usr/src/cmd/bhyve/mem.c

   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  * This file and its contents are supplied under the terms of the
  32  * Common Development and Distribution License ("CDDL"), version 1.0.
  33  * You may only use this file in accordance with the terms of version
  34  * 1.0 of the CDDL.
  35  *
  36  * A full copy of the text of the CDDL should have accompanied this
  37  * source.  A copy of the CDDL is also available via the Internet at
  38  * http://www.illumos.org/license/CDDL.
  39  *
  40  * Copyright 2020 Oxide Computer Company
  41  */
  42
  43 /*
  44  * Memory ranges are represented with an RB tree. On insertion, the range
  45  * is checked for overlaps. On lookup, the key has the same base and limit
  46  * so it can be searched within the range.
  47  */
  48
  49 #include <sys/cdefs.h>
  50 __FBSDID("$FreeBSD$");
  51
  52 #include <sys/types.h>
  53 #include <sys/errno.h>
  54 #include <sys/tree.h>
  55 #include <machine/vmm.h>
  56
  57 #include <assert.h>
  58 #include <err.h>
  59 #include <pthread.h>
  60 #include <stdio.h>
  61 #include <stdlib.h>
  62
  63 #include "mem.h"
  64
  65 struct mmio_rb_range {
  66         RB_ENTRY(mmio_rb_range) mr_link;        /* RB tree links */
  67         struct mem_range        mr_param;
  68         uint64_t                mr_base;
  69         uint64_t                mr_end;
  70 };
  71
  72 struct mmio_rb_tree;
  73 RB_PROTOTYPE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare);
  74
  75 static RB_HEAD(mmio_rb_tree, mmio_rb_range) mmio_rb_root, mmio_rb_fallback;
  76
  77 /*
  78  * Per-vCPU cache. Since most accesses from a vCPU will be to
  79  * consecutive addresses in a range, it makes sense to cache the
  80  * result of a lookup.
  81  */
  82 static struct mmio_rb_range     **mmio_hint;
  83 static int mmio_ncpu;
  84
  85 static pthread_rwlock_t mmio_rwlock;
  86
  87 static int
  88 mmio_rb_range_compare(struct mmio_rb_range *a, struct mmio_rb_range *b)
  89 {
  90         if (a->mr_end < b->mr_base)
  91                 return (-1);
  92         else if (a->mr_base > b->mr_end)
  93                 return (1);
  94         return (0);
  95 }
  96
  97 static int
  98 mmio_rb_lookup(struct mmio_rb_tree *rbt, uint64_t addr,
  99     struct mmio_rb_range **entry)
 100 {
 101         struct mmio_rb_range find, *res;
 102
 103         find.mr_base = find.mr_end = addr;
 104
 105         res = RB_FIND(mmio_rb_tree, rbt, &find);
 106
 107         if (res != NULL) {
 108                 *entry = res;
 109                 return (0);
 110         }
 111
 112         return (ENOENT);
 113 }
 114
 115 static int
 116 mmio_rb_add(struct mmio_rb_tree *rbt, struct mmio_rb_range *new)
 117 {
 118         struct mmio_rb_range *overlap;
 119
 120         overlap = RB_INSERT(mmio_rb_tree, rbt, new);
 121
 122         if (overlap != NULL) {
 123 #ifdef RB_DEBUG
 124                 printf("overlap detected: new %lx:%lx, tree %lx:%lx, '%s' "
 125                        "claims region already claimed for '%s'\n",
 126                        new->mr_base, new->mr_end,
 127                        overlap->mr_base, overlap->mr_end,
 128                        new->mr_param.name, overlap->mr_param.name);
 129 #endif
 130
 131                 return (EEXIST);
 132         }
 133
 134         return (0);
 135 }
 136
 137 #if 0
 138 static void
 139 mmio_rb_dump(struct mmio_rb_tree *rbt)
 140 {
 141         int perror;
 142         struct mmio_rb_range *np;
 143
 144         pthread_rwlock_rdlock(&mmio_rwlock);
 145         RB_FOREACH(np, mmio_rb_tree, rbt) {
 146                 printf(" %lx:%lx, %s\n", np->mr_base, np->mr_end,
 147                        np->mr_param.name);
 148         }
 149         perror = pthread_rwlock_unlock(&mmio_rwlock);
 150         assert(perror == 0);
 151 }
 152 #endif
 153
