]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - include/asm-generic/pgtable.h
tracing: add trace event for memory-failure
[mirror_ubuntu-bionic-kernel.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>
e61ce6ad 9#include <linux/errno.h>
fbd71844 10
235a8f02
KS
11#if 4 - defined(__PAGETABLE_PUD_FOLDED) - defined(__PAGETABLE_PMD_FOLDED) != \
12 CONFIG_PGTABLE_LEVELS
13#error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{PUD,PMD}_FOLDED
14#endif
15
6ee8630e
HD
16/*
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.
21 */
22#ifndef USER_PGTABLES_CEILING
23#define USER_PGTABLES_CEILING 0UL
24#endif
25
1da177e4 26#ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
e2cda322
AA
27extern int ptep_set_access_flags(struct vm_area_struct *vma,
28 unsigned long address, pte_t *ptep,
29 pte_t entry, int dirty);
30#endif
31
32#ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
33extern int pmdp_set_access_flags(struct vm_area_struct *vma,
34 unsigned long address, pmd_t *pmdp,
35 pmd_t entry, int dirty);
1da177e4
LT
36#endif
37
38#ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
e2cda322
AA
39static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
40 unsigned long address,
41 pte_t *ptep)
42{
43 pte_t pte = *ptep;
44 int r = 1;
45 if (!pte_young(pte))
46 r = 0;
47 else
48 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
49 return r;
50}
51#endif
52
53#ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
54#ifdef CONFIG_TRANSPARENT_HUGEPAGE
55static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
56 unsigned long address,
57 pmd_t *pmdp)
58{
59 pmd_t pmd = *pmdp;
60 int r = 1;
61 if (!pmd_young(pmd))
62 r = 0;
63 else
64 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
65 return r;
66}
67#else /* CONFIG_TRANSPARENT_HUGEPAGE */
68static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
69 unsigned long address,
70 pmd_t *pmdp)
71{
72 BUG();
73 return 0;
74}
75#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1da177e4
LT
76#endif
77
78#ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
e2cda322
AA
79int ptep_clear_flush_young(struct vm_area_struct *vma,
80 unsigned long address, pte_t *ptep);
81#endif
82
83#ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
84int pmdp_clear_flush_young(struct vm_area_struct *vma,
85 unsigned long address, pmd_t *pmdp);
1da177e4
LT
86#endif
87
1da177e4 88#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
e2cda322
AA
89static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
90 unsigned long address,
91 pte_t *ptep)
92{
93 pte_t pte = *ptep;
94 pte_clear(mm, address, ptep);
95 return pte;
96}
97#endif
98
99#ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR
100#ifdef CONFIG_TRANSPARENT_HUGEPAGE
101static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
102 unsigned long address,
103 pmd_t *pmdp)
104{
105 pmd_t pmd = *pmdp;
2d28a227 106 pmd_clear(pmdp);
e2cda322 107 return pmd;
49b24d6b 108}
e2cda322 109#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1da177e4
LT
110#endif
111
fcbe08d6
MS
112#ifndef __HAVE_ARCH_PMDP_GET_AND_CLEAR_FULL
113#ifdef CONFIG_TRANSPARENT_HUGEPAGE
114static inline pmd_t pmdp_get_and_clear_full(struct mm_struct *mm,
115 unsigned long address, pmd_t *pmdp,
116 int full)
117{
118 return pmdp_get_and_clear(mm, address, pmdp);
119}
120#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
121#endif
122
a600388d 123#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
e2cda322
AA
124static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
125 unsigned long address, pte_t *ptep,
126 int full)
127{
128 pte_t pte;
129 pte = ptep_get_and_clear(mm, address, ptep);
130 return pte;
131}
a600388d
ZA
132#endif
133
9888a1ca
ZA
134/*
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.
