On 02/19/2012 04:51 PM, Linus Torvalds wrote:
> On Sun, Feb 19, 2012 at 4:07 PM, H. Peter Anvin <hpa@xxxxxxxxx> wrote:
>>
>> Use explicit sizes (__u64) instead of implicit sizes (unsigned long)
>> in the definition for sigcontext.h; this will allow this structure to
>> be shared between the x86-64 native ABI and the x32 ABI.
>
> Btw, since we had this issue just with autofs: what are the x32 ABI
> alignment issues for __u64? Are they like x86-64 ("natural alignment")
> or x86-32 ("4-byte alignment")?
>
> I assume they are natural alignment, and as pointed out by Davem, we
> do have the versions of u64 that make this explicit: "compat_u64" is
> the 4-byte-aligned one, while "__aligned_u64" is the natively aligned
> one.
>
> Just plain "__u64" doesn't tell which it is, which is sad and wrong,
> but we're likely stuck with it forever. Unless some shining knight
> comes and says "__u64 is native alignment, and if you want anything
> else, you need to use __compat_u64", and actually fixes the cases
> where x86-32 depends on the 4-byte aligned one.
>
> Which would be nice, but sounds unlikely. Shining knights tend to be
> rare. But this *could* possibly be automated, so it's not entirely out
> of the question.
>
We are using __u64 as x86-32 compatible since we are sharing most of the
really complex path (like ioctl) with i386 much more so than x86-64. So
it is defined in userspace as:
typedef unsigned long long __u64 __attribute__((aligned(4)));
__aligned_u64 obviously is naturally aligned, which matches uint64_t is
userspace.
-hpa
--
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel. I don't speak on their behalf.
--
To unsubscribe from this list: send the line "unsubscribe linux-arch" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
