Re: [PATCH] Provide better printk() support for SMP machines

[Andrew Morton <akpm@osdl.org>]

David Howells <dhowells@redhat.com> wrote:
>
>  The attached patch prevents oopses interleaving with characters from other
>  printks on other machines by only zapping the locks if the oops is happening
>  on the machine holding the lock.

(s/machine/CPU/)

hm, I guess it adds a theoretical deadlock if some other CPU is in the
middle of printk and is trying to take some_lock and this CPU takes an oops
while holding some_lock.  Probably that's an acceptable tradeoff though.

>  --- linux-2.6.12-mm1/kernel/printk.c	2005-06-22 13:54:08.000000000 +0100
>  +++ linux-2.6.12-mm1-cachefs-wander/kernel/printk.c	2005-06-22 13:57:02.000000000 +0100
>  @@ -514,6 +514,9 @@ asmlinkage int printk(const char *fmt, .
>   	return r;
>   }
>   
>  +/* cpu currently holding logbuf_lock */
>  +static volatile int printk_cpu = -1;
>  +

Does this guy really need to be volatile?  Coud we use atomic_t and lose
that wmb()?

>   asmlinkage int vprintk(const char *fmt, va_list args)
>   {
>   	unsigned long flags;
>  @@ -522,11 +525,15 @@ asmlinkage int vprintk(const char *fmt, 
>   	static char printk_buf[1024];
>   	static int log_level_unknown = 1;
>   
>  -	if (unlikely(oops_in_progress))
>  +	if (unlikely(oops_in_progress) && printk_cpu == smp_processor_id())
>  +		/* If a crash is occurring during printk() on this CPU,
>  +		 * make sure we can't deadlock */

Methinks this should be raw_smp_processor_id().