xref: /openbmc/linux/arch/x86/include/asm/pgtable.h (revision 93d90ad7)
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_file(pte_t pte)
119 {
120 	return pte_flags(pte) & _PAGE_FILE;
121 }
122 
123 static inline int pte_huge(pte_t pte)
124 {
125 	return pte_flags(pte) & _PAGE_PSE;
126 }
127 
128 static inline int pte_global(pte_t pte)
129 {
130 	return pte_flags(pte) & _PAGE_GLOBAL;
131 }
132 
133 static inline int pte_exec(pte_t pte)
134 {
135 	return !(pte_flags(pte) & _PAGE_NX);
136 }
137 
138 static inline int pte_special(pte_t pte)
139 {
140 	/*
141 	 * See CONFIG_NUMA_BALANCING pte_numa in include/asm-generic/pgtable.h.
142 	 * On x86 we have _PAGE_BIT_NUMA == _PAGE_BIT_GLOBAL+1 ==
143 	 * __PAGE_BIT_SOFTW1 == _PAGE_BIT_SPECIAL.
144 	 */
145 	return (pte_flags(pte) & _PAGE_SPECIAL) &&
146 		(pte_flags(pte) & (_PAGE_PRESENT|_PAGE_PROTNONE));
147 }
148 
149 static inline unsigned long pte_pfn(pte_t pte)
150 {
151 	return (pte_val(pte) & PTE_PFN_MASK) >> PAGE_SHIFT;
152 }
153 
154 static inline unsigned long pmd_pfn(pmd_t pmd)
155 {
156 	return (pmd_val(pmd) & PTE_PFN_MASK) >> PAGE_SHIFT;
157 }
158 
159 static inline unsigned long pud_pfn(pud_t pud)
160 {
161 	return (pud_val(pud) & PTE_PFN_MASK) >> PAGE_SHIFT;
162 }
163 
164 #define pte_page(pte)	pfn_to_page(pte_pfn(pte))
165 
166 static inline int pmd_large(pmd_t pte)
167 {
168 	return pmd_flags(pte) & _PAGE_PSE;
169 }
170 
171 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
172 static inline int pmd_trans_splitting(pmd_t pmd)
173 {
174 	return pmd_val(pmd) & _PAGE_SPLITTING;
175 }
176 
177 static inline int pmd_trans_huge(pmd_t pmd)
178 {
179 	return pmd_val(pmd) & _PAGE_PSE;
180 }
181 
182 static inline int has_transparent_hugepage(void)
183 {
184 	return cpu_has_pse;
185 }
186 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
187 
188 static inline pte_t pte_set_flags(pte_t pte, pteval_t set)
189 {
190 	pteval_t v = native_pte_val(pte);
191 
192 	return native_make_pte(v | set);
193 }
194 
195 static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear)
196 {
197 	pteval_t v = native_pte_val(pte);
198 
199 	return native_make_pte(v & ~clear);
200 }
201 
202 static inline pte_t pte_mkclean(pte_t pte)
203 {
204 	return pte_clear_flags(pte, _PAGE_DIRTY);
205 }
206 
207 static inline pte_t pte_mkold(pte_t pte)
208 {
209 	return pte_clear_flags(pte, _PAGE_ACCESSED);
210 }
211 
212 static inline pte_t pte_wrprotect(pte_t pte)
213 {
214 	return pte_clear_flags(pte, _PAGE_RW);
215 }
216 
217 static inline pte_t pte_mkexec(pte_t pte)
218 {
219 	return pte_clear_flags(pte, _PAGE_NX);
220 }
221 
222 static inline pte_t pte_mkdirty(pte_t pte)
223 {
224 	return pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
225 }
226 
227 static inline pte_t pte_mkyoung(pte_t pte)
228 {
229 	return pte_set_flags(pte, _PAGE_ACCESSED);
230 }
231 
