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 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
34 unsigned long address, pmd_t *pmdp,
35 pmd_t entry, int dirty);
38 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
39 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
40 unsigned long address,
48 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
53 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
54 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
55 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
56 unsigned long address,
64 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
67 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
68 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
69 unsigned long address,
75 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
78 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
79 int ptep_clear_flush_young(struct vm_area_struct *vma,
80 unsigned long address, pte_t *ptep);
83 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
84 int pmdp_clear_flush_young(struct vm_area_struct *vma,
85 unsigned long address, pmd_t *pmdp);
88 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
89 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
90 unsigned long address,
94 pte_clear(mm, address, ptep);
99 #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR
100 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
101 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
102 unsigned long address,
109 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
112 #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR_FULL
113 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
114 static inline pmd_t pmdp_get_and_clear_full(struct mm_struct *mm,
115 unsigned long address, pmd_t *pmdp,
118 return pmdp_get_and_clear(mm, address, pmdp);
120 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
123 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
124 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
125 unsigned long address, pte_t *ptep,
129 pte = ptep_get_and_clear(mm, address, ptep);
135 * Some architectures may be able to avoid expensive synchronization
136 * primitives when modifications are made to PTE's which are already
137 * not present, or in the process of an address space destruction.
139 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
140 static inline void pte_clear_not_present_full(struct mm_struct *mm,
141 unsigned long address,
145 pte_clear(mm, address, ptep);
149 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
150 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
151 unsigned long address,
155 #ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
156 extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
157 unsigned long address,
161 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
163 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
165 pte_t old_pte = *ptep;
166 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
170 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
171 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
172 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
173 unsigned long address, pmd_t *pmdp)
175 pmd_t old_pmd = *pmdp;
176 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
178 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
179 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
180 unsigned long address, pmd_t *pmdp)
184 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
187 #ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
188 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
189 unsigned long address, pmd_t *pmdp);
192 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
193 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
197 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
198 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
201 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
202 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
206 #ifndef __HAVE_ARCH_PTE_SAME
207 static inline int pte_same(pte_t pte_a, pte_t pte_b)
209 return pte_val(pte_a) == pte_val(pte_b);
213 #ifndef __HAVE_ARCH_PTE_UNUSED
215 * Some architectures provide facilities to virtualization guests
216 * so that they can flag allocated pages as unused. This allows the
217 * host to transparently reclaim unused pages. This function returns
218 * whether the pte's page is unused.
220 static inline int pte_unused(pte_t pte)
226 #ifndef __HAVE_ARCH_PMD_SAME
227 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
228 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
230 return pmd_val(pmd_a) == pmd_val(pmd_b);
232 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
233 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
238 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
241 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
242 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
245 #ifndef __HAVE_ARCH_MOVE_PTE
246 #define move_pte(pte, prot, old_addr, new_addr) (pte)
249 #ifndef pte_accessible
250 # define pte_accessible(mm, pte) ((void)(pte), 1)
253 #ifndef flush_tlb_fix_spurious_fault
254 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
257 #ifndef pgprot_noncached
258 #define pgprot_noncached(prot) (prot)
261 #ifndef pgprot_writecombine
262 #define pgprot_writecombine pgprot_noncached
265 #ifndef pgprot_writethrough
266 #define pgprot_writethrough pgprot_noncached
269 #ifndef pgprot_device
270 #define pgprot_device pgprot_noncached
273 #ifndef pgprot_modify
274 #define pgprot_modify pgprot_modify
275 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
277 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
278 newprot = pgprot_noncached(newprot);
279 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
280 newprot = pgprot_writecombine(newprot);
281 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
282 newprot = pgprot_device(newprot);
288 * When walking page tables, get the address of the next boundary,
289 * or the end address of the range if that comes earlier. Although no
290 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
293 #define pgd_addr_end(addr, end) \
294 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
295 (__boundary - 1 < (end) - 1)? __boundary: (end); \
299 #define pud_addr_end(addr, end) \
300 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
301 (__boundary - 1 < (end) - 1)? __boundary: (end); \
306 #define pmd_addr_end(addr, end) \
307 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
308 (__boundary - 1 < (end) - 1)? __boundary: (end); \
313 * When walking page tables, we usually want to skip any p?d_none entries;
314 * and any p?d_bad entries - reporting the error before resetting to none.
