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 "hw/hw.h"
  26 #include "gdbstub.h"
  27 #include "kvm.h"
  28
  29 #ifdef KVM_UPSTREAM
  30 /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */
  31 #define PAGE_SIZE TARGET_PAGE_SIZE
  32
  33 //#define DEBUG_KVM
  34
  35 #ifdef DEBUG_KVM
  36 #define dprintf(fmt, ...) \
  37     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  38 #else
  39 #define dprintf(fmt, ...) \
  40     do { } while (0)
  41 #endif
  42
  43 typedef struct KVMSlot
  44 {
  45     target_phys_addr_t start_addr;
  46     ram_addr_t memory_size;
  47     ram_addr_t phys_offset;
  48     int slot;
  49     int flags;
  50 } KVMSlot;
  51
  52 typedef struct kvm_dirty_log KVMDirtyLog;
  53
  54 int kvm_allowed = 0;
  55
  56 struct KVMState
  57 {
  58     KVMSlot slots[32];
  59     int fd;
  60     int vmfd;
  61     int coalesced_mmio;
  62     int broken_set_mem_region;
  63     int migration_log;
  64 #ifdef KVM_CAP_SET_GUEST_DEBUG
  65     struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
  66 #endif
  67 };
  68
  69 static KVMState *kvm_state;
  70
  71 static KVMSlot *kvm_alloc_slot(KVMState *s)
  72 {
  73     int i;
  74
  75     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
  76         /* KVM private memory slots */
  77         if (i >= 8 && i < 12)
  78             continue;
  79         if (s->slots[i].memory_size == 0)
  80             return &s->slots[i];
  81     }
  82
  83     fprintf(stderr, "%s: no free slot available\n", __func__);
  84     abort();
  85 }
  86
  87 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
  88                                          target_phys_addr_t start_addr,
  89                                          target_phys_addr_t end_addr)
  90 {
  91     int i;
  92
  93     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
  94         KVMSlot *mem = &s->slots[i];
  95
  96         if (start_addr == mem->start_addr &&
  97             end_addr == mem->start_addr + mem->memory_size) {
  98             return mem;
  99         }
 100     }
 101
 102     return NULL;
 103 }
 104
 105 /*
 106  * Find overlapping slot with lowest start address
 107  */
 108 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
 109                                             target_phys_addr_t start_addr,
 110                                             target_phys_addr_t end_addr)
 111 {
 112     KVMSlot *found = NULL;
 113     int i;
 114
 115     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
 116         KVMSlot *mem = &s->slots[i];
 117
 118         if (mem->memory_size == 0 ||
 119             (found && found->start_addr < mem->start_addr)) {
 120             continue;
 121         }
 122
 123         if (end_addr > mem->start_addr &&
 124             start_addr < mem->start_addr + mem->memory_size) {
 125             found = mem;
 126         }
 127     }
 128
 129     return found;
 130 }
 131
 132 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
 133 {
 134     struct kvm_userspace_memory_region mem;
 135
 136     mem.slot = slot->slot;
 137     mem.guest_phys_addr = slot->start_addr;
 138     mem.memory_size = slot->memory_size;
 139     mem.userspace_addr = (unsigned long)qemu_get_ram_ptr(slot->phys_offset);
 140     mem.flags = slot->flags;
 141     if (s->migration_log) {
 142         mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
 143     }
 144     return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
 145 }
 146
 147 static void kvm_reset_vcpu(void *opaque)
 148 {
 149     CPUState *env = opaque;
 150
 151     if (kvm_arch_put_registers(env)) {
 152         fprintf(stderr, "Fatal: kvm vcpu reset failed\n");
 153         abort();
 154     }
 155 }
 156
 157 int kvm_init_vcpu(CPUState *env)
 158 {
 159     KVMState *s = kvm_state;
 160     long mmap_size;
 161     int ret;
 162
