xref: /openbmc/linux/arch/x86/include/asm/pgtable.h (revision 2a598d0b)
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 struct seq_file;
31 void ptdump_walk_pgd_level(struct seq_file *m, struct mm_struct *mm);
32 void ptdump_walk_pgd_level_debugfs(struct seq_file *m, struct mm_struct *mm,
33 				   bool user);
34 void ptdump_walk_pgd_level_checkwx(void);
35 void ptdump_walk_user_pgd_level_checkwx(void);
36 
37 /*
38  * Macros to add or remove encryption attribute
39  */
40 #define pgprot_encrypted(prot)	__pgprot(cc_mkenc(pgprot_val(prot)))
41 #define pgprot_decrypted(prot)	__pgprot(cc_mkdec(pgprot_val(prot)))
42 
43 #ifdef CONFIG_DEBUG_WX
44 #define debug_checkwx()		ptdump_walk_pgd_level_checkwx()
45 #define debug_checkwx_user()	ptdump_walk_user_pgd_level_checkwx()
46 #else
47 #define debug_checkwx()		do { } while (0)
48 #define debug_checkwx_user()	do { } while (0)
49 #endif
50 
51 /*
52  * ZERO_PAGE is a global shared page that is always zero: used
53  * for zero-mapped memory areas etc..
54  */
55 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
56 	__visible;
57 #define ZERO_PAGE(vaddr) ((void)(vaddr),virt_to_page(empty_zero_page))
58 
59 extern spinlock_t pgd_lock;
60 extern struct list_head pgd_list;
61 
62 extern struct mm_struct *pgd_page_get_mm(struct page *page);
63 
64 extern pmdval_t early_pmd_flags;
65 
66 #ifdef CONFIG_PARAVIRT_XXL
67 #include <asm/paravirt.h>
68 #else  /* !CONFIG_PARAVIRT_XXL */
69 #define set_pte(ptep, pte)		native_set_pte(ptep, pte)
70 
71 #define set_pte_atomic(ptep, pte)					\
72 	native_set_pte_atomic(ptep, pte)
73 
74 #define set_pmd(pmdp, pmd)		native_set_pmd(pmdp, pmd)
75 
76 #ifndef __PAGETABLE_P4D_FOLDED
77 #define set_pgd(pgdp, pgd)		native_set_pgd(pgdp, pgd)
78 #define pgd_clear(pgd)			(pgtable_l5_enabled() ? native_pgd_clear(pgd) : 0)
79 #endif
80 
81 #ifndef set_p4d
82 # define set_p4d(p4dp, p4d)		native_set_p4d(p4dp, p4d)
83 #endif
84 
85 #ifndef __PAGETABLE_PUD_FOLDED
86 #define p4d_clear(p4d)			native_p4d_clear(p4d)
87 #endif
88 
89 #ifndef set_pud
90 # define set_pud(pudp, pud)		native_set_pud(pudp, pud)
91 #endif
92 
93 #ifndef __PAGETABLE_PUD_FOLDED
94 #define pud_clear(pud)			native_pud_clear(pud)
95 #endif
96 
97 #define pte_clear(mm, addr, ptep)	native_pte_clear(mm, addr, ptep)
98 #define pmd_clear(pmd)			native_pmd_clear(pmd)
99 
100 #define pgd_val(x)	native_pgd_val(x)
101 #define __pgd(x)	native_make_pgd(x)
102 
103 #ifndef __PAGETABLE_P4D_FOLDED
104 #define p4d_val(x)	native_p4d_val(x)
105 #define __p4d(x)	native_make_p4d(x)
106 #endif
107 
108 #ifndef __PAGETABLE_PUD_FOLDED
109 #define pud_val(x)	native_pud_val(x)
110 #define __pud(x)	native_make_pud(x)
111 #endif
112 
113 #ifndef __PAGETABLE_PMD_FOLDED
114 #define pmd_val(x)	native_pmd_val(x)
115 #define __pmd(x)	native_make_pmd(x)
116 #endif
117 
118 #define pte_val(x)	native_pte_val(x)
119 #define __pte(x)	native_make_pte(x)
120 
121 #define arch_end_context_switch(prev)	do {} while(0)
122 #endif	/* CONFIG_PARAVIRT_XXL */
123 
124 /*
125  * The following only work if pte_present() is true.
126  * Undefined behaviour if not..
127  */
128 static inline int pte_dirty(pte_t pte)
129 {
130 	return pte_flags(pte) & _PAGE_DIRTY;
131 }
132 
133 static inline int pte_young(pte_t pte)
134 {
135 	return pte_flags(pte) & _PAGE_ACCESSED;
136 }
137 
138 static inline int pmd_dirty(pmd_t pmd)
139 {
140 	return pmd_flags(pmd) & _PAGE_DIRTY;
141 }
142 
143 #define pmd_young pmd_young
144 static inline int pmd_young(pmd_t pmd)
145 {
146 	return pmd_flags(pmd) & _PAGE_ACCESSED;
147 }
148 
149 static inline int pud_dirty(pud_t pud)
150 {
151 	return pud_flags(pud) & _PAGE_DIRTY;
152 }
153 
154 static inline int pud_young(pud_t pud)
155 {
156 	return pud_flags(pud) & _PAGE_ACCESSED;
157 }
158 
159 static inline int pte_write(pte_t pte)
160 {
161 	return pte_flags(pte) & _PAGE_RW;
162 }
163 
164 static inline int pte_huge(pte_t pte)
165 {
166 	return pte_flags(pte) & _PAGE_PSE;
167 }
168 
169 static inline int pte_global(pte_t pte)
170 {
171 	return pte_flags(pte) & _PAGE_GLOBAL;
172 }
173 
174 static inline int pte_exec(pte_t pte)
175 {
176 	return !(pte_flags(pte) & _PAGE_NX);
177 }
178 
179 static inline int pte_special(pte_t pte)
180 {
181 	return pte_flags(pte) & _PAGE_SPECIAL;
182 }
183 
184 /* Entries that were set to PROT_NONE are inverted */
185 
186 static inline u64 protnone_mask(u64 val);
187 
188 static inline unsigned long pte_pfn(pte_t pte)
189 {
190 	phys_addr_t pfn = pte_val(pte);
191 	pfn ^= protnone_mask(pfn);
192 	return (pfn & PTE_PFN_MASK) >> PAGE_SHIFT;
193 }
194 
195 static inline unsigned long pmd_pfn(pmd_t pmd)
196 {
