xref: /linux-6.15/include/linux/pgtable.h (revision fb23bf6b)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_PGTABLE_H
3 #define _LINUX_PGTABLE_H
4 
5 #include <linux/pfn.h>
6 #include <asm/pgtable.h>
7 
8 #define PMD_ORDER	(PMD_SHIFT - PAGE_SHIFT)
9 #define PUD_ORDER	(PUD_SHIFT - PAGE_SHIFT)
10 
11 #ifndef __ASSEMBLY__
12 #ifdef CONFIG_MMU
13 
14 #include <linux/mm_types.h>
15 #include <linux/bug.h>
16 #include <linux/errno.h>
17 #include <asm-generic/pgtable_uffd.h>
18 #include <linux/page_table_check.h>
19 
20 #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
21 	defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
22 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
23 #endif
24 
25 /*
26  * On almost all architectures and configurations, 0 can be used as the
27  * upper ceiling to free_pgtables(): on many architectures it has the same
28  * effect as using TASK_SIZE.  However, there is one configuration which
29  * must impose a more careful limit, to avoid freeing kernel pgtables.
30  */
31 #ifndef USER_PGTABLES_CEILING
32 #define USER_PGTABLES_CEILING	0UL
33 #endif
34 
35 /*
36  * This defines the first usable user address. Platforms
37  * can override its value with custom FIRST_USER_ADDRESS
38  * defined in their respective <asm/pgtable.h>.
39  */
40 #ifndef FIRST_USER_ADDRESS
41 #define FIRST_USER_ADDRESS	0UL
42 #endif
43 
44 /*
45  * This defines the generic helper for accessing PMD page
46  * table page. Although platforms can still override this
47  * via their respective <asm/pgtable.h>.
48  */
49 #ifndef pmd_pgtable
50 #define pmd_pgtable(pmd) pmd_page(pmd)
51 #endif
52 
53 /*
54  * A page table page can be thought of an array like this: pXd_t[PTRS_PER_PxD]
55  *
56  * The pXx_index() functions return the index of the entry in the page
57  * table page which would control the given virtual address
58  *
59  * As these functions may be used by the same code for different levels of
60  * the page table folding, they are always available, regardless of
61  * CONFIG_PGTABLE_LEVELS value. For the folded levels they simply return 0
62  * because in such cases PTRS_PER_PxD equals 1.
63  */
64 
65 static inline unsigned long pte_index(unsigned long address)
66 {
67 	return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
68 }
69 
70 #ifndef pmd_index
71 static inline unsigned long pmd_index(unsigned long address)
72 {
73 	return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
74 }
75 #define pmd_index pmd_index
76 #endif
77 
78 #ifndef pud_index
79 static inline unsigned long pud_index(unsigned long address)
80 {
81 	return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
82 }
83 #define pud_index pud_index
84 #endif
85 
86 #ifndef pgd_index
87 /* Must be a compile-time constant, so implement it as a macro */
88 #define pgd_index(a)  (((a) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
89 #endif
90 
91 #ifndef pte_offset_kernel
92 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
93 {
94 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
95 }
96 #define pte_offset_kernel pte_offset_kernel
97 #endif
98 
99 #ifdef CONFIG_HIGHPTE
100 #define __pte_map(pmd, address) \
101 	((pte_t *)kmap_local_page(pmd_page(*(pmd))) + pte_index((address)))
102 #define pte_unmap(pte)	do {	\
103 	kunmap_local((pte));	\
104 	rcu_read_unlock();	\
105 } while (0)
106 #else
107 static inline pte_t *__pte_map(pmd_t *pmd, unsigned long address)
108 {
109 	return pte_offset_kernel(pmd, address);
110 }
111 static inline void pte_unmap(pte_t *pte)
112 {
113 	rcu_read_unlock();
114 }
115 #endif
116 
117 void pte_free_defer(struct mm_struct *mm, pgtable_t pgtable);
118 
119 /* Find an entry in the second-level page table.. */
120 #ifndef pmd_offset
121 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
122 {
123 	return pud_pgtable(*pud) + pmd_index(address);
124 }
125 #define pmd_offset pmd_offset
126 #endif
127 
128 #ifndef pud_offset
129 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
130 {
131 	return p4d_pgtable(*p4d) + pud_index(address);
132 }
133 #define pud_offset pud_offset
134 #endif
135 
136 static inline pgd_t *pgd_offset_pgd(pgd_t *pgd, unsigned long address)
137 {
138 	return (pgd + pgd_index(address));
139 };
140 
141 /*
142  * a shortcut to get a pgd_t in a given mm
143  */
144 #ifndef pgd_offset
145 #define pgd_offset(mm, address)		pgd_offset_pgd((mm)->pgd, (address))
146 #endif
147 
148 /*
149  * a shortcut which implies the use of the kernel's pgd, instead
150  * of a process's
151  */
152 #ifndef pgd_offset_k
153 #define pgd_offset_k(address)		pgd_offset(&init_mm, (address))
154 #endif
155 
156 /*
157  * In many cases it is known that a virtual address is mapped at PMD or PTE
158  * level, so instead of traversing all the page table levels, we can get a
159  * pointer to the PMD entry in user or kernel page table or translate a virtual
160  * address to the pointer in the PTE in the kernel page tables with simple
161  * helpers.
162  */
163 static inline pmd_t *pmd_off(struct mm_struct *mm, unsigned long va)
164 {
165 	return pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, va), va), va), va);
166 }
167 
168 static inline pmd_t *pmd_off_k(unsigned long va)
169 {
170 	return pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va);
171 }
172 
173 static inline pte_t *virt_to_kpte(unsigned long vaddr)
174 {
175 	pmd_t *pmd = pmd_off_k(vaddr);
176 
177 	return pmd_none(*pmd) ? NULL : pte_offset_kernel(pmd, vaddr);
178 }
179 
180 #ifndef pmd_young
181 static inline int pmd_young(pmd_t pmd)
182 {
183 	return 0;
184 }
185 #endif
186 
187 #ifndef pmd_dirty
188 static inline int pmd_dirty(pmd_t pmd)
189 {
190 	return 0;
191 }
192 #endif
193 
194 /*
195  * A facility to provide lazy MMU batching.  This allows PTE updates and
196  * page invalidations to be delayed until a call to leave lazy MMU mode
197  * is issued.  Some architectures may benefit from doing this, and it is
198  * beneficial for both shadow and direct mode hypervisors, which may batch
199  * the PTE updates which happen during this window.  Note that using this
200  * interface requires that read hazards be removed from the code.  A read
201  * hazard could result in the direct mode hypervisor case, since the actual
202  * write to the page tables may not yet have taken place, so reads though
203  * a raw PTE pointer after it has been modified are not guaranteed to be
204  * up to date.  This mode can only be entered and left under the protection of
205  * the page table locks for all page tables which may be modified.  In the UP
206  * case, this is required so that preemption is disabled, and in the SMP case,
207  * it must synchronize the delayed page table writes properly on other CPUs.
208  */
209 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
210 #define arch_enter_lazy_mmu_mode()	do {} while (0)
211 #define arch_leave_lazy_mmu_mode()	do {} while (0)
212 #define arch_flush_lazy_mmu_mode()	do {} while (0)
213 #endif
214 
215 #ifndef pte_advance_pfn
216 static inline pte_t pte_advance_pfn(pte_t pte, unsigned long nr)
217 {
218 	return __pte(pte_val(pte) + (nr << PFN_PTE_SHIFT));
219 }
220 #endif
221 
222 #define pte_next_pfn(pte) pte_advance_pfn(pte, 1)
223 
224 #ifndef set_ptes
225 /**
226  * set_ptes - Map consecutive pages to a contiguous range of addresses.
