Re: BUG: unable to handle kernel paging request xfs_is_delayed_page

[Paul E. McKenney]

On Sun, Jul 24, 2011 at 02:06:04PM +0300, Török Edwin wrote:
> On 07/24/2011 10:14 AM, Maciej Rutecki wrote:
> > On czwartek, 21 lipca 2011 o 22:55:04 Török Edwin wrote:
> >> Hi,
> >>
> >> Just got this BUG in my dmesg:
> >> [47504.938446] BUG: unable to handle kernel paging request at
> >> ffff884058ec3270 [47504.938488] IP: [<ffffffff8127baf1>]
> > [...]
> > 
> > 2.6.39 works OK?. It is regression?
> 
> I don't know, I was not able to reproduce the bug on 3.0 either.
> Either the bug was fixed between 3.0-rc7 and 3.0, or it is very hard to reproduce.

There were some regressions in 3.0-rc1 through 3.0-rc7 that are fixed
in 3.0.  If you cannot reproduce in 3.0, then I would guess that you
are hitting one of those bugs.

							Thanx, Paul

> I tried with the attached test program (which creates a mess^H some files in the current directory, performs I/O and dumps core
> from 2 processes in parallel.).
> 
> All I got was 2 hung kernel threads for 2m+ in xfs_evict_inode + xfs_file_sync, trigerring the hung_check timer and NMI backtraces,
> and the process was unkillable (by kill -9) for a while. It eventually recovered though, and its not surprising that this happened
> : the test program generated 100Mb/s - 500Mb/s I/O.
> 
> I'll have to see if I can reproduce the BUG with 3.0-rc7. Although I don't see any XFS changes between 3.0-rc7 and 3.0
> there were some RCU fixes to core VFS code.
> 
> Best regards,
> --Edwin

> #include <stdlib.h>
> #include <stdint.h>
> #include <unistd.h>
> #include <string.h>
> #include <pthread.h>
> #include <stdio.h>
> #include <errno.h>
> 
> void alloc_and_die(void)
> {
>     uint64_t i;
>     uint64_t n = 4*1024*1024*1024ll;
>     char *x = malloc(n);
>     printf("touching pages\n");
>     /* touch each page once */
>     for (i=0;i<n;i += 4096) {
> 	x[i] = 42;
>     }
>     /* wait a bit */
>     printf("sleeping\n");
>     /* parallel coredump */
>     fork();
>     sleep(10);
>     printf("Dumping core...\n");
>     /* now die */
>     abort();
> }
> 
> void *iothread(void *dummy)
> {
>     uint16_t data[4000];
>     char fname[128] = "iothreadXXXXXX";
>     unsigned int seed = 0x42;
>     unsigned i;
>     uint64_t pos = 0;
>     unsigned counter = 0;
> 
>     int fd = mkstemp(fname);
> 
>     if (fd == -1) {
> 	perror("mkstemp");
> 	abort();
>     }
> 
>     for (i=0;i<sizeof(data)/sizeof(data[0]);i++) {
> 	data[i] = rand_r(&seed);
>     }
>     /* continously write to a 1MB sized file */
>     while (1) {
> 	if (write(fd, data, sizeof(data)) != sizeof(data)) {
> 	    perror("write failed");
> 	    abort();
> 	}
> 	pos += sizeof(data);
> 	if (pos > 10*1024*1024ll) {
> 	    counter++;
> 	    if (counter%2) {
> 		fsync(fd);
> 		lseek(fd, 0, SEEK_SET);
> 	    } else {
> 		unlink(fname);
> 		close(fd);
> 		strncpy(fname, "iothreadXXXXXX", sizeof(fname));
> 		fd = mkstemp(fname);
> 		if (fd == -1) {
> 		    perror("mkstemp");
> 		    abort();
> 		}
> 	    }
> 	    for (i=0;i<sizeof(data)/sizeof(data[0]);i++) {
> 		data[i] = rand_r(&seed);
> 	    }
> 	}
>     }
> 
>     return NULL;
> }
> 
> void run_iothread(void)
> {
>     pthread_t thr;
>     int rc;
>     rc = pthread_create(&thr, NULL,
> 			iothread, NULL);
>     if (rc) {
> 	errno = rc;
> 	perror("pthread_create");
> 	abort();
>     }
> }
> 
> int main()
> {
>     switch (fork()) {
> 	case 0:
> 	    run_iothread();
> 	    run_iothread();
> 	    alloc_and_die();
> 	    break;
> 	case -1:
> 	    perror("fork failed\n");
> 	    abort();
> 	    break;
> 	default:
> 	    run_iothread();
> 	    run_iothread();
> 	    run_iothread();
> 	    run_iothread();
> 	    iothread(NULL);
> 	    break;
>     }
>     return 0;
> }
> 

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/