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
121 #define pgd_offset_k(address) pgd_offset(&init_mm, (address))
125 * In many cases it is known that a virtual address is mapped at PMD or PTE
126 * level, so instead of traversing all the page table levels, we can get a
127 * pointer to the PMD entry in user or kernel page table or translate a virtual
128 * address to the pointer in the PTE in the kernel page tables with simple
131 static inline pmd_t *pmd_off(struct mm_struct *mm, unsigned long va)
133 return pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, va), va), va), va);
136 static inline pmd_t *pmd_off_k(unsigned long va)
138 return pmd_offset(pud_offset(p4d_offset(pgd_offset_k(va), va), va), va);
141 static inline pte_t *virt_to_kpte(unsigned long vaddr)
143 pmd_t *pmd = pmd_off_k(vaddr);
145 return pmd_none(*pmd) ? NULL : pte_offset_kernel(pmd, vaddr);
148 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
149 extern int ptep_set_access_flags(struct vm_area_struct *vma,
150 unsigned long address, pte_t *ptep,
151 pte_t entry, int dirty);
154 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
155 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
156 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
157 unsigned long address, pmd_t *pmdp,
158 pmd_t entry, int dirty);
159 extern int pudp_set_access_flags(struct vm_area_struct *vma,
160 unsigned long address, pud_t *pudp,
161 pud_t entry, int dirty);
163 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
164 unsigned long address, pmd_t *pmdp,
165 pmd_t entry, int dirty)
170 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
171 unsigned long address, pud_t *pudp,
172 pud_t entry, int dirty)
177 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
180 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
181 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
182 unsigned long address,
190 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
195 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
196 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
197 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
198 unsigned long address,
206 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
210 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
211 unsigned long address,
217 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
220 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
221 int ptep_clear_flush_young(struct vm_area_struct *vma,
222 unsigned long address, pte_t *ptep);
225 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
226 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
227 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
228 unsigned long address, pmd_t *pmdp);
231 * Despite relevant to THP only, this API is called from generic rmap code
232 * under PageTransHuge(), hence needs a dummy implementation for !THP
234 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
235 unsigned long address, pmd_t *pmdp)
240 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
243 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
244 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
245 unsigned long address,
249 pte_clear(mm, address, ptep);
254 #ifndef __HAVE_ARCH_PTEP_GET
255 static inline pte_t ptep_get(pte_t *ptep)
257 return READ_ONCE(*ptep);
261 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
262 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
263 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
264 unsigned long address,
271 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
272 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
273 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
274 unsigned long address,
282 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
283 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
285 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
286 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
287 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
288 unsigned long address, pmd_t *pmdp,
291 return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
295 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
296 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
297 unsigned long address, pud_t *pudp,
300 return pudp_huge_get_and_clear(mm, address, pudp);
303 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
305 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
306 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
307 unsigned long address, pte_t *ptep,
311 pte = ptep_get_and_clear(mm, address, ptep);
318 * If two threads concurrently fault at the same page, the thread that
319 * won the race updates the PTE and its local TLB/Cache. The other thread
320 * gives up, simply does nothing, and continues; on architectures where
321 * software can update TLB, local TLB can be updated here to avoid next page
322 * fault. This function updates TLB only, do nothing with cache or others.
323 * It is the difference with function update_mmu_cache.
325 #ifndef __HAVE_ARCH_UPDATE_MMU_TLB
326 static inline void update_mmu_tlb(struct vm_area_struct *vma,
327 unsigned long address, pte_t *ptep)
330 #define __HAVE_ARCH_UPDATE_MMU_TLB
334 * Some architectures may be able to avoid expensive synchronization
335 * primitives when modifications are made to PTE's which are already
336 * not present, or in the process of an address space destruction.
338 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
339 static inline void pte_clear_not_present_full(struct mm_struct *mm,
340 unsigned long address,
344 pte_clear(mm, address, ptep);
348 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
349 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
350 unsigned long address,
354 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
355 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
356 unsigned long address,
358 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
359 unsigned long address,
363 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
365 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
367 pte_t old_pte = *ptep;
368 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
373 * On some architectures hardware does not set page access bit when accessing
374 * memory page, it is responsibilty of software setting this bit. It brings
375 * out extra page fault penalty to track page access bit. For optimization page
376 * access bit can be set during all page fault flow on these arches.
