xref: /linux-6.15/include/linux/pgtable.h (revision 2ca7cd80)
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 kernel_pte_init
94 static inline void kernel_pte_init(void *addr)
95 {
96 }
97 #define kernel_pte_init kernel_pte_init
98 #endif
99 
100 #ifndef pmd_init
101 static inline void pmd_init(void *addr)
102 {
103 }
104 #define pmd_init pmd_init
105 #endif
106 
107 #ifndef pud_init
108 static inline void pud_init(void *addr)
109 {
110 }
111 #define pud_init pud_init
112 #endif
113 
114 #ifndef pte_offset_kernel
115 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
116 {
117 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
118 }
119 #define pte_offset_kernel pte_offset_kernel
120 #endif
121 
122 #ifdef CONFIG_HIGHPTE
123 #define __pte_map(pmd, address) \
124 	((pte_t *)kmap_local_page(pmd_page(*(pmd))) + pte_index((address)))
125 #define pte_unmap(pte)	do {	\
126 	kunmap_local((pte));	\
127 	rcu_read_unlock();	\
128 } while (0)
129 #else
130 static inline pte_t *__pte_map(pmd_t *pmd, unsigned long address)
131 {
132 	return pte_offset_kernel(pmd, address);
133 }
134 static inline void pte_unmap(pte_t *pte)
135 {
136 	rcu_read_unlock();
137 }
138 #endif
139 
140 void pte_free_defer(struct mm_struct *mm, pgtable_t pgtable);
141 
142 /* Find an entry in the second-level page table.. */
143 #ifndef pmd_offset
144 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
145 {
146 	return pud_pgtable(*pud) + pmd_index(address);
147 }
148 #define pmd_offset pmd_offset
149 #endif
150 
151 #ifndef pud_offset
152 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
153 {
154 	return p4d_pgtable(*p4d) + pud_index(address);
155 }
156 #define pud_offset pud_offset
157 #endif
158 
159 static inline pgd_t *pgd_offset_pgd(pgd_t *pgd, unsigned long address)
160 {
161 	return (pgd + pgd_index(address));
162 };
163 
164 /*
165  * a shortcut to get a pgd_t in a given mm
166  */
167 #ifndef pgd_offset
168 #define pgd_offset(mm, address)		pgd_offset_pgd((mm)->pgd, (address))
169 #endif
170 
171 /*
172  * a shortcut which implies the use of the kernel's pgd, instead
173  * of a process's
174  */
175 #define pgd_offset_k(address)		pgd_offset(&init_mm, (address))
176 
177 /*
178  * In many cases it is known that a virtual address is mapped at PMD or PTE
179  * level, so instead of traversing all the page table levels, we can get a
180  * pointer to the PMD entry in user or kernel page table or translate a virtual
181  * address to the pointer in the PTE in the kernel page tables with simple
182  * helpers.
183  */
184 static inline pmd_t *pmd_off(struct mm_struct *mm, unsigned long va)
185 {
186 	return pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, va), va), va), va);
187 }
188 
189 static inline pmd_t *pmd_off_k(unsigned long va)
190 {
191 	return pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va);
192 }
193 
194 static inline pte_t *virt_to_kpte(unsigned long vaddr)
195 {
196 	pmd_t *pmd = pmd_off_k(vaddr);
197 
198 	return pmd_none(*pmd) ? NULL : pte_offset_kernel(pmd, vaddr);
199 }
200 
201 #ifndef pmd_young
202 static inline int pmd_young(pmd_t pmd)
203 {
204 	return 0;
205 }
206 #endif
207 
208 #ifndef pmd_dirty
209 static inline int pmd_dirty(pmd_t pmd)
210 {
211 	return 0;
212 }
213 #endif
214 
215 /*
216  * A facility to provide lazy MMU batching.  This allows PTE updates and
217  * page invalidations to be delayed until a call to leave lazy MMU mode
218  * is issued.  Some architectures may benefit from doing this, and it is
219  * beneficial for both shadow and direct mode hypervisors, which may batch
220  * the PTE updates which happen during this window.  Note that using this
221  * interface requires that read hazards be removed from the code.  A read
222  * hazard could result in the direct mode hypervisor case, since the actual
223  * write to the page tables may not yet have taken place, so reads though
224  * a raw PTE pointer after it has been modified are not guaranteed to be
225  * up to date.  This mode can only be entered and left under the protection of
226  * the page table locks for all page tables which may be modified.  In the UP
227  * case, this is required so that preemption is disabled, and in the SMP case,
228  * it must synchronize the delayed page table writes properly on other CPUs.
229  */
230 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
231 #define arch_enter_lazy_mmu_mode()	do {} while (0)
232 #define arch_leave_lazy_mmu_mode()	do {} while (0)
233 #define arch_flush_lazy_mmu_mode()	do {} while (0)
234 #endif
235 
236 #ifndef pte_batch_hint
237 /**
238  * pte_batch_hint - Number of pages that can be added to batch without scanning.
239  * @ptep: Page table pointer for the entry.
240  * @pte: Page table entry.
241  *
242  * Some architectures know that a set of contiguous ptes all map the same
243  * contiguous memory with the same permissions. In this case, it can provide a
244  * hint to aid pte batching without the core code needing to scan every pte.
245  *
246  * An architecture implementation may ignore the PTE accessed state. Further,
247  * the dirty state must apply atomically to all the PTEs described by the hint.
248  *
249  * May be overridden by the architecture, else pte_batch_hint is always 1.
250  */
251 static inline unsigned int pte_batch_hint(pte_t *ptep, pte_t pte)
252 {
253 	return 1;
254 }
255 #endif
256 
257 #ifndef pte_advance_pfn
258 static inline pte_t pte_advance_pfn(pte_t pte, unsigned long nr)
259 {
260 	return __pte(pte_val(pte) + (nr << PFN_PTE_SHIFT));
261 }
262 #endif
263 
264 #define pte_next_pfn(pte) pte_advance_pfn(pte, 1)
265 
266 #ifndef set_ptes
267 /**
268  * set_ptes - Map consecutive pages to a contiguous range of addresses.
269  * @mm: Address space to map the pages into.
270  * @addr: Address to map the first page at.
271  * @ptep: Page table pointer for the first entry.
272  * @pte: Page table entry for the first page.
273  * @nr: Number of pages to map.
274  *
275  * When nr==1, initial state of pte may be present or not present, and new state
276  * may be present or not present. When nr>1, initial state of all ptes must be
277  * not present, and new state must be present.
278  *
279  * May be overridden by the architecture, or the architecture can define
280  * set_pte() and PFN_PTE_SHIFT.
281  *
282  * Context: The caller holds the page table lock.  The pages all belong
283  * to the same folio.  The PTEs are all in the same PMD.
