xref: /openbmc/linux/arch/x86/include/asm/pgtable.h (revision 165f2d28)
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 <asm/x86_init.h>
26 #include <asm/fpu/xstate.h>
27 #include <asm/fpu/api.h>
28 #include <asm-generic/pgtable_uffd.h>
29 
30 extern pgd_t early_top_pgt[PTRS_PER_PGD];
31 int __init __early_make_pgtable(unsigned long address, pmdval_t pmd);
32 
33 void ptdump_walk_pgd_level(struct seq_file *m, struct mm_struct *mm);
34 void ptdump_walk_pgd_level_debugfs(struct seq_file *m, struct mm_struct *mm,
35 				   bool user);
36 void ptdump_walk_pgd_level_checkwx(void);
37 void ptdump_walk_user_pgd_level_checkwx(void);
38 
39 #ifdef CONFIG_DEBUG_WX
40 #define debug_checkwx()		ptdump_walk_pgd_level_checkwx()
41 #define debug_checkwx_user()	ptdump_walk_user_pgd_level_checkwx()
42 #else
43 #define debug_checkwx()		do { } while (0)
44 #define debug_checkwx_user()	do { } while (0)
45 #endif
46 
47 /*
48  * ZERO_PAGE is a global shared page that is always zero: used
49  * for zero-mapped memory areas etc..
50  */
51 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
52 	__visible;
53 #define ZERO_PAGE(vaddr) ((void)(vaddr),virt_to_page(empty_zero_page))
54 
55 extern spinlock_t pgd_lock;
56 extern struct list_head pgd_list;
57 
58 extern struct mm_struct *pgd_page_get_mm(struct page *page);
59 
60 extern pmdval_t early_pmd_flags;
61 
62 #ifdef CONFIG_PARAVIRT_XXL
63 #include <asm/paravirt.h>
64 #else  /* !CONFIG_PARAVIRT_XXL */
65 #define set_pte(ptep, pte)		native_set_pte(ptep, pte)
66 #define set_pte_at(mm, addr, ptep, pte)	native_set_pte_at(mm, addr, 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 
131 static inline u32 read_pkru(void)
132 {
133 	if (boot_cpu_has(X86_FEATURE_OSPKE))
134 		return rdpkru();
135 	return 0;
136 }
137 
138 static inline void write_pkru(u32 pkru)
139 {
140 	struct pkru_state *pk;
141 
142 	if (!boot_cpu_has(X86_FEATURE_OSPKE))
143 		return;
144 
145 	pk = get_xsave_addr(&current->thread.fpu.state.xsave, XFEATURE_PKRU);
146 
147 	/*
148 	 * The PKRU value in xstate needs to be in sync with the value that is
149 	 * written to the CPU. The FPU restore on return to userland would
150 	 * otherwise load the previous value again.
151 	 */
152 	fpregs_lock();
153 	if (pk)
154 		pk->pkru = pkru;
155 	__write_pkru(pkru);
156 	fpregs_unlock();
157 }
158 
159 static inline int pte_young(pte_t pte)
160 {
161 	return pte_flags(pte) & _PAGE_ACCESSED;
162 }
163 
164 static inline int pmd_dirty(pmd_t pmd)
165 {
166 	return pmd_flags(pmd) & _PAGE_DIRTY;
167 }
168 
169 static inline int pmd_young(pmd_t pmd)
170 {
171 	return pmd_flags(pmd) & _PAGE_ACCESSED;
172 }
173 
174 static inline int pud_dirty(pud_t pud)
175 {
176 	return pud_flags(pud) & _PAGE_DIRTY;
177 }
178 
179 static inline int pud_young(pud_t pud)
180 {
181 	return pud_flags(pud) & _PAGE_ACCESSED;
182 }
183 
184 static inline int pte_write(pte_t pte)
185 {
186 	return pte_flags(pte) & _PAGE_RW;
187 }
188 
189 static inline int pte_huge(pte_t pte)
190 {
191 	return pte_flags(pte) & _PAGE_PSE;
192 }
193 
194 static inline int pte_global(pte_t pte)
195 {
196 	return pte_flags(pte) & _PAGE_GLOBAL;
197 }
198 
199 static inline int pte_exec(pte_t pte)
200 {
201 	return !(pte_flags(pte) & _PAGE_NX);
202 }
203 
204 static inline int pte_special(pte_t pte)
205 {
206 	return pte_flags(pte) & _PAGE_SPECIAL;
207 }
208 
209 /* Entries that were set to PROT_NONE are inverted */
210 
211 static inline u64 protnone_mask(u64 val);
212 
213 static inline unsigned long pte_pfn(pte_t pte)
214 {
215 	phys_addr_t pfn = pte_val(pte);
216 	pfn ^= protnone_mask(pfn);
217 	return (pfn & PTE_PFN_MASK) >> PAGE_SHIFT;
218 }
219 
220 static inline unsigned long pmd_pfn(pmd_t pmd)
221 {
222 	phys_addr_t pfn = pmd_val(pmd);
223 	pfn ^= protnone_mask(pfn);
224 	return (pfn & pmd_pfn_mask(pmd)) >> PAGE_SHIFT;
225 }
226 
227 static inline unsigned long pud_pfn(pud_t pud)
228 {
229 	phys_addr_t pfn = pud_val(pud);
230 	pfn ^= protnone_mask(pfn);
231 	return (pfn & pud_pfn_mask(pud)) >> PAGE_SHIFT;
232 }
233 
234 static inline unsigned long p4d_pfn(p4d_t p4d)
235 {
236 	return (p4d_val(p4d) & p4d_pfn_mask(p4d)) >> PAGE_SHIFT;
237 }
238 
239 static inline unsigned long pgd_pfn(pgd_t pgd)
240 {
241 	return (pgd_val(pgd) & PTE_PFN_MASK) >> PAGE_SHIFT;
242 }
243 
244 #define p4d_leaf	p4d_large
245 static inline int p4d_large(p4d_t p4d)
246 {
247 	/* No 512 GiB pages yet */
248 	return 0;
249 }
250 
251 #define pte_page(pte)	pfn_to_page(pte_pfn(pte))
252 
253 #define pmd_leaf	pmd_large
254 static inline int pmd_large(pmd_t pte)
255 {
256 	return pmd_flags(pte) & _PAGE_PSE;
257 }
258 
259 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
260 static inline int pmd_trans_huge(pmd_t pmd)
261 {
262 	return (pmd_val(pmd) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
263 }
264 
265 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
266 static inline int pud_trans_huge(pud_t pud)
267 {
268 	return (pud_val(pud) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
269 }
270 #endif
271 
272 #define has_transparent_hugepage has_transparent_hugepage
273 static inline int has_transparent_hugepage(void)
274 {
275 	return boot_cpu_has(X86_FEATURE_PSE);
276 }
277 
