1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_PGTABLE_H
3 #define _ASM_X86_PGTABLE_H
5 #include <linux/mem_encrypt.h>
7 #include <asm/pgtable_types.h>
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))) \
19 #include <linux/spinlock.h>
20 #include <asm/x86_init.h>
22 #include <asm/fpu/api.h>
24 #include <asm-generic/pgtable_uffd.h>
25 #include <linux/page_table_check.h>
27 extern pgd_t early_top_pgt[PTRS_PER_PGD];
28 bool __init __early_make_pgtable(unsigned long address, pmdval_t pmd);
31 void ptdump_walk_pgd_level(struct seq_file *m, struct mm_struct *mm);
32 void ptdump_walk_pgd_level_debugfs(struct seq_file *m, struct mm_struct *mm,
34 void ptdump_walk_pgd_level_checkwx(void);
35 void ptdump_walk_user_pgd_level_checkwx(void);
38 * Macros to add or remove encryption attribute
40 #define pgprot_encrypted(prot) __pgprot(cc_mkenc(pgprot_val(prot)))
41 #define pgprot_decrypted(prot) __pgprot(cc_mkdec(pgprot_val(prot)))
43 #ifdef CONFIG_DEBUG_WX
44 #define debug_checkwx() ptdump_walk_pgd_level_checkwx()
45 #define debug_checkwx_user() ptdump_walk_user_pgd_level_checkwx()
47 #define debug_checkwx() do { } while (0)
48 #define debug_checkwx_user() do { } while (0)
52 * ZERO_PAGE is a global shared page that is always zero: used
53 * for zero-mapped memory areas etc..
55 extern unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
57 #define ZERO_PAGE(vaddr) ((void)(vaddr),virt_to_page(empty_zero_page))
59 extern spinlock_t pgd_lock;
60 extern struct list_head pgd_list;
62 extern struct mm_struct *pgd_page_get_mm(struct page *page);
64 extern pmdval_t early_pmd_flags;
66 #ifdef CONFIG_PARAVIRT_XXL
67 #include <asm/paravirt.h>
68 #else /* !CONFIG_PARAVIRT_XXL */
69 #define set_pte(ptep, pte) native_set_pte(ptep, pte)
71 #define set_pte_atomic(ptep, pte) \
72 native_set_pte_atomic(ptep, pte)
74 #define set_pmd(pmdp, pmd) native_set_pmd(pmdp, pmd)
76 #ifndef __PAGETABLE_P4D_FOLDED
77 #define set_pgd(pgdp, pgd) native_set_pgd(pgdp, pgd)
78 #define pgd_clear(pgd) (pgtable_l5_enabled() ? native_pgd_clear(pgd) : 0)
82 # define set_p4d(p4dp, p4d) native_set_p4d(p4dp, p4d)
85 #ifndef __PAGETABLE_PUD_FOLDED
86 #define p4d_clear(p4d) native_p4d_clear(p4d)
90 # define set_pud(pudp, pud) native_set_pud(pudp, pud)
93 #ifndef __PAGETABLE_PUD_FOLDED
94 #define pud_clear(pud) native_pud_clear(pud)
97 #define pte_clear(mm, addr, ptep) native_pte_clear(mm, addr, ptep)
98 #define pmd_clear(pmd) native_pmd_clear(pmd)
100 #define pgd_val(x) native_pgd_val(x)
101 #define __pgd(x) native_make_pgd(x)
103 #ifndef __PAGETABLE_P4D_FOLDED
104 #define p4d_val(x) native_p4d_val(x)
105 #define __p4d(x) native_make_p4d(x)
108 #ifndef __PAGETABLE_PUD_FOLDED
109 #define pud_val(x) native_pud_val(x)
110 #define __pud(x) native_make_pud(x)
113 #ifndef __PAGETABLE_PMD_FOLDED
114 #define pmd_val(x) native_pmd_val(x)
115 #define __pmd(x) native_make_pmd(x)
118 #define pte_val(x) native_pte_val(x)
119 #define __pte(x) native_make_pte(x)
121 #define arch_end_context_switch(prev) do {} while(0)
122 #endif /* CONFIG_PARAVIRT_XXL */
125 * The following only work if pte_present() is true.
126 * Undefined behaviour if not..
128 static inline bool pte_dirty(pte_t pte)
130 return pte_flags(pte) & _PAGE_DIRTY_BITS;
133 static inline bool pte_shstk(pte_t pte)
135 return cpu_feature_enabled(X86_FEATURE_SHSTK) &&
136 (pte_flags(pte) & (_PAGE_RW | _PAGE_DIRTY)) == _PAGE_DIRTY;
139 static inline int pte_young(pte_t pte)
141 return pte_flags(pte) & _PAGE_ACCESSED;
144 static inline bool pmd_dirty(pmd_t pmd)
146 return pmd_flags(pmd) & _PAGE_DIRTY_BITS;
149 static inline bool pmd_shstk(pmd_t pmd)
151 return cpu_feature_enabled(X86_FEATURE_SHSTK) &&
152 (pmd_flags(pmd) & (_PAGE_RW | _PAGE_DIRTY | _PAGE_PSE)) ==
153 (_PAGE_DIRTY | _PAGE_PSE);
156 #define pmd_young pmd_young
157 static inline int pmd_young(pmd_t pmd)
159 return pmd_flags(pmd) & _PAGE_ACCESSED;
162 static inline bool pud_dirty(pud_t pud)
164 return pud_flags(pud) & _PAGE_DIRTY_BITS;
167 static inline int pud_young(pud_t pud)
169 return pud_flags(pud) & _PAGE_ACCESSED;
172 static inline int pte_write(pte_t pte)
175 * Shadow stack pages are logically writable, but do not have
176 * _PAGE_RW. Check for them separately from _PAGE_RW itself.
178 return (pte_flags(pte) & _PAGE_RW) || pte_shstk(pte);
181 #define pmd_write pmd_write
182 static inline int pmd_write(pmd_t pmd)
185 * Shadow stack pages are logically writable, but do not have
186 * _PAGE_RW. Check for them separately from _PAGE_RW itself.
