xref: /linux-6.15/include/linux/pgtable.h (revision 23b1b44e)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_PGTABLE_H
3 #define _LINUX_PGTABLE_H
4 
5 #include <linux/pfn.h>
6 #include <asm/pgtable.h>
7 
8 #define PMD_ORDER	(PMD_SHIFT - PAGE_SHIFT)
9 #define PUD_ORDER	(PUD_SHIFT - PAGE_SHIFT)
10 
11 #ifndef __ASSEMBLY__
12 #ifdef CONFIG_MMU
13 
14 #include <linux/mm_types.h>
15 #include <linux/bug.h>
16 #include <linux/errno.h>
17 #include <asm-generic/pgtable_uffd.h>
18 #include <linux/page_table_check.h>
19 
20 #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
21 	defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
22 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
23 #endif
24 
25 /*
26  * On almost all architectures and configurations, 0 can be used as the
27  * upper ceiling to free_pgtables(): on many architectures it has the same
28  * effect as using TASK_SIZE.  However, there is one configuration which
29  * must impose a more careful limit, to avoid freeing kernel pgtables.
30  */
31 #ifndef USER_PGTABLES_CEILING
32 #define USER_PGTABLES_CEILING	0UL
33 #endif
34 
35 /*
36  * This defines the first usable user address. Platforms
37  * can override its value with custom FIRST_USER_ADDRESS
38  * defined in their respective <asm/pgtable.h>.
39  */
40 #ifndef FIRST_USER_ADDRESS
41 #define FIRST_USER_ADDRESS	0UL
42 #endif
43 
44 /*
45  * This defines the generic helper for accessing PMD page
46  * table page. Although platforms can still override this
47  * via their respective <asm/pgtable.h>.
48  */
49 #ifndef pmd_pgtable
50 #define pmd_pgtable(pmd) pmd_page(pmd)
51 #endif
52 
53 #define pmd_folio(pmd) page_folio(pmd_page(pmd))
54 
55 /*
56  * A page table page can be thought of an array like this: pXd_t[PTRS_PER_PxD]
57  *
58  * The pXx_index() functions return the index of the entry in the page
59  * table page which would control the given virtual address
60  *
61  * As these functions may be used by the same code for different levels of
62  * the page table folding, they are always available, regardless of
63  * CONFIG_PGTABLE_LEVELS value. For the folded levels they simply return 0
64  * because in such cases PTRS_PER_PxD equals 1.
65  */
66 
67 static inline unsigned long pte_index(unsigned long address)
68 {
69 	return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
70 }
71 
72 #ifndef pmd_index
73 static inline unsigned long pmd_index(unsigned long address)
74 {
75 	return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
76 }
77 #define pmd_index pmd_index
78 #endif
79 
80 #ifndef pud_index
81 static inline unsigned long pud_index(unsigned long address)
82 {
83 	return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
84 }
85 #define pud_index pud_index
86 #endif
87 
88 #ifndef pgd_index
89 /* Must be a compile-time constant, so implement it as a macro */
90 #define pgd_index(a)  (((a) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
91 #endif
92 
93 #ifndef pte_offset_kernel
94 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
95 {
96 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
97 }
98 #define pte_offset_kernel pte_offset_kernel
99 #endif
100 
101 #ifdef CONFIG_HIGHPTE
102 #define __pte_map(pmd, address) \
103 	((pte_t *)kmap_local_page(pmd_page(*(pmd))) + pte_index((address)))
104 #define pte_unmap(pte)	do {	\
105 	kunmap_local((pte));	\
106 	rcu_read_unlock();	\
107 } while (0)
108 #else
109 static inline pte_t *__pte_map(pmd_t *pmd, unsigned long address)
110 {
111 	return pte_offset_kernel(pmd, address);
112 }
113 static inline void pte_unmap(pte_t *pte)
114 {
115 	rcu_read_unlock();
116 }
117 #endif
118 
119 void pte_free_defer(struct mm_struct *mm, pgtable_t pgtable);
120 
121 /* Find an entry in the second-level page table.. */
122 #ifndef pmd_offset
123 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
124 {
125 	return pud_pgtable(*pud) + pmd_index(address);
126 }
127 #define pmd_offset pmd_offset
128 #endif
129 
130 #ifndef pud_offset
131 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
132 {
133 	return p4d_pgtable(*p4d) + pud_index(address);
134 }
135 #define pud_offset pud_offset
136 #endif
137 
138 static inline pgd_t *pgd_offset_pgd(pgd_t *pgd, unsigned long address)
139 {
140 	return (pgd + pgd_index(address));
141 };
142 
143 /*
144  * a shortcut to get a pgd_t in a given mm
145  */
146 #ifndef pgd_offset
147 #define pgd_offset(mm, address)		pgd_offset_pgd((mm)->pgd, (address))
148 #endif
149 
150 /*
151  * a shortcut which implies the use of the kernel's pgd, instead
152  * of a process's
153  */
154 #define pgd_offset_k(address)		pgd_offset(&init_mm, (address))
155 
156 /*
157  * In many cases it is known that a virtual address is mapped at PMD or PTE
158  * level, so instead of traversing all the page table levels, we can get a
159  * pointer to the PMD entry in user or kernel page table or translate a virtual
160  * address to the pointer in the PTE in the kernel page tables with simple
161  * helpers.
162  */
163 static inline pmd_t *pmd_off(struct mm_struct *mm, unsigned long va)
164 {
165 	return pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, va), va), va), va);
166 }
167 
168 static inline pmd_t *pmd_off_k(unsigned long va)
169 {
170 	return pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va);
171 }
172 
173 static inline pte_t *virt_to_kpte(unsigned long vaddr)
174 {
175 	pmd_t *pmd = pmd_off_k(vaddr);
176 
177 	return pmd_none(*pmd) ? NULL : pte_offset_kernel(pmd, vaddr);
178 }
179 
180 #ifndef pmd_young
181 static inline int pmd_young(pmd_t pmd)
182 {
183 	return 0;
184 }
185 #endif
186 
187 #ifndef pmd_dirty
188 static inline int pmd_dirty(pmd_t pmd)
189 {
190 	return 0;
191 }
192 #endif
193 
194 /*
195  * A facility to provide lazy MMU batching.  This allows PTE updates and
196  * page invalidations to be delayed until a call to leave lazy MMU mode
197  * is issued.  Some architectures may benefit from doing this, and it is
198  * beneficial for both shadow and direct mode hypervisors, which may batch
199  * the PTE updates which happen during this window.  Note that using this
200  * interface requires that read hazards be removed from the code.  A read
201  * hazard could result in the direct mode hypervisor case, since the actual
202  * write to the page tables may not yet have taken place, so reads though
203  * a raw PTE pointer after it has been modified are not guaranteed to be
204  * up to date.  This mode can only be entered and left under the protection of
205  * the page table locks for all page tables which may be modified.  In the UP
206  * case, this is required so that preemption is disabled, and in the SMP case,
207  * it must synchronize the delayed page table writes properly on other CPUs.
208  */
209 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
210 #define arch_enter_lazy_mmu_mode()	do {} while (0)
211 #define arch_leave_lazy_mmu_mode()	do {} while (0)
212 #define arch_flush_lazy_mmu_mode()	do {} while (0)
213 #endif
214 
215 #ifndef pte_batch_hint
216 /**
217  * pte_batch_hint - Number of pages that can be added to batch without scanning.
218  * @ptep: Page table pointer for the entry.
219  * @pte: Page table entry.
220  *
221  * Some architectures know that a set of contiguous ptes all map the same
222  * contiguous memory with the same permissions. In this case, it can provide a
223  * hint to aid pte batching without the core code needing to scan every pte.
224  *
225  * An architecture implementation may ignore the PTE accessed state. Further,
226  * the dirty state must apply atomically to all the PTEs described by the hint.
227  *
228  * May be overridden by the architecture, else pte_batch_hint is always 1.
