xref: /linux-6.15/include/linux/pgtable.h (revision 534aa1dc)
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  * This defines the first usable user address. Platforms
33  * can override its value with custom FIRST_USER_ADDRESS
34  * defined in their respective <asm/pgtable.h>.
35  */
36 #ifndef FIRST_USER_ADDRESS
37 #define FIRST_USER_ADDRESS	0UL
38 #endif
39 
40 /*
41  * This defines the generic helper for accessing PMD page
42  * table page. Although platforms can still override this
43  * via their respective <asm/pgtable.h>.
44  */
45 #ifndef pmd_pgtable
46 #define pmd_pgtable(pmd) pmd_page(pmd)
47 #endif
48 
49 /*
50  * A page table page can be thought of an array like this: pXd_t[PTRS_PER_PxD]
51  *
52  * The pXx_index() functions return the index of the entry in the page
53  * table page which would control the given virtual address
54  *
55  * As these functions may be used by the same code for different levels of
56  * the page table folding, they are always available, regardless of
57  * CONFIG_PGTABLE_LEVELS value. For the folded levels they simply return 0
58  * because in such cases PTRS_PER_PxD equals 1.
59  */
60 
61 static inline unsigned long pte_index(unsigned long address)
62 {
63 	return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
64 }
65 #define pte_index pte_index
66 
67 #ifndef pmd_index
68 static inline unsigned long pmd_index(unsigned long address)
69 {
70 	return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
71 }
72 #define pmd_index pmd_index
73 #endif
74 
75 #ifndef pud_index
76 static inline unsigned long pud_index(unsigned long address)
77 {
78 	return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
79 }
80 #define pud_index pud_index
81 #endif
82 
83 #ifndef pgd_index
84 /* Must be a compile-time constant, so implement it as a macro */
85 #define pgd_index(a)  (((a) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
86 #endif
87 
88 #ifndef pte_offset_kernel
89 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
90 {
91 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
92 }
93 #define pte_offset_kernel pte_offset_kernel
94 #endif
95 
96 #if defined(CONFIG_HIGHPTE)
97 #define pte_offset_map(dir, address)				\
98 	((pte_t *)kmap_atomic(pmd_page(*(dir))) +		\
99 	 pte_index((address)))
100 #define pte_unmap(pte) kunmap_atomic((pte))
101 #else
102 #define pte_offset_map(dir, address)	pte_offset_kernel((dir), (address))
103 #define pte_unmap(pte) ((void)(pte))	/* NOP */
104 #endif
105 
106 /* Find an entry in the second-level page table.. */
107 #ifndef pmd_offset
108 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
109 {
110 	return pud_pgtable(*pud) + pmd_index(address);
111 }
112 #define pmd_offset pmd_offset
113 #endif
114 
115 #ifndef pud_offset
116 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
117 {
118 	return p4d_pgtable(*p4d) + pud_index(address);
119 }
120 #define pud_offset pud_offset
121 #endif
122 
123 static inline pgd_t *pgd_offset_pgd(pgd_t *pgd, unsigned long address)
124 {
125 	return (pgd + pgd_index(address));
126 };
127 
128 /*
129  * a shortcut to get a pgd_t in a given mm
130  */
131 #ifndef pgd_offset
132 #define pgd_offset(mm, address)		pgd_offset_pgd((mm)->pgd, (address))
133 #endif
134 
135 /*
136  * a shortcut which implies the use of the kernel's pgd, instead
137  * of a process's
138  */
139 #ifndef pgd_offset_k
140 #define pgd_offset_k(address)		pgd_offset(&init_mm, (address))
141 #endif
142 
143 /*
144  * In many cases it is known that a virtual address is mapped at PMD or PTE
145  * level, so instead of traversing all the page table levels, we can get a
146  * pointer to the PMD entry in user or kernel page table or translate a virtual
147  * address to the pointer in the PTE in the kernel page tables with simple
148  * helpers.
149  */
150 static inline pmd_t *pmd_off(struct mm_struct *mm, unsigned long va)
151 {
152 	return pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, va), va), va), va);
153 }
154 
155 static inline pmd_t *pmd_off_k(unsigned long va)
156 {
157 	return pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va);
158 }
159 
160 static inline pte_t *virt_to_kpte(unsigned long vaddr)
161 {
162 	pmd_t *pmd = pmd_off_k(vaddr);
163 
164 	return pmd_none(*pmd) ? NULL : pte_offset_kernel(pmd, vaddr);
165 }
166 
167 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
168 extern int ptep_set_access_flags(struct vm_area_struct *vma,
169 				 unsigned long address, pte_t *ptep,
170 				 pte_t entry, int dirty);
171 #endif
172 
173 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
174 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
175 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
176 				 unsigned long address, pmd_t *pmdp,
177 				 pmd_t entry, int dirty);
178 extern int pudp_set_access_flags(struct vm_area_struct *vma,
179 				 unsigned long address, pud_t *pudp,
180 				 pud_t entry, int dirty);
181 #else
182 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
183 					unsigned long address, pmd_t *pmdp,
184 					pmd_t entry, int dirty)
185 {
186 	BUILD_BUG();
187 	return 0;
188 }
189 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
190 					unsigned long address, pud_t *pudp,
191 					pud_t entry, int dirty)
192 {
193 	BUILD_BUG();
194 	return 0;
195 }
196 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
197 #endif
198 
199 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
200 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
201 					    unsigned long address,
202 					    pte_t *ptep)
203 {
204 	pte_t pte = *ptep;
205 	int r = 1;
206 	if (!pte_young(pte))
207 		r = 0;
208 	else
209 		set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
210 	return r;
211 }
212 #endif
213 
214 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
215 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
216 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
217 					    unsigned long address,
218 					    pmd_t *pmdp)
219 {
220 	pmd_t pmd = *pmdp;
221 	int r = 1;
222 	if (!pmd_young(pmd))
223 		r = 0;
224 	else
225 		set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
226 	return r;
227 }
228 #else
229 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
230 					    unsigned long address,
231 					    pmd_t *pmdp)
232 {
233 	BUILD_BUG();
234 	return 0;
235 }
236 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
237 #endif
238 
239 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
240 int ptep_clear_flush_young(struct vm_area_struct *vma,
241 			   unsigned long address, pte_t *ptep);
242 #endif
243 
244 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
245 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
246 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
247 				  unsigned long address, pmd_t *pmdp);
248 #else
249 /*
250  * Despite relevant to THP only, this API is called from generic rmap code
251  * under PageTransHuge(), hence needs a dummy implementation for !THP
252  */
253 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
254 					 unsigned long address, pmd_t *pmdp)
255 {
256 	BUILD_BUG();
257 	return 0;
258 }
259 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
260 #endif
261 
262 #ifndef __HAVE_ARCH_PTEP_CLEAR
263 static inline void ptep_clear(struct mm_struct *mm, unsigned long addr,
264 			      pte_t *ptep)
265 {
266 	pte_clear(mm, addr, ptep);
267 }
268 #endif
269 
270 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
271 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
272 				       unsigned long address,
273 				       pte_t *ptep)
274 {
275 	pte_t pte = *ptep;
276 	pte_clear(mm, address, ptep);
277 	return pte;
278 }
279 #endif
280 
281 #ifndef __HAVE_ARCH_PTEP_GET
282 static inline pte_t ptep_get(pte_t *ptep)
283 {
284 	return READ_ONCE(*ptep);
285 }
286 #endif
287 
288 #ifdef CONFIG_GUP_GET_PTE_LOW_HIGH
289 /*
290  * WARNING: only to be used in the get_user_pages_fast() implementation.
