Re: [RFC][PATCHv4 5/6] printk: use printk_safe buffers

[Petr Mladek <pmladek@suse.com>]

On Fri 2016-10-28 00:49:32, Sergey Senozhatsky wrote:
> Use printk_safe per-CPU buffers in in printk recursion-prone blocks:
> -- around logbuf_lock protected sections in vprintk_emit() and
>    console_unlock()
> -- around down_trylock_console_sem() and up_console_sem()
> 
> diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> index 4675b8d..5907e92 100644
> --- a/kernel/printk/printk.c
> +++ b/kernel/printk/printk.c
> @@ -1823,13 +1837,12 @@ asmlinkage int vprintk_emit(int facility, int level,
>  		 */
>  		if (!oops_in_progress && !lockdep_recursing(current))
> {
>  			recursion_bug = true;
> -			local_irq_restore(flags);
> +			printk_safe_exit(flags);
>  			return 0;
>  		}
>  		zap_locks();
>  	}
>  
> -	lockdep_off();

I really like this patch. The only small problem is that it enables
lockdep and it does not explain why it is safe. The change itself
looks fine but it took me some time to prove why. IMHO, it is
worth a comment.

One thing is printk() recursion caused by lockdep warning
triggered from inside vprintk_emit(). It is safe because
the critical sections are guarded by printk_safe_enter()/exit()
now.

Another thing is lockdep recursion caused by catching another lockdep
issue when printing warning about the first one. This is safe
because lockdep protects itself. First, it sets and checks
current->lockdep_recursion around the critical sections.
Second, further checks are disabled entirely once first
lockdep issue is found.

If you add some comments about lockdep, feel free to use:

Reviewed-by: Petr Mladek <pmladek@suse.com>

Best Regards,
Petr