284  */
285 static inline void set_ptes(struct mm_struct *mm, unsigned long addr,
286 		pte_t *ptep, pte_t pte, unsigned int nr)
287 {
288 	page_table_check_ptes_set(mm, ptep, pte, nr);
289 
290 	arch_enter_lazy_mmu_mode();
291 	for (;;) {
292 		set_pte(ptep, pte);
293 		if (--nr == 0)
294 			break;
295 		ptep++;
296 		pte = pte_next_pfn(pte);
297 	}
298 	arch_leave_lazy_mmu_mode();
299 }
300 #endif
301 #define set_pte_at(mm, addr, ptep, pte) set_ptes(mm, addr, ptep, pte, 1)
302 
303 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
304 extern int ptep_set_access_flags(struct vm_area_struct *vma,
305 				 unsigned long address, pte_t *ptep,
306 				 pte_t entry, int dirty);
307 #endif
308 
309 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
310 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
311 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
312 				 unsigned long address, pmd_t *pmdp,
313 				 pmd_t entry, int dirty);
314 extern int pudp_set_access_flags(struct vm_area_struct *vma,
315 				 unsigned long address, pud_t *pudp,
316 				 pud_t entry, int dirty);
317 #else
318 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
319 					unsigned long address, pmd_t *pmdp,
320 					pmd_t entry, int dirty)
321 {
322 	BUILD_BUG();
323 	return 0;
324 }
325 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
326 					unsigned long address, pud_t *pudp,
327 					pud_t entry, int dirty)
328 {
329 	BUILD_BUG();
330 	return 0;
331 }
332 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
333 #endif
334 
335 #ifndef ptep_get
336 static inline pte_t ptep_get(pte_t *ptep)
337 {
338 	return READ_ONCE(*ptep);
339 }
340 #endif
341 
342 #ifndef pmdp_get
343 static inline pmd_t pmdp_get(pmd_t *pmdp)
344 {
345 	return READ_ONCE(*pmdp);
346 }
347 #endif
348 
349 #ifndef pudp_get
350 static inline pud_t pudp_get(pud_t *pudp)
351 {
352 	return READ_ONCE(*pudp);
353 }
354 #endif
355 
356 #ifndef p4dp_get
357 static inline p4d_t p4dp_get(p4d_t *p4dp)
358 {
359 	return READ_ONCE(*p4dp);
360 }
361 #endif
362 
363 #ifndef pgdp_get
364 static inline pgd_t pgdp_get(pgd_t *pgdp)
365 {
366 	return READ_ONCE(*pgdp);
367 }
368 #endif
369 
370 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
371 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
372 					    unsigned long address,
373 					    pte_t *ptep)
374 {
375 	pte_t pte = ptep_get(ptep);
376 	int r = 1;
377 	if (!pte_young(pte))
378 		r = 0;
379 	else
380 		set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
381 	return r;
382 }
383 #endif
384 
385 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
386 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG)
387 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
388 					    unsigned long address,
389 					    pmd_t *pmdp)
390 {
391 	pmd_t pmd = *pmdp;
392 	int r = 1;
393 	if (!pmd_young(pmd))
394 		r = 0;
395 	else
396 		set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
397 	return r;
398 }
399 #else
400 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
401 					    unsigned long address,
402 					    pmd_t *pmdp)
403 {
404 	BUILD_BUG();
405 	return 0;
406 }
407 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG */
408 #endif
409 
410 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
411 int ptep_clear_flush_young(struct vm_area_struct *vma,
412 			   unsigned long address, pte_t *ptep);
413 #endif
414 
415 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
416 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
417 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
418 				  unsigned long address, pmd_t *pmdp);
419 #else
420 /*
421  * Despite relevant to THP only, this API is called from generic rmap code
422  * under PageTransHuge(), hence needs a dummy implementation for !THP
423  */
424 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
425 					 unsigned long address, pmd_t *pmdp)
426 {
427 	BUILD_BUG();
428 	return 0;
429 }
430 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
431 #endif
432 
433 #ifndef arch_has_hw_nonleaf_pmd_young
434 /*
435  * Return whether the accessed bit in non-leaf PMD entries is supported on the
436  * local CPU.
437  */
438 static inline bool arch_has_hw_nonleaf_pmd_young(void)
439 {
440 	return IS_ENABLED(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG);
441 }
442 #endif
443 
444 #ifndef arch_has_hw_pte_young
445 /*
446  * Return whether the accessed bit is supported on the local CPU.
447  *
448  * This stub assumes accessing through an old PTE triggers a page fault.
449  * Architectures that automatically set the access bit should overwrite it.
450  */
451 static inline bool arch_has_hw_pte_young(void)
452 {
453 	return IS_ENABLED(CONFIG_ARCH_HAS_HW_PTE_YOUNG);
454 }
455 #endif
456 
457 #ifndef arch_check_zapped_pte
458 static inline void arch_check_zapped_pte(struct vm_area_struct *vma,
459 					 pte_t pte)
460 {
461 }
462 #endif
463 
464 #ifndef arch_check_zapped_pmd
465 static inline void arch_check_zapped_pmd(struct vm_area_struct *vma,
466 					 pmd_t pmd)
467 {
468 }
469 #endif
470 
471 #ifndef arch_check_zapped_pud
472 static inline void arch_check_zapped_pud(struct vm_area_struct *vma, pud_t pud)
473 {
474 }
475 #endif
476 
477 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
478 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
479 				       unsigned long address,
480 				       pte_t *ptep)
481 {
482 	pte_t pte = ptep_get(ptep);
483 	pte_clear(mm, address, ptep);
484 	page_table_check_pte_clear(mm, pte);
485 	return pte;
486 }
487 #endif
488 
489 #ifndef clear_young_dirty_ptes
490 /**
491  * clear_young_dirty_ptes - Mark PTEs that map consecutive pages of the
492  *		same folio as old/clean.
493  * @mm: Address space the pages are mapped into.
494  * @addr: Address the first page is mapped at.
495  * @ptep: Page table pointer for the first entry.
496  * @nr: Number of entries to mark old/clean.
497  * @flags: Flags to modify the PTE batch semantics.
498  *
499  * May be overridden by the architecture; otherwise, implemented by
500  * get_and_clear/modify/set for each pte in the range.
501  *
502  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
503  * some PTEs might be write-protected.
504  *
505  * Context: The caller holds the page table lock.  The PTEs map consecutive
506  * pages that belong to the same folio.  The PTEs are all in the same PMD.
507  */
508 static inline void clear_young_dirty_ptes(struct vm_area_struct *vma,
509 					  unsigned long addr, pte_t *ptep,
510 					  unsigned int nr, cydp_t flags)
511 {
512 	pte_t pte;
513 
514 	for (;;) {
515 		if (flags == CYDP_CLEAR_YOUNG)
516 			ptep_test_and_clear_young(vma, addr, ptep);
517 		else {
518 			pte = ptep_get_and_clear(vma->vm_mm, addr, ptep);
519 			if (flags & CYDP_CLEAR_YOUNG)
520 				pte = pte_mkold(pte);
521 			if (flags & CYDP_CLEAR_DIRTY)
522 				pte = pte_mkclean(pte);
523 			set_pte_at(vma->vm_mm, addr, ptep, pte);
524 		}
525 		if (--nr == 0)
526 			break;
527 		ptep++;
528 		addr += PAGE_SIZE;
529 	}
530 }
531 #endif
532 
533 static inline void ptep_clear(struct mm_struct *mm, unsigned long addr,
534 			      pte_t *ptep)
535 {
536 	ptep_get_and_clear(mm, addr, ptep);
537 }
538 
539 #ifdef CONFIG_GUP_GET_PXX_LOW_HIGH
540 /*
541  * For walking the pagetables without holding any locks.  Some architectures
542  * (eg x86-32 PAE) cannot load the entries atomically without using expensive
543  * instructions.  We are guaranteed that a PTE will only either go from not
544  * present to present, or present to not present -- it will not switch to a
545  * completely different present page without a TLB flush inbetween; which we
546  * are blocking by holding interrupts off.
547  *
548  * Setting ptes from not present to present goes:
549  *
550  *   ptep->pte_high = h;
551  *   smp_wmb();
552  *   ptep->pte_low = l;
553  *
554  * And present to not present goes:
555  *
556  *   ptep->pte_low = 0;
557  *   smp_wmb();
558  *   ptep->pte_high = 0;
559  *
560  * We must ensure here that the load of pte_low sees 'l' IFF pte_high sees 'h'.
561  * We load pte_high *after* loading pte_low, which ensures we don't see an older
562  * value of pte_high.  *Then* we recheck pte_low, which ensures that we haven't
563  * picked up a changed pte high. We might have gotten rubbish values from
564  * pte_low and pte_high, but we are guaranteed that pte_low will not have the
565  * present bit set *unless* it is 'l'. Because get_user_pages_fast() only
566  * operates on present ptes we're safe.