315 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
317 void pgd_clear_bad(pgd_t *);
318 void pud_clear_bad(pud_t *);
319 void pmd_clear_bad(pmd_t *);
321 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
325 if (unlikely(pgd_bad(*pgd))) {
332 static inline int pud_none_or_clear_bad(pud_t *pud)
336 if (unlikely(pud_bad(*pud))) {
343 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
347 if (unlikely(pmd_bad(*pmd))) {
354 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
359 * Get the current pte state, but zero it out to make it
360 * non-present, preventing the hardware from asynchronously
363 return ptep_get_and_clear(mm, addr, ptep);
366 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
368 pte_t *ptep, pte_t pte)
371 * The pte is non-present, so there's no hardware state to
374 set_pte_at(mm, addr, ptep, pte);
377 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
379 * Start a pte protection read-modify-write transaction, which
380 * protects against asynchronous hardware modifications to the pte.
381 * The intention is not to prevent the hardware from making pte
382 * updates, but to prevent any updates it may make from being lost.
384 * This does not protect against other software modifications of the
385 * pte; the appropriate pte lock must be held over the transation.
387 * Note that this interface is intended to be batchable, meaning that
388 * ptep_modify_prot_commit may not actually update the pte, but merely
389 * queue the update to be done at some later time. The update must be
390 * actually committed before the pte lock is released, however.
392 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
396 return __ptep_modify_prot_start(mm, addr, ptep);
400 * Commit an update to a pte, leaving any hardware-controlled bits in
401 * the PTE unmodified.
403 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
405 pte_t *ptep, pte_t pte)
407 __ptep_modify_prot_commit(mm, addr, ptep, pte);
409 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
410 #endif /* CONFIG_MMU */
413 * A facility to provide lazy MMU batching. This allows PTE updates and
414 * page invalidations to be delayed until a call to leave lazy MMU mode
415 * is issued. Some architectures may benefit from doing this, and it is
416 * beneficial for both shadow and direct mode hypervisors, which may batch
417 * the PTE updates which happen during this window. Note that using this
418 * interface requires that read hazards be removed from the code. A read
419 * hazard could result in the direct mode hypervisor case, since the actual
420 * write to the page tables may not yet have taken place, so reads though
421 * a raw PTE pointer after it has been modified are not guaranteed to be
422 * up to date. This mode can only be entered and left under the protection of
423 * the page table locks for all page tables which may be modified. In the UP
424 * case, this is required so that preemption is disabled, and in the SMP case,
425 * it must synchronize the delayed page table writes properly on other CPUs.
427 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
428 #define arch_enter_lazy_mmu_mode() do {} while (0)
429 #define arch_leave_lazy_mmu_mode() do {} while (0)
430 #define arch_flush_lazy_mmu_mode() do {} while (0)
434 * A facility to provide batching of the reload of page tables and
435 * other process state with the actual context switch code for
436 * paravirtualized guests. By convention, only one of the batched
437 * update (lazy) modes (CPU, MMU) should be active at any given time,
438 * entry should never be nested, and entry and exits should always be
439 * paired. This is for sanity of maintaining and reasoning about the
440 * kernel code. In this case, the exit (end of the context switch) is
441 * in architecture-specific code, and so doesn't need a generic
444 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
445 #define arch_start_context_switch(prev) do {} while (0)
448 #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
449 static inline int pte_soft_dirty(pte_t pte)
454 static inline int pmd_soft_dirty(pmd_t pmd)
459 static inline pte_t pte_mksoft_dirty(pte_t pte)
464 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
469 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
474 static inline int pte_swp_soft_dirty(pte_t pte)
479 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
485 #ifndef __HAVE_PFNMAP_TRACKING
487 * Interfaces that can be used by architecture code to keep track of
488 * memory type of pfn mappings specified by the remap_pfn_range,
493 * track_pfn_remap is called when a _new_ pfn mapping is being established
494 * by remap_pfn_range() for physical range indicated by pfn and size.
496 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
497 unsigned long pfn, unsigned long addr,
504 * track_pfn_insert is called when a _new_ single pfn is established
505 * by vm_insert_pfn().
507 static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
514 * track_pfn_copy is called when vma that is covering the pfnmap gets
515 * copied through copy_page_range().
517 static inline int track_pfn_copy(struct vm_area_struct *vma)
523 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
524 * untrack can be called for a specific region indicated by pfn and size or
525 * can be for the entire vma (in which case pfn, size are zero).
