xref: /linux-6.15/include/linux/pgtable.h (revision bb8e97e2)
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 /*
188  * A facility to provide lazy MMU batching.  This allows PTE updates and
189  * page invalidations to be delayed until a call to leave lazy MMU mode
190  * is issued.  Some architectures may benefit from doing this, and it is
191  * beneficial for both shadow and direct mode hypervisors, which may batch
192  * the PTE updates which happen during this window.  Note that using this
193  * interface requires that read hazards be removed from the code.  A read
194  * hazard could result in the direct mode hypervisor case, since the actual
195  * write to the page tables may not yet have taken place, so reads though
196  * a raw PTE pointer after it has been modified are not guaranteed to be
197  * up to date.  This mode can only be entered and left under the protection of
198  * the page table locks for all page tables which may be modified.  In the UP
199  * case, this is required so that preemption is disabled, and in the SMP case,
200  * it must synchronize the delayed page table writes properly on other CPUs.
201  */
202 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
203 #define arch_enter_lazy_mmu_mode()	do {} while (0)
204 #define arch_leave_lazy_mmu_mode()	do {} while (0)
205 #define arch_flush_lazy_mmu_mode()	do {} while (0)
206 #endif
207 
208 #ifndef set_ptes
209 /**
210  * set_ptes - Map consecutive pages to a contiguous range of addresses.
211  * @mm: Address space to map the pages into.
212  * @addr: Address to map the first page at.
213  * @ptep: Page table pointer for the first entry.
214  * @pte: Page table entry for the first page.
215  * @nr: Number of pages to map.
216  *
217  * May be overridden by the architecture, or the architecture can define
218  * set_pte() and PFN_PTE_SHIFT.
219  *
220  * Context: The caller holds the page table lock.  The pages all belong
221  * to the same folio.  The PTEs are all in the same PMD.
222  */
223 static inline void set_ptes(struct mm_struct *mm, unsigned long addr,
224 		pte_t *ptep, pte_t pte, unsigned int nr)
225 {
226 	page_table_check_ptes_set(mm, ptep, pte, nr);
227 
228 	arch_enter_lazy_mmu_mode();
229 	for (;;) {
230 		set_pte(ptep, pte);
231 		if (--nr == 0)
232 			break;
233 		ptep++;
234 		pte = __pte(pte_val(pte) + (1UL << PFN_PTE_SHIFT));
235 	}
236 	arch_leave_lazy_mmu_mode();
237 }
238 #endif
239 #define set_pte_at(mm, addr, ptep, pte) set_ptes(mm, addr, ptep, pte, 1)
240 
241 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
242 extern int ptep_set_access_flags(struct vm_area_struct *vma,
243 				 unsigned long address, pte_t *ptep,
244 				 pte_t entry, int dirty);
245 #endif
246 
247 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
248 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
249 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
250 				 unsigned long address, pmd_t *pmdp,
251 				 pmd_t entry, int dirty);
252 extern int pudp_set_access_flags(struct vm_area_struct *vma,
253 				 unsigned long address, pud_t *pudp,
254 				 pud_t entry, int dirty);
255 #else
256 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
257 					unsigned long address, pmd_t *pmdp,
258 					pmd_t entry, int dirty)
259 {
260 	BUILD_BUG();
261 	return 0;
262 }
263 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
264 					unsigned long address, pud_t *pudp,
265 					pud_t entry, int dirty)
266 {
267 	BUILD_BUG();
268 	return 0;
269 }
270 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
271 #endif
272 
273 #ifndef ptep_get
274 static inline pte_t ptep_get(pte_t *ptep)
275 {
276 	return READ_ONCE(*ptep);
277 }
278 #endif
279 
280 #ifndef pmdp_get
281 static inline pmd_t pmdp_get(pmd_t *pmdp)
282 {
283 	return READ_ONCE(*pmdp);
284 }
285 #endif
286 
287 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
288 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
289 					    unsigned long address,
290 					    pte_t *ptep)
291 {
292 	pte_t pte = ptep_get(ptep);
293 	int r = 1;
294 	if (!pte_young(pte))
295 		r = 0;
296 	else
297 		set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
298 	return r;
299 }
300 #endif
301 
302 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
303 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG)
304 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
305 					    unsigned long address,
306 					    pmd_t *pmdp)
307 {
308 	pmd_t pmd = *pmdp;
309 	int r = 1;
310 	if (!pmd_young(pmd))
311 		r = 0;
312 	else
313 		set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
314 	return r;
315 }
316 #else
317 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
318 					    unsigned long address,
319 					    pmd_t *pmdp)
320 {
321 	BUILD_BUG();
322 	return 0;
323 }
324 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG */
325 #endif
326 
327 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
328 int ptep_clear_flush_young(struct vm_area_struct *vma,
329 			   unsigned long address, pte_t *ptep);
330 #endif
331 
332 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
333 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
334 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
335 				  unsigned long address, pmd_t *pmdp);
336 #else
337 /*
338  * Despite relevant to THP only, this API is called from generic rmap code
339  * under PageTransHuge(), hence needs a dummy implementation for !THP
340  */
341 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
342 					 unsigned long address, pmd_t *pmdp)
343 {
344 	BUILD_BUG();
345 	return 0;
346 }
347 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
348 #endif
349 
350 #ifndef arch_has_hw_nonleaf_pmd_young
351 /*
352  * Return whether the accessed bit in non-leaf PMD entries is supported on the
353  * local CPU.
354  */
355 static inline bool arch_has_hw_nonleaf_pmd_young(void)
356 {
357 	return IS_ENABLED(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG);
358 }
359 #endif
360 
361 #ifndef arch_has_hw_pte_young
362 /*
363  * Return whether the accessed bit is supported on the local CPU.
364  *
365  * This stub assumes accessing through an old PTE triggers a page fault.
366  * Architectures that automatically set the access bit should overwrite it.
