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     kvm_arch_reset_vcpu(env);
 154     if (kvm_arch_put_registers(env)) {
 155         fprintf(stderr, "Fatal: kvm vcpu reset failed\n");
 156         abort();
 157     }
 158 }
 159 #endif
 160
 161 int kvm_irqchip_in_kernel(void)
 162 {
 163     return kvm_state->irqchip_in_kernel;
 164 }
 165
 166 #ifdef KVM_UPSTREAM
 167 int kvm_pit_in_kernel(void)
 168 {
 169     return kvm_state->pit_in_kernel;
 170 }
 171
 172
 173 int kvm_init_vcpu(CPUState *env)
 174 {
 175     KVMState *s = kvm_state;
 176     long mmap_size;
 177     int ret;
 178
 179     dprintf("kvm_init_vcpu\n");
 180
 181     ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index);
 182     if (ret < 0) {
 183         dprintf("kvm_create_vcpu failed\n");
 184         goto err;
 185     }
 186
 187     env->kvm_fd = ret;
 188     env->kvm_state = s;
 189
 190     mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
 191     if (mmap_size < 0) {
 192         dprintf("KVM_GET_VCPU_MMAP_SIZE failed\n");
 193         goto err;
 194     }
 195
 196     env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
 197                         env->kvm_fd, 0);
 198     if (env->kvm_run == MAP_FAILED) {
 199         ret = -errno;
 200         dprintf("mmap'ing vcpu state failed\n");
 201         goto err;
 202     }
 203
 204     ret = kvm_arch_init_vcpu(env);
 205     if (ret == 0) {
 206         qemu_register_reset(kvm_reset_vcpu, env);
 207         kvm_arch_reset_vcpu(env);
 208         ret = kvm_arch_put_registers(env);
 209     }
 210 err:
 211     return ret;
 212 }
 213
 214 /*
 215  * dirty pages logging control
 216  */
 217 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr,
 218                                       ram_addr_t size, int flags, int mask)
 219 {
 220     KVMState *s = kvm_state;
 221     KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
 222     int old_flags;
 223
 224     if (mem == NULL)  {
 225             fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
 226                     TARGET_FMT_plx "\n", __func__, phys_addr,
 227                     (target_phys_addr_t)(phys_addr + size - 1));
 228             return -EINVAL;
 229     }
 230
 231     old_flags = mem->flags;
 232
 233     flags = (mem->flags & ~mask) | flags;
 234     mem->flags = flags;
 235
 236     /* If nothing changed effectively, no need to issue ioctl */
 237     if (s->migration_log) {
 238         flags |= KVM_MEM_LOG_DIRTY_PAGES;
 239     }
 240     if (flags == old_flags) {
 241             return 0;
 242     }
 243
 244     return kvm_set_user_memory_region(s, mem);
 245 }
 246
 247 int kvm_log_start(target_phys_addr_t phys_addr, ram_addr_t size)
 248 {
 249         return kvm_dirty_pages_log_change(phys_addr, size,
 250                                           KVM_MEM_LOG_DIRTY_PAGES,
 251                                           KVM_MEM_LOG_DIRTY_PAGES);
 252 }
 253
 254 int kvm_log_stop(target_phys_addr_t phys_addr, ram_addr_t size)
 255 {
 256         return kvm_dirty_pages_log_change(phys_addr, size,
 257                                           0,
 258                                           KVM_MEM_LOG_DIRTY_PAGES);
 259 }
 260
 261 int kvm_set_migration_log(int enable)
 262 {
 263     KVMState *s = kvm_state;
 264     KVMSlot *mem;
 265     int i, err;
 266
 267     s->migration_log = enable;
 268
 269     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
 270         mem = &s->slots[i];
 271
 272         if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
 273             continue;
 274         }
 275         err = kvm_set_user_memory_region(s, mem);
 276         if (err) {
 277             return err;
 278         }
 279     }
 280     return 0;
 281 }
 282
 283 static int test_le_bit(unsigned long nr, unsigned char *addr)
 284 {
 285     return (addr[nr >> 3] >> (nr & 7)) & 1;
 286 }
 287
 288 /**
 289  * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
 290  * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
 291  * This means all bits are set to dirty.
