xref: /linux/arch/arm64/include/asm/pgtable.h (revision 040f404b731207935ed644b14bcc2bb8b8488d00)
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 <asm/por.h>
38 #include <linux/mmdebug.h>
39 #include <linux/mm_types.h>
40 #include <linux/sched.h>
41 #include <linux/page_table_check.h>
42 
43 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
44 #define __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
45 
46 /* Set stride and tlb_level in flush_*_tlb_range */
47 #define flush_pmd_tlb_range(vma, addr, end)	\
48 	__flush_tlb_range(vma, addr, end, PMD_SIZE, false, 2)
49 #define flush_pud_tlb_range(vma, addr, end)	\
50 	__flush_tlb_range(vma, addr, end, PUD_SIZE, false, 1)
51 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
52 
53 /*
54  * Outside of a few very special situations (e.g. hibernation), we always
55  * use broadcast TLB invalidation instructions, therefore a spurious page
56  * fault on one CPU which has been handled concurrently by another CPU
57  * does not need to perform additional invalidation.
58  */
59 #define flush_tlb_fix_spurious_fault(vma, address, ptep) do { } while (0)
60 
61 /*
62  * ZERO_PAGE is a global shared page that is always zero: used
63  * for zero-mapped memory areas etc..
64  */
65 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)];
66 #define ZERO_PAGE(vaddr)	phys_to_page(__pa_symbol(empty_zero_page))
67 
68 #define pte_ERROR(e)	\
69 	pr_err("%s:%d: bad pte %016llx.\n", __FILE__, __LINE__, pte_val(e))
70 
71 #ifdef CONFIG_ARM64_PA_BITS_52
72 static inline phys_addr_t __pte_to_phys(pte_t pte)
73 {
74 	pte_val(pte) &= ~PTE_MAYBE_SHARED;
75 	return (pte_val(pte) & PTE_ADDR_LOW) |
76 		((pte_val(pte) & PTE_ADDR_HIGH) << PTE_ADDR_HIGH_SHIFT);
77 }
78 static inline pteval_t __phys_to_pte_val(phys_addr_t phys)
79 {
80 	return (phys | (phys >> PTE_ADDR_HIGH_SHIFT)) & PHYS_TO_PTE_ADDR_MASK;
81 }
82 #else
83 static inline phys_addr_t __pte_to_phys(pte_t pte)
84 {
85 	return pte_val(pte) & PTE_ADDR_LOW;
86 }
87 
88 static inline pteval_t __phys_to_pte_val(phys_addr_t phys)
89 {
90 	return phys;
91 }
92 #endif
93 
94 #define pte_pfn(pte)		(__pte_to_phys(pte) >> PAGE_SHIFT)
95 #define pfn_pte(pfn,prot)	\
96 	__pte(__phys_to_pte_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
97 
98 #define pte_none(pte)		(!pte_val(pte))
99 #define __pte_clear(mm, addr, ptep) \
100 				__set_pte(ptep, __pte(0))
101 #define pte_page(pte)		(pfn_to_page(pte_pfn(pte)))
102 
103 /*
104  * The following only work if pte_present(). Undefined behaviour otherwise.
105  */
106 #define pte_present(pte)	(pte_valid(pte) || pte_present_invalid(pte))
107 #define pte_young(pte)		(!!(pte_val(pte) & PTE_AF))
108 #define pte_special(pte)	(!!(pte_val(pte) & PTE_SPECIAL))
109 #define pte_write(pte)		(!!(pte_val(pte) & PTE_WRITE))
110 #define pte_rdonly(pte)		(!!(pte_val(pte) & PTE_RDONLY))
111 #define pte_user(pte)		(!!(pte_val(pte) & PTE_USER))
112 #define pte_user_exec(pte)	(!(pte_val(pte) & PTE_UXN))
113 #define pte_cont(pte)		(!!(pte_val(pte) & PTE_CONT))
114 #define pte_devmap(pte)		(!!(pte_val(pte) & PTE_DEVMAP))
115 #define pte_tagged(pte)		((pte_val(pte) & PTE_ATTRINDX_MASK) == \
116 				 PTE_ATTRINDX(MT_NORMAL_TAGGED))
117 
118 #define pte_cont_addr_end(addr, end)						\
119 ({	unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK;	\
120 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
121 })
122 
123 #define pmd_cont_addr_end(addr, end)						\
124 ({	unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK;	\
125 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
126 })
127 
128 #define pte_hw_dirty(pte)	(pte_write(pte) && !pte_rdonly(pte))
129 #define pte_sw_dirty(pte)	(!!(pte_val(pte) & PTE_DIRTY))
130 #define pte_dirty(pte)		(pte_sw_dirty(pte) || pte_hw_dirty(pte))
131 
132 #define pte_valid(pte)		(!!(pte_val(pte) & PTE_VALID))
133 #define pte_present_invalid(pte) \
134 	((pte_val(pte) & (PTE_VALID | PTE_PRESENT_INVALID)) == PTE_PRESENT_INVALID)
135 /*
136  * Execute-only user mappings do not have the PTE_USER bit set. All valid
137  * kernel mappings have the PTE_UXN bit set.
138  */
139 #define pte_valid_not_user(pte) \
140 	((pte_val(pte) & (PTE_VALID | PTE_USER | PTE_UXN)) == (PTE_VALID | PTE_UXN))
141 /*
142  * Returns true if the pte is valid and has the contiguous bit set.
143  */
144 #define pte_valid_cont(pte)	(pte_valid(pte) && pte_cont(pte))
145 /*
146  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
147  * so that we don't erroneously return false for pages that have been
148  * remapped as PROT_NONE but are yet to be flushed from the TLB.
149  * Note that we can't make any assumptions based on the state of the access
150  * flag, since __ptep_clear_flush_young() elides a DSB when invalidating the
151  * TLB.
152  */
153 #define pte_accessible(mm, pte)	\
154 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid(pte))
155 
156 static inline bool por_el0_allows_pkey(u8 pkey, bool write, bool execute)
157 {
158 	u64 por;
159 
160 	if (!system_supports_poe())
161 		return true;
162 
163 	por = read_sysreg_s(SYS_POR_EL0);
164 
165 	if (write)
166 		return por_elx_allows_write(por, pkey);
167 
168 	if (execute)
169 		return por_elx_allows_exec(por, pkey);
170 
171 	return por_elx_allows_read(por, pkey);
172 }
173 
174 /*
175  * p??_access_permitted() is true for valid user mappings (PTE_USER
176  * bit set, subject to the write permission check). For execute-only
177  * mappings, like PROT_EXEC with EPAN (both PTE_USER and PTE_UXN bits
178  * not set) must return false. PROT_NONE mappings do not have the
179  * PTE_VALID bit set.
180  */
181 #define pte_access_permitted_no_overlay(pte, write) \
182 	(((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER)) && (!(write) || pte_write(pte)))
183 #define pte_access_permitted(pte, write) \
184 	(pte_access_permitted_no_overlay(pte, write) && \
185 	por_el0_allows_pkey(FIELD_GET(PTE_PO_IDX_MASK, pte_val(pte)), write, false))
186 #define pmd_access_permitted(pmd, write) \
187 	(pte_access_permitted(pmd_pte(pmd), (write)))
188 #define pud_access_permitted(pud, write) \
189 	(pte_access_permitted(pud_pte(pud), (write)))
190 
191 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
192 {
193 	pte_val(pte) &= ~pgprot_val(prot);
194 	return pte;
195 }
196 
197 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
198 {
199 	pte_val(pte) |= pgprot_val(prot);
200 	return pte;
201 }
202 
203 static inline pmd_t clear_pmd_bit(pmd_t pmd, pgprot_t prot)
204 {
205 	pmd_val(pmd) &= ~pgprot_val(prot);
206 	return pmd;
207 }
208 
209 static inline pmd_t set_pmd_bit(pmd_t pmd, pgprot_t prot)
210 {
211 	pmd_val(pmd) |= pgprot_val(prot);
212 	return pmd;
213 }
214 
215 static inline pte_t pte_mkwrite_novma(pte_t pte)
216 {
217 	pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
218 	pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
219 	return pte;
220 }
221 
222 static inline pte_t pte_mkclean(pte_t pte)
223 {
224 	pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
225 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
226 
227 	return pte;
228 }
229 
230 static inline pte_t pte_mkdirty(pte_t pte)
231 {
232 	pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
233 
234 	if (pte_write(pte))
235 		pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
236 
237 	return pte;
238 }
239 
240 static inline pte_t pte_wrprotect(pte_t pte)
241 {
242 	/*
243 	 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
244 	 * clear), set the PTE_DIRTY bit.
