blob: 0807ddf97b058fb04b1eedfd82c34b76e4253335 [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 Schaefere3ebcf62012-10-08 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 Schaefere3ebcf62012-10-08 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 Schaefere3ebcf62012-10-08 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_PAGE_TEST_AND_CLEAR_YOUNG
Martin Schwidefsky2d425522011-05-23 10:24:39 +0200212#define page_test_and_clear_young(pfn) (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213#endif
214
215#ifndef __HAVE_ARCH_PGD_OFFSET_GATE
216#define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
217#endif
218
David S. Miller0b0968a2006-06-01 17:47:25 -0700219#ifndef __HAVE_ARCH_MOVE_PTE
Nick Piggin8b1f3122005-09-27 21:45:18 -0700220#define move_pte(pte, prot, old_addr, new_addr) (pte)
Nick Piggin8b1f3122005-09-27 21:45:18 -0700221#endif
222
Rik van Riel2c3cf552012-10-09 15:31:12 +0200223#ifndef pte_accessible
224# define pte_accessible(pte) ((void)(pte),1)
225#endif
226
Shaohua Li61c77322010-08-16 09:16:55 +0800227#ifndef flush_tlb_fix_spurious_fault
228#define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
229#endif
230
Paul Mundt0634a632009-06-23 13:51:19 +0200231#ifndef pgprot_noncached
232#define pgprot_noncached(prot) (prot)
233#endif
234
venkatesh.pallipadi@intel.com2520bd32008-12-18 11:41:32 -0800235#ifndef pgprot_writecombine
236#define pgprot_writecombine pgprot_noncached
237#endif
238
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239/*
Hugh Dickins8f6c99c2005-04-19 13:29:17 -0700240 * When walking page tables, get the address of the next boundary,
241 * or the end address of the range if that comes earlier. Although no
242 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 */
244
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245#define pgd_addr_end(addr, end) \
246({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
247 (__boundary - 1 < (end) - 1)? __boundary: (end); \
248})
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250#ifndef pud_addr_end
251#define pud_addr_end(addr, end) \
252({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
253 (__boundary - 1 < (end) - 1)? __boundary: (end); \
254})
255#endif
256
257#ifndef pmd_addr_end
258#define pmd_addr_end(addr, end) \
259({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
260 (__boundary - 1 < (end) - 1)? __boundary: (end); \
261})
262#endif
263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264/*
265 * When walking page tables, we usually want to skip any p?d_none entries;
266 * and any p?d_bad entries - reporting the error before resetting to none.
267 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
268 */
269void pgd_clear_bad(pgd_t *);
270void pud_clear_bad(pud_t *);
271void pmd_clear_bad(pmd_t *);
272
273static inline int pgd_none_or_clear_bad(pgd_t *pgd)
274{
275 if (pgd_none(*pgd))
276 return 1;
277 if (unlikely(pgd_bad(*pgd))) {
278 pgd_clear_bad(pgd);
279 return 1;
280 }
281 return 0;
282}
283
284static inline int pud_none_or_clear_bad(pud_t *pud)
285{
286 if (pud_none(*pud))
287 return 1;
288 if (unlikely(pud_bad(*pud))) {
289 pud_clear_bad(pud);
290 return 1;
291 }
292 return 0;
293}
294
295static inline int pmd_none_or_clear_bad(pmd_t *pmd)
296{
297 if (pmd_none(*pmd))
298 return 1;
299 if (unlikely(pmd_bad(*pmd))) {
300 pmd_clear_bad(pmd);
301 return 1;
302 }
303 return 0;
304}
Greg Ungerer95352392007-08-10 13:01:20 -0700305
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -0700306static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
307 unsigned long addr,
308 pte_t *ptep)
309{
310 /*
311 * Get the current pte state, but zero it out to make it
312 * non-present, preventing the hardware from asynchronously
313 * updating it.