527 static inline void untrack_pfn(struct vm_area_struct *vma,
528 unsigned long pfn, unsigned long size)
532 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
533 unsigned long pfn, unsigned long addr,
535 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
537 extern int track_pfn_copy(struct vm_area_struct *vma);
538 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
542 #ifdef __HAVE_COLOR_ZERO_PAGE
543 static inline int is_zero_pfn(unsigned long pfn)
545 extern unsigned long zero_pfn;
546 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
547 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
550 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
553 static inline int is_zero_pfn(unsigned long pfn)
555 extern unsigned long zero_pfn;
556 return pfn == zero_pfn;
559 static inline unsigned long my_zero_pfn(unsigned long addr)
561 extern unsigned long zero_pfn;
568 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
569 static inline int pmd_trans_huge(pmd_t pmd)
573 static inline int pmd_trans_splitting(pmd_t pmd)
577 #ifndef __HAVE_ARCH_PMD_WRITE
578 static inline int pmd_write(pmd_t pmd)
583 #endif /* __HAVE_ARCH_PMD_WRITE */
584 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
586 #ifndef pmd_read_atomic
587 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
590 * Depend on compiler for an atomic pmd read. NOTE: this is
591 * only going to work, if the pmdval_t isn't larger than
598 #ifndef pmd_move_must_withdraw
599 static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
600 spinlock_t *old_pmd_ptl)
603 * With split pmd lock we also need to move preallocated
604 * PTE page table if new_pmd is on different PMD page table.
606 return new_pmd_ptl != old_pmd_ptl;
611 * This function is meant to be used by sites walking pagetables with
612 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
613 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
614 * into a null pmd and the transhuge page fault can convert a null pmd
615 * into an hugepmd or into a regular pmd (if the hugepage allocation
616 * fails). While holding the mmap_sem in read mode the pmd becomes
617 * stable and stops changing under us only if it's not null and not a
618 * transhuge pmd. When those races occurs and this function makes a
619 * difference vs the standard pmd_none_or_clear_bad, the result is
620 * undefined so behaving like if the pmd was none is safe (because it
621 * can return none anyway). The compiler level barrier() is critically
622 * important to compute the two checks atomically on the same pmdval.
624 * For 32bit kernels with a 64bit large pmd_t this automatically takes
625 * care of reading the pmd atomically to avoid SMP race conditions
626 * against pmd_populate() when the mmap_sem is hold for reading by the
627 * caller (a special atomic read not done by "gcc" as in the generic
628 * version above, is also needed when THP is disabled because the page
629 * fault can populate the pmd from under us).
631 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
633 pmd_t pmdval = pmd_read_atomic(pmd);
635 * The barrier will stabilize the pmdval in a register or on
636 * the stack so that it will stop changing under the code.
638 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
639 * pmd_read_atomic is allowed to return a not atomic pmdval
640 * (for example pointing to an hugepage that has never been
641 * mapped in the pmd). The below checks will only care about
642 * the low part of the pmd with 32bit PAE x86 anyway, with the
643 * exception of pmd_none(). So the important thing is that if
644 * the low part of the pmd is found null, the high part will
645 * be also null or the pmd_none() check below would be
648 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
651 if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
653 if (unlikely(pmd_bad(pmdval))) {
661 * This is a noop if Transparent Hugepage Support is not built into
662 * the kernel. Otherwise it is equivalent to
663 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
664 * places that already verified the pmd is not none and they want to
665 * walk ptes while holding the mmap sem in read mode (write mode don't
666 * need this). If THP is not enabled, the pmd can't go away under the
667 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
668 * run a pmd_trans_unstable before walking the ptes after
669 * split_huge_page_pmd returns (because it may have run when the pmd
670 * become null, but then a page fault can map in a THP and not a
673 static inline int pmd_trans_unstable(pmd_t *pmd)
675 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
676 return pmd_none_or_trans_huge_or_clear_bad(pmd);
682 #ifndef CONFIG_NUMA_BALANCING
684 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
685 * the only case the kernel cares is for NUMA balancing and is only ever set
686 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
687 * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
688 * is the responsibility of the caller to distinguish between PROT_NONE
689 * protections and NUMA hinting fault protections.
691 static inline int pte_protnone(pte_t pte)
696 static inline int pmd_protnone(pmd_t pmd)
700 #endif /* CONFIG_NUMA_BALANCING */
702 #endif /* CONFIG_MMU */
704 #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
705 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
706 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
707 int pud_clear_huge(pud_t *pud);
708 int pmd_clear_huge(pmd_t *pmd);
709 #else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
710 static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
714 static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
718 static inline int pud_clear_huge(pud_t *pud)
722 static inline int pmd_clear_huge(pmd_t *pmd)
726 #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
728 #endif /* !__ASSEMBLY__ */
730 #ifndef io_remap_pfn_range
731 #define io_remap_pfn_range remap_pfn_range
734 #endif /* _ASM_GENERIC_PGTABLE_H */