227  * @mm: Address space to map the pages into.
228  * @addr: Address to map the first page at.
229  * @ptep: Page table pointer for the first entry.
230  * @pte: Page table entry for the first page.
231  * @nr: Number of pages to map.
232  *
233  * When nr==1, initial state of pte may be present or not present, and new state
234  * may be present or not present. When nr>1, initial state of all ptes must be
235  * not present, and new state must be present.
236  *
237  * May be overridden by the architecture, or the architecture can define
238  * set_pte() and PFN_PTE_SHIFT.
239  *
240  * Context: The caller holds the page table lock.  The pages all belong
241  * to the same folio.  The PTEs are all in the same PMD.
242  */
243 static inline void set_ptes(struct mm_struct *mm, unsigned long addr,
244 		pte_t *ptep, pte_t pte, unsigned int nr)
245 {
246 	page_table_check_ptes_set(mm, ptep, pte, nr);
247 
248 	arch_enter_lazy_mmu_mode();
249 	for (;;) {
250 		set_pte(ptep, pte);
251 		if (--nr == 0)
252 			break;
253 		ptep++;
254 		pte = pte_next_pfn(pte);
255 	}
256 	arch_leave_lazy_mmu_mode();
257 }
258 #endif
259 #define set_pte_at(mm, addr, ptep, pte) set_ptes(mm, addr, ptep, pte, 1)
260 
261 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
262 extern int ptep_set_access_flags(struct vm_area_struct *vma,
263 				 unsigned long address, pte_t *ptep,
264 				 pte_t entry, int dirty);
265 #endif
266 
267 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
268 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
269 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
270 				 unsigned long address, pmd_t *pmdp,
271 				 pmd_t entry, int dirty);
272 extern int pudp_set_access_flags(struct vm_area_struct *vma,
273 				 unsigned long address, pud_t *pudp,
274 				 pud_t entry, int dirty);
275 #else
276 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
277 					unsigned long address, pmd_t *pmdp,
278 					pmd_t entry, int dirty)
279 {
280 	BUILD_BUG();
281 	return 0;
282 }
283 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
284 					unsigned long address, pud_t *pudp,
285 					pud_t entry, int dirty)
286 {
287 	BUILD_BUG();
288 	return 0;
289 }
290 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
291 #endif
292 
293 #ifndef ptep_get
294 static inline pte_t ptep_get(pte_t *ptep)
295 {
296 	return READ_ONCE(*ptep);
297 }
298 #endif
299 
300 #ifndef pmdp_get
301 static inline pmd_t pmdp_get(pmd_t *pmdp)
302 {
303 	return READ_ONCE(*pmdp);
304 }
305 #endif
306 
307 #ifndef pudp_get
308 static inline pud_t pudp_get(pud_t *pudp)
309 {
310 	return READ_ONCE(*pudp);
311 }
312 #endif
313 
314 #ifndef p4dp_get
315 static inline p4d_t p4dp_get(p4d_t *p4dp)
316 {
317 	return READ_ONCE(*p4dp);
318 }
319 #endif
320 
321 #ifndef pgdp_get
322 static inline pgd_t pgdp_get(pgd_t *pgdp)
323 {
324 	return READ_ONCE(*pgdp);
325 }
326 #endif
327 
328 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
329 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
330 					    unsigned long address,
331 					    pte_t *ptep)
332 {
333 	pte_t pte = ptep_get(ptep);
334 	int r = 1;
335 	if (!pte_young(pte))
336 		r = 0;
337 	else
338 		set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
339 	return r;
340 }
341 #endif
342 
343 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
344 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG)
345 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
346 					    unsigned long address,
347 					    pmd_t *pmdp)
348 {
349 	pmd_t pmd = *pmdp;
350 	int r = 1;
351 	if (!pmd_young(pmd))
352 		r = 0;
353 	else
354 		set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
355 	return r;
356 }
357 #else
358 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
359 					    unsigned long address,
360 					    pmd_t *pmdp)
361 {
362 	BUILD_BUG();
363 	return 0;
364 }
365 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG */
366 #endif
367 
368 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
369 int ptep_clear_flush_young(struct vm_area_struct *vma,
370 			   unsigned long address, pte_t *ptep);
371 #endif
372 
373 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
374 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
375 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
376 				  unsigned long address, pmd_t *pmdp);
377 #else
378 /*
379  * Despite relevant to THP only, this API is called from generic rmap code
380  * under PageTransHuge(), hence needs a dummy implementation for !THP
381  */
382 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
383 					 unsigned long address, pmd_t *pmdp)
384 {
385 	BUILD_BUG();
386 	return 0;
387 }
388 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
389 #endif
390 
391 #ifndef arch_has_hw_nonleaf_pmd_young
392 /*
393  * Return whether the accessed bit in non-leaf PMD entries is supported on the
394  * local CPU.
395  */
396 static inline bool arch_has_hw_nonleaf_pmd_young(void)
397 {
398 	return IS_ENABLED(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG);
399 }
400 #endif
401 
402 #ifndef arch_has_hw_pte_young
403 /*
404  * Return whether the accessed bit is supported on the local CPU.
405  *
406  * This stub assumes accessing through an old PTE triggers a page fault.
407  * Architectures that automatically set the access bit should overwrite it.
408  */
409 static inline bool arch_has_hw_pte_young(void)
410 {
411 	return IS_ENABLED(CONFIG_ARCH_HAS_HW_PTE_YOUNG);
412 }
413 #endif
414 
415 #ifndef arch_check_zapped_pte
416 static inline void arch_check_zapped_pte(struct vm_area_struct *vma,
417 					 pte_t pte)
418 {
419 }
420 #endif
421 
422 #ifndef arch_check_zapped_pmd
423 static inline void arch_check_zapped_pmd(struct vm_area_struct *vma,
424 					 pmd_t pmd)
425 {
426 }
427 #endif
428 
429 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
430 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
431 				       unsigned long address,
432 				       pte_t *ptep)
433 {
434 	pte_t pte = ptep_get(ptep);
435 	pte_clear(mm, address, ptep);
436 	page_table_check_pte_clear(mm, pte);
437 	return pte;
438 }
439 #endif
440 
441 static inline void ptep_clear(struct mm_struct *mm, unsigned long addr,
442 			      pte_t *ptep)
443 {
444 	ptep_get_and_clear(mm, addr, ptep);
445 }
446 
447 #ifdef CONFIG_GUP_GET_PXX_LOW_HIGH
448 /*
449  * For walking the pagetables without holding any locks.  Some architectures
450  * (eg x86-32 PAE) cannot load the entries atomically without using expensive
451  * instructions.  We are guaranteed that a PTE will only either go from not
452  * present to present, or present to not present -- it will not switch to a
453  * completely different present page without a TLB flush inbetween; which we
454  * are blocking by holding interrupts off.
455  *
456  * Setting ptes from not present to present goes:
457  *
458  *   ptep->pte_high = h;
459  *   smp_wmb();
460  *   ptep->pte_low = l;
461  *
462  * And present to not present goes:
463  *
464  *   ptep->pte_low = 0;
465  *   smp_wmb();
466  *   ptep->pte_high = 0;
467  *
468  * We must ensure here that the load of pte_low sees 'l' IFF pte_high sees 'h'.