278 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
279 static inline int pmd_devmap(pmd_t pmd)
280 {
281 	return !!(pmd_val(pmd) & _PAGE_DEVMAP);
282 }
283 
284 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
285 static inline int pud_devmap(pud_t pud)
286 {
287 	return !!(pud_val(pud) & _PAGE_DEVMAP);
288 }
289 #else
290 static inline int pud_devmap(pud_t pud)
291 {
292 	return 0;
293 }
294 #endif
295 
296 static inline int pgd_devmap(pgd_t pgd)
297 {
298 	return 0;
299 }
300 #endif
301 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
302 
303 static inline pte_t pte_set_flags(pte_t pte, pteval_t set)
304 {
305 	pteval_t v = native_pte_val(pte);
306 
307 	return native_make_pte(v | set);
308 }
309 
310 static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear)
311 {
312 	pteval_t v = native_pte_val(pte);
313 
314 	return native_make_pte(v & ~clear);
315 }
316 
317 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
318 static inline int pte_uffd_wp(pte_t pte)
319 {
320 	return pte_flags(pte) & _PAGE_UFFD_WP;
321 }
322 
323 static inline pte_t pte_mkuffd_wp(pte_t pte)
324 {
325 	return pte_set_flags(pte, _PAGE_UFFD_WP);
326 }
327 
328 static inline pte_t pte_clear_uffd_wp(pte_t pte)
329 {
330 	return pte_clear_flags(pte, _PAGE_UFFD_WP);
331 }
332 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
333 
334 static inline pte_t pte_mkclean(pte_t pte)
335 {
336 	return pte_clear_flags(pte, _PAGE_DIRTY);
337 }
338 
339 static inline pte_t pte_mkold(pte_t pte)
340 {
341 	return pte_clear_flags(pte, _PAGE_ACCESSED);
342 }
343 
344 static inline pte_t pte_wrprotect(pte_t pte)
345 {
346 	return pte_clear_flags(pte, _PAGE_RW);
347 }
348 
349 static inline pte_t pte_mkexec(pte_t pte)
350 {
351 	return pte_clear_flags(pte, _PAGE_NX);
352 }
353 
354 static inline pte_t pte_mkdirty(pte_t pte)
355 {
356 	return pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
357 }
358 
359 static inline pte_t pte_mkyoung(pte_t pte)
360 {
361 	return pte_set_flags(pte, _PAGE_ACCESSED);
362 }
363 
364 static inline pte_t pte_mkwrite(pte_t pte)
365 {
366 	return pte_set_flags(pte, _PAGE_RW);
367 }
368 
369 static inline pte_t pte_mkhuge(pte_t pte)
370 {
371 	return pte_set_flags(pte, _PAGE_PSE);
372 }
373 
374 static inline pte_t pte_clrhuge(pte_t pte)
375 {
376 	return pte_clear_flags(pte, _PAGE_PSE);
377 }
378 
379 static inline pte_t pte_mkglobal(pte_t pte)
380 {
381 	return pte_set_flags(pte, _PAGE_GLOBAL);
382 }
383 
384 static inline pte_t pte_clrglobal(pte_t pte)
385 {
386 	return pte_clear_flags(pte, _PAGE_GLOBAL);
387 }
388 
389 static inline pte_t pte_mkspecial(pte_t pte)
390 {
391 	return pte_set_flags(pte, _PAGE_SPECIAL);
392 }
393 
394 static inline pte_t pte_mkdevmap(pte_t pte)
395 {
396 	return pte_set_flags(pte, _PAGE_SPECIAL|_PAGE_DEVMAP);
397 }
398 
399 static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
400 {
401 	pmdval_t v = native_pmd_val(pmd);
402 
403 	return native_make_pmd(v | set);
404 }
405 
406 static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
407 {
408 	pmdval_t v = native_pmd_val(pmd);
409 
410 	return native_make_pmd(v & ~clear);
411 }
412 
413 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
414 static inline int pmd_uffd_wp(pmd_t pmd)
415 {
416 	return pmd_flags(pmd) & _PAGE_UFFD_WP;
417 }
418 
419 static inline pmd_t pmd_mkuffd_wp(pmd_t pmd)
420 {
421 	return pmd_set_flags(pmd, _PAGE_UFFD_WP);
422 }
423 
424 static inline pmd_t pmd_clear_uffd_wp(pmd_t pmd)
425 {
426 	return pmd_clear_flags(pmd, _PAGE_UFFD_WP);
427 }
428 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
429 
430 static inline pmd_t pmd_mkold(pmd_t pmd)
431 {
432 	return pmd_clear_flags(pmd, _PAGE_ACCESSED);
433 }
434 
435 static inline pmd_t pmd_mkclean(pmd_t pmd)
436 {
437 	return pmd_clear_flags(pmd, _PAGE_DIRTY);
438 }
439 
440 static inline pmd_t pmd_wrprotect(pmd_t pmd)
441 {
442 	return pmd_clear_flags(pmd, _PAGE_RW);
443 }
444 
445 static inline pmd_t pmd_mkdirty(pmd_t pmd)
446 {
447 	return pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
448 }
449 
450 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
451 {
452 	return pmd_set_flags(pmd, _PAGE_DEVMAP);
453 }
454 
455 static inline pmd_t pmd_mkhuge(pmd_t pmd)
456 {
457 	return pmd_set_flags(pmd, _PAGE_PSE);
458 }
459 
460 static inline pmd_t pmd_mkyoung(pmd_t pmd)
461 {
462 	return pmd_set_flags(pmd, _PAGE_ACCESSED);
463 }
464 
465 static inline pmd_t pmd_mkwrite(pmd_t pmd)
466 {
467 	return pmd_set_flags(pmd, _PAGE_RW);
468 }
469 
470 static inline pud_t pud_set_flags(pud_t pud, pudval_t set)
471 {
472 	pudval_t v = native_pud_val(pud);
473 
474 	return native_make_pud(v | set);
475 }
476 
477 static inline pud_t pud_clear_flags(pud_t pud, pudval_t clear)
478 {
479 	pudval_t v = native_pud_val(pud);
480 
481 	return native_make_pud(v & ~clear);
482 }
483 
484 static inline pud_t pud_mkold(pud_t pud)
485 {
486 	return pud_clear_flags(pud, _PAGE_ACCESSED);
487 }
488 
489 static inline pud_t pud_mkclean(pud_t pud)
490 {
491 	return pud_clear_flags(pud, _PAGE_DIRTY);
492 }
493 
494 static inline pud_t pud_wrprotect(pud_t pud)
495 {
496 	return pud_clear_flags(pud, _PAGE_RW);
497 }
498 
499 static inline pud_t pud_mkdirty(pud_t pud)
500 {
