From: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
To: Stanislaw Gruszka <sgruszka@redhat.com>
Cc: "Ingo Molnar" <mingo@elte.hu>,
"Thomas Gleixner" <tglx@linutronix.de>,
"Peter Zijlstra" <peterz@infradead.org>,
"Spencer Candland" <spencer@bluehost.com>,
"Américo Wang" <xiyou.wangcong@gmail.com>,
linux-kernel@vger.kernel.org, "Oleg Nesterov" <oleg@redhat.com>,
"Balbir Singh" <balbir@in.ibm.com>
Subject: reproducer: utime decreasing
Date: Wed, 02 Dec 2009 17:29:05 +0900 [thread overview]
Message-ID: <4B162551.7050303@jp.fujitsu.com> (raw)
In-Reply-To: <20091123101612.GC25978@dhcp-lab-161.englab.brq.redhat.com>
Following is a sample program that can reproduce the problem
in several ways.
I tested my patch using this reproducer, and confirmed that
applying both of Stanislaw's do_sys_times() patch and my
thread_group_times() patch is required to fix the problem.
Thanks,
H.Seto
===
/*
* Sample program to demonstrate time decreasing on thread exit
*/
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <pthread.h>
#include <sys/times.h>
#define DEFAULT_THREADS 500
unsigned long lpt;
int looptype;
int samplesleep;
int threads;
void *child (void *thread_id)
{
struct tms t[12];
int i, j;
unsigned long ret = 0, id = (unsigned long)thread_id;
if (looptype) {
/*
* discont:
*
* Loop tricky, to make a significant gap between
* task->{u,s}time and task_times().
*
* runtime of a thread = 0.5 tick * 1000 = 500 ms
* task->{u,s}time = (??? , 0)
* task_times() = ideally (500, 0)
*/
for (j = 0; j < 1000; j++) {
for (i = 0; i < lpt/2; i++)
;
usleep(0);
}
} else {
/*
* cont:
*
* Loop without tricks.
*
* runtime of a thread = 500 ms
* task->{u,s}time = (500, 0)
* task_times() = (500, 0)
*/
for (i = 0; i < lpt * 500; i++)
;
}
if (!(id % 4))
/* bother other threads */
pthread_exit((void *)ret);
for (i = 0; i < 12; i++) {
times(&t[i]);
if (samplesleep)
usleep(0);
}
for (i = 0; i + 5 < 12; i++) {
/*
* +----+----+----+----+----+
* i+0 i+1 i+2 i+3 i+4 i+5
* ^^^^^^
* check here
*/
if (t[i+2].tms_utime > t[i+3].tms_utime
|| t[i+2].tms_stime > t[i+3].tms_stime) {
printf("[%4ld] %s decreased %3d: "
"(%d %d) (%d %d) [%d %d]<->[%d %d] (%d %d) (%d %d)\n",
id,
t[i+2].tms_utime > t[i+3].tms_utime ?
"utime" : "stime",
t[i+2].tms_utime > t[i+3].tms_utime ?
t[i+3].tms_utime - t[i+2].tms_utime :
t[i+3].tms_stime - t[i+2].tms_stime,
t[i+0].tms_utime, t[i+0].tms_stime,
t[i+1].tms_utime, t[i+1].tms_stime,
t[i+2].tms_utime, t[i+2].tms_stime,
t[i+3].tms_utime, t[i+3].tms_stime,
t[i+4].tms_utime, t[i+4].tms_stime,
t[i+5].tms_utime, t[i+5].tms_stime);
ret = 1;
}
}
pthread_exit((void *)ret);
}
void get_loops_per_tick(void)
{
struct tms t1, t2;
unsigned long i, mloop = 1000 * 1000 * 1000;
times(&t1);
for (i = 0; i < mloop; i++)
;
times(&t2);
lpt = mloop / ((t2.tms_utime - t1.tms_utime) * 10);
}
void do_test(int c)
{
struct tms t1, t2;
clock_t j1, j2;
pthread_t *th;
unsigned long i, ret = 0;
th = calloc(threads, sizeof(pthread_t));
if (!th)
return;
looptype = !!(c & 0x1) ? 1 : 0;
samplesleep = !!(c & 0x2) ? 1 : 0;
printf("looptype : %s\n", looptype ? "discont" : "cont");
printf("samplesleep : %s\n", samplesleep ? "yes" : "no");
printf(" ## start ##\n");
j1 = times(&t1);
for (i = 0; i < threads; i++)
pthread_create (&th[i], NULL, child, (void *)i);
for (i = 0; i < threads; i++) {
int r;
pthread_join(th[i], (void *)&r);
ret += (int)r;
}
j2 = times(&t2);
printf(" ## done. ##\n");
printf(" loop total:\n");
printf(" user : %7d ms\n", (t2.tms_utime - t1.tms_utime) * 10);
printf(" system : %7d ms\n", (t2.tms_stime - t1.tms_stime) * 10);
printf(" elapse : %7d ms\n", (j2 - j1) * 10);
printf(" error : %d\n\n", ret);
printf("result: %s\n\n", ret ? "BAD" : "GOOD");
}
int main(int argc, char **argv)
{
int i;
threads = argc > 1 ? atoi(argv[1]) : DEFAULT_THREADS;
printf("### Prep:\n");
