1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_PGTABLE_H
3 #define _LINUX_PGTABLE_H
6 #include <asm/pgtable.h>
11 #include <linux/mm_types.h>
12 #include <linux/bug.h>
13 #include <linux/errno.h>
14 #include <asm-generic/pgtable_uffd.h>
16 #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
17 defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
18 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
22 * On almost all architectures and configurations, 0 can be used as the
23 * upper ceiling to free_pgtables(): on many architectures it has the same
24 * effect as using TASK_SIZE. However, there is one configuration which
25 * must impose a more careful limit, to avoid freeing kernel pgtables.
27 #ifndef USER_PGTABLES_CEILING
28 #define USER_PGTABLES_CEILING 0UL
32 * A page table page can be thought of an array like this: pXd_t[PTRS_PER_PxD]
34 * The pXx_index() functions return the index of the entry in the page
35 * table page which would control the given virtual address
37 * As these functions may be used by the same code for different levels of
38 * the page table folding, they are always available, regardless of
39 * CONFIG_PGTABLE_LEVELS value. For the folded levels they simply return 0
40 * because in such cases PTRS_PER_PxD equals 1.
43 static inline unsigned long pte_index(unsigned long address)
45 return (address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
49 static inline unsigned long pmd_index(unsigned long address)
51 return (address >> PMD_SHIFT) & (PTRS_PER_PMD - 1);
53 #define pmd_index pmd_index
57 static inline unsigned long pud_index(unsigned long address)
59 return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
61 #define pud_index pud_index
65 /* Must be a compile-time constant, so implement it as a macro */
66 #define pgd_index(a) (((a) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1))
69 #ifndef pte_offset_kernel
70 static inline pte_t *pte_offset_kernel(pmd_t *pmd, unsigned long address)
72 return (pte_t *)pmd_page_vaddr(*pmd) + pte_index(address);
74 #define pte_offset_kernel pte_offset_kernel
77 #if defined(CONFIG_HIGHPTE)
78 #define pte_offset_map(dir, address) \
79 ((pte_t *)kmap_atomic(pmd_page(*(dir))) + \
81 #define pte_unmap(pte) kunmap_atomic((pte))
83 #define pte_offset_map(dir, address) pte_offset_kernel((dir), (address))
84 #define pte_unmap(pte) ((void)(pte)) /* NOP */
87 /* Find an entry in the second-level page table.. */
89 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
91 return (pmd_t *)pud_page_vaddr(*pud) + pmd_index(address);
93 #define pmd_offset pmd_offset
97 static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
99 return (pud_t *)p4d_page_vaddr(*p4d) + pud_index(address);
101 #define pud_offset pud_offset
104 static inline pgd_t *pgd_offset_pgd(pgd_t *pgd, unsigned long address)
106 return (pgd + pgd_index(address));
110 * a shortcut to get a pgd_t in a given mm
113 #define pgd_offset(mm, address) pgd_offset_pgd((mm)->pgd, (address))
117 * a shortcut which implies the use of the kernel's pgd, instead
120 #define pgd_offset_k(address) pgd_offset(&init_mm, (address))
123 * In many cases it is known that a virtual address is mapped at PMD or PTE
124 * level, so instead of traversing all the page table levels, we can get a
125 * pointer to the PMD entry in user or kernel page table or translate a virtual
126 * address to the pointer in the PTE in the kernel page tables with simple
129 static inline pmd_t *pmd_off(struct mm_struct *mm, unsigned long va)
131 return pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, va), va), va), va);
134 static inline pmd_t *pmd_off_k(unsigned long va)
136 return pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va);
139 static inline pte_t *virt_to_kpte(unsigned long vaddr)
141 pmd_t *pmd = pmd_off_k(vaddr);
143 return pmd_none(*pmd) ? NULL : pte_offset_kernel(pmd, vaddr);
146 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
147 extern int ptep_set_access_flags(struct vm_area_struct *vma,
148 unsigned long address, pte_t *ptep,
149 pte_t entry, int dirty);
152 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
153 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
154 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
155 unsigned long address, pmd_t *pmdp,
156 pmd_t entry, int dirty);
157 extern int pudp_set_access_flags(struct vm_area_struct *vma,
158 unsigned long address, pud_t *pudp,
159 pud_t entry, int dirty);
161 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
162 unsigned long address, pmd_t *pmdp,
163 pmd_t entry, int dirty)
168 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
169 unsigned long address, pud_t *pudp,
170 pud_t entry, int dirty)
175 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
178 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
