blob: 8e6d0ca70aba987b4663db6e512d79236400f183 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -07007#ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
8/*
9 * Largely same as above, but only sets the access flags (dirty,
10 * accessed, and writable). Furthermore, we know it always gets set
11 * to a "more permissive" setting, which allows most architectures
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -070012 * to optimize this. We return whether the PTE actually changed, which
13 * in turn instructs the caller to do things like update__mmu_cache.
14 * This used to be done in the caller, but sparc needs minor faults to
15 * force that call on sun4c so we changed this macro slightly
Linus Torvalds1da177e2005-04-16 15:20:36 -070016 */
17#define ptep_set_access_flags(__vma, __address, __ptep, __entry, __dirty) \
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -070018({ \
19 int __changed = !pte_same(*(__ptep), __entry); \
20 if (__changed) { \
21 set_pte_at((__vma)->vm_mm, (__address), __ptep, __entry); \
22 flush_tlb_page(__vma, __address); \
23 } \
24 __changed; \
25})
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#endif
27
28#ifndef __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
29#define ptep_test_and_clear_young(__vma, __address, __ptep) \
30({ \
31 pte_t __pte = *(__ptep); \
32 int r = 1; \
33 if (!pte_young(__pte)) \
34 r = 0; \
35 else \
36 set_pte_at((__vma)->vm_mm, (__address), \
37 (__ptep), pte_mkold(__pte)); \
38 r; \
39})
40#endif
41
42#ifndef __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
43#define ptep_clear_flush_young(__vma, __address, __ptep) \
44({ \
45 int __young; \
46 __young = ptep_test_and_clear_young(__vma, __address, __ptep); \
47 if (__young) \
48 flush_tlb_page(__vma, __address); \
49 __young; \
50})
51#endif
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR
54#define ptep_get_and_clear(__mm, __address, __ptep) \
55({ \
56 pte_t __pte = *(__ptep); \
57 pte_clear((__mm), (__address), (__ptep)); \
58 __pte; \
59})
60#endif
61
Zachary Amsdena6003882005-09-03 15:55:04 -070062#ifndef __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
63#define ptep_get_and_clear_full(__mm, __address, __ptep, __full) \
64({ \
65 pte_t __pte; \
66 __pte = ptep_get_and_clear((__mm), (__address), (__ptep)); \
67 __pte; \
68})
69#endif
70
Zachary Amsden9888a1c2006-09-30 23:29:31 -070071/*
72 * Some architectures may be able to avoid expensive synchronization
73 * primitives when modifications are made to PTE's which are already
74 * not present, or in the process of an address space destruction.
75 */
76#ifndef __HAVE_ARCH_PTE_CLEAR_NOT_PRESENT_FULL
77#define pte_clear_not_present_full(__mm, __address, __ptep, __full) \
Zachary Amsdena6003882005-09-03 15:55:04 -070078do { \
79 pte_clear((__mm), (__address), (__ptep)); \
80} while (0)
81#endif
82
Linus Torvalds1da177e2005-04-16 15:20:36 -070083#ifndef __HAVE_ARCH_PTEP_CLEAR_FLUSH
84#define ptep_clear_flush(__vma, __address, __ptep) \
85({ \
86 pte_t __pte; \
87 __pte = ptep_get_and_clear((__vma)->vm_mm, __address, __ptep); \
88 flush_tlb_page(__vma, __address); \
89 __pte; \
90})
91#endif
92
93#ifndef __HAVE_ARCH_PTEP_SET_WRPROTECT
Tim Schmielau8c65b4a2005-11-07 00:59:43 -080094struct mm_struct;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long address, pte_t *ptep)
96{
97 pte_t old_pte = *ptep;
98 set_pte_at(mm, address, ptep, pte_wrprotect(old_pte));
99}
100#endif
101
102#ifndef __HAVE_ARCH_PTE_SAME
103#define pte_same(A,B) (pte_val(A) == pte_val(B))
104#endif
105
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200106#ifndef __HAVE_ARCH_PAGE_TEST_DIRTY
107#define page_test_dirty(page) (0)
108#endif
109
110#ifndef __HAVE_ARCH_PAGE_CLEAR_DIRTY
111#define page_clear_dirty(page) do { } while (0)
112#endif
113
114#ifndef __HAVE_ARCH_PAGE_TEST_DIRTY
Abhijit Karmarkarb4955ce2005-06-21 17:15:13 -0700115#define pte_maybe_dirty(pte) pte_dirty(pte)
116#else