232 static inline pte_t pte_mkwrite(pte_t pte)
233 {
234 	return pte_set_flags(pte, _PAGE_RW);
235 }
236 
237 static inline pte_t pte_mkhuge(pte_t pte)
238 {
239 	return pte_set_flags(pte, _PAGE_PSE);
240 }
241 
242 static inline pte_t pte_clrhuge(pte_t pte)
243 {
244 	return pte_clear_flags(pte, _PAGE_PSE);
245 }
246 
247 static inline pte_t pte_mkglobal(pte_t pte)
248 {
249 	return pte_set_flags(pte, _PAGE_GLOBAL);
250 }
251 
252 static inline pte_t pte_clrglobal(pte_t pte)
253 {
254 	return pte_clear_flags(pte, _PAGE_GLOBAL);
255 }
256 
257 static inline pte_t pte_mkspecial(pte_t pte)
258 {
259 	return pte_set_flags(pte, _PAGE_SPECIAL);
260 }
261 
262 static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
263 {
264 	pmdval_t v = native_pmd_val(pmd);
265 
266 	return __pmd(v | set);
267 }
268 
269 static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
270 {
271 	pmdval_t v = native_pmd_val(pmd);
272 
273 	return __pmd(v & ~clear);
274 }
275 
276 static inline pmd_t pmd_mkold(pmd_t pmd)
277 {
278 	return pmd_clear_flags(pmd, _PAGE_ACCESSED);
279 }
280 
281 static inline pmd_t pmd_wrprotect(pmd_t pmd)
282 {
283 	return pmd_clear_flags(pmd, _PAGE_RW);
284 }
285 
286 static inline pmd_t pmd_mkdirty(pmd_t pmd)
287 {
288 	return pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
289 }
290 
291 static inline pmd_t pmd_mkhuge(pmd_t pmd)
292 {
293 	return pmd_set_flags(pmd, _PAGE_PSE);
294 }
295 
296 static inline pmd_t pmd_mkyoung(pmd_t pmd)
297 {
298 	return pmd_set_flags(pmd, _PAGE_ACCESSED);
299 }
300 
301 static inline pmd_t pmd_mkwrite(pmd_t pmd)
302 {
303 	return pmd_set_flags(pmd, _PAGE_RW);
304 }
305 
306 static inline pmd_t pmd_mknotpresent(pmd_t pmd)
307 {
308 	return pmd_clear_flags(pmd, _PAGE_PRESENT);
309 }
310 
311 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
312 static inline int pte_soft_dirty(pte_t pte)
313 {
314 	return pte_flags(pte) & _PAGE_SOFT_DIRTY;
315 }
316 
317 static inline int pmd_soft_dirty(pmd_t pmd)
318 {
319 	return pmd_flags(pmd) & _PAGE_SOFT_DIRTY;
320 }
321 
322 static inline pte_t pte_mksoft_dirty(pte_t pte)
323 {
324 	return pte_set_flags(pte, _PAGE_SOFT_DIRTY);
325 }
326 
327 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
328 {
329 	return pmd_set_flags(pmd, _PAGE_SOFT_DIRTY);
330 }
331 
332 static inline pte_t pte_file_clear_soft_dirty(pte_t pte)
333 {
334 	return pte_clear_flags(pte, _PAGE_SOFT_DIRTY);
335 }
336 
337 static inline pte_t pte_file_mksoft_dirty(pte_t pte)
338 {
339 	return pte_set_flags(pte, _PAGE_SOFT_DIRTY);
340 }
341 
342 static inline int pte_file_soft_dirty(pte_t pte)
343 {
344 	return pte_flags(pte) & _PAGE_SOFT_DIRTY;
345 }
346 
347 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
348 
349 /*
350  * Mask out unsupported bits in a present pgprot.  Non-present pgprots
351  * can use those bits for other purposes, so leave them be.
