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 void 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     KVMDirtyLog d;
 280     KVMSlot *mem = kvm_lookup_matching_slot(s, start_addr, end_addr);
 281     unsigned long alloc_size;
 282     ram_addr_t addr;
 283     target_phys_addr_t phys_addr = start_addr;
 284
 285     dprintf("sync addr: " TARGET_FMT_lx " into %lx\n", start_addr,
 286             mem->phys_offset);
 287     if (mem == NULL) {
 288             fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
 289                     TARGET_FMT_plx "\n", __func__, phys_addr, end_addr - 1);
 290             return;
 291     }
 292
 293     alloc_size = ((mem->memory_size >> TARGET_PAGE_BITS) + 7) / 8;
 294     d.dirty_bitmap = qemu_mallocz(alloc_size);
 295
 296     d.slot = mem->slot;
 297     dprintf("slot %d, phys_addr %llx, uaddr: %llx\n",
 298             d.slot, mem->start_addr, mem->phys_offset);
 299
 300     if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
 301         dprintf("ioctl failed %d\n", errno);
 302         goto out;
 303     }
 304
 305     phys_addr = start_addr;
 306     for (addr = mem->phys_offset; phys_addr < end_addr; phys_addr+= TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
 307         unsigned long *bitmap = (unsigned long *)d.dirty_bitmap;
 308         unsigned nr = (phys_addr - start_addr) >> TARGET_PAGE_BITS;
 309         unsigned word = nr / (sizeof(*bitmap) * 8);
 310         unsigned bit = nr % (sizeof(*bitmap) * 8);
 311         if ((bitmap[word] >> bit) & 1)
 312             cpu_physical_memory_set_dirty(addr);
 313     }
 314 out:
 315     qemu_free(d.dirty_bitmap);
 316 }
 317
 318 int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 319 {
 320     int ret = -ENOSYS;
 321 #ifdef KVM_CAP_COALESCED_MMIO
 322     KVMState *s = kvm_state;
 323
 324     if (s->coalesced_mmio) {
 325         struct kvm_coalesced_mmio_zone zone;
 326
 327         zone.addr = start;
 328         zone.size = size;
 329
 330         ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
 331     }
 332 #endif
 333
 334     return ret;
 335 }
 336
 337 int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 338 {
 339     int ret = -ENOSYS;
 340 #ifdef KVM_CAP_COALESCED_MMIO
 341     KVMState *s = kvm_state;
 342
 343     if (s->coalesced_mmio) {
 344         struct kvm_coalesced_mmio_zone zone;
 345
 346         zone.addr = start;
 347         zone.size = size;
 348
 349         ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
 350     }
 351 #endif
 352
 353     return ret;
 354 }
 355
 356 int kvm_check_extension(KVMState *s, unsigned int extension)
 357 {
 358     int ret;
 359
 360     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
 361     if (ret < 0) {
 362         ret = 0;
 363     }
 364
 365     return ret;
 366 }
 367
 368 int kvm_init(int smp_cpus)
 369 {
 370     KVMState *s;
 371     int ret;
 372     int i;
 373
 374     if (smp_cpus > 1) {
 375         fprintf(stderr, "No SMP KVM support, use '-smp 1'\n");
 376         return -EINVAL;
 377     }
 378
 379     s = qemu_mallocz(sizeof(KVMState));
 380
 381 #ifdef KVM_CAP_SET_GUEST_DEBUG
 382     TAILQ_INIT(&s->kvm_sw_breakpoints);
 383 #endif
 384     for (i = 0; i < ARRAY_SIZE(s->slots); i++)
 385         s->slots[i].slot = i;
 386
 387     s->vmfd = -1;
 388     s->fd = open("/dev/kvm", O_RDWR);
 389     if (s->fd == -1) {
 390         fprintf(stderr, "Could not access KVM kernel module: %m\n");
 391         ret = -errno;
 392         goto err;
 393     }
 394
 395     ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
 396     if (ret < KVM_API_VERSION) {
 397         if (ret > 0)
 398             ret = -EINVAL;
 399         fprintf(stderr, "kvm version too old\n");
 400         goto err;
 401     }
 402
 403     if (ret > KVM_API_VERSION) {
 404         ret = -EINVAL;
 405         fprintf(stderr, "kvm version not supported\n");
 406         goto err;
 407     }
 408
 409     s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
 410     if (s->vmfd < 0)
 411         goto err;
 412
 413     /* initially, KVM allocated its own memory and we had to jump through
 414      * hooks to make phys_ram_base point to this.  Modern versions of KVM
 415      * just use a user allocated buffer so we can use regular pages
 416      * unmodified.  Make sure we have a sufficiently modern version of KVM.
