blob: db09234589409760b91cb9e696472a3e828476f9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef _ASM_GENERIC_PGTABLE_H
2#define _ASM_GENERIC_PGTABLE_H
3
Rusty Russell673eae82006-09-25 23:32:29 -07004#ifndef __ASSEMBLY__
Greg Ungerer95352392007-08-10 13:01:20 -07005#ifdef CONFIG_MMU
Rusty Russell673eae82006-09-25 23:32:29 -07006
Ben Hutchingsfbd71842011-02-27 05:41:35 +00007#include <linux/mm_types.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -05008#include <linux/bug.h>
Ben Hutchingsfbd71842011-02-27 05:41:35 +00009
Hugh Dickins6ee86302013-04-29 15:07:44 -070010/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
Andrea Arcangelie2cda322011-01-13 15:46:40 -080021extern 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#endif
31
32#ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
Andrea Arcangelie2cda322011-01-13 15:46:40 -080033static 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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#endif
71
72#ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
Andrea Arcangelie2cda322011-01-13 15:46:40 -080073int 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#endif
81
Linus Torvalds1da177e2005-04-16 15:20:36 -070082#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
Andrea Arcangelie2cda322011-01-13 15:46:40 -080083static 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;
Catalin Marinas2d28a222012-10-08 16:32:59 -0700100 pmd_clear(pmdp);
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800101 return pmd;
Nicolas Kaiser49b24d62011-06-15 15:08:34 -0700102}
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800103#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104#endif
105
Zachary Amsdena6003882005-09-03 15:55:04 -0700106#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800107static 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}
Zachary Amsdena6003882005-09-03 15:55:04 -0700115#endif
116
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700117/*
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
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800123static 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}
Zachary Amsdena6003882005-09-03 15:55:04 -0700130#endif
131
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132#ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800133extern 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
139extern pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
140 unsigned long address,
141 pmd_t *pmdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142#endif
143
144#ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
Tim Schmielau8c65b4a2005-11-07 00:59:43 -0800145struct mm_struct;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146static 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
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800153#ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
154#ifdef CONFIG_TRANSPARENT_HUGEPAGE
155static 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 */
162static 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
Chris Metcalf73636b12012-03-28 13:59:18 -0400171extern void pmdp_splitting_flush(struct vm_area_struct *vma,
172 unsigned long address, pmd_t *pmdp);
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800173#endif
174
Gerald Schaefere3ebcf642012-10-08 16:30:07 -0700175#ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
Aneesh Kumar K.V6b0b50b2013-06-05 17:14:02 -0700176extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
177 pgtable_t pgtable);
Gerald Schaefere3ebcf642012-10-08 16:30:07 -0700178#endif
179
180#ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
Aneesh Kumar K.V6b0b50b2013-06-05 17:14:02 -0700181extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
Gerald Schaefere3ebcf642012-10-08 16:30:07 -0700182#endif
183
Gerald Schaefer46dcde72012-10-08 16:30:09 -0700184#ifndef __HAVE_ARCH_PMDP_INVALIDATE
185extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
186 pmd_t *pmdp);
187#endif
188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189#ifndef __HAVE_ARCH_PTE_SAME
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800190static 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_PMD_SAME
197#ifdef CONFIG_TRANSPARENT_HUGEPAGE
198static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
199{
200 return pmd_val(pmd_a) == pmd_val(pmd_b);
201}
202#else /* CONFIG_TRANSPARENT_HUGEPAGE */
203static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
204{
205 BUG();
206 return 0;
207}
208#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209#endif
210
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211#ifndef __HAVE_ARCH_PGD_OFFSET_GATE
212#define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
213#endif
214
David S. Miller0b0968a2006-06-01 17:47:25 -0700215#ifndef __HAVE_ARCH_MOVE_PTE
Nick Piggin8b1f3122005-09-27 21:45:18 -0700216#define move_pte(pte, prot, old_addr, new_addr) (pte)
Nick Piggin8b1f3122005-09-27 21:45:18 -0700217#endif
218
Rik van Riel2c3cf552012-10-09 15:31:12 +0200219#ifndef pte_accessible
Rik van Riel20841402013-12-18 17:08:44 -0800220# define pte_accessible(mm, pte) ((void)(pte), 1)
Rik van Riel2c3cf552012-10-09 15:31:12 +0200221#endif
222
Shaohua Li61c77322010-08-16 09:16:55 +0800223#ifndef flush_tlb_fix_spurious_fault
224#define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
225#endif
226
Paul Mundt0634a632009-06-23 13:51:19 +0200227#ifndef pgprot_noncached
228#define pgprot_noncached(prot) (prot)
229#endif
230
venkatesh.pallipadi@intel.com2520bd32008-12-18 11:41:32 -0800231#ifndef pgprot_writecombine
232#define pgprot_writecombine pgprot_noncached
233#endif
234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235/*
Hugh Dickins8f6c99c2005-04-19 13:29:17 -0700236 * When walking page tables, get the address of the next boundary,
237 * or the end address of the range if that comes earlier. Although no
238 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 */
240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241#define pgd_addr_end(addr, end) \
242({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
243 (__boundary - 1 < (end) - 1)? __boundary: (end); \
244})
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
246#ifndef pud_addr_end
247#define pud_addr_end(addr, end) \
248({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
249 (__boundary - 1 < (end) - 1)? __boundary: (end); \
250})
251#endif
252
253#ifndef pmd_addr_end
254#define pmd_addr_end(addr, end) \
255({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
256 (__boundary - 1 < (end) - 1)? __boundary: (end); \
257})
258#endif
259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260/*
261 * When walking page tables, we usually want to skip any p?d_none entries;
262 * and any p?d_bad entries - reporting the error before resetting to none.