 292  *
 293  * @start_add: start of logged region.
 294  * @end_addr: end of logged region.
 295  */
 296 int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr,
 297                                    target_phys_addr_t end_addr)
 298 {
 299     KVMState *s = kvm_state;
 300     unsigned long size, allocated_size = 0;
 301     target_phys_addr_t phys_addr;
 302     ram_addr_t addr;
 303     KVMDirtyLog d;
 304     KVMSlot *mem;
 305     int ret = 0;
 306
 307     d.dirty_bitmap = NULL;
 308     while (start_addr < end_addr) {
 309         mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
 310         if (mem == NULL) {
 311             break;
 312         }
 313
 314         size = ((mem->memory_size >> TARGET_PAGE_BITS) + 7) / 8;
 315         if (!d.dirty_bitmap) {
 316             d.dirty_bitmap = qemu_malloc(size);
 317         } else if (size > allocated_size) {
 318             d.dirty_bitmap = qemu_realloc(d.dirty_bitmap, size);
 319         }
 320         allocated_size = size;
 321         memset(d.dirty_bitmap, 0, allocated_size);
 322
 323         d.slot = mem->slot;
 324
 325         if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
 326             dprintf("ioctl failed %d\n", errno);
 327             ret = -1;
 328             break;
 329         }
 330
 331         for (phys_addr = mem->start_addr, addr = mem->phys_offset;
 332              phys_addr < mem->start_addr + mem->memory_size;
 333              phys_addr += TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
 334             unsigned char *bitmap = (unsigned char *)d.dirty_bitmap;
 335             unsigned nr = (phys_addr - mem->start_addr) >> TARGET_PAGE_BITS;
 336
 337             if (test_le_bit(nr, bitmap)) {
 338                 cpu_physical_memory_set_dirty(addr);
 339             }
 340         }
 341         start_addr = phys_addr;
 342     }
 343     qemu_free(d.dirty_bitmap);
 344
 345     return ret;
 346 }
 347 #endif
 348
 349 int kvm_coalesce_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_REGISTER_COALESCED_MMIO, &zone);
 362     }
 363 #endif
 364
 365     return ret;
 366 }
 367
 368 int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 369 {
 370     int ret = -ENOSYS;
 371 #ifdef KVM_CAP_COALESCED_MMIO
 372     KVMState *s = kvm_state;
 373
 374     if (s->coalesced_mmio) {
 375         struct kvm_coalesced_mmio_zone zone;
 376
 377         zone.addr = start;
 378         zone.size = size;
 379
 380         ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
 381     }
 382 #endif
 383
 384     return ret;
 385 }
 386
 387 int kvm_check_extension(KVMState *s, unsigned int extension)
 388 {
 389     int ret;
 390
 391     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
 392     if (ret < 0) {
 393         ret = 0;
 394     }
 395
 396     return ret;
 397 }
 398 #ifdef KVM_UPSTREAM
 399
 400 int kvm_init(int smp_cpus)
 401 {
 402     static const char upgrade_note[] =
 403         "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
 404         "(see http://sourceforge.net/projects/kvm).\n";
 405     KVMState *s;
 406     int ret;
 407     int i;
 408
 409     if (smp_cpus > 1) {
 410         fprintf(stderr, "No SMP KVM support, use '-smp 1'\n");
 411         return -EINVAL;
 412     }
 413
 414     s = qemu_mallocz(sizeof(KVMState));
 415
 416 #ifdef KVM_CAP_SET_GUEST_DEBUG
 417     QTAILQ_INIT(&s->kvm_sw_breakpoints);
 418 #endif
 419     for (i = 0; i < ARRAY_SIZE(s->slots); i++)
 420         s->slots[i].slot = i;
 421
 422     s->vmfd = -1;
 423     s->fd = open("/dev/kvm", O_RDWR);
 424     if (s->fd == -1) {
 425         fprintf(stderr, "Could not access KVM kernel module: %m\n");
 426         ret = -errno;
 427         goto err;
 428     }
 429
 430     ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
 431     if (ret < KVM_API_VERSION) {
 432         if (ret > 0)
 433             ret = -EINVAL;
 434         fprintf(stderr, "kvm version too old\n");
 435         goto err;
 436     }
 437
 438     if (ret > KVM_API_VERSION) {
 439         ret = -EINVAL;
 440         fprintf(stderr, "kvm version not supported\n");
 441         goto err;
 442     }
 443
 444     s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
 445     if (s->vmfd < 0)
 446         goto err;
 447
 448     /* initially, KVM allocated its own memory and we had to jump through
 449      * hooks to make phys_ram_base point to this.  Modern versions of KVM
 450      * just use a user allocated buffer so we can use regular pages
 451      * unmodified.  Make sure we have a sufficiently modern version of KVM.
