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, pgtable_t pgtable);
179 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
180 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm);
183 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
184 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
188 #ifndef __HAVE_ARCH_PTE_SAME
189 static inline int pte_same(pte_t pte_a, pte_t pte_b)
191 return pte_val(pte_a) == pte_val(pte_b);
195 #ifndef __HAVE_ARCH_PMD_SAME
196 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
197 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
199 return pmd_val(pmd_a) == pmd_val(pmd_b);
201 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
202 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
207 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
210 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
211 #define page_test_and_clear_young(pfn) (0)
214 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
215 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
218 #ifndef __HAVE_ARCH_MOVE_PTE
219 #define move_pte(pte, prot, old_addr, new_addr) (pte)
222 #ifndef pte_accessible
223 # define pte_accessible(pte) ((void)(pte),1)
226 #ifndef flush_tlb_fix_spurious_fault
227 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
230 #ifndef pgprot_noncached
231 #define pgprot_noncached(prot) (prot)
234 #ifndef pgprot_writecombine
235 #define pgprot_writecombine pgprot_noncached
239 * When walking page tables, get the address of the next boundary,
240 * or the end address of the range if that comes earlier. Although no
241 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
244 #define pgd_addr_end(addr, end) \
245 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
246 (__boundary - 1 < (end) - 1)? __boundary: (end); \
250 #define pud_addr_end(addr, end) \
251 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
252 (__boundary - 1 < (end) - 1)? __boundary: (end); \
257 #define pmd_addr_end(addr, end) \
258 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
259 (__boundary - 1 < (end) - 1)? __boundary: (end); \
264 * When walking page tables, we usually want to skip any p?d_none entries;
265 * and any p?d_bad entries - reporting the error before resetting to none.
266 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
268 void pgd_clear_bad(pgd_t *);
269 void pud_clear_bad(pud_t *);
270 void pmd_clear_bad(pmd_t *);
272 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
276 if (unlikely(pgd_bad(*pgd))) {
283 static inline int pud_none_or_clear_bad(pud_t *pud)
287 if (unlikely(pud_bad(*pud))) {
294 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
298 if (unlikely(pmd_bad(*pmd))) {
305 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
310 * Get the current pte state, but zero it out to make it
311 * non-present, preventing the hardware from asynchronously
314 return ptep_get_and_clear(mm, addr, ptep);
317 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
319 pte_t *ptep, pte_t pte)
322 * The pte is non-present, so there's no hardware state to
325 set_pte_at(mm, addr, ptep, pte);
328 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
330 * Start a pte protection read-modify-write transaction, which
331 * protects against asynchronous hardware modifications to the pte.
332 * The intention is not to prevent the hardware from making pte
333 * updates, but to prevent any updates it may make from being lost.
335 * This does not protect against other software modifications of the
336 * pte; the appropriate pte lock must be held over the transation.
338 * Note that this interface is intended to be batchable, meaning that
339 * ptep_modify_prot_commit may not actually update the pte, but merely
340 * queue the update to be done at some later time. The update must be
341 * actually committed before the pte lock is released, however.
343 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
347 return __ptep_modify_prot_start(mm, addr, ptep);
351 * Commit an update to a pte, leaving any hardware-controlled bits in
352 * the PTE unmodified.
354 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
356 pte_t *ptep, pte_t pte)
358 __ptep_modify_prot_commit(mm, addr, ptep, pte);
360 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
361 #endif /* CONFIG_MMU */
364 * A facility to provide lazy MMU batching. This allows PTE updates and
365 * page invalidations to be delayed until a call to leave lazy MMU mode
366 * is issued. Some architectures may benefit from doing this, and it is
367 * beneficial for both shadow and direct mode hypervisors, which may batch
368 * the PTE updates which happen during this window. Note that using this
369 * interface requires that read hazards be removed from the code. A read
370 * hazard could result in the direct mode hypervisor case, since the actual
371 * write to the page tables may not yet have taken place, so reads though
372 * a raw PTE pointer after it has been modified are not guaranteed to be
373 * up to date. This mode can only be entered and left under the protection of
374 * the page table locks for all page tables which may be modified. In the UP
375 * case, this is required so that preemption is disabled, and in the SMP case,
376 * it must synchronize the delayed page table writes properly on other CPUs.
