blob: 977e545a64c33ab0c0d965b881b7870b2fa31889 [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
Konstantin Weitz45961722013-04-17 13:59:32 +0200196#ifndef __HAVE_ARCH_PTE_UNUSED
197/*
198 * Some architectures provide facilities to virtualization guests
199 * so that they can flag allocated pages as unused. This allows the
200 * host to transparently reclaim unused pages. This function returns
201 * whether the pte's page is unused.
202 */
203static inline int pte_unused(pte_t pte)
204{
205 return 0;
206}
207#endif
208
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800209#ifndef __HAVE_ARCH_PMD_SAME
210#ifdef CONFIG_TRANSPARENT_HUGEPAGE
211static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
212{
213 return pmd_val(pmd_a) == pmd_val(pmd_b);
214}
215#else /* CONFIG_TRANSPARENT_HUGEPAGE */
216static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
217{
218 BUG();
219 return 0;
220}
221#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222#endif
223
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224#ifndef __HAVE_ARCH_PGD_OFFSET_GATE
225#define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
226#endif
227
David S. Miller0b0968a2006-06-01 17:47:25 -0700228#ifndef __HAVE_ARCH_MOVE_PTE
Nick Piggin8b1f3122005-09-27 21:45:18 -0700229#define move_pte(pte, prot, old_addr, new_addr) (pte)
Nick Piggin8b1f3122005-09-27 21:45:18 -0700230#endif
231
Rik van Riel2c3cf552012-10-09 15:31:12 +0200232#ifndef pte_accessible
Rik van Riel20841402013-12-18 17:08:44 -0800233# define pte_accessible(mm, pte) ((void)(pte), 1)
Rik van Riel2c3cf552012-10-09 15:31:12 +0200234#endif
235
Mel Gormanc46a7c82014-06-04 16:06:30 -0700236#ifndef pte_present_nonuma
237#define pte_present_nonuma(pte) pte_present(pte)
238#endif
239
Shaohua Li61c77322010-08-16 09:16:55 +0800240#ifndef flush_tlb_fix_spurious_fault
241#define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
242#endif
243
Paul Mundt0634a632009-06-23 13:51:19 +0200244#ifndef pgprot_noncached
245#define pgprot_noncached(prot) (prot)
246#endif
247
venkatesh.pallipadi@intel.com2520bd32008-12-18 11:41:32 -0800248#ifndef pgprot_writecombine
249#define pgprot_writecombine pgprot_noncached
250#endif
251
Liviu Dudau8b921ac2014-09-29 15:29:30 +0100252#ifndef pgprot_device
253#define pgprot_device pgprot_noncached
254#endif
255
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256/*
Hugh Dickins8f6c99c2005-04-19 13:29:17 -0700257 * When walking page tables, get the address of the next boundary,
258 * or the end address of the range if that comes earlier. Although no
259 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 */
261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262#define pgd_addr_end(addr, end) \
263({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
264 (__boundary - 1 < (end) - 1)? __boundary: (end); \
265})
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
267#ifndef pud_addr_end
268#define pud_addr_end(addr, end) \
269({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
270 (__boundary - 1 < (end) - 1)? __boundary: (end); \
271})
272#endif
273
274#ifndef pmd_addr_end
275#define pmd_addr_end(addr, end) \
276({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
277 (__boundary - 1 < (end) - 1)? __boundary: (end); \
278})
279#endif
280
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281/*
282 * When walking page tables, we usually want to skip any p?d_none entries;
283 * and any p?d_bad entries - reporting the error before resetting to none.
