]> git.ipfire.org Git - thirdparty/linux.git/blob - include/asm-generic/pgtable.h
mm: enforce that vmap can't map pages executable
[thirdparty/linux.git] / include / asm-generic / pgtable.h
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_GENERIC_PGTABLE_H
3 #define _ASM_GENERIC_PGTABLE_H
4
5 #include <linux/pfn.h>
6
7 #ifndef __ASSEMBLY__
8 #ifdef CONFIG_MMU
9
10 #include <linux/mm_types.h>
11 #include <linux/bug.h>
12 #include <linux/errno.h>
13 #include <asm-generic/pgtable_uffd.h>
14
15 #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
16 defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
17 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
18 #endif
19
20 /*
21 * On almost all architectures and configurations, 0 can be used as the
22 * upper ceiling to free_pgtables(): on many architectures it has the same
23 * effect as using TASK_SIZE. However, there is one configuration which
24 * must impose a more careful limit, to avoid freeing kernel pgtables.
25 */
26 #ifndef USER_PGTABLES_CEILING
27 #define USER_PGTABLES_CEILING 0UL
28 #endif
29
30 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
31 extern int ptep_set_access_flags(struct vm_area_struct *vma,
32 unsigned long address, pte_t *ptep,
33 pte_t entry, int dirty);
34 #endif
35
36 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
37 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
38 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
39 unsigned long address, pmd_t *pmdp,
40 pmd_t entry, int dirty);
41 extern int pudp_set_access_flags(struct vm_area_struct *vma,
42 unsigned long address, pud_t *pudp,
43 pud_t entry, int dirty);
44 #else
45 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
46 unsigned long address, pmd_t *pmdp,
47 pmd_t entry, int dirty)
48 {
49 BUILD_BUG();
50 return 0;
51 }
52 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
53 unsigned long address, pud_t *pudp,
54 pud_t entry, int dirty)
55 {
56 BUILD_BUG();
57 return 0;
58 }
59 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
60 #endif
61
62 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
63 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
64 unsigned long address,
65 pte_t *ptep)
66 {
67 pte_t pte = *ptep;
68 int r = 1;
69 if (!pte_young(pte))
70 r = 0;
71 else
72 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
73 return r;
74 }
75 #endif
76
77 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
78 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
79 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
80 unsigned long address,
81 pmd_t *pmdp)
82 {
83 pmd_t pmd = *pmdp;
84 int r = 1;
85 if (!pmd_young(pmd))
86 r = 0;
87 else
88 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
89 return r;
90 }
91 #else
92 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
93 unsigned long address,
94 pmd_t *pmdp)
95 {
96 BUILD_BUG();
97 return 0;
98 }
99 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
100 #endif
101
102 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
103 int ptep_clear_flush_young(struct vm_area_struct *vma,
104 unsigned long address, pte_t *ptep);
105 #endif
106
107 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
108 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
109 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
110 unsigned long address, pmd_t *pmdp);
111 #else
112 /*
113 * Despite relevant to THP only, this API is called from generic rmap code
114 * under PageTransHuge(), hence needs a dummy implementation for !THP
115 */
116 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
117 unsigned long address, pmd_t *pmdp)
118 {
119 BUILD_BUG();
120 return 0;
121 }
122 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
123 #endif
124
125 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
126 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
127 unsigned long address,
128 pte_t *ptep)
129 {
130 pte_t pte = *ptep;
131 pte_clear(mm, address, ptep);
132 return pte;
133 }
134 #endif
135
136 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
137 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
138 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
139 unsigned long address,
140 pmd_t *pmdp)
141 {
142 pmd_t pmd = *pmdp;
143 pmd_clear(pmdp);
144 return pmd;
145 }
146 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
147 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
148 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
149 unsigned long address,
150 pud_t *pudp)
151 {
152 pud_t pud = *pudp;
153
154 pud_clear(pudp);
155 return pud;
156 }
157 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
158 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
159
160 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
161 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
162 static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
163 unsigned long address, pmd_t *pmdp,
164 int full)
165 {
166 return pmdp_huge_get_and_clear(mm, address, pmdp);
167 }
168 #endif
169
170 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
171 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
172 unsigned long address, pud_t *pudp,
173 int full)
174 {
175 return pudp_huge_get_and_clear(mm, address, pudp);
176 }
177 #endif
178 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
179
180 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
181 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
182 unsigned long address, pte_t *ptep,
183 int full)
184 {
185 pte_t pte;
186 pte = ptep_get_and_clear(mm, address, ptep);
187 return pte;
188 }
189 #endif
190
191 /*
192 * Some architectures may be able to avoid expensive synchronization
193 * primitives when modifications are made to PTE's which are already
194 * not present, or in the process of an address space destruction.