138 */
139#ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
e2cda322
AA
140static inline void pte_clear_not_present_full(struct mm_struct *mm,
141 unsigned long address,
142 pte_t *ptep,
143 int full)
144{
145 pte_clear(mm, address, ptep);
146}
a600388d
ZA
147#endif
148
1da177e4 149#ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
e2cda322
AA
150extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
151 unsigned long address,
152 pte_t *ptep);
153#endif
154
155#ifndef __HAVE_ARCH_PMDP_CLEAR_FLUSH
156extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
157 unsigned long address,
158 pmd_t *pmdp);
1da177e4
LT
159#endif
160
161#ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
8c65b4a6 162struct mm_struct;
1da177e4
LT
163static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
164{
165 pte_t old_pte = *ptep;
166 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
167}
168#endif
169
e2cda322
AA
170#ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
171#ifdef CONFIG_TRANSPARENT_HUGEPAGE
172static inline void pmdp_set_wrprotect(struct mm_struct *mm,
173 unsigned long address, pmd_t *pmdp)
174{
175 pmd_t old_pmd = *pmdp;
176 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
177}
178#else /* CONFIG_TRANSPARENT_HUGEPAGE */
179static inline void pmdp_set_wrprotect(struct mm_struct *mm,
180 unsigned long address, pmd_t *pmdp)
181{
182 BUG();
183}
184#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
185#endif
186
187#ifndef __HAVE_ARCH_PMDP_SPLITTING_FLUSH
73636b1a
CM
188extern void pmdp_splitting_flush(struct vm_area_struct *vma,
189 unsigned long address, pmd_t *pmdp);
e2cda322
AA
190#endif
191
e3ebcf64 192#ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
6b0b50b0
AK
193extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
194 pgtable_t pgtable);
e3ebcf64
GS
195#endif
196
197#ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
6b0b50b0 198extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
e3ebcf64
GS
199#endif
200
46dcde73
GS
201#ifndef __HAVE_ARCH_PMDP_INVALIDATE
202extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
203 pmd_t *pmdp);
204#endif
205
1da177e4 206#ifndef __HAVE_ARCH_PTE_SAME
e2cda322
AA
207static inline int pte_same(pte_t pte_a, pte_t pte_b)
208{
209 return pte_val(pte_a) == pte_val(pte_b);
210}
211#endif
212
45961722
KW
213#ifndef __HAVE_ARCH_PTE_UNUSED
214/*
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.
219 */
220static inline int pte_unused(pte_t pte)
221{
222 return 0;
223}
224#endif
225
e2cda322
AA
226#ifndef __HAVE_ARCH_PMD_SAME
227#ifdef CONFIG_TRANSPARENT_HUGEPAGE
228static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
229{
230 return pmd_val(pmd_a) == pmd_val(pmd_b);
231}
232#else /* CONFIG_TRANSPARENT_HUGEPAGE */
233static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
234{
235 BUG();
236 return 0;
237}
238#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1da177e4
LT
239#endif
240
1da177e4
LT
241#ifndef __HAVE_ARCH_PGD_OFFSET_GATE
242#define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
243#endif
244
0b0968a3 245#ifndef __HAVE_ARCH_MOVE_PTE
8b1f3124 246#define move_pte(pte, prot, old_addr, new_addr) (pte)
8b1f3124
NP
247#endif
248
2c3cf556 249#ifndef pte_accessible
20841405 250# define pte_accessible(mm, pte) ((void)(pte), 1)
2c3cf556
RR
251#endif
252
61c77326
SL
253#ifndef flush_tlb_fix_spurious_fault
254#define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
255#endif
256
0634a632
PM
257#ifndef pgprot_noncached
258#define pgprot_noncached(prot) (prot)
259#endif
260
2520bd31 261#ifndef pgprot_writecombine
262#define pgprot_writecombine pgprot_noncached
263#endif
264
d1b4bfbf
TK
265#ifndef pgprot_writethrough
266#define pgprot_writethrough pgprot_noncached
267#endif
268
8b921acf
LD
269#ifndef pgprot_device
270#define pgprot_device pgprot_noncached
271#endif
272
64e45507
PF
273#ifndef pgprot_modify
274#define pgprot_modify pgprot_modify
275static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
276{
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);
283 return newprot;
284}
285#endif
286
1da177e4 287/*
8f6c99c1
HD
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.
1da177e4
LT
291 */
292
1da177e4
LT
293#define pgd_addr_end(addr, end) \
294({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
295 (__boundary - 1 < (end) - 1)? __boundary: (end); \
296})
1da177e4
LT
297
298#ifndef pud_addr_end
299#define pud_addr_end(addr, end) \
300({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
301 (__boundary - 1 < (end) - 1)? __boundary: (end); \
302})
303#endif
304
305#ifndef pmd_addr_end
306#define pmd_addr_end(addr, end) \
307({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
308 (__boundary - 1 < (end) - 1)? __boundary: (end); \
309})
310#endif
311
1da177e4
LT
312/*
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.