567  */
568 static inline pte_t ptep_get_lockless(pte_t *ptep)
569 {
570 	pte_t pte;
571 
572 	do {
573 		pte.pte_low = ptep->pte_low;
574 		smp_rmb();
575 		pte.pte_high = ptep->pte_high;
576 		smp_rmb();
577 	} while (unlikely(pte.pte_low != ptep->pte_low));
578 
579 	return pte;
580 }
581 #define ptep_get_lockless ptep_get_lockless
582 
583 #if CONFIG_PGTABLE_LEVELS > 2
584 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
585 {
586 	pmd_t pmd;
587 
588 	do {
589 		pmd.pmd_low = pmdp->pmd_low;
590 		smp_rmb();
591 		pmd.pmd_high = pmdp->pmd_high;
592 		smp_rmb();
593 	} while (unlikely(pmd.pmd_low != pmdp->pmd_low));
594 
595 	return pmd;
596 }
597 #define pmdp_get_lockless pmdp_get_lockless
598 #define pmdp_get_lockless_sync() tlb_remove_table_sync_one()
599 #endif /* CONFIG_PGTABLE_LEVELS > 2 */
600 #endif /* CONFIG_GUP_GET_PXX_LOW_HIGH */
601 
602 /*
603  * We require that the PTE can be read atomically.
604  */
605 #ifndef ptep_get_lockless
606 static inline pte_t ptep_get_lockless(pte_t *ptep)
607 {
608 	return ptep_get(ptep);
609 }
610 #endif
611 
612 #ifndef pmdp_get_lockless
613 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
614 {
615 	return pmdp_get(pmdp);
616 }
617 static inline void pmdp_get_lockless_sync(void)
618 {
619 }
620 #endif
621 
622 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
623 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
624 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
625 					    unsigned long address,
626 					    pmd_t *pmdp)
627 {
628 	pmd_t pmd = *pmdp;
629 
630 	pmd_clear(pmdp);
631 	page_table_check_pmd_clear(mm, pmd);
632 
633 	return pmd;
634 }
635 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
636 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
637 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
638 					    unsigned long address,
639 					    pud_t *pudp)
640 {
641 	pud_t pud = *pudp;
642 
643 	pud_clear(pudp);
644 	page_table_check_pud_clear(mm, pud);
645 
646 	return pud;
647 }
648 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
649 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
650 
651 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
652 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
653 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
654 					    unsigned long address, pmd_t *pmdp,
655 					    int full)
656 {
657 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
658 }
659 #endif
660 
661 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
662 static inline pud_t pudp_huge_get_and_clear_full(struct vm_area_struct *vma,
663 					    unsigned long address, pud_t *pudp,
664 					    int full)
665 {
666 	return pudp_huge_get_and_clear(vma->vm_mm, address, pudp);
667 }
668 #endif
669 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
670 
671 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
672 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
673 					    unsigned long address, pte_t *ptep,
674 					    int full)
675 {
676 	return ptep_get_and_clear(mm, address, ptep);
677 }
678 #endif
679 
680 #ifndef get_and_clear_full_ptes
681 /**
682  * get_and_clear_full_ptes - Clear present PTEs that map consecutive pages of
683  *			     the same folio, collecting dirty/accessed bits.
684  * @mm: Address space the pages are mapped into.
685  * @addr: Address the first page is mapped at.
686  * @ptep: Page table pointer for the first entry.
687  * @nr: Number of entries to clear.
688  * @full: Whether we are clearing a full mm.
689  *
690  * May be overridden by the architecture; otherwise, implemented as a simple
691  * loop over ptep_get_and_clear_full(), merging dirty/accessed bits into the
692  * returned PTE.
693  *
694  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
695  * some PTEs might be write-protected.
696  *
697  * Context: The caller holds the page table lock.  The PTEs map consecutive
698  * pages that belong to the same folio.  The PTEs are all in the same PMD.
699  */
700 static inline pte_t get_and_clear_full_ptes(struct mm_struct *mm,
701 		unsigned long addr, pte_t *ptep, unsigned int nr, int full)
702 {
703 	pte_t pte, tmp_pte;
704 
705 	pte = ptep_get_and_clear_full(mm, addr, ptep, full);
706 	while (--nr) {
707 		ptep++;
708 		addr += PAGE_SIZE;
709 		tmp_pte = ptep_get_and_clear_full(mm, addr, ptep, full);
710 		if (pte_dirty(tmp_pte))
711 			pte = pte_mkdirty(pte);
712 		if (pte_young(tmp_pte))
713 			pte = pte_mkyoung(pte);
714 	}
715 	return pte;
716 }
717 #endif
718 
719 #ifndef clear_full_ptes
720 /**
721  * clear_full_ptes - Clear present PTEs that map consecutive pages of the same
722  *		     folio.
723  * @mm: Address space the pages are mapped into.
724  * @addr: Address the first page is mapped at.
725  * @ptep: Page table pointer for the first entry.
726  * @nr: Number of entries to clear.
727  * @full: Whether we are clearing a full mm.
728  *
729  * May be overridden by the architecture; otherwise, implemented as a simple
730  * loop over ptep_get_and_clear_full().
731  *
732  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
733  * some PTEs might be write-protected.
734  *
735  * Context: The caller holds the page table lock.  The PTEs map consecutive
736  * pages that belong to the same folio.  The PTEs are all in the same PMD.
737  */
738 static inline void clear_full_ptes(struct mm_struct *mm, unsigned long addr,
739 		pte_t *ptep, unsigned int nr, int full)
740 {
741 	for (;;) {
742 		ptep_get_and_clear_full(mm, addr, ptep, full);
743 		if (--nr == 0)
744 			break;
745 		ptep++;
746 		addr += PAGE_SIZE;
747 	}
748 }
749 #endif
750 
751 /*
752  * If two threads concurrently fault at the same page, the thread that
753  * won the race updates the PTE and its local TLB/Cache. The other thread
754  * gives up, simply does nothing, and continues; on architectures where
755  * software can update TLB,  local TLB can be updated here to avoid next page
756  * fault. This function updates TLB only, do nothing with cache or others.
757  * It is the difference with function update_mmu_cache.
758  */
759 #ifndef update_mmu_tlb_range
760 static inline void update_mmu_tlb_range(struct vm_area_struct *vma,
761 				unsigned long address, pte_t *ptep, unsigned int nr)
762 {
763 }
764 #endif
765 
766 static inline void update_mmu_tlb(struct vm_area_struct *vma,
767 				unsigned long address, pte_t *ptep)
768 {
769 	update_mmu_tlb_range(vma, address, ptep, 1);
770 }
771 
772 /*
773  * Some architectures may be able to avoid expensive synchronization
774  * primitives when modifications are made to PTE's which are already
775  * not present, or in the process of an address space destruction.
776  */
777 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
778 static inline void pte_clear_not_present_full(struct mm_struct *mm,
779 					      unsigned long address,
780 					      pte_t *ptep,
781 					      int full)
782 {
783 	pte_clear(mm, address, ptep);
784 }
785 #endif
786 
787 #ifndef clear_not_present_full_ptes
788 /**
789  * clear_not_present_full_ptes - Clear multiple not present PTEs which are
790  *				 consecutive in the pgtable.
791  * @mm: Address space the ptes represent.
792  * @addr: Address of the first pte.
793  * @ptep: Page table pointer for the first entry.
794  * @nr: Number of entries to clear.
795  * @full: Whether we are clearing a full mm.
796  *
797  * May be overridden by the architecture; otherwise, implemented as a simple
798  * loop over pte_clear_not_present_full().
799  *
800  * Context: The caller holds the page table lock.  The PTEs are all not present.
