1 #ifndef _ASM_GENERIC_PGTABLE_H
2 #define _ASM_GENERIC_PGTABLE_H
9 #include <linux/mm_types.h>
10 #include <linux/bug.h>
11 #include <linux/errno.h>
13 #if 5 - defined(__PAGETABLE_P4D_FOLDED) - defined(__PAGETABLE_PUD_FOLDED) - \
14 defined(__PAGETABLE_PMD_FOLDED) != CONFIG_PGTABLE_LEVELS
15 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{P4D,PUD,PMD}_FOLDED
19 * On almost all architectures and configurations, 0 can be used as the
20 * upper ceiling to free_pgtables(): on many architectures it has the same
21 * effect as using TASK_SIZE. However, there is one configuration which
22 * must impose a more careful limit, to avoid freeing kernel pgtables.
24 #ifndef USER_PGTABLES_CEILING
25 #define USER_PGTABLES_CEILING 0UL
28 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
29 extern int ptep_set_access_flags(struct vm_area_struct *vma,
30 unsigned long address, pte_t *ptep,
31 pte_t entry, int dirty);
34 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
35 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
36 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
37 unsigned long address, pmd_t *pmdp,
38 pmd_t entry, int dirty);
39 extern int pudp_set_access_flags(struct vm_area_struct *vma,
40 unsigned long address, pud_t *pudp,
41 pud_t entry, int dirty);
43 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
44 unsigned long address, pmd_t *pmdp,
45 pmd_t entry, int dirty)
50 static inline int pudp_set_access_flags(struct vm_area_struct *vma,
51 unsigned long address, pud_t *pudp,
52 pud_t entry, int dirty)
57 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
60 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
61 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
62 unsigned long address,
70 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
75 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
76 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
77 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
78 unsigned long address,
86 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
90 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
91 unsigned long address,
97 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
100 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
101 int ptep_clear_flush_young(struct vm_area_struct *vma,
102 unsigned long address, pte_t *ptep);
105 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
106 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
107 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
108 unsigned long address, pmd_t *pmdp);
111 * Despite relevant to THP only, this API is called from generic rmap code
112 * under PageTransHuge(), hence needs a dummy implementation for !THP
114 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
115 unsigned long address, pmd_t *pmdp)
120 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
123 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
124 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
125 unsigned long address,
129 pte_clear(mm, address, ptep);
134 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
135 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
136 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
137 unsigned long address,
144 #endif /* __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR */
145 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR
146 static inline pud_t pudp_huge_get_and_clear(struct mm_struct *mm,
147 unsigned long address,
155 #endif /* __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR */
156 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
158 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
159 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
160 static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
161 unsigned long address, pmd_t *pmdp,
164 return pmdp_huge_get_and_clear(mm, address, pmdp);
168 #ifndef __HAVE_ARCH_PUDP_HUGE_GET_AND_CLEAR_FULL
169 static inline pud_t pudp_huge_get_and_clear_full(struct mm_struct *mm,
170 unsigned long address, pud_t *pudp,
173 return pudp_huge_get_and_clear(mm, address, pudp);
176 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
178 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
179 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
180 unsigned long address, pte_t *ptep,
184 pte = ptep_get_and_clear(mm, address, ptep);
190 * Some architectures may be able to avoid expensive synchronization
191 * primitives when modifications are made to PTE's which are already
192 * not present, or in the process of an address space destruction.
