mm: cleanup follow_page_mask()
authorKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Wed, 4 Jun 2014 23:08:11 +0000 (16:08 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 4 Jun 2014 23:54:04 +0000 (16:54 -0700)
Cleanups:
 - move pte-related code to separate function. It's about half of the
   function;
 - get rid of some goto-logic;
 - use 'return NULL' instead of 'return page' where page can only be
   NULL;

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/gup.c

index 0bf127b332e7e3e5f9c65dba4da4974438c50739..406367845dedb948ef4251d0e388de18ff14cbbf 100644 (file)
--- a/mm/gup.c
+++ b/mm/gup.c
 
 #include "internal.h"
 
-/**
- * follow_page_mask - look up a page descriptor from a user-virtual address
- * @vma: vm_area_struct mapping @address
- * @address: virtual address to look up
- * @flags: flags modifying lookup behaviour
- * @page_mask: on output, *page_mask is set according to the size of the page
- *
- * @flags can have FOLL_ flags set, defined in <linux/mm.h>
- *
- * Returns the mapped (struct page *), %NULL if no mapping exists, or
- * an error pointer if there is a mapping to something not represented
- * by a page descriptor (see also vm_normal_page()).
- */
-struct page *follow_page_mask(struct vm_area_struct *vma,
-                             unsigned long address, unsigned int flags,
-                             unsigned int *page_mask)
+static struct page *no_page_table(struct vm_area_struct *vma,
+               unsigned int flags)
 {
-       pgd_t *pgd;
-       pud_t *pud;
-       pmd_t *pmd;
-       pte_t *ptep, pte;
-       spinlock_t *ptl;
-       struct page *page;
-       struct mm_struct *mm = vma->vm_mm;
-
-       *page_mask = 0;
-
-       page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
-       if (!IS_ERR(page)) {
-               BUG_ON(flags & FOLL_GET);
-               goto out;
-       }
-
-       page = NULL;
-       pgd = pgd_offset(mm, address);
-       if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
-               goto no_page_table;
+       /*
+        * When core dumping an enormous anonymous area that nobody
+        * has touched so far, we don't want to allocate unnecessary pages or
+        * page tables.  Return error instead of NULL to skip handle_mm_fault,
+        * then get_dump_page() will return NULL to leave a hole in the dump.
+        * But we can only make this optimization where a hole would surely
+        * be zero-filled if handle_mm_fault() actually did handle it.
+        */
+       if ((flags & FOLL_DUMP) && (!vma->vm_ops || !vma->vm_ops->fault))
+               return ERR_PTR(-EFAULT);
+       return NULL;
+}
 
-       pud = pud_offset(pgd, address);
-       if (pud_none(*pud))
-               goto no_page_table;
-       if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) {
-               if (flags & FOLL_GET)
-                       goto out;
-               page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
-               goto out;
-       }
-       if (unlikely(pud_bad(*pud)))
-               goto no_page_table;
+static struct page *follow_page_pte(struct vm_area_struct *vma,
+               unsigned long address, pmd_t *pmd, unsigned int flags)
+{
+       struct mm_struct *mm = vma->vm_mm;
+       struct page *page;
+       spinlock_t *ptl;
+       pte_t *ptep, pte;
 
-       pmd = pmd_offset(pud, address);
-       if (pmd_none(*pmd))
-               goto no_page_table;
-       if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) {
-               page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
-               if (flags & FOLL_GET) {
-                       /*
-                        * Refcount on tail pages are not well-defined and
-                        * shouldn't be taken. The caller should handle a NULL
-                        * return when trying to follow tail pages.
-                        */
-                       if (PageHead(page))
-                               get_page(page);
-                       else {
-                               page = NULL;
-                               goto out;
-                       }
-               }
-               goto out;
-       }
-       if ((flags & FOLL_NUMA) && pmd_numa(*pmd))
-               goto no_page_table;
-       if (pmd_trans_huge(*pmd)) {
-               if (flags & FOLL_SPLIT) {
-                       split_huge_page_pmd(vma, address, pmd);
-                       goto split_fallthrough;
-               }
-               ptl = pmd_lock(mm, pmd);
-               if (likely(pmd_trans_huge(*pmd))) {
-                       if (unlikely(pmd_trans_splitting(*pmd))) {
-                               spin_unlock(ptl);
-                               wait_split_huge_page(vma->anon_vma, pmd);
-                       } else {
-                               page = follow_trans_huge_pmd(vma, address,
-                                                            pmd, flags);
-                               spin_unlock(ptl);
-                               *page_mask = HPAGE_PMD_NR - 1;
-                               goto out;
-                       }
-               } else
-                       spin_unlock(ptl);
-               /* fall through */
-       }
-split_fallthrough:
+retry:
        if (unlikely(pmd_bad(*pmd)))
-               goto no_page_table;
+               return no_page_table(vma, flags);
 
        ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
-
        pte = *ptep;
        if (!pte_present(pte)) {
                swp_entry_t entry;
@@ -128,12 +58,14 @@ split_fallthrough:
                        goto no_page;
                pte_unmap_unlock(ptep, ptl);
                migration_entry_wait(mm, pmd, address);
-               goto split_fallthrough;
+               goto retry;
        }
        if ((flags & FOLL_NUMA) && pte_numa(pte))
                goto no_page;
-       if ((flags & FOLL_WRITE) && !pte_write(pte))
-               goto unlock;
+       if ((flags & FOLL_WRITE) && !pte_write(pte)) {
+               pte_unmap_unlock(ptep, ptl);
+               return NULL;
+       }
 
        page = vm_normal_page(vma, address, pte);
        if (unlikely(!page)) {
@@ -178,11 +110,8 @@ split_fallthrough:
                        unlock_page(page);
                }
        }
-unlock:
        pte_unmap_unlock(ptep, ptl);
-out:
        return page;
-
 bad_page:
        pte_unmap_unlock(ptep, ptl);
        return ERR_PTR(-EFAULT);
@@ -190,21 +119,99 @@ bad_page:
 no_page:
        pte_unmap_unlock(ptep, ptl);
        if (!pte_none(pte))
+               return NULL;
+       return no_page_table(vma, flags);
+}
+
+/**
+ * follow_page_mask - look up a page descriptor from a user-virtual address
+ * @vma: vm_area_struct mapping @address
+ * @address: virtual address to look up
+ * @flags: flags modifying lookup behaviour
+ * @page_mask: on output, *page_mask is set according to the size of the page
+ *
+ * @flags can have FOLL_ flags set, defined in <linux/mm.h>
+ *
+ * Returns the mapped (struct page *), %NULL if no mapping exists, or
+ * an error pointer if there is a mapping to something not represented
+ * by a page descriptor (see also vm_normal_page()).
+ */
+struct page *follow_page_mask(struct vm_area_struct *vma,
+                             unsigned long address, unsigned int flags,
+                             unsigned int *page_mask)
+{
+       pgd_t *pgd;
+       pud_t *pud;
+       pmd_t *pmd;
+       spinlock_t *ptl;
+       struct page *page;
+       struct mm_struct *mm = vma->vm_mm;
+
+       *page_mask = 0;
+
+       page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
+       if (!IS_ERR(page)) {
+               BUG_ON(flags & FOLL_GET);
                return page;
+       }
 
-no_page_table:
-       /*
-        * When core dumping an enormous anonymous area that nobody
-        * has touched so far, we don't want to allocate unnecessary pages or
-        * page tables.  Return error instead of NULL to skip handle_mm_fault,
-        * then get_dump_page() will return NULL to leave a hole in the dump.
-        * But we can only make this optimization where a hole would surely
-        * be zero-filled if handle_mm_fault() actually did handle it.
-        */
-       if ((flags & FOLL_DUMP) &&
-           (!vma->vm_ops || !vma->vm_ops->fault))
-               return ERR_PTR(-EFAULT);
-       return page;
+       pgd = pgd_offset(mm, address);
+       if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
+               return no_page_table(vma, flags);
+
+       pud = pud_offset(pgd, address);
+       if (pud_none(*pud))
+               return no_page_table(vma, flags);
+       if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) {
+               if (flags & FOLL_GET)
+                       return NULL;
+               page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
+               return page;
+       }
+       if (unlikely(pud_bad(*pud)))
+               return no_page_table(vma, flags);
+
+       pmd = pmd_offset(pud, address);
+       if (pmd_none(*pmd))
+               return no_page_table(vma, flags);
+       if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) {
+               page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
+               if (flags & FOLL_GET) {
+                       /*
+                        * Refcount on tail pages are not well-defined and
+                        * shouldn't be taken. The caller should handle a NULL
+                        * return when trying to follow tail pages.
+                        */
+                       if (PageHead(page))
+                               get_page(page);
+                       else
+                               page = NULL;
+               }
+               return page;
+       }
+       if ((flags & FOLL_NUMA) && pmd_numa(*pmd))
+               return no_page_table(vma, flags);
+       if (pmd_trans_huge(*pmd)) {
+               if (flags & FOLL_SPLIT) {
+                       split_huge_page_pmd(vma, address, pmd);
+                       return follow_page_pte(vma, address, pmd, flags);
+               }
+               ptl = pmd_lock(mm, pmd);
+               if (likely(pmd_trans_huge(*pmd))) {
+                       if (unlikely(pmd_trans_splitting(*pmd))) {
+                               spin_unlock(ptl);
+                               wait_split_huge_page(vma->anon_vma, pmd);
+                       } else {
+                               page = follow_trans_huge_pmd(vma, address,
+                                                            pmd, flags);
+                               spin_unlock(ptl);
+                               *page_mask = HPAGE_PMD_NR - 1;
+                               return page;
+                       }
+               } else
+                       spin_unlock(ptl);
+       }
+       return follow_page_pte(vma, address, pmd, flags);
 }
 
 static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr)