kvm-all.c

   1 /*
   2  * QEMU KVM support
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *           Red Hat, Inc. 2008
   6  *
   7  * Authors:
   8  *  Anthony Liguori   <aliguori@us.ibm.com>
   9  *  Glauber Costa     <gcosta@redhat.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  12  * See the COPYING file in the top-level directory.
  13  *
  14  */
  15
  16 #include <sys/types.h>
  17 #include <sys/ioctl.h>
  18 #include <sys/mman.h>
  19 #include <stdarg.h>
  20
  21 #include <linux/kvm.h>
  22
  23 #include "qemu-common.h"
  24 #include "sysemu.h"
  25 #include "hw/hw.h"
  26 #include "gdbstub.h"
  27 #include "kvm.h"
  28
  29 #ifdef KVM_UPSTREAM
  30 /* KVM uses PAGE_SIZE in it's definition of COALESCED_MMIO_MAX */
  31 #define PAGE_SIZE TARGET_PAGE_SIZE
  32
  33 //#define DEBUG_KVM
  34
  35 #ifdef DEBUG_KVM
  36 #define dprintf(fmt, ...) \
  37     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  38 #else
  39 #define dprintf(fmt, ...) \
  40     do { } while (0)
  41 #endif
  42
  43 typedef struct KVMSlot
  44 {
  45     target_phys_addr_t start_addr;
  46     ram_addr_t memory_size;
  47     ram_addr_t phys_offset;
  48     int slot;
  49     int flags;
  50 } KVMSlot;
  51
  52 typedef struct kvm_dirty_log KVMDirtyLog;
  53
  54 int kvm_allowed = 0;
  55
  56 struct KVMState
  57 {
  58     KVMSlot slots[32];
  59     int fd;
  60     int vmfd;
  61     int coalesced_mmio;
  62     int broken_set_mem_region;
  63     int migration_log;
  64 #ifdef KVM_CAP_SET_GUEST_DEBUG
  65     struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
  66 #endif
  67     int irqchip_in_kernel;
  68     int pit_in_kernel;
  69 };
  70
  71 static KVMState *kvm_state;
  72
  73 static KVMSlot *kvm_alloc_slot(KVMState *s)
  74 {
  75     int i;
  76
  77     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
  78         /* KVM private memory slots */
  79         if (i >= 8 && i < 12)
  80             continue;
  81         if (s->slots[i].memory_size == 0)
  82             return &s->slots[i];
  83     }
  84
  85     fprintf(stderr, "%s: no free slot available\n", __func__);
  86     abort();
  87 }
  88
  89 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
  90                                          target_phys_addr_t start_addr,
  91                                          target_phys_addr_t end_addr)
  92 {
  93     int i;
  94
  95     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
  96         KVMSlot *mem = &s->slots[i];
  97
  98         if (start_addr == mem->start_addr &&
  99             end_addr == mem->start_addr + mem->memory_size) {
 100             return mem;
 101         }
 102     }
 103
 104     return NULL;
 105 }
 106
 107 /*
 108  * Find overlapping slot with lowest start address
 109  */
 110 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
 111                                             target_phys_addr_t start_addr,
 112                                             target_phys_addr_t end_addr)
 113 {
 114     KVMSlot *found = NULL;
 115     int i;
 116
 117     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
 118         KVMSlot *mem = &s->slots[i];
 119
 120         if (mem->memory_size == 0 ||
 121             (found && found->start_addr < mem->start_addr)) {
 122             continue;
 123         }
 124
 125         if (end_addr > mem->start_addr &&
 126             start_addr < mem->start_addr + mem->memory_size) {
 127             found = mem;
 128         }
 129     }
 130
 131     return found;
 132 }
 133
 134 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
 135 {
 136     struct kvm_userspace_memory_region mem;
 137
 138     mem.slot = slot->slot;
 139     mem.guest_phys_addr = slot->start_addr;
 140     mem.memory_size = slot->memory_size;
 141     mem.userspace_addr = (unsigned long)qemu_get_ram_ptr(slot->phys_offset);
 142     mem.flags = slot->flags;
 143     if (s->migration_log) {
 144         mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
 145     }
 146     return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
 147 }
 148
 149 static void kvm_reset_vcpu(void *opaque)
 150 {
 151     CPUState *env = opaque;
 152
 153     if (kvm_arch_put_registers(env)) {
 154         fprintf(stderr, "Fatal: kvm vcpu reset failed\n");
 155         abort();
 156     }
