On Mon, 27 Aug 2018 09:47:01 +0200
Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> On Mon, Aug 27, 2018 at 03:00:08PM +1000, Nicholas Piggin wrote:
> > On Fri, 24 Aug 2018 13:39:53 +0200
> > Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > > On Fri, Aug 24, 2018 at 01:32:14PM +0200, Peter Zijlstra wrote:
>
> > > > Hurm.. look at commit:
> > > >
> > > > e77b0852b551 ("mm/mmu_gather: track page size with mmu gather and force flush if page size change")
> > >
> > > Ah, good, it seems that already got cleaned up a lot. But it all moved
> > > into the power code.. blergh.
> >
> > I lost track of what the problem is here?
>
> Aside from the commit above being absolute crap (which did get fixed up,
> luckily) I would really like to get rid of all arch specific mmu_gather.
>
> We can have opt-in bits to the generic code, but the endless back and
> forth between common and arch code is an utter pain in the arse.
>
> And there's only like 4 architectures that still have a custom
> mmu_gather:
>
> - sh
> - arm
> - ia64
> - s390
>
> sh is trivial, arm seems doable, with a bit of luck we can do 'rm -rf
> arch/ia64' leaving us with s390.
Well I don't see a big problem in having an arch_mmu_gather field
or small bits. powerpc would actually like that rather than trying
to add things it wants into generic code (and it wants more than
just a few flags bits, ideally).
> After that everyone uses the common code and we can clean up.
>
> > For powerpc, tlb_start_vma is not the right API to use for this because
> > it wants to deal with different page sizes within a vma.
>
> Yes.. I see that. tlb_remove_check_page_size_change() really is a rather
> ugly thing, it can cause loads of TLB flushes. Do you really _have_ to
> do that? The way ARM and x86 work is that using INVLPG in a 4K stride is
> still correct for huge pages, inefficient maybe, but so is flushing
> every other page because 'sparse' transparant-huge-pages.
It could do that. It requires a tlbie that matches the page size,
so it means 3 sizes. I think possibly even that would be better
than current code, but we could do better if we had a few specific
fields in there.
Thanks,
Nick