291  *
292  * With get_user_pages_fast(), we walk down the pagetables without taking any
293  * locks.  For this we would like to load the pointers atomically, but sometimes
294  * that is not possible (e.g. without expensive cmpxchg8b on x86_32 PAE).  What
295  * we do have is the guarantee that a PTE will only either go from not present
296  * to present, or present to not present or both -- it will not switch to a
297  * completely different present page without a TLB flush in between; something
298  * that we are blocking by holding interrupts off.
299  *
300  * Setting ptes from not present to present goes:
301  *
302  *   ptep->pte_high = h;
303  *   smp_wmb();
304  *   ptep->pte_low = l;
305  *
306  * And present to not present goes:
307  *
308  *   ptep->pte_low = 0;
309  *   smp_wmb();
310  *   ptep->pte_high = 0;
311  *
312  * We must ensure here that the load of pte_low sees 'l' IFF pte_high sees 'h'.
313  * We load pte_high *after* loading pte_low, which ensures we don't see an older
314  * value of pte_high.  *Then* we recheck pte_low, which ensures that we haven't
315  * picked up a changed pte high. We might have gotten rubbish values from
316  * pte_low and pte_high, but we are guaranteed that pte_low will not have the
317  * present bit set *unless* it is 'l'. Because get_user_pages_fast() only
318  * operates on present ptes we're safe.
319  */
320 static inline pte_t ptep_get_lockless(pte_t *ptep)
321 {
322 	pte_t pte;
323 
324 	do {
325 		pte.pte_low = ptep->pte_low;
326 		smp_rmb();
327 		pte.pte_high = ptep->pte_high;
328 		smp_rmb();
329 	} while (unlikely(pte.pte_low != ptep->pte_low));
330 
331 	return pte;
332 }
333 #else /* CONFIG_GUP_GET_PTE_LOW_HIGH */
334 /*
335  * We require that the PTE can be read atomically.
336  */
337 static inline pte_t ptep_get_lockless(pte_t *ptep)
338 {
339 	return ptep_get(ptep);
340 }
341 #endif /* CONFIG_GUP_GET_PTE_LOW_HIGH */
342 
343 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
344 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
345 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
346 					    unsigned long address,
347 					    pmd_t *pmdp)
348 {
349 	pmd_t pmd = *pmdp;
350 	pmd_clear(pmdp);
351 	return pmd;
352 }
353 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
354 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
355 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
356 					    unsigned long address,
357 					    pud_t *pudp)
358 {
359 	pud_t pud = *pudp;
360 
361 	pud_clear(pudp);
362 	return pud;
363 }
364 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
365 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
366 
367 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
368 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
369 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
370 					    unsigned long address, pmd_t *pmdp,
371 					    int full)
372 {
373 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
374 }
375 #endif
376 
377 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
378 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
379 					    unsigned long address, pud_t *pudp,
380 					    int full)
381 {
382 	return pudp_huge_get_and_clear(mm, address, pudp);
383 }
384 #endif
385 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
386 
387 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
388 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
389 					    unsigned long address, pte_t *ptep,
390 					    int full)
391 {
392 	pte_t pte;
393 	pte = ptep_get_and_clear(mm, address, ptep);
394 	return pte;
395 }
396 #endif
397 
398 
399 /*
400  * If two threads concurrently fault at the same page, the thread that
401  * won the race updates the PTE and its local TLB/Cache. The other thread
402  * gives up, simply does nothing, and continues; on architectures where
403  * software can update TLB,  local TLB can be updated here to avoid next page
404  * fault. This function updates TLB only, do nothing with cache or others.
405  * It is the difference with function update_mmu_cache.
406  */
407 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
408 static inline void update_mmu_tlb(struct vm_area_struct *vma,
409 				unsigned long address, pte_t *ptep)
410 {
411 }
412 #define __HAVE_ARCH_UPDATE_MMU_TLB
413 #endif
414 
415 /*
416  * Some architectures may be able to avoid expensive synchronization
417  * primitives when modifications are made to PTE's which are already
418  * not present, or in the process of an address space destruction.
419  */
420 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
421 static inline void pte_clear_not_present_full(struct mm_struct *mm,
422 					      unsigned long address,
423 					      pte_t *ptep,
424 					      int full)
425 {
426 	pte_clear(mm, address, ptep);
427 }
428 #endif
429 
430 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
431 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
432 			      unsigned long address,
433 			      pte_t *ptep);
434 #endif
435 
436 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
437 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
438 			      unsigned long address,
439 			      pmd_t *pmdp);
440 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
441 			      unsigned long address,
442 			      pud_t *pudp);
443 #endif
444 
445 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
446 struct mm_struct;
447 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
448 {
449 	pte_t old_pte = *ptep;
450 	set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
451 }
452 #endif
453 
454 /*
455  * On some architectures hardware does not set page access bit when accessing
456  * memory page, it is responsibility of software setting this bit. It brings
457  * out extra page fault penalty to track page access bit. For optimization page
458  * access bit can be set during all page fault flow on these arches.
459  * To be differentiate with macro pte_mkyoung, this macro is used on platforms
460  * where software maintains page access bit.