284 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
285 */
286void pgd_clear_bad(pgd_t *);
287void pud_clear_bad(pud_t *);
288void pmd_clear_bad(pmd_t *);
289
290static inline int pgd_none_or_clear_bad(pgd_t *pgd)
291{
292 if (pgd_none(*pgd))
293 return 1;
294 if (unlikely(pgd_bad(*pgd))) {
295 pgd_clear_bad(pgd);
296 return 1;
297 }
298 return 0;
299}
300
301static inline int pud_none_or_clear_bad(pud_t *pud)
302{
303 if (pud_none(*pud))
304 return 1;
305 if (unlikely(pud_bad(*pud))) {
306 pud_clear_bad(pud);
307 return 1;
308 }
309 return 0;
310}
311
312static inline int pmd_none_or_clear_bad(pmd_t *pmd)
313{
314 if (pmd_none(*pmd))
315 return 1;
316 if (unlikely(pmd_bad(*pmd))) {
317 pmd_clear_bad(pmd);
318 return 1;
319 }
320 return 0;
321}
Greg Ungerer95352392007-08-10 13:01:20 -0700322
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -0700323static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
324 unsigned long addr,
325 pte_t *ptep)
326{
327 /*
328 * Get the current pte state, but zero it out to make it
329 * non-present, preventing the hardware from asynchronously
330 * updating it.
331 */
332 return ptep_get_and_clear(mm, addr, ptep);
333}
334
335static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
336 unsigned long addr,
337 pte_t *ptep, pte_t pte)
338{
339 /*
340 * The pte is non-present, so there's no hardware state to
341 * preserve.
342 */
343 set_pte_at(mm, addr, ptep, pte);
344}
345
346#ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
347/*
348 * Start a pte protection read-modify-write transaction, which
349 * protects against asynchronous hardware modifications to the pte.
350 * The intention is not to prevent the hardware from making pte
351 * updates, but to prevent any updates it may make from being lost.
352 *
353 * This does not protect against other software modifications of the
354 * pte; the appropriate pte lock must be held over the transation.
355 *
356 * Note that this interface is intended to be batchable, meaning that
357 * ptep_modify_prot_commit may not actually update the pte, but merely
358 * queue the update to be done at some later time. The update must be
359 * actually committed before the pte lock is released, however.
360 */
361static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
362 unsigned long addr,
363 pte_t *ptep)
364{
365 return __ptep_modify_prot_start(mm, addr, ptep);
366}
367
368/*
369 * Commit an update to a pte, leaving any hardware-controlled bits in
370 * the PTE unmodified.
371 */
372static inline void ptep_modify_prot_commit(struct mm_struct *mm,
373 unsigned long addr,
374 pte_t *ptep, pte_t pte)
375{
376 __ptep_modify_prot_commit(mm, addr, ptep, pte);
377}
378#endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
Sebastian Siewiorfe1a6872008-07-15 22:28:46 +0200379#endif /* CONFIG_MMU */
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -0700380
Greg Ungerer95352392007-08-10 13:01:20 -0700381/*
382 * A facility to provide lazy MMU batching. This allows PTE updates and
383 * page invalidations to be delayed until a call to leave lazy MMU mode
384 * is issued. Some architectures may benefit from doing this, and it is
385 * beneficial for both shadow and direct mode hypervisors, which may batch
386 * the PTE updates which happen during this window. Note that using this
387 * interface requires that read hazards be removed from the code. A read
388 * hazard could result in the direct mode hypervisor case, since the actual
389 * write to the page tables may not yet have taken place, so reads though
390 * a raw PTE pointer after it has been modified are not guaranteed to be
391 * up to date. This mode can only be entered and left under the protection of
392 * the page table locks for all page tables which may be modified. In the UP
393 * case, this is required so that preemption is disabled, and in the SMP case,
394 * it must synchronize the delayed page table writes properly on other CPUs.
