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 __HAVE_ARCH_PTEP_GET_AND_CLEAR
125 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
126 unsigned long address,
130 pte_clear(mm, address, ptep);
135 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
136 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
137 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
138 unsigned long address,
145 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
146 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
147 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
148 unsigned long address,
156 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
157 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
159 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
160 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
161 static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
162 unsigned long address, pmd_t *pmdp,
165 return pmdp_huge_get_and_clear(mm, address, pmdp);
169 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
170 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
171 unsigned long address, pud_t *pudp,
174 return pudp_huge_get_and_clear(mm, address, pudp);
177 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
179 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
180 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
181 unsigned long address, pte_t *ptep,
185 pte = ptep_get_and_clear(mm, address, ptep);
191 * Some architectures may be able to avoid expensive synchronization
192 * primitives when modifications are made to PTE's which are already
193 * not present, or in the process of an address space destruction.
195 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
196 static inline void pte_clear_not_present_full(struct mm_struct *mm,
197 unsigned long address,
201 pte_clear(mm, address, ptep);
205 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
206 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
207 unsigned long address,
211 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
212 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
213 unsigned long address,
215 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
216 unsigned long address,
220 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
222 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
224 pte_t old_pte = *ptep;
225 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
229 #ifndef pte_savedwrite
230 #define pte_savedwrite pte_write
233 #ifndef pte_mk_savedwrite
234 #define pte_mk_savedwrite pte_mkwrite
237 #ifndef pte_clear_savedwrite
238 #define pte_clear_savedwrite pte_wrprotect
241 #ifndef pmd_savedwrite
242 #define pmd_savedwrite pmd_write
245 #ifndef pmd_mk_savedwrite
246 #define pmd_mk_savedwrite pmd_mkwrite
249 #ifndef pmd_clear_savedwrite
250 #define pmd_clear_savedwrite pmd_wrprotect
253 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
254 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
255 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
256 unsigned long address, pmd_t *pmdp)
258 pmd_t old_pmd = *pmdp;
259 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
262 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
263 unsigned long address, pmd_t *pmdp)
267 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
269 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
270 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
271 static inline void pudp_set_wrprotect(struct mm_struct *mm,
272 unsigned long address, pud_t *pudp)
274 pud_t old_pud = *pudp;
276 set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
279 static inline void pudp_set_wrprotect(struct mm_struct *mm,
280 unsigned long address, pud_t *pudp)
284 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
287 #ifndef pmdp_collapse_flush
288 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
289 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
290 unsigned long address, pmd_t *pmdp);
292 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
293 unsigned long address,
299 #define pmdp_collapse_flush pmdp_collapse_flush
300 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
303 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
304 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
308 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
309 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
312 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
314 * This is an implementation of pmdp_establish() that is only suitable for an
315 * architecture that doesn't have hardware dirty/accessed bits. In this case we
316 * can't race with CPU which sets these bits and non-atomic aproach is fine.
318 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
319 unsigned long address, pmd_t *pmdp, pmd_t pmd)
321 pmd_t old_pmd = *pmdp;
322 set_pmd_at(vma->vm_mm, address, pmdp, pmd);
327 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
328 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
332 #ifndef __HAVE_ARCH_PTE_SAME
333 static inline int pte_same(pte_t pte_a, pte_t pte_b)
335 return pte_val(pte_a) == pte_val(pte_b);
339 #ifndef __HAVE_ARCH_PTE_UNUSED
341 * Some architectures provide facilities to virtualization guests
342 * so that they can flag allocated pages as unused. This allows the
343 * host to transparently reclaim unused pages. This function returns
344 * whether the pte's page is unused.
