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 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
253 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
254 static inline pmd_t
pmdp_huge_get_and_clear(struct mm_struct
*mm
,
255 unsigned long address
,
262 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
263 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
264 static inline pud_t
pudp_huge_get_and_clear(struct mm_struct
*mm
,
265 unsigned long address
,
273 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
274 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
276 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
277 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
278 static inline pmd_t
pmdp_huge_get_and_clear_full(struct vm_area_struct
*vma
,
279 unsigned long address
, pmd_t
*pmdp
,
282 return pmdp_huge_get_and_clear(vma
->vm_mm
, address
, pmdp
);
286 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
287 static inline pud_t
pudp_huge_get_and_clear_full(struct mm_struct
*mm
,
288 unsigned long address
, pud_t
*pudp
,
291 return pudp_huge_get_and_clear(mm
, address
, pudp
);
294 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
296 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
297 static inline pte_t
ptep_get_and_clear_full(struct mm_struct
*mm
,
298 unsigned long address
, pte_t
*ptep
,
302 pte
= ptep_get_and_clear(mm
, address
, ptep
);
309 * If two threads concurrently fault at the same page, the thread that
310 * won the race updates the PTE and its local TLB/Cache. The other thread
311 * gives up, simply does nothing, and continues; on architectures where
312 * software can update TLB, local TLB can be updated here to avoid next page
313 * fault. This function updates TLB only, do nothing with cache or others.
314 * It is the difference with function update_mmu_cache.
316 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
317 static inline void update_mmu_tlb(struct vm_area_struct
*vma
,
318 unsigned long address
, pte_t
*ptep
)
321 #define __HAVE_ARCH_UPDATE_MMU_TLB
325 * Some architectures may be able to avoid expensive synchronization
326 * primitives when modifications are made to PTE's which are already
327 * not present, or in the process of an address space destruction.
329 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
330 static inline void pte_clear_not_present_full(struct mm_struct
*mm
,
331 unsigned long address
,
335 pte_clear(mm
, address
, ptep
);
339 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
340 extern pte_t
ptep_clear_flush(struct vm_area_struct
*vma
,
341 unsigned long address
,
345 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
346 extern pmd_t
pmdp_huge_clear_flush(struct vm_area_struct
*vma
,
347 unsigned long address
,
349 extern pud_t
pudp_huge_clear_flush(struct vm_area_struct
*vma
,
350 unsigned long address
,
354 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
356 static inline void ptep_set_wrprotect(struct mm_struct
*mm
, unsigned long address
, pte_t
*ptep
)
358 pte_t old_pte
= *ptep
;
359 set_pte_at(mm
, address
, ptep
, pte_wrprotect(old_pte
));
364 * On some architectures hardware does not set page access bit when accessing
365 * memory page, it is responsibilty of software setting this bit. It brings
366 * out extra page fault penalty to track page access bit. For optimization page
367 * access bit can be set during all page fault flow on these arches.
368 * To be differentiate with macro pte_mkyoung, this macro is used on platforms
369 * where software maintains page access bit.