314 */
315 return ptep_get_and_clear(mm, addr, ptep);
316}
317
318static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
319 unsigned long addr,
320 pte_t *ptep, pte_t pte)
321{
322 /*
323 * The pte is non-present, so there's no hardware state to
324 * preserve.
325 */
326 set_pte_at(mm, addr, ptep, pte);
327}
328
329#ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
330/*
331 * Start a pte protection read-modify-write transaction, which
332 * protects against asynchronous hardware modifications to the pte.
333 * The intention is not to prevent the hardware from making pte
334 * updates, but to prevent any updates it may make from being lost.
335 *
336 * This does not protect against other software modifications of the
337 * pte; the appropriate pte lock must be held over the transation.
338 *
339 * Note that this interface is intended to be batchable, meaning that
340 * ptep_modify_prot_commit may not actually update the pte, but merely
341 * queue the update to be done at some later time. The update must be
342 * actually committed before the pte lock is released, however.
343 */
344static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
345 unsigned long addr,
346 pte_t *ptep)
347{
348 return __ptep_modify_prot_start(mm, addr, ptep);
349}
350
351/*
352 * Commit an update to a pte, leaving any hardware-controlled bits in
353 * the PTE unmodified.
354 */
355static inline void ptep_modify_prot_commit(struct mm_struct *mm,
356 unsigned long addr,
357 pte_t *ptep, pte_t pte)
358{
359 __ptep_modify_prot_commit(mm, addr, ptep, pte);
360}
361#endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
Sebastian Siewiorfe1a6872008-07-15 22:28:46 +0200362#endif /* CONFIG_MMU */
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -0700363
Greg Ungerer95352392007-08-10 13:01:20 -0700364/*
365 * A facility to provide lazy MMU batching. This allows PTE updates and
366 * page invalidations to be delayed until a call to leave lazy MMU mode
367 * is issued. Some architectures may benefit from doing this, and it is
368 * beneficial for both shadow and direct mode hypervisors, which may batch
369 * the PTE updates which happen during this window. Note that using this
370 * interface requires that read hazards be removed from the code. A read
371 * hazard could result in the direct mode hypervisor case, since the actual
372 * write to the page tables may not yet have taken place, so reads though
373 * a raw PTE pointer after it has been modified are not guaranteed to be
374 * up to date. This mode can only be entered and left under the protection of
375 * the page table locks for all page tables which may be modified. In the UP
376 * case, this is required so that preemption is disabled, and in the SMP case,
377 * it must synchronize the delayed page table writes properly on other CPUs.
378 */
379#ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
380#define arch_enter_lazy_mmu_mode() do {} while (0)
381#define arch_leave_lazy_mmu_mode() do {} while (0)
382#define arch_flush_lazy_mmu_mode() do {} while (0)
383#endif
384
385/*
Jeremy Fitzhardinge7fd7d832009-02-17 23:24:03 -0800386 * A facility to provide batching of the reload of page tables and
387 * other process state with the actual context switch code for
388 * paravirtualized guests. By convention, only one of the batched
389 * update (lazy) modes (CPU, MMU) should be active at any given time,
390 * entry should never be nested, and entry and exits should always be
391 * paired. This is for sanity of maintaining and reasoning about the
392 * kernel code. In this case, the exit (end of the context switch) is
393 * in architecture-specific code, and so doesn't need a generic
394 * definition.