188 return (pmd_flags(pmd) & _PAGE_RW) || pmd_shstk(pmd);
191 #define pud_write pud_write
192 static inline int pud_write(pud_t pud)
194 return pud_flags(pud) & _PAGE_RW;
197 static inline int pte_huge(pte_t pte)
199 return pte_flags(pte) & _PAGE_PSE;
202 static inline int pte_global(pte_t pte)
204 return pte_flags(pte) & _PAGE_GLOBAL;
207 static inline int pte_exec(pte_t pte)
209 return !(pte_flags(pte) & _PAGE_NX);
212 static inline int pte_special(pte_t pte)
214 return pte_flags(pte) & _PAGE_SPECIAL;
217 /* Entries that were set to PROT_NONE are inverted */
219 static inline u64 protnone_mask(u64 val);
221 #define PFN_PTE_SHIFT PAGE_SHIFT
223 static inline unsigned long pte_pfn(pte_t pte)
225 phys_addr_t pfn = pte_val(pte);
226 pfn ^= protnone_mask(pfn);
227 return (pfn & PTE_PFN_MASK) >> PAGE_SHIFT;
230 static inline unsigned long pmd_pfn(pmd_t pmd)
232 phys_addr_t pfn = pmd_val(pmd);
233 pfn ^= protnone_mask(pfn);
234 return (pfn & pmd_pfn_mask(pmd)) >> PAGE_SHIFT;
237 static inline unsigned long pud_pfn(pud_t pud)
239 phys_addr_t pfn = pud_val(pud);
240 pfn ^= protnone_mask(pfn);
241 return (pfn & pud_pfn_mask(pud)) >> PAGE_SHIFT;
244 static inline unsigned long p4d_pfn(p4d_t p4d)
246 return (p4d_val(p4d) & p4d_pfn_mask(p4d)) >> PAGE_SHIFT;
249 static inline unsigned long pgd_pfn(pgd_t pgd)
251 return (pgd_val(pgd) & PTE_PFN_MASK) >> PAGE_SHIFT;
254 #define p4d_leaf p4d_large
255 static inline int p4d_large(p4d_t p4d)
257 /* No 512 GiB pages yet */
261 #define pte_page(pte) pfn_to_page(pte_pfn(pte))
263 #define pmd_leaf pmd_large
264 static inline int pmd_large(pmd_t pte)
266 return pmd_flags(pte) & _PAGE_PSE;
269 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
270 /* NOTE: when predicate huge page, consider also pmd_devmap, or use pmd_large */
271 static inline int pmd_trans_huge(pmd_t pmd)
273 return (pmd_val(pmd) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
276 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
277 static inline int pud_trans_huge(pud_t pud)
279 return (pud_val(pud) & (_PAGE_PSE|_PAGE_DEVMAP)) == _PAGE_PSE;
283 #define has_transparent_hugepage has_transparent_hugepage
284 static inline int has_transparent_hugepage(void)
286 return boot_cpu_has(X86_FEATURE_PSE);
289 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
290 static inline int pmd_devmap(pmd_t pmd)
292 return !!(pmd_val(pmd) & _PAGE_DEVMAP);
295 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
296 static inline int pud_devmap(pud_t pud)
298 return !!(pud_val(pud) & _PAGE_DEVMAP);
301 static inline int pud_devmap(pud_t pud)
307 static inline int pgd_devmap(pgd_t pgd)
312 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
314 static inline pte_t pte_set_flags(pte_t pte, pteval_t set)
316 pteval_t v = native_pte_val(pte);
318 return native_make_pte(v | set);
321 static inline pte_t pte_clear_flags(pte_t pte, pteval_t clear)
323 pteval_t v = native_pte_val(pte);
325 return native_make_pte(v & ~clear);
329 * Write protection operations can result in Dirty=1,Write=0 PTEs. But in the
330 * case of X86_FEATURE_USER_SHSTK, these PTEs denote shadow stack memory. So
331 * when creating dirty, write-protected memory, a software bit is used:
332 * _PAGE_BIT_SAVED_DIRTY. The following functions take a PTE and transition the
333 * Dirty bit to SavedDirty, and vice-vesra.
335 * This shifting is only done if needed. In the case of shifting
336 * Dirty->SavedDirty, the condition is if the PTE is Write=0. In the case of
337 * shifting SavedDirty->Dirty, the condition is Write=1.
339 static inline pgprotval_t mksaveddirty_shift(pgprotval_t v)
341 pgprotval_t cond = (~v >> _PAGE_BIT_RW) & 1;
343 v |= ((v >> _PAGE_BIT_DIRTY) & cond) << _PAGE_BIT_SAVED_DIRTY;
344 v &= ~(cond << _PAGE_BIT_DIRTY);
349 static inline pgprotval_t clear_saveddirty_shift(pgprotval_t v)
351 pgprotval_t cond = (v >> _PAGE_BIT_RW) & 1;
353 v |= ((v >> _PAGE_BIT_SAVED_DIRTY) & cond) << _PAGE_BIT_DIRTY;
354 v &= ~(cond << _PAGE_BIT_SAVED_DIRTY);
359 static inline pte_t pte_mksaveddirty(pte_t pte)
361 pteval_t v = native_pte_val(pte);
363 v = mksaveddirty_shift(v);
364 return native_make_pte(v);
367 static inline pte_t pte_clear_saveddirty(pte_t pte)
369 pteval_t v = native_pte_val(pte);
371 v = clear_saveddirty_shift(v);
372 return native_make_pte(v);
375 static inline pte_t pte_wrprotect(pte_t pte)
377 pte = pte_clear_flags(pte, _PAGE_RW);
380 * Blindly clearing _PAGE_RW might accidentally create
381 * a shadow stack PTE (Write=0,Dirty=1). Move the hardware
382 * dirty value to the software bit, if present.
384 return pte_mksaveddirty(pte);
387 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
388 static inline int pte_uffd_wp(pte_t pte)
390 bool wp = pte_flags(pte) & _PAGE_UFFD_WP;
392 #ifdef CONFIG_DEBUG_VM
394 * Having write bit for wr-protect-marked present ptes is fatal,
395 * because it means the uffd-wp bit will be ignored and write will
398 * Use any chance of pgtable walking to verify this (e.g., when
399 * page swapped out or being migrated for all purposes). It means
400 * something is already wrong. Tell the admin even before the
401 * process crashes. We also nail it with wrong pgtable setup.
