Re: [ros-dev] [ros-diffs] [jgardou] 50991: [NTOSKRNL] - Print supported CPU features. - Simplify detection of non working but claimed SYSENTER support for intel processors - Workaround a virtualbox "feature" : claim that SY...

[Aleksey Bragin]

On Mar 7, 2011, at 4:33 PM, jgardou(a)svn.reactos.org wrote:

>   Author: jgardou
>               /* Check if the CPU is too old to support SYSENTER */
>  -            if ((Prcb->CpuType < 6) ||
>  -                ((Prcb->CpuType == 6) && (Prcb->CpuStep < 0x0303)))
>  +            if ((Reg[0] & 0x0FFF3FFF) < 0x00000633)
>               { 
It was intentionally done over CpuType and CpuStep. Now you  
criptified it back to a raw Reg[0] value and weird hardcoded  
constant. Could you please 1) elaborate this change and 2) if you  
still think it's correct, and if you still tihnk(!) Windows does it  
same way, then rework it to use Prcb->Cpu* instead of these magic  
values?

Not to say adding a hack for VirtualBox is far from being a  
beautiful, and would need to be *at least* marked so that it's not  
forgotten.

Thanks,
Aleksey Bragin.

[Jérôme Gardou]

Official intel documentation :
http://www.intel.com/Assets/PDF/appnote/241618.pdf, page 30.

As to know how it's made on windows, well, I hope they follow intel docs ;-)

Le 07/03/2011 21:49, Aleksey Bragin a écrit :
>  
>
>  On Mar 7, 2011, at 4:33 PM, jgardou(a)svn.reactos.org wrote:
>
> >   Author: jgardou
> >               /* Check if the CPU is too old to support SYSENTER */
> >  -            if ((Prcb->CpuType < 6) ||
> >  -                ((Prcb->CpuType == 6) && (Prcb->CpuStep < 0x0303)))
> >  +            if ((Reg[0] & 0x0FFF3FFF) < 0x00000633)
> >               { 
>  It was intentionally done over CpuType and CpuStep. Now you
> criptified 
>  it back to a raw Reg[0] value and weird hardcoded constant. Could you 
>  please 1) elaborate this change and 2) if you still think it's 
>  correct, and if you still tihnk(!) Windows does it same way, then 
>  rework it to use Prcb->Cpu* instead of these magic values?
>
>  Not to say adding a hack for VirtualBox is far from being a beautiful, 
>  and would need to be *at least* marked so that it's not forgotten.
>
>  Thanks,
>  Aleksey Bragin.
>
>  _______________________________________________
>  Ros-dev mailing list
>  Ros-dev(a)reactos.org
>  http://www.reactos.org/mailman/listinfo/ros-dev
>