xref: /linux/arch/arm64/include/asm/pgtable.h (revision 113d1dd9c8ea2186d56a641a787e2588673c9c32)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Copyright (C) 2012 ARM Ltd.
4  */
5 #ifndef __ASM_PGTABLE_H
6 #define __ASM_PGTABLE_H
7 
8 #include <asm/bug.h>
9 #include <asm/proc-fns.h>
10 
11 #include <asm/memory.h>
12 #include <asm/mte.h>
13 #include <asm/pgtable-hwdef.h>
14 #include <asm/pgtable-prot.h>
15 #include <asm/tlbflush.h>
16 
17 /*
18  * VMALLOC range.
19  *
20  * VMALLOC_START: beginning of the kernel vmalloc space
21  * VMALLOC_END: extends to the available space below vmemmap
22  */
23 #define VMALLOC_START		(MODULES_END)
24 #if VA_BITS == VA_BITS_MIN
25 #define VMALLOC_END		(VMEMMAP_START - SZ_8M)
26 #else
27 #define VMEMMAP_UNUSED_NPAGES	((_PAGE_OFFSET(vabits_actual) - PAGE_OFFSET) >> PAGE_SHIFT)
28 #define VMALLOC_END		(VMEMMAP_START + VMEMMAP_UNUSED_NPAGES * sizeof(struct page) - SZ_8M)
29 #endif
30 
31 #define vmemmap			((struct page *)VMEMMAP_START - (memstart_addr >> PAGE_SHIFT))
32 
33 #ifndef __ASSEMBLY__
34 
35 #include <asm/cmpxchg.h>
36 #include <asm/fixmap.h>
37 #include <linux/mmdebug.h>
38 #include <linux/mm_types.h>
39 #include <linux/sched.h>
40 #include <linux/page_table_check.h>
41 
42 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
43 #define __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
44 
45 /* Set stride and tlb_level in flush_*_tlb_range */
46 #define flush_pmd_tlb_range(vma, addr, end)	\
47 	__flush_tlb_range(vma, addr, end, PMD_SIZE, false, 2)
48 #define flush_pud_tlb_range(vma, addr, end)	\
49 	__flush_tlb_range(vma, addr, end, PUD_SIZE, false, 1)
50 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
51 
52 static inline bool arch_thp_swp_supported(void)
53 {
54 	return !system_supports_mte();
55 }
56 #define arch_thp_swp_supported arch_thp_swp_supported
57 
58 /*
59  * Outside of a few very special situations (e.g. hibernation), we always
60  * use broadcast TLB invalidation instructions, therefore a spurious page
61  * fault on one CPU which has been handled concurrently by another CPU
62  * does not need to perform additional invalidation.
63  */
64 #define flush_tlb_fix_spurious_fault(vma, address, ptep) do { } while (0)
65 
66 /*
67  * ZERO_PAGE is a global shared page that is always zero: used
68  * for zero-mapped memory areas etc..
69  */
70 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
71 #define ZERO_PAGE(vaddr)	phys_to_page(__pa_symbol(empty_zero_page))
72 
73 #define pte_ERROR(e)	\
74 	pr_err("%s:%d: bad pte %016llx.\n", __FILE__, __LINE__, pte_val(e))
75 
76 /*
77  * Macros to convert between a physical address and its placement in a
78  * page table entry, taking care of 52-bit addresses.
79  */
80 #ifdef CONFIG_ARM64_PA_BITS_52
81 static inline phys_addr_t __pte_to_phys(pte_t pte)
82 {
83 	pte_val(pte) &= ~PTE_MAYBE_SHARED;
84 	return (pte_val(pte) & PTE_ADDR_LOW) |
85 		((pte_val(pte) & PTE_ADDR_HIGH) << PTE_ADDR_HIGH_SHIFT);
86 }
87 static inline pteval_t __phys_to_pte_val(phys_addr_t phys)
88 {
89 	return (phys | (phys >> PTE_ADDR_HIGH_SHIFT)) & PHYS_TO_PTE_ADDR_MASK;
90 }
91 #else
92 #define __pte_to_phys(pte)	(pte_val(pte) & PTE_ADDR_LOW)
93 #define __phys_to_pte_val(phys)	(phys)
94 #endif
95 
96 #define pte_pfn(pte)		(__pte_to_phys(pte) >> PAGE_SHIFT)
97 #define pfn_pte(pfn,prot)	\
98 	__pte(__phys_to_pte_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
99 
100 #define pte_none(pte)		(!pte_val(pte))
101 #define __pte_clear(mm, addr, ptep) \
102 				__set_pte(ptep, __pte(0))
103 #define pte_page(pte)		(pfn_to_page(pte_pfn(pte)))
104 
105 /*
106  * The following only work if pte_present(). Undefined behaviour otherwise.
107  */
108 #define pte_present(pte)	(pte_valid(pte) || pte_present_invalid(pte))
109 #define pte_young(pte)		(!!(pte_val(pte) & PTE_AF))
110 #define pte_special(pte)	(!!(pte_val(pte) & PTE_SPECIAL))
111 #define pte_write(pte)		(!!(pte_val(pte) & PTE_WRITE))
112 #define pte_rdonly(pte)		(!!(pte_val(pte) & PTE_RDONLY))
113 #define pte_user(pte)		(!!(pte_val(pte) & PTE_USER))
114 #define pte_user_exec(pte)	(!(pte_val(pte) & PTE_UXN))
115 #define pte_cont(pte)		(!!(pte_val(pte) & PTE_CONT))
116 #define pte_devmap(pte)		(!!(pte_val(pte) & PTE_DEVMAP))
117 #define pte_tagged(pte)		((pte_val(pte) & PTE_ATTRINDX_MASK) == \
118 				 PTE_ATTRINDX(MT_NORMAL_TAGGED))
119 
120 #define pte_cont_addr_end(addr, end)						\
121 ({	unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK;	\
122 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
123 })
124 
125 #define pmd_cont_addr_end(addr, end)						\
126 ({	unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK;	\
127 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
128 })
129 
130 #define pte_hw_dirty(pte)	(pte_write(pte) && !pte_rdonly(pte))
131 #define pte_sw_dirty(pte)	(!!(pte_val(pte) & PTE_DIRTY))
132 #define pte_dirty(pte)		(pte_sw_dirty(pte) || pte_hw_dirty(pte))
133 
134 #define pte_valid(pte)		(!!(pte_val(pte) & PTE_VALID))
135 #define pte_present_invalid(pte) \
136 	((pte_val(pte) & (PTE_VALID | PTE_PRESENT_INVALID)) == PTE_PRESENT_INVALID)
137 /*
138  * Execute-only user mappings do not have the PTE_USER bit set. All valid
139  * kernel mappings have the PTE_UXN bit set.
140  */
141 #define pte_valid_not_user(pte) \
142 	((pte_val(pte) & (PTE_VALID | PTE_USER | PTE_UXN)) == (PTE_VALID | PTE_UXN))
143 /*
144  * Returns true if the pte is valid and has the contiguous bit set.
145  */
146 #define pte_valid_cont(pte)	(pte_valid(pte) && pte_cont(pte))
147 /*
148  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
149  * so that we don't erroneously return false for pages that have been
150  * remapped as PROT_NONE but are yet to be flushed from the TLB.
151  * Note that we can't make any assumptions based on the state of the access
152  * flag, since __ptep_clear_flush_young() elides a DSB when invalidating the
153  * TLB.
154  */
155 #define pte_accessible(mm, pte)	\
156 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid(pte))
157 
158 /*
159  * p??_access_permitted() is true for valid user mappings (PTE_USER
160  * bit set, subject to the write permission check). For execute-only
161  * mappings, like PROT_EXEC with EPAN (both PTE_USER and PTE_UXN bits
162  * not set) must return false. PROT_NONE mappings do not have the
163  * PTE_VALID bit set.
