blob: 0b3c0d39ef753053bb26c1b9fb4979e706240a58 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef _ASM_GENERIC_PGTABLE_H
2#define _ASM_GENERIC_PGTABLE_H
3
Dan Williamsf25748e32016-01-15 16:56:43 -08004#include <linux/pfn.h>
5
Rusty Russell673eae82006-09-25 23:32:29 -07006#ifndef __ASSEMBLY__
Greg Ungerer95352392007-08-10 13:01:20 -07007#ifdef CONFIG_MMU
Rusty Russell673eae82006-09-25 23:32:29 -07008
Ben Hutchingsfbd71842011-02-27 05:41:35 +00009#include <linux/mm_types.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050010#include <linux/bug.h>
Toshi Kanie61ce6a2015-04-14 15:47:23 -070011#include <linux/errno.h>
Ben Hutchingsfbd71842011-02-27 05:41:35 +000012
Kirill A. Shutemov235a8f02015-04-14 15:46:17 -070013#if 4 - defined(__PAGETABLE_PUD_FOLDED) - defined(__PAGETABLE_PMD_FOLDED) != \
14 CONFIG_PGTABLE_LEVELS
15#error CONFIG_PGTABLE_LEVELS is not consistent with __PAGETABLE_{PUD,PMD}_FOLDED
16#endif
17
Hugh Dickins6ee86302013-04-29 15:07:44 -070018/*
19 * On almost all architectures and configurations, 0 can be used as the
20 * upper ceiling to free_pgtables(): on many architectures it has the same
21 * effect as using TASK_SIZE. However, there is one configuration which
22 * must impose a more careful limit, to avoid freeing kernel pgtables.
23 */
24#ifndef USER_PGTABLES_CEILING
25#define USER_PGTABLES_CEILING 0UL
26#endif
27
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
Andrea Arcangelie2cda322011-01-13 15:46:40 -080029extern int ptep_set_access_flags(struct vm_area_struct *vma,
30 unsigned long address, pte_t *ptep,
31 pte_t entry, int dirty);
32#endif
33
34#ifndef __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
Vineet Guptabd5e88a2015-07-09 17:22:44 +053035#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Andrea Arcangelie2cda322011-01-13 15:46:40 -080036extern int pmdp_set_access_flags(struct vm_area_struct *vma,
37 unsigned long address, pmd_t *pmdp,
38 pmd_t entry, int dirty);
Vineet Guptabd5e88a2015-07-09 17:22:44 +053039#else
40static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
41 unsigned long address, pmd_t *pmdp,
42 pmd_t entry, int dirty)
43{
44 BUILD_BUG();
45 return 0;
46}
47#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#endif
49
50#ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
Andrea Arcangelie2cda322011-01-13 15:46:40 -080051static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
52 unsigned long address,
53 pte_t *ptep)
54{
55 pte_t pte = *ptep;
56 int r = 1;
57 if (!pte_young(pte))
58 r = 0;
59 else
60 set_pte_at(vma->vm_mm, address, ptep, pte_mkold(pte));
61 return r;
62}
63#endif
64
65#ifndef __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
66#ifdef CONFIG_TRANSPARENT_HUGEPAGE
67static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
68 unsigned long address,
69 pmd_t *pmdp)
70{
71 pmd_t pmd = *pmdp;
72 int r = 1;
73 if (!pmd_young(pmd))
74 r = 0;
75 else
76 set_pmd_at(vma->vm_mm, address, pmdp, pmd_mkold(pmd));
77 return r;
78}
Vineet Guptabd5e88a2015-07-09 17:22:44 +053079#else
Andrea Arcangelie2cda322011-01-13 15:46:40 -080080static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
81 unsigned long address,
82 pmd_t *pmdp)
83{
Vineet Guptabd5e88a2015-07-09 17:22:44 +053084 BUILD_BUG();
Andrea Arcangelie2cda322011-01-13 15:46:40 -080085 return 0;
86}
87#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Linus Torvalds1da177e2005-04-16 15:20:36 -070088#endif
89
90#ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
Andrea Arcangelie2cda322011-01-13 15:46:40 -080091int ptep_clear_flush_young(struct vm_area_struct *vma,
92 unsigned long address, pte_t *ptep);
93#endif
94
95#ifndef __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
Vineet Guptabd5e88a2015-07-09 17:22:44 +053096#ifdef CONFIG_TRANSPARENT_HUGEPAGE
97extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
