xref: /linux/arch/x86/include/asm/pgtable.h (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 #ifndef _ASM_X86_PGTABLE_H
2 #define _ASM_X86_PGTABLE_H
3 
4 #include <asm/page.h>
5 #include <asm/e820.h>
6 
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 <asm/x86_init.h>
20 
21 void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd);
22 
23 /*
24  * ZERO_PAGE is a global shared page that is always zero: used
25  * for zero-mapped memory areas etc..
26  */
27 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
28 	__visible;
29 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
30 
31 extern spinlock_t pgd_lock;
32 extern struct list_head pgd_list;
33 
34 extern struct mm_struct *pgd_page_get_mm(struct page *page);
35 
36 #ifdef CONFIG_PARAVIRT
37 #include <asm/paravirt.h>
38 #else  /* !CONFIG_PARAVIRT */
39 #define set_pte(ptep, pte)		native_set_pte(ptep, pte)
40 #define set_pte_at(mm, addr, ptep, pte)	native_set_pte_at(mm, addr, ptep, pte)
41 #define set_pmd_at(mm, addr, pmdp, pmd)	native_set_pmd_at(mm, addr, pmdp, pmd)
42 
43 #define set_pte_atomic(ptep, pte)					\
44 	native_set_pte_atomic(ptep, pte)
45 
46 #define set_pmd(pmdp, pmd)		native_set_pmd(pmdp, pmd)
47 
48 #ifndef __PAGETABLE_PUD_FOLDED
49 #define set_pgd(pgdp, pgd)		native_set_pgd(pgdp, pgd)
50 #define pgd_clear(pgd)			native_pgd_clear(pgd)
51 #endif
52 
53 #ifndef set_pud
54 # define set_pud(pudp, pud)		native_set_pud(pudp, pud)
55 #endif
56 
57 #ifndef __PAGETABLE_PMD_FOLDED
58 #define pud_clear(pud)			native_pud_clear(pud)
59 #endif
60 
61 #define pte_clear(mm, addr, ptep)	native_pte_clear(mm, addr, ptep)
62 #define pmd_clear(pmd)			native_pmd_clear(pmd)
63 
64 #define pte_update(mm, addr, ptep)              do { } while (0)
65 #define pte_update_defer(mm, addr, ptep)        do { } while (0)
66 #define pmd_update(mm, addr, ptep)              do { } while (0)
67 #define pmd_update_defer(mm, addr, ptep)        do { } while (0)
68 
69 #define pgd_val(x)	native_pgd_val(x)
70 #define __pgd(x)	native_make_pgd(x)
71 
72 #ifndef __PAGETABLE_PUD_FOLDED
73 #define pud_val(x)	native_pud_val(x)
74 #define __pud(x)	native_make_pud(x)
75 #endif
76 
77 #ifndef __PAGETABLE_PMD_FOLDED
78 #define pmd_val(x)	native_pmd_val(x)
79 #define __pmd(x)	native_make_pmd(x)
80 #endif
81 
82 #define pte_val(x)	native_pte_val(x)
83 #define __pte(x)	native_make_pte(x)
84 
85 #define arch_end_context_switch(prev)	do {} while(0)
86 
87 #endif	/* CONFIG_PARAVIRT */
88 
89 /*
90  * The following only work if pte_present() is true.
91  * Undefined behaviour if not..