164  */
165 #define pte_access_permitted(pte, write) \
166 	(((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER)) && (!(write) || pte_write(pte)))
167 #define pmd_access_permitted(pmd, write) \
168 	(pte_access_permitted(pmd_pte(pmd), (write)))
169 #define pud_access_permitted(pud, write) \
170 	(pte_access_permitted(pud_pte(pud), (write)))
171 
172 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
173 {
174 	pte_val(pte) &= ~pgprot_val(prot);
175 	return pte;
176 }
177 
178 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
179 {
180 	pte_val(pte) |= pgprot_val(prot);
181 	return pte;
182 }
183 
184 static inline pmd_t clear_pmd_bit(pmd_t pmd, pgprot_t prot)
185 {
186 	pmd_val(pmd) &= ~pgprot_val(prot);
187 	return pmd;
188 }
189 
190 static inline pmd_t set_pmd_bit(pmd_t pmd, pgprot_t prot)
191 {
192 	pmd_val(pmd) |= pgprot_val(prot);
193 	return pmd;
194 }
195 
196 static inline pte_t pte_mkwrite_novma(pte_t pte)
197 {
198 	pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
199 	pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
200 	return pte;
201 }
202 
203 static inline pte_t pte_mkclean(pte_t pte)
204 {
205 	pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
206 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
207 
208 	return pte;
209 }
210 
211 static inline pte_t pte_mkdirty(pte_t pte)
212 {
213 	pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
214 
215 	if (pte_write(pte))
216 		pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
217 
218 	return pte;
219 }
220 
221 static inline pte_t pte_wrprotect(pte_t pte)
222 {
223 	/*
224 	 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
225 	 * clear), set the PTE_DIRTY bit.
226 	 */
227 	if (pte_hw_dirty(pte))
228 		pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
229 
230 	pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
231 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
232 	return pte;
233 }
234 
235 static inline pte_t pte_mkold(pte_t pte)
236 {
237 	return clear_pte_bit(pte, __pgprot(PTE_AF));
238 }
239 
240 static inline pte_t pte_mkyoung(pte_t pte)
241 {
242 	return set_pte_bit(pte, __pgprot(PTE_AF));
243 }
244 
245 static inline pte_t pte_mkspecial(pte_t pte)
246 {
247 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
248 }
249 
250 static inline pte_t pte_mkcont(pte_t pte)
251 {
252 	pte = set_pte_bit(pte, __pgprot(PTE_CONT));
253 	return set_pte_bit(pte, __pgprot(PTE_TYPE_PAGE));
254 }
255 
256 static inline pte_t pte_mknoncont(pte_t pte)
257 {
258 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
259 }
260 
261 static inline pte_t pte_mkpresent(pte_t pte)
262 {
263 	return set_pte_bit(pte, __pgprot(PTE_VALID));
264 }
265 
266 static inline pte_t pte_mkinvalid(pte_t pte)
267 {
268 	pte = set_pte_bit(pte, __pgprot(PTE_PRESENT_INVALID));
269 	pte = clear_pte_bit(pte, __pgprot(PTE_VALID));
270 	return pte;
271 }
272 
273 static inline pmd_t pmd_mkcont(pmd_t pmd)
274 {
275 	return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
276 }
277 
278 static inline pte_t pte_mkdevmap(pte_t pte)
279 {
280 	return set_pte_bit(pte, __pgprot(PTE_DEVMAP | PTE_SPECIAL));
281 }
282 
283 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
284 static inline int pte_uffd_wp(pte_t pte)
285 {
286 	return !!(pte_val(pte) & PTE_UFFD_WP);
287 }
288 
289 static inline pte_t pte_mkuffd_wp(pte_t pte)
290 {
291 	return pte_wrprotect(set_pte_bit(pte, __pgprot(PTE_UFFD_WP)));
292 }
293 
294 static inline pte_t pte_clear_uffd_wp(pte_t pte)
295 {
296 	return clear_pte_bit(pte, __pgprot(PTE_UFFD_WP));
297 }
298 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
299 
300 static inline void __set_pte_nosync(pte_t *ptep, pte_t pte)
301 {
302 	WRITE_ONCE(*ptep, pte);
303 }
304 
305 static inline void __set_pte(pte_t *ptep, pte_t pte)
306 {
307 	__set_pte_nosync(ptep, pte);
308 
309 	/*
310 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
311 	 * or update_mmu_cache() have the necessary barriers.
312 	 */
313 	if (pte_valid_not_user(pte)) {
314 		dsb(ishst);
315 		isb();
316 	}
317 }
318 
319 static inline pte_t __ptep_get(pte_t *ptep)
320 {
321 	return READ_ONCE(*ptep);
322 }
323 
324 extern void __sync_icache_dcache(pte_t pteval);
325 bool pgattr_change_is_safe(u64 old, u64 new);
326 
327 /*
328  * PTE bits configuration in the presence of hardware Dirty Bit Management
329  * (PTE_WRITE == PTE_DBM):
330  *
331  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
332  *   0      0      |   1           0          0
333  *   0      1      |   1           1          0
334  *   1      0      |   1           0          1
335  *   1      1      |   0           1          x
336  *
337  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
338  * the page fault mechanism. Checking the dirty status of a pte becomes:
339  *
340  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
341  */
342 
343 static inline void __check_safe_pte_update(struct mm_struct *mm, pte_t *ptep,
344 					   pte_t pte)
345 {
346 	pte_t old_pte;
347 
348 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
349 		return;
350 
351 	old_pte = __ptep_get(ptep);
352 
353 	if (!pte_valid(old_pte) || !pte_valid(pte))
354 		return;
355 	if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1)
356 		return;
357 
358 	/*
359 	 * Check for potential race with hardware updates of the pte
360 	 * (__ptep_set_access_flags safely changes valid ptes without going
361 	 * through an invalid entry).
362 	 */
363 	VM_WARN_ONCE(!pte_young(pte),
364 		     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
365 		     __func__, pte_val(old_pte), pte_val(pte));
366 	VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
367 		     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
368 		     __func__, pte_val(old_pte), pte_val(pte));
369 	VM_WARN_ONCE(!pgattr_change_is_safe(pte_val(old_pte), pte_val(pte)),
370 		     "%s: unsafe attribute change: 0x%016llx -> 0x%016llx",
371 		     __func__, pte_val(old_pte), pte_val(pte));
372 }
373 
374 static inline void __sync_cache_and_tags(pte_t pte, unsigned int nr_pages)
375 {
376 	if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
377 		__sync_icache_dcache(pte);
378 
379 	/*
380 	 * If the PTE would provide user space access to the tags associated
381 	 * with it then ensure that the MTE tags are synchronised.  Although
382 	 * pte_access_permitted() returns false for exec only mappings, they
383 	 * don't expose tags (instruction fetches don't check tags).
384 	 */
385 	if (system_supports_mte() && pte_access_permitted(pte, false) &&
386 	    !pte_special(pte) && pte_tagged(pte))
387 		mte_sync_tags(pte, nr_pages);
388 }
389 
390 /*
391  * Select all bits except the pfn
392  */
393 static inline pgprot_t pte_pgprot(pte_t pte)
394 {
395 	unsigned long pfn = pte_pfn(pte);
396 
397 	return __pgprot(pte_val(pfn_pte(pfn, __pgprot(0))) ^ pte_val(pte));
398 }
399 
400 #define pte_advance_pfn pte_advance_pfn
401 static inline pte_t pte_advance_pfn(pte_t pte, unsigned long nr)
402 {
403 	return pfn_pte(pte_pfn(pte) + nr, pte_pgprot(pte));
404 }
405 
406 static inline void __set_ptes(struct mm_struct *mm,
407 			      unsigned long __always_unused addr,
408 			      pte_t *ptep, pte_t pte, unsigned int nr)
409 {
410 	page_table_check_ptes_set(mm, ptep, pte, nr);
411 	__sync_cache_and_tags(pte, nr);
412 
413 	for (;;) {
414 		__check_safe_pte_update(mm, ptep, pte);
415 		__set_pte(ptep, pte);
416 		if (--nr == 0)
417 			break;
418 		ptep++;
419 		pte = pte_advance_pfn(pte, 1);
420 	}
421 }
422 
423 /*
424  * Huge pte definitions.
425  */
426 #define pte_mkhuge(pte)		(__pte(pte_val(pte) & ~PTE_TABLE_BIT))
427 
428 /*
429  * Hugetlb definitions.
