xref: /linux-6.15/include/linux/pgtable.h (revision c074e146)
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 mkold_ptes
365 /**
366  * mkold_ptes - Mark PTEs that map consecutive pages of the same folio as old.
367  * @vma: VMA the pages are mapped into.
368  * @addr: Address the first page is mapped at.
369  * @ptep: Page table pointer for the first entry.
370  * @nr: Number of entries to mark old.
371  *
372  * May be overridden by the architecture; otherwise, implemented as a simple
373  * loop over ptep_test_and_clear_young().
374  *
375  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
376  * some PTEs might be write-protected.
377  *
378  * Context: The caller holds the page table lock.  The PTEs map consecutive
379  * pages that belong to the same folio.  The PTEs are all in the same PMD.
380  */
381 static inline void mkold_ptes(struct vm_area_struct *vma, unsigned long addr,
382 		pte_t *ptep, unsigned int nr)
383 {
384 	for (;;) {
385 		ptep_test_and_clear_young(vma, addr, ptep);
386 		if (--nr == 0)
387 			break;
388 		ptep++;
389 		addr += PAGE_SIZE;
390 	}
391 }
392 #endif
393 
394 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
395 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG)
396 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
397 					    unsigned long address,
398 					    pmd_t *pmdp)
399 {
400 	pmd_t pmd = *pmdp;
401 	int r = 1;
402 	if (!pmd_young(pmd))
403 		r = 0;
404 	else
405 		set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
406 	return r;
407 }
408 #else
409 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
410 					    unsigned long address,
411 					    pmd_t *pmdp)
412 {
413 	BUILD_BUG();
414 	return 0;
415 }
416 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG */
417 #endif
418 
419 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
420 int ptep_clear_flush_young(struct vm_area_struct *vma,
421 			   unsigned long address, pte_t *ptep);
422 #endif
423 
424 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
425 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
426 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
427 				  unsigned long address, pmd_t *pmdp);
428 #else
429 /*
430  * Despite relevant to THP only, this API is called from generic rmap code
431  * under PageTransHuge(), hence needs a dummy implementation for !THP
432  */
433 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
434 					 unsigned long address, pmd_t *pmdp)
435 {
436 	BUILD_BUG();
437 	return 0;
438 }
439 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
440 #endif
441 
442 #ifndef arch_has_hw_nonleaf_pmd_young
443 /*
444  * Return whether the accessed bit in non-leaf PMD entries is supported on the
445  * local CPU.
446  */
447 static inline bool arch_has_hw_nonleaf_pmd_young(void)
448 {
449 	return IS_ENABLED(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG);
450 }
451 #endif
452 
453 #ifndef arch_has_hw_pte_young
454 /*
455  * Return whether the accessed bit is supported on the local CPU.
456  *
457  * This stub assumes accessing through an old PTE triggers a page fault.
458  * Architectures that automatically set the access bit should overwrite it.
459  */
460 static inline bool arch_has_hw_pte_young(void)
461 {
462 	return IS_ENABLED(CONFIG_ARCH_HAS_HW_PTE_YOUNG);
463 }
464 #endif
465 
466 #ifndef arch_check_zapped_pte
467 static inline void arch_check_zapped_pte(struct vm_area_struct *vma,
468 					 pte_t pte)
469 {
470 }
471 #endif
472 
473 #ifndef arch_check_zapped_pmd
474 static inline void arch_check_zapped_pmd(struct vm_area_struct *vma,
475 					 pmd_t pmd)
476 {
477 }
478 #endif
479 
480 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
481 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
482 				       unsigned long address,
483 				       pte_t *ptep)
484 {
485 	pte_t pte = ptep_get(ptep);
486 	pte_clear(mm, address, ptep);
487 	page_table_check_pte_clear(mm, pte);
488 	return pte;
489 }
490 #endif
491 
492 static inline void ptep_clear(struct mm_struct *mm, unsigned long addr,
493 			      pte_t *ptep)
494 {
495 	ptep_get_and_clear(mm, addr, ptep);
496 }
497 
498 #ifdef CONFIG_GUP_GET_PXX_LOW_HIGH
499 /*
500  * For walking the pagetables without holding any locks.  Some architectures
501  * (eg x86-32 PAE) cannot load the entries atomically without using expensive
502  * instructions.  We are guaranteed that a PTE will only either go from not
503  * present to present, or present to not present -- it will not switch to a
504  * completely different present page without a TLB flush inbetween; which we
505  * are blocking by holding interrupts off.
506  *
507  * Setting ptes from not present to present goes:
508  *
509  *   ptep->pte_high = h;
510  *   smp_wmb();
511  *   ptep->pte_low = l;
512  *
513  * And present to not present goes:
514  *
515  *   ptep->pte_low = 0;
516  *   smp_wmb();
517  *   ptep->pte_high = 0;
518  *
519  * We must ensure here that the load of pte_low sees 'l' IFF pte_high sees 'h'.
520  * We load pte_high *after* loading pte_low, which ensures we don't see an older
521  * value of pte_high.  *Then* we recheck pte_low, which ensures that we haven't
522  * picked up a changed pte high. We might have gotten rubbish values from
523  * pte_low and pte_high, but we are guaranteed that pte_low will not have the
524  * present bit set *unless* it is 'l'. Because get_user_pages_fast() only
525  * operates on present ptes we're safe.
526  */
527 static inline pte_t ptep_get_lockless(pte_t *ptep)
528 {
529 	pte_t pte;
530 
531 	do {
532 		pte.pte_low = ptep->pte_low;
533 		smp_rmb();
534 		pte.pte_high = ptep->pte_high;
535 		smp_rmb();
536 	} while (unlikely(pte.pte_low != ptep->pte_low));
537 
538 	return pte;
539 }
540 #define ptep_get_lockless ptep_get_lockless
541 
542 #if CONFIG_PGTABLE_LEVELS > 2
543 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
544 {
545 	pmd_t pmd;
546 
547 	do {
548 		pmd.pmd_low = pmdp->pmd_low;
549 		smp_rmb();
550 		pmd.pmd_high = pmdp->pmd_high;
551 		smp_rmb();
552 	} while (unlikely(pmd.pmd_low != pmdp->pmd_low));
553 
554 	return pmd;
555 }
556 #define pmdp_get_lockless pmdp_get_lockless
557 #define pmdp_get_lockless_sync() tlb_remove_table_sync_one()
558 #endif /* CONFIG_PGTABLE_LEVELS > 2 */
559 #endif /* CONFIG_GUP_GET_PXX_LOW_HIGH */
560 
561 /*
562  * We require that the PTE can be read atomically.