377 * To be differentiate with macro pte_mkyoung, this macro is used on platforms
378 * where software maintains page access bit.
380 #ifndef pte_sw_mkyoung
381 static inline pte_t pte_sw_mkyoung(pte_t pte)
385 #define pte_sw_mkyoung pte_sw_mkyoung
388 #ifndef pte_savedwrite
389 #define pte_savedwrite pte_write
392 #ifndef pte_mk_savedwrite
393 #define pte_mk_savedwrite pte_mkwrite
396 #ifndef pte_clear_savedwrite
397 #define pte_clear_savedwrite pte_wrprotect
400 #ifndef pmd_savedwrite
401 #define pmd_savedwrite pmd_write
404 #ifndef pmd_mk_savedwrite
405 #define pmd_mk_savedwrite pmd_mkwrite
408 #ifndef pmd_clear_savedwrite
409 #define pmd_clear_savedwrite pmd_wrprotect
412 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
413 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
414 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
415 unsigned long address, pmd_t *pmdp)
417 pmd_t old_pmd = *pmdp;
418 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
421 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
422 unsigned long address, pmd_t *pmdp)
426 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
428 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
429 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
430 static inline void pudp_set_wrprotect(struct mm_struct *mm,
431 unsigned long address, pud_t *pudp)
433 pud_t old_pud = *pudp;
435 set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
438 static inline void pudp_set_wrprotect(struct mm_struct *mm,
439 unsigned long address, pud_t *pudp)
443 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
446 #ifndef pmdp_collapse_flush
447 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
448 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
449 unsigned long address, pmd_t *pmdp);
451 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
452 unsigned long address,
458 #define pmdp_collapse_flush pmdp_collapse_flush
459 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
462 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
463 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
467 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
468 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
471 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
473 * This is an implementation of pmdp_establish() that is only suitable for an
474 * architecture that doesn't have hardware dirty/accessed bits. In this case we
475 * can't race with CPU which sets these bits and non-atomic aproach is fine.
477 static inline pmd_t generic_pmdp_establish(struct vm_area_struct *vma,
478 unsigned long address, pmd_t *pmdp, pmd_t pmd)
480 pmd_t old_pmd = *pmdp;
481 set_pmd_at(vma->vm_mm, address, pmdp, pmd);
486 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
487 extern pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
491 #ifndef __HAVE_ARCH_PTE_SAME
492 static inline int pte_same(pte_t pte_a, pte_t pte_b)
494 return pte_val(pte_a) == pte_val(pte_b);
498 #ifndef __HAVE_ARCH_PTE_UNUSED
500 * Some architectures provide facilities to virtualization guests
501 * so that they can flag allocated pages as unused. This allows the
502 * host to transparently reclaim unused pages. This function returns
503 * whether the pte's page is unused.
505 static inline int pte_unused(pte_t pte)
511 #ifndef pte_access_permitted
512 #define pte_access_permitted(pte, write) \
513 (pte_present(pte) && (!(write) || pte_write(pte)))
516 #ifndef pmd_access_permitted
517 #define pmd_access_permitted(pmd, write) \
518 (pmd_present(pmd) && (!(write) || pmd_write(pmd)))
521 #ifndef pud_access_permitted
522 #define pud_access_permitted(pud, write) \
523 (pud_present(pud) && (!(write) || pud_write(pud)))
526 #ifndef p4d_access_permitted
527 #define p4d_access_permitted(p4d, write) \
528 (p4d_present(p4d) && (!(write) || p4d_write(p4d)))
531 #ifndef pgd_access_permitted
532 #define pgd_access_permitted(pgd, write) \
533 (pgd_present(pgd) && (!(write) || pgd_write(pgd)))
