Avoid process cputimer state oscillation
PATCH 1/1 Avoid process cputimer state oscillation
When a periodic process timer is fired, a signal is generated. Rearming the timer, if necessary, will be performed in a second step when the signal will be delivered.
Hence, checking the list of process timers list to decide to stop to cpu timer right after generating the signal is premature and is leading the cputimer to oscillate between the on/off states when a single process timer is present.
The following posix-cpu-timers module intrumentation
In void thread_group_cputimer(struct task_s
raw_spin_lock_irqsave(&cputimer->lock, flags);
cputimer->running = 1; update_gt_cputime(&cputimer->cputime, &sum);
+ printk(KERN_DEBUG "cputimer started\n");
} else
raw_spin_lock_irqsave(&cputimer->lock, flags);
*times = cputimer->cputime;
and in static void stop_process_timers(struct s
raw_spin_lock_irqsave(&cputimer->lock, flags); cputimer->running = 0;
+ printk(KERN_DEBUG "cputimer stopped\n");
raw_spin_unlock_irqrestore(&cputimer->lock, flags);
}
with the small program:
\#include <signal.h>
\#include <time.h>
\#include <stddef.h>
\#include <stdio.h>
\#include <string.h>
\#include <fcntl.h>
\#include <unistd.h>
static void chew_cpu()
{
static volatile char buf[4096];
for (size_t i = 0; i < 100; ++i)
for (size_t j = 0; j < sizeof buf; ++j)
buf[j] = 0xaa;
int nullfd = open("/dev/null", O_WRONLY);
for (size_t i = 0; i < 100; ++ i)
for (size_t j = 0; j < sizeof buf; ++j)
buf[j] = 0xbb;
write(nullfd, (char*)buf, sizeof buf);
close(nullfd);
}timer_t t;
volatile int sig_cnt;
static voidsig_handler (int sig, siginfo_t *info, void *ctx)
{
if (sig_cnt >= 5) {
struct itimerspec it = {};
timer_settime(t, 0, &it, NULL);
}
++sig_cnt;
}int main( int argc, char *argv[] )
{
clockid_t my_process_clock;
int e;
struct sigaction sa = { .sa_sigaction = sig_handler,
.sa_flags = SA_SIGINFO };
sigemptyset(&sa.sa_mask);
sigaction(SIGRTMIN, &sa, NULL);
struct sigevent ev;
memset( &ev, 0, sizeof (ev));
ev.sigev_notify = SIGEV_SIGNAL;
ev.sigev_signo = SIGRTMIN;
ev.sigev_value.sival_ptr = &ev;
e = clock_getcpuclockid( 0, &my_process_clock);
if (e) {
printf("clock_getcpuclockid: (%d) %s\n", e, strerror(e));
return 1;
}
if (timer_create(my_process_clock, &ev, &t) != 0) {
printf("timer_create: %m\n");
return 1;
}
struct itimerspec it;
it.it_value.tv_sec = 0;
it.it_value.tv_nsec = 100000;
it.it_interval.tv_sec = 0;
it.it_interval.tv_nsec = 100000;
timer_settime(t,0,&it,NULL);
while (sig_cnt < 5)
chew_cpu();
// process timer should stop
struct timespec req = { .tv_sec = 0, .tv_nsec = 400000 };
nanosleep(&req,NULL);
// restart timer.