263 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
264 */
265void pgd_clear_bad(pgd_t *);
266void pud_clear_bad(pud_t *);
267void pmd_clear_bad(pmd_t *);
268
269static inline int pgd_none_or_clear_bad(pgd_t *pgd)
270{
271 if (pgd_none(*pgd))
272 return 1;
273 if (unlikely(pgd_bad(*pgd))) {
274 pgd_clear_bad(pgd);
275 return 1;
276 }
277 return 0;
278}
279
280static inline int pud_none_or_clear_bad(pud_t *pud)
281{
282 if (pud_none(*pud))
283 return 1;
284 if (unlikely(pud_bad(*pud))) {
285 pud_clear_bad(pud);
286 return 1;
287 }
288 return 0;
289}
290
291static inline int pmd_none_or_clear_bad(pmd_t *pmd)
292{
293 if (pmd_none(*pmd))
294 return 1;
295 if (unlikely(pmd_bad(*pmd))) {
296 pmd_clear_bad(pmd);
297 return 1;
298 }
299 return 0;
300}
Greg Ungerer95352392007-08-10 13:01:20 -0700301
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -0700302static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
303 unsigned long addr,
304 pte_t *ptep)
305{
306 /*
307 * Get the current pte state, but zero it out to make it
308 * non-present, preventing the hardware from asynchronously
309 * updating it.
310 */
311 return ptep_get_and_clear(mm, addr, ptep);
312}
313
314static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
315 unsigned long addr,
316 pte_t *ptep, pte_t pte)
317{
318 /*
319 * The pte is non-present, so there's no hardware state to
320 * preserve.
321 */
322 set_pte_at(mm, addr, ptep, pte);
323}
324
325#ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
326/*
327 * Start a pte protection read-modify-write transaction, which
328 * protects against asynchronous hardware modifications to the pte.
329 * The intention is not to prevent the hardware from making pte
330 * updates, but to prevent any updates it may make from being lost.
331 *
332 * This does not protect against other software modifications of the
333 * pte; the appropriate pte lock must be held over the transation.
334 *
335 * Note that this interface is intended to be batchable, meaning that
336 * ptep_modify_prot_commit may not actually update the pte, but merely
337 * queue the update to be done at some later time. The update must be
338 * actually committed before the pte lock is released, however.
339 */
340static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
341 unsigned long addr,
342 pte_t *ptep)
343{
344 return __ptep_modify_prot_start(mm, addr, ptep);
345}
346
347/*
348 * Commit an update to a pte, leaving any hardware-controlled bits in
349 * the PTE unmodified.
350 */
351static inline void ptep_modify_prot_commit(struct mm_struct *mm,
352 unsigned long addr,
353 pte_t *ptep, pte_t pte)
354{
355 __ptep_modify_prot_commit(mm, addr, ptep, pte);
356}
357#endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
Sebastian Siewiorfe1a6872008-07-15 22:28:46 +0200358#endif /* CONFIG_MMU */
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -0700359
Greg Ungerer95352392007-08-10 13:01:20 -0700360/*
361 * A facility to provide lazy MMU batching. This allows PTE updates and
362 * page invalidations to be delayed until a call to leave lazy MMU mode
363 * is issued. Some architectures may benefit from doing this, and it is
364 * beneficial for both shadow and direct mode hypervisors, which may batch
365 * the PTE updates which happen during this window. Note that using this
366 * interface requires that read hazards be removed from the code. A read
367 * hazard could result in the direct mode hypervisor case, since the actual
368 * write to the page tables may not yet have taken place, so reads though
369 * a raw PTE pointer after it has been modified are not guaranteed to be
370 * up to date. This mode can only be entered and left under the protection of
371 * the page table locks for all page tables which may be modified. In the UP
372 * case, this is required so that preemption is disabled, and in the SMP case,
373 * it must synchronize the delayed page table writes properly on other CPUs.