 163     dprintf("kvm_init_vcpu\n");
 164
 165     ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index);
 166     if (ret < 0) {
 167         dprintf("kvm_create_vcpu failed\n");
 168         goto err;
 169     }
 170
 171     env->kvm_fd = ret;
 172     env->kvm_state = s;
 173
 174     mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
 175     if (mmap_size < 0) {
 176         dprintf("KVM_GET_VCPU_MMAP_SIZE failed\n");
 177         goto err;
 178     }
 179
 180     env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
 181                         env->kvm_fd, 0);
 182     if (env->kvm_run == MAP_FAILED) {
 183         ret = -errno;
 184         dprintf("mmap'ing vcpu state failed\n");
 185         goto err;
 186     }
 187
 188     ret = kvm_arch_init_vcpu(env);
 189     if (ret == 0) {
 190         qemu_register_reset(kvm_reset_vcpu, env);
 191         ret = kvm_arch_put_registers(env);
 192     }
 193 err:
 194     return ret;
 195 }
 196
 197 int kvm_put_mp_state(CPUState *env)
 198 {
 199     struct kvm_mp_state mp_state = { .mp_state = env->mp_state };
 200
 201     return kvm_vcpu_ioctl(env, KVM_SET_MP_STATE, &mp_state);
 202 }
 203
 204 int kvm_get_mp_state(CPUState *env)
 205 {
 206     struct kvm_mp_state mp_state;
 207     int ret;
 208
 209     ret = kvm_vcpu_ioctl(env, KVM_GET_MP_STATE, &mp_state);
 210     if (ret < 0) {
 211         return ret;
 212     }
 213     env->mp_state = mp_state.mp_state;
 214     return 0;
 215 }
 216
 217 /*
 218  * dirty pages logging control
 219  */
 220 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr,
 221                                       ram_addr_t size, int flags, int mask)
 222 {
 223     KVMState *s = kvm_state;
 224     KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
 225     int old_flags;
 226
 227     if (mem == NULL)  {
 228             fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
 229                     TARGET_FMT_plx "\n", __func__, phys_addr,
 230                     phys_addr + size - 1);
 231             return -EINVAL;
 232     }
 233
 234     old_flags = mem->flags;
 235
 236     flags = (mem->flags & ~mask) | flags;
 237     mem->flags = flags;
 238
 239     /* If nothing changed effectively, no need to issue ioctl */
 240     if (s->migration_log) {
 241         flags |= KVM_MEM_LOG_DIRTY_PAGES;
 242     }
 243     if (flags == old_flags) {
 244             return 0;
 245     }
 246
 247     return kvm_set_user_memory_region(s, mem);
 248 }
 249
 250 int kvm_log_start(target_phys_addr_t phys_addr, ram_addr_t size)
 251 {
 252         return kvm_dirty_pages_log_change(phys_addr, size,
 253                                           KVM_MEM_LOG_DIRTY_PAGES,
 254                                           KVM_MEM_LOG_DIRTY_PAGES);
 255 }
 256
 257 int kvm_log_stop(target_phys_addr_t phys_addr, ram_addr_t size)
 258 {
 259         return kvm_dirty_pages_log_change(phys_addr, size,
 260                                           0,
 261                                           KVM_MEM_LOG_DIRTY_PAGES);
 262 }
 263
 264 int kvm_set_migration_log(int enable)
 265 {
 266     KVMState *s = kvm_state;
 267     KVMSlot *mem;
 268     int i, err;
 269
 270     s->migration_log = enable;
 271
 272     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
 273         mem = &s->slots[i];
 274
 275         if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
 276             continue;
 277         }
 278         err = kvm_set_user_memory_region(s, mem);
 279         if (err) {
 280             return err;
 281         }
 282     }
 283     return 0;
 284 }
 285
 286 /**
 287  * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
 288  * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
 289  * This means all bits are set to dirty.
 290  *
 291  * @start_add: start of logged region.
 292  * @end_addr: end of logged region.