229  */
230 static inline unsigned int pte_batch_hint(pte_t *ptep, pte_t pte)
231 {
232 	return 1;
233 }
234 #endif
235 
236 #ifndef pte_advance_pfn
237 static inline pte_t pte_advance_pfn(pte_t pte, unsigned long nr)
238 {
239 	return __pte(pte_val(pte) + (nr << PFN_PTE_SHIFT));
240 }
241 #endif
242 
243 #define pte_next_pfn(pte) pte_advance_pfn(pte, 1)
244 
245 #ifndef set_ptes
246 /**
247  * set_ptes - Map consecutive pages to a contiguous range of addresses.
248  * @mm: Address space to map the pages into.
249  * @addr: Address to map the first page at.
250  * @ptep: Page table pointer for the first entry.
251  * @pte: Page table entry for the first page.
252  * @nr: Number of pages to map.
253  *
254  * When nr==1, initial state of pte may be present or not present, and new state
255  * may be present or not present. When nr>1, initial state of all ptes must be
256  * not present, and new state must be present.
257  *
258  * May be overridden by the architecture, or the architecture can define
259  * set_pte() and PFN_PTE_SHIFT.
260  *
261  * Context: The caller holds the page table lock.  The pages all belong
262  * to the same folio.  The PTEs are all in the same PMD.
263  */
264 static inline void set_ptes(struct mm_struct *mm, unsigned long addr,
265 		pte_t *ptep, pte_t pte, unsigned int nr)
266 {
267 	page_table_check_ptes_set(mm, ptep, pte, nr);
268 
269 	arch_enter_lazy_mmu_mode();
270 	for (;;) {
271 		set_pte(ptep, pte);
272 		if (--nr == 0)
273 			break;
274 		ptep++;
275 		pte = pte_next_pfn(pte);
276 	}
277 	arch_leave_lazy_mmu_mode();
278 }
279 #endif
280 #define set_pte_at(mm, addr, ptep, pte) set_ptes(mm, addr, ptep, pte, 1)
281 
282 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
283 extern int ptep_set_access_flags(struct vm_area_struct *vma,
284 				 unsigned long address, pte_t *ptep,
285 				 pte_t entry, int dirty);
286 #endif
287 
288 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
289 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
290 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
291 				 unsigned long address, pmd_t *pmdp,
292 				 pmd_t entry, int dirty);
293 extern int pudp_set_access_flags(struct vm_area_struct *vma,
294 				 unsigned long address, pud_t *pudp,
295 				 pud_t entry, int dirty);
296 #else
297 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
298 					unsigned long address, pmd_t *pmdp,
299 					pmd_t entry, int dirty)
300 {
301 	BUILD_BUG();
302 	return 0;
303 }
304 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
305 					unsigned long address, pud_t *pudp,
306 					pud_t entry, int dirty)
307 {
308 	BUILD_BUG();
309 	return 0;
310 }
311 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
312 #endif
313 
314 #ifndef ptep_get
315 static inline pte_t ptep_get(pte_t *ptep)
316 {
317 	return READ_ONCE(*ptep);
318 }
319 #endif
320 
321 #ifndef pmdp_get
322 static inline pmd_t pmdp_get(pmd_t *pmdp)
323 {
324 	return READ_ONCE(*pmdp);
325 }
326 #endif
327 
328 #ifndef pudp_get
329 static inline pud_t pudp_get(pud_t *pudp)
330 {
331 	return READ_ONCE(*pudp);
332 }
333 #endif
334 
335 #ifndef p4dp_get
336 static inline p4d_t p4dp_get(p4d_t *p4dp)
337 {
338 	return READ_ONCE(*p4dp);
339 }
340 #endif
341 
342 #ifndef pgdp_get
343 static inline pgd_t pgdp_get(pgd_t *pgdp)
344 {
345 	return READ_ONCE(*pgdp);
346 }
347 #endif
348 
349 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
350 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
351 					    unsigned long address,
352 					    pte_t *ptep)
353 {
354 	pte_t pte = ptep_get(ptep);
355 	int r = 1;
356 	if (!pte_young(pte))
357 		r = 0;
358 	else
359 		set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
360 	return r;
361 }
362 #endif
363 
364 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
365 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG)
366 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
367 					    unsigned long address,
368 					    pmd_t *pmdp)
369 {
370 	pmd_t pmd = *pmdp;
371 	int r = 1;
372 	if (!pmd_young(pmd))
373 		r = 0;
374 	else
375 		set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
376 	return r;
377 }
378 #else
379 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
380 					    unsigned long address,
381 					    pmd_t *pmdp)
382 {
383 	BUILD_BUG();
384 	return 0;
385 }
386 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG */
387 #endif
388 
389 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
390 int ptep_clear_flush_young(struct vm_area_struct *vma,
391 			   unsigned long address, pte_t *ptep);
392 #endif
393 
394 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
395 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
396 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
397 				  unsigned long address, pmd_t *pmdp);
398 #else
399 /*
400  * Despite relevant to THP only, this API is called from generic rmap code
401  * under PageTransHuge(), hence needs a dummy implementation for !THP
402  */
403 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
404 					 unsigned long address, pmd_t *pmdp)
405 {
406 	BUILD_BUG();
407 	return 0;
408 }
409 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
410 #endif
411 
412 #ifndef arch_has_hw_nonleaf_pmd_young
413 /*
414  * Return whether the accessed bit in non-leaf PMD entries is supported on the
415  * local CPU.
416  */
417 static inline bool arch_has_hw_nonleaf_pmd_young(void)
418 {
419 	return IS_ENABLED(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG);
420 }
421 #endif
422 
423 #ifndef arch_has_hw_pte_young
424 /*
425  * Return whether the accessed bit is supported on the local CPU.
426  *
427  * This stub assumes accessing through an old PTE triggers a page fault.
428  * Architectures that automatically set the access bit should overwrite it.
429  */
430 static inline bool arch_has_hw_pte_young(void)
431 {
432 	return IS_ENABLED(CONFIG_ARCH_HAS_HW_PTE_YOUNG);
433 }
434 #endif
435 
436 #ifndef arch_check_zapped_pte
437 static inline void arch_check_zapped_pte(struct vm_area_struct *vma,
438 					 pte_t pte)
439 {
440 }
441 #endif
442 
443 #ifndef arch_check_zapped_pmd
444 static inline void arch_check_zapped_pmd(struct vm_area_struct *vma,
445 					 pmd_t pmd)
446 {
447 }
448 #endif
449 
450 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
451 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
452 				       unsigned long address,
453 				       pte_t *ptep)
454 {
455 	pte_t pte = ptep_get(ptep);
456 	pte_clear(mm, address, ptep);
457 	page_table_check_pte_clear(mm, pte);
458 	return pte;
459 }
460 #endif
461 
462 #ifndef clear_young_dirty_ptes
463 /**
464  * clear_young_dirty_ptes - Mark PTEs that map consecutive pages of the
465  *		same folio as old/clean.
466  * @mm: Address space the pages are mapped into.
467  * @addr: Address the first page is mapped at.
468  * @ptep: Page table pointer for the first entry.
469  * @nr: Number of entries to mark old/clean.
470  * @flags: Flags to modify the PTE batch semantics.
471  *
472  * May be overridden by the architecture; otherwise, implemented by
473  * get_and_clear/modify/set for each pte in the range.
474  *
475  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
476  * some PTEs might be write-protected.
477  *
478  * Context: The caller holds the page table lock.  The PTEs map consecutive
479  * pages that belong to the same folio.  The PTEs are all in the same PMD.
