xref: /openbmc/linux/arch/x86/include/asm/pgtable.h (revision 82e6fdd6)
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 
27 extern pgd_t early_top_pgt[PTRS_PER_PGD];
28 int __init __early_make_pgtable(unsigned long address, pmdval_t pmd);
29 
30 void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd);
31 void ptdump_walk_pgd_level_debugfs(struct seq_file *m, pgd_t *pgd, bool user);
32 void ptdump_walk_pgd_level_checkwx(void);
33 
34 #ifdef CONFIG_DEBUG_WX
35 #define debug_checkwx() ptdump_walk_pgd_level_checkwx()
36 #else
37 #define debug_checkwx() do { } while (0)
38 #endif
39 
40 /*
41  * ZERO_PAGE is a global shared page that is always zero: used
42  * for zero-mapped memory areas etc..
43  */
44 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
45 	__visible;
46 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
47 
48 extern spinlock_t pgd_lock;
49 extern struct list_head pgd_list;
50 
51 extern struct mm_struct *pgd_page_get_mm(struct page *page);
52 
53 extern pmdval_t early_pmd_flags;
54 
55 #ifdef CONFIG_PARAVIRT
56 #include <asm/paravirt.h>
57 #else  /* !CONFIG_PARAVIRT */
58 #define set_pte(ptep, pte)		native_set_pte(ptep, pte)
59 #define set_pte_at(mm, addr, ptep, pte)	native_set_pte_at(mm, addr, ptep, pte)
60 
61 #define set_pte_atomic(ptep, pte)					\
62 	native_set_pte_atomic(ptep, pte)
63 
64 #define set_pmd(pmdp, pmd)		native_set_pmd(pmdp, pmd)
65 
66 #ifndef __PAGETABLE_P4D_FOLDED
67 #define set_pgd(pgdp, pgd)		native_set_pgd(pgdp, pgd)
68 #define pgd_clear(pgd)			native_pgd_clear(pgd)
69 #endif
70 
71 #ifndef set_p4d
72 # define set_p4d(p4dp, p4d)		native_set_p4d(p4dp, p4d)
73 #endif
74 
75 #ifndef __PAGETABLE_PUD_FOLDED
76 #define p4d_clear(p4d)			native_p4d_clear(p4d)
77 #endif
78 
79 #ifndef set_pud
80 # define set_pud(pudp, pud)		native_set_pud(pudp, pud)
81 #endif
82 
83 #ifndef __PAGETABLE_PUD_FOLDED
84 #define pud_clear(pud)			native_pud_clear(pud)
85 #endif
86 
87 #define pte_clear(mm, addr, ptep)	native_pte_clear(mm, addr, ptep)
88 #define pmd_clear(pmd)			native_pmd_clear(pmd)
89 
90 #define pgd_val(x)	native_pgd_val(x)
91 #define __pgd(x)	native_make_pgd(x)
92 
93 #ifndef __PAGETABLE_P4D_FOLDED
94 #define p4d_val(x)	native_p4d_val(x)
95 #define __p4d(x)	native_make_p4d(x)
96 #endif
97 
98 #ifndef __PAGETABLE_PUD_FOLDED
99 #define pud_val(x)	native_pud_val(x)
100 #define __pud(x)	native_make_pud(x)
101 #endif
102 
103 #ifndef __PAGETABLE_PMD_FOLDED
104 #define pmd_val(x)	native_pmd_val(x)
105 #define __pmd(x)	native_make_pmd(x)
106 #endif
107 
108 #define pte_val(x)	native_pte_val(x)
109 #define __pte(x)	native_make_pte(x)
110 
111 #define arch_end_context_switch(prev)	do {} while(0)
112 
113 #endif	/* CONFIG_PARAVIRT */
114 
115 /*
116  * The following only work if pte_present() is true.
117  * Undefined behaviour if not..