801  * The PTEs are all in the same PMD.
802  */
803 static inline void clear_not_present_full_ptes(struct mm_struct *mm,
804 		unsigned long addr, pte_t *ptep, unsigned int nr, int full)
805 {
806 	for (;;) {
807 		pte_clear_not_present_full(mm, addr, ptep, full);
808 		if (--nr == 0)
809 			break;
810 		ptep++;
811 		addr += PAGE_SIZE;
812 	}
813 }
814 #endif
815 
816 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
817 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
818 			      unsigned long address,
819 			      pte_t *ptep);
820 #endif
821 
822 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
823 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
824 			      unsigned long address,
825 			      pmd_t *pmdp);
826 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
827 			      unsigned long address,
828 			      pud_t *pudp);
829 #endif
830 
831 #ifndef pte_mkwrite
832 static inline pte_t pte_mkwrite(pte_t pte, struct vm_area_struct *vma)
833 {
834 	return pte_mkwrite_novma(pte);
835 }
836 #endif
837 
838 #if defined(CONFIG_ARCH_WANT_PMD_MKWRITE) && !defined(pmd_mkwrite)
839 static inline pmd_t pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma)
840 {
841 	return pmd_mkwrite_novma(pmd);
842 }
843 #endif
844 
845 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
846 struct mm_struct;
847 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
848 {
849 	pte_t old_pte = ptep_get(ptep);
850 	set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
851 }
852 #endif
853 
854 #ifndef wrprotect_ptes
855 /**
856  * wrprotect_ptes - Write-protect PTEs that map consecutive pages of the same
857  *		    folio.
858  * @mm: Address space the pages are mapped into.
859  * @addr: Address the first page is mapped at.
860  * @ptep: Page table pointer for the first entry.
861  * @nr: Number of entries to write-protect.
862  *
863  * May be overridden by the architecture; otherwise, implemented as a simple
864  * loop over ptep_set_wrprotect().
865  *
866  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
867  * some PTEs might be write-protected.
868  *
869  * Context: The caller holds the page table lock.  The PTEs map consecutive
870  * pages that belong to the same folio.  The PTEs are all in the same PMD.
871  */
872 static inline void wrprotect_ptes(struct mm_struct *mm, unsigned long addr,
873 		pte_t *ptep, unsigned int nr)
874 {
875 	for (;;) {
876 		ptep_set_wrprotect(mm, addr, ptep);
877 		if (--nr == 0)
878 			break;
879 		ptep++;
880 		addr += PAGE_SIZE;
881 	}
882 }
883 #endif
884 
885 /*
886  * On some architectures hardware does not set page access bit when accessing
887  * memory page, it is responsibility of software setting this bit. It brings
888  * out extra page fault penalty to track page access bit. For optimization page
889  * access bit can be set during all page fault flow on these arches.
890  * To be differentiate with macro pte_mkyoung, this macro is used on platforms
891  * where software maintains page access bit.
892  */
893 #ifndef pte_sw_mkyoung
894 static inline pte_t pte_sw_mkyoung(pte_t pte)
895 {
896 	return pte;
897 }
898 #define pte_sw_mkyoung	pte_sw_mkyoung
899 #endif
900 
901 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
902 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
903 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
904 				      unsigned long address, pmd_t *pmdp)
905 {
906 	pmd_t old_pmd = *pmdp;
907 	set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
908 }
909 #else
910 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
911 				      unsigned long address, pmd_t *pmdp)
912 {
913 	BUILD_BUG();
914 }
915 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
916 #endif
917 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
918 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
919 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
920 static inline void pudp_set_wrprotect(struct mm_struct *mm,
921 				      unsigned long address, pud_t *pudp)
922 {
923 	pud_t old_pud = *pudp;
924 
925 	set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
926 }
927 #else
928 static inline void pudp_set_wrprotect(struct mm_struct *mm,
929 				      unsigned long address, pud_t *pudp)
930 {
931 	BUILD_BUG();
932 }
933 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
934 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
935 #endif
936 
937 #ifndef pmdp_collapse_flush
938 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
939 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
940 				 unsigned long address, pmd_t *pmdp);
941 #else
942 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
943 					unsigned long address,
944 					pmd_t *pmdp)
945 {
946 	BUILD_BUG();
947 	return *pmdp;
948 }
949 #define pmdp_collapse_flush pmdp_collapse_flush
950 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
951 #endif
952 
953 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
954 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
955 				       pgtable_t pgtable);
956 #endif
957 
958 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
959 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
960 #endif
961 
962 #ifndef arch_needs_pgtable_deposit
963 #define arch_needs_pgtable_deposit() (false)
964 #endif
965 
966 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
967 /*
968  * This is an implementation of pmdp_establish() that is only suitable for an
969  * architecture that doesn't have hardware dirty/accessed bits. In this case we
970  * can't race with CPU which sets these bits and non-atomic approach is fine.
971  */
972 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
973 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
974 {
975 	pmd_t old_pmd = *pmdp;
976 	set_pmd_at(vma->vm_mm, address, pmdp, pmd);
977 	return old_pmd;
978 }
979 #endif
980 
981 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
982 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
983 			    pmd_t *pmdp);
984 #endif
985 
986 #ifndef __HAVE_ARCH_PMDP_INVALIDATE_AD
987 
988 /*
989  * pmdp_invalidate_ad() invalidates the PMD while changing a transparent
990  * hugepage mapping in the page tables. This function is similar to
991  * pmdp_invalidate(), but should only be used if the access and dirty bits would
992  * not be cleared by the software in the new PMD value. The function ensures
993  * that hardware changes of the access and dirty bits updates would not be lost.
994  *
995  * Doing so can allow in certain architectures to avoid a TLB flush in most
996  * cases. Yet, another TLB flush might be necessary later if the PMD update
997  * itself requires such flush (e.g., if protection was set to be stricter). Yet,
998  * even when a TLB flush is needed because of the update, the caller may be able
999  * to batch these TLB flushing operations, so fewer TLB flush operations are
1000  * needed.
1001  */
1002 extern pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma,
1003 				unsigned long address, pmd_t *pmdp);
1004 #endif
1005 
1006 #ifndef __HAVE_ARCH_PTE_SAME
1007 static inline int pte_same(pte_t pte_a, pte_t pte_b)
1008 {
1009 	return pte_val(pte_a) == pte_val(pte_b);
1010 }
1011 #endif
1012 
1013 #ifndef __HAVE_ARCH_PTE_UNUSED
1014 /*
1015  * Some architectures provide facilities to virtualization guests
1016  * so that they can flag allocated pages as unused. This allows the
1017  * host to transparently reclaim unused pages. This function returns
1018  * whether the pte's page is unused.
