xref: /linux/arch/x86/include/asm/pgtable.h (revision c288ea679840de4dee2ce6da5d0f139e3774ad86)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_PGTABLE_H
3 #define _ASM_X86_PGTABLE_H
4 
5 #include <linux/mem_encrypt.h>
6 #include <asm/page.h>
7 #include <asm/pgtable_types.h>
8 
9 /*
10  * Macro to mark a page protection value as UC-
11  */
12 #define pgprot_noncached(prot)						\
13 	((boot_cpu_data.x86 > 3)					\
14 	 ? (__pgprot(pgprot_val(prot) |					\
15 		     cachemode2protval(_PAGE_CACHE_MODE_UC_MINUS)))	\
16 	 : (prot))
17 
18 /*
19  * Macros to add or remove encryption attribute
20  */
21 #define pgprot_encrypted(prot)	__pgprot(__sme_set(pgprot_val(prot)))
22 #define pgprot_decrypted(prot)	__pgprot(__sme_clr(pgprot_val(prot)))
23 
24 #ifndef __ASSEMBLY__
25 #include <linux/spinlock.h>
26 #include <asm/x86_init.h>
27 #include <asm/pkru.h>
28 #include <asm/fpu/api.h>
29 #include <asm-generic/pgtable_uffd.h>
30 
31 extern pgd_t early_top_pgt[PTRS_PER_PGD];
32 bool __init __early_make_pgtable(unsigned long address, pmdval_t pmd);
33 
34 void ptdump_walk_pgd_level(struct seq_file *m, struct mm_struct *mm);
35 void ptdump_walk_pgd_level_debugfs(struct seq_file *m, struct mm_struct *mm,
36 				   bool user);
37 void ptdump_walk_pgd_level_checkwx(void);
38 void ptdump_walk_user_pgd_level_checkwx(void);
39 
40 #ifdef CONFIG_DEBUG_WX
41 #define debug_checkwx()		ptdump_walk_pgd_level_checkwx()
42 #define debug_checkwx_user()	ptdump_walk_user_pgd_level_checkwx()
43 #else
44 #define debug_checkwx()		do { } while (0)
45 #define debug_checkwx_user()	do { } while (0)
46 #endif
47 
48 /*
49  * ZERO_PAGE is a global shared page that is always zero: used
50  * for zero-mapped memory areas etc..
51  */
52 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
53 	__visible;
54 #define ZERO_PAGE(vaddr) ((void)(vaddr),virt_to_page(empty_zero_page))
55 
56 extern spinlock_t pgd_lock;
57 extern struct list_head pgd_list;
58 
59 extern struct mm_struct *pgd_page_get_mm(struct page *page);
60 
61 extern pmdval_t early_pmd_flags;
62 
63 #ifdef CONFIG_PARAVIRT_XXL
64 #include <asm/paravirt.h>
65 #else  /* !CONFIG_PARAVIRT_XXL */
66 #define set_pte(ptep, pte)		native_set_pte(ptep, pte)
67 
68 #define set_pte_atomic(ptep, pte)					\
69 	native_set_pte_atomic(ptep, pte)
70 
71 #define set_pmd(pmdp, pmd)		native_set_pmd(pmdp, pmd)
72 
73 #ifndef __PAGETABLE_P4D_FOLDED
74 #define set_pgd(pgdp, pgd)		native_set_pgd(pgdp, pgd)
75 #define pgd_clear(pgd)			(pgtable_l5_enabled() ? native_pgd_clear(pgd) : 0)
76 #endif
77 
78 #ifndef set_p4d
79 # define set_p4d(p4dp, p4d)		native_set_p4d(p4dp, p4d)
80 #endif
81 
82 #ifndef __PAGETABLE_PUD_FOLDED
83 #define p4d_clear(p4d)			native_p4d_clear(p4d)
84 #endif
85 
86 #ifndef set_pud
87 # define set_pud(pudp, pud)		native_set_pud(pudp, pud)
88 #endif
89 
90 #ifndef __PAGETABLE_PUD_FOLDED
91 #define pud_clear(pud)			native_pud_clear(pud)
92 #endif
93 
94 #define pte_clear(mm, addr, ptep)	native_pte_clear(mm, addr, ptep)
95 #define pmd_clear(pmd)			native_pmd_clear(pmd)
96 
97 #define pgd_val(x)	native_pgd_val(x)
98 #define __pgd(x)	native_make_pgd(x)
99 
100 #ifndef __PAGETABLE_P4D_FOLDED
101 #define p4d_val(x)	native_p4d_val(x)
102 #define __p4d(x)	native_make_p4d(x)
103 #endif
104 
105 #ifndef __PAGETABLE_PUD_FOLDED
106 #define pud_val(x)	native_pud_val(x)
107 #define __pud(x)	native_make_pud(x)
108 #endif
109 
110 #ifndef __PAGETABLE_PMD_FOLDED
111 #define pmd_val(x)	native_pmd_val(x)
112 #define __pmd(x)	native_make_pmd(x)
113 #endif
114 
115 #define pte_val(x)	native_pte_val(x)
116 #define __pte(x)	native_make_pte(x)
117 
118 #define arch_end_context_switch(prev)	do {} while(0)
119 #endif	/* CONFIG_PARAVIRT_XXL */
120 
121 /*
122  * The following only work if pte_present() is true.
123  * Undefined behaviour if not..
