xref: /linux-6.15/include/linux/pgtable.h (revision a62fb92a)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_PGTABLE_H
3 #define _LINUX_PGTABLE_H
4 
5 #include <linux/pfn.h>
6 #include <asm/pgtable.h>
7 
8 #define PMD_ORDER	(PMD_SHIFT - PAGE_SHIFT)
9 #define PUD_ORDER	(PUD_SHIFT - PAGE_SHIFT)
10 
11 #ifndef __ASSEMBLY__
12 #ifdef CONFIG_MMU
13 
14 #include <linux/mm_types.h>
15 #include <linux/bug.h>
16 #include <linux/errno.h>
17 #include <asm-generic/pgtable_uffd.h>
18 #include <linux/page_table_check.h>
19 
20 #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
21 	defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
22 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
23 #endif
24 
25 /*
26  * On almost all architectures and configurations, 0 can be used as the
27  * upper ceiling to free_pgtables(): on many architectures it has the same
28  * effect as using TASK_SIZE.  However, there is one configuration which
29  * must impose a more careful limit, to avoid freeing kernel pgtables.
30  */
31 #ifndef USER_PGTABLES_CEILING
32 #define USER_PGTABLES_CEILING	0UL
33 #endif
34 
35 /*
36  * This defines the first usable user address. Platforms
37  * can override its value with custom FIRST_USER_ADDRESS
38  * defined in their respective <asm/pgtable.h>.
39  */
40 #ifndef FIRST_USER_ADDRESS
41 #define FIRST_USER_ADDRESS	0UL
42 #endif
43 
44 /*
45  * This defines the generic helper for accessing PMD page
46  * table page. Although platforms can still override this
47  * via their respective <asm/pgtable.h>.
48  */
49 #ifndef pmd_pgtable
50 #define pmd_pgtable(pmd) pmd_page(pmd)
51 #endif
52 
53 #define pmd_folio(pmd) page_folio(pmd_page(pmd))
54 
55 /*
56  * A page table page can be thought of an array like this: pXd_t[PTRS_PER_PxD]
57  *
58  * The pXx_index() functions return the index of the entry in the page
59  * table page which would control the given virtual address
60  *
61  * As these functions may be used by the same code for different levels of
62  * the page table folding, they are always available, regardless of
63  * CONFIG_PGTABLE_LEVELS value. For the folded levels they simply return 0
64  * because in such cases PTRS_PER_PxD equals 1.
65  */
66 
67 static inline unsigned long pte_index(unsigned long address)
68 {
69 	return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
70 }
71 
72 #ifndef pmd_index
73 static inline unsigned long pmd_index(unsigned long address)
74 {
75 	return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
76 }
77 #define pmd_index pmd_index
78 #endif
79 
80 #ifndef pud_index
81 static inline unsigned long pud_index(unsigned long address)
82 {
83 	return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
84 }
85 #define pud_index pud_index
86 #endif
87 
88 #ifndef pgd_index
89 /* Must be a compile-time constant, so implement it as a macro */
90 #define pgd_index(a)  (((a) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
91 #endif
92 
93 #ifndef pte_offset_kernel
94 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
95 {
96 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
97 }
98 #define pte_offset_kernel pte_offset_kernel
99 #endif
100 
101 #ifdef CONFIG_HIGHPTE
102 #define __pte_map(pmd, address) \
103 	((pte_t *)kmap_local_page(pmd_page(*(pmd))) + pte_index((address)))
104 #define pte_unmap(pte)	do {	\
105 	kunmap_local((pte));	\
106 	rcu_read_unlock();	\
107 } while (0)
108 #else
109 static inline pte_t *__pte_map(pmd_t *pmd, unsigned long address)
110 {
111 	return pte_offset_kernel(pmd, address);
112 }
113 static inline void pte_unmap(pte_t *pte)
114 {
115 	rcu_read_unlock();
116 }
117 #endif
118 
119 void pte_free_defer(struct mm_struct *mm, pgtable_t pgtable);
120 
121 /* Find an entry in the second-level page table.. */
122 #ifndef pmd_offset
123 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
124 {
125 	return pud_pgtable(*pud) + pmd_index(address);
126 }
127 #define pmd_offset pmd_offset
128 #endif
129 
130 #ifndef pud_offset
131 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
132 {
133 	return p4d_pgtable(*p4d) + pud_index(address);
134 }
135 #define pud_offset pud_offset
136 #endif
137 
138 static inline pgd_t *pgd_offset_pgd(pgd_t *pgd, unsigned long address)
139 {
140 	return (pgd + pgd_index(address));
141 };
142 
143 /*
144  * a shortcut to get a pgd_t in a given mm
145  */
146 #ifndef pgd_offset
147 #define pgd_offset(mm, address)		pgd_offset_pgd((mm)->pgd, (address))
148 #endif
149 
150 /*
151  * a shortcut which implies the use of the kernel's pgd, instead
152  * of a process's
153  */
154 #define pgd_offset_k(address)		pgd_offset(&init_mm, (address))
155 
156 /*
157  * In many cases it is known that a virtual address is mapped at PMD or PTE
158  * level, so instead of traversing all the page table levels, we can get a
159  * pointer to the PMD entry in user or kernel page table or translate a virtual
160  * address to the pointer in the PTE in the kernel page tables with simple
161  * helpers.
162  */
163 static inline pmd_t *pmd_off(struct mm_struct *mm, unsigned long va)
164 {
165 	return pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, va), va), va), va);
166 }
167 
168 static inline pmd_t *pmd_off_k(unsigned long va)
169 {
170 	return pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va);
171 }
172 
173 static inline pte_t *virt_to_kpte(unsigned long vaddr)
174 {
175 	pmd_t *pmd = pmd_off_k(vaddr);
176 
177 	return pmd_none(*pmd) ? NULL : pte_offset_kernel(pmd, vaddr);
178 }
179 
180 #ifndef pmd_young
181 static inline int pmd_young(pmd_t pmd)
182 {
183 	return 0;
184 }
185 #endif
186 
187 #ifndef pmd_dirty
188 static inline int pmd_dirty(pmd_t pmd)
189 {
190 	return 0;
191 }
192 #endif
193 
194 /*
195  * A facility to provide lazy MMU batching.  This allows PTE updates and
196  * page invalidations to be delayed until a call to leave lazy MMU mode
197  * is issued.  Some architectures may benefit from doing this, and it is
198  * beneficial for both shadow and direct mode hypervisors, which may batch
199  * the PTE updates which happen during this window.  Note that using this
200  * interface requires that read hazards be removed from the code.  A read
201  * hazard could result in the direct mode hypervisor case, since the actual
202  * write to the page tables may not yet have taken place, so reads though
203  * a raw PTE pointer after it has been modified are not guaranteed to be
204  * up to date.  This mode can only be entered and left under the protection of
205  * the page table locks for all page tables which may be modified.  In the UP
206  * case, this is required so that preemption is disabled, and in the SMP case,
207  * it must synchronize the delayed page table writes properly on other CPUs.
208  */
209 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
210 #define arch_enter_lazy_mmu_mode()	do {} while (0)
211 #define arch_leave_lazy_mmu_mode()	do {} while (0)
212 #define arch_flush_lazy_mmu_mode()	do {} while (0)
213 #endif
214 
215 #ifndef pte_batch_hint
216 /**
217  * pte_batch_hint - Number of pages that can be added to batch without scanning.
218  * @ptep: Page table pointer for the entry.
219  * @pte: Page table entry.
220  *
221  * Some architectures know that a set of contiguous ptes all map the same
222  * contiguous memory with the same permissions. In this case, it can provide a
223  * hint to aid pte batching without the core code needing to scan every pte.
224  *
225  * An architecture implementation may ignore the PTE accessed state. Further,
226  * the dirty state must apply atomically to all the PTEs described by the hint.
