kvm-all.c

   1 /*
   2  * QEMU KVM support
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *           Red Hat, Inc. 2008
   6  *
   7  * Authors:
   8  *  Anthony Liguori   <aliguori@us.ibm.com>
   9  *  Glauber Costa     <gcosta@redhat.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  12  * See the COPYING file in the top-level directory.
  13  *
  14  */
  15
  16 #include <sys/types.h>
  17 #include <sys/ioctl.h>
  18 #include <sys/mman.h>
  19 #include <stdarg.h>
  20
  21 #include <linux/kvm.h>
  22
  23 #include "qemu-common.h"
  24 #include "sysemu.h"
  25 #include "gdbstub.h"
  26 #include "kvm.h"
  27
  28 /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */
  29 #define PAGE_SIZE TARGET_PAGE_SIZE
  30
  31 //#define DEBUG_KVM
  32
  33 #ifdef DEBUG_KVM
  34 #define dprintf(fmt, ...) \
  35     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  36 #else
  37 #define dprintf(fmt, ...) \
  38     do { } while (0)
  39 #endif
  40
  41 typedef struct KVMSlot
  42 {
  43     target_phys_addr_t start_addr;
  44     ram_addr_t memory_size;
  45     ram_addr_t phys_offset;
  46     int slot;
  47     int flags;
  48 } KVMSlot;
  49
  50 typedef struct kvm_dirty_log KVMDirtyLog;
  51
  52 int kvm_allowed = 0;
  53
  54 struct KVMState
  55 {
  56     KVMSlot slots[32];
  57     int fd;
  58     int vmfd;
  59     int coalesced_mmio;
  60     int broken_set_mem_region;
  61     int migration_log;
  62 #ifdef KVM_CAP_SET_GUEST_DEBUG
  63     struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
  64 #endif
  65 };
  66
  67 static KVMState *kvm_state;
  68
  69 static KVMSlot *kvm_alloc_slot(KVMState *s)
  70 {
  71     int i;
  72
  73     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
  74         /* KVM private memory slots */
  75         if (i >= 8 && i < 12)
  76             continue;
  77         if (s->slots[i].memory_size == 0)
  78             return &s->slots[i];
  79     }
  80
  81     fprintf(stderr, "%s: no free slot available\n", __func__);
  82     abort();
  83 }
  84
  85 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
  86                                          target_phys_addr_t start_addr,
  87                                          target_phys_addr_t end_addr)
  88 {
  89     int i;
  90
  91     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
  92         KVMSlot *mem = &s->slots[i];
  93
  94         if (start_addr == mem->start_addr &&
  95             end_addr == mem->start_addr + mem->memory_size) {
  96             return mem;
  97         }
  98     }
  99
 100     return NULL;
 101 }
 102
 103 /*
 104  * Find overlapping slot with lowest start address
 105  */
 106 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
 107                                             target_phys_addr_t start_addr,
 108                                             target_phys_addr_t end_addr)
 109 {
 110     KVMSlot *found = NULL;
 111     int i;
 112
 113     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
 114         KVMSlot *mem = &s->slots[i];
 115
 116         if (mem->memory_size == 0 ||
 117             (found && found->start_addr < mem->start_addr)) {
 118             continue;
 119         }
 120
 121         if (end_addr > mem->start_addr &&
 122             start_addr < mem->start_addr + mem->memory_size) {
 123             found = mem;
 124         }
 125     }
 126
 127     return found;
 128 }
 129
 130 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
 131 {
 132     struct kvm_userspace_memory_region mem;
 133
 134     mem.slot = slot->slot;
 135     mem.guest_phys_addr = slot->start_addr;
 136     mem.memory_size = slot->memory_size;
 137     mem.userspace_addr = (unsigned long)qemu_get_ram_ptr(slot->phys_offset);
 138     mem.flags = slot->flags;
 139     if (s->migration_log) {