346 static inline int pte_unused(pte_t pte)
352 #ifndef pte_access_permitted
353 #define pte_access_permitted(pte, write) \
354 (pte_present(pte) && (!(write) || pte_write(pte)))
357 #ifndef pmd_access_permitted
358 #define pmd_access_permitted(pmd, write) \
359 (pmd_present(pmd) && (!(write) || pmd_write(pmd)))
362 #ifndef pud_access_permitted
363 #define pud_access_permitted(pud, write) \
364 (pud_present(pud) && (!(write) || pud_write(pud)))
367 #ifndef p4d_access_permitted
368 #define p4d_access_permitted(p4d, write) \
369 (p4d_present(p4d) && (!(write) || p4d_write(p4d)))
372 #ifndef pgd_access_permitted
373 #define pgd_access_permitted(pgd, write) \
374 (pgd_present(pgd) && (!(write) || pgd_write(pgd)))
377 #ifndef __HAVE_ARCH_PMD_SAME
378 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
380 return pmd_val(pmd_a) == pmd_val(pmd_b);
383 static inline int pud_same(pud_t pud_a, pud_t pud_b)
385 return pud_val(pud_a) == pud_val(pud_b);
389 #ifndef __HAVE_ARCH_P4D_SAME
390 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
392 return p4d_val(p4d_a) == p4d_val(p4d_b);
396 #ifndef __HAVE_ARCH_PGD_SAME
397 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
399 return pgd_val(pgd_a) == pgd_val(pgd_b);
404 * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
405 * TLB flush will be required as a result of the "set". For example, use
406 * in scenarios where it is known ahead of time that the routine is
407 * setting non-present entries, or re-setting an existing entry to the
408 * same value. Otherwise, use the typical "set" helpers and flush the
411 #define set_pte_safe(ptep, pte) \
413 WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
414 set_pte(ptep, pte); \
417 #define set_pmd_safe(pmdp, pmd) \
419 WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
420 set_pmd(pmdp, pmd); \
423 #define set_pud_safe(pudp, pud) \
425 WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
426 set_pud(pudp, pud); \
429 #define set_p4d_safe(p4dp, p4d) \
431 WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
432 set_p4d(p4dp, p4d); \
435 #define set_pgd_safe(pgdp, pgd) \
437 WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
438 set_pgd(pgdp, pgd); \
441 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
443 * Some architectures support metadata associated with a page. When a
444 * page is being swapped out, this metadata must be saved so it can be
445 * restored when the page is swapped back in. SPARC M7 and newer
446 * processors support an ADI (Application Data Integrity) tag for the
447 * page as metadata for the page. arch_do_swap_page() can restore this
448 * metadata when a page is swapped back in.
450 static inline void arch_do_swap_page(struct mm_struct *mm,
451 struct vm_area_struct *vma,
453 pte_t pte, pte_t oldpte)
459 #ifndef __HAVE_ARCH_UNMAP_ONE
461 * Some architectures support metadata associated with a page. When a
462 * page is being swapped out, this metadata must be saved so it can be
463 * restored when the page is swapped back in. SPARC M7 and newer
464 * processors support an ADI (Application Data Integrity) tag for the
465 * page as metadata for the page. arch_unmap_one() can save this
466 * metadata on a swap-out of a page.
468 static inline int arch_unmap_one(struct mm_struct *mm,
469 struct vm_area_struct *vma,
477 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
478 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
481 #ifndef __HAVE_ARCH_MOVE_PTE
482 #define move_pte(pte, prot, old_addr, new_addr) (pte)
485 #ifndef pte_accessible
486 # define pte_accessible(mm, pte) ((void)(pte), 1)
489 #ifndef flush_tlb_fix_spurious_fault
490 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
493 #ifndef pgprot_noncached
494 #define pgprot_noncached(prot) (prot)
497 #ifndef pgprot_writecombine
498 #define pgprot_writecombine pgprot_noncached
501 #ifndef pgprot_writethrough
502 #define pgprot_writethrough pgprot_noncached
505 #ifndef pgprot_device
506 #define pgprot_device pgprot_noncached
509 #ifndef pgprot_modify
510 #define pgprot_modify pgprot_modify
511 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
513 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
514 newprot = pgprot_noncached(newprot);
515 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
516 newprot = pgprot_writecombine(newprot);
517 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
518 newprot = pgprot_device(newprot);
524 * When walking page tables, get the address of the next boundary,
525 * or the end address of the range if that comes earlier. Although no
526 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
529 #define pgd_addr_end(addr, end) \
530 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
531 (__boundary - 1 < (end) - 1)? __boundary: (end); \
535 #define p4d_addr_end(addr, end) \
536 ({ unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK; \
537 (__boundary - 1 < (end) - 1)? __boundary: (end); \
542 #define pud_addr_end(addr, end) \
543 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
544 (__boundary - 1 < (end) - 1)? __boundary: (end); \
549 #define pmd_addr_end(addr, end) \
550 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
551 (__boundary - 1 < (end) - 1)? __boundary: (end); \
556 * When walking page tables, we usually want to skip any p?d_none entries;
557 * and any p?d_bad entries - reporting the error before resetting to none.