98 unsigned long address, pmd_t *pmdp);
99#else
100/*
101 * Despite relevant to THP only, this API is called from generic rmap code
102 * under PageTransHuge(), hence needs a dummy implementation for !THP
103 */
104static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
105 unsigned long address, pmd_t *pmdp)
106{
107 BUILD_BUG();
108 return 0;
109}
110#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111#endif
112
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800114static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
115 unsigned long address,
116 pte_t *ptep)
117{
118 pte_t pte = *ptep;
119 pte_clear(mm, address, ptep);
120 return pte;
121}
122#endif
123
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700124#ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800125#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700126static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
127 unsigned long address,
128 pmd_t *pmdp)
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800129{
130 pmd_t pmd = *pmdp;
Catalin Marinas2d28a222012-10-08 16:32:59 -0700131 pmd_clear(pmdp);
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800132 return pmd;
Nicolas Kaiser49b24d62011-06-15 15:08:34 -0700133}
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800134#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135#endif
136
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700137#ifndef __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
Martin Schwidefskyfcbe08d62014-10-24 10:52:29 +0200138#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700139static inline pmd_t pmdp_huge_get_and_clear_full(struct mm_struct *mm,
Martin Schwidefskyfcbe08d62014-10-24 10:52:29 +0200140 unsigned long address, pmd_t *pmdp,
141 int full)
142{
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700143 return pmdp_huge_get_and_clear(mm, address, pmdp);
Martin Schwidefskyfcbe08d62014-10-24 10:52:29 +0200144}
145#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
146#endif
147
Zachary Amsdena6003882005-09-03 15:55:04 -0700148#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800149static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
150 unsigned long address, pte_t *ptep,
151 int full)
152{
153 pte_t pte;
154 pte = ptep_get_and_clear(mm, address, ptep);
155 return pte;
156}
Zachary Amsdena6003882005-09-03 15:55:04 -0700157#endif
158
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700159/*
160 * Some architectures may be able to avoid expensive synchronization
161 * primitives when modifications are made to PTE's which are already
162 * not present, or in the process of an address space destruction.
163 */
164#ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800165static inline void pte_clear_not_present_full(struct mm_struct *mm,
166 unsigned long address,
167 pte_t *ptep,
168 int full)
169{
170 pte_clear(mm, address, ptep);
171}
Zachary Amsdena6003882005-09-03 15:55:04 -0700172#endif
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174#ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800175extern pte_t ptep_clear_flush(struct vm_area_struct *vma,
176 unsigned long address,
177 pte_t *ptep);
178#endif
179
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -0700180#ifndef __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
181extern pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800182 unsigned long address,
183 pmd_t *pmdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184#endif
185
186#ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
Tim Schmielau8c65b4a2005-11-07 00:59:43 -0800187struct mm_struct;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
189{
190 pte_t old_pte = *ptep;
191 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
192}
193#endif
194
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800195#ifndef __HAVE_ARCH_PMDP_SET_WRPROTECT