92  */
93 static inline int pte_dirty(pte_t pte)
94 {
95 	return pte_flags(pte) & _PAGE_DIRTY;
96 }
97 
98 static inline int pte_young(pte_t pte)
99 {
100 	return pte_flags(pte) & _PAGE_ACCESSED;
101 }
102 
103 static inline int pmd_dirty(pmd_t pmd)
104 {
105 	return pmd_flags(pmd) & _PAGE_DIRTY;
106 }
107 
108 static inline int pmd_young(pmd_t pmd)
109 {
110 	return pmd_flags(pmd) & _PAGE_ACCESSED;
111 }
112 
113 static inline int pte_write(pte_t pte)
114 {
115 	return pte_flags(pte) & _PAGE_RW;
116 }
117 
118 static inline int pte_huge(pte_t pte)
119 {
120 	return pte_flags(pte) & _PAGE_PSE;
121 }
122 
123 static inline int pte_global(pte_t pte)
124 {
125 	return pte_flags(pte) & _PAGE_GLOBAL;
126 }
127 
128 static inline int pte_exec(pte_t pte)
129 {
130 	return !(pte_flags(pte) & _PAGE_NX);
131 }
132 
133 static inline int pte_special(pte_t pte)
134 {
135 	return pte_flags(pte) & _PAGE_SPECIAL;
136 }
137 
138 static inline unsigned long pte_pfn(pte_t pte)
139 {
140 	return (pte_val(pte) & PTE_PFN_MASK) >> PAGE_SHIFT;
141 }
142 
143 static inline unsigned long pmd_pfn(pmd_t pmd)
144 {
145 	return (pmd_val(pmd) & PTE_PFN_MASK) >> PAGE_SHIFT;
146 }
147 
148 static inline unsigned long pud_pfn(pud_t pud)
149 {
150 	return (pud_val(pud) & PTE_PFN_MASK) >> PAGE_SHIFT;
151 }
152 
153 #define pte_page(pte)	pfn_to_page(pte_pfn(pte))
154 
155 static inline int pmd_large(pmd_t pte)
156 {
157 	return pmd_flags(pte) & _PAGE_PSE;
158 }
159 
160 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
161 static inline int pmd_trans_splitting(pmd_t pmd)
162 {
163 	return pmd_val(pmd) & _PAGE_SPLITTING;
164 }
165 
166 static inline int pmd_trans_huge(pmd_t pmd)
167 {
168 	return pmd_val(pmd) & _PAGE_PSE;
169 }
170 
171 static inline int has_transparent_hugepage(void)
172 {
173 	return cpu_has_pse;
174 }
175 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
176 
177 static inline pte_t pte_set_flags(pte_t pte, pteval_t set)
178 {
179 	pteval_t v = native_pte_val(pte);
180 
181 	return native_make_pte(v | set);
182 }
183 
184 static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear)
185 {
186 	pteval_t v = native_pte_val(pte);
187 
188 	return native_make_pte(v & ~clear);
189 }
190 
191 static inline pte_t pte_mkclean(pte_t pte)
192 {
193 	return pte_clear_flags(pte, _PAGE_DIRTY);
194 }
195 
196 static inline pte_t pte_mkold(pte_t pte)
197 {
198 	return pte_clear_flags(pte, _PAGE_ACCESSED);
199 }
200 
201 static inline pte_t pte_wrprotect(pte_t pte)
202 {
203 	return pte_clear_flags(pte, _PAGE_RW);
204 }
205 
206 static inline pte_t pte_mkexec(pte_t pte)
207 {
208 	return pte_clear_flags(pte, _PAGE_NX);
209 }
210 
211 static inline pte_t pte_mkdirty(pte_t pte)
212 {
213 	return pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
214 }
215 
216 static inline pte_t pte_mkyoung(pte_t pte)
217 {
218 	return pte_set_flags(pte, _PAGE_ACCESSED);
219 }
220 
221 static inline pte_t pte_mkwrite(pte_t pte)
222 {
223 	return pte_set_flags(pte, _PAGE_RW);
224 }
225 
226 static inline pte_t pte_mkhuge(pte_t pte)
227 {
228 	return pte_set_flags(pte, _PAGE_PSE);
229 }
230 
231 static inline pte_t pte_clrhuge(pte_t pte)
232 {
233 	return pte_clear_flags(pte, _PAGE_PSE);
234 }
235 
236 static inline pte_t pte_mkglobal(pte_t pte)
237 {
238 	return pte_set_flags(pte, _PAGE_GLOBAL);
239 }
240 
241 static inline pte_t pte_clrglobal(pte_t pte)
242 {
243 	return pte_clear_flags(pte, _PAGE_GLOBAL);
244 }
245 
246 static inline pte_t pte_mkspecial(pte_t pte)
247 {
248 	return pte_set_flags(pte, _PAGE_SPECIAL);
249 }
250 
251 static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
252 {
