xref: /linux-6.15/include/linux/pgtable.h (revision 8ca4fc32)
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 #include <linux/page_table_check.h>
16 
17 #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
18 	defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
19 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
20 #endif
21 
22 /*
23  * On almost all architectures and configurations, 0 can be used as the
24  * upper ceiling to free_pgtables(): on many architectures it has the same
25  * effect as using TASK_SIZE.  However, there is one configuration which
26  * must impose a more careful limit, to avoid freeing kernel pgtables.
27  */
28 #ifndef USER_PGTABLES_CEILING
29 #define USER_PGTABLES_CEILING	0UL
30 #endif
31 
32 /*
33  * This defines the first usable user address. Platforms
34  * can override its value with custom FIRST_USER_ADDRESS
35  * defined in their respective <asm/pgtable.h>.
36  */
37 #ifndef FIRST_USER_ADDRESS
38 #define FIRST_USER_ADDRESS	0UL
39 #endif
40 
41 /*
42  * This defines the generic helper for accessing PMD page
43  * table page. Although platforms can still override this
44  * via their respective <asm/pgtable.h>.
45  */
46 #ifndef pmd_pgtable
47 #define pmd_pgtable(pmd) pmd_page(pmd)
48 #endif
49 
50 /*
51  * A page table page can be thought of an array like this: pXd_t[PTRS_PER_PxD]
52  *
53  * The pXx_index() functions return the index of the entry in the page
54  * table page which would control the given virtual address
55  *
56  * As these functions may be used by the same code for different levels of
57  * the page table folding, they are always available, regardless of
58  * CONFIG_PGTABLE_LEVELS value. For the folded levels they simply return 0
59  * because in such cases PTRS_PER_PxD equals 1.
60  */
61 
62 static inline unsigned long pte_index(unsigned long address)
63 {
64 	return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
65 }
66 #define pte_index pte_index
67 
68 #ifndef pmd_index
69 static inline unsigned long pmd_index(unsigned long address)
70 {
71 	return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
72 }
73 #define pmd_index pmd_index
74 #endif
75 
76 #ifndef pud_index
77 static inline unsigned long pud_index(unsigned long address)
78 {
79 	return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
80 }
81 #define pud_index pud_index
82 #endif
83 
84 #ifndef pgd_index
85 /* Must be a compile-time constant, so implement it as a macro */
86 #define pgd_index(a)  (((a) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
87 #endif
88 
89 #ifndef pte_offset_kernel
90 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
91 {
92 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
93 }
94 #define pte_offset_kernel pte_offset_kernel
95 #endif
96 
97 #if defined(CONFIG_HIGHPTE)
98 #define pte_offset_map(dir, address)				\
99 	((pte_t *)kmap_atomic(pmd_page(*(dir))) +		\
100 	 pte_index((address)))
101 #define pte_unmap(pte) kunmap_atomic((pte))
102 #else
103 #define pte_offset_map(dir, address)	pte_offset_kernel((dir), (address))
104 #define pte_unmap(pte) ((void)(pte))	/* NOP */
105 #endif
106 
107 /* Find an entry in the second-level page table.. */
108 #ifndef pmd_offset
109 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
110 {
111 	return pud_pgtable(*pud) + pmd_index(address);
112 }
113 #define pmd_offset pmd_offset
114 #endif
115 
116 #ifndef pud_offset
117 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
118 {
119 	return p4d_pgtable(*p4d) + pud_index(address);
120 }
121 #define pud_offset pud_offset
122 #endif
123 
124 static inline pgd_t *pgd_offset_pgd(pgd_t *pgd, unsigned long address)
125 {
126 	return (pgd + pgd_index(address));
127 };
128 
129 /*
130  * a shortcut to get a pgd_t in a given mm
131  */
132 #ifndef pgd_offset
133 #define pgd_offset(mm, address)		pgd_offset_pgd((mm)->pgd, (address))
134 #endif
135 
136 /*
137  * a shortcut which implies the use of the kernel's pgd, instead
138  * of a process's
139  */
140 #ifndef pgd_offset_k
141 #define pgd_offset_k(address)		pgd_offset(&init_mm, (address))
142 #endif
143 
144 /*
145  * In many cases it is known that a virtual address is mapped at PMD or PTE
146  * level, so instead of traversing all the page table levels, we can get a
147  * pointer to the PMD entry in user or kernel page table or translate a virtual
148  * address to the pointer in the PTE in the kernel page tables with simple
149  * helpers.
150  */
151 static inline pmd_t *pmd_off(struct mm_struct *mm, unsigned long va)
152 {
153 	return pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, va), va), va), va);
154 }
155 
156 static inline pmd_t *pmd_off_k(unsigned long va)
157 {
158 	return pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va);
159 }
160 
161 static inline pte_t *virt_to_kpte(unsigned long vaddr)
162 {
163 	pmd_t *pmd = pmd_off_k(vaddr);
164 
165 	return pmd_none(*pmd) ? NULL : pte_offset_kernel(pmd, vaddr);
166 }
167 
168 #ifndef pmd_young
169 static inline int pmd_young(pmd_t pmd)
170 {
171 	return 0;
172 }
173 #endif
174 
175 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
176 extern int ptep_set_access_flags(struct vm_area_struct *vma,
177 				 unsigned long address, pte_t *ptep,
178 				 pte_t entry, int dirty);
179 #endif
180 
181 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
182 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
183 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
184 				 unsigned long address, pmd_t *pmdp,
185 				 pmd_t entry, int dirty);
186 extern int pudp_set_access_flags(struct vm_area_struct *vma,
187 				 unsigned long address, pud_t *pudp,
188 				 pud_t entry, int dirty);
189 #else
190 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
191 					unsigned long address, pmd_t *pmdp,
192 					pmd_t entry, int dirty)
193 {
194 	BUILD_BUG();
195 	return 0;
196 }
197 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
198 					unsigned long address, pud_t *pudp,
199 					pud_t entry, int dirty)
200 {
201 	BUILD_BUG();
202 	return 0;
203 }
204 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
205 #endif
206 
207 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
208 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
209 					    unsigned long address,
210 					    pte_t *ptep)
211 {
212 	pte_t pte = *ptep;
213 	int r = 1;
214 	if (!pte_young(pte))
215 		r = 0;
216 	else
217 		set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
218 	return r;
219 }
220 #endif
221 
222 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
223 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG)
224 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
225 					    unsigned long address,
226 					    pmd_t *pmdp)
227 {
228 	pmd_t pmd = *pmdp;
229 	int r = 1;
230 	if (!pmd_young(pmd))
231 		r = 0;
232 	else
233 		set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
234 	return r;
235 }
236 #else
237 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
238 					    unsigned long address,
239 					    pmd_t *pmdp)
240 {
241 	BUILD_BUG();
242 	return 0;
243 }
244 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG */
245 #endif
246 
247 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
248 int ptep_clear_flush_young(struct vm_area_struct *vma,
249 			   unsigned long address, pte_t *ptep);
250 #endif
251 
252 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
253 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
254 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
255 				  unsigned long address, pmd_t *pmdp);
256 #else
257 /*
258  * Despite relevant to THP only, this API is called from generic rmap code
259  * under PageTransHuge(), hence needs a dummy implementation for !THP
260  */
261 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
262 					 unsigned long address, pmd_t *pmdp)
263 {
264 	BUILD_BUG();
265 	return 0;
266 }
267 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
268 #endif
269 
270 #ifndef arch_has_hw_nonleaf_pmd_young
271 /*
272  * Return whether the accessed bit in non-leaf PMD entries is supported on the
273  * local CPU.
