1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_GENERIC_PGTABLE_H
3 #define _ASM_GENERIC_PGTABLE_H
10 #include <linux/mm_types.h>
11 #include <linux/bug.h>
12 #include <linux/errno.h>
14 #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
15 defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
16 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
20 * On almost all architectures and configurations, 0 can be used as the
21 * upper ceiling to free_pgtables(): on many architectures it has the same
22 * effect as using TASK_SIZE. However, there is one configuration which
23 * must impose a more careful limit, to avoid freeing kernel pgtables.
25 #ifndef USER_PGTABLES_CEILING
26 #define USER_PGTABLES_CEILING 0UL
29 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
30 extern int ptep_set_access_flags(struct vm_area_struct *vma,
31 unsigned long address, pte_t *ptep,
32 pte_t entry, int dirty);
35 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
36 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
37 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
38 unsigned long address, pmd_t *pmdp,
39 pmd_t entry, int dirty);
40 extern int pudp_set_access_flags(struct vm_area_struct *vma,
41 unsigned long address, pud_t *pudp,
42 pud_t entry, int dirty);
44 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
45 unsigned long address, pmd_t *pmdp,
46 pmd_t entry, int dirty)
51 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
52 unsigned long address, pud_t *pudp,
53 pud_t entry, int dirty)
58 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
61 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
62 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
63 unsigned long address,
71 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
76 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
77 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
78 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
79 unsigned long address,
87 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
91 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
92 unsigned long address,
98 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
101 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
102 int ptep_clear_flush_young(struct vm_area_struct *vma,
103 unsigned long address, pte_t *ptep);
106 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
107 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
108 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
109 unsigned long address, pmd_t *pmdp);
112 * Despite relevant to THP only, this API is called from generic rmap code
113 * under PageTransHuge(), hence needs a dummy implementation for !THP
115 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
116 unsigned long address, pmd_t *pmdp)
121 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
124 #ifndef arch_has_hw_pte_young
126 * Return whether the accessed bit is supported on the local CPU.
128 * This stub assumes accessing through an old PTE triggers a page fault.
129 * Architectures that automatically set the access bit should overwrite it.
131 static inline bool arch_has_hw_pte_young(void)
137 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
138 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
139 unsigned long address,
143 pte_clear(mm, address, ptep);
148 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
149 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
150 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
151 unsigned long address,
158 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
159 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
160 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
161 unsigned long address,
169 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
170 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
172 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
173 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
174 static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
175 unsigned long address, pmd_t *pmdp,
178 return pmdp_huge_get_and_clear(mm, address, pmdp);
182 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
183 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
184 unsigned long address, pud_t *pudp,
187 return pudp_huge_get_and_clear(mm, address, pudp);
190 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
192 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
193 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
194 unsigned long address, pte_t *ptep,
198 pte = ptep_get_and_clear(mm, address, ptep);
204 * Some architectures may be able to avoid expensive synchronization
205 * primitives when modifications are made to PTE's which are already
206 * not present, or in the process of an address space destruction.
208 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
209 static inline void pte_clear_not_present_full(struct mm_struct *mm,
210 unsigned long address,
214 pte_clear(mm, address, ptep);
218 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
219 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
220 unsigned long address,
224 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
225 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
226 unsigned long address,
228 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
229 unsigned long address,
233 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
235 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
237 pte_t old_pte = *ptep;
238 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
242 #ifndef pte_savedwrite
243 #define pte_savedwrite pte_write
246 #ifndef pte_mk_savedwrite
247 #define pte_mk_savedwrite pte_mkwrite
250 #ifndef pte_clear_savedwrite
251 #define pte_clear_savedwrite pte_wrprotect
254 #ifndef pmd_savedwrite
255 #define pmd_savedwrite pmd_write
258 #ifndef pmd_mk_savedwrite
259 #define pmd_mk_savedwrite pmd_mkwrite
262 #ifndef pmd_clear_savedwrite
263 #define pmd_clear_savedwrite pmd_wrprotect
266 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
267 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
268 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
269 unsigned long address, pmd_t *pmdp)
271 pmd_t old_pmd = *pmdp;
272 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
275 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
276 unsigned long address, pmd_t *pmdp)
280 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
282 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
283 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
284 static inline void pudp_set_wrprotect(struct mm_struct *mm,
285 unsigned long address, pud_t *pudp)
287 pud_t old_pud = *pudp;
289 set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
292 static inline void pudp_set_wrprotect(struct mm_struct *mm,
293 unsigned long address, pud_t *pudp)
297 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
300 #ifndef pmdp_collapse_flush
301 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
302 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
303 unsigned long address, pmd_t *pmdp);
305 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
306 unsigned long address,
312 #define pmdp_collapse_flush pmdp_collapse_flush
313 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
316 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
317 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
321 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
322 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
325 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
327 * This is an implementation of pmdp_establish() that is only suitable for an
328 * architecture that doesn't have hardware dirty/accessed bits. In this case we
329 * can't race with CPU which sets these bits and non-atomic aproach is fine.