118  */
119 static inline int pte_dirty(pte_t pte)
120 {
121 	return pte_flags(pte) & _PAGE_DIRTY;
122 }
123 
124 
125 static inline u32 read_pkru(void)
126 {
127 	if (boot_cpu_has(X86_FEATURE_OSPKE))
128 		return __read_pkru();
129 	return 0;
130 }
131 
132 static inline void write_pkru(u32 pkru)
133 {
134 	if (boot_cpu_has(X86_FEATURE_OSPKE))
135 		__write_pkru(pkru);
136 }
137 
138 static inline int pte_young(pte_t pte)
139 {
140 	return pte_flags(pte) & _PAGE_ACCESSED;
141 }
142 
143 static inline int pmd_dirty(pmd_t pmd)
144 {
145 	return pmd_flags(pmd) & _PAGE_DIRTY;
146 }
147 
148 static inline int pmd_young(pmd_t pmd)
149 {
150 	return pmd_flags(pmd) & _PAGE_ACCESSED;
151 }
152 
153 static inline int pud_dirty(pud_t pud)
154 {
155 	return pud_flags(pud) & _PAGE_DIRTY;
156 }
157 
158 static inline int pud_young(pud_t pud)
159 {
160 	return pud_flags(pud) & _PAGE_ACCESSED;
161 }
162 
163 static inline int pte_write(pte_t pte)
164 {
165 	return pte_flags(pte) & _PAGE_RW;
166 }
167 
168 static inline int pte_huge(pte_t pte)
169 {
170 	return pte_flags(pte) & _PAGE_PSE;
171 }
172 
173 static inline int pte_global(pte_t pte)
174 {
175 	return pte_flags(pte) & _PAGE_GLOBAL;
176 }
177 
178 static inline int pte_exec(pte_t pte)
179 {
180 	return !(pte_flags(pte) & _PAGE_NX);
181 }
182 
183 static inline int pte_special(pte_t pte)
184 {
185 	return pte_flags(pte) & _PAGE_SPECIAL;
186 }
187 
188 static inline unsigned long pte_pfn(pte_t pte)
189 {
190 	return (pte_val(pte) & PTE_PFN_MASK) >> PAGE_SHIFT;
191 }
192 
193 static inline unsigned long pmd_pfn(pmd_t pmd)
194 {
195 	return (pmd_val(pmd) & pmd_pfn_mask(pmd)) >> PAGE_SHIFT;
196 }
197 
198 static inline unsigned long pud_pfn(pud_t pud)
199 {
200 	return (pud_val(pud) & pud_pfn_mask(pud)) >> PAGE_SHIFT;
201 }
202 
203 static inline unsigned long p4d_pfn(p4d_t p4d)
204 {
205 	return (p4d_val(p4d) & p4d_pfn_mask(p4d)) >> PAGE_SHIFT;
206 }
207 
208 static inline unsigned long pgd_pfn(pgd_t pgd)
209 {
210 	return (pgd_val(pgd) & PTE_PFN_MASK) >> PAGE_SHIFT;
211 }
212 
213 static inline int p4d_large(p4d_t p4d)
214 {
215 	/* No 512 GiB pages yet */
216 	return 0;
217 }
218 
219 #define pte_page(pte)	pfn_to_page(pte_pfn(pte))
220 
221 static inline int pmd_large(pmd_t pte)
222 {
223 	return pmd_flags(pte) & _PAGE_PSE;
224 }
225 
226 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
227 static inline int pmd_trans_huge(pmd_t pmd)
228 {
229 	return (pmd_val(pmd) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
230 }
231 
232 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
233 static inline int pud_trans_huge(pud_t pud)
234 {
235 	return (pud_val(pud) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
236 }
237 #endif
238 
239 #define has_transparent_hugepage has_transparent_hugepage
240 static inline int has_transparent_hugepage(void)
241 {
242 	return boot_cpu_has(X86_FEATURE_PSE);
243 }
244 
245 #ifdef __HAVE_ARCH_PTE_DEVMAP
246 static inline int pmd_devmap(pmd_t pmd)
247 {
248 	return !!(pmd_val(pmd) & _PAGE_DEVMAP);
249 }
250 
251 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
252 static inline int pud_devmap(pud_t pud)
253 {
254 	return !!(pud_val(pud) & _PAGE_DEVMAP);
255 }
256 #else
257 static inline int pud_devmap(pud_t pud)
258 {
259 	return 0;
260 }
261 #endif
262 
263 static inline int pgd_devmap(pgd_t pgd)
264 {
265 	return 0;
266 }
267 #endif
268 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
269 
270 static inline pte_t pte_set_flags(pte_t pte, pteval_t set)
271 {
272 	pteval_t v = native_pte_val(pte);
273 
274 	return native_make_pte(v | set);
275 }
276 
277 static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear)
278 {
279 	pteval_t v = native_pte_val(pte);
280 
281 	return native_make_pte(v & ~clear);
282 }
283 
284 static inline pte_t pte_mkclean(pte_t pte)
285 {
286 	return pte_clear_flags(pte, _PAGE_DIRTY);
287 }
288 
289 static inline pte_t pte_mkold(pte_t pte)
290 {
291 	return pte_clear_flags(pte, _PAGE_ACCESSED);
292 }
293 
294 static inline pte_t pte_wrprotect(pte_t pte)
295 {
296 	return pte_clear_flags(pte, _PAGE_RW);
297 }
298 
299 static inline pte_t pte_mkexec(pte_t pte)
300 {
301 	return pte_clear_flags(pte, _PAGE_NX);
302 }
303 
304 static inline pte_t pte_mkdirty(pte_t pte)
305 {
306 	return pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
307 }
308 
309 static inline pte_t pte_mkyoung(pte_t pte)
310 {
311 	return pte_set_flags(pte, _PAGE_ACCESSED);
312 }
313 
314 static inline pte_t pte_mkwrite(pte_t pte)
315 {
316 	return pte_set_flags(pte, _PAGE_RW);
317 }
318 
319 static inline pte_t pte_mkhuge(pte_t pte)
320 {
321 	return pte_set_flags(pte, _PAGE_PSE);
322 }
323 