274  */
275 static inline bool arch_has_hw_nonleaf_pmd_young(void)
276 {
277 	return IS_ENABLED(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG);
278 }
279 #endif
280 
281 #ifndef arch_has_hw_pte_young
282 /*
283  * Return whether the accessed bit is supported on the local CPU.
284  *
285  * This stub assumes accessing through an old PTE triggers a page fault.
286  * Architectures that automatically set the access bit should overwrite it.
287  */
288 static inline bool arch_has_hw_pte_young(void)
289 {
290 	return false;
291 }
292 #endif
293 
294 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
295 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
296 				       unsigned long address,
297 				       pte_t *ptep)
298 {
299 	pte_t pte = *ptep;
300 	pte_clear(mm, address, ptep);
301 	page_table_check_pte_clear(mm, address, pte);
302 	return pte;
303 }
304 #endif
305 
306 static inline void ptep_clear(struct mm_struct *mm, unsigned long addr,
307 			      pte_t *ptep)
308 {
309 	ptep_get_and_clear(mm, addr, ptep);
310 }
311 
312 #ifndef __HAVE_ARCH_PTEP_GET
313 static inline pte_t ptep_get(pte_t *ptep)
314 {
315 	return READ_ONCE(*ptep);
316 }
317 #endif
318 
319 #ifdef CONFIG_GUP_GET_PTE_LOW_HIGH
320 /*
321  * WARNING: only to be used in the get_user_pages_fast() implementation.
322  *
323  * With get_user_pages_fast(), we walk down the pagetables without taking any
324  * locks.  For this we would like to load the pointers atomically, but sometimes
325  * that is not possible (e.g. without expensive cmpxchg8b on x86_32 PAE).  What
326  * we do have is the guarantee that a PTE will only either go from not present
327  * to present, or present to not present or both -- it will not switch to a
328  * completely different present page without a TLB flush in between; something
329  * that we are blocking by holding interrupts off.
330  *
331  * Setting ptes from not present to present goes:
332  *
333  *   ptep->pte_high = h;
334  *   smp_wmb();
335  *   ptep->pte_low = l;
336  *
337  * And present to not present goes:
338  *
339  *   ptep->pte_low = 0;
340  *   smp_wmb();
341  *   ptep->pte_high = 0;
342  *
343  * We must ensure here that the load of pte_low sees 'l' IFF pte_high sees 'h'.
344  * We load pte_high *after* loading pte_low, which ensures we don't see an older
345  * value of pte_high.  *Then* we recheck pte_low, which ensures that we haven't
346  * picked up a changed pte high. We might have gotten rubbish values from
347  * pte_low and pte_high, but we are guaranteed that pte_low will not have the
348  * present bit set *unless* it is 'l'. Because get_user_pages_fast() only
349  * operates on present ptes we're safe.
350  */
351 static inline pte_t ptep_get_lockless(pte_t *ptep)
352 {
353 	pte_t pte;
354 
355 	do {
356 		pte.pte_low = ptep->pte_low;
357 		smp_rmb();
358 		pte.pte_high = ptep->pte_high;
359 		smp_rmb();
360 	} while (unlikely(pte.pte_low != ptep->pte_low));
361 
362 	return pte;
363 }
364 #else /* CONFIG_GUP_GET_PTE_LOW_HIGH */
365 /*
366  * We require that the PTE can be read atomically.
367  */
368 static inline pte_t ptep_get_lockless(pte_t *ptep)
369 {
370 	return ptep_get(ptep);
371 }
372 #endif /* CONFIG_GUP_GET_PTE_LOW_HIGH */
373 
374 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
375 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
376 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
377 					    unsigned long address,
378 					    pmd_t *pmdp)
379 {
380 	pmd_t pmd = *pmdp;
381 
382 	pmd_clear(pmdp);
383 	page_table_check_pmd_clear(mm, address, pmd);
384 
385 	return pmd;
386 }
387 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
388 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
389 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
390 					    unsigned long address,
391 					    pud_t *pudp)
392 {
393 	pud_t pud = *pudp;
394 
395 	pud_clear(pudp);
396 	page_table_check_pud_clear(mm, address, pud);
397 
398 	return pud;
399 }
400 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
401 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
402 
403 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
404 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
405 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
406 					    unsigned long address, pmd_t *pmdp,
407 					    int full)
408 {
409 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
410 }
411 #endif
412 
413 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
414 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
415 					    unsigned long address, pud_t *pudp,
416 					    int full)
417 {
418 	return pudp_huge_get_and_clear(mm, address, pudp);
419 }
420 #endif
421 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
422 
423 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
424 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
425 					    unsigned long address, pte_t *ptep,
426 					    int full)
427 {
428 	return ptep_get_and_clear(mm, address, ptep);
429 }
430 #endif
431 
432 
433 /*
434  * If two threads concurrently fault at the same page, the thread that
435  * won the race updates the PTE and its local TLB/Cache. The other thread
436  * gives up, simply does nothing, and continues; on architectures where
437  * software can update TLB,  local TLB can be updated here to avoid next page
438  * fault. This function updates TLB only, do nothing with cache or others.
439  * It is the difference with function update_mmu_cache.
440  */
441 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
442 static inline void update_mmu_tlb(struct vm_area_struct *vma,
443 				unsigned long address, pte_t *ptep)
444 {
445 }
446 #define __HAVE_ARCH_UPDATE_MMU_TLB
447 #endif
448 
449 /*
450  * Some architectures may be able to avoid expensive synchronization
451  * primitives when modifications are made to PTE's which are already
452  * not present, or in the process of an address space destruction.