371 #ifndef pte_sw_mkyoung
372 static inline pte_t
pte_sw_mkyoung(pte_t pte
)
376 #define pte_sw_mkyoung pte_sw_mkyoung
379 #ifndef pte_savedwrite
380 #define pte_savedwrite pte_write
383 #ifndef pte_mk_savedwrite
384 #define pte_mk_savedwrite pte_mkwrite
387 #ifndef pte_clear_savedwrite
388 #define pte_clear_savedwrite pte_wrprotect
391 #ifndef pmd_savedwrite
392 #define pmd_savedwrite pmd_write
395 #ifndef pmd_mk_savedwrite
396 #define pmd_mk_savedwrite pmd_mkwrite
399 #ifndef pmd_clear_savedwrite
400 #define pmd_clear_savedwrite pmd_wrprotect
403 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
404 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
405 static inline void pmdp_set_wrprotect(struct mm_struct
*mm
,
406 unsigned long address
, pmd_t
*pmdp
)
408 pmd_t old_pmd
= *pmdp
;
409 set_pmd_at(mm
, address
, pmdp
, pmd_wrprotect(old_pmd
));
412 static inline void pmdp_set_wrprotect(struct mm_struct
*mm
,
413 unsigned long address
, pmd_t
*pmdp
)
417 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
419 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
420 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
421 static inline void pudp_set_wrprotect(struct mm_struct
*mm
,
422 unsigned long address
, pud_t
*pudp
)
424 pud_t old_pud
= *pudp
;
426 set_pud_at(mm
, address
, pudp
, pud_wrprotect(old_pud
));
429 static inline void pudp_set_wrprotect(struct mm_struct
*mm
,
430 unsigned long address
, pud_t
*pudp
)
434 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
437 #ifndef pmdp_collapse_flush
438 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
439 extern pmd_t
pmdp_collapse_flush(struct vm_area_struct
*vma
,
440 unsigned long address
, pmd_t
*pmdp
);
442 static inline pmd_t
pmdp_collapse_flush(struct vm_area_struct
*vma
,
443 unsigned long address
,
449 #define pmdp_collapse_flush pmdp_collapse_flush
450 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
453 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
454 extern void pgtable_trans_huge_deposit(struct mm_struct
*mm
, pmd_t
*pmdp
,
458 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
459 extern pgtable_t
pgtable_trans_huge_withdraw(struct mm_struct
*mm
, pmd_t
*pmdp
);
462 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
464 * This is an implementation of pmdp_establish() that is only suitable for an
465 * architecture that doesn't have hardware dirty/accessed bits. In this case we
466 * can't race with CPU which sets these bits and non-atomic aproach is fine.
468 static inline pmd_t
generic_pmdp_establish(struct vm_area_struct
*vma
,
469 unsigned long address
, pmd_t
*pmdp
, pmd_t pmd
)
471 pmd_t old_pmd
= *pmdp
;
472 set_pmd_at(vma
->vm_mm
, address
, pmdp
, pmd
);
477 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
478 extern pmd_t
pmdp_invalidate(struct vm_area_struct
*vma
, unsigned long address
,
482 #ifndef __HAVE_ARCH_PTE_SAME
483 static inline int pte_same(pte_t pte_a
, pte_t pte_b
)
485 return pte_val(pte_a
) == pte_val(pte_b
);
489 #ifndef __HAVE_ARCH_PTE_UNUSED
491 * Some architectures provide facilities to virtualization guests
492 * so that they can flag allocated pages as unused. This allows the
493 * host to transparently reclaim unused pages. This function returns
494 * whether the pte's page is unused.
496 static inline int pte_unused(pte_t pte
)
502 #ifndef pte_access_permitted
503 #define pte_access_permitted(pte, write) \
504 (pte_present(pte) && (!(write) || pte_write(pte)))
507 #ifndef pmd_access_permitted
508 #define pmd_access_permitted(pmd, write) \
509 (pmd_present(pmd) && (!(write) || pmd_write(pmd)))
512 #ifndef pud_access_permitted
513 #define pud_access_permitted(pud, write) \
514 (pud_present(pud) && (!(write) || pud_write(pud)))
517 #ifndef p4d_access_permitted
518 #define p4d_access_permitted(p4d, write) \
519 (p4d_present(p4d) && (!(write) || p4d_write(p4d)))
522 #ifndef pgd_access_permitted
523 #define pgd_access_permitted(pgd, write) \
524 (pgd_present(pgd) && (!(write) || pgd_write(pgd)))
527 #ifndef __HAVE_ARCH_PMD_SAME
528 static inline int pmd_same(pmd_t pmd_a
, pmd_t pmd_b
)
530 return pmd_val(pmd_a
) == pmd_val(pmd_b
);
533 static inline int pud_same(pud_t pud_a
, pud_t pud_b
)
535 return pud_val(pud_a
) == pud_val(pud_b
);
539 #ifndef __HAVE_ARCH_P4D_SAME
540 static inline int p4d_same(p4d_t p4d_a
, p4d_t p4d_b
)
542 return p4d_val(p4d_a
) == p4d_val(p4d_b
);