536 #ifndef __HAVE_ARCH_PMD_SAME
537 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
539 return pmd_val(pmd_a) == pmd_val(pmd_b);
542 static inline int pud_same(pud_t pud_a, pud_t pud_b)
544 return pud_val(pud_a) == pud_val(pud_b);
548 #ifndef __HAVE_ARCH_P4D_SAME
549 static inline int p4d_same(p4d_t p4d_a, p4d_t p4d_b)
551 return p4d_val(p4d_a) == p4d_val(p4d_b);
555 #ifndef __HAVE_ARCH_PGD_SAME
556 static inline int pgd_same(pgd_t pgd_a, pgd_t pgd_b)
558 return pgd_val(pgd_a) == pgd_val(pgd_b);
563 * Use set_p*_safe(), and elide TLB flushing, when confident that *no*
564 * TLB flush will be required as a result of the "set". For example, use
565 * in scenarios where it is known ahead of time that the routine is
566 * setting non-present entries, or re-setting an existing entry to the
567 * same value. Otherwise, use the typical "set" helpers and flush the
570 #define set_pte_safe(ptep, pte) \
572 WARN_ON_ONCE(pte_present(*ptep) && !pte_same(*ptep, pte)); \
573 set_pte(ptep, pte); \
576 #define set_pmd_safe(pmdp, pmd) \
578 WARN_ON_ONCE(pmd_present(*pmdp) && !pmd_same(*pmdp, pmd)); \
579 set_pmd(pmdp, pmd); \
582 #define set_pud_safe(pudp, pud) \
584 WARN_ON_ONCE(pud_present(*pudp) && !pud_same(*pudp, pud)); \
585 set_pud(pudp, pud); \
588 #define set_p4d_safe(p4dp, p4d) \
590 WARN_ON_ONCE(p4d_present(*p4dp) && !p4d_same(*p4dp, p4d)); \
591 set_p4d(p4dp, p4d); \
594 #define set_pgd_safe(pgdp, pgd) \
596 WARN_ON_ONCE(pgd_present(*pgdp) && !pgd_same(*pgdp, pgd)); \
597 set_pgd(pgdp, pgd); \
600 #ifndef __HAVE_ARCH_DO_SWAP_PAGE
602 * Some architectures support metadata associated with a page. When a
603 * page is being swapped out, this metadata must be saved so it can be
604 * restored when the page is swapped back in. SPARC M7 and newer
605 * processors support an ADI (Application Data Integrity) tag for the
606 * page as metadata for the page. arch_do_swap_page() can restore this
607 * metadata when a page is swapped back in.
609 static inline void arch_do_swap_page(struct mm_struct *mm,
610 struct vm_area_struct *vma,
612 pte_t pte, pte_t oldpte)
618 #ifndef __HAVE_ARCH_UNMAP_ONE
620 * Some architectures support metadata associated with a page. When a
621 * page is being swapped out, this metadata must be saved so it can be
622 * restored when the page is swapped back in. SPARC M7 and newer
623 * processors support an ADI (Application Data Integrity) tag for the
624 * page as metadata for the page. arch_unmap_one() can save this
625 * metadata on a swap-out of a page.
627 static inline int arch_unmap_one(struct mm_struct *mm,
628 struct vm_area_struct *vma,
636 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
637 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
640 #ifndef __HAVE_ARCH_MOVE_PTE
641 #define move_pte(pte, prot, old_addr, new_addr) (pte)
644 #ifndef pte_accessible
645 # define pte_accessible(mm, pte) ((void)(pte), 1)
648 #ifndef flush_tlb_fix_spurious_fault
649 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
653 * When walking page tables, get the address of the next boundary,
654 * or the end address of the range if that comes earlier. Although no
655 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
658 #define pgd_addr_end(addr, end) \
659 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
660 (__boundary - 1 < (end) - 1)? __boundary: (end); \
664 #define p4d_addr_end(addr, end) \
665 ({ unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK; \
666 (__boundary - 1 < (end) - 1)? __boundary: (end); \
671 #define pud_addr_end(addr, end) \
672 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
673 (__boundary - 1 < (end) - 1)? __boundary: (end); \
678 #define pmd_addr_end(addr, end) \
679 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
680 (__boundary - 1 < (end) - 1)? __boundary: (end); \
685 * When walking page tables, we usually want to skip any p?d_none entries;
686 * and any p?d_bad entries - reporting the error before resetting to none.
