xref: /linux/arch/arm64/include/asm/pgtable.h (revision 6215d9f4470fbb48245ffdfade821685e2728c65)
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 __ASSEMBLER__
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 static inline void emit_pte_barriers(void)
44 {
45 	/*
46 	 * These barriers are emitted under certain conditions after a pte entry
47 	 * was modified (see e.g. __set_pte_complete()). The dsb makes the store
48 	 * visible to the table walker. The isb ensures that any previous
49 	 * speculative "invalid translation" marker that is in the CPU's
50 	 * pipeline gets cleared, so that any access to that address after
51 	 * setting the pte to valid won't cause a spurious fault. If the thread
52 	 * gets preempted after storing to the pgtable but before emitting these
53 	 * barriers, __switch_to() emits a dsb which ensure the walker gets to
54 	 * see the store. There is no guarantee of an isb being issued though.
55 	 * This is safe because it will still get issued (albeit on a
56 	 * potentially different CPU) when the thread starts running again,
57 	 * before any access to the address.
58 	 */
59 	dsb(ishst);
60 	isb();
61 }
62 
63 static inline void queue_pte_barriers(void)
64 {
65 	if (is_lazy_mmu_mode_active()) {
66 		/* Avoid the atomic op if already set. */
67 		if (!test_thread_flag(TIF_LAZY_MMU_PENDING))
68 			set_thread_flag(TIF_LAZY_MMU_PENDING);
69 	} else {
70 		emit_pte_barriers();
71 	}
72 }
73 
74 static inline void arch_enter_lazy_mmu_mode(void) {}
75 
76 static inline void arch_flush_lazy_mmu_mode(void)
77 {
78 	if (test_and_clear_thread_flag(TIF_LAZY_MMU_PENDING))
79 		emit_pte_barriers();
80 }
81 
82 static inline void arch_leave_lazy_mmu_mode(void)
83 {
84 	arch_flush_lazy_mmu_mode();
85 }
86 
87 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
88 #define __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
89 
90 /* Set stride and tlb_level in flush_*_tlb_range */
91 #define flush_pmd_tlb_range(vma, addr, end)	\
92 	__flush_tlb_range(vma, addr, end, PMD_SIZE, false, 2)
93 #define flush_pud_tlb_range(vma, addr, end)	\
94 	__flush_tlb_range(vma, addr, end, PUD_SIZE, false, 1)
95 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
96 
97 /*
98  * We use local TLB invalidation instruction when reusing page in
99  * write protection fault handler to avoid TLBI broadcast in the hot
100  * path.  This will cause spurious page faults if stale read-only TLB
101  * entries exist.
102  */
103 #define flush_tlb_fix_spurious_fault(vma, address, ptep)	\
104 	local_flush_tlb_page_nonotify(vma, address)
105 
106 #define flush_tlb_fix_spurious_fault_pmd(vma, address, pmdp)	\
107 	local_flush_tlb_page_nonotify(vma, address)
108 
109 /*
110  * ZERO_PAGE is a global shared page that is always zero: used
111  * for zero-mapped memory areas etc..
112  */
113 #define ZERO_PAGE(vaddr)	phys_to_page(__pa_symbol(empty_zero_page))
114 
115 #define pte_ERROR(e)	\
116 	pr_err("%s:%d: bad pte %016llx.\n", __FILE__, __LINE__, pte_val(e))
117 
118 #ifdef CONFIG_ARM64_PA_BITS_52
119 static inline phys_addr_t __pte_to_phys(pte_t pte)
120 {
121 	pte_val(pte) &= ~PTE_MAYBE_SHARED;
122 	return (pte_val(pte) & PTE_ADDR_LOW) |
123 		((pte_val(pte) & PTE_ADDR_HIGH) << PTE_ADDR_HIGH_SHIFT);
124 }
125 static inline pteval_t __phys_to_pte_val(phys_addr_t phys)
126 {
127 	return (phys | (phys >> PTE_ADDR_HIGH_SHIFT)) & PHYS_TO_PTE_ADDR_MASK;
128 }
129 #else
130 static inline phys_addr_t __pte_to_phys(pte_t pte)
131 {
132 	return pte_val(pte) & PTE_ADDR_LOW;
133 }
134 
135 static inline pteval_t __phys_to_pte_val(phys_addr_t phys)
136 {
137 	return phys;
138 }
139 #endif
140 
141 #define pte_pfn(pte)		(__pte_to_phys(pte) >> PAGE_SHIFT)
142 #define pfn_pte(pfn,prot)	\
143 	__pte(__phys_to_pte_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
144 
145 #define pte_none(pte)		(!pte_val(pte))
146 #define pte_page(pte)		(pfn_to_page(pte_pfn(pte)))
147 
148 /*
149  * The following only work if pte_present(). Undefined behaviour otherwise.
150  */
151 #define pte_present(pte)	(pte_valid(pte) || pte_present_invalid(pte))
152 #define pte_young(pte)		(!!(pte_val(pte) & PTE_AF))
153 #define pte_special(pte)	(!!(pte_val(pte) & PTE_SPECIAL))
154 #define pte_write(pte)		(!!(pte_val(pte) & PTE_WRITE))
155 #define pte_rdonly(pte)		(!!(pte_val(pte) & PTE_RDONLY))
156 #define pte_user(pte)		(!!(pte_val(pte) & PTE_USER))
157 #define pte_user_exec(pte)	(!(pte_val(pte) & PTE_UXN))
158 #define pte_cont(pte)		(!!(pte_val(pte) & PTE_CONT))
159 #define pte_tagged(pte)		((pte_val(pte) & PTE_ATTRINDX_MASK) == \
160 				 PTE_ATTRINDX(MT_NORMAL_TAGGED))
161 
162 #define pte_cont_addr_end(addr, end)						\
163 ({	unsigned long __boundary = ((addr) + CONT_PTE_SIZE) & CONT_PTE_MASK;	\
164 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
165 })
166 
167 #define pmd_cont_addr_end(addr, end)						\
168 ({	unsigned long __boundary = ((addr) + CONT_PMD_SIZE) & CONT_PMD_MASK;	\
169 	(__boundary - 1 < (end) - 1) ? __boundary : (end);			\
170 })
171 
172 #define pte_hw_dirty(pte)	(pte_write(pte) && !pte_rdonly(pte))
173 #define pte_sw_dirty(pte)	(!!(pte_val(pte) & PTE_DIRTY))
174 #define pte_dirty(pte)		(pte_sw_dirty(pte) || pte_hw_dirty(pte))
175 
176 #define pte_valid(pte)		(!!(pte_val(pte) & PTE_VALID))
177 #define pte_present_invalid(pte) \
178 	((pte_val(pte) & (PTE_VALID | PTE_PRESENT_INVALID)) == PTE_PRESENT_INVALID)
179 /*
180  * Execute-only user mappings do not have the PTE_USER bit set. All valid
181  * kernel mappings have the PTE_UXN bit set.
182  */
183 #define pte_valid_not_user(pte) \
184 	((pte_val(pte) & (PTE_VALID | PTE_USER | PTE_UXN)) == (PTE_VALID | PTE_UXN))
185 /*
186  * Returns true if the pte is valid and has the contiguous bit set.
187  */
188 #define pte_valid_cont(pte)	(pte_valid(pte) && pte_cont(pte))
189 /*
190  * Could the pte be present in the TLB? We must check mm_tlb_flush_pending
191  * so that we don't erroneously return false for pages that have been
192  * remapped as PROT_NONE but are yet to be flushed from the TLB.
193  * Note that we can't make any assumptions based on the state of the access
194  * flag, since __ptep_clear_flush_young() elides a DSB when invalidating the
195  * TLB.
196  */
197 #define pte_accessible(mm, pte)	\
198 	(mm_tlb_flush_pending(mm) ? pte_present(pte) : pte_valid(pte))
199 
200 static inline bool por_el0_allows_pkey(u8 pkey, bool write, bool execute)
201 {
202 	u64 por;
203 
204 	if (!system_supports_poe())
205 		return true;
206 
207 	por = read_sysreg_s(SYS_POR_EL0);
208 
209 	if (write)
210 		return por_elx_allows_write(por, pkey);
211 
212 	if (execute)
213 		return por_elx_allows_exec(por, pkey);
214 
215 	return por_elx_allows_read(por, pkey);
216 }
217 
218 /*
219  * p??_access_permitted() is true for valid user mappings (PTE_USER
220  * bit set, subject to the write permission check). For execute-only
221  * mappings, like PROT_EXEC with EPAN (both PTE_USER and PTE_UXN bits
222  * not set) must return false. PROT_NONE mappings do not have the
223  * PTE_VALID bit set.
224  */
225 #define pte_access_permitted_no_overlay(pte, write) \
226 	(((pte_val(pte) & (PTE_VALID | PTE_USER)) == (PTE_VALID | PTE_USER)) && (!(write) || pte_write(pte)))
227 #define pte_access_permitted(pte, write) \
228 	(pte_access_permitted_no_overlay(pte, write) && \
229 	por_el0_allows_pkey(FIELD_GET(PTE_PO_IDX_MASK, pte_val(pte)), write, false))
230 #define pmd_access_permitted(pmd, write) \
231 	(pte_access_permitted(pmd_pte(pmd), (write)))
232 #define pud_access_permitted(pud, write) \
233 	(pte_access_permitted(pud_pte(pud), (write)))
234 
235 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
236 {
237 	pte_val(pte) &= ~pgprot_val(prot);
238 	return pte;
239 }
240 
241 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
242 {
243 	pte_val(pte) |= pgprot_val(prot);
244 	return pte;
245 }
246 
247 static inline pmd_t clear_pmd_bit(pmd_t pmd, pgprot_t prot)
248 {
249 	pmd_val(pmd) &= ~pgprot_val(prot);
250 	return pmd;
251 }
252 
253 static inline pmd_t set_pmd_bit(pmd_t pmd, pgprot_t prot)
254 {
255 	pmd_val(pmd) |= pgprot_val(prot);
256 	return pmd;
257 }
258 
259 static inline pte_t pte_mkwrite_novma(pte_t pte)
260 {
261 	pte = set_pte_bit(pte, __pgprot(PTE_WRITE));
262 	if (pte_sw_dirty(pte))
263 		pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
264 	return pte;
265 }
266 
267 static inline pte_t pte_mkclean(pte_t pte)
268 {
269 	pte = clear_pte_bit(pte, __pgprot(PTE_DIRTY));
270 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
271 
272 	return pte;
273 }
274 
275 static inline pte_t pte_mkdirty(pte_t pte)
276 {
277 	pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
278 
279 	if (pte_write(pte))
280 		pte = clear_pte_bit(pte, __pgprot(PTE_RDONLY));
281 
282 	return pte;
283 }
284 
285 static inline pte_t pte_wrprotect(pte_t pte)
286 {
287 	/*
288 	 * If hardware-dirty (PTE_WRITE/DBM bit set and PTE_RDONLY
289 	 * clear), set the PTE_DIRTY bit.
290 	 */
291 	if (pte_hw_dirty(pte))
292 		pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
293 
294 	pte = clear_pte_bit(pte, __pgprot(PTE_WRITE));
295 	pte = set_pte_bit(pte, __pgprot(PTE_RDONLY));
296 	return pte;
297 }
298 
299 static inline pte_t pte_mkold(pte_t pte)
300 {
301 	return clear_pte_bit(pte, __pgprot(PTE_AF));
302 }
303 
304 static inline pte_t pte_mkyoung(pte_t pte)
305 {
306 	return set_pte_bit(pte, __pgprot(PTE_AF));
307 }
308 
309 static inline pte_t pte_mkspecial(pte_t pte)
310 {
311 	return set_pte_bit(pte, __pgprot(PTE_SPECIAL));
312 }
313 
314 static inline pte_t pte_mkcont(pte_t pte)
315 {
316 	return set_pte_bit(pte, __pgprot(PTE_CONT));
317 }
318 
319 static inline pte_t pte_mknoncont(pte_t pte)
320 {
321 	return clear_pte_bit(pte, __pgprot(PTE_CONT));
322 }
323 
324 static inline pte_t pte_mkvalid(pte_t pte)
325 {
326 	return set_pte_bit(pte, __pgprot(PTE_VALID));
327 }
328 
329 static inline pte_t pte_mkinvalid(pte_t pte)
330 {
331 	pte = set_pte_bit(pte, __pgprot(PTE_PRESENT_INVALID));
332 	pte = clear_pte_bit(pte, __pgprot(PTE_VALID));
333 	return pte;
334 }
335 
336 static inline pmd_t pmd_mkcont(pmd_t pmd)
337 {
338 	return __pmd(pmd_val(pmd) | PMD_SECT_CONT);
339 }
340 
341 static inline pmd_t pmd_mknoncont(pmd_t pmd)
342 {
343 	return __pmd(pmd_val(pmd) & ~PMD_SECT_CONT);
344 }
345 
346 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
347 static inline int pte_uffd_wp(pte_t pte)
348 {
349 	return !!(pte_val(pte) & PTE_UFFD_WP);
350 }
351 
352 static inline pte_t pte_mkuffd_wp(pte_t pte)
353 {
354 	return pte_wrprotect(set_pte_bit(pte, __pgprot(PTE_UFFD_WP)));
355 }
356 
357 static inline pte_t pte_clear_uffd_wp(pte_t pte)
358 {
359 	return clear_pte_bit(pte, __pgprot(PTE_UFFD_WP));
360 }
361 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
362 
363 static inline void __set_pte_nosync(pte_t *ptep, pte_t pte)
364 {
365 	WRITE_ONCE(*ptep, pte);
366 }
367 
368 static inline void __set_pte_complete(pte_t pte)
369 {
370 	/*
371 	 * Only if the new pte is valid and kernel, otherwise TLB maintenance
372 	 * has the necessary barriers.
373 	 */
374 	if (pte_valid_not_user(pte))
375 		queue_pte_barriers();
376 }
377 
378 static inline void __set_pte(pte_t *ptep, pte_t pte)
379 {
380 	__set_pte_nosync(ptep, pte);
381 	__set_pte_complete(pte);
382 }
383 
384 static inline pte_t __ptep_get(pte_t *ptep)
385 {
386 	return READ_ONCE(*ptep);
387 }
388 
389 extern void __sync_icache_dcache(pte_t pteval);
390 bool pgattr_change_is_safe(pteval_t old, pteval_t new);
391 
392 /*
393  * PTE bits configuration in the presence of hardware Dirty Bit Management
394  * (PTE_WRITE == PTE_DBM):
395  *
396  * Dirty  Writable | PTE_RDONLY  PTE_WRITE  PTE_DIRTY (sw)
397  *   0      0      |   1           0          0
398  *   0      1      |   1           1          0
399  *   1      0      |   1           0          1
400  *   1      1      |   0           1          x
401  *
402  * When hardware DBM is not present, the software PTE_DIRTY bit is updated via
403  * the page fault mechanism. Checking the dirty status of a pte becomes:
404  *
405  *   PTE_DIRTY || (PTE_WRITE && !PTE_RDONLY)
406  */
407 
408 static inline void __check_safe_pte_update(struct mm_struct *mm, pte_t *ptep,
409 					   pte_t pte)
410 {
411 	pte_t old_pte;
412 
413 	if (!IS_ENABLED(CONFIG_DEBUG_VM))
414 		return;
415 
416 	old_pte = __ptep_get(ptep);
417 
418 	if (!pte_valid(old_pte) || !pte_valid(pte))
419 		return;
420 	if (mm != current->active_mm && atomic_read(&mm->mm_users) <= 1)
421 		return;
422 
423 	/*
424 	 * Check for potential race with hardware updates of the pte
425 	 * (__ptep_set_access_flags safely changes valid ptes without going
426 	 * through an invalid entry).
427 	 */
428 	VM_WARN_ONCE(!pte_young(pte),
429 		     "%s: racy access flag clearing: 0x%016llx -> 0x%016llx",
430 		     __func__, pte_val(old_pte), pte_val(pte));
431 	VM_WARN_ONCE(pte_write(old_pte) && !pte_dirty(pte),
432 		     "%s: racy dirty state clearing: 0x%016llx -> 0x%016llx",
433 		     __func__, pte_val(old_pte), pte_val(pte));
434 	VM_WARN_ONCE(!pgattr_change_is_safe(pte_val(old_pte), pte_val(pte)),
435 		     "%s: unsafe attribute change: 0x%016llx -> 0x%016llx",
436 		     __func__, pte_val(old_pte), pte_val(pte));
437 }
438 
439 static inline void __sync_cache_and_tags(pte_t pte, unsigned int nr_pages)
440 {
441 	if (pte_present(pte) && pte_user_exec(pte) && !pte_special(pte))
442 		__sync_icache_dcache(pte);
443 
444 	/*
445 	 * If the PTE would provide user space access to the tags associated
446 	 * with it then ensure that the MTE tags are synchronised.  Although
447 	 * pte_access_permitted_no_overlay() returns false for exec only
448 	 * mappings, they don't expose tags (instruction fetches don't check
449 	 * tags).
450 	 */
451 	if (system_supports_mte() && pte_access_permitted_no_overlay(pte, false) &&
452 	    !pte_special(pte) && pte_tagged(pte))
453 		mte_sync_tags(pte, nr_pages);
454 }
455 
456 /*
457  * Select all bits except the pfn
458  */
459 #define pte_pgprot pte_pgprot
460 static inline pgprot_t pte_pgprot(pte_t pte)
461 {
462 	unsigned long pfn = pte_pfn(pte);
463 
464 	return __pgprot(pte_val(pfn_pte(pfn, __pgprot(0))) ^ pte_val(pte));
465 }
466 
467 #define pte_advance_pfn pte_advance_pfn
468 static inline pte_t pte_advance_pfn(pte_t pte, unsigned long nr)
469 {
470 	return pfn_pte(pte_pfn(pte) + nr, pte_pgprot(pte));
471 }
472 
473 /*
474  * Hugetlb definitions.
475  */
476 #define HUGE_MAX_HSTATE		4
477 #define HPAGE_SHIFT		PMD_SHIFT
478 #define HPAGE_SIZE		(_AC(1, UL) << HPAGE_SHIFT)
479 #define HPAGE_MASK		(~(HPAGE_SIZE - 1))
480 #define HUGETLB_PAGE_ORDER	(HPAGE_SHIFT - PAGE_SHIFT)
481 
482 static inline pte_t pgd_pte(pgd_t pgd)
483 {
484 	return __pte(pgd_val(pgd));
485 }
486 
487 static inline pte_t p4d_pte(p4d_t p4d)
488 {
489 	return __pte(p4d_val(p4d));
490 }
491 
492 static inline pte_t pud_pte(pud_t pud)
493 {
494 	return __pte(pud_val(pud));
495 }
496 
497 static inline pud_t pte_pud(pte_t pte)
498 {
499 	return __pud(pte_val(pte));
500 }
501 
502 static inline pmd_t pud_pmd(pud_t pud)
503 {
504 	return __pmd(pud_val(pud));
505 }
506 
507 static inline pte_t pmd_pte(pmd_t pmd)
508 {
509 	return __pte(pmd_val(pmd));
510 }
511 
512 static inline pmd_t pte_pmd(pte_t pte)
513 {
514 	return __pmd(pte_val(pte));
515 }
516 
517 static inline pgprot_t mk_pud_sect_prot(pgprot_t prot)
518 {
519 	return __pgprot((pgprot_val(prot) & ~PUD_TYPE_MASK) | PUD_TYPE_SECT);
520 }
521 
522 static inline pgprot_t mk_pmd_sect_prot(pgprot_t prot)
523 {
524 	return __pgprot((pgprot_val(prot) & ~PMD_TYPE_MASK) | PMD_TYPE_SECT);
525 }
526 
527 static inline pte_t pte_swp_mkexclusive(pte_t pte)
528 {
529 	return set_pte_bit(pte, __pgprot(PTE_SWP_EXCLUSIVE));
530 }
531 
532 static inline bool pte_swp_exclusive(pte_t pte)
533 {
534 	return pte_val(pte) & PTE_SWP_EXCLUSIVE;
535 }
536 
537 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
538 {
539 	return clear_pte_bit(pte, __pgprot(PTE_SWP_EXCLUSIVE));
540 }
541 
542 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
543 static inline pte_t pte_swp_mkuffd_wp(pte_t pte)
544 {
545 	return set_pte_bit(pte, __pgprot(PTE_SWP_UFFD_WP));
546 }
547 
548 static inline int pte_swp_uffd_wp(pte_t pte)
549 {
550 	return !!(pte_val(pte) & PTE_SWP_UFFD_WP);
551 }
552 
553 static inline pte_t pte_swp_clear_uffd_wp(pte_t pte)
554 {
555 	return clear_pte_bit(pte, __pgprot(PTE_SWP_UFFD_WP));
556 }
557 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
558 
559 #ifdef CONFIG_NUMA_BALANCING
560 /*
561  * See the comment in include/linux/pgtable.h
562  */
563 static inline int pte_protnone(pte_t pte)
564 {
565 	/*
566 	 * pte_present_invalid() tells us that the pte is invalid from HW
567 	 * perspective but present from SW perspective, so the fields are to be
568 	 * interpreted as per the HW layout. The second 2 checks are the unique
569 	 * encoding that we use for PROT_NONE. It is insufficient to only use
570 	 * the first check because we share the same encoding scheme with pmds
571 	 * which support pmd_mkinvalid(), so can be present-invalid without
572 	 * being PROT_NONE.
573 	 */
574 	return pte_present_invalid(pte) && !pte_user(pte) && !pte_user_exec(pte);
575 }
576 
577 static inline int pmd_protnone(pmd_t pmd)
578 {
579 	return pte_protnone(pmd_pte(pmd));
580 }
581 #endif
582 
583 #define pmd_present(pmd)	pte_present(pmd_pte(pmd))
584 #define pmd_dirty(pmd)		pte_dirty(pmd_pte(pmd))
585 #define pmd_young(pmd)		pte_young(pmd_pte(pmd))
586 #define pmd_valid(pmd)		pte_valid(pmd_pte(pmd))
587 #define pmd_user(pmd)		pte_user(pmd_pte(pmd))
588 #define pmd_user_exec(pmd)	pte_user_exec(pmd_pte(pmd))
589 #define pmd_cont(pmd)		pte_cont(pmd_pte(pmd))
590 #define pmd_wrprotect(pmd)	pte_pmd(pte_wrprotect(pmd_pte(pmd)))
591 #define pmd_mkold(pmd)		pte_pmd(pte_mkold(pmd_pte(pmd)))
592 #define pmd_mkwrite_novma(pmd)	pte_pmd(pte_mkwrite_novma(pmd_pte(pmd)))
593 #define pmd_mkclean(pmd)	pte_pmd(pte_mkclean(pmd_pte(pmd)))
594 #define pmd_mkdirty(pmd)	pte_pmd(pte_mkdirty(pmd_pte(pmd)))
595 #define pmd_mkyoung(pmd)	pte_pmd(pte_mkyoung(pmd_pte(pmd)))
596 #define pmd_mkinvalid(pmd)	pte_pmd(pte_mkinvalid(pmd_pte(pmd)))
597 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
598 #define pmd_uffd_wp(pmd)	pte_uffd_wp(pmd_pte(pmd))
599 #define pmd_mkuffd_wp(pmd)	pte_pmd(pte_mkuffd_wp(pmd_pte(pmd)))
600 #define pmd_clear_uffd_wp(pmd)	pte_pmd(pte_clear_uffd_wp(pmd_pte(pmd)))
601 #define pmd_swp_uffd_wp(pmd)	pte_swp_uffd_wp(pmd_pte(pmd))
602 #define pmd_swp_mkuffd_wp(pmd)	pte_pmd(pte_swp_mkuffd_wp(pmd_pte(pmd)))
603 #define pmd_swp_clear_uffd_wp(pmd) \
604 				pte_pmd(pte_swp_clear_uffd_wp(pmd_pte(pmd)))
605 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
606 
607 #define pmd_write(pmd)		pte_write(pmd_pte(pmd))
608 
609 static inline pmd_t pmd_mkhuge(pmd_t pmd)
610 {
611 	/*
612 	 * It's possible that the pmd is present-invalid on entry
613 	 * and in that case it needs to remain present-invalid on
614 	 * exit. So ensure the VALID bit does not get modified.
615 	 */
616 	pmdval_t mask = PMD_TYPE_MASK & ~PTE_VALID;
617 	pmdval_t val = PMD_TYPE_SECT & ~PTE_VALID;
618 
619 	return __pmd((pmd_val(pmd) & ~mask) | val);
620 }
621 
622 #ifdef CONFIG_ARCH_SUPPORTS_PMD_PFNMAP
623 #define pmd_special(pte)	(!!((pmd_val(pte) & PTE_SPECIAL)))
624 static inline pmd_t pmd_mkspecial(pmd_t pmd)
625 {
626 	return set_pmd_bit(pmd, __pgprot(PTE_SPECIAL));
627 }
628 #endif
629 
630 #define __pmd_to_phys(pmd)	__pte_to_phys(pmd_pte(pmd))
631 #define __phys_to_pmd_val(phys)	__phys_to_pte_val(phys)
632 #define pmd_pfn(pmd)		((__pmd_to_phys(pmd) & PMD_MASK) >> PAGE_SHIFT)
633 #define pfn_pmd(pfn,prot)	__pmd(__phys_to_pmd_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
634 
635 #define pud_young(pud)		pte_young(pud_pte(pud))
636 #define pud_mkyoung(pud)	pte_pud(pte_mkyoung(pud_pte(pud)))
637 #define pud_write(pud)		pte_write(pud_pte(pud))
638 
639 static inline pud_t pud_mkhuge(pud_t pud)
640 {
641 	/*
642 	 * It's possible that the pud is present-invalid on entry
643 	 * and in that case it needs to remain present-invalid on
644 	 * exit. So ensure the VALID bit does not get modified.
645 	 */
646 	pudval_t mask = PUD_TYPE_MASK & ~PTE_VALID;
647 	pudval_t val = PUD_TYPE_SECT & ~PTE_VALID;
648 
649 	return __pud((pud_val(pud) & ~mask) | val);
650 }
651 
652 #define __pud_to_phys(pud)	__pte_to_phys(pud_pte(pud))
653 #define __phys_to_pud_val(phys)	__phys_to_pte_val(phys)
654 #define pud_pfn(pud)		((__pud_to_phys(pud) & PUD_MASK) >> PAGE_SHIFT)
655 #define pfn_pud(pfn,prot)	__pud(__phys_to_pud_val((phys_addr_t)(pfn) << PAGE_SHIFT) | pgprot_val(prot))
656 
657 #define pmd_pgprot pmd_pgprot
658 static inline pgprot_t pmd_pgprot(pmd_t pmd)
659 {
660 	unsigned long pfn = pmd_pfn(pmd);
661 
662 	return __pgprot(pmd_val(pfn_pmd(pfn, __pgprot(0))) ^ pmd_val(pmd));
663 }
664 
665 #define pud_pgprot pud_pgprot
666 static inline pgprot_t pud_pgprot(pud_t pud)
667 {
668 	unsigned long pfn = pud_pfn(pud);
669 
670 	return __pgprot(pud_val(pfn_pud(pfn, __pgprot(0))) ^ pud_val(pud));
671 }
672 
673 static inline void __set_ptes_anysz(struct mm_struct *mm, unsigned long addr,
674 				    pte_t *ptep, pte_t pte, unsigned int nr,
675 				    unsigned long pgsize)
676 {
677 	unsigned long stride = pgsize >> PAGE_SHIFT;
678 
679 	switch (pgsize) {
680 	case PAGE_SIZE:
681 		page_table_check_ptes_set(mm, addr, ptep, pte, nr);
682 		break;
683 	case PMD_SIZE:
684 		page_table_check_pmds_set(mm, addr, (pmd_t *)ptep,
685 					  pte_pmd(pte), nr);
686 		break;
687 #ifndef __PAGETABLE_PMD_FOLDED
688 	case PUD_SIZE:
689 		page_table_check_puds_set(mm, addr, (pud_t *)ptep,
690 					  pte_pud(pte), nr);
691 		break;
692 #endif
693 	default:
694 		VM_WARN_ON(1);
695 	}
696 
697 	__sync_cache_and_tags(pte, nr * stride);
698 
699 	for (;;) {
700 		__check_safe_pte_update(mm, ptep, pte);
701 		__set_pte_nosync(ptep, pte);
702 		if (--nr == 0)
703 			break;
704 		ptep++;
705 		pte = pte_advance_pfn(pte, stride);
706 	}
707 
708 	__set_pte_complete(pte);
709 }
710 
711 static inline void __set_ptes(struct mm_struct *mm, unsigned long addr,
712 			      pte_t *ptep, pte_t pte, unsigned int nr)
713 {
714 	__set_ptes_anysz(mm, addr, ptep, pte, nr, PAGE_SIZE);
715 }
716 
717 static inline void __set_pmds(struct mm_struct *mm, unsigned long addr,
718 			      pmd_t *pmdp, pmd_t pmd, unsigned int nr)
719 {
720 	__set_ptes_anysz(mm, addr, (pte_t *)pmdp, pmd_pte(pmd), nr, PMD_SIZE);
721 }
722 #define set_pmd_at(mm, addr, pmdp, pmd) __set_pmds(mm, addr, pmdp, pmd, 1)
723 
724 static inline void __set_puds(struct mm_struct *mm, unsigned long addr,
725 			      pud_t *pudp, pud_t pud, unsigned int nr)
726 {
727 	__set_ptes_anysz(mm, addr, (pte_t *)pudp, pud_pte(pud), nr, PUD_SIZE);
728 }
729 #define set_pud_at(mm, addr, pudp, pud) __set_puds(mm, addr, pudp, pud, 1)
730 
731 #define __p4d_to_phys(p4d)	__pte_to_phys(p4d_pte(p4d))
732 #define __phys_to_p4d_val(phys)	__phys_to_pte_val(phys)
733 
734 #define __pgd_to_phys(pgd)	__pte_to_phys(pgd_pte(pgd))
735 #define __phys_to_pgd_val(phys)	__phys_to_pte_val(phys)
736 
737 #define __pgprot_modify(prot,mask,bits) \
738 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
739 
740 #define pgprot_nx(prot) \
741 	__pgprot_modify(prot, PTE_MAYBE_GP, PTE_PXN)
742 
743 #define pgprot_decrypted(prot) \
744 	__pgprot_modify(prot, PROT_NS_SHARED, PROT_NS_SHARED)
745 #define pgprot_encrypted(prot) \
746 	__pgprot_modify(prot, PROT_NS_SHARED, 0)
747 
748 /*
749  * Mark the prot value as uncacheable and unbufferable.