194 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
195 static inline void pte_clear_not_present_full(struct mm_struct *mm,
196 unsigned long address,
200 pte_clear(mm, address, ptep);
204 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
205 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
206 unsigned long address,
210 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
211 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
212 unsigned long address,
214 extern pud_t pudp_huge_clear_flush(struct vm_area_struct *vma,
215 unsigned long address,
219 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
221 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
223 pte_t old_pte = *ptep;
224 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
228 #ifndef pte_savedwrite
229 #define pte_savedwrite pte_write
232 #ifndef pte_mk_savedwrite
233 #define pte_mk_savedwrite pte_mkwrite
236 #ifndef pte_clear_savedwrite
237 #define pte_clear_savedwrite pte_wrprotect
240 #ifndef pmd_savedwrite
241 #define pmd_savedwrite pmd_write
244 #ifndef pmd_mk_savedwrite
245 #define pmd_mk_savedwrite pmd_mkwrite
248 #ifndef pmd_clear_savedwrite
249 #define pmd_clear_savedwrite pmd_wrprotect
252 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
253 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
254 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
255 unsigned long address, pmd_t *pmdp)
257 pmd_t old_pmd = *pmdp;
258 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
261 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
262 unsigned long address, pmd_t *pmdp)
266 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
268 #ifndef __HAVE_ARCH_PUDP_SET_WRPROTECT
269 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
270 static inline void pudp_set_wrprotect(struct mm_struct *mm,
271 unsigned long address, pud_t *pudp)
273 pud_t old_pud = *pudp;
275 set_pud_at(mm, address, pudp, pud_wrprotect(old_pud));
278 static inline void pudp_set_wrprotect(struct mm_struct *mm,
279 unsigned long address, pud_t *pudp)
283 #endif /* CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD */
286 #ifndef pmdp_collapse_flush
287 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
288 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
289 unsigned long address, pmd_t *pmdp);
291 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
292 unsigned long address,
298 #define pmdp_collapse_flush pmdp_collapse_flush
299 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
302 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
303 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
307 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
308 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
311 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
312 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
316 #ifndef __HAVE_ARCH_PMDP_HUGE_SPLIT_PREPARE
317 static inline void pmdp_huge_split_prepare(struct vm_area_struct *vma,
318 unsigned long address, pmd_t *pmdp)
324 #ifndef __HAVE_ARCH_PTE_SAME
325 static inline int pte_same(pte_t pte_a, pte_t pte_b)
327 return pte_val(pte_a) == pte_val(pte_b);
331 #ifndef __HAVE_ARCH_PTE_UNUSED
333 * Some architectures provide facilities to virtualization guests
334 * so that they can flag allocated pages as unused. This allows the
335 * host to transparently reclaim unused pages. This function returns
336 * whether the pte's page is unused.
338 static inline int pte_unused(pte_t pte)
344 #ifndef pte_access_permitted
345 #define pte_access_permitted(pte, write) \
346 (pte_present(pte) && (!(write) || pte_write(pte)))
349 #ifndef pmd_access_permitted
350 #define pmd_access_permitted(pmd, write) \
351 (pmd_present(pmd) && (!(write) || pmd_write(pmd)))
354 #ifndef pud_access_permitted
355 #define pud_access_permitted(pud, write) \
356 (pud_present(pud) && (!(write) || pud_write(pud)))
359 #ifndef p4d_access_permitted
360 #define p4d_access_permitted(p4d, write) \
361 (p4d_present(p4d) && (!(write) || p4d_write(p4d)))
364 #ifndef pgd_access_permitted
365 #define pgd_access_permitted(pgd, write) \
366 (pgd_present(pgd) && (!(write) || pgd_write(pgd)))
369 #ifndef __HAVE_ARCH_PMD_SAME
370 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
371 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
373 return pmd_val(pmd_a) == pmd_val(pmd_b);
376 static inline int pud_same(pud_t pud_a, pud_t pud_b)
378 return pud_val(pud_a) == pud_val(pud_b);
380 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
381 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
387 static inline int pud_same(pud_t pud_a, pud_t pud_b)
392 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
395 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
396 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
399 #ifndef __HAVE_ARCH_MOVE_PTE
400 #define move_pte(pte, prot, old_addr, new_addr) (pte)
403 #ifndef pte_accessible
404 # define pte_accessible(mm, pte) ((void)(pte), 1)
407 #ifndef flush_tlb_fix_spurious_fault
408 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
411 #ifndef pgprot_noncached
412 #define pgprot_noncached(prot) (prot)
415 #ifndef pgprot_writecombine
416 #define pgprot_writecombine pgprot_noncached
419 #ifndef pgprot_writethrough
420 #define pgprot_writethrough pgprot_noncached
423 #ifndef pgprot_device
424 #define pgprot_device pgprot_noncached
427 #ifndef pgprot_modify
428 #define pgprot_modify pgprot_modify
429 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
431 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
432 newprot = pgprot_noncached(newprot);
433 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
434 newprot = pgprot_writecombine(newprot);
435 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
436 newprot = pgprot_device(newprot);
442 * When walking page tables, get the address of the next boundary,
443 * or the end address of the range if that comes earlier. Although no
444 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
447 #define pgd_addr_end(addr, end) \
448 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
449 (__boundary - 1 < (end) - 1)? __boundary: (end); \
453 #define p4d_addr_end(addr, end) \
454 ({ unsigned long __boundary = ((addr) + P4D_SIZE) & P4D_MASK; \
455 (__boundary - 1 < (end) - 1)? __boundary: (end); \
460 #define pud_addr_end(addr, end) \
461 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
462 (__boundary - 1 < (end) - 1)? __boundary: (end); \
467 #define pmd_addr_end(addr, end) \
468 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
469 (__boundary - 1 < (end) - 1)? __boundary: (end); \
474 * When walking page tables, we usually want to skip any p?d_none entries;
475 * and any p?d_bad entries - reporting the error before resetting to none.