331 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
332 unsigned long address, pmd_t *pmdp, pmd_t pmd)
334 pmd_t old_pmd = *pmdp;
335 set_pmd_at(vma->vm_mm, address, pmdp, pmd);
340 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
341 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
345 #ifndef __HAVE_ARCH_PTE_SAME
346 static inline int pte_same(pte_t pte_a, pte_t pte_b)
348 return pte_val(pte_a) == pte_val(pte_b);
352 #ifndef __HAVE_ARCH_PTE_UNUSED
354 * Some architectures provide facilities to virtualization guests
355 * so that they can flag allocated pages as unused. This allows the
356 * host to transparently reclaim unused pages. This function returns
357 * whether the pte's page is unused.
359 static inline int pte_unused(pte_t pte)
365 #ifndef pte_access_permitted
366 #define pte_access_permitted(pte, write) \
367 (pte_present(pte) && (!(write) || pte_write(pte)))
370 #ifndef pmd_access_permitted
371 #define pmd_access_permitted(pmd, write) \
372 (pmd_present(pmd) && (!(write) || pmd_write(pmd)))
375 #ifndef pud_access_permitted
376 #define pud_access_permitted(pud, write) \
377 (pud_present(pud) && (!(write) || pud_write(pud)))
380 #ifndef p4d_access_permitted
381 #define p4d_access_permitted(p4d, write) \
382 (p4d_present(p4d) && (!(write) || p4d_write(p4d)))
385 #ifndef pgd_access_permitted
386 #define pgd_access_permitted(pgd, write) \
387 (pgd_present(pgd) && (!(write) || pgd_write(pgd)))
390 #ifndef __HAVE_ARCH_PMD_SAME
391 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
393 return pmd_val(pmd_a) == pmd_val(pmd_b);
396 static inline int pud_same(pud_t pud_a, pud_t pud_b)
398 return pud_val(pud_a) == pud_val(pud_b);
402 #ifndef __HAVE_ARCH_P4D_SAME
403 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
405 return p4d_val(p4d_a) == p4d_val(p4d_b);
409 #ifndef __HAVE_ARCH_PGD_SAME
410 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
412 return pgd_val(pgd_a) == pgd_val(pgd_b);
417 * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
418 * TLB flush will be required as a result of the "set". For example, use
419 * in scenarios where it is known ahead of time that the routine is
420 * setting non-present entries, or re-setting an existing entry to the
421 * same value. Otherwise, use the typical "set" helpers and flush the
424 #define set_pte_safe(ptep, pte) \
426 WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
427 set_pte(ptep, pte); \
430 #define set_pmd_safe(pmdp, pmd) \
432 WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
433 set_pmd(pmdp, pmd); \
436 #define set_pud_safe(pudp, pud) \
438 WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
439 set_pud(pudp, pud); \
442 #define set_p4d_safe(p4dp, p4d) \
444 WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
445 set_p4d(p4dp, p4d); \
448 #define set_pgd_safe(pgdp, pgd) \
450 WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
451 set_pgd(pgdp, pgd); \
454 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
456 * Some architectures support metadata associated with a page. When a
457 * page is being swapped out, this metadata must be saved so it can be
458 * restored when the page is swapped back in. SPARC M7 and newer
459 * processors support an ADI (Application Data Integrity) tag for the
460 * page as metadata for the page. arch_do_swap_page() can restore this
461 * metadata when a page is swapped back in.
