Re: Re: pv_ops & gntdev?

[Gerd Hoffmann <kraxel@redhat.com>]

Gerd Hoffmann wrote:
> Jeremy Fitzhardinge wrote:
>> Do you have a test program?

> usage:
> qemu -M xenpv -xen-create -uuid $(uuidgen) \
>   -kernel <bzImage> -initrd <initrd> \
>   -drive media=disk,if=xen,file=<diskimage> \
>   -net nic,model=xen,macaddr=<addr> \
>   -serial <yourconsolehere>

"-m 256" is needed too.

>>> +    BUG_ON(pgnr >= map->count);
>>> +    mpte  = (u64)pfn_to_mfn(page_to_pfn(token)) << PAGE_SHIFT;
>>> +    mpte |= (unsigned long)pte & ~PAGE_MASK;
>>>   
>> (!) You're casting pte_t * to unsigned long, masking off the lower bits
>> then oring them into your pte.  That doesn't look like it will mean very
>> much...  I guess this explains your not-unmapping bug.
>>
>> This should probably be using mfn_pte() anyway:
>>
>>     mfn = pfn_to_mfn(page_to_pfn(token));
>>     pgprot = __pgprot(pte->pte & PTE_FLAGS_MASK);
>>     mpte = mfn_pte(mfn, pgprot);
> 
> Looks better indeed.  Unclean stuff carried over from 2.6.18 ...

Well, no.  mpte (bad name indeed) isn't a pte, but the *pointer*
(machine address) to the pte.

I've fixed that and did a few more cleanups, patch comes later today.

cheers,
  Gerd