461  */
462 #ifndef pte_sw_mkyoung
463 static inline pte_t pte_sw_mkyoung(pte_t pte)
464 {
465 	return pte;
466 }
467 #define pte_sw_mkyoung	pte_sw_mkyoung
468 #endif
469 
470 #ifndef pte_savedwrite
471 #define pte_savedwrite pte_write
472 #endif
473 
474 #ifndef pte_mk_savedwrite
475 #define pte_mk_savedwrite pte_mkwrite
476 #endif
477 
478 #ifndef pte_clear_savedwrite
479 #define pte_clear_savedwrite pte_wrprotect
480 #endif
481 
482 #ifndef pmd_savedwrite
483 #define pmd_savedwrite pmd_write
484 #endif
485 
486 #ifndef pmd_mk_savedwrite
487 #define pmd_mk_savedwrite pmd_mkwrite
488 #endif
489 
490 #ifndef pmd_clear_savedwrite
491 #define pmd_clear_savedwrite pmd_wrprotect
492 #endif
493 
494 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
495 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
496 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
497 				      unsigned long address, pmd_t *pmdp)
498 {
499 	pmd_t old_pmd = *pmdp;
500 	set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
501 }
502 #else
503 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
504 				      unsigned long address, pmd_t *pmdp)
505 {
506 	BUILD_BUG();
507 }
508 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
509 #endif
510 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
511 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
512 static inline void pudp_set_wrprotect(struct mm_struct *mm,
513 				      unsigned long address, pud_t *pudp)
514 {
515 	pud_t old_pud = *pudp;
516 
517 	set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
518 }
519 #else
520 static inline void pudp_set_wrprotect(struct mm_struct *mm,
521 				      unsigned long address, pud_t *pudp)
522 {
523 	BUILD_BUG();
524 }
525 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
526 #endif
527 
528 #ifndef pmdp_collapse_flush
529 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
530 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
531 				 unsigned long address, pmd_t *pmdp);
532 #else
533 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
534 					unsigned long address,
535 					pmd_t *pmdp)
536 {
537 	BUILD_BUG();
538 	return *pmdp;
539 }
540 #define pmdp_collapse_flush pmdp_collapse_flush
541 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
542 #endif
543 
544 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
545 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
546 				       pgtable_t pgtable);
547 #endif
548 
549 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
550 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
551 #endif
552 
553 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
554 /*
555  * This is an implementation of pmdp_establish() that is only suitable for an
556  * architecture that doesn't have hardware dirty/accessed bits. In this case we
557  * can't race with CPU which sets these bits and non-atomic approach is fine.
558  */
559 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
560 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
561 {
562 	pmd_t old_pmd = *pmdp;
563 	set_pmd_at(vma->vm_mm, address, pmdp, pmd);
564 	return old_pmd;
565 }
566 #endif
567 
568 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
569 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
570 			    pmd_t *pmdp);
571 #endif
572 
573 #ifndef __HAVE_ARCH_PMDP_INVALIDATE_AD
574 
575 /*
576  * pmdp_invalidate_ad() invalidates the PMD while changing a transparent
577  * hugepage mapping in the page tables. This function is similar to
578  * pmdp_invalidate(), but should only be used if the access and dirty bits would
579  * not be cleared by the software in the new PMD value. The function ensures
580  * that hardware changes of the access and dirty bits updates would not be lost.
581  *
582  * Doing so can allow in certain architectures to avoid a TLB flush in most
583  * cases. Yet, another TLB flush might be necessary later if the PMD update
584  * itself requires such flush (e.g., if protection was set to be stricter). Yet,
585  * even when a TLB flush is needed because of the update, the caller may be able
586  * to batch these TLB flushing operations, so fewer TLB flush operations are
587  * needed.
588  */
589 extern pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma,
590 				unsigned long address, pmd_t *pmdp);
591 #endif
592 
593 #ifndef __HAVE_ARCH_PTE_SAME
594 static inline int pte_same(pte_t pte_a, pte_t pte_b)
595 {
596 	return pte_val(pte_a) == pte_val(pte_b);
597 }
598 #endif
599 
600 #ifndef __HAVE_ARCH_PTE_UNUSED
601 /*
602  * Some architectures provide facilities to virtualization guests
603  * so that they can flag allocated pages as unused. This allows the
604  * host to transparently reclaim unused pages. This function returns
605  * whether the pte's page is unused.
606  */
607 static inline int pte_unused(pte_t pte)
608 {
609 	return 0;
610 }
611 #endif
612 
613 #ifndef pte_access_permitted
614 #define pte_access_permitted(pte, write) \
615 	(pte_present(pte) && (!(write) || pte_write(pte)))
616 #endif
617 
618 #ifndef pmd_access_permitted
619 #define pmd_access_permitted(pmd, write) \
620 	(pmd_present(pmd) && (!(write) || pmd_write(pmd)))
621 #endif
622 
623 #ifndef pud_access_permitted
624 #define pud_access_permitted(pud, write) \
625 	(pud_present(pud) && (!(write) || pud_write(pud)))
626 #endif
627 
628 #ifndef p4d_access_permitted
629 #define p4d_access_permitted(p4d, write) \
630 	(p4d_present(p4d) && (!(write) || p4d_write(p4d)))
631 #endif
632 
633 #ifndef pgd_access_permitted
634 #define pgd_access_permitted(pgd, write) \
635 	(pgd_present(pgd) && (!(write) || pgd_write(pgd)))
636 #endif
637 
638 #ifndef __HAVE_ARCH_PMD_SAME
639 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
640 {
641 	return pmd_val(pmd_a) == pmd_val(pmd_b);
642 }
643 
644 static inline int pud_same(pud_t pud_a, pud_t pud_b)
645 {
646 	return pud_val(pud_a) == pud_val(pud_b);
647 }
648 #endif
649 
650 #ifndef __HAVE_ARCH_P4D_SAME
651 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
652 {
653 	return p4d_val(p4d_a) == p4d_val(p4d_b);
654 }
655 #endif
656 
657 #ifndef __HAVE_ARCH_PGD_SAME
658 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
659 {
660 	return pgd_val(pgd_a) == pgd_val(pgd_b);
661 }
662 #endif
663 
664 /*
665  * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
666  * TLB flush will be required as a result of the "set". For example, use
667  * in scenarios where it is known ahead of time that the routine is
668  * setting non-present entries, or re-setting an existing entry to the
669  * same value. Otherwise, use the typical "set" helpers and flush the
670  * TLB.