563  */
564 #ifndef ptep_get_lockless
565 static inline pte_t ptep_get_lockless(pte_t *ptep)
566 {
567 	return ptep_get(ptep);
568 }
569 #endif
570 
571 #ifndef pmdp_get_lockless
572 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
573 {
574 	return pmdp_get(pmdp);
575 }
576 static inline void pmdp_get_lockless_sync(void)
577 {
578 }
579 #endif
580 
581 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
582 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
583 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
584 					    unsigned long address,
585 					    pmd_t *pmdp)
586 {
587 	pmd_t pmd = *pmdp;
588 
589 	pmd_clear(pmdp);
590 	page_table_check_pmd_clear(mm, pmd);
591 
592 	return pmd;
593 }
594 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
595 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
596 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
597 					    unsigned long address,
598 					    pud_t *pudp)
599 {
600 	pud_t pud = *pudp;
601 
602 	pud_clear(pudp);
603 	page_table_check_pud_clear(mm, pud);
604 
605 	return pud;
606 }
607 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
608 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
609 
610 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
611 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
612 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
613 					    unsigned long address, pmd_t *pmdp,
614 					    int full)
615 {
616 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
617 }
618 #endif
619 
620 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
621 static inline pud_t pudp_huge_get_and_clear_full(struct vm_area_struct *vma,
622 					    unsigned long address, pud_t *pudp,
623 					    int full)
624 {
625 	return pudp_huge_get_and_clear(vma->vm_mm, address, pudp);
626 }
627 #endif
628 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
629 
630 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
631 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
632 					    unsigned long address, pte_t *ptep,
633 					    int full)
634 {
635 	return ptep_get_and_clear(mm, address, ptep);
636 }
637 #endif
638 
639 #ifndef get_and_clear_full_ptes
640 /**
641  * get_and_clear_full_ptes - Clear present PTEs that map consecutive pages of
642  *			     the same folio, collecting dirty/accessed bits.
643  * @mm: Address space the pages are mapped into.
644  * @addr: Address the first page is mapped at.
645  * @ptep: Page table pointer for the first entry.
646  * @nr: Number of entries to clear.
647  * @full: Whether we are clearing a full mm.
648  *
649  * May be overridden by the architecture; otherwise, implemented as a simple
650  * loop over ptep_get_and_clear_full(), merging dirty/accessed bits into the
651  * returned PTE.
652  *
653  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
654  * some PTEs might be write-protected.
655  *
656  * Context: The caller holds the page table lock.  The PTEs map consecutive
657  * pages that belong to the same folio.  The PTEs are all in the same PMD.
658  */
659 static inline pte_t get_and_clear_full_ptes(struct mm_struct *mm,
660 		unsigned long addr, pte_t *ptep, unsigned int nr, int full)
661 {
662 	pte_t pte, tmp_pte;
663 
664 	pte = ptep_get_and_clear_full(mm, addr, ptep, full);
665 	while (--nr) {
666 		ptep++;
667 		addr += PAGE_SIZE;
668 		tmp_pte = ptep_get_and_clear_full(mm, addr, ptep, full);
669 		if (pte_dirty(tmp_pte))
670 			pte = pte_mkdirty(pte);
671 		if (pte_young(tmp_pte))
672 			pte = pte_mkyoung(pte);
673 	}
674 	return pte;
675 }
676 #endif
677 
678 #ifndef clear_full_ptes
679 /**
680  * clear_full_ptes - Clear present PTEs that map consecutive pages of the same
681  *		     folio.
682  * @mm: Address space the pages are mapped into.
683  * @addr: Address the first page is mapped at.
684  * @ptep: Page table pointer for the first entry.
685  * @nr: Number of entries to clear.
686  * @full: Whether we are clearing a full mm.
687  *
688  * May be overridden by the architecture; otherwise, implemented as a simple
689  * loop over ptep_get_and_clear_full().
690  *
691  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
692  * some PTEs might be write-protected.
693  *
694  * Context: The caller holds the page table lock.  The PTEs map consecutive
695  * pages that belong to the same folio.  The PTEs are all in the same PMD.
696  */
697 static inline void clear_full_ptes(struct mm_struct *mm, unsigned long addr,
698 		pte_t *ptep, unsigned int nr, int full)
699 {
700 	for (;;) {
701 		ptep_get_and_clear_full(mm, addr, ptep, full);
702 		if (--nr == 0)
703 			break;
704 		ptep++;
705 		addr += PAGE_SIZE;
706 	}
707 }
708 #endif
709 
710 /*
711  * If two threads concurrently fault at the same page, the thread that
712  * won the race updates the PTE and its local TLB/Cache. The other thread
713  * gives up, simply does nothing, and continues; on architectures where
714  * software can update TLB,  local TLB can be updated here to avoid next page
715  * fault. This function updates TLB only, do nothing with cache or others.
716  * It is the difference with function update_mmu_cache.
717  */
718 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
719 static inline void update_mmu_tlb(struct vm_area_struct *vma,
720 				unsigned long address, pte_t *ptep)
721 {
722 }
723 #define __HAVE_ARCH_UPDATE_MMU_TLB
724 #endif
725 
726 /*
727  * Some architectures may be able to avoid expensive synchronization
728  * primitives when modifications are made to PTE's which are already
729  * not present, or in the process of an address space destruction.
730  */
731 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
732 static inline void pte_clear_not_present_full(struct mm_struct *mm,
733 					      unsigned long address,
734 					      pte_t *ptep,
735 					      int full)
736 {
737 	pte_clear(mm, address, ptep);
738 }
739 #endif
740 
741 #ifndef clear_not_present_full_ptes
742 /**
743  * clear_not_present_full_ptes - Clear multiple not present PTEs which are
744  *				 consecutive in the pgtable.
745  * @mm: Address space the ptes represent.
746  * @addr: Address of the first pte.
747  * @ptep: Page table pointer for the first entry.
748  * @nr: Number of entries to clear.
749  * @full: Whether we are clearing a full mm.
750  *
751  * May be overridden by the architecture; otherwise, implemented as a simple
752  * loop over pte_clear_not_present_full().
753  *
754  * Context: The caller holds the page table lock.  The PTEs are all not present.
755  * The PTEs are all in the same PMD.