501 	return pud_set_flags(pud, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
502 }
503 
504 static inline pud_t pud_mkdevmap(pud_t pud)
505 {
506 	return pud_set_flags(pud, _PAGE_DEVMAP);
507 }
508 
509 static inline pud_t pud_mkhuge(pud_t pud)
510 {
511 	return pud_set_flags(pud, _PAGE_PSE);
512 }
513 
514 static inline pud_t pud_mkyoung(pud_t pud)
515 {
516 	return pud_set_flags(pud, _PAGE_ACCESSED);
517 }
518 
519 static inline pud_t pud_mkwrite(pud_t pud)
520 {
521 	return pud_set_flags(pud, _PAGE_RW);
522 }
523 
524 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
525 static inline int pte_soft_dirty(pte_t pte)
526 {
527 	return pte_flags(pte) & _PAGE_SOFT_DIRTY;
528 }
529 
530 static inline int pmd_soft_dirty(pmd_t pmd)
531 {
532 	return pmd_flags(pmd) & _PAGE_SOFT_DIRTY;
533 }
534 
535 static inline int pud_soft_dirty(pud_t pud)
536 {
537 	return pud_flags(pud) & _PAGE_SOFT_DIRTY;
538 }
539 
540 static inline pte_t pte_mksoft_dirty(pte_t pte)
541 {
542 	return pte_set_flags(pte, _PAGE_SOFT_DIRTY);
543 }
544 
545 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
546 {
547 	return pmd_set_flags(pmd, _PAGE_SOFT_DIRTY);
548 }
549 
550 static inline pud_t pud_mksoft_dirty(pud_t pud)
551 {
552 	return pud_set_flags(pud, _PAGE_SOFT_DIRTY);
553 }
554 
555 static inline pte_t pte_clear_soft_dirty(pte_t pte)
556 {
557 	return pte_clear_flags(pte, _PAGE_SOFT_DIRTY);
558 }
559 
560 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
561 {
562 	return pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY);
563 }
564 
565 static inline pud_t pud_clear_soft_dirty(pud_t pud)
566 {
567 	return pud_clear_flags(pud, _PAGE_SOFT_DIRTY);
568 }
569 
570 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
571 
572 /*
573  * Mask out unsupported bits in a present pgprot.  Non-present pgprots
574  * can use those bits for other purposes, so leave them be.
575  */
576 static inline pgprotval_t massage_pgprot(pgprot_t pgprot)
577 {
578 	pgprotval_t protval = pgprot_val(pgprot);
579 
580 	if (protval & _PAGE_PRESENT)
581 		protval &= __supported_pte_mask;
582 
583 	return protval;
584 }
585 
586 static inline pgprotval_t check_pgprot(pgprot_t pgprot)
587 {
588 	pgprotval_t massaged_val = massage_pgprot(pgprot);
589 
590 	/* mmdebug.h can not be included here because of dependencies */
591 #ifdef CONFIG_DEBUG_VM
592 	WARN_ONCE(pgprot_val(pgprot) != massaged_val,
593 		  "attempted to set unsupported pgprot: %016llx "
594 		  "bits: %016llx supported: %016llx\n",
595 		  (u64)pgprot_val(pgprot),
596 		  (u64)pgprot_val(pgprot) ^ massaged_val,
597 		  (u64)__supported_pte_mask);
598 #endif
599 
600 	return massaged_val;
601 }
602 
603 static inline pte_t pfn_pte(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 &= PTE_PFN_MASK;
608 	return __pte(pfn | check_pgprot(pgprot));
609 }
610 
611 static inline pmd_t pfn_pmd(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_PMD_PAGE_MASK;
616 	return __pmd(pfn | check_pgprot(pgprot));
617 }
618 
619 static inline pud_t pfn_pud(unsigned long page_nr, pgprot_t pgprot)
620 {
621 	phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
622 	pfn ^= protnone_mask(pgprot_val(pgprot));
623 	pfn &= PHYSICAL_PUD_PAGE_MASK;
624 	return __pud(pfn | check_pgprot(pgprot));
625 }
626 
627 static inline pmd_t pmd_mknotpresent(pmd_t pmd)
628 {
629 	return pfn_pmd(pmd_pfn(pmd),
630 		      __pgprot(pmd_flags(pmd) & ~(_PAGE_PRESENT|_PAGE_PROTNONE)));
631 }
632 
633 static inline u64 flip_protnone_guard(u64 oldval, u64 val, u64 mask);
634 
635 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
636 {
637 	pteval_t val = pte_val(pte), oldval = val;
638 
639 	/*
640 	 * Chop off the NX bit (if present), and add the NX portion of
641 	 * the newprot (if present):
642 	 */
643 	val &= _PAGE_CHG_MASK;
644 	val |= check_pgprot(newprot) & ~_PAGE_CHG_MASK;
645 	val = flip_protnone_guard(oldval, val, PTE_PFN_MASK);
646 	return __pte(val);
647 }
648 
649 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
650 {
651 	pmdval_t val = pmd_val(pmd), oldval = val;
652 
653 	val &= _HPAGE_CHG_MASK;
654 	val |= check_pgprot(newprot) & ~_HPAGE_CHG_MASK;
655 	val = flip_protnone_guard(oldval, val, PHYSICAL_PMD_PAGE_MASK);
656 	return __pmd(val);
657 }
658 
659 /*
660  * mprotect needs to preserve PAT and encryption bits when updating
661  * vm_page_prot
662  */
663 #define pgprot_modify pgprot_modify
664 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
665 {
666 	pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK;
667 	pgprotval_t addbits = pgprot_val(newprot) & ~_PAGE_CHG_MASK;
668 	return __pgprot(preservebits | addbits);
669 }
670 
671 #define pte_pgprot(x) __pgprot(pte_flags(x))
672 #define pmd_pgprot(x) __pgprot(pmd_flags(x))
673 #define pud_pgprot(x) __pgprot(pud_flags(x))
674 #define p4d_pgprot(x) __pgprot(p4d_flags(x))
675 
676 #define canon_pgprot(p) __pgprot(massage_pgprot(p))
677 
678 static inline pgprot_t arch_filter_pgprot(pgprot_t prot)
679 {
680 	return canon_pgprot(prot);
681 }
682 
683 static inline int is_new_memtype_allowed(u64 paddr, unsigned long size,
684 					 enum page_cache_mode pcm,
685 					 enum page_cache_mode new_pcm)
686 {
687 	/*
688 	 * PAT type is always WB for untracked ranges, so no need to check.
