Re: [PATCH 04/10] locking: export osq_lock()/osq_unlock()

From: Kent Overstreet
Date: Fri May 18 2018 - 06:36:01 EST


On Fri, May 18, 2018 at 01:08:08PM +0200, Peter Zijlstra wrote:
> On Fri, May 18, 2018 at 06:18:04AM -0400, Kent Overstreet wrote:
> > On Fri, May 18, 2018 at 11:52:04AM +0200, Peter Zijlstra wrote:
> > > On Fri, May 18, 2018 at 03:49:06AM -0400, Kent Overstreet wrote:
> > >
> > > No.. and most certainly not without a _very_ good reason.
> >
> > Ok, can I ask why?
>
> Because it is an internal helper for lock implementations that want to
> do optimistic spinning, it isn't a lock on its own and lacks several
> things you would expect.
>
> Using it is tricky and I don't trust random module authors to get 1+1
> right, let alone use this thing correctly (no judgement on your code,
> just in general).

Yeah, that's true. I just modelled my usage on the rwsem code.

It does strike me that the whole optimistic spin algorithm
(mutex_optimistic_spin() and rwsem_optimistic_spin()) are ripe for factoring
out. They've been growing more optimizations I see, and the optimizations mostly
aren't specific to either locks.

> > Here's what it's for:
>
> I'll try and have a look soon :-) But does that really _have_ to live in
> a module?

No, I'd be completely fine with moving six locks out of bcachefs, just don't
know that there'd be any other users. But I suppose we do have other filesystems
that use btrees, and that's what they're for.