1 #ifndef _ASM_GENERIC_PGTABLE_H
2 #define _ASM_GENERIC_PGTABLE_H
7 #include <linux/mm_types.h>
9 #include <linux/errno.h>
11 #if 4 - defined(__PAGETABLE_PUD_FOLDED) - defined(__PAGETABLE_PMD_FOLDED) != \
13 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{PUD,PMD}_FOLDED
17 * On almost all architectures and configurations, 0 can be used as the
18 * upper ceiling to free_pgtables(): on many architectures it has the same
19 * effect as using TASK_SIZE. However, there is one configuration which
20 * must impose a more careful limit, to avoid freeing kernel pgtables.
22 #ifndef USER_PGTABLES_CEILING
23 #define USER_PGTABLES_CEILING 0UL
26 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
27 extern int ptep_set_access_flags(struct vm_area_struct *vma,
28 unsigned long address, pte_t *ptep,
29 pte_t entry, int dirty);
32 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
33 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
34 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
35 unsigned long address, pmd_t *pmdp,
36 pmd_t entry, int dirty);
38 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
39 unsigned long address, pmd_t *pmdp,
40 pmd_t entry, int dirty)
45 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
48 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
49 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
50 unsigned long address,
58 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
63 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
64 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
65 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
66 unsigned long address,
74 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
78 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
79 unsigned long address,
85 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
88 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
89 int ptep_clear_flush_young(struct vm_area_struct *vma,
90 unsigned long address, pte_t *ptep);
93 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
94 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
95 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
96 unsigned long address, pmd_t *pmdp);
99 * Despite relevant to THP only, this API is called from generic rmap code
100 * under PageTransHuge(), hence needs a dummy implementation for !THP
102 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
103 unsigned long address, pmd_t *pmdp)
108 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
111 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
112 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
113 unsigned long address,
117 pte_clear(mm, address, ptep);
122 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
123 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
124 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
125 unsigned long address,
132 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
135 #ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
136 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
137 static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
138 unsigned long address, pmd_t *pmdp,
141 return pmdp_huge_get_and_clear(mm, address, pmdp);
143 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
146 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
147 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
148 unsigned long address, pte_t *ptep,
152 pte = ptep_get_and_clear(mm, address, ptep);
158 * Some architectures may be able to avoid expensive synchronization
159 * primitives when modifications are made to PTE's which are already
160 * not present, or in the process of an address space destruction.
162 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
163 static inline void pte_clear_not_present_full(struct mm_struct *mm,
164 unsigned long address,
168 pte_clear(mm, address, ptep);
172 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
173 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
174 unsigned long address,
178 #ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
179 extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
180 unsigned long address,
184 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
186 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
188 pte_t old_pte = *ptep;
189 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
193 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
194 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
195 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
196 unsigned long address, pmd_t *pmdp)
198 pmd_t old_pmd = *pmdp;
199 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
202 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
203 unsigned long address, pmd_t *pmdp)
207 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
210 #ifndef pmdp_collapse_flush
211 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
212 extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
213 unsigned long address, pmd_t *pmdp);
215 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
216 unsigned long address,
222 #define pmdp_collapse_flush pmdp_collapse_flush
223 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
226 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
227 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
231 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
232 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
235 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
236 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
240 #ifndef __HAVE_ARCH_PTE_SAME
241 static inline int pte_same(pte_t pte_a, pte_t pte_b)
243 return pte_val(pte_a) == pte_val(pte_b);
247 #ifndef __HAVE_ARCH_PTE_UNUSED
249 * Some architectures provide facilities to virtualization guests
250 * so that they can flag allocated pages as unused. This allows the
251 * host to transparently reclaim unused pages. This function returns
252 * whether the pte's page is unused.
254 static inline int pte_unused(pte_t pte)
260 #ifndef __HAVE_ARCH_PMD_SAME
261 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
262 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
264 return pmd_val(pmd_a) == pmd_val(pmd_b);
266 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
267 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
272 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
275 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
276 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
279 #ifndef __HAVE_ARCH_MOVE_PTE
280 #define move_pte(pte, prot, old_addr, new_addr) (pte)
283 #ifndef pte_accessible
284 # define pte_accessible(mm, pte) ((void)(pte), 1)
287 #ifndef flush_tlb_fix_spurious_fault
288 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
291 #ifndef pgprot_noncached
292 #define pgprot_noncached(prot) (prot)
295 #ifndef pgprot_writecombine
296 #define pgprot_writecombine pgprot_noncached
299 #ifndef pgprot_writethrough
300 #define pgprot_writethrough pgprot_noncached
303 #ifndef pgprot_device
304 #define pgprot_device pgprot_noncached
307 #ifndef pgprot_modify
308 #define pgprot_modify pgprot_modify
309 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
311 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
312 newprot = pgprot_noncached(newprot);
313 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
314 newprot = pgprot_writecombine(newprot);
315 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
316 newprot = pgprot_device(newprot);
322 * When walking page tables, get the address of the next boundary,
323 * or the end address of the range if that comes earlier. Although no
324 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
327 #define pgd_addr_end(addr, end) \
328 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
329 (__boundary - 1 < (end) - 1)? __boundary: (end); \
333 #define pud_addr_end(addr, end) \
334 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
335 (__boundary - 1 < (end) - 1)? __boundary: (end); \
340 #define pmd_addr_end(addr, end) \
341 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
342 (__boundary - 1 < (end) - 1)? __boundary: (end); \
347 * When walking page tables, we usually want to skip any p?d_none entries;
348 * and any p?d_bad entries - reporting the error before resetting to none.