124  */
125 static inline int pte_dirty(pte_t pte)
126 {
127 	return pte_flags(pte) & _PAGE_DIRTY;
128 }
129 
130 static inline int pte_young(pte_t pte)
131 {
132 	return pte_flags(pte) & _PAGE_ACCESSED;
133 }
134 
135 static inline int pmd_dirty(pmd_t pmd)
136 {
137 	return pmd_flags(pmd) & _PAGE_DIRTY;
138 }
139 
140 static inline int pmd_young(pmd_t pmd)
141 {
142 	return pmd_flags(pmd) & _PAGE_ACCESSED;
143 }
144 
145 static inline int pud_dirty(pud_t pud)
146 {
147 	return pud_flags(pud) & _PAGE_DIRTY;
148 }
149 
150 static inline int pud_young(pud_t pud)
151 {
152 	return pud_flags(pud) & _PAGE_ACCESSED;
153 }
154 
155 static inline int pte_write(pte_t pte)
156 {
157 	return pte_flags(pte) & _PAGE_RW;
158 }
159 
160 static inline int pte_huge(pte_t pte)
161 {
162 	return pte_flags(pte) & _PAGE_PSE;
163 }
164 
165 static inline int pte_global(pte_t pte)
166 {
167 	return pte_flags(pte) & _PAGE_GLOBAL;
168 }
169 
170 static inline int pte_exec(pte_t pte)
171 {
172 	return !(pte_flags(pte) & _PAGE_NX);
173 }
174 
175 static inline int pte_special(pte_t pte)
176 {
177 	return pte_flags(pte) & _PAGE_SPECIAL;
178 }
179 
180 /* Entries that were set to PROT_NONE are inverted */
181 
182 static inline u64 protnone_mask(u64 val);
183 
184 static inline unsigned long pte_pfn(pte_t pte)
185 {
186 	phys_addr_t pfn = pte_val(pte);
187 	pfn ^= protnone_mask(pfn);
188 	return (pfn & PTE_PFN_MASK) >> PAGE_SHIFT;
189 }
190 
191 static inline unsigned long pmd_pfn(pmd_t pmd)
192 {
193 	phys_addr_t pfn = pmd_val(pmd);
194 	pfn ^= protnone_mask(pfn);
195 	return (pfn & pmd_pfn_mask(pmd)) >> PAGE_SHIFT;
196 }
197 
198 static inline unsigned long pud_pfn(pud_t pud)
199 {
200 	phys_addr_t pfn = pud_val(pud);
201 	pfn ^= protnone_mask(pfn);
202 	return (pfn & pud_pfn_mask(pud)) >> PAGE_SHIFT;
203 }
204 
205 static inline unsigned long p4d_pfn(p4d_t p4d)
206 {
207 	return (p4d_val(p4d) & p4d_pfn_mask(p4d)) >> PAGE_SHIFT;
208 }
209 
210 static inline unsigned long pgd_pfn(pgd_t pgd)
211 {
212 	return (pgd_val(pgd) & PTE_PFN_MASK) >> PAGE_SHIFT;
213 }
214 
215 #define p4d_leaf	p4d_large
216 static inline int p4d_large(p4d_t p4d)
217 {
218 	/* No 512 GiB pages yet */
219 	return 0;
220 }
221 
222 #define pte_page(pte)	pfn_to_page(pte_pfn(pte))
223 
224 #define pmd_leaf	pmd_large
225 static inline int pmd_large(pmd_t pte)
226 {
227 	return pmd_flags(pte) & _PAGE_PSE;
228 }
229 
230 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
231 /* NOTE: when predicate huge page, consider also pmd_devmap, or use pmd_large */
232 static inline int pmd_trans_huge(pmd_t pmd)
233 {
234 	return (pmd_val(pmd) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
235 }
236 
237 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
238 static inline int pud_trans_huge(pud_t pud)
239 {
240 	return (pud_val(pud) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
241 }
242 #endif
243 
244 #define has_transparent_hugepage has_transparent_hugepage
245 static inline int has_transparent_hugepage(void)
246 {
247 	return boot_cpu_has(X86_FEATURE_PSE);
248 }
249 
250 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
251 static inline int pmd_devmap(pmd_t pmd)
252 {
253 	return !!(pmd_val(pmd) & _PAGE_DEVMAP);
254 }
255 
256 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
257 static inline int pud_devmap(pud_t pud)
258 {
259 	return !!(pud_val(pud) & _PAGE_DEVMAP);
260 }
261 #else
262 static inline int pud_devmap(pud_t pud)
263 {
264 	return 0;
265 }
266 #endif
267 
268 static inline int pgd_devmap(pgd_t pgd)
269 {
270 	return 0;
271 }
272 #endif
273 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
274 
275 static inline pte_t pte_set_flags(pte_t pte, pteval_t set)
276 {
277 	pteval_t v = native_pte_val(pte);
278 
279 	return native_make_pte(v | set);
280 }
281 
282 static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear)
283 {
284 	pteval_t v = native_pte_val(pte);
285 
286 	return native_make_pte(v & ~clear);
287 }
288 
289 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
290 static inline int pte_uffd_wp(pte_t pte)
291 {
292 	return pte_flags(pte) & _PAGE_UFFD_WP;
293 }
294 
295 static inline pte_t pte_mkuffd_wp(pte_t pte)
296 {
297 	return pte_set_flags(pte, _PAGE_UFFD_WP);
298 }
299 
300 static inline pte_t pte_clear_uffd_wp(pte_t pte)
301 {
302 	return pte_clear_flags(pte, _PAGE_UFFD_WP);
303 }
304 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
305 
306 static inline pte_t pte_mkclean(pte_t pte)
307 {
308 	return pte_clear_flags(pte, _PAGE_DIRTY);
309 }
310 
311 static inline pte_t pte_mkold(pte_t pte)
312 {
313 	return pte_clear_flags(pte, _PAGE_ACCESSED);
314 }
315 
316 static inline pte_t pte_wrprotect(pte_t pte)
317 {
318 	return pte_clear_flags(pte, _PAGE_RW);
319 }
320 
321 static inline pte_t pte_mkexec(pte_t pte)
322 {
323 	return pte_clear_flags(pte, _PAGE_NX);
324 }
325 
326 static inline pte_t pte_mkdirty(pte_t pte)
327 {
328 	return pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
329 }
330 
331 static inline pte_t pte_mkyoung(pte_t pte)
332 {
333 	return pte_set_flags(pte, _PAGE_ACCESSED);
