Re: bad scheduling and an BUG between them

[Oleg Drokin <green@namesys.com>]

Hello!

On Wed, Aug 06, 2003 at 01:24:05PM +0300, Lenar L?hmus wrote:

> with kernel-2.6.0-test2-mm3+A3-O12.2int i found this in logs:
> bad: scheduling while atomic!
> Call Trace:
>  [<c011c87d>] schedule+0x56d/0x580
>  [<c0144b33>] unmap_page_range+0x43/0x70
>  [<c0144d15>] unmap_vmas+0x1b5/0x210
>  [<c014893b>] exit_mmap+0x7b/0x190
>  [<c011e399>] mmput+0x79/0xf0
>  [<c0122082>] do_exit+0x122/0x3f0
>  [<c010b7c0>] do_invalid_op+0x0/0xd0
>  [<c010b4d9>] die+0xf9/0x100
>  [<c010b889>] do_invalid_op+0xc9/0xd0
>  [<c01436ba>] kunmap_high+0x1a/0xa0
>  [<c02a0f9f>] error_code+0x2f/0x38
>  [<c01436ba>] kunmap_high+0x1a/0xa0
>  [<c01a6040>] reiserfs_unprepare_pages+0x30/0x70
>  [<c01a72b5>] reiserfs_file_write+0x4e5/0x595
>  [<c011aa31>] do_page_fault+0x251/0x454
>  [<c011c941>] __wake_up_common+0x31/0x60
>  [<c01a6dd0>] reiserfs_file_write+0x0/0x595
>  [<c0153e38>] vfs_write+0xb8/0x130
>  [<c0153f62>] sys_write+0x42/0x70
>  [<c02a0593>] syscall_call+0x7/0xb

I am really unsure who is at fault here. At least reiserfs does not hold any locks
at this point. Not even BKL.

> and one one kernel BUG report in the middle of them:
> kernel BUG at mm/highmem.c:178!
> invalid operand: 0000 [#2]
> Call Trace:
>  [<c01a6040>] reiserfs_unprepare_pages+0x30/0x70
>  [<c01a72b5>] reiserfs_file_write+0x4e5/0x595

Hm, this one is real.
Try the patch below.
I wonder how you was able to hit this reiserfs_unprepare_pages() codepath at all.
Were there any messages prior to the bug? Or was there out of space situation?

Thanks for the report.

Bye,
    Oleg

===== fs/reiserfs/file.c 1.20 vs edited =====
--- 1.20/fs/reiserfs/file.c	Wed Jun  4 11:50:34 2003
+++ edited/fs/reiserfs/file.c	Wed Aug  6 19:11:01 2003
@@ -555,7 +555,6 @@
 	struct page *page = prepared_pages[i];
 
 	try_to_free_buffers(page);
-	kunmap(page);
 	unlock_page(page);
 	page_cache_release(page);
     }