367  */
368 static inline bool arch_has_hw_pte_young(void)
369 {
370 	return false;
371 }
372 #endif
373 
374 #ifndef arch_check_zapped_pte
375 static inline void arch_check_zapped_pte(struct vm_area_struct *vma,
376 					 pte_t pte)
377 {
378 }
379 #endif
380 
381 #ifndef arch_check_zapped_pmd
382 static inline void arch_check_zapped_pmd(struct vm_area_struct *vma,
383 					 pmd_t pmd)
384 {
385 }
386 #endif
387 
388 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
389 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
390 				       unsigned long address,
391 				       pte_t *ptep)
392 {
393 	pte_t pte = ptep_get(ptep);
394 	pte_clear(mm, address, ptep);
395 	page_table_check_pte_clear(mm, pte);
396 	return pte;
397 }
398 #endif
399 
400 static inline void ptep_clear(struct mm_struct *mm, unsigned long addr,
401 			      pte_t *ptep)
402 {
403 	ptep_get_and_clear(mm, addr, ptep);
404 }
405 
406 #ifdef CONFIG_GUP_GET_PXX_LOW_HIGH
407 /*
408  * For walking the pagetables without holding any locks.  Some architectures
409  * (eg x86-32 PAE) cannot load the entries atomically without using expensive
410  * instructions.  We are guaranteed that a PTE will only either go from not
411  * present to present, or present to not present -- it will not switch to a
412  * completely different present page without a TLB flush inbetween; which we
413  * are blocking by holding interrupts off.
414  *
415  * Setting ptes from not present to present goes:
416  *
417  *   ptep->pte_high = h;
418  *   smp_wmb();
419  *   ptep->pte_low = l;
420  *
421  * And present to not present goes:
422  *
423  *   ptep->pte_low = 0;
424  *   smp_wmb();
425  *   ptep->pte_high = 0;
426  *
427  * We must ensure here that the load of pte_low sees 'l' IFF pte_high sees 'h'.
428  * We load pte_high *after* loading pte_low, which ensures we don't see an older
429  * value of pte_high.  *Then* we recheck pte_low, which ensures that we haven't
430  * picked up a changed pte high. We might have gotten rubbish values from
431  * pte_low and pte_high, but we are guaranteed that pte_low will not have the
432  * present bit set *unless* it is 'l'. Because get_user_pages_fast() only
433  * operates on present ptes we're safe.
434  */
435 static inline pte_t ptep_get_lockless(pte_t *ptep)
436 {
437 	pte_t pte;
438 
439 	do {
440 		pte.pte_low = ptep->pte_low;
441 		smp_rmb();
442 		pte.pte_high = ptep->pte_high;
443 		smp_rmb();
444 	} while (unlikely(pte.pte_low != ptep->pte_low));
445 
446 	return pte;
447 }
448 #define ptep_get_lockless ptep_get_lockless
449 
450 #if CONFIG_PGTABLE_LEVELS > 2
451 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
452 {
453 	pmd_t pmd;
454 
455 	do {
456 		pmd.pmd_low = pmdp->pmd_low;
457 		smp_rmb();
458 		pmd.pmd_high = pmdp->pmd_high;
459 		smp_rmb();
460 	} while (unlikely(pmd.pmd_low != pmdp->pmd_low));
461 
462 	return pmd;
463 }
464 #define pmdp_get_lockless pmdp_get_lockless
465 #define pmdp_get_lockless_sync() tlb_remove_table_sync_one()
466 #endif /* CONFIG_PGTABLE_LEVELS > 2 */
467 #endif /* CONFIG_GUP_GET_PXX_LOW_HIGH */
468 
469 /*
470  * We require that the PTE can be read atomically.
471  */
472 #ifndef ptep_get_lockless
473 static inline pte_t ptep_get_lockless(pte_t *ptep)
474 {
475 	return ptep_get(ptep);
476 }
477 #endif
478 
479 #ifndef pmdp_get_lockless
480 static inline pmd_t pmdp_get_lockless(pmd_t *pmdp)
481 {
482 	return pmdp_get(pmdp);
483 }
484 static inline void pmdp_get_lockless_sync(void)
485 {
486 }
487 #endif
488 
489 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
490 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
491 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
492 					    unsigned long address,
493 					    pmd_t *pmdp)
494 {
495 	pmd_t pmd = *pmdp;
496 
497 	pmd_clear(pmdp);
498 	page_table_check_pmd_clear(mm, pmd);
499 
500 	return pmd;
501 }
502 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
503 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
504 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
505 					    unsigned long address,
506 					    pud_t *pudp)
507 {
508 	pud_t pud = *pudp;
509 
510 	pud_clear(pudp);
511 	page_table_check_pud_clear(mm, pud);
512 
513 	return pud;
514 }
515 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
516 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
517 
518 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
519 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
520 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
521 					    unsigned long address, pmd_t *pmdp,
522 					    int full)
523 {
524 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
525 }
526 #endif
527 
528 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
529 static inline pud_t pudp_huge_get_and_clear_full(struct vm_area_struct *vma,
530 					    unsigned long address, pud_t *pudp,
531 					    int full)
532 {
533 	return pudp_huge_get_and_clear(vma->vm_mm, address, pudp);
534 }
535 #endif
536 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
537 
538 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
539 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
540 					    unsigned long address, pte_t *ptep,
541 					    int full)
542 {
543 	return ptep_get_and_clear(mm, address, ptep);
544 }
545 #endif
546 
547 
548 /*
549  * If two threads concurrently fault at the same page, the thread that
550  * won the race updates the PTE and its local TLB/Cache. The other thread
551  * gives up, simply does nothing, and continues; on architectures where
552  * software can update TLB,  local TLB can be updated here to avoid next page
553  * fault. This function updates TLB only, do nothing with cache or others.
554  * It is the difference with function update_mmu_cache.
555  */
556 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
557 static inline void update_mmu_tlb(struct vm_area_struct *vma,
558 				unsigned long address, pte_t *ptep)
559 {
560 }
561 #define __HAVE_ARCH_UPDATE_MMU_TLB
562 #endif
563 
564 /*
565  * Some architectures may be able to avoid expensive synchronization
566  * primitives when modifications are made to PTE's which are already
567  * not present, or in the process of an address space destruction.
