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 /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */
  30 #define PAGE_SIZE TARGET_PAGE_SIZE
  31
  32 //#define DEBUG_KVM
  33
  34 #ifdef DEBUG_KVM
  35 #define dprintf(fmt, ...) \
  36     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  37 #else
  38 #define dprintf(fmt, ...) \
  39     do { } while (0)
  40 #endif
  41
  42 typedef struct KVMSlot
  43 {
  44     target_phys_addr_t start_addr;
  45     ram_addr_t memory_size;
  46     ram_addr_t phys_offset;
  47     int slot;
  48     int flags;
  49 } KVMSlot;
  50
  51 typedef struct kvm_dirty_log KVMDirtyLog;
  52
  53 int kvm_allowed = 0;
  54
  55 struct KVMState
  56 {
  57     KVMSlot slots[32];
  58     int fd;
  59     int vmfd;
  60     int coalesced_mmio;
  61     int broken_set_mem_region;
  62     int migration_log;
  63 #ifdef KVM_CAP_SET_GUEST_DEBUG
  64     struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
  65 #endif
  66     int irqchip_in_kernel;
  67     int pit_in_kernel;
  68 };
  69
  70 static KVMState *kvm_state;
  71
  72 static KVMSlot *kvm_alloc_slot(KVMState *s)
  73 {
  74     int i;
  75
  76     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
  77         /* KVM private memory slots */
  78         if (i >= 8 && i < 12)
  79             continue;
  80         if (s->slots[i].memory_size == 0)
  81             return &s->slots[i];
  82     }
  83
  84     fprintf(stderr, "%s: no free slot available\n", __func__);
  85     abort();
  86 }
  87
  88 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
  89                                          target_phys_addr_t start_addr,
  90                                          target_phys_addr_t end_addr)
  91 {
  92     int i;
  93
  94     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
  95         KVMSlot *mem = &s->slots[i];
  96
  97         if (start_addr == mem->start_addr &&
  98             end_addr == mem->start_addr + mem->memory_size) {
  99             return mem;
 100         }
 101     }
 102
 103     return NULL;
 104 }
 105
 106 /*
 107  * Find overlapping slot with lowest start address
 108  */
 109 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
 110                                             target_phys_addr_t start_addr,
 111                                             target_phys_addr_t end_addr)
 112 {
 113     KVMSlot *found = NULL;
 114     int i;
 115
 116     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
 117         KVMSlot *mem = &s->slots[i];
 118
 119         if (mem->memory_size == 0 ||
 120             (found && found->start_addr < mem->start_addr)) {
 121             continue;
 122         }
 123
 124         if (end_addr > mem->start_addr &&
 125             start_addr < mem->start_addr + mem->memory_size) {
 126             found = mem;
 127         }
 128     }
 129
 130     return found;
 131 }
 132
 133 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
 134 {
 135     struct kvm_userspace_memory_region mem;
 136
 137     mem.slot = slot->slot;
 138     mem.guest_phys_addr = slot->start_addr;
 139     mem.memory_size = slot->memory_size;
 140     mem.userspace_addr = (unsigned long)qemu_get_ram_ptr(slot->phys_offset);
 141     mem.flags = slot->flags;
 142     if (s->migration_log) {
 143         mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
 144     }
 145     return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
 146 }
 147
 148 static void kvm_reset_vcpu(void *opaque)
 149 {
 150     CPUState *env = opaque;
 151
 152     if (kvm_arch_put_registers(env)) {
 153         fprintf(stderr, "Fatal: kvm vcpu reset failed\n");
 154         abort();
 155     }
 156 }
 157
 158 static void on_vcpu(CPUState *env, void (*func)(void *data), void *data)
 159 {
 160     if (env == cpu_single_env) {
 161         func(data);
 162         return;
 163     }
 164     abort();
 165 }
 166
 167 int kvm_irqchip_in_kernel(void)
 168 {
 169     return kvm_state->irqchip_in_kernel;
 170 }
 171
 172 int kvm_pit_in_kernel(void)
 173 {
 174     return kvm_state->pit_in_kernel;
