Re: [External] Re: [PATCH] mm/z3fold: remove unneeded spinlock

[Zhongkun He <hezhongkun.hzk@bytedance.com>]

On Thu, Feb 8, 2024 at 11:29 AM Johannes Weiner <hannes@cmpxchg.org> wrote:
>
> On Mon, Feb 05, 2024 at 09:08:05AM +0800, Zhongkun He wrote:
> > On Mon, Feb 5, 2024 at 2:46 AM Matthew Wilcox <willy@infradead.org> wrote:
> > >
> > > On Sun, Feb 04, 2024 at 08:54:04PM +0800, Zhongkun He wrote:
> > > > There is no need to use spinlock in this section, so
> > > > remove it.
> > >
> > > I don't know this code at all, but the idiom is (relatively) common.
> > > It waits until anybody _currently_ holding the lock has released it.
> > >
> > > That would, eg, make it safe to free the 'pool' memory.
> > >
> > > > -     spin_lock(&pool->lock);
> > > > -     spin_unlock(&pool->lock);
> > >
> >
> > no,  please see the commit 'e774a7bc7f0adb'.
> >
> >         spin_lock(&pool->lock);
> > -       if (!list_empty(&page->lru))
> > -               list_del_init(&page->lru);
> >         spin_unlock(&pool->lock);
> >
> > The original purpose of this lock was to protect  page->lru,
> > which was removed now, so the spinlock is unnecessary.
>
> But pool->lock protects other stuff too? This doesn't rule out that
> there is some other ordering dependency on cycling the lock before
> freeing the entry. The person who would know best is the maintainer of
> this code, Vitaly. Let's CC him.

Thank you for your reply and look forward to hearing from Vitaly.