558 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
560 void pgd_clear_bad(pgd_t *);
561 void p4d_clear_bad(p4d_t *);
562 void pud_clear_bad(pud_t *);
563 void pmd_clear_bad(pmd_t *);
565 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
569 if (unlikely(pgd_bad(*pgd))) {
576 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
580 if (unlikely(p4d_bad(*p4d))) {
587 static inline int pud_none_or_clear_bad(pud_t *pud)
591 if (unlikely(pud_bad(*pud))) {
598 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
602 if (unlikely(pmd_bad(*pmd))) {
609 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
614 * Get the current pte state, but zero it out to make it
615 * non-present, preventing the hardware from asynchronously
618 return ptep_get_and_clear(vma->vm_mm, addr, ptep);
621 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
623 pte_t *ptep, pte_t pte)
626 * The pte is non-present, so there's no hardware state to
629 set_pte_at(vma->vm_mm, addr, ptep, pte);
632 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
634 * Start a pte protection read-modify-write transaction, which
635 * protects against asynchronous hardware modifications to the pte.
636 * The intention is not to prevent the hardware from making pte
637 * updates, but to prevent any updates it may make from being lost.
639 * This does not protect against other software modifications of the
640 * pte; the appropriate pte lock must be held over the transation.
642 * Note that this interface is intended to be batchable, meaning that
643 * ptep_modify_prot_commit may not actually update the pte, but merely
644 * queue the update to be done at some later time. The update must be
645 * actually committed before the pte lock is released, however.
647 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
651 return __ptep_modify_prot_start(vma, addr, ptep);
655 * Commit an update to a pte, leaving any hardware-controlled bits in
656 * the PTE unmodified.
658 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
660 pte_t *ptep, pte_t pte)
662 __ptep_modify_prot_commit(vma, addr, ptep, pte);
664 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
665 #endif /* CONFIG_MMU */
668 * No-op macros that just return the current protection value. Defined here
669 * because these macros can be used used even if CONFIG_MMU is not defined.
671 #ifndef pgprot_encrypted
672 #define pgprot_encrypted(prot) (prot)
675 #ifndef pgprot_decrypted
676 #define pgprot_decrypted(prot) (prot)
680 * A facility to provide lazy MMU batching. This allows PTE updates and
681 * page invalidations to be delayed until a call to leave lazy MMU mode
682 * is issued. Some architectures may benefit from doing this, and it is
683 * beneficial for both shadow and direct mode hypervisors, which may batch
684 * the PTE updates which happen during this window. Note that using this
685 * interface requires that read hazards be removed from the code. A read
686 * hazard could result in the direct mode hypervisor case, since the actual
687 * write to the page tables may not yet have taken place, so reads though
688 * a raw PTE pointer after it has been modified are not guaranteed to be
689 * up to date. This mode can only be entered and left under the protection of
690 * the page table locks for all page tables which may be modified. In the UP
691 * case, this is required so that preemption is disabled, and in the SMP case,
692 * it must synchronize the delayed page table writes properly on other CPUs.
694 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
695 #define arch_enter_lazy_mmu_mode() do {} while (0)
696 #define arch_leave_lazy_mmu_mode() do {} while (0)
697 #define arch_flush_lazy_mmu_mode() do {} while (0)
701 * A facility to provide batching of the reload of page tables and
702 * other process state with the actual context switch code for
703 * paravirtualized guests. By convention, only one of the batched
704 * update (lazy) modes (CPU, MMU) should be active at any given time,
705 * entry should never be nested, and entry and exits should always be
706 * paired. This is for sanity of maintaining and reasoning about the
707 * kernel code. In this case, the exit (end of the context switch) is
708 * in architecture-specific code, and so doesn't need a generic
711 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
712 #define arch_start_context_switch(prev) do {} while (0)
715 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
716 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
717 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
722 static inline int pmd_swp_soft_dirty(pmd_t pmd)
727 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
732 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
733 static inline int pte_soft_dirty(pte_t pte)
738 static inline int pmd_soft_dirty(pmd_t pmd)
743 static inline pte_t pte_mksoft_dirty(pte_t pte)
748 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
753 static inline pte_t pte_clear_soft_dirty(pte_t pte)
758 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
763 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
768 static inline int pte_swp_soft_dirty(pte_t pte)
773 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
778 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
783 static inline int pmd_swp_soft_dirty(pmd_t pmd)
788 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
794 #ifndef __HAVE_PFNMAP_TRACKING
796 * Interfaces that can be used by architecture code to keep track of
797 * memory type of pfn mappings specified by the remap_pfn_range,
802 * track_pfn_remap is called when a _new_ pfn mapping is being established
803 * by remap_pfn_range() for physical range indicated by pfn and size.