671  */
672 #define set_pte_safe(ptep, pte) \
673 ({ \
674 	WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
675 	set_pte(ptep, pte); \
676 })
677 
678 #define set_pmd_safe(pmdp, pmd) \
679 ({ \
680 	WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
681 	set_pmd(pmdp, pmd); \
682 })
683 
684 #define set_pud_safe(pudp, pud) \
685 ({ \
686 	WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
687 	set_pud(pudp, pud); \
688 })
689 
690 #define set_p4d_safe(p4dp, p4d) \
691 ({ \
692 	WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
693 	set_p4d(p4dp, p4d); \
694 })
695 
696 #define set_pgd_safe(pgdp, pgd) \
697 ({ \
698 	WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
699 	set_pgd(pgdp, pgd); \
700 })
701 
702 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
703 /*
704  * Some architectures support metadata associated with a page. When a
705  * page is being swapped out, this metadata must be saved so it can be
706  * restored when the page is swapped back in. SPARC M7 and newer
707  * processors support an ADI (Application Data Integrity) tag for the
708  * page as metadata for the page. arch_do_swap_page() can restore this
709  * metadata when a page is swapped back in.
710  */
711 static inline void arch_do_swap_page(struct mm_struct *mm,
712 				     struct vm_area_struct *vma,
713 				     unsigned long addr,
714 				     pte_t pte, pte_t oldpte)
715 {
716 
717 }
718 #endif
719 
720 #ifndef __HAVE_ARCH_UNMAP_ONE
721 /*
722  * Some architectures support metadata associated with a page. When a
723  * page is being swapped out, this metadata must be saved so it can be
724  * restored when the page is swapped back in. SPARC M7 and newer
725  * processors support an ADI (Application Data Integrity) tag for the
726  * page as metadata for the page. arch_unmap_one() can save this
727  * metadata on a swap-out of a page.
728  */
729 static inline int arch_unmap_one(struct mm_struct *mm,
730 				  struct vm_area_struct *vma,
731 				  unsigned long addr,
732 				  pte_t orig_pte)
733 {
734 	return 0;
735 }
736 #endif
737 
738 /*
739  * Allow architectures to preserve additional metadata associated with
740  * swapped-out pages. The corresponding __HAVE_ARCH_SWAP_* macros and function
741  * prototypes must be defined in the arch-specific asm/pgtable.h file.
742  */
743 #ifndef __HAVE_ARCH_PREPARE_TO_SWAP
744 static inline int arch_prepare_to_swap(struct page *page)
745 {
746 	return 0;
747 }
748 #endif
749 
750 #ifndef __HAVE_ARCH_SWAP_INVALIDATE
751 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
752 {
753 }
754 
755 static inline void arch_swap_invalidate_area(int type)
756 {
757 }
758 #endif
759 
760 #ifndef __HAVE_ARCH_SWAP_RESTORE
761 static inline void arch_swap_restore(swp_entry_t entry, struct page *page)
762 {
763 }
764 #endif
765 
766 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
767 #define pgd_offset_gate(mm, addr)	pgd_offset(mm, addr)
768 #endif
769 
770 #ifndef __HAVE_ARCH_MOVE_PTE
771 #define move_pte(pte, prot, old_addr, new_addr)	(pte)
772 #endif
773 
774 #ifndef pte_accessible
775 # define pte_accessible(mm, pte)	((void)(pte), 1)
776 #endif
777 
778 #ifndef flush_tlb_fix_spurious_fault
779 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
780 #endif
781 
782 /*
783  * When walking page tables, get the address of the next boundary,
784  * or the end address of the range if that comes earlier.  Although no
785  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
786  */
787 
788 #define pgd_addr_end(addr, end)						\
789 ({	unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;	\
790 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
791 })
792 
793 #ifndef p4d_addr_end
794 #define p4d_addr_end(addr, end)						\
795 ({	unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;	\
796 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
797 })
798 #endif
799 
800 #ifndef pud_addr_end
801 #define pud_addr_end(addr, end)						\
802 ({	unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;	\
803 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
804 })
805 #endif
806 
807 #ifndef pmd_addr_end
808 #define pmd_addr_end(addr, end)						\
809 ({	unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;	\
810 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
811 })
812 #endif
813 
814 /*
815  * When walking page tables, we usually want to skip any p?d_none entries;
816  * and any p?d_bad entries - reporting the error before resetting to none.
817  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
818  */
819 void pgd_clear_bad(pgd_t *);
820 
821 #ifndef __PAGETABLE_P4D_FOLDED
822 void p4d_clear_bad(p4d_t *);
823 #else
824 #define p4d_clear_bad(p4d)        do { } while (0)
825 #endif
826 
827 #ifndef __PAGETABLE_PUD_FOLDED
828 void pud_clear_bad(pud_t *);
829 #else
830 #define pud_clear_bad(p4d)        do { } while (0)
831 #endif
832 
833 void pmd_clear_bad(pmd_t *);
834 
835 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
836 {
837 	if (pgd_none(*pgd))
838 		return 1;
839 	if (unlikely(pgd_bad(*pgd))) {
840 		pgd_clear_bad(pgd);
841 		return 1;
842 	}
843 	return 0;
844 }
845 
846 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
847 {
848 	if (p4d_none(*p4d))
849 		return 1;
850 	if (unlikely(p4d_bad(*p4d))) {
851 		p4d_clear_bad(p4d);
852 		return 1;
853 	}
854 	return 0;
855 }
856 
857 static inline int pud_none_or_clear_bad(pud_t *pud)
858 {
859 	if (pud_none(*pud))
860 		return 1;
861 	if (unlikely(pud_bad(*pud))) {
862 		pud_clear_bad(pud);
863 		return 1;
864 	}
865 	return 0;
866 }
867 
868 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
869 {
870 	if (pmd_none(*pmd))
871 		return 1;
872 	if (unlikely(pmd_bad(*pmd))) {
873 		pmd_clear_bad(pmd);
874 		return 1;
875 	}
876 	return 0;
877 }
878 
879 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
880 					     unsigned long addr,
881 					     pte_t *ptep)
882 {
883 	/*
884 	 * Get the current pte state, but zero it out to make it
885 	 * non-present, preventing the hardware from asynchronously
886 	 * updating it.
887 	 */
888 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
889 }
890 
891 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
892 					     unsigned long addr,
893 					     pte_t *ptep, pte_t pte)
894 {
895 	/*
896 	 * The pte is non-present, so there's no hardware state to
897 	 * preserve.
898 	 */
899 	set_pte_at(vma->vm_mm, addr, ptep, pte);
900 }
901 
902 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
903 /*
904  * Start a pte protection read-modify-write transaction, which
905  * protects against asynchronous hardware modifications to the pte.
906  * The intention is not to prevent the hardware from making pte
907  * updates, but to prevent any updates it may make from being lost.
908  *
909  * This does not protect against other software modifications of the
910  * pte; the appropriate pte lock must be held over the transaction.
911  *
912  * Note that this interface is intended to be batchable, meaning that
913  * ptep_modify_prot_commit may not actually update the pte, but merely
914  * queue the update to be done at some later time.  The update must be
915  * actually committed before the pte lock is released, however.