 293  */
 294 int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr,
 295                                    target_phys_addr_t end_addr)
 296 {
 297     KVMState *s = kvm_state;
 298     unsigned long size, allocated_size = 0;
 299     target_phys_addr_t phys_addr;
 300     ram_addr_t addr;
 301     KVMDirtyLog d;
 302     KVMSlot *mem;
 303     int ret = 0;
 304
 305     d.dirty_bitmap = NULL;
 306     while (start_addr < end_addr) {
 307         mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
 308         if (mem == NULL) {
 309             break;
 310         }
 311
 312         size = ((mem->memory_size >> TARGET_PAGE_BITS) + 7) / 8;
 313         if (!d.dirty_bitmap) {
 314             d.dirty_bitmap = qemu_malloc(size);
 315         } else if (size > allocated_size) {
 316             d.dirty_bitmap = qemu_realloc(d.dirty_bitmap, size);
 317         }
 318         allocated_size = size;
 319         memset(d.dirty_bitmap, 0, allocated_size);
 320
 321         d.slot = mem->slot;
 322
 323         if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
 324             dprintf("ioctl failed %d\n", errno);
 325             ret = -1;
 326             break;
 327         }
 328
 329         for (phys_addr = mem->start_addr, addr = mem->phys_offset;
 330              phys_addr < mem->start_addr + mem->memory_size;
 331              phys_addr += TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
 332             unsigned long *bitmap = (unsigned long *)d.dirty_bitmap;
 333             unsigned nr = (phys_addr - mem->start_addr) >> TARGET_PAGE_BITS;
 334             unsigned word = nr / (sizeof(*bitmap) * 8);
 335             unsigned bit = nr % (sizeof(*bitmap) * 8);
 336
 337             if ((bitmap[word] >> bit) & 1) {
 338                 cpu_physical_memory_set_dirty(addr);
 339             }
 340         }
 341         start_addr = phys_addr;
 342     }
 343     qemu_free(d.dirty_bitmap);
 344
 345     return ret;
 346 }
 347
 348 int kvm_coalesce_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_REGISTER_COALESCED_MMIO, &zone);
 361     }
 362 #endif
 363
 364     return ret;
 365 }
 366
 367 int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 368 {
 369     int ret = -ENOSYS;
 370 #ifdef KVM_CAP_COALESCED_MMIO
 371     KVMState *s = kvm_state;
 372
 373     if (s->coalesced_mmio) {
 374         struct kvm_coalesced_mmio_zone zone;
 375
 376         zone.addr = start;
 377         zone.size = size;
 378
 379         ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
 380     }
 381 #endif
 382
 383     return ret;
 384 }
 385
 386 int kvm_check_extension(KVMState *s, unsigned int extension)
 387 {
 388     int ret;
 389
 390     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
 391     if (ret < 0) {
 392         ret = 0;
 393     }
 394
 395     return ret;
 396 }
 397
 398 int kvm_init(int smp_cpus)
 399 {
 400     static const char upgrade_note[] =
 401         "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
 402         "(see http://sourceforge.net/projects/kvm).\n";
 403     KVMState *s;
 404     int ret;
 405     int i;
 406
 407     if (smp_cpus > 1) {
 408         fprintf(stderr, "No SMP KVM support, use '-smp 1'\n");
 409         return -EINVAL;
 410     }
 411
 412     s = qemu_mallocz(sizeof(KVMState));
 413
 414 #ifdef KVM_CAP_SET_GUEST_DEBUG
 415     TAILQ_INIT(&s->kvm_sw_breakpoints);
 416 #endif
 417     for (i = 0; i < ARRAY_SIZE(s->slots); i++)
 418         s->slots[i].slot = i;
 419
 420     s->vmfd = -1;
 421     s->fd = open("/dev/kvm", O_RDWR);
 422     if (s->fd == -1) {
 423         fprintf(stderr, "Could not access KVM kernel module: %m\n");
 424         ret = -errno;
 425         goto err;
 426     }
 427
 428     ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
 429     if (ret < KVM_API_VERSION) {
 430         if (ret > 0)
 431             ret = -EINVAL;
 432         fprintf(stderr, "kvm version too old\n");
 433         goto err;
 434     }
 435
 436     if (ret > KVM_API_VERSION) {
 437         ret = -EINVAL;
 438         fprintf(stderr, "kvm version not supported\n");
 439         goto err;
 440     }
 441
 442     s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
 443     if (s->vmfd < 0)
 444         goto err;
 445
 446     /* initially, KVM allocated its own memory and we had to jump through
 447      * hooks to make phys_ram_base point to this.  Modern versions of KVM
 448      * just use a user allocated buffer so we can use regular pages
 449      * unmodified.  Make sure we have a sufficiently modern version of KVM.