568  */
569 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
570 static inline void pte_clear_not_present_full(struct mm_struct *mm,
571 					      unsigned long address,
572 					      pte_t *ptep,
573 					      int full)
574 {
575 	pte_clear(mm, address, ptep);
576 }
577 #endif
578 
579 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
580 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
581 			      unsigned long address,
582 			      pte_t *ptep);
583 #endif
584 
585 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
586 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
587 			      unsigned long address,
588 			      pmd_t *pmdp);
589 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
590 			      unsigned long address,
591 			      pud_t *pudp);
592 #endif
593 
594 #ifndef pte_mkwrite
595 static inline pte_t pte_mkwrite(pte_t pte, struct vm_area_struct *vma)
596 {
597 	return pte_mkwrite_novma(pte);
598 }
599 #endif
600 
601 #if defined(CONFIG_ARCH_WANT_PMD_MKWRITE) && !defined(pmd_mkwrite)
602 static inline pmd_t pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma)
603 {
604 	return pmd_mkwrite_novma(pmd);
605 }
606 #endif
607 
608 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
609 struct mm_struct;
610 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
611 {
612 	pte_t old_pte = ptep_get(ptep);
613 	set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
614 }
615 #endif
616 
617 /*
618  * On some architectures hardware does not set page access bit when accessing
619  * memory page, it is responsibility of software setting this bit. It brings
620  * out extra page fault penalty to track page access bit. For optimization page
621  * access bit can be set during all page fault flow on these arches.
622  * To be differentiate with macro pte_mkyoung, this macro is used on platforms
623  * where software maintains page access bit.
624  */
625 #ifndef pte_sw_mkyoung
626 static inline pte_t pte_sw_mkyoung(pte_t pte)
627 {
628 	return pte;
629 }
630 #define pte_sw_mkyoung	pte_sw_mkyoung
631 #endif
632 
633 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
634 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
635 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
636 				      unsigned long address, pmd_t *pmdp)
637 {
638 	pmd_t old_pmd = *pmdp;
639 	set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
640 }
641 #else
642 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
643 				      unsigned long address, pmd_t *pmdp)
644 {
645 	BUILD_BUG();
646 }
647 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
648 #endif
649 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
650 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
651 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
652 static inline void pudp_set_wrprotect(struct mm_struct *mm,
653 				      unsigned long address, pud_t *pudp)
654 {
655 	pud_t old_pud = *pudp;
656 
657 	set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
658 }
659 #else
660 static inline void pudp_set_wrprotect(struct mm_struct *mm,
661 				      unsigned long address, pud_t *pudp)
662 {
663 	BUILD_BUG();
664 }
665 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
666 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
667 #endif
668 
669 #ifndef pmdp_collapse_flush
670 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
671 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
672 				 unsigned long address, pmd_t *pmdp);
673 #else
674 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
675 					unsigned long address,
676 					pmd_t *pmdp)
677 {
678 	BUILD_BUG();
679 	return *pmdp;
680 }
681 #define pmdp_collapse_flush pmdp_collapse_flush
682 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
683 #endif
684 
685 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
686 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
687 				       pgtable_t pgtable);
688 #endif
689 
690 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
691 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
692 #endif
693 
694 #ifndef arch_needs_pgtable_deposit
695 #define arch_needs_pgtable_deposit() (false)
696 #endif
697 
698 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
699 /*
700  * This is an implementation of pmdp_establish() that is only suitable for an
701  * architecture that doesn't have hardware dirty/accessed bits. In this case we
702  * can't race with CPU which sets these bits and non-atomic approach is fine.
703  */
704 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
705 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
706 {
707 	pmd_t old_pmd = *pmdp;
708 	set_pmd_at(vma->vm_mm, address, pmdp, pmd);
709 	return old_pmd;
710 }
711 #endif
712 
713 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
714 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
715 			    pmd_t *pmdp);
716 #endif
717 
718 #ifndef __HAVE_ARCH_PMDP_INVALIDATE_AD
719 
720 /*
721  * pmdp_invalidate_ad() invalidates the PMD while changing a transparent
722  * hugepage mapping in the page tables. This function is similar to
723  * pmdp_invalidate(), but should only be used if the access and dirty bits would
724  * not be cleared by the software in the new PMD value. The function ensures
725  * that hardware changes of the access and dirty bits updates would not be lost.
726  *
727  * Doing so can allow in certain architectures to avoid a TLB flush in most
728  * cases. Yet, another TLB flush might be necessary later if the PMD update
729  * itself requires such flush (e.g., if protection was set to be stricter). Yet,
730  * even when a TLB flush is needed because of the update, the caller may be able
731  * to batch these TLB flushing operations, so fewer TLB flush operations are
732  * needed.
733  */
734 extern pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma,
735 				unsigned long address, pmd_t *pmdp);
736 #endif
737 
738 #ifndef __HAVE_ARCH_PTE_SAME
739 static inline int pte_same(pte_t pte_a, pte_t pte_b)
740 {
741 	return pte_val(pte_a) == pte_val(pte_b);
742 }
743 #endif
744 
745 #ifndef __HAVE_ARCH_PTE_UNUSED
746 /*
747  * Some architectures provide facilities to virtualization guests
748  * so that they can flag allocated pages as unused. This allows the
749  * host to transparently reclaim unused pages. This function returns
750  * whether the pte's page is unused.