196#ifdef CONFIG_TRANSPARENT_HUGEPAGE
197static inline void pmdp_set_wrprotect(struct mm_struct *mm,
198 unsigned long address, pmd_t *pmdp)
199{
200 pmd_t old_pmd = *pmdp;
201 set_pmd_at(mm, address, pmdp, pmd_wrprotect(old_pmd));
202}
Vineet Guptabd5e88a2015-07-09 17:22:44 +0530203#else
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800204static inline void pmdp_set_wrprotect(struct mm_struct *mm,
205 unsigned long address, pmd_t *pmdp)
206{
Vineet Guptabd5e88a2015-07-09 17:22:44 +0530207 BUILD_BUG();
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800208}
209#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
210#endif
211
Aneesh Kumar K.V15a25b22015-06-24 16:57:39 -0700212#ifndef pmdp_collapse_flush
213#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Aneesh Kumar K.Vf28b6ff2015-06-24 16:57:42 -0700214extern pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
215 unsigned long address, pmd_t *pmdp);
Aneesh Kumar K.V15a25b22015-06-24 16:57:39 -0700216#else
217static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
218 unsigned long address,
219 pmd_t *pmdp)
220{
221 BUILD_BUG();
222 return *pmdp;
223}
224#define pmdp_collapse_flush pmdp_collapse_flush
225#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
226#endif
227
Gerald Schaefere3ebcf642012-10-08 16:30:07 -0700228#ifndef __HAVE_ARCH_PGTABLE_DEPOSIT
Aneesh Kumar K.V6b0b50b2013-06-05 17:14:02 -0700229extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
230 pgtable_t pgtable);
Gerald Schaefere3ebcf642012-10-08 16:30:07 -0700231#endif
232
233#ifndef __HAVE_ARCH_PGTABLE_WITHDRAW
Aneesh Kumar K.V6b0b50b2013-06-05 17:14:02 -0700234extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
Gerald Schaefere3ebcf642012-10-08 16:30:07 -0700235#endif
236
Gerald Schaefer46dcde72012-10-08 16:30:09 -0700237#ifndef __HAVE_ARCH_PMDP_INVALIDATE
238extern void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
239 pmd_t *pmdp);
240#endif
241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242#ifndef __HAVE_ARCH_PTE_SAME
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800243static inline int pte_same(pte_t pte_a, pte_t pte_b)
244{
245 return pte_val(pte_a) == pte_val(pte_b);
246}
247#endif
248
Konstantin Weitz45961722013-04-17 13:59:32 +0200249#ifndef __HAVE_ARCH_PTE_UNUSED
250/*
251 * Some architectures provide facilities to virtualization guests
252 * so that they can flag allocated pages as unused. This allows the
253 * host to transparently reclaim unused pages. This function returns
254 * whether the pte's page is unused.
255 */
256static inline int pte_unused(pte_t pte)
257{
258 return 0;
259}
260#endif
261
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800262#ifndef __HAVE_ARCH_PMD_SAME
263#ifdef CONFIG_TRANSPARENT_HUGEPAGE
264static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
265{
266 return pmd_val(pmd_a) == pmd_val(pmd_b);
267}
268#else /* CONFIG_TRANSPARENT_HUGEPAGE */
269static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
270{
Vineet Guptabd5e88a2015-07-09 17:22:44 +0530271 BUILD_BUG();
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800272 return 0;
273}
274#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275#endif
276
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277#ifndef __HAVE_ARCH_PGD_OFFSET_GATE
278#define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
279#endif
280
David S. Miller0b0968a2006-06-01 17:47:25 -0700281#ifndef __HAVE_ARCH_MOVE_PTE
Nick Piggin8b1f3122005-09-27 21:45:18 -0700282#define move_pte(pte, prot, old_addr, new_addr) (pte)
Nick Piggin8b1f3122005-09-27 21:45:18 -0700283#endif
284
Rik van Riel2c3cf552012-10-09 15:31:12 +0200285#ifndef pte_accessible
Rik van Riel20841402013-12-18 17:08:44 -0800286# define pte_accessible(mm, pte) ((void)(pte), 1)
Rik van Riel2c3cf552012-10-09 15:31:12 +0200287#endif