546 #ifndef __HAVE_ARCH_PGD_SAME
547 static inline int pgd_same(pgd_t pgd_a
, pgd_t pgd_b
)
549 return pgd_val(pgd_a
) == pgd_val(pgd_b
);
554 * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
555 * TLB flush will be required as a result of the "set". For example, use
556 * in scenarios where it is known ahead of time that the routine is
557 * setting non-present entries, or re-setting an existing entry to the
558 * same value. Otherwise, use the typical "set" helpers and flush the
561 #define set_pte_safe(ptep, pte) \
563 WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
564 set_pte(ptep, pte); \
567 #define set_pmd_safe(pmdp, pmd) \
569 WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
570 set_pmd(pmdp, pmd); \
573 #define set_pud_safe(pudp, pud) \
575 WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
576 set_pud(pudp, pud); \
579 #define set_p4d_safe(p4dp, p4d) \
581 WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
582 set_p4d(p4dp, p4d); \
585 #define set_pgd_safe(pgdp, pgd) \
587 WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
588 set_pgd(pgdp, pgd); \
591 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
593 * Some architectures support metadata associated with a page. When a
594 * page is being swapped out, this metadata must be saved so it can be
595 * restored when the page is swapped back in. SPARC M7 and newer
596 * processors support an ADI (Application Data Integrity) tag for the
597 * page as metadata for the page. arch_do_swap_page() can restore this
598 * metadata when a page is swapped back in.
600 static inline void arch_do_swap_page(struct mm_struct
*mm
,
601 struct vm_area_struct
*vma
,
603 pte_t pte
, pte_t oldpte
)
609 #ifndef __HAVE_ARCH_UNMAP_ONE
611 * Some architectures support metadata associated with a page. When a
612 * page is being swapped out, this metadata must be saved so it can be
613 * restored when the page is swapped back in. SPARC M7 and newer
614 * processors support an ADI (Application Data Integrity) tag for the
615 * page as metadata for the page. arch_unmap_one() can save this
616 * metadata on a swap-out of a page.
618 static inline int arch_unmap_one(struct mm_struct
*mm
,
619 struct vm_area_struct
*vma
,
627 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
628 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
631 #ifndef __HAVE_ARCH_MOVE_PTE
632 #define move_pte(pte, prot, old_addr, new_addr) (pte)
635 #ifndef pte_accessible
636 # define pte_accessible(mm, pte) ((void)(pte), 1)
639 #ifndef flush_tlb_fix_spurious_fault
640 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
644 #define pgprot_nx(prot) (prot)
647 #ifndef pgprot_noncached
648 #define pgprot_noncached(prot) (prot)
651 #ifndef pgprot_writecombine
652 #define pgprot_writecombine pgprot_noncached
655 #ifndef pgprot_writethrough
656 #define pgprot_writethrough pgprot_noncached
659 #ifndef pgprot_device
660 #define pgprot_device pgprot_noncached
663 #ifndef pgprot_modify
664 #define pgprot_modify pgprot_modify
665 static inline pgprot_t
pgprot_modify(pgprot_t oldprot
, pgprot_t newprot
)
667 if (pgprot_val(oldprot
) == pgprot_val(pgprot_noncached(oldprot
)))
668 newprot
= pgprot_noncached(newprot
);
669 if (pgprot_val(oldprot
) == pgprot_val(pgprot_writecombine(oldprot
)))
670 newprot
= pgprot_writecombine(newprot
);
671 if (pgprot_val(oldprot
) == pgprot_val(pgprot_device(oldprot
)))
672 newprot
= pgprot_device(newprot
);
678 * When walking page tables, get the address of the next boundary,
679 * or the end address of the range if that comes earlier. Although no
680 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
683 #define pgd_addr_end(addr, end) \
684 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
685 (__boundary - 1 < (end) - 1)? __boundary: (end); \
689 #define p4d_addr_end(addr, end) \
690 ({ unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK; \
691 (__boundary - 1 < (end) - 1)? __boundary: (end); \
696 #define pud_addr_end(addr, end) \
697 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
698 (__boundary - 1 < (end) - 1)? __boundary: (end); \
703 #define pmd_addr_end(addr, end) \
704 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
705 (__boundary - 1 < (end) - 1)? __boundary: (end); \
710 * When walking page tables, we usually want to skip any p?d_none entries;
711 * and any p?d_bad entries - reporting the error before resetting to none.
