mm/rmap: update to new mmu_notifier semantic v2
authorJérôme Glisse <jglisse@redhat.com>
Thu, 31 Aug 2017 21:17:27 +0000 (17:17 -0400)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 31 Aug 2017 23:12:59 +0000 (16:12 -0700)
Replace all mmu_notifier_invalidate_page() calls by *_invalidate_range()
and make sure it is bracketed by calls to *_invalidate_range_start()/end().

Note that because we can not presume the pmd value or pte value we have
to assume the worst and unconditionaly report an invalidation as
happening.

Changed since v2:
  - try_to_unmap_one() only one call to mmu_notifier_invalidate_range()
  - compute end with PAGE_SIZE << compound_order(page)
  - fix PageHuge() case in try_to_unmap_one()

Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Bernhard Held <berny156@gmx.de>
Cc: Adam Borowski <kilobyte@angband.pl>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Takashi Iwai <tiwai@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: axie <axie@amd.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/rmap.c

index c8993c63eb259b3a5302a058ce231d1290fc9b66..c570f82e6827153316465b9e18f0fca376a1c1a1 100644 (file)
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -887,11 +887,21 @@ static bool page_mkclean_one(struct page *page, struct vm_area_struct *vma,
                .address = address,
                .flags = PVMW_SYNC,
        };
+       unsigned long start = address, end;
        int *cleaned = arg;
 
+       /*
+        * We have to assume the worse case ie pmd for invalidation. Note that
+        * the page can not be free from this function.
+        */
+       end = min(vma->vm_end, start + (PAGE_SIZE << compound_order(page)));
+       mmu_notifier_invalidate_range_start(vma->vm_mm, start, end);
+
        while (page_vma_mapped_walk(&pvmw)) {
+               unsigned long cstart, cend;
                int ret = 0;
-               address = pvmw.address;
+
+               cstart = address = pvmw.address;
                if (pvmw.pte) {
                        pte_t entry;
                        pte_t *pte = pvmw.pte;
@@ -904,6 +914,7 @@ static bool page_mkclean_one(struct page *page, struct vm_area_struct *vma,
                        entry = pte_wrprotect(entry);
                        entry = pte_mkclean(entry);
                        set_pte_at(vma->vm_mm, address, pte, entry);
+                       cend = cstart + PAGE_SIZE;
                        ret = 1;
                } else {
 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
@@ -918,6 +929,8 @@ static bool page_mkclean_one(struct page *page, struct vm_area_struct *vma,
                        entry = pmd_wrprotect(entry);
                        entry = pmd_mkclean(entry);
                        set_pmd_at(vma->vm_mm, address, pmd, entry);
+                       cstart &= PMD_MASK;
+                       cend = cstart + PMD_SIZE;
                        ret = 1;
 #else
                        /* unexpected pmd-mapped page? */
@@ -926,11 +939,13 @@ static bool page_mkclean_one(struct page *page, struct vm_area_struct *vma,
                }
 
                if (ret) {
-                       mmu_notifier_invalidate_page(vma->vm_mm, address);
+                       mmu_notifier_invalidate_range(vma->vm_mm, cstart, cend);
                        (*cleaned)++;
                }
        }
 
+       mmu_notifier_invalidate_range_end(vma->vm_mm, start, end);
+
        return true;
 }
 
@@ -1324,6 +1339,7 @@ static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
        pte_t pteval;
        struct page *subpage;
        bool ret = true;
+       unsigned long start = address, end;
        enum ttu_flags flags = (enum ttu_flags)arg;
 
        /* munlock has nothing to gain from examining un-locked vmas */
@@ -1335,6 +1351,14 @@ static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
                                flags & TTU_MIGRATION, page);
        }
 
+       /*
+        * We have to assume the worse case ie pmd for invalidation. Note that
+        * the page can not be free in this function as call of try_to_unmap()
+        * must hold a reference on the page.
+        */
+       end = min(vma->vm_end, start + (PAGE_SIZE << compound_order(page)));
+       mmu_notifier_invalidate_range_start(vma->vm_mm, start, end);
+
        while (page_vma_mapped_walk(&pvmw)) {
                /*
                 * If the page is mlock()d, we cannot swap it out.
@@ -1445,6 +1469,7 @@ static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
                        if (unlikely(PageSwapBacked(page) != PageSwapCache(page))) {
                                WARN_ON_ONCE(1);
                                ret = false;
+                               /* We have to invalidate as we cleared the pte */
                                page_vma_mapped_walk_done(&pvmw);
                                break;
                        }
@@ -1490,8 +1515,12 @@ static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
 discard:
                page_remove_rmap(subpage, PageHuge(page));
                put_page(page);
-               mmu_notifier_invalidate_page(mm, address);
+               mmu_notifier_invalidate_range(mm, address,
+                                             address + PAGE_SIZE);
        }
+
+       mmu_notifier_invalidate_range_end(vma->vm_mm, start, end);
+
        return ret;
 }