197 	phys_addr_t pfn = pmd_val(pmd);
198 	pfn ^= protnone_mask(pfn);
199 	return (pfn & pmd_pfn_mask(pmd)) >> PAGE_SHIFT;
200 }
201 
202 static inline unsigned long pud_pfn(pud_t pud)
203 {
204 	phys_addr_t pfn = pud_val(pud);
205 	pfn ^= protnone_mask(pfn);
206 	return (pfn & pud_pfn_mask(pud)) >> PAGE_SHIFT;
207 }
208 
209 static inline unsigned long p4d_pfn(p4d_t p4d)
210 {
211 	return (p4d_val(p4d) & p4d_pfn_mask(p4d)) >> PAGE_SHIFT;
212 }
213 
214 static inline unsigned long pgd_pfn(pgd_t pgd)
215 {
216 	return (pgd_val(pgd) & PTE_PFN_MASK) >> PAGE_SHIFT;
217 }
218 
219 #define p4d_leaf	p4d_large
220 static inline int p4d_large(p4d_t p4d)
221 {
222 	/* No 512 GiB pages yet */
223 	return 0;
224 }
225 
226 #define pte_page(pte)	pfn_to_page(pte_pfn(pte))
227 
228 #define pmd_leaf	pmd_large
229 static inline int pmd_large(pmd_t pte)
230 {
231 	return pmd_flags(pte) & _PAGE_PSE;
232 }
233 
234 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
235 /* NOTE: when predicate huge page, consider also pmd_devmap, or use pmd_large */
236 static inline int pmd_trans_huge(pmd_t pmd)
237 {
238 	return (pmd_val(pmd) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
239 }
240 
241 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
242 static inline int pud_trans_huge(pud_t pud)
243 {
244 	return (pud_val(pud) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
245 }
246 #endif
247 
248 #define has_transparent_hugepage has_transparent_hugepage
249 static inline int has_transparent_hugepage(void)
250 {
251 	return boot_cpu_has(X86_FEATURE_PSE);
252 }
253 
254 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
255 static inline int pmd_devmap(pmd_t pmd)
256 {
257 	return !!(pmd_val(pmd) & _PAGE_DEVMAP);
258 }
259 
260 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
261 static inline int pud_devmap(pud_t pud)
262 {
263 	return !!(pud_val(pud) & _PAGE_DEVMAP);
264 }
265 #else
266 static inline int pud_devmap(pud_t pud)
267 {
268 	return 0;
269 }
270 #endif
271 
272 static inline int pgd_devmap(pgd_t pgd)
273 {
274 	return 0;
275 }
276 #endif
277 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
278 
279 static inline pte_t pte_set_flags(pte_t pte, pteval_t set)
280 {
281 	pteval_t v = native_pte_val(pte);
282 
283 	return native_make_pte(v | set);
284 }
285 
286 static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear)
287 {
288 	pteval_t v = native_pte_val(pte);
289 
290 	return native_make_pte(v & ~clear);
291 }
292 
293 static inline pte_t pte_wrprotect(pte_t pte)
294 {
295 	return pte_clear_flags(pte, _PAGE_RW);
296 }
297 
298 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
299 static inline int pte_uffd_wp(pte_t pte)
300 {
301 	bool wp = pte_flags(pte) & _PAGE_UFFD_WP;
302 
303 #ifdef CONFIG_DEBUG_VM
304 	/*
305 	 * Having write bit for wr-protect-marked present ptes is fatal,
306 	 * because it means the uffd-wp bit will be ignored and write will
307 	 * just go through.
308 	 *
309 	 * Use any chance of pgtable walking to verify this (e.g., when
310 	 * page swapped out or being migrated for all purposes). It means
311 	 * something is already wrong.  Tell the admin even before the
312 	 * process crashes. We also nail it with wrong pgtable setup.
313 	 */
314 	WARN_ON_ONCE(wp && pte_write(pte));
315 #endif
316 
317 	return wp;
318 }
319 
320 static inline pte_t pte_mkuffd_wp(pte_t pte)
321 {
322 	return pte_wrprotect(pte_set_flags(pte, _PAGE_UFFD_WP));
323 }
324 
325 static inline pte_t pte_clear_uffd_wp(pte_t pte)
326 {
327 	return pte_clear_flags(pte, _PAGE_UFFD_WP);
328 }
329 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
330 
331 static inline pte_t pte_mkclean(pte_t pte)
332 {
333 	return pte_clear_flags(pte, _PAGE_DIRTY);
334 }
335 
336 static inline pte_t pte_mkold(pte_t pte)
337 {
338 	return pte_clear_flags(pte, _PAGE_ACCESSED);
339 }
340 
341 static inline pte_t pte_mkexec(pte_t pte)
342 {
343 	return pte_clear_flags(pte, _PAGE_NX);
344 }
345 
346 static inline pte_t pte_mkdirty(pte_t pte)
347 {
348 	return pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
349 }
350 
351 static inline pte_t pte_mkyoung(pte_t pte)
352 {
353 	return pte_set_flags(pte, _PAGE_ACCESSED);
354 }
355 
356 static inline pte_t pte_mkwrite(pte_t pte)
357 {
358 	return pte_set_flags(pte, _PAGE_RW);
359 }
360 
361 static inline pte_t pte_mkhuge(pte_t pte)
362 {
363 	return pte_set_flags(pte, _PAGE_PSE);
364 }
365 
366 static inline pte_t pte_clrhuge(pte_t pte)
367 {
368 	return pte_clear_flags(pte, _PAGE_PSE);
369 }
370 
371 static inline pte_t pte_mkglobal(pte_t pte)
372 {
373 	return pte_set_flags(pte, _PAGE_GLOBAL);
374 }
375 
