__add("pid", &sched->cmp_pid);
|
2009-09-13 01:36:29 +00:00
|
|
|
}
|
|
|
|
|
2009-09-13 07:44:29 +00:00
|
|
|
static int __cmd_record(int argc, const char **argv)
|
|
|
|
{
|
|
|
|
unsigned int rec_argc, i, j;
|
|
|
|
const char **rec_argv;
|
2012-09-11 20:29:27 +00:00
|
|
|
const char * const record_args[] = {
|
|
|
|
"record",
|
|
|
|
"-a",
|
|
|
|
"-R",
|
|
|
|
"-m", "1024",
|
|
|
|
"-c", "1",
|
|
|
|
"-e", "sched:sched_switch",
|
|
|
|
"-e", "sched:sched_stat_wait",
|
|
|
|
"-e", "sched:sched_stat_sleep",
|
|
|
|
"-e", "sched:sched_stat_iowait",
|
|
|
|
"-e", "sched:sched_stat_runtime",
|
|
|
|
"-e", "sched:sched_process_fork",
|
|
|
|
"-e", "sched:sched_wakeup",
|
2014-05-05 07:05:53 +00:00
|
|
|
"-e", "sched:sched_wakeup_new",
|
2012-09-11 20:29:27 +00:00
|
|
|
"-e", "sched:sched_migrate_task",
|
|
|
|
};
|
2009-09-13 07:44:29 +00:00
|
|
|
|
|
|
|
rec_argc = ARRAY_SIZE(record_args) + argc - 1;
|
|
|
|
rec_argv = calloc(rec_argc + 1, sizeof(char *));
|
|
|
|
|
2011-01-10 12:48:47 +00:00
|
|
|
if (rec_argv == NULL)
|
2010-11-13 02:35:06 +00:00
|
|
|
return -ENOMEM;
|
|
|
|
|
2009-09-13 07:44:29 +00:00
|
|
|
for (i = 0; i < ARRAY_SIZE(record_args); i++)
|
|
|
|
rec_argv[i] = strdup(record_args[i]);
|
|
|
|
|
|
|
|
for (j = 1; j < (unsigned int)argc; j++, i++)
|
|
|
|
rec_argv[i] = argv[j];
|
|
|
|
|
|
|
|
BUG_ON(i != rec_argc);
|
|
|
|
|
|
|
|
return cmd_record(i, rec_argv, NULL);
|
|
|
|
}
|
|
|
|
|
2012-09-10 22:15:03 +00:00
|
|
|
int cmd_sched(int argc, const char **argv, const char *prefix __maybe_unused)
|
2009-09-11 10:12:54 +00:00
|
|
|
{
|
2013-10-22 07:34:15 +00:00
|
|
|
const char default_sort_order[] = "avg, max, switch, runtime";
|
|
|
|
struct perf_sched sched = {
|
|
|
|
.tool = {
|
|
|
|
.sample = perf_sched__process_tracepoint_sample,
|
|
|
|
.comm = perf_event__process_comm,
|
|
|
|
.lost = perf_event__process_lost,
|
|
|
|
.fork = perf_sched__process_fork_event,
|
2014-07-06 12:18:21 +00:00
|
|
|
.ordered_events = true,
|
2013-10-22 07:34:15 +00:00
|
|
|
},
|
|
|
|
.cmp_pid = LIST_HEAD_INIT(sched.cmp_pid),
|
|
|
|
.sort_list = LIST_HEAD_INIT(sched.sort_list),
|
|
|
|
.start_work_mutex = PTHREAD_MUTEX_INITIALIZER,
|
|
|
|
.work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER,
|
|
|
|
.sort_order = default_sort_order,
|
|
|
|
.replay_repeat = 10,
|
|
|
|
.profile_cpu = -1,
|
|
|
|
.next_shortname1 = 'A',
|
|
|
|
.next_shortname2 = '0',
|
|
|
|
};
|
2012-09-11 20:29:27 +00:00
|
|
|
const struct option latency_options[] = {
|
|
|
|
OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]",
|
|
|
|
"sort by key(s): runtime, switch, avg, max"),
|
|
|
|
OPT_INCR('v', "verbose", &verbose,
|
|
|
|
"be more verbose (show symbol address, etc)"),
|
|
|
|
OPT_INTEGER('C', "CPU", &sched.profile_cpu,
|
|
|
|
"CPU to profile on"),
|
|
|
|
OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
|
|
|
|
"dump raw trace in ASCII"),
|
|
|
|
OPT_END()
|
|
|
|
};
|
|
|
|
const struct option replay_options[] = {
|
|
|
|
OPT_UINTEGER('r', "repeat", &sched.replay_repeat,
|
|
|
|
"repeat the workload replay N times (-1: infinite)"),
|
|
|
|
OPT_INCR('v', "verbose", &verbose,
|
|
|
|
"be more verbose (show symbol address, etc)"),
|
|
|
|
OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
|
|
|
|
"dump raw trace in ASCII"),
|
perf sched replay: Fix the EMFILE error caused by the limitation of the maximum open files
The soft maximum number of open files for a calling process is 1024,
which is defined as INR_OPEN_CUR in include/uapi/linux/fs.h, and the
hard maximum number of open files for a calling process is 4096, which
is defined as INR_OPEN_MAX in include/uapi/linux/fs.h.