469  * We load pte_high *after* loading pte_low, which ensures we don't see an older
470  * value of pte_high.  *Then* we recheck pte_low, which ensures that we haven't
471  * picked up a changed pte high. We might have gotten rubbish values from
472  * pte_low and pte_high, but we are guaranteed that pte_low will not have the
473  * present bit set *unless* it is 'l'. Because get_user_pages_fast() only
474  * operates on present ptes we're safe.
475  */
476 static inline pte_t ptep_get_lockless(pte_t *ptep)
477 {
478 	pte_t pte;
479 
480 	do {
481 		pte.pte_low = ptep->pte_low;
482 		smp_rmb();
483 		pte.pte_high = ptep->pte_high;
484 		smp_rmb();
485 	} while (unlikely(pte.pte_low != ptep->pte_low));
486 
487 	return pte;
488 }
489 #define ptep_get_lockless ptep_get_lockless
490 
491 #if CONFIG_PGTABLE_LEVELS > 2
492 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
493 {
494 	pmd_t pmd;
495 
496 	do {
497 		pmd.pmd_low = pmdp->pmd_low;
498 		smp_rmb();
499 		pmd.pmd_high = pmdp->pmd_high;
500 		smp_rmb();
501 	} while (unlikely(pmd.pmd_low != pmdp->pmd_low));
502 
503 	return pmd;
504 }
505 #define pmdp_get_lockless pmdp_get_lockless
506 #define pmdp_get_lockless_sync() tlb_remove_table_sync_one()
507 #endif /* CONFIG_PGTABLE_LEVELS > 2 */
508 #endif /* CONFIG_GUP_GET_PXX_LOW_HIGH */
509 
510 /*
511  * We require that the PTE can be read atomically.
512  */
513 #ifndef ptep_get_lockless
514 static inline pte_t ptep_get_lockless(pte_t *ptep)
515 {
516 	return ptep_get(ptep);
517 }
518 #endif
519 
520 #ifndef pmdp_get_lockless
521 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
522 {
523 	return pmdp_get(pmdp);
524 }
525 static inline void pmdp_get_lockless_sync(void)
526 {
527 }
528 #endif
529 
530 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
531 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
532 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
533 					    unsigned long address,
534 					    pmd_t *pmdp)
535 {
536 	pmd_t pmd = *pmdp;
537 
538 	pmd_clear(pmdp);
539 	page_table_check_pmd_clear(mm, pmd);
540 
541 	return pmd;
542 }
543 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
544 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
545 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
546 					    unsigned long address,
547 					    pud_t *pudp)
548 {
549 	pud_t pud = *pudp;
550 
551 	pud_clear(pudp);
552 	page_table_check_pud_clear(mm, pud);
553 
554 	return pud;
555 }
556 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
557 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
558 
559 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
560 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
561 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
562 					    unsigned long address, pmd_t *pmdp,
563 					    int full)
564 {
565 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
566 }
567 #endif
568 
569 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
570 static inline pud_t pudp_huge_get_and_clear_full(struct vm_area_struct *vma,
571 					    unsigned long address, pud_t *pudp,
572 					    int full)
573 {
574 	return pudp_huge_get_and_clear(vma->vm_mm, address, pudp);
575 }
576 #endif
577 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
578 
579 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
580 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
581 					    unsigned long address, pte_t *ptep,
582 					    int full)
583 {
584 	return ptep_get_and_clear(mm, address, ptep);
585 }
586 #endif
587 
588 #ifndef get_and_clear_full_ptes
589 /**
590  * get_and_clear_full_ptes - Clear present PTEs that map consecutive pages of
591  *			     the same folio, collecting dirty/accessed bits.
592  * @mm: Address space the pages are mapped into.
593  * @addr: Address the first page is mapped at.
594  * @ptep: Page table pointer for the first entry.
595  * @nr: Number of entries to clear.
596  * @full: Whether we are clearing a full mm.
597  *
598  * May be overridden by the architecture; otherwise, implemented as a simple
599  * loop over ptep_get_and_clear_full(), merging dirty/accessed bits into the
600  * returned PTE.
601  *
602  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
603  * some PTEs might be write-protected.
604  *
605  * Context: The caller holds the page table lock.  The PTEs map consecutive
606  * pages that belong to the same folio.  The PTEs are all in the same PMD.
607  */
608 static inline pte_t get_and_clear_full_ptes(struct mm_struct *mm,
609 		unsigned long addr, pte_t *ptep, unsigned int nr, int full)
610 {
611 	pte_t pte, tmp_pte;
612 
613 	pte = ptep_get_and_clear_full(mm, addr, ptep, full);
614 	while (--nr) {
615 		ptep++;
616 		addr += PAGE_SIZE;
617 		tmp_pte = ptep_get_and_clear_full(mm, addr, ptep, full);
618 		if (pte_dirty(tmp_pte))
619 			pte = pte_mkdirty(pte);
620 		if (pte_young(tmp_pte))
621 			pte = pte_mkyoung(pte);
622 	}
623 	return pte;
624 }
625 #endif
626 
627 #ifndef clear_full_ptes
628 /**
629  * clear_full_ptes - Clear present PTEs that map consecutive pages of the same
630  *		     folio.
631  * @mm: Address space the pages are mapped into.
632  * @addr: Address the first page is mapped at.
633  * @ptep: Page table pointer for the first entry.
634  * @nr: Number of entries to clear.
635  * @full: Whether we are clearing a full mm.
636  *
637  * May be overridden by the architecture; otherwise, implemented as a simple
638  * loop over ptep_get_and_clear_full().
639  *
640  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
641  * some PTEs might be write-protected.
642  *
643  * Context: The caller holds the page table lock.  The PTEs map consecutive
644  * pages that belong to the same folio.  The PTEs are all in the same PMD.
645  */
646 static inline void clear_full_ptes(struct mm_struct *mm, unsigned long addr,
647 		pte_t *ptep, unsigned int nr, int full)
648 {
649 	for (;;) {
650 		ptep_get_and_clear_full(mm, addr, ptep, full);
651 		if (--nr == 0)
652 			break;
653 		ptep++;
654 		addr += PAGE_SIZE;
655 	}
656 }
657 #endif
658 
659 /*
660  * If two threads concurrently fault at the same page, the thread that
661  * won the race updates the PTE and its local TLB/Cache. The other thread
662  * gives up, simply does nothing, and continues; on architectures where
663  * software can update TLB,  local TLB can be updated here to avoid next page
664  * fault. This function updates TLB only, do nothing with cache or others.
665  * It is the difference with function update_mmu_cache.
666  */
667 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
668 static inline void update_mmu_tlb(struct vm_area_struct *vma,
669 				unsigned long address, pte_t *ptep)
670 {
671 }
672 #define __HAVE_ARCH_UPDATE_MMU_TLB
673 #endif
674 
675 /*
676  * Some architectures may be able to avoid expensive synchronization
677  * primitives when modifications are made to PTE's which are already
678  * not present, or in the process of an address space destruction.