316 */
317void pgd_clear_bad(pgd_t *);
318void pud_clear_bad(pud_t *);
319void pmd_clear_bad(pmd_t *);
320
321static inline int pgd_none_or_clear_bad(pgd_t *pgd)
322{
323 if (pgd_none(*pgd))
324 return 1;
325 if (unlikely(pgd_bad(*pgd))) {
326 pgd_clear_bad(pgd);
327 return 1;
328 }
329 return 0;
330}
331
332static inline int pud_none_or_clear_bad(pud_t *pud)
333{
334 if (pud_none(*pud))
335 return 1;
336 if (unlikely(pud_bad(*pud))) {
337 pud_clear_bad(pud);
338 return 1;
339 }
340 return 0;
341}
342
343static inline int pmd_none_or_clear_bad(pmd_t *pmd)
344{
345 if (pmd_none(*pmd))
346 return 1;
347 if (unlikely(pmd_bad(*pmd))) {
348 pmd_clear_bad(pmd);
349 return 1;
350 }
351 return 0;
352}
9535239f 353
1ea0704e
JF
354static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
355 unsigned long addr,
356 pte_t *ptep)
357{
358 /*
359 * Get the current pte state, but zero it out to make it
360 * non-present, preventing the hardware from asynchronously
361 * updating it.
362 */
363 return ptep_get_and_clear(mm, addr, ptep);
364}
365
366static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
367 unsigned long addr,
368 pte_t *ptep, pte_t pte)
369{
370 /*
371 * The pte is non-present, so there's no hardware state to
372 * preserve.
373 */
374 set_pte_at(mm, addr, ptep, pte);
375}
376
377#ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
378/*
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.
383 *
384 * This does not protect against other software modifications of the
385 * pte; the appropriate pte lock must be held over the transation.
386 *
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.
391 */
392static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
393 unsigned long addr,
394 pte_t *ptep)
395{
396 return __ptep_modify_prot_start(mm, addr, ptep);
397}
398
399/*
400 * Commit an update to a pte, leaving any hardware-controlled bits in
401 * the PTE unmodified.
402 */
403static inline void ptep_modify_prot_commit(struct mm_struct *mm,
404 unsigned long addr,
405 pte_t *ptep, pte_t pte)
406{
407 __ptep_modify_prot_commit(mm, addr, ptep, pte);
408}
409#endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
fe1a6875 410#endif /* CONFIG_MMU */
1ea0704e 411
9535239f
GU
412/*
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.
426 */
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)
431#endif
432
433/*
7fd7d83d
JF
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
442 * definition.
9535239f 443 */
7fd7d83d 444#ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
224101ed 445#define arch_start_context_switch(prev) do {} while (0)
9535239f
GU
446#endif
447
0f8975ec
PE
448#ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
449static inline int pte_soft_dirty(pte_t pte)
450{
451 return 0;
452}
453
454static inline int pmd_soft_dirty(pmd_t pmd)
455{
456 return 0;
457}
458
459static inline pte_t pte_mksoft_dirty(pte_t pte)
460{
461 return pte;
462}
463
464static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
465{
466 return pmd;
467}
179ef71c
CG
468
469static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
470{
471 return pte;
472}
473
474static inline int pte_swp_soft_dirty(pte_t pte)
475{
476 return 0;
477}
478
479static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
480{
481 return pte;
482}
0f8975ec
PE
483#endif
484
34801ba9 485#ifndef __HAVE_PFNMAP_TRACKING
486/*
5180da41
SS
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,
489 * vm_insert_pfn.
490 */
491
492/*
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.
34801ba9 495 */
5180da41 496static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
b3b9c293
KK
497 unsigned long pfn, unsigned long addr,
498 unsigned long size)
34801ba9 499{
500 return 0;
501}
502
503/*
5180da41
SS
504 * track_pfn_insert is called when a _new_ single pfn is established
505 * by vm_insert_pfn().
506 */
507static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
508 unsigned long pfn)
509{
510 return 0;
511}
512
513/*
514 * track_pfn_copy is called when vma that is covering the pfnmap gets
34801ba9 515 * copied through copy_page_range().
516 */
5180da41 517static inline int track_pfn_copy(struct vm_area_struct *vma)
34801ba9 518{
519 return 0;
520}
521
522/*
34801ba9 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
5180da41 525 * can be for the entire vma (in which case pfn, size are zero).
