1 #ifndef _ASM_GENERIC_PGTABLE_H
2 #define _ASM_GENERIC_PGTABLE_H
7 #include <linux/mm_types.h>
11 * On almost all architectures and configurations, 0 can be used as the
12 * upper ceiling to free_pgtables(): on many architectures it has the same
13 * effect as using TASK_SIZE. However, there is one configuration which
14 * must impose a more careful limit, to avoid freeing kernel pgtables.
16 #ifndef USER_PGTABLES_CEILING
17 #define USER_PGTABLES_CEILING 0UL
20 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
21 extern int ptep_set_access_flags(struct vm_area_struct
*vma
,
22 unsigned long address
, pte_t
*ptep
,
23 pte_t entry
, int dirty
);
26 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
27 extern int pmdp_set_access_flags(struct vm_area_struct
*vma
,
28 unsigned long address
, pmd_t
*pmdp
,
29 pmd_t entry
, int dirty
);
32 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
33 static inline int ptep_test_and_clear_young(struct vm_area_struct
*vma
,
34 unsigned long address
,
42 set_pte_at(vma
->vm_mm
, address
, ptep
, pte_mkold(pte
));
47 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
48 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
49 static inline int pmdp_test_and_clear_young(struct vm_area_struct
*vma
,
50 unsigned long address
,
58 set_pmd_at(vma
->vm_mm
, address
, pmdp
, pmd_mkold(pmd
));
61 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
62 static inline int pmdp_test_and_clear_young(struct vm_area_struct
*vma
,
63 unsigned long address
,
69 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
72 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
73 int ptep_clear_flush_young(struct vm_area_struct
*vma
,
74 unsigned long address
, pte_t
*ptep
);
77 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
78 int pmdp_clear_flush_young(struct vm_area_struct
*vma
,
79 unsigned long address
, pmd_t
*pmdp
);
82 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
83 static inline pte_t
ptep_get_and_clear(struct mm_struct
*mm
,
84 unsigned long address
,
88 pte_clear(mm
, address
, ptep
);
93 #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR
94 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
95 static inline pmd_t
pmdp_get_and_clear(struct mm_struct
*mm
,
96 unsigned long address
,
103 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
106 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
107 static inline pte_t
ptep_get_and_clear_full(struct mm_struct
*mm
,
108 unsigned long address
, pte_t
*ptep
,
112 pte
= ptep_get_and_clear(mm
, address
, ptep
);
118 * Some architectures may be able to avoid expensive synchronization
119 * primitives when modifications are made to PTE's which are already
120 * not present, or in the process of an address space destruction.
122 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
123 static inline void pte_clear_not_present_full(struct mm_struct
*mm
,
124 unsigned long address
,
128 pte_clear(mm
, address
, ptep
);
132 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
133 extern pte_t
ptep_clear_flush(struct vm_area_struct
*vma
,
134 unsigned long address
,
138 #ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
139 extern pmd_t
pmdp_clear_flush(struct vm_area_struct
*vma
,
140 unsigned long address
,
144 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
146 static inline void ptep_set_wrprotect(struct mm_struct
*mm
, unsigned long address
, pte_t
*ptep
)
148 pte_t old_pte
= *ptep
;
149 set_pte_at(mm
, address
, ptep
, pte_wrprotect(old_pte
));
153 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
154 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
155 static inline void pmdp_set_wrprotect(struct mm_struct
*mm
,
156 unsigned long address
, pmd_t
*pmdp
)
158 pmd_t old_pmd
= *pmdp
;
159 set_pmd_at(mm
, address
, pmdp
, pmd_wrprotect(old_pmd
));
161 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
162 static inline void pmdp_set_wrprotect(struct mm_struct
*mm
,
163 unsigned long address
, pmd_t
*pmdp
)
167 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
