Jeremy first noticed this
http://marc.info/?l=3Dlinux-kernel&m=3D121783008503477&w=3D2

[    3.132333] ------------[ cut here ]------------
[    3.132343] kernel BUG at /home/ijc/development/kernel/2.6.git/lib/radix=
-tree.c:473!
[    3.132348] invalid opcode: 0000 [#1] SMP=20
[    3.132352] Modules linked in:
[    3.132356]=20
[    3.132363] Pid: 580, comm: debconf Tainted: G        W (2.6.26 #27)
[    3.132368] EIP: 0061:[<c01f5279>] EFLAGS: 00010002 CPU: 0
[    3.132375] EIP is at radix_tree_tag_set+0x1d/0x9f
[    3.132379] EAX: c203af30 EBX: c261b8c0 ECX: 00000000 EDX: 00000001
[    3.132383] ESI: 00000000 EDI: 00000001 EBP: c7977ce8 ESP: c7977cc8
[    3.132387]  DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0069
[    3.132392] Process debconf (pid: 580, ti=3Dc7976000 task=3Dc538a240 tas=
k.ti=3Dc7976000)
[    3.132396] Stack: fffede22 00000000 00000001 c203af30 c203af2c c261b8c0=
 c203af2c 00000001=20
[    3.132406]        c7977cfc c01a1570 c261b8c0 c203af2c c2563000 c7977d0c=
 c01a1a24 c261b8c0=20
[    3.132416]        00000001 c7977d1c c01682fc c261b8c0 00000001 c7977d2c=
 c0169185 c261b8c0=20
[    3.132425] Call Trace:
[    3.132428]  [<c01a1570>] ? __set_page_dirty+0xdf/0x11f
[    3.132434]  [<c01a1a24>] ? __set_page_dirty_buffers+0x68/0x6c
[    3.132441]  [<c01682fc>] ? set_page_dirty+0x34/0x94
[    3.132446]  [<c0169185>] ? set_page_dirty_balance+0xe/0x3c
[    3.132452]  [<c016eec6>] ? __do_fault+0x35d/0x37e
[    3.132458]  [<c01707a4>] ? handle_mm_fault+0x45d/0x9c9
[    3.132463]  [<c01939a0>] ? __d_lookup+0xb7/0xeb
[    3.132469]  [<c0180d18>] ? kfree+0x81/0x88
[    3.132474]  [<c02dfaf7>] ? _spin_unlock_irqrestore+0x19/0x1f
[    3.132481]  [<c02dfaf7>] ? _spin_unlock_irqrestore+0x19/0x1f
[    3.132487]  [<c011b5ee>] ? do_page_fault+0x3be/0x8d0
[    3.132493]  [<c02061ef>] ? fb_ioctl+0x1a2/0x2de
[    3.132499]  [<c011ad1c>] ? pvclock_clocksource_read+0x48/0xa3
[    3.132506]  [<c02dfaf7>] ? _spin_unlock_irqrestore+0x19/0x1f
[    3.132512]  [<c013e02b>] ? hrtimer_start+0x12a/0x144
[    3.132519]  [<c0104682>] ? xen_mc_flush+0x123/0x160
[    3.132525]  [<c0104699>] ? xen_mc_flush+0x13a/0x160
[    3.136027]  [<c010436c>] ? xen_leave_lazy+0x12/0x14
[    3.136027]  [<c01075bc>] ? __switch_to+0xec/0x126
[    3.136027]  [<c012755c>] ? finish_task_switch+0x32/0xa5
[    3.136027]  [<c02de601>] ? schedule+0x6cc/0x735
[    3.136027]  [<c018e377>] ? vfs_ioctl+0x57/0x69
[    3.136027]  [<c018e636>] ? sys_ioctl+0x50/0x5a
[    3.136027]  [<c011b230>] ? do_page_fault+0x0/0x8d0
[    3.136027]  [<c02dfd3a>] ? error_code+0x72/0x78
[    3.136027]  =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D
[    3.136027] Code: b4 89 42 04 83 c4 50 89 d8 5b 5e 5f 5d c3 55 89 e5 57 =
56 53 83 ec 14 89 45 ec 89 55 e8 89 4d e4 8b 30 3b 14 b5 88 52 3a c0 76 04 =
<0f> 0b eb fe 8b 45 ec 8b 4d e4 8b 58 08 6b c6 06 c1 e1 03=20
[    3.136027] EIP: [<c01f5279>] radix_tree_tag_set+0x1d/0x9f SS:ESP 0069:c=
7977cc8
[    3.136027] ---[ end trace 991579adcab01bbf ]---

I've bisected it down to:
commit 14fcc23fdc78e9d32372553ccf21758a9bd56fa1
Author: Hugh Dickins <hugh@veritas.com>
Date:   Mon Jul 28 15:46:19 2008 -0700

    tmpfs: fix kernel BUG in shmem_delete_inode
   =20
    SuSE's insserve initscript ordering program hits kernel BUG at mm/shmem=
.c:814
    on 2.6.26.  It's using posix_fadvise on directories, and the shmem_read=
page
    method added in 2.6.23 is letting POSIX_FADV_WILLNEED allocate useless =
pages
    to a tmpfs directory, incrementing i_blocks count but never decrementin=
g it.
   =20
    Fix this by assigning shmem_aops (pointing to readpage and writepage an=
d
    set_page_dirty) only when it's needed, on a regular file or a long syml=
ink.
   =20
    Many thanks to Kel for outstanding bugreport and steps to reproduce it.
   =20
    Reported-by: Kel Modderman <kel@otaku42.de>
    Tested-by: Kel Modderman <kel@otaku42.de>
    Signed-off-by: Hugh Dickins <hugh@veritas.com>
    Cc: <stable@kernel.org>             [2.6.25.x, 2.6.26.x]
    Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
    Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Reverting this patch from current Linus tree
(b635acec48bcaa9183fcbf4e3955616b0d4119b5) causes the problem to go
away. I haven't yet seen the link between the backtrace and this
changeset though.

Ian.
--=20
Ian Campbell

Preserve wildlife -- pickle a squirrel today!