NetBSD Problem Report #46581

From www@NetBSD.org  Mon Jun 11 03:50:51 2012
Return-Path: <www@NetBSD.org>
Received: from mail.netbsd.org (mail.netbsd.org [149.20.53.66])
	by www.NetBSD.org (Postfix) with ESMTP id 1996C63B89C
	for <gnats-bugs@gnats.NetBSD.org>; Mon, 11 Jun 2012 03:50:51 +0000 (UTC)
Message-Id: <20120611035050.5CE0863B882@www.NetBSD.org>
Date: Mon, 11 Jun 2012 03:50:50 +0000 (UTC)
From: naruse@airemix.jp
Reply-To: naruse@airemix.jp
To: gnats-bugs@NetBSD.org
Subject: Continuous pthread_create may fail on amd64
X-Send-Pr-Version: www-1.0

>Number:         46581
>Category:       port-amd64
>Synopsis:       Continuous pthread_create may fail on amd64
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    port-amd64-maintainer
>State:          closed
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Mon Jun 11 03:55:00 +0000 2012
>Closed-Date:    Tue Aug 15 09:56:22 +0000 2017
>Last-Modified:  Tue Aug 15 09:56:22 +0000 2017
>Originator:     Yui NARUSE
>Release:        6.99.7
>Organization:
>Environment:
NetBSD kelvena.airemix.net 6.99.7 NetBSD 6.99.7 (GENERIC) #2: Thu May 17 14:42:34 JST 2012  naruse@kelvena.airemix.net:/usr/obj/sys/arch/amd64/compile/GENERIC amd64
>Description:
Below program fails on amd64 as following:

% cc -pthread thr.c&&./a.out
pthread_create: Resource temporarily unavailable
zsh: abort (core dumped)  ./a.out

On the environment, ulimit -u is 160.
But it won't fail on i386.
It won't fail if sched_yield() is not commented out.
So it seems something is wrong.

#include <pthread.h>
#include <stdlib.h>
#define THREAD_COUNT 161
void *dummy(void *p) {
    return NULL;
}
int main(void) {
    int i;
    pthread_t *threads = malloc(sizeof(pthread_t) * THREAD_COUNT);
    for (i = 0; i < THREAD_COUNT; i++) {
        if (pthread_create(&threads[i], NULL, dummy, &i)) {
            perror("pthread_create");
            abort();
        }   
        //sched_yield();
    }   
    return 0;
} 
>How-To-Repeat:

>Fix:

>Release-Note:

>Audit-Trail:
From: Martin Husemann <martin@duskware.de>
To: gnats-bugs@NetBSD.org
Cc: 
Subject: Re: port-amd64/46581: Continuous pthread_create may fail on amd64
Date: Mon, 11 Jun 2012 11:23:37 +0200

 I don't see what's wrong with this behaviour - you can not rely on your new
 threads being scheduled before running into the resource limit on threads
 in the creating for loop.

 Am I missing something?

 Martin

From: "NARUSE, Yui" <naruse@airemix.jp>
To: gnats-bugs@netbsd.org
Cc: 
Subject: Re: port-amd64/46581: Continuous pthread_create may fail on amd64
Date: Tue, 12 Jun 2012 01:48:12 +0900

 2012/6/11 Martin Husemann <martin@duskware.de>:
 > The following reply was made to PR port-amd64/46581; it has been noted by=
  GNATS.
 >
 > From: Martin Husemann <martin@duskware.de>
 > To: gnats-bugs@NetBSD.org
 > Cc:
 > Subject: Re: port-amd64/46581: Continuous pthread_create may fail on amd6=
 4
 > Date: Mon, 11 Jun 2012 11:23:37 +0200
 >
 > =A0I don't see what's wrong with this behaviour - you can not rely on you=
 r new
 > =A0threads being scheduled before running into the resource limit on thre=
 ads
 > =A0in the creating for loop.
 >
 > =A0Am I missing something?

 I'm sorry about confusing post.
 What I described about sched_yield is useless.

 The correct reproducible program is below and wrong platform is i386.
 On amd64, it fails as ulimit specified.
 But on i386, it completes working.

 #include <pthread.h>
 #include <stdlib.h>
 #define THREAD_COUNT 161
 void *dummy(void *p) {
     sleep(100);
     return NULL;
 }
 int main(void) {
     int i;
     pthread_t *threads =3D malloc(sizeof(pthread_t) * THREAD_COUNT);
     for (i =3D 0; i < THREAD_COUNT; i++) {
         if (pthread_create(&threads[i], NULL, dummy, &i)) {
             perror("pthread_create");
             abort();
         }
         sched_yield();
     }
     return 0;
 }

 --=20
 NARUSE, Yui =A0<naruse@airemix.jp>

State-Changed-From-To: open->closed
State-Changed-By: maxv@NetBSD.org
State-Changed-When: Tue, 15 Aug 2017 09:56:22 +0000
State-Changed-Why:
There is no issue here. On NetBSD it's 'ulimit -r', and it does work
as expected on both i386 and amd64. I'm closing this PR.


>Unformatted:

Home	PR Database Search