1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2012 NetApp, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 /* 30 * Memory ranges are represented with an RB tree. On insertion, the range 31 * is checked for overlaps. On lookup, the key has the same base and limit 32 * so it can be searched within the range. 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 37 38 #include <sys/types.h> 39 #include <sys/errno.h> 40 #include <sys/tree.h> 41 #include <machine/vmm.h> 42 #include <machine/vmm_instruction_emul.h> 43 44 #include <assert.h> 45 #include <err.h> 46 #include <pthread.h> 47 #include <stdio.h> 48 #include <stdlib.h> 49 #include <vmmapi.h> 50 51 #include "mem.h" 52 53 struct mmio_rb_range { 54 RB_ENTRY(mmio_rb_range) mr_link; /* RB tree links */ 55 struct mem_range mr_param; 56 uint64_t mr_base; 57 uint64_t mr_end; 58 }; 59 60 struct mmio_rb_tree; 61 RB_PROTOTYPE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare); 62 63 static RB_HEAD(mmio_rb_tree, mmio_rb_range) mmio_rb_root, mmio_rb_fallback; 64 65 /* 66 * Per-vCPU cache. Since most accesses from a vCPU will be to 67 * consecutive addresses in a range, it makes sense to cache the 68 * result of a lookup. 69 */ 70 static struct mmio_rb_range **mmio_hint; 71 static int mmio_ncpu; 72 73 static pthread_rwlock_t mmio_rwlock; 74 75 static int 76 mmio_rb_range_compare(struct mmio_rb_range *a, struct mmio_rb_range *b) 77 { 78 if (a->mr_end < b->mr_base) 79 return (-1); 80 else if (a->mr_base > b->mr_end) 81 return (1); 82 return (0); 83 } 84 85 static int 86 mmio_rb_lookup(struct mmio_rb_tree *rbt, uint64_t addr, 87 struct mmio_rb_range **entry) 88 { 89 struct mmio_rb_range find, *res; 90 91 find.mr_base = find.mr_end = addr; 92 93 res = RB_FIND(mmio_rb_tree, rbt, &find); 94 95 if (res != NULL) { 96 *entry = res; 97 return (0); 98 } 99 100 return (ENOENT); 101 } 102 103 static int 104 mmio_rb_add(struct mmio_rb_tree *rbt, struct mmio_rb_range *new) 105 { 106 struct mmio_rb_range *overlap; 107 108 overlap = RB_INSERT(mmio_rb_tree, rbt, new); 109 110 if (overlap != NULL) { 111 #ifdef RB_DEBUG 112 printf("overlap detected: new %lx:%lx, tree %lx:%lx, '%s' " 113 "claims region already claimed for '%s'\n", 114 new->mr_base, new->mr_end, 115 overlap->mr_base, overlap->mr_end, 116 new->mr_param.name, overlap->mr_param.name); 117 #endif 118 119 return (EEXIST); 120 } 121 122 return (0); 123 } 124 125 #if 0 126 static void 127 mmio_rb_dump(struct mmio_rb_tree *rbt) 128 { 129 int perror; 130 struct mmio_rb_range *np; 131 132 pthread_rwlock_rdlock(&mmio_rwlock); 133 RB_FOREACH(np, mmio_rb_tree, rbt) { 134 printf(" %lx:%lx, %s\n", np->mr_base, np->mr_end, 135 np->mr_param.name); 136 } 137 perror = pthread_rwlock_unlock(&mmio_rwlock); 138 assert(perror == 0); 139 } 140 #endif 141 142 RB_GENERATE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare); 143 144 typedef int (mem_cb_t)(struct vcpu *vcpu, uint64_t gpa, struct mem_range *mr, 145 void *arg); 146 147 static int 148 mem_read(struct vcpu *vcpu, uint64_t gpa, uint64_t *rval, int size, void *arg) 149 { 150 int error; 151 struct mem_range *mr = arg; 152 153 error = (*mr->handler)(vcpu, MEM_F_READ, gpa, size, rval, mr->arg1, 154 mr->arg2); 155 return (error); 156 } 157 158 static int 159 mem_write(struct vcpu *vcpu, uint64_t gpa, uint64_t wval, int size, void *arg) 160 { 161 int error; 162 struct mem_range *mr = arg; 163 164 error = (*mr->handler)(vcpu, MEM_F_WRITE, gpa, size, &wval, mr->arg1, 165 mr->arg2); 166 return (error); 167 } 168 169 static int 170 access_memory(struct vcpu *vcpu, uint64_t paddr, mem_cb_t *cb, void *arg) 171 { 172 struct mmio_rb_range *entry; 173 int err, perror, immutable, vcpuid; 174 175 vcpuid = vcpu_id(vcpu); 176 pthread_rwlock_rdlock(&mmio_rwlock); 177 /* 178 * First check the per-vCPU cache 179 */ 180 if (mmio_hint[vcpuid] && 181 paddr >= mmio_hint[vcpuid]->mr_base && 182 paddr <= mmio_hint[vcpuid]->mr_end) { 183 entry = mmio_hint[vcpuid]; 184 } else 185 entry = NULL; 186 187 if (entry == NULL) { 188 if (mmio_rb_lookup(&mmio_rb_root, paddr, &entry) == 0) { 189 /* Update the per-vCPU cache */ 190 mmio_hint[vcpuid] = entry; 191 } else if (mmio_rb_lookup(&mmio_rb_fallback, paddr, &entry)) { 192 perror = pthread_rwlock_unlock(&mmio_rwlock); 193 assert(perror == 0); 194 return (ESRCH); 195 } 196 } 197 198 assert(entry != NULL); 199 200 /* 201 * An 'immutable' memory range is guaranteed to be never removed 202 * so there is no need to hold 'mmio_rwlock' while calling the 203 * handler. 