Re: [RFC PATCH v2 0/8] Removal of AioContext lock, bs->parents and ->children: new rwlock

[Emanuele Giuseppe Esposito]

Am 28/04/2022 um 12:34 schrieb Stefan Hajnoczi:
> On Tue, Apr 26, 2022 at 04:51:06AM -0400, Emanuele Giuseppe Esposito wrote:
>> Next step is the most complex one: figure where to put the rdlocks.
> 
> I got a little lost at this step but will hopefully understand more when
> reading the patches. Earlier it said there is a counter for readers, so
> it seems trivial to make it a recursive lock. I don't understand why you
> want to avoid a recursive locking - the complications you described stem
> from the fact that you want to avoid recursive locks?

I'll let Paolo to further elaborate on this. I think it would be cleaner
if we try at least not to have recursive locks.

> 
> Secondly, can you describe the role of the lock? The text mentions the
> old AioContext lock that is being replaced but doesn't define this new
> lock in its own right. Is the idea to make ->parents and ->children list
> accesses thread-safe and all loops "safe" so they don't break when a BDS
> is removed?

Yes. Replace what AioContext lock is doing right now.

Emanuele
> 
> Stefan
>