324 static inline pte_t pte_clrhuge(pte_t pte)
325 {
326 	return pte_clear_flags(pte, _PAGE_PSE);
327 }
328 
329 static inline pte_t pte_mkglobal(pte_t pte)
330 {
331 	return pte_set_flags(pte, _PAGE_GLOBAL);
332 }
333 
334 static inline pte_t pte_clrglobal(pte_t pte)
335 {
336 	return pte_clear_flags(pte, _PAGE_GLOBAL);
337 }
338 
339 static inline pte_t pte_mkspecial(pte_t pte)
340 {
341 	return pte_set_flags(pte, _PAGE_SPECIAL);
342 }
343 
344 static inline pte_t pte_mkdevmap(pte_t pte)
345 {
346 	return pte_set_flags(pte, _PAGE_SPECIAL|_PAGE_DEVMAP);
347 }
348 
349 static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
350 {
351 	pmdval_t v = native_pmd_val(pmd);
352 
353 	return native_make_pmd(v | set);
354 }
355 
356 static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
357 {
358 	pmdval_t v = native_pmd_val(pmd);
359 
360 	return native_make_pmd(v & ~clear);
361 }
362 
363 static inline pmd_t pmd_mkold(pmd_t pmd)
364 {
365 	return pmd_clear_flags(pmd, _PAGE_ACCESSED);
366 }
367 
368 static inline pmd_t pmd_mkclean(pmd_t pmd)
369 {
370 	return pmd_clear_flags(pmd, _PAGE_DIRTY);
371 }
372 
373 static inline pmd_t pmd_wrprotect(pmd_t pmd)
374 {
375 	return pmd_clear_flags(pmd, _PAGE_RW);
376 }
377 
378 static inline pmd_t pmd_mkdirty(pmd_t pmd)
379 {
380 	return pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
381 }
382 
383 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
384 {
385 	return pmd_set_flags(pmd, _PAGE_DEVMAP);
386 }
387 
388 static inline pmd_t pmd_mkhuge(pmd_t pmd)
389 {
390 	return pmd_set_flags(pmd, _PAGE_PSE);
391 }
392 
393 static inline pmd_t pmd_mkyoung(pmd_t pmd)
394 {
395 	return pmd_set_flags(pmd, _PAGE_ACCESSED);
396 }
397 
398 static inline pmd_t pmd_mkwrite(pmd_t pmd)
399 {
400 	return pmd_set_flags(pmd, _PAGE_RW);
401 }
402 
403 static inline pmd_t pmd_mknotpresent(pmd_t pmd)
404 {
405 	return pmd_clear_flags(pmd, _PAGE_PRESENT | _PAGE_PROTNONE);
406 }
407 
408 static inline pud_t pud_set_flags(pud_t pud, pudval_t set)
409 {
410 	pudval_t v = native_pud_val(pud);
411 
412 	return native_make_pud(v | set);
413 }
414 
415 static inline pud_t pud_clear_flags(pud_t pud, pudval_t clear)
416 {
417 	pudval_t v = native_pud_val(pud);
418 
419 	return native_make_pud(v & ~clear);
420 }
421 
422 static inline pud_t pud_mkold(pud_t pud)
423 {
424 	return pud_clear_flags(pud, _PAGE_ACCESSED);
425 }
426 
427 static inline pud_t pud_mkclean(pud_t pud)
428 {
429 	return pud_clear_flags(pud, _PAGE_DIRTY);
430 }
431 
432 static inline pud_t pud_wrprotect(pud_t pud)
433 {
434 	return pud_clear_flags(pud, _PAGE_RW);
435 }
436 
437 static inline pud_t pud_mkdirty(pud_t pud)
438 {
439 	return pud_set_flags(pud, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
440 }
441 
442 static inline pud_t pud_mkdevmap(pud_t pud)
443 {
444 	return pud_set_flags(pud, _PAGE_DEVMAP);
445 }
446 
447 static inline pud_t pud_mkhuge(pud_t pud)
448 {
449 	return pud_set_flags(pud, _PAGE_PSE);
450 }
451 
452 static inline pud_t pud_mkyoung(pud_t pud)
453 {
454 	return pud_set_flags(pud, _PAGE_ACCESSED);
455 }
456 
457 static inline pud_t pud_mkwrite(pud_t pud)
458 {
459 	return pud_set_flags(pud, _PAGE_RW);
460 }
461 
462 static inline pud_t pud_mknotpresent(pud_t pud)
463 {
464 	return pud_clear_flags(pud, _PAGE_PRESENT | _PAGE_PROTNONE);
465 }
466 
467 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
468 static inline int pte_soft_dirty(pte_t pte)
469 {
470 	return pte_flags(pte) & _PAGE_SOFT_DIRTY;
471 }
472 
473 static inline int pmd_soft_dirty(pmd_t pmd)
474 {
475 	return pmd_flags(pmd) & _PAGE_SOFT_DIRTY;
476 }
477 
478 static inline int pud_soft_dirty(pud_t pud)
479 {
480 	return pud_flags(pud) & _PAGE_SOFT_DIRTY;
481 }
482 
483 static inline pte_t pte_mksoft_dirty(pte_t pte)
484 {
485 	return pte_set_flags(pte, _PAGE_SOFT_DIRTY);
486 }
487 
488 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
489 {
490 	return pmd_set_flags(pmd, _PAGE_SOFT_DIRTY);
491 }
492 
493 static inline pud_t pud_mksoft_dirty(pud_t pud)
494 {
495 	return pud_set_flags(pud, _PAGE_SOFT_DIRTY);
496 }
497 
498 static inline pte_t pte_clear_soft_dirty(pte_t pte)
499 {
500 	return pte_clear_flags(pte, _PAGE_SOFT_DIRTY);
501 }
502 
503 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
504 {
505 	return pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY);
506 }
507 
508 static inline pud_t pud_clear_soft_dirty(pud_t pud)
509 {
510 	return pud_clear_flags(pud, _PAGE_SOFT_DIRTY);
511 }
512 
513 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
514 
515 /*
516  * Mask out unsupported bits in a present pgprot.  Non-present pgprots
517  * can use those bits for other purposes, so leave them be.