480  */
481 static inline void clear_young_dirty_ptes(struct vm_area_struct *vma,
482 					  unsigned long addr, pte_t *ptep,
483 					  unsigned int nr, cydp_t flags)
484 {
485 	pte_t pte;
486 
487 	for (;;) {
488 		if (flags == CYDP_CLEAR_YOUNG)
489 			ptep_test_and_clear_young(vma, addr, ptep);
490 		else {
491 			pte = ptep_get_and_clear(vma->vm_mm, addr, ptep);
492 			if (flags & CYDP_CLEAR_YOUNG)
493 				pte = pte_mkold(pte);
494 			if (flags & CYDP_CLEAR_DIRTY)
495 				pte = pte_mkclean(pte);
496 			set_pte_at(vma->vm_mm, addr, ptep, pte);
497 		}
498 		if (--nr == 0)
499 			break;
500 		ptep++;
501 		addr += PAGE_SIZE;
502 	}
503 }
504 #endif
505 
506 static inline void ptep_clear(struct mm_struct *mm, unsigned long addr,
507 			      pte_t *ptep)
508 {
509 	ptep_get_and_clear(mm, addr, ptep);
510 }
511 
512 #ifdef CONFIG_GUP_GET_PXX_LOW_HIGH
513 /*
514  * For walking the pagetables without holding any locks.  Some architectures
515  * (eg x86-32 PAE) cannot load the entries atomically without using expensive
516  * instructions.  We are guaranteed that a PTE will only either go from not
517  * present to present, or present to not present -- it will not switch to a
518  * completely different present page without a TLB flush inbetween; which we
519  * are blocking by holding interrupts off.
520  *
521  * Setting ptes from not present to present goes:
522  *
523  *   ptep->pte_high = h;
524  *   smp_wmb();
525  *   ptep->pte_low = l;
526  *
527  * And present to not present goes:
528  *
529  *   ptep->pte_low = 0;
530  *   smp_wmb();
531  *   ptep->pte_high = 0;
532  *
533  * We must ensure here that the load of pte_low sees 'l' IFF pte_high sees 'h'.
534  * We load pte_high *after* loading pte_low, which ensures we don't see an older
535  * value of pte_high.  *Then* we recheck pte_low, which ensures that we haven't
536  * picked up a changed pte high. We might have gotten rubbish values from
537  * pte_low and pte_high, but we are guaranteed that pte_low will not have the
538  * present bit set *unless* it is 'l'. Because get_user_pages_fast() only
539  * operates on present ptes we're safe.
540  */
541 static inline pte_t ptep_get_lockless(pte_t *ptep)
542 {
543 	pte_t pte;
544 
545 	do {
546 		pte.pte_low = ptep->pte_low;
547 		smp_rmb();
548 		pte.pte_high = ptep->pte_high;
549 		smp_rmb();
550 	} while (unlikely(pte.pte_low != ptep->pte_low));
551 
552 	return pte;
553 }
554 #define ptep_get_lockless ptep_get_lockless
555 
556 #if CONFIG_PGTABLE_LEVELS > 2
557 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
558 {
559 	pmd_t pmd;
560 
561 	do {
562 		pmd.pmd_low = pmdp->pmd_low;
563 		smp_rmb();
564 		pmd.pmd_high = pmdp->pmd_high;
565 		smp_rmb();
566 	} while (unlikely(pmd.pmd_low != pmdp->pmd_low));
567 
568 	return pmd;
569 }
570 #define pmdp_get_lockless pmdp_get_lockless
571 #define pmdp_get_lockless_sync() tlb_remove_table_sync_one()
572 #endif /* CONFIG_PGTABLE_LEVELS > 2 */
573 #endif /* CONFIG_GUP_GET_PXX_LOW_HIGH */
574 
575 /*
576  * We require that the PTE can be read atomically.
577  */
578 #ifndef ptep_get_lockless
579 static inline pte_t ptep_get_lockless(pte_t *ptep)
580 {
581 	return ptep_get(ptep);
582 }
583 #endif
584 
585 #ifndef pmdp_get_lockless
586 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
587 {
588 	return pmdp_get(pmdp);
589 }
590 static inline void pmdp_get_lockless_sync(void)
591 {
592 }
593 #endif
594 
595 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
596 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
597 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
598 					    unsigned long address,
599 					    pmd_t *pmdp)
600 {
601 	pmd_t pmd = *pmdp;
602 
603 	pmd_clear(pmdp);
604 	page_table_check_pmd_clear(mm, pmd);
605 
606 	return pmd;
607 }
608 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
609 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
610 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
611 					    unsigned long address,
612 					    pud_t *pudp)
613 {
614 	pud_t pud = *pudp;
615 
616 	pud_clear(pudp);
617 	page_table_check_pud_clear(mm, pud);
618 
619 	return pud;
620 }
621 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
622 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
623 
624 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
625 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
626 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
627 					    unsigned long address, pmd_t *pmdp,
628 					    int full)
629 {
630 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
631 }
632 #endif
633 
634 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
635 static inline pud_t pudp_huge_get_and_clear_full(struct vm_area_struct *vma,
636 					    unsigned long address, pud_t *pudp,
637 					    int full)
638 {
639 	return pudp_huge_get_and_clear(vma->vm_mm, address, pudp);
640 }
641 #endif
642 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
643 
644 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
645 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
646 					    unsigned long address, pte_t *ptep,
647 					    int full)
648 {
649 	return ptep_get_and_clear(mm, address, ptep);
650 }
651 #endif
652 
653 #ifndef get_and_clear_full_ptes
654 /**
655  * get_and_clear_full_ptes - Clear present PTEs that map consecutive pages of
656  *			     the same folio, collecting dirty/accessed bits.
657  * @mm: Address space the pages are mapped into.
658  * @addr: Address the first page is mapped at.
659  * @ptep: Page table pointer for the first entry.
660  * @nr: Number of entries to clear.
661  * @full: Whether we are clearing a full mm.
662  *
663  * May be overridden by the architecture; otherwise, implemented as a simple
664  * loop over ptep_get_and_clear_full(), merging dirty/accessed bits into the
665  * returned PTE.
666  *
667  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
668  * some PTEs might be write-protected.
669  *
670  * Context: The caller holds the page table lock.  The PTEs map consecutive
671  * pages that belong to the same folio.  The PTEs are all in the same PMD.
672  */
673 static inline pte_t get_and_clear_full_ptes(struct mm_struct *mm,
674 		unsigned long addr, pte_t *ptep, unsigned int nr, int full)
675 {
676 	pte_t pte, tmp_pte;
677 
678 	pte = ptep_get_and_clear_full(mm, addr, ptep, full);
679 	while (--nr) {
680 		ptep++;
681 		addr += PAGE_SIZE;
682 		tmp_pte = ptep_get_and_clear_full(mm, addr, ptep, full);
683 		if (pte_dirty(tmp_pte))
684 			pte = pte_mkdirty(pte);
685 		if (pte_young(tmp_pte))
686 			pte = pte_mkyoung(pte);
687 	}
688 	return pte;
689 }
690 #endif
691 
692 #ifndef clear_full_ptes
693 /**
694  * clear_full_ptes - Clear present PTEs that map consecutive pages of the same
695  *		     folio.
696  * @mm: Address space the pages are mapped into.
697  * @addr: Address the first page is mapped at.
698  * @ptep: Page table pointer for the first entry.
699  * @nr: Number of entries to clear.
700  * @full: Whether we are clearing a full mm.
701  *
702  * May be overridden by the architecture; otherwise, implemented as a simple
703  * loop over ptep_get_and_clear_full().
704  *
705  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
706  * some PTEs might be write-protected.
707  *
708  * Context: The caller holds the page table lock.  The PTEs map consecutive
709  * pages that belong to the same folio.  The PTEs are all in the same PMD.
710  */
711 static inline void clear_full_ptes(struct mm_struct *mm, unsigned long addr,
712 		pte_t *ptep, unsigned int nr, int full)
713 {
714 	for (;;) {
715 		ptep_get_and_clear_full(mm, addr, ptep, full);
716 		if (--nr == 0)
717 			break;
718 		ptep++;
719 		addr += PAGE_SIZE;
720 	}
721 }
722 #endif
723 
724 /*
725  * If two threads concurrently fault at the same page, the thread that
726  * won the race updates the PTE and its local TLB/Cache. The other thread
727  * gives up, simply does nothing, and continues; on architectures where
728  * software can update TLB,  local TLB can be updated here to avoid next page
729  * fault. This function updates TLB only, do nothing with cache or others.
730  * It is the difference with function update_mmu_cache.