195 */
196 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
197 static inline void pte_clear_not_present_full(struct mm_struct *mm,
198 unsigned long address,
199 pte_t *ptep,
200 int full)
201 {
202 pte_clear(mm, address, ptep);
203 }
204 #endif
205
206 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
207 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
208 unsigned long address,
209 pte_t *ptep);
210 #endif
211
212 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
213 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
214 unsigned long address,
215 pmd_t *pmdp);
216 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
217 unsigned long address,
218 pud_t *pudp);
219 #endif
220
221 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
222 struct mm_struct;
223 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
224 {
225 pte_t old_pte = *ptep;
226 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
227 }
228 #endif
229
230 #ifndef pte_savedwrite
231 #define pte_savedwrite pte_write
232 #endif
233
234 #ifndef pte_mk_savedwrite
235 #define pte_mk_savedwrite pte_mkwrite
236 #endif
237
238 #ifndef pte_clear_savedwrite
239 #define pte_clear_savedwrite pte_wrprotect
240 #endif
241
242 #ifndef pmd_savedwrite
243 #define pmd_savedwrite pmd_write
244 #endif
245
246 #ifndef pmd_mk_savedwrite
247 #define pmd_mk_savedwrite pmd_mkwrite
248 #endif
249
250 #ifndef pmd_clear_savedwrite
251 #define pmd_clear_savedwrite pmd_wrprotect
252 #endif
253
254 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
255 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
256 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
257 unsigned long address, pmd_t *pmdp)
258 {
259 pmd_t old_pmd = *pmdp;
260 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
261 }
262 #else
263 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
264 unsigned long address, pmd_t *pmdp)
265 {
266 BUILD_BUG();
267 }
268 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
269 #endif
270 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
271 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
272 static inline void pudp_set_wrprotect(struct mm_struct *mm,
273 unsigned long address, pud_t *pudp)
274 {
275 pud_t old_pud = *pudp;
276
277 set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
278 }
279 #else
280 static inline void pudp_set_wrprotect(struct mm_struct *mm,
281 unsigned long address, pud_t *pudp)
282 {
283 BUILD_BUG();
284 }
285 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
286 #endif
287
288 #ifndef pmdp_collapse_flush
289 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
290 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
291 unsigned long address, pmd_t *pmdp);
292 #else
293 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
294 unsigned long address,
295 pmd_t *pmdp)
296 {
297 BUILD_BUG();
298 return *pmdp;
299 }
300 #define pmdp_collapse_flush pmdp_collapse_flush
301 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
302 #endif
303
304 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
305 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
306 pgtable_t pgtable);
307 #endif
308
309 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
310 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
311 #endif
312
313 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
314 /*
315 * This is an implementation of pmdp_establish() that is only suitable for an
316 * architecture that doesn't have hardware dirty/accessed bits. In this case we
317 * can't race with CPU which sets these bits and non-atomic aproach is fine.
318 */
319 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
320 unsigned long address, pmd_t *pmdp, pmd_t pmd)
321 {
322 pmd_t old_pmd = *pmdp;
323 set_pmd_at(vma->vm_mm, address, pmdp, pmd);
324 return old_pmd;
325 }
326 #endif
327
328 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
329 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
330 pmd_t *pmdp);
331 #endif
332
333 #ifndef __HAVE_ARCH_PTE_SAME
334 static inline int pte_same(pte_t pte_a, pte_t pte_b)
335 {
336 return pte_val(pte_a) == pte_val(pte_b);
337 }
338 #endif
339
340 #ifndef __HAVE_ARCH_PTE_UNUSED
341 /*
342 * Some architectures provide facilities to virtualization guests
343 * so that they can flag allocated pages as unused. This allows the
344 * host to transparently reclaim unused pages. This function returns
345 * whether the pte's page is unused.