374 */
375#ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
376#define arch_enter_lazy_mmu_mode() do {} while (0)
377#define arch_leave_lazy_mmu_mode() do {} while (0)
378#define arch_flush_lazy_mmu_mode() do {} while (0)
379#endif
380
381/*
Jeremy Fitzhardinge7fd7d832009-02-17 23:24:03 -0800382 * A facility to provide batching of the reload of page tables and
383 * other process state with the actual context switch code for
384 * paravirtualized guests. By convention, only one of the batched
385 * update (lazy) modes (CPU, MMU) should be active at any given time,
386 * entry should never be nested, and entry and exits should always be
387 * paired. This is for sanity of maintaining and reasoning about the
388 * kernel code. In this case, the exit (end of the context switch) is
389 * in architecture-specific code, and so doesn't need a generic
390 * definition.
Greg Ungerer95352392007-08-10 13:01:20 -0700391 */
Jeremy Fitzhardinge7fd7d832009-02-17 23:24:03 -0800392#ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800393#define arch_start_context_switch(prev) do {} while (0)
Greg Ungerer95352392007-08-10 13:01:20 -0700394#endif
395
Pavel Emelyanov0f8975e2013-07-03 15:01:20 -0700396#ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
397static inline int pte_soft_dirty(pte_t pte)
398{
399 return 0;
400}
401
402static inline int pmd_soft_dirty(pmd_t pmd)
403{
404 return 0;
405}
406
407static inline pte_t pte_mksoft_dirty(pte_t pte)
408{
409 return pte;
410}
411
412static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
413{
414 return pmd;
415}
Cyrill Gorcunov179ef712013-08-13 16:00:49 -0700416
417static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
418{
419 return pte;
420}
421
422static inline int pte_swp_soft_dirty(pte_t pte)
423{
424 return 0;
425}
426
427static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
428{
429 return pte;
430}
Cyrill Gorcunov41bb3472013-08-13 16:00:51 -0700431
432static inline pte_t pte_file_clear_soft_dirty(pte_t pte)
433{
434 return pte;
435}
436
437static inline pte_t pte_file_mksoft_dirty(pte_t pte)
438{
439 return pte;
440}
441
442static inline int pte_file_soft_dirty(pte_t pte)
443{
444 return 0;
445}
Pavel Emelyanov0f8975e2013-07-03 15:01:20 -0700446#endif
447
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800448#ifndef __HAVE_PFNMAP_TRACKING
449/*
Suresh Siddha5180da42012-10-08 16:28:29 -0700450 * Interfaces that can be used by architecture code to keep track of
451 * memory type of pfn mappings specified by the remap_pfn_range,
452 * vm_insert_pfn.
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800453 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700454
455/*
456 * track_pfn_remap is called when a _new_ pfn mapping is being established
457 * by remap_pfn_range() for physical range indicated by pfn and size.
458 */
459static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -0700460 unsigned long pfn, unsigned long addr,
461 unsigned long size)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800462{
463 return 0;
464}
465
466/*
Suresh Siddha5180da42012-10-08 16:28:29 -0700467 * track_pfn_insert is called when a _new_ single pfn is established
468 * by vm_insert_pfn().
469 */
470static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
471 unsigned long pfn)
472{
473 return 0;
474}
475
476/*
477 * track_pfn_copy is called when vma that is covering the pfnmap gets
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800478 * copied through copy_page_range().
479 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700480static inline int track_pfn_copy(struct vm_area_struct *vma)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800481{
482 return 0;
483}
484
485/*
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800486 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
487 * untrack can be called for a specific region indicated by pfn and size or
Suresh Siddha5180da42012-10-08 16:28:29 -0700488 * can be for the entire vma (in which case pfn, size are zero).