731  */
732 #ifndef update_mmu_tlb_range
733 static inline void update_mmu_tlb_range(struct vm_area_struct *vma,
734 				unsigned long address, pte_t *ptep, unsigned int nr)
735 {
736 }
737 #endif
738 
739 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
740 static inline void update_mmu_tlb(struct vm_area_struct *vma,
741 				unsigned long address, pte_t *ptep)
742 {
743 }
744 #define __HAVE_ARCH_UPDATE_MMU_TLB
745 #endif
746 
747 /*
748  * Some architectures may be able to avoid expensive synchronization
749  * primitives when modifications are made to PTE's which are already
750  * not present, or in the process of an address space destruction.
751  */
752 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
753 static inline void pte_clear_not_present_full(struct mm_struct *mm,
754 					      unsigned long address,
755 					      pte_t *ptep,
756 					      int full)
757 {
758 	pte_clear(mm, address, ptep);
759 }
760 #endif
761 
762 #ifndef clear_not_present_full_ptes
763 /**
764  * clear_not_present_full_ptes - Clear multiple not present PTEs which are
765  *				 consecutive in the pgtable.
766  * @mm: Address space the ptes represent.
767  * @addr: Address of the first pte.
768  * @ptep: Page table pointer for the first entry.
769  * @nr: Number of entries to clear.
770  * @full: Whether we are clearing a full mm.
771  *
772  * May be overridden by the architecture; otherwise, implemented as a simple
773  * loop over pte_clear_not_present_full().
774  *
775  * Context: The caller holds the page table lock.  The PTEs are all not present.
776  * The PTEs are all in the same PMD.
777  */
778 static inline void clear_not_present_full_ptes(struct mm_struct *mm,
779 		unsigned long addr, pte_t *ptep, unsigned int nr, int full)
780 {
781 	for (;;) {
782 		pte_clear_not_present_full(mm, addr, ptep, full);
783 		if (--nr == 0)
784 			break;
785 		ptep++;
786 		addr += PAGE_SIZE;
787 	}
788 }
789 #endif
790 
791 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
792 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
793 			      unsigned long address,
794 			      pte_t *ptep);
795 #endif
796 
797 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
798 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
799 			      unsigned long address,
800 			      pmd_t *pmdp);
801 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
802 			      unsigned long address,
803 			      pud_t *pudp);
804 #endif
805 
806 #ifndef pte_mkwrite
807 static inline pte_t pte_mkwrite(pte_t pte, struct vm_area_struct *vma)
808 {
809 	return pte_mkwrite_novma(pte);
810 }
811 #endif
812 
813 #if defined(CONFIG_ARCH_WANT_PMD_MKWRITE) && !defined(pmd_mkwrite)
814 static inline pmd_t pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma)
815 {
816 	return pmd_mkwrite_novma(pmd);
817 }
818 #endif
819 
820 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
821 struct mm_struct;
822 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
823 {
824 	pte_t old_pte = ptep_get(ptep);
825 	set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
826 }
827 #endif
828 
829 #ifndef wrprotect_ptes
830 /**
831  * wrprotect_ptes - Write-protect PTEs that map consecutive pages of the same
832  *		    folio.
833  * @mm: Address space the pages are mapped into.
834  * @addr: Address the first page is mapped at.
835  * @ptep: Page table pointer for the first entry.
836  * @nr: Number of entries to write-protect.
837  *
838  * May be overridden by the architecture; otherwise, implemented as a simple
839  * loop over ptep_set_wrprotect().
840  *
841  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
842  * some PTEs might be write-protected.
843  *
844  * Context: The caller holds the page table lock.  The PTEs map consecutive
845  * pages that belong to the same folio.  The PTEs are all in the same PMD.
846  */
847 static inline void wrprotect_ptes(struct mm_struct *mm, unsigned long addr,
848 		pte_t *ptep, unsigned int nr)
849 {
850 	for (;;) {
851 		ptep_set_wrprotect(mm, addr, ptep);
852 		if (--nr == 0)
853 			break;
854 		ptep++;
855 		addr += PAGE_SIZE;
856 	}
857 }
858 #endif
859 
860 /*
861  * On some architectures hardware does not set page access bit when accessing
862  * memory page, it is responsibility of software setting this bit. It brings
863  * out extra page fault penalty to track page access bit. For optimization page
864  * access bit can be set during all page fault flow on these arches.
865  * To be differentiate with macro pte_mkyoung, this macro is used on platforms
866  * where software maintains page access bit.
867  */
868 #ifndef pte_sw_mkyoung
869 static inline pte_t pte_sw_mkyoung(pte_t pte)
870 {
871 	return pte;
872 }
873 #define pte_sw_mkyoung	pte_sw_mkyoung
874 #endif
875 
876 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
877 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
878 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
879 				      unsigned long address, pmd_t *pmdp)
880 {
881 	pmd_t old_pmd = *pmdp;
882 	set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
883 }
884 #else
885 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
886 				      unsigned long address, pmd_t *pmdp)
887 {
888 	BUILD_BUG();
889 }
890 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
891 #endif
892 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
893 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
894 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
895 static inline void pudp_set_wrprotect(struct mm_struct *mm,
896 				      unsigned long address, pud_t *pudp)
897 {
898 	pud_t old_pud = *pudp;
899 
900 	set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
901 }
902 #else
903 static inline void pudp_set_wrprotect(struct mm_struct *mm,
904 				      unsigned long address, pud_t *pudp)
905 {
906 	BUILD_BUG();
907 }
908 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
909 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
910 #endif
911 
912 #ifndef pmdp_collapse_flush
913 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
914 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
915 				 unsigned long address, pmd_t *pmdp);
916 #else
917 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
918 					unsigned long address,
919 					pmd_t *pmdp)
920 {
921 	BUILD_BUG();
922 	return *pmdp;
923 }
924 #define pmdp_collapse_flush pmdp_collapse_flush
925 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
926 #endif
927 
928 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
929 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
930 				       pgtable_t pgtable);
931 #endif
932 
933 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
934 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
935 #endif
936 
937 #ifndef arch_needs_pgtable_deposit
938 #define arch_needs_pgtable_deposit() (false)
939 #endif
940 
941 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
942 /*
943  * This is an implementation of pmdp_establish() that is only suitable for an
944  * architecture that doesn't have hardware dirty/accessed bits. In this case we
945  * can't race with CPU which sets these bits and non-atomic approach is fine.
946  */
947 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
948 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
949 {
950 	pmd_t old_pmd = *pmdp;
951 	set_pmd_at(vma->vm_mm, address, pmdp, pmd);
952 	return old_pmd;
953 }
954 #endif
955 
956 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
957 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
958 			    pmd_t *pmdp);
959 #endif
960 
961 #ifndef __HAVE_ARCH_PMDP_INVALIDATE_AD
962 
963 /*
964  * pmdp_invalidate_ad() invalidates the PMD while changing a transparent
965  * hugepage mapping in the page tables. This function is similar to
966  * pmdp_invalidate(), but should only be used if the access and dirty bits would
967  * not be cleared by the software in the new PMD value. The function ensures
968  * that hardware changes of the access and dirty bits updates would not be lost.
969  *
970  * Doing so can allow in certain architectures to avoid a TLB flush in most
971  * cases. Yet, another TLB flush might be necessary later if the PMD update
972  * itself requires such flush (e.g., if protection was set to be stricter). Yet,
973  * even when a TLB flush is needed because of the update, the caller may be able
974  * to batch these TLB flushing operations, so fewer TLB flush operations are
975  * needed.
976  */
977 extern pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma,
978 				unsigned long address, pmd_t *pmdp);
979 #endif
980 
981 #ifndef __HAVE_ARCH_PTE_SAME
982 static inline int pte_same(pte_t pte_a, pte_t pte_b)
983 {
984 	return pte_val(pte_a) == pte_val(pte_b);
985 }
986 #endif
987 
988 #ifndef __HAVE_ARCH_PTE_UNUSED
989 /*
990  * Some architectures provide facilities to virtualization guests
991  * so that they can flag allocated pages as unused. This allows the
992  * host to transparently reclaim unused pages. This function returns
993  * whether the pte's page is unused.
