Re: oops in old isdn4linux and 2.6.2-rc3 (was in -rc2 too)

[Karsten Keil]

On Tue, Feb 10, 2004 at 01:53:42AM +0000, David Woodhouse wrote:
> On Mon, 2004-02-09 at 12:13 +0000, David Woodhouse wrote:
> > Dies horribly upon an incoming v.110 call though...
> 
> > NMI Watchdog detected LOCKUP on CPU1, eip f8a5cd65, registers:
> > CPU:    1
> > EIP:    0060:[<f8a5cd65>]    Not tainted
> > EFLAGS: 00000086
> > EIP is at .text.lock.hfc_pci+0xfb/0x186 [hisax]
> 
> This is line 1391 of hfc_pci.c, in hfcpci_l2l1().
> It's trying to relock &bcs->cs->lock which is evidently already locked.
> 
> This is because isdn_v110_stat_callback() is being called with the
> spinlock held when we've transmitted a skb... and it's trying to
> transmit another immediately.

Your analyse is correct, but the error is, that
ll_writewakeup is called with the lock.
Upper layers should not be called in this path.

> 
> > eax: f6531000   ebx: 00000120   ecx: 00000000   edx: f6288a40
> > esi: f6531614   edi: f652ac00   ebp: f6288a40   esp: f7f81b40
> > ds: 007b   es: 007b   ss: 0068
> > Process events/1 (pid: 7, threadinfo=f7f80000 task=f7f94ce0)
> > Stack: f6288800 00000028 c0224fea 00000097 f6531210 f652ac00 00000028 f6288800
> >        f8a533e6 f652ac00 00000120 f6288a40 0000003c 00000020 f75b7810 f6d52000
> >        f6288800 00000000 00000001 f89fb58f 00000000 00000001 00000001 f6288800
> > Call Trace:
> 	<hfcpci_l2l1>
> >  [<f8a533e6>] HiSax_writebuf_skb+0x156/0x230 [hisax]
> >  [<f89fb58f>] isdn_v110_stat_callback+0x2bf/0x330 [isdn]
> >  [<f89fc57e>] isdn_status_callback+0x76e/0x8e0 [isdn]
> >  [<f89fc09c>] isdn_status_callback+0x28c/0x8e0 [isdn]
> >  [<f8a52231>] ll_writewakeup+0x61/0xa0 [hisax]
> >  [<f8a5af40>] hfcpci_fill_fifo+0x4e0/0x6a0 [hisax]
> >  [<f8a5c2d2>] hfcpci_send_data+0x22/0x50 [hisax]
> >  [<f8a5c5d5>] hfcpci_l2l1+0xa5/0x200 [hisax]
> >  [<f8a533e6>] HiSax_writebuf_skb+0x156/0x230 [hisax]
> >  [<f89fb3dc>] isdn_v110_stat_callback+0x10c/0x330 [isdn]
> >  [<f89fc109>] isdn_status_callback+0x2f9/0x8e0 [isdn]
> 
> I'm going to try an _evil_ hack like this...
> 

Yes this should help you for the moment.


> --- drivers/isdn/hisax/hfc_pci.c.orig	2004-02-10 01:49:49.278608994 +0000
> +++ drivers/isdn/hisax/hfc_pci.c	2004-02-10 01:50:15.404811276 +0000
> @@ -597,7 +597,7 @@
>  	u_char new_f1, *src, *dst;
>  	unsigned short *z1t, *z2t;
>  
> -	if (!bcs->tx_skb)
> +	if (!bcs->tx_skb && !(bcs->tx_skb = skb_dequeue(&bcs->squeue)))
>  		return;
>  	if (bcs->tx_skb->len <= 0)
>  		return;
> @@ -1388,7 +1388,13 @@
>  
>  	switch (pr) {
>  		case (PH_DATA | REQUEST):
> -			spin_lock_irqsave(&bcs->cs->lock, flags);
> +			local_irq_save(flags);
> +			if (!spin_trylock(&bcs->cs->lock)) {
> +				/* Possible recursion. Queue it and return */
> +				local_irq_restore(flags);
> +				skb_queue_tail(&bcs->squeue, skb);
> +				break;
> +			}
>  			if (bcs->tx_skb) {
>  				skb_queue_tail(&bcs->squeue, skb);
>  			} else {
> 
> 
> -- 
> dwmw2
> 

-- 
Karsten Keil
SuSE Labs
ISDN development
-
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/