463 static inline void arch_do_swap_page(struct mm_struct *mm,
464 struct vm_area_struct *vma,
466 pte_t pte, pte_t oldpte)
472 #ifndef __HAVE_ARCH_UNMAP_ONE
474 * Some architectures support metadata associated with a page. When a
475 * page is being swapped out, this metadata must be saved so it can be
476 * restored when the page is swapped back in. SPARC M7 and newer
477 * processors support an ADI (Application Data Integrity) tag for the
478 * page as metadata for the page. arch_unmap_one() can save this
479 * metadata on a swap-out of a page.
481 static inline int arch_unmap_one(struct mm_struct *mm,
482 struct vm_area_struct *vma,
490 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
491 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
494 #ifndef __HAVE_ARCH_MOVE_PTE
495 #define move_pte(pte, prot, old_addr, new_addr) (pte)
498 #ifndef pte_accessible
499 # define pte_accessible(mm, pte) ((void)(pte), 1)
502 #ifndef flush_tlb_fix_spurious_fault
503 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
506 #ifndef pgprot_noncached
507 #define pgprot_noncached(prot) (prot)
510 #ifndef pgprot_writecombine
511 #define pgprot_writecombine pgprot_noncached
514 #ifndef pgprot_writethrough
515 #define pgprot_writethrough pgprot_noncached
518 #ifndef pgprot_device
519 #define pgprot_device pgprot_noncached
522 #ifndef pgprot_modify
523 #define pgprot_modify pgprot_modify
524 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
526 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
527 newprot = pgprot_noncached(newprot);
528 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
529 newprot = pgprot_writecombine(newprot);
530 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
531 newprot = pgprot_device(newprot);
537 * When walking page tables, get the address of the next boundary,
538 * or the end address of the range if that comes earlier. Although no
539 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
542 #define pgd_addr_end(addr, end) \
543 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
544 (__boundary - 1 < (end) - 1)? __boundary: (end); \
548 #define p4d_addr_end(addr, end) \
549 ({ unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK; \
550 (__boundary - 1 < (end) - 1)? __boundary: (end); \
555 #define pud_addr_end(addr, end) \
556 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
557 (__boundary - 1 < (end) - 1)? __boundary: (end); \
562 #define pmd_addr_end(addr, end) \
563 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
564 (__boundary - 1 < (end) - 1)? __boundary: (end); \
569 * When walking page tables, we usually want to skip any p?d_none entries;
570 * and any p?d_bad entries - reporting the error before resetting to none.
571 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
573 void pgd_clear_bad(pgd_t *);
574 void p4d_clear_bad(p4d_t *);
575 void pud_clear_bad(pud_t *);
576 void pmd_clear_bad(pmd_t *);
578 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
582 if (unlikely(pgd_bad(*pgd))) {
589 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
593 if (unlikely(p4d_bad(*p4d))) {
600 static inline int pud_none_or_clear_bad(pud_t *pud)
604 if (unlikely(pud_bad(*pud))) {
611 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
615 if (unlikely(pmd_bad(*pmd))) {
622 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
627 * Get the current pte state, but zero it out to make it
628 * non-present, preventing the hardware from asynchronously
631 return ptep_get_and_clear(vma->vm_mm, addr, ptep);
634 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
636 pte_t *ptep, pte_t pte)
639 * The pte is non-present, so there's no hardware state to
642 set_pte_at(vma->vm_mm, addr, ptep, pte);
645 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
647 * Start a pte protection read-modify-write transaction, which
648 * protects against asynchronous hardware modifications to the pte.
649 * The intention is not to prevent the hardware from making pte
650 * updates, but to prevent any updates it may make from being lost.
652 * This does not protect against other software modifications of the
653 * pte; the appropriate pte lock must be held over the transation.
655 * Note that this interface is intended to be batchable, meaning that
656 * ptep_modify_prot_commit may not actually update the pte, but merely
657 * queue the update to be done at some later time. The update must be
658 * actually committed before the pte lock is released, however.
660 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
664 return __ptep_modify_prot_start(vma, addr, ptep);
668 * Commit an update to a pte, leaving any hardware-controlled bits in
669 * the PTE unmodified.
