Re: [PATCH v3] mm: improve mprotect(R|W) efficiency on pages referenced once
From: Peter Xu <peterx@redhat.com>
Date: 2021-05-21 18:58:04
Subsystem:
memory management, memory mapping, the rest · Maintainers:
Andrew Morton, Liam R. Howlett, Lorenzo Stoakes, Linus Torvalds
On Thu, Apr 29, 2021 at 02:48:01PM -0700, Peter Collingbourne wrote:
quoted hunk ↗ jump to hunk
@@ -132,9 +134,12 @@ static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd, } /* Avoid taking write faults for known dirty pages */ - if (dirty_accountable && pte_dirty(ptent) && - (pte_soft_dirty(ptent) || - !(vma->vm_flags & VM_SOFTDIRTY))) { + if ((dirty_accountable || + (anon_writable && + page_mapcount(pte_page(ptent)) == 1)) && + pte_dirty(ptent) && + (pte_soft_dirty(ptent) || + !(vma->vm_flags & VM_SOFTDIRTY))) { ptent = pte_mkwrite(ptent); } ptep_modify_prot_commit(vma, addr, pte, oldpte, ptent);
This breaks uffd-wp anonymous page. I think we'd at least need below to be squashed:
diff --git a/mm/mprotect.c b/mm/mprotect.c
index ed10a790933e..6cb531eab811 100644
--- a/mm/mprotect.c
+++ b/mm/mprotect.c@@ -139,6 +139,7 @@ static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd, (anon_writable && page_mapcount(pte_page(ptent)) == 1)) && pte_dirty(ptent) && + !pte_uffd_wp(ptent) && (pte_soft_dirty(ptent) || !(vma->vm_flags & VM_SOFTDIRTY))) { ptent = pte_mkwrite(ptent);
It'll be good to start thinking about how to not continue growing this if clause condition, though.. -- Peter Xu