1019  */
1020 static inline int pte_unused(pte_t pte)
1021 {
1022 	return 0;
1023 }
1024 #endif
1025 
1026 #ifndef pte_access_permitted
1027 #define pte_access_permitted(pte, write) \
1028 	(pte_present(pte) && (!(write) || pte_write(pte)))
1029 #endif
1030 
1031 #ifndef pmd_access_permitted
1032 #define pmd_access_permitted(pmd, write) \
1033 	(pmd_present(pmd) && (!(write) || pmd_write(pmd)))
1034 #endif
1035 
1036 #ifndef pud_access_permitted
1037 #define pud_access_permitted(pud, write) \
1038 	(pud_present(pud) && (!(write) || pud_write(pud)))
1039 #endif
1040 
1041 #ifndef p4d_access_permitted
1042 #define p4d_access_permitted(p4d, write) \
1043 	(p4d_present(p4d) && (!(write) || p4d_write(p4d)))
1044 #endif
1045 
1046 #ifndef pgd_access_permitted
1047 #define pgd_access_permitted(pgd, write) \
1048 	(pgd_present(pgd) && (!(write) || pgd_write(pgd)))
1049 #endif
1050 
1051 #ifndef __HAVE_ARCH_PMD_SAME
1052 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
1053 {
1054 	return pmd_val(pmd_a) == pmd_val(pmd_b);
1055 }
1056 #endif
1057 
1058 #ifndef pud_same
1059 static inline int pud_same(pud_t pud_a, pud_t pud_b)
1060 {
1061 	return pud_val(pud_a) == pud_val(pud_b);
1062 }
1063 #define pud_same pud_same
1064 #endif
1065 
1066 #ifndef __HAVE_ARCH_P4D_SAME
1067 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
1068 {
1069 	return p4d_val(p4d_a) == p4d_val(p4d_b);
1070 }
1071 #endif
1072 
1073 #ifndef __HAVE_ARCH_PGD_SAME
1074 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
1075 {
1076 	return pgd_val(pgd_a) == pgd_val(pgd_b);
1077 }
1078 #endif
1079 
1080 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
1081 static inline void arch_do_swap_page_nr(struct mm_struct *mm,
1082 				     struct vm_area_struct *vma,
1083 				     unsigned long addr,
1084 				     pte_t pte, pte_t oldpte,
1085 				     int nr)
1086 {
1087 
1088 }
1089 #else
1090 /*
1091  * Some architectures support metadata associated with a page. When a
1092  * page is being swapped out, this metadata must be saved so it can be
1093  * restored when the page is swapped back in. SPARC M7 and newer
1094  * processors support an ADI (Application Data Integrity) tag for the
1095  * page as metadata for the page. arch_do_swap_page() can restore this
1096  * metadata when a page is swapped back in.
1097  */
1098 static inline void arch_do_swap_page_nr(struct mm_struct *mm,
1099 					struct vm_area_struct *vma,
1100 					unsigned long addr,
1101 					pte_t pte, pte_t oldpte,
1102 					int nr)
1103 {
1104 	for (int i = 0; i < nr; i++) {
1105 		arch_do_swap_page(vma->vm_mm, vma, addr + i * PAGE_SIZE,
1106 				pte_advance_pfn(pte, i),
1107 				pte_advance_pfn(oldpte, i));
1108 	}
1109 }
1110 #endif
1111 
1112 #ifndef __HAVE_ARCH_UNMAP_ONE
1113 /*
1114  * Some architectures support metadata associated with a page. When a
1115  * page is being swapped out, this metadata must be saved so it can be
1116  * restored when the page is swapped back in. SPARC M7 and newer
1117  * processors support an ADI (Application Data Integrity) tag for the
1118  * page as metadata for the page. arch_unmap_one() can save this
1119  * metadata on a swap-out of a page.
1120  */
1121 static inline int arch_unmap_one(struct mm_struct *mm,
1122 				  struct vm_area_struct *vma,
1123 				  unsigned long addr,
1124 				  pte_t orig_pte)
1125 {
1126 	return 0;
1127 }
1128 #endif
1129 
1130 /*
1131  * Allow architectures to preserve additional metadata associated with
1132  * swapped-out pages. The corresponding __HAVE_ARCH_SWAP_* macros and function
1133  * prototypes must be defined in the arch-specific asm/pgtable.h file.
1134  */
1135 #ifndef __HAVE_ARCH_PREPARE_TO_SWAP
1136 static inline int arch_prepare_to_swap(struct folio *folio)
1137 {
1138 	return 0;
1139 }
1140 #endif
1141 
1142 #ifndef __HAVE_ARCH_SWAP_INVALIDATE
1143 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
1144 {
1145 }
1146 
1147 static inline void arch_swap_invalidate_area(int type)
1148 {
1149 }
1150 #endif
1151 
1152 #ifndef __HAVE_ARCH_SWAP_RESTORE
1153 static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
1154 {
1155 }
1156 #endif
1157 
1158 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
1159 #define pgd_offset_gate(mm, addr)	pgd_offset(mm, addr)
1160 #endif
1161 
1162 #ifndef __HAVE_ARCH_MOVE_PTE
1163 #define move_pte(pte, old_addr, new_addr)	(pte)
1164 #endif
1165 
1166 #ifndef pte_accessible
1167 # define pte_accessible(mm, pte)	((void)(pte), 1)
1168 #endif
1169 
1170 #ifndef flush_tlb_fix_spurious_fault
1171 #define flush_tlb_fix_spurious_fault(vma, address, ptep) flush_tlb_page(vma, address)
1172 #endif
1173 
1174 /*
1175  * When walking page tables, get the address of the next boundary,
1176  * or the end address of the range if that comes earlier.  Although no
1177  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
1178  */
1179 
1180 #define pgd_addr_end(addr, end)						\
1181 ({	unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;	\
1182 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1183 })
1184 
1185 #ifndef p4d_addr_end
1186 #define p4d_addr_end(addr, end)						\
1187 ({	unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;	\
1188 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1189 })
1190 #endif
1191 
1192 #ifndef pud_addr_end
1193 #define pud_addr_end(addr, end)						\
1194 ({	unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;	\
1195 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1196 })
1197 #endif
1198 
1199 #ifndef pmd_addr_end
1200 #define pmd_addr_end(addr, end)						\
1201 ({	unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;	\
1202 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1203 })
1204 #endif
1205 
1206 /*
1207  * When walking page tables, we usually want to skip any p?d_none entries;
1208  * and any p?d_bad entries - reporting the error before resetting to none.
1209  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
1210  */
1211 void pgd_clear_bad(pgd_t *);
1212 
1213 #ifndef __PAGETABLE_P4D_FOLDED
1214 void p4d_clear_bad(p4d_t *);
1215 #else
1216 #define p4d_clear_bad(p4d)        do { } while (0)
1217 #endif
1218 
1219 #ifndef __PAGETABLE_PUD_FOLDED
1220 void pud_clear_bad(pud_t *);
1221 #else
1222 #define pud_clear_bad(p4d)        do { } while (0)
1223 #endif
1224 
1225 void pmd_clear_bad(pmd_t *);
1226 
1227 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
1228 {
1229 	if (pgd_none(*pgd))
1230 		return 1;
1231 	if (unlikely(pgd_bad(*pgd))) {
1232 		pgd_clear_bad(pgd);
1233 		return 1;
1234 	}
1235 	return 0;
1236 }
1237 
1238 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
1239 {
1240 	if (p4d_none(*p4d))
1241 		return 1;
1242 	if (unlikely(p4d_bad(*p4d))) {
1243 		p4d_clear_bad(p4d);
1244 		return 1;
1245 	}
1246 	return 0;
1247 }
1248 
1249 static inline int pud_none_or_clear_bad(pud_t *pud)
1250 {
1251 	if (pud_none(*pud))
1252 		return 1;
1253 	if (unlikely(pud_bad(*pud))) {
1254 		pud_clear_bad(pud);
1255 		return 1;
1256 	}
1257 	return 0;
1258 }
1259 
1260 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
1261 {
1262 	if (pmd_none(*pmd))
1263 		return 1;
1264 	if (unlikely(pmd_bad(*pmd))) {
1265 		pmd_clear_bad(pmd);
1266 		return 1;
1267 	}
1268 	return 0;
1269 }
1270 
1271 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
1272 					     unsigned long addr,
1273 					     pte_t *ptep)
1274 {
1275 	/*
1276 	 * Get the current pte state, but zero it out to make it
1277 	 * non-present, preventing the hardware from asynchronously
1278 	 * updating it.
1279 	 */
1280 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
1281 }
1282 
1283 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
1284 					     unsigned long addr,
1285 					     pte_t *ptep, pte_t pte)
1286 {
1287 	/*
1288 	 * The pte is non-present, so there's no hardware state to
1289 	 * preserve.