403 WARN_ON_ONCE(wp && pte_write(pte));
409 static inline pte_t pte_mkuffd_wp(pte_t pte)
411 return pte_wrprotect(pte_set_flags(pte, _PAGE_UFFD_WP));
414 static inline pte_t pte_clear_uffd_wp(pte_t pte)
416 return pte_clear_flags(pte, _PAGE_UFFD_WP);
418 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
420 static inline pte_t pte_mkclean(pte_t pte)
422 return pte_clear_flags(pte, _PAGE_DIRTY_BITS);
425 static inline pte_t pte_mkold(pte_t pte)
427 return pte_clear_flags(pte, _PAGE_ACCESSED);
430 static inline pte_t pte_mkexec(pte_t pte)
432 return pte_clear_flags(pte, _PAGE_NX);
435 static inline pte_t pte_mkdirty(pte_t pte)
437 pte = pte_set_flags(pte, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
439 return pte_mksaveddirty(pte);
442 static inline pte_t pte_mkwrite_shstk(pte_t pte)
444 pte = pte_clear_flags(pte, _PAGE_RW);
446 return pte_set_flags(pte, _PAGE_DIRTY);
449 static inline pte_t pte_mkyoung(pte_t pte)
451 return pte_set_flags(pte, _PAGE_ACCESSED);
454 static inline pte_t pte_mkwrite_novma(pte_t pte)
456 return pte_set_flags(pte, _PAGE_RW);
459 struct vm_area_struct;
460 pte_t pte_mkwrite(pte_t pte, struct vm_area_struct *vma);
461 #define pte_mkwrite pte_mkwrite
463 static inline pte_t pte_mkhuge(pte_t pte)
465 return pte_set_flags(pte, _PAGE_PSE);
468 static inline pte_t pte_clrhuge(pte_t pte)
470 return pte_clear_flags(pte, _PAGE_PSE);
473 static inline pte_t pte_mkglobal(pte_t pte)
475 return pte_set_flags(pte, _PAGE_GLOBAL);
478 static inline pte_t pte_clrglobal(pte_t pte)
480 return pte_clear_flags(pte, _PAGE_GLOBAL);
483 static inline pte_t pte_mkspecial(pte_t pte)
485 return pte_set_flags(pte, _PAGE_SPECIAL);
488 static inline pte_t pte_mkdevmap(pte_t pte)
490 return pte_set_flags(pte, _PAGE_SPECIAL|_PAGE_DEVMAP);
493 static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
495 pmdval_t v = native_pmd_val(pmd);
497 return native_make_pmd(v | set);
500 static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
502 pmdval_t v = native_pmd_val(pmd);
504 return native_make_pmd(v & ~clear);
507 /* See comments above mksaveddirty_shift() */
508 static inline pmd_t pmd_mksaveddirty(pmd_t pmd)
510 pmdval_t v = native_pmd_val(pmd);
512 v = mksaveddirty_shift(v);
513 return native_make_pmd(v);
516 /* See comments above mksaveddirty_shift() */
517 static inline pmd_t pmd_clear_saveddirty(pmd_t pmd)
519 pmdval_t v = native_pmd_val(pmd);
521 v = clear_saveddirty_shift(v);
522 return native_make_pmd(v);
525 static inline pmd_t pmd_wrprotect(pmd_t pmd)
527 pmd = pmd_clear_flags(pmd, _PAGE_RW);
530 * Blindly clearing _PAGE_RW might accidentally create
531 * a shadow stack PMD (RW=0, Dirty=1). Move the hardware
532 * dirty value to the software bit.
534 return pmd_mksaveddirty(pmd);
537 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
538 static inline int pmd_uffd_wp(pmd_t pmd)
540 return pmd_flags(pmd) & _PAGE_UFFD_WP;
543 static inline pmd_t pmd_mkuffd_wp(pmd_t pmd)
545 return pmd_wrprotect(pmd_set_flags(pmd, _PAGE_UFFD_WP));
548 static inline pmd_t pmd_clear_uffd_wp(pmd_t pmd)
550 return pmd_clear_flags(pmd, _PAGE_UFFD_WP);
552 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
554 static inline pmd_t pmd_mkold(pmd_t pmd)
556 return pmd_clear_flags(pmd, _PAGE_ACCESSED);
559 static inline pmd_t pmd_mkclean(pmd_t pmd)
561 return pmd_clear_flags(pmd, _PAGE_DIRTY_BITS);
564 static inline pmd_t pmd_mkdirty(pmd_t pmd)
566 pmd = pmd_set_flags(pmd, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
568 return pmd_mksaveddirty(pmd);
571 static inline pmd_t pmd_mkwrite_shstk(pmd_t pmd)
573 pmd = pmd_clear_flags(pmd, _PAGE_RW);
575 return pmd_set_flags(pmd, _PAGE_DIRTY);
578 static inline pmd_t pmd_mkdevmap(pmd_t pmd)
580 return pmd_set_flags(pmd, _PAGE_DEVMAP);
583 static inline pmd_t pmd_mkhuge(pmd_t pmd)
585 return pmd_set_flags(pmd, _PAGE_PSE);
588 static inline pmd_t pmd_mkyoung(pmd_t pmd)
590 return pmd_set_flags(pmd, _PAGE_ACCESSED);
593 static inline pmd_t pmd_mkwrite_novma(pmd_t pmd)
595 return pmd_set_flags(pmd, _PAGE_RW);
598 pmd_t pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma);
599 #define pmd_mkwrite pmd_mkwrite
601 static inline pud_t pud_set_flags(pud_t pud, pudval_t set)
603 pudval_t v = native_pud_val(pud);
605 return native_make_pud(v | set);
608 static inline pud_t pud_clear_flags(pud_t pud, pudval_t clear)
610 pudval_t v = native_pud_val(pud);
612 return native_make_pud(v & ~clear);
615 /* See comments above mksaveddirty_shift() */
616 static inline pud_t pud_mksaveddirty(pud_t pud)
618 pudval_t v = native_pud_val(pud);
620 v = mksaveddirty_shift(v);
621 return native_make_pud(v);
624 /* See comments above mksaveddirty_shift() */
625 static inline pud_t pud_clear_saveddirty(pud_t pud)
627 pudval_t v = native_pud_val(pud);
629 v = clear_saveddirty_shift(v);
630 return native_make_pud(v);
633 static inline pud_t pud_mkold(pud_t pud)
635 return pud_clear_flags(pud, _PAGE_ACCESSED);
638 static inline pud_t pud_mkclean(pud_t pud)
640 return pud_clear_flags(pud, _PAGE_DIRTY_BITS);
643 static inline pud_t pud_wrprotect(pud_t pud)
645 pud = pud_clear_flags(pud, _PAGE_RW);
648 * Blindly clearing _PAGE_RW might accidentally create
649 * a shadow stack PUD (RW=0, Dirty=1). Move the hardware
650 * dirty value to the software bit.