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800489 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700490static inline void untrack_pfn(struct vm_area_struct *vma,
491 unsigned long pfn, unsigned long size)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800492{
493}
494#else
Suresh Siddha5180da42012-10-08 16:28:29 -0700495extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -0700496 unsigned long pfn, unsigned long addr,
497 unsigned long size);
Suresh Siddha5180da42012-10-08 16:28:29 -0700498extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
499 unsigned long pfn);
500extern int track_pfn_copy(struct vm_area_struct *vma);
501extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
502 unsigned long size);
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800503#endif
504
Kirill A. Shutemov816422a2012-12-12 13:52:36 -0800505#ifdef __HAVE_COLOR_ZERO_PAGE
506static inline int is_zero_pfn(unsigned long pfn)
507{
508 extern unsigned long zero_pfn;
509 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
510 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
511}
512
Kirill A. Shutemov2f91ec82012-12-26 03:19:55 +0300513#define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
514
Kirill A. Shutemov816422a2012-12-12 13:52:36 -0800515#else
516static inline int is_zero_pfn(unsigned long pfn)
517{
518 extern unsigned long zero_pfn;
519 return pfn == zero_pfn;
520}
521
522static inline unsigned long my_zero_pfn(unsigned long addr)
523{
524 extern unsigned long zero_pfn;
525 return zero_pfn;
526}
527#endif
528
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700529#ifdef CONFIG_MMU
530
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800531#ifndef CONFIG_TRANSPARENT_HUGEPAGE
532static inline int pmd_trans_huge(pmd_t pmd)
533{
534 return 0;
535}
536static inline int pmd_trans_splitting(pmd_t pmd)
537{
538 return 0;
539}
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800540#ifndef __HAVE_ARCH_PMD_WRITE
541static inline int pmd_write(pmd_t pmd)
542{
543 BUG();
544 return 0;
545}
546#endif /* __HAVE_ARCH_PMD_WRITE */
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700547#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
548
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700549#ifndef pmd_read_atomic
550static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
551{
552 /*
553 * Depend on compiler for an atomic pmd read. NOTE: this is
554 * only going to work, if the pmdval_t isn't larger than
555 * an unsigned long.
556 */
557 return *pmdp;
558}
559#endif
560
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700561/*
562 * This function is meant to be used by sites walking pagetables with
563 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
564 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
565 * into a null pmd and the transhuge page fault can convert a null pmd
566 * into an hugepmd or into a regular pmd (if the hugepage allocation
567 * fails). While holding the mmap_sem in read mode the pmd becomes
568 * stable and stops changing under us only if it's not null and not a
569 * transhuge pmd. When those races occurs and this function makes a
570 * difference vs the standard pmd_none_or_clear_bad, the result is
571 * undefined so behaving like if the pmd was none is safe (because it
572 * can return none anyway). The compiler level barrier() is critically
573 * important to compute the two checks atomically on the same pmdval.
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700574 *
575 * For 32bit kernels with a 64bit large pmd_t this automatically takes
576 * care of reading the pmd atomically to avoid SMP race conditions
577 * against pmd_populate() when the mmap_sem is hold for reading by the
578 * caller (a special atomic read not done by "gcc" as in the generic
579 * version above, is also needed when THP is disabled because the page
580 * fault can populate the pmd from under us).
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700581 */
582static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
583{
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700584 pmd_t pmdval = pmd_read_atomic(pmd);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700585 /*
586 * The barrier will stabilize the pmdval in a register or on
587 * the stack so that it will stop changing under the code.
Andrea Arcangelie4eed032012-06-20 12:52:57 -0700588 *
589 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
590 * pmd_read_atomic is allowed to return a not atomic pmdval
591 * (for example pointing to an hugepage that has never been
592 * mapped in the pmd). The below checks will only care about
593 * the low part of the pmd with 32bit PAE x86 anyway, with the
594 * exception of pmd_none(). So the important thing is that if
595 * the low part of the pmd is found null, the high part will
596 * be also null or the pmd_none() check below would be
597 * confused.
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700598 */
599#ifdef CONFIG_TRANSPARENT_HUGEPAGE
600 barrier();
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800601#endif
Kirill A. Shutemovee536642013-12-20 15:10:03 +0200602 if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700603 return 1;
604 if (unlikely(pmd_bad(pmdval))) {
Kirill A. Shutemovee536642013-12-20 15:10:03 +0200605 pmd_clear_bad(pmd);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700606 return 1;
607 }
608 return 0;
609}
610
611/*
612 * This is a noop if Transparent Hugepage Support is not built into
613 * the kernel. Otherwise it is equivalent to
614 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
615 * places that already verified the pmd is not none and they want to
616 * walk ptes while holding the mmap sem in read mode (write mode don't
617 * need this). If THP is not enabled, the pmd can't go away under the
618 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
619 * run a pmd_trans_unstable before walking the ptes after
620 * split_huge_page_pmd returns (because it may have run when the pmd
621 * become null, but then a page fault can map in a THP and not a
622 * regular page).