 157 }
 158 #endif
 159
 160 int kvm_irqchip_in_kernel(void)
 161 {
 162     return kvm_state->irqchip_in_kernel;
 163 }
 164
 165 #ifdef KVM_UPSTREAM
 166 int kvm_pit_in_kernel(void)
 167 {
 168     return kvm_state->pit_in_kernel;
 169 }
 170
 171
 172 int kvm_init_vcpu(CPUState *env)
 173 {
 174     KVMState *s = kvm_state;
 175     long mmap_size;
 176     int ret;
 177
 178     dprintf("kvm_init_vcpu\n");
 179
 180     ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index);
 181     if (ret < 0) {
 182         dprintf("kvm_create_vcpu failed\n");
 183         goto err;
 184     }
 185
 186     env->kvm_fd = ret;
 187     env->kvm_state = s;
 188
 189     mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
 190     if (mmap_size < 0) {
 191         dprintf("KVM_GET_VCPU_MMAP_SIZE failed\n");
 192         goto err;
 193     }
 194
 195     env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
 196                         env->kvm_fd, 0);
 197     if (env->kvm_run == MAP_FAILED) {
 198         ret = -errno;
 199         dprintf("mmap'ing vcpu state failed\n");
 200         goto err;
 201     }
 202
 203     ret = kvm_arch_init_vcpu(env);
 204     if (ret == 0) {
 205         qemu_register_reset(kvm_reset_vcpu, env);
 206         ret = kvm_arch_put_registers(env);
 207     }
 208 err:
 209     return ret;
 210 }
 211
 212 int kvm_put_mp_state(CPUState *env)
 213 {
 214     struct kvm_mp_state mp_state = { .mp_state = env->mp_state };
 215
 216     return kvm_vcpu_ioctl(env, KVM_SET_MP_STATE, &mp_state);
 217 }
 218
 219 int kvm_get_mp_state(CPUState *env)
 220 {
 221     struct kvm_mp_state mp_state;
 222     int ret;
 223
 224     ret = kvm_vcpu_ioctl(env, KVM_GET_MP_STATE, &mp_state);
 225     if (ret < 0) {
 226         return ret;
 227     }
 228     env->mp_state = mp_state.mp_state;
 229     return 0;
 230 }
 231
 232 /*
 233  * dirty pages logging control
 234  */
 235 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr,
 236                                       ram_addr_t size, int flags, int mask)
 237 {
 238     KVMState *s = kvm_state;
 239     KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
 240     int old_flags;
 241
 242     if (mem == NULL)  {
 243             fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
 244                     TARGET_FMT_plx "\n", __func__, phys_addr,
 245                     (target_phys_addr_t)(phys_addr + size - 1));
 246             return -EINVAL;
 247     }
 248
 249     old_flags = mem->flags;
 250
 251     flags = (mem->flags & ~mask) | flags;
 252     mem->flags = flags;
 253
 254     /* If nothing changed effectively, no need to issue ioctl */
 255     if (s->migration_log) {
 256         flags |= KVM_MEM_LOG_DIRTY_PAGES;
 257     }
 258     if (flags == old_flags) {
 259             return 0;
 260     }
 261
 262     return kvm_set_user_memory_region(s, mem);
 263 }
 264
 265 int kvm_log_start(target_phys_addr_t phys_addr, ram_addr_t size)
 266 {
 267         return kvm_dirty_pages_log_change(phys_addr, size,
 268                                           KVM_MEM_LOG_DIRTY_PAGES,
 269                                           KVM_MEM_LOG_DIRTY_PAGES);
 270 }
 271
 272 int kvm_log_stop(target_phys_addr_t phys_addr, ram_addr_t size)
 273 {
 274         return kvm_dirty_pages_log_change(phys_addr, size,
 275                                           0,
 276                                           KVM_MEM_LOG_DIRTY_PAGES);
 277 }
 278
 279 int kvm_set_migration_log(int enable)
 280 {
 281     KVMState *s = kvm_state;
 282     KVMSlot *mem;
 283     int i, err;
 284
 285     s->migration_log = enable;
 286
 287     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
 288         mem = &s->slots[i];
 289
 290         if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
 291             continue;
 292         }
 293         err = kvm_set_user_memory_region(s, mem);
 294         if (err) {
 295             return err;
 296         }
 297     }
 298     return 0;
 299 }
 300
 301 static int test_le_bit(unsigned long nr, unsigned char *addr)
 302 {
 303     return (addr[nr >> 3] >> (nr & 7)) & 1;
 304 }
 305
 306 /**
 307  * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
 308  * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
 309  * This means all bits are set to dirty.