652 return pud_mksaveddirty(pud);
655 static inline pud_t pud_mkdirty(pud_t pud)
657 pud = pud_set_flags(pud, _PAGE_DIRTY | _PAGE_SOFT_DIRTY);
659 return pud_mksaveddirty(pud);
662 static inline pud_t pud_mkdevmap(pud_t pud)
664 return pud_set_flags(pud, _PAGE_DEVMAP);
667 static inline pud_t pud_mkhuge(pud_t pud)
669 return pud_set_flags(pud, _PAGE_PSE);
672 static inline pud_t pud_mkyoung(pud_t pud)
674 return pud_set_flags(pud, _PAGE_ACCESSED);
677 static inline pud_t pud_mkwrite(pud_t pud)
679 pud = pud_set_flags(pud, _PAGE_RW);
681 return pud_clear_saveddirty(pud);
684 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
685 static inline int pte_soft_dirty(pte_t pte)
687 return pte_flags(pte) & _PAGE_SOFT_DIRTY;
690 static inline int pmd_soft_dirty(pmd_t pmd)
692 return pmd_flags(pmd) & _PAGE_SOFT_DIRTY;
695 static inline int pud_soft_dirty(pud_t pud)
697 return pud_flags(pud) & _PAGE_SOFT_DIRTY;
700 static inline pte_t pte_mksoft_dirty(pte_t pte)
702 return pte_set_flags(pte, _PAGE_SOFT_DIRTY);
705 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
707 return pmd_set_flags(pmd, _PAGE_SOFT_DIRTY);
710 static inline pud_t pud_mksoft_dirty(pud_t pud)
712 return pud_set_flags(pud, _PAGE_SOFT_DIRTY);
715 static inline pte_t pte_clear_soft_dirty(pte_t pte)
717 return pte_clear_flags(pte, _PAGE_SOFT_DIRTY);
720 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
722 return pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY);
725 static inline pud_t pud_clear_soft_dirty(pud_t pud)
727 return pud_clear_flags(pud, _PAGE_SOFT_DIRTY);
730 #endif /* CONFIG_HAVE_ARCH_SOFT_DIRTY */
733 * Mask out unsupported bits in a present pgprot. Non-present pgprots
734 * can use those bits for other purposes, so leave them be.
736 static inline pgprotval_t massage_pgprot(pgprot_t pgprot)
738 pgprotval_t protval = pgprot_val(pgprot);
740 if (protval & _PAGE_PRESENT)
741 protval &= __supported_pte_mask;
746 static inline pgprotval_t check_pgprot(pgprot_t pgprot)
748 pgprotval_t massaged_val = massage_pgprot(pgprot);
750 /* mmdebug.h can not be included here because of dependencies */
751 #ifdef CONFIG_DEBUG_VM
752 WARN_ONCE(pgprot_val(pgprot) != massaged_val,
753 "attempted to set unsupported pgprot: %016llx "
754 "bits: %016llx supported: %016llx\n",
755 (u64)pgprot_val(pgprot),
756 (u64)pgprot_val(pgprot) ^ massaged_val,
757 (u64)__supported_pte_mask);
763 static inline pte_t pfn_pte(unsigned long page_nr, pgprot_t pgprot)
765 phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
766 pfn ^= protnone_mask(pgprot_val(pgprot));
768 return __pte(pfn | check_pgprot(pgprot));
771 static inline pmd_t pfn_pmd(unsigned long page_nr, pgprot_t pgprot)
773 phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
774 pfn ^= protnone_mask(pgprot_val(pgprot));
775 pfn &= PHYSICAL_PMD_PAGE_MASK;
776 return __pmd(pfn | check_pgprot(pgprot));
779 static inline pud_t pfn_pud(unsigned long page_nr, pgprot_t pgprot)
781 phys_addr_t pfn = (phys_addr_t)page_nr << PAGE_SHIFT;
782 pfn ^= protnone_mask(pgprot_val(pgprot));
783 pfn &= PHYSICAL_PUD_PAGE_MASK;
784 return __pud(pfn | check_pgprot(pgprot));
787 static inline pmd_t pmd_mkinvalid(pmd_t pmd)
789 return pfn_pmd(pmd_pfn(pmd),
790 __pgprot(pmd_flags(pmd) & ~(_PAGE_PRESENT|_PAGE_PROTNONE)));
793 static inline u64 flip_protnone_guard(u64 oldval, u64 val, u64 mask);
795 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
797 pteval_t val = pte_val(pte), oldval = val;
801 * Chop off the NX bit (if present), and add the NX portion of
802 * the newprot (if present):
804 val &= _PAGE_CHG_MASK;
805 val |= check_pgprot(newprot) & ~_PAGE_CHG_MASK;
806 val = flip_protnone_guard(oldval, val, PTE_PFN_MASK);
808 pte_result = __pte(val);
811 * To avoid creating Write=0,Dirty=1 PTEs, pte_modify() needs to avoid:
812 * 1. Marking Write=0 PTEs Dirty=1
813 * 2. Marking Dirty=1 PTEs Write=0
815 * The first case cannot happen because the _PAGE_CHG_MASK will filter
816 * out any Dirty bit passed in newprot. Handle the second case by
817 * going through the mksaveddirty exercise. Only do this if the old
818 * value was Write=1 to avoid doing this on Shadow Stack PTEs.
820 if (oldval & _PAGE_RW)
821 pte_result = pte_mksaveddirty(pte_result);
823 pte_result = pte_clear_saveddirty(pte_result);
828 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
830 pmdval_t val = pmd_val(pmd), oldval = val;
833 val &= (_HPAGE_CHG_MASK & ~_PAGE_DIRTY);
834 val |= check_pgprot(newprot) & ~_HPAGE_CHG_MASK;
835 val = flip_protnone_guard(oldval, val, PHYSICAL_PMD_PAGE_MASK);
837 pmd_result = __pmd(val);
840 * To avoid creating Write=0,Dirty=1 PMDs, pte_modify() needs to avoid:
841 * 1. Marking Write=0 PMDs Dirty=1
842 * 2. Marking Dirty=1 PMDs Write=0
844 * The first case cannot happen because the _PAGE_CHG_MASK will filter
845 * out any Dirty bit passed in newprot. Handle the second case by
846 * going through the mksaveddirty exercise. Only do this if the old
847 * value was Write=1 to avoid doing this on Shadow Stack PTEs.