751  */
752 static inline int pte_unused(pte_t pte)
753 {
754 	return 0;
755 }
756 #endif
757 
758 #ifndef pte_access_permitted
759 #define pte_access_permitted(pte, write) \
760 	(pte_present(pte) && (!(write) || pte_write(pte)))
761 #endif
762 
763 #ifndef pmd_access_permitted
764 #define pmd_access_permitted(pmd, write) \
765 	(pmd_present(pmd) && (!(write) || pmd_write(pmd)))
766 #endif
767 
768 #ifndef pud_access_permitted
769 #define pud_access_permitted(pud, write) \
770 	(pud_present(pud) && (!(write) || pud_write(pud)))
771 #endif
772 
773 #ifndef p4d_access_permitted
774 #define p4d_access_permitted(p4d, write) \
775 	(p4d_present(p4d) && (!(write) || p4d_write(p4d)))
776 #endif
777 
778 #ifndef pgd_access_permitted
779 #define pgd_access_permitted(pgd, write) \
780 	(pgd_present(pgd) && (!(write) || pgd_write(pgd)))
781 #endif
782 
783 #ifndef __HAVE_ARCH_PMD_SAME
784 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
785 {
786 	return pmd_val(pmd_a) == pmd_val(pmd_b);
787 }
788 #endif
789 
790 #ifndef pud_same
791 static inline int pud_same(pud_t pud_a, pud_t pud_b)
792 {
793 	return pud_val(pud_a) == pud_val(pud_b);
794 }
795 #define pud_same pud_same
796 #endif
797 
798 #ifndef __HAVE_ARCH_P4D_SAME
799 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
800 {
801 	return p4d_val(p4d_a) == p4d_val(p4d_b);
802 }
803 #endif
804 
805 #ifndef __HAVE_ARCH_PGD_SAME
806 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
807 {
808 	return pgd_val(pgd_a) == pgd_val(pgd_b);
809 }
810 #endif
811 
812 /*
813  * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
814  * TLB flush will be required as a result of the "set". For example, use
815  * in scenarios where it is known ahead of time that the routine is
816  * setting non-present entries, or re-setting an existing entry to the
817  * same value. Otherwise, use the typical "set" helpers and flush the
818  * TLB.
819  */
820 #define set_pte_safe(ptep, pte) \
821 ({ \
822 	WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
823 	set_pte(ptep, pte); \
824 })
825 
826 #define set_pmd_safe(pmdp, pmd) \
827 ({ \
828 	WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
829 	set_pmd(pmdp, pmd); \
830 })
831 
832 #define set_pud_safe(pudp, pud) \
833 ({ \
834 	WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
835 	set_pud(pudp, pud); \
836 })
837 
838 #define set_p4d_safe(p4dp, p4d) \
839 ({ \
840 	WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
841 	set_p4d(p4dp, p4d); \
842 })
843 
844 #define set_pgd_safe(pgdp, pgd) \
845 ({ \
846 	WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
847 	set_pgd(pgdp, pgd); \
848 })
849 
850 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
851 /*
852  * Some architectures support metadata associated with a page. When a
853  * page is being swapped out, this metadata must be saved so it can be
854  * restored when the page is swapped back in. SPARC M7 and newer
855  * processors support an ADI (Application Data Integrity) tag for the
856  * page as metadata for the page. arch_do_swap_page() can restore this
857  * metadata when a page is swapped back in.
858  */
859 static inline void arch_do_swap_page(struct mm_struct *mm,
860 				     struct vm_area_struct *vma,
861 				     unsigned long addr,
862 				     pte_t pte, pte_t oldpte)
863 {
864 
865 }
866 #endif
867 
868 #ifndef __HAVE_ARCH_UNMAP_ONE
869 /*
870  * Some architectures support metadata associated with a page. When a
871  * page is being swapped out, this metadata must be saved so it can be
872  * restored when the page is swapped back in. SPARC M7 and newer
873  * processors support an ADI (Application Data Integrity) tag for the
874  * page as metadata for the page. arch_unmap_one() can save this
875  * metadata on a swap-out of a page.
876  */
877 static inline int arch_unmap_one(struct mm_struct *mm,
878 				  struct vm_area_struct *vma,
879 				  unsigned long addr,
880 				  pte_t orig_pte)
881 {
882 	return 0;
883 }
884 #endif
885 
886 /*
887  * Allow architectures to preserve additional metadata associated with
888  * swapped-out pages. The corresponding __HAVE_ARCH_SWAP_* macros and function
889  * prototypes must be defined in the arch-specific asm/pgtable.h file.
890  */
891 #ifndef __HAVE_ARCH_PREPARE_TO_SWAP
892 static inline int arch_prepare_to_swap(struct page *page)
893 {
894 	return 0;
895 }
896 #endif
897 
898 #ifndef __HAVE_ARCH_SWAP_INVALIDATE
899 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
900 {
901 }
902 
903 static inline void arch_swap_invalidate_area(int type)
904 {
905 }
906 #endif
907 
908 #ifndef __HAVE_ARCH_SWAP_RESTORE
909 static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
910 {
911 }
912 #endif
913 
914 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
915 #define pgd_offset_gate(mm, addr)	pgd_offset(mm, addr)
916 #endif
917 
918 #ifndef __HAVE_ARCH_MOVE_PTE
919 #define move_pte(pte, prot, old_addr, new_addr)	(pte)
920 #endif
921 
922 #ifndef pte_accessible
923 # define pte_accessible(mm, pte)	((void)(pte), 1)
924 #endif
925 
926 #ifndef flush_tlb_fix_spurious_fault
927 #define flush_tlb_fix_spurious_fault(vma, address, ptep) flush_tlb_page(vma, address)
928 #endif
929 
930 /*
931  * When walking page tables, get the address of the next boundary,
932  * or the end address of the range if that comes earlier.  Although no
933  * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
934  */
935 
936 #define pgd_addr_end(addr, end)						\
937 ({	unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK;	\
938 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
939 })
940 
941 #ifndef p4d_addr_end
942 #define p4d_addr_end(addr, end)						\
943 ({	unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK;	\
944 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
945 })
946 #endif
947 
948 #ifndef pud_addr_end
949 #define pud_addr_end(addr, end)						\
950 ({	unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK;	\
951 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
952 })
953 #endif
954 
955 #ifndef pmd_addr_end
956 #define pmd_addr_end(addr, end)						\
957 ({	unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK;	\
958 	(__boundary - 1 < (end) - 1)? __boundary: (end);		\
959 })
960 #endif
961 
962 /*
963  * When walking page tables, we usually want to skip any p?d_none entries;
964  * and any p?d_bad entries - reporting the error before resetting to none.