227  *
228  * May be overridden by the architecture, else pte_batch_hint is always 1.
229  */
230 static inline unsigned int pte_batch_hint(pte_t *ptep, pte_t pte)
231 {
232 	return 1;
233 }
234 #endif
235 
236 #ifndef pte_advance_pfn
237 static inline pte_t pte_advance_pfn(pte_t pte, unsigned long nr)
238 {
239 	return __pte(pte_val(pte) + (nr << PFN_PTE_SHIFT));
240 }
241 #endif
242 
243 #define pte_next_pfn(pte) pte_advance_pfn(pte, 1)
244 
245 #ifndef set_ptes
246 /**
247  * set_ptes - Map consecutive pages to a contiguous range of addresses.
248  * @mm: Address space to map the pages into.
249  * @addr: Address to map the first page at.
250  * @ptep: Page table pointer for the first entry.
251  * @pte: Page table entry for the first page.
252  * @nr: Number of pages to map.
253  *
254  * When nr==1, initial state of pte may be present or not present, and new state
255  * may be present or not present. When nr>1, initial state of all ptes must be
256  * not present, and new state must be present.
257  *
258  * May be overridden by the architecture, or the architecture can define
259  * set_pte() and PFN_PTE_SHIFT.
260  *
261  * Context: The caller holds the page table lock.  The pages all belong
262  * to the same folio.  The PTEs are all in the same PMD.
263  */
264 static inline void set_ptes(struct mm_struct *mm, unsigned long addr,
265 		pte_t *ptep, pte_t pte, unsigned int nr)
266 {
267 	page_table_check_ptes_set(mm, ptep, pte, nr);
268 
269 	arch_enter_lazy_mmu_mode();
270 	for (;;) {
271 		set_pte(ptep, pte);
272 		if (--nr == 0)
273 			break;
274 		ptep++;
275 		pte = pte_next_pfn(pte);
276 	}
277 	arch_leave_lazy_mmu_mode();
278 }
279 #endif
280 #define set_pte_at(mm, addr, ptep, pte) set_ptes(mm, addr, ptep, pte, 1)
281 
282 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
283 extern int ptep_set_access_flags(struct vm_area_struct *vma,
284 				 unsigned long address, pte_t *ptep,
285 				 pte_t entry, int dirty);
286 #endif
287 
288 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
289 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
290 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
291 				 unsigned long address, pmd_t *pmdp,
292 				 pmd_t entry, int dirty);
293 extern int pudp_set_access_flags(struct vm_area_struct *vma,
294 				 unsigned long address, pud_t *pudp,
295 				 pud_t entry, int dirty);
296 #else
297 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
298 					unsigned long address, pmd_t *pmdp,
299 					pmd_t entry, int dirty)
300 {
301 	BUILD_BUG();
302 	return 0;
303 }
304 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
305 					unsigned long address, pud_t *pudp,
306 					pud_t entry, int dirty)
307 {
308 	BUILD_BUG();
309 	return 0;
310 }
311 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
312 #endif
313 
314 #ifndef ptep_get
315 static inline pte_t ptep_get(pte_t *ptep)
316 {
317 	return READ_ONCE(*ptep);
318 }
319 #endif
320 
321 #ifndef pmdp_get
322 static inline pmd_t pmdp_get(pmd_t *pmdp)
323 {
324 	return READ_ONCE(*pmdp);
325 }
326 #endif
327 
328 #ifndef pudp_get
329 static inline pud_t pudp_get(pud_t *pudp)
330 {
331 	return READ_ONCE(*pudp);
332 }
333 #endif
334 
335 #ifndef p4dp_get
336 static inline p4d_t p4dp_get(p4d_t *p4dp)
337 {
338 	return READ_ONCE(*p4dp);
339 }
340 #endif
341 
342 #ifndef pgdp_get
343 static inline pgd_t pgdp_get(pgd_t *pgdp)
344 {
345 	return READ_ONCE(*pgdp);
346 }
347 #endif
348 
349 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
350 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
351 					    unsigned long address,
352 					    pte_t *ptep)
353 {
354 	pte_t pte = ptep_get(ptep);
355 	int r = 1;
356 	if (!pte_young(pte))
357 		r = 0;
358 	else
359 		set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
360 	return r;
361 }
362 #endif
363 
364 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
365 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG)
366 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
367 					    unsigned long address,
368 					    pmd_t *pmdp)
369 {
370 	pmd_t pmd = *pmdp;
371 	int r = 1;
372 	if (!pmd_young(pmd))
373 		r = 0;
374 	else
375 		set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
376 	return r;
377 }
378 #else
379 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
380 					    unsigned long address,
381 					    pmd_t *pmdp)
382 {
383 	BUILD_BUG();
384 	return 0;
385 }
386 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG */
387 #endif
388 
389 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
390 int ptep_clear_flush_young(struct vm_area_struct *vma,
391 			   unsigned long address, pte_t *ptep);
392 #endif
393 
394 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
395 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
396 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
397 				  unsigned long address, pmd_t *pmdp);
398 #else
399 /*
400  * Despite relevant to THP only, this API is called from generic rmap code
401  * under PageTransHuge(), hence needs a dummy implementation for !THP
402  */
403 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
404 					 unsigned long address, pmd_t *pmdp)
405 {
406 	BUILD_BUG();
407 	return 0;
408 }
409 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
410 #endif
411 
412 #ifndef arch_has_hw_nonleaf_pmd_young
413 /*
414  * Return whether the accessed bit in non-leaf PMD entries is supported on the
415  * local CPU.
416  */
417 static inline bool arch_has_hw_nonleaf_pmd_young(void)
418 {
419 	return IS_ENABLED(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG);
420 }
421 #endif
422 
423 #ifndef arch_has_hw_pte_young
424 /*
425  * Return whether the accessed bit is supported on the local CPU.
426  *
427  * This stub assumes accessing through an old PTE triggers a page fault.
428  * Architectures that automatically set the access bit should overwrite it.
429  */
430 static inline bool arch_has_hw_pte_young(void)
431 {
432 	return IS_ENABLED(CONFIG_ARCH_HAS_HW_PTE_YOUNG);
433 }
434 #endif
435 
436 #ifndef arch_check_zapped_pte
437 static inline void arch_check_zapped_pte(struct vm_area_struct *vma,
438 					 pte_t pte)
439 {
440 }
441 #endif
442 
443 #ifndef arch_check_zapped_pmd
444 static inline void arch_check_zapped_pmd(struct vm_area_struct *vma,
445 					 pmd_t pmd)
446 {
447 }
448 #endif
449 
450 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
451 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
452 				       unsigned long address,
453 				       pte_t *ptep)
454 {
455 	pte_t pte = ptep_get(ptep);
456 	pte_clear(mm, address, ptep);
457 	page_table_check_pte_clear(mm, pte);
458 	return pte;
459 }
460 #endif
461 
462 static inline void ptep_clear(struct mm_struct *mm, unsigned long addr,
463 			      pte_t *ptep)
464 {
465 	ptep_get_and_clear(mm, addr, ptep);
466 }
467 
468 #ifdef CONFIG_GUP_GET_PXX_LOW_HIGH
469 /*
470  * For walking the pagetables without holding any locks.  Some architectures
471  * (eg x86-32 PAE) cannot load the entries atomically without using expensive
472  * instructions.  We are guaranteed that a PTE will only either go from not
473  * present to present, or present to not present -- it will not switch to a
474  * completely different present page without a TLB flush inbetween; which we
475  * are blocking by holding interrupts off.
476  *
477  * Setting ptes from not present to present goes:
478  *
479  *   ptep->pte_high = h;
480  *   smp_wmb();
481  *   ptep->pte_low = l;
482  *
483  * And present to not present goes:
484  *
485  *   ptep->pte_low = 0;
486  *   smp_wmb();
487  *   ptep->pte_high = 0;
488  *
489  * We must ensure here that the load of pte_low sees 'l' IFF pte_high sees 'h'.
