Re: Locking problem reported for mainline

[Larry Finger <Larry.Finger@lwfinger.net>]

On 01/11/2011 01:45 PM, Bob Copeland wrote:
> On Tue, Jan 11, 2011 at 11:31 AM, Larry Finger
> <Larry.Finger@lwfinger.net> wrote:
>> Is there a document that explains what the meaning of these semantics?
>>
>> inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage.
>> kdostartupconfi/3502 [HC1[1]:SC0[0]:HE0:SE1] takes:
>>  (&(&list->lock)->rlock#5){?.-...}, at: [<ffffffff812995c6>]
>> skb_queue_tail+0x26/0x60
>> {HARDIRQ-ON-W} state was registered at:
> 
> I'm not sure about all the HC1[1]:SC0[0] etc stuff, but check out
> Documentation/lockdep-design.txt for the basics.

That one I had read.

> In this case, someone took a lock with interrupts enabled (HARDIRQ-ON-W)
> while someone else took it in a hard IRQ context (IN-HARDIRQ-W) where
> they are normally disabled.  The problem of course is:
> 
> cpu0:
> spin_lock(&foo);
> do some stuff protected by foo;
> 
> ----> interrupt happens here
>    spin_lock(&foo);  /* darn, deadlock! */
>    other stuff;
>    spin_unlock(&foo);
> <----
> 
> spin_unlock(&foo);
> 
> Could be a missing _irqsave() if it's not, as Stanislaw suggested, a false
> positive.

I suspected the message meant mixed interrupts disabled/enabled, but thanks for
the confirmation.

Thanks,

Larry