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, Version 1.0 only 6 * (the "License"). You may not use this file except in compliance 7 * with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or http://www.opensolaris.org/os/licensing. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22 /* 23 * Copyright 2005 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #ifndef _VM_HTABLE_H 28 #define _VM_HTABLE_H 29 30 #pragma ident "%Z%%M% %I% %E% SMI" 31 32 #ifdef __cplusplus 33 extern "C" { 34 #endif 35 36 #if defined(__GNUC__) && defined(_ASM_INLINES) && defined(_KERNEL) 37 #include <asm/htable.h> 38 #endif 39 40 extern void atomic_andb(uint8_t *addr, uint8_t value); 41 extern void atomic_orb(uint8_t *addr, uint8_t value); 42 extern void atomic_inc16(uint16_t *addr); 43 extern void atomic_dec16(uint16_t *addr); 44 extern void mmu_tlbflush_entry(caddr_t addr); 45 46 /* 47 * Each hardware page table has an htable_t describing it. 48 * 49 * We use a reference counter mechanism to detect when we can free an htable. 50 * In the implmentation the reference count is split into 2 separate counters: 51 * 52 * ht_busy is a traditional reference count of uses of the htable pointer 53 * 54 * ht_valid_cnt is a count of how references are implied by valid PTE/PTP 55 * entries in the pagetable 56 * 57 * ht_busy is only incremented by htable_lookup() or htable_create() 58 * while holding the appropriate hash_table mutex. While installing a new 59 * valid PTE or PTP, in order to increment ht_valid_cnt a thread must have 60 * done an htable_lookup() or htable_create() but not the htable_release yet. 61 * 62 * htable_release(), while holding the mutex, can know that if 63 * busy == 1 and valid_cnt == 0, the htable can be free'd. 64 * 65 * The fields have been ordered to make htable_lookup() fast. Hence, 66 * ht_hat, ht_vaddr, ht_level and ht_next need to be clustered together. 67 */ 68 struct htable { 69 struct htable *ht_next; /* forward link for hash table */ 70 struct hat *ht_hat; /* hat this mapping comes from */ 71 uintptr_t ht_vaddr; /* virt addr at start of this table */ 72 level_t ht_level; /* page table level: 0=4K, 1=2M, ... */ 73 uint16_t ht_flags; /* see below */ 74 int16_t ht_busy; /* implements locking protocol */ 75 uint16_t ht_num_ptes; /* # of PTEs in page table */ 76 int16_t ht_valid_cnt; /* # of valid entries in this table */ 77 uint32_t ht_lock_cnt; /* # of locked entries in this table */ 78 /* never used for kernel hat */ 79 pfn_t ht_pfn; /* pfn of page of the pagetable */ 80 struct htable *ht_prev; /* backward link for hash table */ 81 struct htable *ht_parent; /* htable that points to this htable */ 82 struct htable *ht_shares; /* for HTABLE_SHARED_PFN only */ 83 }; 84 typedef struct htable htable_t; 85 86 /* 87 * Flags values for htable ht_flags field: 88 * 89 * HTABLE_VLP - this is the top level htable of a VLP HAT. 90 * 91 * HTABLE_SHARED_PFN - this htable had it's PFN assigned from sharing another 92 * htable. Used by hat_share() for ISM. 93 */ 94 #define HTABLE_VLP (0x0001) 95 #define HTABLE_SHARED_PFN (0x0002) 96 97 /* 98 * The htable hash table hashing function. The 28 is so that high 99 * order bits are include in the hash index to skew the wrap 100 * around of addresses. 