334 }
335 
336 static inline pte_t pte_mkwrite(pte_t pte)
337 {
338 	return pte_set_flags(pte, _PAGE_RW);
339 }
340 
341 static inline pte_t pte_mkhuge(pte_t pte)
342 {
343 	return pte_set_flags(pte, _PAGE_PSE);
344 }
345 
346 static inline pte_t pte_clrhuge(pte_t pte)
347 {
348 	return pte_clear_flags(pte, _PAGE_PSE);
349 }
350 
351 static inline pte_t pte_mkglobal(pte_t pte)
352 {
353 	return pte_set_flags(pte, _PAGE_GLOBAL);
354 }
355 
356 static inline pte_t pte_clrglobal(pte_t pte)
357 {
358 	return pte_clear_flags(pte, _PAGE_GLOBAL);
359 }
360 
361 static inline pte_t pte_mkspecial(pte_t pte)
362 {
363 	return pte_set_flags(pte, _PAGE_SPECIAL);
364 }
365 
366 static inline pte_t pte_mkdevmap(pte_t pte)
367 {
368 	return pte_set_flags(pte, _PAGE_SPECIAL|_PAGE_DEVMAP);
369 }
370 
371 static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
372 {
373 	pmdval_t v = native_pmd_val(pmd);
374 
375 	return native_make_pmd(v | set);
376 }
377 
378 static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
379 {
380 	pmdval_t v = native_pmd_val(pmd);
381 
382 	return native_make_pmd(v & ~clear);
383 }
384 
385 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
386 static inline int pmd_uffd_wp(pmd_t pmd)
387 {
388 	return pmd_flags(pmd) & _PAGE_UFFD_WP;
389 }
390 
391 static inline pmd_t pmd_mkuffd_wp(pmd_t pmd)
392 {
393 	return pmd_set_flags(pmd, _PAGE_UFFD_WP);
394 }
395 
396 static inline pmd_t pmd_clear_uffd_wp(pmd_t pmd)
397 {
398 	return pmd_clear_flags(pmd, _PAGE_UFFD_WP);
399 }
400 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
401 
402 static inline pmd_t pmd_mkold(pmd_t pmd)
403 {
404 	return pmd_clear_flags(pmd, _PAGE_ACCESSED);
405 }
406 
407 static inline pmd_t pmd_mkclean(pmd_t pmd)
408 {
409 	return pmd_clear_flags(pmd, _PAGE_DIRTY);
410 }
411 
412 static inline pmd_t pmd_wrprotect(pmd_t pmd)
413 {
414 	return pmd_clear_flags(pmd, _PAGE_RW);
415 }
416 
417 static inline pmd_t pmd_mkdirty(pmd_t pmd)
418 {
419 	return pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
420 }
421 
422 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
423 {
424 	return pmd_set_flags(pmd, _PAGE_DEVMAP);
425 }
426 
427 static inline pmd_t pmd_mkhuge(pmd_t pmd)
428 {
429 	return pmd_set_flags(pmd, _PAGE_PSE);
430 }
431 
432 static inline pmd_t pmd_mkyoung(pmd_t pmd)
433 {
434 	return pmd_set_flags(pmd, _PAGE_ACCESSED);
435 }
436 
437 static inline pmd_t pmd_mkwrite(pmd_t pmd)
438 {
439 	return pmd_set_flags(pmd, _PAGE_RW);
440 }
441 
442 static inline pud_t pud_set_flags(pud_t pud, pudval_t set)
443 {
444 	pudval_t v = native_pud_val(pud);
445 
446 	return native_make_pud(v | set);
447 }
448 
449 static inline pud_t pud_clear_flags(pud_t pud, pudval_t clear)
450 {
451 	pudval_t v = native_pud_val(pud);
452 
453 	return native_make_pud(v & ~clear);
454 }
455 
456 static inline pud_t pud_mkold(pud_t pud)
457 {
458 	return pud_clear_flags(pud, _PAGE_ACCESSED);
459 }
460 
461 static inline pud_t pud_mkclean(pud_t pud)
462 {
463 	return pud_clear_flags(pud, _PAGE_DIRTY);
464 }
465 
466 static inline pud_t pud_wrprotect(pud_t pud)
467 {
468 	return pud_clear_flags(pud, _PAGE_RW);
469 }
470 
471 static inline pud_t pud_mkdirty(pud_t pud)
472 {
473 	return pud_set_flags(pud, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
474 }
475 
476 static inline pud_t pud_mkdevmap(pud_t pud)
477 {
478 	return pud_set_flags(pud, _PAGE_DEVMAP);
479 }
480 
481 static inline pud_t pud_mkhuge(pud_t pud)
482 {
483 	return pud_set_flags(pud, _PAGE_PSE);
484 }
485 
486 static inline pud_t pud_mkyoung(pud_t pud)
487 {
488 	return pud_set_flags(pud, _PAGE_ACCESSED);
489 }
490 
491 static inline pud_t pud_mkwrite(pud_t pud)
492 {
493 	return pud_set_flags(pud, _PAGE_RW);
494 }
495 
496 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
497 static inline int pte_soft_dirty(pte_t pte)
498 {
499 	return pte_flags(pte) & _PAGE_SOFT_DIRTY;
500 }
501 
502 static inline int pmd_soft_dirty(pmd_t pmd)
503 {
504 	return pmd_flags(pmd) & _PAGE_SOFT_DIRTY;
505 }
506 
507 static inline int pud_soft_dirty(pud_t pud)
508 {
509 	return pud_flags(pud) & _PAGE_SOFT_DIRTY;
510 }
511 
512 static inline pte_t pte_mksoft_dirty(pte_t pte)
513 {
514 	return pte_set_flags(pte, _PAGE_SOFT_DIRTY);
515 }
516 
517 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
518 {
519 	return pmd_set_flags(pmd, _PAGE_SOFT_DIRTY);
520 }
521 
522 static inline pud_t pud_mksoft_dirty(pud_t pud)
523 {
524 	return pud_set_flags(pud, _PAGE_SOFT_DIRTY);
525 }
526 
527 static inline pte_t pte_clear_soft_dirty(pte_t pte)
528 {
529 	return pte_clear_flags(pte, _PAGE_SOFT_DIRTY);
530 }
531 
532 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
533 {
534 	return pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY);
535 }
536 
537 static inline pud_t pud_clear_soft_dirty(pud_t pud)
538 {
539 	return pud_clear_flags(pud, _PAGE_SOFT_DIRTY);
540 }
541 
542 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
543 
544 /*
545  * Mask out unsupported bits in a present pgprot.  Non-present pgprots
546  * can use those bits for other purposes, so leave them be.