679  */
680 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
681 static inline void pte_clear_not_present_full(struct mm_struct *mm,
682 					      unsigned long address,
683 					      pte_t *ptep,
684 					      int full)
685 {
686 	pte_clear(mm, address, ptep);
687 }
688 #endif
689 
690 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
691 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
692 			      unsigned long address,
693 			      pte_t *ptep);
694 #endif
695 
696 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
697 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
698 			      unsigned long address,
699 			      pmd_t *pmdp);
700 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
701 			      unsigned long address,
702 			      pud_t *pudp);
703 #endif
704 
705 #ifndef pte_mkwrite
706 static inline pte_t pte_mkwrite(pte_t pte, struct vm_area_struct *vma)
707 {
708 	return pte_mkwrite_novma(pte);
709 }
710 #endif
711 
712 #if defined(CONFIG_ARCH_WANT_PMD_MKWRITE) && !defined(pmd_mkwrite)
713 static inline pmd_t pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma)
714 {
715 	return pmd_mkwrite_novma(pmd);
716 }
717 #endif
718 
719 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
720 struct mm_struct;
721 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
722 {
723 	pte_t old_pte = ptep_get(ptep);
724 	set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
725 }
726 #endif
727 
728 #ifndef wrprotect_ptes
729 /**
730  * wrprotect_ptes - Write-protect PTEs that map consecutive pages of the same
731  *		    folio.
732  * @mm: Address space the pages are mapped into.
733  * @addr: Address the first page is mapped at.
734  * @ptep: Page table pointer for the first entry.
735  * @nr: Number of entries to write-protect.
736  *
737  * May be overridden by the architecture; otherwise, implemented as a simple
738  * loop over ptep_set_wrprotect().
739  *
740  * Note that PTE bits in the PTE range besides the PFN can differ. For example,
741  * some PTEs might be write-protected.
742  *
743  * Context: The caller holds the page table lock.  The PTEs map consecutive
744  * pages that belong to the same folio.  The PTEs are all in the same PMD.
745  */
746 static inline void wrprotect_ptes(struct mm_struct *mm, unsigned long addr,
747 		pte_t *ptep, unsigned int nr)
748 {
749 	for (;;) {
750 		ptep_set_wrprotect(mm, addr, ptep);
751 		if (--nr == 0)
752 			break;
753 		ptep++;
754 		addr += PAGE_SIZE;
755 	}
756 }
757 #endif
758 
759 /*
760  * On some architectures hardware does not set page access bit when accessing
761  * memory page, it is responsibility of software setting this bit. It brings
762  * out extra page fault penalty to track page access bit. For optimization page
763  * access bit can be set during all page fault flow on these arches.
764  * To be differentiate with macro pte_mkyoung, this macro is used on platforms
765  * where software maintains page access bit.
766  */
767 #ifndef pte_sw_mkyoung
768 static inline pte_t pte_sw_mkyoung(pte_t pte)
769 {
770 	return pte;
771 }
772 #define pte_sw_mkyoung	pte_sw_mkyoung
773 #endif
774 
775 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
776 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
777 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
778 				      unsigned long address, pmd_t *pmdp)
779 {
780 	pmd_t old_pmd = *pmdp;
781 	set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
782 }
783 #else
784 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
785 				      unsigned long address, pmd_t *pmdp)
786 {
787 	BUILD_BUG();
788 }
789 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
790 #endif
791 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
792 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
793 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
794 static inline void pudp_set_wrprotect(struct mm_struct *mm,
795 				      unsigned long address, pud_t *pudp)
796 {
797 	pud_t old_pud = *pudp;
798 
799 	set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
800 }
801 #else
802 static inline void pudp_set_wrprotect(struct mm_struct *mm,
803 				      unsigned long address, pud_t *pudp)
804 {
805 	BUILD_BUG();
806 }
807 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
808 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
809 #endif
810 
811 #ifndef pmdp_collapse_flush
812 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
813 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
814 				 unsigned long address, pmd_t *pmdp);
815 #else
816 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
817 					unsigned long address,
818 					pmd_t *pmdp)
819 {
820 	BUILD_BUG();
821 	return *pmdp;
822 }
823 #define pmdp_collapse_flush pmdp_collapse_flush
824 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
825 #endif
826 
827 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
828 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
829 				       pgtable_t pgtable);
830 #endif
831 
832 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
833 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
834 #endif
835 
836 #ifndef arch_needs_pgtable_deposit
837 #define arch_needs_pgtable_deposit() (false)
838 #endif
839 
840 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
841 /*
842  * This is an implementation of pmdp_establish() that is only suitable for an
843  * architecture that doesn't have hardware dirty/accessed bits. In this case we
844  * can't race with CPU which sets these bits and non-atomic approach is fine.
845  */
846 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
847 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
848 {
849 	pmd_t old_pmd = *pmdp;
850 	set_pmd_at(vma->vm_mm, address, pmdp, pmd);
851 	return old_pmd;
852 }
853 #endif
854 
855 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
856 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
857 			    pmd_t *pmdp);
858 #endif
859 
860 #ifndef __HAVE_ARCH_PMDP_INVALIDATE_AD
861 
862 /*
863  * pmdp_invalidate_ad() invalidates the PMD while changing a transparent
864  * hugepage mapping in the page tables. This function is similar to
865  * pmdp_invalidate(), but should only be used if the access and dirty bits would
866  * not be cleared by the software in the new PMD value. The function ensures
867  * that hardware changes of the access and dirty bits updates would not be lost.
868  *
869  * Doing so can allow in certain architectures to avoid a TLB flush in most
870  * cases. Yet, another TLB flush might be necessary later if the PMD update
871  * itself requires such flush (e.g., if protection was set to be stricter). Yet,
872  * even when a TLB flush is needed because of the update, the caller may be able
873  * to batch these TLB flushing operations, so fewer TLB flush operations are
874  * needed.
875  */
876 extern pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma,
877 				unsigned long address, pmd_t *pmdp);
878 #endif
879 
880 #ifndef __HAVE_ARCH_PTE_SAME
881 static inline int pte_same(pte_t pte_a, pte_t pte_b)
882 {
883 	return pte_val(pte_a) == pte_val(pte_b);
884 }
885 #endif
886 
887 #ifndef __HAVE_ARCH_PTE_UNUSED
888 /*
889  * Some architectures provide facilities to virtualization guests
890  * so that they can flag allocated pages as unused. This allows the
891  * host to transparently reclaim unused pages. This function returns
892  * whether the pte's page is unused.