346 */
347 static inline int pte_unused(pte_t pte)
348 {
349 return 0;
350 }
351 #endif
352
353 #ifndef pte_access_permitted
354 #define pte_access_permitted(pte, write) \
355 (pte_present(pte) && (!(write) || pte_write(pte)))
356 #endif
357
358 #ifndef pmd_access_permitted
359 #define pmd_access_permitted(pmd, write) \
360 (pmd_present(pmd) && (!(write) || pmd_write(pmd)))
361 #endif
362
363 #ifndef pud_access_permitted
364 #define pud_access_permitted(pud, write) \
365 (pud_present(pud) && (!(write) || pud_write(pud)))
366 #endif
367
368 #ifndef p4d_access_permitted
369 #define p4d_access_permitted(p4d, write) \
370 (p4d_present(p4d) && (!(write) || p4d_write(p4d)))
371 #endif
372
373 #ifndef pgd_access_permitted
374 #define pgd_access_permitted(pgd, write) \
375 (pgd_present(pgd) && (!(write) || pgd_write(pgd)))
376 #endif
377
378 #ifndef __HAVE_ARCH_PMD_SAME
379 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
380 {
381 return pmd_val(pmd_a) == pmd_val(pmd_b);
382 }
383
384 static inline int pud_same(pud_t pud_a, pud_t pud_b)
385 {
386 return pud_val(pud_a) == pud_val(pud_b);
387 }
388 #endif
389
390 #ifndef __HAVE_ARCH_P4D_SAME
391 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
392 {
393 return p4d_val(p4d_a) == p4d_val(p4d_b);
394 }
395 #endif
396
397 #ifndef __HAVE_ARCH_PGD_SAME
398 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
399 {
400 return pgd_val(pgd_a) == pgd_val(pgd_b);
401 }
402 #endif
403
404 /*
405 * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
406 * TLB flush will be required as a result of the "set". For example, use
407 * in scenarios where it is known ahead of time that the routine is
408 * setting non-present entries, or re-setting an existing entry to the
409 * same value. Otherwise, use the typical "set" helpers and flush the
410 * TLB.
411 */
412 #define set_pte_safe(ptep, pte) \
413 ({ \
414 WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
415 set_pte(ptep, pte); \
416 })
417
418 #define set_pmd_safe(pmdp, pmd) \
419 ({ \
420 WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
421 set_pmd(pmdp, pmd); \
422 })
423
424 #define set_pud_safe(pudp, pud) \
425 ({ \
426 WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
427 set_pud(pudp, pud); \
428 })
429
430 #define set_p4d_safe(p4dp, p4d) \
431 ({ \
432 WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
433 set_p4d(p4dp, p4d); \
434 })
435
436 #define set_pgd_safe(pgdp, pgd) \
437 ({ \
438 WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
439 set_pgd(pgdp, pgd); \
440 })
441
442 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
443 /*
444 * Some architectures support metadata associated with a page. When a
445 * page is being swapped out, this metadata must be saved so it can be
446 * restored when the page is swapped back in. SPARC M7 and newer
447 * processors support an ADI (Application Data Integrity) tag for the
448 * page as metadata for the page. arch_do_swap_page() can restore this
449 * metadata when a page is swapped back in.
450 */
451 static inline void arch_do_swap_page(struct mm_struct *mm,
452 struct vm_area_struct *vma,
453 unsigned long addr,
454 pte_t pte, pte_t oldpte)
455 {
456
457 }
458 #endif
459
460 #ifndef __HAVE_ARCH_UNMAP_ONE
461 /*
462 * Some architectures support metadata associated with a page. When a
463 * page is being swapped out, this metadata must be saved so it can be
464 * restored when the page is swapped back in. SPARC M7 and newer
465 * processors support an ADI (Application Data Integrity) tag for the
466 * page as metadata for the page. arch_unmap_one() can save this
467 * metadata on a swap-out of a page.
