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 #endif
 387 int kvm_check_extension(KVMState *s, unsigned int extension)
 388 {
 389     int ret;
 390
 391     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
 392     if (ret < 0) {
 393         ret = 0;
 394     }
 395
 396     return ret;
 397 }
 398 #ifdef KVM_UPSTREAM
 399
 400 int kvm_init(int smp_cpus)
 401 {
 402     static const char upgrade_note[] =
 403         "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
 404         "(see http://sourceforge.net/projects/kvm).\n";
 405     KVMState *s;
 406     int ret;
 407     int i;
 408
 409     if (smp_cpus > 1) {
 410         fprintf(stderr, "No SMP KVM support, use '-smp 1'\n");
 411         return -EINVAL;
 412     }
 413
 414     s = qemu_mallocz(sizeof(KVMState));
 415
 416 #ifdef KVM_CAP_SET_GUEST_DEBUG
 417     TAILQ_INIT(&s->kvm_sw_breakpoints);
 418 #endif
 419     for (i = 0; i < ARRAY_SIZE(s->slots); i++)
 420         s->slots[i].slot = i;
 421
 422     s->vmfd = -1;
 423     s->fd = open("/dev/kvm", O_RDWR);
 424     if (s->fd == -1) {
 425         fprintf(stderr, "Could not access KVM kernel module: %m\n");
 426         ret = -errno;
 427         goto err;
 428     }
 429
 430     ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
 431     if (ret < KVM_API_VERSION) {
 432         if (ret > 0)
 433             ret = -EINVAL;
 434         fprintf(stderr, "kvm version too old\n");
 435         goto err;
 436     }
 437
 438     if (ret > KVM_API_VERSION) {
 439         ret = -EINVAL;
 440         fprintf(stderr, "kvm version not supported\n");
 441         goto err;
 442     }
 443
 444     s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
 445     if (s->vmfd < 0)
 446         goto err;
 447
 448     /* initially, KVM allocated its own memory and we had to jump through
 449      * hooks to make phys_ram_base point to this.  Modern versions of KVM
 450      * just use a user allocated buffer so we can use regular pages
 451      * unmodified.  Make sure we have a sufficiently modern version of KVM.
 452      */
 453     if (!kvm_check_extension(s, KVM_CAP_USER_MEMORY)) {
 454         ret = -EINVAL;
 455         fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
 456                 upgrade_note);
 457         goto err;
 458     }
 459
 460     /* There was a nasty bug in < kvm-80 that prevents memory slots from being
 461      * destroyed properly.  Since we rely on this capability, refuse to work
 462      * with any kernel without this capability. */
 463     if (!kvm_check_extension(s, KVM_CAP_DESTROY_MEMORY_REGION_WORKS)) {
 464         ret = -EINVAL;
 465
 466         fprintf(stderr,
 467                 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
 468                 upgrade_note);
 469         goto err;
 470     }
 471
 472 #ifdef KVM_CAP_COALESCED_MMIO
 473     s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
 474 #else
 475     s->coalesced_mmio = 0;
 476 #endif
 477
 478     s->broken_set_mem_region = 1;
 479 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
 480     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
 481     if (ret > 0) {
 482         s->broken_set_mem_region = 0;
 483     }
 484 #endif
 485
 486     ret = kvm_arch_init(s, smp_cpus);
 487     if (ret < 0)
 488         goto err;
 489
 490     kvm_state = s;
 491
 492     return 0;
 493
 494 err:
 495     if (s) {
 496         if (s->vmfd != -1)
 497             close(s->vmfd);
 498         if (s->fd != -1)
 499             close(s->fd);
 500     }
 501     qemu_free(s);
 502
 503     return ret;
 504 }
 505
 506 static int kvm_handle_io(CPUState *env, uint16_t port, void *data,
 507                          int direction, int size, uint32_t count)
 508 {
 509     int i;
 510     uint8_t *ptr = data;
 511
 512     for (i = 0; i < count; i++) {
 513         if (direction == KVM_EXIT_IO_IN) {
 514             switch (size) {
 515             case 1:
 516                 stb_p(ptr, cpu_inb(env, port));
 517                 break;
 518             case 2:
 519                 stw_p(ptr, cpu_inw(env, port));
