linux-kernel - Re: [PATCH v4 08/10] mm/gup: limit number of gup migration failures, honor failures

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CA+CK2bB9REcktfq=kh9MhvAwg8GPpK_tHx_LiruF0DCkMWHXmA@mail.gmail.com>
Date:   Wed, 13 Jan 2021 14:49:41 -0500
From:   Pavel Tatashin <pasha.tatashin@...een.com>
To:     Michal Hocko <mhocko@...e.com>
Cc:     LKML <linux-kernel@...r.kernel.org>, linux-mm <linux-mm@...ck.org>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Vlastimil Babka <vbabka@...e.cz>,
        David Hildenbrand <david@...hat.com>,
        Oscar Salvador <osalvador@...e.de>,
        Dan Williams <dan.j.williams@...el.com>,
        Sasha Levin <sashal@...nel.org>,
        Tyler Hicks <tyhicks@...ux.microsoft.com>,
        Joonsoo Kim <iamjoonsoo.kim@....com>, mike.kravetz@...cle.com,
        Steven Rostedt <rostedt@...dmis.org>,
        Ingo Molnar <mingo@...hat.com>, Jason Gunthorpe <jgg@...pe.ca>,
        Peter Zijlstra <peterz@...radead.org>,
        Mel Gorman <mgorman@...e.de>,
        Matthew Wilcox <willy@...radead.org>,
        David Rientjes <rientjes@...gle.com>,
        John Hubbard <jhubbard@...dia.com>,
        Linux Doc Mailing List <linux-doc@...r.kernel.org>,
        Ira Weiny <ira.weiny@...el.com>,
        linux-kselftest@...r.kernel.org
Subject: Re: [PATCH v4 08/10] mm/gup: limit number of gup migration failures,
 honor failures

On Fri, Dec 18, 2020 at 8:14 AM Michal Hocko <mhocko@...e.com> wrote:
>
> On Fri 18-12-20 07:43:15, Pavel Tatashin wrote:
> > On Fri, Dec 18, 2020 at 5:46 AM Michal Hocko <mhocko@...e.com> wrote:
> > >
> > > On Thu 17-12-20 13:52:41, Pavel Tatashin wrote:
> > > [...]
> > > > +#define PINNABLE_MIGRATE_MAX 10
> > > > +#define PINNABLE_ISOLATE_MAX 100
> > >
> > > Why would we need to limit the isolation retries. Those should always be
> > > temporary failure unless I am missing something.
> >
> > Actually, during development, I was retrying isolate errors
> > infinitely, but during testing found a hung where when FOLL_TOUCH
> > without FOLL_WRITE is passed (fault in kernel without write flag), the
> > zero page is faulted. The isolation of the zero page was failing every
> > time, therefore the process was hanging.
>
> Why would you migrate zero page in the first place? Simply instantiate
> it.

This is exactly the idea behind FOLL_WRITE; it causes zero pages to be
created in the right zone right away, and no migration is necessary.

>
> > Since then, I fixed this problem by adding FOLL_WRITE unconditionally
> > to FOLL_LONGTERM, but I was worried about other possible bugs that
> > would cause hangs, so decided to limit isolation errors. If you think
> > it its not necessary, I can unlimit isolate retires.
>
> It should have a really good reason to exist. Worries about some corner
> cases is definitely not a reason to put some awkward retry mechanism.
> My historical experience is that these things are extremely hard to get
> rid of later.
>
> > > I am not sure about the
> > > PINNABLE_MIGRATE_MAX either. Why do we want to limit that? migrate_pages
> > > already implements its retry logic why do you want to count retries on
> > > top of that? I do agree that the existing logic is suboptimal because
> >
> > True, but again, just recently, I worked on a race bug where pages can
> > end up in per-cpu list after lru_add_drain_all() but before isolation,
> > so I think retry is necessary.
>
> There are ways to make sure pages are not ending on pcp list. Have a
> look at how hotplug does that.

Sounds good to me, I will remove PINNABLE_MIGRATE_MAX, and leave only
PINNABLE_ISOLATE_MAX for transient isolation errors.

>
> > > the migration failure might be ephemeral or permanent but that should be
> > > IMHO addressed at migrate_pages (resp. unmap_and_move) and simply report
> > > failures that are permanent - e.g. any potential pre-existing long term
> > > pin - if that is possible at all. If not what would cause permanent
> > > migration failure? OOM?
> >
> > Yes, OOM is the main cause for migration failures.
>
> Then you can treat ENOMEM as a permanent failure.

Sounds good.