From nobody Tue Jun 23 08:17:07 2026 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id B81C9C433EF for ; Wed, 9 Mar 2022 02:57:35 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S231560AbiCIC6b (ORCPT ); Tue, 8 Mar 2022 21:58:31 -0500 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:58220 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231497AbiCIC6a (ORCPT ); Tue, 8 Mar 2022 21:58:30 -0500 Received: from loongson.cn (mail.loongson.cn [114.242.206.163]) by lindbergh.monkeyblade.net (Postfix) with ESMTP id 288EF3EAB6 for ; Tue, 8 Mar 2022 18:57:28 -0800 (PST) Received: from localhost.localdomain (unknown [10.2.5.185]) by mail.loongson.cn (Coremail) with SMTP id AQAAf9DxvxORFyhi1W0FAA--.123S2; Wed, 09 Mar 2022 10:57:21 +0800 (CST) From: Bibo Mao To: Andrew Morton Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org Subject: [PATCH] mm: reduce tlb flush range when changing vma protection Date: Tue, 8 Mar 2022 21:57:21 -0500 Message-Id: <20220309025721.3051365-1-maobibo@loongson.cn> X-Mailer: git-send-email 2.31.1 MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable X-CM-TRANSID: AQAAf9DxvxORFyhi1W0FAA--.123S2 X-Coremail-Antispam: 1UD129KBjvJXoWxZrWkZryfZr45uFWfZw17Awb_yoW7JF1UpF 48Ka4ktF42q3yqgF9xZrW5Z343Jw17Xa1xAa9FgasYqFn8ta43XFy3GayF9r15AFykZF9I yayaqw1vkw47Z3JanT9S1TB71UUUUU7qnTZGkaVYY2UrUUUUjbIjqfuFe4nvWSU5nxnvy2 9KBjDU0xBIdaVrnUUvcSsGvfC2KfnxnUUI43ZEXa7xR_UUUUUUUUU== X-CM-SenderInfo: xpdruxter6z05rqj20fqof0/ Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Type: text/plain; charset="utf-8" numa worker will periodically change vma prot with PROT_NONE, by default it will scan 256M vma memory size with pmd stepping size. If there are fewer pages changed with PROT_NONE, tlb flush is called with pmd size. This patch will calculate flush range for those pages with pte prot changed, it will reduce size for tlb flush. Signed-off-by: Bibo Mao --- mm/mprotect.c | 39 +++++++++++++++++++++++++++++---------- 1 file changed, 29 insertions(+), 10 deletions(-) diff --git a/mm/mprotect.c b/mm/mprotect.c index 2887644fd150..a9f51a998dc8 100644 --- a/mm/mprotect.c +++ b/mm/mprotect.c @@ -35,9 +35,23 @@ =20 #include "internal.h" =20 +typedef struct { + unsigned long start; + unsigned long end; +} tlb_range; + +static inline void add_tlb_range(tlb_range *range, unsigned long start, + unsigned long end) +{ + if (start < range->start) + range->start =3D start; + if (end > range->end) + range->end =3D end; +} + static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *p= md, unsigned long addr, unsigned long end, pgprot_t newprot, - unsigned long cp_flags) + unsigned long cp_flags, tlb_range *range) { pte_t *pte, oldpte; spinlock_t *ptl; @@ -138,6 +152,7 @@ static unsigned long change_pte_range(struct vm_area_st= ruct *vma, pmd_t *pmd, ptent =3D pte_mkwrite(ptent); } ptep_modify_prot_commit(vma, addr, pte, oldpte, ptent); + add_tlb_range(range, addr, addr + PAGE_SIZE); pages++; } else if (is_swap_pte(oldpte)) { swp_entry_t entry =3D pte_to_swp_entry(oldpte); @@ -184,6 +199,7 @@ static unsigned long change_pte_range(struct vm_area_st= ruct *vma, pmd_t *pmd, =20 if (!pte_same(oldpte, newpte)) { set_pte_at(vma->vm_mm, addr, pte, newpte); + add_tlb_range(range, addr, addr + PAGE_SIZE); pages++; } } @@ -221,7 +237,7 @@ static inline int pmd_none_or_clear_bad_unless_trans_hu= ge(pmd_t *pmd) =20 static inline unsigned long change_pmd_range(struct vm_area_struct *vma, pud_t *pud, unsigned long addr, unsigned long end, - pgprot_t newprot, unsigned long cp_flags) + pgprot_t newprot, unsigned long cp_flags, tlb_range *tlb) { pmd_t *pmd; unsigned long next; @@ -267,6 +283,7 @@ static inline unsigned long change_pmd_range(struct vm_= area_struct *vma, if (nr_ptes) { if (nr_ptes =3D=3D HPAGE_PMD_NR) { pages +=3D HPAGE_PMD_NR; + add_tlb_range(tlb, addr, next); nr_huge_updates++; } =20 @@ -277,7 +294,7 @@ static inline unsigned long change_pmd_range(struct vm_= area_struct *vma, /* fall through, the trans huge pmd just split */ } this_pages =3D change_pte_range(vma, pmd, addr, next, newprot, - cp_flags); + cp_flags, tlb); pages +=3D this_pages; next: cond_resched(); @@ -293,7 +310,7 @@ static inline unsigned long change_pmd_range(struct vm_= area_struct *vma, =20 static inline unsigned long change_pud_range(struct vm_area_struct *vma, p4d_t *p4d, unsigned long addr, unsigned long end, - pgprot_t newprot, unsigned long cp_flags) + pgprot_t newprot, unsigned long cp_flags, tlb_range *range) { pud_t *pud; unsigned long next; @@ -305,7 +322,7 @@ static inline unsigned long change_pud_range(struct vm_= area_struct *vma, if (pud_none_or_clear_bad(pud)) continue; pages +=3D change_pmd_range(vma, pud, addr, next, newprot, - cp_flags); + cp_flags, range); } while (pud++, addr =3D next, addr !=3D end); =20 return pages; @@ -313,7 +330,7 @@ static inline unsigned long change_pud_range(struct vm_= area_struct *vma, =20 static inline unsigned long change_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, unsigned long addr, unsigned long end, - pgprot_t newprot, unsigned long cp_flags) + pgprot_t newprot, unsigned long cp_flags, tlb_range *range) { p4d_t *p4d; unsigned long next; @@ -325,7 +342,7 @@ static inline unsigned long change_p4d_range(struct vm_= area_struct *vma, if (p4d_none_or_clear_bad(p4d)) continue; pages +=3D change_pud_range(vma, p4d, addr, next, newprot, - cp_flags); + cp_flags, range); } while (p4d++, addr =3D next, addr !=3D end); =20 return pages; @@ -338,24 +355,26 @@ static unsigned long change_protection_range(struct v= m_area_struct *vma, struct mm_struct *mm =3D vma->vm_mm; pgd_t *pgd; unsigned long next; - unsigned long start =3D addr; unsigned long pages =3D 0; + tlb_range range; =20 BUG_ON(addr >=3D end); pgd =3D pgd_offset(mm, addr); flush_cache_range(vma, addr, end); + range.start =3D end; + range.end =3D addr; inc_tlb_flush_pending(mm); do { next =3D pgd_addr_end(addr, end); if (pgd_none_or_clear_bad(pgd)) continue; pages +=3D change_p4d_range(vma, pgd, addr, next, newprot, - cp_flags); + cp_flags, &range); } while (pgd++, addr =3D next, addr !=3D end); =20 /* Only flush the TLB if we actually modified any entries: */ if (pages) - flush_tlb_range(vma, start, end); + flush_tlb_range(vma, range.start, range.end); dec_tlb_flush_pending(mm); =20 return pages; --=20 2.31.1