376 static inline pte_t pte_clrglobal(pte_t pte)
377 {
378 	return pte_clear_flags(pte, _PAGE_GLOBAL);
379 }
380 
381 static inline pte_t pte_mkspecial(pte_t pte)
382 {
383 	return pte_set_flags(pte, _PAGE_SPECIAL);
384 }
385 
386 static inline pte_t pte_mkdevmap(pte_t pte)
387 {
388 	return pte_set_flags(pte, _PAGE_SPECIAL|_PAGE_DEVMAP);
389 }
390 
391 static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
392 {
393 	pmdval_t v = native_pmd_val(pmd);
394 
395 	return native_make_pmd(v | set);
396 }
397 
398 static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
399 {
400 	pmdval_t v = native_pmd_val(pmd);
401 
402 	return native_make_pmd(v & ~clear);
403 }
404 
405 static inline pmd_t pmd_wrprotect(pmd_t pmd)
406 {
407 	return pmd_clear_flags(pmd, _PAGE_RW);
408 }
409 
410 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
411 static inline int pmd_uffd_wp(pmd_t pmd)
412 {
413 	return pmd_flags(pmd) & _PAGE_UFFD_WP;
414 }
415 
416 static inline pmd_t pmd_mkuffd_wp(pmd_t pmd)
417 {
418 	return pmd_wrprotect(pmd_set_flags(pmd, _PAGE_UFFD_WP));
419 }
420 
421 static inline pmd_t pmd_clear_uffd_wp(pmd_t pmd)
422 {
423 	return pmd_clear_flags(pmd, _PAGE_UFFD_WP);
424 }
425 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
426 
427 static inline pmd_t pmd_mkold(pmd_t pmd)
428 {
429 	return pmd_clear_flags(pmd, _PAGE_ACCESSED);
430 }
431 
432 static inline pmd_t pmd_mkclean(pmd_t pmd)
433 {
434 	return pmd_clear_flags(pmd, _PAGE_DIRTY);
435 }
436 
437 static inline pmd_t pmd_mkdirty(pmd_t pmd)
438 {
439 	return pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
440 }
441 
442 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
443 {
444 	return pmd_set_flags(pmd, _PAGE_DEVMAP);
445 }
446 
447 static inline pmd_t pmd_mkhuge(pmd_t pmd)
448 {
449 	return pmd_set_flags(pmd, _PAGE_PSE);
450 }
451 
452 static inline pmd_t pmd_mkyoung(pmd_t pmd)
453 {
454 	return pmd_set_flags(pmd, _PAGE_ACCESSED);
455 }
456 
457 static inline pmd_t pmd_mkwrite(pmd_t pmd)
458 {
459 	return pmd_set_flags(pmd, _PAGE_RW);
460 }
461 
462 static inline pud_t pud_set_flags(pud_t pud, pudval_t set)
463 {
464 	pudval_t v = native_pud_val(pud);
465 
466 	return native_make_pud(v | set);
467 }
468 
469 static inline pud_t pud_clear_flags(pud_t pud, pudval_t clear)
470 {
471 	pudval_t v = native_pud_val(pud);
472 
473 	return native_make_pud(v & ~clear);
474 }
475 
476 static inline pud_t pud_mkold(pud_t pud)
477 {
478 	return pud_clear_flags(pud, _PAGE_ACCESSED);
479 }
480 
481 static inline pud_t pud_mkclean(pud_t pud)
482 {
483 	return pud_clear_flags(pud, _PAGE_DIRTY);
484 }
485 
486 static inline pud_t pud_wrprotect(pud_t pud)
487 {
488 	return pud_clear_flags(pud, _PAGE_RW);
489 }
490 
491 static inline pud_t pud_mkdirty(pud_t pud)
492 {
493 	return pud_set_flags(pud, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
494 }
495 
496 static inline pud_t pud_mkdevmap(pud_t pud)
497 {
498 	return pud_set_flags(pud, _PAGE_DEVMAP);
499 }
500 
501 static inline pud_t pud_mkhuge(pud_t pud)
502 {
503 	return pud_set_flags(pud, _PAGE_PSE);
504 }
505 
506 static inline pud_t pud_mkyoung(pud_t pud)
507 {
508 	return pud_set_flags(pud, _PAGE_ACCESSED);
509 }
510 
511 static inline pud_t pud_mkwrite(pud_t pud)
512 {
513 	return pud_set_flags(pud, _PAGE_RW);
514 }
515 
516 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
517 static inline int pte_soft_dirty(pte_t pte)
518 {
519 	return pte_flags(pte) & _PAGE_SOFT_DIRTY;
520 }
521 
522 static inline int pmd_soft_dirty(pmd_t pmd)
523 {
524 	return pmd_flags(pmd) & _PAGE_SOFT_DIRTY;
525 }
526 
527 static inline int pud_soft_dirty(pud_t pud)
528 {
529 	return pud_flags(pud) & _PAGE_SOFT_DIRTY;
530 }
531 
532 static inline pte_t pte_mksoft_dirty(pte_t pte)
533 {
534 	return pte_set_flags(pte, _PAGE_SOFT_DIRTY);
535 }
536 
537 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
538 {
539 	return pmd_set_flags(pmd, _PAGE_SOFT_DIRTY);
540 }
541 
542 static inline pud_t pud_mksoft_dirty(pud_t pud)
543 {
544 	return pud_set_flags(pud, _PAGE_SOFT_DIRTY);
545 }
546 
547 static inline pte_t pte_clear_soft_dirty(pte_t pte)
548 {
549 	return pte_clear_flags(pte, _PAGE_SOFT_DIRTY);
550 }
551 
552 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
553 {
554 	return pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY);
555 }
556 
557 static inline pud_t pud_clear_soft_dirty(pud_t pud)
558 {
559 	return pud_clear_flags(pud, _PAGE_SOFT_DIRTY);
560 }
561 
562 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
563 
564 /*
565  * Mask out unsupported bits in a present pgprot.  Non-present pgprots
566  * can use those bits for other purposes, so leave them be.