 417      */
 418     if (!kvm_check_extension(s, KVM_CAP_USER_MEMORY)) {
 419         ret = -EINVAL;
 420         fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n");
 421         goto err;
 422     }
 423
 424     /* There was a nasty bug in < kvm-80 that prevents memory slots from being
 425      * destroyed properly.  Since we rely on this capability, refuse to work
 426      * with any kernel without this capability. */
 427     if (!kvm_check_extension(s, KVM_CAP_DESTROY_MEMORY_REGION_WORKS)) {
 428         ret = -EINVAL;
 429
 430         fprintf(stderr,
 431                 "KVM kernel module broken (DESTROY_MEMORY_REGION)\n"
 432                 "Please upgrade to at least kvm-81.\n");
 433         goto err;
 434     }
 435
 436 #ifdef KVM_CAP_COALESCED_MMIO
 437     s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
 438 #else
 439     s->coalesced_mmio = 0;
 440 #endif
 441
 442     s->broken_set_mem_region = 1;
 443 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
 444     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
 445     if (ret > 0) {
 446         s->broken_set_mem_region = 0;
 447     }
 448 #endif
 449
 450     ret = kvm_arch_init(s, smp_cpus);
 451     if (ret < 0)
 452         goto err;
 453
 454     kvm_state = s;
 455
 456     return 0;
 457
 458 err:
 459     if (s) {
 460         if (s->vmfd != -1)
 461             close(s->vmfd);
 462         if (s->fd != -1)
 463             close(s->fd);
 464     }
 465     qemu_free(s);
 466
 467     return ret;
 468 }
 469
 470 static int kvm_handle_io(CPUState *env, uint16_t port, void *data,
 471                          int direction, int size, uint32_t count)
 472 {
 473     int i;
 474     uint8_t *ptr = data;
 475
 476     for (i = 0; i < count; i++) {
 477         if (direction == KVM_EXIT_IO_IN) {
 478             switch (size) {
 479             case 1:
 480                 stb_p(ptr, cpu_inb(env, port));
 481                 break;
 482             case 2:
 483                 stw_p(ptr, cpu_inw(env, port));
 484                 break;
 485             case 4:
 486                 stl_p(ptr, cpu_inl(env, port));
 487                 break;
 488             }
 489         } else {
 490             switch (size) {
 491             case 1:
 492                 cpu_outb(env, port, ldub_p(ptr));
 493                 break;
 494             case 2:
 495                 cpu_outw(env, port, lduw_p(ptr));
 496                 break;
 497             case 4:
 498                 cpu_outl(env, port, ldl_p(ptr));
 499                 break;
 500             }
 501         }
 502
 503         ptr += size;
 504     }
 505
 506     return 1;
 507 }
 508
 509 static void kvm_run_coalesced_mmio(CPUState *env, struct kvm_run *run)
 510 {
 511 #ifdef KVM_CAP_COALESCED_MMIO
 512     KVMState *s = kvm_state;
 513     if (s->coalesced_mmio) {
 514         struct kvm_coalesced_mmio_ring *ring;
 515
 516         ring = (void *)run + (s->coalesced_mmio * TARGET_PAGE_SIZE);
 517         while (ring->first != ring->last) {
 518             struct kvm_coalesced_mmio *ent;
 519
 520             ent = &ring->coalesced_mmio[ring->first];
 521
 522             cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
 523             /* FIXME smp_wmb() */
 524             ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
 525         }
 526     }
 527 #endif
 528 }
 529
 530 int kvm_cpu_exec(CPUState *env)
 531 {
 532     struct kvm_run *run = env->kvm_run;
 533     int ret;
 534
 535     dprintf("kvm_cpu_exec()\n");
 536
 537     do {
 538         kvm_arch_pre_run(env, run);
 539
 540         if (env->exit_request) {
 541             dprintf("interrupt exit requested\n");
 542             ret = 0;
 543             break;
 544         }
 545
 546         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
 547         kvm_arch_post_run(env, run);
 548
 549         if (ret == -EINTR || ret == -EAGAIN) {
 550             dprintf("io window exit\n");
 551             ret = 0;
 552             break;
 553         }
 554
 555         if (ret < 0) {
 556             dprintf("kvm run failed %s\n", strerror(-ret));
 557             abort();
 558         }
 559
 560         kvm_run_coalesced_mmio(env, run);
 561
 562         ret = 0; /* exit loop */
 563         switch (run->exit_reason) {