 175 }
 176
 177
 178 int kvm_init_vcpu(CPUState *env)
 179 {
 180     KVMState *s = kvm_state;
 181     long mmap_size;
 182     int ret;
 183
 184     dprintf("kvm_init_vcpu\n");
 185
 186     ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index);
 187     if (ret < 0) {
 188         dprintf("kvm_create_vcpu failed\n");
 189         goto err;
 190     }
 191
 192     env->kvm_fd = ret;
 193     env->kvm_state = s;
 194
 195     mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
 196     if (mmap_size < 0) {
 197         dprintf("KVM_GET_VCPU_MMAP_SIZE failed\n");
 198         goto err;
 199     }
 200
 201     env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
 202                         env->kvm_fd, 0);
 203     if (env->kvm_run == MAP_FAILED) {
 204         ret = -errno;
 205         dprintf("mmap'ing vcpu state failed\n");
 206         goto err;
 207     }
 208
 209     ret = kvm_arch_init_vcpu(env);
 210     if (ret == 0) {
 211         qemu_register_reset(kvm_reset_vcpu, env);
 212         ret = kvm_arch_put_registers(env);
 213     }
 214 err:
 215     return ret;
 216 }
 217
 218 int kvm_put_mp_state(CPUState *env)
 219 {
 220     struct kvm_mp_state mp_state = { .mp_state = env->mp_state };
 221
 222     return kvm_vcpu_ioctl(env, KVM_SET_MP_STATE, &mp_state);
 223 }
 224
 225 int kvm_get_mp_state(CPUState *env)
 226 {
 227     struct kvm_mp_state mp_state;
 228     int ret;
 229
 230     ret = kvm_vcpu_ioctl(env, KVM_GET_MP_STATE, &mp_state);
 231     if (ret < 0) {
 232         return ret;
 233     }
 234     env->mp_state = mp_state.mp_state;
 235     return 0;
 236 }
 237
 238 /*
 239  * dirty pages logging control
 240  */
 241 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr,
 242                                       ram_addr_t size, int flags, int mask)
 243 {
 244     KVMState *s = kvm_state;
 245     KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
 246     int old_flags;
 247
 248     if (mem == NULL)  {
 249             fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
 250                     TARGET_FMT_plx "\n", __func__, phys_addr,
 251                     (target_phys_addr_t)(phys_addr + size - 1));
 252             return -EINVAL;
 253     }
 254
 255     old_flags = mem->flags;
 256
 257     flags = (mem->flags & ~mask) | flags;
 258     mem->flags = flags;
 259
 260     /* If nothing changed effectively, no need to issue ioctl */
 261     if (s->migration_log) {
 262         flags |= KVM_MEM_LOG_DIRTY_PAGES;
 263     }
 264     if (flags == old_flags) {
 265             return 0;
 266     }
 267
 268     return kvm_set_user_memory_region(s, mem);
 269 }
 270
 271 int kvm_log_start(target_phys_addr_t phys_addr, ram_addr_t size)
 272 {
 273         return kvm_dirty_pages_log_change(phys_addr, size,
 274                                           KVM_MEM_LOG_DIRTY_PAGES,
 275                                           KVM_MEM_LOG_DIRTY_PAGES);
 276 }
 277
 278 int kvm_log_stop(target_phys_addr_t phys_addr, ram_addr_t size)
 279 {
 280         return kvm_dirty_pages_log_change(phys_addr, size,
 281                                           0,
 282                                           KVM_MEM_LOG_DIRTY_PAGES);
 283 }
 284
 285 int kvm_set_migration_log(int enable)
 286 {
 287     KVMState *s = kvm_state;
 288     KVMSlot *mem;
 289     int i, err;
 290
 291     s->migration_log = enable;
 292
 293     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
 294         mem = &s->slots[i];
 295
 296         if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
 297             continue;
 298         }
 299         err = kvm_set_user_memory_region(s, mem);
 300         if (err) {
 301             return err;
 302         }
 303     }
 304     return 0;
 305 }
 306
 307 static int test_le_bit(unsigned long nr, unsigned char *addr)
 308 {
 309     return (addr[nr >> 3] >> (nr & 7)) & 1;
 310 }
 311
 312 /**
 313  * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
 314  * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
 315  * This means all bits are set to dirty.
