Re: [PATCH] Make printk lock when appropriate...

[Andrew Morton]

Keith Owens wrote:
> 
> What about this code instead of spin_lock_irqsave?  Pity we do not have
> spin_trylock_irqsave();
> 
> local_irq_save(flags);
> if (!spin_trylock(&console_lock)) {
>         local_irq_restore(flags);
>         return(0);
> }

If somebody tries to call printk with console_lock held then something
very bad is happening.  That message wants to come out.

I suggest that rather than returning we should bust the spinlock, as is
done in the NMI handler (arch/i386/kernel/traps.c).  Move that code to
printk.

And let's move that dopey `buf' declaration inside printk() and make it
static.