490  * We load pte_high *after* loading pte_low, which ensures we don't see an older
491  * value of pte_high.  *Then* we recheck pte_low, which ensures that we haven't
492  * picked up a changed pte high. We might have gotten rubbish values from
493  * pte_low and pte_high, but we are guaranteed that pte_low will not have the
494  * present bit set *unless* it is 'l'. Because get_user_pages_fast() only
495  * operates on present ptes we're safe.
496  */
497 static inline pte_t ptep_get_lockless(pte_t *ptep)
498 {
499 	pte_t pte;
500 
501 	do {
502 		pte.pte_low = ptep->pte_low;
503 		smp_rmb();
504 		pte.pte_high = ptep->pte_high;
505 		smp_rmb();
506 	} while (unlikely(pte.pte_low != ptep->pte_low));
507 
508 	return pte;
509 }
510 #define ptep_get_lockless ptep_get_lockless
511 
512 #if CONFIG_PGTABLE_LEVELS > 2
513 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
514 {
515 	pmd_t pmd;
516 
517 	do {
518 		pmd.pmd_low = pmdp->pmd_low;
519 		smp_rmb();
520 		pmd.pmd_high = pmdp->pmd_high;
521 		smp_rmb();
522 	} while (unlikely(pmd.pmd_low != pmdp->pmd_low));
523 
524 	return pmd;
525 }
526 #define pmdp_get_lockless pmdp_get_lockless
527 #define pmdp_get_lockless_sync() tlb_remove_table_sync_one()
528 #endif /* CONFIG_PGTABLE_LEVELS > 2 */
529 #endif /* CONFIG_GUP_GET_PXX_LOW_HIGH */
530 
531 /*
532  * We require that the PTE can be read atomically.
533  */
534 #ifndef ptep_get_lockless
535 static inline pte_t ptep_get_lockless(pte_t *ptep)
536 {
537 	return ptep_get(ptep);
538 }
539 #endif
540 
541 #ifndef pmdp_get_lockless
542 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
543 {
544 	return pmdp_get(pmdp);
545 }
546 static inline void pmdp_get_lockless_sync(void)
547 {
548 }
549 #endif
550 
551 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
552 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
553 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
554 					    unsigned long address,
555 					    pmd_t *pmdp)
556 {
557 	pmd_t pmd = *pmdp;
558 
559 	pmd_clear(pmdp);
560 	page_table_check_pmd_clear(mm, pmd);
561 
562 	return pmd;
563 }
564 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
565 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
566 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
567 					    unsigned long address,
568 					    pud_t *pudp)
569 {
570 	pud_t pud = *pudp;
571 
572 	pud_clear(pudp);
573 	page_table_check_pud_clear(mm, pud);
574 
575 	return pud;
576 }
577 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
578 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
579 
580 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
581 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
582 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
583 					    unsigned long address, pmd_t *pmdp,
584 					    int full)
585 {
586 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
587 }
588 #endif
589 
590 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
591 static inline pud_t pudp_huge_get_and_clear_full(struct vm_area_struct *vma,
592 					    unsigned long address, pud_t *pudp,
593 					    int full)
594 {
595 	return pudp_huge_get_and_clear(vma->vm_mm, address, pudp);
596 }
597 #endif
598 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
599 
600 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
601 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
602 					    unsigned long address, pte_t *ptep,
603 					    int full)
604 {
605 	return ptep_get_and_clear(mm, address, ptep);
606 }
607 #endif
608 
609 #ifndef get_and_clear_full_ptes
610 /**
611  * get_and_clear_full_ptes - Clear present PTEs that map consecutive pages of
612  *			     the same folio, collecting dirty/accessed bits.
613  * @mm: Address space the pages are mapped into.
614  * @addr: Address the first page is mapped at.
615  * @ptep: Page table pointer for the first entry.
616  * @nr: Number of entries to clear.
617  * @full: Whether we are clearing a full mm.
618  *
619  * May be overridden by the architecture; otherwise, implemented as a simple
620  * loop over ptep_get_and_clear_full(), merging dirty/accessed bits into the
621  * returned PTE.
622  *
623  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
624  * some PTEs might be write-protected.
625  *
626  * Context: The caller holds the page table lock.  The PTEs map consecutive
627  * pages that belong to the same folio.  The PTEs are all in the same PMD.
628  */
629 static inline pte_t get_and_clear_full_ptes(struct mm_struct *mm,
630 		unsigned long addr, pte_t *ptep, unsigned int nr, int full)
631 {
632 	pte_t pte, tmp_pte;
633 
634 	pte = ptep_get_and_clear_full(mm, addr, ptep, full);
635 	while (--nr) {
636 		ptep++;
637 		addr += PAGE_SIZE;
638 		tmp_pte = ptep_get_and_clear_full(mm, addr, ptep, full);
639 		if (pte_dirty(tmp_pte))
640 			pte = pte_mkdirty(pte);
641 		if (pte_young(tmp_pte))
642 			pte = pte_mkyoung(pte);
643 	}
644 	return pte;
645 }
646 #endif
647 
648 #ifndef clear_full_ptes
649 /**
650  * clear_full_ptes - Clear present PTEs that map consecutive pages of the same
651  *		     folio.
652  * @mm: Address space the pages are mapped into.
653  * @addr: Address the first page is mapped at.
654  * @ptep: Page table pointer for the first entry.
655  * @nr: Number of entries to clear.
656  * @full: Whether we are clearing a full mm.
657  *
658  * May be overridden by the architecture; otherwise, implemented as a simple
659  * loop over ptep_get_and_clear_full().
660  *
661  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
662  * some PTEs might be write-protected.
663  *
664  * Context: The caller holds the page table lock.  The PTEs map consecutive
665  * pages that belong to the same folio.  The PTEs are all in the same PMD.
666  */
667 static inline void clear_full_ptes(struct mm_struct *mm, unsigned long addr,
668 		pte_t *ptep, unsigned int nr, int full)
669 {
670 	for (;;) {
671 		ptep_get_and_clear_full(mm, addr, ptep, full);
672 		if (--nr == 0)
673 			break;
674 		ptep++;
675 		addr += PAGE_SIZE;
676 	}
677 }
678 #endif
679 
680 /*
681  * If two threads concurrently fault at the same page, the thread that
682  * won the race updates the PTE and its local TLB/Cache. The other thread
683  * gives up, simply does nothing, and continues; on architectures where
684  * software can update TLB,  local TLB can be updated here to avoid next page
685  * fault. This function updates TLB only, do nothing with cache or others.
686  * It is the difference with function update_mmu_cache.
687  */
688 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
689 static inline void update_mmu_tlb(struct vm_area_struct *vma,
690 				unsigned long address, pte_t *ptep)
691 {
692 }
693 #define __HAVE_ARCH_UPDATE_MMU_TLB
694 #endif
695 
696 /*
697  * Some architectures may be able to avoid expensive synchronization
698  * primitives when modifications are made to PTE's which are already
699  * not present, or in the process of an address space destruction.
700  */
701 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
702 static inline void pte_clear_not_present_full(struct mm_struct *mm,
703 					      unsigned long address,
704 					      pte_t *ptep,
705 					      int full)
706 {
707 	pte_clear(mm, address, ptep);
708 }
709 #endif
710 
711 #ifndef clear_not_present_full_ptes
712 /**
713  * clear_not_present_full_ptes - Clear multiple not present PTEs which are
714  *				 consecutive in the pgtable.
715  * @mm: Address space the ptes represent.
716  * @addr: Address of the first pte.
717  * @ptep: Page table pointer for the first entry.