712 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
714 void pgd_clear_bad(pgd_t
*);
716 #ifndef __PAGETABLE_P4D_FOLDED
717 void p4d_clear_bad(p4d_t
*);
719 #define p4d_clear_bad(p4d) do { } while (0)
722 #ifndef __PAGETABLE_PUD_FOLDED
723 void pud_clear_bad(pud_t
*);
725 #define pud_clear_bad(p4d) do { } while (0)
728 void pmd_clear_bad(pmd_t
*);
730 static inline int pgd_none_or_clear_bad(pgd_t
*pgd
)
734 if (unlikely(pgd_bad(*pgd
))) {
741 static inline int p4d_none_or_clear_bad(p4d_t
*p4d
)
745 if (unlikely(p4d_bad(*p4d
))) {
752 static inline int pud_none_or_clear_bad(pud_t
*pud
)
756 if (unlikely(pud_bad(*pud
))) {
763 static inline int pmd_none_or_clear_bad(pmd_t
*pmd
)
767 if (unlikely(pmd_bad(*pmd
))) {
774 static inline pte_t
__ptep_modify_prot_start(struct vm_area_struct
*vma
,
779 * Get the current pte state, but zero it out to make it
780 * non-present, preventing the hardware from asynchronously
783 return ptep_get_and_clear(vma
->vm_mm
, addr
, ptep
);
786 static inline void __ptep_modify_prot_commit(struct vm_area_struct
*vma
,
788 pte_t
*ptep
, pte_t pte
)
791 * The pte is non-present, so there's no hardware state to
794 set_pte_at(vma
->vm_mm
, addr
, ptep
, pte
);
797 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
799 * Start a pte protection read-modify-write transaction, which
800 * protects against asynchronous hardware modifications to the pte.
801 * The intention is not to prevent the hardware from making pte
802 * updates, but to prevent any updates it may make from being lost.
804 * This does not protect against other software modifications of the
805 * pte; the appropriate pte lock must be held over the transation.
807 * Note that this interface is intended to be batchable, meaning that
808 * ptep_modify_prot_commit may not actually update the pte, but merely
809 * queue the update to be done at some later time. The update must be
810 * actually committed before the pte lock is released, however.
812 static inline pte_t
ptep_modify_prot_start(struct vm_area_struct
*vma
,
816 return __ptep_modify_prot_start(vma
, addr
, ptep
);
820 * Commit an update to a pte, leaving any hardware-controlled bits in
821 * the PTE unmodified.
823 static inline void ptep_modify_prot_commit(struct vm_area_struct
*vma
,
825 pte_t
*ptep
, pte_t old_pte
, pte_t pte
)
827 __ptep_modify_prot_commit(vma
, addr
, ptep
, pte
);
829 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
830 #endif /* CONFIG_MMU */
833 * No-op macros that just return the current protection value. Defined here
834 * because these macros can be used used even if CONFIG_MMU is not defined.
836 #ifndef pgprot_encrypted
837 #define pgprot_encrypted(prot) (prot)
840 #ifndef pgprot_decrypted
841 #define pgprot_decrypted(prot) (prot)
845 * A facility to provide lazy MMU batching. This allows PTE updates and
846 * page invalidations to be delayed until a call to leave lazy MMU mode
847 * is issued. Some architectures may benefit from doing this, and it is
848 * beneficial for both shadow and direct mode hypervisors, which may batch
849 * the PTE updates which happen during this window. Note that using this
850 * interface requires that read hazards be removed from the code. A read
851 * hazard could result in the direct mode hypervisor case, since the actual
852 * write to the page tables may not yet have taken place, so reads though
853 * a raw PTE pointer after it has been modified are not guaranteed to be
854 * up to date. This mode can only be entered and left under the protection of
855 * the page table locks for all page tables which may be modified. In the UP
856 * case, this is required so that preemption is disabled, and in the SMP case,
857 * it must synchronize the delayed page table writes properly on other CPUs.
