mm: numa: do not dereference pmd outside of the lock during NUMA hinting fault
[linux-block.git] / include / asm-generic / pgtable.h
CommitLineData
1da177e4
LT
1#ifndef _ASM_GENERIC_PGTABLE_H
2#define _ASM_GENERIC_PGTABLE_H
3
673eae82 4#ifndef __ASSEMBLY__
9535239f 5#ifdef CONFIG_MMU
673eae82 6
fbd71844 7#include <linux/mm_types.h>
187f1882 8#include <linux/bug.h>
fbd71844 9
6ee8630e
HD
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
1da177e4 20#ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
e2cda322
AA
21extern 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
27extern int pmdp_set_access_flags(struct vm_area_struct *vma,
28 unsigned long address, pmd_t *pmdp,
29 pmd_t entry, int dirty);
1da177e4
LT
30#endif
31
32#ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
e2cda322
AA
33static 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
49static 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 */
62static 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 */
1da177e4
LT
70#endif
71
72#ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
e2cda322
AA
73int 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
78int pmdp_clear_flush_young(struct vm_area_struct *vma,
79 unsigned long address, pmd_t *pmdp);
1da177e4
LT
80#endif
81
1da177e4 82#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
e2cda322
AA
83static 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
95static 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;
2d28a227 100 pmd_clear(pmdp);
e2cda322 101 return pmd;
49b24d6b 102}
e2cda322 103#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1da177e4
LT
104#endif
105
fcbe08d6
MS
106#ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR_FULL
107#ifdef CONFIG_TRANSPARENT_HUGEPAGE
108static inline pmd_t pmdp_get_and_clear_full(struct mm_struct *mm,
109 unsigned long address, pmd_t *pmdp,
110 int full)
111{
112 return pmdp_get_and_clear(mm, address, pmdp);
113}
114#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
115#endif
116
a600388d 117#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
e2cda322
AA
118static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
119 unsigned long address, pte_t *ptep,
120 int full)
121{
122 pte_t pte;
123 pte = ptep_get_and_clear(mm, address, ptep);
124 return pte;
125}
a600388d
ZA
126#endif
127
9888a1ca
ZA
128/*
129 * Some architectures may be able to avoid expensive synchronization
130 * primitives when modifications are made to PTE's which are already
131 * not present, or in the process of an address space destruction.
132 */
133#ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
e2cda322
AA
134static inline void pte_clear_not_present_full(struct mm_struct *mm,
135 unsigned long address,
136 pte_t *ptep,
137 int full)
138{
139 pte_clear(mm, address, ptep);
140}
a600388d
ZA
141#endif
142
1da177e4 143#ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
e2cda322
AA
144extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
145 unsigned long address,
146 pte_t *ptep);
147#endif
148
149#ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
150extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
151 unsigned long address,
152 pmd_t *pmdp);
1da177e4
LT
153#endif
154
155#ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
8c65b4a6 156struct mm_struct;
1da177e4
LT
157static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
158{
159 pte_t old_pte = *ptep;
160 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
161}
162#endif
163
e2cda322
AA
164#ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
165#ifdef CONFIG_TRANSPARENT_HUGEPAGE
166static inline void pmdp_set_wrprotect(struct mm_struct *mm,
167 unsigned long address, pmd_t *pmdp)
168{
169 pmd_t old_pmd = *pmdp;
170 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
171}
172#else /* CONFIG_TRANSPARENT_HUGEPAGE */
173static inline void pmdp_set_wrprotect(struct mm_struct *mm,
174 unsigned long address, pmd_t *pmdp)
175{
176 BUG();
177}
178#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
179#endif
180
181#ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
73636b1a
CM
182extern void pmdp_splitting_flush(struct vm_area_struct *vma,
183 unsigned long address, pmd_t *pmdp);
e2cda322
AA
184#endif
185
e3ebcf64 186#ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
6b0b50b0
AK
187extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
188 pgtable_t pgtable);
e3ebcf64
GS
189#endif
190
191#ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
6b0b50b0 192extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
e3ebcf64
GS
193#endif
194
46dcde73
GS
195#ifndef __HAVE_ARCH_PMDP_INVALIDATE
196extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
197 pmd_t *pmdp);
198#endif
199
1da177e4 200#ifndef __HAVE_ARCH_PTE_SAME
e2cda322
AA
201static inline int pte_same(pte_t pte_a, pte_t pte_b)
202{
203 return pte_val(pte_a) == pte_val(pte_b);
204}
205#endif
206
45961722
KW
207#ifndef __HAVE_ARCH_PTE_UNUSED
208/*
209 * Some architectures provide facilities to virtualization guests
210 * so that they can flag allocated pages as unused. This allows the
211 * host to transparently reclaim unused pages. This function returns
212 * whether the pte's page is unused.