689 	 */
690 	if (x86_platform.is_untracked_pat_range(paddr, paddr + size))
691 		return 1;
692 
693 	/*
694 	 * Certain new memtypes are not allowed with certain
695 	 * requested memtype:
696 	 * - request is uncached, return cannot be write-back
697 	 * - request is write-combine, return cannot be write-back
698 	 * - request is write-through, return cannot be write-back
699 	 * - request is write-through, return cannot be write-combine
700 	 */
701 	if ((pcm == _PAGE_CACHE_MODE_UC_MINUS &&
702 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
703 	    (pcm == _PAGE_CACHE_MODE_WC &&
704 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
705 	    (pcm == _PAGE_CACHE_MODE_WT &&
706 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
707 	    (pcm == _PAGE_CACHE_MODE_WT &&
708 	     new_pcm == _PAGE_CACHE_MODE_WC)) {
709 		return 0;
710 	}
711 
712 	return 1;
713 }
714 
715 pmd_t *populate_extra_pmd(unsigned long vaddr);
716 pte_t *populate_extra_pte(unsigned long vaddr);
717 
718 #ifdef CONFIG_PAGE_TABLE_ISOLATION
719 pgd_t __pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd);
720 
721 /*
722  * Take a PGD location (pgdp) and a pgd value that needs to be set there.
723  * Populates the user and returns the resulting PGD that must be set in
724  * the kernel copy of the page tables.
725  */
726 static inline pgd_t pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd)
727 {
728 	if (!static_cpu_has(X86_FEATURE_PTI))
729 		return pgd;
730 	return __pti_set_user_pgtbl(pgdp, pgd);
731 }
732 #else   /* CONFIG_PAGE_TABLE_ISOLATION */
733 static inline pgd_t pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd)
734 {
735 	return pgd;
736 }
737 #endif  /* CONFIG_PAGE_TABLE_ISOLATION */
738 
739 #endif	/* __ASSEMBLY__ */
740 
741 
742 #ifdef CONFIG_X86_32
743 # include <asm/pgtable_32.h>
744 #else
745 # include <asm/pgtable_64.h>
746 #endif
747 
748 #ifndef __ASSEMBLY__
749 #include <linux/mm_types.h>
750 #include <linux/mmdebug.h>
751 #include <linux/log2.h>
752 #include <asm/fixmap.h>
753 
754 static inline int pte_none(pte_t pte)
755 {
756 	return !(pte.pte & ~(_PAGE_KNL_ERRATUM_MASK));
757 }
758 
759 #define __HAVE_ARCH_PTE_SAME
760 static inline int pte_same(pte_t a, pte_t b)
761 {
762 	return a.pte == b.pte;
763 }
764 
765 static inline int pte_present(pte_t a)
766 {
767 	return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE);
768 }
769 
770 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
771 static inline int pte_devmap(pte_t a)
772 {
773 	return (pte_flags(a) & _PAGE_DEVMAP) == _PAGE_DEVMAP;
774 }
775 #endif
776 
777 #define pte_accessible pte_accessible
778 static inline bool pte_accessible(struct mm_struct *mm, pte_t a)
779 {
780 	if (pte_flags(a) & _PAGE_PRESENT)
781 		return true;
782 
783 	if ((pte_flags(a) & _PAGE_PROTNONE) &&
784 			mm_tlb_flush_pending(mm))
785 		return true;
786 
787 	return false;
788 }
789 
790 static inline int pmd_present(pmd_t pmd)
791 {
792 	/*
793 	 * Checking for _PAGE_PSE is needed too because
794 	 * split_huge_page will temporarily clear the present bit (but
795 	 * the _PAGE_PSE flag will remain set at all times while the
796 	 * _PAGE_PRESENT bit is clear).
797 	 */
798 	return pmd_flags(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PSE);
799 }
800 
801 #ifdef CONFIG_NUMA_BALANCING
802 /*
803  * These work without NUMA balancing but the kernel does not care. See the
804  * comment in include/asm-generic/pgtable.h
805  */
806 static inline int pte_protnone(pte_t pte)
807 {
808 	return (pte_flags(pte) & (_PAGE_PROTNONE | _PAGE_PRESENT))
809 		== _PAGE_PROTNONE;
810 }
811 
812 static inline int pmd_protnone(pmd_t pmd)
813 {
814 	return (pmd_flags(pmd) & (_PAGE_PROTNONE | _PAGE_PRESENT))
815 		== _PAGE_PROTNONE;
816 }
817 #endif /* CONFIG_NUMA_BALANCING */
818 
819 static inline int pmd_none(pmd_t pmd)
820 {
821 	/* Only check low word on 32-bit platforms, since it might be
822 	   out of sync with upper half. */
823 	unsigned long val = native_pmd_val(pmd);
824 	return (val & ~_PAGE_KNL_ERRATUM_MASK) == 0;
825 }
826 
827 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
828 {
829 	return (unsigned long)__va(pmd_val(pmd) & pmd_pfn_mask(pmd));
830 }
831 
832 /*
833  * Currently stuck as a macro due to indirect forward reference to
834  * linux/mmzone.h's __section_mem_map_addr() definition:
835  */
836 #define pmd_page(pmd)	pfn_to_page(pmd_pfn(pmd))
837 
838 /*
839  * the pmd page can be thought of an array like this: pmd_t[PTRS_PER_PMD]
840  *
841  * this macro returns the index of the entry in the pmd page which would
842  * control the given virtual address
843  */
844 static inline unsigned long pmd_index(unsigned long address)
845 {
846 	return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
847 }
848 
849 /*
850  * Conversion functions: convert a page and protection to a page entry,
851  * and a page entry and page directory to the page they refer to.