 140         mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
 141     }
 142     return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
 143 }
 144
 145
 146 int kvm_init_vcpu(CPUState *env)
 147 {
 148     KVMState *s = kvm_state;
 149     long mmap_size;
 150     int ret;
 151
 152     dprintf("kvm_init_vcpu\n");
 153
 154     ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index);
 155     if (ret < 0) {
 156         dprintf("kvm_create_vcpu failed\n");
 157         goto err;
 158     }
 159
 160     env->kvm_fd = ret;
 161     env->kvm_state = s;
 162
 163     mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
 164     if (mmap_size < 0) {
 165         dprintf("KVM_GET_VCPU_MMAP_SIZE failed\n");
 166         goto err;
 167     }
 168
 169     env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
 170                         env->kvm_fd, 0);
 171     if (env->kvm_run == MAP_FAILED) {
 172         ret = -errno;
 173         dprintf("mmap'ing vcpu state failed\n");
 174         goto err;
 175     }
 176
 177     ret = kvm_arch_init_vcpu(env);
 178
 179 err:
 180     return ret;
 181 }
 182
 183 int kvm_sync_vcpus(void)
 184 {
 185     CPUState *env;
 186
 187     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 188         int ret;
 189
 190         ret = kvm_arch_put_registers(env);
 191         if (ret)
 192             return ret;
 193     }
 194
 195     return 0;
 196 }
 197
 198 /*
 199  * dirty pages logging control
 200  */
 201 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr,
 202                                       ram_addr_t size, int flags, int mask)
 203 {
 204     KVMState *s = kvm_state;
 205     KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
 206     int old_flags;
 207
 208     if (mem == NULL)  {
 209             fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
 210                     TARGET_FMT_plx "\n", __func__, phys_addr,
 211                     phys_addr + size - 1);
 212             return -EINVAL;
 213     }
 214
 215     old_flags = mem->flags;
 216
 217     flags = (mem->flags & ~mask) | flags;
 218     mem->flags = flags;
 219
 220     /* If nothing changed effectively, no need to issue ioctl */
 221     if (s->migration_log) {
 222         flags |= KVM_MEM_LOG_DIRTY_PAGES;
 223     }
 224     if (flags == old_flags) {
 225             return 0;
 226     }
 227
 228     return kvm_set_user_memory_region(s, mem);
 229 }
 230
 231 int kvm_log_start(target_phys_addr_t phys_addr, ram_addr_t size)
 232 {
 233         return kvm_dirty_pages_log_change(phys_addr, size,
 234                                           KVM_MEM_LOG_DIRTY_PAGES,
 235                                           KVM_MEM_LOG_DIRTY_PAGES);
 236 }
 237
 238 int kvm_log_stop(target_phys_addr_t phys_addr, ram_addr_t size)
 239 {
 240         return kvm_dirty_pages_log_change(phys_addr, size,
 241                                           0,
 242                                           KVM_MEM_LOG_DIRTY_PAGES);
 243 }
 244
 245 int kvm_set_migration_log(int enable)
 246 {
 247     KVMState *s = kvm_state;
 248     KVMSlot *mem;
 249     int i, err;
 250
 251     s->migration_log = enable;
 252
 253     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
 254         mem = &s->slots[i];
 255
 256         if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
 257             continue;
 258         }
 259         err = kvm_set_user_memory_region(s, mem);
 260         if (err) {
 261             return err;
 262         }
 263     }
 264     return 0;
 265 }
 266
 267 /**
 268  * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
 269  * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
 270  * This means all bits are set to dirty.
 271  *
 272  * @start_add: start of logged region.
 273  * @end_addr: end of logged region.