213 */
214static inline int pte_unused(pte_t pte)
215{
216 return 0;
217}
218#endif
219
e2cda322
AA
220#ifndef __HAVE_ARCH_PMD_SAME
221#ifdef CONFIG_TRANSPARENT_HUGEPAGE
222static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
223{
224 return pmd_val(pmd_a) == pmd_val(pmd_b);
225}
226#else /* CONFIG_TRANSPARENT_HUGEPAGE */
227static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
228{
229 BUG();
230 return 0;
231}
232#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1da177e4
LT
233#endif
234
1da177e4
LT
235#ifndef __HAVE_ARCH_PGD_OFFSET_GATE
236#define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
237#endif
238
0b0968a3 239#ifndef __HAVE_ARCH_MOVE_PTE
8b1f3124 240#define move_pte(pte, prot, old_addr, new_addr) (pte)
8b1f3124
NP
241#endif
242
2c3cf556 243#ifndef pte_accessible
20841405 244# define pte_accessible(mm, pte) ((void)(pte), 1)
2c3cf556
RR
245#endif
246
c46a7c81
MG
247#ifndef pte_present_nonuma
248#define pte_present_nonuma(pte) pte_present(pte)
249#endif
250
61c77326
SL
251#ifndef flush_tlb_fix_spurious_fault
252#define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
253#endif
254
0634a632
PM
255#ifndef pgprot_noncached
256#define pgprot_noncached(prot) (prot)
257#endif
258
2520bd31 259#ifndef pgprot_writecombine
260#define pgprot_writecombine pgprot_noncached
261#endif
262
8b921acf
LD
263#ifndef pgprot_device
264#define pgprot_device pgprot_noncached
265#endif
266
64e45507
PF
267#ifndef pgprot_modify
268#define pgprot_modify pgprot_modify
269static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
270{
271 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
272 newprot = pgprot_noncached(newprot);
273 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
274 newprot = pgprot_writecombine(newprot);
275 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
276 newprot = pgprot_device(newprot);
277 return newprot;
278}
279#endif
280
1da177e4 281/*
8f6c99c1
HD
282 * When walking page tables, get the address of the next boundary,
283 * or the end address of the range if that comes earlier. Although no
284 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
1da177e4
LT
285 */
286
1da177e4
LT
287#define pgd_addr_end(addr, end) \
288({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
289 (__boundary - 1 < (end) - 1)? __boundary: (end); \
290})
1da177e4
LT
291
292#ifndef pud_addr_end
293#define pud_addr_end(addr, end) \
294({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
295 (__boundary - 1 < (end) - 1)? __boundary: (end); \
296})
297#endif
298
299#ifndef pmd_addr_end
300#define pmd_addr_end(addr, end) \
301({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
302 (__boundary - 1 < (end) - 1)? __boundary: (end); \
303})
304#endif
305
1da177e4
LT
306/*
307 * When walking page tables, we usually want to skip any p?d_none entries;
308 * and any p?d_bad entries - reporting the error before resetting to none.
309 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
310 */
311void pgd_clear_bad(pgd_t *);
312void pud_clear_bad(pud_t *);
313void pmd_clear_bad(pmd_t *);
314
315static inline int pgd_none_or_clear_bad(pgd_t *pgd)
316{
317 if (pgd_none(*pgd))
318 return 1;
319 if (unlikely(pgd_bad(*pgd))) {
320 pgd_clear_bad(pgd);
321 return 1;
322 }
323 return 0;
324}
325
326static inline int pud_none_or_clear_bad(pud_t *pud)
327{
328 if (pud_none(*pud))
329 return 1;
330 if (unlikely(pud_bad(*pud))) {
331 pud_clear_bad(pud);
332 return 1;
333 }
334 return 0;
335}
336
337static inline int pmd_none_or_clear_bad(pmd_t *pmd)
338{
339 if (pmd_none(*pmd))
340 return 1;
341 if (unlikely(pmd_bad(*pmd))) {
342 pmd_clear_bad(pmd);
343 return 1;
344 }
345 return 0;
346}
9535239f 347
1ea0704e
JF
348static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
349 unsigned long addr,
350 pte_t *ptep)
351{
352 /*
353 * Get the current pte state, but zero it out to make it
354 * non-present, preventing the hardware from asynchronously
355 * updating it.