430  */
431 #define HUGE_MAX_HSTATE		4
432 #define HPAGE_SHIFT		PMD_SHIFT
433 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
434 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
435 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
436 
437 static inline pte_t pgd_pte(pgd_t pgd)
438 {
439 	return __pte(pgd_val(pgd));
440 }
441 
442 static inline pte_t p4d_pte(p4d_t p4d)
443 {
444 	return __pte(p4d_val(p4d));
445 }
446 
447 static inline pte_t pud_pte(pud_t pud)
448 {
449 	return __pte(pud_val(pud));
450 }
451 
452 static inline pud_t pte_pud(pte_t pte)
453 {
454 	return __pud(pte_val(pte));
455 }
456 
457 static inline pmd_t pud_pmd(pud_t pud)
458 {
459 	return __pmd(pud_val(pud));
460 }
461 
462 static inline pte_t pmd_pte(pmd_t pmd)
463 {
464 	return __pte(pmd_val(pmd));
465 }
466 
467 static inline pmd_t pte_pmd(pte_t pte)
468 {
469 	return __pmd(pte_val(pte));
470 }
471 
472 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot)
473 {
474 	return __pgprot((pgprot_val(prot) & ~PUD_TABLE_BIT) | PUD_TYPE_SECT);
475 }
476 
477 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot)
478 {
479 	return __pgprot((pgprot_val(prot) & ~PMD_TABLE_BIT) | PMD_TYPE_SECT);
480 }
481 
482 static inline pte_t pte_swp_mkexclusive(pte_t pte)
483 {
484 	return set_pte_bit(pte, __pgprot(PTE_SWP_EXCLUSIVE));
485 }
486 
487 static inline int pte_swp_exclusive(pte_t pte)
488 {
489 	return pte_val(pte) & PTE_SWP_EXCLUSIVE;
490 }
491 
492 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
493 {
494 	return clear_pte_bit(pte, __pgprot(PTE_SWP_EXCLUSIVE));
495 }
496 
497 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
498 static inline pte_t pte_swp_mkuffd_wp(pte_t pte)
499 {
500 	return set_pte_bit(pte, __pgprot(PTE_SWP_UFFD_WP));
501 }
502 
503 static inline int pte_swp_uffd_wp(pte_t pte)
504 {
505 	return !!(pte_val(pte) & PTE_SWP_UFFD_WP);
506 }
507 
508 static inline pte_t pte_swp_clear_uffd_wp(pte_t pte)
509 {
510 	return clear_pte_bit(pte, __pgprot(PTE_SWP_UFFD_WP));
511 }
512 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
513 
514 #ifdef CONFIG_NUMA_BALANCING
515 /*
516  * See the comment in include/linux/pgtable.h
517  */
518 static inline int pte_protnone(pte_t pte)
519 {
520 	/*
521 	 * pte_present_invalid() tells us that the pte is invalid from HW
522 	 * perspective but present from SW perspective, so the fields are to be
523 	 * interpretted as per the HW layout. The second 2 checks are the unique
524 	 * encoding that we use for PROT_NONE. It is insufficient to only use
525 	 * the first check because we share the same encoding scheme with pmds
526 	 * which support pmd_mkinvalid(), so can be present-invalid without
527 	 * being PROT_NONE.
528 	 */
529 	return pte_present_invalid(pte) && !pte_user(pte) && !pte_user_exec(pte);
530 }
531 
532 static inline int pmd_protnone(pmd_t pmd)
533 {
534 	return pte_protnone(pmd_pte(pmd));
535 }
536 #endif
537 
538 #define pmd_present(pmd)	pte_present(pmd_pte(pmd))
539 
540 /*
541  * THP definitions.
542  */
543 
544 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
545 static inline int pmd_trans_huge(pmd_t pmd)
546 {
547 	return pmd_val(pmd) && pmd_present(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT);
548 }
549 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
550 
551 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
552 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
553 #define pmd_valid(pmd)		pte_valid(pmd_pte(pmd))
554 #define pmd_user(pmd)		pte_user(pmd_pte(pmd))
555 #define pmd_user_exec(pmd)	pte_user_exec(pmd_pte(pmd))
556 #define pmd_cont(pmd)		pte_cont(pmd_pte(pmd))
557 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
558 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
559 #define pmd_mkwrite_novma(pmd)	pte_pmd(pte_mkwrite_novma(pmd_pte(pmd)))
560 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
561 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
562 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
563 #define pmd_mkinvalid(pmd)	pte_pmd(pte_mkinvalid(pmd_pte(pmd)))
564 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
565 #define pmd_uffd_wp(pmd)	pte_uffd_wp(pmd_pte(pmd))
566 #define pmd_mkuffd_wp(pmd)	pte_pmd(pte_mkuffd_wp(pmd_pte(pmd)))
567 #define pmd_clear_uffd_wp(pmd)	pte_pmd(pte_clear_uffd_wp(pmd_pte(pmd)))
568 #define pmd_swp_uffd_wp(pmd)	pte_swp_uffd_wp(pmd_pte(pmd))
569 #define pmd_swp_mkuffd_wp(pmd)	pte_pmd(pte_swp_mkuffd_wp(pmd_pte(pmd)))
570 #define pmd_swp_clear_uffd_wp(pmd) \
571 				pte_pmd(pte_swp_clear_uffd_wp(pmd_pte(pmd)))
572 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
573 
574 #define pmd_thp_or_huge(pmd)	(pmd_huge(pmd) || pmd_trans_huge(pmd))
575 
576 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
577 
578 #define pmd_mkhuge(pmd)		(__pmd(pmd_val(pmd) & ~PMD_TABLE_BIT))
579 
580 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
581 #define pmd_devmap(pmd)		pte_devmap(pmd_pte(pmd))
582 #endif
583 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
584 {
585 	return pte_pmd(set_pte_bit(pmd_pte(pmd), __pgprot(PTE_DEVMAP)));
586 }
587 
588 #define __pmd_to_phys(pmd)	__pte_to_phys(pmd_pte(pmd))
589 #define __phys_to_pmd_val(phys)	__phys_to_pte_val(phys)
590 #define pmd_pfn(pmd)		((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
591 #define pfn_pmd(pfn,prot)	__pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
592 #define mk_pmd(page,prot)	pfn_pmd(page_to_pfn(page),prot)
593 
594 #define pud_young(pud)		pte_young(pud_pte(pud))
595 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
596 #define pud_write(pud)		pte_write(pud_pte(pud))
597 
598 #define pud_mkhuge(pud)		(__pud(pud_val(pud) & ~PUD_TABLE_BIT))
599 
600 #define __pud_to_phys(pud)	__pte_to_phys(pud_pte(pud))
601 #define __phys_to_pud_val(phys)	__phys_to_pte_val(phys)
602 #define pud_pfn(pud)		((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
603 #define pfn_pud(pfn,prot)	__pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
604 
605 static inline void __set_pte_at(struct mm_struct *mm,
606 				unsigned long __always_unused addr,
607 				pte_t *ptep, pte_t pte, unsigned int nr)
608 {
609 	__sync_cache_and_tags(pte, nr);
610 	__check_safe_pte_update(mm, ptep, pte);
611 	__set_pte(ptep, pte);
612 }
613 
614 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
615 			      pmd_t *pmdp, pmd_t pmd)
616 {
617 	page_table_check_pmd_set(mm, pmdp, pmd);
618 	return __set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd),
619 						PMD_SIZE >> PAGE_SHIFT);
620 }
621 
622 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
623 			      pud_t *pudp, pud_t pud)
624 {
625 	page_table_check_pud_set(mm, pudp, pud);
626 	return __set_pte_at(mm, addr, (pte_t *)pudp, pud_pte(pud),
627 						PUD_SIZE >> PAGE_SHIFT);
628 }
629 
630 #define __p4d_to_phys(p4d)	__pte_to_phys(p4d_pte(p4d))
631 #define __phys_to_p4d_val(phys)	__phys_to_pte_val(phys)
632 
633 #define __pgd_to_phys(pgd)	__pte_to_phys(pgd_pte(pgd))
634 #define __phys_to_pgd_val(phys)	__phys_to_pte_val(phys)
635 
636 #define __pgprot_modify(prot,mask,bits) \
637 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
638 
639 #define pgprot_nx(prot) \
640 	__pgprot_modify(prot, PTE_MAYBE_GP, PTE_PXN)
641 
642 /*
643  * Mark the prot value as uncacheable and unbufferable.
