xref: /linux/arch/x86/include/asm/pgtable.h (revision 34f7c6e7d4396090692a09789db231e12cb4762b)
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 #ifndef __ASSEMBLY__
19 #include <linux/spinlock.h>
20 #include <asm/x86_init.h>
21 #include <asm/pkru.h>
22 #include <asm/fpu/api.h>
23 #include <asm/coco.h>
24 #include <asm-generic/pgtable_uffd.h>
25 #include <linux/page_table_check.h>
26 
27 extern pgd_t early_top_pgt[PTRS_PER_PGD];
28 bool __init __early_make_pgtable(unsigned long address, pmdval_t pmd);
29 
30 void ptdump_walk_pgd_level(struct seq_file *m, struct mm_struct *mm);
31 void ptdump_walk_pgd_level_debugfs(struct seq_file *m, struct mm_struct *mm,
32 				   bool user);
33 void ptdump_walk_pgd_level_checkwx(void);
34 void ptdump_walk_user_pgd_level_checkwx(void);
35 
36 /*
37  * Macros to add or remove encryption attribute
38  */
39 #define pgprot_encrypted(prot)	__pgprot(cc_mkenc(pgprot_val(prot)))
40 #define pgprot_decrypted(prot)	__pgprot(cc_mkdec(pgprot_val(prot)))
41 
42 #ifdef CONFIG_DEBUG_WX
43 #define debug_checkwx()		ptdump_walk_pgd_level_checkwx()
44 #define debug_checkwx_user()	ptdump_walk_user_pgd_level_checkwx()
45 #else
46 #define debug_checkwx()		do { } while (0)
47 #define debug_checkwx_user()	do { } while (0)
48 #endif
49 
50 /*
51  * ZERO_PAGE is a global shared page that is always zero: used
52  * for zero-mapped memory areas etc..
53  */
54 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
55 	__visible;
56 #define ZERO_PAGE(vaddr) ((void)(vaddr),virt_to_page(empty_zero_page))
57 
58 extern spinlock_t pgd_lock;
59 extern struct list_head pgd_list;
60 
61 extern struct mm_struct *pgd_page_get_mm(struct page *page);
62 
63 extern pmdval_t early_pmd_flags;
64 
65 #ifdef CONFIG_PARAVIRT_XXL
66 #include <asm/paravirt.h>
67 #else  /* !CONFIG_PARAVIRT_XXL */
68 #define set_pte(ptep, pte)		native_set_pte(ptep, pte)
69 
70 #define set_pte_atomic(ptep, pte)					\
71 	native_set_pte_atomic(ptep, pte)
72 
73 #define set_pmd(pmdp, pmd)		native_set_pmd(pmdp, pmd)
74 
75 #ifndef __PAGETABLE_P4D_FOLDED
76 #define set_pgd(pgdp, pgd)		native_set_pgd(pgdp, pgd)
77 #define pgd_clear(pgd)			(pgtable_l5_enabled() ? native_pgd_clear(pgd) : 0)
78 #endif
79 
80 #ifndef set_p4d
81 # define set_p4d(p4dp, p4d)		native_set_p4d(p4dp, p4d)
82 #endif
83 
84 #ifndef __PAGETABLE_PUD_FOLDED
85 #define p4d_clear(p4d)			native_p4d_clear(p4d)
86 #endif
87 
88 #ifndef set_pud
89 # define set_pud(pudp, pud)		native_set_pud(pudp, pud)
90 #endif
91 
92 #ifndef __PAGETABLE_PUD_FOLDED
93 #define pud_clear(pud)			native_pud_clear(pud)
94 #endif
95 
96 #define pte_clear(mm, addr, ptep)	native_pte_clear(mm, addr, ptep)
97 #define pmd_clear(pmd)			native_pmd_clear(pmd)
98 
99 #define pgd_val(x)	native_pgd_val(x)
100 #define __pgd(x)	native_make_pgd(x)
101 
102 #ifndef __PAGETABLE_P4D_FOLDED
103 #define p4d_val(x)	native_p4d_val(x)
104 #define __p4d(x)	native_make_p4d(x)
105 #endif
106 
107 #ifndef __PAGETABLE_PUD_FOLDED
108 #define pud_val(x)	native_pud_val(x)
109 #define __pud(x)	native_make_pud(x)
110 #endif
111 
112 #ifndef __PAGETABLE_PMD_FOLDED
113 #define pmd_val(x)	native_pmd_val(x)
114 #define __pmd(x)	native_make_pmd(x)
115 #endif
116 
117 #define pte_val(x)	native_pte_val(x)
118 #define __pte(x)	native_make_pte(x)
119 
120 #define arch_end_context_switch(prev)	do {} while(0)
121 #endif	/* CONFIG_PARAVIRT_XXL */
122 
123 /*
124  * The following only work if pte_present() is true.
125  * Undefined behaviour if not..
