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 #pragma ident "%Z%%M% %I% %E% SMI"
30
31 #ifdef __cplusplus
32 extern "C" {
33 #endif
34
35 #if defined(__GNUC__) && defined(_ASM_INLINES) && defined(_KERNEL)
36 #include <asm/htable.h>
37 #endif
38
39 extern void atomic_andb(uint8_t *addr, uint8_t value);
40 extern void atomic_orb(uint8_t *addr, uint8_t value);
41 extern void atomic_inc16(uint16_t *addr);
42 extern void atomic_dec16(uint16_t *addr);
43 extern void mmu_tlbflush_entry(caddr_t addr);
44
45 /*
46 * Each hardware page table has an htable_t describing it.
47 *
48 * We use a reference counter mechanism to detect when we can free an htable.
49 * In the implmentation the reference count is split into 2 separate counters:
50 *
51 * ht_busy is a traditional reference count of uses of the htable pointer
52 *
53 * ht_valid_cnt is a count of how references are implied by valid PTE/PTP
54 * entries in the pagetable
55 *
56 * ht_busy is only incremented by htable_lookup() or htable_create()
57 * while holding the appropriate hash_table mutex. While installing a new
58 * valid PTE or PTP, in order to increment ht_valid_cnt a thread must have
59 * done an htable_lookup() or htable_create() but not the htable_release yet.
60 *
61 * htable_release(), while holding the mutex, can know that if
62 * busy == 1 and valid_cnt == 0, the htable can be free'd.
63 *
64 * The fields have been ordered to make htable_lookup() fast. Hence,
65 * ht_hat, ht_vaddr, ht_level and ht_next need to be clustered together.
66 */
67 struct htable {
68 struct htable *ht_next; /* forward link for hash table */
69 struct hat *ht_hat; /* hat this mapping comes from */
70 uintptr_t ht_vaddr; /* virt addr at start of this table */
71 int8_t ht_level; /* page table level: 0=4K, 1=2M, ... */
72 uint8_t ht_flags; /* see below */
73 int16_t ht_busy; /* implements locking protocol */
74 int16_t ht_valid_cnt; /* # of valid entries in this table */
75 uint32_t ht_lock_cnt; /* # of locked entries in this table */
76 /* never used for kernel hat */
77 pfn_t ht_pfn; /* pfn of page of the pagetable */
78 struct htable *ht_prev; /* backward link for hash table */
79 struct htable *ht_parent; /* htable that points to this htable */
80 struct htable *ht_shares; /* for HTABLE_SHARED_PFN only */
81 };
82 typedef struct htable htable_t;
83
84 /*
85 * Flags values for htable ht_flags field:
86 *
87 * HTABLE_VLP - this is the top level htable of a VLP HAT.
88 *
89 * HTABLE_SHARED_PFN - this htable had its PFN assigned from sharing another
90 * htable. Used by hat_share() for ISM.
91 */
92 #define HTABLE_VLP (0x01)
93 #define HTABLE_SHARED_PFN (0x02)
94
95 /*
96 * The htable hash table hashing function. The 28 is so that high
97 * order bits are include in the hash index to skew the wrap
98 * around of addresses. Even though the hash buckets are stored per
99 * hat we include the value of hat pointer in the hash function so
100 * that the secondary hash for the htable mutex winds up begin different in
101 * every address space.
102 */
103 #define HTABLE_HASH(hat, va, lvl) \
104 ((((va) >> LEVEL_SHIFT(1)) + ((va) >> 28) + (lvl) + \
105 ((uintptr_t)(hat) >> 4)) & ((hat)->hat_num_hash - 1))
106
107 /*
108 * Each CPU gets a unique hat_cpu_info structure in cpu_hat_info.
109 */
110 struct hat_cpu_info {
111 kmutex_t hci_mutex; /* mutex to ensure sequential usage */
112 #if defined(__amd64)
113 pfn_t hci_vlp_pfn; /* pfn of hci_vlp_l3ptes */
114 x86pte_t *hci_vlp_l3ptes; /* VLP Level==3 pagetable (top) */
115 x86pte_t *hci_vlp_l2ptes; /* VLP Level==2 pagetable */
116 #endif /* __amd64 */
117 };
118
119
120 /*
121 * Compute the last page aligned VA mapped by an htable.
122 *
123 * Given a va and a level, compute the virtual address of the start of the
124 * next page at that level.
125 *
126 * XX64 - The check for the VA hole needs to be better generalized.