547  */
548 static inline pgprotval_t massage_pgprot(pgprot_t pgprot)
549 {
550 	pgprotval_t protval = pgprot_val(pgprot);
551 
552 	if (protval & _PAGE_PRESENT)
553 		protval &= __supported_pte_mask;
554 
555 	return protval;
556 }
557 
558 static inline pgprotval_t check_pgprot(pgprot_t pgprot)
559 {
560 	pgprotval_t massaged_val = massage_pgprot(pgprot);
561 
562 	/* mmdebug.h can not be included here because of dependencies */
563 #ifdef CONFIG_DEBUG_VM
564 	WARN_ONCE(pgprot_val(pgprot) != massaged_val,
565 		  "attempted to set unsupported pgprot: %016llx "
566 		  "bits: %016llx supported: %016llx\n",
567 		  (u64)pgprot_val(pgprot),
568 		  (u64)pgprot_val(pgprot) ^ massaged_val,
569 		  (u64)__supported_pte_mask);
570 #endif
571 
572 	return massaged_val;
573 }
574 
575 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
576 {
577 	phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
578 	pfn ^= protnone_mask(pgprot_val(pgprot));
579 	pfn &= PTE_PFN_MASK;
580 	return __pte(pfn | check_pgprot(pgprot));
581 }
582 
583 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
584 {
585 	phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
586 	pfn ^= protnone_mask(pgprot_val(pgprot));
587 	pfn &= PHYSICAL_PMD_PAGE_MASK;
588 	return __pmd(pfn | check_pgprot(pgprot));
589 }
590 
591 static inline pud_t pfn_pud(unsigned long page_nr, pgprot_t pgprot)
592 {
593 	phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
594 	pfn ^= protnone_mask(pgprot_val(pgprot));
595 	pfn &= PHYSICAL_PUD_PAGE_MASK;
596 	return __pud(pfn | check_pgprot(pgprot));
597 }
598 
599 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
600 {
601 	return pfn_pmd(pmd_pfn(pmd),
602 		      __pgprot(pmd_flags(pmd) & ~(_PAGE_PRESENT|_PAGE_PROTNONE)));
603 }
604 
605 static inline u64 flip_protnone_guard(u64 oldval, u64 val, u64 mask);
606 
607 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
608 {
609 	pteval_t val = pte_val(pte), oldval = val;
610 
611 	/*
612 	 * Chop off the NX bit (if present), and add the NX portion of
613 	 * the newprot (if present):
614 	 */
615 	val &= _PAGE_CHG_MASK;
616 	val |= check_pgprot(newprot) & ~_PAGE_CHG_MASK;
617 	val = flip_protnone_guard(oldval, val, PTE_PFN_MASK);
618 	return __pte(val);
619 }
620 
621 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
622 {
623 	pmdval_t val = pmd_val(pmd), oldval = val;
624 
625 	val &= _HPAGE_CHG_MASK;
626 	val |= check_pgprot(newprot) & ~_HPAGE_CHG_MASK;
627 	val = flip_protnone_guard(oldval, val, PHYSICAL_PMD_PAGE_MASK);
628 	return __pmd(val);
629 }
630 
631 /*
632  * mprotect needs to preserve PAT and encryption bits when updating
633  * vm_page_prot
634  */
635 #define pgprot_modify pgprot_modify
636 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
637 {
638 	pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK;
639 	pgprotval_t addbits = pgprot_val(newprot) & ~_PAGE_CHG_MASK;
640 	return __pgprot(preservebits | addbits);
641 }
642 
643 #define pte_pgprot(x) __pgprot(pte_flags(x))
644 #define pmd_pgprot(x) __pgprot(pmd_flags(x))
645 #define pud_pgprot(x) __pgprot(pud_flags(x))
646 #define p4d_pgprot(x) __pgprot(p4d_flags(x))
647 
648 #define canon_pgprot(p) __pgprot(massage_pgprot(p))
649 
650 static inline pgprot_t arch_filter_pgprot(pgprot_t prot)
651 {
652 	return canon_pgprot(prot);
653 }
654 
655 static inline int is_new_memtype_allowed(u64 paddr, unsigned long size,
656 					 enum page_cache_mode pcm,
657 					 enum page_cache_mode new_pcm)
658 {
659 	/*
660 	 * PAT type is always WB for untracked ranges, so no need to check.
661 	 */
662 	if (x86_platform.is_untracked_pat_range(paddr, paddr + size))
663 		return 1;
664 
665 	/*
666 	 * Certain new memtypes are not allowed with certain
667 	 * requested memtype:
668 	 * - request is uncached, return cannot be write-back
669 	 * - request is write-combine, return cannot be write-back
670 	 * - request is write-through, return cannot be write-back
671 	 * - request is write-through, return cannot be write-combine
672 	 */
673 	if ((pcm == _PAGE_CACHE_MODE_UC_MINUS &&
674 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
675 	    (pcm == _PAGE_CACHE_MODE_WC &&
676 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
677 	    (pcm == _PAGE_CACHE_MODE_WT &&
678 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
679 	    (pcm == _PAGE_CACHE_MODE_WT &&
680 	     new_pcm == _PAGE_CACHE_MODE_WC)) {
681 		return 0;
682 	}
683 
684 	return 1;
685 }
686 
687 pmd_t *populate_extra_pmd(unsigned long vaddr);
688 pte_t *populate_extra_pte(unsigned long vaddr);
689 
690 #ifdef CONFIG_PAGE_TABLE_ISOLATION
691 pgd_t __pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd);
692 
693 /*
694  * Take a PGD location (pgdp) and a pgd value that needs to be set there.
695  * Populates the user and returns the resulting PGD that must be set in
696  * the kernel copy of the page tables.