750  */
751 #define pgprot_noncached(prot) \
752 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRnE) | PTE_PXN | PTE_UXN)
753 #define pgprot_writecombine(prot) \
754 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
755 #define pgprot_device(prot) \
756 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_DEVICE_nGnRE) | PTE_PXN | PTE_UXN)
757 #define pgprot_tagged(prot) \
758 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, PTE_ATTRINDX(MT_NORMAL_TAGGED))
759 #define pgprot_mhp	pgprot_tagged
760 /*
761  * DMA allocations for non-coherent devices use what the Arm architecture calls
762  * "Normal non-cacheable" memory, which permits speculation, unaligned accesses
763  * and merging of writes.  This is different from "Device-nGnR[nE]" memory which
764  * is intended for MMIO and thus forbids speculation, preserves access size,
765  * requires strict alignment and can also force write responses to come from the
766  * endpoint.
767  */
768 #define pgprot_dmacoherent(prot) \
769 	__pgprot_modify(prot, PTE_ATTRINDX_MASK, \
770 			PTE_ATTRINDX(MT_NORMAL_NC) | PTE_PXN | PTE_UXN)
771 
772 #define __HAVE_PHYS_MEM_ACCESS_PROT
773 struct file;
774 extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
775 				     unsigned long size, pgprot_t vma_prot);
776 
777 #define pmd_none(pmd)		(!pmd_val(pmd))
778 
779 #define pmd_table(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
780 				 PMD_TYPE_TABLE)
781 #define pmd_sect(pmd)		((pmd_val(pmd) & PMD_TYPE_MASK) == \
782 				 PMD_TYPE_SECT)
783 #define pmd_leaf(pmd)		(pmd_present(pmd) && !pmd_table(pmd))
784 #define pmd_bad(pmd)		(!pmd_table(pmd))
785 
786 #define pmd_leaf_size(pmd)	(pmd_cont(pmd) ? CONT_PMD_SIZE : PMD_SIZE)
787 #define pte_leaf_size(pte)	(pte_cont(pte) ? CONT_PTE_SIZE : PAGE_SIZE)
788 
789 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
790 static inline int pmd_trans_huge(pmd_t pmd)
791 {
792 	/*
793 	 * If pmd is present-invalid, pmd_table() won't detect it
794 	 * as a table, so force the valid bit for the comparison.
795 	 */
796 	return pmd_present(pmd) && !pmd_table(__pmd(pmd_val(pmd) | PTE_VALID));
797 }
798 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
799 
800 #if defined(CONFIG_ARM64_64K_PAGES) || CONFIG_PGTABLE_LEVELS < 3
801 static inline bool pud_sect(pud_t pud) { return false; }
802 static inline bool pud_table(pud_t pud) { return true; }
803 #else
804 #define pud_sect(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
805 				 PUD_TYPE_SECT)
806 #define pud_table(pud)		((pud_val(pud) & PUD_TYPE_MASK) == \
807 				 PUD_TYPE_TABLE)
808 #endif
809 
810 extern pgd_t swapper_pg_dir[];
811 extern pgd_t idmap_pg_dir[];
812 extern pgd_t tramp_pg_dir[];
813 extern pgd_t reserved_pg_dir[];
814 
815 extern void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd);
816 
817 static inline bool in_swapper_pgdir(void *addr)
818 {
819 	return ((unsigned long)addr & PAGE_MASK) ==
820 	        ((unsigned long)swapper_pg_dir & PAGE_MASK);
821 }
822 
823 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
824 {
825 #ifdef __PAGETABLE_PMD_FOLDED
826 	if (in_swapper_pgdir(pmdp)) {
827 		set_swapper_pgd((pgd_t *)pmdp, __pgd(pmd_val(pmd)));
828 		return;
829 	}
830 #endif /* __PAGETABLE_PMD_FOLDED */
831 
832 	WRITE_ONCE(*pmdp, pmd);
833 
834 	if (pmd_valid(pmd))
835 		queue_pte_barriers();
836 }
837 
838 static inline void pmd_clear(pmd_t *pmdp)
839 {
840 	set_pmd(pmdp, __pmd(0));
841 }
842 
843 static inline phys_addr_t pmd_page_paddr(pmd_t pmd)
844 {
845 	return __pmd_to_phys(pmd);
846 }
847 
848 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
849 {
850 	return (unsigned long)__va(pmd_page_paddr(pmd));
851 }
852 
853 /* Find an entry in the third-level page table. */
854 #define pte_offset_phys(dir,addr)	(pmd_page_paddr(READ_ONCE(*(dir))) + pte_index(addr) * sizeof(pte_t))
855 
856 #define pte_set_fixmap(addr)		((pte_t *)set_fixmap_offset(FIX_PTE, addr))
857 #define pte_set_fixmap_offset(pmd, addr)	pte_set_fixmap(pte_offset_phys(pmd, addr))
858 #define pte_clear_fixmap()		clear_fixmap(FIX_PTE)
859 
860 #define pmd_page(pmd)			phys_to_page(__pmd_to_phys(pmd))
861 
862 /* use ONLY for statically allocated translation tables */
863 #define pte_offset_kimg(dir,addr)	((pte_t *)__phys_to_kimg(pte_offset_phys((dir), (addr))))
864 
865 #if CONFIG_PGTABLE_LEVELS > 2
866 
867 #define pmd_ERROR(e)	\
868 	pr_err("%s:%d: bad pmd %016llx.\n", __FILE__, __LINE__, pmd_val(e))
869 
870 #define pud_none(pud)		(!pud_val(pud))
871 #define pud_bad(pud)		((pud_val(pud) & PUD_TYPE_MASK) != \
872 				 PUD_TYPE_TABLE)
873 #define pud_present(pud)	pte_present(pud_pte(pud))
874 #ifndef __PAGETABLE_PMD_FOLDED
875 #define pud_leaf(pud)		(pud_present(pud) && !pud_table(pud))
876 #else
877 #define pud_leaf(pud)		false
878 #endif
879 #define pud_valid(pud)		pte_valid(pud_pte(pud))
880 #define pud_user(pud)		pte_user(pud_pte(pud))
881 #define pud_user_exec(pud)	pte_user_exec(pud_pte(pud))
882 
883 static inline bool pgtable_l4_enabled(void);
884 
885 static inline void set_pud(pud_t *pudp, pud_t pud)
886 {
887 	if (!pgtable_l4_enabled() && in_swapper_pgdir(pudp)) {
888 		set_swapper_pgd((pgd_t *)pudp, __pgd(pud_val(pud)));
889 		return;
890 	}
891 
892 	WRITE_ONCE(*pudp, pud);
893 
894 	if (pud_valid(pud))
895 		queue_pte_barriers();
896 }
897 
898 static inline void pud_clear(pud_t *pudp)
899 {
900 	set_pud(pudp, __pud(0));
901 }
902 
903 static inline phys_addr_t pud_page_paddr(pud_t pud)
904 {
905 	return __pud_to_phys(pud);
906 }
907 
908 static inline pmd_t *pud_pgtable(pud_t pud)
909 {
910 	return (pmd_t *)__va(pud_page_paddr(pud));
911 }
912 
913 /* Find an entry in the second-level page table. */
914 #define pmd_offset_phys(dir, addr)	(pud_page_paddr(READ_ONCE(*(dir))) + pmd_index(addr) * sizeof(pmd_t))
915 
916 #define pmd_set_fixmap(addr)		((pmd_t *)set_fixmap_offset(FIX_PMD, addr))
917 #define pmd_set_fixmap_offset(pud, addr)	pmd_set_fixmap(pmd_offset_phys(pud, addr))
918 #define pmd_clear_fixmap()		clear_fixmap(FIX_PMD)
919 
920 #define pud_page(pud)			phys_to_page(__pud_to_phys(pud))
921 
922 /* use ONLY for statically allocated translation tables */
923 #define pmd_offset_kimg(dir,addr)	((pmd_t *)__phys_to_kimg(pmd_offset_phys((dir), (addr))))
924 
925 #else
926 
927 #define pud_valid(pud)		false
928 #define pud_page_paddr(pud)	({ BUILD_BUG(); 0; })
929 #define pud_user_exec(pud)	pud_user(pud) /* Always 0 with folding */
930 
931 /* Match pmd_offset folding in <asm/generic/pgtable-nopmd.h> */
932 #define pmd_set_fixmap(addr)		NULL
933 #define pmd_set_fixmap_offset(pudp, addr)	((pmd_t *)pudp)
934 #define pmd_clear_fixmap()
935 
936 #define pmd_offset_kimg(dir,addr)	((pmd_t *)dir)
937 
938 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
939 
940 #if CONFIG_PGTABLE_LEVELS > 3
941 
942 static __always_inline bool pgtable_l4_enabled(void)
943 {
944 	if (CONFIG_PGTABLE_LEVELS > 4 || !IS_ENABLED(CONFIG_ARM64_LPA2))
945 		return true;
946 	if (!alternative_has_cap_likely(ARM64_ALWAYS_BOOT))
947 		return vabits_actual == VA_BITS;
948 	return alternative_has_cap_unlikely(ARM64_HAS_VA52);
949 }
950 
951 static inline bool mm_pud_folded(const struct mm_struct *mm)
952 {
953 	return !pgtable_l4_enabled();
954 }
955 #define mm_pud_folded  mm_pud_folded
956 
957 #define pud_ERROR(e)	\
958 	pr_err("%s:%d: bad pud %016llx.\n", __FILE__, __LINE__, pud_val(e))
959 
960 #define p4d_none(p4d)		(pgtable_l4_enabled() && !p4d_val(p4d))
961 #define p4d_bad(p4d)		(pgtable_l4_enabled() && \
962 				((p4d_val(p4d) & P4D_TYPE_MASK) != \
963 				 P4D_TYPE_TABLE))
