x86/mm/pat: Don't report PAT on CPUs that don't support it
[linux/fpc-iii.git] / arch / x86 / include / asm / pgtable.h
blob585ee0d42d18fc162601ff0d8a53827f0d011f5e
1 #ifndef _ASM_X86_PGTABLE_H
2 #define _ASM_X86_PGTABLE_H
4 #include <asm/page.h>
5 #include <asm/e820.h>
7 #include <asm/pgtable_types.h>
9 /*
10 * Macro to mark a page protection value as UC-
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))
18 #ifndef __ASSEMBLY__
19 #include <asm/x86_init.h>
21 void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd);
22 void ptdump_walk_pgd_level_checkwx(void);
24 #ifdef CONFIG_DEBUG_WX
25 #define debug_checkwx() ptdump_walk_pgd_level_checkwx()
26 #else
27 #define debug_checkwx() do { } while (0)
28 #endif
31 * ZERO_PAGE is a global shared page that is always zero: used
32 * for zero-mapped memory areas etc..
34 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
35 __visible;
36 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
38 extern spinlock_t pgd_lock;
39 extern struct list_head pgd_list;
41 extern struct mm_struct *pgd_page_get_mm(struct page *page);
43 #ifdef CONFIG_PARAVIRT
44 #include <asm/paravirt.h>
45 #else /* !CONFIG_PARAVIRT */
46 #define set_pte(ptep, pte) native_set_pte(ptep, pte)
47 #define set_pte_at(mm, addr, ptep, pte) native_set_pte_at(mm, addr, ptep, pte)
48 #define set_pmd_at(mm, addr, pmdp, pmd) native_set_pmd_at(mm, addr, pmdp, pmd)
49 #define set_pud_at(mm, addr, pudp, pud) native_set_pud_at(mm, addr, pudp, pud)
51 #define set_pte_atomic(ptep, pte) \
52 native_set_pte_atomic(ptep, pte)
54 #define set_pmd(pmdp, pmd) native_set_pmd(pmdp, pmd)
56 #ifndef __PAGETABLE_PUD_FOLDED
57 #define set_pgd(pgdp, pgd) native_set_pgd(pgdp, pgd)
58 #define pgd_clear(pgd) native_pgd_clear(pgd)
59 #endif
61 #ifndef set_pud
62 # define set_pud(pudp, pud) native_set_pud(pudp, pud)
63 #endif
65 #ifndef __PAGETABLE_PUD_FOLDED
66 #define pud_clear(pud) native_pud_clear(pud)
67 #endif
69 #define pte_clear(mm, addr, ptep) native_pte_clear(mm, addr, ptep)
70 #define pmd_clear(pmd) native_pmd_clear(pmd)
72 #define pte_update(mm, addr, ptep) do { } while (0)
74 #define pgd_val(x) native_pgd_val(x)
75 #define __pgd(x) native_make_pgd(x)
77 #ifndef __PAGETABLE_PUD_FOLDED
78 #define pud_val(x) native_pud_val(x)
79 #define __pud(x) native_make_pud(x)
80 #endif
82 #ifndef __PAGETABLE_PMD_FOLDED
83 #define pmd_val(x) native_pmd_val(x)
84 #define __pmd(x) native_make_pmd(x)
85 #endif
87 #define pte_val(x) native_pte_val(x)
88 #define __pte(x) native_make_pte(x)
90 #define arch_end_context_switch(prev) do {} while(0)
92 #endif /* CONFIG_PARAVIRT */
95 * The following only work if pte_present() is true.
96 * Undefined behaviour if not..