916  */
917 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
918 					   unsigned long addr,
919 					   pte_t *ptep)
920 {
921 	return __ptep_modify_prot_start(vma, addr, ptep);
922 }
923 
924 /*
925  * Commit an update to a pte, leaving any hardware-controlled bits in
926  * the PTE unmodified.
927  */
928 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
929 					   unsigned long addr,
930 					   pte_t *ptep, pte_t old_pte, pte_t pte)
931 {
932 	__ptep_modify_prot_commit(vma, addr, ptep, pte);
933 }
934 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
935 #endif /* CONFIG_MMU */
936 
937 /*
938  * No-op macros that just return the current protection value. Defined here
939  * because these macros can be used even if CONFIG_MMU is not defined.
940  */
941 
942 #ifndef pgprot_nx
943 #define pgprot_nx(prot)	(prot)
944 #endif
945 
946 #ifndef pgprot_noncached
947 #define pgprot_noncached(prot)	(prot)
948 #endif
949 
950 #ifndef pgprot_writecombine
951 #define pgprot_writecombine pgprot_noncached
952 #endif
953 
954 #ifndef pgprot_writethrough
955 #define pgprot_writethrough pgprot_noncached
956 #endif
957 
958 #ifndef pgprot_device
959 #define pgprot_device pgprot_noncached
960 #endif
961 
962 #ifndef pgprot_mhp
963 #define pgprot_mhp(prot)	(prot)
964 #endif
965 
966 #ifdef CONFIG_MMU
967 #ifndef pgprot_modify
968 #define pgprot_modify pgprot_modify
969 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
970 {
971 	if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
972 		newprot = pgprot_noncached(newprot);
973 	if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
974 		newprot = pgprot_writecombine(newprot);
975 	if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
976 		newprot = pgprot_device(newprot);
977 	return newprot;
978 }
979 #endif
980 #endif /* CONFIG_MMU */
981 
982 #ifndef pgprot_encrypted
983 #define pgprot_encrypted(prot)	(prot)
984 #endif
985 
986 #ifndef pgprot_decrypted
987 #define pgprot_decrypted(prot)	(prot)
988 #endif
989 
990 /*
991  * A facility to provide lazy MMU batching.  This allows PTE updates and
992  * page invalidations to be delayed until a call to leave lazy MMU mode
993  * is issued.  Some architectures may benefit from doing this, and it is
994  * beneficial for both shadow and direct mode hypervisors, which may batch
995  * the PTE updates which happen during this window.  Note that using this
996  * interface requires that read hazards be removed from the code.  A read
997  * hazard could result in the direct mode hypervisor case, since the actual
998  * write to the page tables may not yet have taken place, so reads though
999  * a raw PTE pointer after it has been modified are not guaranteed to be
1000  * up to date.  This mode can only be entered and left under the protection of
1001  * the page table locks for all page tables which may be modified.  In the UP
1002  * case, this is required so that preemption is disabled, and in the SMP case,
1003  * it must synchronize the delayed page table writes properly on other CPUs.
1004  */
1005 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
1006 #define arch_enter_lazy_mmu_mode()	do {} while (0)
1007 #define arch_leave_lazy_mmu_mode()	do {} while (0)
1008 #define arch_flush_lazy_mmu_mode()	do {} while (0)
1009 #endif
1010 
1011 /*
1012  * A facility to provide batching of the reload of page tables and
1013  * other process state with the actual context switch code for
1014  * paravirtualized guests.  By convention, only one of the batched
1015  * update (lazy) modes (CPU, MMU) should be active at any given time,
1016  * entry should never be nested, and entry and exits should always be
1017  * paired.  This is for sanity of maintaining and reasoning about the
1018  * kernel code.  In this case, the exit (end of the context switch) is
1019  * in architecture-specific code, and so doesn't need a generic
1020  * definition.
1021  */
1022 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
1023 #define arch_start_context_switch(prev)	do {} while (0)
1024 #endif
1025 
1026 /*
1027  * When replacing an anonymous page by a real (!non) swap entry, we clear
1028  * PG_anon_exclusive from the page and instead remember whether the flag was
1029  * set in the swp pte. During fork(), we have to mark the entry as !exclusive
1030  * (possibly shared). On swapin, we use that information to restore
1031  * PG_anon_exclusive, which is very helpful in cases where we might have
1032  * additional (e.g., FOLL_GET) references on a page and wouldn't be able to
1033  * detect exclusivity.
1034  *
1035  * These functions don't apply to non-swap entries (e.g., migration, hwpoison,
1036  * ...).
1037  */
1038 #ifndef __HAVE_ARCH_PTE_SWP_EXCLUSIVE
1039 static inline pte_t pte_swp_mkexclusive(pte_t pte)
1040 {
1041 	return pte;
1042 }
1043 
1044 static inline int pte_swp_exclusive(pte_t pte)
1045 {
1046 	return false;
1047 }
1048 
1049 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
1050 {
1051 	return pte;
1052 }
1053 #endif
1054 
1055 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1056 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
1057 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1058 {
1059 	return pmd;
1060 }
1061 
1062 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1063 {
1064 	return 0;
1065 }
1066 
1067 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1068 {
1069 	return pmd;
1070 }
1071 #endif
1072 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
1073 static inline int pte_soft_dirty(pte_t pte)
1074 {
1075 	return 0;
1076 }
1077 
1078 static inline int pmd_soft_dirty(pmd_t pmd)
1079 {
1080 	return 0;
1081 }
1082 
1083 static inline pte_t pte_mksoft_dirty(pte_t pte)
1084 {
1085 	return pte;
1086 }
1087 
1088 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
1089 {
1090 	return pmd;
1091 }
1092 
1093 static inline pte_t pte_clear_soft_dirty(pte_t pte)
1094 {
1095 	return pte;
1096 }
1097 
1098 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
1099 {
1100 	return pmd;
1101 }
1102 
1103 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1104 {
1105 	return pte;
1106 }
1107 
1108 static inline int pte_swp_soft_dirty(pte_t pte)
1109 {
1110 	return 0;
1111 }
1112 
1113 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1114 {
1115 	return pte;
1116 }
1117 
1118 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1119 {
1120 	return pmd;
1121 }
1122 
1123 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1124 {
1125 	return 0;
1126 }
1127 
1128 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1129 {
1130 	return pmd;
1131 }
1132 #endif
1133 
1134 #ifndef __HAVE_PFNMAP_TRACKING
1135 /*
1136  * Interfaces that can be used by architecture code to keep track of
1137  * memory type of pfn mappings specified by the remap_pfn_range,
1138  * vmf_insert_pfn.