644  */
645 #define pgprot_noncached(prot) \
646 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
647 #define pgprot_writecombine(prot) \
648 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
649 #define pgprot_device(prot) \
650 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
651 #define pgprot_tagged(prot) \
652 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_TAGGED))
653 #define pgprot_mhp	pgprot_tagged
654 /*
655  * DMA allocations for non-coherent devices use what the Arm architecture calls
656  * "Normal non-cacheable" memory, which permits speculation, unaligned accesses
657  * and merging of writes.  This is different from "Device-nGnR[nE]" memory which
658  * is intended for MMIO and thus forbids speculation, preserves access size,
659  * requires strict alignment and can also force write responses to come from the
660  * endpoint.
661  */
662 #define pgprot_dmacoherent(prot) \
663 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, \
664 			PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
665 
666 #define __HAVE_PHYS_MEM_ACCESS_PROT
667 struct file;
668 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
669 				     unsigned long size, pgprot_t vma_prot);
670 
671 #define pmd_none(pmd)		(!pmd_val(pmd))
672 
673 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
674 				 PMD_TYPE_TABLE)
675 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
676 				 PMD_TYPE_SECT)
677 #define pmd_leaf(pmd)		(pmd_present(pmd) && !pmd_table(pmd))
678 #define pmd_bad(pmd)		(!pmd_table(pmd))
679 
680 #define pmd_leaf_size(pmd)	(pmd_cont(pmd) ? CONT_PMD_SIZE : PMD_SIZE)
681 #define pte_leaf_size(pte)	(pte_cont(pte) ? CONT_PTE_SIZE : PAGE_SIZE)
682 
683 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
684 static inline bool pud_sect(pud_t pud) { return false; }
685 static inline bool pud_table(pud_t pud) { return true; }
686 #else
687 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
688 				 PUD_TYPE_SECT)
689 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
690 				 PUD_TYPE_TABLE)
691 #endif
692 
693 extern pgd_t init_pg_dir[];
694 extern pgd_t init_pg_end[];
695 extern pgd_t swapper_pg_dir[];
696 extern pgd_t idmap_pg_dir[];
697 extern pgd_t tramp_pg_dir[];
698 extern pgd_t reserved_pg_dir[];
699 
700 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
701 
702 static inline bool in_swapper_pgdir(void *addr)
703 {
704 	return ((unsigned long)addr & PAGE_MASK) ==
705 	        ((unsigned long)swapper_pg_dir & PAGE_MASK);
706 }
707 
708 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
709 {
710 #ifdef __PAGETABLE_PMD_FOLDED
711 	if (in_swapper_pgdir(pmdp)) {
712 		set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
713 		return;
714 	}
715 #endif /* __PAGETABLE_PMD_FOLDED */
716 
717 	WRITE_ONCE(*pmdp, pmd);
718 
719 	if (pmd_valid(pmd)) {
720 		dsb(ishst);
721 		isb();
722 	}
723 }
724 
725 static inline void pmd_clear(pmd_t *pmdp)
726 {
727 	set_pmd(pmdp, __pmd(0));
728 }
729 
730 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
731 {
732 	return __pmd_to_phys(pmd);
733 }
734 
735 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
736 {
737 	return (unsigned long)__va(pmd_page_paddr(pmd));
738 }
739 
740 /* Find an entry in the third-level page table. */
741 #define pte_offset_phys(dir,addr)	(pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
742 
743 #define pte_set_fixmap(addr)		((pte_t *)set_fixmap_offset(FIX_PTE, addr))
744 #define pte_set_fixmap_offset(pmd, addr)	pte_set_fixmap(pte_offset_phys(pmd, addr))
745 #define pte_clear_fixmap()		clear_fixmap(FIX_PTE)
746 
747 #define pmd_page(pmd)			phys_to_page(__pmd_to_phys(pmd))
748 
749 /* use ONLY for statically allocated translation tables */
750 #define pte_offset_kimg(dir,addr)	((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
751 
752 /*
753  * Conversion functions: convert a page and protection to a page entry,
754  * and a page entry and page directory to the page they refer to.
755  */
756 #define mk_pte(page,prot)	pfn_pte(page_to_pfn(page),prot)
757 
758 #if CONFIG_PGTABLE_LEVELS > 2
759 
760 #define pmd_ERROR(e)	\
761 	pr_err("%s:%d: bad pmd %016llx.\n", __FILE__, __LINE__, pmd_val(e))
762 
763 #define pud_none(pud)		(!pud_val(pud))
764 #define pud_bad(pud)		(!pud_table(pud))
765 #define pud_present(pud)	pte_present(pud_pte(pud))
766 #define pud_leaf(pud)		(pud_present(pud) && !pud_table(pud))
767 #define pud_valid(pud)		pte_valid(pud_pte(pud))
768 #define pud_user(pud)		pte_user(pud_pte(pud))
769 #define pud_user_exec(pud)	pte_user_exec(pud_pte(pud))
770 
771 static inline bool pgtable_l4_enabled(void);
772 
773 static inline void set_pud(pud_t *pudp, pud_t pud)
774 {
775 	if (!pgtable_l4_enabled() && in_swapper_pgdir(pudp)) {
776 		set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
777 		return;
778 	}
779 
780 	WRITE_ONCE(*pudp, pud);
781 
782 	if (pud_valid(pud)) {
783 		dsb(ishst);
784 		isb();
785 	}
786 }
787 
788 static inline void pud_clear(pud_t *pudp)
789 {
790 	set_pud(pudp, __pud(0));
791 }
792 
793 static inline phys_addr_t pud_page_paddr(pud_t pud)
794 {
795 	return __pud_to_phys(pud);
796 }
797 
798 static inline pmd_t *pud_pgtable(pud_t pud)
799 {
800 	return (pmd_t *)__va(pud_page_paddr(pud));
801 }
802 
803 /* Find an entry in the second-level page table. */
804 #define pmd_offset_phys(dir, addr)	(pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
805 
806 #define pmd_set_fixmap(addr)		((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
807 #define pmd_set_fixmap_offset(pud, addr)	pmd_set_fixmap(pmd_offset_phys(pud, addr))
808 #define pmd_clear_fixmap()		clear_fixmap(FIX_PMD)
809 
810 #define pud_page(pud)			phys_to_page(__pud_to_phys(pud))
811 
812 /* use ONLY for statically allocated translation tables */
813 #define pmd_offset_kimg(dir,addr)	((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
814 
815 #else
816 
817 #define pud_valid(pud)		false
818 #define pud_page_paddr(pud)	({ BUILD_BUG(); 0; })
819 #define pud_user_exec(pud)	pud_user(pud) /* Always 0 with folding */
820 
821 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
822 #define pmd_set_fixmap(addr)		NULL
823 #define pmd_set_fixmap_offset(pudp, addr)	((pmd_t *)pudp)
824 #define pmd_clear_fixmap()
825 
826 #define pmd_offset_kimg(dir,addr)	((pmd_t *)dir)
827 
828 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
829 
830 #if CONFIG_PGTABLE_LEVELS > 3
831 
832 static __always_inline bool pgtable_l4_enabled(void)
833 {
834 	if (CONFIG_PGTABLE_LEVELS > 4 || !IS_ENABLED(CONFIG_ARM64_LPA2))
835 		return true;
836 	if (!alternative_has_cap_likely(ARM64_ALWAYS_BOOT))
837 		return vabits_actual == VA_BITS;
838 	return alternative_has_cap_unlikely(ARM64_HAS_VA52);
839 }
840 
841 static inline bool mm_pud_folded(const struct mm_struct *mm)
842 {
843 	return !pgtable_l4_enabled();
844 }
845 #define mm_pud_folded  mm_pud_folded
846 
847 #define pud_ERROR(e)	\
848 	pr_err("%s:%d: bad pud %016llx.\n", __FILE__, __LINE__, pud_val(e))
849 
850 #define p4d_none(p4d)		(pgtable_l4_enabled() && !p4d_val(p4d))