756  */
757 static inline void clear_not_present_full_ptes(struct mm_struct *mm,
758 		unsigned long addr, pte_t *ptep, unsigned int nr, int full)
759 {
760 	for (;;) {
761 		pte_clear_not_present_full(mm, addr, ptep, full);
762 		if (--nr == 0)
763 			break;
764 		ptep++;
765 		addr += PAGE_SIZE;
766 	}
767 }
768 #endif
769 
770 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
771 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
772 			      unsigned long address,
773 			      pte_t *ptep);
774 #endif
775 
776 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
777 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
778 			      unsigned long address,
779 			      pmd_t *pmdp);
780 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
781 			      unsigned long address,
782 			      pud_t *pudp);
783 #endif
784 
785 #ifndef pte_mkwrite
786 static inline pte_t pte_mkwrite(pte_t pte, struct vm_area_struct *vma)
787 {
788 	return pte_mkwrite_novma(pte);
789 }
790 #endif
791 
792 #if defined(CONFIG_ARCH_WANT_PMD_MKWRITE) && !defined(pmd_mkwrite)
793 static inline pmd_t pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma)
794 {
795 	return pmd_mkwrite_novma(pmd);
796 }
797 #endif
798 
799 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
800 struct mm_struct;
801 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
802 {
803 	pte_t old_pte = ptep_get(ptep);
804 	set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
805 }
806 #endif
807 
808 #ifndef wrprotect_ptes
809 /**
810  * wrprotect_ptes - Write-protect PTEs that map consecutive pages of the same
811  *		    folio.
812  * @mm: Address space the pages are mapped into.
813  * @addr: Address the first page is mapped at.
814  * @ptep: Page table pointer for the first entry.
815  * @nr: Number of entries to write-protect.
816  *
817  * May be overridden by the architecture; otherwise, implemented as a simple
818  * loop over ptep_set_wrprotect().
819  *
820  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
821  * some PTEs might be write-protected.
822  *
823  * Context: The caller holds the page table lock.  The PTEs map consecutive
824  * pages that belong to the same folio.  The PTEs are all in the same PMD.
825  */
826 static inline void wrprotect_ptes(struct mm_struct *mm, unsigned long addr,
827 		pte_t *ptep, unsigned int nr)
828 {
829 	for (;;) {
830 		ptep_set_wrprotect(mm, addr, ptep);
831 		if (--nr == 0)
832 			break;
833 		ptep++;
834 		addr += PAGE_SIZE;
835 	}
836 }
837 #endif
838 
839 /*
840  * On some architectures hardware does not set page access bit when accessing
841  * memory page, it is responsibility of software setting this bit. It brings
842  * out extra page fault penalty to track page access bit. For optimization page
843  * access bit can be set during all page fault flow on these arches.
844  * To be differentiate with macro pte_mkyoung, this macro is used on platforms
845  * where software maintains page access bit.
846  */
847 #ifndef pte_sw_mkyoung
848 static inline pte_t pte_sw_mkyoung(pte_t pte)
849 {
850 	return pte;
851 }
852 #define pte_sw_mkyoung	pte_sw_mkyoung
853 #endif
854 
855 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
856 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
857 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
858 				      unsigned long address, pmd_t *pmdp)
859 {
860 	pmd_t old_pmd = *pmdp;
861 	set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
862 }
863 #else
864 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
865 				      unsigned long address, pmd_t *pmdp)
866 {
867 	BUILD_BUG();
868 }
869 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
870 #endif
871 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
872 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
873 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
874 static inline void pudp_set_wrprotect(struct mm_struct *mm,
875 				      unsigned long address, pud_t *pudp)
876 {
877 	pud_t old_pud = *pudp;
878 
879 	set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
880 }
881 #else
882 static inline void pudp_set_wrprotect(struct mm_struct *mm,
883 				      unsigned long address, pud_t *pudp)
884 {
885 	BUILD_BUG();
886 }
887 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
888 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
889 #endif
890 
891 #ifndef pmdp_collapse_flush
892 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
893 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
894 				 unsigned long address, pmd_t *pmdp);
895 #else
896 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
897 					unsigned long address,
898 					pmd_t *pmdp)
899 {
900 	BUILD_BUG();
901 	return *pmdp;
902 }
903 #define pmdp_collapse_flush pmdp_collapse_flush
904 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
905 #endif
906 
907 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
908 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
909 				       pgtable_t pgtable);
910 #endif
911 
912 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
913 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
914 #endif
915 
916 #ifndef arch_needs_pgtable_deposit
917 #define arch_needs_pgtable_deposit() (false)
918 #endif
919 
920 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
921 /*
922  * This is an implementation of pmdp_establish() that is only suitable for an
923  * architecture that doesn't have hardware dirty/accessed bits. In this case we
924  * can't race with CPU which sets these bits and non-atomic approach is fine.
925  */
926 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
927 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
928 {
929 	pmd_t old_pmd = *pmdp;
930 	set_pmd_at(vma->vm_mm, address, pmdp, pmd);
931 	return old_pmd;
932 }
933 #endif
934 
935 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
936 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
937 			    pmd_t *pmdp);
938 #endif
939 
940 #ifndef __HAVE_ARCH_PMDP_INVALIDATE_AD
941 
942 /*
943  * pmdp_invalidate_ad() invalidates the PMD while changing a transparent
944  * hugepage mapping in the page tables. This function is similar to
945  * pmdp_invalidate(), but should only be used if the access and dirty bits would
946  * not be cleared by the software in the new PMD value. The function ensures
947  * that hardware changes of the access and dirty bits updates would not be lost.
948  *
949  * Doing so can allow in certain architectures to avoid a TLB flush in most
950  * cases. Yet, another TLB flush might be necessary later if the PMD update
951  * itself requires such flush (e.g., if protection was set to be stricter). Yet,
952  * even when a TLB flush is needed because of the update, the caller may be able
953  * to batch these TLB flushing operations, so fewer TLB flush operations are
954  * needed.
955  */
956 extern pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma,
957 				unsigned long address, pmd_t *pmdp);
958 #endif
959 
960 #ifndef __HAVE_ARCH_PTE_SAME
961 static inline int pte_same(pte_t pte_a, pte_t pte_b)
962 {
963 	return pte_val(pte_a) == pte_val(pte_b);
964 }
965 #endif
966 
967 #ifndef __HAVE_ARCH_PTE_UNUSED
968 /*
969  * Some architectures provide facilities to virtualization guests
970  * so that they can flag allocated pages as unused. This allows the
971  * host to transparently reclaim unused pages. This function returns
972  * whether the pte's page is unused.