126  */
127 static inline int pte_dirty(pte_t pte)
128 {
129 	return pte_flags(pte) & _PAGE_DIRTY;
130 }
131 
132 static inline int pte_young(pte_t pte)
133 {
134 	return pte_flags(pte) & _PAGE_ACCESSED;
135 }
136 
137 static inline int pmd_dirty(pmd_t pmd)
138 {
139 	return pmd_flags(pmd) & _PAGE_DIRTY;
140 }
141 
142 static inline int pmd_young(pmd_t pmd)
143 {
144 	return pmd_flags(pmd) & _PAGE_ACCESSED;
145 }
146 
147 static inline int pud_dirty(pud_t pud)
148 {
149 	return pud_flags(pud) & _PAGE_DIRTY;
150 }
151 
152 static inline int pud_young(pud_t pud)
153 {
154 	return pud_flags(pud) & _PAGE_ACCESSED;
155 }
156 
157 static inline int pte_write(pte_t pte)
158 {
159 	return pte_flags(pte) & _PAGE_RW;
160 }
161 
162 static inline int pte_huge(pte_t pte)
163 {
164 	return pte_flags(pte) & _PAGE_PSE;
165 }
166 
167 static inline int pte_global(pte_t pte)
168 {
169 	return pte_flags(pte) & _PAGE_GLOBAL;
170 }
171 
172 static inline int pte_exec(pte_t pte)
173 {
174 	return !(pte_flags(pte) & _PAGE_NX);
175 }
176 
177 static inline int pte_special(pte_t pte)
178 {
179 	return pte_flags(pte) & _PAGE_SPECIAL;
180 }
181 
182 /* Entries that were set to PROT_NONE are inverted */
183 
184 static inline u64 protnone_mask(u64 val);
185 
186 static inline unsigned long pte_pfn(pte_t pte)
187 {
188 	phys_addr_t pfn = pte_val(pte);
189 	pfn ^= protnone_mask(pfn);
190 	return (pfn & PTE_PFN_MASK) >> PAGE_SHIFT;
191 }
192 
193 static inline unsigned long pmd_pfn(pmd_t pmd)
194 {
195 	phys_addr_t pfn = pmd_val(pmd);
196 	pfn ^= protnone_mask(pfn);
197 	return (pfn & pmd_pfn_mask(pmd)) >> PAGE_SHIFT;
198 }
199 
200 static inline unsigned long pud_pfn(pud_t pud)
201 {
202 	phys_addr_t pfn = pud_val(pud);
203 	pfn ^= protnone_mask(pfn);
204 	return (pfn & pud_pfn_mask(pud)) >> PAGE_SHIFT;
205 }
206 
207 static inline unsigned long p4d_pfn(p4d_t p4d)
208 {
209 	return (p4d_val(p4d) & p4d_pfn_mask(p4d)) >> PAGE_SHIFT;
210 }
211 
212 static inline unsigned long pgd_pfn(pgd_t pgd)
213 {
214 	return (pgd_val(pgd) & PTE_PFN_MASK) >> PAGE_SHIFT;
215 }
216 
217 #define p4d_leaf	p4d_large
218 static inline int p4d_large(p4d_t p4d)
219 {
220 	/* No 512 GiB pages yet */
221 	return 0;
222 }
223 
224 #define pte_page(pte)	pfn_to_page(pte_pfn(pte))
225 
226 #define pmd_leaf	pmd_large
227 static inline int pmd_large(pmd_t pte)
228 {
229 	return pmd_flags(pte) & _PAGE_PSE;
230 }
231 
232 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
233 /* NOTE: when predicate huge page, consider also pmd_devmap, or use pmd_large */
234 static inline int pmd_trans_huge(pmd_t pmd)
235 {
236 	return (pmd_val(pmd) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
237 }
238 
239 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
240 static inline int pud_trans_huge(pud_t pud)
241 {
242 	return (pud_val(pud) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
243 }
244 #endif
245 
246 #define has_transparent_hugepage has_transparent_hugepage
247 static inline int has_transparent_hugepage(void)
248 {
249 	return boot_cpu_has(X86_FEATURE_PSE);
250 }
251 
252 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
253 static inline int pmd_devmap(pmd_t pmd)
254 {
255 	return !!(pmd_val(pmd) & _PAGE_DEVMAP);
256 }
257 
258 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
259 static inline int pud_devmap(pud_t pud)
260 {
261 	return !!(pud_val(pud) & _PAGE_DEVMAP);
262 }
263 #else
264 static inline int pud_devmap(pud_t pud)
265 {
266 	return 0;
267 }
268 #endif
269 
270 static inline int pgd_devmap(pgd_t pgd)
271 {
272 	return 0;
273 }
274 #endif
275 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
276 
277 static inline pte_t pte_set_flags(pte_t pte, pteval_t set)
278 {
279 	pteval_t v = native_pte_val(pte);
280 
281 	return native_make_pte(v | set);
282 }
283 
284 static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear)
285 {
286 	pteval_t v = native_pte_val(pte);
287 
288 	return native_make_pte(v & ~clear);
289 }
290 
291 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
292 static inline int pte_uffd_wp(pte_t pte)
293 {
294 	return pte_flags(pte) & _PAGE_UFFD_WP;
295 }
296 
297 static inline pte_t pte_mkuffd_wp(pte_t pte)
298 {
299 	return pte_set_flags(pte, _PAGE_UFFD_WP);
300 }
301 
302 static inline pte_t pte_clear_uffd_wp(pte_t pte)
303 {
304 	return pte_clear_flags(pte, _PAGE_UFFD_WP);
305 }
306 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
307 
308 static inline pte_t pte_mkclean(pte_t pte)
309 {
310 	return pte_clear_flags(pte, _PAGE_DIRTY);
311 }
312 
313 static inline pte_t pte_mkold(pte_t pte)
314 {
315 	return pte_clear_flags(pte, _PAGE_ACCESSED);
316 }
317 
318 static inline pte_t pte_wrprotect(pte_t pte)
319 {
320 	return pte_clear_flags(pte, _PAGE_RW);
321 }
322 
323 static inline pte_t pte_mkexec(pte_t pte)
324 {
325 	return pte_clear_flags(pte, _PAGE_NX);
326 }
327 
328 static inline pte_t pte_mkdirty(pte_t pte)
329 {
330 	return pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
331 }
332 
333 static inline pte_t pte_mkyoung(pte_t pte)
334 {
335 	return pte_set_flags(pte, _PAGE_ACCESSED);