101 */ 102 #define HTABLE_HASH(hat, va, lvl) \ 103 ((((va) >> LEVEL_SHIFT(1)) + ((va) >> 28) + (lvl)) & \ 104 ((hat)->hat_num_hash - 1)) 105 106 /* 107 * For 32 bit, access to page table entries is done via the page table's PFN and 108 * the index of the PTE. We use a CPU specific mapping (a la ppcopy) to map 109 * in page tables on an "as needed" basis. 110 * 111 * 64 bit kernels will use seg_kpm style mappings and avoid any overhead. 112 * 113 * The code uses compare and swap instructions to read/write PTE's to 114 * avoid atomicity problems, since PTEs can be 8 bytes on 32 bit systems. 115 * Again this can be optimized on 64 bit systems, since aligned load/store 116 * will naturally be atomic. 117 * 118 * Each CPU gets a unique hat_cpu_info structure in cpu_hat_info. 119 */ 120 struct hat_cpu_info { 121 pfn_t hci_mapped_pfn; /* pfn of currently mapped page table */ 122 x86pte_t *hci_pagetable_va; /* VA to use for mappings */ 123 x86pte_t *hci_kernel_pte; /* kernel PTE for cpu_pagetable_va */ 124 kmutex_t hci_mutex; /* mutex to ensure sequential usage */ 125 #if defined(__amd64) 126 pfn_t hci_vlp_pfn; /* pfn of hci_vlp_l3ptes */ 127 x86pte_t *hci_vlp_l3ptes; /* VLP Level==3 pagetable (top) */ 128 x86pte_t *hci_vlp_l2ptes; /* VLP Level==2 pagetable */ 129 #endif /* __amd64 */ 130 }; 131 132 133 /* 134 * Compute the last page aligned VA mapped by an htable. 135 * 136 * Given a va and a level, compute the virtual address of the start of the 137 * next page at that level. 138 * 139 * XX64 - The check for the VA hole needs to be better generalized. 140 */ 141 #if defined(__amd64) 142 143 #define HTABLE_LAST_PAGE(ht) \ 144 ((ht)->ht_level == mmu.max_level ? ((uintptr_t)0UL - MMU_PAGESIZE) :\ 145 ((ht)->ht_vaddr - MMU_PAGESIZE + \ 146 ((uintptr_t)((ht)->ht_num_ptes) << LEVEL_SHIFT((ht)->ht_level)))) 147 148 #define NEXT_ENTRY_VA(va, l) \ 149 ((va & LEVEL_MASK(l)) + LEVEL_SIZE(l) == mmu.hole_start ? \ 150 mmu.hole_end : (va & LEVEL_MASK(l)) + LEVEL_SIZE(l)) 151 152 #elif defined(__i386) 153 154 #define HTABLE_LAST_PAGE(ht) ((ht)->ht_vaddr - MMU_PAGESIZE + \ 155 ((uintptr_t)((ht)->ht_num_ptes) << LEVEL_SHIFT((ht)->ht_level))) 156 157 #define NEXT_ENTRY_VA(va, l) ((va & LEVEL_MASK(l)) + LEVEL_SIZE(l)) 158 159 #endif 160 161 #if defined(_KERNEL) 162 163 /* 164 * initialization function called from hat_init() 165 */ 166 extern void htable_init(void); 167 168 /* 169 * Functions to lookup, or "lookup and create", the htable corresponding 170 * to the virtual address "vaddr" in the "hat" at the given "level" of 171 * page tables. htable_lookup() may return NULL if no such entry exists. 172 * 173 * On return the given htable is marked busy (a shared lock) - this prevents 174 * the htable from being stolen or freed) until htable_release() is called. 175 * 176 * If kalloc_flag is set on an htable_create() we can't call kmem allocation 177 * routines for this htable, since it's for the kernel hat itself. 178 * 179 * htable_acquire() is used when an htable pointer has been extracted from 180 * an hment and we need to get a reference to the htable. 181 */ 182 extern htable_t *htable_lookup(struct hat *hat, uintptr_t vaddr, level_t level); 183 extern htable_t *htable_create(struct hat *hat, uintptr_t vaddr, level_t level, 184 htable_t *shared); 185 extern void htable_acquire(htable_t *); 186 187 extern void htable_release(htable_t *ht); 188 189 /* 190 * Code to free all remaining htables for a hat. Called after the hat is no 191 * longer in use by any thread. 