[IRDA]: Lockdep fix.
authorPeter Zijlstra <a.p.zijlstra@chello.nl>
Wed, 22 Nov 2006 01:33:01 +0000 (17:33 -0800)
committerDavid S. Miller <davem@davemloft.net>
Wed, 22 Nov 2006 01:33:01 +0000 (17:33 -0800)
On Sat, 2006-11-18 at 16:12 +0300, Andrey Borzenkov wrote:

> =============================================
> [ INFO: possible recursive locking detected ]
> 2.6.19-rc5-2avb #2
> - ---------------------------------------------
> pppd/26425 is trying to acquire lock:
>  (&hashbin->hb_spinlock){....}, at: [<dfdea87a>] irlmp_slsap_inuse+0x5a/0x170
> [irda]
>
> but task is already holding lock:
>  (&hashbin->hb_spinlock){....}, at: [<dfdea857>] irlmp_slsap_inuse+0x37/0x170
> [irda]
>
> other info that might help us debug this:
> 1 lock held by pppd/26425:
>  #0:  (&hashbin->hb_spinlock){....}, at: [<dfdea857>]
> irlmp_slsap_inuse+0x37/0x170 [irda]
>
> stack backtrace:
>  [<c010413c>] dump_trace+0x1cc/0x200
>  [<c010418a>] show_trace_log_lvl+0x1a/0x30
>  [<c01047f2>] show_trace+0x12/0x20
>  [<c01048c9>] dump_stack+0x19/0x20
>  [<c01346ca>] __lock_acquire+0x8fa/0xc20
>  [<c0134d2d>] lock_acquire+0x5d/0x80
>  [<c02a851c>] _spin_lock+0x2c/0x40
>  [<dfdea87a>] irlmp_slsap_inuse+0x5a/0x170 [irda]
>  [<dfdebab2>] irlmp_open_lsap+0x62/0x180 [irda]
>  [<dfdf35d1>] irttp_open_tsap+0x181/0x230 [irda]
>  [<dfdc0c3d>] ircomm_open_tsap+0x5d/0xa0 [ircomm]
>  [<dfdc05d8>] ircomm_open+0xb8/0xd0 [ircomm]
>  [<dfdd0477>] ircomm_tty_open+0x4f7/0x570 [ircomm_tty]
>  [<c020bbe4>] tty_open+0x174/0x340
>  [<c016bd69>] chrdev_open+0x89/0x170
>  [<c0167bd6>] __dentry_open+0xa6/0x1d0
>  [<c0167da5>] nameidata_to_filp+0x35/0x40
>  [<c0167df9>] do_filp_open+0x49/0x50
>  [<c0167e47>] do_sys_open+0x47/0xd0
>  [<c0167f0c>] sys_open+0x1c/0x20
>  [<c010307d>] sysenter_past_esp+0x56/0x8d
>  [<b7f86410>] 0xb7f86410
>  =======================

The comment at the nesting lock says:

/* Careful for priority inversions here !
 * irlmp->links is never taken while another IrDA
 * spinlock is held, so we are safe. Jean II */

So, under the assumption the author was right, it just needs a lockdep
annotation.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/irda/irlmp.c

index 5073261..fede837 100644 (file)
@@ -1678,7 +1678,8 @@ static int irlmp_slsap_inuse(__u8 slsap_sel)
         *  every IrLAP connection and check every LSAP associated with each
         *  the connection.
         */
-       spin_lock_irqsave(&irlmp->links->hb_spinlock, flags);
+       spin_lock_irqsave_nested(&irlmp->links->hb_spinlock, flags,
+                       SINGLE_DEPTH_NESTING);
        lap = (struct lap_cb *) hashbin_get_first(irlmp->links);
        while (lap != NULL) {
                IRDA_ASSERT(lap->magic == LMP_LAP_MAGIC, goto errlap;);