xref: /linux-6.15/include/linux/pgtable.h (revision dfbe56bf)
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 #ifndef __ASSEMBLY__
9 #ifdef CONFIG_MMU
10 
11 #include <linux/mm_types.h>
12 #include <linux/bug.h>
13 #include <linux/errno.h>
14 #include <asm-generic/pgtable_uffd.h>
15 
16 #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
17 	defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
18 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
19 #endif
20 
21 /*
22  * On almost all architectures and configurations, 0 can be used as the
23  * upper ceiling to free_pgtables(): on many architectures it has the same
24  * effect as using TASK_SIZE.  However, there is one configuration which
25  * must impose a more careful limit, to avoid freeing kernel pgtables.
26  */
27 #ifndef USER_PGTABLES_CEILING
28 #define USER_PGTABLES_CEILING	0UL
29 #endif
30 
31 /*
32  * A page table page can be thought of an array like this: pXd_t[PTRS_PER_PxD]
33  *
34  * The pXx_index() functions return the index of the entry in the page
35  * table page which would control the given virtual address
36  *
37  * As these functions may be used by the same code for different levels of
38  * the page table folding, they are always available, regardless of
39  * CONFIG_PGTABLE_LEVELS value. For the folded levels they simply return 0
40  * because in such cases PTRS_PER_PxD equals 1.
41  */
42 
43 static inline unsigned long pte_index(unsigned long address)
44 {
45 	return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
46 }
47 
48 #ifndef pmd_index
49 static inline unsigned long pmd_index(unsigned long address)
50 {
51 	return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
52 }
53 #define pmd_index pmd_index
54 #endif
55 
56 #ifndef pud_index
57 static inline unsigned long pud_index(unsigned long address)
58 {
59 	return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
60 }
61 #define pud_index pud_index
62 #endif
63 
64 #ifndef pgd_index
65 /* Must be a compile-time constant, so implement it as a macro */
66 #define pgd_index(a)  (((a) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
67 #endif
68 
69 #ifndef pte_offset_kernel
70 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
71 {
72 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
73 }
74 #define pte_offset_kernel pte_offset_kernel
75 #endif
76 
77 #if defined(CONFIG_HIGHPTE)
78 #define pte_offset_map(dir, address)				\
79 	((pte_t *)kmap_atomic(pmd_page(*(dir))) +		\
80 	 pte_index((address)))
81 #define pte_unmap(pte) kunmap_atomic((pte))
82 #else
83 #define pte_offset_map(dir, address)	pte_offset_kernel((dir), (address))
84 #define pte_unmap(pte) ((void)(pte))	/* NOP */
85 #endif
86 
87 /* Find an entry in the second-level page table.. */
88 #ifndef pmd_offset
89 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
90 {
91 	return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(address);
92 }
93 #define pmd_offset pmd_offset
94 #endif
95 
96 #ifndef pud_offset
97 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
98 {
99 	return (pud_t *)p4d_page_vaddr(*p4d) + pud_index(address);
100 }
101 #define pud_offset pud_offset
102 #endif
103 
104 static inline pgd_t *pgd_offset_pgd(pgd_t *pgd, unsigned long address)
105 {
106 	return (pgd + pgd_index(address));
107 };
108 
109 /*
110  * a shortcut to get a pgd_t in a given mm
111  */
112 #ifndef pgd_offset
113 #define pgd_offset(mm, address)		pgd_offset_pgd((mm)->pgd, (address))
114 #endif
115 
116 /*
117  * a shortcut which implies the use of the kernel's pgd, instead
118  * of a process's
119  */
120 #ifndef pgd_offset_k
121 #define pgd_offset_k(address)		pgd_offset(&init_mm, (address))
122 #endif
123 
124 /*
125  * In many cases it is known that a virtual address is mapped at PMD or PTE
126  * level, so instead of traversing all the page table levels, we can get a
127  * pointer to the PMD entry in user or kernel page table or translate a virtual
128  * address to the pointer in the PTE in the kernel page tables with simple
129  * helpers.
130  */
131 static inline pmd_t *pmd_off(struct mm_struct *mm, unsigned long va)
132 {
133 	return pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, va), va), va), va);
134 }
135 
136 static inline pmd_t *pmd_off_k(unsigned long va)
137 {
138 	return pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va);
139 }
140 
141 static inline pte_t *virt_to_kpte(unsigned long vaddr)
142 {
143 	pmd_t *pmd = pmd_off_k(vaddr);
144 
145 	return pmd_none(*pmd) ? NULL : pte_offset_kernel(pmd, vaddr);
146 }
147 
148 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
149 extern int ptep_set_access_flags(struct vm_area_struct *vma,
150 				 unsigned long address, pte_t *ptep,
151 				 pte_t entry, int dirty);
152 #endif
153 
154 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
155 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
156 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
157 				 unsigned long address, pmd_t *pmdp,
158 				 pmd_t entry, int dirty);
159 extern int pudp_set_access_flags(struct vm_area_struct *vma,
160 				 unsigned long address, pud_t *pudp,
161 				 pud_t entry, int dirty);
162 #else
163 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
164 					unsigned long address, pmd_t *pmdp,
165 					pmd_t entry, int dirty)
166 {
167 	BUILD_BUG();
168 	return 0;
169 }
170 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
171 					unsigned long address, pud_t *pudp,
172 					pud_t entry, int dirty)
173 {
174 	BUILD_BUG();
175 	return 0;
176 }
177 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
178 #endif
179 
180 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
181 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
182 					    unsigned long address,
183 					    pte_t *ptep)
184 {
185 	pte_t pte = *ptep;
186 	int r = 1;
187 	if (!pte_young(pte))
188 		r = 0;
189 	else
190 		set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
191 	return r;
192 }
193 #endif
194 
195 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
196 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
197 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
198 					    unsigned long address,
199 					    pmd_t *pmdp)
200 {
201 	pmd_t pmd = *pmdp;
202 	int r = 1;
203 	if (!pmd_young(pmd))
204 		r = 0;
205 	else
206 		set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
207 	return r;
208 }
209 #else
210 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
211 					    unsigned long address,
212 					    pmd_t *pmdp)
213 {
214 	BUILD_BUG();
215 	return 0;
216 }
217 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
218 #endif
219 
220 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
221 int ptep_clear_flush_young(struct vm_area_struct *vma,
222 			   unsigned long address, pte_t *ptep);
223 #endif
224 
225 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
226 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
227 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
228 				  unsigned long address, pmd_t *pmdp);
229 #else
230 /*
231  * Despite relevant to THP only, this API is called from generic rmap code
232  * under PageTransHuge(), hence needs a dummy implementation for !THP
233  */
234 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
235 					 unsigned long address, pmd_t *pmdp)
236 {
237 	BUILD_BUG();
238 	return 0;
239 }
240 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
241 #endif
242 
243 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
244 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
245 				       unsigned long address,
246 				       pte_t *ptep)
247 {
248 	pte_t pte = *ptep;
249 	pte_clear(mm, address, ptep);
250 	return pte;
251 }
252 #endif
253 
254 #ifndef __HAVE_ARCH_PTEP_GET
255 static inline pte_t ptep_get(pte_t *ptep)
256 {
257 	return READ_ONCE(*ptep);
258 }
259 #endif
260 
261 #ifdef CONFIG_GUP_GET_PTE_LOW_HIGH
262 /*
263  * WARNING: only to be used in the get_user_pages_fast() implementation.
