Re: [PATCH 1/1] mm: protect xa split stuff under lruvec->lru_lock during migration

[Zhaoyang Huang <huangzhaoyang@gmail.com>]

On Fri, Apr 12, 2024 at 8:24 PM Matthew Wilcox <willy@infradead.org> wrote:
>
> On Fri, Apr 12, 2024 at 02:43:53PM +0800, zhaoyang.huang wrote:
> > From: Zhaoyang Huang <zhaoyang.huang@unisoc.com>
> >
> > Livelock in [1] is reported multitimes since v515, where the zero-ref
> > folio is repeatly found on the page cache by find_get_entry. A possible
> > timing sequence is proposed in [2], which can be described briefly as
>
> I have no patience for going through another one of your "analyses".
>
> 1. Can you reproduce this bug without this patch?
> 2. Does the reproducer stop working after this patch?
>
> Otherwise I'm not interested.  Sorry.  You burnt all my good will.

This bug has been reported many times by three people including me as
below for at least two years, have you ever tried to solve it? Do Dave
and Brian also burnt your good will if you ever have? Be aware that
you are the maintainer who has the responsibility for maintaining the
code but not us. "To wear crowns shall bear the heavy or give up". Put
me on your SPAM list, thank you.

https://lore.kernel.org/linux-mm/20221018223042.GJ2703033@dread.disaster.area/
https://lore.kernel.org/linux-mm/Y0%2FkZbIvMgkNhWpM@bfoster/

>
> > the lockless xarray operation could get harmed by an illegal folio
> > remaining on the slot[offset]. This commit would like to protect
> > the xa split stuff(folio_ref_freeze and __split_huge_page) under
> > lruvec->lock to remove the race window.