687 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
689 void pgd_clear_bad(pgd_t *);
691 #ifndef __PAGETABLE_P4D_FOLDED
692 void p4d_clear_bad(p4d_t *);
694 #define p4d_clear_bad(p4d) do { } while (0)
697 #ifndef __PAGETABLE_PUD_FOLDED
698 void pud_clear_bad(pud_t *);
700 #define pud_clear_bad(p4d) do { } while (0)
703 void pmd_clear_bad(pmd_t *);
705 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
709 if (unlikely(pgd_bad(*pgd))) {
716 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
720 if (unlikely(p4d_bad(*p4d))) {
727 static inline int pud_none_or_clear_bad(pud_t *pud)
731 if (unlikely(pud_bad(*pud))) {
738 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
742 if (unlikely(pmd_bad(*pmd))) {
749 static inline pte_t __ptep_modify_prot_start(struct vm_area_struct *vma,
754 * Get the current pte state, but zero it out to make it
755 * non-present, preventing the hardware from asynchronously
758 return ptep_get_and_clear(vma->vm_mm, addr, ptep);
761 static inline void __ptep_modify_prot_commit(struct vm_area_struct *vma,
763 pte_t *ptep, pte_t pte)
766 * The pte is non-present, so there's no hardware state to
769 set_pte_at(vma->vm_mm, addr, ptep, pte);
772 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
774 * Start a pte protection read-modify-write transaction, which
775 * protects against asynchronous hardware modifications to the pte.
776 * The intention is not to prevent the hardware from making pte
777 * updates, but to prevent any updates it may make from being lost.
779 * This does not protect against other software modifications of the
780 * pte; the appropriate pte lock must be held over the transation.
782 * Note that this interface is intended to be batchable, meaning that
783 * ptep_modify_prot_commit may not actually update the pte, but merely
784 * queue the update to be done at some later time. The update must be
785 * actually committed before the pte lock is released, however.
787 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
791 return __ptep_modify_prot_start(vma, addr, ptep);
795 * Commit an update to a pte, leaving any hardware-controlled bits in
796 * the PTE unmodified.
798 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma,
800 pte_t *ptep, pte_t old_pte, pte_t pte)
802 __ptep_modify_prot_commit(vma, addr, ptep, pte);
804 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
805 #endif /* CONFIG_MMU */
808 * No-op macros that just return the current protection value. Defined here
809 * because these macros can be used even if CONFIG_MMU is not defined.
813 #define pgprot_nx(prot) (prot)
816 #ifndef pgprot_noncached
817 #define pgprot_noncached(prot) (prot)
820 #ifndef pgprot_writecombine
821 #define pgprot_writecombine pgprot_noncached
824 #ifndef pgprot_writethrough
825 #define pgprot_writethrough pgprot_noncached
828 #ifndef pgprot_device
829 #define pgprot_device pgprot_noncached
833 #ifndef pgprot_modify
834 #define pgprot_modify pgprot_modify
835 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
837 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
838 newprot = pgprot_noncached(newprot);
839 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
840 newprot = pgprot_writecombine(newprot);
841 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
842 newprot = pgprot_device(newprot);
846 #endif /* CONFIG_MMU */
848 #ifndef pgprot_encrypted
849 #define pgprot_encrypted(prot) (prot)
852 #ifndef pgprot_decrypted
853 #define pgprot_decrypted(prot) (prot)
857 * A facility to provide lazy MMU batching. This allows PTE updates and
858 * page invalidations to be delayed until a call to leave lazy MMU mode
859 * is issued. Some architectures may benefit from doing this, and it is
860 * beneficial for both shadow and direct mode hypervisors, which may batch
861 * the PTE updates which happen during this window. Note that using this
862 * interface requires that read hazards be removed from the code. A read
863 * hazard could result in the direct mode hypervisor case, since the actual
864 * write to the page tables may not yet have taken place, so reads though
865 * a raw PTE pointer after it has been modified are not guaranteed to be
866 * up to date. This mode can only be entered and left under the protection of
867 * the page table locks for all page tables which may be modified. In the UP
868 * case, this is required so that preemption is disabled, and in the SMP case,
869 * it must synchronize the delayed page table writes properly on other CPUs.
871 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
872 #define arch_enter_lazy_mmu_mode() do {} while (0)
873 #define arch_leave_lazy_mmu_mode() do {} while (0)
874 #define arch_flush_lazy_mmu_mode() do {} while (0)
878 * A facility to provide batching of the reload of page tables and
879 * other process state with the actual context switch code for
880 * paravirtualized guests. By convention, only one of the batched
881 * update (lazy) modes (CPU, MMU) should be active at any given time,
882 * entry should never be nested, and entry and exits should always be
883 * paired. This is for sanity of maintaining and reasoning about the
884 * kernel code. In this case, the exit (end of the context switch) is
885 * in architecture-specific code, and so doesn't need a generic
888 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
889 #define arch_start_context_switch(prev) do {} while (0)
892 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
893 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
894 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
899 static inline int pmd_swp_soft_dirty(pmd_t pmd)
904 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
909 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
910 static inline int pte_soft_dirty(pte_t pte)
915 static inline int pmd_soft_dirty(pmd_t pmd)
920 static inline pte_t pte_mksoft_dirty(pte_t pte)
925 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
930 static inline pte_t pte_clear_soft_dirty(pte_t pte)
935 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
940 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
945 static inline int pte_swp_soft_dirty(pte_t pte)
950 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
955 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
960 static inline int pmd_swp_soft_dirty(pmd_t pmd)
965 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
971 #ifndef __HAVE_PFNMAP_TRACKING
973 * Interfaces that can be used by architecture code to keep track of
974 * memory type of pfn mappings specified by the remap_pfn_range,
979 * track_pfn_remap is called when a _new_ pfn mapping is being established
980 * by remap_pfn_range() for physical range indicated by pfn and size.