352  */
353 static inline pgprotval_t massage_pgprot(pgprot_t pgprot)
354 {
355 	pgprotval_t protval = pgprot_val(pgprot);
356 
357 	if (protval & _PAGE_PRESENT)
358 		protval &= __supported_pte_mask;
359 
360 	return protval;
361 }
362 
363 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
364 {
365 	return __pte(((phys_addr_t)page_nr << PAGE_SHIFT) |
366 		     massage_pgprot(pgprot));
367 }
368 
369 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
370 {
371 	return __pmd(((phys_addr_t)page_nr << PAGE_SHIFT) |
372 		     massage_pgprot(pgprot));
373 }
374 
375 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
376 {
377 	pteval_t val = pte_val(pte);
378 
379 	/*
380 	 * Chop off the NX bit (if present), and add the NX portion of
381 	 * the newprot (if present):
382 	 */
383 	val &= _PAGE_CHG_MASK;
384 	val |= massage_pgprot(newprot) & ~_PAGE_CHG_MASK;
385 
386 	return __pte(val);
387 }
388 
389 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
390 {
391 	pmdval_t val = pmd_val(pmd);
392 
393 	val &= _HPAGE_CHG_MASK;
394 	val |= massage_pgprot(newprot) & ~_HPAGE_CHG_MASK;
395 
396 	return __pmd(val);
397 }
398 
399 /* mprotect needs to preserve PAT bits when updating vm_page_prot */
400 #define pgprot_modify pgprot_modify
401 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
402 {
403 	pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK;
404 	pgprotval_t addbits = pgprot_val(newprot);
405 	return __pgprot(preservebits | addbits);
406 }
407 
408 #define pte_pgprot(x) __pgprot(pte_flags(x) & PTE_FLAGS_MASK)
409 
410 #define canon_pgprot(p) __pgprot(massage_pgprot(p))
411 
412 static inline int is_new_memtype_allowed(u64 paddr, unsigned long size,
413 					 enum page_cache_mode pcm,
414 					 enum page_cache_mode new_pcm)
415 {
416 	/*
417 	 * PAT type is always WB for untracked ranges, so no need to check.
418 	 */
419 	if (x86_platform.is_untracked_pat_range(paddr, paddr + size))
420 		return 1;
421 
422 	/*
423 	 * Certain new memtypes are not allowed with certain
424 	 * requested memtype:
425 	 * - request is uncached, return cannot be write-back
426 	 * - request is write-combine, return cannot be write-back
427 	 */
428 	if ((pcm == _PAGE_CACHE_MODE_UC_MINUS &&
429 	     new_pcm == _PAGE_CACHE_MODE_WB) ||
430 	    (pcm == _PAGE_CACHE_MODE_WC &&
431 	     new_pcm == _PAGE_CACHE_MODE_WB)) {
432 		return 0;
433 	}
434 
435 	return 1;
436 }
437 
438 pmd_t *populate_extra_pmd(unsigned long vaddr);
439 pte_t *populate_extra_pte(unsigned long vaddr);
440 #endif	/* __ASSEMBLY__ */
441 
442 #ifdef CONFIG_X86_32
443 # include <asm/pgtable_32.h>
444 #else
445 # include <asm/pgtable_64.h>
446 #endif
447 
448 #ifndef __ASSEMBLY__
449 #include <linux/mm_types.h>
450 #include <linux/mmdebug.h>
451 #include <linux/log2.h>
452 
453 static inline int pte_none(pte_t pte)
454 {
455 	return !pte.pte;
456 }
457 
458 #define __HAVE_ARCH_PTE_SAME
459 static inline int pte_same(pte_t a, pte_t b)
460 {
461 	return a.pte == b.pte;
462 }
463 
464 static inline int pte_present(pte_t a)
465 {
466 	return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE |
467 			       _PAGE_NUMA);
468 }
469 
470 #define pte_present_nonuma pte_present_nonuma
471 static inline int pte_present_nonuma(pte_t a)
472 {
473 	return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE);
474 }
475 
476 #define pte_accessible pte_accessible
477 static inline bool pte_accessible(struct mm_struct *mm, pte_t a)
478 {
479 	if (pte_flags(a) & _PAGE_PRESENT)
480 		return true;
481 
482 	if ((pte_flags(a) & (_PAGE_PROTNONE | _PAGE_NUMA)) &&
483 			mm_tlb_flush_pending(mm))
484 		return true;
485 
486 	return false;
487 }
488 
489 static inline int pte_hidden(pte_t pte)
490 {
491 	return pte_flags(pte) & _PAGE_HIDDEN;
492 }
493 
494 static inline int pmd_present(pmd_t pmd)
495 {
496 	/*
497 	 * Checking for _PAGE_PSE is needed too because
498 	 * split_huge_page will temporarily clear the present bit (but
499 	 * the _PAGE_PSE flag will remain set at all times while the
500 	 * _PAGE_PRESENT bit is clear).