179 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
180 unsigned long address,
188 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
193 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
194 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
195 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
196 unsigned long address,
204 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
208 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
209 unsigned long address,
215 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
218 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
219 int ptep_clear_flush_young(struct vm_area_struct *vma,
220 unsigned long address, pte_t *ptep);
223 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
224 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
225 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
226 unsigned long address, pmd_t *pmdp);
229 * Despite relevant to THP only, this API is called from generic rmap code
230 * under PageTransHuge(), hence needs a dummy implementation for !THP
232 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
233 unsigned long address, pmd_t *pmdp)
238 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
241 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
242 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
243 unsigned long address,
247 pte_clear(mm, address, ptep);
252 #ifndef __HAVE_ARCH_PTEP_GET
253 static inline pte_t ptep_get(pte_t *ptep)
255 return READ_ONCE(*ptep);
259 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
260 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
261 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
262 unsigned long address,
269 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
270 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
271 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
272 unsigned long address,
280 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
281 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
283 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
284 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
285 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
286 unsigned long address, pmd_t *pmdp,
289 return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
293 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
294 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
295 unsigned long address, pud_t *pudp,
298 return pudp_huge_get_and_clear(mm, address, pudp);
301 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
303 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
304 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
305 unsigned long address, pte_t *ptep,
309 pte = ptep_get_and_clear(mm, address, ptep);
316 * If two threads concurrently fault at the same page, the thread that
317 * won the race updates the PTE and its local TLB/Cache. The other thread
318 * gives up, simply does nothing, and continues; on architectures where
319 * software can update TLB, local TLB can be updated here to avoid next page
320 * fault. This function updates TLB only, do nothing with cache or others.
321 * It is the difference with function update_mmu_cache.
323 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
324 static inline void update_mmu_tlb(struct vm_area_struct *vma,
325 unsigned long address, pte_t *ptep)
328 #define __HAVE_ARCH_UPDATE_MMU_TLB
332 * Some architectures may be able to avoid expensive synchronization
333 * primitives when modifications are made to PTE's which are already
334 * not present, or in the process of an address space destruction.
336 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
337 static inline void pte_clear_not_present_full(struct mm_struct *mm,
338 unsigned long address,
342 pte_clear(mm, address, ptep);
346 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
347 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
348 unsigned long address,
352 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
353 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
354 unsigned long address,
356 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
357 unsigned long address,
361 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
363 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
365 pte_t old_pte = *ptep;
366 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
371 * On some architectures hardware does not set page access bit when accessing
372 * memory page, it is responsibilty of software setting this bit. It brings
373 * out extra page fault penalty to track page access bit. For optimization page
374 * access bit can be set during all page fault flow on these arches.
375 * To be differentiate with macro pte_mkyoung, this macro is used on platforms
376 * where software maintains page access bit.