718  * @nr: Number of entries to clear.
719  * @full: Whether we are clearing a full mm.
720  *
721  * May be overridden by the architecture; otherwise, implemented as a simple
722  * loop over pte_clear_not_present_full().
723  *
724  * Context: The caller holds the page table lock.  The PTEs are all not present.
725  * The PTEs are all in the same PMD.
726  */
727 static inline void clear_not_present_full_ptes(struct mm_struct *mm,
728 		unsigned long addr, pte_t *ptep, unsigned int nr, int full)
729 {
730 	for (;;) {
731 		pte_clear_not_present_full(mm, addr, ptep, full);
732 		if (--nr == 0)
733 			break;
734 		ptep++;
735 		addr += PAGE_SIZE;
736 	}
737 }
738 #endif
739 
740 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
741 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
742 			      unsigned long address,
743 			      pte_t *ptep);
744 #endif
745 
746 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
747 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
748 			      unsigned long address,
749 			      pmd_t *pmdp);
750 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
751 			      unsigned long address,
752 			      pud_t *pudp);
753 #endif
754 
755 #ifndef pte_mkwrite
756 static inline pte_t pte_mkwrite(pte_t pte, struct vm_area_struct *vma)
757 {
758 	return pte_mkwrite_novma(pte);
759 }
760 #endif
761 
762 #if defined(CONFIG_ARCH_WANT_PMD_MKWRITE) && !defined(pmd_mkwrite)
763 static inline pmd_t pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma)
764 {
765 	return pmd_mkwrite_novma(pmd);
766 }
767 #endif
768 
769 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
770 struct mm_struct;
771 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
772 {
773 	pte_t old_pte = ptep_get(ptep);
774 	set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
775 }
776 #endif
777 
778 #ifndef wrprotect_ptes
779 /**
780  * wrprotect_ptes - Write-protect PTEs that map consecutive pages of the same
781  *		    folio.
782  * @mm: Address space the pages are mapped into.
783  * @addr: Address the first page is mapped at.
784  * @ptep: Page table pointer for the first entry.
785  * @nr: Number of entries to write-protect.
786  *
787  * May be overridden by the architecture; otherwise, implemented as a simple
788  * loop over ptep_set_wrprotect().
789  *
790  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
791  * some PTEs might be write-protected.
792  *
793  * Context: The caller holds the page table lock.  The PTEs map consecutive
794  * pages that belong to the same folio.  The PTEs are all in the same PMD.
795  */
796 static inline void wrprotect_ptes(struct mm_struct *mm, unsigned long addr,
797 		pte_t *ptep, unsigned int nr)
798 {
799 	for (;;) {
800 		ptep_set_wrprotect(mm, addr, ptep);
801 		if (--nr == 0)
802 			break;
803 		ptep++;
804 		addr += PAGE_SIZE;
805 	}
806 }
807 #endif
808 
809 /*
810  * On some architectures hardware does not set page access bit when accessing
811  * memory page, it is responsibility of software setting this bit. It brings
812  * out extra page fault penalty to track page access bit. For optimization page
813  * access bit can be set during all page fault flow on these arches.
814  * To be differentiate with macro pte_mkyoung, this macro is used on platforms
815  * where software maintains page access bit.
816  */
817 #ifndef pte_sw_mkyoung
818 static inline pte_t pte_sw_mkyoung(pte_t pte)
819 {
820 	return pte;
821 }
822 #define pte_sw_mkyoung	pte_sw_mkyoung
823 #endif
824 
825 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
826 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
827 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
828 				      unsigned long address, pmd_t *pmdp)
829 {
830 	pmd_t old_pmd = *pmdp;
831 	set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
832 }
833 #else
834 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
835 				      unsigned long address, pmd_t *pmdp)
836 {
837 	BUILD_BUG();
838 }
839 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
840 #endif
841 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
842 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
843 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
844 static inline void pudp_set_wrprotect(struct mm_struct *mm,
845 				      unsigned long address, pud_t *pudp)
846 {
847 	pud_t old_pud = *pudp;
848 
849 	set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
850 }
851 #else
852 static inline void pudp_set_wrprotect(struct mm_struct *mm,
853 				      unsigned long address, pud_t *pudp)
854 {
855 	BUILD_BUG();
856 }
857 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
858 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
859 #endif
860 
861 #ifndef pmdp_collapse_flush
862 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
863 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
864 				 unsigned long address, pmd_t *pmdp);
865 #else
866 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
867 					unsigned long address,
868 					pmd_t *pmdp)
869 {
870 	BUILD_BUG();
871 	return *pmdp;
872 }
873 #define pmdp_collapse_flush pmdp_collapse_flush
874 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
875 #endif
876 
877 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
878 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
879 				       pgtable_t pgtable);
880 #endif
881 
882 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
883 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
884 #endif
885 
886 #ifndef arch_needs_pgtable_deposit
887 #define arch_needs_pgtable_deposit() (false)
888 #endif
889 
890 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
891 /*
892  * This is an implementation of pmdp_establish() that is only suitable for an
893  * architecture that doesn't have hardware dirty/accessed bits. In this case we
894  * can't race with CPU which sets these bits and non-atomic approach is fine.
895  */
896 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
897 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
898 {
899 	pmd_t old_pmd = *pmdp;
900 	set_pmd_at(vma->vm_mm, address, pmdp, pmd);
901 	return old_pmd;
902 }
903 #endif
904 
905 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
906 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
907 			    pmd_t *pmdp);
908 #endif
909 
910 #ifndef __HAVE_ARCH_PMDP_INVALIDATE_AD
911 
912 /*
913  * pmdp_invalidate_ad() invalidates the PMD while changing a transparent
914  * hugepage mapping in the page tables. This function is similar to
915  * pmdp_invalidate(), but should only be used if the access and dirty bits would
916  * not be cleared by the software in the new PMD value. The function ensures
917  * that hardware changes of the access and dirty bits updates would not be lost.
918  *
919  * Doing so can allow in certain architectures to avoid a TLB flush in most
920  * cases. Yet, another TLB flush might be necessary later if the PMD update
921  * itself requires such flush (e.g., if protection was set to be stricter). Yet,
922  * even when a TLB flush is needed because of the update, the caller may be able
923  * to batch these TLB flushing operations, so fewer TLB flush operations are
924  * needed.
925  */
926 extern pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma,
927 				unsigned long address, pmd_t *pmdp);
928 #endif
929 
930 #ifndef __HAVE_ARCH_PTE_SAME
931 static inline int pte_same(pte_t pte_a, pte_t pte_b)
932 {
933 	return pte_val(pte_a) == pte_val(pte_b);
934 }
935 #endif
936 
937 #ifndef __HAVE_ARCH_PTE_UNUSED
938 /*
939  * Some architectures provide facilities to virtualization guests
940  * so that they can flag allocated pages as unused. This allows the
941  * host to transparently reclaim unused pages. This function returns
942  * whether the pte's page is unused.