395 */
396#ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
397#define arch_enter_lazy_mmu_mode() do {} while (0)
398#define arch_leave_lazy_mmu_mode() do {} while (0)
399#define arch_flush_lazy_mmu_mode() do {} while (0)
400#endif
401
402/*
Jeremy Fitzhardinge7fd7d832009-02-17 23:24:03 -0800403 * A facility to provide batching of the reload of page tables and
404 * other process state with the actual context switch code for
405 * paravirtualized guests. By convention, only one of the batched
406 * update (lazy) modes (CPU, MMU) should be active at any given time,
407 * entry should never be nested, and entry and exits should always be
408 * paired. This is for sanity of maintaining and reasoning about the
409 * kernel code. In this case, the exit (end of the context switch) is
410 * in architecture-specific code, and so doesn't need a generic
411 * definition.
Greg Ungerer95352392007-08-10 13:01:20 -0700412 */
Jeremy Fitzhardinge7fd7d832009-02-17 23:24:03 -0800413#ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800414#define arch_start_context_switch(prev) do {} while (0)
Greg Ungerer95352392007-08-10 13:01:20 -0700415#endif
416
Pavel Emelyanov0f8975e2013-07-03 15:01:20 -0700417#ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
418static inline int pte_soft_dirty(pte_t pte)
419{
420 return 0;
421}
422
423static inline int pmd_soft_dirty(pmd_t pmd)
424{
425 return 0;
426}
427
428static inline pte_t pte_mksoft_dirty(pte_t pte)
429{
430 return pte;
431}
432
433static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
434{
435 return pmd;
436}
Cyrill Gorcunov179ef712013-08-13 16:00:49 -0700437
438static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
439{
440 return pte;
441}
442
443static inline int pte_swp_soft_dirty(pte_t pte)
444{
445 return 0;
446}
447
448static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
449{
450 return pte;
451}
Cyrill Gorcunov41bb3472013-08-13 16:00:51 -0700452
453static inline pte_t pte_file_clear_soft_dirty(pte_t pte)
454{
455 return pte;
456}
457
458static inline pte_t pte_file_mksoft_dirty(pte_t pte)
459{
460 return pte;
461}
462
463static inline int pte_file_soft_dirty(pte_t pte)
464{
465 return 0;
466}
Pavel Emelyanov0f8975e2013-07-03 15:01:20 -0700467#endif
468
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800469#ifndef __HAVE_PFNMAP_TRACKING
470/*
Suresh Siddha5180da42012-10-08 16:28:29 -0700471 * Interfaces that can be used by architecture code to keep track of
472 * memory type of pfn mappings specified by the remap_pfn_range,
473 * vm_insert_pfn.
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800474 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700475
476/*
477 * track_pfn_remap is called when a _new_ pfn mapping is being established
478 * by remap_pfn_range() for physical range indicated by pfn and size.
479 */
480static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -0700481 unsigned long pfn, unsigned long addr,
482 unsigned long size)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800483{
484 return 0;
485}
486
487/*
Suresh Siddha5180da42012-10-08 16:28:29 -0700488 * track_pfn_insert is called when a _new_ single pfn is established
489 * by vm_insert_pfn().
490 */
491static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
492 unsigned long pfn)
493{
494 return 0;
495}
496
497/*
498 * track_pfn_copy is called when vma that is covering the pfnmap gets
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800499 * copied through copy_page_range().
500 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700501static inline int track_pfn_copy(struct vm_area_struct *vma)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800502{
503 return 0;
504}
505
506/*
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800507 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
508 * untrack can be called for a specific region indicated by pfn and size or
Suresh Siddha5180da42012-10-08 16:28:29 -0700509 * can be for the entire vma (in which case pfn, size are zero).