1290 	 */
1291 	set_pte_at(vma->vm_mm, addr, ptep, pte);
1292 }
1293 
1294 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1295 /*
1296  * Start a pte protection read-modify-write transaction, which
1297  * protects against asynchronous hardware modifications to the pte.
1298  * The intention is not to prevent the hardware from making pte
1299  * updates, but to prevent any updates it may make from being lost.
1300  *
1301  * This does not protect against other software modifications of the
1302  * pte; the appropriate pte lock must be held over the transaction.
1303  *
1304  * Note that this interface is intended to be batchable, meaning that
1305  * ptep_modify_prot_commit may not actually update the pte, but merely
1306  * queue the update to be done at some later time.  The update must be
1307  * actually committed before the pte lock is released, however.
1308  */
1309 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
1310 					   unsigned long addr,
1311 					   pte_t *ptep)
1312 {
1313 	return __ptep_modify_prot_start(vma, addr, ptep);
1314 }
1315 
1316 /*
1317  * Commit an update to a pte, leaving any hardware-controlled bits in
1318  * the PTE unmodified.
1319  */
1320 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
1321 					   unsigned long addr,
1322 					   pte_t *ptep, pte_t old_pte, pte_t pte)
1323 {
1324 	__ptep_modify_prot_commit(vma, addr, ptep, pte);
1325 }
1326 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
1327 #endif /* CONFIG_MMU */
1328 
1329 /*
1330  * No-op macros that just return the current protection value. Defined here
1331  * because these macros can be used even if CONFIG_MMU is not defined.
1332  */
1333 
1334 #ifndef pgprot_nx
1335 #define pgprot_nx(prot)	(prot)
1336 #endif
1337 
1338 #ifndef pgprot_noncached
1339 #define pgprot_noncached(prot)	(prot)
1340 #endif
1341 
1342 #ifndef pgprot_writecombine
1343 #define pgprot_writecombine pgprot_noncached
1344 #endif
1345 
1346 #ifndef pgprot_writethrough
1347 #define pgprot_writethrough pgprot_noncached
1348 #endif
1349 
1350 #ifndef pgprot_device
1351 #define pgprot_device pgprot_noncached
1352 #endif
1353 
1354 #ifndef pgprot_mhp
1355 #define pgprot_mhp(prot)	(prot)
1356 #endif
1357 
1358 #ifdef CONFIG_MMU
1359 #ifndef pgprot_modify
1360 #define pgprot_modify pgprot_modify
1361 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
1362 {
1363 	if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
1364 		newprot = pgprot_noncached(newprot);
1365 	if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
1366 		newprot = pgprot_writecombine(newprot);
1367 	if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
1368 		newprot = pgprot_device(newprot);
1369 	return newprot;
1370 }
1371 #endif
1372 #endif /* CONFIG_MMU */
1373 
1374 #ifndef pgprot_encrypted
1375 #define pgprot_encrypted(prot)	(prot)
1376 #endif
1377 
1378 #ifndef pgprot_decrypted
1379 #define pgprot_decrypted(prot)	(prot)
1380 #endif
1381 
1382 /*
1383  * A facility to provide batching of the reload of page tables and
1384  * other process state with the actual context switch code for
1385  * paravirtualized guests.  By convention, only one of the batched
1386  * update (lazy) modes (CPU, MMU) should be active at any given time,
1387  * entry should never be nested, and entry and exits should always be
1388  * paired.  This is for sanity of maintaining and reasoning about the
1389  * kernel code.  In this case, the exit (end of the context switch) is
1390  * in architecture-specific code, and so doesn't need a generic
1391  * definition.
1392  */
1393 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
1394 #define arch_start_context_switch(prev)	do {} while (0)
1395 #endif
1396 
1397 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1398 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
1399 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1400 {
1401 	return pmd;
1402 }
1403 
1404 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1405 {
1406 	return 0;
1407 }
1408 
1409 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1410 {
1411 	return pmd;
1412 }
1413 #endif
1414 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
1415 static inline int pte_soft_dirty(pte_t pte)
1416 {
1417 	return 0;
1418 }
1419 
1420 static inline int pmd_soft_dirty(pmd_t pmd)
1421 {
1422 	return 0;
1423 }
1424 
1425 static inline pte_t pte_mksoft_dirty(pte_t pte)
1426 {
1427 	return pte;
1428 }
1429 
1430 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
1431 {
1432 	return pmd;
1433 }
1434 
1435 static inline pte_t pte_clear_soft_dirty(pte_t pte)
1436 {
1437 	return pte;
1438 }
1439 
1440 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
1441 {
1442 	return pmd;
1443 }
1444 
1445 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1446 {
1447 	return pte;
1448 }
1449 
1450 static inline int pte_swp_soft_dirty(pte_t pte)
1451 {
1452 	return 0;
1453 }
1454 
1455 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1456 {
1457 	return pte;
1458 }
1459 
1460 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1461 {
1462 	return pmd;
1463 }
1464 
1465 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1466 {
1467 	return 0;
1468 }
1469 
1470 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1471 {
1472 	return pmd;
1473 }
1474 #endif
1475 
1476 #ifndef __HAVE_PFNMAP_TRACKING
1477 /*
1478  * Interfaces that can be used by architecture code to keep track of
1479  * memory type of pfn mappings specified by the remap_pfn_range,
1480  * vmf_insert_pfn.
1481  */
1482 
1483 /*
1484  * track_pfn_remap is called when a _new_ pfn mapping is being established
1485  * by remap_pfn_range() for physical range indicated by pfn and size.
1486  */
1487 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1488 				  unsigned long pfn, unsigned long addr,
1489 				  unsigned long size)
1490 {
1491 	return 0;
1492 }
1493 
1494 /*
1495  * track_pfn_insert is called when a _new_ single pfn is established
1496  * by vmf_insert_pfn().
1497  */
1498 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1499 				    pfn_t pfn)
1500 {
1501 }
1502 
1503 /*
1504  * track_pfn_copy is called when vma that is covering the pfnmap gets
1505  * copied through copy_page_range().
1506  */
1507 static inline int track_pfn_copy(struct vm_area_struct *vma)
1508 {
1509 	return 0;
1510 }
1511 
1512 /*
1513  * untrack_pfn is called while unmapping a pfnmap for a region.
1514  * untrack can be called for a specific region indicated by pfn and size or
1515  * can be for the entire vma (in which case pfn, size are zero).
1516  */
1517 static inline void untrack_pfn(struct vm_area_struct *vma,
1518 			       unsigned long pfn, unsigned long size,
1519 			       bool mm_wr_locked)
1520 {
1521 }
1522 
1523 /*
1524  * untrack_pfn_clear is called while mremapping a pfnmap for a new region
1525  * or fails to copy pgtable during duplicate vm area.