Both INR_OPEN_CUR and INR_OPEN_MAX are used to limit the value of
RLIMIT_NOFILE in include/asm-generic/resource.h.
And the soft maximum number finally decides the limitation of the
maximum files which are allowed to be opened.
That is to say a process can use at most 1024 file descriptors for its
o pened files, or an EMFILE error will happen.
This error can be fixed by increasing the soft maximum number, under the
constraint that the soft maximum number can not exceed the hard maximum
number, or both soft and hard maximum number should be increased
simultaneously with privilege.
For perf sched replay, it uses sys_perf_event_open to create the file
descriptor for each of the tasks in order to handle information of perf
events.
That is to say each task needs a unique file descriptor. In x86_64,
there may be over 1024 or 4096 tasks correspoinding to the record in
perf.data, which causes that no enough file descriptors can be used.
As a result, EMFILE error happens and stops the replay process. To solve
this problem, we adaptively increase the soft and hard maximum number of
open files with a '-f' option.
Example:
Test environment: x86_64 with 160 cores
$ cat /proc/sys/kernel/pid_max
163840
$ cat /proc/sys/fs/file-max
6815744
$ ulimit -Sn
1024
$ ulimit -Hn
4096
Before this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
After this patch:
$ perf sched replay
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
Error: sys_perf_event_open() syscall returned with -1 (Too many open
files)
Have a try with -f option
$ perf sched replay -f
...
task 1549 ( :163132: 163132), nr_events: 1
task 1550 ( :163540: 163540), nr_events: 1
task 1551 ( <unknown>: 0), nr_events: 10
------------------------------------------------------------
#1 : 54.401, ravg: 54.40, cpu: 3285.21 / 3285.21
#2 : 199.548, ravg: 68.92, cpu: 4999.65 / 3456.66
#3 : 170.483, ravg: 79.07, cpu: 1349.94 / 3245.99
#4 : 192.034, ravg: 90.37, cpu: 1322.88 / 3053.67
#5 : 182.929, ravg: 99.62, cpu: 1406.51 / 2888.96
#6 : 152.974, ravg: 104.96, cpu: 1167.54 / 2716.82
#7 : 155.579, ravg: 110.02, cpu: 2992.53 / 2744.39
#8 : 130.557, ravg: 112.08, cpu: 1126.43 / 2582.59
#9 : 138.520, ravg: 114.72, cpu: 1253.22 / 2449.65
#10 : 134.328, ravg: 116.68, cpu: 1587.95 / 2363.48
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-8-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-03-31 13:46:34 +00:00
|
|
|
OPT_BOOLEAN('f', "force", &sched.force, "don't complain, do it"),
|
2012-09-11 20:29:27 +00:00
|
|
|
OPT_END()
|
|
|
|
};
|
|
|
|
const struct option sched_options[] = {
|
2012-10-30 03:56:02 +00:00
|
|
|
OPT_STRING('i', "input", &input_name, "file",
|
2012-09-11 20:29:27 +00:00
|
|
|
"input file name"),
|
|
|
|
OPT_INCR('v', "verbose", &verbose,
|
|
|
|
"be more verbose (show symbol address, etc)"),
|
|
|
|
OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
|
|
|
|
"dump raw trace in ASCII"),
|
|
|
|
OPT_END()
|
|
|
|
};
|
|
|
|
const char * const latency_usage[] = {
|
|
|
|
"perf sched latency [<options>]",
|
|
|
|
NULL
|
|
|
|
};
|
|
|
|
const char * const replay_usage[] = {