 310  *
 311  * @start_add: start of logged region.
 312  * @end_addr: end of logged region.
 313  */
 314 int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr,
 315                                    target_phys_addr_t end_addr)
 316 {
 317     KVMState *s = kvm_state;
 318     unsigned long size, allocated_size = 0;
 319     target_phys_addr_t phys_addr;
 320     ram_addr_t addr;
 321     KVMDirtyLog d;
 322     KVMSlot *mem;
 323     int ret = 0;
 324
 325     d.dirty_bitmap = NULL;
 326     while (start_addr < end_addr) {
 327         mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
 328         if (mem == NULL) {
 329             break;
 330         }
 331
 332         size = ((mem->memory_size >> TARGET_PAGE_BITS) + 7) / 8;
 333         if (!d.dirty_bitmap) {
 334             d.dirty_bitmap = qemu_malloc(size);
 335         } else if (size > allocated_size) {
 336             d.dirty_bitmap = qemu_realloc(d.dirty_bitmap, size);
 337         }
 338         allocated_size = size;
 339         memset(d.dirty_bitmap, 0, allocated_size);
 340
 341         d.slot = mem->slot;
 342
 343         if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
 344             dprintf("ioctl failed %d\n", errno);
 345             ret = -1;
 346             break;
 347         }
 348
 349         for (phys_addr = mem->start_addr, addr = mem->phys_offset;
 350              phys_addr < mem->start_addr + mem->memory_size;
 351              phys_addr += TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
 352             unsigned char *bitmap = (unsigned char *)d.dirty_bitmap;
 353             unsigned nr = (phys_addr - mem->start_addr) >> TARGET_PAGE_BITS;
 354
 355             if (test_le_bit(nr, bitmap)) {
 356                 cpu_physical_memory_set_dirty(addr);
 357             }
 358         }
 359         start_addr = phys_addr;
 360     }
 361     qemu_free(d.dirty_bitmap);
 362
 363     return ret;
 364 }
 365 #endif
 366
 367 int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 368 {
 369     int ret = -ENOSYS;
 370 #ifdef KVM_CAP_COALESCED_MMIO
 371     KVMState *s = kvm_state;
 372
 373     if (s->coalesced_mmio) {
 374         struct kvm_coalesced_mmio_zone zone;
 375
 376         zone.addr = start;
 377         zone.size = size;
 378
 379         ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
 380     }
 381 #endif
 382
 383     return ret;
 384 }
 385
 386 int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 387 {
 388     int ret = -ENOSYS;
 389 #ifdef KVM_CAP_COALESCED_MMIO
 390     KVMState *s = kvm_state;
 391
 392     if (s->coalesced_mmio) {
 393         struct kvm_coalesced_mmio_zone zone;
 394
 395         zone.addr = start;
 396         zone.size = size;
 397
 398         ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
 399     }
 400 #endif
 401
 402     return ret;
 403 }
 404
 405 int kvm_check_extension(KVMState *s, unsigned int extension)
 406 {
 407     int ret;
 408
 409     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
 410     if (ret < 0) {
 411         ret = 0;
 412     }
 413
 414     return ret;
 415 }
 416 #ifdef KVM_UPSTREAM
 417
 418 int kvm_init(int smp_cpus)
 419 {
 420     static const char upgrade_note[] =
 421         "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
 422         "(see http://sourceforge.net/projects/kvm).\n";
 423     KVMState *s;
 424     int ret;
 425     int i;
 426
 427     if (smp_cpus > 1) {
 428         fprintf(stderr, "No SMP KVM support, use '-smp 1'\n");
 429         return -EINVAL;
 430     }
 431
 432     s = qemu_mallocz(sizeof(KVMState));
 433
 434 #ifdef KVM_CAP_SET_GUEST_DEBUG
 435     QTAILQ_INIT(&s->kvm_sw_breakpoints);
 436 #endif
 437     for (i = 0; i < ARRAY_SIZE(s->slots); i++)
 438         s->slots[i].slot = i;
 439
 440     s->vmfd = -1;
 441     s->fd = open("/dev/kvm", O_RDWR);
 442     if (s->fd == -1) {
 443         fprintf(stderr, "Could not access KVM kernel module: %m\n");
 444         ret = -errno;
 445         goto err;
 446     }
 447
 448     ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
 449     if (ret < KVM_API_VERSION) {
 450         if (ret > 0)
 451             ret = -EINVAL;
 452         fprintf(stderr, "kvm version too old\n");
 453         goto err;
 454     }
 455
 456     if (ret > KVM_API_VERSION) {
