[PATCH] mbind: check_range use standard ptwalk
Hugh Dickins [Wed, 22 Jun 2005 00:15:07 +0000 (17:15 -0700)]
Strict mbind's check for currently mapped pages being on node has been
using a slow loop which re-evaluates pgd, pud, pmd, pte for each entry:
replace that by a standard four-level page table walk like others in mm.
Since mmap_sem is held for writing, page_table_lock can be taken at the
inner level to limit latency.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

mm/mempolicy.c

index c512cc9..cb41c31 100644 (file)
@@ -238,56 +238,81 @@ static struct mempolicy *mpol_new(int mode, unsigned long *nodes)
 }
 
 /* Ensure all existing pages follow the policy. */
-static int
-verify_pages(struct mm_struct *mm,
-            unsigned long addr, unsigned long end, unsigned long *nodes)
+static int check_pte_range(struct mm_struct *mm, pmd_t *pmd,
+               unsigned long addr, unsigned long end, unsigned long *nodes)
 {
-       int err = 0;
+       pte_t *orig_pte;
+       pte_t *pte;
 
        spin_lock(&mm->page_table_lock);
-       while (addr < end) {
-               struct page *p;
-               pte_t *pte;
-               pmd_t *pmd;
-               pud_t *pud;
-               pgd_t *pgd;
-               pgd = pgd_offset(mm, addr);
-               if (pgd_none(*pgd)) {
-                       unsigned long next = (addr + PGDIR_SIZE) & PGDIR_MASK;
-                       if (next > addr)
-                               break;
-                       addr = next;
-                       continue;
-               }
-               pud = pud_offset(pgd, addr);
-               if (pud_none(*pud)) {
-                       addr = (addr + PUD_SIZE) & PUD_MASK;
+       orig_pte = pte = pte_offset_map(pmd, addr);
+       do {
+               unsigned long pfn;
+               unsigned int nid;
+
+               if (!pte_present(*pte))
                        continue;
-               }
-               pmd = pmd_offset(pud, addr);
-               if (pmd_none(*pmd)) {
-                       addr = (addr + PMD_SIZE) & PMD_MASK;
+               pfn = pte_pfn(*pte);
+               if (!pfn_valid(pfn))
                        continue;
-               }
-               p = NULL;
-               pte = pte_offset_map(pmd, addr);
-               if (pte_present(*pte)) {
-                       unsigned long pfn = pte_pfn(*pte);
-                       if (pfn_valid(pfn))
-                               p = pfn_to_page(pfn);
-               }
-               pte_unmap(pte);
-               if (p) {
-                       unsigned nid = page_to_nid(p);
-                       if (!test_bit(nid, nodes)) {
-                               err = -EIO;
-                               break;
-                       }
-               }
-               addr += PAGE_SIZE;
-       }
+               nid = pfn_to_nid(pfn);
+               if (!test_bit(nid, nodes))
+                       break;
+       } while (pte++, addr += PAGE_SIZE, addr != end);
+       pte_unmap(orig_pte);
        spin_unlock(&mm->page_table_lock);
-       return err;
+       return addr != end;
+}
+
+static inline int check_pmd_range(struct mm_struct *mm, pud_t *pud,
+               unsigned long addr, unsigned long end, unsigned long *nodes)
+{
+       pmd_t *pmd;
+       unsigned long next;
+
+       pmd = pmd_offset(pud, addr);
+       do {
+               next = pmd_addr_end(addr, end);
+               if (pmd_none_or_clear_bad(pmd))
+                       continue;
+               if (check_pte_range(mm, pmd, addr, next, nodes))
+                       return -EIO;
+       } while (pmd++, addr = next, addr != end);
+       return 0;
+}
+
+static inline int check_pud_range(struct mm_struct *mm, pgd_t *pgd,
+               unsigned long addr, unsigned long end, unsigned long *nodes)
+{
+       pud_t *pud;
+       unsigned long next;
+
+       pud = pud_offset(pgd, addr);
+       do {
+               next = pud_addr_end(addr, end);
+               if (pud_none_or_clear_bad(pud))
+                       continue;
+               if (check_pmd_range(mm, pud, addr, next, nodes))
+                       return -EIO;
+       } while (pud++, addr = next, addr != end);
+       return 0;
+}
+
+static inline int check_pgd_range(struct mm_struct *mm,
+               unsigned long addr, unsigned long end, unsigned long *nodes)
+{
+       pgd_t *pgd;
+       unsigned long next;
+
+       pgd = pgd_offset(mm, addr);
+       do {
+               next = pgd_addr_end(addr, end);
+               if (pgd_none_or_clear_bad(pgd))
+                       continue;
+               if (check_pud_range(mm, pgd, addr, next, nodes))
+                       return -EIO;
+       } while (pgd++, addr = next, addr != end);
+       return 0;
 }
 
 /* Step 1: check the range */
@@ -308,7 +333,7 @@ check_range(struct mm_struct *mm, unsigned long start, unsigned long end,
                if (prev && prev->vm_end < vma->vm_start)
                        return ERR_PTR(-EFAULT);
                if ((flags & MPOL_MF_STRICT) && !is_vm_hugetlb_page(vma)) {
-                       err = verify_pages(vma->vm_mm,
+                       err = check_pgd_range(vma->vm_mm,
                                           vma->vm_start, vma->vm_end, nodes);
                        if (err) {
                                first = ERR_PTR(err);