 564         case KVM_EXIT_IO:
 565             dprintf("handle_io\n");
 566             ret = kvm_handle_io(env, run->io.port,
 567                                 (uint8_t *)run + run->io.data_offset,
 568                                 run->io.direction,
 569                                 run->io.size,
 570                                 run->io.count);
 571             break;
 572         case KVM_EXIT_MMIO:
 573             dprintf("handle_mmio\n");
 574             cpu_physical_memory_rw(run->mmio.phys_addr,
 575                                    run->mmio.data,
 576                                    run->mmio.len,
 577                                    run->mmio.is_write);
 578             ret = 1;
 579             break;
 580         case KVM_EXIT_IRQ_WINDOW_OPEN:
 581             dprintf("irq_window_open\n");
 582             break;
 583         case KVM_EXIT_SHUTDOWN:
 584             dprintf("shutdown\n");
 585             qemu_system_reset_request();
 586             ret = 1;
 587             break;
 588         case KVM_EXIT_UNKNOWN:
 589             dprintf("kvm_exit_unknown\n");
 590             break;
 591         case KVM_EXIT_FAIL_ENTRY:
 592             dprintf("kvm_exit_fail_entry\n");
 593             break;
 594         case KVM_EXIT_EXCEPTION:
 595             dprintf("kvm_exit_exception\n");
 596             break;
 597         case KVM_EXIT_DEBUG:
 598             dprintf("kvm_exit_debug\n");
 599 #ifdef KVM_CAP_SET_GUEST_DEBUG
 600             if (kvm_arch_debug(&run->debug.arch)) {
 601                 gdb_set_stop_cpu(env);
 602                 vm_stop(EXCP_DEBUG);
 603                 env->exception_index = EXCP_DEBUG;
 604                 return 0;
 605             }
 606             /* re-enter, this exception was guest-internal */
 607             ret = 1;
 608 #endif /* KVM_CAP_SET_GUEST_DEBUG */
 609             break;
 610         default:
 611             dprintf("kvm_arch_handle_exit\n");
 612             ret = kvm_arch_handle_exit(env, run);
 613             break;
 614         }
 615     } while (ret > 0);
 616
 617     if (env->exit_request) {
 618         env->exit_request = 0;
 619         env->exception_index = EXCP_INTERRUPT;
 620     }
 621
 622     return ret;
 623 }
 624
 625 void kvm_set_phys_mem(target_phys_addr_t start_addr,
 626                       ram_addr_t size,
 627                       ram_addr_t phys_offset)
 628 {
 629     KVMState *s = kvm_state;
 630     ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK;
 631     KVMSlot *mem, old;
 632     int err;
 633
 634     if (start_addr & ~TARGET_PAGE_MASK) {
 635         if (flags >= IO_MEM_UNASSIGNED) {
 636             if (!kvm_lookup_overlapping_slot(s, start_addr,
 637                                              start_addr + size)) {
 638                 return;
 639             }
 640             fprintf(stderr, "Unaligned split of a KVM memory slot\n");
 641         } else {
 642             fprintf(stderr, "Only page-aligned memory slots supported\n");
 643         }
 644         abort();
 645     }
 646
 647     /* KVM does not support read-only slots */
 648     phys_offset &= ~IO_MEM_ROM;
 649
 650     while (1) {
 651         mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
 652         if (!mem) {
 653             break;
 654         }
 655
 656         if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr &&
 657             (start_addr + size <= mem->start_addr + mem->memory_size) &&
 658             (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) {
 659             /* The new slot fits into the existing one and comes with
 660              * identical parameters - nothing to be done. */
 661             return;
 662         }
 663
 664         old = *mem;
 665
 666         /* unregister the overlapping slot */
 667         mem->memory_size = 0;
 668         err = kvm_set_user_memory_region(s, mem);
 669         if (err) {
 670             fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
 671                     __func__, strerror(-err));
 672             abort();
 673         }
 674
 675         /* Workaround for older KVM versions: we can't join slots, even not by
 676          * unregistering the previous ones and then registering the larger
 677          * slot. We have to maintain the existing fragmentation. Sigh.