 450      */
 451     if (!kvm_check_extension(s, KVM_CAP_USER_MEMORY)) {
 452         ret = -EINVAL;
 453         fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
 454                 upgrade_note);
 455         goto err;
 456     }
 457
 458     /* There was a nasty bug in < kvm-80 that prevents memory slots from being
 459      * destroyed properly.  Since we rely on this capability, refuse to work
 460      * with any kernel without this capability. */
 461     if (!kvm_check_extension(s, KVM_CAP_DESTROY_MEMORY_REGION_WORKS)) {
 462         ret = -EINVAL;
 463
 464         fprintf(stderr,
 465                 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
 466                 upgrade_note);
 467         goto err;
 468     }
 469
 470 #ifdef KVM_CAP_COALESCED_MMIO
 471     s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
 472 #else
 473     s->coalesced_mmio = 0;
 474 #endif
 475
 476     s->broken_set_mem_region = 1;
 477 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
 478     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
 479     if (ret > 0) {
 480         s->broken_set_mem_region = 0;
 481     }
 482 #endif
 483
 484     ret = kvm_arch_init(s, smp_cpus);
 485     if (ret < 0)
 486         goto err;
 487
 488     kvm_state = s;
 489
 490     return 0;
 491
 492 err:
 493     if (s) {
 494         if (s->vmfd != -1)
 495             close(s->vmfd);
 496         if (s->fd != -1)
 497             close(s->fd);
 498     }
 499     qemu_free(s);
 500
 501     return ret;
 502 }
 503
 504 static int kvm_handle_io(CPUState *env, uint16_t port, void *data,
 505                          int direction, int size, uint32_t count)
 506 {
 507     int i;
 508     uint8_t *ptr = data;
 509
 510     for (i = 0; i < count; i++) {
 511         if (direction == KVM_EXIT_IO_IN) {
 512             switch (size) {
 513             case 1:
 514                 stb_p(ptr, cpu_inb(env, port));
 515                 break;
 516             case 2:
 517                 stw_p(ptr, cpu_inw(env, port));
 518                 break;
 519             case 4:
 520                 stl_p(ptr, cpu_inl(env, port));
 521                 break;
 522             }
 523         } else {
 524             switch (size) {
 525             case 1:
 526                 cpu_outb(env, port, ldub_p(ptr));
 527                 break;
 528             case 2:
 529                 cpu_outw(env, port, lduw_p(ptr));
 530                 break;
 531             case 4:
 532                 cpu_outl(env, port, ldl_p(ptr));
 533                 break;
 534             }
 535         }
 536
 537         ptr += size;
 538     }
 539
 540     return 1;
 541 }
 542
 543 static void kvm_run_coalesced_mmio(CPUState *env, struct kvm_run *run)
 544 {
 545 #ifdef KVM_CAP_COALESCED_MMIO
 546     KVMState *s = kvm_state;
 547     if (s->coalesced_mmio) {
 548         struct kvm_coalesced_mmio_ring *ring;
 549
 550         ring = (void *)run + (s->coalesced_mmio * TARGET_PAGE_SIZE);
 551         while (ring->first != ring->last) {
 552             struct kvm_coalesced_mmio *ent;
 553
 554             ent = &ring->coalesced_mmio[ring->first];
 555
 556             cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
 557             /* FIXME smp_wmb() */
 558             ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
 559         }
 560     }
 561 #endif
 562 }
 563
 564 int kvm_cpu_exec(CPUState *env)
 565 {
 566     struct kvm_run *run = env->kvm_run;
 567     int ret;
 568
 569     dprintf("kvm_cpu_exec()\n");
 570
 571     do {
 572         if (env->exit_request) {
 573             dprintf("interrupt exit requested\n");
 574             ret = 0;
 575             break;
 576         }
 577
 578         kvm_arch_pre_run(env, run);
 579         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
 580         kvm_arch_post_run(env, run);
 581
 582         if (ret == -EINTR || ret == -EAGAIN) {
 583             dprintf("io window exit\n");
 584             ret = 0;
 585             break;
 586         }
 587
 588         if (ret < 0) {
 589             dprintf("kvm run failed %s\n", strerror(-ret));
 590             abort();
 591         }
 592
 593         kvm_run_coalesced_mmio(env, run);
 594
 595         ret = 0; /* exit loop */
 596         switch (run->exit_reason) {
