Re: [PATCH 1/3] stop_machine: kill __stop_machine()

[Tejun Heo <tj@kernel.org>]

On Thu, Jun 16, 2011 at 02:12:52PM +0200, Peter Zijlstra wrote:
> On Tue, 2011-06-14 at 19:06 +0200, Tejun Heo wrote:
> > +++ b/arch/x86/kernel/alternative.c
> > @@ -719,8 +719,7 @@ void *__kprobes text_poke_smp(void *addr, const void *opcode, size_t len)
> >         tpp.nparams = 1;
> >         atomic_set(&stop_machine_first, 1);
> >         wrote_text = 0;
> > -       /* Use __stop_machine() because the caller already got online_cpus. */
> > -       __stop_machine(stop_machine_text_poke, (void *)&tpp, cpu_online_mask);
> > +       stop_machine(stop_machine_text_poke, (void *)&tpp, cpu_online_mask);
> >         return addr;
> >  }
> 
> Please have a look at: 
> 
> ---
> commit d91309f69b7bdb64aeb30106fde8d18c5dd354b5
> Author: Peter Zijlstra <peterz@infradead.org>
> Date:   Fri Feb 11 22:07:46 2011 +0100
> 
>     x86: Fix text_poke_smp_batch() deadlock
>     
>     Fix this deadlock - we are already holding the mutex:

Ah, right, I thought cpu_hotplug_begin() sets active_writer and then
returns with cpu_hotplug.lock released.  Weird locking there.
Anyways, this one should be dropped then and I don't think it affects
the rest of the series much.

Thanks.

-- 
tejun