805 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
806 unsigned long pfn, unsigned long addr,
813 * track_pfn_insert is called when a _new_ single pfn is established
814 * by vmf_insert_pfn().
816 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
822 * track_pfn_copy is called when vma that is covering the pfnmap gets
823 * copied through copy_page_range().
825 static inline int track_pfn_copy(struct vm_area_struct *vma)
831 * untrack_pfn is called while unmapping a pfnmap for a region.
832 * untrack can be called for a specific region indicated by pfn and size or
833 * can be for the entire vma (in which case pfn, size are zero).
835 static inline void untrack_pfn(struct vm_area_struct *vma,
836 unsigned long pfn, unsigned long size)
841 * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
843 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
847 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
848 unsigned long pfn, unsigned long addr,
850 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
852 extern int track_pfn_copy(struct vm_area_struct *vma);
853 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
855 extern void untrack_pfn_moved(struct vm_area_struct *vma);
858 #ifdef __HAVE_COLOR_ZERO_PAGE
859 static inline int is_zero_pfn(unsigned long pfn)
861 extern unsigned long zero_pfn;
862 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
863 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
866 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
869 static inline int is_zero_pfn(unsigned long pfn)
871 extern unsigned long zero_pfn;
872 return pfn == zero_pfn;
875 static inline unsigned long my_zero_pfn(unsigned long addr)
877 extern unsigned long zero_pfn;
884 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
885 static inline int pmd_trans_huge(pmd_t pmd)
890 static inline int pmd_write(pmd_t pmd)
895 #endif /* pmd_write */
896 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
899 static inline int pud_write(pud_t pud)
904 #endif /* pud_write */
906 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
907 (defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
908 !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
909 static inline int pud_trans_huge(pud_t pud)
915 #ifndef pmd_read_atomic
916 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
919 * Depend on compiler for an atomic pmd read. NOTE: this is
920 * only going to work, if the pmdval_t isn't larger than
927 #ifndef arch_needs_pgtable_deposit
928 #define arch_needs_pgtable_deposit() (false)
931 * This function is meant to be used by sites walking pagetables with
932 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
933 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
934 * into a null pmd and the transhuge page fault can convert a null pmd
935 * into an hugepmd or into a regular pmd (if the hugepage allocation
936 * fails). While holding the mmap_sem in read mode the pmd becomes
937 * stable and stops changing under us only if it's not null and not a
938 * transhuge pmd. When those races occurs and this function makes a
939 * difference vs the standard pmd_none_or_clear_bad, the result is
940 * undefined so behaving like if the pmd was none is safe (because it
941 * can return none anyway). The compiler level barrier() is critically
942 * important to compute the two checks atomically on the same pmdval.
944 * For 32bit kernels with a 64bit large pmd_t this automatically takes
945 * care of reading the pmd atomically to avoid SMP race conditions
946 * against pmd_populate() when the mmap_sem is hold for reading by the
947 * caller (a special atomic read not done by "gcc" as in the generic
948 * version above, is also needed when THP is disabled because the page
949 * fault can populate the pmd from under us).
951 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
953 pmd_t pmdval = pmd_read_atomic(pmd);
955 * The barrier will stabilize the pmdval in a register or on
956 * the stack so that it will stop changing under the code.
958 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
959 * pmd_read_atomic is allowed to return a not atomic pmdval
960 * (for example pointing to an hugepage that has never been
961 * mapped in the pmd). The below checks will only care about
962 * the low part of the pmd with 32bit PAE x86 anyway, with the
963 * exception of pmd_none(). So the important thing is that if
964 * the low part of the pmd is found null, the high part will
965 * be also null or the pmd_none() check below would be
968 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
972 * !pmd_present() checks for pmd migration entries
974 * The complete check uses is_pmd_migration_entry() in linux/swapops.h
975 * But using that requires moving current function and pmd_trans_unstable()
976 * to linux/swapops.h to resovle dependency, which is too much code move.
978 * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
979 * because !pmd_present() pages can only be under migration not swapped
982 * pmd_none() is preseved for future condition checks on pmd migration
983 * entries and not confusing with this function name, although it is
984 * redundant with !pmd_present().