336 }
337 
338 static inline pte_t pte_mkwrite(pte_t pte)
339 {
340 	return pte_set_flags(pte, _PAGE_RW);
341 }
342 
343 static inline pte_t pte_mkhuge(pte_t pte)
344 {
345 	return pte_set_flags(pte, _PAGE_PSE);
346 }
347 
348 static inline pte_t pte_clrhuge(pte_t pte)
349 {
350 	return pte_clear_flags(pte, _PAGE_PSE);
351 }
352 
353 static inline pte_t pte_mkglobal(pte_t pte)
354 {
355 	return pte_set_flags(pte, _PAGE_GLOBAL);
356 }
357 
358 static inline pte_t pte_clrglobal(pte_t pte)
359 {
360 	return pte_clear_flags(pte, _PAGE_GLOBAL);
361 }
362 
363 static inline pte_t pte_mkspecial(pte_t pte)
364 {
365 	return pte_set_flags(pte, _PAGE_SPECIAL);
366 }
367 
368 static inline pte_t pte_mkdevmap(pte_t pte)
369 {
370 	return pte_set_flags(pte, _PAGE_SPECIAL|_PAGE_DEVMAP);
371 }
372 
373 static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
374 {
375 	pmdval_t v = native_pmd_val(pmd);
376 
377 	return native_make_pmd(v | set);
378 }
379 
380 static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
381 {
382 	pmdval_t v = native_pmd_val(pmd);
383 
384 	return native_make_pmd(v & ~clear);
385 }
386 
387 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
388 static inline int pmd_uffd_wp(pmd_t pmd)
389 {
390 	return pmd_flags(pmd) & _PAGE_UFFD_WP;
391 }
392 
393 static inline pmd_t pmd_mkuffd_wp(pmd_t pmd)
394 {
395 	return pmd_set_flags(pmd, _PAGE_UFFD_WP);
396 }
397 
398 static inline pmd_t pmd_clear_uffd_wp(pmd_t pmd)
399 {
400 	return pmd_clear_flags(pmd, _PAGE_UFFD_WP);
401 }
402 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
403 
404 static inline pmd_t pmd_mkold(pmd_t pmd)
405 {
406 	return pmd_clear_flags(pmd, _PAGE_ACCESSED);
407 }
408 
409 static inline pmd_t pmd_mkclean(pmd_t pmd)
410 {
411 	return pmd_clear_flags(pmd, _PAGE_DIRTY);
412 }
413 
414 static inline pmd_t pmd_wrprotect(pmd_t pmd)
415 {
416 	return pmd_clear_flags(pmd, _PAGE_RW);
417 }
418 
419 static inline pmd_t pmd_mkdirty(pmd_t pmd)
420 {
421 	return pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
422 }
423 
424 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
425 {
426 	return pmd_set_flags(pmd, _PAGE_DEVMAP);
427 }
428 
429 static inline pmd_t pmd_mkhuge(pmd_t pmd)
430 {
431 	return pmd_set_flags(pmd, _PAGE_PSE);
432 }
433 
434 static inline pmd_t pmd_mkyoung(pmd_t pmd)
435 {
436 	return pmd_set_flags(pmd, _PAGE_ACCESSED);
437 }
438 
439 static inline pmd_t pmd_mkwrite(pmd_t pmd)
440 {
441 	return pmd_set_flags(pmd, _PAGE_RW);
442 }
443 
444 static inline pud_t pud_set_flags(pud_t pud, pudval_t set)
445 {
446 	pudval_t v = native_pud_val(pud);
447 
448 	return native_make_pud(v | set);
449 }
450 
451 static inline pud_t pud_clear_flags(pud_t pud, pudval_t clear)
452 {
453 	pudval_t v = native_pud_val(pud);
454 
455 	return native_make_pud(v & ~clear);
456 }
457 
458 static inline pud_t pud_mkold(pud_t pud)
459 {
460 	return pud_clear_flags(pud, _PAGE_ACCESSED);
461 }
462 
463 static inline pud_t pud_mkclean(pud_t pud)
464 {
465 	return pud_clear_flags(pud, _PAGE_DIRTY);
466 }
467 
468 static inline pud_t pud_wrprotect(pud_t pud)
469 {
470 	return pud_clear_flags(pud, _PAGE_RW);
471 }
472 
473 static inline pud_t pud_mkdirty(pud_t pud)
474 {
475 	return pud_set_flags(pud, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
476 }
477 
478 static inline pud_t pud_mkdevmap(pud_t pud)
479 {
480 	return pud_set_flags(pud, _PAGE_DEVMAP);
481 }
482 
483 static inline pud_t pud_mkhuge(pud_t pud)
484 {
485 	return pud_set_flags(pud, _PAGE_PSE);
486 }
487 
488 static inline pud_t pud_mkyoung(pud_t pud)
489 {
490 	return pud_set_flags(pud, _PAGE_ACCESSED);
491 }
492 
493 static inline pud_t pud_mkwrite(pud_t pud)
494 {
495 	return pud_set_flags(pud, _PAGE_RW);
496 }
497 
498 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
499 static inline int pte_soft_dirty(pte_t pte)
500 {
501 	return pte_flags(pte) & _PAGE_SOFT_DIRTY;
502 }
503 
504 static inline int pmd_soft_dirty(pmd_t pmd)
505 {
506 	return pmd_flags(pmd) & _PAGE_SOFT_DIRTY;
507 }
508 
509 static inline int pud_soft_dirty(pud_t pud)
510 {
511 	return pud_flags(pud) & _PAGE_SOFT_DIRTY;
512 }
513 
514 static inline pte_t pte_mksoft_dirty(pte_t pte)
515 {
516 	return pte_set_flags(pte, _PAGE_SOFT_DIRTY);
517 }
518 
519 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
520 {
521 	return pmd_set_flags(pmd, _PAGE_SOFT_DIRTY);
522 }
523 
524 static inline pud_t pud_mksoft_dirty(pud_t pud)
525 {
526 	return pud_set_flags(pud, _PAGE_SOFT_DIRTY);
527 }
528 
529 static inline pte_t pte_clear_soft_dirty(pte_t pte)
530 {
531 	return pte_clear_flags(pte, _PAGE_SOFT_DIRTY);
532 }
533 
534 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
535 {
536 	return pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY);
537 }
538 
539 static inline pud_t pud_clear_soft_dirty(pud_t pud)
540 {
541 	return pud_clear_flags(pud, _PAGE_SOFT_DIRTY);
542 }
543 
544 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
545 
546 /*
547  * Mask out unsupported bits in a present pgprot.  Non-present pgprots
548  * can use those bits for other purposes, so leave them be.
