1 #ifndef _ASM_GENERIC_PGTABLE_H
2 #define _ASM_GENERIC_PGTABLE_H
7 #include <linux/mm_types.h>
10 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
11 extern int ptep_set_access_flags(struct vm_area_struct *vma,
12 unsigned long address, pte_t *ptep,
13 pte_t entry, int dirty);
16 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
17 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
18 unsigned long address, pmd_t *pmdp,
19 pmd_t entry, int dirty);
22 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
23 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
24 unsigned long address,
32 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
37 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
38 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
39 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
40 unsigned long address,
48 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
51 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
52 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
53 unsigned long address,
59 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
62 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
63 int ptep_clear_flush_young(struct vm_area_struct *vma,
64 unsigned long address, pte_t *ptep);
67 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
68 int pmdp_clear_flush_young(struct vm_area_struct *vma,
69 unsigned long address, pmd_t *pmdp);
72 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
73 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
74 unsigned long address,
78 pte_clear(mm, address, ptep);
83 #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR
84 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
85 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
86 unsigned long address,
93 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
96 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
97 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
98 unsigned long address, pte_t *ptep,
102 pte = ptep_get_and_clear(mm, address, ptep);
108 * Some architectures may be able to avoid expensive synchronization
109 * primitives when modifications are made to PTE's which are already
110 * not present, or in the process of an address space destruction.
112 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
113 static inline void pte_clear_not_present_full(struct mm_struct *mm,
114 unsigned long address,
118 pte_clear(mm, address, ptep);
122 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
123 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
124 unsigned long address,
128 #ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
129 extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
130 unsigned long address,
134 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
136 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
138 pte_t old_pte = *ptep;
139 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
143 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
144 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
145 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
146 unsigned long address, pmd_t *pmdp)
148 pmd_t old_pmd = *pmdp;
149 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
151 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
152 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
153 unsigned long address, pmd_t *pmdp)
157 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
160 #ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
161 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
162 unsigned long address, pmd_t *pmdp);
165 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
166 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pgtable_t pgtable);
169 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
170 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm);
173 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
174 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
178 #ifndef __HAVE_ARCH_PTE_SAME
179 static inline int pte_same(pte_t pte_a, pte_t pte_b)
181 return pte_val(pte_a) == pte_val(pte_b);
185 #ifndef __HAVE_ARCH_PMD_SAME
186 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
187 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
189 return pmd_val(pmd_a) == pmd_val(pmd_b);
191 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
192 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
197 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
200 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
201 #define page_test_and_clear_young(pfn) (0)
204 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
205 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
208 #ifndef __HAVE_ARCH_MOVE_PTE
209 #define move_pte(pte, prot, old_addr, new_addr) (pte)
212 #ifndef pte_accessible
213 # define pte_accessible(pte) ((void)(pte),1)
216 #ifndef flush_tlb_fix_spurious_fault
217 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
220 #ifndef pgprot_noncached
221 #define pgprot_noncached(prot) (prot)
224 #ifndef pgprot_writecombine
225 #define pgprot_writecombine pgprot_noncached
229 * When walking page tables, get the address of the next boundary,
230 * or the end address of the range if that comes earlier. Although no
231 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
234 #define pgd_addr_end(addr, end) \
235 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
236 (__boundary - 1 < (end) - 1)? __boundary: (end); \
240 #define pud_addr_end(addr, end) \
241 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
242 (__boundary - 1 < (end) - 1)? __boundary: (end); \
247 #define pmd_addr_end(addr, end) \
248 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
249 (__boundary - 1 < (end) - 1)? __boundary: (end); \
254 * When walking page tables, we usually want to skip any p?d_none entries;
255 * and any p?d_bad entries - reporting the error before resetting to none.
256 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
258 void pgd_clear_bad(pgd_t *);
259 void pud_clear_bad(pud_t *);
260 void pmd_clear_bad(pmd_t *);
262 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
266 if (unlikely(pgd_bad(*pgd))) {
273 static inline int pud_none_or_clear_bad(pud_t *pud)
277 if (unlikely(pud_bad(*pud))) {
284 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
288 if (unlikely(pmd_bad(*pmd))) {
295 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
300 * Get the current pte state, but zero it out to make it
301 * non-present, preventing the hardware from asynchronously
304 return ptep_get_and_clear(mm, addr, ptep);
307 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
309 pte_t *ptep, pte_t pte)
312 * The pte is non-present, so there's no hardware state to
315 set_pte_at(mm, addr, ptep, pte);
318 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
320 * Start a pte protection read-modify-write transaction, which
321 * protects against asynchronous hardware modifications to the pte.
322 * The intention is not to prevent the hardware from making pte
323 * updates, but to prevent any updates it may make from being lost.
325 * This does not protect against other software modifications of the
326 * pte; the appropriate pte lock must be held over the transation.