288
Shaohua Li61c77322010-08-16 09:16:55 +0800289#ifndef flush_tlb_fix_spurious_fault
290#define flush_tlb_fix_spurious_fault(vma, address) flush_tlb_page(vma, address)
291#endif
292
Paul Mundt0634a632009-06-23 13:51:19 +0200293#ifndef pgprot_noncached
294#define pgprot_noncached(prot) (prot)
295#endif
296
venkatesh.pallipadi@intel.com2520bd32008-12-18 11:41:32 -0800297#ifndef pgprot_writecombine
298#define pgprot_writecombine pgprot_noncached
299#endif
300
Toshi Kanid1b4bfb2015-06-04 18:55:18 +0200301#ifndef pgprot_writethrough
302#define pgprot_writethrough pgprot_noncached
303#endif
304
Liviu Dudau8b921ac2014-09-29 15:29:30 +0100305#ifndef pgprot_device
306#define pgprot_device pgprot_noncached
307#endif
308
Peter Feiner64e45502014-10-13 15:55:46 -0700309#ifndef pgprot_modify
310#define pgprot_modify pgprot_modify
311static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
312{
313 if (pgprot_val(oldprot) == pgprot_val(pgprot_noncached(oldprot)))
314 newprot = pgprot_noncached(newprot);
315 if (pgprot_val(oldprot) == pgprot_val(pgprot_writecombine(oldprot)))
316 newprot = pgprot_writecombine(newprot);
317 if (pgprot_val(oldprot) == pgprot_val(pgprot_device(oldprot)))
318 newprot = pgprot_device(newprot);
319 return newprot;
320}
321#endif
322
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323/*
Hugh Dickins8f6c99c2005-04-19 13:29:17 -0700324 * When walking page tables, get the address of the next boundary,
325 * or the end address of the range if that comes earlier. Although no
326 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 */
328
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329#define pgd_addr_end(addr, end) \
330({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
331 (__boundary - 1 < (end) - 1)? __boundary: (end); \
332})
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
334#ifndef pud_addr_end
335#define pud_addr_end(addr, end) \
336({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
337 (__boundary - 1 < (end) - 1)? __boundary: (end); \
338})
339#endif
340
341#ifndef pmd_addr_end
342#define pmd_addr_end(addr, end) \
343({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
344 (__boundary - 1 < (end) - 1)? __boundary: (end); \
345})
346#endif
347
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348/*
349 * When walking page tables, we usually want to skip any p?d_none entries;
350 * and any p?d_bad entries - reporting the error before resetting to none.
351 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
352 */
353void pgd_clear_bad(pgd_t *);
354void pud_clear_bad(pud_t *);
355void pmd_clear_bad(pmd_t *);
356
357static inline int pgd_none_or_clear_bad(pgd_t *pgd)
358{
359 if (pgd_none(*pgd))
360 return 1;
361 if (unlikely(pgd_bad(*pgd))) {
362 pgd_clear_bad(pgd);
363 return 1;
364 }
365 return 0;
366}
367
368static inline int pud_none_or_clear_bad(pud_t *pud)
369{
370 if (pud_none(*pud))
371 return 1;
372 if (unlikely(pud_bad(*pud))) {
373 pud_clear_bad(pud);
374 return 1;
375 }
376 return 0;
377}
378
379static inline int pmd_none_or_clear_bad(pmd_t *pmd)
380{
381 if (pmd_none(*pmd))
382 return 1;
383 if (unlikely(pmd_bad(*pmd))) {
384 pmd_clear_bad(pmd);
385 return 1;
386 }
387 return 0;
388}
Greg Ungerer95352392007-08-10 13:01:20 -0700389
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -0700390static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
391 unsigned long addr,
392 pte_t *ptep)
393{
394 /*
395 * Get the current pte state, but zero it out to make it
396 * non-present, preventing the hardware from asynchronously
397 * updating it.
398 */
399 return ptep_get_and_clear(mm, addr, ptep);
400}
401
402static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
403 unsigned long addr,
404 pte_t *ptep, pte_t pte)
405{
406 /*
407 * The pte is non-present, so there's no hardware state to
408 * preserve.