264  *
265  * With get_user_pages_fast(), we walk down the pagetables without taking any
266  * locks.  For this we would like to load the pointers atomically, but sometimes
267  * that is not possible (e.g. without expensive cmpxchg8b on x86_32 PAE).  What
268  * we do have is the guarantee that a PTE will only either go from not present
269  * to present, or present to not present or both -- it will not switch to a
270  * completely different present page without a TLB flush in between; something
271  * that we are blocking by holding interrupts off.
272  *
273  * Setting ptes from not present to present goes:
274  *
275  *   ptep->pte_high = h;
276  *   smp_wmb();
277  *   ptep->pte_low = l;
278  *
279  * And present to not present goes:
280  *
281  *   ptep->pte_low = 0;
282  *   smp_wmb();
283  *   ptep->pte_high = 0;
284  *
285  * We must ensure here that the load of pte_low sees 'l' IFF pte_high sees 'h'.
286  * We load pte_high *after* loading pte_low, which ensures we don't see an older
287  * value of pte_high.  *Then* we recheck pte_low, which ensures that we haven't
288  * picked up a changed pte high. We might have gotten rubbish values from
289  * pte_low and pte_high, but we are guaranteed that pte_low will not have the
290  * present bit set *unless* it is 'l'. Because get_user_pages_fast() only
291  * operates on present ptes we're safe.
292  */
293 static inline pte_t ptep_get_lockless(pte_t *ptep)
294 {
295 	pte_t pte;
296 
297 	do {
298 		pte.pte_low = ptep->pte_low;
299 		smp_rmb();
300 		pte.pte_high = ptep->pte_high;
301 		smp_rmb();
302 	} while (unlikely(pte.pte_low != ptep->pte_low));
303 
304 	return pte;
305 }
306 #else /* CONFIG_GUP_GET_PTE_LOW_HIGH */
307 /*
308  * We require that the PTE can be read atomically.
309  */
310 static inline pte_t ptep_get_lockless(pte_t *ptep)
311 {
312 	return ptep_get(ptep);
313 }
314 #endif /* CONFIG_GUP_GET_PTE_LOW_HIGH */
315 
316 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
317 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
318 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
319 					    unsigned long address,
320 					    pmd_t *pmdp)
321 {
322 	pmd_t pmd = *pmdp;
323 	pmd_clear(pmdp);
324 	return pmd;
325 }
326 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
327 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
328 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
329 					    unsigned long address,
330 					    pud_t *pudp)
331 {
332 	pud_t pud = *pudp;
333 
334 	pud_clear(pudp);
335 	return pud;
336 }
337 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
338 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
339 
340 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
341 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
342 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
343 					    unsigned long address, pmd_t *pmdp,
344 					    int full)
345 {
346 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
347 }
348 #endif
349 
350 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
351 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
352 					    unsigned long address, pud_t *pudp,
353 					    int full)
354 {
355 	return pudp_huge_get_and_clear(mm, address, pudp);
356 }
357 #endif
358 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
359 
360 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
361 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
362 					    unsigned long address, pte_t *ptep,
363 					    int full)
364 {
365 	pte_t pte;
366 	pte = ptep_get_and_clear(mm, address, ptep);
367 	return pte;
368 }
369 #endif
370 
371 
372 /*
373  * If two threads concurrently fault at the same page, the thread that
374  * won the race updates the PTE and its local TLB/Cache. The other thread
375  * gives up, simply does nothing, and continues; on architectures where
376  * software can update TLB,  local TLB can be updated here to avoid next page
377  * fault. This function updates TLB only, do nothing with cache or others.
378  * It is the difference with function update_mmu_cache.
379  */
380 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
381 static inline void update_mmu_tlb(struct vm_area_struct *vma,
382 				unsigned long address, pte_t *ptep)
383 {
384 }
385 #define __HAVE_ARCH_UPDATE_MMU_TLB
386 #endif
387 
388 /*
389  * Some architectures may be able to avoid expensive synchronization
390  * primitives when modifications are made to PTE's which are already
391  * not present, or in the process of an address space destruction.
392  */
393 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
394 static inline void pte_clear_not_present_full(struct mm_struct *mm,
395 					      unsigned long address,
396 					      pte_t *ptep,
397 					      int full)
398 {
399 	pte_clear(mm, address, ptep);
400 }
401 #endif
402 
403 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
404 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
405 			      unsigned long address,
406 			      pte_t *ptep);
407 #endif
408 
409 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
410 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
411 			      unsigned long address,
412 			      pmd_t *pmdp);
413 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
414 			      unsigned long address,
415 			      pud_t *pudp);
416 #endif
417 
418 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
419 struct mm_struct;
420 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
421 {
422 	pte_t old_pte = *ptep;
423 	set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
424 }
425 #endif
426 
427 /*
428  * On some architectures hardware does not set page access bit when accessing
429  * memory page, it is responsibilty of software setting this bit. It brings
430  * out extra page fault penalty to track page access bit. For optimization page
431  * access bit can be set during all page fault flow on these arches.