671 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
673 pte_t *ptep, pte_t old_pte, pte_t pte)
675 __ptep_modify_prot_commit(vma, addr, ptep, pte);
677 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
678 #endif /* CONFIG_MMU */
681 * No-op macros that just return the current protection value. Defined here
682 * because these macros can be used used even if CONFIG_MMU is not defined.
684 #ifndef pgprot_encrypted
685 #define pgprot_encrypted(prot) (prot)
688 #ifndef pgprot_decrypted
689 #define pgprot_decrypted(prot) (prot)
693 * A facility to provide lazy MMU batching. This allows PTE updates and
694 * page invalidations to be delayed until a call to leave lazy MMU mode
695 * is issued. Some architectures may benefit from doing this, and it is
696 * beneficial for both shadow and direct mode hypervisors, which may batch
697 * the PTE updates which happen during this window. Note that using this
698 * interface requires that read hazards be removed from the code. A read
699 * hazard could result in the direct mode hypervisor case, since the actual
700 * write to the page tables may not yet have taken place, so reads though
701 * a raw PTE pointer after it has been modified are not guaranteed to be
702 * up to date. This mode can only be entered and left under the protection of
703 * the page table locks for all page tables which may be modified. In the UP
704 * case, this is required so that preemption is disabled, and in the SMP case,
705 * it must synchronize the delayed page table writes properly on other CPUs.
707 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
708 #define arch_enter_lazy_mmu_mode() do {} while (0)
709 #define arch_leave_lazy_mmu_mode() do {} while (0)
710 #define arch_flush_lazy_mmu_mode() do {} while (0)
714 * A facility to provide batching of the reload of page tables and
715 * other process state with the actual context switch code for
716 * paravirtualized guests. By convention, only one of the batched
717 * update (lazy) modes (CPU, MMU) should be active at any given time,
718 * entry should never be nested, and entry and exits should always be
719 * paired. This is for sanity of maintaining and reasoning about the
720 * kernel code. In this case, the exit (end of the context switch) is
721 * in architecture-specific code, and so doesn't need a generic
724 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
725 #define arch_start_context_switch(prev) do {} while (0)
728 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
729 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
730 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
735 static inline int pmd_swp_soft_dirty(pmd_t pmd)
740 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
745 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
746 static inline int pte_soft_dirty(pte_t pte)
751 static inline int pmd_soft_dirty(pmd_t pmd)
756 static inline pte_t pte_mksoft_dirty(pte_t pte)
761 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
766 static inline pte_t pte_clear_soft_dirty(pte_t pte)
771 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
776 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
781 static inline int pte_swp_soft_dirty(pte_t pte)
786 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
791 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
796 static inline int pmd_swp_soft_dirty(pmd_t pmd)
801 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
807 #ifndef __HAVE_PFNMAP_TRACKING
809 * Interfaces that can be used by architecture code to keep track of
810 * memory type of pfn mappings specified by the remap_pfn_range,
815 * track_pfn_remap is called when a _new_ pfn mapping is being established
816 * by remap_pfn_range() for physical range indicated by pfn and size.
818 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
819 unsigned long pfn, unsigned long addr,
826 * track_pfn_insert is called when a _new_ single pfn is established
827 * by vmf_insert_pfn().
829 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
835 * track_pfn_copy is called when vma that is covering the pfnmap gets
836 * copied through copy_page_range().
838 static inline int track_pfn_copy(struct vm_area_struct *vma)
844 * untrack_pfn is called while unmapping a pfnmap for a region.
845 * untrack can be called for a specific region indicated by pfn and size or
846 * can be for the entire vma (in which case pfn, size are zero).
