lkml.org 
[lkml]   [2007]   [Apr]   [30]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
Subject2.6.21-mm1: many processes end up in D state
Hi,

I have a problem with higher disk loads (e.g. running git-log or yum update).
Many processes end up in D state and system is unusable -- I'm not able to run
anything but smooth mouse moving when this happens.

If I wait for a 20-30sec it becomes usable. This happens in 2.6.21-rc7-mm2 and
also in 2007-04-28-05-06 broken-out snapshot. I think 2.6.21-rc6-mm1 worked
fine, but I'm uncertain. If it is important, let me know to re-test.

sysrq-t:

yum D 0000004C 0 2976 2955 (NOTLB)
c455bd00 00200082 8c68ea9e 0000004c 00000000 00000000 c455bcdc c455bc9c
c455bc9c c2d6a580 8c68ea9e 0000004c c455bcc0 c2d6a694 c1809980 00000000
c455bcd0 c4640c80 000036a0 00000001 c455bcf0 c02a7132 c3682164 00001cde
Call Trace:
[<c037834c>] io_schedule+0x21/0x2e
[<c014d505>] sync_page+0x38/0x43
[<c0378581>] __wait_on_bit_lock+0x40/0x63
[<c014d4b6>] __lock_page+0x54/0x5c
[<c014da81>] do_generic_mapping_read+0x1dc/0x57e
[<c014fc84>] generic_file_aio_read+0xc9/0x1b5
[<c016c31c>] do_sync_read+0xd0/0x106
[<c016cada>] vfs_read+0x89/0x11d
[<c016d02c>] sys_pread64+0x64/0x68
[<c0103ffc>] syscall_call+0x7/0xb
[<47176012>] 0x47176012
=======================
xterm S 00000051 0 2980 1 (NOTLB)
c4678ae0 00000082 b2fbf0ce 00000051 00000000 00000000 00000000 00000000
c4678b30 c469d540 b2fbf0ce 00000051 c4678ab0 c469d654 c1813980 00000001
c4678ad0 c47f0e40 c4678b00 00000286 c4678ae0 c012c077 00000001 00003286
Call Trace:
[<c03784ad>] schedule_timeout+0x44/0xa4
[<c0177ee3>] do_select+0x46d/0x55b
[<c017816f>] core_sys_select+0x19e/0x2c0
[<c01786f3>] sys_select+0xd3/0x18d
[<c0103ffc>] syscall_call+0x7/0xb
[<47180f98>] 0x47180f98
=======================
bash S 0000004A 0 2982 2980 (NOTLB)
c50b3e80 00000086 ccf4cc7e 0000004a 00000000 c021bed9 00000000 c45c0030
c50b3e30 c45c0030 ccf4cc7e 0000004a c50b3e50 c45c0144 c1809980 00000000
c1809e44 c4517900 00000000 000000b3 c50b3e90 c014f841 c477cb40 c2d79000
Call Trace:
[<c03784d3>] schedule_timeout+0x6a/0xa4
[<c0221960>] read_chan+0x1b9/0x5b9
[<c021e48f>] tty_read+0x75/0xaa
[<c016cada>] vfs_read+0x89/0x11d
[<c016cf3d>] sys_read+0x3d/0x64
[<c0103ffc>] syscall_call+0x7/0xb
[<4717883e>] 0x4717883e
=======================
metacity S 00000050 0 3005 2429 (NOTLB)
c2fd5bb0 00000086 5205656d 00000050 00000000 00000000 00000000 c441c51c
00000246 c45c0540 5205656d 00000050 c2fd5b80 c45c0654 c1809980 00000000
c1813e44 c27f4900 00002cb8 00000246 c2fd5ba0 c0135a80 c2fd5ba8 00002cb2
Call Trace:
[<c03784d3>] schedule_timeout+0x6a/0xa4
[<c0177905>] do_sys_poll+0x33f/0x46d
[<c0177a74>] sys_poll+0x41/0x43
[<c0103ffc>] syscall_call+0x7/0xb
[<4717e3a6>] 0x4717e3a6
=======================
ccpd D 0000004C 0 3009 2010 (NOTLB)
c50abbb0 00000082 f08ef585 0000004c 00000000 00000000 c180abc0 c22fc788
c50abba0 c505d070 f08ef585 0000004c 00000000 c505d184 c1809980 00000000
c21167f0 c2fc6740 c5794e3c c31eff74 c5794310 c22fc788 00000002 00001e64
Call Trace:
[<c01b64d9>] do_get_write_access+0x310/0x4f5
[<c01b66d9>] journal_get_write_access+0x1b/0x2a
[<c01b59ce>] __ext3_journal_get_write_access+0x19/0x3f
[<c01aa5f6>] ext3_reserve_inode_write+0x53/0x6c