849 if (oldval & _PAGE_RW)
850 pmd_result = pmd_mksaveddirty(pmd_result);
852 pmd_result = pmd_clear_saveddirty(pmd_result);
858 * mprotect needs to preserve PAT and encryption bits when updating
861 #define pgprot_modify pgprot_modify
862 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
864 pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK;
865 pgprotval_t addbits = pgprot_val(newprot) & ~_PAGE_CHG_MASK;
866 return __pgprot(preservebits | addbits);
869 #define pte_pgprot(x) __pgprot(pte_flags(x))
870 #define pmd_pgprot(x) __pgprot(pmd_flags(x))
871 #define pud_pgprot(x) __pgprot(pud_flags(x))
872 #define p4d_pgprot(x) __pgprot(p4d_flags(x))
874 #define canon_pgprot(p) __pgprot(massage_pgprot(p))
876 static inline int is_new_memtype_allowed(u64 paddr, unsigned long size,
877 enum page_cache_mode pcm,
878 enum page_cache_mode new_pcm)
881 * PAT type is always WB for untracked ranges, so no need to check.
883 if (x86_platform.is_untracked_pat_range(paddr, paddr + size))
887 * Certain new memtypes are not allowed with certain
889 * - request is uncached, return cannot be write-back
890 * - request is write-combine, return cannot be write-back
891 * - request is write-through, return cannot be write-back
892 * - request is write-through, return cannot be write-combine
894 if ((pcm == _PAGE_CACHE_MODE_UC_MINUS &&
895 new_pcm == _PAGE_CACHE_MODE_WB) ||
896 (pcm == _PAGE_CACHE_MODE_WC &&
897 new_pcm == _PAGE_CACHE_MODE_WB) ||
898 (pcm == _PAGE_CACHE_MODE_WT &&
899 new_pcm == _PAGE_CACHE_MODE_WB) ||
900 (pcm == _PAGE_CACHE_MODE_WT &&
901 new_pcm == _PAGE_CACHE_MODE_WC)) {
908 pmd_t *populate_extra_pmd(unsigned long vaddr);
909 pte_t *populate_extra_pte(unsigned long vaddr);
911 #ifdef CONFIG_PAGE_TABLE_ISOLATION
912 pgd_t __pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd);
915 * Take a PGD location (pgdp) and a pgd value that needs to be set there.
916 * Populates the user and returns the resulting PGD that must be set in
917 * the kernel copy of the page tables.
919 static inline pgd_t pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd)
921 if (!static_cpu_has(X86_FEATURE_PTI))
923 return __pti_set_user_pgtbl(pgdp, pgd);
925 #else /* CONFIG_PAGE_TABLE_ISOLATION */
926 static inline pgd_t pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd)
930 #endif /* CONFIG_PAGE_TABLE_ISOLATION */
932 #endif /* __ASSEMBLY__ */
936 # include <asm/pgtable_32.h>
938 # include <asm/pgtable_64.h>
942 #include <linux/mm_types.h>
943 #include <linux/mmdebug.h>
944 #include <linux/log2.h>
945 #include <asm/fixmap.h>
947 static inline int pte_none(pte_t pte)
949 return !(pte.pte & ~(_PAGE_KNL_ERRATUM_MASK));
952 #define __HAVE_ARCH_PTE_SAME
953 static inline int pte_same(pte_t a, pte_t b)
955 return a.pte == b.pte;
958 static inline pte_t pte_next_pfn(pte_t pte)
960 if (__pte_needs_invert(pte_val(pte)))
961 return __pte(pte_val(pte) - (1UL << PFN_PTE_SHIFT));
962 return __pte(pte_val(pte) + (1UL << PFN_PTE_SHIFT));
964 #define pte_next_pfn pte_next_pfn
966 static inline int pte_present(pte_t a)
968 return pte_flags(a) & (_PAGE_PRESENT | _PAGE_PROTNONE);
971 #ifdef CONFIG_ARCH_HAS_PTE_DEVMAP
972 static inline int pte_devmap(pte_t a)
974 return (pte_flags(a) & _PAGE_DEVMAP) == _PAGE_DEVMAP;
978 #define pte_accessible pte_accessible
979 static inline bool pte_accessible(struct mm_struct *mm, pte_t a)
981 if (pte_flags(a) & _PAGE_PRESENT)
984 if ((pte_flags(a) & _PAGE_PROTNONE) &&
985 atomic_read(&mm->tlb_flush_pending))
991 static inline int pmd_present(pmd_t pmd)
994 * Checking for _PAGE_PSE is needed too because
995 * split_huge_page will temporarily clear the present bit (but
996 * the _PAGE_PSE flag will remain set at all times while the
997 * _PAGE_PRESENT bit is clear).
999 return pmd_flags(pmd) & (_PAGE_PRESENT | _PAGE_PROTNONE | _PAGE_PSE);
1002 #ifdef CONFIG_NUMA_BALANCING
1004 * These work without NUMA balancing but the kernel does not care. See the
1005 * comment in include/linux/pgtable.h
1007 static inline int pte_protnone(pte_t pte)
1009 return (pte_flags(pte) & (_PAGE_PROTNONE | _PAGE_PRESENT))
1013 static inline int pmd_protnone(pmd_t pmd)
1015 return (pmd_flags(pmd) & (_PAGE_PROTNONE | _PAGE_PRESENT))
1018 #endif /* CONFIG_NUMA_BALANCING */
1020 static inline int pmd_none(pmd_t pmd)
1022 /* Only check low word on 32-bit platforms, since it might be
1023 out of sync with upper half. */
1024 unsigned long val = native_pmd_val(pmd);
1025 return (val & ~_PAGE_KNL_ERRATUM_MASK) == 0;
1028 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
1030 return (unsigned long)__va(pmd_val(pmd) & pmd_pfn_mask(pmd));
1034 * Currently stuck as a macro due to indirect forward reference to
1035 * linux/mmzone.h's __section_mem_map_addr() definition:
1037 #define pmd_page(pmd) pfn_to_page(pmd_pfn(pmd))
1040 * Conversion functions: convert a page and protection to a page entry,
1041 * and a page entry and page directory to the page they refer to.