409 */
410 set_pte_at(mm, addr, ptep, pte);
411}
412
413#ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
414/*
415 * Start a pte protection read-modify-write transaction, which
416 * protects against asynchronous hardware modifications to the pte.
417 * The intention is not to prevent the hardware from making pte
418 * updates, but to prevent any updates it may make from being lost.
419 *
420 * This does not protect against other software modifications of the
421 * pte; the appropriate pte lock must be held over the transation.
422 *
423 * Note that this interface is intended to be batchable, meaning that
424 * ptep_modify_prot_commit may not actually update the pte, but merely
425 * queue the update to be done at some later time. The update must be
426 * actually committed before the pte lock is released, however.
427 */
428static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
429 unsigned long addr,
430 pte_t *ptep)
431{
432 return __ptep_modify_prot_start(mm, addr, ptep);
433}
434
435/*
436 * Commit an update to a pte, leaving any hardware-controlled bits in
437 * the PTE unmodified.
438 */
439static inline void ptep_modify_prot_commit(struct mm_struct *mm,
440 unsigned long addr,
441 pte_t *ptep, pte_t pte)
442{
443 __ptep_modify_prot_commit(mm, addr, ptep, pte);
444}
445#endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
Sebastian Siewiorfe1a6872008-07-15 22:28:46 +0200446#endif /* CONFIG_MMU */
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -0700447
Greg Ungerer95352392007-08-10 13:01:20 -0700448/*
449 * A facility to provide lazy MMU batching. This allows PTE updates and
450 * page invalidations to be delayed until a call to leave lazy MMU mode
451 * is issued. Some architectures may benefit from doing this, and it is
452 * beneficial for both shadow and direct mode hypervisors, which may batch
453 * the PTE updates which happen during this window. Note that using this
454 * interface requires that read hazards be removed from the code. A read
455 * hazard could result in the direct mode hypervisor case, since the actual
456 * write to the page tables may not yet have taken place, so reads though
457 * a raw PTE pointer after it has been modified are not guaranteed to be
458 * up to date. This mode can only be entered and left under the protection of
459 * the page table locks for all page tables which may be modified. In the UP
460 * case, this is required so that preemption is disabled, and in the SMP case,
461 * it must synchronize the delayed page table writes properly on other CPUs.
462 */
463#ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
464#define arch_enter_lazy_mmu_mode() do {} while (0)
465#define arch_leave_lazy_mmu_mode() do {} while (0)
466#define arch_flush_lazy_mmu_mode() do {} while (0)
467#endif
468
469/*
Jeremy Fitzhardinge7fd7d832009-02-17 23:24:03 -0800470 * A facility to provide batching of the reload of page tables and
471 * other process state with the actual context switch code for
472 * paravirtualized guests. By convention, only one of the batched
473 * update (lazy) modes (CPU, MMU) should be active at any given time,
474 * entry should never be nested, and entry and exits should always be
475 * paired. This is for sanity of maintaining and reasoning about the
476 * kernel code. In this case, the exit (end of the context switch) is
477 * in architecture-specific code, and so doesn't need a generic
478 * definition.