 520                 break;
 521             case 4:
 522                 stl_p(ptr, cpu_inl(env, port));
 523                 break;
 524             }
 525         } else {
 526             switch (size) {
 527             case 1:
 528                 cpu_outb(env, port, ldub_p(ptr));
 529                 break;
 530             case 2:
 531                 cpu_outw(env, port, lduw_p(ptr));
 532                 break;
 533             case 4:
 534                 cpu_outl(env, port, ldl_p(ptr));
 535                 break;
 536             }
 537         }
 538
 539         ptr += size;
 540     }
 541
 542     return 1;
 543 }
 544
 545 static void kvm_run_coalesced_mmio(CPUState *env, struct kvm_run *run)
 546 {
 547 #ifdef KVM_CAP_COALESCED_MMIO
 548     KVMState *s = kvm_state;
 549     if (s->coalesced_mmio) {
 550         struct kvm_coalesced_mmio_ring *ring;
 551
 552         ring = (void *)run + (s->coalesced_mmio * TARGET_PAGE_SIZE);
 553         while (ring->first != ring->last) {
 554             struct kvm_coalesced_mmio *ent;
 555
 556             ent = &ring->coalesced_mmio[ring->first];
 557
 558             cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
 559             /* FIXME smp_wmb() */
 560             ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
 561         }
 562     }
 563 #endif
 564 }
 565
 566 int kvm_cpu_exec(CPUState *env)
 567 {
 568     struct kvm_run *run = env->kvm_run;
 569     int ret;
 570
 571     dprintf("kvm_cpu_exec()\n");
 572
 573     do {
 574         if (env->exit_request) {
 575             dprintf("interrupt exit requested\n");
 576             ret = 0;
 577             break;
 578         }
 579
 580         kvm_arch_pre_run(env, run);
 581         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
 582         kvm_arch_post_run(env, run);
 583
 584         if (ret == -EINTR || ret == -EAGAIN) {
 585             dprintf("io window exit\n");
 586             ret = 0;
 587             break;
 588         }
 589
 590         if (ret < 0) {
 591             dprintf("kvm run failed %s\n", strerror(-ret));
 592             abort();
 593         }
 594
 595         kvm_run_coalesced_mmio(env, run);
 596
 597         ret = 0; /* exit loop */
 598         switch (run->exit_reason) {
 599         case KVM_EXIT_IO:
 600             dprintf("handle_io\n");
 601             ret = kvm_handle_io(env, run->io.port,
 602                                 (uint8_t *)run + run->io.data_offset,
 603                                 run->io.direction,
 604                                 run->io.size,
 605                                 run->io.count);
 606             break;
 607         case KVM_EXIT_MMIO:
 608             dprintf("handle_mmio\n");
 609             cpu_physical_memory_rw(run->mmio.phys_addr,
 610                                    run->mmio.data,
 611                                    run->mmio.len,
 612                                    run->mmio.is_write);
 613             ret = 1;
 614             break;
 615         case KVM_EXIT_IRQ_WINDOW_OPEN:
 616             dprintf("irq_window_open\n");
 617             break;
 618         case KVM_EXIT_SHUTDOWN:
 619             dprintf("shutdown\n");
 620             qemu_system_reset_request();
 621             ret = 1;
 622             break;
 623         case KVM_EXIT_UNKNOWN:
 624             dprintf("kvm_exit_unknown\n");
 625             break;
 626         case KVM_EXIT_FAIL_ENTRY:
 627             dprintf("kvm_exit_fail_entry\n");
 628             break;
 629         case KVM_EXIT_EXCEPTION:
 630             dprintf("kvm_exit_exception\n");
 631             break;
 632         case KVM_EXIT_DEBUG:
 633             dprintf("kvm_exit_debug\n");
 634 #ifdef KVM_CAP_SET_GUEST_DEBUG
 635             if (kvm_arch_debug(&run->debug.arch)) {
 636                 gdb_set_stop_cpu(env);
 637                 vm_stop(EXCP_DEBUG);
 638                 env->exception_index = EXCP_DEBUG;
 639                 return 0;
 640             }
 641             /* re-enter, this exception was guest-internal */
 642             ret = 1;