567  */
568 static inline pgprotval_t massage_pgprot(pgprot_t pgprot)
569 {
570 	pgprotval_t protval = pgprot_val(pgprot);
571 
572 	if (protval & _PAGE_PRESENT)
573 		protval &= __supported_pte_mask;
574 
575 	return protval;
576 }
577 
578 static inline pgprotval_t check_pgprot(pgprot_t pgprot)
579 {
580 	pgprotval_t massaged_val = massage_pgprot(pgprot);
581 
582 	/* mmdebug.h can not be included here because of dependencies */
583 #ifdef CONFIG_DEBUG_VM
584 	WARN_ONCE(pgprot_val(pgprot) != massaged_val,
585 		  "attempted to set unsupported pgprot: %016llx "
586 		  "bits: %016llx supported: %016llx\n",
587 		  (u64)pgprot_val(pgprot),
588 		  (u64)pgprot_val(pgprot) ^ massaged_val,
589 		  (u64)__supported_pte_mask);
590 #endif
591 
592 	return massaged_val;
593 }
594 
595 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
596 {
597 	phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
598 	pfn ^= protnone_mask(pgprot_val(pgprot));
599 	pfn &= PTE_PFN_MASK;
600 	return __pte(pfn | check_pgprot(pgprot));
601 }
602 
603 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
604 {
605 	phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
606 	pfn ^= protnone_mask(pgprot_val(pgprot));
607 	pfn &= PHYSICAL_PMD_PAGE_MASK;
608 	return __pmd(pfn | check_pgprot(pgprot));
609 }
610 
611 static inline pud_t pfn_pud(unsigned long page_nr, pgprot_t pgprot)
612 {
613 	phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
614 	pfn ^= protnone_mask(pgprot_val(pgprot));
615 	pfn &= PHYSICAL_PUD_PAGE_MASK;
616 	return __pud(pfn | check_pgprot(pgprot));
617 }
618 
619 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
620 {
621 	return pfn_pmd(pmd_pfn(pmd),
622 		      __pgprot(pmd_flags(pmd) & ~(_PAGE_PRESENT|_PAGE_PROTNONE)));
623 }
624 
625 static inline u64 flip_protnone_guard(u64 oldval, u64 val, u64 mask);
626 
627 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
628 {
629 	pteval_t val = pte_val(pte), oldval = val;
630 
631 	/*
632 	 * Chop off the NX bit (if present), and add the NX portion of
633 	 * the newprot (if present):
634 	 */
635 	val &= _PAGE_CHG_MASK;
636 	val |= check_pgprot(newprot) & ~_PAGE_CHG_MASK;
637 	val = flip_protnone_guard(oldval, val, PTE_PFN_MASK);
638 	return __pte(val);
639 }
640 
641 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
642 {
643 	pmdval_t val = pmd_val(pmd), oldval = val;
644 
645 	val &= _HPAGE_CHG_MASK;
646 	val |= check_pgprot(newprot) & ~_HPAGE_CHG_MASK;
647 	val = flip_protnone_guard(oldval, val, PHYSICAL_PMD_PAGE_MASK);
648 	return __pmd(val);
649 }
650 
651 /*
652  * mprotect needs to preserve PAT and encryption bits when updating
653  * vm_page_prot
654  */
655 #define pgprot_modify pgprot_modify
656 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
657 {
658 	pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK;
659 	pgprotval_t addbits = pgprot_val(newprot) & ~_PAGE_CHG_MASK;
660 	return __pgprot(preservebits | addbits);
661 }
662 
663 #define pte_pgprot(x) __pgprot(pte_flags(x))
664 #define pmd_pgprot(x) __pgprot(pmd_flags(x))
665 #define pud_pgprot(x) __pgprot(pud_flags(x))
666 #define p4d_pgprot(x) __pgprot(p4d_flags(x))
667 
668 #define canon_pgprot(p) __pgprot(massage_pgprot(p))
669 
670 static inline int is_new_memtype_allowed(u64 paddr, unsigned long size,
671 					 enum page_cache_mode pcm,
672 					 enum page_cache_mode new_pcm)
673 {
674 	/*
675 	 * PAT type is always WB for untracked ranges, so no need to check.
676 	 */
677 	if (x86_platform.is_untracked_pat_range(paddr, paddr + size))
678 		return 1;
679 
680 	/*
681 	 * Certain new memtypes are not allowed with certain
682 	 * requested memtype:
683 	 * - request is uncached, return cannot be write-back
684 	 * - request is write-combine, return cannot be write-back
685 	 * - request is write-through, return cannot be write-back
686 	 * - request is write-through, return cannot be write-combine
687 	 */
688 	if ((pcm == _PAGE_CACHE_MODE_UC_MINUS &&
689 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
690 	    (pcm == _PAGE_CACHE_MODE_WC &&
691 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
692 	    (pcm == _PAGE_CACHE_MODE_WT &&
693 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
694 	    (pcm == _PAGE_CACHE_MODE_WT &&
695 	     new_pcm == _PAGE_CACHE_MODE_WC)) {
696 		return 0;
697 	}
698 
699 	return 1;
700 }
701 
702 pmd_t *populate_extra_pmd(unsigned long vaddr);
703 pte_t *populate_extra_pte(unsigned long vaddr);
704 
705 #ifdef CONFIG_PAGE_TABLE_ISOLATION
706 pgd_t __pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd);
707 
708 /*
709  * Take a PGD location (pgdp) and a pgd value that needs to be set there.
710  * Populates the user and returns the resulting PGD that must be set in
711  * the kernel copy of the page tables.
712  */
713 static inline pgd_t pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd)
714 {
715 	if (!static_cpu_has(X86_FEATURE_PTI))
716 		return pgd;
717 	return __pti_set_user_pgtbl(pgdp, pgd);
718 }
719 #else   /* CONFIG_PAGE_TABLE_ISOLATION */
720 static inline pgd_t pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd)
721 {
722 	return pgd;
723 }
724 #endif  /* CONFIG_PAGE_TABLE_ISOLATION */
725 
726 #endif	/* __ASSEMBLY__ */
727 
728 
729 #ifdef CONFIG_X86_32
730 # include <asm/pgtable_32.h>
731 #else
732 # include <asm/pgtable_64.h>
733 #endif
734 
735 #ifndef __ASSEMBLY__
736 #include <linux/mm_types.h>
737 #include <linux/mmdebug.h>
738 #include <linux/log2.h>
739 #include <asm/fixmap.h>
740 
741 static inline int pte_none(pte_t pte)
742 {
743 	return !(pte.pte & ~(_PAGE_KNL_ERRATUM_MASK));
744 }
745 
746 #define __HAVE_ARCH_PTE_SAME
747 static inline int pte_same(pte_t a, pte_t b)
748 {
749 	return a.pte == b.pte;
750 }
751 
752 static inline int pte_present(pte_t a)
753 {
754 	return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE);
755 }
756 
757 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
758 static inline int pte_devmap(pte_t a)
759 {
760 	return (pte_flags(a) & _PAGE_DEVMAP) == _PAGE_DEVMAP;
761 }
762 #endif
763 
764 #define pte_accessible pte_accessible
765 static inline bool pte_accessible(struct mm_struct *mm, pte_t a)
766 {
767 	if (pte_flags(a) & _PAGE_PRESENT)
768 		return true;
769 
770 	if ((pte_flags(a) & _PAGE_PROTNONE) &&
771 			atomic_read(&mm->tlb_flush_pending))
772 		return true;
773 
774 	return false;
775 }
776 
777 static inline int pmd_present(pmd_t pmd)
778 {
779 	/*
780 	 * Checking for _PAGE_PSE is needed too because
781 	 * split_huge_page will temporarily clear the present bit (but
782 	 * the _PAGE_PSE flag will remain set at all times while the
783 	 * _PAGE_PRESENT bit is clear).