1139  */
1140 
1141 /*
1142  * track_pfn_remap is called when a _new_ pfn mapping is being established
1143  * by remap_pfn_range() for physical range indicated by pfn and size.
1144  */
1145 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1146 				  unsigned long pfn, unsigned long addr,
1147 				  unsigned long size)
1148 {
1149 	return 0;
1150 }
1151 
1152 /*
1153  * track_pfn_insert is called when a _new_ single pfn is established
1154  * by vmf_insert_pfn().
1155  */
1156 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1157 				    pfn_t pfn)
1158 {
1159 }
1160 
1161 /*
1162  * track_pfn_copy is called when vma that is covering the pfnmap gets
1163  * copied through copy_page_range().
1164  */
1165 static inline int track_pfn_copy(struct vm_area_struct *vma)
1166 {
1167 	return 0;
1168 }
1169 
1170 /*
1171  * untrack_pfn is called while unmapping a pfnmap for a region.
1172  * untrack can be called for a specific region indicated by pfn and size or
1173  * can be for the entire vma (in which case pfn, size are zero).
1174  */
1175 static inline void untrack_pfn(struct vm_area_struct *vma,
1176 			       unsigned long pfn, unsigned long size)
1177 {
1178 }
1179 
1180 /*
1181  * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
1182  */
1183 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
1184 {
1185 }
1186 #else
1187 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1188 			   unsigned long pfn, unsigned long addr,
1189 			   unsigned long size);
1190 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1191 			     pfn_t pfn);
1192 extern int track_pfn_copy(struct vm_area_struct *vma);
1193 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1194 			unsigned long size);
1195 extern void untrack_pfn_moved(struct vm_area_struct *vma);
1196 #endif
1197 
1198 #ifdef CONFIG_MMU
1199 #ifdef __HAVE_COLOR_ZERO_PAGE
1200 static inline int is_zero_pfn(unsigned long pfn)
1201 {
1202 	extern unsigned long zero_pfn;
1203 	unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1204 	return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1205 }
1206 
1207 #define my_zero_pfn(addr)	page_to_pfn(ZERO_PAGE(addr))
1208 
1209 #else
1210 static inline int is_zero_pfn(unsigned long pfn)
1211 {
1212 	extern unsigned long zero_pfn;
1213 	return pfn == zero_pfn;
1214 }
1215 
1216 static inline unsigned long my_zero_pfn(unsigned long addr)
1217 {
1218 	extern unsigned long zero_pfn;
1219 	return zero_pfn;
1220 }
1221 #endif
1222 #else
1223 static inline int is_zero_pfn(unsigned long pfn)
1224 {
1225 	return 0;
1226 }
1227 
1228 static inline unsigned long my_zero_pfn(unsigned long addr)
1229 {
1230 	return 0;
1231 }
1232 #endif /* CONFIG_MMU */
1233 
1234 #ifdef CONFIG_MMU
1235 
1236 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1237 static inline int pmd_trans_huge(pmd_t pmd)
1238 {
1239 	return 0;
1240 }
1241 #ifndef pmd_write
1242 static inline int pmd_write(pmd_t pmd)
1243 {
1244 	BUG();
1245 	return 0;
1246 }
1247 #endif /* pmd_write */
1248 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1249 
1250 #ifndef pud_write
1251 static inline int pud_write(pud_t pud)
1252 {
1253 	BUG();
1254 	return 0;
1255 }
1256 #endif /* pud_write */
1257 
1258 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1259 static inline int pmd_devmap(pmd_t pmd)
1260 {
1261 	return 0;
1262 }
1263 static inline int pud_devmap(pud_t pud)
1264 {
1265 	return 0;
1266 }
1267 static inline int pgd_devmap(pgd_t pgd)
1268 {
1269 	return 0;
1270 }
1271 #endif
1272 
1273 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1274 	(defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1275 	 !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
1276 static inline int pud_trans_huge(pud_t pud)
1277 {
1278 	return 0;
1279 }
1280 #endif
1281 
1282 /* See pmd_none_or_trans_huge_or_clear_bad for discussion. */
1283 static inline int pud_none_or_trans_huge_or_dev_or_clear_bad(pud_t *pud)
1284 {
1285 	pud_t pudval = READ_ONCE(*pud);
1286 
1287 	if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1288 		return 1;
1289 	if (unlikely(pud_bad(pudval))) {
1290 		pud_clear_bad(pud);
1291 		return 1;
1292 	}
1293 	return 0;
1294 }
1295 
1296 /* See pmd_trans_unstable for discussion. */
1297 static inline int pud_trans_unstable(pud_t *pud)
1298 {
1299 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) &&			\
1300 	defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1301 	return pud_none_or_trans_huge_or_dev_or_clear_bad(pud);
1302 #else
1303 	return 0;
1304 #endif
1305 }
1306 
1307 #ifndef pmd_read_atomic
1308 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
1309 {
1310 	/*
1311 	 * Depend on compiler for an atomic pmd read. NOTE: this is
1312 	 * only going to work, if the pmdval_t isn't larger than
1313 	 * an unsigned long.
1314 	 */
1315 	return *pmdp;
1316 }
1317 #endif
1318 
1319 #ifndef arch_needs_pgtable_deposit
1320 #define arch_needs_pgtable_deposit() (false)
1321 #endif
1322 /*
1323  * This function is meant to be used by sites walking pagetables with
1324  * the mmap_lock held in read mode to protect against MADV_DONTNEED and
1325  * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
1326  * into a null pmd and the transhuge page fault can convert a null pmd
1327  * into an hugepmd or into a regular pmd (if the hugepage allocation
1328  * fails). While holding the mmap_lock in read mode the pmd becomes
1329  * stable and stops changing under us only if it's not null and not a
1330  * transhuge pmd. When those races occurs and this function makes a
1331  * difference vs the standard pmd_none_or_clear_bad, the result is
1332  * undefined so behaving like if the pmd was none is safe (because it
1333  * can return none anyway). The compiler level barrier() is critically
1334  * important to compute the two checks atomically on the same pmdval.
1335  *
1336  * For 32bit kernels with a 64bit large pmd_t this automatically takes
1337  * care of reading the pmd atomically to avoid SMP race conditions
1338  * against pmd_populate() when the mmap_lock is hold for reading by the
1339  * caller (a special atomic read not done by "gcc" as in the generic
1340  * version above, is also needed when THP is disabled because the page
1341  * fault can populate the pmd from under us).
