On Wed, Apr 26, 2017 at 12:31 PM, Matthias Kaehlcke <[email protected]> wrote:
> Hi Kees,
>
> El Wed, Apr 26, 2017 at 12:09:13PM -0700 Kees Cook ha dit:
>
>> On Wed, Apr 26, 2017 at 12:01 PM, Matthias Kaehlcke <[email protected]> 
>> wrote:
>> > In difference to gas clang doesn't seem to infer the size from the
>> > operands. Adding the suffix fixes the following error when building
>> > with clang:
>> >
>> > CC      arch/x86/lib/kaslr.o
>> > /tmp/kaslr-dfe1ad.s: Assembler messages:
>> > /tmp/kaslr-dfe1ad.s:182: Error: no instruction mnemonic suffix given and
>> > no register operands; can't size instruction
>> >
>> > Signed-off-by: Matthias Kaehlcke <[email protected]>
>> > ---
>> >  arch/x86/lib/kaslr.c | 6 +++++-
>> >  1 file changed, 5 insertions(+), 1 deletion(-)
>> >
>> > diff --git a/arch/x86/lib/kaslr.c b/arch/x86/lib/kaslr.c
>> > index 121f59c6ee54..947d4aa92ff7 100644
>> > --- a/arch/x86/lib/kaslr.c
>> > +++ b/arch/x86/lib/kaslr.c
>> > @@ -79,7 +79,11 @@ unsigned long kaslr_get_random_long(const char *purpose)
>> >         }
>> >
>> >         /* Circular multiply for better bit diffusion */
>> > -       asm("mul %3"
>> > +#ifdef CONFIG_X86_64
>> > +       asm("mulq %3"
>> > +#else
>> > +       asm("mull %3"
>> > +#endif
>> >             : "=a" (random), "=d" (raw)
>> >             : "a" (random), "rm" (mix_const));
>> >         random += raw;
>> >
>>
>> Since there may be more of these and we try to avoid #ifdef in .c
>> files, I think we should create an isns helper macro and use it here
>> instead. Like:
>>
>> #ifdef CONFIG_X86_64
>> # define ASM_MUL "mulq"
>> #else
>> # define ASM_MUL "mull"
>> #endif
>>
>> ...
>>
>>
>> asm(ASM_MUL " %3" ...
>
> Sounds very reasonable. Thanks for the suggestion, I'll rework the patch.

Oh, hah, this almost already exists. I went looking for where it might
best live and I see this is already built up in
arch/x86/include/asm/asm.h

I think you just need to add:

#define _ASM_MUL __ASM_SIZE(mul)

to that header, and use _ASM_MUL in kaslr.c.

-Kees

-- 
Kees Cook
Pixel Security

Reply via email to