 274  */
 275 int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr,
 276                                    target_phys_addr_t end_addr)
 277 {
 278     KVMState *s = kvm_state;
 279     unsigned long size, allocated_size = 0;
 280     target_phys_addr_t phys_addr;
 281     ram_addr_t addr;
 282     KVMDirtyLog d;
 283     KVMSlot *mem;
 284     int ret = 0;
 285
 286     d.dirty_bitmap = NULL;
 287     while (start_addr < end_addr) {
 288         mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
 289         if (mem == NULL) {
 290             break;
 291         }
 292
 293         size = ((mem->memory_size >> TARGET_PAGE_BITS) + 7) / 8;
 294         if (!d.dirty_bitmap) {
 295             d.dirty_bitmap = qemu_malloc(size);
 296         } else if (size > allocated_size) {
 297             d.dirty_bitmap = qemu_realloc(d.dirty_bitmap, size);
 298         }
 299         allocated_size = size;
 300         memset(d.dirty_bitmap, 0, allocated_size);
 301
 302         d.slot = mem->slot;
 303
 304         if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
 305             dprintf("ioctl failed %d\n", errno);
 306             ret = -1;
 307             break;
 308         }
 309
 310         for (phys_addr = mem->start_addr, addr = mem->phys_offset;
 311              phys_addr < mem->start_addr + mem->memory_size;
 312              phys_addr += TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
 313             unsigned long *bitmap = (unsigned long *)d.dirty_bitmap;
 314             unsigned nr = (phys_addr - mem->start_addr) >> TARGET_PAGE_BITS;
 315             unsigned word = nr / (sizeof(*bitmap) * 8);
 316             unsigned bit = nr % (sizeof(*bitmap) * 8);
 317
 318             if ((bitmap[word] >> bit) & 1) {
 319                 cpu_physical_memory_set_dirty(addr);
 320             }
 321         }
 322         start_addr = phys_addr;
 323     }
 324     qemu_free(d.dirty_bitmap);
 325
 326     return ret;
 327 }
 328
 329 int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 330 {
 331     int ret = -ENOSYS;
 332 #ifdef KVM_CAP_COALESCED_MMIO
 333     KVMState *s = kvm_state;
 334
 335     if (s->coalesced_mmio) {
 336         struct kvm_coalesced_mmio_zone zone;
 337
 338         zone.addr = start;
 339         zone.size = size;
 340
 341         ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
 342     }
 343 #endif
 344
 345     return ret;
 346 }
 347
 348 int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 349 {
 350     int ret = -ENOSYS;
 351 #ifdef KVM_CAP_COALESCED_MMIO
 352     KVMState *s = kvm_state;
 353
 354     if (s->coalesced_mmio) {
 355         struct kvm_coalesced_mmio_zone zone;
 356
 357         zone.addr = start;
 358         zone.size = size;
 359
 360         ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
 361     }
 362 #endif
 363
 364     return ret;
 365 }
 366
 367 int kvm_check_extension(KVMState *s, unsigned int extension)
 368 {
 369     int ret;
 370
 371     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
 372     if (ret < 0) {
 373         ret = 0;
 374     }
 375
 376     return ret;
 377 }
 378
 379 int kvm_init(int smp_cpus)
 380 {
 381     KVMState *s;
 382     int ret;
 383     int i;
 384
 385     if (smp_cpus > 1) {
 386         fprintf(stderr, "No SMP KVM support, use '-smp 1'\n");
 387         return -EINVAL;
 388     }
 389
 390     s = qemu_mallocz(sizeof(KVMState));
 391
 392 #ifdef KVM_CAP_SET_GUEST_DEBUG
 393     TAILQ_INIT(&s->kvm_sw_breakpoints);
 394 #endif
 395     for (i = 0; i < ARRAY_SIZE(s->slots); i++)
 396         s->slots[i].slot = i;
 397
 398     s->vmfd = -1;
 399     s->fd = open("/dev/kvm", O_RDWR);
 400     if (s->fd == -1) {
 401         fprintf(stderr, "Could not access KVM kernel module: %m\n");
 402         ret = -errno;
 403         goto err;
 404     }
 405
 406     ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
 407     if (ret < KVM_API_VERSION) {
 408         if (ret > 0)
 409             ret = -EINVAL;
 410         fprintf(stderr, "kvm version too old\n");
 411         goto err;
 412     }
 413
 414     if (ret > KVM_API_VERSION) {
 415         ret = -EINVAL;
 416         fprintf(stderr, "kvm version not supported\n");
 417         goto err;
 418     }
 419
 420     s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
 421     if (s->vmfd < 0)
 422         goto err;
 423
 424     /* initially, KVM allocated its own memory and we had to jump through
 425      * hooks to make phys_ram_base point to this.  Modern versions of KVM
 426      * just use a user allocated buffer so we can use regular pages
 427      * unmodified.  Make sure we have a sufficiently modern version of KVM.