 452      */
 453     if (!kvm_check_extension(s, KVM_CAP_USER_MEMORY)) {
 454         ret = -EINVAL;
 455         fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
 456                 upgrade_note);
 457         goto err;
 458     }
 459
 460     /* There was a nasty bug in < kvm-80 that prevents memory slots from being
 461      * destroyed properly.  Since we rely on this capability, refuse to work
 462      * with any kernel without this capability. */
 463     if (!kvm_check_extension(s, KVM_CAP_DESTROY_MEMORY_REGION_WORKS)) {
 464         ret = -EINVAL;
 465
 466         fprintf(stderr,
 467                 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
 468                 upgrade_note);
 469         goto err;
 470     }
 471
 472 #ifdef KVM_CAP_COALESCED_MMIO
 473     s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
 474 #else
 475     s->coalesced_mmio = 0;
 476 #endif
 477
 478     s->broken_set_mem_region = 1;
 479 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
 480     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
 481     if (ret > 0) {
 482         s->broken_set_mem_region = 0;
 483     }
 484 #endif
 485
 486     ret = kvm_arch_init(s, smp_cpus);
 487     if (ret < 0)
 488         goto err;
 489
 490     kvm_state = s;
 491
 492     return 0;
 493
 494 err:
 495     if (s) {
 496         if (s->vmfd != -1)
 497             close(s->vmfd);
 498         if (s->fd != -1)
 499             close(s->fd);
 500     }
 501     qemu_free(s);
 502
 503     return ret;
 504 }
 505 #endif
 506
 507 static int kvm_handle_io(uint16_t port, void *data, int direction, int size,
 508                          uint32_t count)
 509 {
 510     int i;
 511     uint8_t *ptr = data;
 512
 513     for (i = 0; i < count; i++) {
 514         if (direction == KVM_EXIT_IO_IN) {
 515             switch (size) {
 516             case 1:
 517                 stb_p(ptr, cpu_inb(port));
 518                 break;
 519             case 2:
 520                 stw_p(ptr, cpu_inw(port));
 521                 break;
 522             case 4:
 523                 stl_p(ptr, cpu_inl(port));
 524                 break;
 525             }
 526         } else {
 527             switch (size) {
 528             case 1:
 529                 cpu_outb(port, ldub_p(ptr));
 530                 break;
 531             case 2:
 532                 cpu_outw(port, lduw_p(ptr));
 533                 break;
 534             case 4:
 535                 cpu_outl(port, ldl_p(ptr));
 536                 break;
 537             }
 538         }
 539
 540         ptr += size;
 541     }
 542
 543     return 1;
 544 }
 545
 546 #ifdef KVM_UPSTREAM
 547 static void kvm_run_coalesced_mmio(CPUState *env, struct kvm_run *run)
 548 {
 549 #ifdef KVM_CAP_COALESCED_MMIO
 550     KVMState *s = kvm_state;
 551     if (s->coalesced_mmio) {
 552         struct kvm_coalesced_mmio_ring *ring;
 553
 554         ring = (void *)run + (s->coalesced_mmio * TARGET_PAGE_SIZE);
 555         while (ring->first != ring->last) {
 556             struct kvm_coalesced_mmio *ent;
 557
 558             ent = &ring->coalesced_mmio[ring->first];
 559
 560             cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
 561             /* FIXME smp_wmb() */
 562             ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
 563         }
 564     }
 565 #endif
 566 }
 567
 568 void kvm_cpu_synchronize_state(CPUState *env)
 569 {
 570     if (!env->kvm_state->regs_modified) {
 571         kvm_arch_get_registers(env);
 572         env->kvm_state->regs_modified = 1;
 573     }
 574 }
 575
 576 int kvm_cpu_exec(CPUState *env)
 577 {
 578     struct kvm_run *run = env->kvm_run;
 579     int ret;
 580
 581     dprintf("kvm_cpu_exec()\n");