378 #ifndef pte_sw_mkyoung
379 static inline pte_t pte_sw_mkyoung(pte_t pte)
383 #define pte_sw_mkyoung pte_sw_mkyoung
386 #ifndef pte_savedwrite
387 #define pte_savedwrite pte_write
390 #ifndef pte_mk_savedwrite
391 #define pte_mk_savedwrite pte_mkwrite
394 #ifndef pte_clear_savedwrite
395 #define pte_clear_savedwrite pte_wrprotect
398 #ifndef pmd_savedwrite
399 #define pmd_savedwrite pmd_write
402 #ifndef pmd_mk_savedwrite
403 #define pmd_mk_savedwrite pmd_mkwrite
406 #ifndef pmd_clear_savedwrite
407 #define pmd_clear_savedwrite pmd_wrprotect
410 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
411 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
412 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
413 unsigned long address, pmd_t *pmdp)
415 pmd_t old_pmd = *pmdp;
416 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
419 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
420 unsigned long address, pmd_t *pmdp)
424 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
426 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
427 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
428 static inline void pudp_set_wrprotect(struct mm_struct *mm,
429 unsigned long address, pud_t *pudp)
431 pud_t old_pud = *pudp;
433 set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
436 static inline void pudp_set_wrprotect(struct mm_struct *mm,
437 unsigned long address, pud_t *pudp)
441 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
444 #ifndef pmdp_collapse_flush
445 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
446 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
447 unsigned long address, pmd_t *pmdp);
449 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
450 unsigned long address,
456 #define pmdp_collapse_flush pmdp_collapse_flush
457 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
460 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
461 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
465 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
466 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
469 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
471 * This is an implementation of pmdp_establish() that is only suitable for an
472 * architecture that doesn't have hardware dirty/accessed bits. In this case we
473 * can't race with CPU which sets these bits and non-atomic aproach is fine.
475 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
476 unsigned long address, pmd_t *pmdp, pmd_t pmd)
478 pmd_t old_pmd = *pmdp;
479 set_pmd_at(vma->vm_mm, address, pmdp, pmd);
484 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
485 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
489 #ifndef __HAVE_ARCH_PTE_SAME
490 static inline int pte_same(pte_t pte_a, pte_t pte_b)
492 return pte_val(pte_a) == pte_val(pte_b);
496 #ifndef __HAVE_ARCH_PTE_UNUSED
498 * Some architectures provide facilities to virtualization guests
499 * so that they can flag allocated pages as unused. This allows the
500 * host to transparently reclaim unused pages. This function returns
501 * whether the pte's page is unused.
503 static inline int pte_unused(pte_t pte)
509 #ifndef pte_access_permitted
510 #define pte_access_permitted(pte, write) \
511 (pte_present(pte) && (!(write) || pte_write(pte)))
514 #ifndef pmd_access_permitted
515 #define pmd_access_permitted(pmd, write) \
516 (pmd_present(pmd) && (!(write) || pmd_write(pmd)))
519 #ifndef pud_access_permitted
520 #define pud_access_permitted(pud, write) \
521 (pud_present(pud) && (!(write) || pud_write(pud)))
524 #ifndef p4d_access_permitted
525 #define p4d_access_permitted(p4d, write) \
526 (p4d_present(p4d) && (!(write) || p4d_write(p4d)))
529 #ifndef pgd_access_permitted
530 #define pgd_access_permitted(pgd, write) \
531 (pgd_present(pgd) && (!(write) || pgd_write(pgd)))
534 #ifndef __HAVE_ARCH_PMD_SAME
535 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
537 return pmd_val(pmd_a) == pmd_val(pmd_b);
540 static inline int pud_same(pud_t pud_a, pud_t pud_b)
542 return pud_val(pud_a) == pud_val(pud_b);
546 #ifndef __HAVE_ARCH_P4D_SAME
547 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
549 return p4d_val(p4d_a) == p4d_val(p4d_b);
553 #ifndef __HAVE_ARCH_PGD_SAME
554 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
556 return pgd_val(pgd_a) == pgd_val(pgd_b);
561 * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
562 * TLB flush will be required as a result of the "set". For example, use
563 * in scenarios where it is known ahead of time that the routine is
564 * setting non-present entries, or re-setting an existing entry to the
565 * same value. Otherwise, use the typical "set" helpers and flush the
568 #define set_pte_safe(ptep, pte) \
570 WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
571 set_pte(ptep, pte); \
574 #define set_pmd_safe(pmdp, pmd) \
576 WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
577 set_pmd(pmdp, pmd); \
580 #define set_pud_safe(pudp, pud) \
582 WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
583 set_pud(pudp, pud); \
586 #define set_p4d_safe(p4dp, p4d) \
588 WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
589 set_p4d(p4dp, p4d); \
592 #define set_pgd_safe(pgdp, pgd) \
594 WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
595 set_pgd(pgdp, pgd); \
598 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
600 * Some architectures support metadata associated with a page. When a
601 * page is being swapped out, this metadata must be saved so it can be
602 * restored when the page is swapped back in. SPARC M7 and newer
603 * processors support an ADI (Application Data Integrity) tag for the
604 * page as metadata for the page. arch_do_swap_page() can restore this
605 * metadata when a page is swapped back in.