468 */
469 static inline int arch_unmap_one(struct mm_struct *mm,
470 struct vm_area_struct *vma,
471 unsigned long addr,
472 pte_t orig_pte)
473 {
474 return 0;
475 }
476 #endif
477
478 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
479 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
480 #endif
481
482 #ifndef __HAVE_ARCH_MOVE_PTE
483 #define move_pte(pte, prot, old_addr, new_addr) (pte)
484 #endif
485
486 #ifndef pte_accessible
487 # define pte_accessible(mm, pte) ((void)(pte), 1)
488 #endif
489
490 #ifndef flush_tlb_fix_spurious_fault
491 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
492 #endif
493
494 #ifndef pgprot_nx
495 #define pgprot_nx(prot) (prot)
496 #endif
497
498 #ifndef pgprot_noncached
499 #define pgprot_noncached(prot) (prot)
500 #endif
501
502 #ifndef pgprot_writecombine
503 #define pgprot_writecombine pgprot_noncached
504 #endif
505
506 #ifndef pgprot_writethrough
507 #define pgprot_writethrough pgprot_noncached
508 #endif
509
510 #ifndef pgprot_device
511 #define pgprot_device pgprot_noncached
512 #endif
513
514 #ifndef pgprot_modify
515 #define pgprot_modify pgprot_modify
516 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
517 {
518 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
519 newprot = pgprot_noncached(newprot);
520 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
521 newprot = pgprot_writecombine(newprot);
522 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
523 newprot = pgprot_device(newprot);
524 return newprot;
525 }
526 #endif
527
528 /*
529 * When walking page tables, get the address of the next boundary,
530 * or the end address of the range if that comes earlier. Although no
531 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
532 */
533
534 #define pgd_addr_end(addr, end) \
535 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
536 (__boundary - 1 < (end) - 1)? __boundary: (end); \
537 })
538
539 #ifndef p4d_addr_end
540 #define p4d_addr_end(addr, end) \
541 ({ unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK; \
542 (__boundary - 1 < (end) - 1)? __boundary: (end); \
543 })
544 #endif
545
546 #ifndef pud_addr_end
547 #define pud_addr_end(addr, end) \
548 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
549 (__boundary - 1 < (end) - 1)? __boundary: (end); \
550 })
551 #endif
552
553 #ifndef pmd_addr_end
554 #define pmd_addr_end(addr, end) \
555 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
556 (__boundary - 1 < (end) - 1)? __boundary: (end); \
557 })
558 #endif
559
560 /*
561 * When walking page tables, we usually want to skip any p?d_none entries;
562 * and any p?d_bad entries - reporting the error before resetting to none.
563 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
564 */
565 void pgd_clear_bad(pgd_t *);
566
567 #ifndef __PAGETABLE_P4D_FOLDED
568 void p4d_clear_bad(p4d_t *);
569 #else
570 #define p4d_clear_bad(p4d) do { } while (0)
571 #endif
572
573 #ifndef __PAGETABLE_PUD_FOLDED
574 void pud_clear_bad(pud_t *);
575 #else
576 #define pud_clear_bad(p4d) do { } while (0)
577 #endif
578
579 void pmd_clear_bad(pmd_t *);
580
581 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
582 {
583 if (pgd_none(*pgd))
584 return 1;
585 if (unlikely(pgd_bad(*pgd))) {
586 pgd_clear_bad(pgd);
587 return 1;
588 }
589 return 0;
590 }
591
592 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
593 {
594 if (p4d_none(*p4d))
595 return 1;
596 if (unlikely(p4d_bad(*p4d))) {
597 p4d_clear_bad(p4d);
598 return 1;
599 }
600 return 0;
601 }
602
603 static inline int pud_none_or_clear_bad(pud_t *pud)
604 {
605 if (pud_none(*pud))
606 return 1;
607 if (unlikely(pud_bad(*pud))) {
608 pud_clear_bad(pud);
609 return 1;
610 }
611 return 0;
612 }
613
614 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
615 {
616 if (pmd_none(*pmd))
617 return 1;
618 if (unlikely(pmd_bad(*pmd))) {
619 pmd_clear_bad(pmd);
620 return 1;
621 }
622 return 0;
623 }
624
625 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
626 unsigned long addr,
627 pte_t *ptep)
628 {
629 /*
630 * Get the current pte state, but zero it out to make it
631 * non-present, preventing the hardware from asynchronously
632 * updating it.
633 */
634 return ptep_get_and_clear(vma->vm_mm, addr, ptep);
635 }
636
637 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
638 unsigned long addr,
639 pte_t *ptep, pte_t pte)
640 {
641 /*
642 * The pte is non-present, so there's no hardware state to
643 * preserve.
644 */
645 set_pte_at(vma->vm_mm, addr, ptep, pte);
646 }
647
648 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
649 /*
650 * Start a pte protection read-modify-write transaction, which
651 * protects against asynchronous hardware modifications to the pte.
652 * The intention is not to prevent the hardware from making pte
653 * updates, but to prevent any updates it may make from being lost.
654 *
655 * This does not protect against other software modifications of the
656 * pte; the appropriate pte lock must be held over the transation.
657 *
658 * Note that this interface is intended to be batchable, meaning that
659 * ptep_modify_prot_commit may not actually update the pte, but merely
660 * queue the update to be done at some later time. The update must be
661 * actually committed before the pte lock is released, however.