453  */
454 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
455 static inline void pte_clear_not_present_full(struct mm_struct *mm,
456 					      unsigned long address,
457 					      pte_t *ptep,
458 					      int full)
459 {
460 	pte_clear(mm, address, ptep);
461 }
462 #endif
463 
464 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
465 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
466 			      unsigned long address,
467 			      pte_t *ptep);
468 #endif
469 
470 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
471 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
472 			      unsigned long address,
473 			      pmd_t *pmdp);
474 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
475 			      unsigned long address,
476 			      pud_t *pudp);
477 #endif
478 
479 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
480 struct mm_struct;
481 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
482 {
483 	pte_t old_pte = *ptep;
484 	set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
485 }
486 #endif
487 
488 /*
489  * On some architectures hardware does not set page access bit when accessing
490  * memory page, it is responsibility of software setting this bit. It brings
491  * out extra page fault penalty to track page access bit. For optimization page
492  * access bit can be set during all page fault flow on these arches.
493  * To be differentiate with macro pte_mkyoung, this macro is used on platforms
494  * where software maintains page access bit.
495  */
496 #ifndef pte_sw_mkyoung
497 static inline pte_t pte_sw_mkyoung(pte_t pte)
498 {
499 	return pte;
500 }
501 #define pte_sw_mkyoung	pte_sw_mkyoung
502 #endif
503 
504 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
505 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
506 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
507 				      unsigned long address, pmd_t *pmdp)
508 {
509 	pmd_t old_pmd = *pmdp;
510 	set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
511 }
512 #else
513 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
514 				      unsigned long address, pmd_t *pmdp)
515 {
516 	BUILD_BUG();
517 }
518 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
519 #endif
520 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
521 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
522 static inline void pudp_set_wrprotect(struct mm_struct *mm,
523 				      unsigned long address, pud_t *pudp)
524 {
525 	pud_t old_pud = *pudp;
526 
527 	set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
528 }
529 #else
530 static inline void pudp_set_wrprotect(struct mm_struct *mm,
531 				      unsigned long address, pud_t *pudp)
532 {
533 	BUILD_BUG();
534 }
535 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
536 #endif
537 
538 #ifndef pmdp_collapse_flush
539 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
540 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
541 				 unsigned long address, pmd_t *pmdp);
542 #else
543 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
544 					unsigned long address,
545 					pmd_t *pmdp)
546 {
547 	BUILD_BUG();
548 	return *pmdp;
549 }
550 #define pmdp_collapse_flush pmdp_collapse_flush
551 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
552 #endif
553 
554 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
555 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
556 				       pgtable_t pgtable);
557 #endif
558 
559 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
560 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
561 #endif
562 
563 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
564 /*
565  * This is an implementation of pmdp_establish() that is only suitable for an
566  * architecture that doesn't have hardware dirty/accessed bits. In this case we
567  * can't race with CPU which sets these bits and non-atomic approach is fine.
568  */
569 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
570 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
571 {
572 	pmd_t old_pmd = *pmdp;
573 	set_pmd_at(vma->vm_mm, address, pmdp, pmd);
574 	return old_pmd;
575 }
576 #endif
577 
578 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
579 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
580 			    pmd_t *pmdp);
581 #endif
582 
583 #ifndef __HAVE_ARCH_PMDP_INVALIDATE_AD
584 
585 /*
586  * pmdp_invalidate_ad() invalidates the PMD while changing a transparent
587  * hugepage mapping in the page tables. This function is similar to
588  * pmdp_invalidate(), but should only be used if the access and dirty bits would
589  * not be cleared by the software in the new PMD value. The function ensures
590  * that hardware changes of the access and dirty bits updates would not be lost.
591  *
592  * Doing so can allow in certain architectures to avoid a TLB flush in most
593  * cases. Yet, another TLB flush might be necessary later if the PMD update
594  * itself requires such flush (e.g., if protection was set to be stricter). Yet,
595  * even when a TLB flush is needed because of the update, the caller may be able
596  * to batch these TLB flushing operations, so fewer TLB flush operations are
597  * needed.
598  */
599 extern pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma,
600 				unsigned long address, pmd_t *pmdp);
601 #endif
602 
603 #ifndef __HAVE_ARCH_PTE_SAME
604 static inline int pte_same(pte_t pte_a, pte_t pte_b)
605 {
606 	return pte_val(pte_a) == pte_val(pte_b);
607 }
608 #endif
609 
610 #ifndef __HAVE_ARCH_PTE_UNUSED
611 /*
612  * Some architectures provide facilities to virtualization guests
613  * so that they can flag allocated pages as unused. This allows the
614  * host to transparently reclaim unused pages. This function returns
615  * whether the pte's page is unused.
616  */
617 static inline int pte_unused(pte_t pte)
618 {
619 	return 0;
620 }
621 #endif
622 
623 #ifndef pte_access_permitted
624 #define pte_access_permitted(pte, write) \
625 	(pte_present(pte) && (!(write) || pte_write(pte)))
626 #endif
627 
628 #ifndef pmd_access_permitted
629 #define pmd_access_permitted(pmd, write) \
630 	(pmd_present(pmd) && (!(write) || pmd_write(pmd)))
631 #endif
632 
633 #ifndef pud_access_permitted
634 #define pud_access_permitted(pud, write) \
635 	(pud_present(pud) && (!(write) || pud_write(pud)))
636 #endif
637 
638 #ifndef p4d_access_permitted
639 #define p4d_access_permitted(p4d, write) \
640 	(p4d_present(p4d) && (!(write) || p4d_write(p4d)))
641 #endif
642 
643 #ifndef pgd_access_permitted
644 #define pgd_access_permitted(pgd, write) \
645 	(pgd_present(pgd) && (!(write) || pgd_write(pgd)))
646 #endif
647 
648 #ifndef __HAVE_ARCH_PMD_SAME
649 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
650 {
651 	return pmd_val(pmd_a) == pmd_val(pmd_b);
652 }
653 
654 static inline int pud_same(pud_t pud_a, pud_t pud_b)
655 {
656 	return pud_val(pud_a) == pud_val(pud_b);
657 }
658 #endif
659 
660 #ifndef __HAVE_ARCH_P4D_SAME
661 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
662 {
663 	return p4d_val(p4d_a) == p4d_val(p4d_b);
664 }
665 #endif
666 
667 #ifndef __HAVE_ARCH_PGD_SAME
668 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
669 {
670 	return pgd_val(pgd_a) == pgd_val(pgd_b);
671 }
672 #endif
673 
674 /*
675  * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
676  * TLB flush will be required as a result of the "set". For example, use
677  * in scenarios where it is known ahead of time that the routine is
678  * setting non-present entries, or re-setting an existing entry to the
679  * same value. Otherwise, use the typical "set" helpers and flush the
680  * TLB.