253 	pmdval_t v = native_pmd_val(pmd);
254 
255 	return __pmd(v | set);
256 }
257 
258 static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
259 {
260 	pmdval_t v = native_pmd_val(pmd);
261 
262 	return __pmd(v & ~clear);
263 }
264 
265 static inline pmd_t pmd_mkold(pmd_t pmd)
266 {
267 	return pmd_clear_flags(pmd, _PAGE_ACCESSED);
268 }
269 
270 static inline pmd_t pmd_wrprotect(pmd_t pmd)
271 {
272 	return pmd_clear_flags(pmd, _PAGE_RW);
273 }
274 
275 static inline pmd_t pmd_mkdirty(pmd_t pmd)
276 {
277 	return pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
278 }
279 
280 static inline pmd_t pmd_mkhuge(pmd_t pmd)
281 {
282 	return pmd_set_flags(pmd, _PAGE_PSE);
283 }
284 
285 static inline pmd_t pmd_mkyoung(pmd_t pmd)
286 {
287 	return pmd_set_flags(pmd, _PAGE_ACCESSED);
288 }
289 
290 static inline pmd_t pmd_mkwrite(pmd_t pmd)
291 {
292 	return pmd_set_flags(pmd, _PAGE_RW);
293 }
294 
295 static inline pmd_t pmd_mknotpresent(pmd_t pmd)
296 {
297 	return pmd_clear_flags(pmd, _PAGE_PRESENT | _PAGE_PROTNONE);
298 }
299 
300 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
301 static inline int pte_soft_dirty(pte_t pte)
302 {
303 	return pte_flags(pte) & _PAGE_SOFT_DIRTY;
304 }
305 
306 static inline int pmd_soft_dirty(pmd_t pmd)
307 {
308 	return pmd_flags(pmd) & _PAGE_SOFT_DIRTY;
309 }
310 
311 static inline pte_t pte_mksoft_dirty(pte_t pte)
312 {
313 	return pte_set_flags(pte, _PAGE_SOFT_DIRTY);
314 }
315 
316 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
317 {
318 	return pmd_set_flags(pmd, _PAGE_SOFT_DIRTY);
319 }
320 
321 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
322 
323 /*
324  * Mask out unsupported bits in a present pgprot.  Non-present pgprots
325  * can use those bits for other purposes, so leave them be.
326  */
327 static inline pgprotval_t massage_pgprot(pgprot_t pgprot)
328 {
329 	pgprotval_t protval = pgprot_val(pgprot);
330 
331 	if (protval & _PAGE_PRESENT)
332 		protval &= __supported_pte_mask;
333 
334 	return protval;
335 }
336 
337 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
338 {
339 	return __pte(((phys_addr_t)page_nr << PAGE_SHIFT) |
340 		     massage_pgprot(pgprot));
341 }
342 
343 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
344 {
345 	return __pmd(((phys_addr_t)page_nr << PAGE_SHIFT) |
346 		     massage_pgprot(pgprot));
347 }
348 
349 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
350 {
351 	pteval_t val = pte_val(pte);
352 
353 	/*
354 	 * Chop off the NX bit (if present), and add the NX portion of
355 	 * the newprot (if present):
356 	 */
357 	val &= _PAGE_CHG_MASK;
358 	val |= massage_pgprot(newprot) & ~_PAGE_CHG_MASK;
359 
360 	return __pte(val);
361 }
362 
363 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
364 {
365 	pmdval_t val = pmd_val(pmd);
366 
367 	val &= _HPAGE_CHG_MASK;
368 	val |= massage_pgprot(newprot) & ~_HPAGE_CHG_MASK;
369 
370 	return __pmd(val);
371 }
372 
373 /* mprotect needs to preserve PAT bits when updating vm_page_prot */
374 #define pgprot_modify pgprot_modify
375 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
376 {
377 	pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK;
378 	pgprotval_t addbits = pgprot_val(newprot);
379 	return __pgprot(preservebits | addbits);
380 }
381 
382 #define pte_pgprot(x) __pgprot(pte_flags(x) & PTE_FLAGS_MASK)
383 
384 #define canon_pgprot(p) __pgprot(massage_pgprot(p))
385 
386 static inline int is_new_memtype_allowed(u64 paddr, unsigned long size,
387 					 enum page_cache_mode pcm,
388 					 enum page_cache_mode new_pcm)
389 {
390 	/*
391 	 * PAT type is always WB for untracked ranges, so no need to check.