 428      */
 429     if (!kvm_check_extension(s, KVM_CAP_USER_MEMORY)) {
 430         ret = -EINVAL;
 431         fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n");
 432         goto err;
 433     }
 434
 435     /* There was a nasty bug in < kvm-80 that prevents memory slots from being
 436      * destroyed properly.  Since we rely on this capability, refuse to work
 437      * with any kernel without this capability. */
 438     if (!kvm_check_extension(s, KVM_CAP_DESTROY_MEMORY_REGION_WORKS)) {
 439         ret = -EINVAL;
 440
 441         fprintf(stderr,
 442                 "KVM kernel module broken (DESTROY_MEMORY_REGION)\n"
 443                 "Please upgrade to at least kvm-81.\n");
 444         goto err;
 445     }
 446
 447 #ifdef KVM_CAP_COALESCED_MMIO
 448     s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
 449 #else
 450     s->coalesced_mmio = 0;
 451 #endif
 452
 453     s->broken_set_mem_region = 1;
 454 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
 455     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
 456     if (ret > 0) {
 457         s->broken_set_mem_region = 0;
 458     }
 459 #endif
 460
 461     ret = kvm_arch_init(s, smp_cpus);
 462     if (ret < 0)
 463         goto err;
 464
 465     kvm_state = s;
 466
 467     return 0;
 468
 469 err:
 470     if (s) {
 471         if (s->vmfd != -1)
 472             close(s->vmfd);
 473         if (s->fd != -1)
 474             close(s->fd);
 475     }
 476     qemu_free(s);
 477
 478     return ret;
 479 }
 480
 481 static int kvm_handle_io(CPUState *env, uint16_t port, void *data,
 482                          int direction, int size, uint32_t count)
 483 {
 484     int i;
 485     uint8_t *ptr = data;
 486
 487     for (i = 0; i < count; i++) {
 488         if (direction == KVM_EXIT_IO_IN) {
 489             switch (size) {
 490             case 1:
 491                 stb_p(ptr, cpu_inb(env, port));
 492                 break;
 493             case 2:
 494                 stw_p(ptr, cpu_inw(env, port));
 495                 break;
 496             case 4:
 497                 stl_p(ptr, cpu_inl(env, port));
 498                 break;
 499             }
 500         } else {
 501             switch (size) {
 502             case 1:
 503                 cpu_outb(env, port, ldub_p(ptr));
 504                 break;
 505             case 2:
 506                 cpu_outw(env, port, lduw_p(ptr));
 507                 break;
 508             case 4:
 509                 cpu_outl(env, port, ldl_p(ptr));
 510                 break;
 511             }
 512         }
 513
 514         ptr += size;
 515     }
 516
 517     return 1;
 518 }
 519
 520 static void kvm_run_coalesced_mmio(CPUState *env, struct kvm_run *run)
 521 {
 522 #ifdef KVM_CAP_COALESCED_MMIO
 523     KVMState *s = kvm_state;
 524     if (s->coalesced_mmio) {
 525         struct kvm_coalesced_mmio_ring *ring;
 526
 527         ring = (void *)run + (s->coalesced_mmio * TARGET_PAGE_SIZE);
 528         while (ring->first != ring->last) {
 529             struct kvm_coalesced_mmio *ent;
 530
 531             ent = &ring->coalesced_mmio[ring->first];
 532
 533             cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
 534             /* FIXME smp_wmb() */
 535             ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
 536         }
 537     }
 538 #endif
 539 }
 540
 541 int kvm_cpu_exec(CPUState *env)
 542 {
 543     struct kvm_run *run = env->kvm_run;
 544     int ret;
 545
 546     dprintf("kvm_cpu_exec()\n");
 547
 548     do {
 549         kvm_arch_pre_run(env, run);
 550
 551         if (env->exit_request) {
 552             dprintf("interrupt exit requested\n");
 553             ret = 0;
 554             break;
 555         }
 556
 557         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
 558         kvm_arch_post_run(env, run);
 559
 560         if (ret == -EINTR || ret == -EAGAIN) {
 561             dprintf("io window exit\n");
 562             ret = 0;
 563             break;
 564         }
 565
 566         if (ret < 0) {
 567             dprintf("kvm run failed %s\n", strerror(-ret));
 568             abort();
 569         }
 570
 571         kvm_run_coalesced_mmio(env, run);
 572
 573         ret = 0; /* exit loop */
 574         switch (run->exit_reason) {