681  */
682 #define set_pte_safe(ptep, pte) \
683 ({ \
684 	WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
685 	set_pte(ptep, pte); \
686 })
687 
688 #define set_pmd_safe(pmdp, pmd) \
689 ({ \
690 	WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
691 	set_pmd(pmdp, pmd); \
692 })
693 
694 #define set_pud_safe(pudp, pud) \
695 ({ \
696 	WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
697 	set_pud(pudp, pud); \
698 })
699 
700 #define set_p4d_safe(p4dp, p4d) \
701 ({ \
702 	WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
703 	set_p4d(p4dp, p4d); \
704 })
705 
706 #define set_pgd_safe(pgdp, pgd) \
707 ({ \
708 	WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
709 	set_pgd(pgdp, pgd); \
710 })
711 
712 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
713 /*
714  * Some architectures support metadata associated with a page. When a
715  * page is being swapped out, this metadata must be saved so it can be
716  * restored when the page is swapped back in. SPARC M7 and newer
717  * processors support an ADI (Application Data Integrity) tag for the
718  * page as metadata for the page. arch_do_swap_page() can restore this
719  * metadata when a page is swapped back in.
720  */
721 static inline void arch_do_swap_page(struct mm_struct *mm,
722 				     struct vm_area_struct *vma,
723 				     unsigned long addr,
724 				     pte_t pte, pte_t oldpte)
725 {
726 
727 }
728 #endif
729 
730 #ifndef __HAVE_ARCH_UNMAP_ONE
731 /*
732  * Some architectures support metadata associated with a page. When a
733  * page is being swapped out, this metadata must be saved so it can be
734  * restored when the page is swapped back in. SPARC M7 and newer
735  * processors support an ADI (Application Data Integrity) tag for the
736  * page as metadata for the page. arch_unmap_one() can save this
737  * metadata on a swap-out of a page.
738  */
739 static inline int arch_unmap_one(struct mm_struct *mm,
740 				  struct vm_area_struct *vma,
741 				  unsigned long addr,
742 				  pte_t orig_pte)
743 {
744 	return 0;
745 }
746 #endif
747 
748 /*
749  * Allow architectures to preserve additional metadata associated with
750  * swapped-out pages. The corresponding __HAVE_ARCH_SWAP_* macros and function
751  * prototypes must be defined in the arch-specific asm/pgtable.h file.
752  */
753 #ifndef __HAVE_ARCH_PREPARE_TO_SWAP
754 static inline int arch_prepare_to_swap(struct page *page)
755 {
756 	return 0;
757 }
758 #endif
759 
760 #ifndef __HAVE_ARCH_SWAP_INVALIDATE
761 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
762 {
763 }
764 
765 static inline void arch_swap_invalidate_area(int type)
766 {
767 }
768 #endif
769 
770 #ifndef __HAVE_ARCH_SWAP_RESTORE
771 static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
772 {
773 }
774 #endif
775 
776 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
777 #define pgd_offset_gate(mm, addr)	pgd_offset(mm, addr)
778 #endif
779 
780 #ifndef __HAVE_ARCH_MOVE_PTE
781 #define move_pte(pte, prot, old_addr, new_addr)	(pte)
782 #endif
783 
784 #ifndef pte_accessible
785 # define pte_accessible(mm, pte)	((void)(pte), 1)
786 #endif
787 
788 #ifndef flush_tlb_fix_spurious_fault
789 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
790 #endif
791 
792 /*
793  * When walking page tables, get the address of the next boundary,
794  * or the end address of the range if that comes earlier.  Although no
795  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
796  */
797 
798 #define pgd_addr_end(addr, end)						\
799 ({	unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;	\
800 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
801 })
802 
803 #ifndef p4d_addr_end
804 #define p4d_addr_end(addr, end)						\
805 ({	unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;	\
806 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
807 })
808 #endif
809 
810 #ifndef pud_addr_end
811 #define pud_addr_end(addr, end)						\
812 ({	unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;	\
813 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
814 })
815 #endif
816 
817 #ifndef pmd_addr_end
818 #define pmd_addr_end(addr, end)						\
819 ({	unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;	\
820 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
821 })
822 #endif
823 
824 /*
825  * When walking page tables, we usually want to skip any p?d_none entries;
826  * and any p?d_bad entries - reporting the error before resetting to none.
827  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
828  */
829 void pgd_clear_bad(pgd_t *);
830 
831 #ifndef __PAGETABLE_P4D_FOLDED
832 void p4d_clear_bad(p4d_t *);
833 #else
834 #define p4d_clear_bad(p4d)        do { } while (0)
835 #endif
836 
837 #ifndef __PAGETABLE_PUD_FOLDED
838 void pud_clear_bad(pud_t *);
839 #else
840 #define pud_clear_bad(p4d)        do { } while (0)
841 #endif
842 
843 void pmd_clear_bad(pmd_t *);
844 
845 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
846 {
847 	if (pgd_none(*pgd))
848 		return 1;
849 	if (unlikely(pgd_bad(*pgd))) {
850 		pgd_clear_bad(pgd);
851 		return 1;
852 	}
853 	return 0;
854 }
855 
856 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
857 {
858 	if (p4d_none(*p4d))
859 		return 1;
860 	if (unlikely(p4d_bad(*p4d))) {
861 		p4d_clear_bad(p4d);
862 		return 1;
863 	}
864 	return 0;
865 }
866 
867 static inline int pud_none_or_clear_bad(pud_t *pud)
868 {
869 	if (pud_none(*pud))
870 		return 1;
871 	if (unlikely(pud_bad(*pud))) {
872 		pud_clear_bad(pud);
873 		return 1;
874 	}
875 	return 0;
876 }
877 
878 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
879 {
880 	if (pmd_none(*pmd))
881 		return 1;
882 	if (unlikely(pmd_bad(*pmd))) {
883 		pmd_clear_bad(pmd);
884 		return 1;
885 	}
886 	return 0;
887 }
888 
889 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
890 					     unsigned long addr,
891 					     pte_t *ptep)
892 {
893 	/*
894 	 * Get the current pte state, but zero it out to make it
895 	 * non-present, preventing the hardware from asynchronously
896 	 * updating it.
897 	 */
898 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
899 }
900 
901 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
902 					     unsigned long addr,
903 					     pte_t *ptep, pte_t pte)
904 {
905 	/*
906 	 * The pte is non-present, so there's no hardware state to
907 	 * preserve.
908 	 */
909 	set_pte_at(vma->vm_mm, addr, ptep, pte);
910 }
911 
912 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
913 /*
914  * Start a pte protection read-modify-write transaction, which
915  * protects against asynchronous hardware modifications to the pte.
916  * The intention is not to prevent the hardware from making pte
917  * updates, but to prevent any updates it may make from being lost.
918  *
919  * This does not protect against other software modifications of the
920  * pte; the appropriate pte lock must be held over the transaction.
921  *
922  * Note that this interface is intended to be batchable, meaning that
923  * ptep_modify_prot_commit may not actually update the pte, but merely
924  * queue the update to be done at some later time.  The update must be
925  * actually committed before the pte lock is released, however.