 154 RB_GENERATE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare);
 155
 156 typedef int (mem_cb_t)(struct vmctx *ctx, int vcpu, uint64_t gpa,
 157     struct mem_range *mr, void *arg);
 158
 159 static int
 160 mem_read(void *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size, void *arg)
 161 {
 162         int error;
 163         struct mem_range *mr = arg;
 164
 165         error = (*mr->handler)(ctx, vcpu, MEM_F_READ, gpa, size,
 166                                rval, mr->arg1, mr->arg2);
 167         return (error);
 168 }
 169
 170 static int
 171 mem_write(void *ctx, int vcpu, uint64_t gpa, uint64_t wval, int size, void *arg)
 172 {
 173         int error;
 174         struct mem_range *mr = arg;
 175
 176         error = (*mr->handler)(ctx, vcpu, MEM_F_WRITE, gpa, size,
 177                                &wval, mr->arg1, mr->arg2);
 178         return (error);
 179 }
 180
 181 static int
 182 access_memory(struct vmctx *ctx, int vcpu, uint64_t paddr, mem_cb_t *cb,
 183     void *arg)
 184 {
 185         struct mmio_rb_range *entry;
 186         int err, perror, immutable;
 187
 188         pthread_rwlock_rdlock(&mmio_rwlock);
 189         /*
 190          * First check the per-vCPU cache
 191          */
 192         if (mmio_hint[vcpu] &&
 193             paddr >= mmio_hint[vcpu]->mr_base &&
 194             paddr <= mmio_hint[vcpu]->mr_end) {
 195                 entry = mmio_hint[vcpu];
 196         } else
 197                 entry = NULL;
 198
 199         if (entry == NULL) {
 200                 if (mmio_rb_lookup(&mmio_rb_root, paddr, &entry) == 0) {
 201                         /* Update the per-vCPU cache */
 202                         mmio_hint[vcpu] = entry;
 203                 } else if (mmio_rb_lookup(&mmio_rb_fallback, paddr, &entry)) {
 204                         perror = pthread_rwlock_unlock(&mmio_rwlock);
 205                         assert(perror == 0);
 206                         return (ESRCH);
 207                 }
 208         }
 209
 210         assert(entry != NULL);
 211
 212         /*
 213          * An 'immutable' memory range is guaranteed to be never removed
 214          * so there is no need to hold 'mmio_rwlock' while calling the
 215          * handler.
 216          *
 217          * XXX writes to the PCIR_COMMAND register can cause register_mem()
 218          * to be called. If the guest is using PCI extended config space
 219          * to modify the PCIR_COMMAND register then register_mem() can
 220          * deadlock on 'mmio_rwlock'. However by registering the extended
 221          * config space window as 'immutable' the deadlock can be avoided.
 222          */
 223         immutable = (entry->mr_param.flags & MEM_F_IMMUTABLE);
 224         if (immutable) {
 225                 perror = pthread_rwlock_unlock(&mmio_rwlock);
 226                 assert(perror == 0);
 227         }
 228
 229         err = cb(ctx, vcpu, paddr, &entry->mr_param, arg);
 230
 231         if (!immutable) {
 232                 perror = pthread_rwlock_unlock(&mmio_rwlock);
 233                 assert(perror == 0);
 234         }
 235
 236
 237         return (err);
 238 }
 239
 240 static int
 241 emulate_mem_cb(struct vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr,
 242     void *arg)
 243 {
 244         struct vm_mmio *mmio;
 245         int err = 0;
 246
 247         mmio = arg;
 248
 249         if (mmio->read != 0) {
 250                 err = mem_read(ctx, vcpu, paddr, &mmio->data, mmio->bytes, mr);
 251         } else {
 252                 err = mem_write(ctx, vcpu, paddr, mmio->data, mmio->bytes, mr);
 253         }
 254
 255         return (err);
 256 }
 257
 258 int
 259 emulate_mem(struct vmctx *ctx, int vcpu, struct vm_mmio *mmio)
 260 {
 261         return (access_memory(ctx, vcpu, mmio->gpa, emulate_mem_cb, mmio));
 262 }
 263
 264 struct rw_mem_args {
 265         uint64_t *val;