392 	 */
393 	if (x86_platform.is_untracked_pat_range(paddr, paddr + size))
394 		return 1;
395 
396 	/*
397 	 * Certain new memtypes are not allowed with certain
398 	 * requested memtype:
399 	 * - request is uncached, return cannot be write-back
400 	 * - request is write-combine, return cannot be write-back
401 	 * - request is write-through, return cannot be write-back
402 	 * - request is write-through, return cannot be write-combine
403 	 */
404 	if ((pcm == _PAGE_CACHE_MODE_UC_MINUS &&
405 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
406 	    (pcm == _PAGE_CACHE_MODE_WC &&
407 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
408 	    (pcm == _PAGE_CACHE_MODE_WT &&
409 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
410 	    (pcm == _PAGE_CACHE_MODE_WT &&
411 	     new_pcm == _PAGE_CACHE_MODE_WC)) {
412 		return 0;
413 	}
414 
415 	return 1;
416 }
417 
418 pmd_t *populate_extra_pmd(unsigned long vaddr);
419 pte_t *populate_extra_pte(unsigned long vaddr);
420 #endif	/* __ASSEMBLY__ */
421 
422 #ifdef CONFIG_X86_32
423 # include <asm/pgtable_32.h>
424 #else
425 # include <asm/pgtable_64.h>
426 #endif
427 
428 #ifndef __ASSEMBLY__
429 #include <linux/mm_types.h>
430 #include <linux/mmdebug.h>
431 #include <linux/log2.h>
432 
433 static inline int pte_none(pte_t pte)
434 {
435 	return !pte.pte;
436 }
437 
438 #define __HAVE_ARCH_PTE_SAME
439 static inline int pte_same(pte_t a, pte_t b)
440 {
441 	return a.pte == b.pte;
442 }
443 
444 static inline int pte_present(pte_t a)
445 {
446 	return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE);
447 }
448 
449 #define pte_accessible pte_accessible
450 static inline bool pte_accessible(struct mm_struct *mm, pte_t a)
451 {
452 	if (pte_flags(a) & _PAGE_PRESENT)
453 		return true;
454 
455 	if ((pte_flags(a) & _PAGE_PROTNONE) &&
456 			mm_tlb_flush_pending(mm))
457 		return true;
458 
459 	return false;
460 }
461 
462 static inline int pte_hidden(pte_t pte)
463 {
464 	return pte_flags(pte) & _PAGE_HIDDEN;
465 }
466 
467 static inline int pmd_present(pmd_t pmd)
468 {
469 	/*
470 	 * Checking for _PAGE_PSE is needed too because
471 	 * split_huge_page will temporarily clear the present bit (but
472 	 * the _PAGE_PSE flag will remain set at all times while the
473 	 * _PAGE_PRESENT bit is clear).
474 	 */
475 	return pmd_flags(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PSE);
476 }
477 
478 #ifdef CONFIG_NUMA_BALANCING
479 /*
480  * These work without NUMA balancing but the kernel does not care. See the
481  * comment in include/asm-generic/pgtable.h
482  */
483 static inline int pte_protnone(pte_t pte)
484 {
485 	return (pte_flags(pte) & (_PAGE_PROTNONE | _PAGE_PRESENT))
486 		== _PAGE_PROTNONE;
487 }
488 
489 static inline int pmd_protnone(pmd_t pmd)
490 {
491 	return (pmd_flags(pmd) & (_PAGE_PROTNONE | _PAGE_PRESENT))
492 		== _PAGE_PROTNONE;
493 }
494 #endif /* CONFIG_NUMA_BALANCING */
495 
496 static inline int pmd_none(pmd_t pmd)
497 {
498 	/* Only check low word on 32-bit platforms, since it might be
499 	   out of sync with upper half. */
500 	return (unsigned long)native_pmd_val(pmd) == 0;
501 }
502 
503 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
504 {
505 	return (unsigned long)__va(pmd_val(pmd) & PTE_PFN_MASK);
506 }
507 
508 /*
509  * Currently stuck as a macro due to indirect forward reference to
510  * linux/mmzone.h's __section_mem_map_addr() definition:
511  */
512 #define pmd_page(pmd)	pfn_to_page((pmd_val(pmd) & PTE_PFN_MASK) >> PAGE_SHIFT)
513 
514 /*
515  * the pmd page can be thought of an array like this: pmd_t[PTRS_PER_PMD]
516  *
517  * this macro returns the index of the entry in the pmd page which would
518  * control the given virtual address
519  */
520 static inline unsigned long pmd_index(unsigned long address)
521 {
522 	return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
523 }
524 
525 /*
526  * Conversion functions: convert a page and protection to a page entry,
527  * and a page entry and page directory to the page they refer to.