Greg Ungerer95352392007-08-10 13:01:20 -0700479 */
Jeremy Fitzhardinge7fd7d832009-02-17 23:24:03 -0800480#ifndef __HAVE_ARCH_START_CONTEXT_SWITCH
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800481#define arch_start_context_switch(prev) do {} while (0)
Greg Ungerer95352392007-08-10 13:01:20 -0700482#endif
483
Pavel Emelyanov0f8975e2013-07-03 15:01:20 -0700484#ifndef CONFIG_HAVE_ARCH_SOFT_DIRTY
485static inline int pte_soft_dirty(pte_t pte)
486{
487 return 0;
488}
489
490static inline int pmd_soft_dirty(pmd_t pmd)
491{
492 return 0;
493}
494
495static inline pte_t pte_mksoft_dirty(pte_t pte)
496{
497 return pte;
498}
499
500static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
501{
502 return pmd;
503}
Cyrill Gorcunov179ef712013-08-13 16:00:49 -0700504
Martin Schwidefskya7b76172015-04-22 14:20:47 +0200505static inline pte_t pte_clear_soft_dirty(pte_t pte)
506{
507 return pte;
508}
509
510static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
511{
512 return pmd;
513}
514
Cyrill Gorcunov179ef712013-08-13 16:00:49 -0700515static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
516{
517 return pte;
518}
519
520static inline int pte_swp_soft_dirty(pte_t pte)
521{
522 return 0;
523}
524
525static inline pte_t pte_swp_clear_soft_dirty(pte_t pte)
526{
527 return pte;
528}
Pavel Emelyanov0f8975e2013-07-03 15:01:20 -0700529#endif
530
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800531#ifndef __HAVE_PFNMAP_TRACKING
532/*
Suresh Siddha5180da42012-10-08 16:28:29 -0700533 * Interfaces that can be used by architecture code to keep track of
534 * memory type of pfn mappings specified by the remap_pfn_range,
535 * vm_insert_pfn.
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800536 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700537
538/*
539 * track_pfn_remap is called when a _new_ pfn mapping is being established
540 * by remap_pfn_range() for physical range indicated by pfn and size.
541 */
542static inline int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -0700543 unsigned long pfn, unsigned long addr,
544 unsigned long size)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800545{
546 return 0;
547}
548
549/*
Suresh Siddha5180da42012-10-08 16:28:29 -0700550 * track_pfn_insert is called when a _new_ single pfn is established
551 * by vm_insert_pfn().
552 */
553static inline int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
Dan Williamsf25748e32016-01-15 16:56:43 -0800554 pfn_t pfn)
Suresh Siddha5180da42012-10-08 16:28:29 -0700555{
556 return 0;
557}
558
559/*
560 * track_pfn_copy is called when vma that is covering the pfnmap gets
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800561 * copied through copy_page_range().
562 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700563static inline int track_pfn_copy(struct vm_area_struct *vma)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800564{
565 return 0;
566}
567
568/*
Toshi Kanid9fe4fa2015-12-22 17:54:23 -0700569 * untrack_pfn is called while unmapping a pfnmap for a region.
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800570 * untrack can be called for a specific region indicated by pfn and size or
Suresh Siddha5180da42012-10-08 16:28:29 -0700571 * can be for the entire vma (in which case pfn, size are zero).
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800572 */
Suresh Siddha5180da42012-10-08 16:28:29 -0700573static inline void untrack_pfn(struct vm_area_struct *vma,
574 unsigned long pfn, unsigned long size)
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800575{
576}
Toshi Kanid9fe4fa2015-12-22 17:54:23 -0700577
578/*
579 * untrack_pfn_moved is called while mremapping a pfnmap for a new region.