607 static inline void arch_do_swap_page(struct mm_struct *mm,
608 struct vm_area_struct *vma,
610 pte_t pte, pte_t oldpte)
616 #ifndef __HAVE_ARCH_UNMAP_ONE
618 * Some architectures support metadata associated with a page. When a
619 * page is being swapped out, this metadata must be saved so it can be
620 * restored when the page is swapped back in. SPARC M7 and newer
621 * processors support an ADI (Application Data Integrity) tag for the
622 * page as metadata for the page. arch_unmap_one() can save this
623 * metadata on a swap-out of a page.
625 static inline int arch_unmap_one(struct mm_struct *mm,
626 struct vm_area_struct *vma,
634 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
635 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
638 #ifndef __HAVE_ARCH_MOVE_PTE
639 #define move_pte(pte, prot, old_addr, new_addr) (pte)
642 #ifndef pte_accessible
643 # define pte_accessible(mm, pte) ((void)(pte), 1)
646 #ifndef flush_tlb_fix_spurious_fault
647 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
651 #define pgprot_nx(prot) (prot)
654 #ifndef pgprot_noncached
655 #define pgprot_noncached(prot) (prot)
658 #ifndef pgprot_writecombine
659 #define pgprot_writecombine pgprot_noncached
662 #ifndef pgprot_writethrough
663 #define pgprot_writethrough pgprot_noncached
666 #ifndef pgprot_device
667 #define pgprot_device pgprot_noncached
670 #ifndef pgprot_modify
671 #define pgprot_modify pgprot_modify
672 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
674 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
675 newprot = pgprot_noncached(newprot);
676 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
677 newprot = pgprot_writecombine(newprot);
678 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
679 newprot = pgprot_device(newprot);
685 * When walking page tables, get the address of the next boundary,
686 * or the end address of the range if that comes earlier. Although no
687 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
690 #define pgd_addr_end(addr, end) \
691 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
692 (__boundary - 1 < (end) - 1)? __boundary: (end); \
696 #define p4d_addr_end(addr, end) \
697 ({ unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK; \
698 (__boundary - 1 < (end) - 1)? __boundary: (end); \
703 #define pud_addr_end(addr, end) \
704 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
705 (__boundary - 1 < (end) - 1)? __boundary: (end); \
710 #define pmd_addr_end(addr, end) \
711 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
712 (__boundary - 1 < (end) - 1)? __boundary: (end); \
717 * When walking page tables, we usually want to skip any p?d_none entries;
718 * and any p?d_bad entries - reporting the error before resetting to none.
719 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
721 void pgd_clear_bad(pgd_t *);
723 #ifndef __PAGETABLE_P4D_FOLDED
724 void p4d_clear_bad(p4d_t *);
726 #define p4d_clear_bad(p4d) do { } while (0)
729 #ifndef __PAGETABLE_PUD_FOLDED
730 void pud_clear_bad(pud_t *);
732 #define pud_clear_bad(p4d) do { } while (0)
735 void pmd_clear_bad(pmd_t *);
737 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
741 if (unlikely(pgd_bad(*pgd))) {
748 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
752 if (unlikely(p4d_bad(*p4d))) {
759 static inline int pud_none_or_clear_bad(pud_t *pud)
763 if (unlikely(pud_bad(*pud))) {
770 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
774 if (unlikely(pmd_bad(*pmd))) {
781 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
786 * Get the current pte state, but zero it out to make it
787 * non-present, preventing the hardware from asynchronously
790 return ptep_get_and_clear(vma->vm_mm, addr, ptep);
793 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
795 pte_t *ptep, pte_t pte)
798 * The pte is non-present, so there's no hardware state to
801 set_pte_at(vma->vm_mm, addr, ptep, pte);
804 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
806 * Start a pte protection read-modify-write transaction, which
807 * protects against asynchronous hardware modifications to the pte.