501 	 */
502 	return pmd_flags(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PSE |
503 				 _PAGE_NUMA);
504 }
505 
506 static inline int pmd_none(pmd_t pmd)
507 {
508 	/* Only check low word on 32-bit platforms, since it might be
509 	   out of sync with upper half. */
510 	return (unsigned long)native_pmd_val(pmd) == 0;
511 }
512 
513 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
514 {
515 	return (unsigned long)__va(pmd_val(pmd) & PTE_PFN_MASK);
516 }
517 
518 /*
519  * Currently stuck as a macro due to indirect forward reference to
520  * linux/mmzone.h's __section_mem_map_addr() definition:
521  */
522 #define pmd_page(pmd)	pfn_to_page((pmd_val(pmd) & PTE_PFN_MASK) >> PAGE_SHIFT)
523 
524 /*
525  * the pmd page can be thought of an array like this: pmd_t[PTRS_PER_PMD]
526  *
527  * this macro returns the index of the entry in the pmd page which would
528  * control the given virtual address
529  */
530 static inline unsigned long pmd_index(unsigned long address)
531 {
532 	return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
533 }
534 
535 /*
536  * Conversion functions: convert a page and protection to a page entry,
537  * and a page entry and page directory to the page they refer to.
538  *
539  * (Currently stuck as a macro because of indirect forward reference
540  * to linux/mm.h:page_to_nid())
541  */
542 #define mk_pte(page, pgprot)   pfn_pte(page_to_pfn(page), (pgprot))
543 
544 /*
545  * the pte page can be thought of an array like this: pte_t[PTRS_PER_PTE]
546  *
547  * this function returns the index of the entry in the pte page which would
548  * control the given virtual address
549  */
550 static inline unsigned long pte_index(unsigned long address)
551 {
552 	return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
553 }
554 
555 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
556 {
557 	return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
558 }
559 
560 static inline int pmd_bad(pmd_t pmd)
561 {
562 #ifdef CONFIG_NUMA_BALANCING
563 	/* pmd_numa check */
564 	if ((pmd_flags(pmd) & (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA)
565 		return 0;
566 #endif
567 	return (pmd_flags(pmd) & ~_PAGE_USER) != _KERNPG_TABLE;
568 }
569 
570 static inline unsigned long pages_to_mb(unsigned long npg)
571 {
572 	return npg >> (20 - PAGE_SHIFT);
573 }
574 
575 #if PAGETABLE_LEVELS > 2
576 static inline int pud_none(pud_t pud)
577 {
578 	return native_pud_val(pud) == 0;
579 }
580 
581 static inline int pud_present(pud_t pud)
582 {
583 	return pud_flags(pud) & _PAGE_PRESENT;
584 }
585 
586 static inline unsigned long pud_page_vaddr(pud_t pud)
587 {
588 	return (unsigned long)__va((unsigned long)pud_val(pud) & PTE_PFN_MASK);
589 }
590 
591 /*
592  * Currently stuck as a macro due to indirect forward reference to
593  * linux/mmzone.h's __section_mem_map_addr() definition:
594  */
595 #define pud_page(pud)		pfn_to_page(pud_val(pud) >> PAGE_SHIFT)
596 
597 /* Find an entry in the second-level page table.. */
598 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
599 {
600 	return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(address);
601 }
602 
603 static inline int pud_large(pud_t pud)
604 {
605 	return (pud_val(pud) & (_PAGE_PSE | _PAGE_PRESENT)) ==
606 		(_PAGE_PSE | _PAGE_PRESENT);
607 }
608 
609 static inline int pud_bad(pud_t pud)
610 {
611 	return (pud_flags(pud) & ~(_KERNPG_TABLE | _PAGE_USER)) != 0;
612 }
613 #else
614 static inline int pud_large(pud_t pud)
615 {
616 	return 0;
617 }
618 #endif	/* PAGETABLE_LEVELS > 2 */
619 
620 #if PAGETABLE_LEVELS > 3
621 static inline int pgd_present(pgd_t pgd)
622 {
623 	return pgd_flags(pgd) & _PAGE_PRESENT;
624 }
625 
626 static inline unsigned long pgd_page_vaddr(pgd_t pgd)
627 {
628 	return (unsigned long)__va((unsigned long)pgd_val(pgd) & PTE_PFN_MASK);
629 }
630 
631 /*