697  */
698 static inline pgd_t pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd)
699 {
700 	if (!static_cpu_has(X86_FEATURE_PTI))
701 		return pgd;
702 	return __pti_set_user_pgtbl(pgdp, pgd);
703 }
704 #else   /* CONFIG_PAGE_TABLE_ISOLATION */
705 static inline pgd_t pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd)
706 {
707 	return pgd;
708 }
709 #endif  /* CONFIG_PAGE_TABLE_ISOLATION */
710 
711 #endif	/* __ASSEMBLY__ */
712 
713 
714 #ifdef CONFIG_X86_32
715 # include <asm/pgtable_32.h>
716 #else
717 # include <asm/pgtable_64.h>
718 #endif
719 
720 #ifndef __ASSEMBLY__
721 #include <linux/mm_types.h>
722 #include <linux/mmdebug.h>
723 #include <linux/log2.h>
724 #include <asm/fixmap.h>
725 
726 static inline int pte_none(pte_t pte)
727 {
728 	return !(pte.pte & ~(_PAGE_KNL_ERRATUM_MASK));
729 }
730 
731 #define __HAVE_ARCH_PTE_SAME
732 static inline int pte_same(pte_t a, pte_t b)
733 {
734 	return a.pte == b.pte;
735 }
736 
737 static inline int pte_present(pte_t a)
738 {
739 	return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE);
740 }
741 
742 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
743 static inline int pte_devmap(pte_t a)
744 {
745 	return (pte_flags(a) & _PAGE_DEVMAP) == _PAGE_DEVMAP;
746 }
747 #endif
748 
749 #define pte_accessible pte_accessible
750 static inline bool pte_accessible(struct mm_struct *mm, pte_t a)
751 {
752 	if (pte_flags(a) & _PAGE_PRESENT)
753 		return true;
754 
755 	if ((pte_flags(a) & _PAGE_PROTNONE) &&
756 			mm_tlb_flush_pending(mm))
757 		return true;
758 
759 	return false;
760 }
761 
762 static inline int pmd_present(pmd_t pmd)
763 {
764 	/*
765 	 * Checking for _PAGE_PSE is needed too because
766 	 * split_huge_page will temporarily clear the present bit (but
767 	 * the _PAGE_PSE flag will remain set at all times while the
768 	 * _PAGE_PRESENT bit is clear).
769 	 */
770 	return pmd_flags(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PSE);
771 }
772 
773 #ifdef CONFIG_NUMA_BALANCING
774 /*
775  * These work without NUMA balancing but the kernel does not care. See the
776  * comment in include/linux/pgtable.h
777  */
778 static inline int pte_protnone(pte_t pte)
779 {
780 	return (pte_flags(pte) & (_PAGE_PROTNONE | _PAGE_PRESENT))
781 		== _PAGE_PROTNONE;
782 }
783 
784 static inline int pmd_protnone(pmd_t pmd)
785 {
786 	return (pmd_flags(pmd) & (_PAGE_PROTNONE | _PAGE_PRESENT))
787 		== _PAGE_PROTNONE;
788 }
789 #endif /* CONFIG_NUMA_BALANCING */
790 
791 static inline int pmd_none(pmd_t pmd)
792 {
793 	/* Only check low word on 32-bit platforms, since it might be
794 	   out of sync with upper half. */
795 	unsigned long val = native_pmd_val(pmd);
796 	return (val & ~_PAGE_KNL_ERRATUM_MASK) == 0;
797 }
798 
799 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
800 {
801 	return (unsigned long)__va(pmd_val(pmd) & pmd_pfn_mask(pmd));
802 }
803 
804 /*
805  * Currently stuck as a macro due to indirect forward reference to
806  * linux/mmzone.h's __section_mem_map_addr() definition:
807  */
808 #define pmd_page(pmd)	pfn_to_page(pmd_pfn(pmd))
809 
810 /*
811  * Conversion functions: convert a page and protection to a page entry,
812  * and a page entry and page directory to the page they refer to.
813  *
814  * (Currently stuck as a macro because of indirect forward reference
815  * to linux/mm.h:page_to_nid())
816  */
817 #define mk_pte(page, pgprot)   pfn_pte(page_to_pfn(page), (pgprot))
818 
819 static inline int pmd_bad(pmd_t pmd)
820 {
821 	return (pmd_flags(pmd) & ~_PAGE_USER) != _KERNPG_TABLE;
822 }
823 
824 static inline unsigned long pages_to_mb(unsigned long npg)
825 {
826 	return npg >> (20 - PAGE_SHIFT);
827 }
828 
829 #if CONFIG_PGTABLE_LEVELS > 2
830 static inline int pud_none(pud_t pud)
831 {
832 	return (native_pud_val(pud) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
833 }
834 
835 static inline int pud_present(pud_t pud)
836 {
837 	return pud_flags(pud) & _PAGE_PRESENT;
838 }
839 
840 static inline pmd_t *pud_pgtable(pud_t pud)
841 {
842 	return (pmd_t *)__va(pud_val(pud) & pud_pfn_mask(pud));
843 }
844 
845 /*
846  * Currently stuck as a macro due to indirect forward reference to
847  * linux/mmzone.h's __section_mem_map_addr() definition:
848  */
849 #define pud_page(pud)	pfn_to_page(pud_pfn(pud))
850 
851 #define pud_leaf	pud_large
852 static inline int pud_large(pud_t pud)
853 {
854 	return (pud_val(pud) & (_PAGE_PSE | _PAGE_PRESENT)) ==
855 		(_PAGE_PSE | _PAGE_PRESENT);
856 }
857 
858 static inline int pud_bad(pud_t pud)
859 {
860 	return (pud_flags(pud) & ~(_KERNPG_TABLE | _PAGE_USER)) != 0;
861 }
862 #else
863 #define pud_leaf	pud_large