808 * The intention is not to prevent the hardware from making pte
809 * updates, but to prevent any updates it may make from being lost.
811 * This does not protect against other software modifications of the
812 * pte; the appropriate pte lock must be held over the transation.
814 * Note that this interface is intended to be batchable, meaning that
815 * ptep_modify_prot_commit may not actually update the pte, but merely
816 * queue the update to be done at some later time. The update must be
817 * actually committed before the pte lock is released, however.
819 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
823 return __ptep_modify_prot_start(vma, addr, ptep);
827 * Commit an update to a pte, leaving any hardware-controlled bits in
828 * the PTE unmodified.
830 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
832 pte_t *ptep, pte_t old_pte, pte_t pte)
834 __ptep_modify_prot_commit(vma, addr, ptep, pte);
836 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
837 #endif /* CONFIG_MMU */
840 * No-op macros that just return the current protection value. Defined here
841 * because these macros can be used used even if CONFIG_MMU is not defined.
843 #ifndef pgprot_encrypted
844 #define pgprot_encrypted(prot) (prot)
847 #ifndef pgprot_decrypted
848 #define pgprot_decrypted(prot) (prot)
852 * A facility to provide lazy MMU batching. This allows PTE updates and
853 * page invalidations to be delayed until a call to leave lazy MMU mode
854 * is issued. Some architectures may benefit from doing this, and it is
855 * beneficial for both shadow and direct mode hypervisors, which may batch
856 * the PTE updates which happen during this window. Note that using this
857 * interface requires that read hazards be removed from the code. A read
858 * hazard could result in the direct mode hypervisor case, since the actual
859 * write to the page tables may not yet have taken place, so reads though
860 * a raw PTE pointer after it has been modified are not guaranteed to be
861 * up to date. This mode can only be entered and left under the protection of
862 * the page table locks for all page tables which may be modified. In the UP
863 * case, this is required so that preemption is disabled, and in the SMP case,
864 * it must synchronize the delayed page table writes properly on other CPUs.
866 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
867 #define arch_enter_lazy_mmu_mode() do {} while (0)
868 #define arch_leave_lazy_mmu_mode() do {} while (0)
869 #define arch_flush_lazy_mmu_mode() do {} while (0)
873 * A facility to provide batching of the reload of page tables and
874 * other process state with the actual context switch code for
875 * paravirtualized guests. By convention, only one of the batched
876 * update (lazy) modes (CPU, MMU) should be active at any given time,
877 * entry should never be nested, and entry and exits should always be
878 * paired. This is for sanity of maintaining and reasoning about the
879 * kernel code. In this case, the exit (end of the context switch) is
880 * in architecture-specific code, and so doesn't need a generic
883 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
884 #define arch_start_context_switch(prev) do {} while (0)
887 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
888 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
889 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
894 static inline int pmd_swp_soft_dirty(pmd_t pmd)
899 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
904 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
905 static inline int pte_soft_dirty(pte_t pte)
910 static inline int pmd_soft_dirty(pmd_t pmd)
915 static inline pte_t pte_mksoft_dirty(pte_t pte)
920 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
925 static inline pte_t pte_clear_soft_dirty(pte_t pte)
930 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
935 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
940 static inline int pte_swp_soft_dirty(pte_t pte)
945 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
950 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
955 static inline int pmd_swp_soft_dirty(pmd_t pmd)
960 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
966 #ifndef __HAVE_PFNMAP_TRACKING
968 * Interfaces that can be used by architecture code to keep track of
969 * memory type of pfn mappings specified by the remap_pfn_range,
974 * track_pfn_remap is called when a _new_ pfn mapping is being established
975 * by remap_pfn_range() for physical range indicated by pfn and size.
