Re: [GIT PULL] x86/paravirt for v2.6.33

[Brian Gerst]

On Wed, Dec 9, 2009 at 1:54 PM, H. Peter Anvin <hpa@xxxxxxxxx> wrote:
> On 12/09/2009 10:47 AM, Linus Torvalds wrote:
>>
>> NOTE NOTE NOTE! Totally untested. Is the second argument even in %edx? I
>> don't remember, I didn't check, I'm just throwing this out as a "hey,
>> maybe something _like_ this can work" patch, and will be immediately
>> removing it from my machine after sending this email.
>>
>
> The second argument is in %edx, but unlike 64 bits, it is not loaded
> into that register a priory ("asmlinkage" means arguments are on the stack.)
>
> As such, we need something looking like:
>
> #define PTREGSCALL0(name) Â Â Â \
> ptregs_##name: Â Â Â Â Â Â Â Â Â\
>    Âleal  Â4(%esp),%eax;  \
> Â Â Â Âjmp sys_##name
>
> #define PTREGSCALL1(name) Â Â Â \
> ptregs_##name: Â Â Â Â Â Â Â Â Â\
>    Âmovl  Â4(%esp),%eax;  \
>    Âleal  Â4(%esp),%edx;  \
> Â Â Â Âjmp sys_##name
>
> #define PTREGSCALL2(name) Â Â Â \
> ptregs_##name: Â Â Â Â Â Â Â Â Â\
>    Âmovl  Â4(%esp),%eax;  \
>    Âmovl  Â8(%esp),%edx;  \
>    Âleal  Â4(%esp),%ecx;  \
> Â Â Â Âjmp sys_##name
>
> If we need more than two arguments + pt_regs, then we have to set up a
> temporary stack frame.
>
> [Sorry, I'm sitting in a meeting so I can't actually write up a real patch]
>
> Â Â Â Â-hpa

I had started working on a patchset like this when this issue came up
the first time.  I'll get it up to date and send it out.  It would be
nice if this could be done for all syscalls, so that
asmlinkage_protect() isn't needed.

--
Brian Gerst
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/