994  */
995 static inline int pte_unused(pte_t pte)
996 {
997 	return 0;
998 }
999 #endif
1000 
1001 #ifndef pte_access_permitted
1002 #define pte_access_permitted(pte, write) \
1003 	(pte_present(pte) && (!(write) || pte_write(pte)))
1004 #endif
1005 
1006 #ifndef pmd_access_permitted
1007 #define pmd_access_permitted(pmd, write) \
1008 	(pmd_present(pmd) && (!(write) || pmd_write(pmd)))
1009 #endif
1010 
1011 #ifndef pud_access_permitted
1012 #define pud_access_permitted(pud, write) \
1013 	(pud_present(pud) && (!(write) || pud_write(pud)))
1014 #endif
1015 
1016 #ifndef p4d_access_permitted
1017 #define p4d_access_permitted(p4d, write) \
1018 	(p4d_present(p4d) && (!(write) || p4d_write(p4d)))
1019 #endif
1020 
1021 #ifndef pgd_access_permitted
1022 #define pgd_access_permitted(pgd, write) \
1023 	(pgd_present(pgd) && (!(write) || pgd_write(pgd)))
1024 #endif
1025 
1026 #ifndef __HAVE_ARCH_PMD_SAME
1027 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
1028 {
1029 	return pmd_val(pmd_a) == pmd_val(pmd_b);
1030 }
1031 #endif
1032 
1033 #ifndef pud_same
1034 static inline int pud_same(pud_t pud_a, pud_t pud_b)
1035 {
1036 	return pud_val(pud_a) == pud_val(pud_b);
1037 }
1038 #define pud_same pud_same
1039 #endif
1040 
1041 #ifndef __HAVE_ARCH_P4D_SAME
1042 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
1043 {
1044 	return p4d_val(p4d_a) == p4d_val(p4d_b);
1045 }
1046 #endif
1047 
1048 #ifndef __HAVE_ARCH_PGD_SAME
1049 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
1050 {
1051 	return pgd_val(pgd_a) == pgd_val(pgd_b);
1052 }
1053 #endif
1054 
1055 /*
1056  * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
1057  * TLB flush will be required as a result of the "set". For example, use
1058  * in scenarios where it is known ahead of time that the routine is
1059  * setting non-present entries, or re-setting an existing entry to the
1060  * same value. Otherwise, use the typical "set" helpers and flush the
1061  * TLB.
1062  */
1063 #define set_pte_safe(ptep, pte) \
1064 ({ \
1065 	WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
1066 	set_pte(ptep, pte); \
1067 })
1068 
1069 #define set_pmd_safe(pmdp, pmd) \
1070 ({ \
1071 	WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
1072 	set_pmd(pmdp, pmd); \
1073 })
1074 
1075 #define set_pud_safe(pudp, pud) \
1076 ({ \
1077 	WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
1078 	set_pud(pudp, pud); \
1079 })
1080 
1081 #define set_p4d_safe(p4dp, p4d) \
1082 ({ \
1083 	WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
1084 	set_p4d(p4dp, p4d); \
1085 })
1086 
1087 #define set_pgd_safe(pgdp, pgd) \
1088 ({ \
1089 	WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
1090 	set_pgd(pgdp, pgd); \
1091 })
1092 
1093 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
1094 /*
1095  * Some architectures support metadata associated with a page. When a
1096  * page is being swapped out, this metadata must be saved so it can be
1097  * restored when the page is swapped back in. SPARC M7 and newer
1098  * processors support an ADI (Application Data Integrity) tag for the
1099  * page as metadata for the page. arch_do_swap_page() can restore this
1100  * metadata when a page is swapped back in.
1101  */
1102 static inline void arch_do_swap_page(struct mm_struct *mm,
1103 				     struct vm_area_struct *vma,
1104 				     unsigned long addr,
1105 				     pte_t pte, pte_t oldpte)
1106 {
1107 
1108 }
1109 #endif
1110 
1111 #ifndef __HAVE_ARCH_UNMAP_ONE
1112 /*
1113  * Some architectures support metadata associated with a page. When a
1114  * page is being swapped out, this metadata must be saved so it can be
1115  * restored when the page is swapped back in. SPARC M7 and newer
1116  * processors support an ADI (Application Data Integrity) tag for the
1117  * page as metadata for the page. arch_unmap_one() can save this
1118  * metadata on a swap-out of a page.
1119  */
1120 static inline int arch_unmap_one(struct mm_struct *mm,
1121 				  struct vm_area_struct *vma,
1122 				  unsigned long addr,
1123 				  pte_t orig_pte)
1124 {
1125 	return 0;
1126 }
1127 #endif
1128 
1129 /*
1130  * Allow architectures to preserve additional metadata associated with
1131  * swapped-out pages. The corresponding __HAVE_ARCH_SWAP_* macros and function
1132  * prototypes must be defined in the arch-specific asm/pgtable.h file.
1133  */
1134 #ifndef __HAVE_ARCH_PREPARE_TO_SWAP
1135 static inline int arch_prepare_to_swap(struct folio *folio)
1136 {
1137 	return 0;
1138 }
1139 #endif
1140 
1141 #ifndef __HAVE_ARCH_SWAP_INVALIDATE
1142 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
1143 {
1144 }
1145 
1146 static inline void arch_swap_invalidate_area(int type)
1147 {
1148 }
1149 #endif
1150 
1151 #ifndef __HAVE_ARCH_SWAP_RESTORE
1152 static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
1153 {
1154 }
1155 #endif
1156 
1157 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
1158 #define pgd_offset_gate(mm, addr)	pgd_offset(mm, addr)
1159 #endif
1160 
1161 #ifndef __HAVE_ARCH_MOVE_PTE
1162 #define move_pte(pte, old_addr, new_addr)	(pte)
1163 #endif
1164 
1165 #ifndef pte_accessible
1166 # define pte_accessible(mm, pte)	((void)(pte), 1)
1167 #endif
1168 
1169 #ifndef flush_tlb_fix_spurious_fault
1170 #define flush_tlb_fix_spurious_fault(vma, address, ptep) flush_tlb_page(vma, address)
1171 #endif
1172 
1173 /*
1174  * When walking page tables, get the address of the next boundary,
1175  * or the end address of the range if that comes earlier.  Although no
1176  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
1177  */
1178 
1179 #define pgd_addr_end(addr, end)						\
1180 ({	unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;	\
1181 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1182 })
1183 
1184 #ifndef p4d_addr_end
1185 #define p4d_addr_end(addr, end)						\
1186 ({	unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;	\
1187 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1188 })
1189 #endif
1190 
1191 #ifndef pud_addr_end
1192 #define pud_addr_end(addr, end)						\
1193 ({	unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;	\
1194 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1195 })
1196 #endif
1197 
1198 #ifndef pmd_addr_end
1199 #define pmd_addr_end(addr, end)						\
1200 ({	unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;	\
1201 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1202 })
1203 #endif
1204 
1205 /*
1206  * When walking page tables, we usually want to skip any p?d_none entries;
1207  * and any p?d_bad entries - reporting the error before resetting to none.