98 static inline int pte_dirty(pte_t pte)
100 return pte_flags(pte) & _PAGE_DIRTY;
104 static inline u32 read_pkru(void)
106 if (boot_cpu_has(X86_FEATURE_OSPKE))
107 return __read_pkru();
108 return 0;
111 static inline void write_pkru(u32 pkru)
113 if (boot_cpu_has(X86_FEATURE_OSPKE))
114 __write_pkru(pkru);
117 static inline int pte_young(pte_t pte)
119 return pte_flags(pte) & _PAGE_ACCESSED;
122 static inline int pmd_dirty(pmd_t pmd)
124 return pmd_flags(pmd) & _PAGE_DIRTY;
127 static inline int pmd_young(pmd_t pmd)
129 return pmd_flags(pmd) & _PAGE_ACCESSED;
132 static inline int pud_dirty(pud_t pud)
134 return pud_flags(pud) & _PAGE_DIRTY;
137 static inline int pud_young(pud_t pud)
139 return pud_flags(pud) & _PAGE_ACCESSED;
142 static inline int pte_write(pte_t pte)
144 return pte_flags(pte) & _PAGE_RW;
147 static inline int pte_huge(pte_t pte)
149 return pte_flags(pte) & _PAGE_PSE;
152 static inline int pte_global(pte_t pte)
154 return pte_flags(pte) & _PAGE_GLOBAL;
157 static inline int pte_exec(pte_t pte)
159 return !(pte_flags(pte) & _PAGE_NX);
162 static inline int pte_special(pte_t pte)
164 return pte_flags(pte) & _PAGE_SPECIAL;
167 static inline unsigned long pte_pfn(pte_t pte)
169 return (pte_val(pte) & PTE_PFN_MASK) >> PAGE_SHIFT;
172 static inline unsigned long pmd_pfn(pmd_t pmd)
174 return (pmd_val(pmd) & pmd_pfn_mask(pmd)) >> PAGE_SHIFT;
177 static inline unsigned long pud_pfn(pud_t pud)
179 return (pud_val(pud) & pud_pfn_mask(pud)) >> PAGE_SHIFT;
182 #define pte_page(pte) pfn_to_page(pte_pfn(pte))
184 static inline int pmd_large(pmd_t pte)
186 return pmd_flags(pte) & _PAGE_PSE;
189 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
190 static inline int pmd_trans_huge(pmd_t pmd)
192 return (pmd_val(pmd) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
195 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
196 static inline int pud_trans_huge(pud_t pud)
198 return (pud_val(pud) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
200 #endif
202 #define has_transparent_hugepage has_transparent_hugepage
203 static inline int has_transparent_hugepage(void)
205 return boot_cpu_has(X86_FEATURE_PSE);
208 #ifdef __HAVE_ARCH_PTE_DEVMAP
209 static inline int pmd_devmap(pmd_t pmd)
211 return !!(pmd_val(pmd) & _PAGE_DEVMAP);
214 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
215 static inline int pud_devmap(pud_t pud)
217 return !!(pud_val(pud) & _PAGE_DEVMAP);
219 #else
220 static inline int pud_devmap(pud_t pud)
222 return 0;
224 #endif
225 #endif
226 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
228 static inline pte_t pte_set_flags(pte_t pte, pteval_t set)
230 pteval_t v = native_pte_val(pte);
232 return native_make_pte(v | set);
235 static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear)
237 pteval_t v = native_pte_val(pte);
239 return native_make_pte(v & ~clear);
242 static inline pte_t pte_mkclean(pte_t pte)
244 return pte_clear_flags(pte, _PAGE_DIRTY);
247 static inline pte_t pte_mkold(pte_t pte)
249 return pte_clear_flags(pte, _PAGE_ACCESSED);
252 static inline pte_t pte_wrprotect(pte_t pte)
254 return pte_clear_flags(pte, _PAGE_RW);
257 static inline pte_t pte_mkexec(pte_t pte)
259 return pte_clear_flags(pte, _PAGE_NX);
262 static inline pte_t pte_mkdirty(pte_t pte)
264 return pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
267 static inline pte_t pte_mkyoung(pte_t pte)
269 return pte_set_flags(pte, _PAGE_ACCESSED);
272 static inline pte_t pte_mkwrite(pte_t pte)
274 return pte_set_flags(pte, _PAGE_RW);
277 static inline pte_t pte_mkhuge(pte_t pte)
279 return pte_set_flags(pte, _PAGE_PSE);
282 static inline pte_t pte_clrhuge(pte_t pte)
284 return pte_clear_flags(pte, _PAGE_PSE);
287 static inline pte_t pte_mkglobal(pte_t pte)
289 return pte_set_flags(pte, _PAGE_GLOBAL);