 582
 583     do {
 584         if (env->exit_request) {
 585             dprintf("interrupt exit requested\n");
 586             ret = 0;
 587             break;
 588         }
 589
 590         if (env->kvm_state->regs_modified) {
 591             kvm_arch_put_registers(env);
 592             env->kvm_state->regs_modified = 0;
 593         }
 594
 595         kvm_arch_pre_run(env, run);
 596         qemu_mutex_unlock_iothread();
 597         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
 598         qemu_mutex_lock_iothread();
 599         kvm_arch_post_run(env, run);
 600
 601         if (ret == -EINTR || ret == -EAGAIN) {
 602             dprintf("io window exit\n");
 603             ret = 0;
 604             break;
 605         }
 606
 607         if (ret < 0) {
 608             dprintf("kvm run failed %s\n", strerror(-ret));
 609             abort();
 610         }
 611
 612         kvm_run_coalesced_mmio(env, run);
 613
 614         ret = 0; /* exit loop */
 615         switch (run->exit_reason) {
 616         case KVM_EXIT_IO:
 617             dprintf("handle_io\n");
 618             ret = kvm_handle_io(run->io.port,
 619                                 (uint8_t *)run + run->io.data_offset,
 620                                 run->io.direction,
 621                                 run->io.size,
 622                                 run->io.count);
 623             break;
 624         case KVM_EXIT_MMIO:
 625             dprintf("handle_mmio\n");
 626             cpu_physical_memory_rw(run->mmio.phys_addr,
 627                                    run->mmio.data,
 628                                    run->mmio.len,
 629                                    run->mmio.is_write);
 630             ret = 1;
 631             break;
 632         case KVM_EXIT_IRQ_WINDOW_OPEN:
 633             dprintf("irq_window_open\n");
 634             break;
 635         case KVM_EXIT_SHUTDOWN:
 636             dprintf("shutdown\n");
 637             qemu_system_reset_request();
 638             ret = 1;
 639             break;
 640         case KVM_EXIT_UNKNOWN:
 641             dprintf("kvm_exit_unknown\n");
 642             break;
 643         case KVM_EXIT_FAIL_ENTRY:
 644             dprintf("kvm_exit_fail_entry\n");
 645             break;
 646         case KVM_EXIT_EXCEPTION:
 647             dprintf("kvm_exit_exception\n");
 648             break;
 649         case KVM_EXIT_DEBUG:
 650             dprintf("kvm_exit_debug\n");
 651 #ifdef KVM_CAP_SET_GUEST_DEBUG
 652             if (kvm_arch_debug(&run->debug.arch)) {
 653                 gdb_set_stop_cpu(env);
 654                 vm_stop(EXCP_DEBUG);
 655                 env->exception_index = EXCP_DEBUG;
 656                 return 0;
 657             }
 658             /* re-enter, this exception was guest-internal */
 659             ret = 1;
 660 #endif /* KVM_CAP_SET_GUEST_DEBUG */
 661             break;
 662         default:
 663             dprintf("kvm_arch_handle_exit\n");
 664             ret = kvm_arch_handle_exit(env, run);
 665             break;
 666         }
 667     } while (ret > 0);
 668
 669     if (env->exit_request) {
 670         env->exit_request = 0;
 671         env->exception_index = EXCP_INTERRUPT;
 672     }
 673
 674     return ret;
 675 }
 676
 677 void kvm_set_phys_mem(target_phys_addr_t start_addr,
 678                       ram_addr_t size,
 679                       ram_addr_t phys_offset)
 680 {
 681     KVMState *s = kvm_state;
 682     ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK;
 683     KVMSlot *mem, old;
 684     int err;
 685
 686     if (start_addr & ~TARGET_PAGE_MASK) {
 687         if (flags >= IO_MEM_UNASSIGNED) {
 688             if (!kvm_lookup_overlapping_slot(s, start_addr,