859 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
860 #define arch_enter_lazy_mmu_mode() do {} while (0)
861 #define arch_leave_lazy_mmu_mode() do {} while (0)
862 #define arch_flush_lazy_mmu_mode() do {} while (0)
866 * A facility to provide batching of the reload of page tables and
867 * other process state with the actual context switch code for
868 * paravirtualized guests. By convention, only one of the batched
869 * update (lazy) modes (CPU, MMU) should be active at any given time,
870 * entry should never be nested, and entry and exits should always be
871 * paired. This is for sanity of maintaining and reasoning about the
872 * kernel code. In this case, the exit (end of the context switch) is
873 * in architecture-specific code, and so doesn't need a generic
876 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
877 #define arch_start_context_switch(prev) do {} while (0)
880 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
881 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
882 static inline pmd_t
pmd_swp_mksoft_dirty(pmd_t pmd
)
887 static inline int pmd_swp_soft_dirty(pmd_t pmd
)
892 static inline pmd_t
pmd_swp_clear_soft_dirty(pmd_t pmd
)
897 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
898 static inline int pte_soft_dirty(pte_t pte
)
903 static inline int pmd_soft_dirty(pmd_t pmd
)
908 static inline pte_t
pte_mksoft_dirty(pte_t pte
)
913 static inline pmd_t
pmd_mksoft_dirty(pmd_t pmd
)
918 static inline pte_t
pte_clear_soft_dirty(pte_t pte
)
923 static inline pmd_t
pmd_clear_soft_dirty(pmd_t pmd
)
928 static inline pte_t
pte_swp_mksoft_dirty(pte_t pte
)
933 static inline int pte_swp_soft_dirty(pte_t pte
)
938 static inline pte_t
pte_swp_clear_soft_dirty(pte_t pte
)
943 static inline pmd_t
pmd_swp_mksoft_dirty(pmd_t pmd
)
948 static inline int pmd_swp_soft_dirty(pmd_t pmd
)
953 static inline pmd_t
pmd_swp_clear_soft_dirty(pmd_t pmd
)
959 #ifndef __HAVE_PFNMAP_TRACKING
961 * Interfaces that can be used by architecture code to keep track of
962 * memory type of pfn mappings specified by the remap_pfn_range,
967 * track_pfn_remap is called when a _new_ pfn mapping is being established
968 * by remap_pfn_range() for physical range indicated by pfn and size.
970 static inline int track_pfn_remap(struct vm_area_struct
*vma
, pgprot_t
*prot
,
971 unsigned long pfn
, unsigned long addr
,
978 * track_pfn_insert is called when a _new_ single pfn is established
979 * by vmf_insert_pfn().
981 static inline void track_pfn_insert(struct vm_area_struct
*vma
, pgprot_t
*prot
,
987 * track_pfn_copy is called when vma that is covering the pfnmap gets
988 * copied through copy_page_range().
990 static inline int track_pfn_copy(struct vm_area_struct
*vma
)
996 * untrack_pfn is called while unmapping a pfnmap for a region.
997 * untrack can be called for a specific region indicated by pfn and size or
998 * can be for the entire vma (in which case pfn, size are zero).
