1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
26 #ifndef _VM_HTABLE_H
27 #define _VM_HTABLE_H
28
29 #ifdef __cplusplus
30 extern "C" {
31 #endif
32
33 #if defined(__GNUC__) && defined(_ASM_INLINES) && defined(_KERNEL)
34 #include <asm/htable.h>
35 #endif
36
37 extern void atomic_andb(uint8_t *addr, uint8_t value);
38 extern void atomic_orb(uint8_t *addr, uint8_t value);
39 extern void atomic_inc16(uint16_t *addr);
40 extern void atomic_dec16(uint16_t *addr);
41 extern void mmu_tlbflush_entry(caddr_t addr);
42
43 /*
44 * Each hardware page table has an htable_t describing it.
45 *
46 * We use a reference counter mechanism to detect when we can free an htable.
47 * In the implmentation the reference count is split into 2 separate counters:
48 *
49 * ht_busy is a traditional reference count of uses of the htable pointer
50 *
51 * ht_valid_cnt is a count of how references are implied by valid PTE/PTP
52 * entries in the pagetable
53 *
54 * ht_busy is only incremented by htable_lookup() or htable_create()
55 * while holding the appropriate hash_table mutex. While installing a new
56 * valid PTE or PTP, in order to increment ht_valid_cnt a thread must have
57 * done an htable_lookup() or htable_create() but not the htable_release yet.
58 *
59 * htable_release(), while holding the mutex, can know that if
60 * busy == 1 and valid_cnt == 0, the htable can be free'd.
61 *
62 * The fields have been ordered to make htable_lookup() fast. Hence,
63 * ht_hat, ht_vaddr, ht_level and ht_next need to be clustered together.
64 */
65 struct htable {
66 struct htable *ht_next; /* forward link for hash table */
67 struct hat *ht_hat; /* hat this mapping comes from */
68 uintptr_t ht_vaddr; /* virt addr at start of this table */
69 int8_t ht_level; /* page table level: 0=4K, 1=2M, ... */
70 uint8_t ht_flags; /* see below */
71 int16_t ht_busy; /* implements locking protocol */
72 int16_t ht_valid_cnt; /* # of valid entries in this table */
73 uint32_t ht_lock_cnt; /* # of locked entries in this table */
74 /* never used for kernel hat */
75 pfn_t ht_pfn; /* pfn of page of the pagetable */
76 struct htable *ht_prev; /* backward link for hash table */
77 struct htable *ht_parent; /* htable that points to this htable */
78 struct htable *ht_shares; /* for HTABLE_SHARED_PFN only */
79 };
80 typedef struct htable htable_t;
81
82 /*
83 * Flags values for htable ht_flags field:
84 *
85 * HTABLE_VLP - this is the top level htable of a VLP HAT.
86 *
87 * HTABLE_SHARED_PFN - this htable had its PFN assigned from sharing another
88 * htable. Used by hat_share() for ISM.
89 */
90 #define HTABLE_VLP (0x01)
91 #define HTABLE_SHARED_PFN (0x02)
92
93 /*
94 * The htable hash table hashing function. The 28 is so that high
95 * order bits are include in the hash index to skew the wrap
96 * around of addresses. Even though the hash buckets are stored per
97 * hat we include the value of hat pointer in the hash function so
98 * that the secondary hash for the htable mutex winds up begin different in
99 * every address space.
100 */
101 #define HTABLE_HASH(hat, va, lvl) \
102 ((((va) >> LEVEL_SHIFT(1)) + ((va) >> 28) + (lvl) + \
103 ((uintptr_t)(hat) >> 4)) & ((hat)->hat_num_hash - 1))
104
105 /*
106 * Each CPU gets a unique hat_cpu_info structure in cpu_hat_info.
107 */
108 struct hat_cpu_info {
109 kmutex_t hci_mutex; /* mutex to ensure sequential usage */
110 #if defined(__amd64)
111 pfn_t hci_vlp_pfn; /* pfn of hci_vlp_l3ptes */
112 x86pte_t *hci_vlp_l3ptes; /* VLP Level==3 pagetable (top) */
113 x86pte_t *hci_vlp_l2ptes; /* VLP Level==2 pagetable */
114 #endif /* __amd64 */
115 };
116
117
118 /*
119 * Compute the last page aligned VA mapped by an htable.
120 *
121 * Given a va and a level, compute the virtual address of the start of the
122 * next page at that level.
123 *
124 * XX64 - The check for the VA hole needs to be better generalized.