292 static inline pte_t pte_clrglobal(pte_t pte)
294 return pte_clear_flags(pte, _PAGE_GLOBAL);
297 static inline pte_t pte_mkspecial(pte_t pte)
299 return pte_set_flags(pte, _PAGE_SPECIAL);
302 static inline pte_t pte_mkdevmap(pte_t pte)
304 return pte_set_flags(pte, _PAGE_SPECIAL|_PAGE_DEVMAP);
307 static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
309 pmdval_t v = native_pmd_val(pmd);
311 return __pmd(v | set);
314 static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
316 pmdval_t v = native_pmd_val(pmd);
318 return __pmd(v & ~clear);
321 static inline pmd_t pmd_mkold(pmd_t pmd)
323 return pmd_clear_flags(pmd, _PAGE_ACCESSED);
326 static inline pmd_t pmd_mkclean(pmd_t pmd)
328 return pmd_clear_flags(pmd, _PAGE_DIRTY);
331 static inline pmd_t pmd_wrprotect(pmd_t pmd)
333 return pmd_clear_flags(pmd, _PAGE_RW);
336 static inline pmd_t pmd_mkdirty(pmd_t pmd)
338 return pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
341 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
343 return pmd_set_flags(pmd, _PAGE_DEVMAP);
346 static inline pmd_t pmd_mkhuge(pmd_t pmd)
348 return pmd_set_flags(pmd, _PAGE_PSE);
351 static inline pmd_t pmd_mkyoung(pmd_t pmd)
353 return pmd_set_flags(pmd, _PAGE_ACCESSED);
356 static inline pmd_t pmd_mkwrite(pmd_t pmd)
358 return pmd_set_flags(pmd, _PAGE_RW);
361 static inline pmd_t pmd_mknotpresent(pmd_t pmd)
363 return pmd_clear_flags(pmd, _PAGE_PRESENT | _PAGE_PROTNONE);
366 static inline pud_t pud_set_flags(pud_t pud, pudval_t set)
368 pudval_t v = native_pud_val(pud);
370 return __pud(v | set);
373 static inline pud_t pud_clear_flags(pud_t pud, pudval_t clear)
375 pudval_t v = native_pud_val(pud);
377 return __pud(v & ~clear);
380 static inline pud_t pud_mkold(pud_t pud)
382 return pud_clear_flags(pud, _PAGE_ACCESSED);
385 static inline pud_t pud_mkclean(pud_t pud)
387 return pud_clear_flags(pud, _PAGE_DIRTY);
390 static inline pud_t pud_wrprotect(pud_t pud)
392 return pud_clear_flags(pud, _PAGE_RW);
395 static inline pud_t pud_mkdirty(pud_t pud)
397 return pud_set_flags(pud, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
400 static inline pud_t pud_mkdevmap(pud_t pud)
402 return pud_set_flags(pud, _PAGE_DEVMAP);
405 static inline pud_t pud_mkhuge(pud_t pud)
407 return pud_set_flags(pud, _PAGE_PSE);
410 static inline pud_t pud_mkyoung(pud_t pud)
412 return pud_set_flags(pud, _PAGE_ACCESSED);
415 static inline pud_t pud_mkwrite(pud_t pud)
417 return pud_set_flags(pud, _PAGE_RW);
420 static inline pud_t pud_mknotpresent(pud_t pud)
422 return pud_clear_flags(pud, _PAGE_PRESENT | _PAGE_PROTNONE);
425 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
426 static inline int pte_soft_dirty(pte_t pte)
428 return pte_flags(pte) & _PAGE_SOFT_DIRTY;
431 static inline int pmd_soft_dirty(pmd_t pmd)
433 return pmd_flags(pmd) & _PAGE_SOFT_DIRTY;
436 static inline int pud_soft_dirty(pud_t pud)
438 return pud_flags(pud) & _PAGE_SOFT_DIRTY;
441 static inline pte_t pte_mksoft_dirty(pte_t pte)
443 return pte_set_flags(pte, _PAGE_SOFT_DIRTY);
446 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
448 return pmd_set_flags(pmd, _PAGE_SOFT_DIRTY);
451 static inline pud_t pud_mksoft_dirty(pud_t pud)
453 return pud_set_flags(pud, _PAGE_SOFT_DIRTY);
456 static inline pte_t pte_clear_soft_dirty(pte_t pte)
458 return pte_clear_flags(pte, _PAGE_SOFT_DIRTY);
461 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
463 return pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY);
466 static inline pud_t pud_clear_soft_dirty(pud_t pud)
468 return pud_clear_flags(pud, _PAGE_SOFT_DIRTY);
471 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
474 * Mask out unsupported bits in a present pgprot. Non-present pgprots
475 * can use those bits for other purposes, so leave them be.