549  */
550 static inline pgprotval_t massage_pgprot(pgprot_t pgprot)
551 {
552 	pgprotval_t protval = pgprot_val(pgprot);
553 
554 	if (protval & _PAGE_PRESENT)
555 		protval &= __supported_pte_mask;
556 
557 	return protval;
558 }
559 
560 static inline pgprotval_t check_pgprot(pgprot_t pgprot)
561 {
562 	pgprotval_t massaged_val = massage_pgprot(pgprot);
563 
564 	/* mmdebug.h can not be included here because of dependencies */
565 #ifdef CONFIG_DEBUG_VM
566 	WARN_ONCE(pgprot_val(pgprot) != massaged_val,
567 		  "attempted to set unsupported pgprot: %016llx "
568 		  "bits: %016llx supported: %016llx\n",
569 		  (u64)pgprot_val(pgprot),
570 		  (u64)pgprot_val(pgprot) ^ massaged_val,
571 		  (u64)__supported_pte_mask);
572 #endif
573 
574 	return massaged_val;
575 }
576 
577 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
578 {
579 	phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
580 	pfn ^= protnone_mask(pgprot_val(pgprot));
581 	pfn &= PTE_PFN_MASK;
582 	return __pte(pfn | check_pgprot(pgprot));
583 }
584 
585 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
586 {
587 	phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
588 	pfn ^= protnone_mask(pgprot_val(pgprot));
589 	pfn &= PHYSICAL_PMD_PAGE_MASK;
590 	return __pmd(pfn | check_pgprot(pgprot));
591 }
592 
593 static inline pud_t pfn_pud(unsigned long page_nr, pgprot_t pgprot)
594 {
595 	phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
596 	pfn ^= protnone_mask(pgprot_val(pgprot));
597 	pfn &= PHYSICAL_PUD_PAGE_MASK;
598 	return __pud(pfn | check_pgprot(pgprot));
599 }
600 
601 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
602 {
603 	return pfn_pmd(pmd_pfn(pmd),
604 		      __pgprot(pmd_flags(pmd) & ~(_PAGE_PRESENT|_PAGE_PROTNONE)));
605 }
606 
607 static inline u64 flip_protnone_guard(u64 oldval, u64 val, u64 mask);
608 
609 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
610 {
611 	pteval_t val = pte_val(pte), oldval = val;
612 
613 	/*
614 	 * Chop off the NX bit (if present), and add the NX portion of
615 	 * the newprot (if present):
616 	 */
617 	val &= _PAGE_CHG_MASK;
618 	val |= check_pgprot(newprot) & ~_PAGE_CHG_MASK;
619 	val = flip_protnone_guard(oldval, val, PTE_PFN_MASK);
620 	return __pte(val);
621 }
622 
623 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
624 {
625 	pmdval_t val = pmd_val(pmd), oldval = val;
626 
627 	val &= _HPAGE_CHG_MASK;
628 	val |= check_pgprot(newprot) & ~_HPAGE_CHG_MASK;
629 	val = flip_protnone_guard(oldval, val, PHYSICAL_PMD_PAGE_MASK);
630 	return __pmd(val);
631 }
632 
633 /*
634  * mprotect needs to preserve PAT and encryption bits when updating
635  * vm_page_prot
636  */
637 #define pgprot_modify pgprot_modify
638 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
639 {
640 	pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK;
641 	pgprotval_t addbits = pgprot_val(newprot) & ~_PAGE_CHG_MASK;
642 	return __pgprot(preservebits | addbits);
643 }
644 
645 #define pte_pgprot(x) __pgprot(pte_flags(x))
646 #define pmd_pgprot(x) __pgprot(pmd_flags(x))
647 #define pud_pgprot(x) __pgprot(pud_flags(x))
648 #define p4d_pgprot(x) __pgprot(p4d_flags(x))
649 
650 #define canon_pgprot(p) __pgprot(massage_pgprot(p))
651 
652 static inline pgprot_t arch_filter_pgprot(pgprot_t prot)
653 {
654 	return canon_pgprot(prot);
655 }
656 
657 static inline int is_new_memtype_allowed(u64 paddr, unsigned long size,
658 					 enum page_cache_mode pcm,
659 					 enum page_cache_mode new_pcm)
660 {
661 	/*
662 	 * PAT type is always WB for untracked ranges, so no need to check.
663 	 */
664 	if (x86_platform.is_untracked_pat_range(paddr, paddr + size))
665 		return 1;
666 
667 	/*
668 	 * Certain new memtypes are not allowed with certain
669 	 * requested memtype:
670 	 * - request is uncached, return cannot be write-back
671 	 * - request is write-combine, return cannot be write-back
672 	 * - request is write-through, return cannot be write-back
673 	 * - request is write-through, return cannot be write-combine
674 	 */
675 	if ((pcm == _PAGE_CACHE_MODE_UC_MINUS &&
676 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
677 	    (pcm == _PAGE_CACHE_MODE_WC &&
678 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
679 	    (pcm == _PAGE_CACHE_MODE_WT &&
680 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
681 	    (pcm == _PAGE_CACHE_MODE_WT &&
682 	     new_pcm == _PAGE_CACHE_MODE_WC)) {
683 		return 0;
684 	}
685 
686 	return 1;
687 }
688 
689 pmd_t *populate_extra_pmd(unsigned long vaddr);
690 pte_t *populate_extra_pte(unsigned long vaddr);
691 
692 #ifdef CONFIG_PAGE_TABLE_ISOLATION
693 pgd_t __pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd);
694 
695 /*
696  * Take a PGD location (pgdp) and a pgd value that needs to be set there.
697  * Populates the user and returns the resulting PGD that must be set in
698  * the kernel copy of the page tables.