943  */
944 static inline int pte_unused(pte_t pte)
945 {
946 	return 0;
947 }
948 #endif
949 
950 #ifndef pte_access_permitted
951 #define pte_access_permitted(pte, write) \
952 	(pte_present(pte) && (!(write) || pte_write(pte)))
953 #endif
954 
955 #ifndef pmd_access_permitted
956 #define pmd_access_permitted(pmd, write) \
957 	(pmd_present(pmd) && (!(write) || pmd_write(pmd)))
958 #endif
959 
960 #ifndef pud_access_permitted
961 #define pud_access_permitted(pud, write) \
962 	(pud_present(pud) && (!(write) || pud_write(pud)))
963 #endif
964 
965 #ifndef p4d_access_permitted
966 #define p4d_access_permitted(p4d, write) \
967 	(p4d_present(p4d) && (!(write) || p4d_write(p4d)))
968 #endif
969 
970 #ifndef pgd_access_permitted
971 #define pgd_access_permitted(pgd, write) \
972 	(pgd_present(pgd) && (!(write) || pgd_write(pgd)))
973 #endif
974 
975 #ifndef __HAVE_ARCH_PMD_SAME
976 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
977 {
978 	return pmd_val(pmd_a) == pmd_val(pmd_b);
979 }
980 #endif
981 
982 #ifndef pud_same
983 static inline int pud_same(pud_t pud_a, pud_t pud_b)
984 {
985 	return pud_val(pud_a) == pud_val(pud_b);
986 }
987 #define pud_same pud_same
988 #endif
989 
990 #ifndef __HAVE_ARCH_P4D_SAME
991 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
992 {
993 	return p4d_val(p4d_a) == p4d_val(p4d_b);
994 }
995 #endif
996 
997 #ifndef __HAVE_ARCH_PGD_SAME
998 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
999 {
1000 	return pgd_val(pgd_a) == pgd_val(pgd_b);
1001 }
1002 #endif
1003 
1004 /*
1005  * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
1006  * TLB flush will be required as a result of the "set". For example, use
1007  * in scenarios where it is known ahead of time that the routine is
1008  * setting non-present entries, or re-setting an existing entry to the
1009  * same value. Otherwise, use the typical "set" helpers and flush the
1010  * TLB.
1011  */
1012 #define set_pte_safe(ptep, pte) \
1013 ({ \
1014 	WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
1015 	set_pte(ptep, pte); \
1016 })
1017 
1018 #define set_pmd_safe(pmdp, pmd) \
1019 ({ \
1020 	WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
1021 	set_pmd(pmdp, pmd); \
1022 })
1023 
1024 #define set_pud_safe(pudp, pud) \
1025 ({ \
1026 	WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
1027 	set_pud(pudp, pud); \
1028 })
1029 
1030 #define set_p4d_safe(p4dp, p4d) \
1031 ({ \
1032 	WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
1033 	set_p4d(p4dp, p4d); \
1034 })
1035 
1036 #define set_pgd_safe(pgdp, pgd) \
1037 ({ \
1038 	WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
1039 	set_pgd(pgdp, pgd); \
1040 })
1041 
1042 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
1043 /*
1044  * Some architectures support metadata associated with a page. When a
1045  * page is being swapped out, this metadata must be saved so it can be
1046  * restored when the page is swapped back in. SPARC M7 and newer
1047  * processors support an ADI (Application Data Integrity) tag for the
1048  * page as metadata for the page. arch_do_swap_page() can restore this
1049  * metadata when a page is swapped back in.
1050  */
1051 static inline void arch_do_swap_page(struct mm_struct *mm,
1052 				     struct vm_area_struct *vma,
1053 				     unsigned long addr,
1054 				     pte_t pte, pte_t oldpte)
1055 {
1056 
1057 }
1058 #endif
1059 
1060 #ifndef __HAVE_ARCH_UNMAP_ONE
1061 /*
1062  * Some architectures support metadata associated with a page. When a
1063  * page is being swapped out, this metadata must be saved so it can be
1064  * restored when the page is swapped back in. SPARC M7 and newer
1065  * processors support an ADI (Application Data Integrity) tag for the
1066  * page as metadata for the page. arch_unmap_one() can save this
1067  * metadata on a swap-out of a page.
1068  */
1069 static inline int arch_unmap_one(struct mm_struct *mm,
1070 				  struct vm_area_struct *vma,
1071 				  unsigned long addr,
1072 				  pte_t orig_pte)
1073 {
1074 	return 0;
1075 }
1076 #endif
1077 
1078 /*
1079  * Allow architectures to preserve additional metadata associated with
1080  * swapped-out pages. The corresponding __HAVE_ARCH_SWAP_* macros and function
1081  * prototypes must be defined in the arch-specific asm/pgtable.h file.
1082  */
1083 #ifndef __HAVE_ARCH_PREPARE_TO_SWAP
1084 static inline int arch_prepare_to_swap(struct folio *folio)
1085 {
1086 	return 0;
1087 }
1088 #endif
1089 
1090 #ifndef __HAVE_ARCH_SWAP_INVALIDATE
1091 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
1092 {
1093 }
1094 
1095 static inline void arch_swap_invalidate_area(int type)
1096 {
1097 }
1098 #endif
1099 
1100 #ifndef __HAVE_ARCH_SWAP_RESTORE
1101 static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
1102 {
1103 }
1104 #endif
1105 
1106 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
1107 #define pgd_offset_gate(mm, addr)	pgd_offset(mm, addr)
1108 #endif
1109 
1110 #ifndef __HAVE_ARCH_MOVE_PTE
1111 #define move_pte(pte, old_addr, new_addr)	(pte)
1112 #endif
1113 
1114 #ifndef pte_accessible
1115 # define pte_accessible(mm, pte)	((void)(pte), 1)
1116 #endif
1117 
1118 #ifndef flush_tlb_fix_spurious_fault
1119 #define flush_tlb_fix_spurious_fault(vma, address, ptep) flush_tlb_page(vma, address)
1120 #endif
1121 
1122 /*
1123  * When walking page tables, get the address of the next boundary,
1124  * or the end address of the range if that comes earlier.  Although no
1125  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
1126  */
1127 
1128 #define pgd_addr_end(addr, end)						\
1129 ({	unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;	\
1130 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1131 })
1132 
1133 #ifndef p4d_addr_end
1134 #define p4d_addr_end(addr, end)						\
1135 ({	unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;	\
1136 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1137 })
1138 #endif
1139 
1140 #ifndef pud_addr_end
1141 #define pud_addr_end(addr, end)						\
1142 ({	unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;	\
1143 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1144 })
1145 #endif
1146 
1147 #ifndef pmd_addr_end
1148 #define pmd_addr_end(addr, end)						\
1149 ({	unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;	\
1150 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1151 })
1152 #endif
1153 
1154 /*
1155  * When walking page tables, we usually want to skip any p?d_none entries;
1156  * and any p?d_bad entries - reporting the error before resetting to none.
1157  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
1158  */
1159 void pgd_clear_bad(pgd_t *);
1160 
1161 #ifndef __PAGETABLE_P4D_FOLDED
1162 void p4d_clear_bad(p4d_t *);
1163 #else
1164 #define p4d_clear_bad(p4d)        do { } while (0)
1165 #endif
1166 
1167 #ifndef __PAGETABLE_PUD_FOLDED
1168 void pud_clear_bad(pud_t *);
1169 #else
1170 #define pud_clear_bad(p4d)        do { } while (0)
1171 #endif
1172 
1173 void pmd_clear_bad(pmd_t *);
1174 
1175 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
1176 {
1177 	if (pgd_none(*pgd))
1178 		return 1;
1179 	if (unlikely(pgd_bad(*pgd))) {
1180 		pgd_clear_bad(pgd);
1181 		return 1;
1182 	}
1183 	return 0;
1184 }
1185 
1186 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
1187 {
1188 	if (p4d_none(*p4d))
1189 		return 1;
1190 	if (unlikely(p4d_bad(*p4d))) {
1191 		p4d_clear_bad(p4d);
1192 		return 1;
1193 	}
1194 	return 0;
1195 }
1196 
1197 static inline int pud_none_or_clear_bad(pud_t *pud)
1198 {
1199 	if (pud_none(*pud))
1200 		return 1;
1201 	if (unlikely(pud_bad(*pud))) {
1202 		pud_clear_bad(pud);
1203 		return 1;
1204 	}
1205 	return 0;
1206 }
1207 
1208 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
1209 {
1210 	if (pmd_none(*pmd))
1211 		return 1;
1212 	if (unlikely(pmd_bad(*pmd))) {
1213 		pmd_clear_bad(pmd);
1214 		return 1;
1215 	}
1216 	return 0;
1217 }
1218 
1219 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
1220 					     unsigned long addr,
1221 					     pte_t *ptep)
1222 {
1223 	/*
1224 	 * Get the current pte state, but zero it out to make it
1225 	 * non-present, preventing the hardware from asynchronously
1226 	 * updating it.