477 static inline pgprotval_t massage_pgprot(pgprot_t pgprot)
479 pgprotval_t protval = pgprot_val(pgprot);
481 if (protval & _PAGE_PRESENT)
482 protval &= __supported_pte_mask;
484 return protval;
487 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
489 return __pte(((phys_addr_t)page_nr << PAGE_SHIFT) |
490 massage_pgprot(pgprot));
493 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
495 return __pmd(((phys_addr_t)page_nr << PAGE_SHIFT) |
496 massage_pgprot(pgprot));
499 static inline pud_t pfn_pud(unsigned long page_nr, pgprot_t pgprot)
501 return __pud(((phys_addr_t)page_nr << PAGE_SHIFT) |
502 massage_pgprot(pgprot));
505 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
507 pteval_t val = pte_val(pte);
510 * Chop off the NX bit (if present), and add the NX portion of
511 * the newprot (if present):
513 val &= _PAGE_CHG_MASK;
514 val |= massage_pgprot(newprot) & ~_PAGE_CHG_MASK;
516 return __pte(val);
519 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
521 pmdval_t val = pmd_val(pmd);
523 val &= _HPAGE_CHG_MASK;
524 val |= massage_pgprot(newprot) & ~_HPAGE_CHG_MASK;
526 return __pmd(val);
529 /* mprotect needs to preserve PAT bits when updating vm_page_prot */
530 #define pgprot_modify pgprot_modify
531 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
533 pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK;
534 pgprotval_t addbits = pgprot_val(newprot);
535 return __pgprot(preservebits | addbits);
538 #define pte_pgprot(x) __pgprot(pte_flags(x))
539 #define pmd_pgprot(x) __pgprot(pmd_flags(x))
540 #define pud_pgprot(x) __pgprot(pud_flags(x))
542 #define canon_pgprot(p) __pgprot(massage_pgprot(p))
544 static inline int is_new_memtype_allowed(u64 paddr, unsigned long size,
545 enum page_cache_mode pcm,
546 enum page_cache_mode new_pcm)
549 * PAT type is always WB for untracked ranges, so no need to check.