986 if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
987 (IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
989 if (unlikely(pmd_bad(pmdval))) {
997 * This is a noop if Transparent Hugepage Support is not built into
998 * the kernel. Otherwise it is equivalent to
999 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
1000 * places that already verified the pmd is not none and they want to
1001 * walk ptes while holding the mmap sem in read mode (write mode don't
1002 * need this). If THP is not enabled, the pmd can't go away under the
1003 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
1004 * run a pmd_trans_unstable before walking the ptes after
1005 * split_huge_page_pmd returns (because it may have run when the pmd
1006 * become null, but then a page fault can map in a THP and not a
1009 static inline int pmd_trans_unstable(pmd_t *pmd)
1011 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1012 return pmd_none_or_trans_huge_or_clear_bad(pmd);
1018 #ifndef CONFIG_NUMA_BALANCING
1020 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
1021 * the only case the kernel cares is for NUMA balancing and is only ever set
1022 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
1023 * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
1024 * is the responsibility of the caller to distinguish between PROT_NONE
1025 * protections and NUMA hinting fault protections.
1027 static inline int pte_protnone(pte_t pte)
1032 static inline int pmd_protnone(pmd_t pmd)
1036 #endif /* CONFIG_NUMA_BALANCING */
1038 #endif /* CONFIG_MMU */
1040 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1042 #ifndef __PAGETABLE_P4D_FOLDED
1043 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1044 int p4d_clear_huge(p4d_t *p4d);
1046 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1050 static inline int p4d_clear_huge(p4d_t *p4d)
1054 #endif /* !__PAGETABLE_P4D_FOLDED */
1056 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1057 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1058 int pud_clear_huge(pud_t *pud);
1059 int pmd_clear_huge(pmd_t *pmd);
1060 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1061 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1062 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1063 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1064 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1068 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1072 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1076 static inline int p4d_clear_huge(p4d_t *p4d)
1080 static inline int pud_clear_huge(pud_t *pud)
1084 static inline int pmd_clear_huge(pmd_t *pmd)
1088 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1092 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1096 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1100 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
1102 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1103 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1105 * ARCHes with special requirements for evicting THP backing TLB entries can
1106 * implement this. Otherwise also, it can help optimize normal TLB flush in
1107 * THP regime. stock flush_tlb_range() typically has optimization to nuke the
1108 * entire TLB TLB if flush span is greater than a threshold, which will
1109 * likely be true for a single huge page. Thus a single thp flush will
1110 * invalidate the entire TLB which is not desitable.
1111 * e.g. see arch/arc: flush_pmd_tlb_range
1113 #define flush_pmd_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1114 #define flush_pud_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1116 #define flush_pmd_tlb_range(vma, addr, end) BUILD_BUG()
1117 #define flush_pud_tlb_range(vma, addr, end) BUILD_BUG()
1122 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1123 unsigned long size, pgprot_t *vma_prot);
1125 #ifndef CONFIG_X86_ESPFIX64
1126 static inline void init_espfix_bsp(void) { }
1129 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1130 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1135 static inline bool arch_has_pfn_modify_check(void)
1139 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1142 * Architecture PAGE_KERNEL_* fallbacks
1144 * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1145 * because they really don't support them, or the port needs to be updated to
1146 * reflect the required functionality. Below are a set of relatively safe
1147 * fallbacks, as best effort, which we can count on in lieu of the architectures
1148 * not defining them on their own yet.
1151 #ifndef PAGE_KERNEL_RO
1152 # define PAGE_KERNEL_RO PAGE_KERNEL
1155 #ifndef PAGE_KERNEL_EXEC
1156 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1159 #endif /* !__ASSEMBLY__ */
1161 #ifndef io_remap_pfn_range
1162 #define io_remap_pfn_range remap_pfn_range
1165 #ifndef has_transparent_hugepage
1166 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1167 #define has_transparent_hugepage() 1
1169 #define has_transparent_hugepage() 0
1174 * On some architectures it depends on the mm if the p4d/pud or pmd
1175 * layer of the page table hierarchy is folded or not.
1177 #ifndef mm_p4d_folded
1178 #define mm_p4d_folded(mm) __is_defined(__PAGETABLE_P4D_FOLDED)
1181 #ifndef mm_pud_folded
1182 #define mm_pud_folded(mm) __is_defined(__PAGETABLE_PUD_FOLDED)
1185 #ifndef mm_pmd_folded
1186 #define mm_pmd_folded(mm) __is_defined(__PAGETABLE_PMD_FOLDED)
1189 #endif /* _ASM_GENERIC_PGTABLE_H */