 316  *
 317  * @start_add: start of logged region.
 318  * @end_addr: end of logged region.
 319  */
 320 int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr,
 321                                    target_phys_addr_t end_addr)
 322 {
 323     KVMState *s = kvm_state;
 324     unsigned long size, allocated_size = 0;
 325     target_phys_addr_t phys_addr;
 326     ram_addr_t addr;
 327     KVMDirtyLog d;
 328     KVMSlot *mem;
 329     int ret = 0;
 330     int r;
 331
 332     d.dirty_bitmap = NULL;
 333     while (start_addr < end_addr) {
 334         mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
 335         if (mem == NULL) {
 336             break;
 337         }
 338
 339         /* We didn't activate dirty logging? Don't care then. */
 340         if(!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES)) {
 341             continue;
 342         }
 343
 344         size = ((mem->memory_size >> TARGET_PAGE_BITS) + 7) / 8;
 345         if (!d.dirty_bitmap) {
 346             d.dirty_bitmap = qemu_malloc(size);
 347         } else if (size > allocated_size) {
 348             d.dirty_bitmap = qemu_realloc(d.dirty_bitmap, size);
 349         }
 350         allocated_size = size;
 351         memset(d.dirty_bitmap, 0, allocated_size);
 352
 353         d.slot = mem->slot;
 354
 355         r = kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d);
 356         if (r == -EINVAL) {
 357             dprintf("ioctl failed %d\n", errno);
 358             ret = -1;
 359             break;
 360         }
 361
 362         for (phys_addr = mem->start_addr, addr = mem->phys_offset;
 363              phys_addr < mem->start_addr + mem->memory_size;
 364              phys_addr += TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
 365             unsigned char *bitmap = (unsigned char *)d.dirty_bitmap;
 366             unsigned nr = (phys_addr - mem->start_addr) >> TARGET_PAGE_BITS;
 367
 368             if (test_le_bit(nr, bitmap)) {
 369                 cpu_physical_memory_set_dirty(addr);
 370             } else if (r < 0) {
 371                 /* When our KVM implementation doesn't know about dirty logging
 372                  * we can just assume it's always dirty and be fine. */
 373                 cpu_physical_memory_set_dirty(addr);
 374             }
 375         }
 376         start_addr = phys_addr;
 377     }
 378     qemu_free(d.dirty_bitmap);
 379
 380     return ret;
 381 }
 382
 383 int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 384 {
 385     int ret = -ENOSYS;
 386 #ifdef KVM_CAP_COALESCED_MMIO
 387     KVMState *s = kvm_state;
 388
 389     if (s->coalesced_mmio) {
 390         struct kvm_coalesced_mmio_zone zone;
 391
 392         zone.addr = start;
 393         zone.size = size;
 394
 395         ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
 396     }
 397 #endif
 398
 399     return ret;
 400 }
 401
 402 int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 403 {
 404     int ret = -ENOSYS;
 405 #ifdef KVM_CAP_COALESCED_MMIO
 406     KVMState *s = kvm_state;
 407
 408     if (s->coalesced_mmio) {
 409         struct kvm_coalesced_mmio_zone zone;
 410
 411         zone.addr = start;
 412         zone.size = size;
 413
 414         ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
 415     }
 416 #endif
 417
 418     return ret;
 419 }
 420
 421 int kvm_check_extension(KVMState *s, unsigned int extension)
 422 {
 423     int ret;
 424
 425     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
 426     if (ret < 0) {
 427         ret = 0;
 428     }
 429
 430     return ret;
 431 }
 432
 433 int kvm_init(int smp_cpus)
 434 {
 435     static const char upgrade_note[] =
 436         "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
 437         "(see http://sourceforge.net/projects/kvm).\n";
 438     KVMState *s;
 439     int ret;
 440     int i;
 441
 442     if (smp_cpus > 1) {
 443         fprintf(stderr, "No SMP KVM support, use '-smp 1'\n");
 444         return -EINVAL;
 445     }
 446
 447     s = qemu_mallocz(sizeof(KVMState));
 448
 449 #ifdef KVM_CAP_SET_GUEST_DEBUG
 450     TAILQ_INIT(&s->kvm_sw_breakpoints);
 451 #endif
 452     for (i = 0; i < ARRAY_SIZE(s->slots); i++)
 453         s->slots[i].slot = i;
 454
 455     s->vmfd = -1;
 456     s->fd = open("/dev/kvm", O_RDWR);
