Re: "BUG: scheduling while atomic: pdflush/30/0x00000002" in latest git

[Arjan van de Ven <arjan@infradead.org>]

On Thu, 08 Jan 2009 08:37:52 -0600
Dave Kleikamp <shaggy@linux.vnet.ibm.com> wrote:

> Adding cc:lix-fsdevel
> 
> On Thu, 2009-01-08 at 16:07 +0800, Grissiom wrote:
> > When I using the latest git version, I occasionally got this in
> > dmesg:
> > 
> > [ 2008.237234] BUG: scheduling while atomic: pdflush/30/0x00000002
> > [ 2008.237240] 2 locks held by pdflush/30:
> > [ 2008.237244]  #0:  (mutex){--..}, at: [<c01a57a1>]
> > sync_filesystems+0x11/0x120 [ 2008.237258]  #1:  (sb_lock){--..},
> > at: [<c01a57ab>] sync_filesystems+0x1b/0x120
> > [ 2008.237277] Modules linked in: fuse ricoh_mmc b43
> > [ 2008.237288] Pid: 30, comm: pdflush Not tainted
> > 2.6.28-g14-rfkill-nophy-ledon-07485-g9e42d0c #62
> > [ 2008.237294] Call Trace:
> > [ 2008.237303]  [<c04d7576>] schedule+0x326/0x8e0
> > [ 2008.237311]  [<c01500d3>] __lock_acquire+0x293/0xa20
> > [ 2008.237321]  [<c014e307>] mark_held_locks+0x67/0x80
> > [ 2008.237330]  [<c04da58c>] _spin_unlock_irqrestore+0x4c/0x60
> > [ 2008.237339]  [<c014e519>] trace_hardirqs_on_caller+0x149/0x1a0
> > [ 2008.237351]  [<c0143d55>]
> > async_synchronize_cookie_special+0xb5/0x140 [ 2008.237362]
> > [<c013e1a0>] autoremove_wake_function+0x0/0x40 [ 2008.237372]
> > [<c01a57fc>] sync_filesystems+0x6c/0x120 [ 2008.237381]
> > [<c017ee90>] pdflush+0x0/0x1e0 [ 2008.237392]  [<c01c0b90>]
> > do_sync+0x20/0x60 [ 2008.237402]  [<c01c0bda>] sys_sync+0xa/0x10
> > [ 2008.237412]  [<c017ef9e>] pdflush+0x10e/0x1e0
> > [ 2008.237420]  [<c014e519>] trace_hardirqs_on_caller+0x149/0x1a0
> > [ 2008.237429]  [<c017de80>] laptop_flush+0x0/0x10
> > [ 2008.237437]  [<c013dea2>] kthread+0x42/0x70
> > [ 2008.237444]  [<c013de60>] kthread+0x0/0x70
> > [ 2008.237452]  [<c0103c03>] kernel_thread_helper+0x7/0x14
> > (repeat some times)
> > 
> 
> The offender is
> http://git.kernel.org/gitweb.cgi?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=efaee192
> 
> async_synchronize_full_special() shouldn't be called while holding a
> spinlock, sb_lock.
> 
> I think this patch should fix it.  Arjan, would this work?

yeah it will

woops/