1208  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
1209  */
1210 void pgd_clear_bad(pgd_t *);
1211 
1212 #ifndef __PAGETABLE_P4D_FOLDED
1213 void p4d_clear_bad(p4d_t *);
1214 #else
1215 #define p4d_clear_bad(p4d)        do { } while (0)
1216 #endif
1217 
1218 #ifndef __PAGETABLE_PUD_FOLDED
1219 void pud_clear_bad(pud_t *);
1220 #else
1221 #define pud_clear_bad(p4d)        do { } while (0)
1222 #endif
1223 
1224 void pmd_clear_bad(pmd_t *);
1225 
1226 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
1227 {
1228 	if (pgd_none(*pgd))
1229 		return 1;
1230 	if (unlikely(pgd_bad(*pgd))) {
1231 		pgd_clear_bad(pgd);
1232 		return 1;
1233 	}
1234 	return 0;
1235 }
1236 
1237 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
1238 {
1239 	if (p4d_none(*p4d))
1240 		return 1;
1241 	if (unlikely(p4d_bad(*p4d))) {
1242 		p4d_clear_bad(p4d);
1243 		return 1;
1244 	}
1245 	return 0;
1246 }
1247 
1248 static inline int pud_none_or_clear_bad(pud_t *pud)
1249 {
1250 	if (pud_none(*pud))
1251 		return 1;
1252 	if (unlikely(pud_bad(*pud))) {
1253 		pud_clear_bad(pud);
1254 		return 1;
1255 	}
1256 	return 0;
1257 }
1258 
1259 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
1260 {
1261 	if (pmd_none(*pmd))
1262 		return 1;
1263 	if (unlikely(pmd_bad(*pmd))) {
1264 		pmd_clear_bad(pmd);
1265 		return 1;
1266 	}
1267 	return 0;
1268 }
1269 
1270 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
1271 					     unsigned long addr,
1272 					     pte_t *ptep)
1273 {
1274 	/*
1275 	 * Get the current pte state, but zero it out to make it
1276 	 * non-present, preventing the hardware from asynchronously
1277 	 * updating it.
1278 	 */
1279 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
1280 }
1281 
1282 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
1283 					     unsigned long addr,
1284 					     pte_t *ptep, pte_t pte)
1285 {
1286 	/*
1287 	 * The pte is non-present, so there's no hardware state to
1288 	 * preserve.
1289 	 */
1290 	set_pte_at(vma->vm_mm, addr, ptep, pte);
1291 }
1292 
1293 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1294 /*
1295  * Start a pte protection read-modify-write transaction, which
1296  * protects against asynchronous hardware modifications to the pte.
1297  * The intention is not to prevent the hardware from making pte
1298  * updates, but to prevent any updates it may make from being lost.
1299  *
1300  * This does not protect against other software modifications of the
1301  * pte; the appropriate pte lock must be held over the transaction.
1302  *
1303  * Note that this interface is intended to be batchable, meaning that
1304  * ptep_modify_prot_commit may not actually update the pte, but merely
1305  * queue the update to be done at some later time.  The update must be
1306  * actually committed before the pte lock is released, however.
1307  */
1308 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
1309 					   unsigned long addr,
1310 					   pte_t *ptep)
1311 {
1312 	return __ptep_modify_prot_start(vma, addr, ptep);
1313 }
1314 
1315 /*
1316  * Commit an update to a pte, leaving any hardware-controlled bits in
1317  * the PTE unmodified.
1318  */
1319 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
1320 					   unsigned long addr,
1321 					   pte_t *ptep, pte_t old_pte, pte_t pte)
1322 {
1323 	__ptep_modify_prot_commit(vma, addr, ptep, pte);
1324 }
1325 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
1326 #endif /* CONFIG_MMU */
1327 
1328 /*
1329  * No-op macros that just return the current protection value. Defined here
1330  * because these macros can be used even if CONFIG_MMU is not defined.
1331  */
1332 
1333 #ifndef pgprot_nx
1334 #define pgprot_nx(prot)	(prot)
1335 #endif
1336 
1337 #ifndef pgprot_noncached
1338 #define pgprot_noncached(prot)	(prot)
1339 #endif
1340 
1341 #ifndef pgprot_writecombine
1342 #define pgprot_writecombine pgprot_noncached
1343 #endif
1344 
1345 #ifndef pgprot_writethrough
1346 #define pgprot_writethrough pgprot_noncached
1347 #endif
1348 
1349 #ifndef pgprot_device
1350 #define pgprot_device pgprot_noncached
1351 #endif
1352 
1353 #ifndef pgprot_mhp
1354 #define pgprot_mhp(prot)	(prot)
1355 #endif
1356 
1357 #ifdef CONFIG_MMU
1358 #ifndef pgprot_modify
1359 #define pgprot_modify pgprot_modify
1360 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
1361 {
1362 	if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
1363 		newprot = pgprot_noncached(newprot);
1364 	if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
1365 		newprot = pgprot_writecombine(newprot);
1366 	if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
1367 		newprot = pgprot_device(newprot);
1368 	return newprot;
1369 }
1370 #endif
1371 #endif /* CONFIG_MMU */
1372 
1373 #ifndef pgprot_encrypted
1374 #define pgprot_encrypted(prot)	(prot)
1375 #endif
1376 
1377 #ifndef pgprot_decrypted
1378 #define pgprot_decrypted(prot)	(prot)
1379 #endif
1380 
1381 /*
1382  * A facility to provide batching of the reload of page tables and
1383  * other process state with the actual context switch code for
1384  * paravirtualized guests.  By convention, only one of the batched
1385  * update (lazy) modes (CPU, MMU) should be active at any given time,
1386  * entry should never be nested, and entry and exits should always be
1387  * paired.  This is for sanity of maintaining and reasoning about the
1388  * kernel code.  In this case, the exit (end of the context switch) is
1389  * in architecture-specific code, and so doesn't need a generic
1390  * definition.
1391  */
1392 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
1393 #define arch_start_context_switch(prev)	do {} while (0)
1394 #endif
1395 
1396 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1397 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
1398 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1399 {
1400 	return pmd;
1401 }
1402 
1403 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1404 {
1405 	return 0;
1406 }
1407 
1408 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1409 {
1410 	return pmd;
1411 }
1412 #endif
1413 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
1414 static inline int pte_soft_dirty(pte_t pte)
1415 {
1416 	return 0;
1417 }
1418 
1419 static inline int pmd_soft_dirty(pmd_t pmd)
1420 {
1421 	return 0;
1422 }
1423 
1424 static inline pte_t pte_mksoft_dirty(pte_t pte)
1425 {
1426 	return pte;
1427 }
1428 
1429 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
1430 {
1431 	return pmd;
1432 }
1433 
1434 static inline pte_t pte_clear_soft_dirty(pte_t pte)
1435 {
1436 	return pte;
1437 }
1438 
1439 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
1440 {
1441 	return pmd;
1442 }
1443 
1444 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1445 {
1446 	return pte;
1447 }
1448 
1449 static inline int pte_swp_soft_dirty(pte_t pte)
1450 {
1451 	return 0;
1452 }
1453 
1454 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1455 {
1456 	return pte;
1457 }
1458 
1459 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1460 {
1461 	return pmd;
1462 }
1463 
1464 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1465 {
1466 	return 0;
1467 }
1468 
1469 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1470 {
1471 	return pmd;
1472 }
1473 #endif
1474 
1475 #ifndef __HAVE_PFNMAP_TRACKING
1476 /*
1477  * Interfaces that can be used by architecture code to keep track of
1478  * memory type of pfn mappings specified by the remap_pfn_range,
1479  * vmf_insert_pfn.
1480  */
1481 
1482 /*
1483  * track_pfn_remap is called when a _new_ pfn mapping is being established
1484  * by remap_pfn_range() for physical range indicated by pfn and size.
1485  */
1486 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1487 				  unsigned long pfn, unsigned long addr,
1488 				  unsigned long size)
1489 {
1490 	return 0;
1491 }
1492 
1493 /*
1494  * track_pfn_insert is called when a _new_ single pfn is established
1495  * by vmf_insert_pfn().
1496  */
1497 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1498 				    pfn_t pfn)
1499 {
1500 }
1501 
1502 /*
1503  * track_pfn_copy is called when vma that is covering the pfnmap gets
1504  * copied through copy_page_range().
1505  */
1506 static inline int track_pfn_copy(struct vm_area_struct *vma)
1507 {
1508 	return 0;
1509 }
1510 
1511 /*
1512  * untrack_pfn is called while unmapping a pfnmap for a region.
1513  * untrack can be called for a specific region indicated by pfn and size or
1514  * can be for the entire vma (in which case pfn, size are zero).
1515  */
1516 static inline void untrack_pfn(struct vm_area_struct *vma,
1517 			       unsigned long pfn, unsigned long size,
1518 			       bool mm_wr_locked)
1519 {
1520 }
1521 
1522 /*
1523  * untrack_pfn_clear is called while mremapping a pfnmap for a new region
1524  * or fails to copy pgtable during duplicate vm area.