 457     if (s->fd == -1) {
 458         fprintf(stderr, "Could not access KVM kernel module: %m\n");
 459         ret = -errno;
 460         goto err;
 461     }
 462
 463     ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
 464     if (ret < KVM_API_VERSION) {
 465         if (ret > 0)
 466             ret = -EINVAL;
 467         fprintf(stderr, "kvm version too old\n");
 468         goto err;
 469     }
 470
 471     if (ret > KVM_API_VERSION) {
 472         ret = -EINVAL;
 473         fprintf(stderr, "kvm version not supported\n");
 474         goto err;
 475     }
 476
 477     s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
 478     if (s->vmfd < 0)
 479         goto err;
 480
 481     /* initially, KVM allocated its own memory and we had to jump through
 482      * hooks to make phys_ram_base point to this.  Modern versions of KVM
 483      * just use a user allocated buffer so we can use regular pages
 484      * unmodified.  Make sure we have a sufficiently modern version of KVM.
 485      */
 486     if (!kvm_check_extension(s, KVM_CAP_USER_MEMORY)) {
 487         ret = -EINVAL;
 488         fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
 489                 upgrade_note);
 490         goto err;
 491     }
 492
 493     /* There was a nasty bug in < kvm-80 that prevents memory slots from being
 494      * destroyed properly.  Since we rely on this capability, refuse to work
 495      * with any kernel without this capability. */
 496     if (!kvm_check_extension(s, KVM_CAP_DESTROY_MEMORY_REGION_WORKS)) {
 497         ret = -EINVAL;
 498
 499         fprintf(stderr,
 500                 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
 501                 upgrade_note);
 502         goto err;
 503     }
 504
 505 #ifdef KVM_CAP_COALESCED_MMIO
 506     s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
 507 #else
 508     s->coalesced_mmio = 0;
 509 #endif
 510
 511     s->broken_set_mem_region = 1;
 512 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
 513     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
 514     if (ret > 0) {
 515         s->broken_set_mem_region = 0;
 516     }
 517 #endif
 518
 519     ret = kvm_arch_init(s, smp_cpus);
 520     if (ret < 0)
 521         goto err;
 522
 523     kvm_state = s;
 524
 525     return 0;
 526
 527 err:
 528     if (s) {
 529         if (s->vmfd != -1)
 530             close(s->vmfd);
 531         if (s->fd != -1)
 532             close(s->fd);
 533     }
 534     qemu_free(s);
 535
 536     return ret;
 537 }
 538
 539 static int kvm_handle_io(CPUState *env, uint16_t port, void *data,
 540                          int direction, int size, uint32_t count)
 541 {
 542     int i;
 543     uint8_t *ptr = data;
 544
 545     for (i = 0; i < count; i++) {
 546         if (direction == KVM_EXIT_IO_IN) {
 547             switch (size) {
 548             case 1:
 549                 stb_p(ptr, cpu_inb(env, port));
 550                 break;
 551             case 2:
 552                 stw_p(ptr, cpu_inw(env, port));
 553                 break;
 554             case 4:
 555                 stl_p(ptr, cpu_inl(env, port));
 556                 break;
 557             }
 558         } else {
 559             switch (size) {
 560             case 1:
 561                 cpu_outb(env, port, ldub_p(ptr));
 562                 break;
 563             case 2:
 564                 cpu_outw(env, port, lduw_p(ptr));
 565                 break;
 566             case 4:
 567                 cpu_outl(env, port, ldl_p(ptr));
 568                 break;
 569             }
 570         }
 571
 572         ptr += size;
 573     }
 574
 575     return 1;
 576 }
 577
 578 static void kvm_run_coalesced_mmio(CPUState *env, struct kvm_run *run)
 579 {
 580 #ifdef KVM_CAP_COALESCED_MMIO
 581     KVMState *s = kvm_state;
 582     if (s->coalesced_mmio) {
 583         struct kvm_coalesced_mmio_ring *ring;
 584
 585         ring = (void *)run + (s->coalesced_mmio * TARGET_PAGE_SIZE);
 586         while (ring->first != ring->last) {
 587             struct kvm_coalesced_mmio *ent;
 588
 589             ent = &ring->coalesced_mmio[ring->first];
 590
 591             cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
 592             /* FIXME smp_wmb() */
