linux-kernel - Re: [v2 PATCH] mm: gup: do not call try_grab

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: <20240627163242.39b0a716bd950a895c032136@linux-foundation.org>
Date: Thu, 27 Jun 2024 16:32:42 -0700
From: Andrew Morton <akpm@...ux-foundation.org>
To: Peter Xu <peterx@...hat.com>
Cc: Yang Shi <yang@...amperecomputing.com>, yangge1116@....com,
 david@...hat.com, linux-mm@...ck.org, linux-kernel@...r.kernel.org,
 stable@...r.kernel.org
Subject: Re: [v2 PATCH] mm: gup: do not call try_grab_folio() in slow path

On Thu, 27 Jun 2024 19:19:40 -0400 Peter Xu <peterx@...hat.com> wrote:

> Yang,
> 
> On Thu, Jun 27, 2024 at 03:14:13PM -0700, Yang Shi wrote:
> > The try_grab_folio() is supposed to be used in fast path and it elevates
> > folio refcount by using add ref unless zero.  We are guaranteed to have
> > at least one stable reference in slow path, so the simple atomic add
> > could be used.  The performance difference should be trivial, but the
> > misuse may be confusing and misleading.
> 
> This first paragraph is IMHO misleading itself..
> 
> I think we should mention upfront the important bit, on the user impact.
> 
> Here IMO the user impact should be: Linux may fail longterm pin in some
> releavnt paths when applied over CMA reserved blocks.  And if to extend a
> bit, that include not only slow-gup but also the new memfd pinning, because
> both of them used try_grab_folio() which used to be only for fast-gup.

It's still unclear how users will be affected.  What do the *users*
see?  If it's a slight slowdown, do we need to backport this at all?

> 
> The patch itself looks mostly ok to me.
> 
> There's still some "cleanup" part mangled together, e.g., the real meat
> should be avoiding the folio_is_longterm_pinnable() check in relevant
> paths.  The rest (e.g. switch slow-gup / memfd pin to use folio_ref_add()
> not try_get_folio(), and renames) could be good cleanups.
> 
> So a smaller fix might be doable, but again I don't have a strong opinion
> here.

The smaller the better for backporting, of course.