551 if (x86_platform.is_untracked_pat_range(paddr, paddr + size))
552 return 1;
555 * Certain new memtypes are not allowed with certain
556 * requested memtype:
557 * - request is uncached, return cannot be write-back
558 * - request is write-combine, return cannot be write-back
559 * - request is write-through, return cannot be write-back
560 * - request is write-through, return cannot be write-combine
562 if ((pcm == _PAGE_CACHE_MODE_UC_MINUS &&
563 new_pcm == _PAGE_CACHE_MODE_WB) ||
564 (pcm == _PAGE_CACHE_MODE_WC &&
565 new_pcm == _PAGE_CACHE_MODE_WB) ||
566 (pcm == _PAGE_CACHE_MODE_WT &&
567 new_pcm == _PAGE_CACHE_MODE_WB) ||
568 (pcm == _PAGE_CACHE_MODE_WT &&
569 new_pcm == _PAGE_CACHE_MODE_WC)) {
570 return 0;
573 return 1;
576 pmd_t *populate_extra_pmd(unsigned long vaddr);
577 pte_t *populate_extra_pte(unsigned long vaddr);
578 #endif /* __ASSEMBLY__ */
580 #ifdef CONFIG_X86_32
581 # include <asm/pgtable_32.h>
582 #else
583 # include <asm/pgtable_64.h>
584 #endif
586 #ifndef __ASSEMBLY__
587 #include <linux/mm_types.h>
588 #include <linux/mmdebug.h>
589 #include <linux/log2.h>
591 static inline int pte_none(pte_t pte)
593 return !(pte.pte & ~(_PAGE_KNL_ERRATUM_MASK));
596 #define __HAVE_ARCH_PTE_SAME
597 static inline int pte_same(pte_t a, pte_t b)
599 return a.pte == b.pte;
602 static inline int pte_present(pte_t a)
604 return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE);
607 #ifdef __HAVE_ARCH_PTE_DEVMAP
608 static inline int pte_devmap(pte_t a)
610 return (pte_flags(a) & _PAGE_DEVMAP) == _PAGE_DEVMAP;
612 #endif
614 #define pte_accessible pte_accessible
615 static inline bool pte_accessible(struct mm_struct *mm, pte_t a)
617 if (pte_flags(a) & _PAGE_PRESENT)
618 return true;
620 if ((pte_flags(a) & _PAGE_PROTNONE) &&
621 mm_tlb_flush_pending(mm))
622 return true;
624 return false;
627 static inline int pte_hidden(pte_t pte)
629 return pte_flags(pte) & _PAGE_HIDDEN;
632 static inline int pmd_present(pmd_t pmd)
635 * Checking for _PAGE_PSE is needed too because
636 * split_huge_page will temporarily clear the present bit (but
637 * the _PAGE_PSE flag will remain set at all times while the
638 * _PAGE_PRESENT bit is clear).
640 return pmd_flags(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PSE);
643 #ifdef CONFIG_NUMA_BALANCING
645 * These work without NUMA balancing but the kernel does not care. See the
646 * comment in include/asm-generic/pgtable.h
648 static inline int pte_protnone(pte_t pte)
650 return (pte_flags(pte) & (_PAGE_PROTNONE | _PAGE_PRESENT))
651 == _PAGE_PROTNONE;
654 static inline int pmd_protnone(pmd_t pmd)
656 return (pmd_flags(pmd) & (_PAGE_PROTNONE | _PAGE_PRESENT))
657 == _PAGE_PROTNONE;
659 #endif /* CONFIG_NUMA_BALANCING */
661 static inline int pmd_none(pmd_t pmd)
663 /* Only check low word on 32-bit platforms, since it might be
664 out of sync with upper half. */
665 unsigned long val = native_pmd_val(pmd);
666 return (val & ~_PAGE_KNL_ERRATUM_MASK) == 0;
669 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
671 return (unsigned long)__va(pmd_val(pmd) & pmd_pfn_mask(pmd));
675 * Currently stuck as a macro due to indirect forward reference to
676 * linux/mmzone.h's __section_mem_map_addr() definition:
678 #define pmd_page(pmd) \
679 pfn_to_page((pmd_val(pmd) & pmd_pfn_mask(pmd)) >> PAGE_SHIFT)
682 * the pmd page can be thought of an array like this: pmd_t[PTRS_PER_PMD]
684 * this macro returns the index of the entry in the pmd page which would
685 * control the given virtual address
687 static inline unsigned long pmd_index(unsigned long address)
689 return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
693 * Conversion functions: convert a page and protection to a page entry,
694 * and a page entry and page directory to the page they refer to.