|
|
|
|
"perf sched replay [<options>]",
|
|
|
|
NULL
|
|
|
|
};
|
2014-03-15 03:17:54 +00:00
|
|
|
const char *const sched_subcommands[] = { "record", "latency", "map",
|
|
|
|
"replay", "script", NULL };
|
|
|
|
const char *sched_usage[] = {
|
|
|
|
NULL,
|
2012-09-11 20:29:27 +00:00
|
|
|
NULL
|
|
|
|
};
|
|
|
|
struct trace_sched_handler lat_ops = {
|
|
|
|
.wakeup_event = latency_wakeup_event,
|
|
|
|
.switch_event = latency_switch_event,
|
|
|
|
.runtime_event = latency_runtime_event,
|
|
|
|
.migrate_task_event = latency_migrate_task_event,
|
|
|
|
};
|
|
|
|
struct trace_sched_handler map_ops = {
|
|
|
|
.switch_event = map_switch_event,
|
|
|
|
};
|
|
|
|
struct trace_sched_handler replay_ops = {
|
|
|
|
.wakeup_event = replay_wakeup_event,
|
|
|
|
.switch_event = replay_switch_event,
|
|
|
|
.fork_event = replay_fork_event,
|
|
|
|
};
|
2013-10-22 07:34:16 +00:00
|
|
|
unsigned int i;
|
|
|
|
|
|
|
|
for (i = 0; i < ARRAY_SIZE(sched.curr_pid); i++)
|
|
|
|
sched.curr_pid[i] = -1;
|
2012-09-11 20:29:27 +00:00
|
|
|
|
2014-03-15 03:17:54 +00:00
|
|
|
argc = parse_options_subcommand(argc, argv, sched_options, sched_subcommands,
|
|
|
|
sched_usage, PARSE_OPT_STOP_AT_NON_OPTION);
|
2009-09-11 10:12:54 +00:00
|
|
|
if (!argc)
|
|
|
|
usage_with_options(sched_usage, sched_options);
|
2009-09-11 10:12:54 +00:00
|
|
|
|
2009-12-07 04:04:49 +00:00
|
|
|
/*
|
2010-11-16 17:45:39 +00:00
|
|
|
* Aliased to 'perf script' for now:
|
2009-12-07 04:04:49 +00:00
|
|
|
*/
|
2010-11-16 17:45:39 +00:00
|
|
|
if (!strcmp(argv[0], "script"))
|
|
|
|
return cmd_script(argc, argv, prefix);
|
2009-12-07 04:04:49 +00:00
|
|
|
|
2009-09-13 07:44:29 +00:00
|
|
|
if (!strncmp(argv[0], "rec", 3)) {
|
|
|
|
return __cmd_record(argc, argv);
|
|
|
|
} else if (!strncmp(argv[0], "lat", 3)) {
|
2012-09-11 20:29:27 +00:00
|
|
|
sched.tp_handler = &lat_ops;
|
2009-09-11 10:12:54 +00:00
|
|
|
if (argc > 1) {
|
|
|
|
argc = parse_options(argc, argv, latency_options, latency_usage, 0);
|
|
|
|
if (argc)
|
|
|
|
usage_with_options(latency_usage, latency_options);
|
|
|
|
}
|
2012-09-11 20:29:27 +00:00
|
|
|
setup_sorting(&sched, latency_options, latency_usage);
|
|
|
|
return perf_sched__lat(&sched);
|
2009-09-16 15:40:48 +00:00
|
|
|
} else if (!strcmp(argv[0], "map")) {
|
2012-09-11 20:29:27 +00:00
|
|
|
sched.tp_handler = &map_ops;
|
|
|
|
setup_sorting(&sched, latency_options, latency_usage);
|
|
|
|
return perf_sched__map(&sched);
|
2009-09-11 10:12:54 +00:00
|
|
|
} else if (!strncmp(argv[0], "rep", 3)) {
|
2012-09-11 20:29:27 +00:00
|
|
|
sched.tp_handler = &replay_ops;
|
2009-09-11 10:12:54 +00:00
|
|
|
if (argc) {
|
|
|
|
argc = parse_options(argc, argv, replay_options, replay_usage, 0);
|
|
|
|
if (argc)
|
|
|
|
usage_with_options(replay_usage, replay_options);
|
|
|
|
}
|
2012-09-11 20:29:27 +00:00
|
|
|
return perf_sched__replay(&sched);
|
2009-09-11 10:12:54 +00:00
|
|
|
} else {
|
|
|
|
usage_with_options(sched_usage, sched_options);
|
|
|
|
}
|
|
|
|
|
2009-09-11 10:12:54 +00:00
|
|
|
return 0;
|
2009-09-11 10:12:54 +00:00
|
|
|
}
|