245 	 */
246 	if (pte_hw_dirty(pte))
247 		pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
248 
249 	pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
250 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
251 	return pte;
252 }
253 
254 static inline pte_t pte_mkold(pte_t pte)
255 {
256 	return clear_pte_bit(pte, __pgprot(PTE_AF));
257 }
258 
259 static inline pte_t pte_mkyoung(pte_t pte)
260 {
261 	return set_pte_bit(pte, __pgprot(PTE_AF));
262 }
263 
264 static inline pte_t pte_mkspecial(pte_t pte)
265 {
266 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
267 }
268 
269 static inline pte_t pte_mkcont(pte_t pte)
270 {
271 	return set_pte_bit(pte, __pgprot(PTE_CONT));
272 }
273 
274 static inline pte_t pte_mknoncont(pte_t pte)
275 {
276 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
277 }
278 
279 static inline pte_t pte_mkvalid(pte_t pte)
280 {
281 	return set_pte_bit(pte, __pgprot(PTE_VALID));
282 }
283 
284 static inline pte_t pte_mkinvalid(pte_t pte)
285 {
286 	pte = set_pte_bit(pte, __pgprot(PTE_PRESENT_INVALID));
287 	pte = clear_pte_bit(pte, __pgprot(PTE_VALID));
288 	return pte;
289 }
290 
291 static inline pmd_t pmd_mkcont(pmd_t pmd)
292 {
293 	return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
294 }
295 
296 static inline pte_t pte_mkdevmap(pte_t pte)
297 {
298 	return set_pte_bit(pte, __pgprot(PTE_DEVMAP | PTE_SPECIAL));
299 }
300 
301 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
302 static inline int pte_uffd_wp(pte_t pte)
303 {
304 	return !!(pte_val(pte) & PTE_UFFD_WP);
305 }
306 
307 static inline pte_t pte_mkuffd_wp(pte_t pte)
308 {
309 	return pte_wrprotect(set_pte_bit(pte, __pgprot(PTE_UFFD_WP)));
310 }
311 
312 static inline pte_t pte_clear_uffd_wp(pte_t pte)
313 {
314 	return clear_pte_bit(pte, __pgprot(PTE_UFFD_WP));
315 }
316 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
317 
318 static inline void __set_pte_nosync(pte_t *ptep, pte_t pte)
319 {
320 	WRITE_ONCE(*ptep, pte);
321 }
322 
323 static inline void __set_pte(pte_t *ptep, pte_t pte)
324 {
325 	__set_pte_nosync(ptep, pte);
326 
327 	/*
328 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
329 	 * or update_mmu_cache() have the necessary barriers.
330 	 */
331 	if (pte_valid_not_user(pte)) {
332 		dsb(ishst);
333 		isb();
334 	}
335 }
336 
337 static inline pte_t __ptep_get(pte_t *ptep)
338 {
339 	return READ_ONCE(*ptep);
340 }
341 
342 extern void __sync_icache_dcache(pte_t pteval);
343 bool pgattr_change_is_safe(pteval_t old, pteval_t new);
344 
345 /*
346  * PTE bits configuration in the presence of hardware Dirty Bit Management
347  * (PTE_WRITE == PTE_DBM):
348  *
349  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
350  *   0      0      |   1           0          0
351  *   0      1      |   1           1          0
352  *   1      0      |   1           0          1
353  *   1      1      |   0           1          x
354  *
355  * When hardware DBM is not present, the sofware PTE_DIRTY bit is updated via
356  * the page fault mechanism. Checking the dirty status of a pte becomes:
357  *
358  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
359  */
360 
361 static inline void __check_safe_pte_update(struct mm_struct *mm, pte_t *ptep,
362 					   pte_t pte)
363 {
364 	pte_t old_pte;
365 
366 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
367 		return;
368 
369 	old_pte = __ptep_get(ptep);
370 
371 	if (!pte_valid(old_pte) || !pte_valid(pte))
372 		return;
373 	if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1)
374 		return;
375 
376 	/*
377 	 * Check for potential race with hardware updates of the pte
378 	 * (__ptep_set_access_flags safely changes valid ptes without going
379 	 * through an invalid entry).
380 	 */
381 	VM_WARN_ONCE(!pte_young(pte),
382 		     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
383 		     __func__, pte_val(old_pte), pte_val(pte));
384 	VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
385 		     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
386 		     __func__, pte_val(old_pte), pte_val(pte));
387 	VM_WARN_ONCE(!pgattr_change_is_safe(pte_val(old_pte), pte_val(pte)),
388 		     "%s: unsafe attribute change: 0x%016llx -> 0x%016llx",
389 		     __func__, pte_val(old_pte), pte_val(pte));
390 }
391 
392 static inline void __sync_cache_and_tags(pte_t pte, unsigned int nr_pages)
393 {
394 	if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
395 		__sync_icache_dcache(pte);
396 
397 	/*
398 	 * If the PTE would provide user space access to the tags associated
399 	 * with it then ensure that the MTE tags are synchronised.  Although
400 	 * pte_access_permitted_no_overlay() returns false for exec only
401 	 * mappings, they don't expose tags (instruction fetches don't check
402 	 * tags).
403 	 */
404 	if (system_supports_mte() && pte_access_permitted_no_overlay(pte, false) &&
405 	    !pte_special(pte) && pte_tagged(pte))
406 		mte_sync_tags(pte, nr_pages);
407 }
408 
409 /*
410  * Select all bits except the pfn
411  */
412 #define pte_pgprot pte_pgprot
413 static inline pgprot_t pte_pgprot(pte_t pte)
414 {
415 	unsigned long pfn = pte_pfn(pte);
416 
417 	return __pgprot(pte_val(pfn_pte(pfn, __pgprot(0))) ^ pte_val(pte));
418 }
419 
420 #define pte_advance_pfn pte_advance_pfn
421 static inline pte_t pte_advance_pfn(pte_t pte, unsigned long nr)
422 {
423 	return pfn_pte(pte_pfn(pte) + nr, pte_pgprot(pte));
424 }
425 
426 static inline void __set_ptes(struct mm_struct *mm,
427 			      unsigned long __always_unused addr,
428 			      pte_t *ptep, pte_t pte, unsigned int nr)
429 {
430 	page_table_check_ptes_set(mm, ptep, pte, nr);
431 	__sync_cache_and_tags(pte, nr);
432 
433 	for (;;) {
434 		__check_safe_pte_update(mm, ptep, pte);
435 		__set_pte(ptep, pte);
436 		if (--nr == 0)
437 			break;
438 		ptep++;
439 		pte = pte_advance_pfn(pte, 1);
440 	}
441 }
442 
443 /*
444  * Hugetlb definitions.
445  */
446 #define HUGE_MAX_HSTATE		4
447 #define HPAGE_SHIFT		PMD_SHIFT
448 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
449 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
450 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
451 
452 static inline pte_t pgd_pte(pgd_t pgd)
453 {
454 	return __pte(pgd_val(pgd));
455 }
456 
457 static inline pte_t p4d_pte(p4d_t p4d)
458 {
459 	return __pte(p4d_val(p4d));
460 }
461 
462 static inline pte_t pud_pte(pud_t pud)
463 {
464 	return __pte(pud_val(pud));
465 }
466 
467 static inline pud_t pte_pud(pte_t pte)
468 {
469 	return __pud(pte_val(pte));
470 }
471 
472 static inline pmd_t pud_pmd(pud_t pud)
473 {
474 	return __pmd(pud_val(pud));
475 }
476 
477 static inline pte_t pmd_pte(pmd_t pmd)
478 {
479 	return __pte(pmd_val(pmd));
480 }
481 
482 static inline pmd_t pte_pmd(pte_t pte)
483 {
484 	return __pmd(pte_val(pte));
485 }
486 
487 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot)
488 {
489 	return __pgprot((pgprot_val(prot) & ~PUD_TYPE_MASK) | PUD_TYPE_SECT);
490 }
491 
492 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot)
493 {
494 	return __pgprot((pgprot_val(prot) & ~PMD_TYPE_MASK) | PMD_TYPE_SECT);
495 }
496 
497 static inline pte_t pte_swp_mkexclusive(pte_t pte)
498 {
499 	return set_pte_bit(pte, __pgprot(PTE_SWP_EXCLUSIVE));
500 }
501 
502 static inline int pte_swp_exclusive(pte_t pte)
503 {
504 	return pte_val(pte) & PTE_SWP_EXCLUSIVE;
505 }
506 
507 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
508 {
509 	return clear_pte_bit(pte, __pgprot(PTE_SWP_EXCLUSIVE));
510 }
511 
512 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
513 static inline pte_t pte_swp_mkuffd_wp(pte_t pte)
514 {
515 	return set_pte_bit(pte, __pgprot(PTE_SWP_UFFD_WP));
516 }
517 
518 static inline int pte_swp_uffd_wp(pte_t pte)
519 {
520 	return !!(pte_val(pte) & PTE_SWP_UFFD_WP);
521 }
522 
523 static inline pte_t pte_swp_clear_uffd_wp(pte_t pte)
524 {
525 	return clear_pte_bit(pte, __pgprot(PTE_SWP_UFFD_WP));
526 }
527 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
528 
529 #ifdef CONFIG_NUMA_BALANCING
530 /*
531  * See the comment in include/linux/pgtable.h
532  */
533 static inline int pte_protnone(pte_t pte)
534 {
535 	/*
536 	 * pte_present_invalid() tells us that the pte is invalid from HW
537 	 * perspective but present from SW perspective, so the fields are to be
538 	 * interpretted as per the HW layout. The second 2 checks are the unique
539 	 * encoding that we use for PROT_NONE. It is insufficient to only use
540 	 * the first check because we share the same encoding scheme with pmds
541 	 * which support pmd_mkinvalid(), so can be present-invalid without
542 	 * being PROT_NONE.