699  */
700 static inline pgd_t pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd)
701 {
702 	if (!static_cpu_has(X86_FEATURE_PTI))
703 		return pgd;
704 	return __pti_set_user_pgtbl(pgdp, pgd);
705 }
706 #else   /* CONFIG_PAGE_TABLE_ISOLATION */
707 static inline pgd_t pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd)
708 {
709 	return pgd;
710 }
711 #endif  /* CONFIG_PAGE_TABLE_ISOLATION */
712 
713 #endif	/* __ASSEMBLY__ */
714 
715 
716 #ifdef CONFIG_X86_32
717 # include <asm/pgtable_32.h>
718 #else
719 # include <asm/pgtable_64.h>
720 #endif
721 
722 #ifndef __ASSEMBLY__
723 #include <linux/mm_types.h>
724 #include <linux/mmdebug.h>
725 #include <linux/log2.h>
726 #include <asm/fixmap.h>
727 
728 static inline int pte_none(pte_t pte)
729 {
730 	return !(pte.pte & ~(_PAGE_KNL_ERRATUM_MASK));
731 }
732 
733 #define __HAVE_ARCH_PTE_SAME
734 static inline int pte_same(pte_t a, pte_t b)
735 {
736 	return a.pte == b.pte;
737 }
738 
739 static inline int pte_present(pte_t a)
740 {
741 	return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE);
742 }
743 
744 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
745 static inline int pte_devmap(pte_t a)
746 {
747 	return (pte_flags(a) & _PAGE_DEVMAP) == _PAGE_DEVMAP;
748 }
749 #endif
750 
751 #define pte_accessible pte_accessible
752 static inline bool pte_accessible(struct mm_struct *mm, pte_t a)
753 {
754 	if (pte_flags(a) & _PAGE_PRESENT)
755 		return true;
756 
757 	if ((pte_flags(a) & _PAGE_PROTNONE) &&
758 			atomic_read(&mm->tlb_flush_pending))
759 		return true;
760 
761 	return false;
762 }
763 
764 static inline int pmd_present(pmd_t pmd)
765 {
766 	/*
767 	 * Checking for _PAGE_PSE is needed too because
768 	 * split_huge_page will temporarily clear the present bit (but
769 	 * the _PAGE_PSE flag will remain set at all times while the
770 	 * _PAGE_PRESENT bit is clear).
771 	 */
772 	return pmd_flags(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PSE);
773 }
774 
775 #ifdef CONFIG_NUMA_BALANCING
776 /*
777  * These work without NUMA balancing but the kernel does not care. See the
778  * comment in include/linux/pgtable.h
779  */
780 static inline int pte_protnone(pte_t pte)
781 {
782 	return (pte_flags(pte) & (_PAGE_PROTNONE | _PAGE_PRESENT))
783 		== _PAGE_PROTNONE;
784 }
785 
786 static inline int pmd_protnone(pmd_t pmd)
787 {
788 	return (pmd_flags(pmd) & (_PAGE_PROTNONE | _PAGE_PRESENT))
789 		== _PAGE_PROTNONE;
790 }
791 #endif /* CONFIG_NUMA_BALANCING */
792 
793 static inline int pmd_none(pmd_t pmd)
794 {
795 	/* Only check low word on 32-bit platforms, since it might be
796 	   out of sync with upper half. */
797 	unsigned long val = native_pmd_val(pmd);
798 	return (val & ~_PAGE_KNL_ERRATUM_MASK) == 0;
799 }
800 
801 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
802 {
803 	return (unsigned long)__va(pmd_val(pmd) & pmd_pfn_mask(pmd));
804 }
805 
806 /*
807  * Currently stuck as a macro due to indirect forward reference to
808  * linux/mmzone.h's __section_mem_map_addr() definition:
809  */
810 #define pmd_page(pmd)	pfn_to_page(pmd_pfn(pmd))
811 
812 /*
813  * Conversion functions: convert a page and protection to a page entry,
814  * and a page entry and page directory to the page they refer to.
815  *
816  * (Currently stuck as a macro because of indirect forward reference
817  * to linux/mm.h:page_to_nid())
818  */
819 #define mk_pte(page, pgprot)   pfn_pte(page_to_pfn(page), (pgprot))
820 
821 static inline int pmd_bad(pmd_t pmd)
822 {
823 	return (pmd_flags(pmd) & ~_PAGE_USER) != _KERNPG_TABLE;
824 }
825 
826 static inline unsigned long pages_to_mb(unsigned long npg)
827 {
828 	return npg >> (20 - PAGE_SHIFT);
829 }
830 
831 #if CONFIG_PGTABLE_LEVELS > 2
832 static inline int pud_none(pud_t pud)
833 {
834 	return (native_pud_val(pud) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
835 }
836 
837 static inline int pud_present(pud_t pud)
838 {
839 	return pud_flags(pud) & _PAGE_PRESENT;
840 }
841 
842 static inline pmd_t *pud_pgtable(pud_t pud)
843 {
844 	return (pmd_t *)__va(pud_val(pud) & pud_pfn_mask(pud));
845 }
846 
847 /*
848  * Currently stuck as a macro due to indirect forward reference to
849  * linux/mmzone.h's __section_mem_map_addr() definition:
850  */
851 #define pud_page(pud)	pfn_to_page(pud_pfn(pud))
852 
853 #define pud_leaf	pud_large
854 static inline int pud_large(pud_t pud)
855 {
856 	return (pud_val(pud) & (_PAGE_PSE | _PAGE_PRESENT)) ==
857 		(_PAGE_PSE | _PAGE_PRESENT);
858 }
859 
860 static inline int pud_bad(pud_t pud)
861 {
862 	return (pud_flags(pud) & ~(_KERNPG_TABLE | _PAGE_USER)) != 0;
863 }
864 #else
865 #define pud_leaf	pud_large
866 static inline int pud_large(pud_t pud)
867 {
868 	return 0;
869 }