977 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
978 unsigned long pfn, unsigned long addr,
985 * track_pfn_insert is called when a _new_ single pfn is established
986 * by vmf_insert_pfn().
988 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
994 * track_pfn_copy is called when vma that is covering the pfnmap gets
995 * copied through copy_page_range().
997 static inline int track_pfn_copy(struct vm_area_struct *vma)
1003 * untrack_pfn is called while unmapping a pfnmap for a region.
1004 * untrack can be called for a specific region indicated by pfn and size or
1005 * can be for the entire vma (in which case pfn, size are zero).
1007 static inline void untrack_pfn(struct vm_area_struct *vma,
1008 unsigned long pfn, unsigned long size)
1013 * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
1015 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
1019 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1020 unsigned long pfn, unsigned long addr,
1021 unsigned long size);
1022 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1024 extern int track_pfn_copy(struct vm_area_struct *vma);
1025 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1026 unsigned long size);
1027 extern void untrack_pfn_moved(struct vm_area_struct *vma);
1030 #ifdef __HAVE_COLOR_ZERO_PAGE
1031 static inline int is_zero_pfn(unsigned long pfn)
1033 extern unsigned long zero_pfn;
1034 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1035 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1038 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
1041 static inline int is_zero_pfn(unsigned long pfn)
1043 extern unsigned long zero_pfn;
1044 return pfn == zero_pfn;
1047 static inline unsigned long my_zero_pfn(unsigned long addr)
1049 extern unsigned long zero_pfn;
1056 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1057 static inline int pmd_trans_huge(pmd_t pmd)
1062 static inline int pmd_write(pmd_t pmd)
1067 #endif /* pmd_write */
1068 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1071 static inline int pud_write(pud_t pud)
1076 #endif /* pud_write */
1078 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1079 static inline int pmd_devmap(pmd_t pmd)
1083 static inline int pud_devmap(pud_t pud)
1087 static inline int pgd_devmap(pgd_t pgd)
1093 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1094 (defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1095 !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
1096 static inline int pud_trans_huge(pud_t pud)
1102 /* See pmd_none_or_trans_huge_or_clear_bad for discussion. */
1103 static inline int pud_none_or_trans_huge_or_dev_or_clear_bad(pud_t *pud)
1105 pud_t pudval = READ_ONCE(*pud);
1107 if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1109 if (unlikely(pud_bad(pudval))) {
1116 /* See pmd_trans_unstable for discussion. */
1117 static inline int pud_trans_unstable(pud_t *pud)
1119 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1120 defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1121 return pud_none_or_trans_huge_or_dev_or_clear_bad(pud);
1127 #ifndef pmd_read_atomic
1128 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
1131 * Depend on compiler for an atomic pmd read. NOTE: this is
1132 * only going to work, if the pmdval_t isn't larger than
1139 #ifndef arch_needs_pgtable_deposit
1140 #define arch_needs_pgtable_deposit() (false)
1143 * This function is meant to be used by sites walking pagetables with
1144 * the mmap_lock held in read mode to protect against MADV_DONTNEED and
1145 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
1146 * into a null pmd and the transhuge page fault can convert a null pmd
1147 * into an hugepmd or into a regular pmd (if the hugepage allocation
1148 * fails). While holding the mmap_lock in read mode the pmd becomes
1149 * stable and stops changing under us only if it's not null and not a
1150 * transhuge pmd. When those races occurs and this function makes a
1151 * difference vs the standard pmd_none_or_clear_bad, the result is
1152 * undefined so behaving like if the pmd was none is safe (because it
1153 * can return none anyway). The compiler level barrier() is critically
1154 * important to compute the two checks atomically on the same pmdval.