528  *
529  * (Currently stuck as a macro because of indirect forward reference
530  * to linux/mm.h:page_to_nid())
531  */
532 #define mk_pte(page, pgprot)   pfn_pte(page_to_pfn(page), (pgprot))
533 
534 /*
535  * the pte page can be thought of an array like this: pte_t[PTRS_PER_PTE]
536  *
537  * this function returns the index of the entry in the pte page which would
538  * control the given virtual address
539  */
540 static inline unsigned long pte_index(unsigned long address)
541 {
542 	return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
543 }
544 
545 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
546 {
547 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
548 }
549 
550 static inline int pmd_bad(pmd_t pmd)
551 {
552 	return (pmd_flags(pmd) & ~_PAGE_USER) != _KERNPG_TABLE;
553 }
554 
555 static inline unsigned long pages_to_mb(unsigned long npg)
556 {
557 	return npg >> (20 - PAGE_SHIFT);
558 }
559 
560 #if CONFIG_PGTABLE_LEVELS > 2
561 static inline int pud_none(pud_t pud)
562 {
563 	return native_pud_val(pud) == 0;
564 }
565 
566 static inline int pud_present(pud_t pud)
567 {
568 	return pud_flags(pud) & _PAGE_PRESENT;
569 }
570 
571 static inline unsigned long pud_page_vaddr(pud_t pud)
572 {
573 	return (unsigned long)__va((unsigned long)pud_val(pud) & PTE_PFN_MASK);
574 }
575 
576 /*
577  * Currently stuck as a macro due to indirect forward reference to
578  * linux/mmzone.h's __section_mem_map_addr() definition:
579  */
580 #define pud_page(pud)		pfn_to_page(pud_val(pud) >> PAGE_SHIFT)
581 
582 /* Find an entry in the second-level page table.. */
583 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
584 {
585 	return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(address);
586 }
587 
588 static inline int pud_large(pud_t pud)
589 {
590 	return (pud_val(pud) & (_PAGE_PSE | _PAGE_PRESENT)) ==
591 		(_PAGE_PSE | _PAGE_PRESENT);
592 }
593 
594 static inline int pud_bad(pud_t pud)
595 {
596 	return (pud_flags(pud) & ~(_KERNPG_TABLE | _PAGE_USER)) != 0;
597 }
598 #else
599 static inline int pud_large(pud_t pud)
600 {
601 	return 0;
602 }
603 #endif	/* CONFIG_PGTABLE_LEVELS > 2 */
604 
605 #if CONFIG_PGTABLE_LEVELS > 3
606 static inline int pgd_present(pgd_t pgd)
607 {
608 	return pgd_flags(pgd) & _PAGE_PRESENT;
609 }
610 
611 static inline unsigned long pgd_page_vaddr(pgd_t pgd)
612 {
613 	return (unsigned long)__va((unsigned long)pgd_val(pgd) & PTE_PFN_MASK);
614 }
615 
616 /*
617  * Currently stuck as a macro due to indirect forward reference to
618  * linux/mmzone.h's __section_mem_map_addr() definition:
619  */
620 #define pgd_page(pgd)		pfn_to_page(pgd_val(pgd) >> PAGE_SHIFT)
621 
622 /* to find an entry in a page-table-directory. */
623 static inline unsigned long pud_index(unsigned long address)
624 {
625 	return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
626 }
627 
628 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
629 {
630 	return (pud_t *)pgd_page_vaddr(*pgd) + pud_index(address);
631 }
632 
633 static inline int pgd_bad(pgd_t pgd)
634 {
635 	return (pgd_flags(pgd) & ~_PAGE_USER) != _KERNPG_TABLE;
636 }
637 
638 static inline int pgd_none(pgd_t pgd)
