Re: [PATCH 2/3] pagewalk: separate function pointers from iterator data

[Jason Gunthorpe]

On Mon, Sep 02, 2019 at 09:58:59AM +0200, Christoph Hellwig wrote:
> On Mon, Sep 02, 2019 at 05:51:58AM +0000, Jason Gunthorpe wrote:
> > On Sun, Sep 01, 2019 at 01:35:16PM -0700, Guenter Roeck wrote:
> > > > I belive the macros above are missing brackets.. Can you confirm the
> > > > below takes care of things? I'll add a patch if so
> > > > 
> > > 
> > > Good catch. Yes, that fixes the build problem.
> > 
> > I added this to the hmm tree to fix it:
> 
> This looks good.  Although I still haven't figure out how this is
> related to the pagewalk changes to start with..

It is this hunk:

@@ -481,7 +461,10 @@ static int madvise_free_single_vma(struct
vm_area_struct *vma,
 	       update_hiwater_rss(mm);
 
	mmu_notifier_invalidate_range_start(&range);
-	madvise_free_page_range(&tlb, vma, range.start, range.end);
+	tlb_start_vma(&tlb, vma);
+	walk_page_range(vma->vm_mm, range.start, range.end,
+                       &madvise_free_walk_ops, &tlb);
+			tlb_end_vma(&tlb, vma);

&tlb does not expand properly in the csky tlb_start_vma macro, and
previously it was just tlb

Jason