973  */
974 static inline int pte_unused(pte_t pte)
975 {
976 	return 0;
977 }
978 #endif
979 
980 #ifndef pte_access_permitted
981 #define pte_access_permitted(pte, write) \
982 	(pte_present(pte) && (!(write) || pte_write(pte)))
983 #endif
984 
985 #ifndef pmd_access_permitted
986 #define pmd_access_permitted(pmd, write) \
987 	(pmd_present(pmd) && (!(write) || pmd_write(pmd)))
988 #endif
989 
990 #ifndef pud_access_permitted
991 #define pud_access_permitted(pud, write) \
992 	(pud_present(pud) && (!(write) || pud_write(pud)))
993 #endif
994 
995 #ifndef p4d_access_permitted
996 #define p4d_access_permitted(p4d, write) \
997 	(p4d_present(p4d) && (!(write) || p4d_write(p4d)))
998 #endif
999 
1000 #ifndef pgd_access_permitted
1001 #define pgd_access_permitted(pgd, write) \
1002 	(pgd_present(pgd) && (!(write) || pgd_write(pgd)))
1003 #endif
1004 
1005 #ifndef __HAVE_ARCH_PMD_SAME
1006 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
1007 {
1008 	return pmd_val(pmd_a) == pmd_val(pmd_b);
1009 }
1010 #endif
1011 
1012 #ifndef pud_same
1013 static inline int pud_same(pud_t pud_a, pud_t pud_b)
1014 {
1015 	return pud_val(pud_a) == pud_val(pud_b);
1016 }
1017 #define pud_same pud_same
1018 #endif
1019 
1020 #ifndef __HAVE_ARCH_P4D_SAME
1021 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
1022 {
1023 	return p4d_val(p4d_a) == p4d_val(p4d_b);
1024 }
1025 #endif
1026 
1027 #ifndef __HAVE_ARCH_PGD_SAME
1028 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
1029 {
1030 	return pgd_val(pgd_a) == pgd_val(pgd_b);
1031 }
1032 #endif
1033 
1034 /*
1035  * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
1036  * TLB flush will be required as a result of the "set". For example, use
1037  * in scenarios where it is known ahead of time that the routine is
1038  * setting non-present entries, or re-setting an existing entry to the
1039  * same value. Otherwise, use the typical "set" helpers and flush the
1040  * TLB.
1041  */
1042 #define set_pte_safe(ptep, pte) \
1043 ({ \
1044 	WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
1045 	set_pte(ptep, pte); \
1046 })
1047 
1048 #define set_pmd_safe(pmdp, pmd) \
1049 ({ \
1050 	WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
1051 	set_pmd(pmdp, pmd); \
1052 })
1053 
1054 #define set_pud_safe(pudp, pud) \
1055 ({ \
1056 	WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
1057 	set_pud(pudp, pud); \
1058 })
1059 
1060 #define set_p4d_safe(p4dp, p4d) \
1061 ({ \
1062 	WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
1063 	set_p4d(p4dp, p4d); \
1064 })
1065 
1066 #define set_pgd_safe(pgdp, pgd) \
1067 ({ \
1068 	WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
1069 	set_pgd(pgdp, pgd); \
1070 })
1071 
1072 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
1073 /*
1074  * Some architectures support metadata associated with a page. When a
1075  * page is being swapped out, this metadata must be saved so it can be
1076  * restored when the page is swapped back in. SPARC M7 and newer
1077  * processors support an ADI (Application Data Integrity) tag for the
1078  * page as metadata for the page. arch_do_swap_page() can restore this
1079  * metadata when a page is swapped back in.
1080  */
1081 static inline void arch_do_swap_page(struct mm_struct *mm,
1082 				     struct vm_area_struct *vma,
1083 				     unsigned long addr,
1084 				     pte_t pte, pte_t oldpte)
1085 {
1086 
1087 }
1088 #endif
1089 
1090 #ifndef __HAVE_ARCH_UNMAP_ONE
1091 /*
1092  * Some architectures support metadata associated with a page. When a
1093  * page is being swapped out, this metadata must be saved so it can be
1094  * restored when the page is swapped back in. SPARC M7 and newer
1095  * processors support an ADI (Application Data Integrity) tag for the
1096  * page as metadata for the page. arch_unmap_one() can save this
1097  * metadata on a swap-out of a page.
1098  */
1099 static inline int arch_unmap_one(struct mm_struct *mm,
1100 				  struct vm_area_struct *vma,
1101 				  unsigned long addr,
1102 				  pte_t orig_pte)
1103 {
1104 	return 0;
1105 }
1106 #endif
1107 
1108 /*
1109  * Allow architectures to preserve additional metadata associated with
1110  * swapped-out pages. The corresponding __HAVE_ARCH_SWAP_* macros and function
1111  * prototypes must be defined in the arch-specific asm/pgtable.h file.
1112  */
1113 #ifndef __HAVE_ARCH_PREPARE_TO_SWAP
1114 static inline int arch_prepare_to_swap(struct folio *folio)
1115 {
1116 	return 0;
1117 }
1118 #endif
1119 
1120 #ifndef __HAVE_ARCH_SWAP_INVALIDATE
1121 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
1122 {
1123 }
1124 
1125 static inline void arch_swap_invalidate_area(int type)
1126 {
1127 }
1128 #endif
1129 
1130 #ifndef __HAVE_ARCH_SWAP_RESTORE
1131 static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
1132 {
1133 }
1134 #endif
1135 
1136 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
1137 #define pgd_offset_gate(mm, addr)	pgd_offset(mm, addr)
1138 #endif
1139 
1140 #ifndef __HAVE_ARCH_MOVE_PTE
1141 #define move_pte(pte, old_addr, new_addr)	(pte)
1142 #endif
1143 
1144 #ifndef pte_accessible
1145 # define pte_accessible(mm, pte)	((void)(pte), 1)
1146 #endif
1147 
1148 #ifndef flush_tlb_fix_spurious_fault
1149 #define flush_tlb_fix_spurious_fault(vma, address, ptep) flush_tlb_page(vma, address)
1150 #endif
1151 
1152 /*
1153  * When walking page tables, get the address of the next boundary,
1154  * or the end address of the range if that comes earlier.  Although no
1155  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
1156  */
1157 
1158 #define pgd_addr_end(addr, end)						\
1159 ({	unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;	\
1160 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1161 })
1162 
1163 #ifndef p4d_addr_end
1164 #define p4d_addr_end(addr, end)						\
1165 ({	unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;	\
1166 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1167 })
1168 #endif
1169 
1170 #ifndef pud_addr_end
1171 #define pud_addr_end(addr, end)						\
1172 ({	unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;	\
1173 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1174 })
1175 #endif
1176 
1177 #ifndef pmd_addr_end
1178 #define pmd_addr_end(addr, end)						\
1179 ({	unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;	\
1180 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1181 })
1182 #endif
1183 
1184 /*
1185  * When walking page tables, we usually want to skip any p?d_none entries;
1186  * and any p?d_bad entries - reporting the error before resetting to none.