870 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
871 
872 #if CONFIG_PGTABLE_LEVELS > 3
873 static inline int p4d_none(p4d_t p4d)
874 {
875 	return (native_p4d_val(p4d) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
876 }
877 
878 static inline int p4d_present(p4d_t p4d)
879 {
880 	return p4d_flags(p4d) & _PAGE_PRESENT;
881 }
882 
883 static inline pud_t *p4d_pgtable(p4d_t p4d)
884 {
885 	return (pud_t *)__va(p4d_val(p4d) & p4d_pfn_mask(p4d));
886 }
887 
888 /*
889  * Currently stuck as a macro due to indirect forward reference to
890  * linux/mmzone.h's __section_mem_map_addr() definition:
891  */
892 #define p4d_page(p4d)	pfn_to_page(p4d_pfn(p4d))
893 
894 static inline int p4d_bad(p4d_t p4d)
895 {
896 	unsigned long ignore_flags = _KERNPG_TABLE | _PAGE_USER;
897 
898 	if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
899 		ignore_flags |= _PAGE_NX;
900 
901 	return (p4d_flags(p4d) & ~ignore_flags) != 0;
902 }
903 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
904 
905 static inline unsigned long p4d_index(unsigned long address)
906 {
907 	return (address >> P4D_SHIFT) & (PTRS_PER_P4D - 1);
908 }
909 
910 #if CONFIG_PGTABLE_LEVELS > 4
911 static inline int pgd_present(pgd_t pgd)
912 {
913 	if (!pgtable_l5_enabled())
914 		return 1;
915 	return pgd_flags(pgd) & _PAGE_PRESENT;
916 }
917 
918 static inline unsigned long pgd_page_vaddr(pgd_t pgd)
919 {
920 	return (unsigned long)__va((unsigned long)pgd_val(pgd) & PTE_PFN_MASK);
921 }
922 
923 /*
924  * Currently stuck as a macro due to indirect forward reference to
925  * linux/mmzone.h's __section_mem_map_addr() definition:
926  */
927 #define pgd_page(pgd)	pfn_to_page(pgd_pfn(pgd))
928 
929 /* to find an entry in a page-table-directory. */
930 static inline p4d_t *p4d_offset(pgd_t *pgd, unsigned long address)
931 {
932 	if (!pgtable_l5_enabled())
933 		return (p4d_t *)pgd;
934 	return (p4d_t *)pgd_page_vaddr(*pgd) + p4d_index(address);
935 }
936 
937 static inline int pgd_bad(pgd_t pgd)
938 {
939 	unsigned long ignore_flags = _PAGE_USER;
940 
941 	if (!pgtable_l5_enabled())
942 		return 0;
943 
944 	if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
945 		ignore_flags |= _PAGE_NX;
946 
947 	return (pgd_flags(pgd) & ~ignore_flags) != _KERNPG_TABLE;
948 }
949 
950 static inline int pgd_none(pgd_t pgd)
951 {
952 	if (!pgtable_l5_enabled())
953 		return 0;
954 	/*
955 	 * There is no need to do a workaround for the KNL stray
956 	 * A/D bit erratum here.  PGDs only point to page tables
957 	 * except on 32-bit non-PAE which is not supported on
958 	 * KNL.
959 	 */
960 	return !native_pgd_val(pgd);
961 }
962 #endif	/* CONFIG_PGTABLE_LEVELS > 4 */
963 
964 #endif	/* __ASSEMBLY__ */
965 
966 #define KERNEL_PGD_BOUNDARY	pgd_index(PAGE_OFFSET)
967 #define KERNEL_PGD_PTRS		(PTRS_PER_PGD - KERNEL_PGD_BOUNDARY)
968 
969 #ifndef __ASSEMBLY__
970 
971 extern int direct_gbpages;
972 void init_mem_mapping(void);
973 void early_alloc_pgt_buf(void);
974 extern void memblock_find_dma_reserve(void);
975 void __init poking_init(void);
976 unsigned long init_memory_mapping(unsigned long start,
977 				  unsigned long end, pgprot_t prot);
978 
979 #ifdef CONFIG_X86_64
980 extern pgd_t trampoline_pgd_entry;
981 #endif
982 
983 /* local pte updates need not use xchg for locking */
984 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
985 {
986 	pte_t res = *ptep;
987 
988 	/* Pure native function needs no input for mm, addr */
989 	native_pte_clear(NULL, 0, ptep);
990 	return res;
991 }
992 
993 static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp)
994 {
995 	pmd_t res = *pmdp;
996 
997 	native_pmd_clear(pmdp);
998 	return res;
999 }
1000 
1001 static inline pud_t native_local_pudp_get_and_clear(pud_t *pudp)
1002 {
1003 	pud_t res = *pudp;
1004 
1005 	native_pud_clear(pudp);
1006 	return res;
1007 }
1008 
1009 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
1010 			      pte_t *ptep, pte_t pte)
1011 {
1012 	page_table_check_pte_set(mm, addr, ptep, pte);
1013 	set_pte(ptep, pte);
1014 }
1015 
1016 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1017 			      pmd_t *pmdp, pmd_t pmd)
1018 {
1019 	page_table_check_pmd_set(mm, addr, pmdp, pmd);
1020 	set_pmd(pmdp, pmd);
1021 }
1022 
1023 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
1024 			      pud_t *pudp, pud_t pud)
1025 {
1026 	page_table_check_pud_set(mm, addr, pudp, pud);
1027 	native_set_pud(pudp, pud);
1028 }
1029 
1030 /*
1031  * We only update the dirty/accessed state if we set
1032  * the dirty bit by hand in the kernel, since the hardware
1033  * will do the accessed bit for us, and we don't want to
1034  * race with other CPU's that might be updating the dirty
1035  * bit at the same time.