696 * (Currently stuck as a macro because of indirect forward reference
697 * to linux/mm.h:page_to_nid())
699 #define mk_pte(page, pgprot) pfn_pte(page_to_pfn(page), (pgprot))
702 * the pte page can be thought of an array like this: pte_t[PTRS_PER_PTE]
704 * this function returns the index of the entry in the pte page which would
705 * control the given virtual address
707 static inline unsigned long pte_index(unsigned long address)
709 return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
712 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
714 return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
717 static inline int pmd_bad(pmd_t pmd)
719 return (pmd_flags(pmd) & ~_PAGE_USER) != _KERNPG_TABLE;
722 static inline unsigned long pages_to_mb(unsigned long npg)
724 return npg >> (20 - PAGE_SHIFT);
727 #if CONFIG_PGTABLE_LEVELS > 2
728 static inline int pud_none(pud_t pud)
730 return (native_pud_val(pud) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
733 static inline int pud_present(pud_t pud)
735 return pud_flags(pud) & _PAGE_PRESENT;
738 static inline unsigned long pud_page_vaddr(pud_t pud)
740 return (unsigned long)__va(pud_val(pud) & pud_pfn_mask(pud));
744 * Currently stuck as a macro due to indirect forward reference to
745 * linux/mmzone.h's __section_mem_map_addr() definition:
747 #define pud_page(pud) \
748 pfn_to_page((pud_val(pud) & pud_pfn_mask(pud)) >> PAGE_SHIFT)
750 /* Find an entry in the second-level page table.. */
751 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
753 return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(address);
756 static inline int pud_large(pud_t pud)
758 return (pud_val(pud) & (_PAGE_PSE | _PAGE_PRESENT)) ==
759 (_PAGE_PSE | _PAGE_PRESENT);
762 static inline int pud_bad(pud_t pud)
764 return (pud_flags(pud) & ~(_KERNPG_TABLE | _PAGE_USER)) != 0;
766 #else
767 static inline int pud_large(pud_t pud)
769 return 0;
771 #endif /* CONFIG_PGTABLE_LEVELS > 2 */
773 #if CONFIG_PGTABLE_LEVELS > 3
774 static inline int pgd_present(pgd_t pgd)
776 return pgd_flags(pgd) & _PAGE_PRESENT;
779 static inline unsigned long pgd_page_vaddr(pgd_t pgd)
781 return (unsigned long)__va((unsigned long)pgd_val(pgd) & PTE_PFN_MASK);
785 * Currently stuck as a macro due to indirect forward reference to
786 * linux/mmzone.h's __section_mem_map_addr() definition:
788 #define pgd_page(pgd) pfn_to_page(pgd_val(pgd) >> PAGE_SHIFT)
790 /* to find an entry in a page-table-directory. */
791 static inline unsigned long pud_index(unsigned long address)
793 return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
796 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
798 return (pud_t *)pgd_page_vaddr(*pgd) + pud_index(address);
801 static inline int pgd_bad(pgd_t pgd)
803 return (pgd_flags(pgd) & ~_PAGE_USER) != _KERNPG_TABLE;
806 static inline int pgd_none(pgd_t pgd)
809 * There is no need to do a workaround for the KNL stray
810 * A/D bit erratum here. PGDs only point to page tables
811 * except on 32-bit non-PAE which is not supported on
812 * KNL.