 597         case KVM_EXIT_IO:
 598             dprintf("handle_io\n");
 599             ret = kvm_handle_io(env, run->io.port,
 600                                 (uint8_t *)run + run->io.data_offset,
 601                                 run->io.direction,
 602                                 run->io.size,
 603                                 run->io.count);
 604             break;
 605         case KVM_EXIT_MMIO:
 606             dprintf("handle_mmio\n");
 607             cpu_physical_memory_rw(run->mmio.phys_addr,
 608                                    run->mmio.data,
 609                                    run->mmio.len,
 610                                    run->mmio.is_write);
 611             ret = 1;
 612             break;
 613         case KVM_EXIT_IRQ_WINDOW_OPEN:
 614             dprintf("irq_window_open\n");
 615             break;
 616         case KVM_EXIT_SHUTDOWN:
 617             dprintf("shutdown\n");
 618             qemu_system_reset_request();
 619             ret = 1;
 620             break;
 621         case KVM_EXIT_UNKNOWN:
 622             dprintf("kvm_exit_unknown\n");
 623             break;
 624         case KVM_EXIT_FAIL_ENTRY:
 625             dprintf("kvm_exit_fail_entry\n");
 626             break;
 627         case KVM_EXIT_EXCEPTION:
 628             dprintf("kvm_exit_exception\n");
 629             break;
 630         case KVM_EXIT_DEBUG:
 631             dprintf("kvm_exit_debug\n");
 632 #ifdef KVM_CAP_SET_GUEST_DEBUG
 633             if (kvm_arch_debug(&run->debug.arch)) {
 634                 gdb_set_stop_cpu(env);
 635                 vm_stop(EXCP_DEBUG);
 636                 env->exception_index = EXCP_DEBUG;
 637                 return 0;
 638             }
 639             /* re-enter, this exception was guest-internal */
 640             ret = 1;
 641 #endif /* KVM_CAP_SET_GUEST_DEBUG */
 642             break;
 643         default:
 644             dprintf("kvm_arch_handle_exit\n");
 645             ret = kvm_arch_handle_exit(env, run);
 646             break;
 647         }
 648     } while (ret > 0);
 649
 650     if (env->exit_request) {
 651         env->exit_request = 0;
 652         env->exception_index = EXCP_INTERRUPT;
 653     }
 654
 655     return ret;
 656 }
 657
 658 void kvm_set_phys_mem(target_phys_addr_t start_addr,
 659                       ram_addr_t size,
 660                       ram_addr_t phys_offset)
 661 {
 662     KVMState *s = kvm_state;
 663     ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK;
 664     KVMSlot *mem, old;
 665     int err;
 666
 667     if (start_addr & ~TARGET_PAGE_MASK) {
 668         if (flags >= IO_MEM_UNASSIGNED) {
 669             if (!kvm_lookup_overlapping_slot(s, start_addr,
 670                                              start_addr + size)) {
 671                 return;
 672             }
 673             fprintf(stderr, "Unaligned split of a KVM memory slot\n");
 674         } else {
 675             fprintf(stderr, "Only page-aligned memory slots supported\n");
 676         }
 677         abort();
 678     }
 679
 680     /* KVM does not support read-only slots */
 681     phys_offset &= ~IO_MEM_ROM;
 682
 683     while (1) {
 684         mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
 685         if (!mem) {
 686             break;
 687         }
 688
 689         if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr &&
 690             (start_addr + size <= mem->start_addr + mem->memory_size) &&
 691             (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) {
 692             /* The new slot fits into the existing one and comes with
 693              * identical parameters - nothing to be done. */
 694             return;
 695         }
 696
 697         old = *mem;
 698
 699         /* unregister the overlapping slot */
 700         mem->memory_size = 0;
 701         err = kvm_set_user_memory_region(s, mem);
 702         if (err) {
 703             fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
 704                     __func__, strerror(-err));
 705             abort();
 706         }
 707
 708         /* Workaround for older KVM versions: we can't join slots, even not by
 709          * unregistering the previous ones and then registering the larger
 710          * slot. We have to maintain the existing fragmentation. Sigh.