 643 #endif /* KVM_CAP_SET_GUEST_DEBUG */
 644             break;
 645         default:
 646             dprintf("kvm_arch_handle_exit\n");
 647             ret = kvm_arch_handle_exit(env, run);
 648             break;
 649         }
 650     } while (ret > 0);
 651
 652     if (env->exit_request) {
 653         env->exit_request = 0;
 654         env->exception_index = EXCP_INTERRUPT;
 655     }
 656
 657     return ret;
 658 }
 659
 660 void kvm_set_phys_mem(target_phys_addr_t start_addr,
 661                       ram_addr_t size,
 662                       ram_addr_t phys_offset)
 663 {
 664     KVMState *s = kvm_state;
 665     ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK;
 666     KVMSlot *mem, old;
 667     int err;
 668
 669     if (start_addr & ~TARGET_PAGE_MASK) {
 670         if (flags >= IO_MEM_UNASSIGNED) {
 671             if (!kvm_lookup_overlapping_slot(s, start_addr,
 672                                              start_addr + size)) {
 673                 return;
 674             }
 675             fprintf(stderr, "Unaligned split of a KVM memory slot\n");
 676         } else {
 677             fprintf(stderr, "Only page-aligned memory slots supported\n");
 678         }
 679         abort();
 680     }
 681
 682     /* KVM does not support read-only slots */
 683     phys_offset &= ~IO_MEM_ROM;
 684
 685     while (1) {
 686         mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
 687         if (!mem) {
 688             break;
 689         }
 690
 691         if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr &&
 692             (start_addr + size <= mem->start_addr + mem->memory_size) &&
 693             (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) {
 694             /* The new slot fits into the existing one and comes with
 695              * identical parameters - nothing to be done. */
 696             return;
 697         }
 698
 699         old = *mem;
 700
 701         /* unregister the overlapping slot */
 702         mem->memory_size = 0;
 703         err = kvm_set_user_memory_region(s, mem);
 704         if (err) {
 705             fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
 706                     __func__, strerror(-err));
 707             abort();
 708         }
 709
 710         /* Workaround for older KVM versions: we can't join slots, even not by
 711          * unregistering the previous ones and then registering the larger
 712          * slot. We have to maintain the existing fragmentation. Sigh.
 713          *
 714          * This workaround assumes that the new slot starts at the same
 715          * address as the first existing one. If not or if some overlapping
 716          * slot comes around later, we will fail (not seen in practice so far)
 717          * - and actually require a recent KVM version. */
 718         if (s->broken_set_mem_region &&
 719             old.start_addr == start_addr && old.memory_size < size &&
 720             flags < IO_MEM_UNASSIGNED) {
 721             mem = kvm_alloc_slot(s);
 722             mem->memory_size = old.memory_size;
 723             mem->start_addr = old.start_addr;
 724             mem->phys_offset = old.phys_offset;
 725             mem->flags = 0;
 726
 727             err = kvm_set_user_memory_region(s, mem);
 728             if (err) {
 729                 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
 730                         strerror(-err));
 731                 abort();
 732             }
 733
 734             start_addr += old.memory_size;
 735             phys_offset += old.memory_size;
 736             size -= old.memory_size;
 737             continue;
 738         }
 739
 740         /* register prefix slot */
 741         if (old.start_addr < start_addr) {
 742             mem = kvm_alloc_slot(s);
 743             mem->memory_size = start_addr - old.start_addr;
 744             mem->start_addr = old.start_addr;
 745             mem->phys_offset = old.phys_offset;
 746             mem->flags = 0;
 747
 748             err = kvm_set_user_memory_region(s, mem);
 749             if (err) {