964 #define p4d_present(p4d)	(!p4d_none(p4d))
965 
966 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
967 {
968 	if (in_swapper_pgdir(p4dp)) {
969 		set_swapper_pgd((pgd_t *)p4dp, __pgd(p4d_val(p4d)));
970 		return;
971 	}
972 
973 	WRITE_ONCE(*p4dp, p4d);
974 	queue_pte_barriers();
975 }
976 
977 static inline void p4d_clear(p4d_t *p4dp)
978 {
979 	if (pgtable_l4_enabled())
980 		set_p4d(p4dp, __p4d(0));
981 }
982 
983 static inline phys_addr_t p4d_page_paddr(p4d_t p4d)
984 {
985 	return __p4d_to_phys(p4d);
986 }
987 
988 #define pud_index(addr)		(((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
989 
990 static inline pud_t *p4d_to_folded_pud(p4d_t *p4dp, unsigned long addr)
991 {
992 	/* Ensure that 'p4dp' indexes a page table according to 'addr' */
993 	VM_BUG_ON(((addr >> P4D_SHIFT) ^ ((u64)p4dp >> 3)) % PTRS_PER_P4D);
994 
995 	return (pud_t *)PTR_ALIGN_DOWN(p4dp, PAGE_SIZE) + pud_index(addr);
996 }
997 
998 static inline pud_t *p4d_pgtable(p4d_t p4d)
999 {
1000 	return (pud_t *)__va(p4d_page_paddr(p4d));
1001 }
1002 
1003 static inline phys_addr_t pud_offset_phys(p4d_t *p4dp, unsigned long addr)
1004 {
1005 	BUG_ON(!pgtable_l4_enabled());
1006 
1007 	return p4d_page_paddr(READ_ONCE(*p4dp)) + pud_index(addr) * sizeof(pud_t);
1008 }
1009 
1010 static inline
1011 pud_t *pud_offset_lockless(p4d_t *p4dp, p4d_t p4d, unsigned long addr)
1012 {
1013 	if (!pgtable_l4_enabled())
1014 		return p4d_to_folded_pud(p4dp, addr);
1015 	return (pud_t *)__va(p4d_page_paddr(p4d)) + pud_index(addr);
1016 }
1017 #define pud_offset_lockless pud_offset_lockless
1018 
1019 static inline pud_t *pud_offset(p4d_t *p4dp, unsigned long addr)
1020 {
1021 	return pud_offset_lockless(p4dp, READ_ONCE(*p4dp), addr);
1022 }
1023 #define pud_offset	pud_offset
1024 
1025 static inline pud_t *pud_set_fixmap(unsigned long addr)
1026 {
1027 	if (!pgtable_l4_enabled())
1028 		return NULL;
1029 	return (pud_t *)set_fixmap_offset(FIX_PUD, addr);
1030 }
1031 
1032 static inline pud_t *pud_set_fixmap_offset(p4d_t *p4dp, unsigned long addr)
1033 {
1034 	if (!pgtable_l4_enabled())
1035 		return p4d_to_folded_pud(p4dp, addr);
1036 	return pud_set_fixmap(pud_offset_phys(p4dp, addr));
1037 }
1038 
1039 static inline void pud_clear_fixmap(void)
1040 {
1041 	if (pgtable_l4_enabled())
1042 		clear_fixmap(FIX_PUD);
1043 }
1044 
1045 /* use ONLY for statically allocated translation tables */
1046 static inline pud_t *pud_offset_kimg(p4d_t *p4dp, u64 addr)
1047 {
1048 	if (!pgtable_l4_enabled())
1049 		return p4d_to_folded_pud(p4dp, addr);
1050 	return (pud_t *)__phys_to_kimg(pud_offset_phys(p4dp, addr));
1051 }
1052 
1053 #define p4d_page(p4d)		pfn_to_page(__phys_to_pfn(__p4d_to_phys(p4d)))
1054 
1055 #else
1056 
1057 static inline bool pgtable_l4_enabled(void) { return false; }
1058 
1059 #define p4d_page_paddr(p4d)	({ BUILD_BUG(); 0;})
1060 
1061 /* Match pud_offset folding in <asm/generic/pgtable-nopud.h> */
1062 #define pud_set_fixmap(addr)		NULL
1063 #define pud_set_fixmap_offset(pgdp, addr)	((pud_t *)pgdp)
1064 #define pud_clear_fixmap()
1065 
1066 #define pud_offset_kimg(dir,addr)	((pud_t *)dir)
1067 
1068 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
1069 
1070 #if CONFIG_PGTABLE_LEVELS > 4
1071 
1072 static __always_inline bool pgtable_l5_enabled(void)
1073 {
1074 	if (!alternative_has_cap_likely(ARM64_ALWAYS_BOOT))
1075 		return vabits_actual == VA_BITS;
1076 	return alternative_has_cap_unlikely(ARM64_HAS_VA52);
1077 }
1078 
1079 static inline bool mm_p4d_folded(const struct mm_struct *mm)
1080 {
1081 	return !pgtable_l5_enabled();
1082 }
1083 #define mm_p4d_folded  mm_p4d_folded
1084 
1085 #define p4d_ERROR(e)	\
1086 	pr_err("%s:%d: bad p4d %016llx.\n", __FILE__, __LINE__, p4d_val(e))
1087 
1088 #define pgd_none(pgd)		(pgtable_l5_enabled() && !pgd_val(pgd))
1089 #define pgd_bad(pgd)		(pgtable_l5_enabled() && \
1090 				((pgd_val(pgd) & PGD_TYPE_MASK) != \
1091 				 PGD_TYPE_TABLE))
1092 #define pgd_present(pgd)	(!pgd_none(pgd))
1093 
1094 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
1095 {
1096 	if (in_swapper_pgdir(pgdp)) {
1097 		set_swapper_pgd(pgdp, __pgd(pgd_val(pgd)));
1098 		return;
1099 	}
1100 
1101 	WRITE_ONCE(*pgdp, pgd);
1102 	queue_pte_barriers();
1103 }
1104 
1105 static inline void pgd_clear(pgd_t *pgdp)
1106 {
1107 	if (pgtable_l5_enabled())
1108 		set_pgd(pgdp, __pgd(0));
1109 }
1110 
1111 static inline phys_addr_t pgd_page_paddr(pgd_t pgd)
1112 {
1113 	return __pgd_to_phys(pgd);
1114 }
1115 
1116 #define p4d_index(addr)		(((addr) >> P4D_SHIFT) & (PTRS_PER_P4D - 1))
1117 
1118 static inline p4d_t *pgd_to_folded_p4d(pgd_t *pgdp, unsigned long addr)
1119 {
1120 	/* Ensure that 'pgdp' indexes a page table according to 'addr' */
1121 	VM_BUG_ON(((addr >> PGDIR_SHIFT) ^ ((u64)pgdp >> 3)) % PTRS_PER_PGD);
1122 
1123 	return (p4d_t *)PTR_ALIGN_DOWN(pgdp, PAGE_SIZE) + p4d_index(addr);
1124 }
1125 
1126 static inline phys_addr_t p4d_offset_phys(pgd_t *pgdp, unsigned long addr)
1127 {
1128 	BUG_ON(!pgtable_l5_enabled());
1129 
1130 	return pgd_page_paddr(READ_ONCE(*pgdp)) + p4d_index(addr) * sizeof(p4d_t);
1131 }
1132 
1133 static inline
1134 p4d_t *p4d_offset_lockless(pgd_t *pgdp, pgd_t pgd, unsigned long addr)
1135 {
1136 	if (!pgtable_l5_enabled())
1137 		return pgd_to_folded_p4d(pgdp, addr);
1138 	return (p4d_t *)__va(pgd_page_paddr(pgd)) + p4d_index(addr);
1139 }
1140 #define p4d_offset_lockless p4d_offset_lockless
1141 
1142 static inline p4d_t *p4d_offset(pgd_t *pgdp, unsigned long addr)
1143 {
1144 	return p4d_offset_lockless(pgdp, READ_ONCE(*pgdp), addr);
1145 }
1146 
1147 static inline p4d_t *p4d_set_fixmap(unsigned long addr)
1148 {
1149 	if (!pgtable_l5_enabled())
1150 		return NULL;
1151 	return (p4d_t *)set_fixmap_offset(FIX_P4D, addr);
1152 }
1153 
1154 static inline p4d_t *p4d_set_fixmap_offset(pgd_t *pgdp, unsigned long addr)
1155 {
1156 	if (!pgtable_l5_enabled())
1157 		return pgd_to_folded_p4d(pgdp, addr);
1158 	return p4d_set_fixmap(p4d_offset_phys(pgdp, addr));
1159 }
1160 
1161 static inline void p4d_clear_fixmap(void)
1162 {
1163 	if (pgtable_l5_enabled())
1164 		clear_fixmap(FIX_P4D);
1165 }
1166 
1167 /* use ONLY for statically allocated translation tables */
1168 static inline p4d_t *p4d_offset_kimg(pgd_t *pgdp, u64 addr)
1169 {
1170 	if (!pgtable_l5_enabled())
1171 		return pgd_to_folded_p4d(pgdp, addr);
1172 	return (p4d_t *)__phys_to_kimg(p4d_offset_phys(pgdp, addr));
1173 }
1174 
1175 #define pgd_page(pgd)		pfn_to_page(__phys_to_pfn(__pgd_to_phys(pgd)))
1176 
1177 #else
1178 
1179 static inline bool pgtable_l5_enabled(void) { return false; }
1180 
1181 #define p4d_index(addr)		(((addr) >> P4D_SHIFT) & (PTRS_PER_P4D - 1))
1182 
1183 /* Match p4d_offset folding in <asm/generic/pgtable-nop4d.h> */
1184 #define p4d_set_fixmap(addr)		NULL
1185 #define p4d_set_fixmap_offset(p4dp, addr)	((p4d_t *)p4dp)
1186 #define p4d_clear_fixmap()
1187 
1188 #define p4d_offset_kimg(dir,addr)	((p4d_t *)dir)
1189 
1190 static inline
1191 p4d_t *p4d_offset_lockless_folded(pgd_t *pgdp, pgd_t pgd, unsigned long addr)
1192 {
1193 	/*
1194 	 * With runtime folding of the pud, pud_offset_lockless() passes
1195 	 * the 'pgd_t *' we return here to p4d_to_folded_pud(), which
1196 	 * will offset the pointer assuming that it points into
1197 	 * a page-table page. However, the fast GUP path passes us a
1198 	 * pgd_t allocated on the stack and so we must use the original
1199 	 * pointer in 'pgdp' to construct the p4d pointer instead of
1200 	 * using the generic p4d_offset_lockless() implementation.
