MIME-Version: 1.0
References: <20230125015738.912924-1-zokeefe@google.com> <ecb2cf3-45f0-8aae-3e1-da4e79de9c27@google.com>
 <20230209132846.122ad88e1c2bd0603a630e5c@linux-foundation.org>
 <1f344fc1-834d-f1df-8a2c-79918be5b22@google.com> <20230209141221.b7881fb0783a58571f2bca53@linux-foundation.org>
In-Reply-To: <20230209141221.b7881fb0783a58571f2bca53@linux-foundation.org>
From:   "Zach O'Keefe" <zokeefe@google.com>
Date:   Thu, 9 Feb 2023 14:29:45 -0800
Message-ID: <CAAa6QmRQqvOvEcb3BRsdr8Jb_XE+pFTVzzroXSUskRU=NOpRsw@mail.gmail.com>
Subject: Re: [PATCH 1/2] mm/MADV_COLLAPSE: set EAGAIN on unexpected page refcount
To:     Andrew Morton <akpm@linux-foundation.org>
Cc:     Hugh Dickins <hughd@google.com>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org, Yang Shi <shy828301@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Feb 9, 2023 at 2:12 PM Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Thu, 9 Feb 2023 13:50:30 -0800 (PST) Hugh Dickins <hughd@google.com> wrote:
>
> >
> > >
> > > I'm not seeing anything in the [1/2] changelog which indicates that a
> > > backport is needed.  IOW,
> >
> > Correct: it's just changing the errno for some racy cases from "you're
> > wrong, don't bother me again" to "it might be worth having another go":
> > not fixing an instability, as 2/2 was.
> >
> > >
> > > # cat .signature
> > > When fixing a bug, please describe the end-user visible effects of that bug.
> >
> > If whatever's being run by the end-user is coded to try again on -EAGAIN,
> > then the end-user will less often see occasional unexplained failures.
> >
>
> OK, thanks.  I redid the changelog's final paragraph thusly:
>
> : In this situation, MADV_COLLAPSE returns -EINVAL when it should return
> : -EAGAIN.  This could cause userspace to conclude that the syscall failed,
> : when it in fact could succeed by retrying.
>

This looks good to me. Thanks Andrew! Also thanks Hugh for being on
the lookout for this patch -- I hastily read through my emails
regarding which patches were merged where and had assumed this merged
with 2/2.

Also, apologies about the confusing v1 [1/2] and v2 [2/2] fiasco; in
hindsight that probably wasn't the most decipherable thing to do :)

Best,
Zach