543 	 */
544 	return pte_present_invalid(pte) && !pte_user(pte) && !pte_user_exec(pte);
545 }
546 
547 static inline int pmd_protnone(pmd_t pmd)
548 {
549 	return pte_protnone(pmd_pte(pmd));
550 }
551 #endif
552 
553 #define pmd_present(pmd)	pte_present(pmd_pte(pmd))
554 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
555 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
556 #define pmd_valid(pmd)		pte_valid(pmd_pte(pmd))
557 #define pmd_user(pmd)		pte_user(pmd_pte(pmd))
558 #define pmd_user_exec(pmd)	pte_user_exec(pmd_pte(pmd))
559 #define pmd_cont(pmd)		pte_cont(pmd_pte(pmd))
560 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
561 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
562 #define pmd_mkwrite_novma(pmd)	pte_pmd(pte_mkwrite_novma(pmd_pte(pmd)))
563 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
564 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
565 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
566 #define pmd_mkinvalid(pmd)	pte_pmd(pte_mkinvalid(pmd_pte(pmd)))
567 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
568 #define pmd_uffd_wp(pmd)	pte_uffd_wp(pmd_pte(pmd))
569 #define pmd_mkuffd_wp(pmd)	pte_pmd(pte_mkuffd_wp(pmd_pte(pmd)))
570 #define pmd_clear_uffd_wp(pmd)	pte_pmd(pte_clear_uffd_wp(pmd_pte(pmd)))
571 #define pmd_swp_uffd_wp(pmd)	pte_swp_uffd_wp(pmd_pte(pmd))
572 #define pmd_swp_mkuffd_wp(pmd)	pte_pmd(pte_swp_mkuffd_wp(pmd_pte(pmd)))
573 #define pmd_swp_clear_uffd_wp(pmd) \
574 				pte_pmd(pte_swp_clear_uffd_wp(pmd_pte(pmd)))
575 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
576 
577 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
578 
579 static inline pmd_t pmd_mkhuge(pmd_t pmd)
580 {
581 	/*
582 	 * It's possible that the pmd is present-invalid on entry
583 	 * and in that case it needs to remain present-invalid on
584 	 * exit. So ensure the VALID bit does not get modified.
585 	 */
586 	pmdval_t mask = PMD_TYPE_MASK & ~PTE_VALID;
587 	pmdval_t val = PMD_TYPE_SECT & ~PTE_VALID;
588 
589 	return __pmd((pmd_val(pmd) & ~mask) | val);
590 }
591 
592 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
593 #define pmd_devmap(pmd)		pte_devmap(pmd_pte(pmd))
594 #endif
595 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
596 {
597 	return pte_pmd(set_pte_bit(pmd_pte(pmd), __pgprot(PTE_DEVMAP)));
598 }
599 
600 #ifdef CONFIG_ARCH_SUPPORTS_PMD_PFNMAP
601 #define pmd_special(pte)	(!!((pmd_val(pte) & PTE_SPECIAL)))
602 static inline pmd_t pmd_mkspecial(pmd_t pmd)
603 {
604 	return set_pmd_bit(pmd, __pgprot(PTE_SPECIAL));
605 }
606 #endif
607 
608 #define __pmd_to_phys(pmd)	__pte_to_phys(pmd_pte(pmd))
609 #define __phys_to_pmd_val(phys)	__phys_to_pte_val(phys)
610 #define pmd_pfn(pmd)		((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
611 #define pfn_pmd(pfn,prot)	__pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
612 
613 #define pud_young(pud)		pte_young(pud_pte(pud))
614 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
615 #define pud_write(pud)		pte_write(pud_pte(pud))
616 
617 static inline pud_t pud_mkhuge(pud_t pud)
618 {
619 	/*
620 	 * It's possible that the pud is present-invalid on entry
621 	 * and in that case it needs to remain present-invalid on
622 	 * exit. So ensure the VALID bit does not get modified.
623 	 */
624 	pudval_t mask = PUD_TYPE_MASK & ~PTE_VALID;
625 	pudval_t val = PUD_TYPE_SECT & ~PTE_VALID;
626 
627 	return __pud((pud_val(pud) & ~mask) | val);
628 }
629 
630 #define __pud_to_phys(pud)	__pte_to_phys(pud_pte(pud))
631 #define __phys_to_pud_val(phys)	__phys_to_pte_val(phys)
632 #define pud_pfn(pud)		((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
633 #define pfn_pud(pfn,prot)	__pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
634 
635 #define pmd_pgprot pmd_pgprot
636 static inline pgprot_t pmd_pgprot(pmd_t pmd)
637 {
638 	unsigned long pfn = pmd_pfn(pmd);
639 
640 	return __pgprot(pmd_val(pfn_pmd(pfn, __pgprot(0))) ^ pmd_val(pmd));
641 }
642 
643 #define pud_pgprot pud_pgprot
644 static inline pgprot_t pud_pgprot(pud_t pud)
645 {
646 	unsigned long pfn = pud_pfn(pud);
647 
648 	return __pgprot(pud_val(pfn_pud(pfn, __pgprot(0))) ^ pud_val(pud));
649 }
650 
651 static inline void __set_pte_at(struct mm_struct *mm,
652 				unsigned long __always_unused addr,
653 				pte_t *ptep, pte_t pte, unsigned int nr)
654 {
655 	__sync_cache_and_tags(pte, nr);
656 	__check_safe_pte_update(mm, ptep, pte);
657 	__set_pte(ptep, pte);
658 }
659 
660 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
661 			      pmd_t *pmdp, pmd_t pmd)
662 {
663 	page_table_check_pmd_set(mm, pmdp, pmd);
664 	return __set_pte_at(mm, addr, (pte_t *)pmdp, pmd_pte(pmd),
665 						PMD_SIZE >> PAGE_SHIFT);
666 }
667 
668 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
669 			      pud_t *pudp, pud_t pud)
670 {
671 	page_table_check_pud_set(mm, pudp, pud);
672 	return __set_pte_at(mm, addr, (pte_t *)pudp, pud_pte(pud),
673 						PUD_SIZE >> PAGE_SHIFT);
674 }
675 
676 #define __p4d_to_phys(p4d)	__pte_to_phys(p4d_pte(p4d))
677 #define __phys_to_p4d_val(phys)	__phys_to_pte_val(phys)
678 
679 #define __pgd_to_phys(pgd)	__pte_to_phys(pgd_pte(pgd))
680 #define __phys_to_pgd_val(phys)	__phys_to_pte_val(phys)
681 
682 #define __pgprot_modify(prot,mask,bits) \
683 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
684 
685 #define pgprot_nx(prot) \
686 	__pgprot_modify(prot, PTE_MAYBE_GP, PTE_PXN)
687 
688 #define pgprot_decrypted(prot) \
689 	__pgprot_modify(prot, PROT_NS_SHARED, PROT_NS_SHARED)
690 #define pgprot_encrypted(prot) \
691 	__pgprot_modify(prot, PROT_NS_SHARED, 0)
692 
693 /*
694  * Mark the prot value as uncacheable and unbufferable.
695  */
696 #define pgprot_noncached(prot) \
697 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
698 #define pgprot_writecombine(prot) \
699 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
700 #define pgprot_device(prot) \
701 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
702 #define pgprot_tagged(prot) \
703 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_TAGGED))
704 #define pgprot_mhp	pgprot_tagged
705 /*
706  * DMA allocations for non-coherent devices use what the Arm architecture calls
707  * "Normal non-cacheable" memory, which permits speculation, unaligned accesses
708  * and merging of writes.  This is different from "Device-nGnR[nE]" memory which
709  * is intended for MMIO and thus forbids speculation, preserves access size,
710  * requires strict alignment and can also force write responses to come from the
711  * endpoint.
712  */
713 #define pgprot_dmacoherent(prot) \
714 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, \
715 			PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
716 
717 #define __HAVE_PHYS_MEM_ACCESS_PROT
718 struct file;
719 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
720 				     unsigned long size, pgprot_t vma_prot);
721 
722 #define pmd_none(pmd)		(!pmd_val(pmd))
723 
724 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
725 				 PMD_TYPE_TABLE)
726 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
727 				 PMD_TYPE_SECT)
728 #define pmd_leaf(pmd)		(pmd_present(pmd) && !pmd_table(pmd))
729 #define pmd_bad(pmd)		(!pmd_table(pmd))
730 
731 #define pmd_leaf_size(pmd)	(pmd_cont(pmd) ? CONT_PMD_SIZE : PMD_SIZE)
732 #define pte_leaf_size(pte)	(pte_cont(pte) ? CONT_PTE_SIZE : PAGE_SIZE)
733 
734 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
735 static inline int pmd_trans_huge(pmd_t pmd)
736 {
737 	/*
738 	 * If pmd is present-invalid, pmd_table() won't detect it
739 	 * as a table, so force the valid bit for the comparison.
