2 #include <linux/highmem.h>
3 #include <linux/sched.h>
4 #include <linux/hugetlb.h>
6 static int walk_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
12 pte = pte_offset_map(pmd, addr);
14 err = walk->pte_entry(pte, addr, addr + PAGE_SIZE, walk);
27 static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
34 pmd = pmd_offset(pud, addr);
37 next = pmd_addr_end(addr, end);
38 if (pmd_none(*pmd) || !walk->vma) {
40 err = walk->pte_hole(addr, next, walk);
46 * This implies that each ->pmd_entry() handler
47 * needs to know about pmd_trans_huge() pmds
50 err = walk->pmd_entry(pmd, addr, next, walk);
55 * Check this here so we only break down trans_huge
56 * pages when we _need_ to
61 split_huge_page_pmd_mm(walk->mm, addr, pmd);
62 if (pmd_trans_unstable(pmd))
64 err = walk_pte_range(pmd, addr, next, walk);
67 } while (pmd++, addr = next, addr != end);
72 static int walk_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end,
79 pud = pud_offset(pgd, addr);
81 next = pud_addr_end(addr, end);
82 if (pud_none_or_clear_bad(pud)) {
84 err = walk->pte_hole(addr, next, walk);
89 if (walk->pmd_entry || walk->pte_entry)
90 err = walk_pmd_range(pud, addr, next, walk);
93 } while (pud++, addr = next, addr != end);
98 static int walk_pgd_range(unsigned long addr, unsigned long end,
105 pgd = pgd_offset(walk->mm, addr);
107 next = pgd_addr_end(addr, end);
108 if (pgd_none_or_clear_bad(pgd)) {
110 err = walk->pte_hole(addr, next, walk);
115 if (walk->pmd_entry || walk->pte_entry)
116 err = walk_pud_range(pgd, addr, next, walk);
119 } while (pgd++, addr = next, addr != end);
124 #ifdef CONFIG_HUGETLB_PAGE
125 static unsigned long hugetlb_entry_end(struct hstate *h, unsigned long addr,
128 unsigned long boundary = (addr & huge_page_mask(h)) + huge_page_size(h);
129 return boundary < end ? boundary : end;
132 static int walk_hugetlb_range(unsigned long addr, unsigned long end,
133 struct mm_walk *walk)
135 struct vm_area_struct *vma = walk->vma;
136 struct hstate *h = hstate_vma(vma);
138 unsigned long hmask = huge_page_mask(h);
143 next = hugetlb_entry_end(h, addr, end);
144 pte = huge_pte_offset(walk->mm, addr & hmask);
145 if (pte && walk->hugetlb_entry)
146 err = walk->hugetlb_entry(pte, hmask, addr, next, walk);
149 } while (addr = next, addr != end);
154 #else /* CONFIG_HUGETLB_PAGE */
155 static int walk_hugetlb_range(unsigned long addr, unsigned long end,
156 struct mm_walk *walk)
161 #endif /* CONFIG_HUGETLB_PAGE */
164 * Decide whether we really walk over the current vma on [@start, @end)
165 * or skip it via the returned value. Return 0 if we do walk over the
166 * current vma, and return 1 if we skip the vma. Negative values means
167 * error, where we abort the current walk.
169 static int walk_page_test(unsigned long start, unsigned long end,
170 struct mm_walk *walk)
172 struct vm_area_struct *vma = walk->vma;
175 return walk->test_walk(start, end, walk);
178 * vma(VM_PFNMAP) doesn't have any valid struct pages behind VM_PFNMAP
179 * range, so we don't walk over it as we do for normal vmas. However,
180 * Some callers are interested in handling hole range and they don't
181 * want to just ignore any single address range. Such users certainly
182 * define their ->pte_hole() callbacks, so let's delegate them to handle
185 if (vma->vm_flags & VM_PFNMAP) {
188 err = walk->pte_hole(start, end, walk);
189 return err ? err : 1;
194 static int __walk_page_range(unsigned long start, unsigned long end,
195 struct mm_walk *walk)
198 struct vm_area_struct *vma = walk->vma;
200 if (vma && is_vm_hugetlb_page(vma)) {
201 if (walk->hugetlb_entry)
202 err = walk_hugetlb_range(start, end, walk);
204 err = walk_pgd_range(start, end, walk);
210 * walk_page_range - walk page table with caller specific callbacks
212 * Recursively walk the page table tree of the process represented by @walk->mm
213 * within the virtual address range [@start, @end). During walking, we can do
214 * some caller-specific works for each entry, by setting up pmd_entry(),
215 * pte_entry(), and/or hugetlb_entry(). If you don't set up for some of these
216 * callbacks, the associated entries/pages are just ignored.
217 * The return values of these callbacks are commonly defined like below:
218 * - 0 : succeeded to handle the current entry, and if you don't reach the
219 * end address yet, continue to walk.
220 * - >0 : succeeded to handle the current entry, and return to the caller
221 * with caller specific value.
222 * - <0 : failed to handle the current entry, and return to the caller
225 * Before starting to walk page table, some callers want to check whether
226 * they really want to walk over the current vma, typically by checking
227 * its vm_flags. walk_page_test() and @walk->test_walk() are used for this
230 * struct mm_walk keeps current values of some common data like vma and pmd,
231 * which are useful for the access from callbacks. If you want to pass some
232 * caller-specific data to callbacks, @walk->private should be helpful.
235 * Callers of walk_page_range() and walk_page_vma() should hold
236 * @walk->mm->mmap_sem, because these function traverse vma list and/or
237 * access to vma's data.
239 int walk_page_range(unsigned long start, unsigned long end,
240 struct mm_walk *walk)
244 struct vm_area_struct *vma;
252 VM_BUG_ON_MM(!rwsem_is_locked(&walk->mm->mmap_sem), walk->mm);
254 vma = find_vma(walk->mm, start);
256 if (!vma) { /* after the last vma */
259 } else if (start < vma->vm_start) { /* outside vma */
261 next = min(end, vma->vm_start);
262 } else { /* inside vma */
264 next = min(end, vma->vm_end);
267 err = walk_page_test(start, next, walk);
273 if (walk->vma || walk->pte_hole)
274 err = __walk_page_range(start, next, walk);
277 } while (start = next, start < end);
281 int walk_page_vma(struct vm_area_struct *vma, struct mm_walk *walk)
288 VM_BUG_ON(!rwsem_is_locked(&walk->mm->mmap_sem));
291 err = walk_page_test(vma->vm_start, vma->vm_end, walk);
296 return __walk_page_range(vma->vm_start, vma->vm_end, walk);