356 */
357 return ptep_get_and_clear(mm, addr, ptep);
358}
359
360static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
361 unsigned long addr,
362 pte_t *ptep, pte_t pte)
363{
364 /*
365 * The pte is non-present, so there's no hardware state to
366 * preserve.
367 */
368 set_pte_at(mm, addr, ptep, pte);
369}
370
371#ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
372/*
373 * Start a pte protection read-modify-write transaction, which
374 * protects against asynchronous hardware modifications to the pte.
375 * The intention is not to prevent the hardware from making pte
376 * updates, but to prevent any updates it may make from being lost.
377 *
378 * This does not protect against other software modifications of the
379 * pte; the appropriate pte lock must be held over the transation.
380 *
381 * Note that this interface is intended to be batchable, meaning that
382 * ptep_modify_prot_commit may not actually update the pte, but merely
383 * queue the update to be done at some later time. The update must be
384 * actually committed before the pte lock is released, however.
385 */
386static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
387 unsigned long addr,
388 pte_t *ptep)
389{
390 return __ptep_modify_prot_start(mm, addr, ptep);
391}
392
393/*
394 * Commit an update to a pte, leaving any hardware-controlled bits in
395 * the PTE unmodified.
396 */
397static inline void ptep_modify_prot_commit(struct mm_struct *mm,
398 unsigned long addr,
399 pte_t *ptep, pte_t pte)
400{
401 __ptep_modify_prot_commit(mm, addr, ptep, pte);
402}
403#endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
fe1a6875 404#endif /* CONFIG_MMU */
1ea0704e 405
9535239f
GU
406/*
407 * A facility to provide lazy MMU batching. This allows PTE updates and
408 * page invalidations to be delayed until a call to leave lazy MMU mode
409 * is issued. Some architectures may benefit from doing this, and it is
410 * beneficial for both shadow and direct mode hypervisors, which may batch
411 * the PTE updates which happen during this window. Note that using this
412 * interface requires that read hazards be removed from the code. A read
413 * hazard could result in the direct mode hypervisor case, since the actual
414 * write to the page tables may not yet have taken place, so reads though
415 * a raw PTE pointer after it has been modified are not guaranteed to be
416 * up to date. This mode can only be entered and left under the protection of
417 * the page table locks for all page tables which may be modified. In the UP
418 * case, this is required so that preemption is disabled, and in the SMP case,
419 * it must synchronize the delayed page table writes properly on other CPUs.
420 */
421#ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
422#define arch_enter_lazy_mmu_mode() do {} while (0)
423#define arch_leave_lazy_mmu_mode() do {} while (0)
424#define arch_flush_lazy_mmu_mode() do {} while (0)
425#endif
426
427/*
7fd7d83d
JF
428 * A facility to provide batching of the reload of page tables and
429 * other process state with the actual context switch code for
430 * paravirtualized guests. By convention, only one of the batched
431 * update (lazy) modes (CPU, MMU) should be active at any given time,
432 * entry should never be nested, and entry and exits should always be
433 * paired. This is for sanity of maintaining and reasoning about the
434 * kernel code. In this case, the exit (end of the context switch) is
435 * in architecture-specific code, and so doesn't need a generic
436 * definition.