740 	 */
741 	return pmd_val(pmd) && pmd_present(pmd) &&
742 	       !pmd_table(__pmd(pmd_val(pmd) | PTE_VALID));
743 }
744 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
745 
746 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
747 static inline bool pud_sect(pud_t pud) { return false; }
748 static inline bool pud_table(pud_t pud) { return true; }
749 #else
750 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
751 				 PUD_TYPE_SECT)
752 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
753 				 PUD_TYPE_TABLE)
754 #endif
755 
756 extern pgd_t init_pg_dir[];
757 extern pgd_t init_pg_end[];
758 extern pgd_t swapper_pg_dir[];
759 extern pgd_t idmap_pg_dir[];
760 extern pgd_t tramp_pg_dir[];
761 extern pgd_t reserved_pg_dir[];
762 
763 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
764 
765 static inline bool in_swapper_pgdir(void *addr)
766 {
767 	return ((unsigned long)addr & PAGE_MASK) ==
768 	        ((unsigned long)swapper_pg_dir & PAGE_MASK);
769 }
770 
771 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
772 {
773 #ifdef __PAGETABLE_PMD_FOLDED
774 	if (in_swapper_pgdir(pmdp)) {
775 		set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
776 		return;
777 	}
778 #endif /* __PAGETABLE_PMD_FOLDED */
779 
780 	WRITE_ONCE(*pmdp, pmd);
781 
782 	if (pmd_valid(pmd)) {
783 		dsb(ishst);
784 		isb();
785 	}
786 }
787 
788 static inline void pmd_clear(pmd_t *pmdp)
789 {
790 	set_pmd(pmdp, __pmd(0));
791 }
792 
793 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
794 {
795 	return __pmd_to_phys(pmd);
796 }
797 
798 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
799 {
800 	return (unsigned long)__va(pmd_page_paddr(pmd));
801 }
802 
803 /* Find an entry in the third-level page table. */
804 #define pte_offset_phys(dir,addr)	(pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
805 
806 #define pte_set_fixmap(addr)		((pte_t *)set_fixmap_offset(FIX_PTE, addr))
807 #define pte_set_fixmap_offset(pmd, addr)	pte_set_fixmap(pte_offset_phys(pmd, addr))
808 #define pte_clear_fixmap()		clear_fixmap(FIX_PTE)
809 
810 #define pmd_page(pmd)			phys_to_page(__pmd_to_phys(pmd))
811 
812 /* use ONLY for statically allocated translation tables */
813 #define pte_offset_kimg(dir,addr)	((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
814 
815 #if CONFIG_PGTABLE_LEVELS > 2
816 
817 #define pmd_ERROR(e)	\
818 	pr_err("%s:%d: bad pmd %016llx.\n", __FILE__, __LINE__, pmd_val(e))
819 
820 #define pud_none(pud)		(!pud_val(pud))
821 #define pud_bad(pud)		((pud_val(pud) & PUD_TYPE_MASK) != \
822 				 PUD_TYPE_TABLE)
823 #define pud_present(pud)	pte_present(pud_pte(pud))
824 #ifndef __PAGETABLE_PMD_FOLDED
825 #define pud_leaf(pud)		(pud_present(pud) && !pud_table(pud))
826 #else
827 #define pud_leaf(pud)		false
828 #endif
829 #define pud_valid(pud)		pte_valid(pud_pte(pud))
830 #define pud_user(pud)		pte_user(pud_pte(pud))
831 #define pud_user_exec(pud)	pte_user_exec(pud_pte(pud))
832 
833 static inline bool pgtable_l4_enabled(void);
834 
835 static inline void set_pud(pud_t *pudp, pud_t pud)
836 {
837 	if (!pgtable_l4_enabled() && in_swapper_pgdir(pudp)) {
838 		set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
839 		return;
840 	}
841 
842 	WRITE_ONCE(*pudp, pud);
843 
844 	if (pud_valid(pud)) {
845 		dsb(ishst);
846 		isb();
847 	}
848 }
849 
850 static inline void pud_clear(pud_t *pudp)
851 {
852 	set_pud(pudp, __pud(0));
853 }
854 
855 static inline phys_addr_t pud_page_paddr(pud_t pud)
856 {
857 	return __pud_to_phys(pud);
858 }
859 
860 static inline pmd_t *pud_pgtable(pud_t pud)
861 {
862 	return (pmd_t *)__va(pud_page_paddr(pud));
863 }
864 
865 /* Find an entry in the second-level page table. */
866 #define pmd_offset_phys(dir, addr)	(pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
867 
868 #define pmd_set_fixmap(addr)		((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
869 #define pmd_set_fixmap_offset(pud, addr)	pmd_set_fixmap(pmd_offset_phys(pud, addr))
870 #define pmd_clear_fixmap()		clear_fixmap(FIX_PMD)
871 
872 #define pud_page(pud)			phys_to_page(__pud_to_phys(pud))
873 
874 /* use ONLY for statically allocated translation tables */
875 #define pmd_offset_kimg(dir,addr)	((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
876 
877 #else
878 
879 #define pud_valid(pud)		false
880 #define pud_page_paddr(pud)	({ BUILD_BUG(); 0; })
881 #define pud_user_exec(pud)	pud_user(pud) /* Always 0 with folding */
882 
883 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
884 #define pmd_set_fixmap(addr)		NULL
885 #define pmd_set_fixmap_offset(pudp, addr)	((pmd_t *)pudp)
886 #define pmd_clear_fixmap()
887 
888 #define pmd_offset_kimg(dir,addr)	((pmd_t *)dir)
889 
890 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
891 
892 #if CONFIG_PGTABLE_LEVELS > 3
893 
894 static __always_inline bool pgtable_l4_enabled(void)
895 {
896 	if (CONFIG_PGTABLE_LEVELS > 4 || !IS_ENABLED(CONFIG_ARM64_LPA2))
897 		return true;
898 	if (!alternative_has_cap_likely(ARM64_ALWAYS_BOOT))
899 		return vabits_actual == VA_BITS;
900 	return alternative_has_cap_unlikely(ARM64_HAS_VA52);
901 }
902 
903 static inline bool mm_pud_folded(const struct mm_struct *mm)
904 {
905 	return !pgtable_l4_enabled();
906 }
907 #define mm_pud_folded  mm_pud_folded
908 
909 #define pud_ERROR(e)	\
910 	pr_err("%s:%d: bad pud %016llx.\n", __FILE__, __LINE__, pud_val(e))
911 
912 #define p4d_none(p4d)		(pgtable_l4_enabled() && !p4d_val(p4d))
913 #define p4d_bad(p4d)		(pgtable_l4_enabled() && \
914 				((p4d_val(p4d) & P4D_TYPE_MASK) != \
915 				 P4D_TYPE_TABLE))
916 #define p4d_present(p4d)	(!p4d_none(p4d))
917 
918 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
919 {
920 	if (in_swapper_pgdir(p4dp)) {
921 		set_swapper_pgd((pgd_t *)p4dp, __pgd(p4d_val(p4d)));
922 		return;
923 	}
924 
925 	WRITE_ONCE(*p4dp, p4d);
926 	dsb(ishst);
927 	isb();
928 }
929 
930 static inline void p4d_clear(p4d_t *p4dp)
931 {
932 	if (pgtable_l4_enabled())
933 		set_p4d(p4dp, __p4d(0));
934 }
935 
936 static inline phys_addr_t p4d_page_paddr(p4d_t p4d)
937 {
938 	return __p4d_to_phys(p4d);
939 }
940 
941 #define pud_index(addr)		(((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
942 
943 static inline pud_t *p4d_to_folded_pud(p4d_t *p4dp, unsigned long addr)
944 {
945 	/* Ensure that 'p4dp' indexes a page table according to 'addr' */
946 	VM_BUG_ON(((addr >> P4D_SHIFT) ^ ((u64)p4dp >> 3)) % PTRS_PER_P4D);
947 
948 	return (pud_t *)PTR_ALIGN_DOWN(p4dp, PAGE_SIZE) + pud_index(addr);
949 }
950 
951 static inline pud_t *p4d_pgtable(p4d_t p4d)
952 {
953 	return (pud_t *)__va(p4d_page_paddr(p4d));
954 }
955 
956 static inline phys_addr_t pud_offset_phys(p4d_t *p4dp, unsigned long addr)