1187  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
1188  */
1189 void pgd_clear_bad(pgd_t *);
1190 
1191 #ifndef __PAGETABLE_P4D_FOLDED
1192 void p4d_clear_bad(p4d_t *);
1193 #else
1194 #define p4d_clear_bad(p4d)        do { } while (0)
1195 #endif
1196 
1197 #ifndef __PAGETABLE_PUD_FOLDED
1198 void pud_clear_bad(pud_t *);
1199 #else
1200 #define pud_clear_bad(p4d)        do { } while (0)
1201 #endif
1202 
1203 void pmd_clear_bad(pmd_t *);
1204 
1205 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
1206 {
1207 	if (pgd_none(*pgd))
1208 		return 1;
1209 	if (unlikely(pgd_bad(*pgd))) {
1210 		pgd_clear_bad(pgd);
1211 		return 1;
1212 	}
1213 	return 0;
1214 }
1215 
1216 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
1217 {
1218 	if (p4d_none(*p4d))
1219 		return 1;
1220 	if (unlikely(p4d_bad(*p4d))) {
1221 		p4d_clear_bad(p4d);
1222 		return 1;
1223 	}
1224 	return 0;
1225 }
1226 
1227 static inline int pud_none_or_clear_bad(pud_t *pud)
1228 {
1229 	if (pud_none(*pud))
1230 		return 1;
1231 	if (unlikely(pud_bad(*pud))) {
1232 		pud_clear_bad(pud);
1233 		return 1;
1234 	}
1235 	return 0;
1236 }
1237 
1238 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
1239 {
1240 	if (pmd_none(*pmd))
1241 		return 1;
1242 	if (unlikely(pmd_bad(*pmd))) {
1243 		pmd_clear_bad(pmd);
1244 		return 1;
1245 	}
1246 	return 0;
1247 }
1248 
1249 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
1250 					     unsigned long addr,
1251 					     pte_t *ptep)
1252 {
1253 	/*
1254 	 * Get the current pte state, but zero it out to make it
1255 	 * non-present, preventing the hardware from asynchronously
1256 	 * updating it.
1257 	 */
1258 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
1259 }
1260 
1261 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
1262 					     unsigned long addr,
1263 					     pte_t *ptep, pte_t pte)
1264 {
1265 	/*
1266 	 * The pte is non-present, so there's no hardware state to
1267 	 * preserve.
1268 	 */
1269 	set_pte_at(vma->vm_mm, addr, ptep, pte);
1270 }
1271 
1272 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1273 /*
1274  * Start a pte protection read-modify-write transaction, which
1275  * protects against asynchronous hardware modifications to the pte.
1276  * The intention is not to prevent the hardware from making pte
1277  * updates, but to prevent any updates it may make from being lost.
1278  *
1279  * This does not protect against other software modifications of the
1280  * pte; the appropriate pte lock must be held over the transaction.
1281  *
1282  * Note that this interface is intended to be batchable, meaning that
1283  * ptep_modify_prot_commit may not actually update the pte, but merely
1284  * queue the update to be done at some later time.  The update must be
1285  * actually committed before the pte lock is released, however.
1286  */
1287 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
1288 					   unsigned long addr,
1289 					   pte_t *ptep)
1290 {
1291 	return __ptep_modify_prot_start(vma, addr, ptep);
1292 }
1293 
1294 /*
1295  * Commit an update to a pte, leaving any hardware-controlled bits in
1296  * the PTE unmodified.
1297  */
1298 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
1299 					   unsigned long addr,
1300 					   pte_t *ptep, pte_t old_pte, pte_t pte)
1301 {
1302 	__ptep_modify_prot_commit(vma, addr, ptep, pte);
1303 }
1304 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
1305 #endif /* CONFIG_MMU */
1306 
1307 /*
1308  * No-op macros that just return the current protection value. Defined here
1309  * because these macros can be used even if CONFIG_MMU is not defined.
1310  */
1311 
1312 #ifndef pgprot_nx
1313 #define pgprot_nx(prot)	(prot)
1314 #endif
1315 
1316 #ifndef pgprot_noncached
1317 #define pgprot_noncached(prot)	(prot)
1318 #endif
1319 
1320 #ifndef pgprot_writecombine
1321 #define pgprot_writecombine pgprot_noncached
1322 #endif
1323 
1324 #ifndef pgprot_writethrough
1325 #define pgprot_writethrough pgprot_noncached
1326 #endif
1327 
1328 #ifndef pgprot_device
1329 #define pgprot_device pgprot_noncached
1330 #endif
1331 
1332 #ifndef pgprot_mhp
1333 #define pgprot_mhp(prot)	(prot)
1334 #endif
1335 
1336 #ifdef CONFIG_MMU
1337 #ifndef pgprot_modify
1338 #define pgprot_modify pgprot_modify
1339 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
1340 {
1341 	if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
1342 		newprot = pgprot_noncached(newprot);
1343 	if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
1344 		newprot = pgprot_writecombine(newprot);
1345 	if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
1346 		newprot = pgprot_device(newprot);
1347 	return newprot;
1348 }
1349 #endif
1350 #endif /* CONFIG_MMU */
1351 
1352 #ifndef pgprot_encrypted
1353 #define pgprot_encrypted(prot)	(prot)
1354 #endif
1355 
1356 #ifndef pgprot_decrypted
1357 #define pgprot_decrypted(prot)	(prot)
1358 #endif
1359 
1360 /*
1361  * A facility to provide batching of the reload of page tables and
1362  * other process state with the actual context switch code for
1363  * paravirtualized guests.  By convention, only one of the batched
1364  * update (lazy) modes (CPU, MMU) should be active at any given time,
1365  * entry should never be nested, and entry and exits should always be
1366  * paired.  This is for sanity of maintaining and reasoning about the
1367  * kernel code.  In this case, the exit (end of the context switch) is
1368  * in architecture-specific code, and so doesn't need a generic
1369  * definition.