814 return !native_pgd_val(pgd);
816 #endif /* CONFIG_PGTABLE_LEVELS > 3 */
818 #endif /* __ASSEMBLY__ */
821 * the pgd page can be thought of an array like this: pgd_t[PTRS_PER_PGD]
823 * this macro returns the index of the entry in the pgd page which would
824 * control the given virtual address
826 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
829 * pgd_offset() returns a (pgd_t *)
830 * pgd_index() is used get the offset into the pgd page's array of pgd_t's;
832 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index((address)))
834 * a shortcut which implies the use of the kernel's pgd, instead
835 * of a process's
837 #define pgd_offset_k(address) pgd_offset(&init_mm, (address))
840 #define KERNEL_PGD_BOUNDARY pgd_index(PAGE_OFFSET)
841 #define KERNEL_PGD_PTRS (PTRS_PER_PGD - KERNEL_PGD_BOUNDARY)
843 #ifndef __ASSEMBLY__
845 extern int direct_gbpages;
846 void init_mem_mapping(void);
847 void early_alloc_pgt_buf(void);
849 #ifdef CONFIG_X86_64
850 /* Realmode trampoline initialization. */
851 extern pgd_t trampoline_pgd_entry;
852 static inline void __meminit init_trampoline_default(void)
854 /* Default trampoline pgd value */
855 trampoline_pgd_entry = init_level4_pgt[pgd_index(__PAGE_OFFSET)];
857 # ifdef CONFIG_RANDOMIZE_MEMORY
858 void __meminit init_trampoline(void);
859 # else
860 # define init_trampoline init_trampoline_default
861 # endif
862 #else
863 static inline void init_trampoline(void) { }
864 #endif
866 /* local pte updates need not use xchg for locking */
867 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
869 pte_t res = *ptep;
871 /* Pure native function needs no input for mm, addr */
872 native_pte_clear(NULL, 0, ptep);
873 return res;
876 static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp)
878 pmd_t res = *pmdp;
880 native_pmd_clear(pmdp);
881 return res;
884 static inline pud_t native_local_pudp_get_and_clear(pud_t *pudp)
886 pud_t res = *pudp;
888 native_pud_clear(pudp);
889 return res;
892 static inline void native_set_pte_at(struct mm_struct *mm, unsigned long addr,
893 pte_t *ptep , pte_t pte)
895 native_set_pte(ptep, pte);
898 static inline void native_set_pmd_at(struct mm_struct *mm, unsigned long addr,
899 pmd_t *pmdp , pmd_t pmd)
901 native_set_pmd(pmdp, pmd);
904 static inline void native_set_pud_at(struct mm_struct *mm, unsigned long addr,
905 pud_t *pudp, pud_t pud)
907 native_set_pud(pudp, pud);
910 #ifndef CONFIG_PARAVIRT
912 * Rules for using pte_update - it must be called after any PTE update which
913 * has not been done using the set_pte / clear_pte interfaces. It is used by
914 * shadow mode hypervisors to resynchronize the shadow page tables. Kernel PTE
915 * updates should either be sets, clears, or set_pte_atomic for P->P
916 * transitions, which means this hook should only be called for user PTEs.
917 * This hook implies a P->P protection or access change has taken place, which
918 * requires a subsequent TLB flush.
920 #define pte_update(mm, addr, ptep) do { } while (0)
921 #endif
924 * We only update the dirty/accessed state if we set
925 * the dirty bit by hand in the kernel, since the hardware
926 * will do the accessed bit for us, and we don't want to
927 * race with other CPU's that might be updating the dirty
928 * bit at the same time.
930 struct vm_area_struct;
932 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
933 extern int ptep_set_access_flags(struct vm_area_struct *vma,
934 unsigned long address, pte_t *ptep,
935 pte_t entry, int dirty);
937 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
938 extern int ptep_test_and_clear_young(struct vm_area_struct *vma,
939 unsigned long addr, pte_t *ptep);
941 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
942 extern int ptep_clear_flush_young(struct vm_area_struct *vma,
943 unsigned long address, pte_t *ptep);
945 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
946 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
947 pte_t *ptep)
949 pte_t pte = native_ptep_get_and_clear(ptep);
950 pte_update(mm, addr, ptep);
951 return pte;
954 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
955 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
956 unsigned long addr, pte_t *ptep,
957 int full)
959 pte_t pte;
960 if (full) {
962 * Full address destruction in progress; paravirt does not
963 * care about updates and native needs no locking
965 pte = native_local_ptep_get_and_clear(ptep);
966 } else {
967 pte = ptep_get_and_clear(mm, addr, ptep);