127 */
128 #if defined(__amd64)
129 #define HTABLE_NUM_PTES(ht) (((ht)->ht_flags & HTABLE_VLP) ? 4 : 512)
130
131 #define HTABLE_LAST_PAGE(ht) \
132 ((ht)->ht_level == mmu.max_level ? ((uintptr_t)0UL - MMU_PAGESIZE) :\
133 ((ht)->ht_vaddr - MMU_PAGESIZE + \
134 ((uintptr_t)HTABLE_NUM_PTES(ht) << LEVEL_SHIFT((ht)->ht_level))))
135
136 #define NEXT_ENTRY_VA(va, l) \
137 ((va & LEVEL_MASK(l)) + LEVEL_SIZE(l) == mmu.hole_start ? \
138 mmu.hole_end : (va & LEVEL_MASK(l)) + LEVEL_SIZE(l))
139
140 #elif defined(__i386)
141
142 #define HTABLE_NUM_PTES(ht) \
143 (!mmu.pae_hat ? 1024 : ((ht)->ht_level == 2 ? 4 : 512))
144
145 #define HTABLE_LAST_PAGE(ht) ((ht)->ht_vaddr - MMU_PAGESIZE + \
146 ((uintptr_t)HTABLE_NUM_PTES(ht) << LEVEL_SHIFT((ht)->ht_level)))
147
148 #define NEXT_ENTRY_VA(va, l) ((va & LEVEL_MASK(l)) + LEVEL_SIZE(l))
149
150 #endif
151
152 #if defined(_KERNEL)
153
154 /*
155 * initialization function called from hat_init()
156 */
157 extern void htable_init(void);
158
159 /*
160 * Functions to lookup, or "lookup and create", the htable corresponding
161 * to the virtual address "vaddr" in the "hat" at the given "level" of
162 * page tables. htable_lookup() may return NULL if no such entry exists.
163 *
164 * On return the given htable is marked busy (a shared lock) - this prevents
165 * the htable from being stolen or freed) until htable_release() is called.
166 *
167 * If kalloc_flag is set on an htable_create() we can't call kmem allocation
168 * routines for this htable, since it's for the kernel hat itself.
169 *
170 * htable_acquire() is used when an htable pointer has been extracted from
171 * an hment and we need to get a reference to the htable.
172 */
173 extern htable_t *htable_lookup(struct hat *hat, uintptr_t vaddr, level_t level);
174 extern htable_t *htable_create(struct hat *hat, uintptr_t vaddr, level_t level,
175 htable_t *shared);
176 extern void htable_acquire(htable_t *);
177
178 extern void htable_release(htable_t *ht);
179 extern void htable_destroy(htable_t *ht);
180
181 /*
182 * Code to free all remaining htables for a hat. Called after the hat is no
183 * longer in use by any thread.
184 */
185 extern void htable_purge_hat(struct hat *hat);
186
187 /*
188 * Find the htable, page table entry index, and PTE of the given virtual
189 * address. If not found returns NULL. When found, returns the htable_t *,
190 * sets entry, and has a hold on the htable.
191 */
192 extern htable_t *htable_getpte(struct hat *, uintptr_t, uint_t *, x86pte_t *,
193 level_t);
194
195 /*
196 * Similar to hat_getpte(), except that this only succeeds if a valid
197 * page mapping is present.
198 */
199 extern htable_t *htable_getpage(struct hat *hat, uintptr_t va, uint_t *entry);
200
201 /*
202 * Called to allocate initial/additional htables for reserve.
203 */
204 extern void htable_initial_reserve(uint_t);
205 extern void htable_reserve(uint_t);
206
207 /*
208 * Used to readjust the htable reserve after the reserve list has been used.
209 * Also called after boot to release left over boot reserves.
210 */
211 extern void htable_adjust_reserve(void);
212
213 /*
214 * return number of bytes mapped by all the htables in a given hat
215 */
216 extern size_t htable_mapped(struct hat *);
217
218
219 /*
220 * Attach initial pagetables as htables
221 */
222 extern void htable_attach(struct hat *, uintptr_t, level_t, struct htable *,
223 pfn_t);
224
225 /*
226 * Routine to find the next populated htable at or above a given virtual
227 * address. Can specify an upper limit, or HTABLE_WALK_TO_END to indicate
228 * that it should search the entire address space. Similar to
229 * hat_getpte(), but used for walking through address ranges. It can be
230 * used like this:
231 *
232 * va = ...