623 */
624static inline int pmd_trans_unstable(pmd_t *pmd)
625{
626#ifdef CONFIG_TRANSPARENT_HUGEPAGE
627 return pmd_none_or_trans_huge_or_clear_bad(pmd);
628#else
629 return 0;
630#endif
631}
632
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200633#ifdef CONFIG_NUMA_BALANCING
634#ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE
635/*
636 * _PAGE_NUMA works identical to _PAGE_PROTNONE (it's actually the
637 * same bit too). It's set only when _PAGE_PRESET is not set and it's
638 * never set if _PAGE_PRESENT is set.
639 *
640 * pte/pmd_present() returns true if pte/pmd_numa returns true. Page
641 * fault triggers on those regions if pte/pmd_numa returns true
642 * (because _PAGE_PRESENT is not set).
643 */
644#ifndef pte_numa
645static inline int pte_numa(pte_t pte)
646{
647 return (pte_flags(pte) &
648 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
649}
650#endif
651
652#ifndef pmd_numa
653static inline int pmd_numa(pmd_t pmd)
654{
655 return (pmd_flags(pmd) &
656 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
657}
658#endif
659
660/*
661 * pte/pmd_mknuma sets the _PAGE_ACCESSED bitflag automatically
662 * because they're called by the NUMA hinting minor page fault. If we
663 * wouldn't set the _PAGE_ACCESSED bitflag here, the TLB miss handler
664 * would be forced to set it later while filling the TLB after we
665 * return to userland. That would trigger a second write to memory
666 * that we optimize away by setting _PAGE_ACCESSED here.
667 */
668#ifndef pte_mknonnuma
669static inline pte_t pte_mknonnuma(pte_t pte)
670{
671 pte = pte_clear_flags(pte, _PAGE_NUMA);
672 return pte_set_flags(pte, _PAGE_PRESENT|_PAGE_ACCESSED);
673}
674#endif
675
676#ifndef pmd_mknonnuma
677static inline pmd_t pmd_mknonnuma(pmd_t pmd)
678{
679 pmd = pmd_clear_flags(pmd, _PAGE_NUMA);
680 return pmd_set_flags(pmd, _PAGE_PRESENT|_PAGE_ACCESSED);
681}
682#endif
683
684#ifndef pte_mknuma
685static inline pte_t pte_mknuma(pte_t pte)
686{
687 pte = pte_set_flags(pte, _PAGE_NUMA);
688 return pte_clear_flags(pte, _PAGE_PRESENT);
689}
690#endif
691
692#ifndef pmd_mknuma
693static inline pmd_t pmd_mknuma(pmd_t pmd)
694{
695 pmd = pmd_set_flags(pmd, _PAGE_NUMA);
696 return pmd_clear_flags(pmd, _PAGE_PRESENT);
697}
698#endif
699#else
700extern int pte_numa(pte_t pte);
701extern int pmd_numa(pmd_t pmd);
702extern pte_t pte_mknonnuma(pte_t pte);
703extern pmd_t pmd_mknonnuma(pmd_t pmd);
704extern pte_t pte_mknuma(pte_t pte);
705extern pmd_t pmd_mknuma(pmd_t pmd);
706#endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */
707#else
708static inline int pmd_numa(pmd_t pmd)
709{
710 return 0;
711}
712
713static inline int pte_numa(pte_t pte)
714{
715 return 0;
716}
717
718static inline pte_t pte_mknonnuma(pte_t pte)
719{
720 return pte;
721}
722
723static inline pmd_t pmd_mknonnuma(pmd_t pmd)
724{
725 return pmd;
726}
727
728static inline pte_t pte_mknuma(pte_t pte)
729{
730 return pte;
731}
732
733static inline pmd_t pmd_mknuma(pmd_t pmd)
734{
735 return pmd;
736}
737#endif /* CONFIG_NUMA_BALANCING */
738
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700739#endif /* CONFIG_MMU */
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800740
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741#endif /* !__ASSEMBLY__ */
742
Al Viro40d158e2013-05-11 12:13:10 -0400743#ifndef io_remap_pfn_range
744#define io_remap_pfn_range remap_pfn_range
745#endif
746
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747#endif /* _ASM_GENERIC_PGTABLE_H */