1201 	 *
1202 	 * Note: reusing the original pointer means that we may
1203 	 * dereference the same (live) page-table entry multiple times.
1204 	 * This is safe because it is still only loaded once in the
1205 	 * context of each level and the CPU guarantees same-address
1206 	 * read-after-read ordering.
1207 	 */
1208 	return p4d_offset(pgdp, addr);
1209 }
1210 #define p4d_offset_lockless p4d_offset_lockless_folded
1211 
1212 #endif  /* CONFIG_PGTABLE_LEVELS > 4 */
1213 
1214 #define pgd_ERROR(e)	\
1215 	pr_err("%s:%d: bad pgd %016llx.\n", __FILE__, __LINE__, pgd_val(e))
1216 
1217 #define pgd_set_fixmap(addr)	((pgd_t *)set_fixmap_offset(FIX_PGD, addr))
1218 #define pgd_clear_fixmap()	clear_fixmap(FIX_PGD)
1219 
1220 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
1221 {
1222 	/*
1223 	 * Normal and Normal-Tagged are two different memory types and indices
1224 	 * in MAIR_EL1. The mask below has to include PTE_ATTRINDX_MASK.
1225 	 */
1226 	const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY |
1227 			      PTE_PRESENT_INVALID | PTE_VALID | PTE_WRITE |
1228 			      PTE_GP | PTE_ATTRINDX_MASK | PTE_PO_IDX_MASK;
1229 
1230 	/* preserve the hardware dirty information */
1231 	if (pte_hw_dirty(pte))
1232 		pte = set_pte_bit(pte, __pgprot(PTE_DIRTY));
1233 
1234 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
1235 	/*
1236 	 * If we end up clearing hw dirtiness for a sw-dirty PTE, set hardware
1237 	 * dirtiness again.
1238 	 */
1239 	if (pte_sw_dirty(pte))
1240 		pte = pte_mkdirty(pte);
1241 	return pte;
1242 }
1243 
1244 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1245 {
1246 	return pte_pmd(pte_modify(pmd_pte(pmd), newprot));
1247 }
1248 
1249 extern int __ptep_set_access_flags(struct vm_area_struct *vma,
1250 				 unsigned long address, pte_t *ptep,
1251 				 pte_t entry, int dirty);
1252 
1253 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1254 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1255 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
1256 					unsigned long address, pmd_t *pmdp,
1257 					pmd_t entry, int dirty)
1258 {
1259 	return __ptep_set_access_flags(vma, address, (pte_t *)pmdp,
1260 							pmd_pte(entry), dirty);
1261 }
1262 #endif
1263 
1264 #ifdef CONFIG_PAGE_TABLE_CHECK
1265 static inline bool pte_user_accessible_page(pte_t pte, unsigned long addr)
1266 {
1267 	return pte_valid(pte) && (pte_user(pte) || pte_user_exec(pte));
1268 }
1269 
1270 static inline bool pmd_user_accessible_page(pmd_t pmd, unsigned long addr)
1271 {
1272 	return pmd_valid(pmd) && !pmd_table(pmd) && (pmd_user(pmd) || pmd_user_exec(pmd));
1273 }
1274 
1275 static inline bool pud_user_accessible_page(pud_t pud, unsigned long addr)
1276 {
1277 	return pud_valid(pud) && !pud_table(pud) && (pud_user(pud) || pud_user_exec(pud));
1278 }
1279 #endif
1280 
1281 /*
1282  * Atomic pte/pmd modifications.
1283  */
1284 
1285 static inline void __pte_clear(struct mm_struct *mm,
1286 			       unsigned long addr, pte_t *ptep)
1287 {
1288 	__set_pte(ptep, __pte(0));
1289 }
1290 
1291 static inline int __ptep_test_and_clear_young(struct vm_area_struct *vma,
1292 					      unsigned long address,
1293 					      pte_t *ptep)
1294 {
1295 	pte_t old_pte, pte;
1296 
1297 	pte = __ptep_get(ptep);
1298 	do {
1299 		old_pte = pte;
1300 		pte = pte_mkold(pte);
1301 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
1302 					       pte_val(old_pte), pte_val(pte));
1303 	} while (pte_val(pte) != pte_val(old_pte));
1304 
1305 	return pte_young(pte);
1306 }
1307 
1308 static inline int __ptep_clear_flush_young(struct vm_area_struct *vma,
1309 					 unsigned long address, pte_t *ptep)
1310 {
1311 	int young = __ptep_test_and_clear_young(vma, address, ptep);
1312 
1313 	if (young) {
1314 		/*
1315 		 * We can elide the trailing DSB here since the worst that can
1316 		 * happen is that a CPU continues to use the young entry in its
1317 		 * TLB and we mistakenly reclaim the associated page. The
1318 		 * window for such an event is bounded by the next
1319 		 * context-switch, which provides a DSB to complete the TLB
1320 		 * invalidation.
1321 		 */
1322 		flush_tlb_page_nosync(vma, address);
1323 	}
1324 
1325 	return young;
1326 }
1327 
1328 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG)
1329 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1330 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1331 					    unsigned long address,
1332 					    pmd_t *pmdp)
1333 {
1334 	/* Operation applies to PMD table entry only if FEAT_HAFT is enabled */
1335 	VM_WARN_ON(pmd_table(READ_ONCE(*pmdp)) && !system_supports_haft());
1336 	return __ptep_test_and_clear_young(vma, address, (pte_t *)pmdp);
1337 }
1338 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG */
1339 
1340 static inline pte_t __ptep_get_and_clear_anysz(struct mm_struct *mm,
1341 					       unsigned long address,
1342 					       pte_t *ptep,
1343 					       unsigned long pgsize)
1344 {
1345 	pte_t pte = __pte(xchg_relaxed(&pte_val(*ptep), 0));
1346 
1347 	switch (pgsize) {
1348 	case PAGE_SIZE:
1349 		page_table_check_pte_clear(mm, address, pte);
1350 		break;
1351 	case PMD_SIZE:
1352 		page_table_check_pmd_clear(mm, address, pte_pmd(pte));
1353 		break;
1354 #ifndef __PAGETABLE_PMD_FOLDED
1355 	case PUD_SIZE:
1356 		page_table_check_pud_clear(mm, address, pte_pud(pte));
1357 		break;
1358 #endif
1359 	default:
1360 		VM_WARN_ON(1);
1361 	}
1362 
1363 	return pte;
1364 }
1365 
1366 static inline pte_t __ptep_get_and_clear(struct mm_struct *mm,
1367 				       unsigned long address, pte_t *ptep)
1368 {
1369 	return __ptep_get_and_clear_anysz(mm, address, ptep, PAGE_SIZE);
1370 }
1371 
1372 static inline void __clear_full_ptes(struct mm_struct *mm, unsigned long addr,
1373 				pte_t *ptep, unsigned int nr, int full)
1374 {
1375 	for (;;) {
1376 		__ptep_get_and_clear(mm, addr, ptep);
1377 		if (--nr == 0)
1378 			break;
1379 		ptep++;
1380 		addr += PAGE_SIZE;
1381 	}
1382 }
1383 
1384 static inline pte_t __get_and_clear_full_ptes(struct mm_struct *mm,
1385 				unsigned long addr, pte_t *ptep,
1386 				unsigned int nr, int full)
1387 {
1388 	pte_t pte, tmp_pte;
1389 
1390 	pte = __ptep_get_and_clear(mm, addr, ptep);
1391 	while (--nr) {
1392 		ptep++;
1393 		addr += PAGE_SIZE;
1394 		tmp_pte = __ptep_get_and_clear(mm, addr, ptep);
1395 		if (pte_dirty(tmp_pte))
1396 			pte = pte_mkdirty(pte);
1397 		if (pte_young(tmp_pte))
1398 			pte = pte_mkyoung(pte);
1399 	}
1400 	return pte;
1401 }
1402 
1403 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1404 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1405 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
1406 					    unsigned long address, pmd_t *pmdp)
1407 {
1408 	return pte_pmd(__ptep_get_and_clear_anysz(mm, address, (pte_t *)pmdp, PMD_SIZE));
1409 }
1410 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1411 
1412 static inline void ___ptep_set_wrprotect(struct mm_struct *mm,
1413 					unsigned long address, pte_t *ptep,
1414 					pte_t pte)
1415 {
1416 	pte_t old_pte;
1417 
1418 	do {
1419 		old_pte = pte;
1420 		pte = pte_wrprotect(pte);
1421 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
1422 					       pte_val(old_pte), pte_val(pte));
1423 	} while (pte_val(pte) != pte_val(old_pte));
1424 }
1425 
1426 /*
1427  * __ptep_set_wrprotect - mark read-only while transferring potential hardware
1428  * dirty status (PTE_DBM && !PTE_RDONLY) to the software PTE_DIRTY bit.