632  * Currently stuck as a macro due to indirect forward reference to
633  * linux/mmzone.h's __section_mem_map_addr() definition:
634  */
635 #define pgd_page(pgd)		pfn_to_page(pgd_val(pgd) >> PAGE_SHIFT)
636 
637 /* to find an entry in a page-table-directory. */
638 static inline unsigned long pud_index(unsigned long address)
639 {
640 	return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
641 }
642 
643 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
644 {
645 	return (pud_t *)pgd_page_vaddr(*pgd) + pud_index(address);
646 }
647 
648 static inline int pgd_bad(pgd_t pgd)
649 {
650 	return (pgd_flags(pgd) & ~_PAGE_USER) != _KERNPG_TABLE;
651 }
652 
653 static inline int pgd_none(pgd_t pgd)
654 {
655 	return !native_pgd_val(pgd);
656 }
657 #endif	/* PAGETABLE_LEVELS > 3 */
658 
659 #endif	/* __ASSEMBLY__ */
660 
661 /*
662  * the pgd page can be thought of an array like this: pgd_t[PTRS_PER_PGD]
663  *
664  * this macro returns the index of the entry in the pgd page which would
665  * control the given virtual address
666  */
667 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
668 
669 /*
670  * pgd_offset() returns a (pgd_t *)
671  * pgd_index() is used get the offset into the pgd page's array of pgd_t's;
672  */
673 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index((address)))
674 /*
675  * a shortcut which implies the use of the kernel's pgd, instead
676  * of a process's
677  */
678 #define pgd_offset_k(address) pgd_offset(&init_mm, (address))
679 
680 
681 #define KERNEL_PGD_BOUNDARY	pgd_index(PAGE_OFFSET)
682 #define KERNEL_PGD_PTRS		(PTRS_PER_PGD - KERNEL_PGD_BOUNDARY)
683 
684 #ifndef __ASSEMBLY__
685 
686 extern int direct_gbpages;
687 void init_mem_mapping(void);
688 void early_alloc_pgt_buf(void);
689 
690 /* local pte updates need not use xchg for locking */
691 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
692 {
693 	pte_t res = *ptep;
694 
695 	/* Pure native function needs no input for mm, addr */
696 	native_pte_clear(NULL, 0, ptep);
697 	return res;
698 }
699 
700 static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp)
701 {
702 	pmd_t res = *pmdp;
703 
704 	native_pmd_clear(pmdp);
705 	return res;
706 }
707 
708 static inline void native_set_pte_at(struct mm_struct *mm, unsigned long addr,
709 				     pte_t *ptep , pte_t pte)
710 {
711 	native_set_pte(ptep, pte);
712 }
713 
714 static inline void native_set_pmd_at(struct mm_struct *mm, unsigned long addr,
715 				     pmd_t *pmdp , pmd_t pmd)
716 {
717 	native_set_pmd(pmdp, pmd);
718 }
719 
720 #ifndef CONFIG_PARAVIRT
721 /*
722  * Rules for using pte_update - it must be called after any PTE update which
723  * has not been done using the set_pte / clear_pte interfaces.  It is used by
724  * shadow mode hypervisors to resynchronize the shadow page tables.  Kernel PTE
725  * updates should either be sets, clears, or set_pte_atomic for P->P
726  * transitions, which means this hook should only be called for user PTEs.
727  * This hook implies a P->P protection or access change has taken place, which
728  * requires a subsequent TLB flush.  The notification can optionally be delayed
729  * until the TLB flush event by using the pte_update_defer form of the
730  * interface, but care must be taken to assure that the flush happens while
731  * still holding the same page table lock so that the shadow and primary pages
732  * do not become out of sync on SMP.
733  */
734 #define pte_update(mm, addr, ptep)		do { } while (0)
735 #define pte_update_defer(mm, addr, ptep)	do { } while (0)
736 #endif
737 
738 /*
739  * We only update the dirty/accessed state if we set
740  * the dirty bit by hand in the kernel, since the hardware
741  * will do the accessed bit for us, and we don't want to
742  * race with other CPU's that might be updating the dirty
743  * bit at the same time.
