Kexec failure on RDC (and possibly other early x86) platforms

Kexec failure on RDC (and possibly other early x86) platforms

[bifferos]

I've updated the patch to allow Kexec to work on RDC platforms
here:
http://bifferboard.svn.sourceforge.net/viewvc/bifferboard/slack/kernel/2.6.32/0003-kexec-fix.patch

I'm curious as to how one can detect the presence of the CR4 
register in assembler, as I'm sure it's possible, but I've just 
#ifndefed out the offending instruction in this patch and it
seems to work with this change.

regards,
Biff.


      

Re: Kexec failure on RDC (and possibly other early x86) platforms

[Avi Kivity]

On 12/06/2009 01:33 PM, bifferos wrote:
> I've updated the patch to allow Kexec to work on RDC platforms
> here:
> http://bifferboard.svn.sourceforge.net/viewvc/bifferboard/slack/kernel/2.6.32/0003-kexec-fix.patch
>
> I'm curious as to how one can detect the presence of the CR4
> register in assembler, as I'm sure it's possible, but I've just
> #ifndefed out the offending instruction in this patch and it
> seems to work with this change.
>    

One way is to execute the instruction and trap the #UD exception if it 
is not supported.  Not sure whether you have an IDT set up or whether 
your cpu traps on mov cr4.

If your cpu supports cpuid you can test for features that indicate bits 
in cr4 are available, for example VME, DE, PSE, PVI, and PAE.  If none 
are available you likely don't have cr4 (and even if you do, it's 
pointless to reset it).

-- 
error compiling committee.c: too many arguments to function

Re: Kexec failure on RDC (and possibly other early x86) platforms

[H. Peter Anvin]

On 12/06/2009 08:52 AM, Avi Kivity wrote:
> On 12/06/2009 01:33 PM, bifferos wrote:
>> I've updated the patch to allow Kexec to work on RDC platforms
>> here:
>> http://bifferboard.svn.sourceforge.net/viewvc/bifferboard/slack/kernel/2.6.32/0003-kexec-fix.patch
>>
>>
>> I'm curious as to how one can detect the presence of the CR4
>> register in assembler, as I'm sure it's possible, but I've just
>> #ifndefed out the offending instruction in this patch and it
>> seems to work with this change.
>>    
> 
> One way is to execute the instruction and trap the #UD exception if it
> is not supported.  Not sure whether you have an IDT set up or whether
> your cpu traps on mov cr4.
> 
> If your cpu supports cpuid you can test for features that indicate bits
> in cr4 are available, for example VME, DE, PSE, PVI, and PAE.  If none
> are available you likely don't have cr4 (and even if you do, it's
> pointless to reset it).
> 

In general, I believe, existence of CPUID == existence of CR4.

(There were some late 486's which had CPUID, I believe they also had
CR4.)  The first-principles test of catching the trap is more direct,
though.  Inside the kernel we have read_cr4_safe() for that.

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.