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