linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Jan Stancek <jstancek@redhat.com>
To: Peter Zijlstra <peterz@infradead.org>
Cc: alex shi <alex.shi@intel.com>, guz fnst <guz.fnst@cn.fujitsu.com>,
	mingo@redhat.com, jolsa@redhat.com, riel@redhat.com,
	linux-kernel@vger.kernel.org
Subject: Re: [BUG] scheduler doesn't balance thread to idle cpu for 3 seconds
Date: Mon, 8 Feb 2016 14:40:48 +0100	[thread overview]
Message-ID: <56B89AE0.9090603@redhat.com> (raw)
In-Reply-To: <654964868.14006956.1454063625314.JavaMail.zimbra@redhat.com>

[-- Attachment #1: Type: text/plain, Size: 2600 bytes --]

On 01/29/2016 11:33 AM, Jan Stancek wrote:
>>
>> Also note that I don't think failing this test is a bug per se.
>> Undesirable maybe, but within spec, since SIGALRM is process wide, so it
>> being delivered to the SCHED_OTHER task is accepted, and SCHED_OTHER has
>> no timeliness guarantees.
>>
>> That said; if I could reliably reproduce I'd have a go at fixing this, I
>> suspect there's a 'fun' problem at the bottom of this.
> 
> Thanks for trying, I'll see if I can find some more reliable way.

I think I have found a more reliably way, however it requires an older
stable kernel: 3.12.53 up to 4.1.17.

Consider following scenario:
- all tasks on system have RT sched class
- main thread of reproducer becomes the only SCHED_OTHER task on system
- when alarm(2) expires, main thread is woken up on cpu that is occupied by
  busy looping RT thread (low_priority_thread)
- because main thread was sleeping for 2 seconds, its load has decayed to 0
- the only chance for main thread to run is if it gets balanced to idle CPU
- task_tick_fair() doesn't run, there is RT task running on this CPU
- main thread is on cfs run queue but its load stays 0
- load balancer never sees this CPU (group) as busy

Attached is reproducer and script, which tries to trigger scenario above.
I can reproduce it with 4.1.17 on baremetal 4 CPU x86_64 with about 1:50 chance.

In this setup failure state persists for a long time, perhaps indefinitely.
I tried extending RUNTIME to 10 minutes, main thread still wouldn't run.

One more clue: I could work around this issue if I forced an update_entity_load_avg()
on sched_entities that have not been updated for some time, as part of
periodic rebalance_domains() call.

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index c7c1d28..1b5fe80 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -5264,6 +5264,7 @@ static void update_blocked_averages(int cpu)
 	struct rq *rq = cpu_rq(cpu);
 	struct cfs_rq *cfs_rq;
 	unsigned long flags;
+	struct rb_node *rb;

 	raw_spin_lock_irqsave(&rq->lock, flags);
 	update_rq_clock(rq);
@@ -5281,6 +5282,19 @@ static void update_blocked_averages(int cpu)
 	}

 	raw_spin_unlock_irqrestore(&rq->lock, flags);
+
+	cfs_rq = &(cpu_rq(cpu)->cfs);
+	for (rb = rb_first_postorder(&cfs_rq->tasks_timeline); rb; rb = rb_next_postorder(rb)) {
+		struct sched_entity *se = rb_entry(rb, struct sched_entity, run_node);
+
+		// Task on rq has not been updated for 500ms :-(
+		if ((cfs_rq_clock_task(cfs_rq) - se->avg.last_runnable_update) > 500L * (1 << 20))
+			update_entity_load_avg(se, 1);
+	}
 }

 /*

Regards,
Jan


[-- Attachment #2: pthread_cond_wait_1_v3.c --]
[-- Type: text/plain, Size: 5201 bytes --]

/*
 * reproducer v3 for:
 * [BUG] scheduler doesn't balance thread to idle cpu for 3 seconds
 *
 * Based on LTP's pthread_cond_wait_1.c
 *
 */