1370  */
1371 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
1372 #define arch_start_context_switch(prev)	do {} while (0)
1373 #endif
1374 
1375 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1376 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
1377 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1378 {
1379 	return pmd;
1380 }
1381 
1382 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1383 {
1384 	return 0;
1385 }
1386 
1387 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1388 {
1389 	return pmd;
1390 }
1391 #endif
1392 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
1393 static inline int pte_soft_dirty(pte_t pte)
1394 {
1395 	return 0;
1396 }
1397 
1398 static inline int pmd_soft_dirty(pmd_t pmd)
1399 {
1400 	return 0;
1401 }
1402 
1403 static inline pte_t pte_mksoft_dirty(pte_t pte)
1404 {
1405 	return pte;
1406 }
1407 
1408 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
1409 {
1410 	return pmd;
1411 }
1412 
1413 static inline pte_t pte_clear_soft_dirty(pte_t pte)
1414 {
1415 	return pte;
1416 }
1417 
1418 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
1419 {
1420 	return pmd;
1421 }
1422 
1423 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1424 {
1425 	return pte;
1426 }
1427 
1428 static inline int pte_swp_soft_dirty(pte_t pte)
1429 {
1430 	return 0;
1431 }
1432 
1433 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1434 {
1435 	return pte;
1436 }
1437 
1438 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1439 {
1440 	return pmd;
1441 }
1442 
1443 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1444 {
1445 	return 0;
1446 }
1447 
1448 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1449 {
1450 	return pmd;
1451 }
1452 #endif
1453 
1454 #ifndef __HAVE_PFNMAP_TRACKING
1455 /*
1456  * Interfaces that can be used by architecture code to keep track of
1457  * memory type of pfn mappings specified by the remap_pfn_range,
1458  * vmf_insert_pfn.
1459  */
1460 
1461 /*
1462  * track_pfn_remap is called when a _new_ pfn mapping is being established
1463  * by remap_pfn_range() for physical range indicated by pfn and size.
1464  */
1465 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1466 				  unsigned long pfn, unsigned long addr,
1467 				  unsigned long size)
1468 {
1469 	return 0;
1470 }
1471 
1472 /*
1473  * track_pfn_insert is called when a _new_ single pfn is established
1474  * by vmf_insert_pfn().
1475  */
1476 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1477 				    pfn_t pfn)
1478 {
1479 }
1480 
1481 /*
1482  * track_pfn_copy is called when vma that is covering the pfnmap gets
1483  * copied through copy_page_range().
1484  */
1485 static inline int track_pfn_copy(struct vm_area_struct *vma)
1486 {
1487 	return 0;
1488 }
1489 
1490 /*
1491  * untrack_pfn is called while unmapping a pfnmap for a region.
1492  * untrack can be called for a specific region indicated by pfn and size or
1493  * can be for the entire vma (in which case pfn, size are zero).
1494  */
1495 static inline void untrack_pfn(struct vm_area_struct *vma,
1496 			       unsigned long pfn, unsigned long size,
1497 			       bool mm_wr_locked)
1498 {
1499 }
1500 
1501 /*
1502  * untrack_pfn_clear is called while mremapping a pfnmap for a new region
1503  * or fails to copy pgtable during duplicate vm area.
1504  */
1505 static inline void untrack_pfn_clear(struct vm_area_struct *vma)
1506 {
1507 }
1508 #else
1509 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1510 			   unsigned long pfn, unsigned long addr,
1511 			   unsigned long size);
1512 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1513 			     pfn_t pfn);
1514 extern int track_pfn_copy(struct vm_area_struct *vma);
1515 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1516 			unsigned long size, bool mm_wr_locked);
1517 extern void untrack_pfn_clear(struct vm_area_struct *vma);
1518 #endif
1519 
1520 #ifdef CONFIG_MMU
1521 #ifdef __HAVE_COLOR_ZERO_PAGE
1522 static inline int is_zero_pfn(unsigned long pfn)
1523 {
1524 	extern unsigned long zero_pfn;
1525 	unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1526 	return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1527 }
1528 
1529 #define my_zero_pfn(addr)	page_to_pfn(ZERO_PAGE(addr))
1530 
1531 #else
1532 static inline int is_zero_pfn(unsigned long pfn)
1533 {
1534 	extern unsigned long zero_pfn;
1535 	return pfn == zero_pfn;
1536 }
1537 
1538 static inline unsigned long my_zero_pfn(unsigned long addr)
1539 {
1540 	extern unsigned long zero_pfn;
1541 	return zero_pfn;
1542 }
1543 #endif
1544 #else
1545 static inline int is_zero_pfn(unsigned long pfn)
1546 {
1547 	return 0;
1548 }
1549 
1550 static inline unsigned long my_zero_pfn(unsigned long addr)
1551 {
1552 	return 0;
1553 }
1554 #endif /* CONFIG_MMU */
1555 
1556 #ifdef CONFIG_MMU
1557 
1558 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1559 static inline int pmd_trans_huge(pmd_t pmd)
1560 {
1561 	return 0;
1562 }
1563 #ifndef pmd_write
1564 static inline int pmd_write(pmd_t pmd)
1565 {
1566 	BUG();
1567 	return 0;
1568 }
1569 #endif /* pmd_write */
1570 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1571 
1572 #ifndef pud_write
1573 static inline int pud_write(pud_t pud)
1574 {
1575 	BUG();
1576 	return 0;
1577 }
1578 #endif /* pud_write */
1579 
1580 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1581 static inline int pmd_devmap(pmd_t pmd)
1582 {
1583 	return 0;
1584 }
1585 static inline int pud_devmap(pud_t pud)
1586 {
1587 	return 0;
1588 }
1589 static inline int pgd_devmap(pgd_t pgd)
1590 {
1591 	return 0;
1592 }
1593 #endif
1594 
1595 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1596 	!defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1597 static inline int pud_trans_huge(pud_t pud)
1598 {
1599 	return 0;
1600 }
1601 #endif
1602 
1603 static inline int pud_trans_unstable(pud_t *pud)
1604 {
1605 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1606 	defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1607 	pud_t pudval = READ_ONCE(*pud);
1608 
1609 	if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1610 		return 1;
1611 	if (unlikely(pud_bad(pudval))) {
1612 		pud_clear_bad(pud);
1613 		return 1;
1614 	}
1615 #endif
1616 	return 0;
1617 }
1618 
1619 #ifndef CONFIG_NUMA_BALANCING
1620 /*
1621  * In an inaccessible (PROT_NONE) VMA, pte_protnone() may indicate "yes". It is
1622  * perfectly valid to indicate "no" in that case, which is why our default
1623  * implementation defaults to "always no".