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800510 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700511static inline void untrack_pfn(struct vm_area_struct *vma,
512 unsigned long pfn, unsigned long size)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800513{
514}
515#else
Suresh Siddha5180da42012-10-08 16:28:29 -0700516extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -0700517 unsigned long pfn, unsigned long addr,
518 unsigned long size);
Suresh Siddha5180da42012-10-08 16:28:29 -0700519extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
520 unsigned long pfn);
521extern int track_pfn_copy(struct vm_area_struct *vma);
522extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
523 unsigned long size);
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800524#endif
525
Kirill A. Shutemov816422a2012-12-12 13:52:36 -0800526#ifdef __HAVE_COLOR_ZERO_PAGE
527static inline int is_zero_pfn(unsigned long pfn)
528{
529 extern unsigned long zero_pfn;
530 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
531 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
532}
533
Kirill A. Shutemov2f91ec82012-12-26 03:19:55 +0300534#define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
535
Kirill A. Shutemov816422a2012-12-12 13:52:36 -0800536#else
537static inline int is_zero_pfn(unsigned long pfn)
538{
539 extern unsigned long zero_pfn;
540 return pfn == zero_pfn;
541}
542
543static inline unsigned long my_zero_pfn(unsigned long addr)
544{
545 extern unsigned long zero_pfn;
546 return zero_pfn;
547}
548#endif
549
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700550#ifdef CONFIG_MMU
551
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800552#ifndef CONFIG_TRANSPARENT_HUGEPAGE
553static inline int pmd_trans_huge(pmd_t pmd)
554{
555 return 0;
556}
557static inline int pmd_trans_splitting(pmd_t pmd)
558{
559 return 0;
560}
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800561#ifndef __HAVE_ARCH_PMD_WRITE
562static inline int pmd_write(pmd_t pmd)
563{
564 BUG();
565 return 0;
566}
567#endif /* __HAVE_ARCH_PMD_WRITE */
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700568#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
569
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700570#ifndef pmd_read_atomic
571static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
572{
573 /*
574 * Depend on compiler for an atomic pmd read. NOTE: this is
575 * only going to work, if the pmdval_t isn't larger than
576 * an unsigned long.
577 */
578 return *pmdp;
579}
580#endif
581
Aneesh Kumar K.Vb3084f42014-01-13 11:34:24 +0530582#ifndef pmd_move_must_withdraw
583static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
584 spinlock_t *old_pmd_ptl)
585{
586 /*
587 * With split pmd lock we also need to move preallocated
588 * PTE page table if new_pmd is on different PMD page table.
589 */
590 return new_pmd_ptl != old_pmd_ptl;
591}
592#endif
593
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700594/*
595 * This function is meant to be used by sites walking pagetables with
596 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
597 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
598 * into a null pmd and the transhuge page fault can convert a null pmd
599 * into an hugepmd or into a regular pmd (if the hugepage allocation
600 * fails). While holding the mmap_sem in read mode the pmd becomes
601 * stable and stops changing under us only if it's not null and not a
602 * transhuge pmd. When those races occurs and this function makes a
603 * difference vs the standard pmd_none_or_clear_bad, the result is
604 * undefined so behaving like if the pmd was none is safe (because it
605 * can return none anyway). The compiler level barrier() is critically
606 * important to compute the two checks atomically on the same pmdval.
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700607 *
608 * For 32bit kernels with a 64bit large pmd_t this automatically takes
609 * care of reading the pmd atomically to avoid SMP race conditions
610 * against pmd_populate() when the mmap_sem is hold for reading by the
611 * caller (a special atomic read not done by "gcc" as in the generic
612 * version above, is also needed when THP is disabled because the page
613 * fault can populate the pmd from under us).
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700614 */
615static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
616{
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700617 pmd_t pmdval = pmd_read_atomic(pmd);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700618 /*
619 * The barrier will stabilize the pmdval in a register or on
620 * the stack so that it will stop changing under the code.
Andrea Arcangelie4eed032012-06-20 12:52:57 -0700621 *
622 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
623 * pmd_read_atomic is allowed to return a not atomic pmdval
624 * (for example pointing to an hugepage that has never been
625 * mapped in the pmd). The below checks will only care about
626 * the low part of the pmd with 32bit PAE x86 anyway, with the
627 * exception of pmd_none(). So the important thing is that if
628 * the low part of the pmd is found null, the high part will
629 * be also null or the pmd_none() check below would be
630 * confused.