1227 	 */
1228 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
1229 }
1230 
1231 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
1232 					     unsigned long addr,
1233 					     pte_t *ptep, pte_t pte)
1234 {
1235 	/*
1236 	 * The pte is non-present, so there's no hardware state to
1237 	 * preserve.
1238 	 */
1239 	set_pte_at(vma->vm_mm, addr, ptep, pte);
1240 }
1241 
1242 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1243 /*
1244  * Start a pte protection read-modify-write transaction, which
1245  * protects against asynchronous hardware modifications to the pte.
1246  * The intention is not to prevent the hardware from making pte
1247  * updates, but to prevent any updates it may make from being lost.
1248  *
1249  * This does not protect against other software modifications of the
1250  * pte; the appropriate pte lock must be held over the transaction.
1251  *
1252  * Note that this interface is intended to be batchable, meaning that
1253  * ptep_modify_prot_commit may not actually update the pte, but merely
1254  * queue the update to be done at some later time.  The update must be
1255  * actually committed before the pte lock is released, however.
1256  */
1257 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
1258 					   unsigned long addr,
1259 					   pte_t *ptep)
1260 {
1261 	return __ptep_modify_prot_start(vma, addr, ptep);
1262 }
1263 
1264 /*
1265  * Commit an update to a pte, leaving any hardware-controlled bits in
1266  * the PTE unmodified.
1267  */
1268 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
1269 					   unsigned long addr,
1270 					   pte_t *ptep, pte_t old_pte, pte_t pte)
1271 {
1272 	__ptep_modify_prot_commit(vma, addr, ptep, pte);
1273 }
1274 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
1275 #endif /* CONFIG_MMU */
1276 
1277 /*
1278  * No-op macros that just return the current protection value. Defined here
1279  * because these macros can be used even if CONFIG_MMU is not defined.
1280  */
1281 
1282 #ifndef pgprot_nx
1283 #define pgprot_nx(prot)	(prot)
1284 #endif
1285 
1286 #ifndef pgprot_noncached
1287 #define pgprot_noncached(prot)	(prot)
1288 #endif
1289 
1290 #ifndef pgprot_writecombine
1291 #define pgprot_writecombine pgprot_noncached
1292 #endif
1293 
1294 #ifndef pgprot_writethrough
1295 #define pgprot_writethrough pgprot_noncached
1296 #endif
1297 
1298 #ifndef pgprot_device
1299 #define pgprot_device pgprot_noncached
1300 #endif
1301 
1302 #ifndef pgprot_mhp
1303 #define pgprot_mhp(prot)	(prot)
1304 #endif
1305 
1306 #ifdef CONFIG_MMU
1307 #ifndef pgprot_modify
1308 #define pgprot_modify pgprot_modify
1309 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
1310 {
1311 	if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
1312 		newprot = pgprot_noncached(newprot);
1313 	if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
1314 		newprot = pgprot_writecombine(newprot);
1315 	if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
1316 		newprot = pgprot_device(newprot);
1317 	return newprot;
1318 }
1319 #endif
1320 #endif /* CONFIG_MMU */
1321 
1322 #ifndef pgprot_encrypted
1323 #define pgprot_encrypted(prot)	(prot)
1324 #endif
1325 
1326 #ifndef pgprot_decrypted
1327 #define pgprot_decrypted(prot)	(prot)
1328 #endif
1329 
1330 /*
1331  * A facility to provide batching of the reload of page tables and
1332  * other process state with the actual context switch code for
1333  * paravirtualized guests.  By convention, only one of the batched
1334  * update (lazy) modes (CPU, MMU) should be active at any given time,
1335  * entry should never be nested, and entry and exits should always be
1336  * paired.  This is for sanity of maintaining and reasoning about the
1337  * kernel code.  In this case, the exit (end of the context switch) is
1338  * in architecture-specific code, and so doesn't need a generic
1339  * definition.
1340  */
1341 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
1342 #define arch_start_context_switch(prev)	do {} while (0)
1343 #endif
1344 
1345 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1346 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
1347 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1348 {
1349 	return pmd;
1350 }
1351 
1352 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1353 {
1354 	return 0;
1355 }
1356 
1357 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1358 {
1359 	return pmd;
1360 }
1361 #endif
1362 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
1363 static inline int pte_soft_dirty(pte_t pte)
1364 {
1365 	return 0;
1366 }
1367 
1368 static inline int pmd_soft_dirty(pmd_t pmd)
1369 {
1370 	return 0;
1371 }
1372 
1373 static inline pte_t pte_mksoft_dirty(pte_t pte)
1374 {
1375 	return pte;
1376 }
1377 
1378 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
1379 {
1380 	return pmd;
1381 }
1382 
1383 static inline pte_t pte_clear_soft_dirty(pte_t pte)
1384 {
1385 	return pte;
1386 }
1387 
1388 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
1389 {
1390 	return pmd;
1391 }
1392 
1393 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1394 {
1395 	return pte;
1396 }
1397 
1398 static inline int pte_swp_soft_dirty(pte_t pte)
1399 {
1400 	return 0;
1401 }
1402 
1403 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1404 {
1405 	return pte;
1406 }
1407 
1408 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1409 {
1410 	return pmd;
1411 }
1412 
1413 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1414 {
1415 	return 0;
1416 }
1417 
1418 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1419 {
1420 	return pmd;
1421 }
1422 #endif
1423 
1424 #ifndef __HAVE_PFNMAP_TRACKING
1425 /*
1426  * Interfaces that can be used by architecture code to keep track of
1427  * memory type of pfn mappings specified by the remap_pfn_range,
1428  * vmf_insert_pfn.
1429  */
1430 
1431 /*
1432  * track_pfn_remap is called when a _new_ pfn mapping is being established
1433  * by remap_pfn_range() for physical range indicated by pfn and size.
1434  */
1435 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1436 				  unsigned long pfn, unsigned long addr,
1437 				  unsigned long size)
1438 {
1439 	return 0;
1440 }
1441 
1442 /*
1443  * track_pfn_insert is called when a _new_ single pfn is established
1444  * by vmf_insert_pfn().
1445  */
1446 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1447 				    pfn_t pfn)
1448 {
1449 }
1450 
1451 /*
1452  * track_pfn_copy is called when vma that is covering the pfnmap gets
1453  * copied through copy_page_range().
1454  */
1455 static inline int track_pfn_copy(struct vm_area_struct *vma)
1456 {
1457 	return 0;
1458 }
1459 
1460 /*
1461  * untrack_pfn is called while unmapping a pfnmap for a region.
1462  * untrack can be called for a specific region indicated by pfn and size or
1463  * can be for the entire vma (in which case pfn, size are zero).
1464  */
1465 static inline void untrack_pfn(struct vm_area_struct *vma,
1466 			       unsigned long pfn, unsigned long size,
1467 			       bool mm_wr_locked)
1468 {
1469 }
1470 
1471 /*
1472  * untrack_pfn_clear is called while mremapping a pfnmap for a new region
1473  * or fails to copy pgtable during duplicate vm area.
