Re: [PATCH v4 13/21] IOMMU/x86: prefill newly allocate page tables

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Fri, May 20, 2022 at 01:13:28PM +0200, Jan Beulich wrote:
> On 20.05.2022 13:11, Jan Beulich wrote:
> > On 20.05.2022 12:47, Roger Pau Monné wrote:
> >> On Thu, May 19, 2022 at 02:12:04PM +0200, Jan Beulich wrote:
> >>> On 06.05.2022 13:16, Roger Pau Monné wrote:
> >>>> On Mon, Apr 25, 2022 at 10:40:55AM +0200, Jan Beulich wrote:
> >>>>> --- a/xen/drivers/passthrough/amd/iommu_map.c
> >>>>> +++ b/xen/drivers/passthrough/amd/iommu_map.c
> >>>>> @@ -115,7 +115,19 @@ static void set_iommu_ptes_present(unsig
> >>>>>  
> >>>>>      while ( nr_ptes-- )
> >>>>>      {
> >>>>> -        set_iommu_pde_present(pde, next_mfn, 0, iw, ir);
> >>>>> +        ASSERT(!pde->next_level);
> >>>>> +        ASSERT(!pde->u);
> >>>>> +
> >>>>> +        if ( pde > table )
> >>>>> +            ASSERT(pde->ign0 == find_first_set_bit(pde - table));
> >>>>> +        else
> >>>>> +            ASSERT(pde->ign0 == PAGE_SHIFT - 3);
> >>>>
> >>>> I think PAGETABLE_ORDER would be clearer here.
> >>>
> >>> I disagree - PAGETABLE_ORDER is a CPU-side concept. It's not used anywhere
> >>> in IOMMU code afaics.
> >>
> >> Isn't PAGE_SHIFT also a CPU-side concept in the same way?  I'm not
> >> sure what's the rule for declaring that PAGE_SHIFT is fine to use in
> >> IOMMU code  but not PAGETABLE_ORDER.
> > 
> > Hmm, yes and no. But for consistency with other IOMMU code I may want
> > to switch to PAGE_SHIFT_4K.
> 
> Except that, with the plan to re-use pt_update_contig_markers() for CPU-
> side re-coalescing, there I'd prefer to stick to PAGE_SHIFT.

Then can PAGETABLE_ORDER be used instead of PAGE_SHIFT - 3?

IMO it makes the code quite easier to understand.

Thanks, Roger.