1526  */
1527 static inline void untrack_pfn_clear(struct vm_area_struct *vma)
1528 {
1529 }
1530 #else
1531 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1532 			   unsigned long pfn, unsigned long addr,
1533 			   unsigned long size);
1534 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1535 			     pfn_t pfn);
1536 extern int track_pfn_copy(struct vm_area_struct *vma);
1537 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1538 			unsigned long size, bool mm_wr_locked);
1539 extern void untrack_pfn_clear(struct vm_area_struct *vma);
1540 #endif
1541 
1542 #ifdef CONFIG_MMU
1543 #ifdef __HAVE_COLOR_ZERO_PAGE
1544 static inline int is_zero_pfn(unsigned long pfn)
1545 {
1546 	extern unsigned long zero_pfn;
1547 	unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1548 	return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1549 }
1550 
1551 #define my_zero_pfn(addr)	page_to_pfn(ZERO_PAGE(addr))
1552 
1553 #else
1554 static inline int is_zero_pfn(unsigned long pfn)
1555 {
1556 	extern unsigned long zero_pfn;
1557 	return pfn == zero_pfn;
1558 }
1559 
1560 static inline unsigned long my_zero_pfn(unsigned long addr)
1561 {
1562 	extern unsigned long zero_pfn;
1563 	return zero_pfn;
1564 }
1565 #endif
1566 #else
1567 static inline int is_zero_pfn(unsigned long pfn)
1568 {
1569 	return 0;
1570 }
1571 
1572 static inline unsigned long my_zero_pfn(unsigned long addr)
1573 {
1574 	return 0;
1575 }
1576 #endif /* CONFIG_MMU */
1577 
1578 #ifdef CONFIG_MMU
1579 
1580 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1581 static inline int pmd_trans_huge(pmd_t pmd)
1582 {
1583 	return 0;
1584 }
1585 #ifndef pmd_write
1586 static inline int pmd_write(pmd_t pmd)
1587 {
1588 	BUG();
1589 	return 0;
1590 }
1591 #endif /* pmd_write */
1592 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1593 
1594 #ifndef pud_write
1595 static inline int pud_write(pud_t pud)
1596 {
1597 	BUG();
1598 	return 0;
1599 }
1600 #endif /* pud_write */
1601 
1602 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1603 static inline int pmd_devmap(pmd_t pmd)
1604 {
1605 	return 0;
1606 }
1607 static inline int pud_devmap(pud_t pud)
1608 {
1609 	return 0;
1610 }
1611 static inline int pgd_devmap(pgd_t pgd)
1612 {
1613 	return 0;
1614 }
1615 #endif
1616 
1617 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1618 	!defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1619 static inline int pud_trans_huge(pud_t pud)
1620 {
1621 	return 0;
1622 }
1623 #endif
1624 
1625 static inline int pud_trans_unstable(pud_t *pud)
1626 {
1627 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1628 	defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1629 	pud_t pudval = READ_ONCE(*pud);
1630 
1631 	if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1632 		return 1;
1633 	if (unlikely(pud_bad(pudval))) {
1634 		pud_clear_bad(pud);
1635 		return 1;
1636 	}
1637 #endif
1638 	return 0;
1639 }
1640 
1641 #ifndef CONFIG_NUMA_BALANCING
1642 /*
1643  * In an inaccessible (PROT_NONE) VMA, pte_protnone() may indicate "yes". It is
1644  * perfectly valid to indicate "no" in that case, which is why our default
1645  * implementation defaults to "always no".
1646  *
1647  * In an accessible VMA, however, pte_protnone() reliably indicates PROT_NONE
1648  * page protection due to NUMA hinting. NUMA hinting faults only apply in
1649  * accessible VMAs.
1650  *
1651  * So, to reliably identify PROT_NONE PTEs that require a NUMA hinting fault,
1652  * looking at the VMA accessibility is sufficient.
1653  */
1654 static inline int pte_protnone(pte_t pte)
1655 {
1656 	return 0;
1657 }
1658 
1659 static inline int pmd_protnone(pmd_t pmd)
1660 {
1661 	return 0;
1662 }
1663 #endif /* CONFIG_NUMA_BALANCING */
1664 
1665 #endif /* CONFIG_MMU */
1666 
1667 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1668 
1669 #ifndef __PAGETABLE_P4D_FOLDED
1670 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1671 void p4d_clear_huge(p4d_t *p4d);
1672 #else
1673 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1674 {
1675 	return 0;
1676 }
1677 static inline void p4d_clear_huge(p4d_t *p4d) { }
1678 #endif /* !__PAGETABLE_P4D_FOLDED */
1679 
1680 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1681 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1682 int pud_clear_huge(pud_t *pud);
1683 int pmd_clear_huge(pmd_t *pmd);
1684 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1685 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1686 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1687 #else	/* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1688 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1689 {
1690 	return 0;
1691 }
1692 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1693 {
1694 	return 0;
1695 }
1696 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1697 {
1698 	return 0;
1699 }
1700 static inline void p4d_clear_huge(p4d_t *p4d) { }
1701 static inline int pud_clear_huge(pud_t *pud)
1702 {
1703 	return 0;
1704 }
1705 static inline int pmd_clear_huge(pmd_t *pmd)
1706 {
1707 	return 0;
1708 }
1709 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1710 {
1711 	return 0;
1712 }
1713 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1714 {
1715 	return 0;
1716 }
1717 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1718 {
1719 	return 0;
1720 }
1721 #endif	/* CONFIG_HAVE_ARCH_HUGE_VMAP */
1722 
1723 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1724 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1725 /*
1726  * ARCHes with special requirements for evicting THP backing TLB entries can
1727  * implement this. Otherwise also, it can help optimize normal TLB flush in
1728  * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1729  * entire TLB if flush span is greater than a threshold, which will
1730  * likely be true for a single huge page. Thus a single THP flush will
1731  * invalidate the entire TLB which is not desirable.
1732  * e.g. see arch/arc: flush_pmd_tlb_range
1733  */
1734 #define flush_pmd_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1735 #define flush_pud_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1736 #else
1737 #define flush_pmd_tlb_range(vma, addr, end)	BUILD_BUG()
1738 #define flush_pud_tlb_range(vma, addr, end)	BUILD_BUG()
1739 #endif
1740 #endif
1741 
1742 struct file;
1743 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1744 			unsigned long size, pgprot_t *vma_prot);
1745 
1746 #ifndef CONFIG_X86_ESPFIX64
1747 static inline void init_espfix_bsp(void) { }
1748 #endif
1749 
1750 extern void __init pgtable_cache_init(void);
1751 
1752 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1753 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1754 {
1755 	return true;
1756 }
1757 
1758 static inline bool arch_has_pfn_modify_check(void)
1759 {
1760 	return false;
1761 }
1762 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1763 
1764 /*
1765  * Architecture PAGE_KERNEL_* fallbacks
1766  *
1767  * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1768  * because they really don't support them, or the port needs to be updated to
1769  * reflect the required functionality. Below are a set of relatively safe
1770  * fallbacks, as best effort, which we can count on in lieu of the architectures
1771  * not defining them on their own yet.
1772  */
1773 
1774 #ifndef PAGE_KERNEL_RO
1775 # define PAGE_KERNEL_RO PAGE_KERNEL
1776 #endif
1777 
1778 #ifndef PAGE_KERNEL_EXEC
1779 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1780 #endif
1781 
1782 /*
1783  * Page Table Modification bits for pgtbl_mod_mask.
1784  *
1785  * These are used by the p?d_alloc_track*() set of functions an in the generic
1786  * vmalloc/ioremap code to track at which page-table levels entries have been
1787  * modified. Based on that the code can better decide when vmalloc and ioremap
1788  * mapping changes need to be synchronized to other page-tables in the system.
1789  */
1790 #define		__PGTBL_PGD_MODIFIED	0
1791 #define		__PGTBL_P4D_MODIFIED	1
1792 #define		__PGTBL_PUD_MODIFIED	2
1793 #define		__PGTBL_PMD_MODIFIED	3
1794 #define		__PGTBL_PTE_MODIFIED	4
1795 
1796 #define		PGTBL_PGD_MODIFIED	BIT(__PGTBL_PGD_MODIFIED)
1797 #define		PGTBL_P4D_MODIFIED	BIT(__PGTBL_P4D_MODIFIED)
1798 #define		PGTBL_PUD_MODIFIED	BIT(__PGTBL_PUD_MODIFIED)
1799 #define		PGTBL_PMD_MODIFIED	BIT(__PGTBL_PMD_MODIFIED)
1800 #define		PGTBL_PTE_MODIFIED	BIT(__PGTBL_PTE_MODIFIED)
1801 
1802 /* Page-Table Modification Mask */
1803 typedef unsigned int pgtbl_mod_mask;
1804 
1805 #endif /* !__ASSEMBLY__ */
1806 
1807 #if !defined(MAX_POSSIBLE_PHYSMEM_BITS) && !defined(CONFIG_64BIT)
1808 #ifdef CONFIG_PHYS_ADDR_T_64BIT
1809 /*
1810  * ZSMALLOC needs to know the highest PFN on 32-bit architectures
1811  * with physical address space extension, but falls back to
1812  * BITS_PER_LONG otherwise.
