Re: [PATCH] x86_32: add support for 64 bit __get_user()

[Benjamin LaHaise]

On Wed, Mar 09, 2016 at 10:22:50AM -0800, H. Peter Anvin wrote:
> On 03/09/2016 09:50 AM, Benjamin LaHaise wrote:
> > On Wed, Mar 09, 2016 at 09:36:30AM -0800, H. Peter Anvin wrote:
> >> On March 9, 2016 9:22:25 AM PST, Benjamin LaHaise <bcrl@xxxxxxxxx> wrote:
> >>> The existing __get_user() implementation does not support fetching
> >>> 64 bit values on 32 bit x86.  Implement this in a way that does not 
> >>> generate any incorrect warnings as cautioned by Russell King.  Test 
> >>> code available at http://www.kvack.org/~bcrl/x86_32-get_user.tar .  
> > ...
> >> Weird.  I could swear we had already fixed this a few years ago.
> > 
> > That surprised me as well, but Russell raised the fact that the approaches 
> > previously tried on 32 bit architectures had caused various incorrect 
> > compiler warnings for certain obscure cases -- see the code in test_module.c 
> > in that URL that Russell provided to demonstrate the problem across all the 
> > corner cases.
> 
> Oh, I see... I implemented it for put but not get... weird.  You may
> want to look at the __inttype() macro defined earlier in this file; it
> might be useful.

Ah, interesting.  I'll look at that.

> I presume you have already seen:
> 
> >> fs/select.c:710: Error: operand type mismatch for `movq'
> >> fs/select.c:714: Error: incorrect register `%cx' used with `q' suffix
>    fs/select.c:711: Error: operand type mismatch for `movq'
> >> fs/select.c:715: Error: incorrect register `%si' used with `q' suffix
> --
>    fs/aio.c: Assembler messages:
> >> fs/aio.c:1606: Error: operand type mismatch for `movq'
> >> fs/aio.c:1610: Error: incorrect register `%si' used with `q' suffix
> 
> ... which implies it used 16-bit registers for 64-bit operations when
> compiling for 64 bits.

Yup, will respin shortly.

		-ben
-- 
"Thought is the essence of where you are now."