1000 static inline void untrack_pfn(struct vm_area_struct
*vma
,
1001 unsigned long pfn
, unsigned long size
)
1006 * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
1008 static inline void untrack_pfn_moved(struct vm_area_struct
*vma
)
1012 extern int track_pfn_remap(struct vm_area_struct
*vma
, pgprot_t
*prot
,
1013 unsigned long pfn
, unsigned long addr
,
1014 unsigned long size
);
1015 extern void track_pfn_insert(struct vm_area_struct
*vma
, pgprot_t
*prot
,
1017 extern int track_pfn_copy(struct vm_area_struct
*vma
);
1018 extern void untrack_pfn(struct vm_area_struct
*vma
, unsigned long pfn
,
1019 unsigned long size
);
1020 extern void untrack_pfn_moved(struct vm_area_struct
*vma
);
1023 #ifdef __HAVE_COLOR_ZERO_PAGE
1024 static inline int is_zero_pfn(unsigned long pfn
)
1026 extern unsigned long zero_pfn
;
1027 unsigned long offset_from_zero_pfn
= pfn
- zero_pfn
;
1028 return offset_from_zero_pfn
<= (zero_page_mask
>> PAGE_SHIFT
);
1031 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
1034 static inline int is_zero_pfn(unsigned long pfn
)
1036 extern unsigned long zero_pfn
;
1037 return pfn
== zero_pfn
;
1040 static inline unsigned long my_zero_pfn(unsigned long addr
)
1042 extern unsigned long zero_pfn
;
1049 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1050 static inline int pmd_trans_huge(pmd_t pmd
)
1055 static inline int pmd_write(pmd_t pmd
)
1060 #endif /* pmd_write */
1061 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1064 static inline int pud_write(pud_t pud
)
1069 #endif /* pud_write */
1071 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1072 static inline int pmd_devmap(pmd_t pmd
)
1076 static inline int pud_devmap(pud_t pud
)
1080 static inline int pgd_devmap(pgd_t pgd
)
1086 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1087 (defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1088 !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
1089 static inline int pud_trans_huge(pud_t pud
)
1095 /* See pmd_none_or_trans_huge_or_clear_bad for discussion. */
1096 static inline int pud_none_or_trans_huge_or_dev_or_clear_bad(pud_t
*pud
)
1098 pud_t pudval
= READ_ONCE(*pud
);
1100 if (pud_none(pudval
) || pud_trans_huge(pudval
) || pud_devmap(pudval
))
1102 if (unlikely(pud_bad(pudval
))) {
1109 /* See pmd_trans_unstable for discussion. */
1110 static inline int pud_trans_unstable(pud_t
*pud
)
1112 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1113 defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1114 return pud_none_or_trans_huge_or_dev_or_clear_bad(pud
);
1120 #ifndef pmd_read_atomic
1121 static inline pmd_t
pmd_read_atomic(pmd_t
*pmdp
)
1124 * Depend on compiler for an atomic pmd read. NOTE: this is
1125 * only going to work, if the pmdval_t isn't larger than
1132 #ifndef arch_needs_pgtable_deposit
1133 #define arch_needs_pgtable_deposit() (false)
1136 * This function is meant to be used by sites walking pagetables with
1137 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
1138 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
1139 * into a null pmd and the transhuge page fault can convert a null pmd
1140 * into an hugepmd or into a regular pmd (if the hugepage allocation
1141 * fails). While holding the mmap_sem in read mode the pmd becomes
1142 * stable and stops changing under us only if it's not null and not a
1143 * transhuge pmd. When those races occurs and this function makes a
1144 * difference vs the standard pmd_none_or_clear_bad, the result is
1145 * undefined so behaving like if the pmd was none is safe (because it
1146 * can return none anyway). The compiler level barrier() is critically
1147 * important to compute the two checks atomically on the same pmdval.
1149 * For 32bit kernels with a 64bit large pmd_t this automatically takes
1150 * care of reading the pmd atomically to avoid SMP race conditions
1151 * against pmd_populate() when the mmap_sem is hold for reading by the
1152 * caller (a special atomic read not done by "gcc" as in the generic
1153 * version above, is also needed when THP is disabled because the page
1154 * fault can populate the pmd from under us).
1156 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t
*pmd
)
1158 pmd_t pmdval
= pmd_read_atomic(pmd
);
1160 * The barrier will stabilize the pmdval in a register or on
1161 * the stack so that it will stop changing under the code.
1163 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
1164 * pmd_read_atomic is allowed to return a not atomic pmdval
1165 * (for example pointing to an hugepage that has never been
1166 * mapped in the pmd). The below checks will only care about
1167 * the low part of the pmd with 32bit PAE x86 anyway, with the
1168 * exception of pmd_none(). So the important thing is that if
1169 * the low part of the pmd is found null, the high part will
1170 * be also null or the pmd_none() check below would be
1173 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1177 * !pmd_present() checks for pmd migration entries
1179 * The complete check uses is_pmd_migration_entry() in linux/swapops.h
1180 * But using that requires moving current function and pmd_trans_unstable()
1181 * to linux/swapops.h to resovle dependency, which is too much code move.