926  */
927 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
928 					   unsigned long addr,
929 					   pte_t *ptep)
930 {
931 	return __ptep_modify_prot_start(vma, addr, ptep);
932 }
933 
934 /*
935  * Commit an update to a pte, leaving any hardware-controlled bits in
936  * the PTE unmodified.
937  */
938 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
939 					   unsigned long addr,
940 					   pte_t *ptep, pte_t old_pte, pte_t pte)
941 {
942 	__ptep_modify_prot_commit(vma, addr, ptep, pte);
943 }
944 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
945 #endif /* CONFIG_MMU */
946 
947 /*
948  * No-op macros that just return the current protection value. Defined here
949  * because these macros can be used even if CONFIG_MMU is not defined.
950  */
951 
952 #ifndef pgprot_nx
953 #define pgprot_nx(prot)	(prot)
954 #endif
955 
956 #ifndef pgprot_noncached
957 #define pgprot_noncached(prot)	(prot)
958 #endif
959 
960 #ifndef pgprot_writecombine
961 #define pgprot_writecombine pgprot_noncached
962 #endif
963 
964 #ifndef pgprot_writethrough
965 #define pgprot_writethrough pgprot_noncached
966 #endif
967 
968 #ifndef pgprot_device
969 #define pgprot_device pgprot_noncached
970 #endif
971 
972 #ifndef pgprot_mhp
973 #define pgprot_mhp(prot)	(prot)
974 #endif
975 
976 #ifdef CONFIG_MMU
977 #ifndef pgprot_modify
978 #define pgprot_modify pgprot_modify
979 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
980 {
981 	if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
982 		newprot = pgprot_noncached(newprot);
983 	if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
984 		newprot = pgprot_writecombine(newprot);
985 	if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
986 		newprot = pgprot_device(newprot);
987 	return newprot;
988 }
989 #endif
990 #endif /* CONFIG_MMU */
991 
992 #ifndef pgprot_encrypted
993 #define pgprot_encrypted(prot)	(prot)
994 #endif
995 
996 #ifndef pgprot_decrypted
997 #define pgprot_decrypted(prot)	(prot)
998 #endif
999 
1000 /*
1001  * A facility to provide lazy MMU batching.  This allows PTE updates and
1002  * page invalidations to be delayed until a call to leave lazy MMU mode
1003  * is issued.  Some architectures may benefit from doing this, and it is
1004  * beneficial for both shadow and direct mode hypervisors, which may batch
1005  * the PTE updates which happen during this window.  Note that using this
1006  * interface requires that read hazards be removed from the code.  A read
1007  * hazard could result in the direct mode hypervisor case, since the actual
1008  * write to the page tables may not yet have taken place, so reads though
1009  * a raw PTE pointer after it has been modified are not guaranteed to be
1010  * up to date.  This mode can only be entered and left under the protection of
1011  * the page table locks for all page tables which may be modified.  In the UP
1012  * case, this is required so that preemption is disabled, and in the SMP case,
1013  * it must synchronize the delayed page table writes properly on other CPUs.
1014  */
1015 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
1016 #define arch_enter_lazy_mmu_mode()	do {} while (0)
1017 #define arch_leave_lazy_mmu_mode()	do {} while (0)
1018 #define arch_flush_lazy_mmu_mode()	do {} while (0)
1019 #endif
1020 
1021 /*
1022  * A facility to provide batching of the reload of page tables and
1023  * other process state with the actual context switch code for
1024  * paravirtualized guests.  By convention, only one of the batched
1025  * update (lazy) modes (CPU, MMU) should be active at any given time,
1026  * entry should never be nested, and entry and exits should always be
1027  * paired.  This is for sanity of maintaining and reasoning about the
1028  * kernel code.  In this case, the exit (end of the context switch) is
1029  * in architecture-specific code, and so doesn't need a generic
1030  * definition.
1031  */
1032 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
1033 #define arch_start_context_switch(prev)	do {} while (0)
1034 #endif
1035 
1036 /*
1037  * When replacing an anonymous page by a real (!non) swap entry, we clear
1038  * PG_anon_exclusive from the page and instead remember whether the flag was
1039  * set in the swp pte. During fork(), we have to mark the entry as !exclusive
1040  * (possibly shared). On swapin, we use that information to restore
1041  * PG_anon_exclusive, which is very helpful in cases where we might have
1042  * additional (e.g., FOLL_GET) references on a page and wouldn't be able to
1043  * detect exclusivity.
1044  *
1045  * These functions don't apply to non-swap entries (e.g., migration, hwpoison,
1046  * ...).
1047  */
1048 #ifndef __HAVE_ARCH_PTE_SWP_EXCLUSIVE
1049 static inline pte_t pte_swp_mkexclusive(pte_t pte)
1050 {
1051 	return pte;
1052 }
1053 
1054 static inline int pte_swp_exclusive(pte_t pte)
1055 {
1056 	return false;
1057 }
1058 
1059 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
1060 {
1061 	return pte;
1062 }
1063 #endif
1064 
1065 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1066 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
1067 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1068 {
1069 	return pmd;
1070 }
1071 
1072 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1073 {
1074 	return 0;
1075 }
1076 
1077 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1078 {
1079 	return pmd;
1080 }
1081 #endif
1082 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
1083 static inline int pte_soft_dirty(pte_t pte)
1084 {
1085 	return 0;
1086 }
1087 
1088 static inline int pmd_soft_dirty(pmd_t pmd)
1089 {
1090 	return 0;
1091 }
1092 
1093 static inline pte_t pte_mksoft_dirty(pte_t pte)
1094 {
1095 	return pte;
1096 }
1097 
1098 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
1099 {
1100 	return pmd;
1101 }
1102 
1103 static inline pte_t pte_clear_soft_dirty(pte_t pte)
1104 {
1105 	return pte;
1106 }
1107 
1108 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
1109 {
1110 	return pmd;
1111 }
1112 
1113 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1114 {
1115 	return pte;
1116 }
1117 
1118 static inline int pte_swp_soft_dirty(pte_t pte)
1119 {
1120 	return 0;
1121 }
1122 
1123 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1124 {
1125 	return pte;
1126 }
1127 
1128 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1129 {
1130 	return pmd;
1131 }
1132 
1133 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1134 {
1135 	return 0;
1136 }
1137 
1138 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1139 {
1140 	return pmd;
1141 }
1142 #endif
1143 
1144 #ifndef __HAVE_PFNMAP_TRACKING
1145 /*
1146  * Interfaces that can be used by architecture code to keep track of
1147  * memory type of pfn mappings specified by the remap_pfn_range,
1148  * vmf_insert_pfn.
1149  */
1150 
1151 /*
1152  * track_pfn_remap is called when a _new_ pfn mapping is being established
1153  * by remap_pfn_range() for physical range indicated by pfn and size.