852  *
853  * (Currently stuck as a macro because of indirect forward reference
854  * to linux/mm.h:page_to_nid())
855  */
856 #define mk_pte(page, pgprot)   pfn_pte(page_to_pfn(page), (pgprot))
857 
858 /*
859  * the pte page can be thought of an array like this: pte_t[PTRS_PER_PTE]
860  *
861  * this function returns the index of the entry in the pte page which would
862  * control the given virtual address
863  *
864  * Also define macro so we can test if pte_index is defined for arch.
865  */
866 #define pte_index pte_index
867 static inline unsigned long pte_index(unsigned long address)
868 {
869 	return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
870 }
871 
872 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
873 {
874 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
875 }
876 
877 static inline int pmd_bad(pmd_t pmd)
878 {
879 	return (pmd_flags(pmd) & ~_PAGE_USER) != _KERNPG_TABLE;
880 }
881 
882 static inline unsigned long pages_to_mb(unsigned long npg)
883 {
884 	return npg >> (20 - PAGE_SHIFT);
885 }
886 
887 #if CONFIG_PGTABLE_LEVELS > 2
888 static inline int pud_none(pud_t pud)
889 {
890 	return (native_pud_val(pud) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
891 }
892 
893 static inline int pud_present(pud_t pud)
894 {
895 	return pud_flags(pud) & _PAGE_PRESENT;
896 }
897 
898 static inline unsigned long pud_page_vaddr(pud_t pud)
899 {
900 	return (unsigned long)__va(pud_val(pud) & pud_pfn_mask(pud));
901 }
902 
903 /*
904  * Currently stuck as a macro due to indirect forward reference to
905  * linux/mmzone.h's __section_mem_map_addr() definition:
906  */
907 #define pud_page(pud)	pfn_to_page(pud_pfn(pud))
908 
909 /* Find an entry in the second-level page table.. */
910 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
911 {
912 	return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(address);
913 }
914 
915 #define pud_leaf	pud_large
916 static inline int pud_large(pud_t pud)
917 {
918 	return (pud_val(pud) & (_PAGE_PSE | _PAGE_PRESENT)) ==
919 		(_PAGE_PSE | _PAGE_PRESENT);
920 }
921 
922 static inline int pud_bad(pud_t pud)
923 {
924 	return (pud_flags(pud) & ~(_KERNPG_TABLE | _PAGE_USER)) != 0;
925 }
926 #else
927 #define pud_leaf	pud_large
928 static inline int pud_large(pud_t pud)
929 {
930 	return 0;
931 }
932 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
933 
934 static inline unsigned long pud_index(unsigned long address)
935 {
936 	return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
937 }
938 
939 #if CONFIG_PGTABLE_LEVELS > 3
940 static inline int p4d_none(p4d_t p4d)
941 {
942 	return (native_p4d_val(p4d) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
943 }
944 
945 static inline int p4d_present(p4d_t p4d)
946 {
947 	return p4d_flags(p4d) & _PAGE_PRESENT;
948 }
949 
950 static inline unsigned long p4d_page_vaddr(p4d_t p4d)
951 {
952 	return (unsigned long)__va(p4d_val(p4d) & p4d_pfn_mask(p4d));
953 }
954 
955 /*
956  * Currently stuck as a macro due to indirect forward reference to
957  * linux/mmzone.h's __section_mem_map_addr() definition:
958  */
959 #define p4d_page(p4d)	pfn_to_page(p4d_pfn(p4d))
960 
961 /* Find an entry in the third-level page table.. */
962 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
963 {
964 	return (pud_t *)p4d_page_vaddr(*p4d) + pud_index(address);
965 }
966 
967 static inline int p4d_bad(p4d_t p4d)
968 {
969 	unsigned long ignore_flags = _KERNPG_TABLE | _PAGE_USER;
970 
971 	if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
972 		ignore_flags |= _PAGE_NX;
973 
974 	return (p4d_flags(p4d) & ~ignore_flags) != 0;
975 }
976 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
977 
978 static inline unsigned long p4d_index(unsigned long address)
979 {
980 	return (address >> P4D_SHIFT) & (PTRS_PER_P4D - 1);
981 }
982 
983 #if CONFIG_PGTABLE_LEVELS > 4
984 static inline int pgd_present(pgd_t pgd)
985 {
986 	if (!pgtable_l5_enabled())
987 		return 1;
988 	return pgd_flags(pgd) & _PAGE_PRESENT;
989 }
990 
991 static inline unsigned long pgd_page_vaddr(pgd_t pgd)
992 {
993 	return (unsigned long)__va((unsigned long)pgd_val(pgd) & PTE_PFN_MASK);
994 }
995 
996 /*
997  * Currently stuck as a macro due to indirect forward reference to
998  * linux/mmzone.h's __section_mem_map_addr() definition:
999  */
1000 #define pgd_page(pgd)	pfn_to_page(pgd_pfn(pgd))
1001 
1002 /* to find an entry in a page-table-directory. */
1003 static inline p4d_t *p4d_offset(pgd_t *pgd, unsigned long address)
1004 {
1005 	if (!pgtable_l5_enabled())
1006 		return (p4d_t *)pgd;
1007 	return (p4d_t *)pgd_page_vaddr(*pgd) + p4d_index(address);
1008 }
1009 
1010 static inline int pgd_bad(pgd_t pgd)
1011 {
1012 	unsigned long ignore_flags = _PAGE_USER;
1013 
1014 	if (!pgtable_l5_enabled())
1015 		return 0;
1016 
1017 	if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
1018 		ignore_flags |= _PAGE_NX;
1019 
1020 	return (pgd_flags(pgd) & ~ignore_flags) != _KERNPG_TABLE;
1021 }
1022 
1023 static inline int pgd_none(pgd_t pgd)
1024 {
1025 	if (!pgtable_l5_enabled())
1026 		return 0;
1027 	/*
1028 	 * There is no need to do a workaround for the KNL stray
1029 	 * A/D bit erratum here.  PGDs only point to page tables
1030 	 * except on 32-bit non-PAE which is not supported on
1031 	 * KNL.
