futex priority based wakeup

[Ilya Lipovsky]

Good info. Indeed, looks like this is what glibc actually does:

 

http://www.cygwin.com/ml/libc-alpha/2007-06/msg00097.html

 

 

 

  _____  

From: tinghich at gmail.com [mailto:tinghich at gmail.com] On Behalf Of Nguyen
Nguyen
Sent: Tuesday, September 11, 2007 8:15 PM
To: Ilya Lipovsky
Cc: Benedict, Michael; linuxppc-embedded at ozlabs.org
Subject: Re: futex priority based wakeup

 

I have seen something similar before.  Our fix was to use
pthread_attr_setinheritsched(&attr, PTHREAD_EXPLICIT_SCHED) so child threads
wouldn't inherit attribute from parent.  Hope it helps.



On 9/11/07, Ilya Lipovsky <lipovsky at cs.bu.edu> wrote:

Hmm. Just for kicks - inside the important thread could you add:

int curpolicy;
struct sched_param sp;
pthread_getschedparam (pthread_self (), &curpolicy, &sp)
printf("important's policy is %d and priority is %d\n", curpolicy, 
sp.__sched_priority);

before the very first futex syscall and after your "printf("important got
futex!\n");" line.

Do similar for the unimportant thread, and see if you get anything weird - 
e.g. priorities come out to be the same for threads.


-----Original Message-----
From: linuxppc-embedded-bounces+lipovsky=cs.bu.edu at ozlabs.org
[mailto: <mailto:linuxppc-embedded-bounces+lipovsky=cs.bu.edu at ozlabs.org>
linuxppc-embedded-bounces+lipovsky=cs.bu.edu at ozlabs.org] On Behalf
Of Benedict, Michael
Sent: Monday, September 10, 2007 5:41 PM
To: linuxppc-embedded at ozlabs.org 
Subject: RE: futex priority based wakeup

Ilya Lipovsky wrote:
> Your code looks correct to me, so if the kernel developers
> did their job
> correctly, the only potentially weak link is glibc. 
>

Well, either the kernel developers didn't do their job, or I am missing
something.  The following also fails, and it should be bypassing glibc:

#define _XOPEN_SOURCE 600

#include <linux/futex.h> 
#include <sys/time.h>
#include <asm/atomic.h>

#include <sys/syscall.h>
#include <unistd.h>

#include <sched.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>

int myfutex = 0;

void *important(void *ign)
{
        sleep(1);
        printf("important waiting for futex\n");
        fflush(stdout);
        if(syscall(SYS_futex, &myfutex, FUTEX_WAIT, 0, NULL)) { 
                perror("futex");
                exit(1);
        } else {
                printf("important got futex!\n");
                fflush(stdout);
                syscall(SYS_futex, &myfutex, FUTEX_WAKE, 1, NULL); 
}

        return NULL;
}


void *unimportant(void *ign)
{
        printf("unimportant waiting for futex\n");
        fflush(stdout);
        if(syscall(SYS_futex, &myfutex, FUTEX_WAIT, 0, NULL)) { 
                perror("futex");
                exit(1);
        } else {
                printf("unimportant got futex!\n");
                fflush(stdout);
                syscall(SYS_futex, &myfutex, FUTEX_WAKE, 1, NULL); 
}

        return NULL;
}

int main()
{
        struct sched_param p;
        pthread_attr_t attr;
        pthread_t i, u;

        p.__sched_priority = sched_get_priority_min(SCHED_FIFO); 
        if(-1 == p.__sched_priority) {
                perror("sched_get_priority_min");
                return 1;
        }
        pthread_attr_init(&attr);
        pthread_attr_setschedpolicy(&attr, SCHED_FIFO); 
        pthread_attr_setschedparam(&attr, &p);
        pthread_create(&u, &attr, unimportant, NULL);

        p.__sched_priority = sched_get_priority_max(SCHED_FIFO);
        pthread_attr_setschedparam(&attr, &p); 
        pthread_create(&i, &attr, important, NULL);

        sleep(5);
        printf("futex FUTEX_WAKE\n");
        fflush(stdout);
        syscall(SYS_futex, &myfutex, FUTEX_WAKE, 1, NULL); 

        pthread_join(u, NULL);
        pthread_join(i, NULL);

        return 0;
}

Which produces:
unimportant waiting for futex
important waiting for futex
futex FUTEX_WAKE
unimportant got futex! 
important got futex!


Could someone with 2.6.22 please verify?

_______________________________________________
Linuxppc-embedded mailing list
Linuxppc-embedded at ozlabs.org  <mailto:Linuxppc-embedded at ozlabs.org> 
https://ozlabs.org/mailman/listinfo/linuxppc-embedded

_______________________________________________
Linuxppc-embedded mailing list
Linuxppc-embedded at ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-embedded

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://ozlabs.org/pipermail/linuxppc-embedded/attachments/20070911/f4ed6378/attachment.htm