 711          *
 712          * This workaround assumes that the new slot starts at the same
 713          * address as the first existing one. If not or if some overlapping
 714          * slot comes around later, we will fail (not seen in practice so far)
 715          * - and actually require a recent KVM version. */
 716         if (s->broken_set_mem_region &&
 717             old.start_addr == start_addr && old.memory_size < size &&
 718             flags < IO_MEM_UNASSIGNED) {
 719             mem = kvm_alloc_slot(s);
 720             mem->memory_size = old.memory_size;
 721             mem->start_addr = old.start_addr;
 722             mem->phys_offset = old.phys_offset;
 723             mem->flags = 0;
 724
 725             err = kvm_set_user_memory_region(s, mem);
 726             if (err) {
 727                 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
 728                         strerror(-err));
 729                 abort();
 730             }
 731
 732             start_addr += old.memory_size;
 733             phys_offset += old.memory_size;
 734             size -= old.memory_size;
 735             continue;
 736         }
 737
 738         /* register prefix slot */
 739         if (old.start_addr < start_addr) {
 740             mem = kvm_alloc_slot(s);
 741             mem->memory_size = start_addr - old.start_addr;
 742             mem->start_addr = old.start_addr;
 743             mem->phys_offset = old.phys_offset;
 744             mem->flags = 0;
 745
 746             err = kvm_set_user_memory_region(s, mem);
 747             if (err) {
 748                 fprintf(stderr, "%s: error registering prefix slot: %s\n",
 749                         __func__, strerror(-err));
 750                 abort();
 751             }
 752         }
 753
 754         /* register suffix slot */
 755         if (old.start_addr + old.memory_size > start_addr + size) {
 756             ram_addr_t size_delta;
 757
 758             mem = kvm_alloc_slot(s);
 759             mem->start_addr = start_addr + size;
 760             size_delta = mem->start_addr - old.start_addr;
 761             mem->memory_size = old.memory_size - size_delta;
 762             mem->phys_offset = old.phys_offset + size_delta;
 763             mem->flags = 0;
 764
 765             err = kvm_set_user_memory_region(s, mem);
 766             if (err) {
 767                 fprintf(stderr, "%s: error registering suffix slot: %s\n",
 768                         __func__, strerror(-err));
 769                 abort();
 770             }
 771         }
 772     }
 773
 774     /* in case the KVM bug workaround already "consumed" the new slot */
 775     if (!size)
 776         return;
 777
 778     /* KVM does not need to know about this memory */
 779     if (flags >= IO_MEM_UNASSIGNED)
 780         return;
 781
 782     mem = kvm_alloc_slot(s);
 783     mem->memory_size = size;
 784     mem->start_addr = start_addr;
 785     mem->phys_offset = phys_offset;
 786     mem->flags = 0;
 787
 788     err = kvm_set_user_memory_region(s, mem);
 789     if (err) {
 790         fprintf(stderr, "%s: error registering slot: %s\n", __func__,
 791                 strerror(-err));
 792         abort();
 793     }
 794 }
 795
 796 int kvm_ioctl(KVMState *s, int type, ...)
 797 {
 798     int ret;
 799     void *arg;
 800     va_list ap;
 801
 802     va_start(ap, type);
 803     arg = va_arg(ap, void *);
 804     va_end(ap);
 805
 806     ret = ioctl(s->fd, type, arg);
 807     if (ret == -1)
 808         ret = -errno;
 809
 810     return ret;
 811 }
 812
 813 int kvm_vm_ioctl(KVMState *s, int type, ...)
 814 {
 815     int ret;
 816     void *arg;
 817     va_list ap;
 818
 819     va_start(ap, type);
 820     arg = va_arg(ap, void *);
 821     va_end(ap);
 822
 823     ret = ioctl(s->vmfd, type, arg);
 824     if (ret == -1)
 825         ret = -errno;
 826
 827     return ret;
 828 }
 829
 830 int kvm_vcpu_ioctl(CPUState *env, int type, ...)