117#define pte_maybe_dirty(pte) (1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118#endif
119
120#ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
121#define page_test_and_clear_young(page) (0)
122#endif
123
124#ifndef __HAVE_ARCH_PGD_OFFSET_GATE
125#define pgd_offset_gate(mm, addr) pgd_offset(mm, addr)
126#endif
127
David S. Miller0b0968a2006-06-01 17:47:25 -0700128#ifndef __HAVE_ARCH_MOVE_PTE
Nick Piggin8b1f3122005-09-27 21:45:18 -0700129#define move_pte(pte, prot, old_addr, new_addr) (pte)
Nick Piggin8b1f3122005-09-27 21:45:18 -0700130#endif
131
venkatesh.pallipadi@intel.com2520bd32008-12-18 11:41:32 -0800132#ifndef pgprot_writecombine
133#define pgprot_writecombine pgprot_noncached
134#endif
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136/*
Hugh Dickins8f6c99c2005-04-19 13:29:17 -0700137 * When walking page tables, get the address of the next boundary,
138 * or the end address of the range if that comes earlier. Although no
139 * vma end wraps to 0, rounded up __boundary may wrap to 0 throughout.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 */
141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142#define pgd_addr_end(addr, end) \
143({ unsigned long __boundary = ((addr) + PGDIR_SIZE) & PGDIR_MASK; \
144 (__boundary - 1 < (end) - 1)? __boundary: (end); \
145})
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147#ifndef pud_addr_end
148#define pud_addr_end(addr, end) \
149({ unsigned long __boundary = ((addr) + PUD_SIZE) & PUD_MASK; \
150 (__boundary - 1 < (end) - 1)? __boundary: (end); \
151})
152#endif
153
154#ifndef pmd_addr_end
155#define pmd_addr_end(addr, end) \
156({ unsigned long __boundary = ((addr) + PMD_SIZE) & PMD_MASK; \
157 (__boundary - 1 < (end) - 1)? __boundary: (end); \
158})
159#endif
160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161/*
162 * When walking page tables, we usually want to skip any p?d_none entries;
163 * and any p?d_bad entries - reporting the error before resetting to none.
164 * Do the tests inline, but report and clear the bad entry in mm/memory.c.
165 */
166void pgd_clear_bad(pgd_t *);
167void pud_clear_bad(pud_t *);
168void pmd_clear_bad(pmd_t *);
169
170static inline int pgd_none_or_clear_bad(pgd_t *pgd)
171{
172 if (pgd_none(*pgd))
173 return 1;
174 if (unlikely(pgd_bad(*pgd))) {
175 pgd_clear_bad(pgd);
176 return 1;
177 }
178 return 0;
179}
180
181static inline int pud_none_or_clear_bad(pud_t *pud)
182{
183 if (pud_none(*pud))
184 return 1;
185 if (unlikely(pud_bad(*pud))) {
186 pud_clear_bad(pud);
187 return 1;
188 }
189 return 0;
190}
191
192static inline int pmd_none_or_clear_bad(pmd_t *pmd)
193{
194 if (pmd_none(*pmd))
195 return 1;
196 if (unlikely(pmd_bad(*pmd))) {
197 pmd_clear_bad(pmd);
198 return 1;
199 }
200 return 0;
201}
Greg Ungerer95352392007-08-10 13:01:20 -0700202
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -0700203static inline pte_t __ptep_modify_prot_start(struct mm_struct *mm,
204 unsigned long addr,
205 pte_t *ptep)
206{
207 /*
208 * Get the current pte state, but zero it out to make it
209 * non-present, preventing the hardware from asynchronously
210 * updating it.
211 */
212 return ptep_get_and_clear(mm, addr, ptep);
213}
214
215static inline void __ptep_modify_prot_commit(struct mm_struct *mm,
216 unsigned long addr,
217 pte_t *ptep, pte_t pte)
218{
219 /*
220 * The pte is non-present, so there's no hardware state to
221 * preserve.
222 */
223 set_pte_at(mm, addr, ptep, pte);
224}
225
226#ifndef __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
227/*
228 * Start a pte protection read-modify-write transaction, which
229 * protects against asynchronous hardware modifications to the pte.
230 * The intention is not to prevent the hardware from making pte
231 * updates, but to prevent any updates it may make from being lost.
232 *
233 * This does not protect against other software modifications of the
234 * pte; the appropriate pte lock must be held over the transation.