662 */
663 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
664 unsigned long addr,
665 pte_t *ptep)
666 {
667 return __ptep_modify_prot_start(vma, addr, ptep);
668 }
669
670 /*
671 * Commit an update to a pte, leaving any hardware-controlled bits in
672 * the PTE unmodified.
673 */
674 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
675 unsigned long addr,
676 pte_t *ptep, pte_t old_pte, pte_t pte)
677 {
678 __ptep_modify_prot_commit(vma, addr, ptep, pte);
679 }
680 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
681 #endif /* CONFIG_MMU */
682
683 /*
684 * No-op macros that just return the current protection value. Defined here
685 * because these macros can be used used even if CONFIG_MMU is not defined.
686 */
687 #ifndef pgprot_encrypted
688 #define pgprot_encrypted(prot) (prot)
689 #endif
690
691 #ifndef pgprot_decrypted
692 #define pgprot_decrypted(prot) (prot)
693 #endif
694
695 /*
696 * A facility to provide lazy MMU batching. This allows PTE updates and
697 * page invalidations to be delayed until a call to leave lazy MMU mode
698 * is issued. Some architectures may benefit from doing this, and it is
699 * beneficial for both shadow and direct mode hypervisors, which may batch
700 * the PTE updates which happen during this window. Note that using this
701 * interface requires that read hazards be removed from the code. A read
702 * hazard could result in the direct mode hypervisor case, since the actual
703 * write to the page tables may not yet have taken place, so reads though
704 * a raw PTE pointer after it has been modified are not guaranteed to be
705 * up to date. This mode can only be entered and left under the protection of
706 * the page table locks for all page tables which may be modified. In the UP
707 * case, this is required so that preemption is disabled, and in the SMP case,
708 * it must synchronize the delayed page table writes properly on other CPUs.
709 */
710 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
711 #define arch_enter_lazy_mmu_mode() do {} while (0)
712 #define arch_leave_lazy_mmu_mode() do {} while (0)
713 #define arch_flush_lazy_mmu_mode() do {} while (0)
714 #endif
715
716 /*
717 * A facility to provide batching of the reload of page tables and
718 * other process state with the actual context switch code for
719 * paravirtualized guests. By convention, only one of the batched
720 * update (lazy) modes (CPU, MMU) should be active at any given time,
721 * entry should never be nested, and entry and exits should always be
722 * paired. This is for sanity of maintaining and reasoning about the
723 * kernel code. In this case, the exit (end of the context switch) is
724 * in architecture-specific code, and so doesn't need a generic
725 * definition.
726 */
727 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
728 #define arch_start_context_switch(prev) do {} while (0)
729 #endif
730
731 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
732 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
733 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
734 {
735 return pmd;
736 }
737
738 static inline int pmd_swp_soft_dirty(pmd_t pmd)
739 {
740 return 0;
741 }
742
743 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
744 {
745 return pmd;
746 }
747 #endif
748 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
749 static inline int pte_soft_dirty(pte_t pte)
750 {
751 return 0;
752 }
753
754 static inline int pmd_soft_dirty(pmd_t pmd)
755 {
756 return 0;
757 }
758
759 static inline pte_t pte_mksoft_dirty(pte_t pte)
760 {
761 return pte;
762 }
763
764 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
765 {
766 return pmd;
767 }
768
769 static inline pte_t pte_clear_soft_dirty(pte_t pte)
770 {
771 return pte;
772 }
773
774 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
775 {
776 return pmd;
777 }
778
779 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
780 {
781 return pte;
782 }
783
784 static inline int pte_swp_soft_dirty(pte_t pte)
785 {
786 return 0;
787 }
788
789 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
790 {
791 return pte;
792 }
793
794 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
795 {
796 return pmd;
797 }
798
799 static inline int pmd_swp_soft_dirty(pmd_t pmd)
800 {
801 return 0;
802 }
803
804 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
805 {
806 return pmd;
807 }
808 #endif
809
810 #ifndef __HAVE_PFNMAP_TRACKING
811 /*
812 * Interfaces that can be used by architecture code to keep track of
813 * memory type of pfn mappings specified by the remap_pfn_range,
814 * vmf_insert_pfn.
815 */
816
817 /*
818 * track_pfn_remap is called when a _new_ pfn mapping is being established
819 * by remap_pfn_range() for physical range indicated by pfn and size.