784 	 */
785 	return pmd_flags(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PSE);
786 }
787 
788 #ifdef CONFIG_NUMA_BALANCING
789 /*
790  * These work without NUMA balancing but the kernel does not care. See the
791  * comment in include/linux/pgtable.h
792  */
793 static inline int pte_protnone(pte_t pte)
794 {
795 	return (pte_flags(pte) & (_PAGE_PROTNONE | _PAGE_PRESENT))
796 		== _PAGE_PROTNONE;
797 }
798 
799 static inline int pmd_protnone(pmd_t pmd)
800 {
801 	return (pmd_flags(pmd) & (_PAGE_PROTNONE | _PAGE_PRESENT))
802 		== _PAGE_PROTNONE;
803 }
804 #endif /* CONFIG_NUMA_BALANCING */
805 
806 static inline int pmd_none(pmd_t pmd)
807 {
808 	/* Only check low word on 32-bit platforms, since it might be
809 	   out of sync with upper half. */
810 	unsigned long val = native_pmd_val(pmd);
811 	return (val & ~_PAGE_KNL_ERRATUM_MASK) == 0;
812 }
813 
814 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
815 {
816 	return (unsigned long)__va(pmd_val(pmd) & pmd_pfn_mask(pmd));
817 }
818 
819 /*
820  * Currently stuck as a macro due to indirect forward reference to
821  * linux/mmzone.h's __section_mem_map_addr() definition:
822  */
823 #define pmd_page(pmd)	pfn_to_page(pmd_pfn(pmd))
824 
825 /*
826  * Conversion functions: convert a page and protection to a page entry,
827  * and a page entry and page directory to the page they refer to.
828  *
829  * (Currently stuck as a macro because of indirect forward reference
830  * to linux/mm.h:page_to_nid())
831  */
832 #define mk_pte(page, pgprot)   pfn_pte(page_to_pfn(page), (pgprot))
833 
834 static inline int pmd_bad(pmd_t pmd)
835 {
836 	return (pmd_flags(pmd) & ~(_PAGE_USER | _PAGE_ACCESSED)) !=
837 	       (_KERNPG_TABLE & ~_PAGE_ACCESSED);
838 }
839 
840 static inline unsigned long pages_to_mb(unsigned long npg)
841 {
842 	return npg >> (20 - PAGE_SHIFT);
843 }
844 
845 #if CONFIG_PGTABLE_LEVELS > 2
846 static inline int pud_none(pud_t pud)
847 {
848 	return (native_pud_val(pud) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
849 }
850 
851 static inline int pud_present(pud_t pud)
852 {
853 	return pud_flags(pud) & _PAGE_PRESENT;
854 }
855 
856 static inline pmd_t *pud_pgtable(pud_t pud)
857 {
858 	return (pmd_t *)__va(pud_val(pud) & pud_pfn_mask(pud));
859 }
860 
861 /*
862  * Currently stuck as a macro due to indirect forward reference to
863  * linux/mmzone.h's __section_mem_map_addr() definition:
864  */
865 #define pud_page(pud)	pfn_to_page(pud_pfn(pud))
866 
867 #define pud_leaf	pud_large
868 static inline int pud_large(pud_t pud)
869 {
870 	return (pud_val(pud) & (_PAGE_PSE | _PAGE_PRESENT)) ==
871 		(_PAGE_PSE | _PAGE_PRESENT);
872 }
873 
874 static inline int pud_bad(pud_t pud)
875 {
876 	return (pud_flags(pud) & ~(_KERNPG_TABLE | _PAGE_USER)) != 0;
877 }
878 #else
879 #define pud_leaf	pud_large
880 static inline int pud_large(pud_t pud)
881 {
882 	return 0;
883 }
884 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
885 
886 #if CONFIG_PGTABLE_LEVELS > 3
887 static inline int p4d_none(p4d_t p4d)
888 {
889 	return (native_p4d_val(p4d) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
890 }
891 
892 static inline int p4d_present(p4d_t p4d)
893 {
894 	return p4d_flags(p4d) & _PAGE_PRESENT;
895 }
896 
897 static inline pud_t *p4d_pgtable(p4d_t p4d)
898 {
899 	return (pud_t *)__va(p4d_val(p4d) & p4d_pfn_mask(p4d));
900 }
901 
902 /*
903  * Currently stuck as a macro due to indirect forward reference to
904  * linux/mmzone.h's __section_mem_map_addr() definition:
905  */
906 #define p4d_page(p4d)	pfn_to_page(p4d_pfn(p4d))
907 
908 static inline int p4d_bad(p4d_t p4d)
909 {
910 	unsigned long ignore_flags = _KERNPG_TABLE | _PAGE_USER;
911 
912 	if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
913 		ignore_flags |= _PAGE_NX;
914 
915 	return (p4d_flags(p4d) & ~ignore_flags) != 0;
916 }
917 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
918 
919 static inline unsigned long p4d_index(unsigned long address)
920 {
921 	return (address >> P4D_SHIFT) & (PTRS_PER_P4D - 1);
922 }
923 
924 #if CONFIG_PGTABLE_LEVELS > 4
925 static inline int pgd_present(pgd_t pgd)
926 {
927 	if (!pgtable_l5_enabled())
928 		return 1;
929 	return pgd_flags(pgd) & _PAGE_PRESENT;
930 }
931 
932 static inline unsigned long pgd_page_vaddr(pgd_t pgd)
933 {
934 	return (unsigned long)__va((unsigned long)pgd_val(pgd) & PTE_PFN_MASK);
935 }
936 
937 /*
938  * Currently stuck as a macro due to indirect forward reference to
939  * linux/mmzone.h's __section_mem_map_addr() definition:
940  */
941 #define pgd_page(pgd)	pfn_to_page(pgd_pfn(pgd))
942 
943 /* to find an entry in a page-table-directory. */
944 static inline p4d_t *p4d_offset(pgd_t *pgd, unsigned long address)
945 {
946 	if (!pgtable_l5_enabled())
947 		return (p4d_t *)pgd;
948 	return (p4d_t *)pgd_page_vaddr(*pgd) + p4d_index(address);
949 }
950 
951 static inline int pgd_bad(pgd_t pgd)
952 {
953 	unsigned long ignore_flags = _PAGE_USER;
954 
955 	if (!pgtable_l5_enabled())
956 		return 0;
957 
958 	if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
959 		ignore_flags |= _PAGE_NX;
960 
961 	return (pgd_flags(pgd) & ~ignore_flags) != _KERNPG_TABLE;
962 }
963 
964 static inline int pgd_none(pgd_t pgd)
965 {
966 	if (!pgtable_l5_enabled())
967 		return 0;
968 	/*
969 	 * There is no need to do a workaround for the KNL stray
970 	 * A/D bit erratum here.  PGDs only point to page tables
971 	 * except on 32-bit non-PAE which is not supported on
972 	 * KNL.