864 static inline int pud_large(pud_t pud)
865 {
866 	return 0;
867 }
868 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
869 
870 #if CONFIG_PGTABLE_LEVELS > 3
871 static inline int p4d_none(p4d_t p4d)
872 {
873 	return (native_p4d_val(p4d) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
874 }
875 
876 static inline int p4d_present(p4d_t p4d)
877 {
878 	return p4d_flags(p4d) & _PAGE_PRESENT;
879 }
880 
881 static inline pud_t *p4d_pgtable(p4d_t p4d)
882 {
883 	return (pud_t *)__va(p4d_val(p4d) & p4d_pfn_mask(p4d));
884 }
885 
886 /*
887  * Currently stuck as a macro due to indirect forward reference to
888  * linux/mmzone.h's __section_mem_map_addr() definition:
889  */
890 #define p4d_page(p4d)	pfn_to_page(p4d_pfn(p4d))
891 
892 static inline int p4d_bad(p4d_t p4d)
893 {
894 	unsigned long ignore_flags = _KERNPG_TABLE | _PAGE_USER;
895 
896 	if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
897 		ignore_flags |= _PAGE_NX;
898 
899 	return (p4d_flags(p4d) & ~ignore_flags) != 0;
900 }
901 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
902 
903 static inline unsigned long p4d_index(unsigned long address)
904 {
905 	return (address >> P4D_SHIFT) & (PTRS_PER_P4D - 1);
906 }
907 
908 #if CONFIG_PGTABLE_LEVELS > 4
909 static inline int pgd_present(pgd_t pgd)
910 {
911 	if (!pgtable_l5_enabled())
912 		return 1;
913 	return pgd_flags(pgd) & _PAGE_PRESENT;
914 }
915 
916 static inline unsigned long pgd_page_vaddr(pgd_t pgd)
917 {
918 	return (unsigned long)__va((unsigned long)pgd_val(pgd) & PTE_PFN_MASK);
919 }
920 
921 /*
922  * Currently stuck as a macro due to indirect forward reference to
923  * linux/mmzone.h's __section_mem_map_addr() definition:
924  */
925 #define pgd_page(pgd)	pfn_to_page(pgd_pfn(pgd))
926 
927 /* to find an entry in a page-table-directory. */
928 static inline p4d_t *p4d_offset(pgd_t *pgd, unsigned long address)
929 {
930 	if (!pgtable_l5_enabled())
931 		return (p4d_t *)pgd;
932 	return (p4d_t *)pgd_page_vaddr(*pgd) + p4d_index(address);
933 }
934 
935 static inline int pgd_bad(pgd_t pgd)
936 {
937 	unsigned long ignore_flags = _PAGE_USER;
938 
939 	if (!pgtable_l5_enabled())
940 		return 0;
941 
942 	if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
943 		ignore_flags |= _PAGE_NX;
944 
945 	return (pgd_flags(pgd) & ~ignore_flags) != _KERNPG_TABLE;
946 }
947 
948 static inline int pgd_none(pgd_t pgd)
949 {
950 	if (!pgtable_l5_enabled())
951 		return 0;
952 	/*
953 	 * There is no need to do a workaround for the KNL stray
954 	 * A/D bit erratum here.  PGDs only point to page tables
955 	 * except on 32-bit non-PAE which is not supported on
956 	 * KNL.
957 	 */
958 	return !native_pgd_val(pgd);
959 }
960 #endif	/* CONFIG_PGTABLE_LEVELS > 4 */
961 
962 #endif	/* __ASSEMBLY__ */
963 
964 #define KERNEL_PGD_BOUNDARY	pgd_index(PAGE_OFFSET)
965 #define KERNEL_PGD_PTRS		(PTRS_PER_PGD - KERNEL_PGD_BOUNDARY)
966 
967 #ifndef __ASSEMBLY__
968 
969 extern int direct_gbpages;
970 void init_mem_mapping(void);
971 void early_alloc_pgt_buf(void);
972 extern void memblock_find_dma_reserve(void);
973 void __init poking_init(void);
974 unsigned long init_memory_mapping(unsigned long start,
975 				  unsigned long end, pgprot_t prot);
976 
977 #ifdef CONFIG_X86_64
978 extern pgd_t trampoline_pgd_entry;
979 #endif
980 
981 /* local pte updates need not use xchg for locking */
982 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
983 {
984 	pte_t res = *ptep;
985 
986 	/* Pure native function needs no input for mm, addr */
987 	native_pte_clear(NULL, 0, ptep);
988 	return res;
989 }
990 
991 static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp)
992 {
993 	pmd_t res = *pmdp;
994 
995 	native_pmd_clear(pmdp);
996 	return res;
997 }
998 
999 static inline pud_t native_local_pudp_get_and_clear(pud_t *pudp)
1000 {
1001 	pud_t res = *pudp;
1002 
1003 	native_pud_clear(pudp);
1004 	return res;
1005 }
1006 
1007 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
1008 			      pte_t *ptep, pte_t pte)
1009 {
1010 	set_pte(ptep, pte);
1011 }
1012 
1013 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1014 			      pmd_t *pmdp, pmd_t pmd)
1015 {
1016 	set_pmd(pmdp, pmd);
1017 }
1018 
1019 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
1020 			      pud_t *pudp, pud_t pud)
1021 {
1022 	native_set_pud(pudp, pud);
1023 }
1024 
1025 /*
1026  * We only update the dirty/accessed state if we set
1027  * the dirty bit by hand in the kernel, since the hardware
1028  * will do the accessed bit for us, and we don't want to
1029  * race with other CPU's that might be updating the dirty
1030  * bit at the same time.