1474  */
1475 static inline void untrack_pfn_clear(struct vm_area_struct *vma)
1476 {
1477 }
1478 #else
1479 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1480 			   unsigned long pfn, unsigned long addr,
1481 			   unsigned long size);
1482 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1483 			     pfn_t pfn);
1484 extern int track_pfn_copy(struct vm_area_struct *vma);
1485 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1486 			unsigned long size, bool mm_wr_locked);
1487 extern void untrack_pfn_clear(struct vm_area_struct *vma);
1488 #endif
1489 
1490 #ifdef CONFIG_MMU
1491 #ifdef __HAVE_COLOR_ZERO_PAGE
1492 static inline int is_zero_pfn(unsigned long pfn)
1493 {
1494 	extern unsigned long zero_pfn;
1495 	unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1496 	return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1497 }
1498 
1499 #define my_zero_pfn(addr)	page_to_pfn(ZERO_PAGE(addr))
1500 
1501 #else
1502 static inline int is_zero_pfn(unsigned long pfn)
1503 {
1504 	extern unsigned long zero_pfn;
1505 	return pfn == zero_pfn;
1506 }
1507 
1508 static inline unsigned long my_zero_pfn(unsigned long addr)
1509 {
1510 	extern unsigned long zero_pfn;
1511 	return zero_pfn;
1512 }
1513 #endif
1514 #else
1515 static inline int is_zero_pfn(unsigned long pfn)
1516 {
1517 	return 0;
1518 }
1519 
1520 static inline unsigned long my_zero_pfn(unsigned long addr)
1521 {
1522 	return 0;
1523 }
1524 #endif /* CONFIG_MMU */
1525 
1526 #ifdef CONFIG_MMU
1527 
1528 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1529 static inline int pmd_trans_huge(pmd_t pmd)
1530 {
1531 	return 0;
1532 }
1533 #ifndef pmd_write
1534 static inline int pmd_write(pmd_t pmd)
1535 {
1536 	BUG();
1537 	return 0;
1538 }
1539 #endif /* pmd_write */
1540 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1541 
1542 #ifndef pud_write
1543 static inline int pud_write(pud_t pud)
1544 {
1545 	BUG();
1546 	return 0;
1547 }
1548 #endif /* pud_write */
1549 
1550 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1551 static inline int pmd_devmap(pmd_t pmd)
1552 {
1553 	return 0;
1554 }
1555 static inline int pud_devmap(pud_t pud)
1556 {
1557 	return 0;
1558 }
1559 static inline int pgd_devmap(pgd_t pgd)
1560 {
1561 	return 0;
1562 }
1563 #endif
1564 
1565 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1566 	!defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1567 static inline int pud_trans_huge(pud_t pud)
1568 {
1569 	return 0;
1570 }
1571 #endif
1572 
1573 static inline int pud_trans_unstable(pud_t *pud)
1574 {
1575 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1576 	defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1577 	pud_t pudval = READ_ONCE(*pud);
1578 
1579 	if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1580 		return 1;
1581 	if (unlikely(pud_bad(pudval))) {
1582 		pud_clear_bad(pud);
1583 		return 1;
1584 	}
1585 #endif
1586 	return 0;
1587 }
1588 
1589 #ifndef CONFIG_NUMA_BALANCING
1590 /*
1591  * In an inaccessible (PROT_NONE) VMA, pte_protnone() may indicate "yes". It is
1592  * perfectly valid to indicate "no" in that case, which is why our default
1593  * implementation defaults to "always no".
1594  *
1595  * In an accessible VMA, however, pte_protnone() reliably indicates PROT_NONE
1596  * page protection due to NUMA hinting. NUMA hinting faults only apply in
1597  * accessible VMAs.
1598  *
1599  * So, to reliably identify PROT_NONE PTEs that require a NUMA hinting fault,
1600  * looking at the VMA accessibility is sufficient.
1601  */
1602 static inline int pte_protnone(pte_t pte)
1603 {
1604 	return 0;
1605 }
1606 
1607 static inline int pmd_protnone(pmd_t pmd)
1608 {
1609 	return 0;
1610 }
1611 #endif /* CONFIG_NUMA_BALANCING */
1612 
1613 #endif /* CONFIG_MMU */
1614 
1615 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1616 
1617 #ifndef __PAGETABLE_P4D_FOLDED
1618 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1619 void p4d_clear_huge(p4d_t *p4d);
1620 #else
1621 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1622 {
1623 	return 0;
1624 }
1625 static inline void p4d_clear_huge(p4d_t *p4d) { }
1626 #endif /* !__PAGETABLE_P4D_FOLDED */
1627 
1628 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1629 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1630 int pud_clear_huge(pud_t *pud);
1631 int pmd_clear_huge(pmd_t *pmd);
1632 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1633 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1634 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1635 #else	/* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1636 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1637 {
1638 	return 0;
1639 }
1640 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1641 {
1642 	return 0;
1643 }
1644 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1645 {
1646 	return 0;
1647 }
1648 static inline void p4d_clear_huge(p4d_t *p4d) { }
1649 static inline int pud_clear_huge(pud_t *pud)
1650 {
1651 	return 0;
1652 }
1653 static inline int pmd_clear_huge(pmd_t *pmd)
1654 {
1655 	return 0;
1656 }
1657 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1658 {
1659 	return 0;
1660 }
1661 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1662 {
1663 	return 0;
1664 }
1665 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1666 {
1667 	return 0;
1668 }
1669 #endif	/* CONFIG_HAVE_ARCH_HUGE_VMAP */
1670 
1671 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1672 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1673 /*
1674  * ARCHes with special requirements for evicting THP backing TLB entries can
1675  * implement this. Otherwise also, it can help optimize normal TLB flush in
1676  * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1677  * entire TLB if flush span is greater than a threshold, which will
1678  * likely be true for a single huge page. Thus a single THP flush will
1679  * invalidate the entire TLB which is not desirable.
1680  * e.g. see arch/arc: flush_pmd_tlb_range
1681  */
1682 #define flush_pmd_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1683 #define flush_pud_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1684 #else
1685 #define flush_pmd_tlb_range(vma, addr, end)	BUILD_BUG()
1686 #define flush_pud_tlb_range(vma, addr, end)	BUILD_BUG()
1687 #endif
1688 #endif
1689 
1690 struct file;
1691 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1692 			unsigned long size, pgprot_t *vma_prot);
1693 
1694 #ifndef CONFIG_X86_ESPFIX64
1695 static inline void init_espfix_bsp(void) { }
1696 #endif
1697 
1698 extern void __init pgtable_cache_init(void);
1699 
1700 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1701 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1702 {
1703 	return true;
1704 }
1705 
1706 static inline bool arch_has_pfn_modify_check(void)
1707 {
1708 	return false;
1709 }
1710 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1711 
1712 /*
1713  * Architecture PAGE_KERNEL_* fallbacks
1714  *
1715  * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1716  * because they really don't support them, or the port needs to be updated to
1717  * reflect the required functionality. Below are a set of relatively safe
1718  * fallbacks, as best effort, which we can count on in lieu of the architectures
1719  * not defining them on their own yet.
1720  */
1721 
1722 #ifndef PAGE_KERNEL_RO
1723 # define PAGE_KERNEL_RO PAGE_KERNEL
1724 #endif
1725 
1726 #ifndef PAGE_KERNEL_EXEC
1727 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1728 #endif
1729 
1730 /*
1731  * Page Table Modification bits for pgtbl_mod_mask.
1732  *
1733  * These are used by the p?d_alloc_track*() set of functions an in the generic
1734  * vmalloc/ioremap code to track at which page-table levels entries have been
1735  * modified. Based on that the code can better decide when vmalloc and ioremap
1736  * mapping changes need to be synchronized to other page-tables in the system.