1156 * For 32bit kernels with a 64bit large pmd_t this automatically takes
1157 * care of reading the pmd atomically to avoid SMP race conditions
1158 * against pmd_populate() when the mmap_lock is hold for reading by the
1159 * caller (a special atomic read not done by "gcc" as in the generic
1160 * version above, is also needed when THP is disabled because the page
1161 * fault can populate the pmd from under us).
1163 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
1165 pmd_t pmdval = pmd_read_atomic(pmd);
1167 * The barrier will stabilize the pmdval in a register or on
1168 * the stack so that it will stop changing under the code.
1170 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
1171 * pmd_read_atomic is allowed to return a not atomic pmdval
1172 * (for example pointing to an hugepage that has never been
1173 * mapped in the pmd). The below checks will only care about
1174 * the low part of the pmd with 32bit PAE x86 anyway, with the
1175 * exception of pmd_none(). So the important thing is that if
1176 * the low part of the pmd is found null, the high part will
1177 * be also null or the pmd_none() check below would be
1180 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1184 * !pmd_present() checks for pmd migration entries
1186 * The complete check uses is_pmd_migration_entry() in linux/swapops.h
1187 * But using that requires moving current function and pmd_trans_unstable()
1188 * to linux/swapops.h to resovle dependency, which is too much code move.
1190 * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
1191 * because !pmd_present() pages can only be under migration not swapped
1194 * pmd_none() is preseved for future condition checks on pmd migration
1195 * entries and not confusing with this function name, although it is
1196 * redundant with !pmd_present().
1198 if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
1199 (IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
1201 if (unlikely(pmd_bad(pmdval))) {
1209 * This is a noop if Transparent Hugepage Support is not built into
1210 * the kernel. Otherwise it is equivalent to
1211 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
1212 * places that already verified the pmd is not none and they want to
1213 * walk ptes while holding the mmap sem in read mode (write mode don't
1214 * need this). If THP is not enabled, the pmd can't go away under the
1215 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
1216 * run a pmd_trans_unstable before walking the ptes after
1217 * split_huge_pmd returns (because it may have run when the pmd become
1218 * null, but then a page fault can map in a THP and not a regular page).
1220 static inline int pmd_trans_unstable(pmd_t *pmd)
1222 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1223 return pmd_none_or_trans_huge_or_clear_bad(pmd);
1229 #ifndef CONFIG_NUMA_BALANCING
1231 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
1232 * the only case the kernel cares is for NUMA balancing and is only ever set
1233 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
1234 * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
1235 * is the responsibility of the caller to distinguish between PROT_NONE
1236 * protections and NUMA hinting fault protections.
1238 static inline int pte_protnone(pte_t pte)
1243 static inline int pmd_protnone(pmd_t pmd)
1247 #endif /* CONFIG_NUMA_BALANCING */
1249 #endif /* CONFIG_MMU */
1251 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1253 #ifndef __PAGETABLE_P4D_FOLDED
1254 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1255 int p4d_clear_huge(p4d_t *p4d);
1257 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1261 static inline int p4d_clear_huge(p4d_t *p4d)
1265 #endif /* !__PAGETABLE_P4D_FOLDED */
1267 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1268 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1269 int pud_clear_huge(pud_t *pud);
1270 int pmd_clear_huge(pmd_t *pmd);
1271 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1272 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1273 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1274 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1275 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1279 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1283 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1287 static inline int p4d_clear_huge(p4d_t *p4d)
1291 static inline int pud_clear_huge(pud_t *pud)
1295 static inline int pmd_clear_huge(pmd_t *pmd)
1299 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1303 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1307 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1311 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
1313 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1314 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1316 * ARCHes with special requirements for evicting THP backing TLB entries can
1317 * implement this. Otherwise also, it can help optimize normal TLB flush in
1318 * THP regime. stock flush_tlb_range() typically has optimization to nuke the
1319 * entire TLB TLB if flush span is greater than a threshold, which will
1320 * likely be true for a single huge page. Thus a single thp flush will
1321 * invalidate the entire TLB which is not desitable.