982 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
983 unsigned long pfn, unsigned long addr,
990 * track_pfn_insert is called when a _new_ single pfn is established
991 * by vmf_insert_pfn().
993 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
999 * track_pfn_copy is called when vma that is covering the pfnmap gets
1000 * copied through copy_page_range().
1002 static inline int track_pfn_copy(struct vm_area_struct *vma)
1008 * untrack_pfn is called while unmapping a pfnmap for a region.
1009 * untrack can be called for a specific region indicated by pfn and size or
1010 * can be for the entire vma (in which case pfn, size are zero).
1012 static inline void untrack_pfn(struct vm_area_struct *vma,
1013 unsigned long pfn, unsigned long size)
1018 * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
1020 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
1024 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
1025 unsigned long pfn, unsigned long addr,
1026 unsigned long size);
1027 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
1029 extern int track_pfn_copy(struct vm_area_struct *vma);
1030 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
1031 unsigned long size);
1032 extern void untrack_pfn_moved(struct vm_area_struct *vma);
1035 #ifdef __HAVE_COLOR_ZERO_PAGE
1036 static inline int is_zero_pfn(unsigned long pfn)
1038 extern unsigned long zero_pfn;
1039 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
1040 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
1043 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
1046 static inline int is_zero_pfn(unsigned long pfn)
1048 extern unsigned long zero_pfn;
1049 return pfn == zero_pfn;
1052 static inline unsigned long my_zero_pfn(unsigned long addr)
1054 extern unsigned long zero_pfn;
1061 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
1062 static inline int pmd_trans_huge(pmd_t pmd)
1067 static inline int pmd_write(pmd_t pmd)
1072 #endif /* pmd_write */
1073 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1076 static inline int pud_write(pud_t pud)
1081 #endif /* pud_write */
1083 #if !defined(CONFIG_ARCH_HAS_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
1084 static inline int pmd_devmap(pmd_t pmd)
1088 static inline int pud_devmap(pud_t pud)
1092 static inline int pgd_devmap(pgd_t pgd)
1098 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
1099 (defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1100 !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
1101 static inline int pud_trans_huge(pud_t pud)
1107 /* See pmd_none_or_trans_huge_or_clear_bad for discussion. */
1108 static inline int pud_none_or_trans_huge_or_dev_or_clear_bad(pud_t *pud)
1110 pud_t pudval = READ_ONCE(*pud);
1112 if (pud_none(pudval) || pud_trans_huge(pudval) || pud_devmap(pudval))
1114 if (unlikely(pud_bad(pudval))) {
1121 /* See pmd_trans_unstable for discussion. */
1122 static inline int pud_trans_unstable(pud_t *pud)
1124 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
1125 defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)
1126 return pud_none_or_trans_huge_or_dev_or_clear_bad(pud);
1132 #ifndef pmd_read_atomic
1133 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
1136 * Depend on compiler for an atomic pmd read. NOTE: this is
1137 * only going to work, if the pmdval_t isn't larger than
1144 #ifndef arch_needs_pgtable_deposit
1145 #define arch_needs_pgtable_deposit() (false)
1148 * This function is meant to be used by sites walking pagetables with
1149 * the mmap_lock held in read mode to protect against MADV_DONTNEED and
1150 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
1151 * into a null pmd and the transhuge page fault can convert a null pmd
1152 * into an hugepmd or into a regular pmd (if the hugepage allocation
1153 * fails). While holding the mmap_lock in read mode the pmd becomes
1154 * stable and stops changing under us only if it's not null and not a
1155 * transhuge pmd. When those races occurs and this function makes a
1156 * difference vs the standard pmd_none_or_clear_bad, the result is
1157 * undefined so behaving like if the pmd was none is safe (because it
1158 * can return none anyway). The compiler level barrier() is critically
1159 * important to compute the two checks atomically on the same pmdval.
