[tlinux-users:08146] Re: New release of omnibook module soon: testing welcome

Holger Nelson a écrit :
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On Tue, 12 Sep 2006, Mathieu Bérard wrote:
>
>   
>> Holger Nelson a écrit :
>>     
>> One word: Thanks
>> One more fix like that an you have SVN commit access.
>>
>> Could you please try this and check it doesn't introduce
>> any regression:
>>     
>
> It works, but I doubt it does what you really want:
> The disassembly of your code shows:
> intel_do_smi_call():
> /home/crius/download/xx/xxxx/trunk/nbsmi.c:153
>  3c2:   89 f0                   mov    %esi,%eax
>  3c4:   9c                      pushf
>  3c5:   60                      pusha
>  3c6:   66 e7 b2                out    %ax,$0xb2
>  3c9:   66 83 f8 00             cmp    $0x0,%ax
>  3cd:   74 08                   je     3d7 <nbsmi_smi_command+0xad>
>  3cf:   c7 44 24 0c fb ff ff    movl   $0xfffffffb,0xc(%esp)
>  3d6:   ff
>  3d7:   61                      popa
>  3d8:   9d                      popf
>
> Looking at the stack at 3cf will show something like (I dont know the 
> stack-layout of pusha/popa nor exactly which registers get saved so I just 
> guess as I am to lazy to look it up. ;-) )
>   
My bad I have wrongly implicitly assumed that gcc was aware of
the push/pop operations in inline asm.
>
> While I am at it another simple fix against R166:
> As the structs are put on stack we cannot know that they are initialized 
> with 0 so we need to initialize the used fields ourself.
>   
D'oh! the code is full of this pattern, I will audit all of them.


-- 
Mathieu