#define _GNU_SOURCE
#include <pthread.h>
#include <sched.h>
#include <stdio.h>
#include <stdlib.h>
#include <signal.h>
#include <inttypes.h>
#include <unistd.h>
#include <time.h>
#include <sys/time.h>
#include <sys/resource.h>

#define ERROR_PREFIX "unexpected error: "

#define HIGH_PRIORITY 10
#define LOW_PRIORITY  5
#define RUNTIME       5
#define POLICY        SCHED_RR

#define PTS_PASS 0
#define PTS_FAIL 1
#define PTS_UNRESOLVED 2

pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
pthread_cond_t cond = PTHREAD_COND_INITIALIZER;

/* Flags that the threads use to indicate events */
volatile int woken_up = 0;
volatile int low_done = 0;

/* Signal handler that handle the ALRM and wakes up
 * the high priority thread
 */
void signal_handler(int sig)
{
	(void) sig;
	if (pthread_cond_signal(&cond) != 0) {
		printf(ERROR_PREFIX "pthread_cond_signal\n");
		exit(PTS_UNRESOLVED);
	}
}

/* Utility function to find difference between two time values */
float timediff(struct timespec t2, struct timespec t1)
{
	float diff = t2.tv_sec - t1.tv_sec;
	diff += (t2.tv_nsec - t1.tv_nsec) / 1000000000.0;
	return diff;
}

void *hi_priority_thread(void *tmp)
{
	struct sched_param param;
	int policy;
	int rc = 0;

	(void) tmp;
	param.sched_priority = HIGH_PRIORITY;

	rc = pthread_setschedparam(pthread_self(), POLICY, &param);
	if (rc != 0) {
		printf(ERROR_PREFIX "pthread_setschedparam\n");
		exit(PTS_UNRESOLVED);
	}
	rc = pthread_getschedparam(pthread_self(), &policy, &param);
	if (rc != 0) {
		printf(ERROR_PREFIX "pthread_getschedparam\n");
		exit(PTS_UNRESOLVED);
	}
	if ((policy != POLICY) || (param.sched_priority != HIGH_PRIORITY)) {
		printf("Error: the policy or priority not correct\n");
		exit(PTS_UNRESOLVED);
	}

	/* Install a signal handler for ALRM */
	if (signal(SIGALRM, signal_handler) != 0) {
		perror(ERROR_PREFIX "signal:");
		exit(PTS_UNRESOLVED);
	}

	/* acquire the mutex */
	rc = pthread_mutex_lock(&mutex);
	if (rc != 0) {
		printf(ERROR_PREFIX "pthread_mutex_lock\n");
		exit(PTS_UNRESOLVED);
	}

	/* Setup an alarm to go off in 2 seconds */
	alarm(2);

	/* Block, to be woken up by the signal handler */
	rc = pthread_cond_wait(&cond, &mutex);
	if (rc != 0) {
		printf(ERROR_PREFIX "pthread_cond_wait\n");
		exit(PTS_UNRESOLVED);
	}

	/* This variable is unprotected because the scheduling removes
	 * the contention
	 */
	if (low_done != 1)
		woken_up = 1;

	rc = pthread_mutex_unlock(&mutex);
	if (rc != 0) {
		printf(ERROR_PREFIX "pthread_mutex_unlock\n");
		exit(PTS_UNRESOLVED);
	}
	return NULL;
}

void *low_priority_thread(void *tmp)
{
	struct timespec start_time, current_time;
	struct sched_param param;
	int policy;
	cpu_set_t cpuset;
	int rc = 0, slept_times = 0;
	float slept_for = 0;
	uintptr_t tnum = (uintptr_t)tmp;

	param.sched_priority = LOW_PRIORITY;

	rc = pthread_setschedparam(pthread_self(), POLICY, &param);
	if (rc != 0) {
		printf(ERROR_PREFIX "pthread_setschedparam\n");
		exit(PTS_UNRESOLVED);
	}
	rc = pthread_getschedparam(pthread_self(), &policy, &param);
	if (rc != 0) {
		printf(ERROR_PREFIX "pthread_getschedparam\n");
		exit(PTS_UNRESOLVED);
	}
	if ((policy != POLICY) || (param.sched_priority != LOW_PRIORITY)) {
		printf("Error: the policy or priority not correct\n");
		exit(PTS_UNRESOLVED);
	}

