Re: [RFC][PATCH v2 -next 2/2] Adding lock operations to kmsg_dump()/pstore_dump()

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Fri, 2011-10-21 at 17:21 -0400, Seiji Aguchi wrote:
> +++ b/fs/pstore/platform.c
> @@ -97,6 +97,17 @@ static void pstore_dump(struct kmsg_dumper *dumper,
>         else
>                 why = "Unknown";
>  
> +       /*
> +        * pstore_dump() is called after smp_send_stop() in panic path.
> +        * So, spin_lock should be bust for avoiding deadlock.
> +        */
> +       if (reason == KMSG_DUMP_PANIC)
> +               spin_lock_init(&psinfo->buf_lock);
> +
> +       /*
> +        * While a cpu is in NMI handler, other cpus may be running.
> +        * So, trylock should be called so that lockdep checking works.
> +        */

Don't be silly, lockdep doesn't cover NMI, in fact you shouldn't use
locks from NMI context ever.

>         if (in_nmi()) {
>                 is_locked = spin_trylock(&psinfo->buf_lock);
>                 if (!is_locked)
> diff --git a/kernel/printk.c b/kernel/printk.c
> index 1455a0d..f51f547 100644
> --- a/kernel/printk.c
> +++ b/kernel/printk.c
> @@ -1730,15 +1730,37 @@ void kmsg_dump(enum kmsg_dump_reason reason)
>         struct kmsg_dumper *dumper;
>         const char *s1, *s2;
>         unsigned long l1, l2;
> -       unsigned long flags;
> +       unsigned long flags = 0;
> +       int is_locked = 0;
>  
>         /* Theoretically, the log could move on after we do this, but
>            there's not a lot we can do about that. The new messages
>            will overwrite the start of what we dump. */
> -       raw_spin_lock_irqsave(&logbuf_lock, flags);
> +
> +       /*
> +        * kmsg_dump() is called after smp_send_stop() in panic path.
> +        * So, spin_lock should be bust for avoiding deadlock.
> +        */
> +       if (reason == KMSG_DUMP_PANIC)
> +               raw_spin_lock_init(&logbuf_lock); 

In both cases where you bust the spinlock at least yell loudly and
disable lock debugging.

And I guess this is where Don wants to use NMIs for smp_send_stop() so
what you get around the fact that this lock you're busting disabled
IRQs?

All in all this patch is way ugly and doesn't make me feel all warm and
fuzzy.