851 #define p4d_bad(p4d)		(pgtable_l4_enabled() && !(p4d_val(p4d) & 2))
852 #define p4d_present(p4d)	(!p4d_none(p4d))
853 
854 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
855 {
856 	if (in_swapper_pgdir(p4dp)) {
857 		set_swapper_pgd((pgd_t *)p4dp, __pgd(p4d_val(p4d)));
858 		return;
859 	}
860 
861 	WRITE_ONCE(*p4dp, p4d);
862 	dsb(ishst);
863 	isb();
864 }
865 
866 static inline void p4d_clear(p4d_t *p4dp)
867 {
868 	if (pgtable_l4_enabled())
869 		set_p4d(p4dp, __p4d(0));
870 }
871 
872 static inline phys_addr_t p4d_page_paddr(p4d_t p4d)
873 {
874 	return __p4d_to_phys(p4d);
875 }
876 
877 #define pud_index(addr)		(((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
878 
879 static inline pud_t *p4d_to_folded_pud(p4d_t *p4dp, unsigned long addr)
880 {
881 	return (pud_t *)PTR_ALIGN_DOWN(p4dp, PAGE_SIZE) + pud_index(addr);
882 }
883 
884 static inline pud_t *p4d_pgtable(p4d_t p4d)
885 {
886 	return (pud_t *)__va(p4d_page_paddr(p4d));
887 }
888 
889 static inline phys_addr_t pud_offset_phys(p4d_t *p4dp, unsigned long addr)
890 {
891 	BUG_ON(!pgtable_l4_enabled());
892 
893 	return p4d_page_paddr(READ_ONCE(*p4dp)) + pud_index(addr) * sizeof(pud_t);
894 }
895 
896 static inline
897 pud_t *pud_offset_lockless(p4d_t *p4dp, p4d_t p4d, unsigned long addr)
898 {
899 	if (!pgtable_l4_enabled())
900 		return p4d_to_folded_pud(p4dp, addr);
901 	return (pud_t *)__va(p4d_page_paddr(p4d)) + pud_index(addr);
902 }
903 #define pud_offset_lockless pud_offset_lockless
904 
905 static inline pud_t *pud_offset(p4d_t *p4dp, unsigned long addr)
906 {
907 	return pud_offset_lockless(p4dp, READ_ONCE(*p4dp), addr);
908 }
909 #define pud_offset	pud_offset
910 
911 static inline pud_t *pud_set_fixmap(unsigned long addr)
912 {
913 	if (!pgtable_l4_enabled())
914 		return NULL;
915 	return (pud_t *)set_fixmap_offset(FIX_PUD, addr);
916 }
917 
918 static inline pud_t *pud_set_fixmap_offset(p4d_t *p4dp, unsigned long addr)
919 {
920 	if (!pgtable_l4_enabled())
921 		return p4d_to_folded_pud(p4dp, addr);
922 	return pud_set_fixmap(pud_offset_phys(p4dp, addr));
923 }
924 
925 static inline void pud_clear_fixmap(void)
926 {
927 	if (pgtable_l4_enabled())
928 		clear_fixmap(FIX_PUD);
929 }
930 
931 /* use ONLY for statically allocated translation tables */
932 static inline pud_t *pud_offset_kimg(p4d_t *p4dp, u64 addr)
933 {
934 	if (!pgtable_l4_enabled())
935 		return p4d_to_folded_pud(p4dp, addr);
936 	return (pud_t *)__phys_to_kimg(pud_offset_phys(p4dp, addr));
937 }
938 
939 #define p4d_page(p4d)		pfn_to_page(__phys_to_pfn(__p4d_to_phys(p4d)))
940 
941 #else
942 
943 static inline bool pgtable_l4_enabled(void) { return false; }
944 
945 #define p4d_page_paddr(p4d)	({ BUILD_BUG(); 0;})
946 
947 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
948 #define pud_set_fixmap(addr)		NULL
949 #define pud_set_fixmap_offset(pgdp, addr)	((pud_t *)pgdp)
950 #define pud_clear_fixmap()
951 
952 #define pud_offset_kimg(dir,addr)	((pud_t *)dir)
953 
954 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
955 
956 #if CONFIG_PGTABLE_LEVELS > 4
957 
958 static __always_inline bool pgtable_l5_enabled(void)
959 {
960 	if (!alternative_has_cap_likely(ARM64_ALWAYS_BOOT))
961 		return vabits_actual == VA_BITS;
962 	return alternative_has_cap_unlikely(ARM64_HAS_VA52);
963 }
964 
965 static inline bool mm_p4d_folded(const struct mm_struct *mm)
966 {
967 	return !pgtable_l5_enabled();
968 }
969 #define mm_p4d_folded  mm_p4d_folded
970 
971 #define p4d_ERROR(e)	\
972 	pr_err("%s:%d: bad p4d %016llx.\n", __FILE__, __LINE__, p4d_val(e))
973 
974 #define pgd_none(pgd)		(pgtable_l5_enabled() && !pgd_val(pgd))
975 #define pgd_bad(pgd)		(pgtable_l5_enabled() && !(pgd_val(pgd) & 2))
976 #define pgd_present(pgd)	(!pgd_none(pgd))
977 
978 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
979 {
980 	if (in_swapper_pgdir(pgdp)) {
981 		set_swapper_pgd(pgdp, __pgd(pgd_val(pgd)));
982 		return;
983 	}
984 
985 	WRITE_ONCE(*pgdp, pgd);
986 	dsb(ishst);
987 	isb();
988 }
989 
990 static inline void pgd_clear(pgd_t *pgdp)
991 {
992 	if (pgtable_l5_enabled())
993 		set_pgd(pgdp, __pgd(0));
994 }
995 
996 static inline phys_addr_t pgd_page_paddr(pgd_t pgd)
997 {
998 	return __pgd_to_phys(pgd);
999 }
1000 
1001 #define p4d_index(addr)		(((addr) >> P4D_SHIFT) & (PTRS_PER_P4D - 1))
1002 
1003 static inline p4d_t *pgd_to_folded_p4d(pgd_t *pgdp, unsigned long addr)
1004 {
1005 	return (p4d_t *)PTR_ALIGN_DOWN(pgdp, PAGE_SIZE) + p4d_index(addr);
1006 }
1007 
1008 static inline phys_addr_t p4d_offset_phys(pgd_t *pgdp, unsigned long addr)
1009 {
1010 	BUG_ON(!pgtable_l5_enabled());
1011 
1012 	return pgd_page_paddr(READ_ONCE(*pgdp)) + p4d_index(addr) * sizeof(p4d_t);
1013 }
1014 
1015 static inline
1016 p4d_t *p4d_offset_lockless(pgd_t *pgdp, pgd_t pgd, unsigned long addr)
1017 {
1018 	if (!pgtable_l5_enabled())
1019 		return pgd_to_folded_p4d(pgdp, addr);
1020 	return (p4d_t *)__va(pgd_page_paddr(pgd)) + p4d_index(addr);
1021 }
1022 #define p4d_offset_lockless p4d_offset_lockless
1023 
1024 static inline p4d_t *p4d_offset(pgd_t *pgdp, unsigned long addr)
1025 {
1026 	return p4d_offset_lockless(pgdp, READ_ONCE(*pgdp), addr);
1027 }
1028 
1029 static inline p4d_t *p4d_set_fixmap(unsigned long addr)
1030 {
1031 	if (!pgtable_l5_enabled())
1032 		return NULL;
1033 	return (p4d_t *)set_fixmap_offset(FIX_P4D, addr);
1034 }
1035 
1036 static inline p4d_t *p4d_set_fixmap_offset(pgd_t *pgdp, unsigned long addr)
1037 {
1038 	if (!pgtable_l5_enabled())
1039 		return pgd_to_folded_p4d(pgdp, addr);
1040 	return p4d_set_fixmap(p4d_offset_phys(pgdp, addr));
1041 }
1042 
1043 static inline void p4d_clear_fixmap(void)
1044 {
1045 	if (pgtable_l5_enabled())
1046 		clear_fixmap(FIX_P4D);
1047 }
1048 
1049 /* use ONLY for statically allocated translation tables */
1050 static inline p4d_t *p4d_offset_kimg(pgd_t *pgdp, u64 addr)
1051 {
1052 	if (!pgtable_l5_enabled())
1053 		return pgd_to_folded_p4d(pgdp, addr);
1054 	return (p4d_t *)__phys_to_kimg(p4d_offset_phys(pgdp, addr));
1055 }
1056 
1057 #define pgd_page(pgd)		pfn_to_page(__phys_to_pfn(__pgd_to_phys(pgd)))
1058 
1059 #else
1060 
1061 static inline bool pgtable_l5_enabled(void) { return false; }
1062 
1063 #define p4d_index(addr)		(((addr) >> P4D_SHIFT) & (PTRS_PER_P4D - 1))
1064 
1065 /* Match p4d_offset folding in <asm/generic/pgtable-nop4d.h> */
1066 #define p4d_set_fixmap(addr)		NULL
1067 #define p4d_set_fixmap_offset(p4dp, addr)	((p4d_t *)p4dp)
1068 #define p4d_clear_fixmap()
1069 
1070 #define p4d_offset_kimg(dir,addr)	((p4d_t *)dir)
1071 
1072 #endif  /* CONFIG_PGTABLE_LEVELS > 4 */
1073 
1074 #define pgd_ERROR(e)	\
1075 	pr_err("%s:%d: bad pgd %016llx.\n", __FILE__, __LINE__, pgd_val(e))
1076 
1077 #define pgd_set_fixmap(addr)	((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
1078 #define pgd_clear_fixmap()	clear_fixmap(FIX_PGD)
1079 
1080 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
1081 {
1082 	/*
1083 	 * Normal and Normal-Tagged are two different memory types and indices
1084 	 * in MAIR_EL1. The mask below has to include PTE_ATTRINDX_MASK.
