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 4 - defined(__PAGETABLE_PUD_FOLDED) - defined(__PAGETABLE_PMD_FOLDED) != \
15 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{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);
40 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
41 unsigned long address, pmd_t *pmdp,
42 pmd_t entry, int dirty)
47 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
50 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
51 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
52 unsigned long address,
60 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
65 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
66 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
67 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
68 unsigned long address,
76 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
80 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
81 unsigned long address,
87 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
90 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
91 int ptep_clear_flush_young(struct vm_area_struct *vma,
92 unsigned long address, pte_t *ptep);
95 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
96 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
97 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
98 unsigned long address, pmd_t *pmdp);
101 * Despite relevant to THP only, this API is called from generic rmap code
102 * under PageTransHuge(), hence needs a dummy implementation for !THP
104 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
105 unsigned long address, pmd_t *pmdp)
110 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
113 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
114 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
115 unsigned long address,
119 pte_clear(mm, address, ptep);
124 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
125 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
126 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
127 unsigned long address,
134 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
137 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
138 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
139 static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
140 unsigned long address, pmd_t *pmdp,
143 return pmdp_huge_get_and_clear(mm, address, pmdp);
145 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
148 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
149 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
150 unsigned long address, pte_t *ptep,
154 pte = ptep_get_and_clear(mm, address, ptep);
160 * Some architectures may be able to avoid expensive synchronization
161 * primitives when modifications are made to PTE's which are already
162 * not present, or in the process of an address space destruction.
164 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
165 static inline void pte_clear_not_present_full(struct mm_struct *mm,
166 unsigned long address,
170 pte_clear(mm, address, ptep);
174 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
175 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
176 unsigned long address,
180 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
181 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
182 unsigned long address,
186 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
188 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
190 pte_t old_pte = *ptep;
191 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
195 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
196 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
197 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
198 unsigned long address, pmd_t *pmdp)
200 pmd_t old_pmd = *pmdp;
201 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
204 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
205 unsigned long address, pmd_t *pmdp)
209 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
212 #ifndef pmdp_collapse_flush
213 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
214 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
215 unsigned long address, pmd_t *pmdp);
217 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
218 unsigned long address,
224 #define pmdp_collapse_flush pmdp_collapse_flush
225 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
228 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
229 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
233 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
234 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
237 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
238 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
242 #ifndef __HAVE_ARCH_PTE_SAME
243 static inline int pte_same(pte_t pte_a, pte_t pte_b)
245 return pte_val(pte_a) == pte_val(pte_b);
249 #ifndef __HAVE_ARCH_PTE_UNUSED
251 * Some architectures provide facilities to virtualization guests
252 * so that they can flag allocated pages as unused. This allows the
253 * host to transparently reclaim unused pages. This function returns
254 * whether the pte's page is unused.
256 static inline int pte_unused(pte_t pte)
262 #ifndef __HAVE_ARCH_PMD_SAME
263 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
264 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
266 return pmd_val(pmd_a) == pmd_val(pmd_b);
268 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
269 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
274 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
277 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
278 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
281 #ifndef __HAVE_ARCH_MOVE_PTE
282 #define move_pte(pte, prot, old_addr, new_addr) (pte)
285 #ifndef pte_accessible
286 # define pte_accessible(mm, pte) ((void)(pte), 1)
289 #ifndef flush_tlb_fix_spurious_fault
290 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
293 #ifndef pgprot_noncached
294 #define pgprot_noncached(prot) (prot)
297 #ifndef pgprot_writecombine
298 #define pgprot_writecombine pgprot_noncached
301 #ifndef pgprot_writethrough
302 #define pgprot_writethrough pgprot_noncached
305 #ifndef pgprot_device
306 #define pgprot_device pgprot_noncached
309 #ifndef pgprot_modify
310 #define pgprot_modify pgprot_modify
311 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
313 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
314 newprot = pgprot_noncached(newprot);
315 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
316 newprot = pgprot_writecombine(newprot);
317 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
318 newprot = pgprot_device(newprot);
324 * When walking page tables, get the address of the next boundary,
325 * or the end address of the range if that comes earlier. Although no
326 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
329 #define pgd_addr_end(addr, end) \
330 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
331 (__boundary - 1 < (end) - 1)? __boundary: (end); \
335 #define pud_addr_end(addr, end) \
336 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
337 (__boundary - 1 < (end) - 1)? __boundary: (end); \
342 #define pmd_addr_end(addr, end) \
343 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
344 (__boundary - 1 < (end) - 1)? __boundary: (end); \
349 * When walking page tables, we usually want to skip any p?d_none entries;
350 * and any p?d_bad entries - reporting the error before resetting to none.
