1 #ifndef _ASM_GENERIC_PGTABLE_H
2 #define _ASM_GENERIC_PGTABLE_H
7 #include <linux/mm_types.h>
10 #if 4 - defined(__PAGETABLE_PUD_FOLDED) - defined(__PAGETABLE_PMD_FOLDED) != \
12 #error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{PUD,PMD}_FOLDED
16 * On almost all architectures and configurations, 0 can be used as the
17 * upper ceiling to free_pgtables(): on many architectures it has the same
18 * effect as using TASK_SIZE. However, there is one configuration which
19 * must impose a more careful limit, to avoid freeing kernel pgtables.
21 #ifndef USER_PGTABLES_CEILING
22 #define USER_PGTABLES_CEILING 0UL
25 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
26 extern int ptep_set_access_flags(struct vm_area_struct *vma,
27 unsigned long address, pte_t *ptep,
28 pte_t entry, int dirty);
31 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
32 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
33 unsigned long address, pmd_t *pmdp,
34 pmd_t entry, int dirty);
37 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
38 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
39 unsigned long address,
47 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
52 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
53 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
54 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
55 unsigned long address,
63 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
66 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
67 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
68 unsigned long address,
74 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
77 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
78 int ptep_clear_flush_young(struct vm_area_struct *vma,
79 unsigned long address, pte_t *ptep);
82 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
83 int pmdp_clear_flush_young(struct vm_area_struct *vma,
84 unsigned long address, pmd_t *pmdp);
87 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
88 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
89 unsigned long address,
93 pte_clear(mm, address, ptep);
98 #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR
99 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
100 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
101 unsigned long address,
108 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
111 #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR_FULL
112 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
113 static inline pmd_t pmdp_get_and_clear_full(struct mm_struct *mm,
114 unsigned long address, pmd_t *pmdp,
117 return pmdp_get_and_clear(mm, address, pmdp);
119 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
122 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
123 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
124 unsigned long address, pte_t *ptep,
128 pte = ptep_get_and_clear(mm, address, ptep);
134 * Some architectures may be able to avoid expensive synchronization
135 * primitives when modifications are made to PTE's which are already
136 * not present, or in the process of an address space destruction.
138 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
139 static inline void pte_clear_not_present_full(struct mm_struct *mm,
140 unsigned long address,
144 pte_clear(mm, address, ptep);
148 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
149 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
150 unsigned long address,
154 #ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
155 extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
156 unsigned long address,
160 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
162 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
164 pte_t old_pte = *ptep;
165 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
169 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
170 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
171 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
172 unsigned long address, pmd_t *pmdp)
174 pmd_t old_pmd = *pmdp;
175 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
177 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
178 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
179 unsigned long address, pmd_t *pmdp)
183 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
186 #ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
187 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
188 unsigned long address, pmd_t *pmdp);
191 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
192 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
196 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
197 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
200 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
201 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
205 #ifndef __HAVE_ARCH_PTE_SAME
206 static inline int pte_same(pte_t pte_a, pte_t pte_b)
208 return pte_val(pte_a) == pte_val(pte_b);
212 #ifndef __HAVE_ARCH_PTE_UNUSED
214 * Some architectures provide facilities to virtualization guests
215 * so that they can flag allocated pages as unused. This allows the
216 * host to transparently reclaim unused pages. This function returns
217 * whether the pte's page is unused.
219 static inline int pte_unused(pte_t pte)
225 #ifndef __HAVE_ARCH_PMD_SAME
226 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
227 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
229 return pmd_val(pmd_a) == pmd_val(pmd_b);
231 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
232 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
237 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
240 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
241 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
244 #ifndef __HAVE_ARCH_MOVE_PTE
245 #define move_pte(pte, prot, old_addr, new_addr) (pte)
248 #ifndef pte_accessible
249 # define pte_accessible(mm, pte) ((void)(pte), 1)
252 #ifndef flush_tlb_fix_spurious_fault
253 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
256 #ifndef pgprot_noncached
257 #define pgprot_noncached(prot) (prot)
260 #ifndef pgprot_writecombine
261 #define pgprot_writecombine pgprot_noncached
264 #ifndef pgprot_device
265 #define pgprot_device pgprot_noncached
268 #ifndef pgprot_modify
269 #define pgprot_modify pgprot_modify
270 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
272 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
273 newprot = pgprot_noncached(newprot);
274 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
275 newprot = pgprot_writecombine(newprot);
276 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
277 newprot = pgprot_device(newprot);
283 * When walking page tables, get the address of the next boundary,
284 * or the end address of the range if that comes earlier. Although no
285 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
288 #define pgd_addr_end(addr, end) \
289 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
290 (__boundary - 1 < (end) - 1)? __boundary: (end); \
294 #define pud_addr_end(addr, end) \
295 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
296 (__boundary - 1 < (end) - 1)? __boundary: (end); \
301 #define pmd_addr_end(addr, end) \
302 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
303 (__boundary - 1 < (end) - 1)? __boundary: (end); \
308 * When walking page tables, we usually want to skip any p?d_none entries;
309 * and any p?d_bad entries - reporting the error before resetting to none.