893  */
894 static inline int pte_unused(pte_t pte)
895 {
896 	return 0;
897 }
898 #endif
899 
900 #ifndef pte_access_permitted
901 #define pte_access_permitted(pte, write) \
902 	(pte_present(pte) && (!(write) || pte_write(pte)))
903 #endif
904 
905 #ifndef pmd_access_permitted
906 #define pmd_access_permitted(pmd, write) \
907 	(pmd_present(pmd) && (!(write) || pmd_write(pmd)))
908 #endif
909 
910 #ifndef pud_access_permitted
911 #define pud_access_permitted(pud, write) \
912 	(pud_present(pud) && (!(write) || pud_write(pud)))
913 #endif
914 
915 #ifndef p4d_access_permitted
916 #define p4d_access_permitted(p4d, write) \
917 	(p4d_present(p4d) && (!(write) || p4d_write(p4d)))
918 #endif
919 
920 #ifndef pgd_access_permitted
921 #define pgd_access_permitted(pgd, write) \
922 	(pgd_present(pgd) && (!(write) || pgd_write(pgd)))
923 #endif
924 
925 #ifndef __HAVE_ARCH_PMD_SAME
926 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
927 {
928 	return pmd_val(pmd_a) == pmd_val(pmd_b);
929 }
930 #endif
931 
932 #ifndef pud_same
933 static inline int pud_same(pud_t pud_a, pud_t pud_b)
934 {
935 	return pud_val(pud_a) == pud_val(pud_b);
936 }
937 #define pud_same pud_same
938 #endif
939 
940 #ifndef __HAVE_ARCH_P4D_SAME
941 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
942 {
943 	return p4d_val(p4d_a) == p4d_val(p4d_b);
944 }
945 #endif
946 
947 #ifndef __HAVE_ARCH_PGD_SAME
948 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
949 {
950 	return pgd_val(pgd_a) == pgd_val(pgd_b);
951 }
952 #endif
953 
954 /*
955  * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
956  * TLB flush will be required as a result of the "set". For example, use
957  * in scenarios where it is known ahead of time that the routine is
958  * setting non-present entries, or re-setting an existing entry to the
959  * same value. Otherwise, use the typical "set" helpers and flush the
960  * TLB.
961  */
962 #define set_pte_safe(ptep, pte) \
963 ({ \
964 	WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
965 	set_pte(ptep, pte); \
966 })
967 
968 #define set_pmd_safe(pmdp, pmd) \
969 ({ \
970 	WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
971 	set_pmd(pmdp, pmd); \
972 })
973 
974 #define set_pud_safe(pudp, pud) \
975 ({ \
976 	WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
977 	set_pud(pudp, pud); \
978 })
979 
980 #define set_p4d_safe(p4dp, p4d) \
981 ({ \
982 	WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
983 	set_p4d(p4dp, p4d); \
984 })
985 
986 #define set_pgd_safe(pgdp, pgd) \
987 ({ \
988 	WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
989 	set_pgd(pgdp, pgd); \
990 })
991 
992 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
993 /*
994  * Some architectures support metadata associated with a page. When a
995  * page is being swapped out, this metadata must be saved so it can be
996  * restored when the page is swapped back in. SPARC M7 and newer
997  * processors support an ADI (Application Data Integrity) tag for the
998  * page as metadata for the page. arch_do_swap_page() can restore this
999  * metadata when a page is swapped back in.
1000  */
1001 static inline void arch_do_swap_page(struct mm_struct *mm,
1002 				     struct vm_area_struct *vma,
1003 				     unsigned long addr,
1004 				     pte_t pte, pte_t oldpte)
1005 {
1006 
1007 }
1008 #endif
1009 
1010 #ifndef __HAVE_ARCH_UNMAP_ONE
1011 /*
1012  * Some architectures support metadata associated with a page. When a
1013  * page is being swapped out, this metadata must be saved so it can be
1014  * restored when the page is swapped back in. SPARC M7 and newer
1015  * processors support an ADI (Application Data Integrity) tag for the
1016  * page as metadata for the page. arch_unmap_one() can save this
1017  * metadata on a swap-out of a page.
1018  */
1019 static inline int arch_unmap_one(struct mm_struct *mm,
1020 				  struct vm_area_struct *vma,
1021 				  unsigned long addr,
1022 				  pte_t orig_pte)
1023 {
1024 	return 0;
1025 }
1026 #endif
1027 
1028 /*
1029  * Allow architectures to preserve additional metadata associated with
1030  * swapped-out pages. The corresponding __HAVE_ARCH_SWAP_* macros and function
1031  * prototypes must be defined in the arch-specific asm/pgtable.h file.
1032  */
1033 #ifndef __HAVE_ARCH_PREPARE_TO_SWAP
1034 static inline int arch_prepare_to_swap(struct page *page)
1035 {
1036 	return 0;
1037 }
1038 #endif
1039 
1040 #ifndef __HAVE_ARCH_SWAP_INVALIDATE
1041 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
1042 {
1043 }
1044 
1045 static inline void arch_swap_invalidate_area(int type)
1046 {
1047 }
1048 #endif
1049 
1050 #ifndef __HAVE_ARCH_SWAP_RESTORE
1051 static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
1052 {
1053 }
1054 #endif
1055 
1056 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
1057 #define pgd_offset_gate(mm, addr)	pgd_offset(mm, addr)
1058 #endif
1059 
1060 #ifndef __HAVE_ARCH_MOVE_PTE
1061 #define move_pte(pte, prot, old_addr, new_addr)	(pte)
1062 #endif
1063 
1064 #ifndef pte_accessible
1065 # define pte_accessible(mm, pte)	((void)(pte), 1)
1066 #endif
1067 
1068 #ifndef flush_tlb_fix_spurious_fault
1069 #define flush_tlb_fix_spurious_fault(vma, address, ptep) flush_tlb_page(vma, address)
1070 #endif
1071 
1072 /*
1073  * When walking page tables, get the address of the next boundary,
1074  * or the end address of the range if that comes earlier.  Although no
1075  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
1076  */
1077 
1078 #define pgd_addr_end(addr, end)						\
1079 ({	unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;	\
1080 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1081 })
1082 
1083 #ifndef p4d_addr_end
1084 #define p4d_addr_end(addr, end)						\
1085 ({	unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;	\
1086 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1087 })
1088 #endif
1089 
1090 #ifndef pud_addr_end
1091 #define pud_addr_end(addr, end)						\
1092 ({	unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;	\
1093 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1094 })
1095 #endif
1096 
1097 #ifndef pmd_addr_end
1098 #define pmd_addr_end(addr, end)						\
1099 ({	unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;	\
1100 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
1101 })
1102 #endif
1103 
1104 /*
1105  * When walking page tables, we usually want to skip any p?d_none entries;
1106  * and any p?d_bad entries - reporting the error before resetting to none.
1107  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
1108  */
1109 void pgd_clear_bad(pgd_t *);
1110 
1111 #ifndef __PAGETABLE_P4D_FOLDED
1112 void p4d_clear_bad(p4d_t *);
1113 #else
1114 #define p4d_clear_bad(p4d)        do { } while (0)
1115 #endif
1116 
1117 #ifndef __PAGETABLE_PUD_FOLDED
1118 void pud_clear_bad(pud_t *);
1119 #else
1120 #define pud_clear_bad(p4d)        do { } while (0)
1121 #endif
1122 
1123 void pmd_clear_bad(pmd_t *);
1124 
1125 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
1126 {
1127 	if (pgd_none(*pgd))
1128 		return 1;
1129 	if (unlikely(pgd_bad(*pgd))) {
1130 		pgd_clear_bad(pgd);
1131 		return 1;
1132 	}
1133 	return 0;
1134 }
1135 
1136 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
1137 {
1138 	if (p4d_none(*p4d))
1139 		return 1;
1140 	if (unlikely(p4d_bad(*p4d))) {
1141 		p4d_clear_bad(p4d);
1142 		return 1;
1143 	}
1144 	return 0;
1145 }
1146 
1147 static inline int pud_none_or_clear_bad(pud_t *pud)
1148 {
1149 	if (pud_none(*pud))
1150 		return 1;
1151 	if (unlikely(pud_bad(*pud))) {
1152 		pud_clear_bad(pud);
1153 		return 1;
1154 	}
1155 	return 0;
1156 }
1157 
1158 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
1159 {
1160 	if (pmd_none(*pmd))
1161 		return 1;
1162 	if (unlikely(pmd_bad(*pmd))) {
1163 		pmd_clear_bad(pmd);
1164 		return 1;
1165 	}
1166 	return 0;
1167 }
1168 
1169 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
1170 					     unsigned long addr,
1171 					     pte_t *ptep)
1172 {
1173 	/*
1174 	 * Get the current pte state, but zero it out to make it
1175 	 * non-present, preventing the hardware from asynchronously
1176 	 * updating it.