 678          *
 679          * This workaround assumes that the new slot starts at the same
 680          * address as the first existing one. If not or if some overlapping
 681          * slot comes around later, we will fail (not seen in practice so far)
 682          * - and actually require a recent KVM version. */
 683         if (s->broken_set_mem_region &&
 684             old.start_addr == start_addr && old.memory_size < size &&
 685             flags < IO_MEM_UNASSIGNED) {
 686             mem = kvm_alloc_slot(s);
 687             mem->memory_size = old.memory_size;
 688             mem->start_addr = old.start_addr;
 689             mem->phys_offset = old.phys_offset;
 690             mem->flags = 0;
 691
 692             err = kvm_set_user_memory_region(s, mem);
 693             if (err) {
 694                 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
 695                         strerror(-err));
 696                 abort();
 697             }
 698
 699             start_addr += old.memory_size;
 700             phys_offset += old.memory_size;
 701             size -= old.memory_size;
 702             continue;
 703         }
 704
 705         /* register prefix slot */
 706         if (old.start_addr < start_addr) {
 707             mem = kvm_alloc_slot(s);
 708             mem->memory_size = start_addr - old.start_addr;
 709             mem->start_addr = old.start_addr;
 710             mem->phys_offset = old.phys_offset;
 711             mem->flags = 0;
 712
 713             err = kvm_set_user_memory_region(s, mem);
 714             if (err) {
 715                 fprintf(stderr, "%s: error registering prefix slot: %s\n",
 716                         __func__, strerror(-err));
 717                 abort();
 718             }
 719         }
 720
 721         /* register suffix slot */
 722         if (old.start_addr + old.memory_size > start_addr + size) {
 723             ram_addr_t size_delta;
 724
 725             mem = kvm_alloc_slot(s);
 726             mem->start_addr = start_addr + size;
 727             size_delta = mem->start_addr - old.start_addr;
 728             mem->memory_size = old.memory_size - size_delta;
 729             mem->phys_offset = old.phys_offset + size_delta;
 730             mem->flags = 0;
 731
 732             err = kvm_set_user_memory_region(s, mem);
 733             if (err) {
 734                 fprintf(stderr, "%s: error registering suffix slot: %s\n",
 735                         __func__, strerror(-err));
 736                 abort();
 737             }
 738         }
 739     }
 740
 741     /* in case the KVM bug workaround already "consumed" the new slot */
 742     if (!size)
 743         return;
 744
 745     /* KVM does not need to know about this memory */
 746     if (flags >= IO_MEM_UNASSIGNED)
 747         return;
 748
 749     mem = kvm_alloc_slot(s);
 750     mem->memory_size = size;
 751     mem->start_addr = start_addr;
 752     mem->phys_offset = phys_offset;
 753     mem->flags = 0;
 754
 755     err = kvm_set_user_memory_region(s, mem);
 756     if (err) {
 757         fprintf(stderr, "%s: error registering slot: %s\n", __func__,
 758                 strerror(-err));
 759         abort();
 760     }
 761 }
 762
 763 int kvm_ioctl(KVMState *s, int type, ...)
 764 {
 765     int ret;
 766     void *arg;
 767     va_list ap;
 768
 769     va_start(ap, type);
 770     arg = va_arg(ap, void *);
 771     va_end(ap);
 772
 773     ret = ioctl(s->fd, type, arg);
 774     if (ret == -1)
 775         ret = -errno;
 776
 777     return ret;
 778 }
 779
 780 int kvm_vm_ioctl(KVMState *s, int type, ...)
 781 {
 782     int ret;
 783     void *arg;
 784     va_list ap;
 785
 786     va_start(ap, type);
 787     arg = va_arg(ap, void *);
 788     va_end(ap);
 789
 790     ret = ioctl(s->vmfd, type, arg);
 791     if (ret == -1)
 792         ret = -errno;
 793
 794     return ret;
 795 }
 796
 797 int kvm_vcpu_ioctl(CPUState *env, int type, ...)