965  * Do the tests inline, but report and clear the bad entry in mm/memory.c.
966  */
967 void pgd_clear_bad(pgd_t *);
968 
969 #ifndef __PAGETABLE_P4D_FOLDED
970 void p4d_clear_bad(p4d_t *);
971 #else
972 #define p4d_clear_bad(p4d)        do { } while (0)
973 #endif
974 
975 #ifndef __PAGETABLE_PUD_FOLDED
976 void pud_clear_bad(pud_t *);
977 #else
978 #define pud_clear_bad(p4d)        do { } while (0)
979 #endif
980 
981 void pmd_clear_bad(pmd_t *);
982 
983 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
984 {
985 	if (pgd_none(*pgd))
986 		return 1;
987 	if (unlikely(pgd_bad(*pgd))) {
988 		pgd_clear_bad(pgd);
989 		return 1;
990 	}
991 	return 0;
992 }
993 
994 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
995 {
996 	if (p4d_none(*p4d))
997 		return 1;
998 	if (unlikely(p4d_bad(*p4d))) {
999 		p4d_clear_bad(p4d);
1000 		return 1;
1001 	}
1002 	return 0;
1003 }
1004 
1005 static inline int pud_none_or_clear_bad(pud_t *pud)
1006 {
1007 	if (pud_none(*pud))
1008 		return 1;
1009 	if (unlikely(pud_bad(*pud))) {
1010 		pud_clear_bad(pud);
1011 		return 1;
1012 	}
1013 	return 0;
1014 }
1015 
1016 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
1017 {
1018 	if (pmd_none(*pmd))
1019 		return 1;
1020 	if (unlikely(pmd_bad(*pmd))) {
1021 		pmd_clear_bad(pmd);
1022 		return 1;
1023 	}
1024 	return 0;
1025 }
1026 
1027 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
1028 					     unsigned long addr,
1029 					     pte_t *ptep)
1030 {
1031 	/*
1032 	 * Get the current pte state, but zero it out to make it
1033 	 * non-present, preventing the hardware from asynchronously
1034 	 * updating it.
1035 	 */
1036 	return ptep_get_and_clear(vma->vm_mm, addr, ptep);
1037 }
1038 
1039 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
1040 					     unsigned long addr,
1041 					     pte_t *ptep, pte_t pte)
1042 {
1043 	/*
1044 	 * The pte is non-present, so there's no hardware state to
1045 	 * preserve.
1046 	 */
1047 	set_pte_at(vma->vm_mm, addr, ptep, pte);
1048 }
1049 
1050 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1051 /*
1052  * Start a pte protection read-modify-write transaction, which
1053  * protects against asynchronous hardware modifications to the pte.
1054  * The intention is not to prevent the hardware from making pte
1055  * updates, but to prevent any updates it may make from being lost.
1056  *
1057  * This does not protect against other software modifications of the
1058  * pte; the appropriate pte lock must be held over the transaction.
1059  *
1060  * Note that this interface is intended to be batchable, meaning that
1061  * ptep_modify_prot_commit may not actually update the pte, but merely
1062  * queue the update to be done at some later time.  The update must be
1063  * actually committed before the pte lock is released, however.
1064  */
1065 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
1066 					   unsigned long addr,
1067 					   pte_t *ptep)
1068 {
1069 	return __ptep_modify_prot_start(vma, addr, ptep);
1070 }
1071 
1072 /*
1073  * Commit an update to a pte, leaving any hardware-controlled bits in
1074  * the PTE unmodified.
1075  */
1076 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
1077 					   unsigned long addr,
1078 					   pte_t *ptep, pte_t old_pte, pte_t pte)
1079 {
1080 	__ptep_modify_prot_commit(vma, addr, ptep, pte);
1081 }
1082 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
1083 #endif /* CONFIG_MMU */
1084 
1085 /*
1086  * No-op macros that just return the current protection value. Defined here
1087  * because these macros can be used even if CONFIG_MMU is not defined.
1088  */
1089 
1090 #ifndef pgprot_nx
1091 #define pgprot_nx(prot)	(prot)
1092 #endif
1093 
1094 #ifndef pgprot_noncached
1095 #define pgprot_noncached(prot)	(prot)
1096 #endif
1097 
1098 #ifndef pgprot_writecombine
1099 #define pgprot_writecombine pgprot_noncached
1100 #endif
1101 
1102 #ifndef pgprot_writethrough
1103 #define pgprot_writethrough pgprot_noncached
1104 #endif
1105 
1106 #ifndef pgprot_device
1107 #define pgprot_device pgprot_noncached
1108 #endif
1109 
1110 #ifndef pgprot_mhp
1111 #define pgprot_mhp(prot)	(prot)
1112 #endif
1113 
1114 #ifdef CONFIG_MMU
1115 #ifndef pgprot_modify
1116 #define pgprot_modify pgprot_modify
1117 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
1118 {
1119 	if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
1120 		newprot = pgprot_noncached(newprot);
1121 	if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
1122 		newprot = pgprot_writecombine(newprot);
1123 	if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
1124 		newprot = pgprot_device(newprot);
1125 	return newprot;
1126 }
1127 #endif
1128 #endif /* CONFIG_MMU */
1129 
1130 #ifndef pgprot_encrypted
1131 #define pgprot_encrypted(prot)	(prot)
1132 #endif
1133 
1134 #ifndef pgprot_decrypted
1135 #define pgprot_decrypted(prot)	(prot)
1136 #endif
1137 
1138 /*
1139  * A facility to provide batching of the reload of page tables and
1140  * other process state with the actual context switch code for
1141  * paravirtualized guests.  By convention, only one of the batched
1142  * update (lazy) modes (CPU, MMU) should be active at any given time,
1143  * entry should never be nested, and entry and exits should always be
1144  * paired.  This is for sanity of maintaining and reasoning about the
1145  * kernel code.  In this case, the exit (end of the context switch) is
1146  * in architecture-specific code, and so doesn't need a generic
1147  * definition.