580 */
581static inline void untrack_pfn_moved(struct vm_area_struct *vma)
582{
583}
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800584#else
Suresh Siddha5180da42012-10-08 16:28:29 -0700585extern int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -0700586 unsigned long pfn, unsigned long addr,
587 unsigned long size);
Suresh Siddha5180da42012-10-08 16:28:29 -0700588extern int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
Dan Williamsf25748e32016-01-15 16:56:43 -0800589 pfn_t pfn);
Suresh Siddha5180da42012-10-08 16:28:29 -0700590extern int track_pfn_copy(struct vm_area_struct *vma);
591extern void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
592 unsigned long size);
Toshi Kanid9fe4fa2015-12-22 17:54:23 -0700593extern void untrack_pfn_moved(struct vm_area_struct *vma);
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800594#endif
595
Kirill A. Shutemov816422a2012-12-12 13:52:36 -0800596#ifdef __HAVE_COLOR_ZERO_PAGE
597static inline int is_zero_pfn(unsigned long pfn)
598{
599 extern unsigned long zero_pfn;
600 unsigned long offset_from_zero_pfn = pfn - zero_pfn;
601 return offset_from_zero_pfn <= (zero_page_mask >> PAGE_SHIFT);
602}
603
Kirill A. Shutemov2f91ec82012-12-26 03:19:55 +0300604#define my_zero_pfn(addr) page_to_pfn(ZERO_PAGE(addr))
605
Kirill A. Shutemov816422a2012-12-12 13:52:36 -0800606#else
607static inline int is_zero_pfn(unsigned long pfn)
608{
609 extern unsigned long zero_pfn;
610 return pfn == zero_pfn;
611}
612
613static inline unsigned long my_zero_pfn(unsigned long addr)
614{
615 extern unsigned long zero_pfn;
616 return zero_pfn;
617}
618#endif
619
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700620#ifdef CONFIG_MMU
621
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800622#ifndef CONFIG_TRANSPARENT_HUGEPAGE
623static inline int pmd_trans_huge(pmd_t pmd)
624{
625 return 0;
626}
Andrea Arcangelie2cda322011-01-13 15:46:40 -0800627#ifndef __HAVE_ARCH_PMD_WRITE
628static inline int pmd_write(pmd_t pmd)
629{
630 BUG();
631 return 0;
632}
633#endif /* __HAVE_ARCH_PMD_WRITE */
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700634#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
635
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700636#ifndef pmd_read_atomic
637static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
638{
639 /*
640 * Depend on compiler for an atomic pmd read. NOTE: this is
641 * only going to work, if the pmdval_t isn't larger than
642 * an unsigned long.
643 */
644 return *pmdp;
645}
646#endif
647
Aneesh Kumar K.Vb3084f42014-01-13 11:34:24 +0530648#ifndef pmd_move_must_withdraw
649static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,
650 spinlock_t *old_pmd_ptl)
651{
652 /*
653 * With split pmd lock we also need to move preallocated
654 * PTE page table if new_pmd is on different PMD page table.
655 */
656 return new_pmd_ptl != old_pmd_ptl;
657}
658#endif
659
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700660/*
661 * This function is meant to be used by sites walking pagetables with
662 * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
663 * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
664 * into a null pmd and the transhuge page fault can convert a null pmd
665 * into an hugepmd or into a regular pmd (if the hugepage allocation
666 * fails). While holding the mmap_sem in read mode the pmd becomes
667 * stable and stops changing under us only if it's not null and not a
668 * transhuge pmd. When those races occurs and this function makes a
669 * difference vs the standard pmd_none_or_clear_bad, the result is
670 * undefined so behaving like if the pmd was none is safe (because it
671 * can return none anyway). The compiler level barrier() is critically
672 * important to compute the two checks atomically on the same pmdval.
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700673 *
674 * For 32bit kernels with a 64bit large pmd_t this automatically takes
675 * care of reading the pmd atomically to avoid SMP race conditions
676 * against pmd_populate() when the mmap_sem is hold for reading by the
677 * caller (a special atomic read not done by "gcc" as in the generic
678 * version above, is also needed when THP is disabled because the page
679 * fault can populate the pmd from under us).
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700680 */
681static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
682{
Andrea Arcangeli26c19172012-05-29 15:06:49 -0700683 pmd_t pmdval = pmd_read_atomic(pmd);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700684 /*
685 * The barrier will stabilize the pmdval in a register or on
686 * the stack so that it will stop changing under the code.
Andrea Arcangelie4eed032012-06-20 12:52:57 -0700687 *
688 * When CONFIG_TRANSPARENT_HUGEPAGE=y on x86 32bit PAE,
689 * pmd_read_atomic is allowed to return a not atomic pmdval
690 * (for example pointing to an hugepage that has never been
691 * mapped in the pmd). The below checks will only care about
692 * the low part of the pmd with 32bit PAE x86 anyway, with the
693 * exception of pmd_none(). So the important thing is that if
694 * the low part of the pmd is found null, the high part will
695 * be also null or the pmd_none() check below would be
696 * confused.