1032 	 */
1033 	return !native_pgd_val(pgd);
1034 }
1035 #endif	/* CONFIG_PGTABLE_LEVELS > 4 */
1036 
1037 #endif	/* __ASSEMBLY__ */
1038 
1039 /*
1040  * the pgd page can be thought of an array like this: pgd_t[PTRS_PER_PGD]
1041  *
1042  * this macro returns the index of the entry in the pgd page which would
1043  * control the given virtual address
1044  */
1045 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
1046 
1047 /*
1048  * pgd_offset() returns a (pgd_t *)
1049  * pgd_index() is used get the offset into the pgd page's array of pgd_t's;
1050  */
1051 #define pgd_offset_pgd(pgd, address) (pgd + pgd_index((address)))
1052 /*
1053  * a shortcut to get a pgd_t in a given mm
1054  */
1055 #define pgd_offset(mm, address) pgd_offset_pgd((mm)->pgd, (address))
1056 /*
1057  * a shortcut which implies the use of the kernel's pgd, instead
1058  * of a process's
1059  */
1060 #define pgd_offset_k(address) pgd_offset(&init_mm, (address))
1061 
1062 
1063 #define KERNEL_PGD_BOUNDARY	pgd_index(PAGE_OFFSET)
1064 #define KERNEL_PGD_PTRS		(PTRS_PER_PGD - KERNEL_PGD_BOUNDARY)
1065 
1066 #ifndef __ASSEMBLY__
1067 
1068 extern int direct_gbpages;
1069 void init_mem_mapping(void);
1070 void early_alloc_pgt_buf(void);
1071 extern void memblock_find_dma_reserve(void);
1072 
1073 #ifdef CONFIG_X86_64
1074 /* Realmode trampoline initialization. */
1075 extern pgd_t trampoline_pgd_entry;
1076 static inline void __meminit init_trampoline_default(void)
1077 {
1078 	/* Default trampoline pgd value */
1079 	trampoline_pgd_entry = init_top_pgt[pgd_index(__PAGE_OFFSET)];
1080 }
1081 
1082 void __init poking_init(void);
1083 
1084 unsigned long init_memory_mapping(unsigned long start,
1085 				  unsigned long end, pgprot_t prot);
1086 
1087 # ifdef CONFIG_RANDOMIZE_MEMORY
1088 void __meminit init_trampoline(void);
1089 # else
1090 #  define init_trampoline init_trampoline_default
1091 # endif
1092 #else
1093 static inline void init_trampoline(void) { }
1094 #endif
1095 
1096 /* local pte updates need not use xchg for locking */
1097 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
1098 {
1099 	pte_t res = *ptep;
1100 
1101 	/* Pure native function needs no input for mm, addr */
1102 	native_pte_clear(NULL, 0, ptep);
1103 	return res;
1104 }
1105 
1106 static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp)
1107 {
1108 	pmd_t res = *pmdp;
1109 
1110 	native_pmd_clear(pmdp);
1111 	return res;
1112 }
1113 
1114 static inline pud_t native_local_pudp_get_and_clear(pud_t *pudp)
1115 {
1116 	pud_t res = *pudp;
1117 
1118 	native_pud_clear(pudp);
1119 	return res;
1120 }
1121 
1122 static inline void native_set_pte_at(struct mm_struct *mm, unsigned long addr,
1123 				     pte_t *ptep , pte_t pte)
1124 {
1125 	native_set_pte(ptep, pte);
1126 }
1127 
1128 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1129 			      pmd_t *pmdp, pmd_t pmd)
1130 {
1131 	set_pmd(pmdp, pmd);
1132 }
1133 
1134 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
1135 			      pud_t *pudp, pud_t pud)
1136 {
1137 	native_set_pud(pudp, pud);
1138 }
1139 
1140 /*
1141  * We only update the dirty/accessed state if we set
1142  * the dirty bit by hand in the kernel, since the hardware
1143  * will do the accessed bit for us, and we don't want to
1144  * race with other CPU's that might be updating the dirty
1145  * bit at the same time.