969 return pte;
972 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
973 static inline void ptep_set_wrprotect(struct mm_struct *mm,
974 unsigned long addr, pte_t *ptep)
976 clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte);
977 pte_update(mm, addr, ptep);
980 #define flush_tlb_fix_spurious_fault(vma, address) do { } while (0)
982 #define mk_pmd(page, pgprot) pfn_pmd(page_to_pfn(page), (pgprot))
984 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
985 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
986 unsigned long address, pmd_t *pmdp,
987 pmd_t entry, int dirty);
988 extern int pudp_set_access_flags(struct vm_area_struct *vma,
989 unsigned long address, pud_t *pudp,
990 pud_t entry, int dirty);
992 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
993 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
994 unsigned long addr, pmd_t *pmdp);
995 extern int pudp_test_and_clear_young(struct vm_area_struct *vma,
996 unsigned long addr, pud_t *pudp);
998 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
999 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
1000 unsigned long address, pmd_t *pmdp);
1003 #define __HAVE_ARCH_PMD_WRITE
1004 static inline int pmd_write(pmd_t pmd)
1006 return pmd_flags(pmd) & _PAGE_RW;
1009 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1010 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, unsigned long addr,
1011 pmd_t *pmdp)
1013 return native_pmdp_get_and_clear(pmdp);
1016 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
1017 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
1018 unsigned long addr, pud_t *pudp)
1020 return native_pudp_get_and_clear(pudp);
1023 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1024 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1025 unsigned long addr, pmd_t *pmdp)
1027 clear_bit(_PAGE_BIT_RW, (unsigned long *)pmdp);
1031 * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
1033 * dst - pointer to pgd range anwhere on a pgd page
1034 * src - ""
1035 * count - the number of pgds to copy.
1037 * dst and src can be on the same page, but the range must not overlap,
1038 * and must not cross a page boundary.
1040 static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
1042 memcpy(dst, src, count * sizeof(pgd_t));
1045 #define PTE_SHIFT ilog2(PTRS_PER_PTE)
1046 static inline int page_level_shift(enum pg_level level)
1048 return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT;
1050 static inline unsigned long page_level_size(enum pg_level level)
1052 return 1UL << page_level_shift(level);
1054 static inline unsigned long page_level_mask(enum pg_level level)
1056 return ~(page_level_size(level) - 1);
1060 * The x86 doesn't have any external MMU info: the kernel page
1061 * tables contain all the necessary information.
1063 static inline void update_mmu_cache(struct vm_area_struct *vma,
1064 unsigned long addr, pte_t *ptep)
1067 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1068 unsigned long addr, pmd_t *pmd)
1071 static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
1072 unsigned long addr, pud_t *pud)
1076 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1077 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1079 return pte_set_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1082 static inline int pte_swp_soft_dirty(pte_t pte)
1084 return pte_flags(pte) & _PAGE_SWP_SOFT_DIRTY;
1087 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1089 return pte_clear_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1091 #endif
1093 #define PKRU_AD_BIT 0x1
1094 #define PKRU_WD_BIT 0x2
1095 #define PKRU_BITS_PER_PKEY 2
1097 static inline bool __pkru_allows_read(u32 pkru, u16 pkey)
1099 int pkru_pkey_bits = pkey * PKRU_BITS_PER_PKEY;
1100 return !(pkru & (PKRU_AD_BIT << pkru_pkey_bits));
1103 static inline bool __pkru_allows_write(u32 pkru, u16 pkey)
1105 int pkru_pkey_bits = pkey * PKRU_BITS_PER_PKEY;
1107 * Access-disable disables writes too so we need to check
1108 * both bits here.
1110 return !(pkru & ((PKRU_AD_BIT|PKRU_WD_BIT) << pkru_pkey_bits));
1113 static inline u16 pte_flags_pkey(unsigned long pte_flags)
1115 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1116 /* ifdef to avoid doing 59-bit shift on 32-bit values */
1117 return (pte_flags & _PAGE_PKEY_MASK) >> _PAGE_BIT_PKEY_BIT0;
1118 #else
1119 return 0;
1120 #endif
1123 #include <asm-generic/pgtable.h>
1124 #endif /* __ASSEMBLY__ */
1126 #endif /* _ASM_X86_PGTABLE_H */