get_loops_per_tick();
printf("loops_per_tick: %ld\n", lpt);
printf("threads : %d\n\n", threads);
printf("### Test:\n");
for (i = 0; i < 4; i++)
do_test(i);
}
next prev parent reply other threads:[~2009-12-02 8:29 UTC|newest]
Thread overview: 63+ messages / expand[flat|nested] mbox.gz Atom feed top
2009-11-04 0:23 utime/stime decreasing on thread exit Spencer Candland
2009-11-04 6:49 ` Hidetoshi Seto
2009-11-05 5:24 ` Hidetoshi Seto
2009-11-09 14:49 ` Peter Zijlstra
2009-11-09 17:20 ` Oleg Nesterov
2009-11-09 17:27 ` Oleg Nesterov
2009-11-09 17:31 ` Peter Zijlstra
2009-11-09 19:23 ` Oleg Nesterov
2009-11-09 19:32 ` Peter Zijlstra
2009-11-10 10:44 ` Stanislaw Gruszka
2009-11-10 17:40 ` Oleg Nesterov
2009-11-10 18:24 ` Stanislaw Gruszka
2009-11-10 19:23 ` Oleg Nesterov
2009-11-17 12:48 ` Stanislaw Gruszka
2009-11-17 12:57 ` [PATCH] posix-cpu-timers: reset expire cache when no timer is running Stanislaw Gruszka
2009-11-10 5:42 ` utime/stime decreasing on thread exit Hidetoshi Seto
2009-11-10 5:47 ` [PATCH] fix granularity of task_u/stime() Hidetoshi Seto
2009-11-11 12:11 ` Stanislaw Gruszka
2009-11-12 0:00 ` Hidetoshi Seto
2009-11-12 2:49 ` Hidetoshi Seto
2009-11-12 2:55 ` Américo Wang
2009-11-12 4:16 ` Hidetoshi Seto
2009-11-12 4:33 ` [PATCH] fix granularity of task_u/stime(), v2 Hidetoshi Seto
2009-11-12 14:15 ` Peter Zijlstra
2009-11-12 14:49 ` Stanislaw Gruszka
2009-11-12 15:00 ` Peter Zijlstra
2009-11-12 15:40 ` Stanislaw Gruszka
2009-11-13 12:42 ` [PATCH] sys_times: fix utime/stime decreasing on thread exit Stanislaw Gruszka
2009-11-13 13:16 ` Peter Zijlstra
2009-11-13 14:12 ` Balbir Singh
2009-11-13 15:36 ` Stanislaw Gruszka
2009-11-13 17:05 ` Peter Zijlstra
2009-11-16 19:32 ` [PATCH] fix granularity of task_u/stime(), v2 Spencer Candland
2009-11-17 13:08 ` Stanislaw Gruszka
2009-11-17 13:24 ` Peter Zijlstra
2009-11-19 18:17 ` Stanislaw Gruszka
2009-11-20 2:00 ` Hidetoshi Seto
2009-11-23 10:09 ` Stanislaw Gruszka
2009-11-23 10:16 ` [PATCH] cputime: avoid do_sys_times() races with __exit_signal() Stanislaw Gruszka
2009-11-30 9:20 ` [PATCH 1/2] cputime: remove prev_{u,s}time if VIRT_CPU_ACCOUNTING Hidetoshi Seto
2009-11-30 9:21 ` [PATCH 2/2] cputime: introduce thread_group_times() Hidetoshi Seto
2009-11-30 14:54 ` Stanislaw Gruszka
2009-12-01 1:02 ` Hidetoshi Seto
2009-12-02 8:26 ` [PATCH -v2 1/2] sched, cputime: cleanups related to task_times() Hidetoshi Seto
2009-12-02 15:17 ` Peter Zijlstra
2009-12-02 15:29 ` Balbir Singh
2009-12-03 0:21 ` Hidetoshi Seto
2009-12-02 15:57 ` Peter Zijlstra
2009-12-02 17:33 ` [tip:sched/core] sched, cputime: Cleanups " tip-bot for Hidetoshi Seto
2009-12-02 8:28 ` [PATCH -v2 2/2] sched, cputime: introduce thread_group_times() Hidetoshi Seto
2009-12-02 15:58 ` Peter Zijlstra
2009-12-02 17:33 ` [tip:sched/core] sched, cputime: Introduce thread_group_times() tip-bot for Hidetoshi Seto
2009-12-02 8:29 ` Hidetoshi Seto [this message]
2009-12-02 8:32 ` reproducer: invisible utime Hidetoshi Seto
2009-11-23 10:25 ` [PATCH] fix granularity of task_u/stime(), v2 Balbir Singh
2009-11-23 10:46 ` Stanislaw Gruszka
2009-11-24 5:33 ` Hidetoshi Seto
2009-11-18 22:38 ` Spencer Candland
2009-11-23 9:52 ` Stanislaw Gruszka
2009-11-12 18:12 ` [tip:sched/core] sched: Fix granularity of task_u/stime() tip-bot for Hidetoshi Seto
2009-11-13 9:40 ` Stanislaw Gruszka
2009-11-13 23:09 ` Ingo Molnar
2009-11-16 2:44 ` Hidetoshi Seto
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=4B162551.7050303@jp.fujitsu.com \
--to=seto.hidetoshi@jp.fujitsu.com \
--cc=balbir@in.ibm.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=oleg@redhat.com \
--cc=peterz@infradead.org \
--cc=sgruszka@redhat.com \
--cc=spencer@bluehost.com \
--cc=tglx@linutronix.de \
--cc=xiyou.wangcong@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
Powered by JetHome