Re: 32 bit user space compatibility

From: Yann Droneaud <ydroneaud@opteya.com>
To: One Thousand Gnomes <gnomes@lxorguk.ukuu.org.uk>
Cc: "Skidanov, Alexey" <Alexey.Skidanov@amd.com>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	ydroneaud@opteya.com
Subject: Re: 32 bit user space compatibility
Date: Mon, 03 Nov 2014 13:46:09 +0100	[thread overview]
Message-ID: <1415018769.5491.92.camel@localhost.localdomain> (raw)
In-Reply-To: <20141028223435.73b3778c@alan.etchedpixels.co.uk>

Hi,

Le mardi 28 octobre 2014 à 22:34 +0000, One Thousand Gnomes a écrit :
> On Sun, 26 Oct 2014 12:25:08 +0000
> "Skidanov, Alexey" <Alexey.Skidanov@amd.com> wrote:

>  
> > Running 32 bit user space needs some work to be done with ioctls. I understand that there are two options to implement:
> > 1.       Use only fixed size types. Pad IOCTLS params to multiple of 64 bits - simple; don't know if it covers all compatibility issues; 
> > 2.       32 bit compatibility layer (through compat_ioctl, just like many drivers in kernel implement)  - just a little bit simple code with some translations; really covers all issues;
> >  
> > Which one is preferred by kernel community?
> 
> You shouldn't need to pad paramters in most cases as platform alignment
> rules are usually sane for 32 and 64bit. 

In most case, except i386 (ia32) vs amd64 (x86_64): u64 are going to be
aligned on 4 bytes boundaries for 32bits ABI and 8 bytes boundaries for
64bits ABI.

I've tried to explained this issue in a lightning talk[1][2] I'd given
at Kernel Recipes[3] this year.

[1] http://opteya.com/talks/2014/kernel-recipes/lightning-talk-kernel-userspace-ABI/
[2] https://gitorious.org/opteya/talk-kernel-userspace-abi
[3] http://kernel.recipes/

Regards.

-- 
Yann Droneaud
OPTEYA