1429  */
1430 static inline void __ptep_set_wrprotect(struct mm_struct *mm,
1431 					unsigned long address, pte_t *ptep)
1432 {
1433 	___ptep_set_wrprotect(mm, address, ptep, __ptep_get(ptep));
1434 }
1435 
1436 static inline void __wrprotect_ptes(struct mm_struct *mm, unsigned long address,
1437 				pte_t *ptep, unsigned int nr)
1438 {
1439 	unsigned int i;
1440 
1441 	for (i = 0; i < nr; i++, address += PAGE_SIZE, ptep++)
1442 		__ptep_set_wrprotect(mm, address, ptep);
1443 }
1444 
1445 static inline void __clear_young_dirty_pte(struct vm_area_struct *vma,
1446 					   unsigned long addr, pte_t *ptep,
1447 					   pte_t pte, cydp_t flags)
1448 {
1449 	pte_t old_pte;
1450 
1451 	do {
1452 		old_pte = pte;
1453 
1454 		if (flags & CYDP_CLEAR_YOUNG)
1455 			pte = pte_mkold(pte);
1456 		if (flags & CYDP_CLEAR_DIRTY)
1457 			pte = pte_mkclean(pte);
1458 
1459 		pte_val(pte) = cmpxchg_relaxed(&pte_val(*ptep),
1460 					       pte_val(old_pte), pte_val(pte));
1461 	} while (pte_val(pte) != pte_val(old_pte));
1462 }
1463 
1464 static inline void __clear_young_dirty_ptes(struct vm_area_struct *vma,
1465 					    unsigned long addr, pte_t *ptep,
1466 					    unsigned int nr, cydp_t flags)
1467 {
1468 	pte_t pte;
1469 
1470 	for (;;) {
1471 		pte = __ptep_get(ptep);
1472 
1473 		if (flags == (CYDP_CLEAR_YOUNG | CYDP_CLEAR_DIRTY))
1474 			__set_pte(ptep, pte_mkclean(pte_mkold(pte)));
1475 		else
1476 			__clear_young_dirty_pte(vma, addr, ptep, pte, flags);
1477 
1478 		if (--nr == 0)
1479 			break;
1480 		ptep++;
1481 		addr += PAGE_SIZE;
1482 	}
1483 }
1484 
1485 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1486 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1487 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1488 				      unsigned long address, pmd_t *pmdp)
1489 {
1490 	__ptep_set_wrprotect(mm, address, (pte_t *)pmdp);
1491 }
1492 
1493 #define pmdp_establish pmdp_establish
1494 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
1495 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
1496 {
1497 	page_table_check_pmd_set(vma->vm_mm, address, pmdp, pmd);
1498 	return __pmd(xchg_relaxed(&pmd_val(*pmdp), pmd_val(pmd)));
1499 }
1500 #endif
1501 
1502 /*
1503  * Encode and decode a swap entry:
1504  *	bits 0-1:	present (must be zero)
1505  *	bits 2:		remember PG_anon_exclusive
1506  *	bit  3:		remember uffd-wp state
1507  *	bits 6-10:	swap type
1508  *	bit  11:	PTE_PRESENT_INVALID (must be zero)
1509  *	bits 12-61:	swap offset
1510  */
1511 #define __SWP_TYPE_SHIFT	6
1512 #define __SWP_TYPE_BITS		5
1513 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
1514 #define __SWP_OFFSET_SHIFT	12
1515 #define __SWP_OFFSET_BITS	50
1516 #define __SWP_OFFSET_MASK	((1UL << __SWP_OFFSET_BITS) - 1)
1517 
1518 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
1519 #define __swp_offset(x)		(((x).val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK)
1520 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
1521 
1522 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1523 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
1524 
1525 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1526 #define __pmd_to_swp_entry(pmd)		((swp_entry_t) { pmd_val(pmd) })
1527 #define __swp_entry_to_pmd(swp)		__pmd((swp).val)
1528 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
1529 
1530 /*
1531  * Ensure that there are not more swap files than can be encoded in the kernel
1532  * PTEs.
1533  */
1534 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
1535 
1536 #ifdef CONFIG_ARM64_MTE
1537 
1538 #define __HAVE_ARCH_PREPARE_TO_SWAP
1539 extern int arch_prepare_to_swap(struct folio *folio);
1540 
1541 #define __HAVE_ARCH_SWAP_INVALIDATE
1542 static inline void arch_swap_invalidate_page(int type, pgoff_t offset)
1543 {
1544 	if (system_supports_mte())
1545 		mte_invalidate_tags(type, offset);
1546 }
1547 
1548 static inline void arch_swap_invalidate_area(int type)
1549 {
1550 	if (system_supports_mte())
1551 		mte_invalidate_tags_area(type);
1552 }
1553 
1554 #define __HAVE_ARCH_SWAP_RESTORE
1555 extern void arch_swap_restore(swp_entry_t entry, struct folio *folio);
1556 
1557 #endif /* CONFIG_ARM64_MTE */
1558 
1559 /*
1560  * On AArch64, the cache coherency is handled via the __set_ptes() function.
1561  */
1562 static inline void update_mmu_cache_range(struct vm_fault *vmf,
1563 		struct vm_area_struct *vma, unsigned long addr, pte_t *ptep,
1564 		unsigned int nr)
1565 {
1566 	/*
1567 	 * We don't do anything here, so there's a very small chance of
1568 	 * us retaking a user fault which we just fixed up. The alternative
1569 	 * is doing a dsb(ishst), but that penalises the fastpath.
1570 	 */
1571 }
1572 
1573 #define update_mmu_cache(vma, addr, ptep) \
1574 	update_mmu_cache_range(NULL, vma, addr, ptep, 1)
1575 #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0)
1576 
1577 #ifdef CONFIG_ARM64_PA_BITS_52
1578 #define phys_to_ttbr(addr)	(((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52)
1579 #else
1580 #define phys_to_ttbr(addr)	(addr)
1581 #endif
1582 
1583 /*
1584  * On arm64 without hardware Access Flag, copying from user will fail because
1585  * the pte is old and cannot be marked young. So we always end up with zeroed
1586  * page after fork() + CoW for pfn mappings. We don't always have a
1587  * hardware-managed access flag on arm64.
1588  */
1589 #define arch_has_hw_pte_young		cpu_has_hw_af
1590 
1591 #ifdef CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG
1592 #define arch_has_hw_nonleaf_pmd_young	system_supports_haft
1593 #endif
1594 
1595 /*
1596  * Experimentally, it's cheap to set the access flag in hardware and we
1597  * benefit from prefaulting mappings as 'old' to start with.
1598  */
1599 #define arch_wants_old_prefaulted_pte	cpu_has_hw_af
1600 
1601 /*
1602  * Request exec memory is read into pagecache in at least 64K folios. This size
1603  * can be contpte-mapped when 4K base pages are in use (16 pages into 1 iTLB
1604  * entry), and HPA can coalesce it (4 pages into 1 TLB entry) when 16K base
1605  * pages are in use.
1606  */
1607 #define exec_folio_order() ilog2(SZ_64K >> PAGE_SHIFT)
1608 
1609 static inline bool pud_sect_supported(void)
1610 {
1611 	return PAGE_SIZE == SZ_4K;
1612 }
1613 
1614 
1615 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1616 #define ptep_modify_prot_start ptep_modify_prot_start
1617 extern pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
1618 				    unsigned long addr, pte_t *ptep);
1619 
1620 #define ptep_modify_prot_commit ptep_modify_prot_commit
1621 extern void ptep_modify_prot_commit(struct vm_area_struct *vma,
1622 				    unsigned long addr, pte_t *ptep,
1623 				    pte_t old_pte, pte_t new_pte);
1624 
1625 #define modify_prot_start_ptes modify_prot_start_ptes
1626 extern pte_t modify_prot_start_ptes(struct vm_area_struct *vma,
1627 				    unsigned long addr, pte_t *ptep,
1628 				    unsigned int nr);
1629 
1630 #define modify_prot_commit_ptes modify_prot_commit_ptes
1631 extern void modify_prot_commit_ptes(struct vm_area_struct *vma, unsigned long addr,
1632 				    pte_t *ptep, pte_t old_pte, pte_t pte,
1633 				    unsigned int nr);
1634 
1635 #ifdef CONFIG_ARM64_CONTPTE
1636 
1637 /*
1638  * The contpte APIs are used to transparently manage the contiguous bit in ptes
1639  * where it is possible and makes sense to do so. The PTE_CONT bit is considered
1640  * a private implementation detail of the public ptep API (see below).
1641  */
1642 extern void __contpte_try_fold(struct mm_struct *mm, unsigned long addr,
1643 				pte_t *ptep, pte_t pte);
1644 extern void __contpte_try_unfold(struct mm_struct *mm, unsigned long addr,
1645 				pte_t *ptep, pte_t pte);
1646 extern pte_t contpte_ptep_get(pte_t *ptep, pte_t orig_pte);
1647 extern pte_t contpte_ptep_get_lockless(pte_t *orig_ptep);
1648 extern void contpte_set_ptes(struct mm_struct *mm, unsigned long addr,
1649 				pte_t *ptep, pte_t pte, unsigned int nr);
1650 extern void contpte_clear_full_ptes(struct mm_struct *mm, unsigned long addr,
1651 				pte_t *ptep, unsigned int nr, int full);
1652 extern pte_t contpte_get_and_clear_full_ptes(struct mm_struct *mm,
1653 				unsigned long addr, pte_t *ptep,
1654 				unsigned int nr, int full);
1655 int contpte_test_and_clear_young_ptes(struct vm_area_struct *vma,
1656 				unsigned long addr, pte_t *ptep, unsigned int nr);
1657 int contpte_clear_flush_young_ptes(struct vm_area_struct *vma,
1658 				unsigned long addr, pte_t *ptep, unsigned int nr);
1659 extern void contpte_wrprotect_ptes(struct mm_struct *mm, unsigned long addr,
1660 				pte_t *ptep, unsigned int nr);
1661 extern int contpte_ptep_set_access_flags(struct vm_area_struct *vma,
1662 				unsigned long addr, pte_t *ptep,
1663 				pte_t entry, int dirty);
1664 extern void contpte_clear_young_dirty_ptes(struct vm_area_struct *vma,
1665 				unsigned long addr, pte_t *ptep,
1666 				unsigned int nr, cydp_t flags);
1667 
1668 static __always_inline void contpte_try_fold(struct mm_struct *mm,
1669 				unsigned long addr, pte_t *ptep, pte_t pte)
1670 {
1671 	/*
1672 	 * Only bother trying if both the virtual and physical addresses are
1673 	 * aligned and correspond to the last entry in a contig range. The core
1674 	 * code mostly modifies ranges from low to high, so this is the likely
1675 	 * the last modification in the contig range, so a good time to fold.
1676 	 * We can't fold special mappings, because there is no associated folio.