957 {
958 	BUG_ON(!pgtable_l4_enabled());
959 
960 	return p4d_page_paddr(READ_ONCE(*p4dp)) + pud_index(addr) * sizeof(pud_t);
961 }
962 
963 static inline
964 pud_t *pud_offset_lockless(p4d_t *p4dp, p4d_t p4d, unsigned long addr)
965 {
966 	if (!pgtable_l4_enabled())
967 		return p4d_to_folded_pud(p4dp, addr);
968 	return (pud_t *)__va(p4d_page_paddr(p4d)) + pud_index(addr);
969 }
970 #define pud_offset_lockless pud_offset_lockless
971 
972 static inline pud_t *pud_offset(p4d_t *p4dp, unsigned long addr)
973 {
974 	return pud_offset_lockless(p4dp, READ_ONCE(*p4dp), addr);
975 }
976 #define pud_offset	pud_offset
977 
978 static inline pud_t *pud_set_fixmap(unsigned long addr)
979 {
980 	if (!pgtable_l4_enabled())
981 		return NULL;
982 	return (pud_t *)set_fixmap_offset(FIX_PUD, addr);
983 }
984 
985 static inline pud_t *pud_set_fixmap_offset(p4d_t *p4dp, unsigned long addr)
986 {
987 	if (!pgtable_l4_enabled())
988 		return p4d_to_folded_pud(p4dp, addr);
989 	return pud_set_fixmap(pud_offset_phys(p4dp, addr));
990 }
991 
992 static inline void pud_clear_fixmap(void)
993 {
994 	if (pgtable_l4_enabled())
995 		clear_fixmap(FIX_PUD);
996 }
997 
998 /* use ONLY for statically allocated translation tables */
999 static inline pud_t *pud_offset_kimg(p4d_t *p4dp, u64 addr)
1000 {
1001 	if (!pgtable_l4_enabled())
1002 		return p4d_to_folded_pud(p4dp, addr);
1003 	return (pud_t *)__phys_to_kimg(pud_offset_phys(p4dp, addr));
1004 }
1005 
1006 #define p4d_page(p4d)		pfn_to_page(__phys_to_pfn(__p4d_to_phys(p4d)))
1007 
1008 #else
1009 
1010 static inline bool pgtable_l4_enabled(void) { return false; }
1011 
1012 #define p4d_page_paddr(p4d)	({ BUILD_BUG(); 0;})
1013 
1014 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
1015 #define pud_set_fixmap(addr)		NULL
1016 #define pud_set_fixmap_offset(pgdp, addr)	((pud_t *)pgdp)
1017 #define pud_clear_fixmap()
1018 
1019 #define pud_offset_kimg(dir,addr)	((pud_t *)dir)
1020 
1021 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
1022 
1023 #if CONFIG_PGTABLE_LEVELS > 4
1024 
1025 static __always_inline bool pgtable_l5_enabled(void)
1026 {
1027 	if (!alternative_has_cap_likely(ARM64_ALWAYS_BOOT))
1028 		return vabits_actual == VA_BITS;
1029 	return alternative_has_cap_unlikely(ARM64_HAS_VA52);
1030 }
1031 
1032 static inline bool mm_p4d_folded(const struct mm_struct *mm)
1033 {
1034 	return !pgtable_l5_enabled();
1035 }
1036 #define mm_p4d_folded  mm_p4d_folded
1037 
1038 #define p4d_ERROR(e)	\
1039 	pr_err("%s:%d: bad p4d %016llx.\n", __FILE__, __LINE__, p4d_val(e))
1040 
1041 #define pgd_none(pgd)		(pgtable_l5_enabled() && !pgd_val(pgd))
1042 #define pgd_bad(pgd)		(pgtable_l5_enabled() && \
1043 				((pgd_val(pgd) & PGD_TYPE_MASK) != \
1044 				 PGD_TYPE_TABLE))
1045 #define pgd_present(pgd)	(!pgd_none(pgd))
1046 
1047 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
1048 {
1049 	if (in_swapper_pgdir(pgdp)) {
1050 		set_swapper_pgd(pgdp, __pgd(pgd_val(pgd)));
1051 		return;
1052 	}
1053 
1054 	WRITE_ONCE(*pgdp, pgd);
1055 	dsb(ishst);
1056 	isb();
1057 }
1058 
1059 static inline void pgd_clear(pgd_t *pgdp)
1060 {
1061 	if (pgtable_l5_enabled())
1062 		set_pgd(pgdp, __pgd(0));
1063 }
1064 
1065 static inline phys_addr_t pgd_page_paddr(pgd_t pgd)
1066 {
1067 	return __pgd_to_phys(pgd);
1068 }
1069 
1070 #define p4d_index(addr)		(((addr) >> P4D_SHIFT) & (PTRS_PER_P4D - 1))
1071 
1072 static inline p4d_t *pgd_to_folded_p4d(pgd_t *pgdp, unsigned long addr)
1073 {
1074 	/* Ensure that 'pgdp' indexes a page table according to 'addr' */
1075 	VM_BUG_ON(((addr >> PGDIR_SHIFT) ^ ((u64)pgdp >> 3)) % PTRS_PER_PGD);
1076 
1077 	return (p4d_t *)PTR_ALIGN_DOWN(pgdp, PAGE_SIZE) + p4d_index(addr);
1078 }
1079 
1080 static inline phys_addr_t p4d_offset_phys(pgd_t *pgdp, unsigned long addr)
1081 {
1082 	BUG_ON(!pgtable_l5_enabled());
1083 
1084 	return pgd_page_paddr(READ_ONCE(*pgdp)) + p4d_index(addr) * sizeof(p4d_t);
1085 }
1086 
1087 static inline
1088 p4d_t *p4d_offset_lockless(pgd_t *pgdp, pgd_t pgd, unsigned long addr)
1089 {
1090 	if (!pgtable_l5_enabled())
1091 		return pgd_to_folded_p4d(pgdp, addr);
1092 	return (p4d_t *)__va(pgd_page_paddr(pgd)) + p4d_index(addr);
1093 }
1094 #define p4d_offset_lockless p4d_offset_lockless
1095 
1096 static inline p4d_t *p4d_offset(pgd_t *pgdp, unsigned long addr)
1097 {
1098 	return p4d_offset_lockless(pgdp, READ_ONCE(*pgdp), addr);
1099 }
1100 
1101 static inline p4d_t *p4d_set_fixmap(unsigned long addr)
1102 {
1103 	if (!pgtable_l5_enabled())
1104 		return NULL;
1105 	return (p4d_t *)set_fixmap_offset(FIX_P4D, addr);
1106 }
1107 
1108 static inline p4d_t *p4d_set_fixmap_offset(pgd_t *pgdp, unsigned long addr)
1109 {
1110 	if (!pgtable_l5_enabled())
1111 		return pgd_to_folded_p4d(pgdp, addr);
1112 	return p4d_set_fixmap(p4d_offset_phys(pgdp, addr));
1113 }
1114 
1115 static inline void p4d_clear_fixmap(void)
1116 {
1117 	if (pgtable_l5_enabled())
1118 		clear_fixmap(FIX_P4D);
1119 }
1120 
1121 /* use ONLY for statically allocated translation tables */
1122 static inline p4d_t *p4d_offset_kimg(pgd_t *pgdp, u64 addr)
1123 {
1124 	if (!pgtable_l5_enabled())
1125 		return pgd_to_folded_p4d(pgdp, addr);
1126 	return (p4d_t *)__phys_to_kimg(p4d_offset_phys(pgdp, addr));
1127 }
1128 
1129 #define pgd_page(pgd)		pfn_to_page(__phys_to_pfn(__pgd_to_phys(pgd)))
1130 
1131 #else
1132 
1133 static inline bool pgtable_l5_enabled(void) { return false; }
1134 
1135 #define p4d_index(addr)		(((addr) >> P4D_SHIFT) & (PTRS_PER_P4D - 1))
1136 
1137 /* Match p4d_offset folding in <asm/generic/pgtable-nop4d.h> */
1138 #define p4d_set_fixmap(addr)		NULL
1139 #define p4d_set_fixmap_offset(p4dp, addr)	((p4d_t *)p4dp)
1140 #define p4d_clear_fixmap()
1141 
1142 #define p4d_offset_kimg(dir,addr)	((p4d_t *)dir)
1143 
1144 static inline
1145 p4d_t *p4d_offset_lockless_folded(pgd_t *pgdp, pgd_t pgd, unsigned long addr)
1146 {
1147 	/*
1148 	 * With runtime folding of the pud, pud_offset_lockless() passes
1149 	 * the 'pgd_t *' we return here to p4d_to_folded_pud(), which
1150 	 * will offset the pointer assuming that it points into
1151 	 * a page-table page. However, the fast GUP path passes us a
1152 	 * pgd_t allocated on the stack and so we must use the original
1153 	 * pointer in 'pgdp' to construct the p4d pointer instead of
1154 	 * using the generic p4d_offset_lockless() implementation.
1155 	 *
1156 	 * Note: reusing the original pointer means that we may
1157 	 * dereference the same (live) page-table entry multiple times.
1158 	 * This is safe because it is still only loaded once in the
1159 	 * context of each level and the CPU guarantees same-address
1160 	 * read-after-read ordering.