 593             ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
 594         }
 595     }
 596 #endif
 597 }
 598
 599 int kvm_cpu_exec(CPUState *env)
 600 {
 601     struct kvm_run *run = env->kvm_run;
 602     int ret;
 603
 604     dprintf("kvm_cpu_exec()\n");
 605
 606     do {
 607         if (env->exit_request) {
 608             dprintf("interrupt exit requested\n");
 609             ret = 0;
 610             break;
 611         }
 612
 613         kvm_arch_pre_run(env, run);
 614         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
 615         kvm_arch_post_run(env, run);
 616
 617         if (ret == -EINTR || ret == -EAGAIN) {
 618             dprintf("io window exit\n");
 619             ret = 0;
 620             break;
 621         }
 622
 623         if (ret < 0) {
 624             dprintf("kvm run failed %s\n", strerror(-ret));
 625             abort();
 626         }
 627
 628         kvm_run_coalesced_mmio(env, run);
 629
 630         ret = 0; /* exit loop */
 631         switch (run->exit_reason) {
 632         case KVM_EXIT_IO:
 633             dprintf("handle_io\n");
 634             ret = kvm_handle_io(env, run->io.port,
 635                                 (uint8_t *)run + run->io.data_offset,
 636                                 run->io.direction,
 637                                 run->io.size,
 638                                 run->io.count);
 639             break;
 640         case KVM_EXIT_MMIO:
 641             dprintf("handle_mmio\n");
 642             cpu_physical_memory_rw(run->mmio.phys_addr,
 643                                    run->mmio.data,
 644                                    run->mmio.len,
 645                                    run->mmio.is_write);
 646             ret = 1;
 647             break;
 648         case KVM_EXIT_IRQ_WINDOW_OPEN:
 649             dprintf("irq_window_open\n");
 650             break;
 651         case KVM_EXIT_SHUTDOWN:
 652             dprintf("shutdown\n");
 653             qemu_system_reset_request();
 654             ret = 1;
 655             break;
 656         case KVM_EXIT_UNKNOWN:
 657             dprintf("kvm_exit_unknown\n");
 658             break;
 659         case KVM_EXIT_FAIL_ENTRY:
 660             dprintf("kvm_exit_fail_entry\n");
 661             break;
 662         case KVM_EXIT_EXCEPTION:
 663             dprintf("kvm_exit_exception\n");
 664             break;
 665         case KVM_EXIT_DEBUG:
 666             dprintf("kvm_exit_debug\n");
 667 #ifdef KVM_CAP_SET_GUEST_DEBUG
 668             if (kvm_arch_debug(&run->debug.arch)) {
 669                 gdb_set_stop_cpu(env);
 670                 vm_stop(EXCP_DEBUG);
 671                 env->exception_index = EXCP_DEBUG;
 672                 return 0;
 673             }
 674             /* re-enter, this exception was guest-internal */
 675             ret = 1;
 676 #endif /* KVM_CAP_SET_GUEST_DEBUG */
 677             break;
 678         default:
 679             dprintf("kvm_arch_handle_exit\n");
 680             ret = kvm_arch_handle_exit(env, run);
 681             break;
 682         }
 683     } while (ret > 0);
 684
 685     if (env->exit_request) {
 686         env->exit_request = 0;
 687         env->exception_index = EXCP_INTERRUPT;
 688     }
 689
 690     return ret;
 691 }
 692
 693 void kvm_set_phys_mem(target_phys_addr_t start_addr,
 694                       ram_addr_t size,
 695                       ram_addr_t phys_offset)
 696 {
 697     KVMState *s = kvm_state;
 698     ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK;
 699     KVMSlot *mem, old;
 700     int err;
 701
 702     if (start_addr & ~TARGET_PAGE_MASK) {
 703         if (flags >= IO_MEM_UNASSIGNED) {
 704             if (!kvm_lookup_overlapping_slot(s, start_addr,
 705                                              start_addr + size)) {
 706                 return;
 707             }
 708             fprintf(stderr, "Unaligned split of a KVM memory slot\n");
 709         } else {
 710             fprintf(stderr, "Only page-aligned memory slots supported\n");
 711         }
 712         abort();
 713     }
 714
 715     /* KVM does not support read-only slots */
 716     phys_offset &= ~IO_MEM_ROM;
 717
 718     while (1) {
 719         mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
 720         if (!mem) {
 721             break;