432  * To be differentiate with macro pte_mkyoung, this macro is used on platforms
433  * where software maintains page access bit.
434  */
435 #ifndef pte_savedwrite
436 #define pte_savedwrite pte_write
437 #endif
438 
439 #ifndef pte_mk_savedwrite
440 #define pte_mk_savedwrite pte_mkwrite
441 #endif
442 
443 #ifndef pte_clear_savedwrite
444 #define pte_clear_savedwrite pte_wrprotect
445 #endif
446 
447 #ifndef pmd_savedwrite
448 #define pmd_savedwrite pmd_write
449 #endif
450 
451 #ifndef pmd_mk_savedwrite
452 #define pmd_mk_savedwrite pmd_mkwrite
453 #endif
454 
455 #ifndef pmd_clear_savedwrite
456 #define pmd_clear_savedwrite pmd_wrprotect
457 #endif
458 
459 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
460 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
461 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
462 				      unsigned long address, pmd_t *pmdp)
463 {
464 	pmd_t old_pmd = *pmdp;
465 	set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
466 }
467 #else
468 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
469 				      unsigned long address, pmd_t *pmdp)
470 {
471 	BUILD_BUG();
472 }
473 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
474 #endif
475 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
476 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
477 static inline void pudp_set_wrprotect(struct mm_struct *mm,
478 				      unsigned long address, pud_t *pudp)
479 {
480 	pud_t old_pud = *pudp;
481 
482 	set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
483 }
484 #else
485 static inline void pudp_set_wrprotect(struct mm_struct *mm,
486 				      unsigned long address, pud_t *pudp)
487 {
488 	BUILD_BUG();
489 }
490 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
491 #endif
492 
493 #ifndef pmdp_collapse_flush
494 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
495 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
496 				 unsigned long address, pmd_t *pmdp);
497 #else
498 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
499 					unsigned long address,
500 					pmd_t *pmdp)
501 {
502 	BUILD_BUG();
503 	return *pmdp;
504 }
505 #define pmdp_collapse_flush pmdp_collapse_flush
506 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
507 #endif
508 
509 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
510 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
511 				       pgtable_t pgtable);
512 #endif
513 
514 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
515 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
516 #endif
517 
518 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
519 /*
520  * This is an implementation of pmdp_establish() that is only suitable for an
521  * architecture that doesn't have hardware dirty/accessed bits. In this case we
522  * can't race with CPU which sets these bits and non-atomic aproach is fine.
523  */
524 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
525 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
526 {
527 	pmd_t old_pmd = *pmdp;
528 	set_pmd_at(vma->vm_mm, address, pmdp, pmd);
529 	return old_pmd;
530 }
531 #endif
532 
533 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
534 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
535 			    pmd_t *pmdp);
536 #endif
537 
538 #ifndef __HAVE_ARCH_PTE_SAME
539 static inline int pte_same(pte_t pte_a, pte_t pte_b)
540 {
541 	return pte_val(pte_a) == pte_val(pte_b);
542 }
543 #endif
544 
545 #ifndef __HAVE_ARCH_PTE_UNUSED
546 /*
547  * Some architectures provide facilities to virtualization guests
548  * so that they can flag allocated pages as unused. This allows the
549  * host to transparently reclaim unused pages. This function returns
550  * whether the pte's page is unused.
551  */
552 static inline int pte_unused(pte_t pte)
553 {
554 	return 0;
555 }
556 #endif
557 
558 #ifndef pte_access_permitted
559 #define pte_access_permitted(pte, write) \
560 	(pte_present(pte) && (!(write) || pte_write(pte)))
561 #endif
562 
563 #ifndef pmd_access_permitted
564 #define pmd_access_permitted(pmd, write) \
565 	(pmd_present(pmd) && (!(write) || pmd_write(pmd)))
566 #endif
567 
568 #ifndef pud_access_permitted
569 #define pud_access_permitted(pud, write) \
570 	(pud_present(pud) && (!(write) || pud_write(pud)))
571 #endif
572 
573 #ifndef p4d_access_permitted
574 #define p4d_access_permitted(p4d, write) \
575 	(p4d_present(p4d) && (!(write) || p4d_write(p4d)))
576 #endif
577 
578 #ifndef pgd_access_permitted
579 #define pgd_access_permitted(pgd, write) \
580 	(pgd_present(pgd) && (!(write) || pgd_write(pgd)))
581 #endif
582 
583 #ifndef __HAVE_ARCH_PMD_SAME
584 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
585 {
586 	return pmd_val(pmd_a) == pmd_val(pmd_b);
587 }
588 
589 static inline int pud_same(pud_t pud_a, pud_t pud_b)
590 {
591 	return pud_val(pud_a) == pud_val(pud_b);
592 }
593 #endif
594 
595 #ifndef __HAVE_ARCH_P4D_SAME
596 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
597 {
598 	return p4d_val(p4d_a) == p4d_val(p4d_b);
599 }
600 #endif
601 
602 #ifndef __HAVE_ARCH_PGD_SAME
603 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
604 {
605 	return pgd_val(pgd_a) == pgd_val(pgd_b);
606 }
607 #endif
608 
609 /*
610  * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
611  * TLB flush will be required as a result of the "set". For example, use
612  * in scenarios where it is known ahead of time that the routine is
613  * setting non-present entries, or re-setting an existing entry to the
614  * same value. Otherwise, use the typical "set" helpers and flush the
615  * TLB.