1161 	 */
1162 	return p4d_offset(pgdp, addr);
1163 }
1164 #define p4d_offset_lockless p4d_offset_lockless_folded
1165 
1166 #endif  /* CONFIG_PGTABLE_LEVELS > 4 */
1167 
1168 #define pgd_ERROR(e)	\
1169 	pr_err("%s:%d: bad pgd %016llx.\n", __FILE__, __LINE__, pgd_val(e))
1170 
1171 #define pgd_set_fixmap(addr)	((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
1172 #define pgd_clear_fixmap()	clear_fixmap(FIX_PGD)
1173 
1174 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
1175 {
1176 	/*
1177 	 * Normal and Normal-Tagged are two different memory types and indices
1178 	 * in MAIR_EL1. The mask below has to include PTE_ATTRINDX_MASK.
1179 	 */
1180 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
1181 			      PTE_PRESENT_INVALID | PTE_VALID | PTE_WRITE |
1182 			      PTE_GP | PTE_ATTRINDX_MASK | PTE_PO_IDX_MASK;
1183 
1184 	/* preserve the hardware dirty information */
1185 	if (pte_hw_dirty(pte))
1186 		pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
1187 
1188 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
1189 	/*
1190 	 * If we end up clearing hw dirtiness for a sw-dirty PTE, set hardware
1191 	 * dirtiness again.
1192 	 */
1193 	if (pte_sw_dirty(pte))
1194 		pte = pte_mkdirty(pte);
1195 	return pte;
1196 }
1197 
1198 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1199 {
1200 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
1201 }
1202 
1203 extern int __ptep_set_access_flags(struct vm_area_struct *vma,
1204 				 unsigned long address, pte_t *ptep,
1205 				 pte_t entry, int dirty);
1206 
1207 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1208 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1209 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
1210 					unsigned long address, pmd_t *pmdp,
1211 					pmd_t entry, int dirty)
1212 {
1213 	return __ptep_set_access_flags(vma, address, (pte_t *)pmdp,
1214 							pmd_pte(entry), dirty);
1215 }
1216 
1217 static inline int pud_devmap(pud_t pud)
1218 {
1219 	return 0;
1220 }
1221 
1222 static inline int pgd_devmap(pgd_t pgd)
1223 {
1224 	return 0;
1225 }
1226 #endif
1227 
1228 #ifdef CONFIG_PAGE_TABLE_CHECK
1229 static inline bool pte_user_accessible_page(pte_t pte)
1230 {
1231 	return pte_valid(pte) && (pte_user(pte) || pte_user_exec(pte));
1232 }
1233 
1234 static inline bool pmd_user_accessible_page(pmd_t pmd)
1235 {
1236 	return pmd_valid(pmd) && !pmd_table(pmd) && (pmd_user(pmd) || pmd_user_exec(pmd));
1237 }
1238 
1239 static inline bool pud_user_accessible_page(pud_t pud)
1240 {
1241 	return pud_valid(pud) && !pud_table(pud) && (pud_user(pud) || pud_user_exec(pud));
1242 }
1243 #endif
1244 
1245 /*
1246  * Atomic pte/pmd modifications.
1247  */
1248 static inline int __ptep_test_and_clear_young(struct vm_area_struct *vma,
1249 					      unsigned long address,
1250 					      pte_t *ptep)
1251 {
1252 	pte_t old_pte, pte;
1253 
1254 	pte = __ptep_get(ptep);
1255 	do {
1256 		old_pte = pte;
1257 		pte = pte_mkold(pte);
1258 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
1259 					       pte_val(old_pte), pte_val(pte));
1260 	} while (pte_val(pte) != pte_val(old_pte));
1261 
1262 	return pte_young(pte);
1263 }
1264 
1265 static inline int __ptep_clear_flush_young(struct vm_area_struct *vma,
1266 					 unsigned long address, pte_t *ptep)
1267 {
1268 	int young = __ptep_test_and_clear_young(vma, address, ptep);
1269 
1270 	if (young) {
1271 		/*
1272 		 * We can elide the trailing DSB here since the worst that can
1273 		 * happen is that a CPU continues to use the young entry in its
1274 		 * TLB and we mistakenly reclaim the associated page. The
1275 		 * window for such an event is bounded by the next
1276 		 * context-switch, which provides a DSB to complete the TLB
1277 		 * invalidation.
1278 		 */
1279 		flush_tlb_page_nosync(vma, address);
1280 	}
1281 
1282 	return young;
1283 }
1284 
1285 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG)
1286 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1287 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1288 					    unsigned long address,
1289 					    pmd_t *pmdp)
1290 {
1291 	/* Operation applies to PMD table entry only if FEAT_HAFT is enabled */
1292 	VM_WARN_ON(pmd_table(READ_ONCE(*pmdp)) && !system_supports_haft());
1293 	return __ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
1294 }
1295 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG */
1296 
1297 static inline pte_t __ptep_get_and_clear(struct mm_struct *mm,
1298 				       unsigned long address, pte_t *ptep)
1299 {
1300 	pte_t pte = __pte(xchg_relaxed(&pte_val(*ptep), 0));
1301 
1302 	page_table_check_pte_clear(mm, pte);
1303 
1304 	return pte;
1305 }
1306 
1307 static inline void __clear_full_ptes(struct mm_struct *mm, unsigned long addr,
1308 				pte_t *ptep, unsigned int nr, int full)
1309 {
1310 	for (;;) {
1311 		__ptep_get_and_clear(mm, addr, ptep);
1312 		if (--nr == 0)
1313 			break;
1314 		ptep++;
1315 		addr += PAGE_SIZE;
1316 	}
1317 }
1318 
1319 static inline pte_t __get_and_clear_full_ptes(struct mm_struct *mm,
1320 				unsigned long addr, pte_t *ptep,
1321 				unsigned int nr, int full)
1322 {
1323 	pte_t pte, tmp_pte;
1324 
1325 	pte = __ptep_get_and_clear(mm, addr, ptep);
1326 	while (--nr) {
1327 		ptep++;
1328 		addr += PAGE_SIZE;
1329 		tmp_pte = __ptep_get_and_clear(mm, addr, ptep);
1330 		if (pte_dirty(tmp_pte))
1331 			pte = pte_mkdirty(pte);
1332 		if (pte_young(tmp_pte))
1333 			pte = pte_mkyoung(pte);
1334 	}
1335 	return pte;
1336 }
1337 
1338 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1339 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1340 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
1341 					    unsigned long address, pmd_t *pmdp)
1342 {
1343 	pmd_t pmd = __pmd(xchg_relaxed(&pmd_val(*pmdp), 0));
1344 
1345 	page_table_check_pmd_clear(mm, pmd);
1346 
1347 	return pmd;
1348 }
1349 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1350 
1351 static inline void ___ptep_set_wrprotect(struct mm_struct *mm,
1352 					unsigned long address, pte_t *ptep,
1353 					pte_t pte)
1354 {
1355 	pte_t old_pte;
1356 
1357 	do {
1358 		old_pte = pte;
1359 		pte = pte_wrprotect(pte);
1360 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
1361 					       pte_val(old_pte), pte_val(pte));
1362 	} while (pte_val(pte) != pte_val(old_pte));
1363 }
1364 
1365 /*
1366  * __ptep_set_wrprotect - mark read-only while transferring potential hardware
1367  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
1368  */
1369 static inline void __ptep_set_wrprotect(struct mm_struct *mm,
1370 					unsigned long address, pte_t *ptep)
1371 {
1372 	___ptep_set_wrprotect(mm, address, ptep, __ptep_get(ptep));
1373 }
1374 
1375 static inline void __wrprotect_ptes(struct mm_struct *mm, unsigned long address,
1376 				pte_t *ptep, unsigned int nr)
1377 {
1378 	unsigned int i;
1379 
1380 	for (i = 0; i < nr; i++, address += PAGE_SIZE, ptep++)
1381 		__ptep_set_wrprotect(mm, address, ptep);
1382 }
1383 
1384 static inline void __clear_young_dirty_pte(struct vm_area_struct *vma,
1385 					   unsigned long addr, pte_t *ptep,
1386 					   pte_t pte, cydp_t flags)
1387 {
1388 	pte_t old_pte;
1389 
1390 	do {
1391 		old_pte = pte;
1392 
1393 		if (flags & CYDP_CLEAR_YOUNG)
1394 			pte = pte_mkold(pte);
1395 		if (flags & CYDP_CLEAR_DIRTY)
1396 			pte = pte_mkclean(pte);
1397 
1398 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
1399 					       pte_val(old_pte), pte_val(pte));
1400 	} while (pte_val(pte) != pte_val(old_pte));
1401 }
1402 
1403 static inline void __clear_young_dirty_ptes(struct vm_area_struct *vma,
1404 					    unsigned long addr, pte_t *ptep,
1405 					    unsigned int nr, cydp_t flags)
1406 {
1407 	pte_t pte;
1408 
1409 	for (;;) {
1410 		pte = __ptep_get(ptep);
1411 
1412 		if (flags == (CYDP_CLEAR_YOUNG | CYDP_CLEAR_DIRTY))
1413 			__set_pte(ptep, pte_mkclean(pte_mkold(pte)));
1414 		else
1415 			__clear_young_dirty_pte(vma, addr, ptep, pte, flags);
1416 
1417 		if (--nr == 0)
1418 			break;
1419 		ptep++;
1420 		addr += PAGE_SIZE;
1421 	}
1422 }
1423 
1424 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1425 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1426 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1427 				      unsigned long address, pmd_t *pmdp)
1428 {
1429 	__ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
1430 }
1431 
1432 #define pmdp_establish pmdp_establish
1433 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
1434 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
1435 {
1436 	page_table_check_pmd_set(vma->vm_mm, pmdp, pmd);
1437 	return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
1438 }
1439 #endif
1440 
1441 /*
1442  * Encode and decode a swap entry:
1443  *	bits 0-1:	present (must be zero)
1444  *	bits 2:		remember PG_anon_exclusive
1445  *	bit  3:		remember uffd-wp state
1446  *	bits 6-10:	swap type
1447  *	bit  11:	PTE_PRESENT_INVALID (must be zero)
1448  *	bits 12-61:	swap offset
1449  */
1450 #define __SWP_TYPE_SHIFT	6
1451 #define __SWP_TYPE_BITS		5
1452 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
1453 #define __SWP_OFFSET_SHIFT	12
1454 #define __SWP_OFFSET_BITS	50
1455 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
1456 
1457 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
1458 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
1459 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
1460 
1461 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1462 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
1463 
1464 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1465 #define __pmd_to_swp_entry(pmd)		((swp_entry_t) { pmd_val(pmd) })
1466 #define __swp_entry_to_pmd(swp)		__pmd((swp).val)
1467 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
1468 
1469 /*
1470  * Ensure that there are not more swap files than can be encoded in the kernel
1471  * PTEs.