1342  */
1343 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
1344 {
1345 	pmd_t pmdval = pmd_read_atomic(pmd);
1346 	/*
1347 	 * The barrier will stabilize the pmdval in a register or on
1348 	 * the stack so that it will stop changing under the code.
1349 	 *
1350 	 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
1351 	 * pmd_read_atomic is allowed to return a not atomic pmdval
1352 	 * (for example pointing to an hugepage that has never been
1353 	 * mapped in the pmd). The below checks will only care about
1354 	 * the low part of the pmd with 32bit PAE x86 anyway, with the
1355 	 * exception of pmd_none(). So the important thing is that if
1356 	 * the low part of the pmd is found null, the high part will
1357 	 * be also null or the pmd_none() check below would be
1358 	 * confused.
1359 	 */
1360 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1361 	barrier();
1362 #endif
1363 	/*
1364 	 * !pmd_present() checks for pmd migration entries
1365 	 *
1366 	 * The complete check uses is_pmd_migration_entry() in linux/swapops.h
1367 	 * But using that requires moving current function and pmd_trans_unstable()
1368 	 * to linux/swapops.h to resolve dependency, which is too much code move.
1369 	 *
1370 	 * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
1371 	 * because !pmd_present() pages can only be under migration not swapped
1372 	 * out.
1373 	 *
1374 	 * pmd_none() is preserved for future condition checks on pmd migration
1375 	 * entries and not confusing with this function name, although it is
1376 	 * redundant with !pmd_present().
1377 	 */
1378 	if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
1379 		(IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
1380 		return 1;
1381 	if (unlikely(pmd_bad(pmdval))) {
1382 		pmd_clear_bad(pmd);
1383 		return 1;
1384 	}
1385 	return 0;
1386 }
1387 
1388 /*
1389  * This is a noop if Transparent Hugepage Support is not built into
1390  * the kernel. Otherwise it is equivalent to
1391  * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
1392  * places that already verified the pmd is not none and they want to
1393  * walk ptes while holding the mmap sem in read mode (write mode don't
1394  * need this). If THP is not enabled, the pmd can't go away under the
1395  * code even if MADV_DONTNEED runs, but if THP is enabled we need to
1396  * run a pmd_trans_unstable before walking the ptes after
1397  * split_huge_pmd returns (because it may have run when the pmd become
1398  * null, but then a page fault can map in a THP and not a regular page).
1399  */
1400 static inline int pmd_trans_unstable(pmd_t *pmd)
1401 {
1402 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1403 	return pmd_none_or_trans_huge_or_clear_bad(pmd);
1404 #else
1405 	return 0;
1406 #endif
1407 }
1408 
1409 /*
1410  * the ordering of these checks is important for pmds with _page_devmap set.
1411  * if we check pmd_trans_unstable() first we will trip the bad_pmd() check
1412  * inside of pmd_none_or_trans_huge_or_clear_bad(). this will end up correctly
1413  * returning 1 but not before it spams dmesg with the pmd_clear_bad() output.
1414  */
1415 static inline int pmd_devmap_trans_unstable(pmd_t *pmd)
1416 {
1417 	return pmd_devmap(*pmd) || pmd_trans_unstable(pmd);
1418 }
1419 
1420 #ifndef CONFIG_NUMA_BALANCING
1421 /*
1422  * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
1423  * the only case the kernel cares is for NUMA balancing and is only ever set
1424  * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
1425  * _PAGE_PROTNONE so by default, implement the helper as "always no". It
1426  * is the responsibility of the caller to distinguish between PROT_NONE
1427  * protections and NUMA hinting fault protections.
1428  */
1429 static inline int pte_protnone(pte_t pte)
1430 {
1431 	return 0;
1432 }
1433 
1434 static inline int pmd_protnone(pmd_t pmd)
1435 {
1436 	return 0;
1437 }
1438 #endif /* CONFIG_NUMA_BALANCING */
1439 
1440 #endif /* CONFIG_MMU */
1441 
1442 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1443 
1444 #ifndef __PAGETABLE_P4D_FOLDED
1445 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1446 int p4d_clear_huge(p4d_t *p4d);
1447 #else
1448 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1449 {
1450 	return 0;
1451 }
1452 static inline int p4d_clear_huge(p4d_t *p4d)
1453 {
1454 	return 0;
1455 }
1456 #endif /* !__PAGETABLE_P4D_FOLDED */
1457 
1458 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1459 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1460 int pud_clear_huge(pud_t *pud);
1461 int pmd_clear_huge(pmd_t *pmd);
1462 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1463 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1464 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1465 #else	/* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1466 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1467 {
1468 	return 0;
1469 }
1470 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1471 {
1472 	return 0;
1473 }
1474 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1475 {
1476 	return 0;
1477 }
1478 static inline int p4d_clear_huge(p4d_t *p4d)
1479 {
1480 	return 0;
1481 }
1482 static inline int pud_clear_huge(pud_t *pud)
1483 {
1484 	return 0;
1485 }
1486 static inline int pmd_clear_huge(pmd_t *pmd)
1487 {
1488 	return 0;
1489 }
1490 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1491 {
1492 	return 0;
1493 }
1494 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1495 {
1496 	return 0;
1497 }
1498 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1499 {
1500 	return 0;
1501 }
1502 #endif	/* CONFIG_HAVE_ARCH_HUGE_VMAP */
1503 
1504 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1505 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1506 /*
1507  * ARCHes with special requirements for evicting THP backing TLB entries can
1508  * implement this. Otherwise also, it can help optimize normal TLB flush in
1509  * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1510  * entire TLB if flush span is greater than a threshold, which will
1511  * likely be true for a single huge page. Thus a single THP flush will
1512  * invalidate the entire TLB which is not desirable.
1513  * e.g. see arch/arc: flush_pmd_tlb_range
1514  */
1515 #define flush_pmd_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1516 #define flush_pud_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1517 #else
1518 #define flush_pmd_tlb_range(vma, addr, end)	BUILD_BUG()
1519 #define flush_pud_tlb_range(vma, addr, end)	BUILD_BUG()
1520 #endif
1521 #endif
1522 
1523 struct file;
1524 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1525 			unsigned long size, pgprot_t *vma_prot);
1526 
1527 #ifndef CONFIG_X86_ESPFIX64
1528 static inline void init_espfix_bsp(void) { }
1529 #endif
1530 
1531 extern void __init pgtable_cache_init(void);
1532 
1533 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1534 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1535 {
1536 	return true;
1537 }
1538 
1539 static inline bool arch_has_pfn_modify_check(void)
1540 {
1541 	return false;
1542 }
1543 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1544 
1545 /*
1546  * Architecture PAGE_KERNEL_* fallbacks
1547  *
1548  * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1549  * because they really don't support them, or the port needs to be updated to
1550  * reflect the required functionality. Below are a set of relatively safe
1551  * fallbacks, as best effort, which we can count on in lieu of the architectures
1552  * not defining them on their own yet.