1085 	 */
1086 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
1087 			      PTE_PRESENT_INVALID | PTE_VALID | PTE_WRITE |
1088 			      PTE_GP | PTE_ATTRINDX_MASK;
1089 	/* preserve the hardware dirty information */
1090 	if (pte_hw_dirty(pte))
1091 		pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
1092 
1093 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
1094 	/*
1095 	 * If we end up clearing hw dirtiness for a sw-dirty PTE, set hardware
1096 	 * dirtiness again.
1097 	 */
1098 	if (pte_sw_dirty(pte))
1099 		pte = pte_mkdirty(pte);
1100 	return pte;
1101 }
1102 
1103 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1104 {
1105 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
1106 }
1107 
1108 extern int __ptep_set_access_flags(struct vm_area_struct *vma,
1109 				 unsigned long address, pte_t *ptep,
1110 				 pte_t entry, int dirty);
1111 
1112 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1113 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1114 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
1115 					unsigned long address, pmd_t *pmdp,
1116 					pmd_t entry, int dirty)
1117 {
1118 	return __ptep_set_access_flags(vma, address, (pte_t *)pmdp,
1119 							pmd_pte(entry), dirty);
1120 }
1121 
1122 static inline int pud_devmap(pud_t pud)
1123 {
1124 	return 0;
1125 }
1126 
1127 static inline int pgd_devmap(pgd_t pgd)
1128 {
1129 	return 0;
1130 }
1131 #endif
1132 
1133 #ifdef CONFIG_PAGE_TABLE_CHECK
1134 static inline bool pte_user_accessible_page(pte_t pte)
1135 {
1136 	return pte_valid(pte) && (pte_user(pte) || pte_user_exec(pte));
1137 }
1138 
1139 static inline bool pmd_user_accessible_page(pmd_t pmd)
1140 {
1141 	return pmd_valid(pmd) && !pmd_table(pmd) && (pmd_user(pmd) || pmd_user_exec(pmd));
1142 }
1143 
1144 static inline bool pud_user_accessible_page(pud_t pud)
1145 {
1146 	return pud_valid(pud) && !pud_table(pud) && (pud_user(pud) || pud_user_exec(pud));
1147 }
1148 #endif
1149 
1150 /*
1151  * Atomic pte/pmd modifications.
1152  */
1153 static inline int __ptep_test_and_clear_young(struct vm_area_struct *vma,
1154 					      unsigned long address,
1155 					      pte_t *ptep)
1156 {
1157 	pte_t old_pte, pte;
1158 
1159 	pte = __ptep_get(ptep);
1160 	do {
1161 		old_pte = pte;
1162 		pte = pte_mkold(pte);
1163 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
1164 					       pte_val(old_pte), pte_val(pte));
1165 	} while (pte_val(pte) != pte_val(old_pte));
1166 
1167 	return pte_young(pte);
1168 }
1169 
1170 static inline int __ptep_clear_flush_young(struct vm_area_struct *vma,
1171 					 unsigned long address, pte_t *ptep)
1172 {
1173 	int young = __ptep_test_and_clear_young(vma, address, ptep);
1174 
1175 	if (young) {
1176 		/*
1177 		 * We can elide the trailing DSB here since the worst that can
1178 		 * happen is that a CPU continues to use the young entry in its
1179 		 * TLB and we mistakenly reclaim the associated page. The
1180 		 * window for such an event is bounded by the next
1181 		 * context-switch, which provides a DSB to complete the TLB
1182 		 * invalidation.
1183 		 */
1184 		flush_tlb_page_nosync(vma, address);
1185 	}
1186 
1187 	return young;
1188 }
1189 
1190 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1191 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1192 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1193 					    unsigned long address,
1194 					    pmd_t *pmdp)
1195 {
1196 	return __ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
1197 }
1198 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1199 
1200 static inline pte_t __ptep_get_and_clear(struct mm_struct *mm,
1201 				       unsigned long address, pte_t *ptep)
1202 {
1203 	pte_t pte = __pte(xchg_relaxed(&pte_val(*ptep), 0));
1204 
1205 	page_table_check_pte_clear(mm, pte);
1206 
1207 	return pte;
1208 }
1209 
1210 static inline void __clear_full_ptes(struct mm_struct *mm, unsigned long addr,
1211 				pte_t *ptep, unsigned int nr, int full)
1212 {
1213 	for (;;) {
1214 		__ptep_get_and_clear(mm, addr, ptep);
1215 		if (--nr == 0)
1216 			break;
1217 		ptep++;
1218 		addr += PAGE_SIZE;
1219 	}
1220 }
1221 
1222 static inline pte_t __get_and_clear_full_ptes(struct mm_struct *mm,
1223 				unsigned long addr, pte_t *ptep,
1224 				unsigned int nr, int full)
1225 {
1226 	pte_t pte, tmp_pte;
1227 
1228 	pte = __ptep_get_and_clear(mm, addr, ptep);
1229 	while (--nr) {
1230 		ptep++;
1231 		addr += PAGE_SIZE;
1232 		tmp_pte = __ptep_get_and_clear(mm, addr, ptep);
1233 		if (pte_dirty(tmp_pte))
1234 			pte = pte_mkdirty(pte);
1235 		if (pte_young(tmp_pte))
1236 			pte = pte_mkyoung(pte);
1237 	}
1238 	return pte;
1239 }
1240 
1241 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1242 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1243 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
1244 					    unsigned long address, pmd_t *pmdp)
1245 {
1246 	pmd_t pmd = __pmd(xchg_relaxed(&pmd_val(*pmdp), 0));
1247 
1248 	page_table_check_pmd_clear(mm, pmd);
1249 
1250 	return pmd;
1251 }
1252 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1253 
1254 static inline void ___ptep_set_wrprotect(struct mm_struct *mm,
1255 					unsigned long address, pte_t *ptep,
1256 					pte_t pte)
1257 {
1258 	pte_t old_pte;
1259 
1260 	do {
1261 		old_pte = pte;
1262 		pte = pte_wrprotect(pte);
1263 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
1264 					       pte_val(old_pte), pte_val(pte));
1265 	} while (pte_val(pte) != pte_val(old_pte));
1266 }
1267 
1268 /*
1269  * __ptep_set_wrprotect - mark read-only while trasferring potential hardware
1270  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
1271  */
1272 static inline void __ptep_set_wrprotect(struct mm_struct *mm,
1273 					unsigned long address, pte_t *ptep)
1274 {
1275 	___ptep_set_wrprotect(mm, address, ptep, __ptep_get(ptep));
1276 }
1277 
1278 static inline void __wrprotect_ptes(struct mm_struct *mm, unsigned long address,
1279 				pte_t *ptep, unsigned int nr)
1280 {
1281 	unsigned int i;
1282 
1283 	for (i = 0; i < nr; i++, address += PAGE_SIZE, ptep++)
1284 		__ptep_set_wrprotect(mm, address, ptep);
1285 }
1286 
1287 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1288 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1289 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1290 				      unsigned long address, pmd_t *pmdp)
1291 {
1292 	__ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
1293 }
1294 
1295 #define pmdp_establish pmdp_establish
1296 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
1297 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
1298 {
1299 	page_table_check_pmd_set(vma->vm_mm, pmdp, pmd);
1300 	return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
1301 }
1302 #endif
1303 
1304 /*
1305  * Encode and decode a swap entry:
1306  *	bits 0-1:	present (must be zero)
1307  *	bits 2:		remember PG_anon_exclusive
1308  *	bit  3:		remember uffd-wp state
1309  *	bits 6-10:	swap type
1310  *	bit  11:	PTE_PRESENT_INVALID (must be zero)
1311  *	bits 12-61:	swap offset
1312  */
1313 #define __SWP_TYPE_SHIFT	6
1314 #define __SWP_TYPE_BITS		5
1315 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
1316 #define __SWP_OFFSET_SHIFT	12
1317 #define __SWP_OFFSET_BITS	50
1318 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
1319 
1320 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
1321 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
1322 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
1323 
1324 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1325 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
1326 
1327 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1328 #define __pmd_to_swp_entry(pmd)		((swp_entry_t) { pmd_val(pmd) })
1329 #define __swp_entry_to_pmd(swp)		__pmd((swp).val)
1330 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
1331 
1332 /*
1333  * Ensure that there are not more swap files than can be encoded in the kernel
1334  * PTEs.