639 {
640 	return !native_pgd_val(pgd);
641 }
642 #endif	/* CONFIG_PGTABLE_LEVELS > 3 */
643 
644 #endif	/* __ASSEMBLY__ */
645 
646 /*
647  * the pgd page can be thought of an array like this: pgd_t[PTRS_PER_PGD]
648  *
649  * this macro returns the index of the entry in the pgd page which would
650  * control the given virtual address
651  */
652 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
653 
654 /*
655  * pgd_offset() returns a (pgd_t *)
656  * pgd_index() is used get the offset into the pgd page's array of pgd_t's;
657  */
658 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index((address)))
659 /*
660  * a shortcut which implies the use of the kernel's pgd, instead
661  * of a process's
662  */
663 #define pgd_offset_k(address) pgd_offset(&init_mm, (address))
664 
665 
666 #define KERNEL_PGD_BOUNDARY	pgd_index(PAGE_OFFSET)
667 #define KERNEL_PGD_PTRS		(PTRS_PER_PGD - KERNEL_PGD_BOUNDARY)
668 
669 #ifndef __ASSEMBLY__
670 
671 extern int direct_gbpages;
672 void init_mem_mapping(void);
673 void early_alloc_pgt_buf(void);
674 
675 /* local pte updates need not use xchg for locking */
676 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
677 {
678 	pte_t res = *ptep;
679 
680 	/* Pure native function needs no input for mm, addr */
681 	native_pte_clear(NULL, 0, ptep);
682 	return res;
683 }
684 
685 static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp)
686 {
687 	pmd_t res = *pmdp;
688 
689 	native_pmd_clear(pmdp);
690 	return res;
691 }
692 
693 static inline void native_set_pte_at(struct mm_struct *mm, unsigned long addr,
694 				     pte_t *ptep , pte_t pte)
695 {
696 	native_set_pte(ptep, pte);
697 }
698 
699 static inline void native_set_pmd_at(struct mm_struct *mm, unsigned long addr,
700 				     pmd_t *pmdp , pmd_t pmd)
701 {
702 	native_set_pmd(pmdp, pmd);
703 }
704 
705 #ifndef CONFIG_PARAVIRT
706 /*
707  * Rules for using pte_update - it must be called after any PTE update which
708  * has not been done using the set_pte / clear_pte interfaces.  It is used by
709  * shadow mode hypervisors to resynchronize the shadow page tables.  Kernel PTE
710  * updates should either be sets, clears, or set_pte_atomic for P->P
711  * transitions, which means this hook should only be called for user PTEs.
712  * This hook implies a P->P protection or access change has taken place, which
713  * requires a subsequent TLB flush.  The notification can optionally be delayed
714  * until the TLB flush event by using the pte_update_defer form of the
715  * interface, but care must be taken to assure that the flush happens while
716  * still holding the same page table lock so that the shadow and primary pages
717  * do not become out of sync on SMP.
718  */
719 #define pte_update(mm, addr, ptep)		do { } while (0)
720 #define pte_update_defer(mm, addr, ptep)	do { } while (0)
721 #endif
722 
723 /*
724  * We only update the dirty/accessed state if we set
725  * the dirty bit by hand in the kernel, since the hardware
726  * will do the accessed bit for us, and we don't want to
727  * race with other CPU's that might be updating the dirty
728  * bit at the same time.