1525  */
1526 static inline void untrack_pfn_clear(struct vm_area_struct *vma)
1527 {
1528 }
1529 #else
1530 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1531 			   unsigned long pfn, unsigned long addr,
1532 			   unsigned long size);
1533 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1534 			     pfn_t pfn);
1535 extern int track_pfn_copy(struct vm_area_struct *vma);
1536 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1537 			unsigned long size, bool mm_wr_locked);
1538 extern void untrack_pfn_clear(struct vm_area_struct *vma);
1539 #endif
1540 
1541 #ifdef CONFIG_MMU
1542 #ifdef __HAVE_COLOR_ZERO_PAGE
1543 static inline int is_zero_pfn(unsigned long pfn)
1544 {
1545 	extern unsigned long zero_pfn;
1546 	unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1547 	return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1548 }
1549 
1550 #define my_zero_pfn(addr)	page_to_pfn(ZERO_PAGE(addr))
1551 
1552 #else
1553 static inline int is_zero_pfn(unsigned long pfn)
1554 {
1555 	extern unsigned long zero_pfn;
1556 	return pfn == zero_pfn;
1557 }
1558 
1559 static inline unsigned long my_zero_pfn(unsigned long addr)
1560 {
1561 	extern unsigned long zero_pfn;
1562 	return zero_pfn;
1563 }
1564 #endif
1565 #else
1566 static inline int is_zero_pfn(unsigned long pfn)
1567 {
1568 	return 0;
1569 }
1570 
1571 static inline unsigned long my_zero_pfn(unsigned long addr)
1572 {
1573 	return 0;
1574 }
1575 #endif /* CONFIG_MMU */
1576 
1577 #ifdef CONFIG_MMU
1578 
1579 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1580 static inline int pmd_trans_huge(pmd_t pmd)
1581 {
1582 	return 0;
1583 }
1584 #ifndef pmd_write
1585 static inline int pmd_write(pmd_t pmd)
1586 {
1587 	BUG();
1588 	return 0;
1589 }
1590 #endif /* pmd_write */
1591 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1592 
1593 #ifndef pud_write
1594 static inline int pud_write(pud_t pud)
1595 {
1596 	BUG();
1597 	return 0;
1598 }
1599 #endif /* pud_write */
1600 
1601 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1602 static inline int pmd_devmap(pmd_t pmd)
1603 {
1604 	return 0;
1605 }
1606 static inline int pud_devmap(pud_t pud)
1607 {
1608 	return 0;
1609 }
1610 static inline int pgd_devmap(pgd_t pgd)
1611 {
1612 	return 0;
1613 }
1614 #endif
1615 
1616 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1617 	!defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1618 static inline int pud_trans_huge(pud_t pud)
1619 {
1620 	return 0;
1621 }
1622 #endif
1623 
1624 static inline int pud_trans_unstable(pud_t *pud)
1625 {
1626 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1627 	defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1628 	pud_t pudval = READ_ONCE(*pud);
1629 
1630 	if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1631 		return 1;
1632 	if (unlikely(pud_bad(pudval))) {
1633 		pud_clear_bad(pud);
1634 		return 1;
1635 	}
1636 #endif
1637 	return 0;
1638 }
1639 
1640 #ifndef CONFIG_NUMA_BALANCING
1641 /*
1642  * In an inaccessible (PROT_NONE) VMA, pte_protnone() may indicate "yes". It is
1643  * perfectly valid to indicate "no" in that case, which is why our default
1644  * implementation defaults to "always no".
1645  *
1646  * In an accessible VMA, however, pte_protnone() reliably indicates PROT_NONE
1647  * page protection due to NUMA hinting. NUMA hinting faults only apply in
1648  * accessible VMAs.
1649  *
1650  * So, to reliably identify PROT_NONE PTEs that require a NUMA hinting fault,
1651  * looking at the VMA accessibility is sufficient.
1652  */
1653 static inline int pte_protnone(pte_t pte)
1654 {
1655 	return 0;
1656 }
1657 
1658 static inline int pmd_protnone(pmd_t pmd)
1659 {
1660 	return 0;
1661 }
1662 #endif /* CONFIG_NUMA_BALANCING */
1663 
1664 #endif /* CONFIG_MMU */
1665 
1666 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1667 
1668 #ifndef __PAGETABLE_P4D_FOLDED
1669 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1670 void p4d_clear_huge(p4d_t *p4d);
1671 #else
1672 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1673 {
1674 	return 0;
1675 }
1676 static inline void p4d_clear_huge(p4d_t *p4d) { }
1677 #endif /* !__PAGETABLE_P4D_FOLDED */
1678 
1679 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1680 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1681 int pud_clear_huge(pud_t *pud);
1682 int pmd_clear_huge(pmd_t *pmd);
1683 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1684 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1685 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1686 #else	/* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1687 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1688 {
1689 	return 0;
1690 }
1691 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1692 {
1693 	return 0;
1694 }
1695 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1696 {
1697 	return 0;
1698 }
1699 static inline void p4d_clear_huge(p4d_t *p4d) { }
1700 static inline int pud_clear_huge(pud_t *pud)
1701 {
1702 	return 0;
1703 }
1704 static inline int pmd_clear_huge(pmd_t *pmd)
1705 {
1706 	return 0;
1707 }
1708 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1709 {
1710 	return 0;
1711 }
1712 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1713 {
1714 	return 0;
1715 }
1716 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1717 {
1718 	return 0;
1719 }
1720 #endif	/* CONFIG_HAVE_ARCH_HUGE_VMAP */
1721 
1722 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1723 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1724 /*
1725  * ARCHes with special requirements for evicting THP backing TLB entries can
1726  * implement this. Otherwise also, it can help optimize normal TLB flush in
1727  * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1728  * entire TLB if flush span is greater than a threshold, which will
1729  * likely be true for a single huge page. Thus a single THP flush will
1730  * invalidate the entire TLB which is not desirable.
1731  * e.g. see arch/arc: flush_pmd_tlb_range
1732  */
1733 #define flush_pmd_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1734 #define flush_pud_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1735 #else
1736 #define flush_pmd_tlb_range(vma, addr, end)	BUILD_BUG()
1737 #define flush_pud_tlb_range(vma, addr, end)	BUILD_BUG()
1738 #endif
1739 #endif
1740 
1741 struct file;
1742 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1743 			unsigned long size, pgprot_t *vma_prot);
1744 
1745 #ifndef CONFIG_X86_ESPFIX64
1746 static inline void init_espfix_bsp(void) { }
1747 #endif
1748 
1749 extern void __init pgtable_cache_init(void);
1750 
1751 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1752 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1753 {
1754 	return true;
1755 }
1756 
1757 static inline bool arch_has_pfn_modify_check(void)
1758 {
1759 	return false;
1760 }
1761 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1762 
1763 /*
1764  * Architecture PAGE_KERNEL_* fallbacks
1765  *
1766  * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1767  * because they really don't support them, or the port needs to be updated to
1768  * reflect the required functionality. Below are a set of relatively safe
1769  * fallbacks, as best effort, which we can count on in lieu of the architectures
1770  * not defining them on their own yet.
1771  */
1772 
1773 #ifndef PAGE_KERNEL_RO
1774 # define PAGE_KERNEL_RO PAGE_KERNEL
1775 #endif
1776 
1777 #ifndef PAGE_KERNEL_EXEC
1778 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1779 #endif
1780 
1781 /*
1782  * Page Table Modification bits for pgtbl_mod_mask.
1783  *
1784  * These are used by the p?d_alloc_track*() set of functions an in the generic
1785  * vmalloc/ioremap code to track at which page-table levels entries have been
1786  * modified. Based on that the code can better decide when vmalloc and ioremap
1787  * mapping changes need to be synchronized to other page-tables in the system.