328 * Note that this interface is intended to be batchable, meaning that
329 * ptep_modify_prot_commit may not actually update the pte, but merely
330 * queue the update to be done at some later time. The update must be
331 * actually committed before the pte lock is released, however.
333 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
337 return __ptep_modify_prot_start(mm, addr, ptep);
341 * Commit an update to a pte, leaving any hardware-controlled bits in
342 * the PTE unmodified.
344 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
346 pte_t *ptep, pte_t pte)
348 __ptep_modify_prot_commit(mm, addr, ptep, pte);
350 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
351 #endif /* CONFIG_MMU */
354 * A facility to provide lazy MMU batching. This allows PTE updates and
355 * page invalidations to be delayed until a call to leave lazy MMU mode
356 * is issued. Some architectures may benefit from doing this, and it is
357 * beneficial for both shadow and direct mode hypervisors, which may batch
358 * the PTE updates which happen during this window. Note that using this
359 * interface requires that read hazards be removed from the code. A read
360 * hazard could result in the direct mode hypervisor case, since the actual
361 * write to the page tables may not yet have taken place, so reads though
362 * a raw PTE pointer after it has been modified are not guaranteed to be
363 * up to date. This mode can only be entered and left under the protection of
364 * the page table locks for all page tables which may be modified. In the UP
365 * case, this is required so that preemption is disabled, and in the SMP case,
366 * it must synchronize the delayed page table writes properly on other CPUs.
368 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
369 #define arch_enter_lazy_mmu_mode() do {} while (0)
370 #define arch_leave_lazy_mmu_mode() do {} while (0)
371 #define arch_flush_lazy_mmu_mode() do {} while (0)
375 * A facility to provide batching of the reload of page tables and
376 * other process state with the actual context switch code for
377 * paravirtualized guests. By convention, only one of the batched
378 * update (lazy) modes (CPU, MMU) should be active at any given time,
379 * entry should never be nested, and entry and exits should always be
380 * paired. This is for sanity of maintaining and reasoning about the
381 * kernel code. In this case, the exit (end of the context switch) is
382 * in architecture-specific code, and so doesn't need a generic
385 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
386 #define arch_start_context_switch(prev) do {} while (0)
389 #ifndef __HAVE_PFNMAP_TRACKING
391 * Interfaces that can be used by architecture code to keep track of
392 * memory type of pfn mappings specified by the remap_pfn_range,
397 * track_pfn_remap is called when a _new_ pfn mapping is being established
398 * by remap_pfn_range() for physical range indicated by pfn and size.
400 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
401 unsigned long pfn, unsigned long addr,
408 * track_pfn_insert is called when a _new_ single pfn is established
409 * by vm_insert_pfn().
411 static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
418 * track_pfn_copy is called when vma that is covering the pfnmap gets
419 * copied through copy_page_range().
421 static inline int track_pfn_copy(struct vm_area_struct *vma)
427 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
428 * untrack can be called for a specific region indicated by pfn and size or
429 * can be for the entire vma (in which case pfn, size are zero).
431 static inline void untrack_pfn(struct vm_area_struct *vma,
432 unsigned long pfn, unsigned long size)
436 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
437 unsigned long pfn, unsigned long addr,
439 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
441 extern int track_pfn_copy(struct vm_area_struct *vma);
442 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
446 #ifdef __HAVE_COLOR_ZERO_PAGE
447 static inline int is_zero_pfn(unsigned long pfn)
449 extern unsigned long zero_pfn;
450 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
451 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
454 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
457 static inline int is_zero_pfn(unsigned long pfn)
459 extern unsigned long zero_pfn;
460 return pfn == zero_pfn;
463 static inline unsigned long my_zero_pfn(unsigned long addr)
465 extern unsigned long zero_pfn;
472 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
473 static inline int pmd_trans_huge(pmd_t pmd)
477 static inline int pmd_trans_splitting(pmd_t pmd)
481 #ifndef __HAVE_ARCH_PMD_WRITE
482 static inline int pmd_write(pmd_t pmd)
487 #endif /* __HAVE_ARCH_PMD_WRITE */
488 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
490 #ifndef pmd_read_atomic
491 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
494 * Depend on compiler for an atomic pmd read. NOTE: this is
495 * only going to work, if the pmdval_t isn't larger than
503 * This function is meant to be used by sites walking pagetables with
504 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
505 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
506 * into a null pmd and the transhuge page fault can convert a null pmd
507 * into an hugepmd or into a regular pmd (if the hugepage allocation
508 * fails). While holding the mmap_sem in read mode the pmd becomes
509 * stable and stops changing under us only if it's not null and not a
510 * transhuge pmd. When those races occurs and this function makes a
511 * difference vs the standard pmd_none_or_clear_bad, the result is
512 * undefined so behaving like if the pmd was none is safe (because it
513 * can return none anyway). The compiler level barrier() is critically
514 * important to compute the two checks atomically on the same pmdval.