1154  */
1155 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1156 				  unsigned long pfn, unsigned long addr,
1157 				  unsigned long size)
1158 {
1159 	return 0;
1160 }
1161 
1162 /*
1163  * track_pfn_insert is called when a _new_ single pfn is established
1164  * by vmf_insert_pfn().
1165  */
1166 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1167 				    pfn_t pfn)
1168 {
1169 }
1170 
1171 /*
1172  * track_pfn_copy is called when vma that is covering the pfnmap gets
1173  * copied through copy_page_range().
1174  */
1175 static inline int track_pfn_copy(struct vm_area_struct *vma)
1176 {
1177 	return 0;
1178 }
1179 
1180 /*
1181  * untrack_pfn is called while unmapping a pfnmap for a region.
1182  * untrack can be called for a specific region indicated by pfn and size or
1183  * can be for the entire vma (in which case pfn, size are zero).
1184  */
1185 static inline void untrack_pfn(struct vm_area_struct *vma,
1186 			       unsigned long pfn, unsigned long size)
1187 {
1188 }
1189 
1190 /*
1191  * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
1192  */
1193 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
1194 {
1195 }
1196 #else
1197 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1198 			   unsigned long pfn, unsigned long addr,
1199 			   unsigned long size);
1200 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1201 			     pfn_t pfn);
1202 extern int track_pfn_copy(struct vm_area_struct *vma);
1203 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1204 			unsigned long size);
1205 extern void untrack_pfn_moved(struct vm_area_struct *vma);
1206 #endif
1207 
1208 #ifdef CONFIG_MMU
1209 #ifdef __HAVE_COLOR_ZERO_PAGE
1210 static inline int is_zero_pfn(unsigned long pfn)
1211 {
1212 	extern unsigned long zero_pfn;
1213 	unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1214 	return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1215 }
1216 
1217 #define my_zero_pfn(addr)	page_to_pfn(ZERO_PAGE(addr))
1218 
1219 #else
1220 static inline int is_zero_pfn(unsigned long pfn)
1221 {
1222 	extern unsigned long zero_pfn;
1223 	return pfn == zero_pfn;
1224 }
1225 
1226 static inline unsigned long my_zero_pfn(unsigned long addr)
1227 {
1228 	extern unsigned long zero_pfn;
1229 	return zero_pfn;
1230 }
1231 #endif
1232 #else
1233 static inline int is_zero_pfn(unsigned long pfn)
1234 {
1235 	return 0;
1236 }
1237 
1238 static inline unsigned long my_zero_pfn(unsigned long addr)
1239 {
1240 	return 0;
1241 }
1242 #endif /* CONFIG_MMU */
1243 
1244 #ifdef CONFIG_MMU
1245 
1246 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1247 static inline int pmd_trans_huge(pmd_t pmd)
1248 {
1249 	return 0;
1250 }
1251 #ifndef pmd_write
1252 static inline int pmd_write(pmd_t pmd)
1253 {
1254 	BUG();
1255 	return 0;
1256 }
1257 #endif /* pmd_write */
1258 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1259 
1260 #ifndef pud_write
1261 static inline int pud_write(pud_t pud)
1262 {
1263 	BUG();
1264 	return 0;
1265 }
1266 #endif /* pud_write */
1267 
1268 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1269 static inline int pmd_devmap(pmd_t pmd)
1270 {
1271 	return 0;
1272 }
1273 static inline int pud_devmap(pud_t pud)
1274 {
1275 	return 0;
1276 }
1277 static inline int pgd_devmap(pgd_t pgd)
1278 {
1279 	return 0;
1280 }
1281 #endif
1282 
1283 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1284 	!defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1285 static inline int pud_trans_huge(pud_t pud)
1286 {
1287 	return 0;
1288 }
1289 #endif
1290 
1291 /* See pmd_none_or_trans_huge_or_clear_bad for discussion. */
1292 static inline int pud_none_or_trans_huge_or_dev_or_clear_bad(pud_t *pud)
1293 {
1294 	pud_t pudval = READ_ONCE(*pud);
1295 
1296 	if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1297 		return 1;
1298 	if (unlikely(pud_bad(pudval))) {
1299 		pud_clear_bad(pud);
1300 		return 1;
1301 	}
1302 	return 0;
1303 }
1304 
1305 /* See pmd_trans_unstable for discussion. */
1306 static inline int pud_trans_unstable(pud_t *pud)
1307 {
1308 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) &&			\
1309 	defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1310 	return pud_none_or_trans_huge_or_dev_or_clear_bad(pud);
1311 #else
1312 	return 0;
1313 #endif
1314 }
1315 
1316 #ifndef pmd_read_atomic
1317 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
1318 {
1319 	/*
1320 	 * Depend on compiler for an atomic pmd read. NOTE: this is
1321 	 * only going to work, if the pmdval_t isn't larger than
1322 	 * an unsigned long.
1323 	 */
1324 	return *pmdp;
1325 }
1326 #endif
1327 
1328 #ifndef arch_needs_pgtable_deposit
1329 #define arch_needs_pgtable_deposit() (false)
1330 #endif
1331 /*
1332  * This function is meant to be used by sites walking pagetables with
1333  * the mmap_lock held in read mode to protect against MADV_DONTNEED and
1334  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
1335  * into a null pmd and the transhuge page fault can convert a null pmd
1336  * into an hugepmd or into a regular pmd (if the hugepage allocation
1337  * fails). While holding the mmap_lock in read mode the pmd becomes
1338  * stable and stops changing under us only if it's not null and not a
1339  * transhuge pmd. When those races occurs and this function makes a
1340  * difference vs the standard pmd_none_or_clear_bad, the result is
1341  * undefined so behaving like if the pmd was none is safe (because it
1342  * can return none anyway). The compiler level barrier() is critically
1343  * important to compute the two checks atomically on the same pmdval.
1344  *
1345  * For 32bit kernels with a 64bit large pmd_t this automatically takes
1346  * care of reading the pmd atomically to avoid SMP race conditions
1347  * against pmd_populate() when the mmap_lock is hold for reading by the
1348  * caller (a special atomic read not done by "gcc" as in the generic
1349  * version above, is also needed when THP is disabled because the page
1350  * fault can populate the pmd from under us).
1351  */
1352 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
1353 {
1354 	pmd_t pmdval = pmd_read_atomic(pmd);
1355 	/*
1356 	 * The barrier will stabilize the pmdval in a register or on
1357 	 * the stack so that it will stop changing under the code.
1358 	 *
1359 	 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
1360 	 * pmd_read_atomic is allowed to return a not atomic pmdval
1361 	 * (for example pointing to an hugepage that has never been
1362 	 * mapped in the pmd). The below checks will only care about
1363 	 * the low part of the pmd with 32bit PAE x86 anyway, with the
1364 	 * exception of pmd_none(). So the important thing is that if
1365 	 * the low part of the pmd is found null, the high part will
1366 	 * be also null or the pmd_none() check below would be
1367 	 * confused.
