]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - include/asm-generic/pgtable.h
Merge tag 'iommu-fixes-v3.15-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git...
[mirror_ubuntu-artful-kernel.git] / include / asm-generic / pgtable.h
1 #ifndef _ASM_GENERIC_PGTABLE_H
2 #define _ASM_GENERIC_PGTABLE_H
3
4 #ifndef __ASSEMBLY__
5 #ifdef CONFIG_MMU
6
7 #include <linux/mm_types.h>
8 #include <linux/bug.h>
9
10 /*
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.
15 */
16 #ifndef USER_PGTABLES_CEILING
17 #define USER_PGTABLES_CEILING 0UL
18 #endif
19
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);
24 #endif
25
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);
30 #endif
31
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,
35 pte_t *ptep)
36 {
37 pte_t pte = *ptep;
38 int r = 1;
39 if (!pte_young(pte))
40 r = 0;
41 else
42 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
43 return r;
44 }
45 #endif
46
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,
51 pmd_t *pmdp)
52 {
53 pmd_t pmd = *pmdp;
54 int r = 1;
55 if (!pmd_young(pmd))
56 r = 0;
57 else
58 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
59 return r;
60 }
61 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
62 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
63 unsigned long address,
64 pmd_t *pmdp)
65 {
66 BUG();
67 return 0;
68 }
69 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
70 #endif
71
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);
75 #endif
76
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);
80 #endif
81
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,
85 pte_t *ptep)
86 {
87 pte_t pte = *ptep;
88 pte_clear(mm, address, ptep);
89 return pte;
90 }
91 #endif
92
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,
97 pmd_t *pmdp)
98 {
99 pmd_t pmd = *pmdp;
100 pmd_clear(pmdp);
101 return pmd;
102 }
103 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
104 #endif
105
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,
109 int full)
110 {
111 pte_t pte;
112 pte = ptep_get_and_clear(mm, address, ptep);
113 return pte;
114 }
115 #endif
116
117 /*
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.
121 */
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,
125 pte_t *ptep,
126 int full)
127 {
128 pte_clear(mm, address, ptep);
129 }
130 #endif
131
132 #ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
133 extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
134 unsigned long address,
135 pte_t *ptep);
136 #endif
137
138 #ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
139 extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
140 unsigned long address,
141 pmd_t *pmdp);
142 #endif
143
144 #ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
145 struct mm_struct;
146 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
147 {
148 pte_t old_pte = *ptep;
149 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
150 }
151 #endif
152
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)
157 {
158 pmd_t old_pmd = *pmdp;
159 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
160 }
161 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
162 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
163 unsigned long address, pmd_t *pmdp)
164 {
165 BUG();
166 }
167 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
168 #endif
169
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);
173 #endif
174
175 #ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
176 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
177 pgtable_t pgtable);
178 #endif
179
180 #ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
181 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
182 #endif
183
184 #ifndef __HAVE_ARCH_PMDP_INVALIDATE
185 extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
186 pmd_t *pmdp);
187 #endif
188
189 #ifndef __HAVE_ARCH_PTE_SAME
190 static inline int pte_same(pte_t pte_a, pte_t pte_b)
191 {
192 return pte_val(pte_a) == pte_val(pte_b);
193 }
194 #endif
195
196 #ifndef __HAVE_ARCH_PTE_UNUSED
197 /*
198 * Some architectures provide facilities to virtualization guests
199 * so that they can flag allocated pages as unused. This allows the
200 * host to transparently reclaim unused pages. This function returns
201 * whether the pte's page is unused.
202 */
203 static inline int pte_unused(pte_t pte)
204 {
205 return 0;
206 }
207 #endif
208
209 #ifndef __HAVE_ARCH_PMD_SAME
210 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
211 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
212 {
213 return pmd_val(pmd_a) == pmd_val(pmd_b);
214 }
215 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
216 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
217 {
218 BUG();
219 return 0;
220 }
221 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
222 #endif
223
224 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
225 #define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
226 #endif
227
228 #ifndef __HAVE_ARCH_MOVE_PTE
229 #define move_pte(pte, prot, old_addr, new_addr) (pte)
230 #endif
231
232 #ifndef pte_accessible
233 # define pte_accessible(mm, pte) ((void)(pte), 1)
234 #endif
235
236 #ifndef flush_tlb_fix_spurious_fault
237 #define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
238 #endif
239
240 #ifndef pgprot_noncached
241 #define pgprot_noncached(prot) (prot)
242 #endif
243
244 #ifndef pgprot_writecombine
245 #define pgprot_writecombine pgprot_noncached
246 #endif
247
248 /*
249 * When walking page tables, get the address of the next boundary,
250 * or the end address of the range if that comes earlier. Although no
251 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
252 */
253
254 #define pgd_addr_end(addr, end) \
255 ({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
256 (__boundary - 1 < (end) - 1)? __boundary: (end); \
257 })
258
259 #ifndef pud_addr_end
260 #define pud_addr_end(addr, end) \
261 ({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
262 (__boundary - 1 < (end) - 1)? __boundary: (end); \
263 })
264 #endif
265
266 #ifndef pmd_addr_end
267 #define pmd_addr_end(addr, end) \
268 ({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
269 (__boundary - 1 < (end) - 1)? __boundary: (end); \
270 })
271 #endif
272
273 /*
274 * When walking page tables, we usually want to skip any p?d_none entries;
275 * and any p?d_bad entries - reporting the error before resetting to none.