Greg Ungerer95352392007-08-10 13:01:20 -0700395 */
Jeremy Fitzhardinge7fd7d832009-02-17 23:24:03 -0800396#ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800397#define arch_start_context_switch(prev) do {} while (0)
Greg Ungerer95352392007-08-10 13:01:20 -0700398#endif
399
Pavel Emelyanov0f8975e2013-07-03 15:01:20 -0700400#ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
401static inline int pte_soft_dirty(pte_t pte)
402{
403 return 0;
404}
405
406static inline int pmd_soft_dirty(pmd_t pmd)
407{
408 return 0;
409}
410
411static inline pte_t pte_mksoft_dirty(pte_t pte)
412{
413 return pte;
414}
415
416static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
417{
418 return pmd;
419}
Cyrill Gorcunov179ef712013-08-13 16:00:49 -0700420
421static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
422{
423 return pte;
424}
425
426static inline int pte_swp_soft_dirty(pte_t pte)
427{
428 return 0;
429}
430
431static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
432{
433 return pte;
434}
Cyrill Gorcunov41bb3472013-08-13 16:00:51 -0700435
436static inline pte_t pte_file_clear_soft_dirty(pte_t pte)
437{
438 return pte;
439}
440
441static inline pte_t pte_file_mksoft_dirty(pte_t pte)
442{
443 return pte;
444}
445
446static inline int pte_file_soft_dirty(pte_t pte)
447{
448 return 0;
449}
Pavel Emelyanov0f8975e2013-07-03 15:01:20 -0700450#endif
451
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800452#ifndef __HAVE_PFNMAP_TRACKING
453/*
Suresh Siddha5180da42012-10-08 16:28:29 -0700454 * Interfaces that can be used by architecture code to keep track of
455 * memory type of pfn mappings specified by the remap_pfn_range,
456 * vm_insert_pfn.
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800457 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700458
459/*
460 * track_pfn_remap is called when a _new_ pfn mapping is being established
461 * by remap_pfn_range() for physical range indicated by pfn and size.
462 */
463static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -0700464 unsigned long pfn, unsigned long addr,
465 unsigned long size)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800466{
467 return 0;
468}
469
470/*
Suresh Siddha5180da42012-10-08 16:28:29 -0700471 * track_pfn_insert is called when a _new_ single pfn is established
472 * by vm_insert_pfn().
473 */
474static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
475 unsigned long pfn)
476{
477 return 0;
478}
479
480/*
481 * track_pfn_copy is called when vma that is covering the pfnmap gets
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800482 * copied through copy_page_range().
483 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700484static inline int track_pfn_copy(struct vm_area_struct *vma)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800485{
486 return 0;
487}
488
489/*
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800490 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
491 * untrack can be called for a specific region indicated by pfn and size or
Suresh Siddha5180da42012-10-08 16:28:29 -0700492 * can be for the entire vma (in which case pfn, size are zero).
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800493 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700494static inline void untrack_pfn(struct vm_area_struct *vma,
495 unsigned long pfn, unsigned long size)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800496{
497}
498#else
Suresh Siddha5180da42012-10-08 16:28:29 -0700499extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -0700500 unsigned long pfn, unsigned long addr,
501 unsigned long size);
Suresh Siddha5180da42012-10-08 16:28:29 -0700502extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
503 unsigned long pfn);
504extern int track_pfn_copy(struct vm_area_struct *vma);
505extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
506 unsigned long size);
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800507#endif
508
Kirill A. Shutemov816422a2012-12-12 13:52:36 -0800509#ifdef __HAVE_COLOR_ZERO_PAGE
510static inline int is_zero_pfn(unsigned long pfn)
511{
512 extern unsigned long zero_pfn;
513 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
514 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
515}
516
Kirill A. Shutemov2f91ec82012-12-26 03:19:55 +0300517#define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
518
Kirill A. Shutemov816422a2012-12-12 13:52:36 -0800519#else
520static inline int is_zero_pfn(unsigned long pfn)
521{
522 extern unsigned long zero_pfn;
523 return pfn == zero_pfn;
524}
525
526static inline unsigned long my_zero_pfn(unsigned long addr)
527{
528 extern unsigned long zero_pfn;
529 return zero_pfn;
530}
531#endif
532
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700533#ifdef CONFIG_MMU
534
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800535#ifndef CONFIG_TRANSPARENT_HUGEPAGE
536static inline int pmd_trans_huge(pmd_t pmd)
537{
538 return 0;
539}
540static inline int pmd_trans_splitting(pmd_t pmd)
541{
542 return 0;
543}
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800544#ifndef __HAVE_ARCH_PMD_WRITE
545static inline int pmd_write(pmd_t pmd)
546{
547 BUG();
548 return 0;
549}
550#endif /* __HAVE_ARCH_PMD_WRITE */
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700551#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
552
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700553#ifndef pmd_read_atomic
554static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
555{
556 /*
557 * Depend on compiler for an atomic pmd read. NOTE: this is
558 * only going to work, if the pmdval_t isn't larger than
559 * an unsigned long.