518  */
519 static inline pgprotval_t massage_pgprot(pgprot_t pgprot)
520 {
521 	pgprotval_t protval = pgprot_val(pgprot);
522 
523 	if (protval & _PAGE_PRESENT)
524 		protval &= __supported_pte_mask;
525 
526 	return protval;
527 }
528 
529 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
530 {
531 	return __pte(((phys_addr_t)page_nr << PAGE_SHIFT) |
532 		     massage_pgprot(pgprot));
533 }
534 
535 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
536 {
537 	return __pmd(((phys_addr_t)page_nr << PAGE_SHIFT) |
538 		     massage_pgprot(pgprot));
539 }
540 
541 static inline pud_t pfn_pud(unsigned long page_nr, pgprot_t pgprot)
542 {
543 	return __pud(((phys_addr_t)page_nr << PAGE_SHIFT) |
544 		     massage_pgprot(pgprot));
545 }
546 
547 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
548 {
549 	pteval_t val = pte_val(pte);
550 
551 	/*
552 	 * Chop off the NX bit (if present), and add the NX portion of
553 	 * the newprot (if present):
554 	 */
555 	val &= _PAGE_CHG_MASK;
556 	val |= massage_pgprot(newprot) & ~_PAGE_CHG_MASK;
557 
558 	return __pte(val);
559 }
560 
561 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
562 {
563 	pmdval_t val = pmd_val(pmd);
564 
565 	val &= _HPAGE_CHG_MASK;
566 	val |= massage_pgprot(newprot) & ~_HPAGE_CHG_MASK;
567 
568 	return __pmd(val);
569 }
570 
571 /* mprotect needs to preserve PAT bits when updating vm_page_prot */
572 #define pgprot_modify pgprot_modify
573 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
574 {
575 	pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK;
576 	pgprotval_t addbits = pgprot_val(newprot);
577 	return __pgprot(preservebits | addbits);
578 }
579 
580 #define pte_pgprot(x) __pgprot(pte_flags(x))
581 #define pmd_pgprot(x) __pgprot(pmd_flags(x))
582 #define pud_pgprot(x) __pgprot(pud_flags(x))
583 #define p4d_pgprot(x) __pgprot(p4d_flags(x))
584 
585 #define canon_pgprot(p) __pgprot(massage_pgprot(p))
586 
587 static inline int is_new_memtype_allowed(u64 paddr, unsigned long size,
588 					 enum page_cache_mode pcm,
589 					 enum page_cache_mode new_pcm)
590 {
591 	/*
592 	 * PAT type is always WB for untracked ranges, so no need to check.
593 	 */
594 	if (x86_platform.is_untracked_pat_range(paddr, paddr + size))
595 		return 1;
596 
597 	/*
598 	 * Certain new memtypes are not allowed with certain
599 	 * requested memtype:
600 	 * - request is uncached, return cannot be write-back
601 	 * - request is write-combine, return cannot be write-back
602 	 * - request is write-through, return cannot be write-back
603 	 * - request is write-through, return cannot be write-combine
604 	 */
605 	if ((pcm == _PAGE_CACHE_MODE_UC_MINUS &&
606 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
607 	    (pcm == _PAGE_CACHE_MODE_WC &&
608 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
609 	    (pcm == _PAGE_CACHE_MODE_WT &&
610 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
611 	    (pcm == _PAGE_CACHE_MODE_WT &&
612 	     new_pcm == _PAGE_CACHE_MODE_WC)) {
613 		return 0;
614 	}
615 
616 	return 1;
617 }
618 
619 pmd_t *populate_extra_pmd(unsigned long vaddr);
620 pte_t *populate_extra_pte(unsigned long vaddr);
621 #endif	/* __ASSEMBLY__ */
622 
623 #ifdef CONFIG_X86_32
624 # include <asm/pgtable_32.h>
625 #else
626 # include <asm/pgtable_64.h>
627 #endif
628 
629 #ifndef __ASSEMBLY__
630 #include <linux/mm_types.h>
631 #include <linux/mmdebug.h>
632 #include <linux/log2.h>
633 #include <asm/fixmap.h>
634 
635 static inline int pte_none(pte_t pte)
636 {
637 	return !(pte.pte & ~(_PAGE_KNL_ERRATUM_MASK));
638 }
639 
640 #define __HAVE_ARCH_PTE_SAME
641 static inline int pte_same(pte_t a, pte_t b)
642 {
643 	return a.pte == b.pte;
644 }
645 
646 static inline int pte_present(pte_t a)
647 {
648 	return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE);
649 }
650 
651 #ifdef __HAVE_ARCH_PTE_DEVMAP
652 static inline int pte_devmap(pte_t a)
653 {
654 	return (pte_flags(a) & _PAGE_DEVMAP) == _PAGE_DEVMAP;
655 }
656 #endif
657 
658 #define pte_accessible pte_accessible
659 static inline bool pte_accessible(struct mm_struct *mm, pte_t a)
660 {
661 	if (pte_flags(a) & _PAGE_PRESENT)
662 		return true;
663 
664 	if ((pte_flags(a) & _PAGE_PROTNONE) &&
665 			mm_tlb_flush_pending(mm))
666 		return true;
667 
668 	return false;
669 }
670 
671 static inline int pmd_present(pmd_t pmd)
672 {
673 	/*
674 	 * Checking for _PAGE_PSE is needed too because
675 	 * split_huge_page will temporarily clear the present bit (but
676 	 * the _PAGE_PSE flag will remain set at all times while the
677 	 * _PAGE_PRESENT bit is clear).