 457         ret = -EINVAL;
 458         fprintf(stderr, "kvm version not supported\n");
 459         goto err;
 460     }
 461
 462     s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
 463     if (s->vmfd < 0)
 464         goto err;
 465
 466     /* initially, KVM allocated its own memory and we had to jump through
 467      * hooks to make phys_ram_base point to this.  Modern versions of KVM
 468      * just use a user allocated buffer so we can use regular pages
 469      * unmodified.  Make sure we have a sufficiently modern version of KVM.
 470      */
 471     if (!kvm_check_extension(s, KVM_CAP_USER_MEMORY)) {
 472         ret = -EINVAL;
 473         fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
 474                 upgrade_note);
 475         goto err;
 476     }
 477
 478     /* There was a nasty bug in < kvm-80 that prevents memory slots from being
 479      * destroyed properly.  Since we rely on this capability, refuse to work
 480      * with any kernel without this capability. */
 481     if (!kvm_check_extension(s, KVM_CAP_DESTROY_MEMORY_REGION_WORKS)) {
 482         ret = -EINVAL;
 483
 484         fprintf(stderr,
 485                 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
 486                 upgrade_note);
 487         goto err;
 488     }
 489
 490 #ifdef KVM_CAP_COALESCED_MMIO
 491     s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
 492 #else
 493     s->coalesced_mmio = 0;
 494 #endif
 495
 496     s->broken_set_mem_region = 1;
 497 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
 498     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
 499     if (ret > 0) {
 500         s->broken_set_mem_region = 0;
 501     }
 502 #endif
 503
 504     ret = kvm_arch_init(s, smp_cpus);
 505     if (ret < 0)
 506         goto err;
 507
 508     kvm_state = s;
 509
 510     return 0;
 511
 512 err:
 513     if (s) {
 514         if (s->vmfd != -1)
 515             close(s->vmfd);
 516         if (s->fd != -1)
 517             close(s->fd);
 518     }
 519     qemu_free(s);
 520
 521     return ret;
 522 }
 523 #endif
 524
 525 static int kvm_handle_io(uint16_t port, void *data, int direction, int size,
 526                          uint32_t count)
 527 {
 528     int i;
 529     uint8_t *ptr = data;
 530
 531     for (i = 0; i < count; i++) {
 532         if (direction == KVM_EXIT_IO_IN) {
 533             switch (size) {
 534             case 1:
 535                 stb_p(ptr, cpu_inb(port));
 536                 break;
 537             case 2:
 538                 stw_p(ptr, cpu_inw(port));
 539                 break;
 540             case 4:
 541                 stl_p(ptr, cpu_inl(port));
 542                 break;
 543             }
 544         } else {
 545             switch (size) {
 546             case 1:
 547                 cpu_outb(port, ldub_p(ptr));
 548                 break;
 549             case 2:
 550                 cpu_outw(port, lduw_p(ptr));
 551                 break;
 552             case 4:
 553                 cpu_outl(port, ldl_p(ptr));
 554                 break;
 555             }
 556         }
 557
 558         ptr += size;
 559     }
 560
 561     return 1;
 562 }
 563
 564 #ifdef KVM_UPSTREAM
 565 static void kvm_run_coalesced_mmio(CPUState *env, struct kvm_run *run)
 566 {
 567 #ifdef KVM_CAP_COALESCED_MMIO
 568     KVMState *s = kvm_state;
 569     if (s->coalesced_mmio) {
 570         struct kvm_coalesced_mmio_ring *ring;
 571
 572         ring = (void *)run + (s->coalesced_mmio * TARGET_PAGE_SIZE);
 573         while (ring->first != ring->last) {
 574             struct kvm_coalesced_mmio *ent;
 575
 576             ent = &ring->coalesced_mmio[ring->first];
 577
 578             cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
 579             /* FIXME smp_wmb() */
 580             ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
 581         }
 582     }
 583 #endif
 584 }
 585
 586 void kvm_cpu_synchronize_state(CPUState *env)
 587 {
 588     if (!env->kvm_state->regs_modified) {
 589         kvm_arch_get_registers(env);
 590         env->kvm_state->regs_modified = 1;
 591     }
 592 }
 593
 594 int kvm_cpu_exec(CPUState *env)
 595 {
 596     struct kvm_run *run = env->kvm_run;
 597     int ret;
 598
 599     dprintf("kvm_cpu_exec()\n");
 600
 601     do {
