]> asedeno.scripts.mit.edu Git - linux.git/blobdiff - mm/pagewalk.c
Merge tag 'tegra-for-5.5-cpufreq' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux.git] / mm / pagewalk.c
index d48c2a986ea3f8dfffaed7c2c99ca980375f4726..ea0b9e606ad13525a726899e818cc0104d5185f2 100644 (file)
@@ -10,8 +10,9 @@ static int walk_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
        pte_t *pte;
        int err = 0;
        const struct mm_walk_ops *ops = walk->ops;
+       spinlock_t *ptl;
 
-       pte = pte_offset_map(pmd, addr);
+       pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
        for (;;) {
                err = ops->pte_entry(pte, addr, addr + PAGE_SIZE, walk);
                if (err)
@@ -22,7 +23,7 @@ static int walk_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
                pte++;
        }
 
-       pte_unmap(pte);
+       pte_unmap_unlock(pte, ptl);
        return err;
 }
 
@@ -253,13 +254,23 @@ static int __walk_page_range(unsigned long start, unsigned long end,
 {
        int err = 0;
        struct vm_area_struct *vma = walk->vma;
+       const struct mm_walk_ops *ops = walk->ops;
+
+       if (vma && ops->pre_vma) {
+               err = ops->pre_vma(start, end, walk);
+               if (err)
+                       return err;
+       }
 
        if (vma && is_vm_hugetlb_page(vma)) {
-               if (walk->ops->hugetlb_entry)
+               if (ops->hugetlb_entry)
                        err = walk_hugetlb_range(start, end, walk);
        } else
                err = walk_pgd_range(start, end, walk);
 
+       if (vma && ops->post_vma)
+               ops->post_vma(walk);
+
        return err;
 }
 
@@ -290,6 +301,11 @@ static int __walk_page_range(unsigned long start, unsigned long end,
  * its vm_flags. walk_page_test() and @ops->test_walk() are used for this
  * purpose.
  *
+ * If operations need to be staged before and committed after a vma is walked,
+ * there are two callbacks, pre_vma() and post_vma(). Note that post_vma(),
+ * since it is intended to handle commit-type operations, can't return any
+ * errors.
+ *
  * struct mm_walk keeps current values of some common data like vma and pmd,
  * which are useful for the access from callbacks. If you want to pass some
  * caller-specific data to callbacks, @private should be helpful.
@@ -376,3 +392,80 @@ int walk_page_vma(struct vm_area_struct *vma, const struct mm_walk_ops *ops,
                return err;
        return __walk_page_range(vma->vm_start, vma->vm_end, &walk);
 }
+
+/**
+ * walk_page_mapping - walk all memory areas mapped into a struct address_space.
+ * @mapping: Pointer to the struct address_space
+ * @first_index: First page offset in the address_space
+ * @nr: Number of incremental page offsets to cover
+ * @ops:       operation to call during the walk
+ * @private:   private data for callbacks' usage
+ *
+ * This function walks all memory areas mapped into a struct address_space.
+ * The walk is limited to only the given page-size index range, but if
+ * the index boundaries cross a huge page-table entry, that entry will be
+ * included.
+ *
+ * Also see walk_page_range() for additional information.
+ *
+ * Locking:
+ *   This function can't require that the struct mm_struct::mmap_sem is held,
+ *   since @mapping may be mapped by multiple processes. Instead
+ *   @mapping->i_mmap_rwsem must be held. This might have implications in the
+ *   callbacks, and it's up tho the caller to ensure that the
+ *   struct mm_struct::mmap_sem is not needed.
+ *
+ *   Also this means that a caller can't rely on the struct
+ *   vm_area_struct::vm_flags to be constant across a call,
+ *   except for immutable flags. Callers requiring this shouldn't use
+ *   this function.
+ *
+ * Return: 0 on success, negative error code on failure, positive number on
+ * caller defined premature termination.
+ */
+int walk_page_mapping(struct address_space *mapping, pgoff_t first_index,
+                     pgoff_t nr, const struct mm_walk_ops *ops,
+                     void *private)
+{
+       struct mm_walk walk = {
+               .ops            = ops,
+               .private        = private,
+       };
+       struct vm_area_struct *vma;
+       pgoff_t vba, vea, cba, cea;
+       unsigned long start_addr, end_addr;
+       int err = 0;
+
+       lockdep_assert_held(&mapping->i_mmap_rwsem);
+       vma_interval_tree_foreach(vma, &mapping->i_mmap, first_index,
+                                 first_index + nr - 1) {
+               /* Clip to the vma */
+               vba = vma->vm_pgoff;
+               vea = vba + vma_pages(vma);
+               cba = first_index;
+               cba = max(cba, vba);
+               cea = first_index + nr;
+               cea = min(cea, vea);
+
+               start_addr = ((cba - vba) << PAGE_SHIFT) + vma->vm_start;
+               end_addr = ((cea - vba) << PAGE_SHIFT) + vma->vm_start;
+               if (start_addr >= end_addr)
+                       continue;
+
+               walk.vma = vma;
+               walk.mm = vma->vm_mm;
+
+               err = walk_page_test(vma->vm_start, vma->vm_end, &walk);
+               if (err > 0) {
+                       err = 0;
+                       break;
+               } else if (err < 0)
+                       break;
+
+               err = __walk_page_range(start_addr, end_addr, &walk);
+               if (err)
+                       break;
+       }
+
+       return err;
+}