9535239f 437 */
7fd7d83d 438#ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
224101ed 439#define arch_start_context_switch(prev) do {} while (0)
9535239f
GU
440#endif
441
0f8975ec
PE
442#ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
443static inline int pte_soft_dirty(pte_t pte)
444{
445 return 0;
446}
447
448static inline int pmd_soft_dirty(pmd_t pmd)
449{
450 return 0;
451}
452
453static inline pte_t pte_mksoft_dirty(pte_t pte)
454{
455 return pte;
456}
457
458static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
459{
460 return pmd;
461}
179ef71c
CG
462
463static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
464{
465 return pte;
466}
467
468static inline int pte_swp_soft_dirty(pte_t pte)
469{
470 return 0;
471}
472
473static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
474{
475 return pte;
476}
0f8975ec
PE
477#endif
478
34801ba9 479#ifndef __HAVE_PFNMAP_TRACKING
480/*
5180da41
SS
481 * Interfaces that can be used by architecture code to keep track of
482 * memory type of pfn mappings specified by the remap_pfn_range,
483 * vm_insert_pfn.
484 */
485
486/*
487 * track_pfn_remap is called when a _new_ pfn mapping is being established
488 * by remap_pfn_range() for physical range indicated by pfn and size.
34801ba9 489 */
5180da41 490static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
b3b9c293
KK
491 unsigned long pfn, unsigned long addr,
492 unsigned long size)
34801ba9 493{
494 return 0;
495}
496
497/*
5180da41
SS
498 * track_pfn_insert is called when a _new_ single pfn is established
499 * by vm_insert_pfn().
500 */
501static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
502 unsigned long pfn)
503{
504 return 0;
505}
506
507/*
508 * track_pfn_copy is called when vma that is covering the pfnmap gets
34801ba9 509 * copied through copy_page_range().
510 */
5180da41 511static inline int track_pfn_copy(struct vm_area_struct *vma)
34801ba9 512{
513 return 0;
514}
515
516/*
34801ba9 517 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
518 * untrack can be called for a specific region indicated by pfn and size or
5180da41 519 * can be for the entire vma (in which case pfn, size are zero).
34801ba9 520 */
5180da41
SS
521static inline void untrack_pfn(struct vm_area_struct *vma,
522 unsigned long pfn, unsigned long size)
34801ba9 523{
524}
525#else
5180da41 526extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
b3b9c293
KK
527 unsigned long pfn, unsigned long addr,
528 unsigned long size);
5180da41
SS
529extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
530 unsigned long pfn);
531extern int track_pfn_copy(struct vm_area_struct *vma);
532extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
533 unsigned long size);
34801ba9 534#endif
535
816422ad
KS
536#ifdef __HAVE_COLOR_ZERO_PAGE
537static inline int is_zero_pfn(unsigned long pfn)
538{
539 extern unsigned long zero_pfn;
540 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
541 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
542}
543
2f91ec8c
KS
544#define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
545
816422ad
KS
546#else
547static inline int is_zero_pfn(unsigned long pfn)
548{
549 extern unsigned long zero_pfn;
550 return pfn == zero_pfn;
551}
552
553static inline unsigned long my_zero_pfn(unsigned long addr)
554{
555 extern unsigned long zero_pfn;
556 return zero_pfn;
557}
558#endif
559
1a5a9906
AA
560#ifdef CONFIG_MMU
561
5f6e8da7
AA
562#ifndef CONFIG_TRANSPARENT_HUGEPAGE
563static inline int pmd_trans_huge(pmd_t pmd)
564{
565 return 0;
566}
567static inline int pmd_trans_splitting(pmd_t pmd)
568{
569 return 0;
570}
e2cda322
AA
571#ifndef __HAVE_ARCH_PMD_WRITE
572static inline int pmd_write(pmd_t pmd)
573{
574 BUG();
575 return 0;
576}
577#endif /* __HAVE_ARCH_PMD_WRITE */
1a5a9906
AA
578#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
579
26c19178
AA
580#ifndef pmd_read_atomic
581static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
582{
583 /*
584 * Depend on compiler for an atomic pmd read. NOTE: this is
585 * only going to work, if the pmdval_t isn't larger than
586 * an unsigned long.
587 */
588 return *pmdp;
589}
590#endif
591
b3084f4d
AK
592#ifndef pmd_move_must_withdraw
593static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
594 spinlock_t *old_pmd_ptl)
595{
596 /*
597 * With split pmd lock we also need to move preallocated
598 * PTE page table if new_pmd is on different PMD page table.