1043 * (Currently stuck as a macro because of indirect forward reference
1044 * to linux/mm.h:page_to_nid())
1046 #define mk_pte(page, pgprot) \
1048 pgprot_t __pgprot = pgprot; \
1050 WARN_ON_ONCE((pgprot_val(__pgprot) & (_PAGE_DIRTY | _PAGE_RW)) == \
1052 pfn_pte(page_to_pfn(page), __pgprot); \
1055 static inline int pmd_bad(pmd_t pmd)
1057 return (pmd_flags(pmd) & ~(_PAGE_USER | _PAGE_ACCESSED)) !=
1058 (_KERNPG_TABLE & ~_PAGE_ACCESSED);
1061 static inline unsigned long pages_to_mb(unsigned long npg)
1063 return npg >> (20 - PAGE_SHIFT);
1066 #if CONFIG_PGTABLE_LEVELS > 2
1067 static inline int pud_none(pud_t pud)
1069 return (native_pud_val(pud) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
1072 static inline int pud_present(pud_t pud)
1074 return pud_flags(pud) & _PAGE_PRESENT;
1077 static inline pmd_t *pud_pgtable(pud_t pud)
1079 return (pmd_t *)__va(pud_val(pud) & pud_pfn_mask(pud));
1083 * Currently stuck as a macro due to indirect forward reference to
1084 * linux/mmzone.h's __section_mem_map_addr() definition:
1086 #define pud_page(pud) pfn_to_page(pud_pfn(pud))
1088 #define pud_leaf pud_large
1089 static inline int pud_large(pud_t pud)
1091 return (pud_val(pud) & (_PAGE_PSE | _PAGE_PRESENT)) ==
1092 (_PAGE_PSE | _PAGE_PRESENT);
1095 static inline int pud_bad(pud_t pud)
1097 return (pud_flags(pud) & ~(_KERNPG_TABLE | _PAGE_USER)) != 0;
1100 #define pud_leaf pud_large
1101 static inline int pud_large(pud_t pud)
1105 #endif /* CONFIG_PGTABLE_LEVELS > 2 */
1107 #if CONFIG_PGTABLE_LEVELS > 3
1108 static inline int p4d_none(p4d_t p4d)
1110 return (native_p4d_val(p4d) & ~(_PAGE_KNL_ERRATUM_MASK)) == 0;
1113 static inline int p4d_present(p4d_t p4d)
1115 return p4d_flags(p4d) & _PAGE_PRESENT;
1118 static inline pud_t *p4d_pgtable(p4d_t p4d)
1120 return (pud_t *)__va(p4d_val(p4d) & p4d_pfn_mask(p4d));
1124 * Currently stuck as a macro due to indirect forward reference to
1125 * linux/mmzone.h's __section_mem_map_addr() definition:
1127 #define p4d_page(p4d) pfn_to_page(p4d_pfn(p4d))
1129 static inline int p4d_bad(p4d_t p4d)
1131 unsigned long ignore_flags = _KERNPG_TABLE | _PAGE_USER;
1133 if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
1134 ignore_flags |= _PAGE_NX;
1136 return (p4d_flags(p4d) & ~ignore_flags) != 0;
1138 #endif /* CONFIG_PGTABLE_LEVELS > 3 */
1140 static inline unsigned long p4d_index(unsigned long address)
1142 return (address >> P4D_SHIFT) & (PTRS_PER_P4D - 1);
1145 #if CONFIG_PGTABLE_LEVELS > 4
1146 static inline int pgd_present(pgd_t pgd)
1148 if (!pgtable_l5_enabled())
1150 return pgd_flags(pgd) & _PAGE_PRESENT;
1153 static inline unsigned long pgd_page_vaddr(pgd_t pgd)
1155 return (unsigned long)__va((unsigned long)pgd_val(pgd) & PTE_PFN_MASK);
1159 * Currently stuck as a macro due to indirect forward reference to
1160 * linux/mmzone.h's __section_mem_map_addr() definition:
1162 #define pgd_page(pgd) pfn_to_page(pgd_pfn(pgd))
1164 /* to find an entry in a page-table-directory. */
1165 static inline p4d_t *p4d_offset(pgd_t *pgd, unsigned long address)
1167 if (!pgtable_l5_enabled())
1168 return (p4d_t *)pgd;
1169 return (p4d_t *)pgd_page_vaddr(*pgd) + p4d_index(address);
1172 static inline int pgd_bad(pgd_t pgd)
1174 unsigned long ignore_flags = _PAGE_USER;
1176 if (!pgtable_l5_enabled())
1179 if (IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION))
1180 ignore_flags |= _PAGE_NX;
1182 return (pgd_flags(pgd) & ~ignore_flags) != _KERNPG_TABLE;
1185 static inline int pgd_none(pgd_t pgd)
1187 if (!pgtable_l5_enabled())
1190 * There is no need to do a workaround for the KNL stray
1191 * A/D bit erratum here. PGDs only point to page tables
1192 * except on 32-bit non-PAE which is not supported on
1195 return !native_pgd_val(pgd);
1197 #endif /* CONFIG_PGTABLE_LEVELS > 4 */
1199 #endif /* __ASSEMBLY__ */
1201 #define KERNEL_PGD_BOUNDARY pgd_index(PAGE_OFFSET)
1202 #define KERNEL_PGD_PTRS (PTRS_PER_PGD - KERNEL_PGD_BOUNDARY)
1204 #ifndef __ASSEMBLY__
1206 extern int direct_gbpages;
1207 void init_mem_mapping(void);
1208 void early_alloc_pgt_buf(void);
1209 extern void memblock_find_dma_reserve(void);
1210 void __init poking_init(void);
1211 unsigned long init_memory_mapping(unsigned long start,
1212 unsigned long end, pgprot_t prot);
1214 #ifdef CONFIG_X86_64
1215 extern pgd_t trampoline_pgd_entry;
1218 /* local pte updates need not use xchg for locking */
1219 static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
1223 /* Pure native function needs no input for mm, addr */
1224 native_pte_clear(NULL, 0, ptep);
1228 static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp)
1232 native_pmd_clear(pmdp);
1236 static inline pud_t native_local_pudp_get_and_clear(pud_t *pudp)
1240 native_pud_clear(pudp);
1244 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1245 pmd_t *pmdp, pmd_t pmd)
1247 page_table_check_pmd_set(mm, pmdp, pmd);
1251 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
1252 pud_t *pudp, pud_t pud)
1254 page_table_check_pud_set(mm, pudp, pud);
1255 native_set_pud(pudp, pud);
1259 * We only update the dirty/accessed state if we set
1260 * the dirty bit by hand in the kernel, since the hardware
1261 * will do the accessed bit for us, and we don't want to
1262 * race with other CPU's that might be updating the dirty
1263 * bit at the same time.