sig_cnt = 0;
timer_settime(t,0,&it,NULL);
while (sig_cnt < 5)
chew_cpu();
timer_delete(t);
return 0;
}demonstrate the issue. Here are the results before and after applying the patch:
Before:
[ 181.904571] cputimer started [ 181.905013] cputimer stopped [ 181.905027] cputimer started [ 181.906009] cputimer stopped [ 181.906016] cputimer started [ 181.907008] cputimer stopped [ 181.907014] cputimer started [ 181.908008] cputimer stopped [ 181.908013] cputimer started [ 181.909008] cputimer stopped [ 181.909020] cputimer started [ 181.910007] cputimer stopped [ 181.910050] cputimer started [ 181.911011] cputimer stopped [ 181.913619] cputimer started [ 181.914007] cputimer stopped [ 181.914015] cputimer started [ 181.915006] cputimer stopped [ 181.915011] cputimer started [ 181.916006] cputimer stopped [ 181.916010] cputimer started [ 181.917008] cputimer stopped [ 181.917058] cputimer started [ 181.918009] cputimer stopped [ 181.918029] cputimer started [ 181.919006] cputimer stopped [ 181.919010] cputimer started [ 181.920006] cputimer stopped
After:
[ 859.722119] cputimer started [ 859.730004] cputimer stopped [ 859.731354] cputimer started [ 859.739004] cputimer stopped
--- kernel/posix-cpu-timers.c | 34 +++++++++++++++------------------- 1 file changed, 15 insertions(+), 19 deletions(-)
kernel/posix-cpu-timers.c +15 -19
@@ -1049,8 +1049,6 @@ static void check_process_timers(struct task_struct *tsk, sig->cputime_expires.prof_exp =expires_to_cputime (prof_expires); sig->cputime_expires.virt_exp =expires_to_cputime (virt_expires); sig->cputime_expires.sched_exp = sched_expires; -if (task_cputime_zero (&sig->cputime_expires)) -stop_process_timers (sig); }/*
@@ -1158,34 +1156,32 @@ static inline int task_cputime_expired(const struct task_cputime *sample,*/ static inline int fastpath_timer_check (struct task_struct *tsk) { -struct signal_struct *sig; -cputime_t utime, stime; +struct signal_struct *sig = tsk->signal; -task_cputime (tsk, &utime, &stime); +if (sig->cputimer.running) { +if (likely (!task_cputime_zero (&sig->cputime_expires))) { +struct task_cputime group_sample; + +raw_spin_lock (&sig->cputimer.lock); + group_sample = sig->cputimer.cputime; +raw_spin_unlock (&sig->cputimer.lock); + +if (task_cputime_expired (&group_sample, &sig->cputime_expires)) +return 1 ; + }else +stop_process_timers (sig); + }if (!task_cputime_zero (&tsk->cputime_expires)) {struct task_cputime task_sample = { - .utime = utime, - .stime = stime, .sum_exec_runtime = tsk->se.sum_exec_runtime }; +task_cputime (tsk, &task_sample.utime, &task_sample.stime);if (task_cputime_expired (&task_sample, &tsk->cputime_expires))return 1 ; } - sig = tsk->signal; -if (sig->cputimer.running) { -struct task_cputime group_sample; - -raw_spin_lock (&sig->cputimer.lock); - group_sample = sig->cputimer.cputime; -raw_spin_unlock (&sig->cputimer.lock); - -if (task_cputime_expired (&group_sample, &sig->cputime_expires)) -return 1 ; - } -return 0 ; }
Signature · 1 line
1.8.4.2
One more detail useful to note to appreciate this patch is that the call to posix_cpu_timer_schedule() from cpu_timer_fire() is done exceptionnaly when the timer signal generation fails.
The normal processing will call posix_cpu_timer_schedule() from a different context in do_schedule_next_timer() (posix-timers.c)
I believe this misunderstanding may partly explain the current state of the code.
Quoted 224 lines · Olivier Langlois
When a periodic process timer is fired, a signal is generated. Rearming the timer, if necessary, will be performed in a second step when the signal will be delivered.
Hence, checking the list of process timers list to decide to stop to cpu timer right after generating the signal is premature and is leading the cputimer to oscillate between the on/off states when a single process timer is present.
The following posix-cpu-timers module intrumentation
In void thread_group_cputimer(struct
task_sraw_spin_lock_irqsave(&cputimer->lock, flags);(quoted diff) +1
-0cputimer->running = 1; update_gt_cputime(&cputimer->cputime, &sum); + printk(KERN_DEBUG "cputimer started\n"); } else raw_spin_lock_irqsave(&cputimer->lock, flags); *times = cputimer->cputime;and in static void stop_process_timers(struct s
(quoted diff) +1
-0raw_spin_lock_irqsave(&cputimer->lock, flags); cputimer->running = 0; + printk(KERN_DEBUG "cputimer stopped\n"); raw_spin_unlock_irqrestore(&cputimer->lock, flags); }with the small program:
\#include <
signal.h>
\#include <time.h>
\#include <stddef.h>