1183 * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
1184 * because !pmd_present() pages can only be under migration not swapped
1187 * pmd_none() is preseved for future condition checks on pmd migration
1188 * entries and not confusing with this function name, although it is
1189 * redundant with !pmd_present().
1191 if (pmd_none(pmdval
) || pmd_trans_huge(pmdval
) ||
1192 (IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION
) && !pmd_present(pmdval
)))
1194 if (unlikely(pmd_bad(pmdval
))) {
1202 * This is a noop if Transparent Hugepage Support is not built into
1203 * the kernel. Otherwise it is equivalent to
1204 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
1205 * places that already verified the pmd is not none and they want to
1206 * walk ptes while holding the mmap sem in read mode (write mode don't
1207 * need this). If THP is not enabled, the pmd can't go away under the
1208 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
1209 * run a pmd_trans_unstable before walking the ptes after
1210 * split_huge_pmd returns (because it may have run when the pmd become
1211 * null, but then a page fault can map in a THP and not a regular page).
1213 static inline int pmd_trans_unstable(pmd_t
*pmd
)
1215 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1216 return pmd_none_or_trans_huge_or_clear_bad(pmd
);
1222 #ifndef CONFIG_NUMA_BALANCING
1224 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
1225 * the only case the kernel cares is for NUMA balancing and is only ever set
1226 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
1227 * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
1228 * is the responsibility of the caller to distinguish between PROT_NONE
1229 * protections and NUMA hinting fault protections.
1231 static inline int pte_protnone(pte_t pte
)
1236 static inline int pmd_protnone(pmd_t pmd
)
1240 #endif /* CONFIG_NUMA_BALANCING */
1242 #endif /* CONFIG_MMU */
1244 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1246 #ifndef __PAGETABLE_P4D_FOLDED
1247 int p4d_set_huge(p4d_t
*p4d
, phys_addr_t addr
, pgprot_t prot
);
1248 int p4d_clear_huge(p4d_t
*p4d
);
1250 static inline int p4d_set_huge(p4d_t
*p4d
, phys_addr_t addr
, pgprot_t prot
)
1254 static inline int p4d_clear_huge(p4d_t
*p4d
)
1258 #endif /* !__PAGETABLE_P4D_FOLDED */
1260 int pud_set_huge(pud_t
*pud
, phys_addr_t addr
, pgprot_t prot
);
1261 int pmd_set_huge(pmd_t
*pmd
, phys_addr_t addr
, pgprot_t prot
);
1262 int pud_clear_huge(pud_t
*pud
);
1263 int pmd_clear_huge(pmd_t
*pmd
);
1264 int p4d_free_pud_page(p4d_t
*p4d
, unsigned long addr
);
1265 int pud_free_pmd_page(pud_t
*pud
, unsigned long addr
);
1266 int pmd_free_pte_page(pmd_t
*pmd
, unsigned long addr
);
1267 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1268 static inline int p4d_set_huge(p4d_t
*p4d
, phys_addr_t addr
, pgprot_t prot
)
1272 static inline int pud_set_huge(pud_t
*pud
, phys_addr_t addr
, pgprot_t prot
)
1276 static inline int pmd_set_huge(pmd_t
*pmd
, phys_addr_t addr
, pgprot_t prot
)
1280 static inline int p4d_clear_huge(p4d_t
*p4d
)
1284 static inline int pud_clear_huge(pud_t
*pud
)
1288 static inline int pmd_clear_huge(pmd_t
*pmd
)
1292 static inline int p4d_free_pud_page(p4d_t
*p4d
, unsigned long addr
)
1296 static inline int pud_free_pmd_page(pud_t
*pud
, unsigned long addr
)
1300 static inline int pmd_free_pte_page(pmd_t
*pmd
, unsigned long addr
)
1304 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
1306 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1307 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1309 * ARCHes with special requirements for evicting THP backing TLB entries can
1310 * implement this. Otherwise also, it can help optimize normal TLB flush in
1311 * THP regime. stock flush_tlb_range() typically has optimization to nuke the
1312 * entire TLB TLB if flush span is greater than a threshold, which will
1313 * likely be true for a single huge page. Thus a single thp flush will
1314 * invalidate the entire TLB which is not desitable.
