[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <df9afef5-f5db-4d50-9803-28f2ce30086d@redhat.com>
Date: Thu, 4 Jul 2024 17:41:39 +0200
From: David Hildenbrand <david@...hat.com>
To: Oscar Salvador <osalvador@...e.de>,
Andrew Morton <akpm@...ux-foundation.org>
Cc: linux-kernel@...r.kernel.org, linux-mm@...ck.org,
Peter Xu <peterx@...hat.com>, Muchun Song <muchun.song@...ux.dev>,
SeongJae Park <sj@...nel.org>, Miaohe Lin <linmiaohe@...wei.com>,
Michal Hocko <mhocko@...e.com>, Matthew Wilcox <willy@...radead.org>,
Christophe Leroy <christophe.leroy@...roup.eu>
Subject: Re: [PATCH 05/45] mm/pagewalk: Enable walk_pmd_range to handle
cont-pmds
On 04.07.24 06:30, Oscar Salvador wrote:
> HugeTLB pages can be cont-pmd mapped, so teach walk_pmd_range to
> handle those.
> This will save us some cycles as we do it in one-shot instead of
> calling in multiple times.
>
> Signed-off-by: Oscar Salvador <osalvador@...e.de>
> ---
> include/linux/pgtable.h | 12 ++++++++++++
> mm/pagewalk.c | 12 +++++++++---
> 2 files changed, 21 insertions(+), 3 deletions(-)
>
> diff --git a/include/linux/pgtable.h b/include/linux/pgtable.h
> index 2a6a3cccfc36..3a7b8751747e 100644
> --- a/include/linux/pgtable.h
> +++ b/include/linux/pgtable.h
> @@ -1914,6 +1914,18 @@ typedef unsigned int pgtbl_mod_mask;
> #define __pte_leaf_size(x,y) pte_leaf_size(y)
> #endif
>
> +#ifndef pmd_cont
> +#define pmd_cont(x) false
> +#endif
> +
> +#ifndef CONT_PMD_SIZE
> +#define CONT_PMD_SIZE 0
> +#endif
> +
> +#ifndef CONT_PMDS
> +#define CONT_PMDS 0
> +#endif
> +
> /*
> * We always define pmd_pfn for all archs as it's used in lots of generic
> * code. Now it happens too for pud_pfn (and can happen for larger
> diff --git a/mm/pagewalk.c b/mm/pagewalk.c
> index d93e77411482..a9c36f9e9820 100644
> --- a/mm/pagewalk.c
> +++ b/mm/pagewalk.c
> @@ -81,11 +81,18 @@ static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
> const struct mm_walk_ops *ops = walk->ops;
> int err = 0;
> int depth = real_depth(3);
> + int cont_pmds;
>
> pmd = pmd_offset(pud, addr);
> do {
> again:
> - next = pmd_addr_end(addr, end);
> + if (pmd_cont(*pmd)) {
> + cont_pmds = CONT_PMDS;
> + next = pmd_cont_addr_end(addr, end);
> + } else {
> + cont_pmds = 1;
> + next = pmd_addr_end(addr, end);
> + }
> if (pmd_none(*pmd)) {
> if (ops->pte_hole)
> err = ops->pte_hole(addr, next, depth, walk);
> @@ -126,8 +133,7 @@ static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
>
> if (walk->action == ACTION_AGAIN)
> goto again;
> -
> - } while (pmd++, addr = next, addr != end);
> + } while (pmd += cont_pmds, addr = next, addr != end);
Similar to my other comment regarding PTE batching, this is very
specific to architectures that support cont-pmds.
Yes, right now we only have that on architectures that support
cont-pmd-sized hugetlb, but Willy is interested in us supporting+mapping
folios > PMD_SIZE, whereby we'd want to batch even without arch-specific
cont-pmd bits.
Similar to the other (pte) case, having a way to generically patch
folios will me more beneficial. Note that cont-pmd/cont-pte is only
relevant for present entries (-> mapping folios).
--
Cheers,
David / dhildenb
Powered by blists - more mailing lists