616  */
617 #define set_pte_safe(ptep, pte) \
618 ({ \
619 	WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
620 	set_pte(ptep, pte); \
621 })
622 
623 #define set_pmd_safe(pmdp, pmd) \
624 ({ \
625 	WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
626 	set_pmd(pmdp, pmd); \
627 })
628 
629 #define set_pud_safe(pudp, pud) \
630 ({ \
631 	WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
632 	set_pud(pudp, pud); \
633 })
634 
635 #define set_p4d_safe(p4dp, p4d) \
636 ({ \
637 	WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
638 	set_p4d(p4dp, p4d); \
639 })
640 
641 #define set_pgd_safe(pgdp, pgd) \
642 ({ \
643 	WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
644 	set_pgd(pgdp, pgd); \
645 })
646 
647 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
648 /*
649  * Some architectures support metadata associated with a page. When a
650  * page is being swapped out, this metadata must be saved so it can be
651  * restored when the page is swapped back in. SPARC M7 and newer
652  * processors support an ADI (Application Data Integrity) tag for the
653  * page as metadata for the page. arch_do_swap_page() can restore this
654  * metadata when a page is swapped back in.
655  */
656 static inline void arch_do_swap_page(struct mm_struct *mm,
657 				     struct vm_area_struct *vma,
658 				     unsigned long addr,
659 				     pte_t pte, pte_t oldpte)
660 {
661 
662 }
663 #endif
664 
665 #ifndef __HAVE_ARCH_UNMAP_ONE
666 /*
667  * Some architectures support metadata associated with a page. When a
668  * page is being swapped out, this metadata must be saved so it can be
669  * restored when the page is swapped back in. SPARC M7 and newer
670  * processors support an ADI (Application Data Integrity) tag for the
671  * page as metadata for the page. arch_unmap_one() can save this
672  * metadata on a swap-out of a page.
673  */
674 static inline int arch_unmap_one(struct mm_struct *mm,
675 				  struct vm_area_struct *vma,
676 				  unsigned long addr,
677 				  pte_t orig_pte)
678 {
679 	return 0;
680 }
681 #endif
682 
683 /*
684  * Allow architectures to preserve additional metadata associated with
685  * swapped-out pages. The corresponding __HAVE_ARCH_SWAP_* macros and function
686  * prototypes must be defined in the arch-specific asm/pgtable.h file.
687  */
688 #ifndef __HAVE_ARCH_PREPARE_TO_SWAP
689 static inline int arch_prepare_to_swap(struct page *page)
690 {
691 	return 0;
692 }
693 #endif
694 
695 #ifndef __HAVE_ARCH_SWAP_INVALIDATE
696 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
697 {
698 }
699 
700 static inline void arch_swap_invalidate_area(int type)
701 {
702 }
703 #endif
704 
705 #ifndef __HAVE_ARCH_SWAP_RESTORE
706 static inline void arch_swap_restore(swp_entry_t entry, struct page *page)
707 {
708 }
709 #endif
710 
711 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
712 #define pgd_offset_gate(mm, addr)	pgd_offset(mm, addr)
713 #endif
714 
715 #ifndef __HAVE_ARCH_MOVE_PTE
716 #define move_pte(pte, prot, old_addr, new_addr)	(pte)
717 #endif
718 
719 #ifndef pte_accessible
720 # define pte_accessible(mm, pte)	((void)(pte), 1)
721 #endif
722 
723 #ifndef flush_tlb_fix_spurious_fault
724 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
725 #endif
726 
727 /*
728  * When walking page tables, get the address of the next boundary,
729  * or the end address of the range if that comes earlier.  Although no
730  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
731  */
732 
733 #define pgd_addr_end(addr, end)						\
734 ({	unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;	\
735 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
736 })
737 
738 #ifndef p4d_addr_end
739 #define p4d_addr_end(addr, end)						\
740 ({	unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;	\
741 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
742 })
743 #endif
744 
745 #ifndef pud_addr_end
746 #define pud_addr_end(addr, end)						\
747 ({	unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;	\
748 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
749 })
750 #endif
751 
752 #ifndef pmd_addr_end
753 #define pmd_addr_end(addr, end)						\
754 ({	unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;	\
755 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
756 })
757 #endif
758 
759 /*
760  * When walking page tables, we usually want to skip any p?d_none entries;
761  * and any p?d_bad entries - reporting the error before resetting to none.
762  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
763  */
764 void pgd_clear_bad(pgd_t *);
765 
766 #ifndef __PAGETABLE_P4D_FOLDED
767 void p4d_clear_bad(p4d_t *);
768 #else
769 #define p4d_clear_bad(p4d)        do { } while (0)
770 #endif
771 
772 #ifndef __PAGETABLE_PUD_FOLDED
773 void pud_clear_bad(pud_t *);
774 #else
775 #define pud_clear_bad(p4d)        do { } while (0)
776 #endif
777 
778 void pmd_clear_bad(pmd_t *);
779 
780 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
781 {
782 	if (pgd_none(*pgd))
783 		return 1;
784 	if (unlikely(pgd_bad(*pgd))) {
785 		pgd_clear_bad(pgd);
786 		return 1;
787 	}
788 	return 0;
789 }
790 
791 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
792 {
793 	if (p4d_none(*p4d))
794 		return 1;
795 	if (unlikely(p4d_bad(*p4d))) {
796 		p4d_clear_bad(p4d);
797 		return 1;
798 	}
799 	return 0;
800 }
801 
802 static inline int pud_none_or_clear_bad(pud_t *pud)
803 {
804 	if (pud_none(*pud))
805 		return 1;
806 	if (unlikely(pud_bad(*pud))) {
807 		pud_clear_bad(pud);
808 		return 1;
809 	}
810 	return 0;
811 }
812 
813 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
814 {
815 	if (pmd_none(*pmd))
816 		return 1;
817 	if (unlikely(pmd_bad(*pmd))) {
818 		pmd_clear_bad(pmd);
819 		return 1;
820 	}
821 	return 0;
822 }
823 
824 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
825 					     unsigned long addr,
826 					     pte_t *ptep)
827 {
828 	/*
829 	 * Get the current pte state, but zero it out to make it
830 	 * non-present, preventing the hardware from asynchronously
831 	 * updating it.
832 	 */
833 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
834 }
835 
836 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
837 					     unsigned long addr,
838 					     pte_t *ptep, pte_t pte)
839 {
840 	/*
841 	 * The pte is non-present, so there's no hardware state to
842 	 * preserve.