1161 * For 32bit kernels with a 64bit large pmd_t this automatically takes
1162 * care of reading the pmd atomically to avoid SMP race conditions
1163 * against pmd_populate() when the mmap_lock is hold for reading by the
1164 * caller (a special atomic read not done by "gcc" as in the generic
1165 * version above, is also needed when THP is disabled because the page
1166 * fault can populate the pmd from under us).
1168 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
1170 pmd_t pmdval = pmd_read_atomic(pmd);
1172 * The barrier will stabilize the pmdval in a register or on
1173 * the stack so that it will stop changing under the code.
1175 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
1176 * pmd_read_atomic is allowed to return a not atomic pmdval
1177 * (for example pointing to an hugepage that has never been
1178 * mapped in the pmd). The below checks will only care about
1179 * the low part of the pmd with 32bit PAE x86 anyway, with the
1180 * exception of pmd_none(). So the important thing is that if
1181 * the low part of the pmd is found null, the high part will
1182 * be also null or the pmd_none() check below would be
1185 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1189 * !pmd_present() checks for pmd migration entries
1191 * The complete check uses is_pmd_migration_entry() in linux/swapops.h
1192 * But using that requires moving current function and pmd_trans_unstable()
1193 * to linux/swapops.h to resovle dependency, which is too much code move.
1195 * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
1196 * because !pmd_present() pages can only be under migration not swapped
1199 * pmd_none() is preseved for future condition checks on pmd migration
1200 * entries and not confusing with this function name, although it is
1201 * redundant with !pmd_present().
1203 if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
1204 (IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
1206 if (unlikely(pmd_bad(pmdval))) {
1214 * This is a noop if Transparent Hugepage Support is not built into
1215 * the kernel. Otherwise it is equivalent to
1216 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
1217 * places that already verified the pmd is not none and they want to
1218 * walk ptes while holding the mmap sem in read mode (write mode don't
1219 * need this). If THP is not enabled, the pmd can't go away under the
1220 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
1221 * run a pmd_trans_unstable before walking the ptes after
1222 * split_huge_pmd returns (because it may have run when the pmd become
1223 * null, but then a page fault can map in a THP and not a regular page).
1225 static inline int pmd_trans_unstable(pmd_t *pmd)
1227 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1228 return pmd_none_or_trans_huge_or_clear_bad(pmd);
1234 #ifndef CONFIG_NUMA_BALANCING
1236 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
1237 * the only case the kernel cares is for NUMA balancing and is only ever set
1238 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
1239 * _PAGE_PROTNONE so by default, implement the helper as "always no". It
1240 * is the responsibility of the caller to distinguish between PROT_NONE
1241 * protections and NUMA hinting fault protections.
1243 static inline int pte_protnone(pte_t pte)
1248 static inline int pmd_protnone(pmd_t pmd)
1252 #endif /* CONFIG_NUMA_BALANCING */
1254 #endif /* CONFIG_MMU */
1256 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
1258 #ifndef __PAGETABLE_P4D_FOLDED
1259 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
1260 int p4d_clear_huge(p4d_t *p4d);
1262 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1266 static inline int p4d_clear_huge(p4d_t *p4d)
1270 #endif /* !__PAGETABLE_P4D_FOLDED */
1272 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
1273 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
1274 int pud_clear_huge(pud_t *pud);
1275 int pmd_clear_huge(pmd_t *pmd);
1276 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr);
1277 int pud_free_pmd_page(pud_t *pud, unsigned long addr);
1278 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr);
1279 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
1280 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
1284 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1288 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1292 static inline int p4d_clear_huge(p4d_t *p4d)
1296 static inline int pud_clear_huge(pud_t *pud)
1300 static inline int pmd_clear_huge(pmd_t *pmd)
1304 static inline int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1308 static inline int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1312 static inline int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1316 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
1318 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
1319 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1321 * ARCHes with special requirements for evicting THP backing TLB entries can
1322 * implement this. Otherwise also, it can help optimize normal TLB flush in
1323 * THP regime. Stock flush_tlb_range() typically has optimization to nuke the
1324 * entire TLB if flush span is greater than a threshold, which will
1325 * likely be true for a single huge page. Thus a single THP flush will
1326 * invalidate the entire TLB which is not desirable.
