On Wed, Nov 11, 2020 at 03:30:22PM +0000, Matthew Wilcox wrote:
> On Wed, Nov 11, 2020 at 01:43:57PM +0100, Peter Zijlstra wrote:
> > + if (pud_leaf(pud)) {
> > #ifdef pud_page
> > - page = pud_page(*pud);
> > - if (PageHuge(page))
> > - return page_size(compound_head(page));
> > + if (!pud_devmap(pud)) {
> > + page = pud_page(pud);
> > + if (PageHuge(page))
> > + return page_size(compound_head(page));
> > + }
> > #endif
> > return 1ULL << PUD_SHIFT;
>
> This confuses me. Why only special-case hugetlbfs pages here? Should
> they really be treated differently from THP?
Do we have non-pagetable aligned THP ? I thought THP was always PUD
sized.
> If you want to consider that we might be mapping a page that's twice
> as big as a PUD entry and this is only half of it, then the simple way
> is:
>
> if (pud_leaf(pud)) {
> #ifdef pud_page
> page = compound_head(pud_page(*pud));
> return page_size(page);
> #else
> return 1ULL << PUD_SHIFT;
> #endif
> }
>
> Also, what's up with the special-casing of devmap pages here? Did the
> devmap people fuck up their compound pages? If so, they should fix their
> shit, not expect the rest of the kernel to work around this brokenness.
Well, the PTE code we have today (in tip/perf/core) is:
pte = pte_offset_map(pmd, addr);
if (!pte_present(*pte)) {
pte_unmap(pte);
return 0;
}
page = pte_page(*pte);
if (PageHuge(page)) {
u64 size = page_size(compound_head(page));
pte_unmap(pte);
return size;
}
pte_unmap(pte);
return PAGE_SIZE;
and that's crashing in PageHuge()'s PageCompound() test. Clearly I
should be checking pte_special() here (as well as all the READ_ONCE()s I
added in the patch you just commented on). But I wasn't quite sure about
devmap and paranoia won.
You're saying devmap should be valid compound pages? Then I can remove
all that and only keep pte_special().