1177 	 */
1178 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
1179 }
1180 
1181 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
1182 					     unsigned long addr,
1183 					     pte_t *ptep, pte_t pte)
1184 {
1185 	/*
1186 	 * The pte is non-present, so there's no hardware state to
1187 	 * preserve.
1188 	 */
1189 	set_pte_at(vma->vm_mm, addr, ptep, pte);
1190 }
1191 
1192 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1193 /*
1194  * Start a pte protection read-modify-write transaction, which
1195  * protects against asynchronous hardware modifications to the pte.
1196  * The intention is not to prevent the hardware from making pte
1197  * updates, but to prevent any updates it may make from being lost.
1198  *
1199  * This does not protect against other software modifications of the
1200  * pte; the appropriate pte lock must be held over the transaction.
1201  *
1202  * Note that this interface is intended to be batchable, meaning that
1203  * ptep_modify_prot_commit may not actually update the pte, but merely
1204  * queue the update to be done at some later time.  The update must be
1205  * actually committed before the pte lock is released, however.
1206  */
1207 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
1208 					   unsigned long addr,
1209 					   pte_t *ptep)
1210 {
1211 	return __ptep_modify_prot_start(vma, addr, ptep);
1212 }
1213 
1214 /*
1215  * Commit an update to a pte, leaving any hardware-controlled bits in
1216  * the PTE unmodified.
1217  */
1218 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
1219 					   unsigned long addr,
1220 					   pte_t *ptep, pte_t old_pte, pte_t pte)
1221 {
1222 	__ptep_modify_prot_commit(vma, addr, ptep, pte);
1223 }
1224 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
1225 #endif /* CONFIG_MMU */
1226 
1227 /*
1228  * No-op macros that just return the current protection value. Defined here
1229  * because these macros can be used even if CONFIG_MMU is not defined.
1230  */
1231 
1232 #ifndef pgprot_nx
1233 #define pgprot_nx(prot)	(prot)
1234 #endif
1235 
1236 #ifndef pgprot_noncached
1237 #define pgprot_noncached(prot)	(prot)
1238 #endif
1239 
1240 #ifndef pgprot_writecombine
1241 #define pgprot_writecombine pgprot_noncached
1242 #endif
1243 
1244 #ifndef pgprot_writethrough
1245 #define pgprot_writethrough pgprot_noncached
1246 #endif
1247 
1248 #ifndef pgprot_device
1249 #define pgprot_device pgprot_noncached
1250 #endif
1251 
1252 #ifndef pgprot_mhp
1253 #define pgprot_mhp(prot)	(prot)
1254 #endif
1255 
1256 #ifdef CONFIG_MMU
1257 #ifndef pgprot_modify
1258 #define pgprot_modify pgprot_modify
1259 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
1260 {
1261 	if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
1262 		newprot = pgprot_noncached(newprot);
1263 	if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
1264 		newprot = pgprot_writecombine(newprot);
1265 	if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
1266 		newprot = pgprot_device(newprot);
1267 	return newprot;
1268 }
1269 #endif
1270 #endif /* CONFIG_MMU */
1271 
1272 #ifndef pgprot_encrypted
1273 #define pgprot_encrypted(prot)	(prot)
1274 #endif
1275 
1276 #ifndef pgprot_decrypted
1277 #define pgprot_decrypted(prot)	(prot)
1278 #endif
1279 
1280 /*
1281  * A facility to provide batching of the reload of page tables and
1282  * other process state with the actual context switch code for
1283  * paravirtualized guests.  By convention, only one of the batched
1284  * update (lazy) modes (CPU, MMU) should be active at any given time,
1285  * entry should never be nested, and entry and exits should always be
1286  * paired.  This is for sanity of maintaining and reasoning about the
1287  * kernel code.  In this case, the exit (end of the context switch) is
1288  * in architecture-specific code, and so doesn't need a generic
1289  * definition.
1290  */
1291 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
1292 #define arch_start_context_switch(prev)	do {} while (0)
1293 #endif
1294 
1295 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1296 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
1297 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1298 {
1299 	return pmd;
1300 }
1301 
1302 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1303 {
1304 	return 0;
1305 }
1306 
1307 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1308 {
1309 	return pmd;
1310 }
1311 #endif
1312 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
1313 static inline int pte_soft_dirty(pte_t pte)
1314 {
1315 	return 0;
1316 }
1317 
1318 static inline int pmd_soft_dirty(pmd_t pmd)
1319 {
1320 	return 0;
1321 }
1322 
1323 static inline pte_t pte_mksoft_dirty(pte_t pte)
1324 {
1325 	return pte;
1326 }
1327 
1328 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
1329 {
1330 	return pmd;
1331 }
1332 
1333 static inline pte_t pte_clear_soft_dirty(pte_t pte)
1334 {
1335 	return pte;
1336 }
1337 
1338 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
1339 {
1340 	return pmd;
1341 }
1342 
1343 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1344 {
1345 	return pte;
1346 }
1347 
1348 static inline int pte_swp_soft_dirty(pte_t pte)
1349 {
1350 	return 0;
1351 }
1352 
1353 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1354 {
1355 	return pte;
1356 }
1357 
1358 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1359 {
1360 	return pmd;
1361 }
1362 
1363 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1364 {
1365 	return 0;
1366 }
1367 
1368 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1369 {
1370 	return pmd;
1371 }
1372 #endif
1373 
1374 #ifndef __HAVE_PFNMAP_TRACKING
1375 /*
1376  * Interfaces that can be used by architecture code to keep track of
1377  * memory type of pfn mappings specified by the remap_pfn_range,
1378  * vmf_insert_pfn.
1379  */
1380 
1381 /*
1382  * track_pfn_remap is called when a _new_ pfn mapping is being established
1383  * by remap_pfn_range() for physical range indicated by pfn and size.
1384  */
1385 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1386 				  unsigned long pfn, unsigned long addr,
1387 				  unsigned long size)
1388 {
1389 	return 0;
1390 }
1391 
1392 /*
1393  * track_pfn_insert is called when a _new_ single pfn is established
1394  * by vmf_insert_pfn().
1395  */
1396 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1397 				    pfn_t pfn)
1398 {
1399 }
1400 
1401 /*
1402  * track_pfn_copy is called when vma that is covering the pfnmap gets
1403  * copied through copy_page_range().
1404  */
1405 static inline int track_pfn_copy(struct vm_area_struct *vma)
1406 {
1407 	return 0;
1408 }
1409 
1410 /*
1411  * untrack_pfn is called while unmapping a pfnmap for a region.
1412  * untrack can be called for a specific region indicated by pfn and size or
1413  * can be for the entire vma (in which case pfn, size are zero).
1414  */
1415 static inline void untrack_pfn(struct vm_area_struct *vma,
1416 			       unsigned long pfn, unsigned long size,
1417 			       bool mm_wr_locked)
1418 {
1419 }
1420 
1421 /*
1422  * untrack_pfn_clear is called while mremapping a pfnmap for a new region
1423  * or fails to copy pgtable during duplicate vm area.