678 	 */
679 	return pmd_flags(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PSE);
680 }
681 
682 #ifdef CONFIG_NUMA_BALANCING
683 /*
684  * These work without NUMA balancing but the kernel does not care. See the
685  * comment in include/asm-generic/pgtable.h
686  */
687 static inline int pte_protnone(pte_t pte)
688 {
689 	return (pte_flags(pte) & (_PAGE_PROTNONE | _PAGE_PRESENT))
690 		== _PAGE_PROTNONE;
691 }
692 
693 static inline int pmd_protnone(pmd_t pmd)
694 {
695 	return (pmd_flags(pmd) & (_PAGE_PROTNONE | _PAGE_PRESENT))
696 		== _PAGE_PROTNONE;
697 }
698 #endif /* CONFIG_NUMA_BALANCING */
699 
700 static inline int pmd_none(pmd_t pmd)
701 {
702 	/* Only check low word on 32-bit platforms, since it might be
703 	   out of sync with upper half. */
704 	unsigned long val = native_pmd_val(pmd);
705 	return (val & ~_PAGE_KNL_ERRATUM_MASK) == 0;
706 }
707 
708 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
709 {
710 	return (unsigned long)__va(pmd_val(pmd) & pmd_pfn_mask(pmd));
711 }
712 
713 /*
714  * Currently stuck as a macro due to indirect forward reference to
715  * linux/mmzone.h's __section_mem_map_addr() definition:
716  */
717 #define pmd_page(pmd)	pfn_to_page(pmd_pfn(pmd))
718 
719 /*
720  * the pmd page can be thought of an array like this: pmd_t[PTRS_PER_PMD]
721  *
722  * this macro returns the index of the entry in the pmd page which would
723  * control the given virtual address
724  */
725 static inline unsigned long pmd_index(unsigned long address)
726 {
727 	return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
728 }
729 
730 /*
731  * Conversion functions: convert a page and protection to a page entry,
732  * and a page entry and page directory to the page they refer to.
733  *
734  * (Currently stuck as a macro because of indirect forward reference
735  * to linux/mm.h:page_to_nid())
736  */
737 #define mk_pte(page, pgprot)   pfn_pte(page_to_pfn(page), (pgprot))
738 
739 /*
740  * the pte page can be thought of an array like this: pte_t[PTRS_PER_PTE]
741  *
742  * this function returns the index of the entry in the pte page which would
743  * control the given virtual address
744  */
745 static inline unsigned long pte_index(unsigned long address)
746 {
747 	return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
748 }
749 
750 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
751 {
752 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
753 }
754 
755 static inline int pmd_bad(pmd_t pmd)
756 {
757 	return (pmd_flags(pmd) & ~_PAGE_USER) != _KERNPG_TABLE;
758 }
759 
760 static inline unsigned long pages_to_mb(unsigned long npg)
761 {
762 	return npg >> (20 - PAGE_SHIFT);
763 }
764 
765 #if CONFIG_PGTABLE_LEVELS > 2
766 static inline int pud_none(pud_t pud)
767 {
768 	return (native_pud_val(pud) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
769 }
770 
771 static inline int pud_present(pud_t pud)
772 {
773 	return pud_flags(pud) & _PAGE_PRESENT;
774 }
775 
776 static inline unsigned long pud_page_vaddr(pud_t pud)
777 {
778 	return (unsigned long)__va(pud_val(pud) & pud_pfn_mask(pud));
779 }
780 
781 /*
782  * Currently stuck as a macro due to indirect forward reference to
783  * linux/mmzone.h's __section_mem_map_addr() definition:
784  */
785 #define pud_page(pud)	pfn_to_page(pud_pfn(pud))
786 
787 /* Find an entry in the second-level page table.. */
788 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
789 {
790 	return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(address);
791 }
792 
793 static inline int pud_large(pud_t pud)
794 {
795 	return (pud_val(pud) & (_PAGE_PSE | _PAGE_PRESENT)) ==
796 		(_PAGE_PSE | _PAGE_PRESENT);
797 }
798 
799 static inline int pud_bad(pud_t pud)
800 {
801 	return (pud_flags(pud) & ~(_KERNPG_TABLE | _PAGE_USER)) != 0;
802 }
803 #else
804 static inline int pud_large(pud_t pud)
805 {
806 	return 0;
807 }
808 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
809 
810 static inline unsigned long pud_index(unsigned long address)
811 {
812 	return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
813 }
814 
815 #if CONFIG_PGTABLE_LEVELS > 3
816 static inline int p4d_none(p4d_t p4d)
817 {
818 	return (native_p4d_val(p4d) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
819 }
820 
821 static inline int p4d_present(p4d_t p4d)
822 {
823 	return p4d_flags(p4d) & _PAGE_PRESENT;
824 }
825 
826 static inline unsigned long p4d_page_vaddr(p4d_t p4d)
827 {
828 	return (unsigned long)__va(p4d_val(p4d) & p4d_pfn_mask(p4d));
829 }
830 
831 /*
832  * Currently stuck as a macro due to indirect forward reference to
833  * linux/mmzone.h's __section_mem_map_addr() definition:
834  */
835 #define p4d_page(p4d)	pfn_to_page(p4d_pfn(p4d))
836 
837 /* Find an entry in the third-level page table.. */
838 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
839 {