1315 * e.g. see arch/arc: flush_pmd_tlb_range
1317 #define flush_pmd_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1318 #define flush_pud_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1320 #define flush_pmd_tlb_range(vma, addr, end) BUILD_BUG()
1321 #define flush_pud_tlb_range(vma, addr, end) BUILD_BUG()
1326 int phys_mem_access_prot_allowed(struct file
*file
, unsigned long pfn
,
1327 unsigned long size
, pgprot_t
*vma_prot
);
1329 #ifndef CONFIG_X86_ESPFIX64
1330 static inline void init_espfix_bsp(void) { }
1333 extern void __init
pgtable_cache_init(void);
1335 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1336 static inline bool pfn_modify_allowed(unsigned long pfn
, pgprot_t prot
)
1341 static inline bool arch_has_pfn_modify_check(void)
1345 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1348 * Architecture PAGE_KERNEL_* fallbacks
1350 * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1351 * because they really don't support them, or the port needs to be updated to
1352 * reflect the required functionality. Below are a set of relatively safe
1353 * fallbacks, as best effort, which we can count on in lieu of the architectures
1354 * not defining them on their own yet.
1357 #ifndef PAGE_KERNEL_RO
1358 # define PAGE_KERNEL_RO PAGE_KERNEL
1361 #ifndef PAGE_KERNEL_EXEC
1362 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1366 * Page Table Modification bits for pgtbl_mod_mask.
1368 * These are used by the p?d_alloc_track*() set of functions an in the generic
1369 * vmalloc/ioremap code to track at which page-table levels entries have been
1370 * modified. Based on that the code can better decide when vmalloc and ioremap
1371 * mapping changes need to be synchronized to other page-tables in the system.
1373 #define __PGTBL_PGD_MODIFIED 0
1374 #define __PGTBL_P4D_MODIFIED 1
1375 #define __PGTBL_PUD_MODIFIED 2
1376 #define __PGTBL_PMD_MODIFIED 3
1377 #define __PGTBL_PTE_MODIFIED 4
1379 #define PGTBL_PGD_MODIFIED BIT(__PGTBL_PGD_MODIFIED)
1380 #define PGTBL_P4D_MODIFIED BIT(__PGTBL_P4D_MODIFIED)
1381 #define PGTBL_PUD_MODIFIED BIT(__PGTBL_PUD_MODIFIED)
1382 #define PGTBL_PMD_MODIFIED BIT(__PGTBL_PMD_MODIFIED)
1383 #define PGTBL_PTE_MODIFIED BIT(__PGTBL_PTE_MODIFIED)
1385 /* Page-Table Modification Mask */
1386 typedef unsigned int pgtbl_mod_mask
;
1388 #endif /* !__ASSEMBLY__ */
1390 #ifndef io_remap_pfn_range
1391 #define io_remap_pfn_range remap_pfn_range
1394 #ifndef has_transparent_hugepage
1395 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1396 #define has_transparent_hugepage() 1
1398 #define has_transparent_hugepage() 0
1403 * On some architectures it depends on the mm if the p4d/pud or pmd
1404 * layer of the page table hierarchy is folded or not.
1406 #ifndef mm_p4d_folded
1407 #define mm_p4d_folded(mm) __is_defined(__PAGETABLE_P4D_FOLDED)
1410 #ifndef mm_pud_folded
1411 #define mm_pud_folded(mm) __is_defined(__PAGETABLE_PUD_FOLDED)
1414 #ifndef mm_pmd_folded
1415 #define mm_pmd_folded(mm) __is_defined(__PAGETABLE_PMD_FOLDED)
1419 * p?d_leaf() - true if this entry is a final mapping to a physical address.
1420 * This differs from p?d_huge() by the fact that they are always available (if
1421 * the architecture supports large pages at the appropriate level) even
1422 * if CONFIG_HUGETLB_PAGE is not defined.
1423 * Only meaningful when called on a valid entry.
1426 #define pgd_leaf(x) 0
1429 #define p4d_leaf(x) 0
1432 #define pud_leaf(x) 0
1435 #define pmd_leaf(x) 0
1438 #endif /* _LINUX_PGTABLE_H */