1788  */
1789 #define		__PGTBL_PGD_MODIFIED	0
1790 #define		__PGTBL_P4D_MODIFIED	1
1791 #define		__PGTBL_PUD_MODIFIED	2
1792 #define		__PGTBL_PMD_MODIFIED	3
1793 #define		__PGTBL_PTE_MODIFIED	4
1794 
1795 #define		PGTBL_PGD_MODIFIED	BIT(__PGTBL_PGD_MODIFIED)
1796 #define		PGTBL_P4D_MODIFIED	BIT(__PGTBL_P4D_MODIFIED)
1797 #define		PGTBL_PUD_MODIFIED	BIT(__PGTBL_PUD_MODIFIED)
1798 #define		PGTBL_PMD_MODIFIED	BIT(__PGTBL_PMD_MODIFIED)
1799 #define		PGTBL_PTE_MODIFIED	BIT(__PGTBL_PTE_MODIFIED)
1800 
1801 /* Page-Table Modification Mask */
1802 typedef unsigned int pgtbl_mod_mask;
1803 
1804 #endif /* !__ASSEMBLY__ */
1805 
1806 #if !defined(MAX_POSSIBLE_PHYSMEM_BITS) && !defined(CONFIG_64BIT)
1807 #ifdef CONFIG_PHYS_ADDR_T_64BIT
1808 /*
1809  * ZSMALLOC needs to know the highest PFN on 32-bit architectures
1810  * with physical address space extension, but falls back to
1811  * BITS_PER_LONG otherwise.
1812  */
1813 #error Missing MAX_POSSIBLE_PHYSMEM_BITS definition
1814 #else
1815 #define MAX_POSSIBLE_PHYSMEM_BITS 32
1816 #endif
1817 #endif
1818 
1819 #ifndef has_transparent_hugepage
1820 #define has_transparent_hugepage() IS_BUILTIN(CONFIG_TRANSPARENT_HUGEPAGE)
1821 #endif
1822 
1823 #ifndef has_transparent_pud_hugepage
1824 #define has_transparent_pud_hugepage() IS_BUILTIN(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1825 #endif
1826 /*
1827  * On some architectures it depends on the mm if the p4d/pud or pmd
1828  * layer of the page table hierarchy is folded or not.
1829  */
1830 #ifndef mm_p4d_folded
1831 #define mm_p4d_folded(mm)	__is_defined(__PAGETABLE_P4D_FOLDED)
1832 #endif
1833 
1834 #ifndef mm_pud_folded
1835 #define mm_pud_folded(mm)	__is_defined(__PAGETABLE_PUD_FOLDED)
1836 #endif
1837 
1838 #ifndef mm_pmd_folded
1839 #define mm_pmd_folded(mm)	__is_defined(__PAGETABLE_PMD_FOLDED)
1840 #endif
1841 
1842 #ifndef p4d_offset_lockless
1843 #define p4d_offset_lockless(pgdp, pgd, address) p4d_offset(&(pgd), address)
1844 #endif
1845 #ifndef pud_offset_lockless
1846 #define pud_offset_lockless(p4dp, p4d, address) pud_offset(&(p4d), address)
1847 #endif
1848 #ifndef pmd_offset_lockless
1849 #define pmd_offset_lockless(pudp, pud, address) pmd_offset(&(pud), address)
1850 #endif
1851 
1852 /*
1853  * pXd_leaf() is the API to check whether a pgtable entry is a huge page
1854  * mapping.  It should work globally across all archs, without any
1855  * dependency on CONFIG_* options.  For architectures that do not support
1856  * huge mappings on specific levels, below fallbacks will be used.
1857  *
1858  * A leaf pgtable entry should always imply the following:
1859  *
1860  * - It is a "present" entry.  IOW, before using this API, please check it
1861  *   with pXd_present() first. NOTE: it may not always mean the "present
1862  *   bit" is set.  For example, PROT_NONE entries are always "present".
1863  *
1864  * - It should _never_ be a swap entry of any type.  Above "present" check
1865  *   should have guarded this, but let's be crystal clear on this.
1866  *
1867  * - It should contain a huge PFN, which points to a huge page larger than
1868  *   PAGE_SIZE of the platform.  The PFN format isn't important here.
1869  *
1870  * - It should cover all kinds of huge mappings (e.g., pXd_trans_huge(),
1871  *   pXd_devmap(), or hugetlb mappings).
1872  */
1873 #ifndef pgd_leaf
1874 #define pgd_leaf(x)	false
1875 #endif
1876 #ifndef p4d_leaf
1877 #define p4d_leaf(x)	false
1878 #endif
1879 #ifndef pud_leaf
1880 #define pud_leaf(x)	false
1881 #endif
1882 #ifndef pmd_leaf
1883 #define pmd_leaf(x)	false
1884 #endif
1885 
1886 #ifndef pgd_leaf_size
1887 #define pgd_leaf_size(x) (1ULL << PGDIR_SHIFT)
1888 #endif
1889 #ifndef p4d_leaf_size
1890 #define p4d_leaf_size(x) P4D_SIZE
1891 #endif
1892 #ifndef pud_leaf_size
1893 #define pud_leaf_size(x) PUD_SIZE
1894 #endif
1895 #ifndef pmd_leaf_size
1896 #define pmd_leaf_size(x) PMD_SIZE
1897 #endif
1898 #ifndef pte_leaf_size
1899 #define pte_leaf_size(x) PAGE_SIZE
1900 #endif
1901 
1902 /*
1903  * We always define pmd_pfn for all archs as it's used in lots of generic
1904  * code.  Now it happens too for pud_pfn (and can happen for larger
1905  * mappings too in the future; we're not there yet).  Instead of defining
1906  * it for all archs (like pmd_pfn), provide a fallback.
1907  *
1908  * Note that returning 0 here means any arch that didn't define this can
1909  * get severely wrong when it hits a real pud leaf.  It's arch's
1910  * responsibility to properly define it when a huge pud is possible.
1911  */
1912 #ifndef pud_pfn
1913 #define pud_pfn(x) 0
1914 #endif
1915 
1916 /*
1917  * Some architectures have MMUs that are configurable or selectable at boot
1918  * time. These lead to variable PTRS_PER_x. For statically allocated arrays it
1919  * helps to have a static maximum value.
1920  */
1921 
1922 #ifndef MAX_PTRS_PER_PTE
1923 #define MAX_PTRS_PER_PTE PTRS_PER_PTE
1924 #endif
1925 
1926 #ifndef MAX_PTRS_PER_PMD
1927 #define MAX_PTRS_PER_PMD PTRS_PER_PMD
1928 #endif
1929 
1930 #ifndef MAX_PTRS_PER_PUD
1931 #define MAX_PTRS_PER_PUD PTRS_PER_PUD
1932 #endif
1933 
1934 #ifndef MAX_PTRS_PER_P4D
1935 #define MAX_PTRS_PER_P4D PTRS_PER_P4D
1936 #endif
1937 
1938 /* description of effects of mapping type and prot in current implementation.
1939  * this is due to the limited x86 page protection hardware.  The expected
1940  * behavior is in parens:
1941  *
1942  * map_type	prot
1943  *		PROT_NONE	PROT_READ	PROT_WRITE	PROT_EXEC
1944  * MAP_SHARED	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1945  *		w: (no) no	w: (no) no	w: (yes) yes	w: (no) no
1946  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1947  *
1948  * MAP_PRIVATE	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1949  *		w: (no) no	w: (no) no	w: (copy) copy	w: (no) no
1950  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1951  *
1952  * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
1953  * MAP_PRIVATE (with Enhanced PAN supported):
1954  *								r: (no) no
1955  *								w: (no) no
1956  *								x: (yes) yes
1957  */
1958 #define DECLARE_VM_GET_PAGE_PROT					\
1959 pgprot_t vm_get_page_prot(unsigned long vm_flags)			\
1960 {									\
1961 		return protection_map[vm_flags &			\
1962 			(VM_READ | VM_WRITE | VM_EXEC | VM_SHARED)];	\
1963 }									\
1964 EXPORT_SYMBOL(vm_get_page_prot);
1965 
1966 #endif /* _LINUX_PGTABLE_H */
1967