 602         if (env->exit_request) {
 603             dprintf("interrupt exit requested\n");
 604             ret = 0;
 605             break;
 606         }
 607
 608         if (env->kvm_state->regs_modified) {
 609             kvm_arch_put_registers(env);
 610             env->kvm_state->regs_modified = 0;
 611         }
 612
 613         kvm_arch_pre_run(env, run);
 614         qemu_mutex_unlock_iothread();
 615         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
 616         qemu_mutex_lock_iothread();
 617         kvm_arch_post_run(env, run);
 618
 619         if (ret == -EINTR || ret == -EAGAIN) {
 620             dprintf("io window exit\n");
 621             ret = 0;
 622             break;
 623         }
 624
 625         if (ret < 0) {
 626             dprintf("kvm run failed %s\n", strerror(-ret));
 627             abort();
 628         }
 629
 630         kvm_run_coalesced_mmio(env, run);
 631
 632         ret = 0; /* exit loop */
 633         switch (run->exit_reason) {
 634         case KVM_EXIT_IO:
 635             dprintf("handle_io\n");
 636             ret = kvm_handle_io(run->io.port,
 637                                 (uint8_t *)run + run->io.data_offset,
 638                                 run->io.direction,
 639                                 run->io.size,
 640                                 run->io.count);
 641             break;
 642         case KVM_EXIT_MMIO:
 643             dprintf("handle_mmio\n");
 644             cpu_physical_memory_rw(run->mmio.phys_addr,
 645                                    run->mmio.data,
 646                                    run->mmio.len,
 647                                    run->mmio.is_write);
 648             ret = 1;
 649             break;
 650         case KVM_EXIT_IRQ_WINDOW_OPEN:
 651             dprintf("irq_window_open\n");
 652             break;
 653         case KVM_EXIT_SHUTDOWN:
 654             dprintf("shutdown\n");
 655             qemu_system_reset_request();
 656             ret = 1;
 657             break;
 658         case KVM_EXIT_UNKNOWN:
 659             dprintf("kvm_exit_unknown\n");
 660             break;
 661         case KVM_EXIT_FAIL_ENTRY:
 662             dprintf("kvm_exit_fail_entry\n");
 663             break;
 664         case KVM_EXIT_EXCEPTION:
 665             dprintf("kvm_exit_exception\n");
 666             break;
 667         case KVM_EXIT_DEBUG:
 668             dprintf("kvm_exit_debug\n");
 669 #ifdef KVM_CAP_SET_GUEST_DEBUG
 670             if (kvm_arch_debug(&run->debug.arch)) {
 671                 gdb_set_stop_cpu(env);
 672                 vm_stop(EXCP_DEBUG);
 673                 env->exception_index = EXCP_DEBUG;
 674                 return 0;
 675             }
 676             /* re-enter, this exception was guest-internal */
 677             ret = 1;
 678 #endif /* KVM_CAP_SET_GUEST_DEBUG */
 679             break;
 680         default:
 681             dprintf("kvm_arch_handle_exit\n");
 682             ret = kvm_arch_handle_exit(env, run);
 683             break;
 684         }
 685     } while (ret > 0);
 686
 687     if (env->exit_request) {
 688         env->exit_request = 0;
 689         env->exception_index = EXCP_INTERRUPT;
 690     }
 691
 692     return ret;
 693 }
 694
 695 void kvm_set_phys_mem(target_phys_addr_t start_addr,
 696                       ram_addr_t size,
 697                       ram_addr_t phys_offset)
 698 {
 699     KVMState *s = kvm_state;
 700     ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK;
 701     KVMSlot *mem, old;
 702     int err;
 703
 704     if (start_addr & ~TARGET_PAGE_MASK) {
 705         if (flags >= IO_MEM_UNASSIGNED) {
 706             if (!kvm_lookup_overlapping_slot(s, start_addr,
 707                                              start_addr + size)) {
 708                 return;
 709             }
 710             fprintf(stderr, "Unaligned split of a KVM memory slot\n");
 711         } else {
 712             fprintf(stderr, "Only page-aligned memory slots supported\n");
 713         }
 714         abort();
 715     }
 716
 717     /* KVM does not support read-only slots */
 718     phys_offset &= ~IO_MEM_ROM;
 719
 720     while (1) {
 721         mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
 722         if (!mem) {
 723             break;
 724         }