1327 * e.g. see arch/arc: flush_pmd_tlb_range
1329 #define flush_pmd_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1330 #define flush_pud_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1332 #define flush_pmd_tlb_range(vma, addr, end) BUILD_BUG()
1333 #define flush_pud_tlb_range(vma, addr, end) BUILD_BUG()
1338 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1339 unsigned long size, pgprot_t *vma_prot);
1341 #ifndef CONFIG_X86_ESPFIX64
1342 static inline void init_espfix_bsp(void) { }
1345 extern void __init pgtable_cache_init(void);
1347 #ifndef __HAVE_ARCH_PFN_MODIFY_ALLOWED
1348 static inline bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
1353 static inline bool arch_has_pfn_modify_check(void)
1357 #endif /* !_HAVE_ARCH_PFN_MODIFY_ALLOWED */
1360 * Architecture PAGE_KERNEL_* fallbacks
1362 * Some architectures don't define certain PAGE_KERNEL_* flags. This is either
1363 * because they really don't support them, or the port needs to be updated to
1364 * reflect the required functionality. Below are a set of relatively safe
1365 * fallbacks, as best effort, which we can count on in lieu of the architectures
1366 * not defining them on their own yet.
1369 #ifndef PAGE_KERNEL_RO
1370 # define PAGE_KERNEL_RO PAGE_KERNEL
1373 #ifndef PAGE_KERNEL_EXEC
1374 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1378 * Page Table Modification bits for pgtbl_mod_mask.
1380 * These are used by the p?d_alloc_track*() set of functions an in the generic
1381 * vmalloc/ioremap code to track at which page-table levels entries have been
1382 * modified. Based on that the code can better decide when vmalloc and ioremap
1383 * mapping changes need to be synchronized to other page-tables in the system.
1385 #define __PGTBL_PGD_MODIFIED 0
1386 #define __PGTBL_P4D_MODIFIED 1
1387 #define __PGTBL_PUD_MODIFIED 2
1388 #define __PGTBL_PMD_MODIFIED 3
1389 #define __PGTBL_PTE_MODIFIED 4
1391 #define PGTBL_PGD_MODIFIED BIT(__PGTBL_PGD_MODIFIED)
1392 #define PGTBL_P4D_MODIFIED BIT(__PGTBL_P4D_MODIFIED)
1393 #define PGTBL_PUD_MODIFIED BIT(__PGTBL_PUD_MODIFIED)
1394 #define PGTBL_PMD_MODIFIED BIT(__PGTBL_PMD_MODIFIED)
1395 #define PGTBL_PTE_MODIFIED BIT(__PGTBL_PTE_MODIFIED)
1397 /* Page-Table Modification Mask */
1398 typedef unsigned int pgtbl_mod_mask;
1400 #endif /* !__ASSEMBLY__ */
1402 #ifndef io_remap_pfn_range
1403 #define io_remap_pfn_range remap_pfn_range
1406 #ifndef has_transparent_hugepage
1407 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1408 #define has_transparent_hugepage() 1
1410 #define has_transparent_hugepage() 0
1415 * On some architectures it depends on the mm if the p4d/pud or pmd
1416 * layer of the page table hierarchy is folded or not.
1418 #ifndef mm_p4d_folded
1419 #define mm_p4d_folded(mm) __is_defined(__PAGETABLE_P4D_FOLDED)
1422 #ifndef mm_pud_folded
1423 #define mm_pud_folded(mm) __is_defined(__PAGETABLE_PUD_FOLDED)
1426 #ifndef mm_pmd_folded
1427 #define mm_pmd_folded(mm) __is_defined(__PAGETABLE_PMD_FOLDED)
1431 * p?d_leaf() - true if this entry is a final mapping to a physical address.
1432 * This differs from p?d_huge() by the fact that they are always available (if
1433 * the architecture supports large pages at the appropriate level) even
1434 * if CONFIG_HUGETLB_PAGE is not defined.
1435 * Only meaningful when called on a valid entry.
1438 #define pgd_leaf(x) 0
1441 #define p4d_leaf(x) 0
1444 #define pud_leaf(x) 0
1447 #define pmd_leaf(x) 0
1450 #endif /* _LINUX_PGTABLE_H */