125 */
126 #if defined(__amd64)
127 #define HTABLE_NUM_PTES(ht) (((ht)->ht_flags & HTABLE_VLP) ? 4 : 512)
128
129 #define HTABLE_LAST_PAGE(ht) \
130 ((ht)->ht_level == mmu.max_level ? ((uintptr_t)0UL - MMU_PAGESIZE) :\
131 ((ht)->ht_vaddr - MMU_PAGESIZE + \
132 ((uintptr_t)HTABLE_NUM_PTES(ht) << LEVEL_SHIFT((ht)->ht_level))))
133
134 #define NEXT_ENTRY_VA(va, l) \
135 ((va & LEVEL_MASK(l)) + LEVEL_SIZE(l) == mmu.hole_start ? \
136 mmu.hole_end : (va & LEVEL_MASK(l)) + LEVEL_SIZE(l))
137
138 #elif defined(__i386)
139
140 #define HTABLE_NUM_PTES(ht) \
141 (!mmu.pae_hat ? 1024 : ((ht)->ht_level == 2 ? 4 : 512))
142
143 #define HTABLE_LAST_PAGE(ht) ((ht)->ht_vaddr - MMU_PAGESIZE + \
144 ((uintptr_t)HTABLE_NUM_PTES(ht) << LEVEL_SHIFT((ht)->ht_level)))
145
146 #define NEXT_ENTRY_VA(va, l) ((va & LEVEL_MASK(l)) + LEVEL_SIZE(l))
147
148 #endif
149
150 #if defined(_KERNEL)
151
152 /*
153 * initialization function called from hat_init()
154 */
155 extern void htable_init(void);
156
157 /*
158 * Functions to lookup, or "lookup and create", the htable corresponding
159 * to the virtual address "vaddr" in the "hat" at the given "level" of
160 * page tables. htable_lookup() may return NULL if no such entry exists.
161 *
162 * On return the given htable is marked busy (a shared lock) - this prevents
163 * the htable from being stolen or freed) until htable_release() is called.
164 *
165 * If kalloc_flag is set on an htable_create() we can't call kmem allocation
166 * routines for this htable, since it's for the kernel hat itself.
167 *
168 * htable_acquire() is used when an htable pointer has been extracted from
169 * an hment and we need to get a reference to the htable.
170 */
171 extern htable_t *htable_lookup(struct hat *hat, uintptr_t vaddr, level_t level);
172 extern htable_t *htable_create(struct hat *hat, uintptr_t vaddr, level_t level,
173 htable_t *shared);
174 extern void htable_acquire(htable_t *);
175
176 extern void htable_release(htable_t *ht);
177 extern void htable_destroy(htable_t *ht);
178
179 /*
180 * Code to free all remaining htables for a hat. Called after the hat is no
181 * longer in use by any thread.
182 */
183 extern void htable_purge_hat(struct hat *hat);
184
185 /*
186 * Find the htable, page table entry index, and PTE of the given virtual
187 * address. If not found returns NULL. When found, returns the htable_t *,
188 * sets entry, and has a hold on the htable.
189 */
190 extern htable_t *htable_getpte(struct hat *, uintptr_t, uint_t *, x86pte_t *,
191 level_t);
192
193 /*
194 * Similar to hat_getpte(), except that this only succeeds if a valid
195 * page mapping is present.
196 */
197 extern htable_t *htable_getpage(struct hat *hat, uintptr_t va, uint_t *entry);
198
199 /*
200 * Called to allocate initial/additional htables for reserve.
201 */
202 extern void htable_initial_reserve(uint_t);
203 extern void htable_reserve(uint_t);
204
205 /*
206 * Used to readjust the htable reserve after the reserve list has been used.
207 * Also called after boot to release left over boot reserves.
208 */
209 extern void htable_adjust_reserve(void);
210
211 /*
212 * return number of bytes mapped by all the htables in a given hat
213 */
214 extern size_t htable_mapped(struct hat *);
215
216
217 /*
218 * Attach initial pagetables as htables
219 */
220 extern void htable_attach(struct hat *, uintptr_t, level_t, struct htable *,
221 pfn_t);
222
223 /*
224 * Routine to find the next populated htable at or above a given virtual
225 * address. Can specify an upper limit, or HTABLE_WALK_TO_END to indicate
226 * that it should search the entire address space. Similar to
227 * hat_getpte(), but used for walking through address ranges. It can be
228 * used like this:
229 *
230 * va = ...