744  */
745 struct vm_area_struct;
746 
747 #define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
748 extern int ptep_set_access_flags(struct vm_area_struct *vma,
749 				 unsigned long address, pte_t *ptep,
750 				 pte_t entry, int dirty);
751 
752 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
753 extern int ptep_test_and_clear_young(struct vm_area_struct *vma,
754 				     unsigned long addr, pte_t *ptep);
755 
756 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
757 extern int ptep_clear_flush_young(struct vm_area_struct *vma,
758 				  unsigned long address, pte_t *ptep);
759 
760 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
761 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
762 				       pte_t *ptep)
763 {
764 	pte_t pte = native_ptep_get_and_clear(ptep);
765 	pte_update(mm, addr, ptep);
766 	return pte;
767 }
768 
769 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
770 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
771 					    unsigned long addr, pte_t *ptep,
772 					    int full)
773 {
774 	pte_t pte;
775 	if (full) {
776 		/*
777 		 * Full address destruction in progress; paravirt does not
778 		 * care about updates and native needs no locking
779 		 */
780 		pte = native_local_ptep_get_and_clear(ptep);
781 	} else {
782 		pte = ptep_get_and_clear(mm, addr, ptep);
783 	}
784 	return pte;
785 }
786 
787 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
788 static inline void ptep_set_wrprotect(struct mm_struct *mm,
789 				      unsigned long addr, pte_t *ptep)
790 {
791 	clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte);
792 	pte_update(mm, addr, ptep);
793 }
794 
795 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
796 
797 #define mk_pmd(page, pgprot)   pfn_pmd(page_to_pfn(page), (pgprot))
798 
799 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
800 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
801 				 unsigned long address, pmd_t *pmdp,
802 				 pmd_t entry, int dirty);
803 
804 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
805 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
806 				     unsigned long addr, pmd_t *pmdp);
807 
808 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
809 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
810 				  unsigned long address, pmd_t *pmdp);
811 
812 
813 #define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
814 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
815 				 unsigned long addr, pmd_t *pmdp);
816 
817 #define __HAVE_ARCH_PMD_WRITE
818 static inline int pmd_write(pmd_t pmd)
819 {
820 	return pmd_flags(pmd) & _PAGE_RW;
821 }
822 
823 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR
824 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm, unsigned long addr,
825 				       pmd_t *pmdp)
826 {
827 	pmd_t pmd = native_pmdp_get_and_clear(pmdp);
828 	pmd_update(mm, addr, pmdp);
829 	return pmd;
830 }
831 
832 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
833 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
834 				      unsigned long addr, pmd_t *pmdp)
835 {
836 	clear_bit(_PAGE_BIT_RW, (unsigned long *)pmdp);
837 	pmd_update(mm, addr, pmdp);
838 }
839 
840 /*
841  * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
842  *
843  *  dst - pointer to pgd range anwhere on a pgd page
844  *  src - ""
845  *  count - the number of pgds to copy.
846  *
847  * dst and src can be on the same page, but the range must not overlap,
848  * and must not cross a page boundary.
849  */
850 static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
851 {
852        memcpy(dst, src, count * sizeof(pgd_t));
853 }
854 
855 #define PTE_SHIFT ilog2(PTRS_PER_PTE)
856 static inline int page_level_shift(enum pg_level level)
857 {
858 	return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT;
859 }
860 static inline unsigned long page_level_size(enum pg_level level)
861 {
862 	return 1UL << page_level_shift(level);
863 }
864 static inline unsigned long page_level_mask(enum pg_level level)
865 {
866 	return ~(page_level_size(level) - 1);
867 }
868 
869 /*
870  * The x86 doesn't have any external MMU info: the kernel page
871  * tables contain all the necessary information.
872  */
873 static inline void update_mmu_cache(struct vm_area_struct *vma,
874 		unsigned long addr, pte_t *ptep)
875 {
876 }
877 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
878 		unsigned long addr, pmd_t *pmd)
879 {
880 }
881 
882 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
883 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
884 {
885 	VM_BUG_ON(pte_present_nonuma(pte));
886 	return pte_set_flags(pte, _PAGE_SWP_SOFT_DIRTY);
887 }
888 
889 static inline int pte_swp_soft_dirty(pte_t pte)
890 {
891 	VM_BUG_ON(pte_present_nonuma(pte));
892 	return pte_flags(pte) & _PAGE_SWP_SOFT_DIRTY;
893 }
894 
895 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
896 {
897 	VM_BUG_ON(pte_present_nonuma(pte));
898 	return pte_clear_flags(pte, _PAGE_SWP_SOFT_DIRTY);
899 }
900 #endif
901 
902 #include <asm-generic/pgtable.h>
903 #endif	/* __ASSEMBLY__ */
904 
905 #endif /* _ASM_X86_PGTABLE_H */
906