310 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
312 void pgd_clear_bad(pgd_t *);
313 void pud_clear_bad(pud_t *);
314 void pmd_clear_bad(pmd_t *);
316 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
320 if (unlikely(pgd_bad(*pgd))) {
327 static inline int pud_none_or_clear_bad(pud_t *pud)
331 if (unlikely(pud_bad(*pud))) {
338 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
342 if (unlikely(pmd_bad(*pmd))) {
349 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
354 * Get the current pte state, but zero it out to make it
355 * non-present, preventing the hardware from asynchronously
358 return ptep_get_and_clear(mm, addr, ptep);
361 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
363 pte_t *ptep, pte_t pte)
366 * The pte is non-present, so there's no hardware state to
369 set_pte_at(mm, addr, ptep, pte);
372 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
374 * Start a pte protection read-modify-write transaction, which
375 * protects against asynchronous hardware modifications to the pte.
376 * The intention is not to prevent the hardware from making pte
377 * updates, but to prevent any updates it may make from being lost.
379 * This does not protect against other software modifications of the
380 * pte; the appropriate pte lock must be held over the transation.
382 * Note that this interface is intended to be batchable, meaning that
383 * ptep_modify_prot_commit may not actually update the pte, but merely
384 * queue the update to be done at some later time. The update must be
385 * actually committed before the pte lock is released, however.
387 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
391 return __ptep_modify_prot_start(mm, addr, ptep);
395 * Commit an update to a pte, leaving any hardware-controlled bits in
396 * the PTE unmodified.
398 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
400 pte_t *ptep, pte_t pte)
402 __ptep_modify_prot_commit(mm, addr, ptep, pte);
404 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
405 #endif /* CONFIG_MMU */
408 * A facility to provide lazy MMU batching. This allows PTE updates and
409 * page invalidations to be delayed until a call to leave lazy MMU mode
410 * is issued. Some architectures may benefit from doing this, and it is
411 * beneficial for both shadow and direct mode hypervisors, which may batch
412 * the PTE updates which happen during this window. Note that using this
413 * interface requires that read hazards be removed from the code. A read
414 * hazard could result in the direct mode hypervisor case, since the actual
415 * write to the page tables may not yet have taken place, so reads though
416 * a raw PTE pointer after it has been modified are not guaranteed to be
417 * up to date. This mode can only be entered and left under the protection of
418 * the page table locks for all page tables which may be modified. In the UP
419 * case, this is required so that preemption is disabled, and in the SMP case,
420 * it must synchronize the delayed page table writes properly on other CPUs.
422 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
423 #define arch_enter_lazy_mmu_mode() do {} while (0)
424 #define arch_leave_lazy_mmu_mode() do {} while (0)
425 #define arch_flush_lazy_mmu_mode() do {} while (0)
429 * A facility to provide batching of the reload of page tables and
430 * other process state with the actual context switch code for
431 * paravirtualized guests. By convention, only one of the batched
432 * update (lazy) modes (CPU, MMU) should be active at any given time,
433 * entry should never be nested, and entry and exits should always be
434 * paired. This is for sanity of maintaining and reasoning about the
435 * kernel code. In this case, the exit (end of the context switch) is
436 * in architecture-specific code, and so doesn't need a generic
439 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
440 #define arch_start_context_switch(prev) do {} while (0)
443 #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
444 static inline int pte_soft_dirty(pte_t pte)
449 static inline int pmd_soft_dirty(pmd_t pmd)
454 static inline pte_t pte_mksoft_dirty(pte_t pte)
459 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
464 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
469 static inline int pte_swp_soft_dirty(pte_t pte)
474 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
480 #ifndef __HAVE_PFNMAP_TRACKING
482 * Interfaces that can be used by architecture code to keep track of
483 * memory type of pfn mappings specified by the remap_pfn_range,
488 * track_pfn_remap is called when a _new_ pfn mapping is being established
489 * by remap_pfn_range() for physical range indicated by pfn and size.
491 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
492 unsigned long pfn, unsigned long addr,
499 * track_pfn_insert is called when a _new_ single pfn is established
500 * by vm_insert_pfn().
502 static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
509 * track_pfn_copy is called when vma that is covering the pfnmap gets
510 * copied through copy_page_range().