 750                 fprintf(stderr, "%s: error registering prefix slot: %s\n",
 751                         __func__, strerror(-err));
 752                 abort();
 753             }
 754         }
 755
 756         /* register suffix slot */
 757         if (old.start_addr + old.memory_size > start_addr + size) {
 758             ram_addr_t size_delta;
 759
 760             mem = kvm_alloc_slot(s);
 761             mem->start_addr = start_addr + size;
 762             size_delta = mem->start_addr - old.start_addr;
 763             mem->memory_size = old.memory_size - size_delta;
 764             mem->phys_offset = old.phys_offset + size_delta;
 765             mem->flags = 0;
 766
 767             err = kvm_set_user_memory_region(s, mem);
 768             if (err) {
 769                 fprintf(stderr, "%s: error registering suffix slot: %s\n",
 770                         __func__, strerror(-err));
 771                 abort();
 772             }
 773         }
 774     }
 775
 776     /* in case the KVM bug workaround already "consumed" the new slot */
 777     if (!size)
 778         return;
 779
 780     /* KVM does not need to know about this memory */
 781     if (flags >= IO_MEM_UNASSIGNED)
 782         return;
 783
 784     mem = kvm_alloc_slot(s);
 785     mem->memory_size = size;
 786     mem->start_addr = start_addr;
 787     mem->phys_offset = phys_offset;
 788     mem->flags = 0;
 789
 790     err = kvm_set_user_memory_region(s, mem);
 791     if (err) {
 792         fprintf(stderr, "%s: error registering slot: %s\n", __func__,
 793                 strerror(-err));
 794         abort();
 795     }
 796 }
 797
 798 #endif
 799 int kvm_ioctl(KVMState *s, int type, ...)
 800 {
 801     int ret;
 802     void *arg;
 803     va_list ap;
 804
 805     va_start(ap, type);
 806     arg = va_arg(ap, void *);
 807     va_end(ap);
 808
 809     ret = ioctl(s->fd, type, arg);
 810     if (ret == -1)
 811         ret = -errno;
 812
 813     return ret;
 814 }
 815
 816 int kvm_vm_ioctl(KVMState *s, int type, ...)
 817 {
 818     int ret;
 819     void *arg;
 820     va_list ap;
 821
 822     va_start(ap, type);
 823     arg = va_arg(ap, void *);
 824     va_end(ap);
 825
 826     ret = ioctl(s->vmfd, type, arg);
 827     if (ret == -1)
 828         ret = -errno;
 829
 830     return ret;
 831 }
 832
 833 #ifdef KVM_UPSTREAM
 834 int kvm_vcpu_ioctl(CPUState *env, int type, ...)
 835 {
 836     int ret;
 837     void *arg;
 838     va_list ap;
 839
 840     va_start(ap, type);
 841     arg = va_arg(ap, void *);
 842     va_end(ap);
 843
 844     ret = ioctl(env->kvm_fd, type, arg);
 845     if (ret == -1)
 846         ret = -errno;
 847
 848     return ret;
 849 }
 850
 851 int kvm_has_sync_mmu(void)
 852 {
 853 #ifdef KVM_CAP_SYNC_MMU
 854     KVMState *s = kvm_state;
 855
 856     return kvm_check_extension(s, KVM_CAP_SYNC_MMU);
 857 #else
 858     return 0;
 859 #endif
 860 }
 861
 862 void kvm_setup_guest_memory(void *start, size_t size)
 863 {
 864     if (!kvm_has_sync_mmu()) {
 865 #ifdef MADV_DONTFORK
 866         int ret = madvise(start, size, MADV_DONTFORK);
 867
 868         if (ret) {
 869             perror("madvice");
 870             exit(1);
 871         }
 872 #else
 873         fprintf(stderr,
 874                 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
 875         exit(1);
 876 #endif
 877     }
 878 }
 879
 880 #endif /* KVM_UPSTREAM */
 881
 882 #ifdef KVM_CAP_SET_GUEST_DEBUG
 883 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
 884                                                  target_ulong pc)
 885 {
 886     struct kvm_sw_breakpoint *bp;
 887
 888     TAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) {
 889         if (bp->pc == pc)
 890             return bp;
 891     }
 892     return NULL;
 893 }
 894
 895 int kvm_sw_breakpoints_active(CPUState *env)
 896 {
 897     return !TAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints);
 898 }
 899
 900 #ifdef KVM_UPSTREAM
 901 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 902 {
 903     struct kvm_guest_debug dbg;
 904
 905     dbg.control = 0;