476 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
478 void pgd_clear_bad(pgd_t *);
479 void p4d_clear_bad(p4d_t *);
480 void pud_clear_bad(pud_t *);
481 void pmd_clear_bad(pmd_t *);
483 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
487 if (unlikely(pgd_bad(*pgd))) {
494 static inline int p4d_none_or_clear_bad(p4d_t *p4d)
498 if (unlikely(p4d_bad(*p4d))) {
505 static inline int pud_none_or_clear_bad(pud_t *pud)
509 if (unlikely(pud_bad(*pud))) {
516 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
520 if (unlikely(pmd_bad(*pmd))) {
527 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
532 * Get the current pte state, but zero it out to make it
533 * non-present, preventing the hardware from asynchronously
536 return ptep_get_and_clear(mm, addr, ptep);
539 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
541 pte_t *ptep, pte_t pte)
544 * The pte is non-present, so there's no hardware state to
547 set_pte_at(mm, addr, ptep, pte);
550 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
552 * Start a pte protection read-modify-write transaction, which
553 * protects against asynchronous hardware modifications to the pte.
554 * The intention is not to prevent the hardware from making pte
555 * updates, but to prevent any updates it may make from being lost.
557 * This does not protect against other software modifications of the
558 * pte; the appropriate pte lock must be held over the transation.
560 * Note that this interface is intended to be batchable, meaning that
561 * ptep_modify_prot_commit may not actually update the pte, but merely
562 * queue the update to be done at some later time. The update must be
563 * actually committed before the pte lock is released, however.
565 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
569 return __ptep_modify_prot_start(mm, addr, ptep);
573 * Commit an update to a pte, leaving any hardware-controlled bits in
574 * the PTE unmodified.
576 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
578 pte_t *ptep, pte_t pte)
580 __ptep_modify_prot_commit(mm, addr, ptep, pte);
582 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
583 #endif /* CONFIG_MMU */
586 * No-op macros that just return the current protection value. Defined here
587 * because these macros can be used used even if CONFIG_MMU is not defined.
589 #ifndef pgprot_encrypted
590 #define pgprot_encrypted(prot) (prot)
593 #ifndef pgprot_decrypted
594 #define pgprot_decrypted(prot) (prot)
598 * A facility to provide lazy MMU batching. This allows PTE updates and
599 * page invalidations to be delayed until a call to leave lazy MMU mode
600 * is issued. Some architectures may benefit from doing this, and it is
601 * beneficial for both shadow and direct mode hypervisors, which may batch
602 * the PTE updates which happen during this window. Note that using this
603 * interface requires that read hazards be removed from the code. A read
604 * hazard could result in the direct mode hypervisor case, since the actual
605 * write to the page tables may not yet have taken place, so reads though
606 * a raw PTE pointer after it has been modified are not guaranteed to be
607 * up to date. This mode can only be entered and left under the protection of
608 * the page table locks for all page tables which may be modified. In the UP
609 * case, this is required so that preemption is disabled, and in the SMP case,
610 * it must synchronize the delayed page table writes properly on other CPUs.