	CPU_ZERO(&cpuset);
	CPU_SET(tnum, &cpuset);

	rc = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset);
	if (rc != 0) {
		printf(ERROR_PREFIX "pthread_setaffinity_np\n");
		exit(PTS_UNRESOLVED);
	}

	/* grab the start time and busy loop for 5 seconds */
	clock_gettime(CLOCK_REALTIME, &start_time);
	while (!woken_up && !low_done) {
		clock_gettime(CLOCK_REALTIME, &current_time);
		if (timediff(current_time, start_time) > RUNTIME)
			break;
	}

	low_done = 1;
	return NULL;
}

int main()
{
	pthread_t high_id, *low_id, paused_id;
	struct sched_param param;
	int rc = 0;
	int i, ncpus = sysconf(_SC_NPROCESSORS_ONLN);

	low_id = malloc(ncpus * sizeof(pthread_t));

	/* high prio thread */
	rc = pthread_create(&high_id, NULL, hi_priority_thread, NULL);
	if (rc != 0) {
		printf(ERROR_PREFIX "pthread_create\n");
		exit(PTS_UNRESOLVED);
	}

	/* low prio thread on each cpu except last one */
	for (i = 0; i < ncpus - 1; i++) {
		uintptr_t tnum = i;
		rc = pthread_create(&low_id[i], NULL, low_priority_thread, (void *)tnum);
		if (rc != 0) {
			printf(ERROR_PREFIX "pthread_create\n");
			exit(PTS_UNRESOLVED);
		}
	}

	param.sched_priority = 0;
	rc = pthread_setschedparam(pthread_self(), SCHED_OTHER, &param);
	if (rc != 0) {
		printf(ERROR_PREFIX "pthread_setschedparam\n");
		exit(PTS_UNRESOLVED);
	}

	/* Wait for the threads to exit */
	rc = pthread_join(high_id, NULL);
	if (rc != 0) {
		printf(ERROR_PREFIX "pthread_join\n");
		exit(PTS_UNRESOLVED);
	}

	for (i = 0; i < ncpus - 1; i++) {
		rc = pthread_join(low_id[i], NULL);
		if (rc != 0) {
			printf(ERROR_PREFIX "pthread_join\n");
			exit(PTS_UNRESOLVED);
		}
	}

	if (woken_up == 0) {
		printf("Test FAILED: high priority was not woken up\n");
		exit(PTS_FAIL);
	}

	printf("Test PASSED\n");
	exit(PTS_PASS);
}

[-- Attachment #3: reproduce_v3.sh --]
[-- Type: application/x-shellscript, Size: 260 bytes --]

      reply	other threads:[~2016-02-08 13:40 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-01-27 14:52 [BUG] scheduler doesn't balance thread to idle cpu for 3 seconds Jan Stancek
2016-01-28 15:55 ` Jan Stancek
2016-01-28 17:49   ` Peter Zijlstra
2016-01-28 18:43     ` Jan Stancek
2016-01-29 10:15       ` Peter Zijlstra
2016-01-29 10:33         ` Jan Stancek
2016-02-08 13:40           ` Jan Stancek [this message]

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=56B89AE0.9090603@redhat.com \
    --to=jstancek@redhat.com \
    --cc=alex.shi@intel.com \
    --cc=guz.fnst@cn.fujitsu.com \
    --cc=jolsa@redhat.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mingo@redhat.com \
    --cc=peterz@infradead.org \
    --cc=riel@redhat.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;
as well as URLs for NNTP newsgroup(s).