840 	return (pud_t *)p4d_page_vaddr(*p4d) + pud_index(address);
841 }
842 
843 static inline int p4d_bad(p4d_t p4d)
844 {
845 	unsigned long ignore_flags = _KERNPG_TABLE | _PAGE_USER;
846 
847 	if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
848 		ignore_flags |= _PAGE_NX;
849 
850 	return (p4d_flags(p4d) & ~ignore_flags) != 0;
851 }
852 #endif  /* CONFIG_PGTABLE_LEVELS > 3 */
853 
854 static inline unsigned long p4d_index(unsigned long address)
855 {
856 	return (address >> P4D_SHIFT) & (PTRS_PER_P4D - 1);
857 }
858 
859 #if CONFIG_PGTABLE_LEVELS > 4
860 static inline int pgd_present(pgd_t pgd)
861 {
862 	return pgd_flags(pgd) & _PAGE_PRESENT;
863 }
864 
865 static inline unsigned long pgd_page_vaddr(pgd_t pgd)
866 {
867 	return (unsigned long)__va((unsigned long)pgd_val(pgd) & PTE_PFN_MASK);
868 }
869 
870 /*
871  * Currently stuck as a macro due to indirect forward reference to
872  * linux/mmzone.h's __section_mem_map_addr() definition:
873  */
874 #define pgd_page(pgd)	pfn_to_page(pgd_pfn(pgd))
875 
876 /* to find an entry in a page-table-directory. */
877 static inline p4d_t *p4d_offset(pgd_t *pgd, unsigned long address)
878 {
879 	return (p4d_t *)pgd_page_vaddr(*pgd) + p4d_index(address);
880 }
881 
882 static inline int pgd_bad(pgd_t pgd)
883 {
884 	unsigned long ignore_flags = _PAGE_USER;
885 
886 	if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
887 		ignore_flags |= _PAGE_NX;
888 
889 	return (pgd_flags(pgd) & ~ignore_flags) != _KERNPG_TABLE;
890 }
891 
892 static inline int pgd_none(pgd_t pgd)
893 {
894 	/*
895 	 * There is no need to do a workaround for the KNL stray
896 	 * A/D bit erratum here.  PGDs only point to page tables
897 	 * except on 32-bit non-PAE which is not supported on
898 	 * KNL.
899 	 */
900 	return !native_pgd_val(pgd);
901 }
902 #endif	/* CONFIG_PGTABLE_LEVELS > 4 */
903 
904 #endif	/* __ASSEMBLY__ */
905 
906 /*
907  * the pgd page can be thought of an array like this: pgd_t[PTRS_PER_PGD]
908  *
909  * this macro returns the index of the entry in the pgd page which would
910  * control the given virtual address
911  */
912 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
913 
914 /*
915  * pgd_offset() returns a (pgd_t *)
916  * pgd_index() is used get the offset into the pgd page's array of pgd_t's;
917  */
918 #define pgd_offset_pgd(pgd, address) (pgd + pgd_index((address)))
919 /*
920  * a shortcut to get a pgd_t in a given mm
921  */
922 #define pgd_offset(mm, address) pgd_offset_pgd((mm)->pgd, (address))
923 /*
924  * a shortcut which implies the use of the kernel's pgd, instead
925  * of a process's
926  */
927 #define pgd_offset_k(address) pgd_offset(&init_mm, (address))
928 
929 
930 #define KERNEL_PGD_BOUNDARY	pgd_index(PAGE_OFFSET)
931 #define KERNEL_PGD_PTRS		(PTRS_PER_PGD - KERNEL_PGD_BOUNDARY)
932 
933 #ifndef __ASSEMBLY__
934 
935 extern int direct_gbpages;
936 void init_mem_mapping(void);
937 void early_alloc_pgt_buf(void);
938 extern void memblock_find_dma_reserve(void);
939 
940 #ifdef CONFIG_X86_64
941 /* Realmode trampoline initialization. */
942 extern pgd_t trampoline_pgd_entry;
943 static inline void __meminit init_trampoline_default(void)
944 {
945 	/* Default trampoline pgd value */
946 	trampoline_pgd_entry = init_top_pgt[pgd_index(__PAGE_OFFSET)];
947 }
948 # ifdef CONFIG_RANDOMIZE_MEMORY
949 void __meminit init_trampoline(void);
950 # else
951 #  define init_trampoline init_trampoline_default
952 # endif
953 #else
954 static inline void init_trampoline(void) { }
955 #endif
956 
957 /* local pte updates need not use xchg for locking */
958 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
959 {
960 	pte_t res = *ptep;
961 
962 	/* Pure native function needs no input for mm, addr */
963 	native_pte_clear(NULL, 0, ptep);
964 	return res;
965 }
966 
967 static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp)
968 {
969 	pmd_t res = *pmdp;
970 
971 	native_pmd_clear(pmdp);
972 	return res;
973 }
974 
975 static inline pud_t native_local_pudp_get_and_clear(pud_t *pudp)
976 {
977 	pud_t res = *pudp;
978 
979 	native_pud_clear(pudp);
980 	return res;
981 }
982 
983 static inline void native_set_pte_at(struct mm_struct *mm, unsigned long addr,
984 				     pte_t *ptep , pte_t pte)
985 {
986 	native_set_pte(ptep, pte);
987 }
988 
989 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
990 			      pmd_t *pmdp, pmd_t pmd)
991 {
992 	native_set_pmd(pmdp, pmd);
993 }
994 
995 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
996 			      pud_t *pudp, pud_t pud)
997 {
998 	native_set_pud(pudp, pud);
999 }
1000 
1001 /*
1002  * We only update the dirty/accessed state if we set
1003  * the dirty bit by hand in the kernel, since the hardware
1004  * will do the accessed bit for us, and we don't want to
1005  * race with other CPU's that might be updating the dirty
1006  * bit at the same time.
