1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 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 * $FreeBSD$ 29 */ 30 31 /* 32 * Memory ranges are represented with an RB tree. On insertion, the range 33 * is checked for overlaps. On lookup, the key has the same base and limit 34 * so it can be searched within the range. 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include <sys/types.h> 41 #include <sys/errno.h> 42 #include <sys/tree.h> 43 #include <machine/vmm.h> 44 #include <machine/vmm_instruction_emul.h> 45 46 #include <assert.h> 47 #include <err.h> 48 #include <pthread.h> 49 #include <stdio.h> 50 #include <stdlib.h> 51 52 #include "mem.h" 53 54 struct mmio_rb_range { 55 RB_ENTRY(mmio_rb_range) mr_link; /* RB tree links */ 56 struct mem_range mr_param; 57 uint64_t mr_base; 58 uint64_t mr_end; 59 }; 60 61 struct mmio_rb_tree; 62 RB_PROTOTYPE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare); 63 64 RB_HEAD(mmio_rb_tree, mmio_rb_range) mmio_rb_root, mmio_rb_fallback; 65 66 /* 67 * Per-vCPU cache. Since most accesses from a vCPU will be to 68 * consecutive addresses in a range, it makes sense to cache the 69 * result of a lookup. 70 */ 71 static struct mmio_rb_range *mmio_hint[VM_MAXCPU]; 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\n", 113 new->mr_base, new->mr_end, 114 overlap->mr_base, overlap->mr_end); 115 #endif 116 117 return (EEXIST); 118 } 119 120 return (0); 121 } 122 123 #if 0 124 static void 125 mmio_rb_dump(struct mmio_rb_tree *rbt) 126 { 127 int perror; 128 struct mmio_rb_range *np; 129 130 pthread_rwlock_rdlock(&mmio_rwlock); 131 RB_FOREACH(np, mmio_rb_tree, rbt) { 132 printf(" %lx:%lx, %s\n", np->mr_base, np->mr_end, 133 np->mr_param.name); 134 } 135 perror = pthread_rwlock_unlock(&mmio_rwlock); 136 assert(perror == 0); 137 } 138 #endif 139 140 RB_GENERATE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare); 141 142 typedef int (mem_cb_t)(struct vmctx *ctx, int vcpu, uint64_t gpa, 143 struct mem_range *mr, void *arg); 144 145 static int 146 mem_read(void *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size, void *arg) 147 { 148 int error; 149 struct mem_range *mr = arg; 150 151 error = (*mr->handler)(ctx, vcpu, MEM_F_READ, gpa, size, 152 rval, mr->arg1, mr->arg2); 153 return (error); 154 } 155 156 static int 157 mem_write(void *ctx, int vcpu, uint64_t gpa, uint64_t wval, int size, void *arg) 158 { 159 int error; 160 struct mem_range *mr = arg; 161 162 error = (*mr->handler)(ctx, vcpu, MEM_F_WRITE, gpa, size, 163 &wval, mr->arg1, mr->arg2); 164 return (error); 165 } 166 167 static int 168 access_memory(struct vmctx *ctx, int vcpu, uint64_t paddr, mem_cb_t *cb, 169 void *arg) 170 { 171 struct mmio_rb_range *entry; 172 int err, perror, immutable; 173 174 pthread_rwlock_rdlock(&mmio_rwlock); 175 /* 176 * First check the per-vCPU cache 177 */ 178 if (mmio_hint[vcpu] && 179 paddr >= mmio_hint[vcpu]->mr_base && 180 paddr <= mmio_hint[vcpu]->mr_end) { 181 entry = mmio_hint[vcpu]; 182 } else 183 entry = NULL; 184 185 if (entry == NULL) { 186 if (mmio_rb_lookup(&mmio_rb_root, paddr, &entry) == 0) { 187 /* Update the per-vCPU cache */ 188 mmio_hint[vcpu] = entry; 189 } else if (mmio_rb_lookup(&mmio_rb_fallback, paddr, &entry)) { 190 perror = pthread_rwlock_unlock(&mmio_rwlock); 191 assert(perror == 0); 192 return (ESRCH); 193 } 194 } 195 196 assert(entry != NULL); 197 198 /* 199 * An 'immutable' memory range is guaranteed to be never removed 200 * so there is no need to hold 'mmio_rwlock' while calling the 201 * handler. 202 * 203 * XXX writes to the PCIR_COMMAND register can cause register_mem() 204 * to be called. If the guest is using PCI extended config space 205 * to modify the PCIR_COMMAND register then register_mem() can 206 * deadlock on 'mmio_rwlock'. However by registering the extended 207 * config space window as 'immutable' the deadlock can be avoided. 