1146  */
1147 struct vm_area_struct;
1148 
1149 #define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1150 extern int ptep_set_access_flags(struct vm_area_struct *vma,
1151 				 unsigned long address, pte_t *ptep,
1152 				 pte_t entry, int dirty);
1153 
1154 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1155 extern int ptep_test_and_clear_young(struct vm_area_struct *vma,
1156 				     unsigned long addr, pte_t *ptep);
1157 
1158 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1159 extern int ptep_clear_flush_young(struct vm_area_struct *vma,
1160 				  unsigned long address, pte_t *ptep);
1161 
1162 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1163 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
1164 				       pte_t *ptep)
1165 {
1166 	pte_t pte = native_ptep_get_and_clear(ptep);
1167 	return pte;
1168 }
1169 
1170 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1171 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1172 					    unsigned long addr, pte_t *ptep,
1173 					    int full)
1174 {
1175 	pte_t pte;
1176 	if (full) {
1177 		/*
1178 		 * Full address destruction in progress; paravirt does not
1179 		 * care about updates and native needs no locking
1180 		 */
1181 		pte = native_local_ptep_get_and_clear(ptep);
1182 	} else {
1183 		pte = ptep_get_and_clear(mm, addr, ptep);
1184 	}
1185 	return pte;
1186 }
1187 
1188 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1189 static inline void ptep_set_wrprotect(struct mm_struct *mm,
1190 				      unsigned long addr, pte_t *ptep)
1191 {
1192 	clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte);
1193 }
1194 
1195 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
1196 
1197 #define mk_pmd(page, pgprot)   pfn_pmd(page_to_pfn(page), (pgprot))
1198 
1199 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1200 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
1201 				 unsigned long address, pmd_t *pmdp,
1202 				 pmd_t entry, int dirty);
1203 extern int pudp_set_access_flags(struct vm_area_struct *vma,
1204 				 unsigned long address, pud_t *pudp,
1205 				 pud_t entry, int dirty);
1206 
1207 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1208 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1209 				     unsigned long addr, pmd_t *pmdp);
1210 extern int pudp_test_and_clear_young(struct vm_area_struct *vma,
1211 				     unsigned long addr, pud_t *pudp);
1212 
1213 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
1214 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
1215 				  unsigned long address, pmd_t *pmdp);
1216 
1217 
1218 #define pmd_write pmd_write
1219 static inline int pmd_write(pmd_t pmd)
1220 {
1221 	return pmd_flags(pmd) & _PAGE_RW;
1222 }
1223 
1224 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1225 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, unsigned long addr,
1226 				       pmd_t *pmdp)
1227 {
1228 	return native_pmdp_get_and_clear(pmdp);
1229 }
1230 
1231 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
1232 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
1233 					unsigned long addr, pud_t *pudp)
1234 {
1235 	return native_pudp_get_and_clear(pudp);
1236 }
1237 
1238 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1239 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1240 				      unsigned long addr, pmd_t *pmdp)
1241 {
1242 	clear_bit(_PAGE_BIT_RW, (unsigned long *)pmdp);
1243 }
1244 
1245 #define pud_write pud_write
1246 static inline int pud_write(pud_t pud)
1247 {
1248 	return pud_flags(pud) & _PAGE_RW;
1249 }
1250 
1251 #ifndef pmdp_establish
1252 #define pmdp_establish pmdp_establish
1253 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
1254 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
1255 {
1256 	if (IS_ENABLED(CONFIG_SMP)) {
1257 		return xchg(pmdp, pmd);
1258 	} else {
1259 		pmd_t old = *pmdp;
1260 		WRITE_ONCE(*pmdp, pmd);
1261 		return old;
1262 	}
1263 }
1264 #endif
1265 /*
1266  * Page table pages are page-aligned.  The lower half of the top
1267  * level is used for userspace and the top half for the kernel.
1268  *
1269  * Returns true for parts of the PGD that map userspace and
1270  * false for the parts that map the kernel.
1271  */
1272 static inline bool pgdp_maps_userspace(void *__ptr)
1273 {
1274 	unsigned long ptr = (unsigned long)__ptr;
1275 
1276 	return (((ptr & ~PAGE_MASK) / sizeof(pgd_t)) < PGD_KERNEL_START);
1277 }
1278 
1279 #define pgd_leaf	pgd_large
1280 static inline int pgd_large(pgd_t pgd) { return 0; }
1281 
1282 #ifdef CONFIG_PAGE_TABLE_ISOLATION
1283 /*
1284  * All top-level PAGE_TABLE_ISOLATION page tables are order-1 pages
1285  * (8k-aligned and 8k in size).  The kernel one is at the beginning 4k and
1286  * the user one is in the last 4k.  To switch between them, you
1287  * just need to flip the 12th bit in their addresses.
1288  */
1289 #define PTI_PGTABLE_SWITCH_BIT	PAGE_SHIFT
1290 
1291 /*
1292  * This generates better code than the inline assembly in
1293  * __set_bit().
1294  */
1295 static inline void *ptr_set_bit(void *ptr, int bit)
1296 {
1297 	unsigned long __ptr = (unsigned long)ptr;
1298 
1299 	__ptr |= BIT(bit);
1300 	return (void *)__ptr;
1301 }
1302 static inline void *ptr_clear_bit(void *ptr, int bit)
1303 {
1304 	unsigned long __ptr = (unsigned long)ptr;
1305 
1306 	__ptr &= ~BIT(bit);
1307 	return (void *)__ptr;
1308 }
1309 
1310 static inline pgd_t *kernel_to_user_pgdp(pgd_t *pgdp)
1311 {
1312 	return ptr_set_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
1313 }
1314 
1315 static inline pgd_t *user_to_kernel_pgdp(pgd_t *pgdp)
1316 {
1317 	return ptr_clear_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
1318 }
1319 
1320 static inline p4d_t *kernel_to_user_p4dp(p4d_t *p4dp)
1321 {
1322 	return ptr_set_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
1323 }
1324 
1325 static inline p4d_t *user_to_kernel_p4dp(p4d_t *p4dp)
1326 {
1327 	return ptr_clear_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
1328 }
1329 #endif /* CONFIG_PAGE_TABLE_ISOLATION */
1330 
1331 /*
1332  * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
1333  *
1334  *  dst - pointer to pgd range anwhere on a pgd page
1335  *  src - ""
1336  *  count - the number of pgds to copy.
1337  *
1338  * dst and src can be on the same page, but the range must not overlap,
1339  * and must not cross a page boundary.
1340  */
1341 static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
1342 {
1343 	memcpy(dst, src, count * sizeof(pgd_t));
1344 #ifdef CONFIG_PAGE_TABLE_ISOLATION
1345 	if (!static_cpu_has(X86_FEATURE_PTI))
1346 		return;
1347 	/* Clone the user space pgd as well */
1348 	memcpy(kernel_to_user_pgdp(dst), kernel_to_user_pgdp(src),
1349 	       count * sizeof(pgd_t));
1350 #endif
1351 }
1352 
1353 #define PTE_SHIFT ilog2(PTRS_PER_PTE)
1354 static inline int page_level_shift(enum pg_level level)
1355 {
1356 	return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT;
1357 }
1358 static inline unsigned long page_level_size(enum pg_level level)
1359 {
1360 	return 1UL << page_level_shift(level);
1361 }
1362 static inline unsigned long page_level_mask(enum pg_level level)
1363 {
1364 	return ~(page_level_size(level) - 1);
1365 }
1366 
1367 /*
1368  * The x86 doesn't have any external MMU info: the kernel page
1369  * tables contain all the necessary information.