233 * ht = NULL;
234 * while (va < end_va) {
235 * pte = htable_walk(hat, &ht, &va, end_va);
236 * if (!pte)
237 * break;
238 *
239 * ... code to operate on page at va ...
240 *
241 * va += LEVEL_SIZE(ht->ht_level);
242 * }
243 * if (ht)
244 * htable_release(ht);
245 *
246 */
247 extern x86pte_t htable_walk(struct hat *hat, htable_t **ht, uintptr_t *va,
248 uintptr_t eaddr);
249
250 #define HTABLE_WALK_TO_END ((uintptr_t)-1)
251
252 /*
253 * Utilities convert between virtual addresses and page table entry indeces.
254 */
255 extern uint_t htable_va2entry(uintptr_t va, htable_t *ht);
256 extern uintptr_t htable_e2va(htable_t *ht, uint_t entry);
257
258 /*
259 * Interfaces that provide access to page table entries via the htable.
260 *
261 * Note that all accesses except x86pte_copy() and x86pte_zero() are atomic.
262 */
263 extern void x86pte_cpu_init(cpu_t *);
264 extern void x86pte_cpu_fini(cpu_t *);
265
266 extern x86pte_t x86pte_get(htable_t *, uint_t entry);
267
268 /*
269 * x86pte_set returns LPAGE_ERROR if it's asked to overwrite a page table
270 * link with a large page mapping.
271 */
272 #define LPAGE_ERROR (-(x86pte_t)1)
273 extern x86pte_t x86pte_set(htable_t *, uint_t entry, x86pte_t new, void *);
274
275 extern x86pte_t x86pte_inval(htable_t *ht, uint_t entry,
276 x86pte_t old, x86pte_t *ptr);
277
278 extern x86pte_t x86pte_update(htable_t *ht, uint_t entry,
279 x86pte_t old, x86pte_t new);
280
281 extern void x86pte_copy(htable_t *src, htable_t *dest, uint_t entry,
282 uint_t cnt);
283
284 /*
285 * access to a pagetable knowing only the pfn
286 */
287 extern x86pte_t *x86pte_mapin(pfn_t, uint_t, htable_t *);
288 extern void x86pte_mapout(void);
289
290 /*
291 * these are actually inlines for "lock; incw", "lock; decw", etc. instructions.
292 */
293 #define HTABLE_INC(x) atomic_inc16((uint16_t *)&x)
294 #define HTABLE_DEC(x) atomic_dec16((uint16_t *)&x)
295 #define HTABLE_LOCK_INC(ht) atomic_add_32(&(ht)->ht_lock_cnt, 1)
296 #define HTABLE_LOCK_DEC(ht) atomic_add_32(&(ht)->ht_lock_cnt, -1)
297
298 #ifdef __xpv
299 extern void xen_flush_va(caddr_t va);
300 extern void xen_gflush_va(caddr_t va, cpuset_t);
301 extern void xen_flush_tlb(void);
302 extern void xen_gflush_tlb(cpuset_t);
303 extern void xen_pin(pfn_t, level_t);
304 extern void xen_unpin(pfn_t);
305 extern int xen_kpm_page(pfn_t, uint_t);
306
307 /*
308 * The hypervisor maps all page tables into our address space read-only.
309 * Under normal circumstances, the hypervisor then handles all updates to
310 * the page tables underneath the covers for us. However, when we are
311 * trying to dump core after a hypervisor panic, the hypervisor is no
312 * longer available to do these updates. To work around the protection
313 * problem, we simply disable write-protect checking for the duration of a
314 * pagetable update operation.
315 */
316 #define XPV_ALLOW_PAGETABLE_UPDATES() \
317 { \
318 if (IN_XPV_PANIC()) \
319 setcr0((getcr0() & ~CR0_WP) & 0xffffffff); \
320 }
321 #define XPV_DISALLOW_PAGETABLE_UPDATES() \
322 { \
323 if (IN_XPV_PANIC() > 0) \
324 setcr0((getcr0() | CR0_WP) & 0xffffffff); \
325 }
326
327 #else /* __xpv */
328
329 #define XPV_ALLOW_PAGETABLE_UPDATES()
330 #define XPV_DISALLOW_PAGETABLE_UPDATES()
331
332 #endif
333
334 #endif /* _KERNEL */
335
336
337 #ifdef __cplusplus
338 }
339 #endif
340
341 #endif /* _VM_HTABLE_H */