276 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
277 */
278 void pgd_clear_bad(pgd_t *);
279 void pud_clear_bad(pud_t *);
280 void pmd_clear_bad(pmd_t *);
281
282 static inline int pgd_none_or_clear_bad(pgd_t *pgd)
283 {
284 if (pgd_none(*pgd))
285 return 1;
286 if (unlikely(pgd_bad(*pgd))) {
287 pgd_clear_bad(pgd);
288 return 1;
289 }
290 return 0;
291 }
292
293 static inline int pud_none_or_clear_bad(pud_t *pud)
294 {
295 if (pud_none(*pud))
296 return 1;
297 if (unlikely(pud_bad(*pud))) {
298 pud_clear_bad(pud);
299 return 1;
300 }
301 return 0;
302 }
303
304 static inline int pmd_none_or_clear_bad(pmd_t *pmd)
305 {
306 if (pmd_none(*pmd))
307 return 1;
308 if (unlikely(pmd_bad(*pmd))) {
309 pmd_clear_bad(pmd);
310 return 1;
311 }
312 return 0;
313 }
314
315 static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
316 unsigned long addr,
317 pte_t *ptep)
318 {
319 /*
320 * Get the current pte state, but zero it out to make it
321 * non-present, preventing the hardware from asynchronously
322 * updating it.
323 */
324 return ptep_get_and_clear(mm, addr, ptep);
325 }
326
327 static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
328 unsigned long addr,
329 pte_t *ptep, pte_t pte)
330 {
331 /*
332 * The pte is non-present, so there's no hardware state to
333 * preserve.
334 */
335 set_pte_at(mm, addr, ptep, pte);
336 }
337
338 #ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
339 /*
340 * Start a pte protection read-modify-write transaction, which
341 * protects against asynchronous hardware modifications to the pte.
342 * The intention is not to prevent the hardware from making pte
343 * updates, but to prevent any updates it may make from being lost.
344 *
345 * This does not protect against other software modifications of the
346 * pte; the appropriate pte lock must be held over the transation.
347 *
348 * Note that this interface is intended to be batchable, meaning that
349 * ptep_modify_prot_commit may not actually update the pte, but merely
350 * queue the update to be done at some later time. The update must be
351 * actually committed before the pte lock is released, however.
352 */
353 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
354 unsigned long addr,
355 pte_t *ptep)
356 {
357 return __ptep_modify_prot_start(mm, addr, ptep);
358 }
359
360 /*
361 * Commit an update to a pte, leaving any hardware-controlled bits in
362 * the PTE unmodified.
363 */
364 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
365 unsigned long addr,
366 pte_t *ptep, pte_t pte)
367 {
368 __ptep_modify_prot_commit(mm, addr, ptep, pte);
369 }
370 #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
371 #endif /* CONFIG_MMU */
372
373 /*
374 * A facility to provide lazy MMU batching. This allows PTE updates and
375 * page invalidations to be delayed until a call to leave lazy MMU mode
376 * is issued. Some architectures may benefit from doing this, and it is
377 * beneficial for both shadow and direct mode hypervisors, which may batch
378 * the PTE updates which happen during this window. Note that using this
379 * interface requires that read hazards be removed from the code. A read
380 * hazard could result in the direct mode hypervisor case, since the actual
381 * write to the page tables may not yet have taken place, so reads though
382 * a raw PTE pointer after it has been modified are not guaranteed to be
383 * up to date. This mode can only be entered and left under the protection of
384 * the page table locks for all page tables which may be modified. In the UP
385 * case, this is required so that preemption is disabled, and in the SMP case,
386 * it must synchronize the delayed page table writes properly on other CPUs.