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700697 */
698#ifdef CONFIG_TRANSPARENT_HUGEPAGE
699 barrier();
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800700#endif
Kirill A. Shutemovee536642013-12-20 15:10:03 +0200701 if (pmd_none(pmdval) || pmd_trans_huge(pmdval))
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700702 return 1;
703 if (unlikely(pmd_bad(pmdval))) {
Kirill A. Shutemovee536642013-12-20 15:10:03 +0200704 pmd_clear_bad(pmd);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700705 return 1;
706 }
707 return 0;
708}
709
710/*
711 * This is a noop if Transparent Hugepage Support is not built into
712 * the kernel. Otherwise it is equivalent to
713 * pmd_none_or_trans_huge_or_clear_bad(), and shall only be called in
714 * places that already verified the pmd is not none and they want to
715 * walk ptes while holding the mmap sem in read mode (write mode don't
716 * need this). If THP is not enabled, the pmd can't go away under the
717 * code even if MADV_DONTNEED runs, but if THP is enabled we need to
718 * run a pmd_trans_unstable before walking the ptes after
719 * split_huge_page_pmd returns (because it may have run when the pmd
720 * become null, but then a page fault can map in a THP and not a
721 * regular page).
722 */
723static inline int pmd_trans_unstable(pmd_t *pmd)
724{
725#ifdef CONFIG_TRANSPARENT_HUGEPAGE
726 return pmd_none_or_trans_huge_or_clear_bad(pmd);
727#else
728 return 0;
729#endif
730}
731
Mel Gormane7bb4b6d2015-02-12 14:58:19 -0800732#ifndef CONFIG_NUMA_BALANCING
733/*
734 * Technically a PTE can be PROTNONE even when not doing NUMA balancing but
735 * the only case the kernel cares is for NUMA balancing and is only ever set
736 * when the VMA is accessible. For PROT_NONE VMAs, the PTEs are not marked
737 * _PAGE_PROTNONE so by by default, implement the helper as "always no". It
738 * is the responsibility of the caller to distinguish between PROT_NONE
739 * protections and NUMA hinting fault protections.
740 */
741static inline int pte_protnone(pte_t pte)
742{
743 return 0;
744}
745
746static inline int pmd_protnone(pmd_t pmd)
747{
748 return 0;
749}
750#endif /* CONFIG_NUMA_BALANCING */
751
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700752#endif /* CONFIG_MMU */
Andrea Arcangeli5f6e8da2011-01-13 15:46:40 -0800753
Toshi Kanie61ce6a2015-04-14 15:47:23 -0700754#ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
755int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot);
756int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot);
Toshi Kanib9820d82015-04-14 15:47:26 -0700757int pud_clear_huge(pud_t *pud);
758int pmd_clear_huge(pmd_t *pmd);
Toshi Kanie61ce6a2015-04-14 15:47:23 -0700759#else /* !CONFIG_HAVE_ARCH_HUGE_VMAP */
760static inline int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
761{
762 return 0;
763}
764static inline int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
765{
766 return 0;
767}
Toshi Kanib9820d82015-04-14 15:47:26 -0700768static inline int pud_clear_huge(pud_t *pud)
769{
770 return 0;
771}
772static inline int pmd_clear_huge(pmd_t *pmd)
773{
774 return 0;
775}
Toshi Kanie61ce6a2015-04-14 15:47:23 -0700776#endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
777
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778#endif /* !__ASSEMBLY__ */
779
Al Viro40d158e2013-05-11 12:13:10 -0400780#ifndef io_remap_pfn_range
781#define io_remap_pfn_range remap_pfn_range
782#endif
783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784#endif /* _ASM_GENERIC_PGTABLE_H */