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700631 */
632#ifdef CONFIG_TRANSPARENT_HUGEPAGE
633 barrier();
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800634#endif
Kirill A. Shutemovee536642013-12-20 15:10:03 +0200635 if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700636 return 1;
637 if (unlikely(pmd_bad(pmdval))) {
Kirill A. Shutemovee536642013-12-20 15:10:03 +0200638 pmd_clear_bad(pmd);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700639 return 1;
640 }
641 return 0;
642}
643
644/*
645 * This is a noop if Transparent Hugepage Support is not built into
646 * the kernel. Otherwise it is equivalent to
647 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
648 * places that already verified the pmd is not none and they want to
649 * walk ptes while holding the mmap sem in read mode (write mode don't
650 * need this). If THP is not enabled, the pmd can't go away under the
651 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
652 * run a pmd_trans_unstable before walking the ptes after
653 * split_huge_page_pmd returns (because it may have run when the pmd
654 * become null, but then a page fault can map in a THP and not a
655 * regular page).
656 */
657static inline int pmd_trans_unstable(pmd_t *pmd)
658{
659#ifdef CONFIG_TRANSPARENT_HUGEPAGE
660 return pmd_none_or_trans_huge_or_clear_bad(pmd);
661#else
662 return 0;
663#endif
664}
665
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200666#ifdef CONFIG_NUMA_BALANCING
667#ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE
668/*
669 * _PAGE_NUMA works identical to _PAGE_PROTNONE (it's actually the
670 * same bit too). It's set only when _PAGE_PRESET is not set and it's
671 * never set if _PAGE_PRESENT is set.
672 *
673 * pte/pmd_present() returns true if pte/pmd_numa returns true. Page
674 * fault triggers on those regions if pte/pmd_numa returns true
675 * (because _PAGE_PRESENT is not set).
676 */
677#ifndef pte_numa
678static inline int pte_numa(pte_t pte)
679{
680 return (pte_flags(pte) &
Mel Gormanc46a7c82014-06-04 16:06:30 -0700681 (_PAGE_NUMA|_PAGE_PROTNONE|_PAGE_PRESENT)) == _PAGE_NUMA;
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200682}
683#endif
684
685#ifndef pmd_numa
686static inline int pmd_numa(pmd_t pmd)
687{
688 return (pmd_flags(pmd) &
Mel Gormanc46a7c82014-06-04 16:06:30 -0700689 (_PAGE_NUMA|_PAGE_PROTNONE|_PAGE_PRESENT)) == _PAGE_NUMA;
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200690}
691#endif
692
693/*
694 * pte/pmd_mknuma sets the _PAGE_ACCESSED bitflag automatically
695 * because they're called by the NUMA hinting minor page fault. If we
696 * wouldn't set the _PAGE_ACCESSED bitflag here, the TLB miss handler
697 * would be forced to set it later while filling the TLB after we
698 * return to userland. That would trigger a second write to memory
699 * that we optimize away by setting _PAGE_ACCESSED here.