512 static inline int track_pfn_copy(struct vm_area_struct *vma)
518 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
519 * untrack can be called for a specific region indicated by pfn and size or
520 * can be for the entire vma (in which case pfn, size are zero).
522 static inline void untrack_pfn(struct vm_area_struct *vma,
523 unsigned long pfn, unsigned long size)
527 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
528 unsigned long pfn, unsigned long addr,
530 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
532 extern int track_pfn_copy(struct vm_area_struct *vma);
533 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
537 #ifdef __HAVE_COLOR_ZERO_PAGE
538 static inline int is_zero_pfn(unsigned long pfn)
540 extern unsigned long zero_pfn;
541 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
542 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
545 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
548 static inline int is_zero_pfn(unsigned long pfn)
550 extern unsigned long zero_pfn;
551 return pfn == zero_pfn;
554 static inline unsigned long my_zero_pfn(unsigned long addr)
556 extern unsigned long zero_pfn;
563 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
564 static inline int pmd_trans_huge(pmd_t pmd)
568 static inline int pmd_trans_splitting(pmd_t pmd)
572 #ifndef __HAVE_ARCH_PMD_WRITE
573 static inline int pmd_write(pmd_t pmd)
578 #endif /* __HAVE_ARCH_PMD_WRITE */
579 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
581 #ifndef pmd_read_atomic
582 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
585 * Depend on compiler for an atomic pmd read. NOTE: this is
586 * only going to work, if the pmdval_t isn't larger than
593 #ifndef pmd_move_must_withdraw
594 static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
595 spinlock_t *old_pmd_ptl)
598 * With split pmd lock we also need to move preallocated
599 * PTE page table if new_pmd is on different PMD page table.
601 return new_pmd_ptl != old_pmd_ptl;
606 * This function is meant to be used by sites walking pagetables with
607 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
608 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
609 * into a null pmd and the transhuge page fault can convert a null pmd
610 * into an hugepmd or into a regular pmd (if the hugepage allocation
611 * fails). While holding the mmap_sem in read mode the pmd becomes
612 * stable and stops changing under us only if it's not null and not a
613 * transhuge pmd. When those races occurs and this function makes a
614 * difference vs the standard pmd_none_or_clear_bad, the result is
615 * undefined so behaving like if the pmd was none is safe (because it
616 * can return none anyway). The compiler level barrier() is critically
617 * important to compute the two checks atomically on the same pmdval.
619 * For 32bit kernels with a 64bit large pmd_t this automatically takes
620 * care of reading the pmd atomically to avoid SMP race conditions
621 * against pmd_populate() when the mmap_sem is hold for reading by the
622 * caller (a special atomic read not done by "gcc" as in the generic
623 * version above, is also needed when THP is disabled because the page
624 * fault can populate the pmd from under us).
626 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
628 pmd_t pmdval = pmd_read_atomic(pmd);
630 * The barrier will stabilize the pmdval in a register or on
631 * the stack so that it will stop changing under the code.
633 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
634 * pmd_read_atomic is allowed to return a not atomic pmdval
635 * (for example pointing to an hugepage that has never been
636 * mapped in the pmd). The below checks will only care about
637 * the low part of the pmd with 32bit PAE x86 anyway, with the
638 * exception of pmd_none(). So the important thing is that if
639 * the low part of the pmd is found null, the high part will
640 * be also null or the pmd_none() check below would be
643 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
646 if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
648 if (unlikely(pmd_bad(pmdval))) {
656 * This is a noop if Transparent Hugepage Support is not built into
657 * the kernel. Otherwise it is equivalent to
658 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
659 * places that already verified the pmd is not none and they want to
660 * walk ptes while holding the mmap sem in read mode (write mode don't
661 * need this). If THP is not enabled, the pmd can't go away under the
662 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
663 * run a pmd_trans_unstable before walking the ptes after
664 * split_huge_page_pmd returns (because it may have run when the pmd
665 * become null, but then a page fault can map in a THP and not a
668 static inline int pmd_trans_unstable(pmd_t *pmd)
670 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
671 return pmd_none_or_trans_huge_or_clear_bad(pmd);
677 #ifndef CONFIG_NUMA_BALANCING
679 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
680 * the only case the kernel cares is for NUMA balancing and is only ever set
681 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
682 * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
683 * is the responsibility of the caller to distinguish between PROT_NONE
684 * protections and NUMA hinting fault protections.
686 static inline int pte_protnone(pte_t pte)
691 static inline int pmd_protnone(pmd_t pmd)
695 #endif /* CONFIG_NUMA_BALANCING */
697 #endif /* CONFIG_MMU */
699 #endif /* !__ASSEMBLY__ */
701 #ifndef io_remap_pfn_range
702 #define io_remap_pfn_range remap_pfn_range
705 #endif /* _ASM_GENERIC_PGTABLE_H */