820 */
821 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
822 unsigned long pfn, unsigned long addr,
823 unsigned long size)
824 {
825 return 0;
826 }
827
828 /*
829 * track_pfn_insert is called when a _new_ single pfn is established
830 * by vmf_insert_pfn().
831 */
832 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
833 pfn_t pfn)
834 {
835 }
836
837 /*
838 * track_pfn_copy is called when vma that is covering the pfnmap gets
839 * copied through copy_page_range().
840 */
841 static inline int track_pfn_copy(struct vm_area_struct *vma)
842 {
843 return 0;
844 }
845
846 /*
847 * untrack_pfn is called while unmapping a pfnmap for a region.
848 * untrack can be called for a specific region indicated by pfn and size or
849 * can be for the entire vma (in which case pfn, size are zero).
850 */
851 static inline void untrack_pfn(struct vm_area_struct *vma,
852 unsigned long pfn, unsigned long size)
853 {
854 }
855
856 /*
857 * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
858 */
859 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
860 {
861 }
862 #else
863 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
864 unsigned long pfn, unsigned long addr,
865 unsigned long size);
866 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
867 pfn_t pfn);
868 extern int track_pfn_copy(struct vm_area_struct *vma);
869 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
870 unsigned long size);
871 extern void untrack_pfn_moved(struct vm_area_struct *vma);
872 #endif
873
874 #ifdef __HAVE_COLOR_ZERO_PAGE
875 static inline int is_zero_pfn(unsigned long pfn)
876 {
877 extern unsigned long zero_pfn;
878 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
879 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
880 }
881
882 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
883
884 #else
885 static inline int is_zero_pfn(unsigned long pfn)
886 {
887 extern unsigned long zero_pfn;
888 return pfn == zero_pfn;
889 }
890
891 static inline unsigned long my_zero_pfn(unsigned long addr)
892 {
893 extern unsigned long zero_pfn;
894 return zero_pfn;
895 }
896 #endif
897
898 #ifdef CONFIG_MMU
899
900 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
901 static inline int pmd_trans_huge(pmd_t pmd)
902 {
903 return 0;
904 }
905 #ifndef pmd_write
906 static inline int pmd_write(pmd_t pmd)
907 {
908 BUG();
909 return 0;
910 }
911 #endif /* pmd_write */
912 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
913
914 #ifndef pud_write
915 static inline int pud_write(pud_t pud)
916 {
917 BUG();
918 return 0;
919 }
920 #endif /* pud_write */
921
922 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
923 static inline int pmd_devmap(pmd_t pmd)
924 {
925 return 0;
926 }
927 static inline int pud_devmap(pud_t pud)
928 {
929 return 0;
930 }
931 static inline int pgd_devmap(pgd_t pgd)
932 {
933 return 0;
934 }
935 #endif
936
937 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
938 (defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
939 !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
940 static inline int pud_trans_huge(pud_t pud)
941 {
942 return 0;
943 }
944 #endif
945
946 /* See pmd_none_or_trans_huge_or_clear_bad for discussion. */
947 static inline int pud_none_or_trans_huge_or_dev_or_clear_bad(pud_t *pud)
948 {
949 pud_t pudval = READ_ONCE(*pud);
950
951 if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
952 return 1;
953 if (unlikely(pud_bad(pudval))) {
954 pud_clear_bad(pud);
955 return 1;
956 }
957 return 0;
958 }
959
960 /* See pmd_trans_unstable for discussion. */
961 static inline int pud_trans_unstable(pud_t *pud)
962 {
963 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
964 defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
965 return pud_none_or_trans_huge_or_dev_or_clear_bad(pud);
966 #else
967 return 0;
968 #endif
969 }
970
971 #ifndef pmd_read_atomic
972 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
973 {
974 /*
975 * Depend on compiler for an atomic pmd read. NOTE: this is
976 * only going to work, if the pmdval_t isn't larger than
977 * an unsigned long.
978 */
979 return *pmdp;
980 }
981 #endif
982
983 #ifndef arch_needs_pgtable_deposit
984 #define arch_needs_pgtable_deposit() (false)
985 #endif
986 /*
987 * This function is meant to be used by sites walking pagetables with
988 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
989 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
990 * into a null pmd and the transhuge page fault can convert a null pmd
991 * into an hugepmd or into a regular pmd (if the hugepage allocation
992 * fails). While holding the mmap_sem in read mode the pmd becomes
993 * stable and stops changing under us only if it's not null and not a
994 * transhuge pmd. When those races occurs and this function makes a
995 * difference vs the standard pmd_none_or_clear_bad, the result is
996 * undefined so behaving like if the pmd was none is safe (because it
997 * can return none anyway). The compiler level barrier() is critically
998 * important to compute the two checks atomically on the same pmdval.