560 */
561 return *pmdp;
562}
563#endif
564
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700565/*
566 * This function is meant to be used by sites walking pagetables with
567 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
568 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
569 * into a null pmd and the transhuge page fault can convert a null pmd
570 * into an hugepmd or into a regular pmd (if the hugepage allocation
571 * fails). While holding the mmap_sem in read mode the pmd becomes
572 * stable and stops changing under us only if it's not null and not a
573 * transhuge pmd. When those races occurs and this function makes a
574 * difference vs the standard pmd_none_or_clear_bad, the result is
575 * undefined so behaving like if the pmd was none is safe (because it
576 * can return none anyway). The compiler level barrier() is critically
577 * important to compute the two checks atomically on the same pmdval.
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700578 *
579 * For 32bit kernels with a 64bit large pmd_t this automatically takes
580 * care of reading the pmd atomically to avoid SMP race conditions
581 * against pmd_populate() when the mmap_sem is hold for reading by the
582 * caller (a special atomic read not done by "gcc" as in the generic
583 * version above, is also needed when THP is disabled because the page
584 * fault can populate the pmd from under us).
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700585 */
586static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
587{
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700588 pmd_t pmdval = pmd_read_atomic(pmd);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700589 /*
590 * The barrier will stabilize the pmdval in a register or on
591 * the stack so that it will stop changing under the code.
Andrea Arcangelie4eed032012-06-20 12:52:57 -0700592 *
593 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
594 * pmd_read_atomic is allowed to return a not atomic pmdval
595 * (for example pointing to an hugepage that has never been
596 * mapped in the pmd). The below checks will only care about
597 * the low part of the pmd with 32bit PAE x86 anyway, with the
598 * exception of pmd_none(). So the important thing is that if
599 * the low part of the pmd is found null, the high part will
600 * be also null or the pmd_none() check below would be
601 * confused.
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700602 */
603#ifdef CONFIG_TRANSPARENT_HUGEPAGE
604 barrier();
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800605#endif
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700606 if (pmd_none(pmdval))
607 return 1;
608 if (unlikely(pmd_bad(pmdval))) {
609 if (!pmd_trans_huge(pmdval))
610 pmd_clear_bad(pmd);
611 return 1;
612 }
613 return 0;
614}
615
616/*
617 * This is a noop if Transparent Hugepage Support is not built into
618 * the kernel. Otherwise it is equivalent to
619 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
620 * places that already verified the pmd is not none and they want to
621 * walk ptes while holding the mmap sem in read mode (write mode don't
622 * need this). If THP is not enabled, the pmd can't go away under the
623 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
624 * run a pmd_trans_unstable before walking the ptes after
625 * split_huge_page_pmd returns (because it may have run when the pmd
626 * become null, but then a page fault can map in a THP and not a
627 * regular page).
628 */
629static inline int pmd_trans_unstable(pmd_t *pmd)
630{
631#ifdef CONFIG_TRANSPARENT_HUGEPAGE
632 return pmd_none_or_trans_huge_or_clear_bad(pmd);
633#else
634 return 0;
635#endif
636}
637
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200638#ifdef CONFIG_NUMA_BALANCING
639#ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE
640/*
641 * _PAGE_NUMA works identical to _PAGE_PROTNONE (it's actually the
642 * same bit too). It's set only when _PAGE_PRESET is not set and it's
643 * never set if _PAGE_PRESENT is set.