599 */
600 return new_pmd_ptl != old_pmd_ptl;
601}
602#endif
603
1a5a9906
AA
604/*
605 * This function is meant to be used by sites walking pagetables with
606 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
607 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
608 * into a null pmd and the transhuge page fault can convert a null pmd
609 * into an hugepmd or into a regular pmd (if the hugepage allocation
610 * fails). While holding the mmap_sem in read mode the pmd becomes
611 * stable and stops changing under us only if it's not null and not a
612 * transhuge pmd. When those races occurs and this function makes a
613 * difference vs the standard pmd_none_or_clear_bad, the result is
614 * undefined so behaving like if the pmd was none is safe (because it
615 * can return none anyway). The compiler level barrier() is critically
616 * important to compute the two checks atomically on the same pmdval.
26c19178
AA
617 *
618 * For 32bit kernels with a 64bit large pmd_t this automatically takes
619 * care of reading the pmd atomically to avoid SMP race conditions
620 * against pmd_populate() when the mmap_sem is hold for reading by the
621 * caller (a special atomic read not done by "gcc" as in the generic
622 * version above, is also needed when THP is disabled because the page
623 * fault can populate the pmd from under us).
1a5a9906
AA
624 */
625static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
626{
26c19178 627 pmd_t pmdval = pmd_read_atomic(pmd);
1a5a9906
AA
628 /*
629 * The barrier will stabilize the pmdval in a register or on
630 * the stack so that it will stop changing under the code.
e4eed03f
AA
631 *
632 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
633 * pmd_read_atomic is allowed to return a not atomic pmdval
634 * (for example pointing to an hugepage that has never been
635 * mapped in the pmd). The below checks will only care about
636 * the low part of the pmd with 32bit PAE x86 anyway, with the
637 * exception of pmd_none(). So the important thing is that if
638 * the low part of the pmd is found null, the high part will
639 * be also null or the pmd_none() check below would be
640 * confused.
1a5a9906
AA
641 */
642#ifdef CONFIG_TRANSPARENT_HUGEPAGE
643 barrier();
644#endif
ee53664b 645 if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
1a5a9906
AA
646 return 1;
647 if (unlikely(pmd_bad(pmdval))) {
ee53664b 648 pmd_clear_bad(pmd);
1a5a9906
AA
649 return 1;
650 }
651 return 0;
652}
653
654/*
655 * This is a noop if Transparent Hugepage Support is not built into
656 * the kernel. Otherwise it is equivalent to
657 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
658 * places that already verified the pmd is not none and they want to
659 * walk ptes while holding the mmap sem in read mode (write mode don't
660 * need this). If THP is not enabled, the pmd can't go away under the
661 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
662 * run a pmd_trans_unstable before walking the ptes after
663 * split_huge_page_pmd returns (because it may have run when the pmd
664 * become null, but then a page fault can map in a THP and not a
665 * regular page).
666 */
667static inline int pmd_trans_unstable(pmd_t *pmd)
668{
669#ifdef CONFIG_TRANSPARENT_HUGEPAGE
670 return pmd_none_or_trans_huge_or_clear_bad(pmd);
671#else
672 return 0;
5f6e8da7 673#endif
1a5a9906
AA
674}
675
be3a7284 676#ifdef CONFIG_NUMA_BALANCING
be3a7284 677/*
6a33979d
MG
678 * _PAGE_NUMA distinguishes between an unmapped page table entry, an entry that
679 * is protected for PROT_NONE and a NUMA hinting fault entry. If the
680 * architecture defines __PAGE_PROTNONE then it should take that into account
681 * but those that do not can rely on the fact that the NUMA hinting scanner
682 * skips inaccessible VMAs.
be3a7284
AA
683 *
684 * pte/pmd_present() returns true if pte/pmd_numa returns true. Page
685 * fault triggers on those regions if pte/pmd_numa returns true
686 * (because _PAGE_PRESENT is not set).
687 */
688#ifndef pte_numa
689static inline int pte_numa(pte_t pte)
690{
6a33979d 691 return ptenuma_flags(pte) == _PAGE_NUMA;
be3a7284
AA
692}
693#endif
694
695#ifndef pmd_numa
696static inline int pmd_numa(pmd_t pmd)
697{
6a33979d 698 return pmdnuma_flags(pmd) == _PAGE_NUMA;
be3a7284
AA
699}
700#endif
701
702/*
703 * pte/pmd_mknuma sets the _PAGE_ACCESSED bitflag automatically
704 * because they're called by the NUMA hinting minor page fault. If we
705 * wouldn't set the _PAGE_ACCESSED bitflag here, the TLB miss handler
706 * would be forced to set it later while filling the TLB after we
707 * return to userland. That would trigger a second write to memory
708 * that we optimize away by setting _PAGE_ACCESSED here.