1265 struct vm_area_struct;
1267 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1268 extern int ptep_set_access_flags(struct vm_area_struct *vma,
1269 unsigned long address, pte_t *ptep,
1270 pte_t entry, int dirty);
1272 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1273 extern int ptep_test_and_clear_young(struct vm_area_struct *vma,
1274 unsigned long addr, pte_t *ptep);
1276 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1277 extern int ptep_clear_flush_young(struct vm_area_struct *vma,
1278 unsigned long address, pte_t *ptep);
1280 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1281 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,
1284 pte_t pte = native_ptep_get_and_clear(ptep);
1285 page_table_check_pte_clear(mm, pte);
1289 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1290 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1291 unsigned long addr, pte_t *ptep,
1297 * Full address destruction in progress; paravirt does not
1298 * care about updates and native needs no locking
1300 pte = native_local_ptep_get_and_clear(ptep);
1301 page_table_check_pte_clear(mm, pte);
1303 pte = ptep_get_and_clear(mm, addr, ptep);
1308 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1309 static inline void ptep_set_wrprotect(struct mm_struct *mm,
1310 unsigned long addr, pte_t *ptep)
1313 * Avoid accidentally creating shadow stack PTEs
1314 * (Write=0,Dirty=1). Use cmpxchg() to prevent races with
1315 * the hardware setting Dirty=1.
1317 pte_t old_pte, new_pte;
1319 old_pte = READ_ONCE(*ptep);
1321 new_pte = pte_wrprotect(old_pte);
1322 } while (!try_cmpxchg((long *)&ptep->pte, (long *)&old_pte, *(long *)&new_pte));
1325 #define flush_tlb_fix_spurious_fault(vma, address, ptep) do { } while (0)
1327 #define mk_pmd(page, pgprot) pfn_pmd(page_to_pfn(page), (pgprot))
1329 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1330 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
1331 unsigned long address, pmd_t *pmdp,
1332 pmd_t entry, int dirty);
1333 extern int pudp_set_access_flags(struct vm_area_struct *vma,
1334 unsigned long address, pud_t *pudp,
1335 pud_t entry, int dirty);
1337 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1338 extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1339 unsigned long addr, pmd_t *pmdp);
1340 extern int pudp_test_and_clear_young(struct vm_area_struct *vma,
1341 unsigned long addr, pud_t *pudp);
1343 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
1344 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
1345 unsigned long address, pmd_t *pmdp);
1348 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1349 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm, unsigned long addr,
1352 pmd_t pmd = native_pmdp_get_and_clear(pmdp);
1354 page_table_check_pmd_clear(mm, pmd);
1359 #define __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
1360 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
1361 unsigned long addr, pud_t *pudp)
1363 pud_t pud = native_pudp_get_and_clear(pudp);
1365 page_table_check_pud_clear(mm, pud);
1370 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1371 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1372 unsigned long addr, pmd_t *pmdp)
1375 * Avoid accidentally creating shadow stack PTEs
1376 * (Write=0,Dirty=1). Use cmpxchg() to prevent races with
1377 * the hardware setting Dirty=1.
1379 pmd_t old_pmd, new_pmd;
1381 old_pmd = READ_ONCE(*pmdp);
1383 new_pmd = pmd_wrprotect(old_pmd);
1384 } while (!try_cmpxchg((long *)pmdp, (long *)&old_pmd, *(long *)&new_pmd));
1387 #ifndef pmdp_establish
1388 #define pmdp_establish pmdp_establish
1389 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
1390 unsigned long address, pmd_t *pmdp, pmd_t pmd)
1392 page_table_check_pmd_set(vma->vm_mm, pmdp, pmd);
1393 if (IS_ENABLED(CONFIG_SMP)) {
1394 return xchg(pmdp, pmd);
1397 WRITE_ONCE(*pmdp, pmd);
1403 #define __HAVE_ARCH_PMDP_INVALIDATE_AD
1404 extern pmd_t pmdp_invalidate_ad(struct vm_area_struct *vma,
1405 unsigned long address, pmd_t *pmdp);
1408 * Page table pages are page-aligned. The lower half of the top
1409 * level is used for userspace and the top half for the kernel.
1411 * Returns true for parts of the PGD that map userspace and
1412 * false for the parts that map the kernel.
1414 static inline bool pgdp_maps_userspace(void *__ptr)
1416 unsigned long ptr = (unsigned long)__ptr;
1418 return (((ptr & ~PAGE_MASK) / sizeof(pgd_t)) < PGD_KERNEL_START);
1421 #define pgd_leaf pgd_large
1422 static inline int pgd_large(pgd_t pgd) { return 0; }
1424 #ifdef CONFIG_PAGE_TABLE_ISOLATION
1426 * All top-level PAGE_TABLE_ISOLATION page tables are order-1 pages
1427 * (8k-aligned and 8k in size). The kernel one is at the beginning 4k and
1428 * the user one is in the last 4k. To switch between them, you
1429 * just need to flip the 12th bit in their addresses.
1431 #define PTI_PGTABLE_SWITCH_BIT PAGE_SHIFT
1434 * This generates better code than the inline assembly in
1437 static inline void *ptr_set_bit(void *ptr, int bit)
1439 unsigned long __ptr = (unsigned long)ptr;
1442 return (void *)__ptr;
1444 static inline void *ptr_clear_bit(void *ptr, int bit)
1446 unsigned long __ptr = (unsigned long)ptr;
1449 return (void *)__ptr;
1452 static inline pgd_t *kernel_to_user_pgdp(pgd_t *pgdp)
1454 return ptr_set_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
1457 static inline pgd_t *user_to_kernel_pgdp(pgd_t *pgdp)
1459 return ptr_clear_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
1462 static inline p4d_t *kernel_to_user_p4dp(p4d_t *p4dp)
1464 return ptr_set_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
1467 static inline p4d_t *user_to_kernel_p4dp(p4d_t *p4dp)
1469 return ptr_clear_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
1471 #endif /* CONFIG_PAGE_TABLE_ISOLATION */
1474 * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
1476 * dst - pointer to pgd range anywhere on a pgd page
1478 * count - the number of pgds to copy.
1480 * dst and src can be on the same page, but the range must not overlap,
1481 * and must not cross a page boundary.
1483 static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
1485 memcpy(dst, src, count * sizeof(pgd_t));
1486 #ifdef CONFIG_PAGE_TABLE_ISOLATION
1487 if (!static_cpu_has(X86_FEATURE_PTI))
1489 /* Clone the user space pgd as well */
1490 memcpy(kernel_to_user_pgdp(dst), kernel_to_user_pgdp(src),
1491 count * sizeof(pgd_t));
1495 #define PTE_SHIFT ilog2(PTRS_PER_PTE)
1496 static inline int page_level_shift(enum pg_level level)
1498 return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT;
1500 static inline unsigned long page_level_size(enum pg_level level)
1502 return 1UL << page_level_shift(level);
1504 static inline unsigned long page_level_mask(enum pg_level level)
1506 return ~(page_level_size(level) - 1);
1510 * The x86 doesn't have any external MMU info: the kernel page
1511 * tables contain all the necessary information.
