1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/highmem.h>
4 #include <linux/sched.h>
5 #include <linux/hugetlb.h>
7 static int walk_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
13 pte = pte_offset_map(pmd, addr);
15 err = walk->pte_entry(pte, addr, addr + PAGE_SIZE, walk);
28 static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
35 pmd = pmd_offset(pud, addr);
38 next = pmd_addr_end(addr, end);
39 if (pmd_none(*pmd) || !walk->vma) {
41 err = walk->pte_hole(addr, next, walk);
47 * This implies that each ->pmd_entry() handler
48 * needs to know about pmd_trans_huge() pmds
51 err = walk->pmd_entry(pmd, addr, next, walk);
56 * Check this here so we only break down trans_huge
57 * pages when we _need_ to
62 split_huge_pmd(walk->vma, pmd, addr);
63 if (pmd_trans_unstable(pmd))
65 err = walk_pte_range(pmd, addr, next, walk);
68 } while (pmd++, addr = next, addr != end);
73 static int walk_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end,
80 pud = pud_offset(p4d, addr);
83 next = pud_addr_end(addr, end);
84 if (pud_none(*pud) || !walk->vma) {
86 err = walk->pte_hole(addr, next, walk);
92 if (walk->pud_entry) {
93 spinlock_t *ptl = pud_trans_huge_lock(pud, walk->vma);
96 err = walk->pud_entry(pud, addr, next, walk);
104 split_huge_pud(walk->vma, pud, addr);
108 if (walk->pmd_entry || walk->pte_entry)
109 err = walk_pmd_range(pud, addr, next, walk);
112 } while (pud++, addr = next, addr != end);
117 static int walk_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end,
118 struct mm_walk *walk)
124 p4d = p4d_offset(pgd, addr);
126 next = p4d_addr_end(addr, end);
127 if (p4d_none_or_clear_bad(p4d)) {
129 err = walk->pte_hole(addr, next, walk);
134 if (walk->pmd_entry || walk->pte_entry)
135 err = walk_pud_range(p4d, addr, next, walk);
138 } while (p4d++, addr = next, addr != end);
143 static int walk_pgd_range(unsigned long addr, unsigned long end,
144 struct mm_walk *walk)
150 pgd = pgd_offset(walk->mm, addr);
152 next = pgd_addr_end(addr, end);
153 if (pgd_none_or_clear_bad(pgd)) {
155 err = walk->pte_hole(addr, next, walk);
160 if (walk->pmd_entry || walk->pte_entry)
161 err = walk_p4d_range(pgd, addr, next, walk);
164 } while (pgd++, addr = next, addr != end);
169 #ifdef CONFIG_HUGETLB_PAGE
170 static unsigned long hugetlb_entry_end(struct hstate *h, unsigned long addr,
173 unsigned long boundary = (addr & huge_page_mask(h)) + huge_page_size(h);
174 return boundary < end ? boundary : end;
177 static int walk_hugetlb_range(unsigned long addr, unsigned long end,
178 struct mm_walk *walk)
180 struct vm_area_struct *vma = walk->vma;
181 struct hstate *h = hstate_vma(vma);
183 unsigned long hmask = huge_page_mask(h);
184 unsigned long sz = huge_page_size(h);
189 next = hugetlb_entry_end(h, addr, end);
190 pte = huge_pte_offset(walk->mm, addr & hmask, sz);
191 if (pte && walk->hugetlb_entry)
192 err = walk->hugetlb_entry(pte, hmask, addr, next, walk);
195 } while (addr = next, addr != end);
200 #else /* CONFIG_HUGETLB_PAGE */
201 static int walk_hugetlb_range(unsigned long addr, unsigned long end,
202 struct mm_walk *walk)
207 #endif /* CONFIG_HUGETLB_PAGE */
210 * Decide whether we really walk over the current vma on [@start, @end)
211 * or skip it via the returned value. Return 0 if we do walk over the
212 * current vma, and return 1 if we skip the vma. Negative values means
213 * error, where we abort the current walk.
215 static int walk_page_test(unsigned long start, unsigned long end,
216 struct mm_walk *walk)
218 struct vm_area_struct *vma = walk->vma;
221 return walk->test_walk(start, end, walk);
224 * vma(VM_PFNMAP) doesn't have any valid struct pages behind VM_PFNMAP
225 * range, so we don't walk over it as we do for normal vmas. However,
226 * Some callers are interested in handling hole range and they don't
227 * want to just ignore any single address range. Such users certainly
228 * define their ->pte_hole() callbacks, so let's delegate them to handle
231 if (vma->vm_flags & VM_PFNMAP) {
234 err = walk->pte_hole(start, end, walk);
235 return err ? err : 1;
240 static int __walk_page_range(unsigned long start, unsigned long end,
241 struct mm_walk *walk)
244 struct vm_area_struct *vma = walk->vma;
246 if (vma && is_vm_hugetlb_page(vma)) {
247 if (walk->hugetlb_entry)
248 err = walk_hugetlb_range(start, end, walk);
250 err = walk_pgd_range(start, end, walk);
256 * walk_page_range - walk page table with caller specific callbacks
258 * Recursively walk the page table tree of the process represented by @walk->mm
259 * within the virtual address range [@start, @end). During walking, we can do
260 * some caller-specific works for each entry, by setting up pmd_entry(),
261 * pte_entry(), and/or hugetlb_entry(). If you don't set up for some of these
262 * callbacks, the associated entries/pages are just ignored.
263 * The return values of these callbacks are commonly defined like below:
264 * - 0 : succeeded to handle the current entry, and if you don't reach the
265 * end address yet, continue to walk.
266 * - >0 : succeeded to handle the current entry, and return to the caller
267 * with caller specific value.
268 * - <0 : failed to handle the current entry, and return to the caller
271 * Before starting to walk page table, some callers want to check whether
272 * they really want to walk over the current vma, typically by checking
273 * its vm_flags. walk_page_test() and @walk->test_walk() are used for this
276 * struct mm_walk keeps current values of some common data like vma and pmd,
277 * which are useful for the access from callbacks. If you want to pass some
278 * caller-specific data to callbacks, @walk->private should be helpful.
281 * Callers of walk_page_range() and walk_page_vma() should hold
282 * @walk->mm->mmap_sem, because these function traverse vma list and/or
283 * access to vma's data.
285 int walk_page_range(unsigned long start, unsigned long end,
286 struct mm_walk *walk)
290 struct vm_area_struct *vma;
298 VM_BUG_ON_MM(!rwsem_is_locked(&walk->mm->mmap_sem), walk->mm);
300 vma = find_vma(walk->mm, start);
302 if (!vma) { /* after the last vma */
305 } else if (start < vma->vm_start) { /* outside vma */
307 next = min(end, vma->vm_start);
308 } else { /* inside vma */
310 next = min(end, vma->vm_end);
313 err = walk_page_test(start, next, walk);
316 * positive return values are purely for
317 * controlling the pagewalk, so should never
318 * be passed to the callers.
326 if (walk->vma || walk->pte_hole)
327 err = __walk_page_range(start, next, walk);
330 } while (start = next, start < end);
334 int walk_page_vma(struct vm_area_struct *vma, struct mm_walk *walk)
341 VM_BUG_ON(!rwsem_is_locked(&walk->mm->mmap_sem));
344 err = walk_page_test(vma->vm_start, vma->vm_end, walk);
349 return __walk_page_range(vma->vm_start, vma->vm_end, walk);