 725
 726         if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr &&
 727             (start_addr + size <= mem->start_addr + mem->memory_size) &&
 728             (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) {
 729             /* The new slot fits into the existing one and comes with
 730              * identical parameters - nothing to be done. */
 731             return;
 732         }
 733
 734         old = *mem;
 735
 736         /* unregister the overlapping slot */
 737         mem->memory_size = 0;
 738         err = kvm_set_user_memory_region(s, mem);
 739         if (err) {
 740             fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
 741                     __func__, strerror(-err));
 742             abort();
 743         }
 744
 745         /* Workaround for older KVM versions: we can't join slots, even not by
 746          * unregistering the previous ones and then registering the larger
 747          * slot. We have to maintain the existing fragmentation. Sigh.
 748          *
 749          * This workaround assumes that the new slot starts at the same
 750          * address as the first existing one. If not or if some overlapping
 751          * slot comes around later, we will fail (not seen in practice so far)
 752          * - and actually require a recent KVM version. */
 753         if (s->broken_set_mem_region &&
 754             old.start_addr == start_addr && old.memory_size < size &&
 755             flags < IO_MEM_UNASSIGNED) {
 756             mem = kvm_alloc_slot(s);
 757             mem->memory_size = old.memory_size;
 758             mem->start_addr = old.start_addr;
 759             mem->phys_offset = old.phys_offset;
 760             mem->flags = 0;
 761
 762             err = kvm_set_user_memory_region(s, mem);
 763             if (err) {
 764                 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
 765                         strerror(-err));
 766                 abort();
 767             }
 768
 769             start_addr += old.memory_size;
 770             phys_offset += old.memory_size;
 771             size -= old.memory_size;
 772             continue;
 773         }
 774
 775         /* register prefix slot */
 776         if (old.start_addr < start_addr) {
 777             mem = kvm_alloc_slot(s);
 778             mem->memory_size = start_addr - old.start_addr;
 779             mem->start_addr = old.start_addr;
 780             mem->phys_offset = old.phys_offset;
 781             mem->flags = 0;
 782
 783             err = kvm_set_user_memory_region(s, mem);
 784             if (err) {
 785                 fprintf(stderr, "%s: error registering prefix slot: %s\n",
 786                         __func__, strerror(-err));
 787                 abort();
 788             }
 789         }
 790
 791         /* register suffix slot */
 792         if (old.start_addr + old.memory_size > start_addr + size) {
 793             ram_addr_t size_delta;
 794
 795             mem = kvm_alloc_slot(s);
 796             mem->start_addr = start_addr + size;
 797             size_delta = mem->start_addr - old.start_addr;
 798             mem->memory_size = old.memory_size - size_delta;
 799             mem->phys_offset = old.phys_offset + size_delta;
 800             mem->flags = 0;
 801
 802             err = kvm_set_user_memory_region(s, mem);
 803             if (err) {
 804                 fprintf(stderr, "%s: error registering suffix slot: %s\n",
 805                         __func__, strerror(-err));
 806                 abort();
 807             }
 808         }
 809     }
 810
 811     /* in case the KVM bug workaround already "consumed" the new slot */
 812     if (!size)
 813         return;
 814
 815     /* KVM does not need to know about this memory */
 816     if (flags >= IO_MEM_UNASSIGNED)
 817         return;
 818
 819     mem = kvm_alloc_slot(s);
 820     mem->memory_size = size;
 821     mem->start_addr = start_addr;
 822     mem->phys_offset = phys_offset;
 823     mem->flags = 0;
 824
 825     err = kvm_set_user_memory_region(s, mem);
 826     if (err) {
 827         fprintf(stderr, "%s: error registering slot: %s\n", __func__,
 828                 strerror(-err));
 829         abort();
 830     }
 831 }
 832
 833 #endif
 834 int kvm_ioctl(KVMState *s, int type, ...)