\#include <stdio.h>
\#include <string.h>
\#include <fcntl.h>
\#include <unistd.h>static void
chew_cpu(){ static volatile char buf[4096]; for (size_t i = 0; i < 100; ++i) for (size_t j = 0; j < sizeof buf; ++j) buf[j] = 0xaa; int nullfd = open("/dev/null", O_WRONLY); for (size_t i = 0; i < 100; ++ i) for (size_t j = 0; j < sizeof buf; ++j) buf[j] = 0xbb; write(nullfd, (char*)buf, sizeof buf); close(nullfd); }
timer_tt;
volatile intsig_cnt;static void
sig_handler(int sig,siginfo_t*info, void *ctx){ if (sig_cnt >= 5) { struct itimerspec it = {}; timer_settime(t, 0, &it, NULL); } ++sig_cnt; }int
main( int argc, char *argv[] ){ clockid_t my_process_clock; int e; struct sigaction sa = { .sa_sigaction = sig_handler, .sa_flags = SA_SIGINFO }; sigemptyset(&sa.sa_mask); sigaction(SIGRTMIN, &sa, NULL); struct sigevent ev; memset( &ev, 0, sizeof (ev)); ev.sigev_notify = SIGEV_SIGNAL; ev.sigev_signo = SIGRTMIN; ev.sigev_value.sival_ptr = &ev; e = clock_getcpuclockid( 0, &my_process_clock); if (e) { printf("clock_getcpuclockid: (%d) %s\n", e, strerror(e)); return 1; } if (timer_create(my_process_clock, &ev, &t) != 0) { printf("timer_create: %m\n"); return 1; } struct itimerspec it; it.it_value.tv_sec = 0; it.it_value.tv_nsec = 100000; it.it_interval.tv_sec = 0; it.it_interval.tv_nsec = 100000; timer_settime(t,0,&it,NULL); while (sig_cnt < 5) chew_cpu(); // process timer should stop struct timespec req = { .tv_sec = 0, .tv_nsec = 400000 }; nanosleep(&req,NULL); // restart timer. sig_cnt = 0; timer_settime(t,0,&it,NULL); while (sig_cnt < 5) chew_cpu(); timer_delete(t); return 0; }demonstrate the issue. Here are the results before and after applying the patch:
Before:
[ 181.904571] cputimer started [ 181.905013] cputimer stopped [ 181.905027] cputimer started [ 181.906009] cputimer stopped [ 181.906016] cputimer started [ 181.907008] cputimer stopped [ 181.907014] cputimer started [ 181.908008] cputimer stopped [ 181.908013] cputimer started [ 181.909008] cputimer stopped [ 181.909020] cputimer started [ 181.910007] cputimer stopped [ 181.910050] cputimer started [ 181.911011] cputimer stopped [ 181.913619] cputimer started [ 181.914007] cputimer stopped [ 181.914015] cputimer started [ 181.915006] cputimer stopped [ 181.915011] cputimer started [ 181.916006] cputimer stopped [ 181.916010] cputimer started [ 181.917008] cputimer stopped [ 181.917058] cputimer started [ 181.918009] cputimer stopped [ 181.918029] cputimer started [ 181.919006] cputimer stopped [ 181.919010] cputimer started [ 181.920006] cputimer stoppedAfter:
[ 859.722119] cputimer started [ 859.730004] cputimer stopped [ 859.731354] cputimer started [ 859.739004] cputimer stoppedSigned-off-by Olivier Langlois <[email protected]>--- kernel/posix-cpu-timers.c | 34 +++++++++++++++------------------- 1 file changed, 15 insertions(+), 19 deletions(-)kernel/posix-cpu-timers.c +15
-19@@ -1049,8 +1049,6 @@ static void check_process_timers(struct task_struct *tsk, sig->cputime_expires.prof_exp =expires_to_cputime (prof_expires); sig->cputime_expires.virt_exp =expires_to_cputime (virt_expires); sig->cputime_expires.sched_exp = sched_expires; -if (task_cputime_zero (&sig->cputime_expires)) -stop_process_timers (sig); }/* @@ -1158,34 +1156,32 @@ static inline int task_cputime_expired(const struct task_cputime *sample,*/ static inline int fastpath_timer_check (struct task_struct *tsk) { -struct signal_struct *sig; -cputime_t utime, stime; +struct signal_struct *sig = tsk->signal; -task_cputime (tsk, &utime, &stime); +if (sig->cputimer.running) { +if (likely (!task_cputime_zero (&sig->cputime_expires))) { +struct task_cputime group_sample; + +raw_spin_lock (&sig->cputimer.lock); + group_sample = sig->cputimer.cputime; +raw_spin_unlock (&sig->cputimer.lock); + +if (task_cputime_expired (&group_sample, &sig->cputime_expires)) +return 1 ; + }else +stop_process_timers (sig); + }if (!task_cputime_zero (&tsk->cputime_expires)) {struct task_cputime task_sample = { - .utime = utime, - .stime = stime, .sum_exec_runtime = tsk->se.sum_exec_runtime }; +task_cputime (tsk, &task_sample.utime, &task_sample.stime);if (task_cputime_expired (&task_sample, &tsk->cputime_expires))return 1 ; } - sig = tsk->signal; -if (sig->cputimer.running) { -struct task_cputime group_sample; - -raw_spin_lock (&sig->cputimer.lock); - group_sample = sig->cputimer.cputime; -raw_spin_unlock (&sig->cputimer.lock); - -if (task_cputime_expired (&group_sample, &sig->cputime_expires)) -return 1 ; - } -return 0 ; }