1322 * e.g. see arch/arc: flush_pmd_tlb_range
1324 #define flush_pmd_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1325 #define flush_pud_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1327 #define flush_pmd_tlb_range(vma, addr, end) BUILD_BUG()
1328 #define flush_pud_tlb_range(vma, addr, end) BUILD_BUG()
1333 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1334 unsigned long size, pgprot_t *vma_prot);
1336 #ifndef CONFIG_X86_ESPFIX64
1337 static inline void init_espfix_bsp(void) { }
1340 extern void __init pgtable_cache_init(void);
1342 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1343 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1348 static inline bool arch_has_pfn_modify_check(void)
1352 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1355 * Architecture PAGE_KERNEL_* fallbacks
1357 * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1358 * because they really don't support them, or the port needs to be updated to
1359 * reflect the required functionality. Below are a set of relatively safe
1360 * fallbacks, as best effort, which we can count on in lieu of the architectures
1361 * not defining them on their own yet.
1364 #ifndef PAGE_KERNEL_RO
1365 # define PAGE_KERNEL_RO PAGE_KERNEL
1368 #ifndef PAGE_KERNEL_EXEC
1369 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1373 * Page Table Modification bits for pgtbl_mod_mask.
1375 * These are used by the p?d_alloc_track*() set of functions an in the generic
1376 * vmalloc/ioremap code to track at which page-table levels entries have been
1377 * modified. Based on that the code can better decide when vmalloc and ioremap
1378 * mapping changes need to be synchronized to other page-tables in the system.
1380 #define __PGTBL_PGD_MODIFIED 0
1381 #define __PGTBL_P4D_MODIFIED 1
1382 #define __PGTBL_PUD_MODIFIED 2
1383 #define __PGTBL_PMD_MODIFIED 3
1384 #define __PGTBL_PTE_MODIFIED 4
1386 #define PGTBL_PGD_MODIFIED BIT(__PGTBL_PGD_MODIFIED)
1387 #define PGTBL_P4D_MODIFIED BIT(__PGTBL_P4D_MODIFIED)
1388 #define PGTBL_PUD_MODIFIED BIT(__PGTBL_PUD_MODIFIED)
1389 #define PGTBL_PMD_MODIFIED BIT(__PGTBL_PMD_MODIFIED)
1390 #define PGTBL_PTE_MODIFIED BIT(__PGTBL_PTE_MODIFIED)
1392 /* Page-Table Modification Mask */
1393 typedef unsigned int pgtbl_mod_mask;
1395 #endif /* !__ASSEMBLY__ */
1397 #ifndef io_remap_pfn_range
1398 #define io_remap_pfn_range remap_pfn_range
1401 #ifndef has_transparent_hugepage
1402 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1403 #define has_transparent_hugepage() 1
1405 #define has_transparent_hugepage() 0
1410 * On some architectures it depends on the mm if the p4d/pud or pmd
1411 * layer of the page table hierarchy is folded or not.
1413 #ifndef mm_p4d_folded
1414 #define mm_p4d_folded(mm) __is_defined(__PAGETABLE_P4D_FOLDED)
1417 #ifndef mm_pud_folded
1418 #define mm_pud_folded(mm) __is_defined(__PAGETABLE_PUD_FOLDED)
1421 #ifndef mm_pmd_folded
1422 #define mm_pmd_folded(mm) __is_defined(__PAGETABLE_PMD_FOLDED)
1426 * p?d_leaf() - true if this entry is a final mapping to a physical address.
1427 * This differs from p?d_huge() by the fact that they are always available (if
1428 * the architecture supports large pages at the appropriate level) even
1429 * if CONFIG_HUGETLB_PAGE is not defined.
1430 * Only meaningful when called on a valid entry.
1433 #define pgd_leaf(x) 0
1436 #define p4d_leaf(x) 0
1439 #define pud_leaf(x) 0
1442 #define pmd_leaf(x) 0
1445 #endif /* _LINUX_PGTABLE_H */