 689                                              start_addr + size)) {
 690                 return;
 691             }
 692             fprintf(stderr, "Unaligned split of a KVM memory slot\n");
 693         } else {
 694             fprintf(stderr, "Only page-aligned memory slots supported\n");
 695         }
 696         abort();
 697     }
 698
 699     /* KVM does not support read-only slots */
 700     phys_offset &= ~IO_MEM_ROM;
 701
 702     while (1) {
 703         mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
 704         if (!mem) {
 705             break;
 706         }
 707
 708         if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr &&
 709             (start_addr + size <= mem->start_addr + mem->memory_size) &&
 710             (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) {
 711             /* The new slot fits into the existing one and comes with
 712              * identical parameters - nothing to be done. */
 713             return;
 714         }
 715
 716         old = *mem;
 717
 718         /* unregister the overlapping slot */
 719         mem->memory_size = 0;
 720         err = kvm_set_user_memory_region(s, mem);
 721         if (err) {
 722             fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
 723                     __func__, strerror(-err));
 724             abort();
 725         }
 726
 727         /* Workaround for older KVM versions: we can't join slots, even not by
 728          * unregistering the previous ones and then registering the larger
 729          * slot. We have to maintain the existing fragmentation. Sigh.
 730          *
 731          * This workaround assumes that the new slot starts at the same
 732          * address as the first existing one. If not or if some overlapping
 733          * slot comes around later, we will fail (not seen in practice so far)
 734          * - and actually require a recent KVM version. */
 735         if (s->broken_set_mem_region &&
 736             old.start_addr == start_addr && old.memory_size < size &&
 737             flags < IO_MEM_UNASSIGNED) {
 738             mem = kvm_alloc_slot(s);
 739             mem->memory_size = old.memory_size;
 740             mem->start_addr = old.start_addr;
 741             mem->phys_offset = old.phys_offset;
 742             mem->flags = 0;
 743
 744             err = kvm_set_user_memory_region(s, mem);
 745             if (err) {
 746                 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
 747                         strerror(-err));
 748                 abort();
 749             }
 750
 751             start_addr += old.memory_size;
 752             phys_offset += old.memory_size;
 753             size -= old.memory_size;
 754             continue;
 755         }
 756
 757         /* register prefix slot */
 758         if (old.start_addr < start_addr) {
 759             mem = kvm_alloc_slot(s);
 760             mem->memory_size = start_addr - old.start_addr;
 761             mem->start_addr = old.start_addr;
 762             mem->phys_offset = old.phys_offset;
 763             mem->flags = 0;
 764
 765             err = kvm_set_user_memory_region(s, mem);
 766             if (err) {
 767                 fprintf(stderr, "%s: error registering prefix slot: %s\n",
 768                         __func__, strerror(-err));
 769                 abort();
 770             }
 771         }
 772
 773         /* register suffix slot */
 774         if (old.start_addr + old.memory_size > start_addr + size) {
 775             ram_addr_t size_delta;
 776
 777             mem = kvm_alloc_slot(s);
 778             mem->start_addr = start_addr + size;
 779             size_delta = mem->start_addr - old.start_addr;
 780             mem->memory_size = old.memory_size - size_delta;