1368 	 */
1369 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1370 	barrier();
1371 #endif
1372 	/*
1373 	 * !pmd_present() checks for pmd migration entries
1374 	 *
1375 	 * The complete check uses is_pmd_migration_entry() in linux/swapops.h
1376 	 * But using that requires moving current function and pmd_trans_unstable()
1377 	 * to linux/swapops.h to resolve dependency, which is too much code move.
1378 	 *
1379 	 * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
1380 	 * because !pmd_present() pages can only be under migration not swapped
1381 	 * out.
1382 	 *
1383 	 * pmd_none() is preserved for future condition checks on pmd migration
1384 	 * entries and not confusing with this function name, although it is
1385 	 * redundant with !pmd_present().
1386 	 */
1387 	if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
1388 		(IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
1389 		return 1;
1390 	if (unlikely(pmd_bad(pmdval))) {
1391 		pmd_clear_bad(pmd);
1392 		return 1;
1393 	}
1394 	return 0;
1395 }
1396 
1397 /*
1398  * This is a noop if Transparent Hugepage Support is not built into
1399  * the kernel. Otherwise it is equivalent to
1400  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
1401  * places that already verified the pmd is not none and they want to
1402  * walk ptes while holding the mmap sem in read mode (write mode don't
1403  * need this). If THP is not enabled, the pmd can't go away under the
1404  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
1405  * run a pmd_trans_unstable before walking the ptes after
1406  * split_huge_pmd returns (because it may have run when the pmd become
1407  * null, but then a page fault can map in a THP and not a regular page).
1408  */
1409 static inline int pmd_trans_unstable(pmd_t *pmd)
1410 {
1411 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1412 	return pmd_none_or_trans_huge_or_clear_bad(pmd);
1413 #else
1414 	return 0;
1415 #endif
1416 }
1417 
1418 /*
1419  * the ordering of these checks is important for pmds with _page_devmap set.
1420  * if we check pmd_trans_unstable() first we will trip the bad_pmd() check
1421  * inside of pmd_none_or_trans_huge_or_clear_bad(). this will end up correctly
1422  * returning 1 but not before it spams dmesg with the pmd_clear_bad() output.
1423  */
1424 static inline int pmd_devmap_trans_unstable(pmd_t *pmd)
1425 {
1426 	return pmd_devmap(*pmd) || pmd_trans_unstable(pmd);
1427 }
1428 
1429 #ifndef CONFIG_NUMA_BALANCING
1430 /*
1431  * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
1432  * the only case the kernel cares is for NUMA balancing and is only ever set
1433  * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
1434  * _PAGE_PROTNONE so by default, implement the helper as "always no". It
1435  * is the responsibility of the caller to distinguish between PROT_NONE
1436  * protections and NUMA hinting fault protections.
1437  */
1438 static inline int pte_protnone(pte_t pte)
1439 {
1440 	return 0;
1441 }
1442 
1443 static inline int pmd_protnone(pmd_t pmd)
1444 {
1445 	return 0;
1446 }
1447 #endif /* CONFIG_NUMA_BALANCING */
1448 
1449 #endif /* CONFIG_MMU */
1450 
1451 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1452 
1453 #ifndef __PAGETABLE_P4D_FOLDED
1454 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1455 void p4d_clear_huge(p4d_t *p4d);
1456 #else
1457 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1458 {
1459 	return 0;
1460 }
1461 static inline void p4d_clear_huge(p4d_t *p4d) { }
1462 #endif /* !__PAGETABLE_P4D_FOLDED */
1463 
1464 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1465 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1466 int pud_clear_huge(pud_t *pud);
1467 int pmd_clear_huge(pmd_t *pmd);
1468 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1469 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1470 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1471 #else	/* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1472 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1473 {
1474 	return 0;
1475 }
1476 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1477 {
1478 	return 0;
1479 }
1480 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1481 {
1482 	return 0;
1483 }
1484 static inline void p4d_clear_huge(p4d_t *p4d) { }
1485 static inline int pud_clear_huge(pud_t *pud)
1486 {
1487 	return 0;
1488 }
1489 static inline int pmd_clear_huge(pmd_t *pmd)
1490 {
1491 	return 0;
1492 }
1493 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1494 {
1495 	return 0;
1496 }
1497 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1498 {
1499 	return 0;
1500 }
1501 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1502 {
1503 	return 0;
1504 }
1505 #endif	/* CONFIG_HAVE_ARCH_HUGE_VMAP */
1506 
1507 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1508 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1509 /*
1510  * ARCHes with special requirements for evicting THP backing TLB entries can
1511  * implement this. Otherwise also, it can help optimize normal TLB flush in
1512  * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1513  * entire TLB if flush span is greater than a threshold, which will
1514  * likely be true for a single huge page. Thus a single THP flush will
1515  * invalidate the entire TLB which is not desirable.
1516  * e.g. see arch/arc: flush_pmd_tlb_range
1517  */
1518 #define flush_pmd_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1519 #define flush_pud_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1520 #else
1521 #define flush_pmd_tlb_range(vma, addr, end)	BUILD_BUG()
1522 #define flush_pud_tlb_range(vma, addr, end)	BUILD_BUG()
1523 #endif
1524 #endif
1525 
1526 struct file;
1527 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1528 			unsigned long size, pgprot_t *vma_prot);
1529 
1530 #ifndef CONFIG_X86_ESPFIX64
1531 static inline void init_espfix_bsp(void) { }
1532 #endif
1533 
1534 extern void __init pgtable_cache_init(void);
1535 
1536 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1537 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1538 {
1539 	return true;
1540 }
1541 
1542 static inline bool arch_has_pfn_modify_check(void)
1543 {
1544 	return false;
1545 }
1546 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1547 
1548 /*
1549  * Architecture PAGE_KERNEL_* fallbacks
1550  *
1551  * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1552  * because they really don't support them, or the port needs to be updated to
1553  * reflect the required functionality. Below are a set of relatively safe
1554  * fallbacks, as best effort, which we can count on in lieu of the architectures
1555  * not defining them on their own yet.
1556  */
1557 
1558 #ifndef PAGE_KERNEL_RO
1559 # define PAGE_KERNEL_RO PAGE_KERNEL
1560 #endif
1561 
1562 #ifndef PAGE_KERNEL_EXEC
1563 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1564 #endif
1565 
1566 /*
1567  * Page Table Modification bits for pgtbl_mod_mask.
1568  *
1569  * These are used by the p?d_alloc_track*() set of functions an in the generic
1570  * vmalloc/ioremap code to track at which page-table levels entries have been
1571  * modified. Based on that the code can better decide when vmalloc and ioremap
1572  * mapping changes need to be synchronized to other page-tables in the system.