1553  */
1554 
1555 #ifndef PAGE_KERNEL_RO
1556 # define PAGE_KERNEL_RO PAGE_KERNEL
1557 #endif
1558 
1559 #ifndef PAGE_KERNEL_EXEC
1560 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1561 #endif
1562 
1563 /*
1564  * Page Table Modification bits for pgtbl_mod_mask.
1565  *
1566  * These are used by the p?d_alloc_track*() set of functions an in the generic
1567  * vmalloc/ioremap code to track at which page-table levels entries have been
1568  * modified. Based on that the code can better decide when vmalloc and ioremap
1569  * mapping changes need to be synchronized to other page-tables in the system.
1570  */
1571 #define		__PGTBL_PGD_MODIFIED	0
1572 #define		__PGTBL_P4D_MODIFIED	1
1573 #define		__PGTBL_PUD_MODIFIED	2
1574 #define		__PGTBL_PMD_MODIFIED	3
1575 #define		__PGTBL_PTE_MODIFIED	4
1576 
1577 #define		PGTBL_PGD_MODIFIED	BIT(__PGTBL_PGD_MODIFIED)
1578 #define		PGTBL_P4D_MODIFIED	BIT(__PGTBL_P4D_MODIFIED)
1579 #define		PGTBL_PUD_MODIFIED	BIT(__PGTBL_PUD_MODIFIED)
1580 #define		PGTBL_PMD_MODIFIED	BIT(__PGTBL_PMD_MODIFIED)
1581 #define		PGTBL_PTE_MODIFIED	BIT(__PGTBL_PTE_MODIFIED)
1582 
1583 /* Page-Table Modification Mask */
1584 typedef unsigned int pgtbl_mod_mask;
1585 
1586 #endif /* !__ASSEMBLY__ */
1587 
1588 #if !defined(MAX_POSSIBLE_PHYSMEM_BITS) && !defined(CONFIG_64BIT)
1589 #ifdef CONFIG_PHYS_ADDR_T_64BIT
1590 /*
1591  * ZSMALLOC needs to know the highest PFN on 32-bit architectures
1592  * with physical address space extension, but falls back to
1593  * BITS_PER_LONG otherwise.
1594  */
1595 #error Missing MAX_POSSIBLE_PHYSMEM_BITS definition
1596 #else
1597 #define MAX_POSSIBLE_PHYSMEM_BITS 32
1598 #endif
1599 #endif
1600 
1601 #ifndef has_transparent_hugepage
1602 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1603 #define has_transparent_hugepage() 1
1604 #else
1605 #define has_transparent_hugepage() 0
1606 #endif
1607 #endif
1608 
1609 /*
1610  * On some architectures it depends on the mm if the p4d/pud or pmd
1611  * layer of the page table hierarchy is folded or not.
1612  */
1613 #ifndef mm_p4d_folded
1614 #define mm_p4d_folded(mm)	__is_defined(__PAGETABLE_P4D_FOLDED)
1615 #endif
1616 
1617 #ifndef mm_pud_folded
1618 #define mm_pud_folded(mm)	__is_defined(__PAGETABLE_PUD_FOLDED)
1619 #endif
1620 
1621 #ifndef mm_pmd_folded
1622 #define mm_pmd_folded(mm)	__is_defined(__PAGETABLE_PMD_FOLDED)
1623 #endif
1624 
1625 #ifndef p4d_offset_lockless
1626 #define p4d_offset_lockless(pgdp, pgd, address) p4d_offset(&(pgd), address)
1627 #endif
1628 #ifndef pud_offset_lockless
1629 #define pud_offset_lockless(p4dp, p4d, address) pud_offset(&(p4d), address)
1630 #endif
1631 #ifndef pmd_offset_lockless
1632 #define pmd_offset_lockless(pudp, pud, address) pmd_offset(&(pud), address)
1633 #endif
1634 
1635 /*
1636  * p?d_leaf() - true if this entry is a final mapping to a physical address.
1637  * This differs from p?d_huge() by the fact that they are always available (if
1638  * the architecture supports large pages at the appropriate level) even
1639  * if CONFIG_HUGETLB_PAGE is not defined.
1640  * Only meaningful when called on a valid entry.
1641  */
1642 #ifndef pgd_leaf
1643 #define pgd_leaf(x)	0
1644 #endif
1645 #ifndef p4d_leaf
1646 #define p4d_leaf(x)	0
1647 #endif
1648 #ifndef pud_leaf
1649 #define pud_leaf(x)	0
1650 #endif
1651 #ifndef pmd_leaf
1652 #define pmd_leaf(x)	0
1653 #endif
1654 
1655 #ifndef pgd_leaf_size
1656 #define pgd_leaf_size(x) (1ULL << PGDIR_SHIFT)
1657 #endif
1658 #ifndef p4d_leaf_size
1659 #define p4d_leaf_size(x) P4D_SIZE
1660 #endif
1661 #ifndef pud_leaf_size
1662 #define pud_leaf_size(x) PUD_SIZE
1663 #endif
1664 #ifndef pmd_leaf_size
1665 #define pmd_leaf_size(x) PMD_SIZE
1666 #endif
1667 #ifndef pte_leaf_size
1668 #define pte_leaf_size(x) PAGE_SIZE
1669 #endif
1670 
1671 /*
1672  * Some architectures have MMUs that are configurable or selectable at boot
1673  * time. These lead to variable PTRS_PER_x. For statically allocated arrays it
1674  * helps to have a static maximum value.
1675  */
1676 
1677 #ifndef MAX_PTRS_PER_PTE
1678 #define MAX_PTRS_PER_PTE PTRS_PER_PTE
1679 #endif
1680 
1681 #ifndef MAX_PTRS_PER_PMD
1682 #define MAX_PTRS_PER_PMD PTRS_PER_PMD
1683 #endif
1684 
1685 #ifndef MAX_PTRS_PER_PUD
1686 #define MAX_PTRS_PER_PUD PTRS_PER_PUD
1687 #endif
1688 
1689 #ifndef MAX_PTRS_PER_P4D
1690 #define MAX_PTRS_PER_P4D PTRS_PER_P4D
1691 #endif
1692 
1693 #endif /* _LINUX_PGTABLE_H */
1694