843 	 */
844 	set_pte_at(vma->vm_mm, addr, ptep, pte);
845 }
846 
847 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
848 /*
849  * Start a pte protection read-modify-write transaction, which
850  * protects against asynchronous hardware modifications to the pte.
851  * The intention is not to prevent the hardware from making pte
852  * updates, but to prevent any updates it may make from being lost.
853  *
854  * This does not protect against other software modifications of the
855  * pte; the appropriate pte lock must be held over the transation.
856  *
857  * Note that this interface is intended to be batchable, meaning that
858  * ptep_modify_prot_commit may not actually update the pte, but merely
859  * queue the update to be done at some later time.  The update must be
860  * actually committed before the pte lock is released, however.
861  */
862 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
863 					   unsigned long addr,
864 					   pte_t *ptep)
865 {
866 	return __ptep_modify_prot_start(vma, addr, ptep);
867 }
868 
869 /*
870  * Commit an update to a pte, leaving any hardware-controlled bits in
871  * the PTE unmodified.
872  */
873 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
874 					   unsigned long addr,
875 					   pte_t *ptep, pte_t old_pte, pte_t pte)
876 {
877 	__ptep_modify_prot_commit(vma, addr, ptep, pte);
878 }
879 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
880 #endif /* CONFIG_MMU */
881 
882 /*
883  * No-op macros that just return the current protection value. Defined here
884  * because these macros can be used even if CONFIG_MMU is not defined.
885  */
886 
887 #ifndef pgprot_nx
888 #define pgprot_nx(prot)	(prot)
889 #endif
890 
891 #ifndef pgprot_noncached
892 #define pgprot_noncached(prot)	(prot)
893 #endif
894 
895 #ifndef pgprot_writecombine
896 #define pgprot_writecombine pgprot_noncached
897 #endif
898 
899 #ifndef pgprot_writethrough
900 #define pgprot_writethrough pgprot_noncached
901 #endif
902 
903 #ifndef pgprot_device
904 #define pgprot_device pgprot_noncached
905 #endif
906 
907 #ifndef pgprot_mhp
908 #define pgprot_mhp(prot)	(prot)
909 #endif
910 
911 #ifdef CONFIG_MMU
912 #ifndef pgprot_modify
913 #define pgprot_modify pgprot_modify
914 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
915 {
916 	if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
917 		newprot = pgprot_noncached(newprot);
918 	if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
919 		newprot = pgprot_writecombine(newprot);
920 	if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
921 		newprot = pgprot_device(newprot);
922 	return newprot;
923 }
924 #endif
925 #endif /* CONFIG_MMU */
926 
927 #ifndef pgprot_encrypted
928 #define pgprot_encrypted(prot)	(prot)
929 #endif
930 
931 #ifndef pgprot_decrypted
932 #define pgprot_decrypted(prot)	(prot)
933 #endif
934 
935 /*
936  * A facility to provide lazy MMU batching.  This allows PTE updates and
937  * page invalidations to be delayed until a call to leave lazy MMU mode
938  * is issued.  Some architectures may benefit from doing this, and it is
939  * beneficial for both shadow and direct mode hypervisors, which may batch
940  * the PTE updates which happen during this window.  Note that using this
941  * interface requires that read hazards be removed from the code.  A read
942  * hazard could result in the direct mode hypervisor case, since the actual
943  * write to the page tables may not yet have taken place, so reads though
944  * a raw PTE pointer after it has been modified are not guaranteed to be
945  * up to date.  This mode can only be entered and left under the protection of
946  * the page table locks for all page tables which may be modified.  In the UP
947  * case, this is required so that preemption is disabled, and in the SMP case,
948  * it must synchronize the delayed page table writes properly on other CPUs.
949  */
950 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
951 #define arch_enter_lazy_mmu_mode()	do {} while (0)
952 #define arch_leave_lazy_mmu_mode()	do {} while (0)
953 #define arch_flush_lazy_mmu_mode()	do {} while (0)
954 #endif
955 
956 /*
957  * A facility to provide batching of the reload of page tables and
958  * other process state with the actual context switch code for
959  * paravirtualized guests.  By convention, only one of the batched
960  * update (lazy) modes (CPU, MMU) should be active at any given time,
961  * entry should never be nested, and entry and exits should always be
962  * paired.  This is for sanity of maintaining and reasoning about the
963  * kernel code.  In this case, the exit (end of the context switch) is
964  * in architecture-specific code, and so doesn't need a generic
965  * definition.
966  */
967 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
968 #define arch_start_context_switch(prev)	do {} while (0)
969 #endif
970 
971 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
972 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
973 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
974 {
975 	return pmd;
976 }
977 
978 static inline int pmd_swp_soft_dirty(pmd_t pmd)
979 {
980 	return 0;
981 }
982 
983 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
984 {
985 	return pmd;
986 }
987 #endif
988 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
989 static inline int pte_soft_dirty(pte_t pte)
990 {
991 	return 0;
992 }
993 
994 static inline int pmd_soft_dirty(pmd_t pmd)
995 {
996 	return 0;
997 }
998 
999 static inline pte_t pte_mksoft_dirty(pte_t pte)
1000 {
1001 	return pte;
1002 }
1003 
1004 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
1005 {
1006 	return pmd;
1007 }
1008 
1009 static inline pte_t pte_clear_soft_dirty(pte_t pte)
1010 {
1011 	return pte;
1012 }
1013 
1014 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
1015 {
1016 	return pmd;
1017 }
1018 
1019 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1020 {
1021 	return pte;
1022 }
1023 
1024 static inline int pte_swp_soft_dirty(pte_t pte)
1025 {
1026 	return 0;
1027 }
1028 
1029 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1030 {
1031 	return pte;
1032 }
1033 
1034 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1035 {
1036 	return pmd;
1037 }
1038 
1039 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1040 {
1041 	return 0;
1042 }
1043 
1044 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1045 {
1046 	return pmd;
1047 }
1048 #endif
1049 
1050 #ifndef __HAVE_PFNMAP_TRACKING
1051 /*
1052  * Interfaces that can be used by architecture code to keep track of
1053  * memory type of pfn mappings specified by the remap_pfn_range,
1054  * vmf_insert_pfn.