1148  */
1149 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
1150 #define arch_start_context_switch(prev)	do {} while (0)
1151 #endif
1152 
1153 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1154 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
1155 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1156 {
1157 	return pmd;
1158 }
1159 
1160 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1161 {
1162 	return 0;
1163 }
1164 
1165 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1166 {
1167 	return pmd;
1168 }
1169 #endif
1170 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
1171 static inline int pte_soft_dirty(pte_t pte)
1172 {
1173 	return 0;
1174 }
1175 
1176 static inline int pmd_soft_dirty(pmd_t pmd)
1177 {
1178 	return 0;
1179 }
1180 
1181 static inline pte_t pte_mksoft_dirty(pte_t pte)
1182 {
1183 	return pte;
1184 }
1185 
1186 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
1187 {
1188 	return pmd;
1189 }
1190 
1191 static inline pte_t pte_clear_soft_dirty(pte_t pte)
1192 {
1193 	return pte;
1194 }
1195 
1196 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
1197 {
1198 	return pmd;
1199 }
1200 
1201 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1202 {
1203 	return pte;
1204 }
1205 
1206 static inline int pte_swp_soft_dirty(pte_t pte)
1207 {
1208 	return 0;
1209 }
1210 
1211 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1212 {
1213 	return pte;
1214 }
1215 
1216 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1217 {
1218 	return pmd;
1219 }
1220 
1221 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1222 {
1223 	return 0;
1224 }
1225 
1226 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1227 {
1228 	return pmd;
1229 }
1230 #endif
1231 
1232 #ifndef __HAVE_PFNMAP_TRACKING
1233 /*
1234  * Interfaces that can be used by architecture code to keep track of
1235  * memory type of pfn mappings specified by the remap_pfn_range,
1236  * vmf_insert_pfn.
1237  */
1238 
1239 /*
1240  * track_pfn_remap is called when a _new_ pfn mapping is being established
1241  * by remap_pfn_range() for physical range indicated by pfn and size.
1242  */
1243 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1244 				  unsigned long pfn, unsigned long addr,
1245 				  unsigned long size)
1246 {
1247 	return 0;
1248 }
1249 
1250 /*
1251  * track_pfn_insert is called when a _new_ single pfn is established
1252  * by vmf_insert_pfn().
1253  */
1254 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1255 				    pfn_t pfn)
1256 {
1257 }
1258 
1259 /*
1260  * track_pfn_copy is called when vma that is covering the pfnmap gets
1261  * copied through copy_page_range().
1262  */
1263 static inline int track_pfn_copy(struct vm_area_struct *vma)
1264 {
1265 	return 0;
1266 }
1267 
1268 /*
1269  * untrack_pfn is called while unmapping a pfnmap for a region.
1270  * untrack can be called for a specific region indicated by pfn and size or
1271  * can be for the entire vma (in which case pfn, size are zero).
1272  */
1273 static inline void untrack_pfn(struct vm_area_struct *vma,
1274 			       unsigned long pfn, unsigned long size,
1275 			       bool mm_wr_locked)
1276 {
1277 }
1278 
1279 /*
1280  * untrack_pfn_clear is called while mremapping a pfnmap for a new region
1281  * or fails to copy pgtable during duplicate vm area.
1282  */
1283 static inline void untrack_pfn_clear(struct vm_area_struct *vma)
1284 {
1285 }
1286 #else
1287 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1288 			   unsigned long pfn, unsigned long addr,
1289 			   unsigned long size);
1290 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1291 			     pfn_t pfn);
1292 extern int track_pfn_copy(struct vm_area_struct *vma);
1293 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1294 			unsigned long size, bool mm_wr_locked);
1295 extern void untrack_pfn_clear(struct vm_area_struct *vma);
1296 #endif
1297 
1298 #ifdef CONFIG_MMU
1299 #ifdef __HAVE_COLOR_ZERO_PAGE
1300 static inline int is_zero_pfn(unsigned long pfn)
1301 {
1302 	extern unsigned long zero_pfn;
1303 	unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1304 	return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1305 }
1306 
1307 #define my_zero_pfn(addr)	page_to_pfn(ZERO_PAGE(addr))
1308 
1309 #else
1310 static inline int is_zero_pfn(unsigned long pfn)
1311 {
1312 	extern unsigned long zero_pfn;
1313 	return pfn == zero_pfn;
1314 }
1315 
1316 static inline unsigned long my_zero_pfn(unsigned long addr)
1317 {
1318 	extern unsigned long zero_pfn;
1319 	return zero_pfn;
1320 }
1321 #endif
1322 #else
1323 static inline int is_zero_pfn(unsigned long pfn)
1324 {
1325 	return 0;
1326 }
1327 
1328 static inline unsigned long my_zero_pfn(unsigned long addr)
1329 {
1330 	return 0;
1331 }
1332 #endif /* CONFIG_MMU */
1333 
1334 #ifdef CONFIG_MMU
1335 
1336 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1337 static inline int pmd_trans_huge(pmd_t pmd)
1338 {
1339 	return 0;
1340 }
1341 #ifndef pmd_write
1342 static inline int pmd_write(pmd_t pmd)
1343 {
1344 	BUG();
1345 	return 0;
1346 }
1347 #endif /* pmd_write */
1348 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1349 
1350 #ifndef pud_write
1351 static inline int pud_write(pud_t pud)
1352 {
1353 	BUG();
1354 	return 0;
1355 }
1356 #endif /* pud_write */
1357 
1358 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1359 static inline int pmd_devmap(pmd_t pmd)
1360 {
1361 	return 0;
1362 }
1363 static inline int pud_devmap(pud_t pud)
1364 {
1365 	return 0;
1366 }
1367 static inline int pgd_devmap(pgd_t pgd)
1368 {
1369 	return 0;
1370 }
1371 #endif
1372 
1373 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1374 	!defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1375 static inline int pud_trans_huge(pud_t pud)
1376 {
1377 	return 0;
1378 }
1379 #endif
1380 
1381 static inline int pud_trans_unstable(pud_t *pud)
1382 {
1383 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1384 	defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1385 	pud_t pudval = READ_ONCE(*pud);
1386 
1387 	if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1388 		return 1;
1389 	if (unlikely(pud_bad(pudval))) {
1390 		pud_clear_bad(pud);
1391 		return 1;
1392 	}
1393 #endif
1394 	return 0;
1395 }
1396 
1397 #ifndef CONFIG_NUMA_BALANCING
1398 /*
1399  * In an inaccessible (PROT_NONE) VMA, pte_protnone() may indicate "yes". It is
1400  * perfectly valid to indicate "no" in that case, which is why our default
1401  * implementation defaults to "always no".