1573  */
1574 #define		__PGTBL_PGD_MODIFIED	0
1575 #define		__PGTBL_P4D_MODIFIED	1
1576 #define		__PGTBL_PUD_MODIFIED	2
1577 #define		__PGTBL_PMD_MODIFIED	3
1578 #define		__PGTBL_PTE_MODIFIED	4
1579 
1580 #define		PGTBL_PGD_MODIFIED	BIT(__PGTBL_PGD_MODIFIED)
1581 #define		PGTBL_P4D_MODIFIED	BIT(__PGTBL_P4D_MODIFIED)
1582 #define		PGTBL_PUD_MODIFIED	BIT(__PGTBL_PUD_MODIFIED)
1583 #define		PGTBL_PMD_MODIFIED	BIT(__PGTBL_PMD_MODIFIED)
1584 #define		PGTBL_PTE_MODIFIED	BIT(__PGTBL_PTE_MODIFIED)
1585 
1586 /* Page-Table Modification Mask */
1587 typedef unsigned int pgtbl_mod_mask;
1588 
1589 #endif /* !__ASSEMBLY__ */
1590 
1591 #if !defined(MAX_POSSIBLE_PHYSMEM_BITS) && !defined(CONFIG_64BIT)
1592 #ifdef CONFIG_PHYS_ADDR_T_64BIT
1593 /*
1594  * ZSMALLOC needs to know the highest PFN on 32-bit architectures
1595  * with physical address space extension, but falls back to
1596  * BITS_PER_LONG otherwise.
1597  */
1598 #error Missing MAX_POSSIBLE_PHYSMEM_BITS definition
1599 #else
1600 #define MAX_POSSIBLE_PHYSMEM_BITS 32
1601 #endif
1602 #endif
1603 
1604 #ifndef has_transparent_hugepage
1605 #define has_transparent_hugepage() IS_BUILTIN(CONFIG_TRANSPARENT_HUGEPAGE)
1606 #endif
1607 
1608 /*
1609  * On some architectures it depends on the mm if the p4d/pud or pmd
1610  * layer of the page table hierarchy is folded or not.
1611  */
1612 #ifndef mm_p4d_folded
1613 #define mm_p4d_folded(mm)	__is_defined(__PAGETABLE_P4D_FOLDED)
1614 #endif
1615 
1616 #ifndef mm_pud_folded
1617 #define mm_pud_folded(mm)	__is_defined(__PAGETABLE_PUD_FOLDED)
1618 #endif
1619 
1620 #ifndef mm_pmd_folded
1621 #define mm_pmd_folded(mm)	__is_defined(__PAGETABLE_PMD_FOLDED)
1622 #endif
1623 
1624 #ifndef p4d_offset_lockless
1625 #define p4d_offset_lockless(pgdp, pgd, address) p4d_offset(&(pgd), address)
1626 #endif
1627 #ifndef pud_offset_lockless
1628 #define pud_offset_lockless(p4dp, p4d, address) pud_offset(&(p4d), address)
1629 #endif
1630 #ifndef pmd_offset_lockless
1631 #define pmd_offset_lockless(pudp, pud, address) pmd_offset(&(pud), address)
1632 #endif
1633 
1634 /*
1635  * p?d_leaf() - true if this entry is a final mapping to a physical address.
1636  * This differs from p?d_huge() by the fact that they are always available (if
1637  * the architecture supports large pages at the appropriate level) even
1638  * if CONFIG_HUGETLB_PAGE is not defined.
1639  * Only meaningful when called on a valid entry.
1640  */
1641 #ifndef pgd_leaf
1642 #define pgd_leaf(x)	0
1643 #endif
1644 #ifndef p4d_leaf
1645 #define p4d_leaf(x)	0
1646 #endif
1647 #ifndef pud_leaf
1648 #define pud_leaf(x)	0
1649 #endif
1650 #ifndef pmd_leaf
1651 #define pmd_leaf(x)	0
1652 #endif
1653 
1654 #ifndef pgd_leaf_size
1655 #define pgd_leaf_size(x) (1ULL << PGDIR_SHIFT)
1656 #endif
1657 #ifndef p4d_leaf_size
1658 #define p4d_leaf_size(x) P4D_SIZE
1659 #endif
1660 #ifndef pud_leaf_size
1661 #define pud_leaf_size(x) PUD_SIZE
1662 #endif
1663 #ifndef pmd_leaf_size
1664 #define pmd_leaf_size(x) PMD_SIZE
1665 #endif
1666 #ifndef pte_leaf_size
1667 #define pte_leaf_size(x) PAGE_SIZE
1668 #endif
1669 
1670 /*
1671  * Some architectures have MMUs that are configurable or selectable at boot
1672  * time. These lead to variable PTRS_PER_x. For statically allocated arrays it
1673  * helps to have a static maximum value.
1674  */
1675 
1676 #ifndef MAX_PTRS_PER_PTE
1677 #define MAX_PTRS_PER_PTE PTRS_PER_PTE
1678 #endif
1679 
1680 #ifndef MAX_PTRS_PER_PMD
1681 #define MAX_PTRS_PER_PMD PTRS_PER_PMD
1682 #endif
1683 
1684 #ifndef MAX_PTRS_PER_PUD
1685 #define MAX_PTRS_PER_PUD PTRS_PER_PUD
1686 #endif
1687 
1688 #ifndef MAX_PTRS_PER_P4D
1689 #define MAX_PTRS_PER_P4D PTRS_PER_P4D
1690 #endif
1691 
1692 /* description of effects of mapping type and prot in current implementation.
1693  * this is due to the limited x86 page protection hardware.  The expected
1694  * behavior is in parens:
1695  *
1696  * map_type	prot
1697  *		PROT_NONE	PROT_READ	PROT_WRITE	PROT_EXEC
1698  * MAP_SHARED	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1699  *		w: (no) no	w: (no) no	w: (yes) yes	w: (no) no
1700  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1701  *
1702  * MAP_PRIVATE	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1703  *		w: (no) no	w: (no) no	w: (copy) copy	w: (no) no
1704  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1705  *
1706  * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
1707  * MAP_PRIVATE (with Enhanced PAN supported):
1708  *								r: (no) no
1709  *								w: (no) no
1710  *								x: (yes) yes
1711  */
1712 #define DECLARE_VM_GET_PAGE_PROT					\
1713 pgprot_t vm_get_page_prot(unsigned long vm_flags)			\
1714 {									\
1715 		return protection_map[vm_flags &			\
1716 			(VM_READ | VM_WRITE | VM_EXEC | VM_SHARED)];	\
1717 }									\
1718 EXPORT_SYMBOL(vm_get_page_prot);
1719 
1720 #endif /* _LINUX_PGTABLE_H */
1721