1055  */
1056 
1057 /*
1058  * track_pfn_remap is called when a _new_ pfn mapping is being established
1059  * by remap_pfn_range() for physical range indicated by pfn and size.
1060  */
1061 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1062 				  unsigned long pfn, unsigned long addr,
1063 				  unsigned long size)
1064 {
1065 	return 0;
1066 }
1067 
1068 /*
1069  * track_pfn_insert is called when a _new_ single pfn is established
1070  * by vmf_insert_pfn().
1071  */
1072 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1073 				    pfn_t pfn)
1074 {
1075 }
1076 
1077 /*
1078  * track_pfn_copy is called when vma that is covering the pfnmap gets
1079  * copied through copy_page_range().
1080  */
1081 static inline int track_pfn_copy(struct vm_area_struct *vma)
1082 {
1083 	return 0;
1084 }
1085 
1086 /*
1087  * untrack_pfn is called while unmapping a pfnmap for a region.
1088  * untrack can be called for a specific region indicated by pfn and size or
1089  * can be for the entire vma (in which case pfn, size are zero).
1090  */
1091 static inline void untrack_pfn(struct vm_area_struct *vma,
1092 			       unsigned long pfn, unsigned long size)
1093 {
1094 }
1095 
1096 /*
1097  * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
1098  */
1099 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
1100 {
1101 }
1102 #else
1103 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1104 			   unsigned long pfn, unsigned long addr,
1105 			   unsigned long size);
1106 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1107 			     pfn_t pfn);
1108 extern int track_pfn_copy(struct vm_area_struct *vma);
1109 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1110 			unsigned long size);
1111 extern void untrack_pfn_moved(struct vm_area_struct *vma);
1112 #endif
1113 
1114 #ifdef __HAVE_COLOR_ZERO_PAGE
1115 static inline int is_zero_pfn(unsigned long pfn)
1116 {
1117 	extern unsigned long zero_pfn;
1118 	unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1119 	return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1120 }
1121 
1122 #define my_zero_pfn(addr)	page_to_pfn(ZERO_PAGE(addr))
1123 
1124 #else
1125 static inline int is_zero_pfn(unsigned long pfn)
1126 {
1127 	extern unsigned long zero_pfn;
1128 	return pfn == zero_pfn;
1129 }
1130 
1131 static inline unsigned long my_zero_pfn(unsigned long addr)
1132 {
1133 	extern unsigned long zero_pfn;
1134 	return zero_pfn;
1135 }
1136 #endif
1137 
1138 #ifdef CONFIG_MMU
1139 
1140 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1141 static inline int pmd_trans_huge(pmd_t pmd)
1142 {
1143 	return 0;
1144 }
1145 #ifndef pmd_write
1146 static inline int pmd_write(pmd_t pmd)
1147 {
1148 	BUG();
1149 	return 0;
1150 }
1151 #endif /* pmd_write */
1152 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1153 
1154 #ifndef pud_write
1155 static inline int pud_write(pud_t pud)
1156 {
1157 	BUG();
1158 	return 0;
1159 }
1160 #endif /* pud_write */
1161 
1162 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1163 static inline int pmd_devmap(pmd_t pmd)
1164 {
1165 	return 0;
1166 }
1167 static inline int pud_devmap(pud_t pud)
1168 {
1169 	return 0;
1170 }
1171 static inline int pgd_devmap(pgd_t pgd)
1172 {
1173 	return 0;
1174 }
1175 #endif
1176 
1177 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1178 	(defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1179 	 !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
1180 static inline int pud_trans_huge(pud_t pud)
1181 {
1182 	return 0;
1183 }
1184 #endif
1185 
1186 /* See pmd_none_or_trans_huge_or_clear_bad for discussion. */
1187 static inline int pud_none_or_trans_huge_or_dev_or_clear_bad(pud_t *pud)
1188 {
1189 	pud_t pudval = READ_ONCE(*pud);
1190 
1191 	if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1192 		return 1;
1193 	if (unlikely(pud_bad(pudval))) {
1194 		pud_clear_bad(pud);
1195 		return 1;
1196 	}
1197 	return 0;
1198 }
1199 
1200 /* See pmd_trans_unstable for discussion. */
1201 static inline int pud_trans_unstable(pud_t *pud)
1202 {
1203 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) &&			\
1204 	defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1205 	return pud_none_or_trans_huge_or_dev_or_clear_bad(pud);
1206 #else
1207 	return 0;
1208 #endif
1209 }
1210 
1211 #ifndef pmd_read_atomic
1212 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
1213 {
1214 	/*
1215 	 * Depend on compiler for an atomic pmd read. NOTE: this is
1216 	 * only going to work, if the pmdval_t isn't larger than
1217 	 * an unsigned long.
1218 	 */
1219 	return *pmdp;
1220 }
1221 #endif
1222 
1223 #ifndef arch_needs_pgtable_deposit
1224 #define arch_needs_pgtable_deposit() (false)
1225 #endif
1226 /*
1227  * This function is meant to be used by sites walking pagetables with
1228  * the mmap_lock held in read mode to protect against MADV_DONTNEED and
1229  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
1230  * into a null pmd and the transhuge page fault can convert a null pmd
1231  * into an hugepmd or into a regular pmd (if the hugepage allocation
1232  * fails). While holding the mmap_lock in read mode the pmd becomes
1233  * stable and stops changing under us only if it's not null and not a
1234  * transhuge pmd. When those races occurs and this function makes a
1235  * difference vs the standard pmd_none_or_clear_bad, the result is
1236  * undefined so behaving like if the pmd was none is safe (because it
1237  * can return none anyway). The compiler level barrier() is critically
1238  * important to compute the two checks atomically on the same pmdval.
1239  *
1240  * For 32bit kernels with a 64bit large pmd_t this automatically takes
1241  * care of reading the pmd atomically to avoid SMP race conditions
1242  * against pmd_populate() when the mmap_lock is hold for reading by the
1243  * caller (a special atomic read not done by "gcc" as in the generic
1244  * version above, is also needed when THP is disabled because the page
1245  * fault can populate the pmd from under us).