208 */ 209 immutable = (entry->mr_param.flags & MEM_F_IMMUTABLE); 210 if (immutable) { 211 perror = pthread_rwlock_unlock(&mmio_rwlock); 212 assert(perror == 0); 213 } 214 215 err = cb(ctx, vcpu, paddr, &entry->mr_param, arg); 216 217 if (!immutable) { 218 perror = pthread_rwlock_unlock(&mmio_rwlock); 219 assert(perror == 0); 220 } 221 222 223 return (err); 224 } 225 226 struct emulate_mem_args { 227 struct vie *vie; 228 struct vm_guest_paging *paging; 229 }; 230 231 static int 232 emulate_mem_cb(struct vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr, 233 void *arg) 234 { 235 struct emulate_mem_args *ema; 236 237 ema = arg; 238 return (vmm_emulate_instruction(ctx, vcpu, paddr, ema->vie, ema->paging, 239 mem_read, mem_write, mr)); 240 } 241 242 int 243 emulate_mem(struct vmctx *ctx, int vcpu, uint64_t paddr, struct vie *vie, 244 struct vm_guest_paging *paging) 245 246 { 247 struct emulate_mem_args ema; 248 249 ema.vie = vie; 250 ema.paging = paging; 251 return (access_memory(ctx, 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 vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr, 262 void *arg) 263 { 264 struct rw_mem_args *rma; 265 266 rma = arg; 267 return (mr->handler(ctx, vcpu, rma->operation, paddr, rma->size, 268 rma->val, mr->arg1, mr->arg2)); 269 } 270 271 int 272 read_mem(struct vmctx *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size) 273 { 274 struct rw_mem_args rma; 275 276 rma.val = rval; 277 rma.size = size; 278 rma.operation = MEM_F_READ; 279 return (access_memory(ctx, vcpu, gpa, rw_mem_cb, &rma)); 280 } 281 282 int 283 write_mem(struct vmctx *ctx, int vcpu, uint64_t gpa, uint64_t wval, int size) 284 { 285 struct rw_mem_args rma; 286 287 rma.val = &wval; 288 rma.size = size; 289 rma.operation = MEM_F_WRITE; 290 return (access_memory(ctx, vcpu, gpa, rw_mem_cb, &rma)); 291 } 292 293 static int 294 register_mem_int(struct mmio_rb_tree *rbt, struct mem_range *memp) 295 { 296 struct mmio_rb_range *entry, *mrp; 297 int err, perror; 298 299 err = 0; 300 301 mrp = malloc(sizeof(struct mmio_rb_range)); 302 if (mrp == NULL) { 303 warn("%s: couldn't allocate memory for mrp\n", 304 __func__); 305 err = ENOMEM; 306 } else { 307 mrp->mr_param = *memp; 308 mrp->mr_base = memp->base; 309 mrp->mr_end = memp->base + memp->size - 1; 310 pthread_rwlock_wrlock(&mmio_rwlock); 311 if (mmio_rb_lookup(rbt, memp->base, &entry) != 0) 312 err = mmio_rb_add(rbt, mrp); 313 perror = pthread_rwlock_unlock(&mmio_rwlock); 314 assert(perror == 0); 315 if (err) 316 free(mrp); 317 } 318 319 return (err); 320 } 321 322 int 323 register_mem(struct mem_range *memp) 324 { 325 326 return (register_mem_int(&mmio_rb_root, memp)); 327 } 328 329 int 330 register_mem_fallback(struct mem_range *memp) 331 { 332 333 return (register_mem_int(&mmio_rb_fallback, memp)); 334 } 335 336 int 337 unregister_mem(struct mem_range *memp) 338 { 339 struct mem_range *mr; 340 struct mmio_rb_range *entry = NULL; 341 int err, perror, i; 342 343 pthread_rwlock_wrlock(&mmio_rwlock); 344 err = mmio_rb_lookup(&mmio_rb_root, memp->base, &entry); 345 if (err == 0) { 346 mr = &entry->mr_param; 347 assert(mr->name == memp->name); 348 assert(mr->base == memp->base && mr->size == memp->size); 349 assert((mr->flags & MEM_F_IMMUTABLE) == 0); 350 RB_REMOVE(mmio_rb_tree, &mmio_rb_root, entry); 351 352 /* flush Per-vCPU cache */ 353 for (i=0; i < VM_MAXCPU; i++) { 354 if (mmio_hint[i] == entry) 355 mmio_hint[i] = NULL; 356 } 357 } 358 perror = pthread_rwlock_unlock(&mmio_rwlock); 359 assert(perror == 0); 360 361 if (entry) 362 free(entry); 363 364 return (err); 365 } 366 367 void 368 init_mem(void) 369 { 370 371 RB_INIT(&mmio_rb_root); 372 RB_INIT(&mmio_rb_fallback); 373 pthread_rwlock_init(&mmio_rwlock, NULL); 374 } 375