709 */
710#ifndef pte_mknonnuma
711static inline pte_t pte_mknonnuma(pte_t pte)
712{
29c77870
MG
713 pteval_t val = pte_val(pte);
714
715 val &= ~_PAGE_NUMA;
716 val |= (_PAGE_PRESENT|_PAGE_ACCESSED);
717 return __pte(val);
be3a7284
AA
718}
719#endif
720
721#ifndef pmd_mknonnuma
722static inline pmd_t pmd_mknonnuma(pmd_t pmd)
723{
29c77870
MG
724 pmdval_t val = pmd_val(pmd);
725
726 val &= ~_PAGE_NUMA;
727 val |= (_PAGE_PRESENT|_PAGE_ACCESSED);
728
729 return __pmd(val);
be3a7284
AA
730}
731#endif
732
733#ifndef pte_mknuma
734static inline pte_t pte_mknuma(pte_t pte)
735{
29c77870
MG
736 pteval_t val = pte_val(pte);
737
6a33979d
MG
738 VM_BUG_ON(!(val & _PAGE_PRESENT));
739
29c77870
MG
740 val &= ~_PAGE_PRESENT;
741 val |= _PAGE_NUMA;
742
743 return __pte(val);
be3a7284
AA
744}
745#endif
746
56eecdb9
AK
747#ifndef ptep_set_numa
748static inline void ptep_set_numa(struct mm_struct *mm, unsigned long addr,
749 pte_t *ptep)
750{
751 pte_t ptent = *ptep;
752
753 ptent = pte_mknuma(ptent);
754 set_pte_at(mm, addr, ptep, ptent);
755 return;
756}
757#endif
758
be3a7284
AA
759#ifndef pmd_mknuma
760static inline pmd_t pmd_mknuma(pmd_t pmd)
761{
29c77870
MG
762 pmdval_t val = pmd_val(pmd);
763
764 val &= ~_PAGE_PRESENT;
765 val |= _PAGE_NUMA;
766
767 return __pmd(val);
be3a7284
AA
768}
769#endif
56eecdb9
AK
770
771#ifndef pmdp_set_numa
772static inline void pmdp_set_numa(struct mm_struct *mm, unsigned long addr,
773 pmd_t *pmdp)
774{
775 pmd_t pmd = *pmdp;
776
777 pmd = pmd_mknuma(pmd);
778 set_pmd_at(mm, addr, pmdp, pmd);
779 return;
780}
781#endif
be3a7284 782#else
be3a7284
AA
783static inline int pmd_numa(pmd_t pmd)
784{
785 return 0;
786}
787
788static inline int pte_numa(pte_t pte)
789{
790 return 0;
791}
792
793static inline pte_t pte_mknonnuma(pte_t pte)
794{
795 return pte;
796}
797
798static inline pmd_t pmd_mknonnuma(pmd_t pmd)
799{
800 return pmd;
801}
802
803static inline pte_t pte_mknuma(pte_t pte)
804{
805 return pte;
806}
807
56eecdb9
AK
808static inline void ptep_set_numa(struct mm_struct *mm, unsigned long addr,
809 pte_t *ptep)
810{
811 return;
812}
813
814
be3a7284
AA
815static inline pmd_t pmd_mknuma(pmd_t pmd)
816{
817 return pmd;
818}
56eecdb9
AK
819
820static inline void pmdp_set_numa(struct mm_struct *mm, unsigned long addr,
821 pmd_t *pmdp)
822{
823 return ;
824}
be3a7284
AA
825#endif /* CONFIG_NUMA_BALANCING */
826
1a5a9906 827#endif /* CONFIG_MMU */
5f6e8da7 828
1da177e4
LT
829#endif /* !__ASSEMBLY__ */
830
40d158e6
AV
831#ifndef io_remap_pfn_range
832#define io_remap_pfn_range remap_pfn_range
833#endif
834
1da177e4 835#endif /* _ASM_GENERIC_PGTABLE_H */