1402  *
1403  * In an accessible VMA, however, pte_protnone() reliably indicates PROT_NONE
1404  * page protection due to NUMA hinting. NUMA hinting faults only apply in
1405  * accessible VMAs.
1406  *
1407  * So, to reliably identify PROT_NONE PTEs that require a NUMA hinting fault,
1408  * looking at the VMA accessibility is sufficient.
1409  */
1410 static inline int pte_protnone(pte_t pte)
1411 {
1412 	return 0;
1413 }
1414 
1415 static inline int pmd_protnone(pmd_t pmd)
1416 {
1417 	return 0;
1418 }
1419 #endif /* CONFIG_NUMA_BALANCING */
1420 
1421 #endif /* CONFIG_MMU */
1422 
1423 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1424 
1425 #ifndef __PAGETABLE_P4D_FOLDED
1426 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1427 void p4d_clear_huge(p4d_t *p4d);
1428 #else
1429 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1430 {
1431 	return 0;
1432 }
1433 static inline void p4d_clear_huge(p4d_t *p4d) { }
1434 #endif /* !__PAGETABLE_P4D_FOLDED */
1435 
1436 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1437 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1438 int pud_clear_huge(pud_t *pud);
1439 int pmd_clear_huge(pmd_t *pmd);
1440 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1441 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1442 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1443 #else	/* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1444 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1445 {
1446 	return 0;
1447 }
1448 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1449 {
1450 	return 0;
1451 }
1452 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1453 {
1454 	return 0;
1455 }
1456 static inline void p4d_clear_huge(p4d_t *p4d) { }
1457 static inline int pud_clear_huge(pud_t *pud)
1458 {
1459 	return 0;
1460 }
1461 static inline int pmd_clear_huge(pmd_t *pmd)
1462 {
1463 	return 0;
1464 }
1465 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1466 {
1467 	return 0;
1468 }
1469 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1470 {
1471 	return 0;
1472 }
1473 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1474 {
1475 	return 0;
1476 }
1477 #endif	/* CONFIG_HAVE_ARCH_HUGE_VMAP */
1478 
1479 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1480 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1481 /*
1482  * ARCHes with special requirements for evicting THP backing TLB entries can
1483  * implement this. Otherwise also, it can help optimize normal TLB flush in
1484  * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1485  * entire TLB if flush span is greater than a threshold, which will
1486  * likely be true for a single huge page. Thus a single THP flush will
1487  * invalidate the entire TLB which is not desirable.
1488  * e.g. see arch/arc: flush_pmd_tlb_range
1489  */
1490 #define flush_pmd_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1491 #define flush_pud_tlb_range(vma, addr, end)	flush_tlb_range(vma, addr, end)
1492 #else
1493 #define flush_pmd_tlb_range(vma, addr, end)	BUILD_BUG()
1494 #define flush_pud_tlb_range(vma, addr, end)	BUILD_BUG()
1495 #endif
1496 #endif
1497 
1498 struct file;
1499 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1500 			unsigned long size, pgprot_t *vma_prot);
1501 
1502 #ifndef CONFIG_X86_ESPFIX64
1503 static inline void init_espfix_bsp(void) { }
1504 #endif
1505 
1506 extern void __init pgtable_cache_init(void);
1507 
1508 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1509 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1510 {
1511 	return true;
1512 }
1513 
1514 static inline bool arch_has_pfn_modify_check(void)
1515 {
1516 	return false;
1517 }
1518 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1519 
1520 /*
1521  * Architecture PAGE_KERNEL_* fallbacks
1522  *
1523  * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1524  * because they really don't support them, or the port needs to be updated to
1525  * reflect the required functionality. Below are a set of relatively safe
1526  * fallbacks, as best effort, which we can count on in lieu of the architectures
1527  * not defining them on their own yet.
1528  */
1529 
1530 #ifndef PAGE_KERNEL_RO
1531 # define PAGE_KERNEL_RO PAGE_KERNEL
1532 #endif
1533 
1534 #ifndef PAGE_KERNEL_EXEC
1535 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1536 #endif
1537 
1538 /*
1539  * Page Table Modification bits for pgtbl_mod_mask.
1540  *
1541  * These are used by the p?d_alloc_track*() set of functions an in the generic
1542  * vmalloc/ioremap code to track at which page-table levels entries have been
1543  * modified. Based on that the code can better decide when vmalloc and ioremap
1544  * mapping changes need to be synchronized to other page-tables in the system.
1545  */
1546 #define		__PGTBL_PGD_MODIFIED	0
1547 #define		__PGTBL_P4D_MODIFIED	1
1548 #define		__PGTBL_PUD_MODIFIED	2
1549 #define		__PGTBL_PMD_MODIFIED	3
1550 #define		__PGTBL_PTE_MODIFIED	4
1551 
1552 #define		PGTBL_PGD_MODIFIED	BIT(__PGTBL_PGD_MODIFIED)
1553 #define		PGTBL_P4D_MODIFIED	BIT(__PGTBL_P4D_MODIFIED)
1554 #define		PGTBL_PUD_MODIFIED	BIT(__PGTBL_PUD_MODIFIED)
1555 #define		PGTBL_PMD_MODIFIED	BIT(__PGTBL_PMD_MODIFIED)
1556 #define		PGTBL_PTE_MODIFIED	BIT(__PGTBL_PTE_MODIFIED)
1557 
1558 /* Page-Table Modification Mask */
1559 typedef unsigned int pgtbl_mod_mask;
1560 
1561 #endif /* !__ASSEMBLY__ */
1562 
1563 #if !defined(MAX_POSSIBLE_PHYSMEM_BITS) && !defined(CONFIG_64BIT)
1564 #ifdef CONFIG_PHYS_ADDR_T_64BIT
1565 /*
1566  * ZSMALLOC needs to know the highest PFN on 32-bit architectures
1567  * with physical address space extension, but falls back to
1568  * BITS_PER_LONG otherwise.