973 	 */
974 	return !native_pgd_val(pgd);
975 }
976 #endif	/* CONFIG_PGTABLE_LEVELS > 4 */
977 
978 #endif	/* __ASSEMBLY__ */
979 
980 #define KERNEL_PGD_BOUNDARY	pgd_index(PAGE_OFFSET)
981 #define KERNEL_PGD_PTRS		(PTRS_PER_PGD - KERNEL_PGD_BOUNDARY)
982 
983 #ifndef __ASSEMBLY__
984 
985 extern int direct_gbpages;
986 void init_mem_mapping(void);
987 void early_alloc_pgt_buf(void);
988 extern void memblock_find_dma_reserve(void);
989 void __init poking_init(void);
990 unsigned long init_memory_mapping(unsigned long start,
991 				  unsigned long end, pgprot_t prot);
992 
993 #ifdef CONFIG_X86_64
994 extern pgd_t trampoline_pgd_entry;
995 #endif
996 
997 /* local pte updates need not use xchg for locking */
998 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
999 {
1000 	pte_t res = *ptep;
1001 
1002 	/* Pure native function needs no input for mm, addr */
1003 	native_pte_clear(NULL, 0, ptep);
1004 	return res;
1005 }
1006 
1007 static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp)
1008 {
1009 	pmd_t res = *pmdp;
1010 
1011 	native_pmd_clear(pmdp);
1012 	return res;
1013 }
1014 
1015 static inline pud_t native_local_pudp_get_and_clear(pud_t *pudp)
1016 {
1017 	pud_t res = *pudp;
1018 
1019 	native_pud_clear(pudp);
1020 	return res;
1021 }
1022 
1023 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
1024 			      pte_t *ptep, pte_t pte)
1025 {
1026 	page_table_check_pte_set(mm, addr, ptep, pte);
1027 	set_pte(ptep, pte);
1028 }
1029 
1030 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1031 			      pmd_t *pmdp, pmd_t pmd)
1032 {
1033 	page_table_check_pmd_set(mm, addr, pmdp, pmd);
1034 	set_pmd(pmdp, pmd);
1035 }
1036 
1037 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
1038 			      pud_t *pudp, pud_t pud)
1039 {
1040 	page_table_check_pud_set(mm, addr, pudp, pud);
1041 	native_set_pud(pudp, pud);
1042 }
1043 
1044 /*
1045  * We only update the dirty/accessed state if we set
1046  * the dirty bit by hand in the kernel, since the hardware
1047  * will do the accessed bit for us, and we don't want to
1048  * race with other CPU's that might be updating the dirty
1049  * bit at the same time.
1050  */
1051 struct vm_area_struct;
1052 
1053 #define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1054 extern int ptep_set_access_flags(struct vm_area_struct *vma,
1055 				 unsigned long address, pte_t *ptep,
1056 				 pte_t entry, int dirty);
1057 
1058 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1059 extern int ptep_test_and_clear_young(struct vm_area_struct *vma,
1060 				     unsigned long addr, pte_t *ptep);
1061 
1062 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1063 extern int ptep_clear_flush_young(struct vm_area_struct *vma,
1064 				  unsigned long address, pte_t *ptep);
1065 
1066 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1067 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
1068 				       pte_t *ptep)
1069 {
1070 	pte_t pte = native_ptep_get_and_clear(ptep);
1071 	page_table_check_pte_clear(mm, addr, pte);
1072 	return pte;
1073 }
1074 
1075 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1076 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1077 					    unsigned long addr, pte_t *ptep,
1078 					    int full)
1079 {
1080 	pte_t pte;
1081 	if (full) {
1082 		/*
1083 		 * Full address destruction in progress; paravirt does not
1084 		 * care about updates and native needs no locking
1085 		 */
1086 		pte = native_local_ptep_get_and_clear(ptep);
1087 		page_table_check_pte_clear(mm, addr, pte);
1088 	} else {
1089 		pte = ptep_get_and_clear(mm, addr, ptep);
1090 	}
1091 	return pte;
1092 }
1093 
1094 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1095 static inline void ptep_set_wrprotect(struct mm_struct *mm,
1096 				      unsigned long addr, pte_t *ptep)
1097 {
1098 	clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte);
1099 }
1100 
1101 #define flush_tlb_fix_spurious_fault(vma, address, ptep) do { } while (0)
1102 
1103 #define mk_pmd(page, pgprot)   pfn_pmd(page_to_pfn(page), (pgprot))
1104 
1105 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1106 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
1107 				 unsigned long address, pmd_t *pmdp,
1108 				 pmd_t entry, int dirty);
1109 extern int pudp_set_access_flags(struct vm_area_struct *vma,
1110 				 unsigned long address, pud_t *pudp,
1111 				 pud_t entry, int dirty);
1112 
1113 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1114 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1115 				     unsigned long addr, pmd_t *pmdp);
1116 extern int pudp_test_and_clear_young(struct vm_area_struct *vma,
1117 				     unsigned long addr, pud_t *pudp);
1118 
1119 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
1120 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
1121 				  unsigned long address, pmd_t *pmdp);
1122 
1123 
1124 #define pmd_write pmd_write
1125 static inline int pmd_write(pmd_t pmd)
1126 {
1127 	return pmd_flags(pmd) & _PAGE_RW;
1128 }
1129 