 722         }
 723
 724         if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr &&
 725             (start_addr + size <= mem->start_addr + mem->memory_size) &&
 726             (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) {
 727             /* The new slot fits into the existing one and comes with
 728              * identical parameters - nothing to be done. */
 729             return;
 730         }
 731
 732         old = *mem;
 733
 734         /* unregister the overlapping slot */
 735         mem->memory_size = 0;
 736         err = kvm_set_user_memory_region(s, mem);
 737         if (err) {
 738             fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
 739                     __func__, strerror(-err));
 740             abort();
 741         }
 742
 743         /* Workaround for older KVM versions: we can't join slots, even not by
 744          * unregistering the previous ones and then registering the larger
 745          * slot. We have to maintain the existing fragmentation. Sigh.
 746          *
 747          * This workaround assumes that the new slot starts at the same
 748          * address as the first existing one. If not or if some overlapping
 749          * slot comes around later, we will fail (not seen in practice so far)
 750          * - and actually require a recent KVM version. */
 751         if (s->broken_set_mem_region &&
 752             old.start_addr == start_addr && old.memory_size < size &&
 753             flags < IO_MEM_UNASSIGNED) {
 754             mem = kvm_alloc_slot(s);
 755             mem->memory_size = old.memory_size;
 756             mem->start_addr = old.start_addr;
 757             mem->phys_offset = old.phys_offset;
 758             mem->flags = 0;
 759
 760             err = kvm_set_user_memory_region(s, mem);
 761             if (err) {
 762                 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
 763                         strerror(-err));
 764                 abort();
 765             }
 766
 767             start_addr += old.memory_size;
 768             phys_offset += old.memory_size;
 769             size -= old.memory_size;
 770             continue;
 771         }
 772
 773         /* register prefix slot */
 774         if (old.start_addr < start_addr) {
 775             mem = kvm_alloc_slot(s);
 776             mem->memory_size = start_addr - old.start_addr;
 777             mem->start_addr = old.start_addr;
 778             mem->phys_offset = old.phys_offset;
 779             mem->flags = 0;
 780
 781             err = kvm_set_user_memory_region(s, mem);
 782             if (err) {
 783                 fprintf(stderr, "%s: error registering prefix slot: %s\n",
 784                         __func__, strerror(-err));
 785                 abort();
 786             }
 787         }
 788
 789         /* register suffix slot */
 790         if (old.start_addr + old.memory_size > start_addr + size) {
 791             ram_addr_t size_delta;
 792
 793             mem = kvm_alloc_slot(s);
 794             mem->start_addr = start_addr + size;
 795             size_delta = mem->start_addr - old.start_addr;
 796             mem->memory_size = old.memory_size - size_delta;
 797             mem->phys_offset = old.phys_offset + size_delta;
 798             mem->flags = 0;
 799
 800             err = kvm_set_user_memory_region(s, mem);
 801             if (err) {
 802                 fprintf(stderr, "%s: error registering suffix slot: %s\n",
 803                         __func__, strerror(-err));
 804                 abort();
 805             }
 806         }
 807     }
 808
 809     /* in case the KVM bug workaround already "consumed" the new slot */
 810     if (!size)
 811         return;
 812
 813     /* KVM does not need to know about this memory */
 814     if (flags >= IO_MEM_UNASSIGNED)
 815         return;
 816
 817     mem = kvm_alloc_slot(s);
 818     mem->memory_size = size;
 819     mem->start_addr = start_addr;
 820     mem->phys_offset = phys_offset;
 821     mem->flags = 0;
 822
 823     err = kvm_set_user_memory_region(s, mem);
 824     if (err) {
 825         fprintf(stderr, "%s: error registering slot: %s\n", __func__,
 826                 strerror(-err));
 827         abort();
 828     }
 829 }
 830
 831 int kvm_ioctl(KVMState *s, int type, ...)