1813  */
1814 #error Missing MAX_POSSIBLE_PHYSMEM_BITS definition
1815 #else
1816 #define MAX_POSSIBLE_PHYSMEM_BITS 32
1817 #endif
1818 #endif
1819 
1820 #ifndef has_transparent_hugepage
1821 #define has_transparent_hugepage() IS_BUILTIN(CONFIG_TRANSPARENT_HUGEPAGE)
1822 #endif
1823 
1824 #ifndef has_transparent_pud_hugepage
1825 #define has_transparent_pud_hugepage() IS_BUILTIN(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1826 #endif
1827 /*
1828  * On some architectures it depends on the mm if the p4d/pud or pmd
1829  * layer of the page table hierarchy is folded or not.
1830  */
1831 #ifndef mm_p4d_folded
1832 #define mm_p4d_folded(mm)	__is_defined(__PAGETABLE_P4D_FOLDED)
1833 #endif
1834 
1835 #ifndef mm_pud_folded
1836 #define mm_pud_folded(mm)	__is_defined(__PAGETABLE_PUD_FOLDED)
1837 #endif
1838 
1839 #ifndef mm_pmd_folded
1840 #define mm_pmd_folded(mm)	__is_defined(__PAGETABLE_PMD_FOLDED)
1841 #endif
1842 
1843 #ifndef p4d_offset_lockless
1844 #define p4d_offset_lockless(pgdp, pgd, address) p4d_offset(&(pgd), address)
1845 #endif
1846 #ifndef pud_offset_lockless
1847 #define pud_offset_lockless(p4dp, p4d, address) pud_offset(&(p4d), address)
1848 #endif
1849 #ifndef pmd_offset_lockless
1850 #define pmd_offset_lockless(pudp, pud, address) pmd_offset(&(pud), address)
1851 #endif
1852 
1853 /*
1854  * pXd_leaf() is the API to check whether a pgtable entry is a huge page
1855  * mapping.  It should work globally across all archs, without any
1856  * dependency on CONFIG_* options.  For architectures that do not support
1857  * huge mappings on specific levels, below fallbacks will be used.
1858  *
1859  * A leaf pgtable entry should always imply the following:
1860  *
1861  * - It is a "present" entry.  IOW, before using this API, please check it
1862  *   with pXd_present() first. NOTE: it may not always mean the "present
1863  *   bit" is set.  For example, PROT_NONE entries are always "present".
1864  *
1865  * - It should _never_ be a swap entry of any type.  Above "present" check
1866  *   should have guarded this, but let's be crystal clear on this.
1867  *
1868  * - It should contain a huge PFN, which points to a huge page larger than
1869  *   PAGE_SIZE of the platform.  The PFN format isn't important here.
1870  *
1871  * - It should cover all kinds of huge mappings (e.g., pXd_trans_huge(),
1872  *   pXd_devmap(), or hugetlb mappings).
1873  */
1874 #ifndef pgd_leaf
1875 #define pgd_leaf(x)	false
1876 #endif
1877 #ifndef p4d_leaf
1878 #define p4d_leaf(x)	false
1879 #endif
1880 #ifndef pud_leaf
1881 #define pud_leaf(x)	false
1882 #endif
1883 #ifndef pmd_leaf
1884 #define pmd_leaf(x)	false
1885 #endif
1886 
1887 #ifndef pgd_leaf_size
1888 #define pgd_leaf_size(x) (1ULL << PGDIR_SHIFT)
1889 #endif
1890 #ifndef p4d_leaf_size
1891 #define p4d_leaf_size(x) P4D_SIZE
1892 #endif
1893 #ifndef pud_leaf_size
1894 #define pud_leaf_size(x) PUD_SIZE
1895 #endif
1896 #ifndef pmd_leaf_size
1897 #define pmd_leaf_size(x) PMD_SIZE
1898 #endif
1899 #ifndef __pte_leaf_size
1900 #ifndef pte_leaf_size
1901 #define pte_leaf_size(x) PAGE_SIZE
1902 #endif
1903 #define __pte_leaf_size(x,y) pte_leaf_size(y)
1904 #endif
1905 
1906 /*
1907  * We always define pmd_pfn for all archs as it's used in lots of generic
1908  * code.  Now it happens too for pud_pfn (and can happen for larger
1909  * mappings too in the future; we're not there yet).  Instead of defining
1910  * it for all archs (like pmd_pfn), provide a fallback.
1911  *
1912  * Note that returning 0 here means any arch that didn't define this can
1913  * get severely wrong when it hits a real pud leaf.  It's arch's
1914  * responsibility to properly define it when a huge pud is possible.
1915  */
1916 #ifndef pud_pfn
1917 #define pud_pfn(x) 0
1918 #endif
1919 
1920 /*
1921  * Some architectures have MMUs that are configurable or selectable at boot
1922  * time. These lead to variable PTRS_PER_x. For statically allocated arrays it
1923  * helps to have a static maximum value.
1924  */
1925 
1926 #ifndef MAX_PTRS_PER_PTE
1927 #define MAX_PTRS_PER_PTE PTRS_PER_PTE
1928 #endif
1929 
1930 #ifndef MAX_PTRS_PER_PMD
1931 #define MAX_PTRS_PER_PMD PTRS_PER_PMD
1932 #endif
1933 
1934 #ifndef MAX_PTRS_PER_PUD
1935 #define MAX_PTRS_PER_PUD PTRS_PER_PUD
1936 #endif
1937 
1938 #ifndef MAX_PTRS_PER_P4D
1939 #define MAX_PTRS_PER_P4D PTRS_PER_P4D
1940 #endif
1941 
1942 #ifndef pte_pgprot
1943 #define pte_pgprot(x) ((pgprot_t) {0})
1944 #endif
1945 
1946 #ifndef pmd_pgprot
1947 #define pmd_pgprot(x) ((pgprot_t) {0})
1948 #endif
1949 
1950 #ifndef pud_pgprot
1951 #define pud_pgprot(x) ((pgprot_t) {0})
1952 #endif
1953 
1954 /* description of effects of mapping type and prot in current implementation.
1955  * this is due to the limited x86 page protection hardware.  The expected
1956  * behavior is in parens:
1957  *
1958  * map_type	prot
1959  *		PROT_NONE	PROT_READ	PROT_WRITE	PROT_EXEC
1960  * MAP_SHARED	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1961  *		w: (no) no	w: (no) no	w: (yes) yes	w: (no) no
1962  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1963  *
1964  * MAP_PRIVATE	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1965  *		w: (no) no	w: (no) no	w: (copy) copy	w: (no) no
1966  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1967  *
1968  * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
1969  * MAP_PRIVATE (with Enhanced PAN supported):
1970  *								r: (no) no
1971  *								w: (no) no
1972  *								x: (yes) yes
1973  */
1974 #define DECLARE_VM_GET_PAGE_PROT					\
1975 pgprot_t vm_get_page_prot(unsigned long vm_flags)			\
1976 {									\
1977 		return protection_map[vm_flags &			\
1978 			(VM_READ | VM_WRITE | VM_EXEC | VM_SHARED)];	\
1979 }									\
1980 EXPORT_SYMBOL(vm_get_page_prot);
1981 
1982 #endif /* _LINUX_PGTABLE_H */
1983