 781             mem->phys_offset = old.phys_offset + size_delta;
 782             mem->flags = 0;
 783
 784             err = kvm_set_user_memory_region(s, mem);
 785             if (err) {
 786                 fprintf(stderr, "%s: error registering suffix slot: %s\n",
 787                         __func__, strerror(-err));
 788                 abort();
 789             }
 790         }
 791     }
 792
 793     /* in case the KVM bug workaround already "consumed" the new slot */
 794     if (!size)
 795         return;
 796
 797     /* KVM does not need to know about this memory */
 798     if (flags >= IO_MEM_UNASSIGNED)
 799         return;
 800
 801     mem = kvm_alloc_slot(s);
 802     mem->memory_size = size;
 803     mem->start_addr = start_addr;
 804     mem->phys_offset = phys_offset;
 805     mem->flags = 0;
 806
 807     err = kvm_set_user_memory_region(s, mem);
 808     if (err) {
 809         fprintf(stderr, "%s: error registering slot: %s\n", __func__,
 810                 strerror(-err));
 811         abort();
 812     }
 813 }
 814
 815 #endif
 816 int kvm_ioctl(KVMState *s, 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(s->fd, type, arg);
 827     if (ret == -1)
 828         ret = -errno;
 829
 830     return ret;
 831 }
 832
 833 int kvm_vm_ioctl(KVMState *s, int type, ...)
 834 {
 835     int ret;
 836     void *arg;
 837     va_list ap;
 838
 839     va_start(ap, type);
 840     arg = va_arg(ap, void *);
 841     va_end(ap);
 842
 843     ret = ioctl(s->vmfd, type, arg);
 844     if (ret == -1)
 845         ret = -errno;
 846
 847     return ret;
 848 }
 849
 850 int kvm_vcpu_ioctl(CPUState *env, int type, ...)
 851 {
 852     int ret;
 853     void *arg;
 854     va_list ap;
 855
 856     va_start(ap, type);
 857     arg = va_arg(ap, void *);
 858     va_end(ap);
 859
 860     ret = ioctl(env->kvm_fd, type, arg);
 861     if (ret == -1)
 862         ret = -errno;
 863
 864     return ret;
 865 }
 866
 867 int kvm_has_sync_mmu(void)
 868 {
 869 #ifdef KVM_CAP_SYNC_MMU
 870     KVMState *s = kvm_state;
 871
 872     return kvm_check_extension(s, KVM_CAP_SYNC_MMU);
 873 #else
 874     return 0;
 875 #endif
 876 }
 877
 878 #ifdef KVM_UPSTREAM
 879 void kvm_setup_guest_memory(void *start, size_t size)
 880 {
 881     if (!kvm_has_sync_mmu()) {
 882 #ifdef MADV_DONTFORK
 883         int ret = madvise(start, size, MADV_DONTFORK);
 884
 885         if (ret) {
 886             perror("madvice");
 887             exit(1);
 888         }
 889 #else
 890         fprintf(stderr,
 891                 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
 892         exit(1);
 893 #endif
 894     }
 895 }
 896
 897 #endif /* KVM_UPSTREAM */
 898
 899 #ifdef KVM_CAP_SET_GUEST_DEBUG
 900
 901 #ifdef KVM_UPSTREAM
 902 static void on_vcpu(CPUState *env, void (*func)(void *data), void *data)
 903 {
 904 #ifdef CONFIG_IOTHREAD
 905     if (env == cpu_single_env) {
 906         func(data);
 907         return;
 908     }
 909     abort();
 910 #else
 911     func(data);
 912 #endif
 913 }
 914 #endif /* KVM_UPSTREAM */
 915
 916 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
 917                                                  target_ulong pc)
 918 {
 919     struct kvm_sw_breakpoint *bp;
 920
 921     QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) {
 922         if (bp->pc == pc)
 923             return bp;
 924     }
 925     return NULL;
 926 }
 927
 928 int kvm_sw_breakpoints_active(CPUState *env)
 929 {
 930     return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints);
 931 }
 932
 933 #ifdef KVM_UPSTREAM
 934
 935 struct kvm_set_guest_debug_data {
 936     struct kvm_guest_debug dbg;
 937     CPUState *env;
 938     int err;
 939 };
 940
