On Thu, Dec 17, 2020 at 9:17 AM Mike Kravetz <mike.kravetz@xxxxxxxxxx> wrote:
>
> On 12/13/20 7:45 AM, Muchun Song wrote:
> > When we free a HugeTLB page to the buddy allocator, we should allocate the
> > vmemmap pages associated with it. We can do that in the __free_hugepage()
> > before freeing it to buddy.
>
> ...
>
> > diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
> > index 78c527617e8d..ffcf092c92ed 100644
> > --- a/mm/sparse-vmemmap.c
> > +++ b/mm/sparse-vmemmap.c
> > @@ -29,6 +29,7 @@
> > #include <linux/sched.h>
> > #include <linux/pgtable.h>
> > #include <linux/bootmem_info.h>
> > +#include <linux/delay.h>
> >
> > #include <asm/dma.h>
> > #include <asm/pgalloc.h>
> > @@ -39,7 +40,8 @@
> > *
> > * @rmap_pte: called for each non-empty PTE (lowest-level) entry.
> > * @reuse: the page which is reused for the tail vmemmap pages.
> > - * @vmemmap_pages: the list head of the vmemmap pages that can be freed.
> > + * @vmemmap_pages: the list head of the vmemmap pages that can be freed
> > + * or is mapped from.
> > */
> > struct vmemmap_rmap_walk {
> > void (*rmap_pte)(pte_t *pte, unsigned long addr,
> > @@ -54,6 +56,9 @@ struct vmemmap_rmap_walk {
> > */
> > #define VMEMMAP_TAIL_PAGE_REUSE -1
> >
> > +/* The gfp mask of allocating vmemmap page */
> > +#define GFP_VMEMMAP_PAGE (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
> > +
> > static void vmemmap_pte_range(pmd_t *pmd, unsigned long addr,
> > unsigned long end, struct vmemmap_rmap_walk *walk)
> > {
> > @@ -200,6 +205,68 @@ void vmemmap_remap_reuse(unsigned long start, unsigned long size)
> > free_vmemmap_page_list(&vmemmap_pages);
> > }
> >
> > +static void vmemmap_remap_restore_pte(pte_t *pte, unsigned long addr,
> > + struct vmemmap_rmap_walk *walk)
> > +{
> > + pgprot_t pgprot = PAGE_KERNEL;
> > + struct page *page;
> > + void *to;
> > +
> > + BUG_ON(pte_page(*pte) != walk->reuse);
> > +
> > + page = list_first_entry(walk->vmemmap_pages, struct page, lru);
> > + list_del(&page->lru);
> > + to = page_to_virt(page);
> > + copy_page(to, page_to_virt(walk->reuse));
> > +
> > + set_pte_at(&init_mm, addr, pte, mk_pte(page, pgprot));
> > +}
> > +
> > +static void alloc_vmemmap_page_list(struct list_head *list,
> > + unsigned long nr_pages)
> > +{
> > + while (nr_pages--) {
> > + struct page *page;
> > +
> > +retry:
> > + page = alloc_page(GFP_VMEMMAP_PAGE);
>
> Should we try (or require) the vmemmap page be on the same node as the
> pages they describe? I imagine performance would be impacted if a
> struct page and the page it describes are on different numa nodes.
Yeah, it is a good idea. I also think that we should do this. I will do that in
the next version. Thanks.
>
> > + if (unlikely(!page)) {
> > + msleep(100);
> > + /*
> > + * We should retry infinitely, because we cannot
> > + * handle allocation failures. Once we allocate
> > + * vmemmap pages successfully, then we can free
> > + * a HugeTLB page.
> > + */
> > + goto retry;
> > + }
> > + list_add_tail(&page->lru, list);
> > + }
> > +}
> > +
>
> --
> Mike Kravetz
--
Yours,
Muchun