378 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
379 #define arch_enter_lazy_mmu_mode() do {} while (0)
380 #define arch_leave_lazy_mmu_mode() do {} while (0)
381 #define arch_flush_lazy_mmu_mode() do {} while (0)
385 * A facility to provide batching of the reload of page tables and
386 * other process state with the actual context switch code for
387 * paravirtualized guests. By convention, only one of the batched
388 * update (lazy) modes (CPU, MMU) should be active at any given time,
389 * entry should never be nested, and entry and exits should always be
390 * paired. This is for sanity of maintaining and reasoning about the
391 * kernel code. In this case, the exit (end of the context switch) is
392 * in architecture-specific code, and so doesn't need a generic
395 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
396 #define arch_start_context_switch(prev) do {} while (0)
399 #ifndef __HAVE_PFNMAP_TRACKING
401 * Interfaces that can be used by architecture code to keep track of
402 * memory type of pfn mappings specified by the remap_pfn_range,
407 * track_pfn_remap is called when a _new_ pfn mapping is being established
408 * by remap_pfn_range() for physical range indicated by pfn and size.
410 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
411 unsigned long pfn, unsigned long addr,
418 * track_pfn_insert is called when a _new_ single pfn is established
419 * by vm_insert_pfn().
421 static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
428 * track_pfn_copy is called when vma that is covering the pfnmap gets
429 * copied through copy_page_range().
431 static inline int track_pfn_copy(struct vm_area_struct *vma)
437 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
438 * untrack can be called for a specific region indicated by pfn and size or
439 * can be for the entire vma (in which case pfn, size are zero).
441 static inline void untrack_pfn(struct vm_area_struct *vma,
442 unsigned long pfn, unsigned long size)
446 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
447 unsigned long pfn, unsigned long addr,
449 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
451 extern int track_pfn_copy(struct vm_area_struct *vma);
452 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
456 #ifdef __HAVE_COLOR_ZERO_PAGE
457 static inline int is_zero_pfn(unsigned long pfn)
459 extern unsigned long zero_pfn;
460 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
461 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
464 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
467 static inline int is_zero_pfn(unsigned long pfn)
469 extern unsigned long zero_pfn;
470 return pfn == zero_pfn;
473 static inline unsigned long my_zero_pfn(unsigned long addr)
475 extern unsigned long zero_pfn;
482 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
483 static inline int pmd_trans_huge(pmd_t pmd)
487 static inline int pmd_trans_splitting(pmd_t pmd)
491 #ifndef __HAVE_ARCH_PMD_WRITE
492 static inline int pmd_write(pmd_t pmd)
497 #endif /* __HAVE_ARCH_PMD_WRITE */
498 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
500 #ifndef pmd_read_atomic
501 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
504 * Depend on compiler for an atomic pmd read. NOTE: this is
505 * only going to work, if the pmdval_t isn't larger than
513 * This function is meant to be used by sites walking pagetables with
514 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
515 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
516 * into a null pmd and the transhuge page fault can convert a null pmd
517 * into an hugepmd or into a regular pmd (if the hugepage allocation
518 * fails). While holding the mmap_sem in read mode the pmd becomes
519 * stable and stops changing under us only if it's not null and not a
520 * transhuge pmd. When those races occurs and this function makes a
521 * difference vs the standard pmd_none_or_clear_bad, the result is
522 * undefined so behaving like if the pmd was none is safe (because it
523 * can return none anyway). The compiler level barrier() is critically
524 * important to compute the two checks atomically on the same pmdval.
526 * For 32bit kernels with a 64bit large pmd_t this automatically takes
527 * care of reading the pmd atomically to avoid SMP race conditions
528 * against pmd_populate() when the mmap_sem is hold for reading by the
529 * caller (a special atomic read not done by "gcc" as in the generic
530 * version above, is also needed when THP is disabled because the page
531 * fault can populate the pmd from under us).
533 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
535 pmd_t pmdval = pmd_read_atomic(pmd);
537 * The barrier will stabilize the pmdval in a register or on
538 * the stack so that it will stop changing under the code.