170 #ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
171 extern void pmdp_splitting_flush(struct vm_area_struct
*vma
,
172 unsigned long address
, pmd_t
*pmdp
);
175 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
176 extern void pgtable_trans_huge_deposit(struct mm_struct
*mm
, pmd_t
*pmdp
,
180 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
181 extern pgtable_t
pgtable_trans_huge_withdraw(struct mm_struct
*mm
, pmd_t
*pmdp
);
184 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
185 extern void pmdp_invalidate(struct vm_area_struct
*vma
, unsigned long address
,
189 #ifndef __HAVE_ARCH_PTE_SAME
190 static inline int pte_same(pte_t pte_a
, pte_t pte_b
)
192 return pte_val(pte_a
) == pte_val(pte_b
);
196 #ifndef __HAVE_ARCH_PMD_SAME
197 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
198 static inline int pmd_same(pmd_t pmd_a
, pmd_t pmd_b
)
200 return pmd_val(pmd_a
) == pmd_val(pmd_b
);
202 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
203 static inline int pmd_same(pmd_t pmd_a
, pmd_t pmd_b
)
208 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
211 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
212 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
215 #ifndef __HAVE_ARCH_MOVE_PTE
216 #define move_pte(pte, prot, old_addr, new_addr) (pte)
219 #ifndef pte_accessible
220 # define pte_accessible(pte) ((void)(pte),1)
223 #ifndef flush_tlb_fix_spurious_fault
224 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
227 #ifndef pgprot_noncached
228 #define pgprot_noncached(prot) (prot)
231 #ifndef pgprot_writecombine
232 #define pgprot_writecombine pgprot_noncached
236 * When walking page tables, get the address of the next boundary,
237 * or the end address of the range if that comes earlier. Although no
238 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
241 #define pgd_addr_end(addr, end) \
242 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
243 (__boundary - 1 < (end) - 1)? __boundary: (end); \
247 #define pud_addr_end(addr, end) \
248 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
249 (__boundary - 1 < (end) - 1)? __boundary: (end); \
254 #define pmd_addr_end(addr, end) \
255 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
256 (__boundary - 1 < (end) - 1)? __boundary: (end); \
261 * When walking page tables, we usually want to skip any p?d_none entries;
262 * and any p?d_bad entries - reporting the error before resetting to none.
263 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
265 void pgd_clear_bad(pgd_t
*);
266 void pud_clear_bad(pud_t
*);
267 void pmd_clear_bad(pmd_t
*);
269 static inline int pgd_none_or_clear_bad(pgd_t
*pgd
)
273 if (unlikely(pgd_bad(*pgd
))) {
280 static inline int pud_none_or_clear_bad(pud_t
*pud
)
284 if (unlikely(pud_bad(*pud
))) {
291 static inline int pmd_none_or_clear_bad(pmd_t
*pmd
)
295 if (unlikely(pmd_bad(*pmd
))) {
302 static inline pte_t
__ptep_modify_prot_start(struct mm_struct
*mm
,
307 * Get the current pte state, but zero it out to make it
308 * non-present, preventing the hardware from asynchronously
311 return ptep_get_and_clear(mm
, addr
, ptep
);
314 static inline void __ptep_modify_prot_commit(struct mm_struct
*mm
,
316 pte_t
*ptep
, pte_t pte
)
319 * The pte is non-present, so there's no hardware state to
322 set_pte_at(mm
, addr
, ptep
, pte
);
325 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
327 * Start a pte protection read-modify-write transaction, which
328 * protects against asynchronous hardware modifications to the pte.
329 * The intention is not to prevent the hardware from making pte
330 * updates, but to prevent any updates it may make from being lost.
332 * This does not protect against other software modifications of the
333 * pte; the appropriate pte lock must be held over the transation.
335 * Note that this interface is intended to be batchable, meaning that
336 * ptep_modify_prot_commit may not actually update the pte, but merely
337 * queue the update to be done at some later time. The update must be
338 * actually committed before the pte lock is released, however.
