pfn_pte/pfn_pmd

pfn_pte/pfn_pmd

[Jan Beulich]

Isn't it a bug to have pfn_pte be defined identically as pfn_pte_ma in
i386's pgtable-2level.h? Shouldn't this rather be done the same as in
pgtable-3level.h?

Similarly for pfn_pmd (which BUG()-s in pgtable-3level.h)...

Thanks, Jan

Re: pfn_pte/pfn_pmd

[Jan Beulich]

Apparently it's not, as the translation is done in __pte (and not in
__pte_ma); I was just getting confused by the two different mechanisms
use for non-PAE and PAE, which, while merging, I'd like to get to use as
much unconditional code as possible.

I still wonder, however, whether the BUG()/panic() in PAE's pfn_pmd()
couldn't be removed...

Jan

>>> "Jan Beulich" <JBeulich@novell.com> 21.12.05 11:44:31 >>>
Isn't it a bug to have pfn_pte be defined identically as pfn_pte_ma in
i386's pgtable-2level.h? Shouldn't this rather be done the same as in
pgtable-3level.h?

Similarly for pfn_pmd (which BUG()-s in pgtable-3level.h)...

Thanks, Jan

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com 
http://lists.xensource.com/xen-devel

Re: pfn_pte/pfn_pmd

[Gerd Knorr]

> I still wonder, however, whether the BUG()/panic() in PAE's pfn_pmd()
> couldn't be removed...

Wonder why it is still in there though, seems pfn_pmd() is never ever 
used, checking code ...

Ok, seems they are needed for 2M/4M pages, we don't support these 
anyway, so BUG()'ing in case this is called looks perfectly fine.

The panic msg probably should be changed to something more reasonable, 
such as "Huh? Who tried to use 2M pages?" maybe ;)

cheers,

   Gerd

Re: pfn_pte/pfn_pmd

[Jan Beulich]

>>> Gerd Knorr <kraxel@suse.de> 21.12.05 15:25:56 >>>
>> I still wonder, however, whether the BUG()/panic() in PAE's
pfn_pmd()
>> couldn't be removed...
>
>Wonder why it is still in there though, seems pfn_pmd() is never ever

>used, checking code ...
>
>Ok, seems they are needed for 2M/4M pages, we don't support these 
>anyway, so BUG()'ing in case this is called looks perfectly fine.

I was about to think so, too, but I wasn't able to prove that the code
in arch/i386/mm/init-xen.c that uses it (and that is conditional upon
cpu_has_pse) can't get executed. Specifically, I missed
include/asm-i386/mach-xen/setup_arch_post.h clearing X86_FEATURE_PSE...


>The panic msg probably should be changed to something more reasonable,

>such as "Huh? Who tried to use 2M pages?" maybe ;)

Jan