1737  */
1738 #define		__PGTBL_PGD_MODIFIED	0
1739 #define		__PGTBL_P4D_MODIFIED	1
1740 #define		__PGTBL_PUD_MODIFIED	2
1741 #define		__PGTBL_PMD_MODIFIED	3
1742 #define		__PGTBL_PTE_MODIFIED	4
1743 
1744 #define		PGTBL_PGD_MODIFIED	BIT(__PGTBL_PGD_MODIFIED)
1745 #define		PGTBL_P4D_MODIFIED	BIT(__PGTBL_P4D_MODIFIED)
1746 #define		PGTBL_PUD_MODIFIED	BIT(__PGTBL_PUD_MODIFIED)
1747 #define		PGTBL_PMD_MODIFIED	BIT(__PGTBL_PMD_MODIFIED)
1748 #define		PGTBL_PTE_MODIFIED	BIT(__PGTBL_PTE_MODIFIED)
1749 
1750 /* Page-Table Modification Mask */
1751 typedef unsigned int pgtbl_mod_mask;
1752 
1753 #endif /* !__ASSEMBLY__ */
1754 
1755 #if !defined(MAX_POSSIBLE_PHYSMEM_BITS) && !defined(CONFIG_64BIT)
1756 #ifdef CONFIG_PHYS_ADDR_T_64BIT
1757 /*
1758  * ZSMALLOC needs to know the highest PFN on 32-bit architectures
1759  * with physical address space extension, but falls back to
1760  * BITS_PER_LONG otherwise.
1761  */
1762 #error Missing MAX_POSSIBLE_PHYSMEM_BITS definition
1763 #else
1764 #define MAX_POSSIBLE_PHYSMEM_BITS 32
1765 #endif
1766 #endif
1767 
1768 #ifndef has_transparent_hugepage
1769 #define has_transparent_hugepage() IS_BUILTIN(CONFIG_TRANSPARENT_HUGEPAGE)
1770 #endif
1771 
1772 #ifndef has_transparent_pud_hugepage
1773 #define has_transparent_pud_hugepage() IS_BUILTIN(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1774 #endif
1775 /*
1776  * On some architectures it depends on the mm if the p4d/pud or pmd
1777  * layer of the page table hierarchy is folded or not.
1778  */
1779 #ifndef mm_p4d_folded
1780 #define mm_p4d_folded(mm)	__is_defined(__PAGETABLE_P4D_FOLDED)
1781 #endif
1782 
1783 #ifndef mm_pud_folded
1784 #define mm_pud_folded(mm)	__is_defined(__PAGETABLE_PUD_FOLDED)
1785 #endif
1786 
1787 #ifndef mm_pmd_folded
1788 #define mm_pmd_folded(mm)	__is_defined(__PAGETABLE_PMD_FOLDED)
1789 #endif
1790 
1791 #ifndef p4d_offset_lockless
1792 #define p4d_offset_lockless(pgdp, pgd, address) p4d_offset(&(pgd), address)
1793 #endif
1794 #ifndef pud_offset_lockless
1795 #define pud_offset_lockless(p4dp, p4d, address) pud_offset(&(p4d), address)
1796 #endif
1797 #ifndef pmd_offset_lockless
1798 #define pmd_offset_lockless(pudp, pud, address) pmd_offset(&(pud), address)
1799 #endif
1800 
1801 /*
1802  * pXd_leaf() is the API to check whether a pgtable entry is a huge page
1803  * mapping.  It should work globally across all archs, without any
1804  * dependency on CONFIG_* options.  For architectures that do not support
1805  * huge mappings on specific levels, below fallbacks will be used.
1806  *
1807  * A leaf pgtable entry should always imply the following:
1808  *
1809  * - It is a "present" entry.  IOW, before using this API, please check it
1810  *   with pXd_present() first. NOTE: it may not always mean the "present
1811  *   bit" is set.  For example, PROT_NONE entries are always "present".
1812  *
1813  * - It should _never_ be a swap entry of any type.  Above "present" check
1814  *   should have guarded this, but let's be crystal clear on this.
1815  *
1816  * - It should contain a huge PFN, which points to a huge page larger than
1817  *   PAGE_SIZE of the platform.  The PFN format isn't important here.
1818  *
1819  * - It should cover all kinds of huge mappings (e.g., pXd_trans_huge(),
1820  *   pXd_devmap(), or hugetlb mappings).
1821  */
1822 #ifndef pgd_leaf
1823 #define pgd_leaf(x)	false
1824 #endif
1825 #ifndef p4d_leaf
1826 #define p4d_leaf(x)	false
1827 #endif
1828 #ifndef pud_leaf
1829 #define pud_leaf(x)	false
1830 #endif
1831 #ifndef pmd_leaf
1832 #define pmd_leaf(x)	false
1833 #endif
1834 
1835 #ifndef pgd_leaf_size
1836 #define pgd_leaf_size(x) (1ULL << PGDIR_SHIFT)
1837 #endif
1838 #ifndef p4d_leaf_size
1839 #define p4d_leaf_size(x) P4D_SIZE
1840 #endif
1841 #ifndef pud_leaf_size
1842 #define pud_leaf_size(x) PUD_SIZE
1843 #endif
1844 #ifndef pmd_leaf_size
1845 #define pmd_leaf_size(x) PMD_SIZE
1846 #endif
1847 #ifndef pte_leaf_size
1848 #define pte_leaf_size(x) PAGE_SIZE
1849 #endif
1850 
1851 /*
1852  * We always define pmd_pfn for all archs as it's used in lots of generic
1853  * code.  Now it happens too for pud_pfn (and can happen for larger
1854  * mappings too in the future; we're not there yet).  Instead of defining
1855  * it for all archs (like pmd_pfn), provide a fallback.
1856  *
1857  * Note that returning 0 here means any arch that didn't define this can
1858  * get severely wrong when it hits a real pud leaf.  It's arch's
1859  * responsibility to properly define it when a huge pud is possible.
1860  */
1861 #ifndef pud_pfn
1862 #define pud_pfn(x) 0
1863 #endif
1864 
1865 /*
1866  * Some architectures have MMUs that are configurable or selectable at boot
1867  * time. These lead to variable PTRS_PER_x. For statically allocated arrays it
1868  * helps to have a static maximum value.
1869  */
1870 
1871 #ifndef MAX_PTRS_PER_PTE
1872 #define MAX_PTRS_PER_PTE PTRS_PER_PTE
1873 #endif
1874 
1875 #ifndef MAX_PTRS_PER_PMD
1876 #define MAX_PTRS_PER_PMD PTRS_PER_PMD
1877 #endif
1878 
1879 #ifndef MAX_PTRS_PER_PUD
1880 #define MAX_PTRS_PER_PUD PTRS_PER_PUD
1881 #endif
1882 
1883 #ifndef MAX_PTRS_PER_P4D
1884 #define MAX_PTRS_PER_P4D PTRS_PER_P4D
1885 #endif
1886 
1887 /* description of effects of mapping type and prot in current implementation.
1888  * this is due to the limited x86 page protection hardware.  The expected
1889  * behavior is in parens:
1890  *
1891  * map_type	prot
1892  *		PROT_NONE	PROT_READ	PROT_WRITE	PROT_EXEC
1893  * MAP_SHARED	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1894  *		w: (no) no	w: (no) no	w: (yes) yes	w: (no) no
1895  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1896  *
1897  * MAP_PRIVATE	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1898  *		w: (no) no	w: (no) no	w: (copy) copy	w: (no) no
1899  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1900  *
1901  * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
1902  * MAP_PRIVATE (with Enhanced PAN supported):
1903  *								r: (no) no
1904  *								w: (no) no
1905  *								x: (yes) yes
1906  */
1907 #define DECLARE_VM_GET_PAGE_PROT					\
1908 pgprot_t vm_get_page_prot(unsigned long vm_flags)			\
1909 {									\
1910 		return protection_map[vm_flags &			\
1911 			(VM_READ | VM_WRITE | VM_EXEC | VM_SHARED)];	\
1912 }									\
1913 EXPORT_SYMBOL(vm_get_page_prot);
1914 
1915 #endif /* _LINUX_PGTABLE_H */
1916