1031  */
1032 struct vm_area_struct;
1033 
1034 #define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1035 extern int ptep_set_access_flags(struct vm_area_struct *vma,
1036 				 unsigned long address, pte_t *ptep,
1037 				 pte_t entry, int dirty);
1038 
1039 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1040 extern int ptep_test_and_clear_young(struct vm_area_struct *vma,
1041 				     unsigned long addr, pte_t *ptep);
1042 
1043 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1044 extern int ptep_clear_flush_young(struct vm_area_struct *vma,
1045 				  unsigned long address, pte_t *ptep);
1046 
1047 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1048 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
1049 				       pte_t *ptep)
1050 {
1051 	pte_t pte = native_ptep_get_and_clear(ptep);
1052 	return pte;
1053 }
1054 
1055 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1056 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1057 					    unsigned long addr, pte_t *ptep,
1058 					    int full)
1059 {
1060 	pte_t pte;
1061 	if (full) {
1062 		/*
1063 		 * Full address destruction in progress; paravirt does not
1064 		 * care about updates and native needs no locking
1065 		 */
1066 		pte = native_local_ptep_get_and_clear(ptep);
1067 	} else {
1068 		pte = ptep_get_and_clear(mm, addr, ptep);
1069 	}
1070 	return pte;
1071 }
1072 
1073 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1074 static inline void ptep_set_wrprotect(struct mm_struct *mm,
1075 				      unsigned long addr, pte_t *ptep)
1076 {
1077 	clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte);
1078 }
1079 
1080 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
1081 
1082 #define mk_pmd(page, pgprot)   pfn_pmd(page_to_pfn(page), (pgprot))
1083 
1084 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1085 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
1086 				 unsigned long address, pmd_t *pmdp,
1087 				 pmd_t entry, int dirty);
1088 extern int pudp_set_access_flags(struct vm_area_struct *vma,
1089 				 unsigned long address, pud_t *pudp,
1090 				 pud_t entry, int dirty);
1091 
1092 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1093 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1094 				     unsigned long addr, pmd_t *pmdp);
1095 extern int pudp_test_and_clear_young(struct vm_area_struct *vma,
1096 				     unsigned long addr, pud_t *pudp);
1097 
1098 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
1099 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
1100 				  unsigned long address, pmd_t *pmdp);
1101 
1102 
1103 #define pmd_write pmd_write
1104 static inline int pmd_write(pmd_t pmd)
1105 {
1106 	return pmd_flags(pmd) & _PAGE_RW;
1107 }
1108 
1109 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1110 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, unsigned long addr,
1111 				       pmd_t *pmdp)
1112 {
1113 	return native_pmdp_get_and_clear(pmdp);
1114 }
1115 
1116 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
1117 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
1118 					unsigned long addr, pud_t *pudp)
1119 {
1120 	return native_pudp_get_and_clear(pudp);
1121 }
1122 
1123 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1124 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1125 				      unsigned long addr, pmd_t *pmdp)
1126 {
1127 	clear_bit(_PAGE_BIT_RW, (unsigned long *)pmdp);
1128 }
1129 
1130 #define pud_write pud_write
1131 static inline int pud_write(pud_t pud)
1132 {
1133 	return pud_flags(pud) & _PAGE_RW;
1134 }
1135 
1136 #ifndef pmdp_establish
1137 #define pmdp_establish pmdp_establish
1138 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
1139 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
1140 {
1141 	if (IS_ENABLED(CONFIG_SMP)) {
1142 		return xchg(pmdp, pmd);
1143 	} else {
1144 		pmd_t old = *pmdp;
1145 		WRITE_ONCE(*pmdp, pmd);
1146 		return old;
1147 	}
1148 }
1149 #endif
1150 /*
1151  * Page table pages are page-aligned.  The lower half of the top
1152  * level is used for userspace and the top half for the kernel.
1153  *
1154  * Returns true for parts of the PGD that map userspace and
1155  * false for the parts that map the kernel.
1156  */
1157 static inline bool pgdp_maps_userspace(void *__ptr)
1158 {
1159 	unsigned long ptr = (unsigned long)__ptr;
1160 
1161 	return (((ptr & ~PAGE_MASK) / sizeof(pgd_t)) < PGD_KERNEL_START);
1162 }
1163 
1164 #define pgd_leaf	pgd_large
1165 static inline int pgd_large(pgd_t pgd) { return 0; }
1166 
1167 #ifdef CONFIG_PAGE_TABLE_ISOLATION
1168 /*
1169  * All top-level PAGE_TABLE_ISOLATION page tables are order-1 pages
1170  * (8k-aligned and 8k in size).  The kernel one is at the beginning 4k and
1171  * the user one is in the last 4k.  To switch between them, you
1172  * just need to flip the 12th bit in their addresses.
1173  */
1174 #define PTI_PGTABLE_SWITCH_BIT	PAGE_SHIFT
1175 
1176 /*
1177  * This generates better code than the inline assembly in
1178  * __set_bit().
1179  */
1180 static inline void *ptr_set_bit(void *ptr, int bit)
1181 {
1182 	unsigned long __ptr = (unsigned long)ptr;
1183 
1184 	__ptr |= BIT(bit);
1185 	return (void *)__ptr;
1186 }
1187 static inline void *ptr_clear_bit(void *ptr, int bit)
1188 {
1189 	unsigned long __ptr = (unsigned long)ptr;
1190 
1191 	__ptr &= ~BIT(bit);
1192 	return (void *)__ptr;
1193 }
1194 
1195 static inline pgd_t *kernel_to_user_pgdp(pgd_t *pgdp)
1196 {
1197 	return ptr_set_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
1198 }
1199 
1200 static inline pgd_t *user_to_kernel_pgdp(pgd_t *pgdp)
1201 {
1202 	return ptr_clear_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
1203 }
1204 
1205 static inline p4d_t *kernel_to_user_p4dp(p4d_t *p4dp)
1206 {
1207 	return ptr_set_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
1208 }
1209 
1210 static inline p4d_t *user_to_kernel_p4dp(p4d_t *p4dp)
1211 {
1212 	return ptr_clear_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
1213 }
1214 #endif /* CONFIG_PAGE_TABLE_ISOLATION */
1215 
1216 /*
1217  * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
1218  *
1219  *  dst - pointer to pgd range anywhere on a pgd page
1220  *  src - ""
1221  *  count - the number of pgds to copy.
1222  *
1223  * dst and src can be on the same page, but the range must not overlap,
1224  * and must not cross a page boundary.