1569  */
1570 #error Missing MAX_POSSIBLE_PHYSMEM_BITS definition
1571 #else
1572 #define MAX_POSSIBLE_PHYSMEM_BITS 32
1573 #endif
1574 #endif
1575 
1576 #ifndef has_transparent_hugepage
1577 #define has_transparent_hugepage() IS_BUILTIN(CONFIG_TRANSPARENT_HUGEPAGE)
1578 #endif
1579 
1580 #ifndef has_transparent_pud_hugepage
1581 #define has_transparent_pud_hugepage() IS_BUILTIN(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1582 #endif
1583 /*
1584  * On some architectures it depends on the mm if the p4d/pud or pmd
1585  * layer of the page table hierarchy is folded or not.
1586  */
1587 #ifndef mm_p4d_folded
1588 #define mm_p4d_folded(mm)	__is_defined(__PAGETABLE_P4D_FOLDED)
1589 #endif
1590 
1591 #ifndef mm_pud_folded
1592 #define mm_pud_folded(mm)	__is_defined(__PAGETABLE_PUD_FOLDED)
1593 #endif
1594 
1595 #ifndef mm_pmd_folded
1596 #define mm_pmd_folded(mm)	__is_defined(__PAGETABLE_PMD_FOLDED)
1597 #endif
1598 
1599 #ifndef p4d_offset_lockless
1600 #define p4d_offset_lockless(pgdp, pgd, address) p4d_offset(&(pgd), address)
1601 #endif
1602 #ifndef pud_offset_lockless
1603 #define pud_offset_lockless(p4dp, p4d, address) pud_offset(&(p4d), address)
1604 #endif
1605 #ifndef pmd_offset_lockless
1606 #define pmd_offset_lockless(pudp, pud, address) pmd_offset(&(pud), address)
1607 #endif
1608 
1609 /*
1610  * p?d_leaf() - true if this entry is a final mapping to a physical address.
1611  * This differs from p?d_huge() by the fact that they are always available (if
1612  * the architecture supports large pages at the appropriate level) even
1613  * if CONFIG_HUGETLB_PAGE is not defined.
1614  * Only meaningful when called on a valid entry.
1615  */
1616 #ifndef pgd_leaf
1617 #define pgd_leaf(x)	0
1618 #endif
1619 #ifndef p4d_leaf
1620 #define p4d_leaf(x)	0
1621 #endif
1622 #ifndef pud_leaf
1623 #define pud_leaf(x)	0
1624 #endif
1625 #ifndef pmd_leaf
1626 #define pmd_leaf(x)	0
1627 #endif
1628 
1629 #ifndef pgd_leaf_size
1630 #define pgd_leaf_size(x) (1ULL << PGDIR_SHIFT)
1631 #endif
1632 #ifndef p4d_leaf_size
1633 #define p4d_leaf_size(x) P4D_SIZE
1634 #endif
1635 #ifndef pud_leaf_size
1636 #define pud_leaf_size(x) PUD_SIZE
1637 #endif
1638 #ifndef pmd_leaf_size
1639 #define pmd_leaf_size(x) PMD_SIZE
1640 #endif
1641 #ifndef pte_leaf_size
1642 #define pte_leaf_size(x) PAGE_SIZE
1643 #endif
1644 
1645 /*
1646  * Some architectures have MMUs that are configurable or selectable at boot
1647  * time. These lead to variable PTRS_PER_x. For statically allocated arrays it
1648  * helps to have a static maximum value.
1649  */
1650 
1651 #ifndef MAX_PTRS_PER_PTE
1652 #define MAX_PTRS_PER_PTE PTRS_PER_PTE
1653 #endif
1654 
1655 #ifndef MAX_PTRS_PER_PMD
1656 #define MAX_PTRS_PER_PMD PTRS_PER_PMD
1657 #endif
1658 
1659 #ifndef MAX_PTRS_PER_PUD
1660 #define MAX_PTRS_PER_PUD PTRS_PER_PUD
1661 #endif
1662 
1663 #ifndef MAX_PTRS_PER_P4D
1664 #define MAX_PTRS_PER_P4D PTRS_PER_P4D
1665 #endif
1666 
1667 /* description of effects of mapping type and prot in current implementation.
1668  * this is due to the limited x86 page protection hardware.  The expected
1669  * behavior is in parens:
1670  *
1671  * map_type	prot
1672  *		PROT_NONE	PROT_READ	PROT_WRITE	PROT_EXEC
1673  * MAP_SHARED	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1674  *		w: (no) no	w: (no) no	w: (yes) yes	w: (no) no
1675  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1676  *
1677  * MAP_PRIVATE	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
1678  *		w: (no) no	w: (no) no	w: (copy) copy	w: (no) no
1679  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
1680  *
1681  * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
1682  * MAP_PRIVATE (with Enhanced PAN supported):
1683  *								r: (no) no
1684  *								w: (no) no
1685  *								x: (yes) yes
1686  */
1687 #define DECLARE_VM_GET_PAGE_PROT					\
1688 pgprot_t vm_get_page_prot(unsigned long vm_flags)			\
1689 {									\
1690 		return protection_map[vm_flags &			\
1691 			(VM_READ | VM_WRITE | VM_EXEC | VM_SHARED)];	\
1692 }									\
1693 EXPORT_SYMBOL(vm_get_page_prot);
1694 
1695 #endif /* _LINUX_PGTABLE_H */
1696