1424  */
1425 static inline void untrack_pfn_clear(struct vm_area_struct *vma)
1426 {
1427 }
1428 #else
1429 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1430 			   unsigned long pfn, unsigned long addr,
1431 			   unsigned long size);
1432 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1433 			     pfn_t pfn);
1434 extern int track_pfn_copy(struct vm_area_struct *vma);
1435 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1436 			unsigned long size, bool mm_wr_locked);
1437 extern void untrack_pfn_clear(struct vm_area_struct *vma);
1438 #endif
1439 
1440 #ifdef CONFIG_MMU
1441 #ifdef __HAVE_COLOR_ZERO_PAGE
1442 static inline int is_zero_pfn(unsigned long pfn)
1443 {
1444 	extern unsigned long zero_pfn;
1445 	unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1446 	return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1447 }
1448 
1449 #define my_zero_pfn(addr)	page_to_pfn(ZERO_PAGE(addr))
1450 
1451 #else
1452 static inline int is_zero_pfn(unsigned long pfn)
1453 {
1454 	extern unsigned long zero_pfn;
1455 	return pfn == zero_pfn;
1456 }
1457 
1458 static inline unsigned long my_zero_pfn(unsigned long addr)
1459 {
1460 	extern unsigned long zero_pfn;
1461 	return zero_pfn;
1462 }
1463 #endif
1464 #else
1465 static inline int is_zero_pfn(unsigned long pfn)
1466 {
1467 	return 0;
1468 }
1469 
1470 static inline unsigned long my_zero_pfn(unsigned long addr)
1471 {
1472 	return 0;
1473 }
1474 #endif /* CONFIG_MMU */
1475 
1476 #ifdef CONFIG_MMU
1477 
1478 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1479 static inline int pmd_trans_huge(pmd_t pmd)
1480 {
1481 	return 0;
1482 }
1483 #ifndef pmd_write
1484 static inline int pmd_write(pmd_t pmd)
1485 {
1486 	BUG();
1487 	return 0;
1488 }
1489 #endif /* pmd_write */
1490 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1491 
1492 #ifndef pud_write
1493 static inline int pud_write(pud_t pud)
1494 {
1495 	BUG();
1496 	return 0;
1497 }
1498 #endif /* pud_write */
1499 
1500 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1501 static inline int pmd_devmap(pmd_t pmd)
1502 {
1503 	return 0;
1504 }
1505 static inline int pud_devmap(pud_t pud)
1506 {
1507 	return 0;
1508 }
1509 static inline int pgd_devmap(pgd_t pgd)
1510 {
1511 	return 0;
1512 }
1513 #endif
1514 
1515 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1516 	!defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1517 static inline int pud_trans_huge(pud_t pud)
1518 {
1519 	return 0;
1520 }
1521 #endif
1522 
1523 static inline int pud_trans_unstable(pud_t *pud)
1524 {
1525 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1526 	defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1527 	pud_t pudval = READ_ONCE(*pud);
1528 
1529 	if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1530 		return 1;
1531 	if (unlikely(pud_bad(pudval))) {
1532 		pud_clear_bad(pud);
1533 		return 1;
1534 	}
1535 #endif
1536 	return 0;
1537 }
1538 
1539 #ifndef CONFIG_NUMA_BALANCING
1540 /*
1541  * In an inaccessible (PROT_NONE) VMA, pte_protnone() may indicate "yes". It is
1542  * perfectly valid to indicate "no" in that case, which is why our default
1543  * implementation defaults to "always no".
1544  *
1545  * In an accessible VMA, however, pte_protnone() reliably indicates PROT_NONE
1546  * page protection due to NUMA hinting. NUMA hinting faults only apply in
1547  * accessible VMAs.
1548  *
1549  * So, to reliably identify PROT_NONE PTEs that require a NUMA hinting fault,
1550  * looking at the VMA accessibility is sufficient.
1551  */
1552 static inline int pte_protnone(pte_t pte)
1553 {
1554 	return 0;
1555 }
1556 
1557 static inline int pmd_protnone(pmd_t pmd)
1558 {
1559 	return 0;
1560 }
1561 #endif /* CONFIG_NUMA_BALANCING */
1562 
1563 #endif /* CONFIG_MMU */
1564 
1565 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1566 
1567 #ifndef __PAGETABLE_P4D_FOLDED
1568 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1569 void p4d_clear_huge(p4d_t *p4d);
1570 #else
1571 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1572 {
1573 	return 0;
1574 }
1575 static inline void p4d_clear_huge(p4d_t *p4d) { }
1576 #endif /* !__PAGETABLE_P4D_FOLDED */
1577 
1578 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1579 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1580 int pud_clear_huge(pud_t *pud);
1581 int pmd_clear_huge(pmd_t *pmd);
1582 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1583 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1584 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1585 #else	/* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1586 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1587 {
1588 	return 0;
1589 }
1590 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1591 {
1592 	return 0;
1593 }
1594 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1595 {
1596 	return 0;
1597 }
1598 static inline void p4d_clear_huge(p4d_t *p4d) { }
1599 static inline int pud_clear_huge(pud_t *pud)
1600 {
1601 	return 0;
1602 }
1603 static inline int pmd_clear_huge(pmd_t *pmd)
1604 {
1605 	return 0;
1606 }
1607 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1608 {
1609 	return 0;
1610 }
1611 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1612 {
1613 	return 0;
1614 }
1615 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1616 {
1617 	return 0;
1618 }
1619 #endif	/* CONFIG_HAVE_ARCH_HUGE_VMAP */
1620 
1621 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1622 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1623 /*
1624  * ARCHes with special requirements for evicting THP backing TLB entries can
1625  * implement this. Otherwise also, it can help optimize normal TLB flush in
1626  * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1627  * entire TLB if flush span is greater than a threshold, which will
1628  * likely be true for a single huge page. Thus a single THP flush will
1629  * invalidate the entire TLB which is not desirable.
1630  * e.g. see arch/arc: flush_pmd_tlb_range
1631  */
1632 #define flush_pmd_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1633 #define flush_pud_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1634 #else
1635 #define flush_pmd_tlb_range(vma, addr, end)	BUILD_BUG()
1636 #define flush_pud_tlb_range(vma, addr, end)	BUILD_BUG()
1637 #endif
1638 #endif
1639 
1640 struct file;
1641 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1642 			unsigned long size, pgprot_t *vma_prot);
1643 
1644 #ifndef CONFIG_X86_ESPFIX64
1645 static inline void init_espfix_bsp(void) { }
1646 #endif
1647 
1648 extern void __init pgtable_cache_init(void);
1649 
1650 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1651 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1652 {
1653 	return true;
1654 }
1655 
1656 static inline bool arch_has_pfn_modify_check(void)
1657 {
1658 	return false;
1659 }
1660 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1661 
1662 /*
1663  * Architecture PAGE_KERNEL_* fallbacks
1664  *
1665  * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1666  * because they really don't support them, or the port needs to be updated to
1667  * reflect the required functionality. Below are a set of relatively safe
1668  * fallbacks, as best effort, which we can count on in lieu of the architectures
1669  * not defining them on their own yet.