1036  */
1037 struct vm_area_struct;
1038 
1039 #define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1040 extern int ptep_set_access_flags(struct vm_area_struct *vma,
1041 				 unsigned long address, pte_t *ptep,
1042 				 pte_t entry, int dirty);
1043 
1044 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1045 extern int ptep_test_and_clear_young(struct vm_area_struct *vma,
1046 				     unsigned long addr, pte_t *ptep);
1047 
1048 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1049 extern int ptep_clear_flush_young(struct vm_area_struct *vma,
1050 				  unsigned long address, pte_t *ptep);
1051 
1052 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1053 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
1054 				       pte_t *ptep)
1055 {
1056 	pte_t pte = native_ptep_get_and_clear(ptep);
1057 	page_table_check_pte_clear(mm, addr, pte);
1058 	return pte;
1059 }
1060 
1061 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1062 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1063 					    unsigned long addr, pte_t *ptep,
1064 					    int full)
1065 {
1066 	pte_t pte;
1067 	if (full) {
1068 		/*
1069 		 * Full address destruction in progress; paravirt does not
1070 		 * care about updates and native needs no locking
1071 		 */
1072 		pte = native_local_ptep_get_and_clear(ptep);
1073 		page_table_check_pte_clear(mm, addr, pte);
1074 	} else {
1075 		pte = ptep_get_and_clear(mm, addr, ptep);
1076 	}
1077 	return pte;
1078 }
1079 
1080 #define __HAVE_ARCH_PTEP_CLEAR
1081 static inline void ptep_clear(struct mm_struct *mm, unsigned long addr,
1082 			      pte_t *ptep)
1083 {
1084 	if (IS_ENABLED(CONFIG_PAGE_TABLE_CHECK))
1085 		ptep_get_and_clear(mm, addr, ptep);
1086 	else
1087 		pte_clear(mm, addr, ptep);
1088 }
1089 
1090 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1091 static inline void ptep_set_wrprotect(struct mm_struct *mm,
1092 				      unsigned long addr, pte_t *ptep)
1093 {
1094 	clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte);
1095 }
1096 
1097 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
1098 
1099 #define mk_pmd(page, pgprot)   pfn_pmd(page_to_pfn(page), (pgprot))
1100 
1101 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1102 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
1103 				 unsigned long address, pmd_t *pmdp,
1104 				 pmd_t entry, int dirty);
1105 extern int pudp_set_access_flags(struct vm_area_struct *vma,
1106 				 unsigned long address, pud_t *pudp,
1107 				 pud_t entry, int dirty);
1108 
1109 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1110 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1111 				     unsigned long addr, pmd_t *pmdp);
1112 extern int pudp_test_and_clear_young(struct vm_area_struct *vma,
1113 				     unsigned long addr, pud_t *pudp);
1114 
1115 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
1116 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
1117 				  unsigned long address, pmd_t *pmdp);
1118 
1119 
1120 #define pmd_write pmd_write
1121 static inline int pmd_write(pmd_t pmd)
1122 {
1123 	return pmd_flags(pmd) & _PAGE_RW;
1124 }
1125 
1126 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1127 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, unsigned long addr,
1128 				       pmd_t *pmdp)
1129 {
1130 	pmd_t pmd = native_pmdp_get_and_clear(pmdp);
1131 
1132 	page_table_check_pmd_clear(mm, addr, pmd);
1133 
1134 	return pmd;
1135 }
1136 
1137 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
1138 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
1139 					unsigned long addr, pud_t *pudp)
1140 {
1141 	pud_t pud = native_pudp_get_and_clear(pudp);
1142 
1143 	page_table_check_pud_clear(mm, addr, pud);
1144 
1145 	return pud;
1146 }
1147 
1148 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1149 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1150 				      unsigned long addr, pmd_t *pmdp)
1151 {
1152 	clear_bit(_PAGE_BIT_RW, (unsigned long *)pmdp);
1153 }
1154 
1155 #define pud_write pud_write
1156 static inline int pud_write(pud_t pud)
1157 {
1158 	return pud_flags(pud) & _PAGE_RW;
1159 }
1160 
1161 #ifndef pmdp_establish
1162 #define pmdp_establish pmdp_establish
1163 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
1164 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
1165 {
1166 	page_table_check_pmd_set(vma->vm_mm, address, pmdp, pmd);
1167 	if (IS_ENABLED(CONFIG_SMP)) {
1168 		return xchg(pmdp, pmd);
1169 	} else {
1170 		pmd_t old = *pmdp;
1171 		WRITE_ONCE(*pmdp, pmd);
1172 		return old;
1173 	}
1174 }
1175 #endif
1176 /*
1177  * Page table pages are page-aligned.  The lower half of the top
1178  * level is used for userspace and the top half for the kernel.
1179  *
1180  * Returns true for parts of the PGD that map userspace and
1181  * false for the parts that map the kernel.
1182  */
1183 static inline bool pgdp_maps_userspace(void *__ptr)
1184 {
1185 	unsigned long ptr = (unsigned long)__ptr;
1186 
1187 	return (((ptr & ~PAGE_MASK) / sizeof(pgd_t)) < PGD_KERNEL_START);
1188 }
1189 
1190 #define pgd_leaf	pgd_large
1191 static inline int pgd_large(pgd_t pgd) { return 0; }
1192 
1193 #ifdef CONFIG_PAGE_TABLE_ISOLATION
1194 /*
1195  * All top-level PAGE_TABLE_ISOLATION page tables are order-1 pages
1196  * (8k-aligned and 8k in size).  The kernel one is at the beginning 4k and
1197  * the user one is in the last 4k.  To switch between them, you
1198  * just need to flip the 12th bit in their addresses.
1199  */
1200 #define PTI_PGTABLE_SWITCH_BIT	PAGE_SHIFT
1201 
1202 /*
1203  * This generates better code than the inline assembly in
1204  * __set_bit().
1205  */
1206 static inline void *ptr_set_bit(void *ptr, int bit)
1207 {
1208 	unsigned long __ptr = (unsigned long)ptr;
1209 
1210 	__ptr |= BIT(bit);
1211 	return (void *)__ptr;
1212 }
1213 static inline void *ptr_clear_bit(void *ptr, int bit)
1214 {
1215 	unsigned long __ptr = (unsigned long)ptr;
1216 
1217 	__ptr &= ~BIT(bit);
1218 	return (void *)__ptr;
1219 }
1220 
1221 static inline pgd_t *kernel_to_user_pgdp(pgd_t *pgdp)
1222 {
1223 	return ptr_set_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
1224 }
1225 
1226 static inline pgd_t *user_to_kernel_pgdp(pgd_t *pgdp)
1227 {
1228 	return ptr_clear_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
1229 }
1230 
1231 static inline p4d_t *kernel_to_user_p4dp(p4d_t *p4dp)
1232 {
1233 	return ptr_set_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
1234 }
1235 
1236 static inline p4d_t *user_to_kernel_p4dp(p4d_t *p4dp)
1237 {
1238 	return ptr_clear_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
1239 }
1240 #endif /* CONFIG_PAGE_TABLE_ISOLATION */
1241 
1242 /*
1243  * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
1244  *
1245  *  dst - pointer to pgd range anywhere on a pgd page
1246  *  src - ""
1247  *  count - the number of pgds to copy.
1248  *
1249  * dst and src can be on the same page, but the range must not overlap,
1250  * and must not cross a page boundary.