1624  *
1625  * In an accessible VMA, however, pte_protnone() reliably indicates PROT_NONE
1626  * page protection due to NUMA hinting. NUMA hinting faults only apply in
1627  * accessible VMAs.
1628  *
1629  * So, to reliably identify PROT_NONE PTEs that require a NUMA hinting fault,
1630  * looking at the VMA accessibility is sufficient.
1631  */
1632 static inline int pte_protnone(pte_t pte)
1633 {
1634 	return 0;
1635 }
1636 
1637 static inline int pmd_protnone(pmd_t pmd)
1638 {
1639 	return 0;
1640 }
1641 #endif /* CONFIG_NUMA_BALANCING */
1642 
1643 #endif /* CONFIG_MMU */
1644 
1645 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1646 
1647 #ifndef __PAGETABLE_P4D_FOLDED
1648 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1649 void p4d_clear_huge(p4d_t *p4d);
1650 #else
1651 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1652 {
1653 	return 0;
1654 }
1655 static inline void p4d_clear_huge(p4d_t *p4d) { }
1656 #endif /* !__PAGETABLE_P4D_FOLDED */
1657 
1658 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1659 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1660 int pud_clear_huge(pud_t *pud);
1661 int pmd_clear_huge(pmd_t *pmd);
1662 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1663 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1664 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1665 #else	/* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1666 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1667 {
1668 	return 0;
1669 }
1670 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1671 {
1672 	return 0;
1673 }
1674 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1675 {
1676 	return 0;
1677 }
1678 static inline void p4d_clear_huge(p4d_t *p4d) { }
1679 static inline int pud_clear_huge(pud_t *pud)
1680 {
1681 	return 0;
1682 }
1683 static inline int pmd_clear_huge(pmd_t *pmd)
1684 {
1685 	return 0;
1686 }
1687 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1688 {
1689 	return 0;
1690 }
1691 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1692 {
1693 	return 0;
1694 }
1695 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1696 {
1697 	return 0;
1698 }
1699 #endif	/* CONFIG_HAVE_ARCH_HUGE_VMAP */
1700 
1701 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1702 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1703 /*
1704  * ARCHes with special requirements for evicting THP backing TLB entries can
1705  * implement this. Otherwise also, it can help optimize normal TLB flush in
1706  * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1707  * entire TLB if flush span is greater than a threshold, which will
1708  * likely be true for a single huge page. Thus a single THP flush will
1709  * invalidate the entire TLB which is not desirable.
1710  * e.g. see arch/arc: flush_pmd_tlb_range
1711  */
1712 #define flush_pmd_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1713 #define flush_pud_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1714 #else
1715 #define flush_pmd_tlb_range(vma, addr, end)	BUILD_BUG()
1716 #define flush_pud_tlb_range(vma, addr, end)	BUILD_BUG()
1717 #endif
1718 #endif
1719 
1720 struct file;
1721 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1722 			unsigned long size, pgprot_t *vma_prot);
1723 
1724 #ifndef CONFIG_X86_ESPFIX64
1725 static inline void init_espfix_bsp(void) { }
1726 #endif
1727 
1728 extern void __init pgtable_cache_init(void);
1729 
1730 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1731 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1732 {
1733 	return true;
1734 }
1735 
1736 static inline bool arch_has_pfn_modify_check(void)
1737 {
1738 	return false;
1739 }
1740 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1741 
1742 /*
1743  * Architecture PAGE_KERNEL_* fallbacks
1744  *
1745  * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1746  * because they really don't support them, or the port needs to be updated to
1747  * reflect the required functionality. Below are a set of relatively safe
1748  * fallbacks, as best effort, which we can count on in lieu of the architectures
1749  * not defining them on their own yet.
1750  */
1751 
1752 #ifndef PAGE_KERNEL_RO
1753 # define PAGE_KERNEL_RO PAGE_KERNEL
1754 #endif
1755 
1756 #ifndef PAGE_KERNEL_EXEC
1757 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1758 #endif
1759 
1760 /*
1761  * Page Table Modification bits for pgtbl_mod_mask.
1762  *
1763  * These are used by the p?d_alloc_track*() set of functions an in the generic
1764  * vmalloc/ioremap code to track at which page-table levels entries have been
1765  * modified. Based on that the code can better decide when vmalloc and ioremap
1766  * mapping changes need to be synchronized to other page-tables in the system.
1767  */
1768 #define		__PGTBL_PGD_MODIFIED	0
1769 #define		__PGTBL_P4D_MODIFIED	1
1770 #define		__PGTBL_PUD_MODIFIED	2
1771 #define		__PGTBL_PMD_MODIFIED	3
1772 #define		__PGTBL_PTE_MODIFIED	4
1773 
1774 #define		PGTBL_PGD_MODIFIED	BIT(__PGTBL_PGD_MODIFIED)
1775 #define		PGTBL_P4D_MODIFIED	BIT(__PGTBL_P4D_MODIFIED)
1776 #define		PGTBL_PUD_MODIFIED	BIT(__PGTBL_PUD_MODIFIED)
1777 #define		PGTBL_PMD_MODIFIED	BIT(__PGTBL_PMD_MODIFIED)
1778 #define		PGTBL_PTE_MODIFIED	BIT(__PGTBL_PTE_MODIFIED)
1779 
1780 /* Page-Table Modification Mask */
1781 typedef unsigned int pgtbl_mod_mask;
1782 
1783 #endif /* !__ASSEMBLY__ */
1784 
1785 #if !defined(MAX_POSSIBLE_PHYSMEM_BITS) && !defined(CONFIG_64BIT)
1786 #ifdef CONFIG_PHYS_ADDR_T_64BIT
1787 /*
1788  * ZSMALLOC needs to know the highest PFN on 32-bit architectures
1789  * with physical address space extension, but falls back to
1790  * BITS_PER_LONG otherwise.
