Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Subject: Re: [mm 4.15-rc8] Random oopses under memory pressure.
To:     Linus Torvalds <torvalds@linux-foundation.org>
CC:     "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
        Matthew Wilcox <willy@infradead.org>,
        Al Viro <viro@ZenIV.linux.org.uk>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
References: <201801170233.JDG21842.OFOJMQSHtOFFLV@I-love.SAKURA.ne.jp>
 <CA+55aFyxyjN0Mqnz66B4a0R+uR8DdfxdMhcg5rJVi8LwnpSRfA@mail.gmail.com>
 <201801172008.CHH39543.FFtMHOOVSQJLFO@I-love.SAKURA.ne.jp>
 <201801181712.BFD13039.LtHOSVMFJQFOFO@I-love.SAKURA.ne.jp>
 <20180118122550.2lhsjx7hg5drcjo4@node.shutemov.name>
 <d8347087-18a6-1709-8aa8-3c6f2d16aa94@linux.intel.com>
 <20180118145830.GA6406@redhat.com>
 <20180118165629.kpdkezarsf4qymnw@node.shutemov.name>
 <CA+55aFy43ypm0QvA5SqNR4O0ZJETbkR3NDR=dnSdvejc_nmSJQ@mail.gmail.com>
 <20180118234955.nlo55rw2qsfnavfm@node.shutemov.name>
 <20180119125503.GA2897@bombadil.infradead.org>
 <CA+55aFwWCeFrhN+WJDD8u9nqBzmvknXk428Q0dVwwXAvwhg_-w@mail.gmail.com>
Message-ID: <6f0abb8d-84b3-070e-fca7-c71524311702@prevas.dk>
Date:   Mon, 22 Jan 2018 14:26:02 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.5.0
MIME-Version: 1.0
In-Reply-To: <CA+55aFwWCeFrhN+WJDD8u9nqBzmvknXk428Q0dVwwXAvwhg_-w@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 2018-01-19 19:42, Linus Torvalds wrote:
> 
> I actually asked (long long ago) for an optinal compiler warning for
> "pointer subtraction with non-power-of-2 sizes". Not because of it
> being undefined, but simply because it's expensive. The
> divide->multiply thing doesn't always work,

Huh? If (compile-time constant, positive) d=m*2^k with m odd, and x is
known to be a multiple of d, x/d can always be computed as (x>>k)*m' ==
(x*m')>>k, with m' being the mod 2^N multiplicative inverse of m, right?
This works whether x has signed or unsigned type and whether it indeed
happens to be negative, AFAICT.

Sure, the multiplication by m' may not exactly be cheap, but one only
needs the low N bits of the result.

Rasmus