 575         case KVM_EXIT_IO:
 576             dprintf("handle_io\n");
 577             ret = kvm_handle_io(env, run->io.port,
 578                                 (uint8_t *)run + run->io.data_offset,
 579                                 run->io.direction,
 580                                 run->io.size,
 581                                 run->io.count);
 582             break;
 583         case KVM_EXIT_MMIO:
 584             dprintf("handle_mmio\n");
 585             cpu_physical_memory_rw(run->mmio.phys_addr,
 586                                    run->mmio.data,
 587                                    run->mmio.len,
 588                                    run->mmio.is_write);
 589             ret = 1;
 590             break;
 591         case KVM_EXIT_IRQ_WINDOW_OPEN:
 592             dprintf("irq_window_open\n");
 593             break;
 594         case KVM_EXIT_SHUTDOWN:
 595             dprintf("shutdown\n");
 596             qemu_system_reset_request();
 597             ret = 1;
 598             break;
 599         case KVM_EXIT_UNKNOWN:
 600             dprintf("kvm_exit_unknown\n");
 601             break;
 602         case KVM_EXIT_FAIL_ENTRY:
 603             dprintf("kvm_exit_fail_entry\n");
 604             break;
 605         case KVM_EXIT_EXCEPTION:
 606             dprintf("kvm_exit_exception\n");
 607             break;
 608         case KVM_EXIT_DEBUG:
 609             dprintf("kvm_exit_debug\n");
 610 #ifdef KVM_CAP_SET_GUEST_DEBUG
 611             if (kvm_arch_debug(&run->debug.arch)) {
 612                 gdb_set_stop_cpu(env);
 613                 vm_stop(EXCP_DEBUG);
 614                 env->exception_index = EXCP_DEBUG;
 615                 return 0;
 616             }
 617             /* re-enter, this exception was guest-internal */
 618             ret = 1;
 619 #endif /* KVM_CAP_SET_GUEST_DEBUG */
 620             break;
 621         default:
 622             dprintf("kvm_arch_handle_exit\n");
 623             ret = kvm_arch_handle_exit(env, run);
 624             break;
 625         }
 626     } while (ret > 0);
 627
 628     if (env->exit_request) {
 629         env->exit_request = 0;
 630         env->exception_index = EXCP_INTERRUPT;
 631     }
 632
 633     return ret;
 634 }
 635
 636 void kvm_set_phys_mem(target_phys_addr_t start_addr,
 637                       ram_addr_t size,
 638                       ram_addr_t phys_offset)
 639 {
 640     KVMState *s = kvm_state;
 641     ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK;
 642     KVMSlot *mem, old;
 643     int err;
 644
 645     if (start_addr & ~TARGET_PAGE_MASK) {
 646         if (flags >= IO_MEM_UNASSIGNED) {
 647             if (!kvm_lookup_overlapping_slot(s, start_addr,
 648                                              start_addr + size)) {
 649                 return;
 650             }
 651             fprintf(stderr, "Unaligned split of a KVM memory slot\n");
 652         } else {
 653             fprintf(stderr, "Only page-aligned memory slots supported\n");
 654         }
 655         abort();
 656     }
 657
 658     /* KVM does not support read-only slots */
 659     phys_offset &= ~IO_MEM_ROM;
 660
 661     while (1) {
 662         mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
 663         if (!mem) {
 664             break;
 665         }
 666
 667         if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr &&
 668             (start_addr + size <= mem->start_addr + mem->memory_size) &&
 669             (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) {
 670             /* The new slot fits into the existing one and comes with
 671              * identical parameters - nothing to be done. */
 672             return;
 673         }
 674
 675         old = *mem;
 676
 677         /* unregister the overlapping slot */
 678         mem->memory_size = 0;
 679         err = kvm_set_user_memory_region(s, mem);
 680         if (err) {
 681             fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
 682                     __func__, strerror(-err));
 683             abort();
 684         }
 685
 686         /* Workaround for older KVM versions: we can't join slots, even not by
 687          * unregistering the previous ones and then registering the larger
 688          * slot. We have to maintain the existing fragmentation. Sigh.