351 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
353 void pgd_clear_bad(pgd_t *);
354 void pud_clear_bad(pud_t *);
355 void pmd_clear_bad(pmd_t *);
357 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
361 if (unlikely(pgd_bad(*pgd))) {
368 static inline int pud_none_or_clear_bad(pud_t *pud)
372 if (unlikely(pud_bad(*pud))) {
379 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
383 if (unlikely(pmd_bad(*pmd))) {
390 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
395 * Get the current pte state, but zero it out to make it
396 * non-present, preventing the hardware from asynchronously
399 return ptep_get_and_clear(mm, addr, ptep);
402 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
404 pte_t *ptep, pte_t pte)
407 * The pte is non-present, so there's no hardware state to
410 set_pte_at(mm, addr, ptep, pte);
413 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
415 * Start a pte protection read-modify-write transaction, which
416 * protects against asynchronous hardware modifications to the pte.
417 * The intention is not to prevent the hardware from making pte
418 * updates, but to prevent any updates it may make from being lost.
420 * This does not protect against other software modifications of the
421 * pte; the appropriate pte lock must be held over the transation.
423 * Note that this interface is intended to be batchable, meaning that
424 * ptep_modify_prot_commit may not actually update the pte, but merely
425 * queue the update to be done at some later time. The update must be
426 * actually committed before the pte lock is released, however.
428 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
432 return __ptep_modify_prot_start(mm, addr, ptep);
436 * Commit an update to a pte, leaving any hardware-controlled bits in
437 * the PTE unmodified.
439 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
441 pte_t *ptep, pte_t pte)
443 __ptep_modify_prot_commit(mm, addr, ptep, pte);
445 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
446 #endif /* CONFIG_MMU */
449 * A facility to provide lazy MMU batching. This allows PTE updates and
450 * page invalidations to be delayed until a call to leave lazy MMU mode
451 * is issued. Some architectures may benefit from doing this, and it is
452 * beneficial for both shadow and direct mode hypervisors, which may batch
453 * the PTE updates which happen during this window. Note that using this
454 * interface requires that read hazards be removed from the code. A read
455 * hazard could result in the direct mode hypervisor case, since the actual
456 * write to the page tables may not yet have taken place, so reads though
457 * a raw PTE pointer after it has been modified are not guaranteed to be
458 * up to date. This mode can only be entered and left under the protection of
459 * the page table locks for all page tables which may be modified. In the UP
460 * case, this is required so that preemption is disabled, and in the SMP case,
461 * it must synchronize the delayed page table writes properly on other CPUs.
463 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
464 #define arch_enter_lazy_mmu_mode() do {} while (0)
465 #define arch_leave_lazy_mmu_mode() do {} while (0)
466 #define arch_flush_lazy_mmu_mode() do {} while (0)
470 * A facility to provide batching of the reload of page tables and
471 * other process state with the actual context switch code for
472 * paravirtualized guests. By convention, only one of the batched
473 * update (lazy) modes (CPU, MMU) should be active at any given time,
474 * entry should never be nested, and entry and exits should always be
475 * paired. This is for sanity of maintaining and reasoning about the
476 * kernel code. In this case, the exit (end of the context switch) is
477 * in architecture-specific code, and so doesn't need a generic
480 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
481 #define arch_start_context_switch(prev) do {} while (0)
484 #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
485 static inline int pte_soft_dirty(pte_t pte)
490 static inline int pmd_soft_dirty(pmd_t pmd)
495 static inline pte_t pte_mksoft_dirty(pte_t pte)
500 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
505 static inline pte_t pte_clear_soft_dirty(pte_t pte)
510 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
515 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
520 static inline int pte_swp_soft_dirty(pte_t pte)
525 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
531 #ifndef __HAVE_PFNMAP_TRACKING
533 * Interfaces that can be used by architecture code to keep track of
534 * memory type of pfn mappings specified by the remap_pfn_range,
539 * track_pfn_remap is called when a _new_ pfn mapping is being established
540 * by remap_pfn_range() for physical range indicated by pfn and size.
542 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
543 unsigned long pfn, unsigned long addr,
550 * track_pfn_insert is called when a _new_ single pfn is established
551 * by vm_insert_pfn().
553 static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
560 * track_pfn_copy is called when vma that is covering the pfnmap gets
561 * copied through copy_page_range().
563 static inline int track_pfn_copy(struct vm_area_struct *vma)
569 * untrack_pfn is called while unmapping a pfnmap for a region.
570 * untrack can be called for a specific region indicated by pfn and size or
571 * can be for the entire vma (in which case pfn, size are zero).