 832 {
 833     int ret;
 834     void *arg;
 835     va_list ap;
 836
 837     va_start(ap, type);
 838     arg = va_arg(ap, void *);
 839     va_end(ap);
 840
 841     ret = ioctl(s->fd, type, arg);
 842     if (ret == -1)
 843         ret = -errno;
 844
 845     return ret;
 846 }
 847
 848 int kvm_vm_ioctl(KVMState *s, int type, ...)
 849 {
 850     int ret;
 851     void *arg;
 852     va_list ap;
 853
 854     va_start(ap, type);
 855     arg = va_arg(ap, void *);
 856     va_end(ap);
 857
 858     ret = ioctl(s->vmfd, type, arg);
 859     if (ret == -1)
 860         ret = -errno;
 861
 862     return ret;
 863 }
 864
 865 int kvm_vcpu_ioctl(CPUState *env, int type, ...)
 866 {
 867     int ret;
 868     void *arg;
 869     va_list ap;
 870
 871     va_start(ap, type);
 872     arg = va_arg(ap, void *);
 873     va_end(ap);
 874
 875     ret = ioctl(env->kvm_fd, type, arg);
 876     if (ret == -1)
 877         ret = -errno;
 878
 879     return ret;
 880 }
 881
 882 int kvm_has_sync_mmu(void)
 883 {
 884 #ifdef KVM_CAP_SYNC_MMU
 885     KVMState *s = kvm_state;
 886
 887     return kvm_check_extension(s, KVM_CAP_SYNC_MMU);
 888 #else
 889     return 0;
 890 #endif
 891 }
 892
 893 void kvm_setup_guest_memory(void *start, size_t size)
 894 {
 895     if (!kvm_has_sync_mmu()) {
 896 #ifdef MADV_DONTFORK
 897         int ret = madvise(start, size, MADV_DONTFORK);
 898
 899         if (ret) {
 900             perror("madvice");
 901             exit(1);
 902         }
 903 #else
 904         fprintf(stderr,
 905                 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
 906         exit(1);
 907 #endif
 908     }
 909 }
 910
 911 #ifdef KVM_CAP_SET_GUEST_DEBUG
 912 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
 913                                                  target_ulong pc)
 914 {
 915     struct kvm_sw_breakpoint *bp;
 916
 917     TAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) {
 918         if (bp->pc == pc)
 919             return bp;
 920     }
 921     return NULL;
 922 }
 923
 924 int kvm_sw_breakpoints_active(CPUState *env)
 925 {
 926     return !TAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints);
 927 }
 928
 929 struct kvm_set_guest_debug_data {
 930     struct kvm_guest_debug dbg;
 931     CPUState *env;
 932     int err;
 933 };
 934
 935 static void kvm_invoke_set_guest_debug(void *data)
 936 {
 937     struct kvm_set_guest_debug_data *dbg_data = data;
 938     dbg_data->err = kvm_vcpu_ioctl(dbg_data->env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg);
 939 }
 940
 941 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 942 {
 943     struct kvm_set_guest_debug_data data;
 944
 945     data.dbg.control = 0;
 946     if (env->singlestep_enabled)
 947         data.dbg.control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
 948
 949     kvm_arch_update_guest_debug(env, &data.dbg);
 950     data.dbg.control |= reinject_trap;
 951     data.env = env;
 952
 953     on_vcpu(env, kvm_invoke_set_guest_debug, &data);
 954     return data.err;
 955 }
 956
 957 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 958                           target_ulong len, int type)
 959 {
 960     struct kvm_sw_breakpoint *bp;
 961     CPUState *env;
 962     int err;
 963
 964     if (type == GDB_BREAKPOINT_SW) {
 965         bp = kvm_find_sw_breakpoint(current_env, addr);
 966         if (bp) {