 835 {
 836     int ret;
 837     void *arg;
 838     va_list ap;
 839
 840     va_start(ap, type);
 841     arg = va_arg(ap, void *);
 842     va_end(ap);
 843
 844     ret = ioctl(s->fd, type, arg);
 845     if (ret == -1)
 846         ret = -errno;
 847
 848     return ret;
 849 }
 850
 851 int kvm_vm_ioctl(KVMState *s, int type, ...)
 852 {
 853     int ret;
 854     void *arg;
 855     va_list ap;
 856
 857     va_start(ap, type);
 858     arg = va_arg(ap, void *);
 859     va_end(ap);
 860
 861     ret = ioctl(s->vmfd, type, arg);
 862     if (ret == -1)
 863         ret = -errno;
 864
 865     return ret;
 866 }
 867
 868 int kvm_vcpu_ioctl(CPUState *env, int type, ...)
 869 {
 870     int ret;
 871     void *arg;
 872     va_list ap;
 873
 874     va_start(ap, type);
 875     arg = va_arg(ap, void *);
 876     va_end(ap);
 877
 878     ret = ioctl(env->kvm_fd, type, arg);
 879     if (ret == -1)
 880         ret = -errno;
 881
 882     return ret;
 883 }
 884
 885 int kvm_has_sync_mmu(void)
 886 {
 887 #ifdef KVM_CAP_SYNC_MMU
 888     KVMState *s = kvm_state;
 889
 890     return kvm_check_extension(s, KVM_CAP_SYNC_MMU);
 891 #else
 892     return 0;
 893 #endif
 894 }
 895
 896 #ifdef KVM_UPSTREAM
 897 void kvm_setup_guest_memory(void *start, size_t size)
 898 {
 899     if (!kvm_has_sync_mmu()) {
 900 #ifdef MADV_DONTFORK
 901         int ret = madvise(start, size, MADV_DONTFORK);
 902
 903         if (ret) {
 904             perror("madvice");
 905             exit(1);
 906         }
 907 #else
 908         fprintf(stderr,
 909                 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
 910         exit(1);
 911 #endif
 912     }
 913 }
 914
 915 #endif /* KVM_UPSTREAM */
 916
 917 #ifdef KVM_CAP_SET_GUEST_DEBUG
 918
 919 #ifdef KVM_UPSTREAM
 920 static void on_vcpu(CPUState *env, void (*func)(void *data), void *data)
 921 {
 922 #ifdef CONFIG_IOTHREAD
 923     if (env == cpu_single_env) {
 924         func(data);
 925         return;
 926     }
 927     abort();
 928 #else
 929     func(data);
 930 #endif
 931 }
 932 #endif /* KVM_UPSTREAM */
 933
 934 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
 935                                                  target_ulong pc)
 936 {
 937     struct kvm_sw_breakpoint *bp;
 938
 939     QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) {
 940         if (bp->pc == pc)
 941             return bp;
 942     }
 943     return NULL;
 944 }
 945
 946 int kvm_sw_breakpoints_active(CPUState *env)
 947 {
 948     return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints);
 949 }
 950
 951 #ifdef KVM_UPSTREAM
 952
 953 struct kvm_set_guest_debug_data {
 954     struct kvm_guest_debug dbg;
 955     CPUState *env;
 956     int err;
 957 };
 958
 959 static void kvm_invoke_set_guest_debug(void *data)
 960 {
 961     struct kvm_set_guest_debug_data *dbg_data = data;
 962     CPUState *env = dbg_data->env;
 963
 964     if (env->kvm_state->regs_modified) {
 965         kvm_arch_put_registers(env);
 966         env->kvm_state->regs_modified = 0;
 967     }
 968     dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg);
 969 }
 970
 971 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 972 {
 973     struct kvm_set_guest_debug_data data;
 974
 975     data.dbg.control = 0;
 976     if (env->singlestep_enabled)
 977         data.dbg.control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
 978
 979     kvm_arch_update_guest_debug(env, &data.dbg);
 980     data.dbg.control |= reinject_trap;
 981     data.env = env;
 982
 983     on_vcpu(env, kvm_invoke_set_guest_debug, &data);