1472  */
1473 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
1474 
1475 #ifdef CONFIG_ARM64_MTE
1476 
1477 #define __HAVE_ARCH_PREPARE_TO_SWAP
1478 extern int arch_prepare_to_swap(struct folio *folio);
1479 
1480 #define __HAVE_ARCH_SWAP_INVALIDATE
1481 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
1482 {
1483 	if (system_supports_mte())
1484 		mte_invalidate_tags(type, offset);
1485 }
1486 
1487 static inline void arch_swap_invalidate_area(int type)
1488 {
1489 	if (system_supports_mte())
1490 		mte_invalidate_tags_area(type);
1491 }
1492 
1493 #define __HAVE_ARCH_SWAP_RESTORE
1494 extern void arch_swap_restore(swp_entry_t entry, struct folio *folio);
1495 
1496 #endif /* CONFIG_ARM64_MTE */
1497 
1498 /*
1499  * On AArch64, the cache coherency is handled via the __set_ptes() function.
1500  */
1501 static inline void update_mmu_cache_range(struct vm_fault *vmf,
1502 		struct vm_area_struct *vma, unsigned long addr, pte_t *ptep,
1503 		unsigned int nr)
1504 {
1505 	/*
1506 	 * We don't do anything here, so there's a very small chance of
1507 	 * us retaking a user fault which we just fixed up. The alternative
1508 	 * is doing a dsb(ishst), but that penalises the fastpath.
1509 	 */
1510 }
1511 
1512 #define update_mmu_cache(vma, addr, ptep) \
1513 	update_mmu_cache_range(NULL, vma, addr, ptep, 1)
1514 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
1515 
1516 #ifdef CONFIG_ARM64_PA_BITS_52
1517 #define phys_to_ttbr(addr)	(((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
1518 #else
1519 #define phys_to_ttbr(addr)	(addr)
1520 #endif
1521 
1522 /*
1523  * On arm64 without hardware Access Flag, copying from user will fail because
1524  * the pte is old and cannot be marked young. So we always end up with zeroed
1525  * page after fork() + CoW for pfn mappings. We don't always have a
1526  * hardware-managed access flag on arm64.
1527  */
1528 #define arch_has_hw_pte_young		cpu_has_hw_af
1529 
1530 #ifdef CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG
1531 #define arch_has_hw_nonleaf_pmd_young	system_supports_haft
1532 #endif
1533 
1534 /*
1535  * Experimentally, it's cheap to set the access flag in hardware and we
1536  * benefit from prefaulting mappings as 'old' to start with.
1537  */
1538 #define arch_wants_old_prefaulted_pte	cpu_has_hw_af
1539 
1540 static inline bool pud_sect_supported(void)
1541 {
1542 	return PAGE_SIZE == SZ_4K;
1543 }
1544 
1545 
1546 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1547 #define ptep_modify_prot_start ptep_modify_prot_start
1548 extern pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
1549 				    unsigned long addr, pte_t *ptep);
1550 
1551 #define ptep_modify_prot_commit ptep_modify_prot_commit
1552 extern void ptep_modify_prot_commit(struct vm_area_struct *vma,
1553 				    unsigned long addr, pte_t *ptep,
1554 				    pte_t old_pte, pte_t new_pte);
1555 
1556 #ifdef CONFIG_ARM64_CONTPTE
1557 
1558 /*
1559  * The contpte APIs are used to transparently manage the contiguous bit in ptes
1560  * where it is possible and makes sense to do so. The PTE_CONT bit is considered
1561  * a private implementation detail of the public ptep API (see below).
1562  */
1563 extern void __contpte_try_fold(struct mm_struct *mm, unsigned long addr,
1564 				pte_t *ptep, pte_t pte);
1565 extern void __contpte_try_unfold(struct mm_struct *mm, unsigned long addr,
1566 				pte_t *ptep, pte_t pte);
1567 extern pte_t contpte_ptep_get(pte_t *ptep, pte_t orig_pte);
1568 extern pte_t contpte_ptep_get_lockless(pte_t *orig_ptep);
1569 extern void contpte_set_ptes(struct mm_struct *mm, unsigned long addr,
1570 				pte_t *ptep, pte_t pte, unsigned int nr);
1571 extern void contpte_clear_full_ptes(struct mm_struct *mm, unsigned long addr,
1572 				pte_t *ptep, unsigned int nr, int full);
1573 extern pte_t contpte_get_and_clear_full_ptes(struct mm_struct *mm,
1574 				unsigned long addr, pte_t *ptep,
1575 				unsigned int nr, int full);
1576 extern int contpte_ptep_test_and_clear_young(struct vm_area_struct *vma,
1577 				unsigned long addr, pte_t *ptep);
1578 extern int contpte_ptep_clear_flush_young(struct vm_area_struct *vma,
1579 				unsigned long addr, pte_t *ptep);
1580 extern void contpte_wrprotect_ptes(struct mm_struct *mm, unsigned long addr,
1581 				pte_t *ptep, unsigned int nr);
1582 extern int contpte_ptep_set_access_flags(struct vm_area_struct *vma,
1583 				unsigned long addr, pte_t *ptep,
1584 				pte_t entry, int dirty);
1585 extern void contpte_clear_young_dirty_ptes(struct vm_area_struct *vma,
1586 				unsigned long addr, pte_t *ptep,
1587 				unsigned int nr, cydp_t flags);
1588 
1589 static __always_inline void contpte_try_fold(struct mm_struct *mm,
1590 				unsigned long addr, pte_t *ptep, pte_t pte)
1591 {
1592 	/*
1593 	 * Only bother trying if both the virtual and physical addresses are
1594 	 * aligned and correspond to the last entry in a contig range. The core
1595 	 * code mostly modifies ranges from low to high, so this is the likely
1596 	 * the last modification in the contig range, so a good time to fold.
1597 	 * We can't fold special mappings, because there is no associated folio.
1598 	 */
1599 
1600 	const unsigned long contmask = CONT_PTES - 1;
1601 	bool valign = ((addr >> PAGE_SHIFT) & contmask) == contmask;
1602 
1603 	if (unlikely(valign)) {
1604 		bool palign = (pte_pfn(pte) & contmask) == contmask;
1605 
1606 		if (unlikely(palign &&
1607 		    pte_valid(pte) && !pte_cont(pte) && !pte_special(pte)))
1608 			__contpte_try_fold(mm, addr, ptep, pte);
1609 	}
1610 }
1611 
1612 static __always_inline void contpte_try_unfold(struct mm_struct *mm,
1613 				unsigned long addr, pte_t *ptep, pte_t pte)
1614 {
1615 	if (unlikely(pte_valid_cont(pte)))
1616 		__contpte_try_unfold(mm, addr, ptep, pte);
1617 }
1618 
1619 #define pte_batch_hint pte_batch_hint
1620 static inline unsigned int pte_batch_hint(pte_t *ptep, pte_t pte)
1621 {
1622 	if (!pte_valid_cont(pte))
1623 		return 1;
1624 
1625 	return CONT_PTES - (((unsigned long)ptep >> 3) & (CONT_PTES - 1));
1626 }
1627 
1628 /*
1629  * The below functions constitute the public API that arm64 presents to the
1630  * core-mm to manipulate PTE entries within their page tables (or at least this
1631  * is the subset of the API that arm64 needs to implement). These public
1632  * versions will automatically and transparently apply the contiguous bit where
1633  * it makes sense to do so. Therefore any users that are contig-aware (e.g.