1791  */
1792 #error Missing MAX_POSSIBLE_PHYSMEM_BITS definition
1793 #else
1794 #define MAX_POSSIBLE_PHYSMEM_BITS 32
1795 #endif
1796 #endif
1797 
1798 #ifndef has_transparent_hugepage
1799 #define has_transparent_hugepage() IS_BUILTIN(CONFIG_TRANSPARENT_HUGEPAGE)
1800 #endif
1801 
1802 #ifndef has_transparent_pud_hugepage
1803 #define has_transparent_pud_hugepage() IS_BUILTIN(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1804 #endif
1805 /*
1806  * On some architectures it depends on the mm if the p4d/pud or pmd
1807  * layer of the page table hierarchy is folded or not.
1808  */
1809 #ifndef mm_p4d_folded
1810 #define mm_p4d_folded(mm)	__is_defined(__PAGETABLE_P4D_FOLDED)
1811 #endif
1812 
1813 #ifndef mm_pud_folded
1814 #define mm_pud_folded(mm)	__is_defined(__PAGETABLE_PUD_FOLDED)
1815 #endif
1816 
1817 #ifndef mm_pmd_folded
1818 #define mm_pmd_folded(mm)	__is_defined(__PAGETABLE_PMD_FOLDED)
1819 #endif
1820 
1821 #ifndef p4d_offset_lockless
1822 #define p4d_offset_lockless(pgdp, pgd, address) p4d_offset(&(pgd), address)
1823 #endif
1824 #ifndef pud_offset_lockless
1825 #define pud_offset_lockless(p4dp, p4d, address) pud_offset(&(p4d), address)
1826 #endif
1827 #ifndef pmd_offset_lockless
1828 #define pmd_offset_lockless(pudp, pud, address) pmd_offset(&(pud), address)
1829 #endif
1830 
1831 /*
1832  * pXd_leaf() is the API to check whether a pgtable entry is a huge page
1833  * mapping.  It should work globally across all archs, without any
1834  * dependency on CONFIG_* options.  For architectures that do not support
1835  * huge mappings on specific levels, below fallbacks will be used.
1836  *
1837  * A leaf pgtable entry should always imply the following:
1838  *
1839  * - It is a "present" entry.  IOW, before using this API, please check it
1840  *   with pXd_present() first. NOTE: it may not always mean the "present
1841  *   bit" is set.  For example, PROT_NONE entries are always "present".
1842  *
1843  * - It should _never_ be a swap entry of any type.  Above "present" check
1844  *   should have guarded this, but let's be crystal clear on this.
1845  *
1846  * - It should contain a huge PFN, which points to a huge page larger than
1847  *   PAGE_SIZE of the platform.  The PFN format isn't important here.
1848  *
1849  * - It should cover all kinds of huge mappings (e.g., pXd_trans_huge(),
1850  *   pXd_devmap(), or hugetlb mappings).
1851  */
1852 #ifndef pgd_leaf
1853 #define pgd_leaf(x)	false
1854 #endif
1855 #ifndef p4d_leaf
1856 #define p4d_leaf(x)	false
1857 #endif
1858 #ifndef pud_leaf
1859 #define pud_leaf(x)	false
1860 #endif
1861 #ifndef pmd_leaf
1862 #define pmd_leaf(x)	false
1863 #endif
1864 
1865 #ifndef pgd_leaf_size
1866 #define pgd_leaf_size(x) (1ULL << PGDIR_SHIFT)
1867 #endif
1868 #ifndef p4d_leaf_size
1869 #define p4d_leaf_size(x) P4D_SIZE
1870 #endif
1871 #ifndef pud_leaf_size
1872 #define pud_leaf_size(x) PUD_SIZE
1873 #endif
1874 #ifndef pmd_leaf_size
1875 #define pmd_leaf_size(x) PMD_SIZE
1876 #endif
1877 #ifndef pte_leaf_size
1878 #define pte_leaf_size(x) PAGE_SIZE
1879 #endif
1880 
1881 /*
1882  * We always define pmd_pfn for all archs as it's used in lots of generic
1883  * code.  Now it happens too for pud_pfn (and can happen for larger
1884  * mappings too in the future; we're not there yet).  Instead of defining
1885  * it for all archs (like pmd_pfn), provide a fallback.
1886  *
1887  * Note that returning 0 here means any arch that didn't define this can
1888  * get severely wrong when it hits a real pud leaf.  It's arch's
1889  * responsibility to properly define it when a huge pud is possible.
1890  */
1891 #ifndef pud_pfn
1892 #define pud_pfn(x) 0
1893 #endif
1894 
1895 /*
1896  * Some architectures have MMUs that are configurable or selectable at boot
1897  * time. These lead to variable PTRS_PER_x. For statically allocated arrays it
1898  * helps to have a static maximum value.
1899  */
1900 
1901 #ifndef MAX_PTRS_PER_PTE
1902 #define MAX_PTRS_PER_PTE PTRS_PER_PTE
1903 #endif
1904 
1905 #ifndef MAX_PTRS_PER_PMD
1906 #define MAX_PTRS_PER_PMD PTRS_PER_PMD
1907 #endif
1908 
1909 #ifndef MAX_PTRS_PER_PUD
1910 #define MAX_PTRS_PER_PUD PTRS_PER_PUD
1911 #endif
1912 
1913 #ifndef MAX_PTRS_PER_P4D
1914 #define MAX_PTRS_PER_P4D PTRS_PER_P4D
1915 #endif
1916 
1917 /* description of effects of mapping type and prot in current implementation.
1918  * this is due to the limited x86 page protection hardware.  The expected
1919  * behavior is in parens:
1920  *
1921  * map_type	prot
1922  *		PROT_NONE	PROT_READ	PROT_WRITE	PROT_EXEC
1923  * MAP_SHARED	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1924  *		w: (no) no	w: (no) no	w: (yes) yes	w: (no) no
1925  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1926  *
1927  * MAP_PRIVATE	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1928  *		w: (no) no	w: (no) no	w: (copy) copy	w: (no) no
1929  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1930  *
1931  * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
1932  * MAP_PRIVATE (with Enhanced PAN supported):
1933  *								r: (no) no
1934  *								w: (no) no
1935  *								x: (yes) yes
1936  */
1937 #define DECLARE_VM_GET_PAGE_PROT					\
1938 pgprot_t vm_get_page_prot(unsigned long vm_flags)			\
1939 {									\
1940 		return protection_map[vm_flags &			\
1941 			(VM_READ | VM_WRITE | VM_EXEC | VM_SHARED)];	\
1942 }									\
1943 EXPORT_SYMBOL(vm_get_page_prot);
1944 
1945 #endif /* _LINUX_PGTABLE_H */
1946