 798 {
 799     int ret;
 800     void *arg;
 801     va_list ap;
 802
 803     va_start(ap, type);
 804     arg = va_arg(ap, void *);
 805     va_end(ap);
 806
 807     ret = ioctl(env->kvm_fd, type, arg);
 808     if (ret == -1)
 809         ret = -errno;
 810
 811     return ret;
 812 }
 813
 814 int kvm_has_sync_mmu(void)
 815 {
 816 #ifdef KVM_CAP_SYNC_MMU
 817     KVMState *s = kvm_state;
 818
 819     return kvm_check_extension(s, KVM_CAP_SYNC_MMU);
 820 #else
 821     return 0;
 822 #endif
 823 }
 824
 825 void kvm_setup_guest_memory(void *start, size_t size)
 826 {
 827     if (!kvm_has_sync_mmu()) {
 828 #ifdef MADV_DONTFORK
 829         int ret = madvise(start, size, MADV_DONTFORK);
 830
 831         if (ret) {
 832             perror("madvice");
 833             exit(1);
 834         }
 835 #else
 836         fprintf(stderr,
 837                 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
 838         exit(1);
 839 #endif
 840     }
 841 }
 842
 843 #ifdef KVM_CAP_SET_GUEST_DEBUG
 844 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
 845                                                  target_ulong pc)
 846 {
 847     struct kvm_sw_breakpoint *bp;
 848
 849     TAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) {
 850         if (bp->pc == pc)
 851             return bp;
 852     }
 853     return NULL;
 854 }
 855
 856 int kvm_sw_breakpoints_active(CPUState *env)
 857 {
 858     return !TAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints);
 859 }
 860
 861 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 862 {
 863     struct kvm_guest_debug dbg;
 864
 865     dbg.control = 0;
 866     if (env->singlestep_enabled)
 867         dbg.control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
 868
 869     kvm_arch_update_guest_debug(env, &dbg);
 870     dbg.control |= reinject_trap;
 871
 872     return kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg);
 873 }
 874
 875 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 876                           target_ulong len, int type)
 877 {
 878     struct kvm_sw_breakpoint *bp;
 879     CPUState *env;
 880     int err;
 881
 882     if (type == GDB_BREAKPOINT_SW) {
 883         bp = kvm_find_sw_breakpoint(current_env, addr);
 884         if (bp) {
 885             bp->use_count++;
 886             return 0;
 887         }
 888
 889         bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint));
 890         if (!bp)
 891             return -ENOMEM;
 892
 893         bp->pc = addr;
 894         bp->use_count = 1;
 895         err = kvm_arch_insert_sw_breakpoint(current_env, bp);
 896         if (err) {
 897             free(bp);
 898             return err;
 899         }
 900
 901         TAILQ_INSERT_HEAD(&current_env->kvm_state->kvm_sw_breakpoints,
 902                           bp, entry);
 903     } else {
 904         err = kvm_arch_insert_hw_breakpoint(addr, len, type);
 905         if (err)
 906             return err;
 907     }
 908
 909     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 910         err = kvm_update_guest_debug(env, 0);
 911         if (err)
 912             return err;
 913     }
 914     return 0;
 915 }
 916
 917 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
 918                           target_ulong len, int type)
 919 {
 920     struct kvm_sw_breakpoint *bp;
 921     CPUState *env;
 922     int err;
 923
 924     if (type == GDB_BREAKPOINT_SW) {
 925         bp = kvm_find_sw_breakpoint(current_env, addr);
 926         if (!bp)
 927             return -ENOENT;
 928
 929         if (bp->use_count > 1) {
 930             bp->use_count--;
 931             return 0;
 932         }
 933
 934         err = kvm_arch_remove_sw_breakpoint(current_env, bp);
 935         if (err)
 936             return err;
 937
 938         TAILQ_REMOVE(&current_env->kvm_state->kvm_sw_breakpoints, bp, entry);
 939         qemu_free(bp);
 940     } else {
 941         err = kvm_arch_remove_hw_breakpoint(addr, len, type);
 942         if (err)
 943             return err;
 944     }
 945
 946     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 947         err = kvm_update_guest_debug(env, 0);
 948         if (err)
 949             return err;
 950     }
 951     return 0;
 952 }
 953
 954 void kvm_remove_all_breakpoints(CPUState *current_env)
 955 {
 956     struct kvm_sw_breakpoint *bp, *next;
 957     KVMState *s = current_env->kvm_state;
 958     CPUState *env;
 959
 960     TAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
 961         if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) {
 962             /* Try harder to find a CPU that currently sees the breakpoint. */
 963             for (env = first_cpu; env != NULL; env = env->next_cpu) {
 964                 if (kvm_arch_remove_sw_breakpoint(env, bp) == 0)
 965                     break;
 966             }
 967         }
 968     }
 969     kvm_arch_remove_all_hw_breakpoints();
 970
 971     for (env = first_cpu; env != NULL; env = env->next_cpu)
 972         kvm_update_guest_debug(env, 0);
 973 }
 974
 975 #else /* !KVM_CAP_SET_GUEST_DEBUG */
 976
 977 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 978 {
 979     return -EINVAL;
 980 }
 981
 982 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 983                           target_ulong len, int type)
 984 {
 985     return -EINVAL;
 986 }
 987
 988 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
 989                           target_ulong len, int type)
 990 {
 991     return -EINVAL;
 992 }
 993
 994 void kvm_remove_all_breakpoints(CPUState *current_env)
 995 {
 996 }
 997 #endif /* !KVM_CAP_SET_GUEST_DEBUG */