1251  */
1252 static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
1253 {
1254 	memcpy(dst, src, count * sizeof(pgd_t));
1255 #ifdef CONFIG_PAGE_TABLE_ISOLATION
1256 	if (!static_cpu_has(X86_FEATURE_PTI))
1257 		return;
1258 	/* Clone the user space pgd as well */
1259 	memcpy(kernel_to_user_pgdp(dst), kernel_to_user_pgdp(src),
1260 	       count * sizeof(pgd_t));
1261 #endif
1262 }
1263 
1264 #define PTE_SHIFT ilog2(PTRS_PER_PTE)
1265 static inline int page_level_shift(enum pg_level level)
1266 {
1267 	return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT;
1268 }
1269 static inline unsigned long page_level_size(enum pg_level level)
1270 {
1271 	return 1UL << page_level_shift(level);
1272 }
1273 static inline unsigned long page_level_mask(enum pg_level level)
1274 {
1275 	return ~(page_level_size(level) - 1);
1276 }
1277 
1278 /*
1279  * The x86 doesn't have any external MMU info: the kernel page
1280  * tables contain all the necessary information.
1281  */
1282 static inline void update_mmu_cache(struct vm_area_struct *vma,
1283 		unsigned long addr, pte_t *ptep)
1284 {
1285 }
1286 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1287 		unsigned long addr, pmd_t *pmd)
1288 {
1289 }
1290 static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
1291 		unsigned long addr, pud_t *pud)
1292 {
1293 }
1294 
1295 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1296 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1297 {
1298 	return pte_set_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1299 }
1300 
1301 static inline int pte_swp_soft_dirty(pte_t pte)
1302 {
1303 	return pte_flags(pte) & _PAGE_SWP_SOFT_DIRTY;
1304 }
1305 
1306 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1307 {
1308 	return pte_clear_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1309 }
1310 
1311 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1312 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1313 {
1314 	return pmd_set_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1315 }
1316 
1317 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1318 {
1319 	return pmd_flags(pmd) & _PAGE_SWP_SOFT_DIRTY;
1320 }
1321 
1322 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1323 {
1324 	return pmd_clear_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1325 }
1326 #endif
1327 #endif
1328 
1329 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
1330 static inline pte_t pte_swp_mkuffd_wp(pte_t pte)
1331 {
1332 	return pte_set_flags(pte, _PAGE_SWP_UFFD_WP);
1333 }
1334 
1335 static inline int pte_swp_uffd_wp(pte_t pte)
1336 {
1337 	return pte_flags(pte) & _PAGE_SWP_UFFD_WP;
1338 }
1339 
1340 static inline pte_t pte_swp_clear_uffd_wp(pte_t pte)
1341 {
1342 	return pte_clear_flags(pte, _PAGE_SWP_UFFD_WP);
1343 }
1344 
1345 static inline pmd_t pmd_swp_mkuffd_wp(pmd_t pmd)
1346 {
1347 	return pmd_set_flags(pmd, _PAGE_SWP_UFFD_WP);
1348 }
1349 
1350 static inline int pmd_swp_uffd_wp(pmd_t pmd)
1351 {
1352 	return pmd_flags(pmd) & _PAGE_SWP_UFFD_WP;
1353 }
1354 
1355 static inline pmd_t pmd_swp_clear_uffd_wp(pmd_t pmd)
1356 {
1357 	return pmd_clear_flags(pmd, _PAGE_SWP_UFFD_WP);
1358 }
1359 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
1360 
1361 static inline u16 pte_flags_pkey(unsigned long pte_flags)
1362 {
1363 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1364 	/* ifdef to avoid doing 59-bit shift on 32-bit values */
1365 	return (pte_flags & _PAGE_PKEY_MASK) >> _PAGE_BIT_PKEY_BIT0;
1366 #else
1367 	return 0;
1368 #endif
1369 }
1370 
1371 static inline bool __pkru_allows_pkey(u16 pkey, bool write)
1372 {
1373 	u32 pkru = read_pkru();
1374 
1375 	if (!__pkru_allows_read(pkru, pkey))
1376 		return false;
1377 	if (write && !__pkru_allows_write(pkru, pkey))
1378 		return false;
1379 
1380 	return true;
1381 }
1382 
1383 /*
1384  * 'pteval' can come from a PTE, PMD or PUD.  We only check
1385  * _PAGE_PRESENT, _PAGE_USER, and _PAGE_RW in here which are the
1386  * same value on all 3 types.
1387  */
1388 static inline bool __pte_access_permitted(unsigned long pteval, bool write)
1389 {
1390 	unsigned long need_pte_bits = _PAGE_PRESENT|_PAGE_USER;
1391 
1392 	if (write)
1393 		need_pte_bits |= _PAGE_RW;
1394 
1395 	if ((pteval & need_pte_bits) != need_pte_bits)
1396 		return 0;
1397 
1398 	return __pkru_allows_pkey(pte_flags_pkey(pteval), write);
1399 }
1400 
1401 #define pte_access_permitted pte_access_permitted
1402 static inline bool pte_access_permitted(pte_t pte, bool write)
1403 {
1404 	return __pte_access_permitted(pte_val(pte), write);
1405 }
1406 
1407 #define pmd_access_permitted pmd_access_permitted
1408 static inline bool pmd_access_permitted(pmd_t pmd, bool write)
1409 {
1410 	return __pte_access_permitted(pmd_val(pmd), write);
1411 }
1412 
1413 #define pud_access_permitted pud_access_permitted
1414 static inline bool pud_access_permitted(pud_t pud, bool write)
1415 {
1416 	return __pte_access_permitted(pud_val(pud), write);
1417 }
1418 
1419 #define __HAVE_ARCH_PFN_MODIFY_ALLOWED 1
1420 extern bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot);
1421 
1422 static inline bool arch_has_pfn_modify_check(void)
1423 {
1424 	return boot_cpu_has_bug(X86_BUG_L1TF);
1425 }
1426 
1427 #define arch_faults_on_old_pte arch_faults_on_old_pte
1428 static inline bool arch_faults_on_old_pte(void)
1429 {
1430 	return false;
1431 }
1432 
1433 #endif	/* __ASSEMBLY__ */
1434 
1435 #endif /* _ASM_X86_PGTABLE_H */
1436