Re: [External] Re: [PATCH v3 05/21] mm/hugetlb: Introduce pgtable allocation/freeing helpers

From: Muchun Song
Date: Tue Nov 10 2020 - 01:09:28 EST


On Tue, Nov 10, 2020 at 1:42 PM Oscar Salvador <osalvador@xxxxxxx> wrote:
>
> On Tue, Nov 10, 2020 at 11:49:27AM +0800, Muchun Song wrote:
> > On Tue, Nov 10, 2020 at 1:21 AM Oscar Salvador <osalvador@xxxxxxx> wrote:
> > >
> > > On Sun, Nov 08, 2020 at 10:10:57PM +0800, Muchun Song wrote:
> > > > +static inline unsigned int pgtable_pages_to_prealloc_per_hpage(struct hstate *h)
> > > > +{
> > > > + unsigned long vmemmap_size = vmemmap_pages_size_per_hpage(h);
> > > > +
> > > > + /*
> > > > + * No need pre-allocate page tabels when there is no vmemmap pages
> > > > + * to free.
> > > s /tabels/tables/
> >
> > Thanks.
> >
> > >
> > > > +static int vmemmap_pgtable_prealloc(struct hstate *h, struct page *page)
> > > > +{
> > > > + int i;
> > > > + pgtable_t pgtable;
> > > > + unsigned int nr = pgtable_pages_to_prealloc_per_hpage(h);
> > > > +
> > > > + if (!nr)
> > > > + return 0;
> > > > +
> > > > + vmemmap_pgtable_init(page);
> > > > +
> > > > + for (i = 0; i < nr; i++) {
> > > > + pte_t *pte_p;
> > > > +
> > > > + pte_p = pte_alloc_one_kernel(&init_mm);
> > > > + if (!pte_p)
> > > > + goto out;
> > > > + vmemmap_pgtable_deposit(page, virt_to_page(pte_p));
> > > > + }
> > > > +
> > > > + return 0;
> > > > +out:
> > > > + while (i-- && (pgtable = vmemmap_pgtable_withdraw(page)))
> > > > + pte_free_kernel(&init_mm, page_to_virt(pgtable));
> > >
> > > would not be enough to:
> > >
> > > while (pgtable = vmemmap_pgtable_withdrag(page))
> > > pte_free_kernel(&init_mm, page_to_virt(pgtable));
> >
> > The vmemmap_pgtable_withdraw can not return NULL. So we can not
> > drop the "i--".
>
> Yeah, you are right, I managed to confuse myself.
> But why not make it return null, something like:
>
> static pgtable_t vmemmap_pgtable_withdraw(struct page *page)
> {
> pgtable_t pgtable;
>
> /* FIFO */
> pgtable = page_huge_pte(page);

The check should be added here.

if (!pgtable)
return NULL;

Just like my previous v2 patch does. In this case, we can drop those
checks. What do you think?

> page_huge_pte(page) = list_first_entry_or_null(&pgtable->lru,
> struct page, lru);
> if (page_huge_pte(page))
> list_del(&pgtable->lru);
>
> return page_huge_pte(page) ? pgtable : NULL;
> }
>
> What do you think?
>
>
> --
> Oscar Salvador
> SUSE L3



--
Yours,
Muchun