 967             bp->use_count++;
 968             return 0;
 969         }
 970
 971         bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint));
 972         if (!bp)
 973             return -ENOMEM;
 974
 975         bp->pc = addr;
 976         bp->use_count = 1;
 977         err = kvm_arch_insert_sw_breakpoint(current_env, bp);
 978         if (err) {
 979             free(bp);
 980             return err;
 981         }
 982
 983         TAILQ_INSERT_HEAD(&current_env->kvm_state->kvm_sw_breakpoints,
 984                           bp, entry);
 985     } else {
 986         err = kvm_arch_insert_hw_breakpoint(addr, len, type);
 987         if (err)
 988             return err;
 989     }
 990
 991     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 992         err = kvm_update_guest_debug(env, 0);
 993         if (err)
 994             return err;
 995     }
 996     return 0;
 997 }
 998
 999 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
1000                           target_ulong len, int type)
1001 {
1002     struct kvm_sw_breakpoint *bp;
1003     CPUState *env;
1004     int err;
1005
1006     if (type == GDB_BREAKPOINT_SW) {
1007         bp = kvm_find_sw_breakpoint(current_env, addr);
1008         if (!bp)
1009             return -ENOENT;
1010
1011         if (bp->use_count > 1) {
1012             bp->use_count--;
1013             return 0;
1014         }
1015
1016         err = kvm_arch_remove_sw_breakpoint(current_env, bp);
1017         if (err)
1018             return err;
1019
1020         TAILQ_REMOVE(&current_env->kvm_state->kvm_sw_breakpoints, bp, entry);
1021         qemu_free(bp);
1022     } else {
1023         err = kvm_arch_remove_hw_breakpoint(addr, len, type);
1024         if (err)
1025             return err;
1026     }
1027
1028     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1029         err = kvm_update_guest_debug(env, 0);
1030         if (err)
1031             return err;
1032     }
1033     return 0;
1034 }
1035
1036 void kvm_remove_all_breakpoints(CPUState *current_env)
1037 {
1038     struct kvm_sw_breakpoint *bp, *next;
1039     KVMState *s = current_env->kvm_state;
1040     CPUState *env;
1041
1042     TAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
1043         if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) {
1044             /* Try harder to find a CPU that currently sees the breakpoint. */
1045             for (env = first_cpu; env != NULL; env = env->next_cpu) {
1046                 if (kvm_arch_remove_sw_breakpoint(env, bp) == 0)
1047                     break;
1048             }
1049         }
1050     }
1051     kvm_arch_remove_all_hw_breakpoints();
1052
1053     for (env = first_cpu; env != NULL; env = env->next_cpu)
1054         kvm_update_guest_debug(env, 0);
1055 }
1056
1057 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1058
1059 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
1060 {
1061     return -EINVAL;
1062 }
1063
1064 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
1065                           target_ulong len, int type)
1066 {
1067     return -EINVAL;
1068 }
1069
1070 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
1071                           target_ulong len, int type)
1072 {
1073     return -EINVAL;
1074 }
1075
1076 void kvm_remove_all_breakpoints(CPUState *current_env)
1077 {
1078 }
1079 #endif /* !KVM_CAP_SET_GUEST_DEBUG */