 984     return data.err;
 985 }
 986 #endif
 987
 988 int kvm_insert_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             bp->use_count++;
 999             return 0;
1000         }
1001
1002         bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint));
1003         if (!bp)
1004             return -ENOMEM;
1005
1006         bp->pc = addr;
1007         bp->use_count = 1;
1008         err = kvm_arch_insert_sw_breakpoint(current_env, bp);
1009         if (err) {
1010             free(bp);
1011             return err;
1012         }
1013
1014         QTAILQ_INSERT_HEAD(&current_env->kvm_state->kvm_sw_breakpoints,
1015                           bp, entry);
1016     } else {
1017         err = kvm_arch_insert_hw_breakpoint(addr, len, type);
1018         if (err)
1019             return err;
1020     }
1021
1022     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1023         err = kvm_update_guest_debug(env, 0);
1024         if (err)
1025             return err;
1026     }
1027     return 0;
1028 }
1029
1030 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
1031                           target_ulong len, int type)
1032 {
1033     struct kvm_sw_breakpoint *bp;
1034     CPUState *env;
1035     int err;
1036
1037     if (type == GDB_BREAKPOINT_SW) {
1038         bp = kvm_find_sw_breakpoint(current_env, addr);
1039         if (!bp)
1040             return -ENOENT;
1041
1042         if (bp->use_count > 1) {
1043             bp->use_count--;
1044             return 0;
1045         }
1046
1047         err = kvm_arch_remove_sw_breakpoint(current_env, bp);
1048         if (err)
1049             return err;
1050
1051         QTAILQ_REMOVE(&current_env->kvm_state->kvm_sw_breakpoints, bp, entry);
1052         qemu_free(bp);
1053     } else {
1054         err = kvm_arch_remove_hw_breakpoint(addr, len, type);
1055         if (err)
1056             return err;
1057     }
1058
1059     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1060         err = kvm_update_guest_debug(env, 0);
1061         if (err)
1062             return err;
1063     }
1064     return 0;
1065 }
1066
1067 void kvm_remove_all_breakpoints(CPUState *current_env)
1068 {
1069     struct kvm_sw_breakpoint *bp, *next;
1070     KVMState *s = current_env->kvm_state;
1071     CPUState *env;
1072
1073     QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
1074         if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) {
1075             /* Try harder to find a CPU that currently sees the breakpoint. */
1076             for (env = first_cpu; env != NULL; env = env->next_cpu) {
1077                 if (kvm_arch_remove_sw_breakpoint(env, bp) == 0)
1078                     break;
1079             }
1080         }
1081     }
1082     kvm_arch_remove_all_hw_breakpoints();
1083
1084     for (env = first_cpu; env != NULL; env = env->next_cpu)
1085         kvm_update_guest_debug(env, 0);
1086 }
1087
1088 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1089
1090 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
1091 {
1092     return -EINVAL;
1093 }
1094
1095 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
1096                           target_ulong len, int type)
1097 {
1098     return -EINVAL;
1099 }
1100
1101 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
1102                           target_ulong len, int type)
1103 {
1104     return -EINVAL;
1105 }
1106
1107 void kvm_remove_all_breakpoints(CPUState *current_env)
1108 {
1109 }
1110 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1111
1112 #include "qemu-kvm.c"