netdev - Re: [PATCH v2 6/8] page_frag_cache: Use a mask instead of offset

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] [day] [month] [year] [list]

Message-ID: <20180322173450.GI28468@bombadil.infradead.org>
Date:   Thu, 22 Mar 2018 10:34:50 -0700
From:   Matthew Wilcox <willy@...radead.org>
To:     Alexander Duyck <alexander.duyck@...il.com>
Cc:     Matthew Wilcox <mawilcox@...rosoft.com>,
        Netdev <netdev@...r.kernel.org>, linux-mm <linux-mm@...ck.org>,
        Jesper Dangaard Brouer <brouer@...hat.com>,
        Eric Dumazet <eric.dumazet@...il.com>
Subject: Re: [PATCH v2 6/8] page_frag_cache: Use a mask instead of offset

On Thu, Mar 22, 2018 at 09:41:57AM -0700, Matthew Wilcox wrote:
> On Thu, Mar 22, 2018 at 09:22:31AM -0700, Alexander Duyck wrote:
> > You could just use the pfc->mask here instead of size - 1 just to
> > avoid having to do the subtraction more than once assuming the
> > compiler doesn't optimize it.
> 
> Either way I'm assuming a compiler optimisation -- that it won't reload
> from memory, or that it'll remember the subtraction.  I don't much care
> which, and I'll happily use the page_frag_cache_mask() if that reads better
> for you.

Looks like it does reload from memory if I make that change.  Before:

    37e7:       c7 43 08 ff 7f 00 00    movl   $0x7fff,0x8(%rbx)
    37ee:       b9 00 80 00 00          mov    $0x8000,%ecx
    37f3:       be ff 7f 00 00          mov    $0x7fff,%esi
    37f8:       ba 00 80 00 00          mov    $0x8000,%edx
...
    380b:       01 70 1c                add    %esi,0x1c(%rax)

After:

    37e7:       c7 43 08 ff 7f 00 00    movl   $0x7fff,0x8(%rbx)
    37ee:       b9 00 80 00 00          mov    $0x8000,%ecx
    37f3:       ba 00 80 00 00          mov    $0x8000,%edx
...
    3806:       8b 73 08                mov    0x8(%rbx),%esi
    3809:       01 70 1c                add    %esi,0x1c(%rax)

Of course, it's shorter because it's fewer bytes to reload from memory
than it is to put a 32-bit immediate in the instruction stream, but
it's one additional memory reference (cache-hot, of course).  I don't
really care because it's the cold path.