On Tue, Jun 20, 2023 at 09:01:23AM +0200, David Hildenbrand wrote: > On 19.06.23 23:43, Peter Xu wrote: > > On Tue, Jun 13, 2023 at 05:53:42PM -0400, Peter Xu wrote: > > > @@ -6506,6 +6507,7 @@ struct page *hugetlb_follow_page_mask(struct vm_area_struct *vma, > > > * because we hold the ptl lock and have verified pte_present(). > > > */ > > > WARN_ON_ONCE(try_grab_page(page, flags)); > > > + *page_mask = huge_page_mask(h); > > > > Sorry, I was wrong this line. It should be: > > > > *page_mask = ~huge_page_mask(h) >> PAGE_SHIFT; > > > > That's ... surprising. It feels like either page_mask or huge_page_mask() > has a misleading name .... > > h->mask = ~(huge_page_size(h) - 1); > > > For PMDs, we do > > ctx->page_mask = HPAGE_PMD_NR - 1; > > > Maybe > > *page_mask = PHYS_PFN(huge_page_size(h)) - 1; > > Would be clearer. Since I just posted a new version.. I'll see whether I should get that cleaned up in a new one. > > I guess "page_mask" should actually be "pfn_mask" ... but the meaning > regarding PAGE_MASK are still inverted ... Yes, pfn_mask can be at least slightly better. I can add a patch to rename it, or we can also do it on top as cleanups. Thanks, -- Peter Xu
