Re: Oops at "NET: Registering protocol family 23" at boot with 2.6.0t9-bk

[Jean Tourrilhes <jt@bougret.hpl.hp.com>]

On Tue, Nov 04, 2003 at 12:27:00AM +0100, Martin Diehl wrote:
> On Sat, 1 Nov 2003, Stian Jordet wrote:
> 
> > NET: Registered protocol family 23
> > Unable to handle kernel NULL pointer dereference at virtual address 00000004
> > EIP is at dev_add_pack+0x4d/0xb0
> 
> which is at "next->prev = new" in __list_add_rcu()
> 
> > Call Trace:
> >  [<c05defad>] irda_init+0x3d/0x60
> >  [<c05c094c>] do_initcalls+0x2c/0xa0
> >  [<c01359af>] init_workqueues+0xf/0x24
> >  [<c01050f6>] init+0x56/0x180
> >  [<c01050a0>] init+0x0/0x180
> >  [<c01072b9>] kernel_thread_helper+0x5/0xc
>  
> Looks like irda_init is called before the ptype_base[] list heads are 
> initialized. I believe it was triggered by:
> 
> >>>>>>>>>>
> diff -Nru a/net/core/dev.c b/net/core/dev.c
> --- a/net/core/dev.c	Wed Oct 29 15:08:14 2003
> +++ b/net/core/dev.c	Wed Oct 29 15:08:14 2003
> @@ -3023,7 +3023,7 @@
>  	return rc;
>  }
>  
> -subsys_initcall(net_dev_init);
> +fs_initcall(net_dev_init);
> <<<<<<<<<<
>  
> from: cset-akpm@osdl.org|ChangeSet|20031029192849|64746.txt
> 
> As a workaround you might want to make the irda modular instead of 
> build-in. But I'm just wondering why other protocols shouldn't have the 
> same problem. Maybe they are not subsys_initcall like af_irda.
> 
> Martin

	Networking as a filesystem, that's new to me. And then, we
need irda up strictly before the drivers, so I would need to move irda
to fs_initcall(). Grrr....

	I'm in serious need of catching up. Martin, you sent me
previously two important crasher fixes, one for IrCOMM and one for
sir-dev. And I have my IrNET Oops fix. Anything else ?

	Regards,

	Jean