340 static inline pte_t
ptep_modify_prot_start(struct mm_struct
*mm
,
344 return __ptep_modify_prot_start(mm
, addr
, ptep
);
348 * Commit an update to a pte, leaving any hardware-controlled bits in
349 * the PTE unmodified.
351 static inline void ptep_modify_prot_commit(struct mm_struct
*mm
,
353 pte_t
*ptep
, pte_t pte
)
355 __ptep_modify_prot_commit(mm
, addr
, ptep
, pte
);
357 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
358 #endif /* CONFIG_MMU */
361 * A facility to provide lazy MMU batching. This allows PTE updates and
362 * page invalidations to be delayed until a call to leave lazy MMU mode
363 * is issued. Some architectures may benefit from doing this, and it is
364 * beneficial for both shadow and direct mode hypervisors, which may batch
365 * the PTE updates which happen during this window. Note that using this
366 * interface requires that read hazards be removed from the code. A read
367 * hazard could result in the direct mode hypervisor case, since the actual
368 * write to the page tables may not yet have taken place, so reads though
369 * a raw PTE pointer after it has been modified are not guaranteed to be
370 * up to date. This mode can only be entered and left under the protection of
371 * the page table locks for all page tables which may be modified. In the UP
372 * case, this is required so that preemption is disabled, and in the SMP case,
373 * it must synchronize the delayed page table writes properly on other CPUs.
375 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
376 #define arch_enter_lazy_mmu_mode() do {} while (0)
377 #define arch_leave_lazy_mmu_mode() do {} while (0)
378 #define arch_flush_lazy_mmu_mode() do {} while (0)
382 * A facility to provide batching of the reload of page tables and
383 * other process state with the actual context switch code for
384 * paravirtualized guests. By convention, only one of the batched
385 * update (lazy) modes (CPU, MMU) should be active at any given time,
386 * entry should never be nested, and entry and exits should always be
387 * paired. This is for sanity of maintaining and reasoning about the
388 * kernel code. In this case, the exit (end of the context switch) is
389 * in architecture-specific code, and so doesn't need a generic
392 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
393 #define arch_start_context_switch(prev) do {} while (0)
396 #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
397 static inline int pte_soft_dirty(pte_t pte
)
402 static inline int pmd_soft_dirty(pmd_t pmd
)
407 static inline pte_t
pte_mksoft_dirty(pte_t pte
)
412 static inline pmd_t
pmd_mksoft_dirty(pmd_t pmd
)
417 static inline pte_t
pte_swp_mksoft_dirty(pte_t pte
)
422 static inline int pte_swp_soft_dirty(pte_t pte
)
427 static inline pte_t
pte_swp_clear_soft_dirty(pte_t pte
)
432 static inline pte_t
pte_file_clear_soft_dirty(pte_t pte
)
437 static inline pte_t
pte_file_mksoft_dirty(pte_t pte
)
442 static inline int pte_file_soft_dirty(pte_t pte
)
448 #ifndef __HAVE_PFNMAP_TRACKING
450 * Interfaces that can be used by architecture code to keep track of
451 * memory type of pfn mappings specified by the remap_pfn_range,
456 * track_pfn_remap is called when a _new_ pfn mapping is being established
457 * by remap_pfn_range() for physical range indicated by pfn and size.
459 static inline int track_pfn_remap(struct vm_area_struct
*vma
, pgprot_t
*prot
,
460 unsigned long pfn
, unsigned long addr
,
467 * track_pfn_insert is called when a _new_ single pfn is established
468 * by vm_insert_pfn().
470 static inline int track_pfn_insert(struct vm_area_struct
*vma
, pgprot_t
*prot
,
477 * track_pfn_copy is called when vma that is covering the pfnmap gets
478 * copied through copy_page_range().
480 static inline int track_pfn_copy(struct vm_area_struct
*vma
)
486 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
487 * untrack can be called for a specific region indicated by pfn and size or
488 * can be for the entire vma (in which case pfn, size are zero).