349 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
351 void pgd_clear_bad(pgd_t *);
352 void pud_clear_bad(pud_t *);
353 void pmd_clear_bad(pmd_t *);
355 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
359 if (unlikely(pgd_bad(*pgd))) {
366 static inline int pud_none_or_clear_bad(pud_t *pud)
370 if (unlikely(pud_bad(*pud))) {
377 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
381 if (unlikely(pmd_bad(*pmd))) {
388 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
393 * Get the current pte state, but zero it out to make it
394 * non-present, preventing the hardware from asynchronously
397 return ptep_get_and_clear(mm, addr, ptep);
400 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
402 pte_t *ptep, pte_t pte)
405 * The pte is non-present, so there's no hardware state to
408 set_pte_at(mm, addr, ptep, pte);
411 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
413 * Start a pte protection read-modify-write transaction, which
414 * protects against asynchronous hardware modifications to the pte.
415 * The intention is not to prevent the hardware from making pte
416 * updates, but to prevent any updates it may make from being lost.
418 * This does not protect against other software modifications of the
419 * pte; the appropriate pte lock must be held over the transation.
421 * Note that this interface is intended to be batchable, meaning that
422 * ptep_modify_prot_commit may not actually update the pte, but merely
423 * queue the update to be done at some later time. The update must be
424 * actually committed before the pte lock is released, however.
426 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
430 return __ptep_modify_prot_start(mm, addr, ptep);
434 * Commit an update to a pte, leaving any hardware-controlled bits in
435 * the PTE unmodified.
437 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
439 pte_t *ptep, pte_t pte)
441 __ptep_modify_prot_commit(mm, addr, ptep, pte);
443 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
444 #endif /* CONFIG_MMU */
447 * A facility to provide lazy MMU batching. This allows PTE updates and
448 * page invalidations to be delayed until a call to leave lazy MMU mode
449 * is issued. Some architectures may benefit from doing this, and it is
450 * beneficial for both shadow and direct mode hypervisors, which may batch
451 * the PTE updates which happen during this window. Note that using this
452 * interface requires that read hazards be removed from the code. A read
453 * hazard could result in the direct mode hypervisor case, since the actual
454 * write to the page tables may not yet have taken place, so reads though
455 * a raw PTE pointer after it has been modified are not guaranteed to be
456 * up to date. This mode can only be entered and left under the protection of
457 * the page table locks for all page tables which may be modified. In the UP
458 * case, this is required so that preemption is disabled, and in the SMP case,
459 * it must synchronize the delayed page table writes properly on other CPUs.
461 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
462 #define arch_enter_lazy_mmu_mode() do {} while (0)
463 #define arch_leave_lazy_mmu_mode() do {} while (0)
464 #define arch_flush_lazy_mmu_mode() do {} while (0)
468 * A facility to provide batching of the reload of page tables and
469 * other process state with the actual context switch code for
470 * paravirtualized guests. By convention, only one of the batched
471 * update (lazy) modes (CPU, MMU) should be active at any given time,
472 * entry should never be nested, and entry and exits should always be
473 * paired. This is for sanity of maintaining and reasoning about the
474 * kernel code. In this case, the exit (end of the context switch) is
475 * in architecture-specific code, and so doesn't need a generic
478 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
479 #define arch_start_context_switch(prev) do {} while (0)
482 #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
483 static inline int pte_soft_dirty(pte_t pte)
488 static inline int pmd_soft_dirty(pmd_t pmd)
493 static inline pte_t pte_mksoft_dirty(pte_t pte)
498 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
503 static inline pte_t pte_clear_soft_dirty(pte_t pte)
508 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
513 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
518 static inline int pte_swp_soft_dirty(pte_t pte)
523 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
529 #ifndef __HAVE_PFNMAP_TRACKING
531 * Interfaces that can be used by architecture code to keep track of
532 * memory type of pfn mappings specified by the remap_pfn_range,
537 * track_pfn_remap is called when a _new_ pfn mapping is being established
538 * by remap_pfn_range() for physical range indicated by pfn and size.
540 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
541 unsigned long pfn, unsigned long addr,
548 * track_pfn_insert is called when a _new_ single pfn is established
549 * by vm_insert_pfn().
551 static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
558 * track_pfn_copy is called when vma that is covering the pfnmap gets
559 * copied through copy_page_range().