387 */
388 #ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
389 #define arch_enter_lazy_mmu_mode() do {} while (0)
390 #define arch_leave_lazy_mmu_mode() do {} while (0)
391 #define arch_flush_lazy_mmu_mode() do {} while (0)
392 #endif
393
394 /*
395 * A facility to provide batching of the reload of page tables and
396 * other process state with the actual context switch code for
397 * paravirtualized guests. By convention, only one of the batched
398 * update (lazy) modes (CPU, MMU) should be active at any given time,
399 * entry should never be nested, and entry and exits should always be
400 * paired. This is for sanity of maintaining and reasoning about the
401 * kernel code. In this case, the exit (end of the context switch) is
402 * in architecture-specific code, and so doesn't need a generic
403 * definition.
404 */
405 #ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
406 #define arch_start_context_switch(prev) do {} while (0)
407 #endif
408
409 #ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
410 static inline int pte_soft_dirty(pte_t pte)
411 {
412 return 0;
413 }
414
415 static inline int pmd_soft_dirty(pmd_t pmd)
416 {
417 return 0;
418 }
419
420 static inline pte_t pte_mksoft_dirty(pte_t pte)
421 {
422 return pte;
423 }
424
425 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
426 {
427 return pmd;
428 }
429
430 static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
431 {
432 return pte;
433 }
434
435 static inline int pte_swp_soft_dirty(pte_t pte)
436 {
437 return 0;
438 }
439
440 static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
441 {
442 return pte;
443 }
444
445 static inline pte_t pte_file_clear_soft_dirty(pte_t pte)
446 {
447 return pte;
448 }
449
450 static inline pte_t pte_file_mksoft_dirty(pte_t pte)
451 {
452 return pte;
453 }
454
455 static inline int pte_file_soft_dirty(pte_t pte)
456 {
457 return 0;
458 }
459 #endif
460
461 #ifndef __HAVE_PFNMAP_TRACKING
462 /*
463 * Interfaces that can be used by architecture code to keep track of
464 * memory type of pfn mappings specified by the remap_pfn_range,
465 * vm_insert_pfn.
466 */
467
468 /*
469 * track_pfn_remap is called when a _new_ pfn mapping is being established
470 * by remap_pfn_range() for physical range indicated by pfn and size.
471 */
472 static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
473 unsigned long pfn, unsigned long addr,
474 unsigned long size)
475 {
476 return 0;
477 }
478
479 /*
480 * track_pfn_insert is called when a _new_ single pfn is established
481 * by vm_insert_pfn().
482 */
483 static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
484 unsigned long pfn)
485 {
486 return 0;
487 }
488
489 /*
490 * track_pfn_copy is called when vma that is covering the pfnmap gets
491 * copied through copy_page_range().
492 */
493 static inline int track_pfn_copy(struct vm_area_struct *vma)
494 {
495 return 0;
496 }
497
498 /*
499 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
500 * untrack can be called for a specific region indicated by pfn and size or
501 * can be for the entire vma (in which case pfn, size are zero).
502 */
503 static inline void untrack_pfn(struct vm_area_struct *vma,
504 unsigned long pfn, unsigned long size)
505 {
506 }
507 #else
508 extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
509 unsigned long pfn, unsigned long addr,
510 unsigned long size);
511 extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
512 unsigned long pfn);
513 extern int track_pfn_copy(struct vm_area_struct *vma);
514 extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
515 unsigned long size);
516 #endif
517
518 #ifdef __HAVE_COLOR_ZERO_PAGE
519 static inline int is_zero_pfn(unsigned long pfn)
520 {
521 extern unsigned long zero_pfn;
522 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
523 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
524 }
525
526 #define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
527
528 #else
529 static inline int is_zero_pfn(unsigned long pfn)
530 {
531 extern unsigned long zero_pfn;
532 return pfn == zero_pfn;
533 }
534
535 static inline unsigned long my_zero_pfn(unsigned long addr)
536 {
537 extern unsigned long zero_pfn;
538 return zero_pfn;
539 }
540 #endif
541
542 #ifdef CONFIG_MMU
543
544 #ifndef CONFIG_TRANSPARENT_HUGEPAGE
545 static inline int pmd_trans_huge(pmd_t pmd)
546 {
547 return 0;
548 }
549 static inline int pmd_trans_splitting(pmd_t pmd)
550 {
551 return 0;
552 }
553 #ifndef __HAVE_ARCH_PMD_WRITE
554 static inline int pmd_write(pmd_t pmd)
555 {
556 BUG();
557 return 0;
558 }
559 #endif /* __HAVE_ARCH_PMD_WRITE */
560 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
561
562 #ifndef pmd_read_atomic
563 static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
564 {
565 /*
566 * Depend on compiler for an atomic pmd read. NOTE: this is
567 * only going to work, if the pmdval_t isn't larger than
568 * an unsigned long.