34801ba9 526 */
5180da41
SS
527static inline void untrack_pfn(struct vm_area_struct *vma,
528 unsigned long pfn, unsigned long size)
34801ba9 529{
530}
531#else
5180da41 532extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
b3b9c293
KK
533 unsigned long pfn, unsigned long addr,
534 unsigned long size);
5180da41
SS
535extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
536 unsigned long pfn);
537extern int track_pfn_copy(struct vm_area_struct *vma);
538extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
539 unsigned long size);
34801ba9 540#endif
541
816422ad
KS
542#ifdef __HAVE_COLOR_ZERO_PAGE
543static inline int is_zero_pfn(unsigned long pfn)
544{
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);
548}
549
2f91ec8c
KS
550#define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
551
816422ad
KS
552#else
553static inline int is_zero_pfn(unsigned long pfn)
554{
555 extern unsigned long zero_pfn;
556 return pfn == zero_pfn;
557}
558
559static inline unsigned long my_zero_pfn(unsigned long addr)
560{
561 extern unsigned long zero_pfn;
562 return zero_pfn;
563}
564#endif
565
1a5a9906
AA
566#ifdef CONFIG_MMU
567
5f6e8da7
AA
568#ifndef CONFIG_TRANSPARENT_HUGEPAGE
569static inline int pmd_trans_huge(pmd_t pmd)
570{
571 return 0;
572}
573static inline int pmd_trans_splitting(pmd_t pmd)
574{
575 return 0;
576}
e2cda322
AA
577#ifndef __HAVE_ARCH_PMD_WRITE
578static inline int pmd_write(pmd_t pmd)
579{
580 BUG();
581 return 0;
582}
583#endif /* __HAVE_ARCH_PMD_WRITE */
1a5a9906
AA
584#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
585
26c19178
AA
586#ifndef pmd_read_atomic
587static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
588{
589 /*
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
592 * an unsigned long.
593 */
594 return *pmdp;
595}
596#endif
597
b3084f4d
AK
598#ifndef pmd_move_must_withdraw
599static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
600 spinlock_t *old_pmd_ptl)
601{
602 /*
603 * With split pmd lock we also need to move preallocated
604 * PTE page table if new_pmd is on different PMD page table.
605 */
606 return new_pmd_ptl != old_pmd_ptl;
607}
608#endif
609
1a5a9906
AA
610/*
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.
26c19178
AA
623 *
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).
1a5a9906
AA
630 */
631static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
632{
26c19178 633 pmd_t pmdval = pmd_read_atomic(pmd);
1a5a9906
AA
634 /*
635 * The barrier will stabilize the pmdval in a register or on
636 * the stack so that it will stop changing under the code.
e4eed03f
AA
637 *
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
646 * confused.
1a5a9906
AA
647 */
648#ifdef CONFIG_TRANSPARENT_HUGEPAGE
649 barrier();
650#endif
ee53664b 651 if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
1a5a9906
AA
652 return 1;
653 if (unlikely(pmd_bad(pmdval))) {
ee53664b 654 pmd_clear_bad(pmd);
1a5a9906
AA
655 return 1;
656 }
657 return 0;
658}
659
660/*
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
671 * regular page).
672 */
673static inline int pmd_trans_unstable(pmd_t *pmd)
674{
675#ifdef CONFIG_TRANSPARENT_HUGEPAGE
676 return pmd_none_or_trans_huge_or_clear_bad(pmd);
677#else
678 return 0;
5f6e8da7 679#endif
1a5a9906
AA
680}
681
e7bb4b6d
MG
682#ifndef CONFIG_NUMA_BALANCING
683/*
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.
690 */
691static inline int pte_protnone(pte_t pte)
692{
693 return 0;
694}
695
696static inline int pmd_protnone(pmd_t pmd)
697{
698 return 0;
699}
700#endif /* CONFIG_NUMA_BALANCING */
701
1a5a9906 702#endif /* CONFIG_MMU */
5f6e8da7 703
e61ce6ad
TK
704#ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
705int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
706int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
b9820d8f
TK
707int pud_clear_huge(pud_t *pud);
708int pmd_clear_huge(pmd_t *pmd);
e61ce6ad
TK
709#else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
710static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
711{
712 return 0;
713}
714static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
715{
716 return 0;
717}
b9820d8f
TK
718static inline int pud_clear_huge(pud_t *pud)
719{
720 return 0;
721}
722static inline int pmd_clear_huge(pmd_t *pmd)
723{
724 return 0;
725}
e61ce6ad
TK
726#endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
727
1da177e4
LT
728#endif /* !__ASSEMBLY__ */
729
40d158e6
AV
730#ifndef io_remap_pfn_range
731#define io_remap_pfn_range remap_pfn_range
732#endif
733
1da177e4 734#endif /* _ASM_GENERIC_PGTABLE_H */