700 */
701#ifndef pte_mknonnuma
702static inline pte_t pte_mknonnuma(pte_t pte)
703{
Mel Gorman29c77872014-04-18 15:07:21 -0700704 pteval_t val = pte_val(pte);
705
706 val &= ~_PAGE_NUMA;
707 val |= (_PAGE_PRESENT|_PAGE_ACCESSED);
708 return __pte(val);
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200709}
710#endif
711
712#ifndef pmd_mknonnuma
713static inline pmd_t pmd_mknonnuma(pmd_t pmd)
714{
Mel Gorman29c77872014-04-18 15:07:21 -0700715 pmdval_t val = pmd_val(pmd);
716
717 val &= ~_PAGE_NUMA;
718 val |= (_PAGE_PRESENT|_PAGE_ACCESSED);
719
720 return __pmd(val);
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200721}
722#endif
723
724#ifndef pte_mknuma
725static inline pte_t pte_mknuma(pte_t pte)
726{
Mel Gorman29c77872014-04-18 15:07:21 -0700727 pteval_t val = pte_val(pte);
728
729 val &= ~_PAGE_PRESENT;
730 val |= _PAGE_NUMA;
731
732 return __pte(val);
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200733}
734#endif
735
Aneesh Kumar K.V56eecdb2014-02-12 09:13:38 +0530736#ifndef ptep_set_numa
737static inline void ptep_set_numa(struct mm_struct *mm, unsigned long addr,
738 pte_t *ptep)
739{
740 pte_t ptent = *ptep;
741
742 ptent = pte_mknuma(ptent);
743 set_pte_at(mm, addr, ptep, ptent);
744 return;
745}
746#endif
747
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200748#ifndef pmd_mknuma
749static inline pmd_t pmd_mknuma(pmd_t pmd)
750{
Mel Gorman29c77872014-04-18 15:07:21 -0700751 pmdval_t val = pmd_val(pmd);
752
753 val &= ~_PAGE_PRESENT;
754 val |= _PAGE_NUMA;
755
756 return __pmd(val);
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200757}
758#endif
Aneesh Kumar K.V56eecdb2014-02-12 09:13:38 +0530759
760#ifndef pmdp_set_numa
761static inline void pmdp_set_numa(struct mm_struct *mm, unsigned long addr,
762 pmd_t *pmdp)
763{
764 pmd_t pmd = *pmdp;
765
766 pmd = pmd_mknuma(pmd);
767 set_pmd_at(mm, addr, pmdp, pmd);
768 return;
769}
770#endif
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200771#else
772extern int pte_numa(pte_t pte);
773extern int pmd_numa(pmd_t pmd);
774extern pte_t pte_mknonnuma(pte_t pte);
775extern pmd_t pmd_mknonnuma(pmd_t pmd);
776extern pte_t pte_mknuma(pte_t pte);
777extern pmd_t pmd_mknuma(pmd_t pmd);
Aneesh Kumar K.V56eecdb2014-02-12 09:13:38 +0530778extern void ptep_set_numa(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
779extern void pmdp_set_numa(struct mm_struct *mm, unsigned long addr, pmd_t *pmdp);
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200780#endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */
781#else
782static inline int pmd_numa(pmd_t pmd)
783{
784 return 0;
785}
786
787static inline int pte_numa(pte_t pte)
788{
789 return 0;
790}
791
792static inline pte_t pte_mknonnuma(pte_t pte)
793{
794 return pte;
795}
796
797static inline pmd_t pmd_mknonnuma(pmd_t pmd)
798{
799 return pmd;
800}
801
802static inline pte_t pte_mknuma(pte_t pte)
803{
804 return pte;
805}
806
Aneesh Kumar K.V56eecdb2014-02-12 09:13:38 +0530807static inline void ptep_set_numa(struct mm_struct *mm, unsigned long addr,
808 pte_t *ptep)
809{
810 return;
811}
812
813
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200814static inline pmd_t pmd_mknuma(pmd_t pmd)
815{
816 return pmd;
817}
Aneesh Kumar K.V56eecdb2014-02-12 09:13:38 +0530818
819static inline void pmdp_set_numa(struct mm_struct *mm, unsigned long addr,
820 pmd_t *pmdp)
821{
822 return ;
823}
Andrea Arcangelibe3a7282012-10-04 01:50:47 +0200824#endif /* CONFIG_NUMA_BALANCING */
825
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700826#endif /* CONFIG_MMU */
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800827
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828#endif /* !__ASSEMBLY__ */
829
Al Viro40d158e2013-05-11 12:13:10 -0400830#ifndef io_remap_pfn_range
831#define io_remap_pfn_range remap_pfn_range
832#endif
833
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834#endif /* _ASM_GENERIC_PGTABLE_H */