On Wed, 16 Oct 2013 22:52:07 +0200 Peter Zijlstra <peterz@infradead.org>
wrote:

> Hey Neil;
> 
> it looks like its one of your patches isn't it?
> 
> http://www.spinics.net/lists/raid/msg44100.html
> http://www.spinics.net/lists/raid/msg44101.html
> 
> Given that I can't find them in a lkml archive means nobody's ever seen
> those patches.
> 
> Anyway; has that 3/3 patch ever been ran with lockdep enabled? 

I always run with lockdep enabled, and I have done at least basic testing
(I've been on leave for a few weeks and don't remember exactly where I got
to).
And I haven't seen any lockdep reports.

> 
> Stuff like:
> 
> +	for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
> +		spin_lock_init(conf->hash_locks + i);
> 
> And:
> 
> +static void __lock_all_hash_locks(struct r5conf *conf)
> +{
> +	int i;
> +	for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
> +		spin_lock(conf->hash_locks + i);
> +}
> 
> Tends to complain real loud.

Why is that?
Because "conf->hash_locks + i" gets used as the "name" of the lockdep map for
each one, and when they are all locked it looks like nested locking??

the lock_all_hash_locks doesn't get called very often and my testing mustn't
have got that far.  I just tried something that would trigger the 
"lock_all_device_hash_locks_irq" (as it is in the current version) and it
went "splat" just as you said it would.
Thanks.

Do you have a suggestion for how to make this work?
Would
    spin_lock_nested(conf->hash_locks + i, i)
do the trick?

Thanks,
NeilBrown

> 
> This leaves one to wonder... 
>   'fancy' locking scheme:1, validation effort:0
>