1 #ifndef _ASM_GENERIC_PGTABLE_H
2 #define _ASM_GENERIC_PGTABLE_H
7 #include <linux/mm_types.h>
9 #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
10 extern int ptep_set_access_flags(struct vm_area_struct *vma,
11 unsigned long address, pte_t *ptep,
12 pte_t entry, int dirty);
15 #ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
16 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
17 unsigned long address, pmd_t *pmdp,
18 pmd_t entry, int dirty);
21 #ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
22 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
23 unsigned long address,
31 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
36 #ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
37 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
38 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
39 unsigned long address,
47 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
50 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
51 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
52 unsigned long address,
58 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
61 #ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
62 int ptep_clear_flush_young(struct vm_area_struct *vma,
63 unsigned long address, pte_t *ptep);
66 #ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
67 int pmdp_clear_flush_young(struct vm_area_struct *vma,
68 unsigned long address, pmd_t *pmdp);
71 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
72 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
73 unsigned long address,
77 pte_clear(mm, address, ptep);
82 #ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR
83 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
84 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
85 unsigned long address,
92 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
95 #ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
96 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
97 unsigned long address, pte_t *ptep,
101 pte = ptep_get_and_clear(mm, address, ptep);
107 * Some architectures may be able to avoid expensive synchronization
108 * primitives when modifications are made to PTE's which are already
109 * not present, or in the process of an address space destruction.
111 #ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
112 static inline void pte_clear_not_present_full(struct mm_struct *mm,
113 unsigned long address,
117 pte_clear(mm, address, ptep);
121 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
122 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
123 unsigned long address,
127 #ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
128 extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
129 unsigned long address,
133 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
135 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
137 pte_t old_pte = *ptep;
138 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
142 #ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
143 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
144 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
145 unsigned long address, pmd_t *pmdp)
147 pmd_t old_pmd = *pmdp;
148 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
150 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
151 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
152 unsigned long address, pmd_t *pmdp)
156 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
159 #ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
160 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
161 unsigned long address, pmd_t *pmdp);
164 #ifndef __HAVE_ARCH_PTE_SAME
165 static inline int pte_same(pte_t pte_a, pte_t pte_b)
167 return pte_val(pte_a) == pte_val(pte_b);
171 #ifndef __HAVE_ARCH_PMD_SAME
172 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
173 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
175 return pmd_val(pmd_a) == pmd_val(pmd_b);
177 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
178 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
183 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
186 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_DIRTY
187 #define page_test_and_clear_dirty(pfn, mapped) (0)
190 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_DIRTY
191 #define pte_maybe_dirty(pte) pte_dirty(pte)
193 #define pte_maybe_dirty(pte) (1)
196 #ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
197 #define page_test_and_clear_young(pfn) (0)
200 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
201 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
204 #ifndef __HAVE_ARCH_MOVE_PTE
205 #define move_pte(pte, prot, old_addr, new_addr) (pte)
208 #ifndef flush_tlb_fix_spurious_fault
209 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
212 #ifndef pgprot_noncached
213 #define pgprot_noncached(prot) (prot)
216 #ifndef pgprot_writecombine
217 #define pgprot_writecombine pgprot_noncached
221 * When walking page tables, get the address of the next boundary,
222 * or the end address of the range if that comes earlier. Although no
223 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
226 #define pgd_addr_end(addr, end) \
227 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
228 (__boundary - 1 < (end) - 1)? __boundary: (end); \
232 #define pud_addr_end(addr, end) \
233 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
234 (__boundary - 1 < (end) - 1)? __boundary: (end); \
239 #define pmd_addr_end(addr, end) \
240 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
241 (__boundary - 1 < (end) - 1)? __boundary: (end); \
246 * When walking page tables, we usually want to skip any p?d_none entries;
247 * and any p?d_bad entries - reporting the error before resetting to none.
248 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
250 void pgd_clear_bad(pgd_t *);
251 void pud_clear_bad(pud_t *);
252 void pmd_clear_bad(pmd_t *);
254 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
258 if (unlikely(pgd_bad(*pgd))) {
265 static inline int pud_none_or_clear_bad(pud_t *pud)
269 if (unlikely(pud_bad(*pud))) {
276 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
280 if (unlikely(pmd_bad(*pmd))) {
287 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
292 * Get the current pte state, but zero it out to make it
293 * non-present, preventing the hardware from asynchronously
296 return ptep_get_and_clear(mm, addr, ptep);
299 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
301 pte_t *ptep, pte_t pte)
304 * The pte is non-present, so there's no hardware state to
307 set_pte_at(mm, addr, ptep, pte);
310 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
312 * Start a pte protection read-modify-write transaction, which
313 * protects against asynchronous hardware modifications to the pte.
314 * The intention is not to prevent the hardware from making pte
315 * updates, but to prevent any updates it may make from being lost.
317 * This does not protect against other software modifications of the
318 * pte; the appropriate pte lock must be held over the transation.
320 * Note that this interface is intended to be batchable, meaning that
321 * ptep_modify_prot_commit may not actually update the pte, but merely
322 * queue the update to be done at some later time. The update must be
323 * actually committed before the pte lock is released, however.
325 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
329 return __ptep_modify_prot_start(mm, addr, ptep);
333 * Commit an update to a pte, leaving any hardware-controlled bits in
334 * the PTE unmodified.