1513 static inline void update_mmu_cache(struct vm_area_struct *vma,
1514 unsigned long addr, pte_t *ptep)
1517 static inline void update_mmu_cache_range(struct vm_fault *vmf,
1518 struct vm_area_struct *vma, unsigned long addr,
1519 pte_t *ptep, unsigned int nr)
1522 static inline void update_mmu_cache_pmd(struct vm_area_struct *vma,
1523 unsigned long addr, pmd_t *pmd)
1526 static inline void update_mmu_cache_pud(struct vm_area_struct *vma,
1527 unsigned long addr, pud_t *pud)
1530 static inline pte_t pte_swp_mkexclusive(pte_t pte)
1532 return pte_set_flags(pte, _PAGE_SWP_EXCLUSIVE);
1535 static inline int pte_swp_exclusive(pte_t pte)
1537 return pte_flags(pte) & _PAGE_SWP_EXCLUSIVE;
1540 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
1542 return pte_clear_flags(pte, _PAGE_SWP_EXCLUSIVE);
1545 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
1546 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
1548 return pte_set_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1551 static inline int pte_swp_soft_dirty(pte_t pte)
1553 return pte_flags(pte) & _PAGE_SWP_SOFT_DIRTY;
1556 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
1558 return pte_clear_flags(pte, _PAGE_SWP_SOFT_DIRTY);
1561 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1562 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
1564 return pmd_set_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1567 static inline int pmd_swp_soft_dirty(pmd_t pmd)
1569 return pmd_flags(pmd) & _PAGE_SWP_SOFT_DIRTY;
1572 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
1574 return pmd_clear_flags(pmd, _PAGE_SWP_SOFT_DIRTY);
1579 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_WP
1580 static inline pte_t pte_swp_mkuffd_wp(pte_t pte)
1582 return pte_set_flags(pte, _PAGE_SWP_UFFD_WP);
1585 static inline int pte_swp_uffd_wp(pte_t pte)
1587 return pte_flags(pte) & _PAGE_SWP_UFFD_WP;
1590 static inline pte_t pte_swp_clear_uffd_wp(pte_t pte)
1592 return pte_clear_flags(pte, _PAGE_SWP_UFFD_WP);
1595 static inline pmd_t pmd_swp_mkuffd_wp(pmd_t pmd)
1597 return pmd_set_flags(pmd, _PAGE_SWP_UFFD_WP);
1600 static inline int pmd_swp_uffd_wp(pmd_t pmd)
1602 return pmd_flags(pmd) & _PAGE_SWP_UFFD_WP;
1605 static inline pmd_t pmd_swp_clear_uffd_wp(pmd_t pmd)
1607 return pmd_clear_flags(pmd, _PAGE_SWP_UFFD_WP);
1609 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_WP */
1611 static inline u16 pte_flags_pkey(unsigned long pte_flags)
1613 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1614 /* ifdef to avoid doing 59-bit shift on 32-bit values */
1615 return (pte_flags & _PAGE_PKEY_MASK) >> _PAGE_BIT_PKEY_BIT0;
1621 static inline bool __pkru_allows_pkey(u16 pkey, bool write)
1623 u32 pkru = read_pkru();
1625 if (!__pkru_allows_read(pkru, pkey))
1627 if (write && !__pkru_allows_write(pkru, pkey))
1634 * 'pteval' can come from a PTE, PMD or PUD. We only check
1635 * _PAGE_PRESENT, _PAGE_USER, and _PAGE_RW in here which are the
1636 * same value on all 3 types.
1638 static inline bool __pte_access_permitted(unsigned long pteval, bool write)
1640 unsigned long need_pte_bits = _PAGE_PRESENT|_PAGE_USER;
1643 * Write=0,Dirty=1 PTEs are shadow stack, which the kernel
1644 * shouldn't generally allow access to, but since they
1645 * are already Write=0, the below logic covers both cases.
1648 need_pte_bits |= _PAGE_RW;
1650 if ((pteval & need_pte_bits) != need_pte_bits)
1653 return __pkru_allows_pkey(pte_flags_pkey(pteval), write);
1656 #define pte_access_permitted pte_access_permitted
1657 static inline bool pte_access_permitted(pte_t pte, bool write)
1659 return __pte_access_permitted(pte_val(pte), write);
1662 #define pmd_access_permitted pmd_access_permitted
1663 static inline bool pmd_access_permitted(pmd_t pmd, bool write)
1665 return __pte_access_permitted(pmd_val(pmd), write);
1668 #define pud_access_permitted pud_access_permitted
1669 static inline bool pud_access_permitted(pud_t pud, bool write)
1671 return __pte_access_permitted(pud_val(pud), write);
1674 #define __HAVE_ARCH_PFN_MODIFY_ALLOWED 1
1675 extern bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot);
1677 static inline bool arch_has_pfn_modify_check(void)
1679 return boot_cpu_has_bug(X86_BUG_L1TF);
1682 #define arch_has_hw_pte_young arch_has_hw_pte_young
1683 static inline bool arch_has_hw_pte_young(void)
1688 #define arch_check_zapped_pte arch_check_zapped_pte
1689 void arch_check_zapped_pte(struct vm_area_struct *vma, pte_t pte);
1691 #define arch_check_zapped_pmd arch_check_zapped_pmd
1692 void arch_check_zapped_pmd(struct vm_area_struct *vma, pmd_t pmd);
1694 #ifdef CONFIG_XEN_PV
1695 #define arch_has_hw_nonleaf_pmd_young arch_has_hw_nonleaf_pmd_young
1696 static inline bool arch_has_hw_nonleaf_pmd_young(void)
1698 return !cpu_feature_enabled(X86_FEATURE_XENPV);
1702 #ifdef CONFIG_PAGE_TABLE_CHECK
1703 static inline bool pte_user_accessible_page(pte_t pte)
1705 return (pte_val(pte) & _PAGE_PRESENT) && (pte_val(pte) & _PAGE_USER);
1708 static inline bool pmd_user_accessible_page(pmd_t pmd)
1710 return pmd_leaf(pmd) && (pmd_val(pmd) & _PAGE_PRESENT) && (pmd_val(pmd) & _PAGE_USER);
1713 static inline bool pud_user_accessible_page(pud_t pud)
1715 return pud_leaf(pud) && (pud_val(pud) & _PAGE_PRESENT) && (pud_val(pud) & _PAGE_USER);
1719 #endif /* __ASSEMBLY__ */
1721 #endif /* _ASM_X86_PGTABLE_H */