 266         int size;
 267         int operation;
 268 };
 269
 270 static int
 271 rw_mem_cb(struct vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr,
 272     void *arg)
 273 {
 274         struct rw_mem_args *rma;
 275
 276         rma = arg;
 277         return (mr->handler(ctx, vcpu, rma->operation, paddr, rma->size,
 278             rma->val, mr->arg1, mr->arg2));
 279 }
 280
 281 int
 282 read_mem(struct vmctx *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size)
 283 {
 284         struct rw_mem_args rma;
 285
 286         rma.val = rval;
 287         rma.size = size;
 288         rma.operation = MEM_F_READ;
 289         return (access_memory(ctx, vcpu, gpa, rw_mem_cb, &rma));
 290 }
 291
 292 int
 293 write_mem(struct vmctx *ctx, int vcpu, uint64_t gpa, uint64_t wval, int size)
 294 {
 295         struct rw_mem_args rma;
 296
 297         rma.val = &wval;
 298         rma.size = size;
 299         rma.operation = MEM_F_WRITE;
 300         return (access_memory(ctx, vcpu, gpa, rw_mem_cb, &rma));
 301 }
 302
 303 static int
 304 register_mem_int(struct mmio_rb_tree *rbt, struct mem_range *memp)
 305 {
 306         struct mmio_rb_range *entry, *mrp;
 307         int err, perror;
 308
 309         err = 0;
 310
 311         mrp = malloc(sizeof(struct mmio_rb_range));
 312         if (mrp == NULL) {
 313                 warn("%s: couldn't allocate memory for mrp\n",
 314                      __func__);
 315                 err = ENOMEM;
 316         } else {
 317                 mrp->mr_param = *memp;
 318                 mrp->mr_base = memp->base;
 319                 mrp->mr_end = memp->base + memp->size - 1;
 320                 pthread_rwlock_wrlock(&mmio_rwlock);
 321                 if (mmio_rb_lookup(rbt, memp->base, &entry) != 0)
 322                         err = mmio_rb_add(rbt, mrp);
 323 #ifndef __FreeBSD__
 324                 else /* smatch warn: possible memory leak of 'mrp' */
 325                         free(mrp);
 326 #endif
 327                 perror = pthread_rwlock_unlock(&mmio_rwlock);
 328                 assert(perror == 0);
 329                 if (err)
 330                         free(mrp);
 331         }
 332
 333         return (err);
 334 }
 335
 336 int
 337 register_mem(struct mem_range *memp)
 338 {
 339
 340         return (register_mem_int(&mmio_rb_root, memp));
 341 }
 342
 343 int
 344 register_mem_fallback(struct mem_range *memp)
 345 {
 346
 347         return (register_mem_int(&mmio_rb_fallback, memp));
 348 }
 349
 350 int
 351 unregister_mem(struct mem_range *memp)
 352 {
 353         struct mem_range *mr;
 354         struct mmio_rb_range *entry = NULL;
 355         int err, perror, i;
 356
 357         pthread_rwlock_wrlock(&mmio_rwlock);
 358         err = mmio_rb_lookup(&mmio_rb_root, memp->base, &entry);
 359         if (err == 0) {
 360                 mr = &entry->mr_param;
 361                 assert(mr->name == memp->name);
 362                 assert(mr->base == memp->base && mr->size == memp->size);
 363                 assert((mr->flags & MEM_F_IMMUTABLE) == 0);
 364                 RB_REMOVE(mmio_rb_tree, &mmio_rb_root, entry);
 365
 366                 /* flush Per-vCPU cache */
 367                 for (i = 0; i < mmio_ncpu; i++) {
 368                         if (mmio_hint[i] == entry)
 369                                 mmio_hint[i] = NULL;
 370                 }
 371         }
 372         perror = pthread_rwlock_unlock(&mmio_rwlock);
 373         assert(perror == 0);
 374
 375         if (entry)
 376                 free(entry);
 377
 378         return (err);
 379 }
 380
 381 void
 382 init_mem(int ncpu)
 383 {
 384
 385         mmio_ncpu = ncpu;
 386         mmio_hint = calloc(ncpu, sizeof(*mmio_hint));
 387         RB_INIT(&mmio_rb_root);
 388         RB_INIT(&mmio_rb_fallback);
 389         pthread_rwlock_init(&mmio_rwlock, NULL);
 390 }