1335  */
1336 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
1337 
1338 #ifdef CONFIG_ARM64_MTE
1339 
1340 #define __HAVE_ARCH_PREPARE_TO_SWAP
1341 static inline int arch_prepare_to_swap(struct page *page)
1342 {
1343 	if (system_supports_mte())
1344 		return mte_save_tags(page);
1345 	return 0;
1346 }
1347 
1348 #define __HAVE_ARCH_SWAP_INVALIDATE
1349 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
1350 {
1351 	if (system_supports_mte())
1352 		mte_invalidate_tags(type, offset);
1353 }
1354 
1355 static inline void arch_swap_invalidate_area(int type)
1356 {
1357 	if (system_supports_mte())
1358 		mte_invalidate_tags_area(type);
1359 }
1360 
1361 #define __HAVE_ARCH_SWAP_RESTORE
1362 static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
1363 {
1364 	if (system_supports_mte())
1365 		mte_restore_tags(entry, &folio->page);
1366 }
1367 
1368 #endif /* CONFIG_ARM64_MTE */
1369 
1370 /*
1371  * On AArch64, the cache coherency is handled via the __set_ptes() function.
1372  */
1373 static inline void update_mmu_cache_range(struct vm_fault *vmf,
1374 		struct vm_area_struct *vma, unsigned long addr, pte_t *ptep,
1375 		unsigned int nr)
1376 {
1377 	/*
1378 	 * We don't do anything here, so there's a very small chance of
1379 	 * us retaking a user fault which we just fixed up. The alternative
1380 	 * is doing a dsb(ishst), but that penalises the fastpath.
1381 	 */
1382 }
1383 
1384 #define update_mmu_cache(vma, addr, ptep) \
1385 	update_mmu_cache_range(NULL, vma, addr, ptep, 1)
1386 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
1387 
1388 #ifdef CONFIG_ARM64_PA_BITS_52
1389 #define phys_to_ttbr(addr)	(((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
1390 #else
1391 #define phys_to_ttbr(addr)	(addr)
1392 #endif
1393 
1394 /*
1395  * On arm64 without hardware Access Flag, copying from user will fail because
1396  * the pte is old and cannot be marked young. So we always end up with zeroed
1397  * page after fork() + CoW for pfn mappings. We don't always have a
1398  * hardware-managed access flag on arm64.
1399  */
1400 #define arch_has_hw_pte_young		cpu_has_hw_af
1401 
1402 /*
1403  * Experimentally, it's cheap to set the access flag in hardware and we
1404  * benefit from prefaulting mappings as 'old' to start with.
1405  */
1406 #define arch_wants_old_prefaulted_pte	cpu_has_hw_af
1407 
1408 static inline bool pud_sect_supported(void)
1409 {
1410 	return PAGE_SIZE == SZ_4K;
1411 }
1412 
1413 
1414 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1415 #define ptep_modify_prot_start ptep_modify_prot_start
1416 extern pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
1417 				    unsigned long addr, pte_t *ptep);
1418 
1419 #define ptep_modify_prot_commit ptep_modify_prot_commit
1420 extern void ptep_modify_prot_commit(struct vm_area_struct *vma,
1421 				    unsigned long addr, pte_t *ptep,
1422 				    pte_t old_pte, pte_t new_pte);
1423 
1424 #ifdef CONFIG_ARM64_CONTPTE
1425 
1426 /*
1427  * The contpte APIs are used to transparently manage the contiguous bit in ptes
1428  * where it is possible and makes sense to do so. The PTE_CONT bit is considered
1429  * a private implementation detail of the public ptep API (see below).
1430  */
1431 extern void __contpte_try_fold(struct mm_struct *mm, unsigned long addr,
1432 				pte_t *ptep, pte_t pte);
1433 extern void __contpte_try_unfold(struct mm_struct *mm, unsigned long addr,
1434 				pte_t *ptep, pte_t pte);
1435 extern pte_t contpte_ptep_get(pte_t *ptep, pte_t orig_pte);
1436 extern pte_t contpte_ptep_get_lockless(pte_t *orig_ptep);
1437 extern void contpte_set_ptes(struct mm_struct *mm, unsigned long addr,
1438 				pte_t *ptep, pte_t pte, unsigned int nr);
1439 extern void contpte_clear_full_ptes(struct mm_struct *mm, unsigned long addr,
1440 				pte_t *ptep, unsigned int nr, int full);
1441 extern pte_t contpte_get_and_clear_full_ptes(struct mm_struct *mm,
1442 				unsigned long addr, pte_t *ptep,
1443 				unsigned int nr, int full);
1444 extern int contpte_ptep_test_and_clear_young(struct vm_area_struct *vma,
1445 				unsigned long addr, pte_t *ptep);
1446 extern int contpte_ptep_clear_flush_young(struct vm_area_struct *vma,
1447 				unsigned long addr, pte_t *ptep);
1448 extern void contpte_wrprotect_ptes(struct mm_struct *mm, unsigned long addr,
1449 				pte_t *ptep, unsigned int nr);
1450 extern int contpte_ptep_set_access_flags(struct vm_area_struct *vma,
1451 				unsigned long addr, pte_t *ptep,
1452 				pte_t entry, int dirty);
1453 
1454 static __always_inline void contpte_try_fold(struct mm_struct *mm,
1455 				unsigned long addr, pte_t *ptep, pte_t pte)
1456 {
1457 	/*
1458 	 * Only bother trying if both the virtual and physical addresses are
1459 	 * aligned and correspond to the last entry in a contig range. The core
1460 	 * code mostly modifies ranges from low to high, so this is the likely
1461 	 * the last modification in the contig range, so a good time to fold.
1462 	 * We can't fold special mappings, because there is no associated folio.
1463 	 */
1464 
1465 	const unsigned long contmask = CONT_PTES - 1;
1466 	bool valign = ((addr >> PAGE_SHIFT) & contmask) == contmask;
1467 
1468 	if (unlikely(valign)) {
1469 		bool palign = (pte_pfn(pte) & contmask) == contmask;
1470 
1471 		if (unlikely(palign &&
1472 		    pte_valid(pte) && !pte_cont(pte) && !pte_special(pte)))
1473 			__contpte_try_fold(mm, addr, ptep, pte);
1474 	}
1475 }
1476 
1477 static __always_inline void contpte_try_unfold(struct mm_struct *mm,
1478 				unsigned long addr, pte_t *ptep, pte_t pte)
1479 {
1480 	if (unlikely(pte_valid_cont(pte)))
1481 		__contpte_try_unfold(mm, addr, ptep, pte);
1482 }
1483 
1484 #define pte_batch_hint pte_batch_hint
1485 static inline unsigned int pte_batch_hint(pte_t *ptep, pte_t pte)
1486 {
1487 	if (!pte_valid_cont(pte))
1488 		return 1;
1489 
1490 	return CONT_PTES - (((unsigned long)ptep >> 3) & (CONT_PTES - 1));
1491 }
1492 
1493 /*
1494  * The below functions constitute the public API that arm64 presents to the
1495  * core-mm to manipulate PTE entries within their page tables (or at least this
1496  * is the subset of the API that arm64 needs to implement). These public
1497  * versions will automatically and transparently apply the contiguous bit where
1498  * it makes sense to do so. Therefore any users that are contig-aware (e.g.