1246  */
1247 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
1248 {
1249 	pmd_t pmdval = pmd_read_atomic(pmd);
1250 	/*
1251 	 * The barrier will stabilize the pmdval in a register or on
1252 	 * the stack so that it will stop changing under the code.
1253 	 *
1254 	 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
1255 	 * pmd_read_atomic is allowed to return a not atomic pmdval
1256 	 * (for example pointing to an hugepage that has never been
1257 	 * mapped in the pmd). The below checks will only care about
1258 	 * the low part of the pmd with 32bit PAE x86 anyway, with the
1259 	 * exception of pmd_none(). So the important thing is that if
1260 	 * the low part of the pmd is found null, the high part will
1261 	 * be also null or the pmd_none() check below would be
1262 	 * confused.
1263 	 */
1264 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1265 	barrier();
1266 #endif
1267 	/*
1268 	 * !pmd_present() checks for pmd migration entries
1269 	 *
1270 	 * The complete check uses is_pmd_migration_entry() in linux/swapops.h
1271 	 * But using that requires moving current function and pmd_trans_unstable()
1272 	 * to linux/swapops.h to resovle dependency, which is too much code move.
1273 	 *
1274 	 * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
1275 	 * because !pmd_present() pages can only be under migration not swapped
1276 	 * out.
1277 	 *
1278 	 * pmd_none() is preseved for future condition checks on pmd migration
1279 	 * entries and not confusing with this function name, although it is
1280 	 * redundant with !pmd_present().
1281 	 */
1282 	if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
1283 		(IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
1284 		return 1;
1285 	if (unlikely(pmd_bad(pmdval))) {
1286 		pmd_clear_bad(pmd);
1287 		return 1;
1288 	}
1289 	return 0;
1290 }
1291 
1292 /*
1293  * This is a noop if Transparent Hugepage Support is not built into
1294  * the kernel. Otherwise it is equivalent to
1295  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
1296  * places that already verified the pmd is not none and they want to
1297  * walk ptes while holding the mmap sem in read mode (write mode don't
1298  * need this). If THP is not enabled, the pmd can't go away under the
1299  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
1300  * run a pmd_trans_unstable before walking the ptes after
1301  * split_huge_pmd returns (because it may have run when the pmd become
1302  * null, but then a page fault can map in a THP and not a regular page).
1303  */
1304 static inline int pmd_trans_unstable(pmd_t *pmd)
1305 {
1306 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1307 	return pmd_none_or_trans_huge_or_clear_bad(pmd);
1308 #else
1309 	return 0;
1310 #endif
1311 }
1312 
1313 /*
1314  * the ordering of these checks is important for pmds with _page_devmap set.
1315  * if we check pmd_trans_unstable() first we will trip the bad_pmd() check
1316  * inside of pmd_none_or_trans_huge_or_clear_bad(). this will end up correctly
1317  * returning 1 but not before it spams dmesg with the pmd_clear_bad() output.
1318  */
1319 static inline int pmd_devmap_trans_unstable(pmd_t *pmd)
1320 {
1321 	return pmd_devmap(*pmd) || pmd_trans_unstable(pmd);
1322 }
1323 
1324 #ifndef CONFIG_NUMA_BALANCING
1325 /*
1326  * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
1327  * the only case the kernel cares is for NUMA balancing and is only ever set
1328  * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
1329  * _PAGE_PROTNONE so by default, implement the helper as "always no". It
1330  * is the responsibility of the caller to distinguish between PROT_NONE
1331  * protections and NUMA hinting fault protections.
1332  */
1333 static inline int pte_protnone(pte_t pte)
1334 {
1335 	return 0;
1336 }
1337 
1338 static inline int pmd_protnone(pmd_t pmd)
1339 {
1340 	return 0;
1341 }
1342 #endif /* CONFIG_NUMA_BALANCING */
1343 
1344 #endif /* CONFIG_MMU */
1345 
1346 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1347 
1348 #ifndef __PAGETABLE_P4D_FOLDED
1349 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1350 int p4d_clear_huge(p4d_t *p4d);
1351 #else
1352 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1353 {
1354 	return 0;
1355 }
1356 static inline int p4d_clear_huge(p4d_t *p4d)
1357 {
1358 	return 0;
1359 }
1360 #endif /* !__PAGETABLE_P4D_FOLDED */
1361 
1362 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1363 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1364 int pud_clear_huge(pud_t *pud);
1365 int pmd_clear_huge(pmd_t *pmd);
1366 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1367 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1368 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1369 #else	/* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1370 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1371 {
1372 	return 0;
1373 }
1374 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1375 {
1376 	return 0;
1377 }
1378 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1379 {
1380 	return 0;
1381 }
1382 static inline int p4d_clear_huge(p4d_t *p4d)
1383 {
1384 	return 0;
1385 }
1386 static inline int pud_clear_huge(pud_t *pud)
1387 {
1388 	return 0;
1389 }
1390 static inline int pmd_clear_huge(pmd_t *pmd)
1391 {
1392 	return 0;
1393 }
1394 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1395 {
1396 	return 0;
1397 }
1398 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1399 {
1400 	return 0;
1401 }
1402 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1403 {
1404 	return 0;
1405 }
1406 #endif	/* CONFIG_HAVE_ARCH_HUGE_VMAP */
1407 
1408 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1409 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1410 /*
1411  * ARCHes with special requirements for evicting THP backing TLB entries can
1412  * implement this. Otherwise also, it can help optimize normal TLB flush in
1413  * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1414  * entire TLB if flush span is greater than a threshold, which will
1415  * likely be true for a single huge page. Thus a single THP flush will
1416  * invalidate the entire TLB which is not desirable.