1007  */
1008 struct vm_area_struct;
1009 
1010 #define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1011 extern int ptep_set_access_flags(struct vm_area_struct *vma,
1012 				 unsigned long address, pte_t *ptep,
1013 				 pte_t entry, int dirty);
1014 
1015 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1016 extern int ptep_test_and_clear_young(struct vm_area_struct *vma,
1017 				     unsigned long addr, pte_t *ptep);
1018 
1019 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1020 extern int ptep_clear_flush_young(struct vm_area_struct *vma,
1021 				  unsigned long address, pte_t *ptep);
1022 
1023 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1024 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
1025 				       pte_t *ptep)
1026 {
1027 	pte_t pte = native_ptep_get_and_clear(ptep);
1028 	return pte;
1029 }
1030 
1031 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1032 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1033 					    unsigned long addr, pte_t *ptep,
1034 					    int full)
1035 {
1036 	pte_t pte;
1037 	if (full) {
1038 		/*
1039 		 * Full address destruction in progress; paravirt does not
1040 		 * care about updates and native needs no locking
1041 		 */
1042 		pte = native_local_ptep_get_and_clear(ptep);
1043 	} else {
1044 		pte = ptep_get_and_clear(mm, addr, ptep);
1045 	}
1046 	return pte;
1047 }
1048 
1049 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1050 static inline void ptep_set_wrprotect(struct mm_struct *mm,
1051 				      unsigned long addr, pte_t *ptep)
1052 {
1053 	clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte);
1054 }
1055 
1056 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
1057 
1058 #define mk_pmd(page, pgprot)   pfn_pmd(page_to_pfn(page), (pgprot))
1059 
1060 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1061 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
1062 				 unsigned long address, pmd_t *pmdp,
1063 				 pmd_t entry, int dirty);
1064 extern int pudp_set_access_flags(struct vm_area_struct *vma,
1065 				 unsigned long address, pud_t *pudp,
1066 				 pud_t entry, int dirty);
1067 
1068 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1069 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1070 				     unsigned long addr, pmd_t *pmdp);
1071 extern int pudp_test_and_clear_young(struct vm_area_struct *vma,
1072 				     unsigned long addr, pud_t *pudp);
1073 
1074 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
1075 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
1076 				  unsigned long address, pmd_t *pmdp);
1077 
1078 
1079 #define pmd_write pmd_write
1080 static inline int pmd_write(pmd_t pmd)
1081 {
1082 	return pmd_flags(pmd) & _PAGE_RW;
1083 }
1084 
1085 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1086 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, unsigned long addr,
1087 				       pmd_t *pmdp)
1088 {
1089 	return native_pmdp_get_and_clear(pmdp);
1090 }
1091 
1092 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
1093 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
1094 					unsigned long addr, pud_t *pudp)
1095 {
1096 	return native_pudp_get_and_clear(pudp);
1097 }
1098 
1099 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1100 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1101 				      unsigned long addr, pmd_t *pmdp)
1102 {
1103 	clear_bit(_PAGE_BIT_RW, (unsigned long *)pmdp);
1104 }
1105 
1106 #define pud_write pud_write
1107 static inline int pud_write(pud_t pud)
1108 {
1109 	return pud_flags(pud) & _PAGE_RW;
1110 }
1111 
1112 #ifndef pmdp_establish
1113 #define pmdp_establish pmdp_establish
1114 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
1115 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
1116 {
1117 	if (IS_ENABLED(CONFIG_SMP)) {
1118 		return xchg(pmdp, pmd);
1119 	} else {
1120 		pmd_t old = *pmdp;
1121 		*pmdp = pmd;
1122 		return old;
1123 	}
1124 }
1125 #endif
1126 
1127 /*
1128  * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
1129  *
1130  *  dst - pointer to pgd range anwhere on a pgd page
1131  *  src - ""
1132  *  count - the number of pgds to copy.
1133  *
1134  * dst and src can be on the same page, but the range must not overlap,
1135  * and must not cross a page boundary.