612 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
613 #define arch_enter_lazy_mmu_mode() do {} while (0)
614 #define arch_leave_lazy_mmu_mode() do {} while (0)
615 #define arch_flush_lazy_mmu_mode() do {} while (0)
619 * A facility to provide batching of the reload of page tables and
620 * other process state with the actual context switch code for
621 * paravirtualized guests. By convention, only one of the batched
622 * update (lazy) modes (CPU, MMU) should be active at any given time,
623 * entry should never be nested, and entry and exits should always be
624 * paired. This is for sanity of maintaining and reasoning about the
625 * kernel code. In this case, the exit (end of the context switch) is
626 * in architecture-specific code, and so doesn't need a generic
629 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
630 #define arch_start_context_switch(prev) do {} while (0)
633 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
634 #ifndef CONFIG_ARCH_ENABLE_THP_MIGRATION
635 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
640 static inline int pmd_swp_soft_dirty(pmd_t pmd)
645 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
650 #else /* !CONFIG_HAVE_ARCH_SOFT_DIRTY */
651 static inline int pte_soft_dirty(pte_t pte)
656 static inline int pmd_soft_dirty(pmd_t pmd)
661 static inline pte_t pte_mksoft_dirty(pte_t pte)
666 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
671 static inline pte_t pte_clear_soft_dirty(pte_t pte)
676 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
681 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
686 static inline int pte_swp_soft_dirty(pte_t pte)
691 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
696 static inline pmd_t pmd_swp_mksoft_dirty(pmd_t pmd)
701 static inline int pmd_swp_soft_dirty(pmd_t pmd)
706 static inline pmd_t pmd_swp_clear_soft_dirty(pmd_t pmd)
712 #ifndef __HAVE_PFNMAP_TRACKING
714 * Interfaces that can be used by architecture code to keep track of
715 * memory type of pfn mappings specified by the remap_pfn_range,
720 * track_pfn_remap is called when a _new_ pfn mapping is being established
721 * by remap_pfn_range() for physical range indicated by pfn and size.
723 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
724 unsigned long pfn, unsigned long addr,
731 * track_pfn_insert is called when a _new_ single pfn is established
732 * by vm_insert_pfn().
734 static inline void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
740 * track_pfn_copy is called when vma that is covering the pfnmap gets
741 * copied through copy_page_range().
743 static inline int track_pfn_copy(struct vm_area_struct *vma)
749 * untrack_pfn is called while unmapping a pfnmap for a region.
750 * untrack can be called for a specific region indicated by pfn and size or
751 * can be for the entire vma (in which case pfn, size are zero).
753 static inline void untrack_pfn(struct vm_area_struct *vma,
754 unsigned long pfn, unsigned long size)
759 * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
761 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
765 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
766 unsigned long pfn, unsigned long addr,
768 extern void track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
770 extern int track_pfn_copy(struct vm_area_struct *vma);
771 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
773 extern void untrack_pfn_moved(struct vm_area_struct *vma);
776 #ifdef __HAVE_COLOR_ZERO_PAGE
777 static inline int is_zero_pfn(unsigned long pfn)
779 extern unsigned long zero_pfn;
780 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
781 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
784 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
787 static inline int is_zero_pfn(unsigned long pfn)
789 extern unsigned long zero_pfn;
790 return pfn == zero_pfn;
793 static inline unsigned long my_zero_pfn(unsigned long addr)
795 extern unsigned long zero_pfn;
802 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
803 static inline int pmd_trans_huge(pmd_t pmd)
807 #ifndef __HAVE_ARCH_PMD_WRITE
808 static inline int pmd_write(pmd_t pmd)
813 #endif /* __HAVE_ARCH_PMD_WRITE */
814 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
816 #if !defined(CONFIG_TRANSPARENT_HUGEPAGE) || \
817 (defined(CONFIG_TRANSPARENT_HUGEPAGE) && \
818 !defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
819 static inline int pud_trans_huge(pud_t pud)
825 #ifndef pmd_read_atomic
826 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
829 * Depend on compiler for an atomic pmd read. NOTE: this is
830 * only going to work, if the pmdval_t isn't larger than
837 #ifndef arch_needs_pgtable_deposit
838 #define arch_needs_pgtable_deposit() (false)
841 * This function is meant to be used by sites walking pagetables with
842 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
843 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
844 * into a null pmd and the transhuge page fault can convert a null pmd
845 * into an hugepmd or into a regular pmd (if the hugepage allocation
846 * fails). While holding the mmap_sem in read mode the pmd becomes
847 * stable and stops changing under us only if it's not null and not a
848 * transhuge pmd. When those races occurs and this function makes a
849 * difference vs the standard pmd_none_or_clear_bad, the result is
850 * undefined so behaving like if the pmd was none is safe (because it
851 * can return none anyway). The compiler level barrier() is critically
852 * important to compute the two checks atomically on the same pmdval.
854 * For 32bit kernels with a 64bit large pmd_t this automatically takes
855 * care of reading the pmd atomically to avoid SMP race conditions
856 * against pmd_populate() when the mmap_sem is hold for reading by the
857 * caller (a special atomic read not done by "gcc" as in the generic
858 * version above, is also needed when THP is disabled because the page
859 * fault can populate the pmd from under us).
861 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
863 pmd_t pmdval = pmd_read_atomic(pmd);
865 * The barrier will stabilize the pmdval in a register or on
866 * the stack so that it will stop changing under the code.