 689          *
 690          * This workaround assumes that the new slot starts at the same
 691          * address as the first existing one. If not or if some overlapping
 692          * slot comes around later, we will fail (not seen in practice so far)
 693          * - and actually require a recent KVM version. */
 694         if (s->broken_set_mem_region &&
 695             old.start_addr == start_addr && old.memory_size < size &&
 696             flags < IO_MEM_UNASSIGNED) {
 697             mem = kvm_alloc_slot(s);
 698             mem->memory_size = old.memory_size;
 699             mem->start_addr = old.start_addr;
 700             mem->phys_offset = old.phys_offset;
 701             mem->flags = 0;
 702
 703             err = kvm_set_user_memory_region(s, mem);
 704             if (err) {
 705                 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
 706                         strerror(-err));
 707                 abort();
 708             }
 709
 710             start_addr += old.memory_size;
 711             phys_offset += old.memory_size;
 712             size -= old.memory_size;
 713             continue;
 714         }
 715
 716         /* register prefix slot */
 717         if (old.start_addr < start_addr) {
 718             mem = kvm_alloc_slot(s);
 719             mem->memory_size = start_addr - old.start_addr;
 720             mem->start_addr = old.start_addr;
 721             mem->phys_offset = old.phys_offset;
 722             mem->flags = 0;
 723
 724             err = kvm_set_user_memory_region(s, mem);
 725             if (err) {
 726                 fprintf(stderr, "%s: error registering prefix slot: %s\n",
 727                         __func__, strerror(-err));
 728                 abort();
 729             }
 730         }
 731
 732         /* register suffix slot */
 733         if (old.start_addr + old.memory_size > start_addr + size) {
 734             ram_addr_t size_delta;
 735
 736             mem = kvm_alloc_slot(s);
 737             mem->start_addr = start_addr + size;
 738             size_delta = mem->start_addr - old.start_addr;
 739             mem->memory_size = old.memory_size - size_delta;
 740             mem->phys_offset = old.phys_offset + size_delta;
 741             mem->flags = 0;
 742
 743             err = kvm_set_user_memory_region(s, mem);
 744             if (err) {
 745                 fprintf(stderr, "%s: error registering suffix slot: %s\n",
 746                         __func__, strerror(-err));
 747                 abort();
 748             }
 749         }
 750     }
 751
 752     /* in case the KVM bug workaround already "consumed" the new slot */
 753     if (!size)
 754         return;
 755
 756     /* KVM does not need to know about this memory */
 757     if (flags >= IO_MEM_UNASSIGNED)
 758         return;
 759
 760     mem = kvm_alloc_slot(s);
 761     mem->memory_size = size;
 762     mem->start_addr = start_addr;
 763     mem->phys_offset = phys_offset;
 764     mem->flags = 0;
 765
 766     err = kvm_set_user_memory_region(s, mem);
 767     if (err) {
 768         fprintf(stderr, "%s: error registering slot: %s\n", __func__,
 769                 strerror(-err));
 770         abort();
 771     }
 772 }
 773
 774 int kvm_ioctl(KVMState *s, int type, ...)
 775 {
 776     int ret;
 777     void *arg;
 778     va_list ap;
 779
 780     va_start(ap, type);
 781     arg = va_arg(ap, void *);
 782     va_end(ap);
 783
 784     ret = ioctl(s->fd, type, arg);
 785     if (ret == -1)
 786         ret = -errno;
 787
 788     return ret;
 789 }
 790
 791 int kvm_vm_ioctl(KVMState *s, int type, ...)
 792 {
 793     int ret;
 794     void *arg;
 795     va_list ap;
 796
 797     va_start(ap, type);
 798     arg = va_arg(ap, void *);
 799     va_end(ap);
 800
 801     ret = ioctl(s->vmfd, type, arg);
 802     if (ret == -1)
 803         ret = -errno;
 804
 805     return ret;
 806 }
 807
 808 int kvm_vcpu_ioctl(CPUState *env, int type, ...)