1499  * hugetlb, kernel mapper) should NOT use these APIs, but instead use the
1500  * private versions, which are prefixed with double underscore. All of these
1501  * APIs except for ptep_get_lockless() are expected to be called with the PTL
1502  * held. Although the contiguous bit is considered private to the
1503  * implementation, it is deliberately allowed to leak through the getters (e.g.
1504  * ptep_get()), back to core code. This is required so that pte_leaf_size() can
1505  * provide an accurate size for perf_get_pgtable_size(). But this leakage means
1506  * its possible a pte will be passed to a setter with the contiguous bit set, so
1507  * we explicitly clear the contiguous bit in those cases to prevent accidentally
1508  * setting it in the pgtable.
1509  */
1510 
1511 #define ptep_get ptep_get
1512 static inline pte_t ptep_get(pte_t *ptep)
1513 {
1514 	pte_t pte = __ptep_get(ptep);
1515 
1516 	if (likely(!pte_valid_cont(pte)))
1517 		return pte;
1518 
1519 	return contpte_ptep_get(ptep, pte);
1520 }
1521 
1522 #define ptep_get_lockless ptep_get_lockless
1523 static inline pte_t ptep_get_lockless(pte_t *ptep)
1524 {
1525 	pte_t pte = __ptep_get(ptep);
1526 
1527 	if (likely(!pte_valid_cont(pte)))
1528 		return pte;
1529 
1530 	return contpte_ptep_get_lockless(ptep);
1531 }
1532 
1533 static inline void set_pte(pte_t *ptep, pte_t pte)
1534 {
1535 	/*
1536 	 * We don't have the mm or vaddr so cannot unfold contig entries (since
1537 	 * it requires tlb maintenance). set_pte() is not used in core code, so
1538 	 * this should never even be called. Regardless do our best to service
1539 	 * any call and emit a warning if there is any attempt to set a pte on
1540 	 * top of an existing contig range.
1541 	 */
1542 	pte_t orig_pte = __ptep_get(ptep);
1543 
1544 	WARN_ON_ONCE(pte_valid_cont(orig_pte));
1545 	__set_pte(ptep, pte_mknoncont(pte));
1546 }
1547 
1548 #define set_ptes set_ptes
1549 static __always_inline void set_ptes(struct mm_struct *mm, unsigned long addr,
1550 				pte_t *ptep, pte_t pte, unsigned int nr)
1551 {
1552 	pte = pte_mknoncont(pte);
1553 
1554 	if (likely(nr == 1)) {
1555 		contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1556 		__set_ptes(mm, addr, ptep, pte, 1);
1557 		contpte_try_fold(mm, addr, ptep, pte);
1558 	} else {
1559 		contpte_set_ptes(mm, addr, ptep, pte, nr);
1560 	}
1561 }
1562 
1563 static inline void pte_clear(struct mm_struct *mm,
1564 				unsigned long addr, pte_t *ptep)
1565 {
1566 	contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1567 	__pte_clear(mm, addr, ptep);
1568 }
1569 
1570 #define clear_full_ptes clear_full_ptes
1571 static inline void clear_full_ptes(struct mm_struct *mm, unsigned long addr,
1572 				pte_t *ptep, unsigned int nr, int full)
1573 {
1574 	if (likely(nr == 1)) {
1575 		contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1576 		__clear_full_ptes(mm, addr, ptep, nr, full);
1577 	} else {
1578 		contpte_clear_full_ptes(mm, addr, ptep, nr, full);
1579 	}
1580 }
1581 
1582 #define get_and_clear_full_ptes get_and_clear_full_ptes
1583 static inline pte_t get_and_clear_full_ptes(struct mm_struct *mm,
1584 				unsigned long addr, pte_t *ptep,
1585 				unsigned int nr, int full)
1586 {
1587 	pte_t pte;
1588 
1589 	if (likely(nr == 1)) {
1590 		contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1591 		pte = __get_and_clear_full_ptes(mm, addr, ptep, nr, full);
1592 	} else {
1593 		pte = contpte_get_and_clear_full_ptes(mm, addr, ptep, nr, full);
1594 	}
1595 
1596 	return pte;
1597 }
1598 
1599 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1600 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1601 				unsigned long addr, pte_t *ptep)
1602 {
1603 	contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1604 	return __ptep_get_and_clear(mm, addr, ptep);
1605 }
1606 
1607 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1608 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
1609 				unsigned long addr, pte_t *ptep)
1610 {
1611 	pte_t orig_pte = __ptep_get(ptep);
1612 
1613 	if (likely(!pte_valid_cont(orig_pte)))
1614 		return __ptep_test_and_clear_young(vma, addr, ptep);
1615 
1616 	return contpte_ptep_test_and_clear_young(vma, addr, ptep);
1617 }
1618 
1619 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1620 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
1621 				unsigned long addr, pte_t *ptep)
1622 {
1623 	pte_t orig_pte = __ptep_get(ptep);
1624 
1625 	if (likely(!pte_valid_cont(orig_pte)))
1626 		return __ptep_clear_flush_young(vma, addr, ptep);
1627 
1628 	return contpte_ptep_clear_flush_young(vma, addr, ptep);
1629 }
1630 
1631 #define wrprotect_ptes wrprotect_ptes
1632 static __always_inline void wrprotect_ptes(struct mm_struct *mm,
1633 				unsigned long addr, pte_t *ptep, unsigned int nr)
1634 {
1635 	if (likely(nr == 1)) {
1636 		/*
1637 		 * Optimization: wrprotect_ptes() can only be called for present
1638 		 * ptes so we only need to check contig bit as condition for
1639 		 * unfold, and we can remove the contig bit from the pte we read
1640 		 * to avoid re-reading. This speeds up fork() which is sensitive
1641 		 * for order-0 folios. Equivalent to contpte_try_unfold().
1642 		 */
1643 		pte_t orig_pte = __ptep_get(ptep);
1644 
1645 		if (unlikely(pte_cont(orig_pte))) {
1646 			__contpte_try_unfold(mm, addr, ptep, orig_pte);
1647 			orig_pte = pte_mknoncont(orig_pte);
1648 		}
1649 		___ptep_set_wrprotect(mm, addr, ptep, orig_pte);
1650 	} else {
1651 		contpte_wrprotect_ptes(mm, addr, ptep, nr);
1652 	}
1653 }
1654 
1655 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1656 static inline void ptep_set_wrprotect(struct mm_struct *mm,
1657 				unsigned long addr, pte_t *ptep)
1658 {
1659 	wrprotect_ptes(mm, addr, ptep, 1);
1660 }
1661 
1662 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1663 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1664 				unsigned long addr, pte_t *ptep,
1665 				pte_t entry, int dirty)
1666 {
1667 	pte_t orig_pte = __ptep_get(ptep);
1668 
1669 	entry = pte_mknoncont(entry);
1670 
1671 	if (likely(!pte_valid_cont(orig_pte)))
1672 		return __ptep_set_access_flags(vma, addr, ptep, entry, dirty);
1673 
1674 	return contpte_ptep_set_access_flags(vma, addr, ptep, entry, dirty);
1675 }
1676 
1677 #else /* CONFIG_ARM64_CONTPTE */
1678 
1679 #define ptep_get				__ptep_get
1680 #define set_pte					__set_pte
1681 #define set_ptes				__set_ptes
1682 #define pte_clear				__pte_clear
1683 #define clear_full_ptes				__clear_full_ptes
1684 #define get_and_clear_full_ptes			__get_and_clear_full_ptes
1685 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1686 #define ptep_get_and_clear			__ptep_get_and_clear
1687 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1688 #define ptep_test_and_clear_young		__ptep_test_and_clear_young
1689 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1690 #define ptep_clear_flush_young			__ptep_clear_flush_young
1691 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1692 #define ptep_set_wrprotect			__ptep_set_wrprotect
1693 #define wrprotect_ptes				__wrprotect_ptes
1694 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1695 #define ptep_set_access_flags			__ptep_set_access_flags
1696 
1697 #endif /* CONFIG_ARM64_CONTPTE */
1698 
1699 #endif /* !__ASSEMBLY__ */
1700 
1701 #endif /* __ASM_PGTABLE_H */
1702