569 */
570 return *pmdp;
571 }
572 #endif
573
574 #ifndef pmd_move_must_withdraw
575 static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
576 spinlock_t *old_pmd_ptl)
577 {
578 /*
579 * With split pmd lock we also need to move preallocated
580 * PTE page table if new_pmd is on different PMD page table.
581 */
582 return new_pmd_ptl != old_pmd_ptl;
583 }
584 #endif
585
586 /*
587 * This function is meant to be used by sites walking pagetables with
588 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
589 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
590 * into a null pmd and the transhuge page fault can convert a null pmd
591 * into an hugepmd or into a regular pmd (if the hugepage allocation
592 * fails). While holding the mmap_sem in read mode the pmd becomes
593 * stable and stops changing under us only if it's not null and not a
594 * transhuge pmd. When those races occurs and this function makes a
595 * difference vs the standard pmd_none_or_clear_bad, the result is
596 * undefined so behaving like if the pmd was none is safe (because it
597 * can return none anyway). The compiler level barrier() is critically
598 * important to compute the two checks atomically on the same pmdval.
599 *
600 * For 32bit kernels with a 64bit large pmd_t this automatically takes
601 * care of reading the pmd atomically to avoid SMP race conditions
602 * against pmd_populate() when the mmap_sem is hold for reading by the
603 * caller (a special atomic read not done by "gcc" as in the generic
604 * version above, is also needed when THP is disabled because the page
605 * fault can populate the pmd from under us).
606 */
607 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
608 {
609 pmd_t pmdval = pmd_read_atomic(pmd);
610 /*
611 * The barrier will stabilize the pmdval in a register or on
612 * the stack so that it will stop changing under the code.
613 *
614 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
615 * pmd_read_atomic is allowed to return a not atomic pmdval
616 * (for example pointing to an hugepage that has never been
617 * mapped in the pmd). The below checks will only care about
618 * the low part of the pmd with 32bit PAE x86 anyway, with the
619 * exception of pmd_none(). So the important thing is that if
620 * the low part of the pmd is found null, the high part will
621 * be also null or the pmd_none() check below would be
622 * confused.
623 */
624 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
625 barrier();
626 #endif
627 if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
628 return 1;
629 if (unlikely(pmd_bad(pmdval))) {
630 pmd_clear_bad(pmd);
631 return 1;
632 }
633 return 0;
634 }
635
636 /*
637 * This is a noop if Transparent Hugepage Support is not built into
638 * the kernel. Otherwise it is equivalent to
639 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
640 * places that already verified the pmd is not none and they want to
641 * walk ptes while holding the mmap sem in read mode (write mode don't
642 * need this). If THP is not enabled, the pmd can't go away under the
643 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
644 * run a pmd_trans_unstable before walking the ptes after
645 * split_huge_page_pmd returns (because it may have run when the pmd
646 * become null, but then a page fault can map in a THP and not a
647 * regular page).
648 */
649 static inline int pmd_trans_unstable(pmd_t *pmd)
650 {
651 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
652 return pmd_none_or_trans_huge_or_clear_bad(pmd);
653 #else
654 return 0;
655 #endif
656 }
657
658 #ifdef CONFIG_NUMA_BALANCING
659 #ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE
660 /*
661 * _PAGE_NUMA works identical to _PAGE_PROTNONE (it's actually the
662 * same bit too). It's set only when _PAGE_PRESET is not set and it's
663 * never set if _PAGE_PRESENT is set.
664 *
665 * pte/pmd_present() returns true if pte/pmd_numa returns true. Page
666 * fault triggers on those regions if pte/pmd_numa returns true
667 * (because _PAGE_PRESENT is not set).