490 static inline void untrack_pfn(struct vm_area_struct
*vma
,
491 unsigned long pfn
, unsigned long size
)
495 extern int track_pfn_remap(struct vm_area_struct
*vma
, pgprot_t
*prot
,
496 unsigned long pfn
, unsigned long addr
,
498 extern int track_pfn_insert(struct vm_area_struct
*vma
, pgprot_t
*prot
,
500 extern int track_pfn_copy(struct vm_area_struct
*vma
);
501 extern void untrack_pfn(struct vm_area_struct
*vma
, unsigned long pfn
,
505 #ifdef __HAVE_COLOR_ZERO_PAGE
506 static inline int is_zero_pfn(unsigned long pfn
)
508 extern unsigned long zero_pfn
;
509 unsigned long offset_from_zero_pfn
= pfn
- zero_pfn
;
510 return offset_from_zero_pfn
<= (zero_page_mask
>> PAGE_SHIFT
);
513 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
516 static inline int is_zero_pfn(unsigned long pfn
)
518 extern unsigned long zero_pfn
;
519 return pfn
== zero_pfn
;
522 static inline unsigned long my_zero_pfn(unsigned long addr
)
524 extern unsigned long zero_pfn
;
531 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
532 static inline int pmd_trans_huge(pmd_t pmd
)
536 static inline int pmd_trans_splitting(pmd_t pmd
)
540 #ifndef __HAVE_ARCH_PMD_WRITE
541 static inline int pmd_write(pmd_t pmd
)
546 #endif /* __HAVE_ARCH_PMD_WRITE */
547 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
549 #ifndef pmd_read_atomic
550 static inline pmd_t
pmd_read_atomic(pmd_t
*pmdp
)
553 * Depend on compiler for an atomic pmd read. NOTE: this is
554 * only going to work, if the pmdval_t isn't larger than
562 * This function is meant to be used by sites walking pagetables with
563 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
564 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
565 * into a null pmd and the transhuge page fault can convert a null pmd
566 * into an hugepmd or into a regular pmd (if the hugepage allocation
567 * fails). While holding the mmap_sem in read mode the pmd becomes
568 * stable and stops changing under us only if it's not null and not a
569 * transhuge pmd. When those races occurs and this function makes a
570 * difference vs the standard pmd_none_or_clear_bad, the result is
571 * undefined so behaving like if the pmd was none is safe (because it
572 * can return none anyway). The compiler level barrier() is critically
573 * important to compute the two checks atomically on the same pmdval.
575 * For 32bit kernels with a 64bit large pmd_t this automatically takes
576 * care of reading the pmd atomically to avoid SMP race conditions
577 * against pmd_populate() when the mmap_sem is hold for reading by the
578 * caller (a special atomic read not done by "gcc" as in the generic
579 * version above, is also needed when THP is disabled because the page
580 * fault can populate the pmd from under us).
582 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t
*pmd
)
584 pmd_t pmdval
= pmd_read_atomic(pmd
);
586 * The barrier will stabilize the pmdval in a register or on
587 * the stack so that it will stop changing under the code.
589 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
590 * pmd_read_atomic is allowed to return a not atomic pmdval
591 * (for example pointing to an hugepage that has never been
592 * mapped in the pmd). The below checks will only care about
593 * the low part of the pmd with 32bit PAE x86 anyway, with the
594 * exception of pmd_none(). So the important thing is that if
595 * the low part of the pmd is found null, the high part will
596 * be also null or the pmd_none() check below would be
599 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
602 if (pmd_none(pmdval
))
604 if (unlikely(pmd_bad(pmdval
))) {
605 if (!pmd_trans_huge(pmdval
))
613 * This is a noop if Transparent Hugepage Support is not built into
614 * the kernel. Otherwise it is equivalent to
615 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
616 * places that already verified the pmd is not none and they want to
617 * walk ptes while holding the mmap sem in read mode (write mode don't
618 * need this). If THP is not enabled, the pmd can't go away under the
619 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
620 * run a pmd_trans_unstable before walking the ptes after
621 * split_huge_page_pmd returns (because it may have run when the pmd
622 * become null, but then a page fault can map in a THP and not a
625 static inline int pmd_trans_unstable(pmd_t
*pmd
)
627 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
628 return pmd_none_or_trans_huge_or_clear_bad(pmd
);
634 #ifdef CONFIG_NUMA_BALANCING
635 #ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE
637 * _PAGE_NUMA works identical to _PAGE_PROTNONE (it's actually the
638 * same bit too). It's set only when _PAGE_PRESET is not set and it's
639 * never set if _PAGE_PRESENT is set.