573 static inline void untrack_pfn(struct vm_area_struct *vma,
574 unsigned long pfn, unsigned long size)
579 * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
581 static inline void untrack_pfn_moved(struct vm_area_struct *vma)
585 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
586 unsigned long pfn, unsigned long addr,
588 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
590 extern int track_pfn_copy(struct vm_area_struct *vma);
591 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
593 extern void untrack_pfn_moved(struct vm_area_struct *vma);
596 #ifdef __HAVE_COLOR_ZERO_PAGE
597 static inline int is_zero_pfn(unsigned long pfn)
599 extern unsigned long zero_pfn;
600 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
601 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
604 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
607 static inline int is_zero_pfn(unsigned long pfn)
609 extern unsigned long zero_pfn;
610 return pfn == zero_pfn;
613 static inline unsigned long my_zero_pfn(unsigned long addr)
615 extern unsigned long zero_pfn;
622 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
623 static inline int pmd_trans_huge(pmd_t pmd)
627 #ifndef __HAVE_ARCH_PMD_WRITE
628 static inline int pmd_write(pmd_t pmd)
633 #endif /* __HAVE_ARCH_PMD_WRITE */
634 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
636 #ifndef pmd_read_atomic
637 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
640 * Depend on compiler for an atomic pmd read. NOTE: this is
641 * only going to work, if the pmdval_t isn't larger than
648 #ifndef pmd_move_must_withdraw
649 static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
650 spinlock_t *old_pmd_ptl)
653 * With split pmd lock we also need to move preallocated
654 * PTE page table if new_pmd is on different PMD page table.
656 return new_pmd_ptl != old_pmd_ptl;
661 * This function is meant to be used by sites walking pagetables with
662 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
663 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
664 * into a null pmd and the transhuge page fault can convert a null pmd
665 * into an hugepmd or into a regular pmd (if the hugepage allocation
666 * fails). While holding the mmap_sem in read mode the pmd becomes
667 * stable and stops changing under us only if it's not null and not a
668 * transhuge pmd. When those races occurs and this function makes a
669 * difference vs the standard pmd_none_or_clear_bad, the result is
670 * undefined so behaving like if the pmd was none is safe (because it
671 * can return none anyway). The compiler level barrier() is critically
672 * important to compute the two checks atomically on the same pmdval.
674 * For 32bit kernels with a 64bit large pmd_t this automatically takes
675 * care of reading the pmd atomically to avoid SMP race conditions
676 * against pmd_populate() when the mmap_sem is hold for reading by the
677 * caller (a special atomic read not done by "gcc" as in the generic
678 * version above, is also needed when THP is disabled because the page
679 * fault can populate the pmd from under us).
681 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
683 pmd_t pmdval = pmd_read_atomic(pmd);
685 * The barrier will stabilize the pmdval in a register or on
686 * the stack so that it will stop changing under the code.
688 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
689 * pmd_read_atomic is allowed to return a not atomic pmdval
690 * (for example pointing to an hugepage that has never been
691 * mapped in the pmd). The below checks will only care about
692 * the low part of the pmd with 32bit PAE x86 anyway, with the
693 * exception of pmd_none(). So the important thing is that if
694 * the low part of the pmd is found null, the high part will
695 * be also null or the pmd_none() check below would be
698 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
701 if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
703 if (unlikely(pmd_bad(pmdval))) {
711 * This is a noop if Transparent Hugepage Support is not built into
712 * the kernel. Otherwise it is equivalent to
713 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
714 * places that already verified the pmd is not none and they want to
715 * walk ptes while holding the mmap sem in read mode (write mode don't
716 * need this). If THP is not enabled, the pmd can't go away under the
717 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
718 * run a pmd_trans_unstable before walking the ptes after
719 * split_huge_page_pmd returns (because it may have run when the pmd
720 * become null, but then a page fault can map in a THP and not a
723 static inline int pmd_trans_unstable(pmd_t *pmd)
725 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
726 return pmd_none_or_trans_huge_or_clear_bad(pmd);
732 #ifndef CONFIG_NUMA_BALANCING
734 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
735 * the only case the kernel cares is for NUMA balancing and is only ever set
736 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
737 * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
738 * is the responsibility of the caller to distinguish between PROT_NONE
739 * protections and NUMA hinting fault protections.
741 static inline int pte_protnone(pte_t pte)
746 static inline int pmd_protnone(pmd_t pmd)
750 #endif /* CONFIG_NUMA_BALANCING */
752 #endif /* CONFIG_MMU */
754 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
755 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
756 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
757 int pud_clear_huge(pud_t *pud);
758 int pmd_clear_huge(pmd_t *pmd);
759 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
760 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
764 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
768 static inline int pud_clear_huge(pud_t *pud)
772 static inline int pmd_clear_huge(pmd_t *pmd)
776 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
778 #endif /* !__ASSEMBLY__ */
780 #ifndef io_remap_pfn_range
781 #define io_remap_pfn_range remap_pfn_range
784 #endif /* _ASM_GENERIC_PGTABLE_H */