1130 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1131 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, unsigned long addr,
1132 				       pmd_t *pmdp)
1133 {
1134 	pmd_t pmd = native_pmdp_get_and_clear(pmdp);
1135 
1136 	page_table_check_pmd_clear(mm, addr, pmd);
1137 
1138 	return pmd;
1139 }
1140 
1141 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
1142 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
1143 					unsigned long addr, pud_t *pudp)
1144 {
1145 	pud_t pud = native_pudp_get_and_clear(pudp);
1146 
1147 	page_table_check_pud_clear(mm, addr, pud);
1148 
1149 	return pud;
1150 }
1151 
1152 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1153 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1154 				      unsigned long addr, pmd_t *pmdp)
1155 {
1156 	clear_bit(_PAGE_BIT_RW, (unsigned long *)pmdp);
1157 }
1158 
1159 #define pud_write pud_write
1160 static inline int pud_write(pud_t pud)
1161 {
1162 	return pud_flags(pud) & _PAGE_RW;
1163 }
1164 
1165 #ifndef pmdp_establish
1166 #define pmdp_establish pmdp_establish
1167 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
1168 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
1169 {
1170 	page_table_check_pmd_set(vma->vm_mm, address, pmdp, pmd);
1171 	if (IS_ENABLED(CONFIG_SMP)) {
1172 		return xchg(pmdp, pmd);
1173 	} else {
1174 		pmd_t old = *pmdp;
1175 		WRITE_ONCE(*pmdp, pmd);
1176 		return old;
1177 	}
1178 }
1179 #endif
1180 
1181 #define __HAVE_ARCH_PMDP_INVALIDATE_AD
1182 extern pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma,
1183 				unsigned long address, pmd_t *pmdp);
1184 
1185 /*
1186  * Page table pages are page-aligned.  The lower half of the top
1187  * level is used for userspace and the top half for the kernel.
1188  *
1189  * Returns true for parts of the PGD that map userspace and
1190  * false for the parts that map the kernel.
1191  */
1192 static inline bool pgdp_maps_userspace(void *__ptr)
1193 {
1194 	unsigned long ptr = (unsigned long)__ptr;
1195 
1196 	return (((ptr & ~PAGE_MASK) / sizeof(pgd_t)) < PGD_KERNEL_START);
1197 }
1198 
1199 #define pgd_leaf	pgd_large
1200 static inline int pgd_large(pgd_t pgd) { return 0; }
1201 
1202 #ifdef CONFIG_PAGE_TABLE_ISOLATION
1203 /*
1204  * All top-level PAGE_TABLE_ISOLATION page tables are order-1 pages
1205  * (8k-aligned and 8k in size).  The kernel one is at the beginning 4k and
1206  * the user one is in the last 4k.  To switch between them, you
1207  * just need to flip the 12th bit in their addresses.
1208  */
1209 #define PTI_PGTABLE_SWITCH_BIT	PAGE_SHIFT
1210 
1211 /*
1212  * This generates better code than the inline assembly in
1213  * __set_bit().
1214  */
1215 static inline void *ptr_set_bit(void *ptr, int bit)
1216 {
1217 	unsigned long __ptr = (unsigned long)ptr;
1218 
1219 	__ptr |= BIT(bit);
1220 	return (void *)__ptr;
1221 }
1222 static inline void *ptr_clear_bit(void *ptr, int bit)
1223 {
1224 	unsigned long __ptr = (unsigned long)ptr;
1225 
1226 	__ptr &= ~BIT(bit);
1227 	return (void *)__ptr;
1228 }
1229 
1230 static inline pgd_t *kernel_to_user_pgdp(pgd_t *pgdp)
1231 {
1232 	return ptr_set_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
1233 }
1234 
1235 static inline pgd_t *user_to_kernel_pgdp(pgd_t *pgdp)
1236 {
1237 	return ptr_clear_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
1238 }
1239 
1240 static inline p4d_t *kernel_to_user_p4dp(p4d_t *p4dp)
1241 {
1242 	return ptr_set_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
1243 }
1244 
1245 static inline p4d_t *user_to_kernel_p4dp(p4d_t *p4dp)
1246 {
1247 	return ptr_clear_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
1248 }
1249 #endif /* CONFIG_PAGE_TABLE_ISOLATION */
1250 
1251 /*
1252  * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
1253  *
1254  *  dst - pointer to pgd range anywhere on a pgd page
1255  *  src - ""
1256  *  count - the number of pgds to copy.
1257  *
1258  * dst and src can be on the same page, but the range must not overlap,
1259  * and must not cross a page boundary.
1260  */
1261 static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
1262 {
1263 	memcpy(dst, src, count * sizeof(pgd_t));
1264 #ifdef CONFIG_PAGE_TABLE_ISOLATION
1265 	if (!static_cpu_has(X86_FEATURE_PTI))
1266 		return;
1267 	/* Clone the user space pgd as well */
1268 	memcpy(kernel_to_user_pgdp(dst), kernel_to_user_pgdp(src),
1269 	       count * sizeof(pgd_t));
1270 #endif
1271 }
1272 
1273 #define PTE_SHIFT ilog2(PTRS_PER_PTE)
1274 static inline int page_level_shift(enum pg_level level)
1275 {
1276 	return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT;
1277 }
1278 static inline unsigned long page_level_size(enum pg_level level)
1279 {
1280 	return 1UL << page_level_shift(level);
1281 }
1282 static inline unsigned long page_level_mask(enum pg_level level)
1283 {
1284 	return ~(page_level_size(level) - 1);
1285 }
1286 
1287 /*
1288  * The x86 doesn't have any external MMU info: the kernel page
1289  * tables contain all the necessary information.
