linux-kernel - Re: [PATCH] bitmap: fix memset optimization on big-endian systems

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: <20180403180854.GA17679@vader>
Date:   Tue, 3 Apr 2018 11:08:54 -0700
From:   Omar Sandoval <osandov@...ndov.com>
To:     Linus Torvalds <torvalds@...ux-foundation.org>
Cc:     linux-btrfs <linux-btrfs@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        kernel-team <kernel-team@...com>,
        Matthew Wilcox <mawilcox@...rosoft.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Rasmus Villemoes <linux@...musvillemoes.dk>
Subject: Re: [PATCH] bitmap: fix memset optimization on big-endian systems

On Mon, Apr 02, 2018 at 04:49:39PM -0700, Linus Torvalds wrote:
> On Mon, Apr 2, 2018 at 4:37 PM, Linus Torvalds
> <torvalds@...ux-foundation.org> wrote:
> >
> > We should probably have at least switched it to "unsigned long int"
> 
> I meant just "unsigned int", of course.
> 
> Right now we occasionally have a silly 64-bit field just for a couple of flags.

Not to mention the mix of bit fields, macros, and enums, some of which
are bit masks, some of which are the bit number :)

> Of course, we do have cases where 64-bit architectures happily end up
> using more than 32 bits too, so the "unsigned long" is occasionally
> useful. But it's rare enough that it probably wasn't the right thing
> to do.
> 
> Water under the bridge. Part of it is due to another historical
> accident: the alpha port was one of the early ports, and it didn't do
> atomic byte accesses at all.
> 
> So we had multiple issues that all conspired to "unsigned long arrays
> are the natural for atomic bit operations" even though they have this
> really annoying byte order issue.

Thanks for the historical context, this is exactly what I was wondering
when I spotted this bug and fixed a similar one in Btrfs a couple of
years back.