561 static inline int track_pfn_copy(struct vm_area_struct *vma)
567 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
568 * untrack can be called for a specific region indicated by pfn and size or
569 * can be for the entire vma (in which case pfn, size are zero).
571 static inline void untrack_pfn(struct vm_area_struct *vma,
572 unsigned long pfn, unsigned long size)
576 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
577 unsigned long pfn, unsigned long addr,
579 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
581 extern int track_pfn_copy(struct vm_area_struct *vma);
582 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
586 #ifdef __HAVE_COLOR_ZERO_PAGE
587 static inline int is_zero_pfn(unsigned long pfn)
589 extern unsigned long zero_pfn;
590 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
591 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
594 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
597 static inline int is_zero_pfn(unsigned long pfn)
599 extern unsigned long zero_pfn;
600 return pfn == zero_pfn;
603 static inline unsigned long my_zero_pfn(unsigned long addr)
605 extern unsigned long zero_pfn;
612 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
613 static inline int pmd_trans_huge(pmd_t pmd)
617 #ifndef __HAVE_ARCH_PMD_WRITE
618 static inline int pmd_write(pmd_t pmd)
623 #endif /* __HAVE_ARCH_PMD_WRITE */
624 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
626 #ifndef pmd_read_atomic
627 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
630 * Depend on compiler for an atomic pmd read. NOTE: this is
631 * only going to work, if the pmdval_t isn't larger than
638 #ifndef pmd_move_must_withdraw
639 static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
640 spinlock_t *old_pmd_ptl)
643 * With split pmd lock we also need to move preallocated
644 * PTE page table if new_pmd is on different PMD page table.
646 return new_pmd_ptl != old_pmd_ptl;
651 * This function is meant to be used by sites walking pagetables with
652 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
653 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
654 * into a null pmd and the transhuge page fault can convert a null pmd
655 * into an hugepmd or into a regular pmd (if the hugepage allocation
656 * fails). While holding the mmap_sem in read mode the pmd becomes
657 * stable and stops changing under us only if it's not null and not a
658 * transhuge pmd. When those races occurs and this function makes a
659 * difference vs the standard pmd_none_or_clear_bad, the result is
660 * undefined so behaving like if the pmd was none is safe (because it
661 * can return none anyway). The compiler level barrier() is critically
662 * important to compute the two checks atomically on the same pmdval.
664 * For 32bit kernels with a 64bit large pmd_t this automatically takes
665 * care of reading the pmd atomically to avoid SMP race conditions
666 * against pmd_populate() when the mmap_sem is hold for reading by the
667 * caller (a special atomic read not done by "gcc" as in the generic
668 * version above, is also needed when THP is disabled because the page
669 * fault can populate the pmd from under us).
671 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
673 pmd_t pmdval = pmd_read_atomic(pmd);
675 * The barrier will stabilize the pmdval in a register or on
676 * the stack so that it will stop changing under the code.
678 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
679 * pmd_read_atomic is allowed to return a not atomic pmdval
680 * (for example pointing to an hugepage that has never been
681 * mapped in the pmd). The below checks will only care about
682 * the low part of the pmd with 32bit PAE x86 anyway, with the
683 * exception of pmd_none(). So the important thing is that if
684 * the low part of the pmd is found null, the high part will
685 * be also null or the pmd_none() check below would be
688 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
691 if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
693 if (unlikely(pmd_bad(pmdval))) {
701 * This is a noop if Transparent Hugepage Support is not built into
702 * the kernel. Otherwise it is equivalent to
703 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
704 * places that already verified the pmd is not none and they want to
705 * walk ptes while holding the mmap sem in read mode (write mode don't
706 * need this). If THP is not enabled, the pmd can't go away under the
707 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
708 * run a pmd_trans_unstable before walking the ptes after
709 * split_huge_page_pmd returns (because it may have run when the pmd
710 * become null, but then a page fault can map in a THP and not a
713 static inline int pmd_trans_unstable(pmd_t *pmd)
715 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
716 return pmd_none_or_trans_huge_or_clear_bad(pmd);
722 #ifndef CONFIG_NUMA_BALANCING
724 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
725 * the only case the kernel cares is for NUMA balancing and is only ever set
726 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
727 * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
728 * is the responsibility of the caller to distinguish between PROT_NONE
729 * protections and NUMA hinting fault protections.
731 static inline int pte_protnone(pte_t pte)
736 static inline int pmd_protnone(pmd_t pmd)
740 #endif /* CONFIG_NUMA_BALANCING */
742 #endif /* CONFIG_MMU */
744 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
745 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
746 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
747 int pud_clear_huge(pud_t *pud);
748 int pmd_clear_huge(pmd_t *pmd);
749 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
750 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
754 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
758 static inline int pud_clear_huge(pud_t *pud)
762 static inline int pmd_clear_huge(pmd_t *pmd)
766 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
768 #endif /* !__ASSEMBLY__ */
770 #ifndef io_remap_pfn_range
771 #define io_remap_pfn_range remap_pfn_range
774 #endif /* _ASM_GENERIC_PGTABLE_H */