516 * For 32bit kernels with a 64bit large pmd_t this automatically takes
517 * care of reading the pmd atomically to avoid SMP race conditions
518 * against pmd_populate() when the mmap_sem is hold for reading by the
519 * caller (a special atomic read not done by "gcc" as in the generic
520 * version above, is also needed when THP is disabled because the page
521 * fault can populate the pmd from under us).
523 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
525 pmd_t pmdval = pmd_read_atomic(pmd);
527 * The barrier will stabilize the pmdval in a register or on
528 * the stack so that it will stop changing under the code.
530 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
531 * pmd_read_atomic is allowed to return a not atomic pmdval
532 * (for example pointing to an hugepage that has never been
533 * mapped in the pmd). The below checks will only care about
534 * the low part of the pmd with 32bit PAE x86 anyway, with the
535 * exception of pmd_none(). So the important thing is that if
536 * the low part of the pmd is found null, the high part will
537 * be also null or the pmd_none() check below would be
540 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
543 if (pmd_none(pmdval))
545 if (unlikely(pmd_bad(pmdval))) {
546 if (!pmd_trans_huge(pmdval))
554 * This is a noop if Transparent Hugepage Support is not built into
555 * the kernel. Otherwise it is equivalent to
556 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
557 * places that already verified the pmd is not none and they want to
558 * walk ptes while holding the mmap sem in read mode (write mode don't
559 * need this). If THP is not enabled, the pmd can't go away under the
560 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
561 * run a pmd_trans_unstable before walking the ptes after
562 * split_huge_page_pmd returns (because it may have run when the pmd
563 * become null, but then a page fault can map in a THP and not a
566 static inline int pmd_trans_unstable(pmd_t *pmd)
568 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
569 return pmd_none_or_trans_huge_or_clear_bad(pmd);
575 #ifdef CONFIG_NUMA_BALANCING
576 #ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE
578 * _PAGE_NUMA works identical to _PAGE_PROTNONE (it's actually the
579 * same bit too). It's set only when _PAGE_PRESET is not set and it's
580 * never set if _PAGE_PRESENT is set.
582 * pte/pmd_present() returns true if pte/pmd_numa returns true. Page
583 * fault triggers on those regions if pte/pmd_numa returns true
584 * (because _PAGE_PRESENT is not set).
587 static inline int pte_numa(pte_t pte)
589 return (pte_flags(pte) &
590 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
595 static inline int pmd_numa(pmd_t pmd)
597 return (pmd_flags(pmd) &
598 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
603 * pte/pmd_mknuma sets the _PAGE_ACCESSED bitflag automatically
604 * because they're called by the NUMA hinting minor page fault. If we
605 * wouldn't set the _PAGE_ACCESSED bitflag here, the TLB miss handler
606 * would be forced to set it later while filling the TLB after we
607 * return to userland. That would trigger a second write to memory
608 * that we optimize away by setting _PAGE_ACCESSED here.
610 #ifndef pte_mknonnuma
611 static inline pte_t pte_mknonnuma(pte_t pte)
613 pte = pte_clear_flags(pte, _PAGE_NUMA);
614 return pte_set_flags(pte, _PAGE_PRESENT|_PAGE_ACCESSED);
618 #ifndef pmd_mknonnuma
619 static inline pmd_t pmd_mknonnuma(pmd_t pmd)
621 pmd = pmd_clear_flags(pmd, _PAGE_NUMA);
622 return pmd_set_flags(pmd, _PAGE_PRESENT|_PAGE_ACCESSED);
627 static inline pte_t pte_mknuma(pte_t pte)
629 pte = pte_set_flags(pte, _PAGE_NUMA);
630 return pte_clear_flags(pte, _PAGE_PRESENT);
635 static inline pmd_t pmd_mknuma(pmd_t pmd)
637 pmd = pmd_set_flags(pmd, _PAGE_NUMA);
638 return pmd_clear_flags(pmd, _PAGE_PRESENT);
642 extern int pte_numa(pte_t pte);
643 extern int pmd_numa(pmd_t pmd);
644 extern pte_t pte_mknonnuma(pte_t pte);
645 extern pmd_t pmd_mknonnuma(pmd_t pmd);
646 extern pte_t pte_mknuma(pte_t pte);
647 extern pmd_t pmd_mknuma(pmd_t pmd);
648 #endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */
650 static inline int pmd_numa(pmd_t pmd)
655 static inline int pte_numa(pte_t pte)
660 static inline pte_t pte_mknonnuma(pte_t pte)
665 static inline pmd_t pmd_mknonnuma(pmd_t pmd)
670 static inline pte_t pte_mknuma(pte_t pte)
675 static inline pmd_t pmd_mknuma(pmd_t pmd)
679 #endif /* CONFIG_NUMA_BALANCING */
681 #endif /* CONFIG_MMU */
683 #endif /* !__ASSEMBLY__ */
685 #endif /* _ASM_GENERIC_PGTABLE_H */