 906     if (env->singlestep_enabled)
 907         dbg.control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
 908
 909     kvm_arch_update_guest_debug(env, &dbg);
 910     dbg.control |= reinject_trap;
 911
 912     return kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg);
 913 }
 914 #endif
 915
 916 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 917                           target_ulong len, int type)
 918 {
 919     struct kvm_sw_breakpoint *bp;
 920     CPUState *env;
 921     int err;
 922
 923     if (type == GDB_BREAKPOINT_SW) {
 924         bp = kvm_find_sw_breakpoint(current_env, addr);
 925         if (bp) {
 926             bp->use_count++;
 927             return 0;
 928         }
 929
 930         bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint));
 931         if (!bp)
 932             return -ENOMEM;
 933
 934         bp->pc = addr;
 935         bp->use_count = 1;
 936         err = kvm_arch_insert_sw_breakpoint(current_env, bp);
 937         if (err) {
 938             free(bp);
 939             return err;
 940         }
 941
 942         TAILQ_INSERT_HEAD(&current_env->kvm_state->kvm_sw_breakpoints,
 943                           bp, entry);
 944     } else {
 945         err = kvm_arch_insert_hw_breakpoint(addr, len, type);
 946         if (err)
 947             return err;
 948     }
 949
 950     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 951         err = kvm_update_guest_debug(env, 0);
 952         if (err)
 953             return err;
 954     }
 955     return 0;
 956 }
 957
 958 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
 959                           target_ulong len, int type)
 960 {
 961     struct kvm_sw_breakpoint *bp;
 962     CPUState *env;
 963     int err;
 964
 965     if (type == GDB_BREAKPOINT_SW) {
 966         bp = kvm_find_sw_breakpoint(current_env, addr);
 967         if (!bp)
 968             return -ENOENT;
 969
 970         if (bp->use_count > 1) {
 971             bp->use_count--;
 972             return 0;
 973         }
 974
 975         err = kvm_arch_remove_sw_breakpoint(current_env, bp);
 976         if (err)
 977             return err;
 978
 979         TAILQ_REMOVE(&current_env->kvm_state->kvm_sw_breakpoints, bp, entry);
 980         qemu_free(bp);
 981     } else {
 982         err = kvm_arch_remove_hw_breakpoint(addr, len, type);
 983         if (err)
 984             return err;
 985     }
 986
 987     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 988         err = kvm_update_guest_debug(env, 0);
 989         if (err)
 990             return err;
 991     }
 992     return 0;
 993 }
 994
 995 void kvm_remove_all_breakpoints(CPUState *current_env)
 996 {
 997     struct kvm_sw_breakpoint *bp, *next;
 998     KVMState *s = current_env->kvm_state;
 999     CPUState *env;
1000
1001     TAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
1002         if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) {
1003             /* Try harder to find a CPU that currently sees the breakpoint. */
1004             for (env = first_cpu; env != NULL; env = env->next_cpu) {
1005                 if (kvm_arch_remove_sw_breakpoint(env, bp) == 0)
1006                     break;
1007             }
1008         }
1009     }
1010     kvm_arch_remove_all_hw_breakpoints();
1011
1012     for (env = first_cpu; env != NULL; env = env->next_cpu)
1013         kvm_update_guest_debug(env, 0);
1014 }
1015
1016 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1017
1018 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
1019 {
1020     return -EINVAL;
1021 }
1022
1023 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
1024                           target_ulong len, int type)
1025 {
1026     return -EINVAL;
1027 }
1028
1029 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
1030                           target_ulong len, int type)
1031 {
1032     return -EINVAL;
1033 }
1034
1035 void kvm_remove_all_breakpoints(CPUState *current_env)
1036 {
1037 }
1038 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1039
1040 #include "qemu-kvm.c"