Re: [PATCH v8 07/15] KVM: pfncache: include page offset in uhva and use it consistently

[Paul Durrant <xadimgnik@gmail.com>]

On 21/11/2023 22:35, David Woodhouse wrote:
> On Tue, 2023-11-21 at 18:02 +0000, Paul Durrant wrote:
>> @@ -242,8 +242,7 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa,
>>          }
>>   
>>          old_pfn = gpc->pfn;
>> -       old_khva = gpc->khva - offset_in_page(gpc->khva);
>> -       old_uhva = gpc->uhva;
>> +       old_khva = (void *)PAGE_ALIGN_DOWN((uintptr_t)gpc->khva);
>>   
>>          /* If the userspace HVA is invalid, refresh that first */
>>          if (gpc->gpa != gpa || gpc->generation != slots->generation ||
>> @@ -259,13 +258,25 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa,
>>                          ret = -EFAULT;
>>                          goto out;
>>                  }
> 
> 
> There's a subtle behaviour change here, isn't there? I'd *really* like
> you do say 'No functional change intended' where that is true, and then
> the absence of that sentence in this one would be meaningful.
> 
> You are now calling hva_to_pfn_retry() even when the uhva page hasn't
> changed. Which is harmless and probably not important, but IIUC fixable
> by the addition of:
> 
>   +              if (gpc->uhva != PAGE_ALIGN_DOWN(old_uhva))

True; I can keep that optimization and then I will indeed add 'no 
functional change'... Didn't seem worth it at the time, but no harm.

>> +               hva_change = true;
>> +       } else {
>> +               /*
>> +                * No need to do any re-mapping if the only thing that has
>> +                * changed is the page offset. Just page align it to allow the
>> +                * new offset to be added in.
>> +                */
>> +               gpc->uhva = PAGE_ALIGN_DOWN(gpc->uhva);
>>          }
>>   
>> +       /* Note: the offset must be correct before calling hva_to_pfn_retry() */
>> +       gpc->uhva += page_offset;
>> +
>>          /*
>>           * If the userspace HVA changed or the PFN was already invalid,
>>           * drop the lock and do the HVA to PFN lookup again.
>>           */
>> -       if (!gpc->valid || old_uhva != gpc->uhva) {
>> +       if (!gpc->valid || hva_change) {
>>                  ret = hva_to_pfn_retry(gpc);
>>          } else {
>>                  /*
>> -- 
> 
> But I don't really think it's that important if you can come up with a
> coherent justification for the change and note it in the commit
> message. So either way:
> 
> Reviewed-by: David Woodhouse <dwmw@amazon.co.uk>

Thanks,

   Paul