644 *
645 * pte/pmd_present() returns true if pte/pmd_numa returns true. Page
646 * fault triggers on those regions if pte/pmd_numa returns true
647 * (because _PAGE_PRESENT is not set).
648 */
649#ifndef pte_numa
650static inline int pte_numa(pte_t pte)
651{
652 return (pte_flags(pte) &
653 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
654}
655#endif
656
657#ifndef pmd_numa
658static inline int pmd_numa(pmd_t pmd)
659{
660 return (pmd_flags(pmd) &
661 (_PAGE_NUMA|_PAGE_PRESENT)) == _PAGE_NUMA;
662}
663#endif
664
665/*
666 * pte/pmd_mknuma sets the _PAGE_ACCESSED bitflag automatically
667 * because they're called by the NUMA hinting minor page fault. If we
668 * wouldn't set the _PAGE_ACCESSED bitflag here, the TLB miss handler
669 * would be forced to set it later while filling the TLB after we
670 * return to userland. That would trigger a second write to memory
671 * that we optimize away by setting _PAGE_ACCESSED here.
672 */
673#ifndef pte_mknonnuma
674static inline pte_t pte_mknonnuma(pte_t pte)
675{
676 pte = pte_clear_flags(pte, _PAGE_NUMA);
677 return pte_set_flags(pte, _PAGE_PRESENT|_PAGE_ACCESSED);
678}
679#endif
680
681#ifndef pmd_mknonnuma
682static inline pmd_t pmd_mknonnuma(pmd_t pmd)
683{
684 pmd = pmd_clear_flags(pmd, _PAGE_NUMA);
685 return pmd_set_flags(pmd, _PAGE_PRESENT|_PAGE_ACCESSED);
686}
687#endif
688
689#ifndef pte_mknuma
690static inline pte_t pte_mknuma(pte_t pte)
691{
692 pte = pte_set_flags(pte, _PAGE_NUMA);
693 return pte_clear_flags(pte, _PAGE_PRESENT);
694}
695#endif
696
697#ifndef pmd_mknuma
698static inline pmd_t pmd_mknuma(pmd_t pmd)
699{
700 pmd = pmd_set_flags(pmd, _PAGE_NUMA);
701 return pmd_clear_flags(pmd, _PAGE_PRESENT);
702}
703#endif
704#else
705extern int pte_numa(pte_t pte);
706extern int pmd_numa(pmd_t pmd);
707extern pte_t pte_mknonnuma(pte_t pte);
708extern pmd_t pmd_mknonnuma(pmd_t pmd);
709extern pte_t pte_mknuma(pte_t pte);
710extern pmd_t pmd_mknuma(pmd_t pmd);
711#endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */
712#else
713static inline int pmd_numa(pmd_t pmd)
714{
715 return 0;
716}
717
718static inline int pte_numa(pte_t pte)
719{
720 return 0;
721}
722
723static inline pte_t pte_mknonnuma(pte_t pte)
724{
725 return pte;
726}
727
728static inline pmd_t pmd_mknonnuma(pmd_t pmd)
729{
730 return pmd;
731}
732
733static inline pte_t pte_mknuma(pte_t pte)
734{
735 return pte;
736}
737
738static inline pmd_t pmd_mknuma(pmd_t pmd)
739{
740 return pmd;
741}
742#endif /* CONFIG_NUMA_BALANCING */
743
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700744#endif /* CONFIG_MMU */
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746#endif /* !__ASSEMBLY__ */
747
Al Viro40d158e2013-05-11 12:13:10 -0400748#ifndef io_remap_pfn_range
749#define io_remap_pfn_range remap_pfn_range
750#endif
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752#endif /* _ASM_GENERIC_PGTABLE_H */