641 * pte/pmd_present() returns true if pte/pmd_numa returns true. Page
642 * fault triggers on those regions if pte/pmd_numa returns true
643 * (because _PAGE_PRESENT is not set).
646 static inline int pte_numa(pte_t pte
)
648 return (pte_flags(pte
) &
649 (_PAGE_NUMA
|_PAGE_PRESENT
)) == _PAGE_NUMA
;
654 static inline int pmd_numa(pmd_t pmd
)
656 return (pmd_flags(pmd
) &
657 (_PAGE_NUMA
|_PAGE_PRESENT
)) == _PAGE_NUMA
;
662 * pte/pmd_mknuma sets the _PAGE_ACCESSED bitflag automatically
663 * because they're called by the NUMA hinting minor page fault. If we
664 * wouldn't set the _PAGE_ACCESSED bitflag here, the TLB miss handler
665 * would be forced to set it later while filling the TLB after we
666 * return to userland. That would trigger a second write to memory
667 * that we optimize away by setting _PAGE_ACCESSED here.
669 #ifndef pte_mknonnuma
670 static inline pte_t
pte_mknonnuma(pte_t pte
)
672 pte
= pte_clear_flags(pte
, _PAGE_NUMA
);
673 return pte_set_flags(pte
, _PAGE_PRESENT
|_PAGE_ACCESSED
);
677 #ifndef pmd_mknonnuma
678 static inline pmd_t
pmd_mknonnuma(pmd_t pmd
)
680 pmd
= pmd_clear_flags(pmd
, _PAGE_NUMA
);
681 return pmd_set_flags(pmd
, _PAGE_PRESENT
|_PAGE_ACCESSED
);
686 static inline pte_t
pte_mknuma(pte_t pte
)
688 pte
= pte_set_flags(pte
, _PAGE_NUMA
);
689 return pte_clear_flags(pte
, _PAGE_PRESENT
);
694 static inline pmd_t
pmd_mknuma(pmd_t pmd
)
696 pmd
= pmd_set_flags(pmd
, _PAGE_NUMA
);
697 return pmd_clear_flags(pmd
, _PAGE_PRESENT
);
701 extern int pte_numa(pte_t pte
);
702 extern int pmd_numa(pmd_t pmd
);
703 extern pte_t
pte_mknonnuma(pte_t pte
);
704 extern pmd_t
pmd_mknonnuma(pmd_t pmd
);
705 extern pte_t
pte_mknuma(pte_t pte
);
706 extern pmd_t
pmd_mknuma(pmd_t pmd
);
707 #endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */
709 static inline int pmd_numa(pmd_t pmd
)
714 static inline int pte_numa(pte_t pte
)
719 static inline pte_t
pte_mknonnuma(pte_t pte
)
724 static inline pmd_t
pmd_mknonnuma(pmd_t pmd
)
729 static inline pte_t
pte_mknuma(pte_t pte
)
734 static inline pmd_t
pmd_mknuma(pmd_t pmd
)
738 #endif /* CONFIG_NUMA_BALANCING */
740 #endif /* CONFIG_MMU */
742 #endif /* !__ASSEMBLY__ */
744 #ifndef io_remap_pfn_range
745 #define io_remap_pfn_range remap_pfn_range
748 #endif /* _ASM_GENERIC_PGTABLE_H */