1634  * hugetlb, kernel mapper) should NOT use these APIs, but instead use the
1635  * private versions, which are prefixed with double underscore. All of these
1636  * APIs except for ptep_get_lockless() are expected to be called with the PTL
1637  * held. Although the contiguous bit is considered private to the
1638  * implementation, it is deliberately allowed to leak through the getters (e.g.
1639  * ptep_get()), back to core code. This is required so that pte_leaf_size() can
1640  * provide an accurate size for perf_get_pgtable_size(). But this leakage means
1641  * its possible a pte will be passed to a setter with the contiguous bit set, so
1642  * we explicitly clear the contiguous bit in those cases to prevent accidentally
1643  * setting it in the pgtable.
1644  */
1645 
1646 #define ptep_get ptep_get
1647 static inline pte_t ptep_get(pte_t *ptep)
1648 {
1649 	pte_t pte = __ptep_get(ptep);
1650 
1651 	if (likely(!pte_valid_cont(pte)))
1652 		return pte;
1653 
1654 	return contpte_ptep_get(ptep, pte);
1655 }
1656 
1657 #define ptep_get_lockless ptep_get_lockless
1658 static inline pte_t ptep_get_lockless(pte_t *ptep)
1659 {
1660 	pte_t pte = __ptep_get(ptep);
1661 
1662 	if (likely(!pte_valid_cont(pte)))
1663 		return pte;
1664 
1665 	return contpte_ptep_get_lockless(ptep);
1666 }
1667 
1668 static inline void set_pte(pte_t *ptep, pte_t pte)
1669 {
1670 	/*
1671 	 * We don't have the mm or vaddr so cannot unfold contig entries (since
1672 	 * it requires tlb maintenance). set_pte() is not used in core code, so
1673 	 * this should never even be called. Regardless do our best to service
1674 	 * any call and emit a warning if there is any attempt to set a pte on
1675 	 * top of an existing contig range.
1676 	 */
1677 	pte_t orig_pte = __ptep_get(ptep);
1678 
1679 	WARN_ON_ONCE(pte_valid_cont(orig_pte));
1680 	__set_pte(ptep, pte_mknoncont(pte));
1681 }
1682 
1683 #define set_ptes set_ptes
1684 static __always_inline void set_ptes(struct mm_struct *mm, unsigned long addr,
1685 				pte_t *ptep, pte_t pte, unsigned int nr)
1686 {
1687 	pte = pte_mknoncont(pte);
1688 
1689 	if (likely(nr == 1)) {
1690 		contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1691 		__set_ptes(mm, addr, ptep, pte, 1);
1692 		contpte_try_fold(mm, addr, ptep, pte);
1693 	} else {
1694 		contpte_set_ptes(mm, addr, ptep, pte, nr);
1695 	}
1696 }
1697 
1698 static inline void pte_clear(struct mm_struct *mm,
1699 				unsigned long addr, pte_t *ptep)
1700 {
1701 	contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1702 	__pte_clear(mm, addr, ptep);
1703 }
1704 
1705 #define clear_full_ptes clear_full_ptes
1706 static inline void clear_full_ptes(struct mm_struct *mm, unsigned long addr,
1707 				pte_t *ptep, unsigned int nr, int full)
1708 {
1709 	if (likely(nr == 1)) {
1710 		contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1711 		__clear_full_ptes(mm, addr, ptep, nr, full);
1712 	} else {
1713 		contpte_clear_full_ptes(mm, addr, ptep, nr, full);
1714 	}
1715 }
1716 
1717 #define get_and_clear_full_ptes get_and_clear_full_ptes
1718 static inline pte_t get_and_clear_full_ptes(struct mm_struct *mm,
1719 				unsigned long addr, pte_t *ptep,
1720 				unsigned int nr, int full)
1721 {
1722 	pte_t pte;
1723 
1724 	if (likely(nr == 1)) {
1725 		contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1726 		pte = __get_and_clear_full_ptes(mm, addr, ptep, nr, full);
1727 	} else {
1728 		pte = contpte_get_and_clear_full_ptes(mm, addr, ptep, nr, full);
1729 	}
1730 
1731 	return pte;
1732 }
1733 
1734 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1735 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1736 				unsigned long addr, pte_t *ptep)
1737 {
1738 	contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1739 	return __ptep_get_and_clear(mm, addr, ptep);
1740 }
1741 
1742 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1743 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
1744 				unsigned long addr, pte_t *ptep)
1745 {
1746 	pte_t orig_pte = __ptep_get(ptep);
1747 
1748 	if (likely(!pte_valid_cont(orig_pte)))
1749 		return __ptep_test_and_clear_young(vma, addr, ptep);
1750 
1751 	return contpte_ptep_test_and_clear_young(vma, addr, ptep);
1752 }
1753 
1754 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1755 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
1756 				unsigned long addr, pte_t *ptep)
1757 {
1758 	pte_t orig_pte = __ptep_get(ptep);
1759 
1760 	if (likely(!pte_valid_cont(orig_pte)))
1761 		return __ptep_clear_flush_young(vma, addr, ptep);
1762 
1763 	return contpte_ptep_clear_flush_young(vma, addr, ptep);
1764 }
1765 
1766 #define wrprotect_ptes wrprotect_ptes
1767 static __always_inline void wrprotect_ptes(struct mm_struct *mm,
1768 				unsigned long addr, pte_t *ptep, unsigned int nr)
1769 {
1770 	if (likely(nr == 1)) {
1771 		/*
1772 		 * Optimization: wrprotect_ptes() can only be called for present
1773 		 * ptes so we only need to check contig bit as condition for
1774 		 * unfold, and we can remove the contig bit from the pte we read
1775 		 * to avoid re-reading. This speeds up fork() which is sensitive
1776 		 * for order-0 folios. Equivalent to contpte_try_unfold().
1777 		 */
1778 		pte_t orig_pte = __ptep_get(ptep);
1779 
1780 		if (unlikely(pte_cont(orig_pte))) {
1781 			__contpte_try_unfold(mm, addr, ptep, orig_pte);
1782 			orig_pte = pte_mknoncont(orig_pte);
1783 		}
1784 		___ptep_set_wrprotect(mm, addr, ptep, orig_pte);
1785 	} else {
1786 		contpte_wrprotect_ptes(mm, addr, ptep, nr);
1787 	}
1788 }
1789 
1790 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1791 static inline void ptep_set_wrprotect(struct mm_struct *mm,
1792 				unsigned long addr, pte_t *ptep)
1793 {
1794 	wrprotect_ptes(mm, addr, ptep, 1);
1795 }
1796 
1797 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1798 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1799 				unsigned long addr, pte_t *ptep,
1800 				pte_t entry, int dirty)
1801 {
1802 	pte_t orig_pte = __ptep_get(ptep);
1803 
1804 	entry = pte_mknoncont(entry);
1805 
1806 	if (likely(!pte_valid_cont(orig_pte)))
1807 		return __ptep_set_access_flags(vma, addr, ptep, entry, dirty);
1808 
1809 	return contpte_ptep_set_access_flags(vma, addr, ptep, entry, dirty);
1810 }
1811 
1812 #define clear_young_dirty_ptes clear_young_dirty_ptes
1813 static inline void clear_young_dirty_ptes(struct vm_area_struct *vma,
1814 					  unsigned long addr, pte_t *ptep,
1815 					  unsigned int nr, cydp_t flags)
1816 {
1817 	if (likely(nr == 1 && !pte_cont(__ptep_get(ptep))))
1818 		__clear_young_dirty_ptes(vma, addr, ptep, nr, flags);
1819 	else
1820 		contpte_clear_young_dirty_ptes(vma, addr, ptep, nr, flags);
1821 }
1822 
1823 #else /* CONFIG_ARM64_CONTPTE */
1824 
1825 #define ptep_get				__ptep_get
1826 #define set_pte					__set_pte
1827 #define set_ptes				__set_ptes
1828 #define pte_clear				__pte_clear
1829 #define clear_full_ptes				__clear_full_ptes
1830 #define get_and_clear_full_ptes			__get_and_clear_full_ptes
1831 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1832 #define ptep_get_and_clear			__ptep_get_and_clear
1833 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1834 #define ptep_test_and_clear_young		__ptep_test_and_clear_young
1835 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1836 #define ptep_clear_flush_young			__ptep_clear_flush_young
1837 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1838 #define ptep_set_wrprotect			__ptep_set_wrprotect
1839 #define wrprotect_ptes				__wrprotect_ptes
1840 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1841 #define ptep_set_access_flags			__ptep_set_access_flags
1842 #define clear_young_dirty_ptes			__clear_young_dirty_ptes
1843 
1844 #endif /* CONFIG_ARM64_CONTPTE */
1845 
1846 #endif /* !__ASSEMBLY__ */
1847 
1848 #endif /* __ASM_PGTABLE_H */
1849