999 *
1000 * For 32bit kernels with a 64bit large pmd_t this automatically takes
1001 * care of reading the pmd atomically to avoid SMP race conditions
1002 * against pmd_populate() when the mmap_sem is hold for reading by the
1003 * caller (a special atomic read not done by "gcc" as in the generic
1004 * version above, is also needed when THP is disabled because the page
1005 * fault can populate the pmd from under us).
1006 */
1007 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
1008 {
1009 pmd_t pmdval = pmd_read_atomic(pmd);
1010 /*
1011 * The barrier will stabilize the pmdval in a register or on
1012 * the stack so that it will stop changing under the code.
1013 *
1014 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
1015 * pmd_read_atomic is allowed to return a not atomic pmdval
1016 * (for example pointing to an hugepage that has never been
1017 * mapped in the pmd). The below checks will only care about
1018 * the low part of the pmd with 32bit PAE x86 anyway, with the
1019 * exception of pmd_none(). So the important thing is that if
1020 * the low part of the pmd is found null, the high part will
1021 * be also null or the pmd_none() check below would be
1022 * confused.
1023 */
1024 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1025 barrier();
1026 #endif
1027 /*
1028 * !pmd_present() checks for pmd migration entries
1029 *
1030 * The complete check uses is_pmd_migration_entry() in linux/swapops.h
1031 * But using that requires moving current function and pmd_trans_unstable()
1032 * to linux/swapops.h to resovle dependency, which is too much code move.
1033 *
1034 * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
1035 * because !pmd_present() pages can only be under migration not swapped
1036 * out.
1037 *
1038 * pmd_none() is preseved for future condition checks on pmd migration
1039 * entries and not confusing with this function name, although it is
1040 * redundant with !pmd_present().
1041 */
1042 if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
1043 (IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
1044 return 1;
1045 if (unlikely(pmd_bad(pmdval))) {
1046 pmd_clear_bad(pmd);
1047 return 1;
1048 }
1049 return 0;
1050 }
1051
1052 /*
1053 * This is a noop if Transparent Hugepage Support is not built into
1054 * the kernel. Otherwise it is equivalent to
1055 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
1056 * places that already verified the pmd is not none and they want to
1057 * walk ptes while holding the mmap sem in read mode (write mode don't
1058 * need this). If THP is not enabled, the pmd can't go away under the
1059 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
1060 * run a pmd_trans_unstable before walking the ptes after
1061 * split_huge_pmd returns (because it may have run when the pmd become
1062 * null, but then a page fault can map in a THP and not a regular page).
1063 */
1064 static inline int pmd_trans_unstable(pmd_t *pmd)
1065 {
1066 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1067 return pmd_none_or_trans_huge_or_clear_bad(pmd);
1068 #else
1069 return 0;
1070 #endif
1071 }
1072
1073 #ifndef CONFIG_NUMA_BALANCING
1074 /*
1075 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
1076 * the only case the kernel cares is for NUMA balancing and is only ever set
1077 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
1078 * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
1079 * is the responsibility of the caller to distinguish between PROT_NONE
1080 * protections and NUMA hinting fault protections.
1081 */
1082 static inline int pte_protnone(pte_t pte)
1083 {
1084 return 0;
1085 }
1086
1087 static inline int pmd_protnone(pmd_t pmd)
1088 {
1089 return 0;
1090 }
1091 #endif /* CONFIG_NUMA_BALANCING */
1092
1093 #endif /* CONFIG_MMU */
1094
1095 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1096
1097 #ifndef __PAGETABLE_P4D_FOLDED
1098 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1099 int p4d_clear_huge(p4d_t *p4d);
1100 #else
1101 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1102 {
1103 return 0;
1104 }
1105 static inline int p4d_clear_huge(p4d_t *p4d)
1106 {
1107 return 0;
1108 }
1109 #endif /* !__PAGETABLE_P4D_FOLDED */
1110
1111 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1112 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1113 int pud_clear_huge(pud_t *pud);
1114 int pmd_clear_huge(pmd_t *pmd);
1115 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1116 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1117 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1118 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1119 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1120 {
1121 return 0;
1122 }
1123 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1124 {
1125 return 0;
1126 }
1127 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1128 {
1129 return 0;
1130 }
1131 static inline int p4d_clear_huge(p4d_t *p4d)
1132 {
1133 return 0;
1134 }
1135 static inline int pud_clear_huge(pud_t *pud)
1136 {
1137 return 0;
1138 }
1139 static inline int pmd_clear_huge(pmd_t *pmd)
1140 {
1141 return 0;
1142 }
1143 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1144 {
1145 return 0;
1146 }
1147 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1148 {
1149 return 0;
1150 }
1151 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1152 {
1153 return 0;
1154 }
1155 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
1156
1157 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1158 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1159 /*
1160 * ARCHes with special requirements for evicting THP backing TLB entries can
1161 * implement this. Otherwise also, it can help optimize normal TLB flush in
1162 * THP regime. stock flush_tlb_range() typically has optimization to nuke the
1163 * entire TLB TLB if flush span is greater than a threshold, which will
1164 * likely be true for a single huge page. Thus a single thp flush will
1165 * invalidate the entire TLB which is not desitable.