1290  */
1291 static inline void update_mmu_cache(struct vm_area_struct *vma,
1292 		unsigned long addr, pte_t *ptep)
1293 {
1294 }
1295 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1296 		unsigned long addr, pmd_t *pmd)
1297 {
1298 }
1299 static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
1300 		unsigned long addr, pud_t *pud)
1301 {
1302 }
1303 static inline pte_t pte_swp_mkexclusive(pte_t pte)
1304 {
1305 	return pte_set_flags(pte, _PAGE_SWP_EXCLUSIVE);
1306 }
1307 
1308 static inline int pte_swp_exclusive(pte_t pte)
1309 {
1310 	return pte_flags(pte) & _PAGE_SWP_EXCLUSIVE;
1311 }
1312 
1313 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
1314 {
1315 	return pte_clear_flags(pte, _PAGE_SWP_EXCLUSIVE);
1316 }
1317 
1318 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1319 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1320 {
1321 	return pte_set_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1322 }
1323 
1324 static inline int pte_swp_soft_dirty(pte_t pte)
1325 {
1326 	return pte_flags(pte) & _PAGE_SWP_SOFT_DIRTY;
1327 }
1328 
1329 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1330 {
1331 	return pte_clear_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1332 }
1333 
1334 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1335 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1336 {
1337 	return pmd_set_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1338 }
1339 
1340 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1341 {
1342 	return pmd_flags(pmd) & _PAGE_SWP_SOFT_DIRTY;
1343 }
1344 
1345 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1346 {
1347 	return pmd_clear_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1348 }
1349 #endif
1350 #endif
1351 
1352 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
1353 static inline pte_t pte_swp_mkuffd_wp(pte_t pte)
1354 {
1355 	return pte_set_flags(pte, _PAGE_SWP_UFFD_WP);
1356 }
1357 
1358 static inline int pte_swp_uffd_wp(pte_t pte)
1359 {
1360 	return pte_flags(pte) & _PAGE_SWP_UFFD_WP;
1361 }
1362 
1363 static inline pte_t pte_swp_clear_uffd_wp(pte_t pte)
1364 {
1365 	return pte_clear_flags(pte, _PAGE_SWP_UFFD_WP);
1366 }
1367 
1368 static inline pmd_t pmd_swp_mkuffd_wp(pmd_t pmd)
1369 {
1370 	return pmd_set_flags(pmd, _PAGE_SWP_UFFD_WP);
1371 }
1372 
1373 static inline int pmd_swp_uffd_wp(pmd_t pmd)
1374 {
1375 	return pmd_flags(pmd) & _PAGE_SWP_UFFD_WP;
1376 }
1377 
1378 static inline pmd_t pmd_swp_clear_uffd_wp(pmd_t pmd)
1379 {
1380 	return pmd_clear_flags(pmd, _PAGE_SWP_UFFD_WP);
1381 }
1382 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
1383 
1384 static inline u16 pte_flags_pkey(unsigned long pte_flags)
1385 {
1386 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1387 	/* ifdef to avoid doing 59-bit shift on 32-bit values */
1388 	return (pte_flags & _PAGE_PKEY_MASK) >> _PAGE_BIT_PKEY_BIT0;
1389 #else
1390 	return 0;
1391 #endif
1392 }
1393 
1394 static inline bool __pkru_allows_pkey(u16 pkey, bool write)
1395 {
1396 	u32 pkru = read_pkru();
1397 
1398 	if (!__pkru_allows_read(pkru, pkey))
1399 		return false;
1400 	if (write && !__pkru_allows_write(pkru, pkey))
1401 		return false;
1402 
1403 	return true;
1404 }
1405 
1406 /*
1407  * 'pteval' can come from a PTE, PMD or PUD.  We only check
1408  * _PAGE_PRESENT, _PAGE_USER, and _PAGE_RW in here which are the
1409  * same value on all 3 types.
1410  */
1411 static inline bool __pte_access_permitted(unsigned long pteval, bool write)
1412 {
1413 	unsigned long need_pte_bits = _PAGE_PRESENT|_PAGE_USER;
1414 
1415 	if (write)
1416 		need_pte_bits |= _PAGE_RW;
1417 
1418 	if ((pteval & need_pte_bits) != need_pte_bits)
1419 		return 0;
1420 
1421 	return __pkru_allows_pkey(pte_flags_pkey(pteval), write);
1422 }
1423 
1424 #define pte_access_permitted pte_access_permitted
1425 static inline bool pte_access_permitted(pte_t pte, bool write)
1426 {
1427 	return __pte_access_permitted(pte_val(pte), write);
1428 }
1429 
1430 #define pmd_access_permitted pmd_access_permitted
1431 static inline bool pmd_access_permitted(pmd_t pmd, bool write)
1432 {
1433 	return __pte_access_permitted(pmd_val(pmd), write);
1434 }
1435 
1436 #define pud_access_permitted pud_access_permitted
1437 static inline bool pud_access_permitted(pud_t pud, bool write)
1438 {
1439 	return __pte_access_permitted(pud_val(pud), write);
1440 }
1441 
1442 #define __HAVE_ARCH_PFN_MODIFY_ALLOWED 1
1443 extern bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot);
1444 
1445 static inline bool arch_has_pfn_modify_check(void)
1446 {
1447 	return boot_cpu_has_bug(X86_BUG_L1TF);
1448 }
1449 
1450 #define arch_has_hw_pte_young arch_has_hw_pte_young
1451 static inline bool arch_has_hw_pte_young(void)
1452 {
1453 	return true;
1454 }
1455 
1456 #ifdef CONFIG_XEN_PV
1457 #define arch_has_hw_nonleaf_pmd_young arch_has_hw_nonleaf_pmd_young
1458 static inline bool arch_has_hw_nonleaf_pmd_young(void)
1459 {
1460 	return !cpu_feature_enabled(X86_FEATURE_XENPV);
1461 }
1462 #endif
1463 
1464 #ifdef CONFIG_PAGE_TABLE_CHECK
1465 static inline bool pte_user_accessible_page(pte_t pte)
1466 {
1467 	return (pte_val(pte) & _PAGE_PRESENT) && (pte_val(pte) & _PAGE_USER);
1468 }
1469 
1470 static inline bool pmd_user_accessible_page(pmd_t pmd)
1471 {
1472 	return pmd_leaf(pmd) && (pmd_val(pmd) & _PAGE_PRESENT) && (pmd_val(pmd) & _PAGE_USER);
1473 }
1474 
1475 static inline bool pud_user_accessible_page(pud_t pud)
1476 {
1477 	return pud_leaf(pud) && (pud_val(pud) & _PAGE_PRESENT) && (pud_val(pud) & _PAGE_USER);
1478 }
1479 #endif
1480 
1481 #endif	/* __ASSEMBLY__ */
1482 
1483 #endif /* _ASM_X86_PGTABLE_H */
1484