231 * ht = NULL;
232 * while (va < end_va) {
233 * pte = htable_walk(hat, &ht, &va, end_va);
234 * if (!pte)
235 * break;
236 *
237 * ... code to operate on page at va ...
238 *
239 * va += LEVEL_SIZE(ht->ht_level);
240 * }
241 * if (ht)
242 * htable_release(ht);
243 *
244 */
245 extern x86pte_t htable_walk(struct hat *hat, htable_t **ht, uintptr_t *va,
246 uintptr_t eaddr);
247
248 #define HTABLE_WALK_TO_END ((uintptr_t)-1)
249
250 /*
251 * Utilities convert between virtual addresses and page table entry indeces.
252 */
253 extern uint_t htable_va2entry(uintptr_t va, htable_t *ht);
254 extern uintptr_t htable_e2va(htable_t *ht, uint_t entry);
255
256 /*
257 * Interfaces that provide access to page table entries via the htable.
258 *
259 * Note that all accesses except x86pte_copy() and x86pte_zero() are atomic.
260 */
261 extern void x86pte_cpu_init(cpu_t *);
262 extern void x86pte_cpu_fini(cpu_t *);
263
264 extern x86pte_t x86pte_get(htable_t *, uint_t entry);
265
266 /*
267 * x86pte_set returns LPAGE_ERROR if it's asked to overwrite a page table
268 * link with a large page mapping.
269 */
270 #define LPAGE_ERROR (-(x86pte_t)1)
271 extern x86pte_t x86pte_set(htable_t *, uint_t entry, x86pte_t new, void *);
272
273 extern x86pte_t x86pte_inval(htable_t *ht, uint_t entry,
274 x86pte_t old, x86pte_t *ptr);
275
276 extern x86pte_t x86pte_update(htable_t *ht, uint_t entry,
277 x86pte_t old, x86pte_t new);
278
279 extern void x86pte_copy(htable_t *src, htable_t *dest, uint_t entry,
280 uint_t cnt);
281
282 /*
283 * access to a pagetable knowing only the pfn
284 */
285 extern x86pte_t *x86pte_mapin(pfn_t, uint_t, htable_t *);
286 extern void x86pte_mapout(void);
287
288 /*
289 * these are actually inlines for "lock; incw", "lock; decw", etc. instructions.
290 */
291 #define HTABLE_INC(x) atomic_inc16((uint16_t *)&x)
292 #define HTABLE_DEC(x) atomic_dec16((uint16_t *)&x)
293 #define HTABLE_LOCK_INC(ht) atomic_inc_32(&(ht)->ht_lock_cnt)
294 #define HTABLE_LOCK_DEC(ht) atomic_dec_32(&(ht)->ht_lock_cnt)
295
296 #ifdef __xpv
297 extern void xen_flush_va(caddr_t va);
298 extern void xen_gflush_va(caddr_t va, cpuset_t);
299 extern void xen_flush_tlb(void);
300 extern void xen_gflush_tlb(cpuset_t);
301 extern void xen_pin(pfn_t, level_t);
302 extern void xen_unpin(pfn_t);
303 extern int xen_kpm_page(pfn_t, uint_t);
304
305 /*
306 * The hypervisor maps all page tables into our address space read-only.
307 * Under normal circumstances, the hypervisor then handles all updates to
308 * the page tables underneath the covers for us. However, when we are
309 * trying to dump core after a hypervisor panic, the hypervisor is no
310 * longer available to do these updates. To work around the protection
311 * problem, we simply disable write-protect checking for the duration of a
312 * pagetable update operation.
313 */
314 #define XPV_ALLOW_PAGETABLE_UPDATES() \
315 { \
316 if (IN_XPV_PANIC()) \
317 setcr0((getcr0() & ~CR0_WP) & 0xffffffff); \
318 }
319 #define XPV_DISALLOW_PAGETABLE_UPDATES() \
320 { \
321 if (IN_XPV_PANIC() > 0) \
322 setcr0((getcr0() | CR0_WP) & 0xffffffff); \
323 }
324
325 #else /* __xpv */
326
327 #define XPV_ALLOW_PAGETABLE_UPDATES()
328 #define XPV_DISALLOW_PAGETABLE_UPDATES()
329
330 #endif
331
332 #endif /* _KERNEL */
333
334
335 #ifdef __cplusplus
336 }
337 #endif
338
339 #endif /* _VM_HTABLE_H */