 831 {
 832     int ret;
 833     void *arg;
 834     va_list ap;
 835
 836     va_start(ap, type);
 837     arg = va_arg(ap, void *);
 838     va_end(ap);
 839
 840     ret = ioctl(env->kvm_fd, type, arg);
 841     if (ret == -1)
 842         ret = -errno;
 843
 844     return ret;
 845 }
 846
 847 int kvm_has_sync_mmu(void)
 848 {
 849 #ifdef KVM_CAP_SYNC_MMU
 850     KVMState *s = kvm_state;
 851
 852     return kvm_check_extension(s, KVM_CAP_SYNC_MMU);
 853 #else
 854     return 0;
 855 #endif
 856 }
 857
 858 void kvm_setup_guest_memory(void *start, size_t size)
 859 {
 860     if (!kvm_has_sync_mmu()) {
 861 #ifdef MADV_DONTFORK
 862         int ret = madvise(start, size, MADV_DONTFORK);
 863
 864         if (ret) {
 865             perror("madvice");
 866             exit(1);
 867         }
 868 #else
 869         fprintf(stderr,
 870                 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
 871         exit(1);
 872 #endif
 873     }
 874 }
 875
 876 #ifdef KVM_CAP_SET_GUEST_DEBUG
 877 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
 878                                                  target_ulong pc)
 879 {
 880     struct kvm_sw_breakpoint *bp;
 881
 882     TAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) {
 883         if (bp->pc == pc)
 884             return bp;
 885     }
 886     return NULL;
 887 }
 888
 889 int kvm_sw_breakpoints_active(CPUState *env)
 890 {
 891     return !TAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints);
 892 }
 893
 894 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 895 {
 896     struct kvm_guest_debug dbg;
 897
 898     dbg.control = 0;
 899     if (env->singlestep_enabled)
 900         dbg.control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
 901
 902     kvm_arch_update_guest_debug(env, &dbg);
 903     dbg.control |= reinject_trap;
 904
 905     return kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg);
 906 }
 907
 908 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 909                           target_ulong len, int type)
 910 {
 911     struct kvm_sw_breakpoint *bp;
 912     CPUState *env;
 913     int err;
 914
 915     if (type == GDB_BREAKPOINT_SW) {
 916         bp = kvm_find_sw_breakpoint(current_env, addr);
 917         if (bp) {
 918             bp->use_count++;
 919             return 0;
 920         }
 921
 922         bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint));
 923         if (!bp)
 924             return -ENOMEM;
 925
 926         bp->pc = addr;
 927         bp->use_count = 1;
 928         err = kvm_arch_insert_sw_breakpoint(current_env, bp);
 929         if (err) {
 930             free(bp);
 931             return err;
 932         }
 933
 934         TAILQ_INSERT_HEAD(&current_env->kvm_state->kvm_sw_breakpoints,
 935                           bp, entry);
 936     } else {
 937         err = kvm_arch_insert_hw_breakpoint(addr, len, type);
 938         if (err)
 939             return err;
 940     }
 941
 942     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 943         err = kvm_update_guest_debug(env, 0);
 944         if (err)
 945             return err;
 946     }
 947     return 0;
 948 }
 949
 950 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
 951                           target_ulong len, int type)
 952 {
 953     struct kvm_sw_breakpoint *bp;
 954     CPUState *env;
 955     int err;
 956
 957     if (type == GDB_BREAKPOINT_SW) {
 958         bp = kvm_find_sw_breakpoint(current_env, addr);
 959         if (!bp)
 960             return -ENOENT;
 961
 962         if (bp->use_count > 1) {
 963             bp->use_count--;
 964             return 0;
 965         }
 966
 967         err = kvm_arch_remove_sw_breakpoint(current_env, bp);
 968         if (err)
 969             return err;
 970
 971         TAILQ_REMOVE(&current_env->kvm_state->kvm_sw_breakpoints, bp, entry);
 972         qemu_free(bp);
 973     } else {
 974         err = kvm_arch_remove_hw_breakpoint(addr, len, type);
 975         if (err)
 976             return err;
 977     }
 978
 979     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 980         err = kvm_update_guest_debug(env, 0);
 981         if (err)
 982             return err;
 983     }
 984     return 0;
 985 }
 986
 987 void kvm_remove_all_breakpoints(CPUState *current_env)
 988 {
 989     struct kvm_sw_breakpoint *bp, *next;
 990     KVMState *s = current_env->kvm_state;
 991     CPUState *env;
 992
 993     TAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
 994         if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) {
 995             /* Try harder to find a CPU that currently sees the breakpoint. */
 996             for (env = first_cpu; env != NULL; env = env->next_cpu) {
 997                 if (kvm_arch_remove_sw_breakpoint(env, bp) == 0)
 998                     break;
 999             }
1000         }
1001     }
1002     kvm_arch_remove_all_hw_breakpoints();
1003
1004     for (env = first_cpu; env != NULL; env = env->next_cpu)
1005         kvm_update_guest_debug(env, 0);
1006 }
1007
1008 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1009
1010 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
1011 {
1012     return -EINVAL;
1013 }
1014
1015 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
1016                           target_ulong len, int type)
1017 {
1018     return -EINVAL;
1019 }
1020
1021 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
1022                           target_ulong len, int type)
1023 {
1024     return -EINVAL;
1025 }
1026
1027 void kvm_remove_all_breakpoints(CPUState *current_env)
1028 {
1029 }
1030 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1031 #endif
1032
1033 #include "qemu-kvm.c"