[<c01aa62f>] ext3_mark_inode_dirty+0x20/0x37
[<c01ad350>] ext3_dirty_inode+0x6b/0x6d
[<c01868fb>] __mark_inode_dirty+0x2a/0x16d
[<c017eacc>] touch_atime+0x84/0xd8
[<c014dd08>] do_generic_mapping_read+0x463/0x57e
[<c014fc84>] generic_file_aio_read+0xc9/0x1b5
[<c016c31c>] do_sync_read+0xd0/0x106
[<c016cada>] vfs_read+0x89/0x11d
[<c016fc6e>] kernel_read+0x36/0x48
[<c016fd32>] prepare_binprm+0xb2/0xdf
[<c0171322>] do_execve+0xe6/0x1e1
[<c0102715>] sys_execve+0x2e/0x7f
[<c0103ffc>] syscall_call+0x7/0xb
[<471474da>] 0x471474da
=======================
bash D 0000004D 0 3011 2859 (NOTLB)
c471dc90 00000082 08f8373c 0000004d 00000000 00000000 c471dc30 c471dc2c
c471dc2c c4670540 08f8373c 0000004d c471dc50 c4670654 c1809980 00000000
c471dc60 c2f3f040 00000005 00000001 c471dc80 c02a7132 c591af64 00001eea
Call Trace:
[<c037834c>] io_schedule+0x21/0x2e
[<c014d505>] sync_page+0x38/0x43
[<c0378581>] __wait_on_bit_lock+0x40/0x63
[<c014d4b6>] __lock_page+0x54/0x5c
[<c014da81>] do_generic_mapping_read+0x1dc/0x57e
[<c014fc84>] generic_file_aio_read+0xc9/0x1b5
[<c016c31c>] do_sync_read+0xd0/0x106
[<c016cada>] vfs_read+0x89/0x11d
[<c016fc6e>] kernel_read+0x36/0x48
[<c016fd32>] prepare_binprm+0xb2/0xdf
[<c0171322>] do_execve+0xe6/0x1e1
[<c0102715>] sys_execve+0x2e/0x7f
[<c0103ffc>] syscall_call+0x7/0xb
[<471474da>] 0x471474da
=======================
bash D 0000004D 0 3012 2884 (NOTLB)
c50d5a80 00000082 a21bfc35 0000004d 00000000 00000000 c180abc0 c2033dd4
c50d5a70 c2f4f070 a21bfc35 0000004d 00000000 c2f4f184 c1809980 00000000
c21167bc c47f0c80 c2033310 c22fc788 c5794e3c c2033dd4 00000002 0000216d
Call Trace:
[<c01b64d9>] do_get_write_access+0x310/0x4f5
[<c01b66d9>] journal_get_write_access+0x1b/0x2a
[<c01b59ce>] __ext3_journal_get_write_access+0x19/0x3f
[<c01aa5f6>] ext3_reserve_inode_write+0x53/0x6c
[<c01aa62f>] ext3_mark_inode_dirty+0x20/0x37
[<c01ad350>] ext3_dirty_inode+0x6b/0x6d
[<c01868fb>] __mark_inode_dirty+0x2a/0x16d
[<c017eacc>] touch_atime+0x84/0xd8
[<c017499a>] __link_path_walk+0x893/0xca4
[<c0174df1>] link_path_walk+0x46/0xc3
[<c0175038>] do_path_lookup+0x86/0x1b0
[<c0175ad0>] __path_lookup_intent_open+0x44/0x7f
[<c0175b83>] path_lookup_open+0x21/0x27
[<c016fd86>] open_exec+0x27/0xa2
[<c01993a3>] load_elf_binary+0x1482/0x1acf
[<c016fa1d>] search_binary_handler+0x6e/0x19f
[<c0171386>] do_execve+0x14a/0x1e1
[<c0102715>] sys_execve+0x2e/0x7f
[<c0103ffc>] syscall_call+0x7/0xb
[<471474da>] 0x471474da
=======================

Note that yum works on lvm on raid0 and git too, but on the another md volume.
Both ext3s. Drivers are sata_promise and ata_piix (sata disk); CFQ scheduler.
Using noop is no change (but seems to be harder to reproduce with it). I figured
out that it probably happens when 2+ processes are on both "processors" (HT on
P4) and are IO wait (multiload-applet shows red above the half).

Swap usage is 0 all the time.

regards,
--
http://www.fi.muni.cz/~xslaby/ Jiri Slaby
faculty of informatics, masaryk university, brno, cz
e-mail: jirislaby gmail com, gpg pubkey fingerprint:
B674 9967 0407 CE62 ACC8 22A0 32CC 55C3 39D4 7A7E
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2007-04-30 17:43    [W:0.112 / U:0.332 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site