Re: [PATCH v2] mm: use memory barriers for non PMD-sized folios to sync
From: Lorenzo Stoakes (ARM)
Date: Wed Nov 18 2026 - 07:26:06 EST
On Wed, Nov 18, 2026 at 02:19:56PM +0800, Baolin Wang wrote:
> > Do we have any way to measure the benefit of this? I think we'll need that to
> > convince reviewers (like Lorenzo :) )
>
> As I replied to Lorenzo, I will try, though it is hard to measure the
> performance for this uncommon race case.
I mean it's not a race unless you stop taking the PMD lock (I think the commit
message has to be a lot clearer about that), and I'm not really happy to move
from easy-to-reason-about PMD lock to hard-to-reason-about and fragile memory
barriers unless we have a _really_ good reason.
I'd say given what you've said that we don't have a really good reason :)
--
Cheers, Lorenzo