1370  */
1371 static inline void update_mmu_cache(struct vm_area_struct *vma,
1372 		unsigned long addr, pte_t *ptep)
1373 {
1374 }
1375 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1376 		unsigned long addr, pmd_t *pmd)
1377 {
1378 }
1379 static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
1380 		unsigned long addr, pud_t *pud)
1381 {
1382 }
1383 
1384 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1385 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1386 {
1387 	return pte_set_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1388 }
1389 
1390 static inline int pte_swp_soft_dirty(pte_t pte)
1391 {
1392 	return pte_flags(pte) & _PAGE_SWP_SOFT_DIRTY;
1393 }
1394 
1395 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1396 {
1397 	return pte_clear_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1398 }
1399 
1400 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1401 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1402 {
1403 	return pmd_set_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1404 }
1405 
1406 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1407 {
1408 	return pmd_flags(pmd) & _PAGE_SWP_SOFT_DIRTY;
1409 }
1410 
1411 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1412 {
1413 	return pmd_clear_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1414 }
1415 #endif
1416 #endif
1417 
1418 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
1419 static inline pte_t pte_swp_mkuffd_wp(pte_t pte)
1420 {
1421 	return pte_set_flags(pte, _PAGE_SWP_UFFD_WP);
1422 }
1423 
1424 static inline int pte_swp_uffd_wp(pte_t pte)
1425 {
1426 	return pte_flags(pte) & _PAGE_SWP_UFFD_WP;
1427 }
1428 
1429 static inline pte_t pte_swp_clear_uffd_wp(pte_t pte)
1430 {
1431 	return pte_clear_flags(pte, _PAGE_SWP_UFFD_WP);
1432 }
1433 
1434 static inline pmd_t pmd_swp_mkuffd_wp(pmd_t pmd)
1435 {
1436 	return pmd_set_flags(pmd, _PAGE_SWP_UFFD_WP);
1437 }
1438 
1439 static inline int pmd_swp_uffd_wp(pmd_t pmd)
1440 {
1441 	return pmd_flags(pmd) & _PAGE_SWP_UFFD_WP;
1442 }
1443 
1444 static inline pmd_t pmd_swp_clear_uffd_wp(pmd_t pmd)
1445 {
1446 	return pmd_clear_flags(pmd, _PAGE_SWP_UFFD_WP);
1447 }
1448 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
1449 
1450 #define PKRU_AD_BIT 0x1
1451 #define PKRU_WD_BIT 0x2
1452 #define PKRU_BITS_PER_PKEY 2
1453 
1454 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1455 extern u32 init_pkru_value;
1456 #else
1457 #define init_pkru_value	0
1458 #endif
1459 
1460 static inline bool __pkru_allows_read(u32 pkru, u16 pkey)
1461 {
1462 	int pkru_pkey_bits = pkey * PKRU_BITS_PER_PKEY;
1463 	return !(pkru & (PKRU_AD_BIT << pkru_pkey_bits));
1464 }
1465 
1466 static inline bool __pkru_allows_write(u32 pkru, u16 pkey)
1467 {
1468 	int pkru_pkey_bits = pkey * PKRU_BITS_PER_PKEY;
1469 	/*
1470 	 * Access-disable disables writes too so we need to check
1471 	 * both bits here.
1472 	 */
1473 	return !(pkru & ((PKRU_AD_BIT|PKRU_WD_BIT) << pkru_pkey_bits));
1474 }
1475 
1476 static inline u16 pte_flags_pkey(unsigned long pte_flags)
1477 {
1478 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1479 	/* ifdef to avoid doing 59-bit shift on 32-bit values */
1480 	return (pte_flags & _PAGE_PKEY_MASK) >> _PAGE_BIT_PKEY_BIT0;
1481 #else
1482 	return 0;
1483 #endif
1484 }
1485 
1486 static inline bool __pkru_allows_pkey(u16 pkey, bool write)
1487 {
1488 	u32 pkru = read_pkru();
1489 
1490 	if (!__pkru_allows_read(pkru, pkey))
1491 		return false;
1492 	if (write && !__pkru_allows_write(pkru, pkey))
1493 		return false;
1494 
1495 	return true;
1496 }
1497 
1498 /*
1499  * 'pteval' can come from a PTE, PMD or PUD.  We only check
1500  * _PAGE_PRESENT, _PAGE_USER, and _PAGE_RW in here which are the
1501  * same value on all 3 types.
1502  */
1503 static inline bool __pte_access_permitted(unsigned long pteval, bool write)
1504 {
1505 	unsigned long need_pte_bits = _PAGE_PRESENT|_PAGE_USER;
1506 
1507 	if (write)
1508 		need_pte_bits |= _PAGE_RW;
1509 
1510 	if ((pteval & need_pte_bits) != need_pte_bits)
1511 		return 0;
1512 
1513 	return __pkru_allows_pkey(pte_flags_pkey(pteval), write);
1514 }
1515 
1516 #define pte_access_permitted pte_access_permitted
1517 static inline bool pte_access_permitted(pte_t pte, bool write)
1518 {
1519 	return __pte_access_permitted(pte_val(pte), write);
1520 }
1521 
1522 #define pmd_access_permitted pmd_access_permitted
1523 static inline bool pmd_access_permitted(pmd_t pmd, bool write)
1524 {
1525 	return __pte_access_permitted(pmd_val(pmd), write);
1526 }
1527 
1528 #define pud_access_permitted pud_access_permitted
1529 static inline bool pud_access_permitted(pud_t pud, bool write)
1530 {
1531 	return __pte_access_permitted(pud_val(pud), write);
1532 }
1533 
1534 #define __HAVE_ARCH_PFN_MODIFY_ALLOWED 1
1535 extern bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot);
1536 
1537 static inline bool arch_has_pfn_modify_check(void)
1538 {
1539 	return boot_cpu_has_bug(X86_BUG_L1TF);
1540 }
1541 
1542 #define arch_faults_on_old_pte arch_faults_on_old_pte
1543 static inline bool arch_faults_on_old_pte(void)
1544 {
1545 	return false;
1546 }
1547 
1548 #include <asm-generic/pgtable.h>
1549 #endif	/* __ASSEMBLY__ */
1550 
1551 #endif /* _ASM_X86_PGTABLE_H */
1552