192 */ 193 extern void htable_purge_hat(struct hat *hat); 194 195 /* 196 * Find the htable, page table entry index, and PTE of the given virtual 197 * address. If not found returns NULL. When found, returns the htable_t *, 198 * sets entry, and has a hold on the htable. 199 */ 200 extern htable_t *htable_getpte(struct hat *, uintptr_t, uint_t *, x86pte_t *, 201 level_t); 202 203 /* 204 * Similar to hat_getpte(), except that this only succeeds if a valid 205 * page mapping is present. 206 */ 207 extern htable_t *htable_getpage(struct hat *hat, uintptr_t va, uint_t *entry); 208 209 /* 210 * Called to allocate initial/additional htables for reserve. 211 */ 212 extern void htable_initial_reserve(uint_t); 213 extern void htable_reserve(uint_t); 214 215 /* 216 * Used to readjust the htable reserve after the reserve list has been used. 217 * Also called after boot to release left over boot reserves. 218 */ 219 extern void htable_adjust_reserve(void); 220 221 /* 222 * Routine to find the next populated htable at or above a given virtual 223 * address. Can specify an upper limit, or HTABLE_WALK_TO_END to indicate 224 * that it should search the entire address space. Similar to 225 * hat_getpte(), but used for walking through address ranges. It can be 226 * used like this: 227 * 228 * va = ... 229 * ht = NULL; 230 * while (va < end_va) { 231 * pte = htable_walk(hat, &ht, &va, end_va); 232 * if (!pte) 233 * break; 234 * 235 * ... code to operate on page at va ... 236 * 237 * va += LEVEL_SIZE(ht->ht_level); 238 * } 239 * if (ht) 240 * htable_release(ht); 241 * 242 */ 243 extern x86pte_t htable_walk(struct hat *hat, htable_t **ht, uintptr_t *va, 244 uintptr_t eaddr); 245 246 #define HTABLE_WALK_TO_END ((uintptr_t)-1) 247 248 /* 249 * Utilities convert between virtual addresses and page table entry indeces. 250 */ 251 extern uint_t htable_va2entry(uintptr_t va, htable_t *ht); 252 extern uintptr_t htable_e2va(htable_t *ht, uint_t entry); 253 254 /* 255 * Interfaces that provide access to page table entries via the htable. 256 * 257 * Note that all accesses except x86pte_copy() and x86pte_zero() are atomic. 258 */ 259 extern void x86pte_cpu_init(cpu_t *, void *); 260 261 extern x86pte_t x86pte_get(htable_t *, uint_t entry); 262 263 extern x86pte_t x86pte_set(htable_t *, uint_t entry, x86pte_t new, void *); 264 265 extern x86pte_t x86pte_invalidate_pfn(htable_t *ht, uint_t entry, pfn_t pfn, 266 void *pte_ptr); 267 268 extern x86pte_t x86pte_update(htable_t *ht, uint_t entry, 269 x86pte_t old, x86pte_t new); 270 271 extern void x86pte_copy(htable_t *src, htable_t *dest, uint_t entry, 272 uint_t cnt); 273 274 extern void x86pte_zero(htable_t *ht, uint_t entry, uint_t cnt); 275 276 277 /* 278 * these are actually inlines for "lock; incw", "lock; decw", etc. instructions. 279 */ 280 #define HTABLE_INC(x) atomic_inc16((uint16_t *)&x) 281 #define HTABLE_DEC(x) atomic_dec16((uint16_t *)&x) 282 #define HTABLE_LOCK_INC(ht) atomic_add_32(&(ht)->ht_lock_cnt, 1) 283 #define HTABLE_LOCK_DEC(ht) atomic_add_32(&(ht)->ht_lock_cnt, -1) 284 285 #endif /* _KERNEL */ 286 287 288 #ifdef __cplusplus 289 } 290 #endif 291 292 #endif /* _VM_HTABLE_H */ 293