1166 * e.g. see arch/arc: flush_pmd_tlb_range
1167 */
1168 #define flush_pmd_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1169 #define flush_pud_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1170 #else
1171 #define flush_pmd_tlb_range(vma, addr, end) BUILD_BUG()
1172 #define flush_pud_tlb_range(vma, addr, end) BUILD_BUG()
1173 #endif
1174 #endif
1175
1176 struct file;
1177 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1178 unsigned long size, pgprot_t *vma_prot);
1179
1180 #ifndef CONFIG_X86_ESPFIX64
1181 static inline void init_espfix_bsp(void) { }
1182 #endif
1183
1184 extern void __init pgtable_cache_init(void);
1185
1186 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1187 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1188 {
1189 return true;
1190 }
1191
1192 static inline bool arch_has_pfn_modify_check(void)
1193 {
1194 return false;
1195 }
1196 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1197
1198 /*
1199 * Architecture PAGE_KERNEL_* fallbacks
1200 *
1201 * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1202 * because they really don't support them, or the port needs to be updated to
1203 * reflect the required functionality. Below are a set of relatively safe
1204 * fallbacks, as best effort, which we can count on in lieu of the architectures
1205 * not defining them on their own yet.
1206 */
1207
1208 #ifndef PAGE_KERNEL_RO
1209 # define PAGE_KERNEL_RO PAGE_KERNEL
1210 #endif
1211
1212 #ifndef PAGE_KERNEL_EXEC
1213 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1214 #endif
1215
1216 #endif /* !__ASSEMBLY__ */
1217
1218 #ifndef io_remap_pfn_range
1219 #define io_remap_pfn_range remap_pfn_range
1220 #endif
1221
1222 #ifndef has_transparent_hugepage
1223 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1224 #define has_transparent_hugepage() 1
1225 #else
1226 #define has_transparent_hugepage() 0
1227 #endif
1228 #endif
1229
1230 /*
1231 * On some architectures it depends on the mm if the p4d/pud or pmd
1232 * layer of the page table hierarchy is folded or not.
1233 */
1234 #ifndef mm_p4d_folded
1235 #define mm_p4d_folded(mm) __is_defined(__PAGETABLE_P4D_FOLDED)
1236 #endif
1237
1238 #ifndef mm_pud_folded
1239 #define mm_pud_folded(mm) __is_defined(__PAGETABLE_PUD_FOLDED)
1240 #endif
1241
1242 #ifndef mm_pmd_folded
1243 #define mm_pmd_folded(mm) __is_defined(__PAGETABLE_PMD_FOLDED)
1244 #endif
1245
1246 /*
1247 * p?d_leaf() - true if this entry is a final mapping to a physical address.
1248 * This differs from p?d_huge() by the fact that they are always available (if
1249 * the architecture supports large pages at the appropriate level) even
1250 * if CONFIG_HUGETLB_PAGE is not defined.
1251 * Only meaningful when called on a valid entry.
1252 */
1253 #ifndef pgd_leaf
1254 #define pgd_leaf(x) 0
1255 #endif
1256 #ifndef p4d_leaf
1257 #define p4d_leaf(x) 0
1258 #endif
1259 #ifndef pud_leaf
1260 #define pud_leaf(x) 0
1261 #endif
1262 #ifndef pmd_leaf
1263 #define pmd_leaf(x) 0
1264 #endif
1265
1266 #endif /* _ASM_GENERIC_PGTABLE_H */