235 *
236 * Note that this interface is intended to be batchable, meaning that
237 * ptep_modify_prot_commit may not actually update the pte, but merely
238 * queue the update to be done at some later time. The update must be
239 * actually committed before the pte lock is released, however.
240 */
241static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
242 unsigned long addr,
243 pte_t *ptep)
244{
245 return __ptep_modify_prot_start(mm, addr, ptep);
246}
247
248/*
249 * Commit an update to a pte, leaving any hardware-controlled bits in
250 * the PTE unmodified.
251 */
252static inline void ptep_modify_prot_commit(struct mm_struct *mm,
253 unsigned long addr,
254 pte_t *ptep, pte_t pte)
255{
256 __ptep_modify_prot_commit(mm, addr, ptep, pte);
257}
258#endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
Sebastian Siewiorfe1a6872008-07-15 22:28:46 +0200259#endif /* CONFIG_MMU */
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -0700260
Greg Ungerer95352392007-08-10 13:01:20 -0700261/*
262 * A facility to provide lazy MMU batching. This allows PTE updates and
263 * page invalidations to be delayed until a call to leave lazy MMU mode
264 * is issued. Some architectures may benefit from doing this, and it is
265 * beneficial for both shadow and direct mode hypervisors, which may batch
266 * the PTE updates which happen during this window. Note that using this
267 * interface requires that read hazards be removed from the code. A read
268 * hazard could result in the direct mode hypervisor case, since the actual
269 * write to the page tables may not yet have taken place, so reads though
270 * a raw PTE pointer after it has been modified are not guaranteed to be
271 * up to date. This mode can only be entered and left under the protection of
272 * the page table locks for all page tables which may be modified. In the UP
273 * case, this is required so that preemption is disabled, and in the SMP case,
274 * it must synchronize the delayed page table writes properly on other CPUs.
275 */
276#ifndef __HAVE_ARCH_ENTER_LAZY_MMU_MODE
277#define arch_enter_lazy_mmu_mode() do {} while (0)
278#define arch_leave_lazy_mmu_mode() do {} while (0)
279#define arch_flush_lazy_mmu_mode() do {} while (0)
280#endif
281
282/*
283 * A facility to provide batching of the reload of page tables with the
284 * actual context switch code for paravirtualized guests. By convention,
285 * only one of the lazy modes (CPU, MMU) should be active at any given
286 * time, entry should never be nested, and entry and exits should always
287 * be paired. This is for sanity of maintaining and reasoning about the
288 * kernel code.
289 */
290#ifndef __HAVE_ARCH_ENTER_LAZY_CPU_MODE
291#define arch_enter_lazy_cpu_mode() do {} while (0)
292#define arch_leave_lazy_cpu_mode() do {} while (0)
293#define arch_flush_lazy_cpu_mode() do {} while (0)
294#endif
295
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800296#ifndef __HAVE_PFNMAP_TRACKING
297/*
298 * Interface that can be used by architecture code to keep track of
299 * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
300 *
301 * track_pfn_vma_new is called when a _new_ pfn mapping is being established
302 * for physical range indicated by pfn and size.
303 */
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -0800304static inline int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800305 unsigned long pfn, unsigned long size)
306{
307 return 0;
308}
309
310/*
311 * Interface that can be used by architecture code to keep track of
312 * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
313 *
314 * track_pfn_vma_copy is called when vma that is covering the pfnmap gets
315 * copied through copy_page_range().
316 */
317static inline int track_pfn_vma_copy(struct vm_area_struct *vma)
318{
319 return 0;
320}
321
322/*
323 * Interface that can be used by architecture code to keep track of
324 * memory type of pfn mappings (remap_pfn_range, vm_insert_pfn)
325 *
326 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
327 * untrack can be called for a specific region indicated by pfn and size or
328 * can be for the entire vma (in which case size can be zero).
329 */
330static inline void untrack_pfn_vma(struct vm_area_struct *vma,
331 unsigned long pfn, unsigned long size)
332{
333}
334#else
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -0800335extern int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800336 unsigned long pfn, unsigned long size);
337extern int track_pfn_vma_copy(struct vm_area_struct *vma);
338extern void untrack_pfn_vma(struct vm_area_struct *vma, unsigned long pfn,
339 unsigned long size);
340#endif
341
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342#endif /* !__ASSEMBLY__ */
343
344#endif /* _ASM_GENERIC_PGTABLE_H */