668 */
669 #ifndef pte_numa
670 static inline int pte_numa(pte_t pte)
671 {
672 return (pte_flags(pte) &
673 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
674 }
675 #endif
676
677 #ifndef pmd_numa
678 static inline int pmd_numa(pmd_t pmd)
679 {
680 return (pmd_flags(pmd) &
681 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
682 }
683 #endif
684
685 /*
686 * pte/pmd_mknuma sets the _PAGE_ACCESSED bitflag automatically
687 * because they're called by the NUMA hinting minor page fault. If we
688 * wouldn't set the _PAGE_ACCESSED bitflag here, the TLB miss handler
689 * would be forced to set it later while filling the TLB after we
690 * return to userland. That would trigger a second write to memory
691 * that we optimize away by setting _PAGE_ACCESSED here.
692 */
693 #ifndef pte_mknonnuma
694 static inline pte_t pte_mknonnuma(pte_t pte)
695 {
696 pteval_t val = pte_val(pte);
697
698 val &= ~_PAGE_NUMA;
699 val |= (_PAGE_PRESENT|_PAGE_ACCESSED);
700 return __pte(val);
701 }
702 #endif
703
704 #ifndef pmd_mknonnuma
705 static inline pmd_t pmd_mknonnuma(pmd_t pmd)
706 {
707 pmdval_t val = pmd_val(pmd);
708
709 val &= ~_PAGE_NUMA;
710 val |= (_PAGE_PRESENT|_PAGE_ACCESSED);
711
712 return __pmd(val);
713 }
714 #endif
715
716 #ifndef pte_mknuma
717 static inline pte_t pte_mknuma(pte_t pte)
718 {
719 pteval_t val = pte_val(pte);
720
721 val &= ~_PAGE_PRESENT;
722 val |= _PAGE_NUMA;
723
724 return __pte(val);
725 }
726 #endif
727
728 #ifndef ptep_set_numa
729 static inline void ptep_set_numa(struct mm_struct *mm, unsigned long addr,
730 pte_t *ptep)
731 {
732 pte_t ptent = *ptep;
733
734 ptent = pte_mknuma(ptent);
735 set_pte_at(mm, addr, ptep, ptent);
736 return;
737 }
738 #endif
739
740 #ifndef pmd_mknuma
741 static inline pmd_t pmd_mknuma(pmd_t pmd)
742 {
743 pmdval_t val = pmd_val(pmd);
744
745 val &= ~_PAGE_PRESENT;
746 val |= _PAGE_NUMA;
747
748 return __pmd(val);
749 }
750 #endif
751
752 #ifndef pmdp_set_numa
753 static inline void pmdp_set_numa(struct mm_struct *mm, unsigned long addr,
754 pmd_t *pmdp)
755 {
756 pmd_t pmd = *pmdp;
757
758 pmd = pmd_mknuma(pmd);
759 set_pmd_at(mm, addr, pmdp, pmd);
760 return;
761 }
762 #endif
763 #else
764 extern int pte_numa(pte_t pte);
765 extern int pmd_numa(pmd_t pmd);
766 extern pte_t pte_mknonnuma(pte_t pte);
767 extern pmd_t pmd_mknonnuma(pmd_t pmd);
768 extern pte_t pte_mknuma(pte_t pte);
769 extern pmd_t pmd_mknuma(pmd_t pmd);
770 extern void ptep_set_numa(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
771 extern void pmdp_set_numa(struct mm_struct *mm, unsigned long addr, pmd_t *pmdp);
772 #endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */
773 #else
774 static inline int pmd_numa(pmd_t pmd)
775 {
776 return 0;
777 }
778
779 static inline int pte_numa(pte_t pte)
780 {
781 return 0;
782 }
783
784 static inline pte_t pte_mknonnuma(pte_t pte)
785 {
786 return pte;
787 }
788
789 static inline pmd_t pmd_mknonnuma(pmd_t pmd)
790 {
791 return pmd;
792 }
793
794 static inline pte_t pte_mknuma(pte_t pte)
795 {
796 return pte;
797 }
798
799 static inline void ptep_set_numa(struct mm_struct *mm, unsigned long addr,
800 pte_t *ptep)
801 {
802 return;
803 }
804
805
806 static inline pmd_t pmd_mknuma(pmd_t pmd)
807 {
808 return pmd;
809 }
810
811 static inline void pmdp_set_numa(struct mm_struct *mm, unsigned long addr,
812 pmd_t *pmdp)
813 {
814 return ;
815 }
816 #endif /* CONFIG_NUMA_BALANCING */
817
818 #endif /* CONFIG_MMU */
819
820 #endif /* !__ASSEMBLY__ */
821
822 #ifndef io_remap_pfn_range
823 #define io_remap_pfn_range remap_pfn_range
824 #endif
825
826 #endif /* _ASM_GENERIC_PGTABLE_H */