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