 809 {
 810     int ret;
 811     void *arg;
 812     va_list ap;
 813
 814     va_start(ap, type);
 815     arg = va_arg(ap, void *);
 816     va_end(ap);
 817
 818     ret = ioctl(env->kvm_fd, type, arg);
 819     if (ret == -1)
 820         ret = -errno;
 821
 822     return ret;
 823 }
 824
 825 int kvm_has_sync_mmu(void)
 826 {
 827 #ifdef KVM_CAP_SYNC_MMU
 828     KVMState *s = kvm_state;
 829
 830     return kvm_check_extension(s, KVM_CAP_SYNC_MMU);
 831 #else
 832     return 0;
 833 #endif
 834 }
 835
 836 void kvm_setup_guest_memory(void *start, size_t size)
 837 {
 838     if (!kvm_has_sync_mmu()) {
 839 #ifdef MADV_DONTFORK
 840         int ret = madvise(start, size, MADV_DONTFORK);
 841
 842         if (ret) {
 843             perror("madvice");
 844             exit(1);
 845         }
 846 #else
 847         fprintf(stderr,
 848                 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
 849         exit(1);
 850 #endif
 851     }
 852 }
 853
 854 #ifdef KVM_CAP_SET_GUEST_DEBUG
 855 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
 856                                                  target_ulong pc)
 857 {
 858     struct kvm_sw_breakpoint *bp;
 859
 860     TAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) {
 861         if (bp->pc == pc)
 862             return bp;
 863     }
 864     return NULL;
 865 }
 866
 867 int kvm_sw_breakpoints_active(CPUState *env)
 868 {
 869     return !TAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints);
 870 }
 871
 872 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 873 {
 874     struct kvm_guest_debug dbg;
 875
 876     dbg.control = 0;
 877     if (env->singlestep_enabled)
 878         dbg.control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
 879
 880     kvm_arch_update_guest_debug(env, &dbg);
 881     dbg.control |= reinject_trap;
 882
 883     return kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg);
 884 }
 885
 886 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 887                           target_ulong len, int type)
 888 {
 889     struct kvm_sw_breakpoint *bp;
 890     CPUState *env;
 891     int err;
 892
 893     if (type == GDB_BREAKPOINT_SW) {
 894         bp = kvm_find_sw_breakpoint(current_env, addr);
 895         if (bp) {
 896             bp->use_count++;
 897             return 0;
 898         }
 899
 900         bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint));
 901         if (!bp)
 902             return -ENOMEM;
 903
 904         bp->pc = addr;
 905         bp->use_count = 1;
 906         err = kvm_arch_insert_sw_breakpoint(current_env, bp);
 907         if (err) {
 908             free(bp);
 909             return err;
 910         }
 911
 912         TAILQ_INSERT_HEAD(&current_env->kvm_state->kvm_sw_breakpoints,
 913                           bp, entry);
 914     } else {
 915         err = kvm_arch_insert_hw_breakpoint(addr, len, type);
 916         if (err)
 917             return err;
 918     }
 919
 920     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 921         err = kvm_update_guest_debug(env, 0);
 922         if (err)
 923             return err;
 924     }
 925     return 0;
 926 }
 927
 928 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
 929                           target_ulong len, int type)
 930 {
 931     struct kvm_sw_breakpoint *bp;
 932     CPUState *env;
 933     int err;
 934
 935     if (type == GDB_BREAKPOINT_SW) {
 936         bp = kvm_find_sw_breakpoint(current_env, addr);
 937         if (!bp)
 938             return -ENOENT;
 939
 940         if (bp->use_count > 1) {
 941             bp->use_count--;
 942             return 0;
 943         }
 944
 945         err = kvm_arch_remove_sw_breakpoint(current_env, bp);
 946         if (err)
 947             return err;
 948
 949         TAILQ_REMOVE(&current_env->kvm_state->kvm_sw_breakpoints, bp, entry);
 950         qemu_free(bp);
 951     } else {
 952         err = kvm_arch_remove_hw_breakpoint(addr, len, type);
 953         if (err)
 954             return err;
 955     }
 956
 957     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 958         err = kvm_update_guest_debug(env, 0);
 959         if (err)
 960             return err;
 961     }
 962     return 0;
 963 }
 964
 965 void kvm_remove_all_breakpoints(CPUState *current_env)
 966 {
 967     struct kvm_sw_breakpoint *bp, *next;
 968     KVMState *s = current_env->kvm_state;
 969     CPUState *env;
 970
 971     TAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
 972         if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) {
 973             /* Try harder to find a CPU that currently sees the breakpoint. */
 974             for (env = first_cpu; env != NULL; env = env->next_cpu) {
 975                 if (kvm_arch_remove_sw_breakpoint(env, bp) == 0)
 976                     break;
 977             }
 978         }
 979     }
 980     kvm_arch_remove_all_hw_breakpoints();
 981
 982     for (env = first_cpu; env != NULL; env = env->next_cpu)
 983         kvm_update_guest_debug(env, 0);
 984 }
 985
 986 #else /* !KVM_CAP_SET_GUEST_DEBUG */
 987
 988 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 989 {
 990     return -EINVAL;
 991 }
 992
 993 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 994                           target_ulong len, int type)
 995 {
 996     return -EINVAL;
 997 }
 998
 999 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
1000                           target_ulong len, int type)
1001 {
1002     return -EINVAL;
1003 }
1004
1005 void kvm_remove_all_breakpoints(CPUState *current_env)
1006 {
1007 }
1008 #endif /* !KVM_CAP_SET_GUEST_DEBUG */