729  */
730 struct vm_area_struct;
731 
732 #define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
733 extern int ptep_set_access_flags(struct vm_area_struct *vma,
734 				 unsigned long address, pte_t *ptep,
735 				 pte_t entry, int dirty);
736 
737 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
738 extern int ptep_test_and_clear_young(struct vm_area_struct *vma,
739 				     unsigned long addr, pte_t *ptep);
740 
741 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
742 extern int ptep_clear_flush_young(struct vm_area_struct *vma,
743 				  unsigned long address, pte_t *ptep);
744 
745 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
746 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
747 				       pte_t *ptep)
748 {
749 	pte_t pte = native_ptep_get_and_clear(ptep);
750 	pte_update(mm, addr, ptep);
751 	return pte;
752 }
753 
754 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
755 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
756 					    unsigned long addr, pte_t *ptep,
757 					    int full)
758 {
759 	pte_t pte;
760 	if (full) {
761 		/*
762 		 * Full address destruction in progress; paravirt does not
763 		 * care about updates and native needs no locking
764 		 */
765 		pte = native_local_ptep_get_and_clear(ptep);
766 	} else {
767 		pte = ptep_get_and_clear(mm, addr, ptep);
768 	}
769 	return pte;
770 }
771 
772 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
773 static inline void ptep_set_wrprotect(struct mm_struct *mm,
774 				      unsigned long addr, pte_t *ptep)
775 {
776 	clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte);
777 	pte_update(mm, addr, ptep);
778 }
779 
780 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
781 
782 #define mk_pmd(page, pgprot)   pfn_pmd(page_to_pfn(page), (pgprot))
783 
784 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
785 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
786 				 unsigned long address, pmd_t *pmdp,
787 				 pmd_t entry, int dirty);
788 
789 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
790 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
791 				     unsigned long addr, pmd_t *pmdp);
792 
793 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
794 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
795 				  unsigned long address, pmd_t *pmdp);
796 
797 
798 #define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
799 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
800 				 unsigned long addr, pmd_t *pmdp);
801 
802 #define __HAVE_ARCH_PMD_WRITE
803 static inline int pmd_write(pmd_t pmd)
804 {
805 	return pmd_flags(pmd) & _PAGE_RW;
806 }
807 
808 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
809 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, unsigned long addr,
810 				       pmd_t *pmdp)
811 {
812 	pmd_t pmd = native_pmdp_get_and_clear(pmdp);
813 	pmd_update(mm, addr, pmdp);
814 	return pmd;
815 }
816 
817 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
818 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
819 				      unsigned long addr, pmd_t *pmdp)
820 {
821 	clear_bit(_PAGE_BIT_RW, (unsigned long *)pmdp);
822 	pmd_update(mm, addr, pmdp);
823 }
824 
825 /*
826  * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
827  *
828  *  dst - pointer to pgd range anwhere on a pgd page
829  *  src - ""
830  *  count - the number of pgds to copy.
831  *
832  * dst and src can be on the same page, but the range must not overlap,
833  * and must not cross a page boundary.
834  */
835 static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
836 {
837        memcpy(dst, src, count * sizeof(pgd_t));
838 }
839 
840 #define PTE_SHIFT ilog2(PTRS_PER_PTE)
841 static inline int page_level_shift(enum pg_level level)
842 {
843 	return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT;
844 }
845 static inline unsigned long page_level_size(enum pg_level level)
846 {
847 	return 1UL << page_level_shift(level);
848 }
849 static inline unsigned long page_level_mask(enum pg_level level)
850 {
851 	return ~(page_level_size(level) - 1);
852 }
853 
854 /*
855  * The x86 doesn't have any external MMU info: the kernel page
856  * tables contain all the necessary information.
857  */
858 static inline void update_mmu_cache(struct vm_area_struct *vma,
859 		unsigned long addr, pte_t *ptep)
860 {
861 }
862 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
863 		unsigned long addr, pmd_t *pmd)
864 {
865 }
866 
867 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
868 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
869 {
870 	return pte_set_flags(pte, _PAGE_SWP_SOFT_DIRTY);
871 }
872 
873 static inline int pte_swp_soft_dirty(pte_t pte)
874 {
875 	return pte_flags(pte) & _PAGE_SWP_SOFT_DIRTY;
876 }
877 
878 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
879 {
880 	return pte_clear_flags(pte, _PAGE_SWP_SOFT_DIRTY);
881 }
882 #endif
883 
884 #include <asm-generic/pgtable.h>
885 #endif	/* __ASSEMBLY__ */
886 
887 #endif /* _ASM_X86_PGTABLE_H */
888