Re: 2.6.34-rc1: rcu lockdep bug?

["Américo Wang" <xiyou.wangcong@gmail.com>]

On Mon, Mar 15, 2010 at 6:41 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> Le lundi 15 mars 2010 à 18:12 +0800, Américo Wang a écrit :
>> On Mon, Mar 15, 2010 at 6:04 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
>> > Le lundi 15 mars 2010 à 17:39 +0800, Américo Wang a écrit :
>> >
>> >>
>> >> Ok, I think I found what lockdep really complains about, it is that we took
>> >> spin_lock in netpoll_poll_lock() which is in hardirq-enabled environment,
>> >> later, we took another spin_lock with spin_lock_irqsave() in netpoll_rx(),
>> >> so lockdep thought we broke the locking rule.
>> >>
>> >> I don't know why netpoll_rx() needs irq disabled, it looks like that no one
>> >> takes rx_lock in hardirq context. So can we use spin_lock(&rx_lock)
>> >> instead? Or am I missing something here? Eric? David?
>> >
>> > I am a bit lost.
>> >
>> > Could you give the complete picture, because I cannot find it in my
>> > netdev archives.
>> >
>>
>> Sure, sorry for this.
>>
>> Here is the whole thread:
>>
>> http://lkml.org/lkml/2010/3/11/100
>
> OK thanks
>
> netpoll_rx() can be called from hard irqs (netif_rx()), so rx_lock
> definitly needs irq care.
>
> netpoll_poll_lock() does take a spinlock with irq enabled, but its not
> rx_lock, its napi->poll_lock.

Yeah, I knew, but besides rcu locks, these two locks are the only
locks that can be taken in the call chain. I suspect lockdep got
something wrong.

>
> I dont see what could be the problem, is it reproductible with vanilla
> kernel ?
>

No. I don't know why, my patch doesn't touch any function in the
call chain.

I already "fix" this in another way, so no need to worry this any more.

Thanks for your help!