204 * 205 * XXX writes to the PCIR_COMMAND register can cause register_mem() 206 * to be called. If the guest is using PCI extended config space 207 * to modify the PCIR_COMMAND register then register_mem() can 208 * deadlock on 'mmio_rwlock'. However by registering the extended 209 * config space window as 'immutable' the deadlock can be avoided. 210 */ 211 immutable = (entry->mr_param.flags & MEM_F_IMMUTABLE); 212 if (immutable) { 213 perror = pthread_rwlock_unlock(&mmio_rwlock); 214 assert(perror == 0); 215 } 216 217 err = cb(vcpu, paddr, &entry->mr_param, arg); 218 219 if (!immutable) { 220 perror = pthread_rwlock_unlock(&mmio_rwlock); 221 assert(perror == 0); 222 } 223 224 return (err); 225 } 226 227 struct emulate_mem_args { 228 struct vie *vie; 229 struct vm_guest_paging *paging; 230 }; 231 232 static int 233 emulate_mem_cb(struct vcpu *vcpu, uint64_t paddr, struct mem_range *mr, 234 void *arg) 235 { 236 struct emulate_mem_args *ema; 237 238 ema = arg; 239 return (vmm_emulate_instruction(vcpu, paddr, ema->vie, ema->paging, 240 mem_read, mem_write, mr)); 241 } 242 243 int 244 emulate_mem(struct vcpu *vcpu, uint64_t paddr, struct vie *vie, 245 struct vm_guest_paging *paging) 246 { 247 struct emulate_mem_args ema; 248 249 ema.vie = vie; 250 ema.paging = paging; 251 return (access_memory(vcpu, paddr, emulate_mem_cb, &ema)); 252 } 253 254 struct rw_mem_args { 255 uint64_t *val; 256 int size; 257 int operation; 258 }; 259 260 static int 261 rw_mem_cb(struct vcpu *vcpu, uint64_t paddr, struct mem_range *mr, void *arg) 262 { 263 struct rw_mem_args *rma; 264 265 rma = arg; 266 return (mr->handler(vcpu, rma->operation, paddr, rma->size, 267 rma->val, mr->arg1, mr->arg2)); 268 } 269 270 int 271 read_mem(struct vcpu *vcpu, uint64_t gpa, uint64_t *rval, int size) 272 { 273 struct rw_mem_args rma; 274 275 rma.val = rval; 276 rma.size = size; 277 rma.operation = MEM_F_READ; 278 return (access_memory(vcpu, gpa, rw_mem_cb, &rma)); 279 } 280 281 int 282 write_mem(struct vcpu *vcpu, uint64_t gpa, uint64_t wval, int size) 283 { 284 struct rw_mem_args rma; 285 286 rma.val = &wval; 287 rma.size = size; 288 rma.operation = MEM_F_WRITE; 289 return (access_memory(vcpu, gpa, rw_mem_cb, &rma)); 290 } 291 292 static int 293 register_mem_int(struct mmio_rb_tree *rbt, struct mem_range *memp) 294 { 295 struct mmio_rb_range *entry, *mrp; 296 int err, perror; 297 298 err = 0; 299 300 mrp = malloc(sizeof(struct mmio_rb_range)); 301 if (mrp == NULL) { 302 warn("%s: couldn't allocate memory for mrp\n", 303 __func__); 304 err = ENOMEM; 305 } else { 306 mrp->mr_param = *memp; 307 mrp->mr_base = memp->base; 308 mrp->mr_end = memp->base + memp->size - 1; 309 pthread_rwlock_wrlock(&mmio_rwlock); 310 if (mmio_rb_lookup(rbt, memp->base, &entry) != 0) 311 err = mmio_rb_add(rbt, mrp); 312 perror = pthread_rwlock_unlock(&mmio_rwlock); 313 assert(perror == 0); 314 if (err) 315 free(mrp); 316 } 317 318 return (err); 319 } 320 321 int 322 register_mem(struct mem_range *memp) 323 { 324 325 return (register_mem_int(&mmio_rb_root, memp)); 326 } 327 328 int 329 register_mem_fallback(struct mem_range *memp) 330 { 331 332 return (register_mem_int(&mmio_rb_fallback, memp)); 333 } 334 335 int 336 unregister_mem(struct mem_range *memp) 337 { 338 struct mem_range *mr; 339 struct mmio_rb_range *entry = NULL; 340 int err, perror, i; 341 342 pthread_rwlock_wrlock(&mmio_rwlock); 343 err = mmio_rb_lookup(&mmio_rb_root, memp->base, &entry); 344 if (err == 0) { 345 mr = &entry->mr_param; 346 assert(mr->name == memp->name); 347 assert(mr->base == memp->base && mr->size == memp->size); 348 assert((mr->flags & MEM_F_IMMUTABLE) == 0); 349 RB_REMOVE(mmio_rb_tree, &mmio_rb_root, entry); 350 351 /* flush Per-vCPU cache */ 352 for (i = 0; i < mmio_ncpu; i++) { 353 if (mmio_hint[i] == entry) 354 mmio_hint[i] = NULL; 355 } 356 } 357 perror = pthread_rwlock_unlock(&mmio_rwlock); 358 assert(perror == 0); 359 360 if (entry) 361 free(entry); 362 363 return (err); 364 } 365 366 void 367 init_mem(int ncpu) 368 { 369 370 mmio_ncpu = ncpu; 371 mmio_hint = calloc(ncpu, sizeof(*mmio_hint)); 372 RB_INIT(&mmio_rb_root); 373 RB_INIT(&mmio_rb_fallback); 374 pthread_rwlock_init(&mmio_rwlock, NULL); 375 } 376