1417  * e.g. see arch/arc: flush_pmd_tlb_range
1418  */
1419 #define flush_pmd_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1420 #define flush_pud_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1421 #else
1422 #define flush_pmd_tlb_range(vma, addr, end)	BUILD_BUG()
1423 #define flush_pud_tlb_range(vma, addr, end)	BUILD_BUG()
1424 #endif
1425 #endif
1426 
1427 struct file;
1428 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1429 			unsigned long size, pgprot_t *vma_prot);
1430 
1431 #ifndef CONFIG_X86_ESPFIX64
1432 static inline void init_espfix_bsp(void) { }
1433 #endif
1434 
1435 extern void __init pgtable_cache_init(void);
1436 
1437 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1438 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1439 {
1440 	return true;
1441 }
1442 
1443 static inline bool arch_has_pfn_modify_check(void)
1444 {
1445 	return false;
1446 }
1447 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1448 
1449 /*
1450  * Architecture PAGE_KERNEL_* fallbacks
1451  *
1452  * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1453  * because they really don't support them, or the port needs to be updated to
1454  * reflect the required functionality. Below are a set of relatively safe
1455  * fallbacks, as best effort, which we can count on in lieu of the architectures
1456  * not defining them on their own yet.
1457  */
1458 
1459 #ifndef PAGE_KERNEL_RO
1460 # define PAGE_KERNEL_RO PAGE_KERNEL
1461 #endif
1462 
1463 #ifndef PAGE_KERNEL_EXEC
1464 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1465 #endif
1466 
1467 /*
1468  * Page Table Modification bits for pgtbl_mod_mask.
1469  *
1470  * These are used by the p?d_alloc_track*() set of functions an in the generic
1471  * vmalloc/ioremap code to track at which page-table levels entries have been
1472  * modified. Based on that the code can better decide when vmalloc and ioremap
1473  * mapping changes need to be synchronized to other page-tables in the system.
1474  */
1475 #define		__PGTBL_PGD_MODIFIED	0
1476 #define		__PGTBL_P4D_MODIFIED	1
1477 #define		__PGTBL_PUD_MODIFIED	2
1478 #define		__PGTBL_PMD_MODIFIED	3
1479 #define		__PGTBL_PTE_MODIFIED	4
1480 
1481 #define		PGTBL_PGD_MODIFIED	BIT(__PGTBL_PGD_MODIFIED)
1482 #define		PGTBL_P4D_MODIFIED	BIT(__PGTBL_P4D_MODIFIED)
1483 #define		PGTBL_PUD_MODIFIED	BIT(__PGTBL_PUD_MODIFIED)
1484 #define		PGTBL_PMD_MODIFIED	BIT(__PGTBL_PMD_MODIFIED)
1485 #define		PGTBL_PTE_MODIFIED	BIT(__PGTBL_PTE_MODIFIED)
1486 
1487 /* Page-Table Modification Mask */
1488 typedef unsigned int pgtbl_mod_mask;
1489 
1490 #endif /* !__ASSEMBLY__ */
1491 
1492 #if !defined(MAX_POSSIBLE_PHYSMEM_BITS) && !defined(CONFIG_64BIT)
1493 #ifdef CONFIG_PHYS_ADDR_T_64BIT
1494 /*
1495  * ZSMALLOC needs to know the highest PFN on 32-bit architectures
1496  * with physical address space extension, but falls back to
1497  * BITS_PER_LONG otherwise.
1498  */
1499 #error Missing MAX_POSSIBLE_PHYSMEM_BITS definition
1500 #else
1501 #define MAX_POSSIBLE_PHYSMEM_BITS 32
1502 #endif
1503 #endif
1504 
1505 #ifndef has_transparent_hugepage
1506 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1507 #define has_transparent_hugepage() 1
1508 #else
1509 #define has_transparent_hugepage() 0
1510 #endif
1511 #endif
1512 
1513 /*
1514  * On some architectures it depends on the mm if the p4d/pud or pmd
1515  * layer of the page table hierarchy is folded or not.
1516  */
1517 #ifndef mm_p4d_folded
1518 #define mm_p4d_folded(mm)	__is_defined(__PAGETABLE_P4D_FOLDED)
1519 #endif
1520 
1521 #ifndef mm_pud_folded
1522 #define mm_pud_folded(mm)	__is_defined(__PAGETABLE_PUD_FOLDED)
1523 #endif
1524 
1525 #ifndef mm_pmd_folded
1526 #define mm_pmd_folded(mm)	__is_defined(__PAGETABLE_PMD_FOLDED)
1527 #endif
1528 
1529 #ifndef p4d_offset_lockless
1530 #define p4d_offset_lockless(pgdp, pgd, address) p4d_offset(&(pgd), address)
1531 #endif
1532 #ifndef pud_offset_lockless
1533 #define pud_offset_lockless(p4dp, p4d, address) pud_offset(&(p4d), address)
1534 #endif
1535 #ifndef pmd_offset_lockless
1536 #define pmd_offset_lockless(pudp, pud, address) pmd_offset(&(pud), address)
1537 #endif
1538 
1539 /*
1540  * p?d_leaf() - true if this entry is a final mapping to a physical address.
1541  * This differs from p?d_huge() by the fact that they are always available (if
1542  * the architecture supports large pages at the appropriate level) even
1543  * if CONFIG_HUGETLB_PAGE is not defined.
1544  * Only meaningful when called on a valid entry.
1545  */
1546 #ifndef pgd_leaf
1547 #define pgd_leaf(x)	0
1548 #endif
1549 #ifndef p4d_leaf
1550 #define p4d_leaf(x)	0
1551 #endif
1552 #ifndef pud_leaf
1553 #define pud_leaf(x)	0
1554 #endif
1555 #ifndef pmd_leaf
1556 #define pmd_leaf(x)	0
1557 #endif
1558 
1559 #ifndef pgd_leaf_size
1560 #define pgd_leaf_size(x) (1ULL << PGDIR_SHIFT)
1561 #endif
1562 #ifndef p4d_leaf_size
1563 #define p4d_leaf_size(x) P4D_SIZE
1564 #endif
1565 #ifndef pud_leaf_size
1566 #define pud_leaf_size(x) PUD_SIZE
1567 #endif
1568 #ifndef pmd_leaf_size
1569 #define pmd_leaf_size(x) PMD_SIZE
1570 #endif
1571 #ifndef pte_leaf_size
1572 #define pte_leaf_size(x) PAGE_SIZE
1573 #endif
1574 
1575 #endif /* _LINUX_PGTABLE_H */
1576