540 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
541 * pmd_read_atomic is allowed to return a not atomic pmdval
542 * (for example pointing to an hugepage that has never been
543 * mapped in the pmd). The below checks will only care about
544 * the low part of the pmd with 32bit PAE x86 anyway, with the
545 * exception of pmd_none(). So the important thing is that if
546 * the low part of the pmd is found null, the high part will
547 * be also null or the pmd_none() check below would be
550 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
553 if (pmd_none(pmdval))
555 if (unlikely(pmd_bad(pmdval))) {
556 if (!pmd_trans_huge(pmdval))
564 * This is a noop if Transparent Hugepage Support is not built into
565 * the kernel. Otherwise it is equivalent to
566 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
567 * places that already verified the pmd is not none and they want to
568 * walk ptes while holding the mmap sem in read mode (write mode don't
569 * need this). If THP is not enabled, the pmd can't go away under the
570 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
571 * run a pmd_trans_unstable before walking the ptes after
572 * split_huge_page_pmd returns (because it may have run when the pmd
573 * become null, but then a page fault can map in a THP and not a
576 static inline int pmd_trans_unstable(pmd_t *pmd)
578 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
579 return pmd_none_or_trans_huge_or_clear_bad(pmd);
585 #ifdef CONFIG_NUMA_BALANCING
586 #ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE
588 * _PAGE_NUMA works identical to _PAGE_PROTNONE (it's actually the
589 * same bit too). It's set only when _PAGE_PRESET is not set and it's
590 * never set if _PAGE_PRESENT is set.
592 * pte/pmd_present() returns true if pte/pmd_numa returns true. Page
593 * fault triggers on those regions if pte/pmd_numa returns true
594 * (because _PAGE_PRESENT is not set).
597 static inline int pte_numa(pte_t pte)
599 return (pte_flags(pte) &
600 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
605 static inline int pmd_numa(pmd_t pmd)
607 return (pmd_flags(pmd) &
608 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
613 * pte/pmd_mknuma sets the _PAGE_ACCESSED bitflag automatically
614 * because they're called by the NUMA hinting minor page fault. If we
615 * wouldn't set the _PAGE_ACCESSED bitflag here, the TLB miss handler
616 * would be forced to set it later while filling the TLB after we
617 * return to userland. That would trigger a second write to memory
618 * that we optimize away by setting _PAGE_ACCESSED here.
620 #ifndef pte_mknonnuma
621 static inline pte_t pte_mknonnuma(pte_t pte)
623 pte = pte_clear_flags(pte, _PAGE_NUMA);
624 return pte_set_flags(pte, _PAGE_PRESENT|_PAGE_ACCESSED);
628 #ifndef pmd_mknonnuma
629 static inline pmd_t pmd_mknonnuma(pmd_t pmd)
631 pmd = pmd_clear_flags(pmd, _PAGE_NUMA);
632 return pmd_set_flags(pmd, _PAGE_PRESENT|_PAGE_ACCESSED);
637 static inline pte_t pte_mknuma(pte_t pte)
639 pte = pte_set_flags(pte, _PAGE_NUMA);
640 return pte_clear_flags(pte, _PAGE_PRESENT);
645 static inline pmd_t pmd_mknuma(pmd_t pmd)
647 pmd = pmd_set_flags(pmd, _PAGE_NUMA);
648 return pmd_clear_flags(pmd, _PAGE_PRESENT);
652 extern int pte_numa(pte_t pte);
653 extern int pmd_numa(pmd_t pmd);
654 extern pte_t pte_mknonnuma(pte_t pte);
655 extern pmd_t pmd_mknonnuma(pmd_t pmd);
656 extern pte_t pte_mknuma(pte_t pte);
657 extern pmd_t pmd_mknuma(pmd_t pmd);
658 #endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */
660 static inline int pmd_numa(pmd_t pmd)
665 static inline int pte_numa(pte_t pte)
670 static inline pte_t pte_mknonnuma(pte_t pte)
675 static inline pmd_t pmd_mknonnuma(pmd_t pmd)
680 static inline pte_t pte_mknuma(pte_t pte)
685 static inline pmd_t pmd_mknuma(pmd_t pmd)
689 #endif /* CONFIG_NUMA_BALANCING */
691 #endif /* CONFIG_MMU */
693 #endif /* !__ASSEMBLY__ */
695 #endif /* _ASM_GENERIC_PGTABLE_H */