[Bug 1549] New: /bin/sync hangs running reaim on 8-way x86 using ext2 file system

From: Martin J. Bligh
Date: Mon Nov 17 2003 - 20:07:41 EST


http://bugme.osdl.org/show_bug.cgi?id=1549

Summary: /bin/sync hangs running reaim on 8-way x86 using ext2
file system
Kernel Version: 2.6.0-test9-bk19
Status: NEW
Severity: normal
Owner: akpm@xxxxxxxxx
Submitter: dmo@xxxxxxxx


Distribution:rh9
Hardware Environment:8-processor, 700 mgz pentium III, 8 gibabytes memory
Software Environment:reaim test:http://sourceforge.net/projects/re-aim-7
Problem Description: test hangs because /bin/sync never completes. The sync
binaries are usually hung either trying to lock a page in the page cache (this
is the most common case):
sync D DA1D1E3C 4048 13674 1 13807 13806 (NOTLB)
da1d1e48 00000082 da1d0000 da1d1e3c c0231e48 f70f6c00 f69c6a00 c0226e16
f69c6a00 c633cbc0 f69c6a00 f69c6a00 dd7d0040 c633cbc0 000038ea ef356c53
00000092 f700e040 c633cbc0 c11a2908 c6281bf4 da1d1e54 c011e7f8 00000000
Call Trace:
[<c0231e48>] deadline_next_request+0x28/0x40
[<c0226e16>] elv_next_request+0x16/0x110
[<c011e7f8>] io_schedule+0x28/0x40
[<c013e8ef>] __lock_page+0xbf/0xf0
[<c011fb10>] autoremove_wake_function+0x0/0x50
[<c011fb10>] autoremove_wake_function+0x0/0x50
[<c0185748>] mpage_writepages+0x318/0x350
[<c0167ca0>] blkdev_writepage+0x0/0x30
[<c01691ff>] generic_writepages+0x1f/0x23
[<c014447e>] do_writepages+0x1e/0x40
[<c013e209>] __filemap_fdatawrite+0xe9/0x110
[<c013e247>] filemap_fdatawrite+0x17/0x20
[<c0160ac6>] sync_blockdev+0x26/0x50
[<c01840d9>] sync_inodes+0x89/0xa0
[<c0160c54>] do_sync+0x44/0x70
[<c0160c8f>] sys_sync+0xf/0x20
[<c0109729>] sysenter_past_esp+0x52/0x71

This happens with both deadline and anticipatory IO schedulers.
It happens about once every 3 or 4 runs of the "database" form of the
test.

Steps to reproduce:

submit the reaim test to an stp8 machine. specify no arguments to the reaim
will run the database test form. It may take 3 or 4 runs to observe this
test failure. Watch for tests that run for more than 3 hours. This also happens
less frequently on a four-way machine.

-
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/