848 static inline void untrack_pfn(struct vm_area_struct *vma,
849 unsigned long pfn, unsigned long size)
854 * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
856 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
860 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
861 unsigned long pfn, unsigned long addr,
863 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
865 extern int track_pfn_copy(struct vm_area_struct *vma);
866 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
868 extern void untrack_pfn_moved(struct vm_area_struct *vma);
871 #ifdef __HAVE_COLOR_ZERO_PAGE
872 static inline int is_zero_pfn(unsigned long pfn)
874 extern unsigned long zero_pfn;
875 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
876 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
879 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
882 static inline int is_zero_pfn(unsigned long pfn)
884 extern unsigned long zero_pfn;
885 return pfn == zero_pfn;
888 static inline unsigned long my_zero_pfn(unsigned long addr)
890 extern unsigned long zero_pfn;
897 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
898 static inline int pmd_trans_huge(pmd_t pmd)
903 static inline int pmd_write(pmd_t pmd)
908 #endif /* pmd_write */
909 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
912 static inline int pud_write(pud_t pud)
917 #endif /* pud_write */
919 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
920 (defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
921 !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
922 static inline int pud_trans_huge(pud_t pud)
928 #ifndef pmd_read_atomic
929 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
932 * Depend on compiler for an atomic pmd read. NOTE: this is
933 * only going to work, if the pmdval_t isn't larger than
940 #ifndef arch_needs_pgtable_deposit
941 #define arch_needs_pgtable_deposit() (false)
944 * This function is meant to be used by sites walking pagetables with
945 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
946 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
947 * into a null pmd and the transhuge page fault can convert a null pmd
948 * into an hugepmd or into a regular pmd (if the hugepage allocation
949 * fails). While holding the mmap_sem in read mode the pmd becomes
950 * stable and stops changing under us only if it's not null and not a
951 * transhuge pmd. When those races occurs and this function makes a
952 * difference vs the standard pmd_none_or_clear_bad, the result is
953 * undefined so behaving like if the pmd was none is safe (because it
954 * can return none anyway). The compiler level barrier() is critically
955 * important to compute the two checks atomically on the same pmdval.
957 * For 32bit kernels with a 64bit large pmd_t this automatically takes
958 * care of reading the pmd atomically to avoid SMP race conditions
959 * against pmd_populate() when the mmap_sem is hold for reading by the
960 * caller (a special atomic read not done by "gcc" as in the generic
961 * version above, is also needed when THP is disabled because the page
962 * fault can populate the pmd from under us).
964 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
966 pmd_t pmdval = pmd_read_atomic(pmd);
968 * The barrier will stabilize the pmdval in a register or on
969 * the stack so that it will stop changing under the code.
971 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
972 * pmd_read_atomic is allowed to return a not atomic pmdval
973 * (for example pointing to an hugepage that has never been
974 * mapped in the pmd). The below checks will only care about
975 * the low part of the pmd with 32bit PAE x86 anyway, with the
976 * exception of pmd_none(). So the important thing is that if
977 * the low part of the pmd is found null, the high part will
978 * be also null or the pmd_none() check below would be
981 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
985 * !pmd_present() checks for pmd migration entries
987 * The complete check uses is_pmd_migration_entry() in linux/swapops.h
988 * But using that requires moving current function and pmd_trans_unstable()
989 * to linux/swapops.h to resovle dependency, which is too much code move.
991 * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
992 * because !pmd_present() pages can only be under migration not swapped
995 * pmd_none() is preseved for future condition checks on pmd migration
996 * entries and not confusing with this function name, although it is
997 * redundant with !pmd_present().
999 if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
1000 (IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
1002 if (unlikely(pmd_bad(pmdval))) {
1010 * This is a noop if Transparent Hugepage Support is not built into
1011 * the kernel. Otherwise it is equivalent to
1012 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
1013 * places that already verified the pmd is not none and they want to
1014 * walk ptes while holding the mmap sem in read mode (write mode don't
1015 * need this). If THP is not enabled, the pmd can't go away under the
1016 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
1017 * run a pmd_trans_unstable before walking the ptes after
1018 * split_huge_pmd returns (because it may have run when the pmd become
1019 * null, but then a page fault can map in a THP and not a regular page).
1021 static inline int pmd_trans_unstable(pmd_t *pmd)
1023 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1024 return pmd_none_or_trans_huge_or_clear_bad(pmd);
1030 #ifndef CONFIG_NUMA_BALANCING
1032 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
1033 * the only case the kernel cares is for NUMA balancing and is only ever set
1034 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
1035 * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
1036 * is the responsibility of the caller to distinguish between PROT_NONE
1037 * protections and NUMA hinting fault protections.