336 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
338 pte_t *ptep, pte_t pte)
340 __ptep_modify_prot_commit(mm, addr, ptep, pte);
342 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
343 #endif /* CONFIG_MMU */
346 * A facility to provide lazy MMU batching. This allows PTE updates and
347 * page invalidations to be delayed until a call to leave lazy MMU mode
348 * is issued. Some architectures may benefit from doing this, and it is
349 * beneficial for both shadow and direct mode hypervisors, which may batch
350 * the PTE updates which happen during this window. Note that using this
351 * interface requires that read hazards be removed from the code. A read
352 * hazard could result in the direct mode hypervisor case, since the actual
353 * write to the page tables may not yet have taken place, so reads though
354 * a raw PTE pointer after it has been modified are not guaranteed to be
355 * up to date. This mode can only be entered and left under the protection of
356 * the page table locks for all page tables which may be modified. In the UP
357 * case, this is required so that preemption is disabled, and in the SMP case,
358 * it must synchronize the delayed page table writes properly on other CPUs.
360 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
361 #define arch_enter_lazy_mmu_mode() do {} while (0)
362 #define arch_leave_lazy_mmu_mode() do {} while (0)
363 #define arch_flush_lazy_mmu_mode() do {} while (0)
367 * A facility to provide batching of the reload of page tables and
368 * other process state with the actual context switch code for
369 * paravirtualized guests. By convention, only one of the batched
370 * update (lazy) modes (CPU, MMU) should be active at any given time,
371 * entry should never be nested, and entry and exits should always be
372 * paired. This is for sanity of maintaining and reasoning about the
373 * kernel code. In this case, the exit (end of the context switch) is
374 * in architecture-specific code, and so doesn't need a generic
377 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
378 #define arch_start_context_switch(prev) do {} while (0)
381 #ifndef __HAVE_PFNMAP_TRACKING
383 * Interface that can be used by architecture code to keep track of
384 * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
386 * track_pfn_vma_new is called when a _new_ pfn mapping is being established
387 * for physical range indicated by pfn and size.
389 static inline int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
390 unsigned long pfn, unsigned long size)
396 * Interface that can be used by architecture code to keep track of
397 * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
399 * track_pfn_vma_copy is called when vma that is covering the pfnmap gets
400 * copied through copy_page_range().
402 static inline int track_pfn_vma_copy(struct vm_area_struct *vma)
408 * Interface that can be used by architecture code to keep track of
409 * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
411 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
412 * untrack can be called for a specific region indicated by pfn and size or
413 * can be for the entire vma (in which case size can be zero).
415 static inline void untrack_pfn_vma(struct vm_area_struct *vma,
416 unsigned long pfn, unsigned long size)
420 extern int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
421 unsigned long pfn, unsigned long size);
422 extern int track_pfn_vma_copy(struct vm_area_struct *vma);
423 extern void untrack_pfn_vma(struct vm_area_struct *vma, unsigned long pfn,
429 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
430 static inline int pmd_trans_huge(pmd_t pmd)
434 static inline int pmd_trans_splitting(pmd_t pmd)
438 #ifndef __HAVE_ARCH_PMD_WRITE
439 static inline int pmd_write(pmd_t pmd)
444 #endif /* __HAVE_ARCH_PMD_WRITE */
445 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
447 #ifndef pmd_read_atomic
448 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
451 * Depend on compiler for an atomic pmd read. NOTE: this is
452 * only going to work, if the pmdval_t isn't larger than
460 * This function is meant to be used by sites walking pagetables with
461 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
462 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
463 * into a null pmd and the transhuge page fault can convert a null pmd
464 * into an hugepmd or into a regular pmd (if the hugepage allocation
465 * fails). While holding the mmap_sem in read mode the pmd becomes
466 * stable and stops changing under us only if it's not null and not a
467 * transhuge pmd. When those races occurs and this function makes a
468 * difference vs the standard pmd_none_or_clear_bad, the result is
469 * undefined so behaving like if the pmd was none is safe (because it
470 * can return none anyway). The compiler level barrier() is critically
471 * important to compute the two checks atomically on the same pmdval.
473 * For 32bit kernels with a 64bit large pmd_t this automatically takes
474 * care of reading the pmd atomically to avoid SMP race conditions
475 * against pmd_populate() when the mmap_sem is hold for reading by the
476 * caller (a special atomic read not done by "gcc" as in the generic
477 * version above, is also needed when THP is disabled because the page
478 * fault can populate the pmd from under us).
480 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
482 pmd_t pmdval = pmd_read_atomic(pmd);
484 * The barrier will stabilize the pmdval in a register or on
485 * the stack so that it will stop changing under the code.
487 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
488 * pmd_read_atomic is allowed to return a not atomic pmdval
489 * (for example pointing to an hugepage that has never been
490 * mapped in the pmd). The below checks will only care about
491 * the low part of the pmd with 32bit PAE x86 anyway, with the
492 * exception of pmd_none(). So the important thing is that if
493 * the low part of the pmd is found null, the high part will
494 * be also null or the pmd_none() check below would be
497 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
500 if (pmd_none(pmdval))
502 if (unlikely(pmd_bad(pmdval))) {
503 if (!pmd_trans_huge(pmdval))
511 * This is a noop if Transparent Hugepage Support is not built into
512 * the kernel. Otherwise it is equivalent to
513 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
514 * places that already verified the pmd is not none and they want to
515 * walk ptes while holding the mmap sem in read mode (write mode don't
516 * need this). If THP is not enabled, the pmd can't go away under the
517 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
518 * run a pmd_trans_unstable before walking the ptes after
519 * split_huge_page_pmd returns (because it may have run when the pmd
520 * become null, but then a page fault can map in a THP and not a
523 static inline int pmd_trans_unstable(pmd_t *pmd)
525 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
526 return pmd_none_or_trans_huge_or_clear_bad(pmd);
532 #endif /* CONFIG_MMU */
534 #endif /* !__ASSEMBLY__ */
536 #endif /* _ASM_GENERIC_PGTABLE_H */