1677 	 */
1678 
1679 	const unsigned long contmask = CONT_PTES - 1;
1680 	bool valign = ((addr >> PAGE_SHIFT) & contmask) == contmask;
1681 
1682 	if (unlikely(valign)) {
1683 		bool palign = (pte_pfn(pte) & contmask) == contmask;
1684 
1685 		if (unlikely(palign &&
1686 		    pte_valid(pte) && !pte_cont(pte) && !pte_special(pte)))
1687 			__contpte_try_fold(mm, addr, ptep, pte);
1688 	}
1689 }
1690 
1691 static __always_inline void contpte_try_unfold(struct mm_struct *mm,
1692 				unsigned long addr, pte_t *ptep, pte_t pte)
1693 {
1694 	if (unlikely(pte_valid_cont(pte)))
1695 		__contpte_try_unfold(mm, addr, ptep, pte);
1696 }
1697 
1698 #define pte_batch_hint pte_batch_hint
1699 static inline unsigned int pte_batch_hint(pte_t *ptep, pte_t pte)
1700 {
1701 	if (!pte_valid_cont(pte))
1702 		return 1;
1703 
1704 	return CONT_PTES - (((unsigned long)ptep >> 3) & (CONT_PTES - 1));
1705 }
1706 
1707 /*
1708  * The below functions constitute the public API that arm64 presents to the
1709  * core-mm to manipulate PTE entries within their page tables (or at least this
1710  * is the subset of the API that arm64 needs to implement). These public
1711  * versions will automatically and transparently apply the contiguous bit where
1712  * it makes sense to do so. Therefore any users that are contig-aware (e.g.
1713  * hugetlb, kernel mapper) should NOT use these APIs, but instead use the
1714  * private versions, which are prefixed with double underscore. All of these
1715  * APIs except for ptep_get_lockless() are expected to be called with the PTL
1716  * held. Although the contiguous bit is considered private to the
1717  * implementation, it is deliberately allowed to leak through the getters (e.g.
1718  * ptep_get()), back to core code. This is required so that pte_leaf_size() can
1719  * provide an accurate size for perf_get_pgtable_size(). But this leakage means
1720  * its possible a pte will be passed to a setter with the contiguous bit set, so
1721  * we explicitly clear the contiguous bit in those cases to prevent accidentally
1722  * setting it in the pgtable.
1723  */
1724 
1725 #define ptep_get ptep_get
1726 static inline pte_t ptep_get(pte_t *ptep)
1727 {
1728 	pte_t pte = __ptep_get(ptep);
1729 
1730 	if (likely(!pte_valid_cont(pte)))
1731 		return pte;
1732 
1733 	return contpte_ptep_get(ptep, pte);
1734 }
1735 
1736 #define ptep_get_lockless ptep_get_lockless
1737 static inline pte_t ptep_get_lockless(pte_t *ptep)
1738 {
1739 	pte_t pte = __ptep_get(ptep);
1740 
1741 	if (likely(!pte_valid_cont(pte)))
1742 		return pte;
1743 
1744 	return contpte_ptep_get_lockless(ptep);
1745 }
1746 
1747 static inline void set_pte(pte_t *ptep, pte_t pte)
1748 {
1749 	/*
1750 	 * We don't have the mm or vaddr so cannot unfold contig entries (since
1751 	 * it requires tlb maintenance). set_pte() is not used in core code, so
1752 	 * this should never even be called. Regardless do our best to service
1753 	 * any call and emit a warning if there is any attempt to set a pte on
1754 	 * top of an existing contig range.
1755 	 */
1756 	pte_t orig_pte = __ptep_get(ptep);
1757 
1758 	WARN_ON_ONCE(pte_valid_cont(orig_pte));
1759 	__set_pte(ptep, pte_mknoncont(pte));
1760 }
1761 
1762 #define set_ptes set_ptes
1763 static __always_inline void set_ptes(struct mm_struct *mm, unsigned long addr,
1764 				pte_t *ptep, pte_t pte, unsigned int nr)
1765 {
1766 	pte = pte_mknoncont(pte);
1767 
1768 	if (likely(nr == 1)) {
1769 		contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1770 		__set_ptes(mm, addr, ptep, pte, 1);
1771 		contpte_try_fold(mm, addr, ptep, pte);
1772 	} else {
1773 		contpte_set_ptes(mm, addr, ptep, pte, nr);
1774 	}
1775 }
1776 
1777 static inline void pte_clear(struct mm_struct *mm,
1778 				unsigned long addr, pte_t *ptep)
1779 {
1780 	contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1781 	__pte_clear(mm, addr, ptep);
1782 }
1783 
1784 #define clear_full_ptes clear_full_ptes
1785 static inline void clear_full_ptes(struct mm_struct *mm, unsigned long addr,
1786 				pte_t *ptep, unsigned int nr, int full)
1787 {
1788 	if (likely(nr == 1)) {
1789 		contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1790 		__clear_full_ptes(mm, addr, ptep, nr, full);
1791 	} else {
1792 		contpte_clear_full_ptes(mm, addr, ptep, nr, full);
1793 	}
1794 }
1795 
1796 #define get_and_clear_full_ptes get_and_clear_full_ptes
1797 static inline pte_t get_and_clear_full_ptes(struct mm_struct *mm,
1798 				unsigned long addr, pte_t *ptep,
1799 				unsigned int nr, int full)
1800 {
1801 	pte_t pte;
1802 
1803 	if (likely(nr == 1)) {
1804 		contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1805 		pte = __get_and_clear_full_ptes(mm, addr, ptep, nr, full);
1806 	} else {
1807 		pte = contpte_get_and_clear_full_ptes(mm, addr, ptep, nr, full);
1808 	}
1809 
1810 	return pte;
1811 }
1812 
1813 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1814 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1815 				unsigned long addr, pte_t *ptep)
1816 {
1817 	contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
1818 	return __ptep_get_and_clear(mm, addr, ptep);
1819 }
1820 
1821 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1822 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
1823 				unsigned long addr, pte_t *ptep)
1824 {
1825 	pte_t orig_pte = __ptep_get(ptep);
1826 
1827 	if (likely(!pte_valid_cont(orig_pte)))
1828 		return __ptep_test_and_clear_young(vma, addr, ptep);
1829 
1830 	return contpte_test_and_clear_young_ptes(vma, addr, ptep, 1);
1831 }
1832 
1833 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1834 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
1835 				unsigned long addr, pte_t *ptep)
1836 {
1837 	pte_t orig_pte = __ptep_get(ptep);
1838 
1839 	if (likely(!pte_valid_cont(orig_pte)))
1840 		return __ptep_clear_flush_young(vma, addr, ptep);
1841 
1842 	return contpte_clear_flush_young_ptes(vma, addr, ptep, 1);
1843 }
1844 
1845 #define clear_flush_young_ptes clear_flush_young_ptes
1846 static inline int clear_flush_young_ptes(struct vm_area_struct *vma,
1847 					 unsigned long addr, pte_t *ptep,
1848 					 unsigned int nr)
1849 {
1850 	if (likely(nr == 1 && !pte_cont(__ptep_get(ptep))))
1851 		return __ptep_clear_flush_young(vma, addr, ptep);
1852 
1853 	return contpte_clear_flush_young_ptes(vma, addr, ptep, nr);
1854 }
1855 
1856 #define wrprotect_ptes wrprotect_ptes
1857 static __always_inline void wrprotect_ptes(struct mm_struct *mm,
1858 				unsigned long addr, pte_t *ptep, unsigned int nr)
1859 {
1860 	if (likely(nr == 1)) {
1861 		/*
1862 		 * Optimization: wrprotect_ptes() can only be called for present
1863 		 * ptes so we only need to check contig bit as condition for
1864 		 * unfold, and we can remove the contig bit from the pte we read
1865 		 * to avoid re-reading. This speeds up fork() which is sensitive
1866 		 * for order-0 folios. Equivalent to contpte_try_unfold().
1867 		 */
1868 		pte_t orig_pte = __ptep_get(ptep);
1869 
1870 		if (unlikely(pte_cont(orig_pte))) {
1871 			__contpte_try_unfold(mm, addr, ptep, orig_pte);
1872 			orig_pte = pte_mknoncont(orig_pte);
1873 		}
1874 		___ptep_set_wrprotect(mm, addr, ptep, orig_pte);
1875 	} else {
1876 		contpte_wrprotect_ptes(mm, addr, ptep, nr);
1877 	}
1878 }
1879 
1880 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1881 static inline void ptep_set_wrprotect(struct mm_struct *mm,
1882 				unsigned long addr, pte_t *ptep)
1883 {
1884 	wrprotect_ptes(mm, addr, ptep, 1);
1885 }
1886 
1887 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1888 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1889 				unsigned long addr, pte_t *ptep,
1890 				pte_t entry, int dirty)
1891 {
1892 	pte_t orig_pte = __ptep_get(ptep);
1893 
1894 	entry = pte_mknoncont(entry);
1895 
1896 	if (likely(!pte_valid_cont(orig_pte)))
1897 		return __ptep_set_access_flags(vma, addr, ptep, entry, dirty);
1898 
1899 	return contpte_ptep_set_access_flags(vma, addr, ptep, entry, dirty);
1900 }
1901 
1902 #define clear_young_dirty_ptes clear_young_dirty_ptes
1903 static inline void clear_young_dirty_ptes(struct vm_area_struct *vma,
1904 					  unsigned long addr, pte_t *ptep,
1905 					  unsigned int nr, cydp_t flags)
1906 {
1907 	if (likely(nr == 1 && !pte_cont(__ptep_get(ptep))))
1908 		__clear_young_dirty_ptes(vma, addr, ptep, nr, flags);
1909 	else
1910 		contpte_clear_young_dirty_ptes(vma, addr, ptep, nr, flags);
1911 }
1912 
1913 #else /* CONFIG_ARM64_CONTPTE */
1914 
1915 #define ptep_get				__ptep_get
1916 #define set_pte					__set_pte
1917 #define set_ptes				__set_ptes
1918 #define pte_clear				__pte_clear
1919 #define clear_full_ptes				__clear_full_ptes
1920 #define get_and_clear_full_ptes			__get_and_clear_full_ptes
1921 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1922 #define ptep_get_and_clear			__ptep_get_and_clear
1923 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1924 #define ptep_test_and_clear_young		__ptep_test_and_clear_young
1925 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1926 #define ptep_clear_flush_young			__ptep_clear_flush_young
1927 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1928 #define ptep_set_wrprotect			__ptep_set_wrprotect
1929 #define wrprotect_ptes				__wrprotect_ptes
1930 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1931 #define ptep_set_access_flags			__ptep_set_access_flags
1932 #define clear_young_dirty_ptes			__clear_young_dirty_ptes
1933 
1934 #endif /* CONFIG_ARM64_CONTPTE */
1935 
1936 #endif /* !__ASSEMBLER__ */
1937 
1938 #endif /* __ASM_PGTABLE_H */
1939