1039 static inline int pte_protnone(pte_t pte)
1044 static inline int pmd_protnone(pmd_t pmd)
1048 #endif /* CONFIG_NUMA_BALANCING */
1050 #endif /* CONFIG_MMU */
1052 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1054 #ifndef __PAGETABLE_P4D_FOLDED
1055 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1056 int p4d_clear_huge(p4d_t *p4d);
1058 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1062 static inline int p4d_clear_huge(p4d_t *p4d)
1066 #endif /* !__PAGETABLE_P4D_FOLDED */
1068 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1069 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1070 int pud_clear_huge(pud_t *pud);
1071 int pmd_clear_huge(pmd_t *pmd);
1072 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1073 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1074 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1075 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1076 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1080 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1084 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1088 static inline int p4d_clear_huge(p4d_t *p4d)
1092 static inline int pud_clear_huge(pud_t *pud)
1096 static inline int pmd_clear_huge(pmd_t *pmd)
1100 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1104 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1108 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1112 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
1114 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1115 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1117 * ARCHes with special requirements for evicting THP backing TLB entries can
1118 * implement this. Otherwise also, it can help optimize normal TLB flush in
1119 * THP regime. stock flush_tlb_range() typically has optimization to nuke the
1120 * entire TLB TLB if flush span is greater than a threshold, which will
1121 * likely be true for a single huge page. Thus a single thp flush will
1122 * invalidate the entire TLB which is not desitable.
1123 * e.g. see arch/arc: flush_pmd_tlb_range
1125 #define flush_pmd_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1126 #define flush_pud_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1128 #define flush_pmd_tlb_range(vma, addr, end) BUILD_BUG()
1129 #define flush_pud_tlb_range(vma, addr, end) BUILD_BUG()
1134 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1135 unsigned long size, pgprot_t *vma_prot);
1137 #ifndef CONFIG_X86_ESPFIX64
1138 static inline void init_espfix_bsp(void) { }
1141 extern void __init pgtable_cache_init(void);
1143 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1144 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1149 static inline bool arch_has_pfn_modify_check(void)
1153 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1156 * Architecture PAGE_KERNEL_* fallbacks
1158 * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1159 * because they really don't support them, or the port needs to be updated to
1160 * reflect the required functionality. Below are a set of relatively safe
1161 * fallbacks, as best effort, which we can count on in lieu of the architectures
1162 * not defining them on their own yet.
1165 #ifndef PAGE_KERNEL_RO
1166 # define PAGE_KERNEL_RO PAGE_KERNEL
1169 #ifndef PAGE_KERNEL_EXEC
1170 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1173 #endif /* !__ASSEMBLY__ */
1175 #ifndef io_remap_pfn_range
1176 #define io_remap_pfn_range remap_pfn_range
1179 #ifndef has_transparent_hugepage
1180 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1181 #define has_transparent_hugepage() 1
1183 #define has_transparent_hugepage() 0
1188 * On some architectures it depends on the mm if the p4d/pud or pmd
1189 * layer of the page table hierarchy is folded or not.
1191 #ifndef mm_p4d_folded
1192 #define mm_p4d_folded(mm) __is_defined(__PAGETABLE_P4D_FOLDED)
1195 #ifndef mm_pud_folded
1196 #define mm_pud_folded(mm) __is_defined(__PAGETABLE_PUD_FOLDED)
1199 #ifndef mm_pmd_folded
1200 #define mm_pmd_folded(mm) __is_defined(__PAGETABLE_PMD_FOLDED)
1204 * p?d_leaf() - true if this entry is a final mapping to a physical address.
1205 * This differs from p?d_huge() by the fact that they are always available (if
1206 * the architecture supports large pages at the appropriate level) even
1207 * if CONFIG_HUGETLB_PAGE is not defined.
1208 * Only meaningful when called on a valid entry.
1211 #define pgd_leaf(x) 0
1214 #define p4d_leaf(x) 0
1217 #define pud_leaf(x) 0
1220 #define pmd_leaf(x) 0
1224 static inline unsigned long pud_index(unsigned long address)
1226 return (address >> PUD_SHIFT) & (PTRS_PER_PUD - 1);
1228 #define pud_index pud_index
1231 #endif /* _ASM_GENERIC_PGTABLE_H */