868 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
869 * pmd_read_atomic is allowed to return a not atomic pmdval
870 * (for example pointing to an hugepage that has never been
871 * mapped in the pmd). The below checks will only care about
872 * the low part of the pmd with 32bit PAE x86 anyway, with the
873 * exception of pmd_none(). So the important thing is that if
874 * the low part of the pmd is found null, the high part will
875 * be also null or the pmd_none() check below would be
878 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
882 * !pmd_present() checks for pmd migration entries
884 * The complete check uses is_pmd_migration_entry() in linux/swapops.h
885 * But using that requires moving current function and pmd_trans_unstable()
886 * to linux/swapops.h to resovle dependency, which is too much code move.
888 * !pmd_present() is equivalent to is_pmd_migration_entry() currently,
889 * because !pmd_present() pages can only be under migration not swapped
892 * pmd_none() is preseved for future condition checks on pmd migration
893 * entries and not confusing with this function name, although it is
894 * redundant with !pmd_present().
896 if (pmd_none(pmdval) || pmd_trans_huge(pmdval) ||
897 (IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION) && !pmd_present(pmdval)))
899 if (unlikely(pmd_bad(pmdval))) {
907 * This is a noop if Transparent Hugepage Support is not built into
908 * the kernel. Otherwise it is equivalent to
909 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
910 * places that already verified the pmd is not none and they want to
911 * walk ptes while holding the mmap sem in read mode (write mode don't
912 * need this). If THP is not enabled, the pmd can't go away under the
913 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
914 * run a pmd_trans_unstable before walking the ptes after
915 * split_huge_page_pmd returns (because it may have run when the pmd
916 * become null, but then a page fault can map in a THP and not a
919 static inline int pmd_trans_unstable(pmd_t *pmd)
921 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
922 return pmd_none_or_trans_huge_or_clear_bad(pmd);
928 #ifndef CONFIG_NUMA_BALANCING
930 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
931 * the only case the kernel cares is for NUMA balancing and is only ever set
932 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
933 * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
934 * is the responsibility of the caller to distinguish between PROT_NONE
935 * protections and NUMA hinting fault protections.
937 static inline int pte_protnone(pte_t pte)
942 static inline int pmd_protnone(pmd_t pmd)
946 #endif /* CONFIG_NUMA_BALANCING */
948 #endif /* CONFIG_MMU */
950 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
952 #ifndef __PAGETABLE_P4D_FOLDED
953 int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot);
954 int p4d_clear_huge(p4d_t *p4d);
956 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
960 static inline int p4d_clear_huge(p4d_t *p4d)
964 #endif /* !__PAGETABLE_P4D_FOLDED */
966 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
967 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
968 int pud_clear_huge(pud_t *pud);
969 int pmd_clear_huge(pmd_t *pmd);
970 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
971 static inline int p4d_set_huge(p4d_t *p4d, phys_addr_t addr, pgprot_t prot)
975 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
979 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
983 static inline int p4d_clear_huge(p4d_t *p4d)
987 static inline int pud_clear_huge(pud_t *pud)
991 static inline int pmd_clear_huge(pmd_t *pmd)
995 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
997 #ifndef __HAVE_ARCH_FLUSH_PMD_TLB_RANGE
998 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1000 * ARCHes with special requirements for evicting THP backing TLB entries can
1001 * implement this. Otherwise also, it can help optimize normal TLB flush in
1002 * THP regime. stock flush_tlb_range() typically has optimization to nuke the
1003 * entire TLB TLB if flush span is greater than a threshold, which will
1004 * likely be true for a single huge page. Thus a single thp flush will
1005 * invalidate the entire TLB which is not desitable.
1006 * e.g. see arch/arc: flush_pmd_tlb_range
1008 #define flush_pmd_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1009 #define flush_pud_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1011 #define flush_pmd_tlb_range(vma, addr, end) BUILD_BUG()
1012 #define flush_pud_tlb_range(vma, addr, end) BUILD_BUG()
1017 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
1018 unsigned long size, pgprot_t *vma_prot);
1019 #endif /* !__ASSEMBLY__ */
1021 #ifndef io_remap_pfn_range
1022 #define io_remap_pfn_range remap_pfn_range
1025 #ifndef has_transparent_hugepage
1026 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1027 #define has_transparent_hugepage() 1
1029 #define has_transparent_hugepage() 0
1033 #endif /* _ASM_GENERIC_PGTABLE_H */