1225  */
1226 static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
1227 {
1228 	memcpy(dst, src, count * sizeof(pgd_t));
1229 #ifdef CONFIG_PAGE_TABLE_ISOLATION
1230 	if (!static_cpu_has(X86_FEATURE_PTI))
1231 		return;
1232 	/* Clone the user space pgd as well */
1233 	memcpy(kernel_to_user_pgdp(dst), kernel_to_user_pgdp(src),
1234 	       count * sizeof(pgd_t));
1235 #endif
1236 }
1237 
1238 #define PTE_SHIFT ilog2(PTRS_PER_PTE)
1239 static inline int page_level_shift(enum pg_level level)
1240 {
1241 	return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT;
1242 }
1243 static inline unsigned long page_level_size(enum pg_level level)
1244 {
1245 	return 1UL << page_level_shift(level);
1246 }
1247 static inline unsigned long page_level_mask(enum pg_level level)
1248 {
1249 	return ~(page_level_size(level) - 1);
1250 }
1251 
1252 /*
1253  * The x86 doesn't have any external MMU info: the kernel page
1254  * tables contain all the necessary information.
1255  */
1256 static inline void update_mmu_cache(struct vm_area_struct *vma,
1257 		unsigned long addr, pte_t *ptep)
1258 {
1259 }
1260 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1261 		unsigned long addr, pmd_t *pmd)
1262 {
1263 }
1264 static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
1265 		unsigned long addr, pud_t *pud)
1266 {
1267 }
1268 
1269 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1270 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1271 {
1272 	return pte_set_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1273 }
1274 
1275 static inline int pte_swp_soft_dirty(pte_t pte)
1276 {
1277 	return pte_flags(pte) & _PAGE_SWP_SOFT_DIRTY;
1278 }
1279 
1280 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1281 {
1282 	return pte_clear_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1283 }
1284 
1285 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1286 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1287 {
1288 	return pmd_set_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1289 }
1290 
1291 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1292 {
1293 	return pmd_flags(pmd) & _PAGE_SWP_SOFT_DIRTY;
1294 }
1295 
1296 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1297 {
1298 	return pmd_clear_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1299 }
1300 #endif
1301 #endif
1302 
1303 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
1304 static inline pte_t pte_swp_mkuffd_wp(pte_t pte)
1305 {
1306 	return pte_set_flags(pte, _PAGE_SWP_UFFD_WP);
1307 }
1308 
1309 static inline int pte_swp_uffd_wp(pte_t pte)
1310 {
1311 	return pte_flags(pte) & _PAGE_SWP_UFFD_WP;
1312 }
1313 
1314 static inline pte_t pte_swp_clear_uffd_wp(pte_t pte)
1315 {
1316 	return pte_clear_flags(pte, _PAGE_SWP_UFFD_WP);
1317 }
1318 
1319 static inline pmd_t pmd_swp_mkuffd_wp(pmd_t pmd)
1320 {
1321 	return pmd_set_flags(pmd, _PAGE_SWP_UFFD_WP);
1322 }
1323 
1324 static inline int pmd_swp_uffd_wp(pmd_t pmd)
1325 {
1326 	return pmd_flags(pmd) & _PAGE_SWP_UFFD_WP;
1327 }
1328 
1329 static inline pmd_t pmd_swp_clear_uffd_wp(pmd_t pmd)
1330 {
1331 	return pmd_clear_flags(pmd, _PAGE_SWP_UFFD_WP);
1332 }
1333 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
1334 
1335 static inline u16 pte_flags_pkey(unsigned long pte_flags)
1336 {
1337 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1338 	/* ifdef to avoid doing 59-bit shift on 32-bit values */
1339 	return (pte_flags & _PAGE_PKEY_MASK) >> _PAGE_BIT_PKEY_BIT0;
1340 #else
1341 	return 0;
1342 #endif
1343 }
1344 
1345 static inline bool __pkru_allows_pkey(u16 pkey, bool write)
1346 {
1347 	u32 pkru = read_pkru();
1348 
1349 	if (!__pkru_allows_read(pkru, pkey))
1350 		return false;
1351 	if (write && !__pkru_allows_write(pkru, pkey))
1352 		return false;
1353 
1354 	return true;
1355 }
1356 
1357 /*
1358  * 'pteval' can come from a PTE, PMD or PUD.  We only check
1359  * _PAGE_PRESENT, _PAGE_USER, and _PAGE_RW in here which are the
1360  * same value on all 3 types.
1361  */
1362 static inline bool __pte_access_permitted(unsigned long pteval, bool write)
1363 {
1364 	unsigned long need_pte_bits = _PAGE_PRESENT|_PAGE_USER;
1365 
1366 	if (write)
1367 		need_pte_bits |= _PAGE_RW;
1368 
1369 	if ((pteval & need_pte_bits) != need_pte_bits)
1370 		return 0;
1371 
1372 	return __pkru_allows_pkey(pte_flags_pkey(pteval), write);
1373 }
1374 
1375 #define pte_access_permitted pte_access_permitted
1376 static inline bool pte_access_permitted(pte_t pte, bool write)
1377 {
1378 	return __pte_access_permitted(pte_val(pte), write);
1379 }
1380 
1381 #define pmd_access_permitted pmd_access_permitted
1382 static inline bool pmd_access_permitted(pmd_t pmd, bool write)
1383 {
1384 	return __pte_access_permitted(pmd_val(pmd), write);
1385 }
1386 
1387 #define pud_access_permitted pud_access_permitted
1388 static inline bool pud_access_permitted(pud_t pud, bool write)
1389 {
1390 	return __pte_access_permitted(pud_val(pud), write);
1391 }
1392 
1393 #define __HAVE_ARCH_PFN_MODIFY_ALLOWED 1
1394 extern bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot);
1395 
1396 static inline bool arch_has_pfn_modify_check(void)
1397 {
1398 	return boot_cpu_has_bug(X86_BUG_L1TF);
1399 }
1400 
1401 #define arch_faults_on_old_pte arch_faults_on_old_pte
1402 static inline bool arch_faults_on_old_pte(void)
1403 {
1404 	return false;
1405 }
1406 
1407 #endif	/* __ASSEMBLY__ */
1408 
1409 #endif /* _ASM_X86_PGTABLE_H */
1410