1670  */
1671 
1672 #ifndef PAGE_KERNEL_RO
1673 # define PAGE_KERNEL_RO PAGE_KERNEL
1674 #endif
1675 
1676 #ifndef PAGE_KERNEL_EXEC
1677 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1678 #endif
1679 
1680 /*
1681  * Page Table Modification bits for pgtbl_mod_mask.
1682  *
1683  * These are used by the p?d_alloc_track*() set of functions an in the generic
1684  * vmalloc/ioremap code to track at which page-table levels entries have been
1685  * modified. Based on that the code can better decide when vmalloc and ioremap
1686  * mapping changes need to be synchronized to other page-tables in the system.
1687  */
1688 #define		__PGTBL_PGD_MODIFIED	0
1689 #define		__PGTBL_P4D_MODIFIED	1
1690 #define		__PGTBL_PUD_MODIFIED	2
1691 #define		__PGTBL_PMD_MODIFIED	3
1692 #define		__PGTBL_PTE_MODIFIED	4
1693 
1694 #define		PGTBL_PGD_MODIFIED	BIT(__PGTBL_PGD_MODIFIED)
1695 #define		PGTBL_P4D_MODIFIED	BIT(__PGTBL_P4D_MODIFIED)
1696 #define		PGTBL_PUD_MODIFIED	BIT(__PGTBL_PUD_MODIFIED)
1697 #define		PGTBL_PMD_MODIFIED	BIT(__PGTBL_PMD_MODIFIED)
1698 #define		PGTBL_PTE_MODIFIED	BIT(__PGTBL_PTE_MODIFIED)
1699 
1700 /* Page-Table Modification Mask */
1701 typedef unsigned int pgtbl_mod_mask;
1702 
1703 #endif /* !__ASSEMBLY__ */
1704 
1705 #if !defined(MAX_POSSIBLE_PHYSMEM_BITS) && !defined(CONFIG_64BIT)
1706 #ifdef CONFIG_PHYS_ADDR_T_64BIT
1707 /*
1708  * ZSMALLOC needs to know the highest PFN on 32-bit architectures
1709  * with physical address space extension, but falls back to
1710  * BITS_PER_LONG otherwise.
1711  */
1712 #error Missing MAX_POSSIBLE_PHYSMEM_BITS definition
1713 #else
1714 #define MAX_POSSIBLE_PHYSMEM_BITS 32
1715 #endif
1716 #endif
1717 
1718 #ifndef has_transparent_hugepage
1719 #define has_transparent_hugepage() IS_BUILTIN(CONFIG_TRANSPARENT_HUGEPAGE)
1720 #endif
1721 
1722 #ifndef has_transparent_pud_hugepage
1723 #define has_transparent_pud_hugepage() IS_BUILTIN(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1724 #endif
1725 /*
1726  * On some architectures it depends on the mm if the p4d/pud or pmd
1727  * layer of the page table hierarchy is folded or not.
1728  */
1729 #ifndef mm_p4d_folded
1730 #define mm_p4d_folded(mm)	__is_defined(__PAGETABLE_P4D_FOLDED)
1731 #endif
1732 
1733 #ifndef mm_pud_folded
1734 #define mm_pud_folded(mm)	__is_defined(__PAGETABLE_PUD_FOLDED)
1735 #endif
1736 
1737 #ifndef mm_pmd_folded
1738 #define mm_pmd_folded(mm)	__is_defined(__PAGETABLE_PMD_FOLDED)
1739 #endif
1740 
1741 #ifndef p4d_offset_lockless
1742 #define p4d_offset_lockless(pgdp, pgd, address) p4d_offset(&(pgd), address)
1743 #endif
1744 #ifndef pud_offset_lockless
1745 #define pud_offset_lockless(p4dp, p4d, address) pud_offset(&(p4d), address)
1746 #endif
1747 #ifndef pmd_offset_lockless
1748 #define pmd_offset_lockless(pudp, pud, address) pmd_offset(&(pud), address)
1749 #endif
1750 
1751 /*
1752  * p?d_leaf() - true if this entry is a final mapping to a physical address.
1753  * This differs from p?d_huge() by the fact that they are always available (if
1754  * the architecture supports large pages at the appropriate level) even
1755  * if CONFIG_HUGETLB_PAGE is not defined.
1756  * Only meaningful when called on a valid entry.
1757  */
1758 #ifndef pgd_leaf
1759 #define pgd_leaf(x)	0
1760 #endif
1761 #ifndef p4d_leaf
1762 #define p4d_leaf(x)	0
1763 #endif
1764 #ifndef pud_leaf
1765 #define pud_leaf(x)	0
1766 #endif
1767 #ifndef pmd_leaf
1768 #define pmd_leaf(x)	0
1769 #endif
1770 
1771 #ifndef pgd_leaf_size
1772 #define pgd_leaf_size(x) (1ULL << PGDIR_SHIFT)
1773 #endif
1774 #ifndef p4d_leaf_size
1775 #define p4d_leaf_size(x) P4D_SIZE
1776 #endif
1777 #ifndef pud_leaf_size
1778 #define pud_leaf_size(x) PUD_SIZE
1779 #endif
1780 #ifndef pmd_leaf_size
1781 #define pmd_leaf_size(x) PMD_SIZE
1782 #endif
1783 #ifndef pte_leaf_size
1784 #define pte_leaf_size(x) PAGE_SIZE
1785 #endif
1786 
1787 /*
1788  * Some architectures have MMUs that are configurable or selectable at boot
1789  * time. These lead to variable PTRS_PER_x. For statically allocated arrays it
1790  * helps to have a static maximum value.
1791  */
1792 
1793 #ifndef MAX_PTRS_PER_PTE
1794 #define MAX_PTRS_PER_PTE PTRS_PER_PTE
1795 #endif
1796 
1797 #ifndef MAX_PTRS_PER_PMD
1798 #define MAX_PTRS_PER_PMD PTRS_PER_PMD
1799 #endif
1800 
1801 #ifndef MAX_PTRS_PER_PUD
1802 #define MAX_PTRS_PER_PUD PTRS_PER_PUD
1803 #endif
1804 
1805 #ifndef MAX_PTRS_PER_P4D
1806 #define MAX_PTRS_PER_P4D PTRS_PER_P4D
1807 #endif
1808 
1809 /* description of effects of mapping type and prot in current implementation.
1810  * this is due to the limited x86 page protection hardware.  The expected
1811  * behavior is in parens:
1812  *
1813  * map_type	prot
1814  *		PROT_NONE	PROT_READ	PROT_WRITE	PROT_EXEC
1815  * MAP_SHARED	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1816  *		w: (no) no	w: (no) no	w: (yes) yes	w: (no) no
1817  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1818  *
1819  * MAP_PRIVATE	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1820  *		w: (no) no	w: (no) no	w: (copy) copy	w: (no) no
1821  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1822  *
1823  * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
1824  * MAP_PRIVATE (with Enhanced PAN supported):
1825  *								r: (no) no
1826  *								w: (no) no
1827  *								x: (yes) yes
1828  */
1829 #define DECLARE_VM_GET_PAGE_PROT					\
1830 pgprot_t vm_get_page_prot(unsigned long vm_flags)			\
1831 {									\
1832 		return protection_map[vm_flags &			\
1833 			(VM_READ | VM_WRITE | VM_EXEC | VM_SHARED)];	\
1834 }									\
1835 EXPORT_SYMBOL(vm_get_page_prot);
1836 
1837 #endif /* _LINUX_PGTABLE_H */
1838