1136  */
1137 static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
1138 {
1139 	memcpy(dst, src, count * sizeof(pgd_t));
1140 #ifdef CONFIG_PAGE_TABLE_ISOLATION
1141 	if (!static_cpu_has(X86_FEATURE_PTI))
1142 		return;
1143 	/* Clone the user space pgd as well */
1144 	memcpy(kernel_to_user_pgdp(dst), kernel_to_user_pgdp(src),
1145 	       count * sizeof(pgd_t));
1146 #endif
1147 }
1148 
1149 #define PTE_SHIFT ilog2(PTRS_PER_PTE)
1150 static inline int page_level_shift(enum pg_level level)
1151 {
1152 	return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT;
1153 }
1154 static inline unsigned long page_level_size(enum pg_level level)
1155 {
1156 	return 1UL << page_level_shift(level);
1157 }
1158 static inline unsigned long page_level_mask(enum pg_level level)
1159 {
1160 	return ~(page_level_size(level) - 1);
1161 }
1162 
1163 /*
1164  * The x86 doesn't have any external MMU info: the kernel page
1165  * tables contain all the necessary information.
1166  */
1167 static inline void update_mmu_cache(struct vm_area_struct *vma,
1168 		unsigned long addr, pte_t *ptep)
1169 {
1170 }
1171 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1172 		unsigned long addr, pmd_t *pmd)
1173 {
1174 }
1175 static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
1176 		unsigned long addr, pud_t *pud)
1177 {
1178 }
1179 
1180 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1181 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1182 {
1183 	return pte_set_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1184 }
1185 
1186 static inline int pte_swp_soft_dirty(pte_t pte)
1187 {
1188 	return pte_flags(pte) & _PAGE_SWP_SOFT_DIRTY;
1189 }
1190 
1191 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1192 {
1193 	return pte_clear_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1194 }
1195 
1196 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1197 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1198 {
1199 	return pmd_set_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1200 }
1201 
1202 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1203 {
1204 	return pmd_flags(pmd) & _PAGE_SWP_SOFT_DIRTY;
1205 }
1206 
1207 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1208 {
1209 	return pmd_clear_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1210 }
1211 #endif
1212 #endif
1213 
1214 #define PKRU_AD_BIT 0x1
1215 #define PKRU_WD_BIT 0x2
1216 #define PKRU_BITS_PER_PKEY 2
1217 
1218 static inline bool __pkru_allows_read(u32 pkru, u16 pkey)
1219 {
1220 	int pkru_pkey_bits = pkey * PKRU_BITS_PER_PKEY;
1221 	return !(pkru & (PKRU_AD_BIT << pkru_pkey_bits));
1222 }
1223 
1224 static inline bool __pkru_allows_write(u32 pkru, u16 pkey)
1225 {
1226 	int pkru_pkey_bits = pkey * PKRU_BITS_PER_PKEY;
1227 	/*
1228 	 * Access-disable disables writes too so we need to check
1229 	 * both bits here.
1230 	 */
1231 	return !(pkru & ((PKRU_AD_BIT|PKRU_WD_BIT) << pkru_pkey_bits));
1232 }
1233 
1234 static inline u16 pte_flags_pkey(unsigned long pte_flags)
1235 {
1236 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1237 	/* ifdef to avoid doing 59-bit shift on 32-bit values */
1238 	return (pte_flags & _PAGE_PKEY_MASK) >> _PAGE_BIT_PKEY_BIT0;
1239 #else
1240 	return 0;
1241 #endif
1242 }
1243 
1244 static inline bool __pkru_allows_pkey(u16 pkey, bool write)
1245 {
1246 	u32 pkru = read_pkru();
1247 
1248 	if (!__pkru_allows_read(pkru, pkey))
1249 		return false;
1250 	if (write && !__pkru_allows_write(pkru, pkey))
1251 		return false;
1252 
1253 	return true;
1254 }
1255 
1256 /*
1257  * 'pteval' can come from a PTE, PMD or PUD.  We only check
1258  * _PAGE_PRESENT, _PAGE_USER, and _PAGE_RW in here which are the
1259  * same value on all 3 types.
1260  */
1261 static inline bool __pte_access_permitted(unsigned long pteval, bool write)
1262 {
1263 	unsigned long need_pte_bits = _PAGE_PRESENT|_PAGE_USER;
1264 
1265 	if (write)
1266 		need_pte_bits |= _PAGE_RW;
1267 
1268 	if ((pteval & need_pte_bits) != need_pte_bits)
1269 		return 0;
1270 
1271 	return __pkru_allows_pkey(pte_flags_pkey(pteval), write);
1272 }
1273 
1274 #define pte_access_permitted pte_access_permitted
1275 static inline bool pte_access_permitted(pte_t pte, bool write)
1276 {
1277 	return __pte_access_permitted(pte_val(pte), write);
1278 }
1279 
1280 #define pmd_access_permitted pmd_access_permitted
1281 static inline bool pmd_access_permitted(pmd_t pmd, bool write)
1282 {
1283 	return __pte_access_permitted(pmd_val(pmd), write);
1284 }
1285 
1286 #define pud_access_permitted pud_access_permitted
1287 static inline bool pud_access_permitted(pud_t pud, bool write)
1288 {
1289 	return __pte_access_permitted(pud_val(pud), write);
1290 }
1291 
1292 #include <asm-generic/pgtable.h>
1293 #endif	/* __ASSEMBLY__ */
1294 
1295 #endif /* _ASM_X86_PGTABLE_H */
1296