 941 static void kvm_invoke_set_guest_debug(void *data)
 942 {
 943     struct kvm_set_guest_debug_data *dbg_data = data;
 944     CPUState *env = dbg_data->env;
 945
 946     if (env->kvm_state->regs_modified) {
 947         kvm_arch_put_registers(env);
 948         env->kvm_state->regs_modified = 0;
 949     }
 950     dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg);
 951 }
 952
 953 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 954 {
 955     struct kvm_set_guest_debug_data data;
 956
 957     data.dbg.control = 0;
 958     if (env->singlestep_enabled)
 959         data.dbg.control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
 960
 961     kvm_arch_update_guest_debug(env, &data.dbg);
 962     data.dbg.control |= reinject_trap;
 963     data.env = env;
 964
 965     on_vcpu(env, kvm_invoke_set_guest_debug, &data);
 966     return data.err;
 967 }
 968 #endif
 969
 970 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 971                           target_ulong len, int type)
 972 {
 973     struct kvm_sw_breakpoint *bp;
 974     CPUState *env;
 975     int err;
 976
 977     if (type == GDB_BREAKPOINT_SW) {
 978         bp = kvm_find_sw_breakpoint(current_env, addr);
 979         if (bp) {
 980             bp->use_count++;
 981             return 0;
 982         }
 983
 984         bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint));
 985         if (!bp)
 986             return -ENOMEM;
 987
 988         bp->pc = addr;
 989         bp->use_count = 1;
 990         err = kvm_arch_insert_sw_breakpoint(current_env, bp);
 991         if (err) {
 992             free(bp);
 993             return err;
 994         }
 995
 996         QTAILQ_INSERT_HEAD(&current_env->kvm_state->kvm_sw_breakpoints,
 997                           bp, entry);
 998     } else {
 999         err = kvm_arch_insert_hw_breakpoint(addr, len, type);
1000         if (err)
1001             return err;
1002     }
1003
1004     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1005         err = kvm_update_guest_debug(env, 0);
1006         if (err)
1007             return err;
1008     }
1009     return 0;
1010 }
1011
1012 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
1013                           target_ulong len, int type)
1014 {
1015     struct kvm_sw_breakpoint *bp;
1016     CPUState *env;
1017     int err;
1018
1019     if (type == GDB_BREAKPOINT_SW) {
1020         bp = kvm_find_sw_breakpoint(current_env, addr);
1021         if (!bp)
1022             return -ENOENT;
1023
1024         if (bp->use_count > 1) {
1025             bp->use_count--;
1026             return 0;
1027         }
1028
1029         err = kvm_arch_remove_sw_breakpoint(current_env, bp);
1030         if (err)
1031             return err;
1032
1033         QTAILQ_REMOVE(&current_env->kvm_state->kvm_sw_breakpoints, bp, entry);
1034         qemu_free(bp);
1035     } else {
1036         err = kvm_arch_remove_hw_breakpoint(addr, len, type);
1037         if (err)
1038             return err;
1039     }
1040
1041     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1042         err = kvm_update_guest_debug(env, 0);
1043         if (err)
1044             return err;
1045     }
1046     return 0;
1047 }
1048
1049 void kvm_remove_all_breakpoints(CPUState *current_env)
1050 {
1051     struct kvm_sw_breakpoint *bp, *next;
1052     KVMState *s = current_env->kvm_state;
1053     CPUState *env;
1054
1055     QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
1056         if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) {
1057             /* Try harder to find a CPU that currently sees the breakpoint. */
1058             for (env = first_cpu; env != NULL; env = env->next_cpu) {
1059                 if (kvm_arch_remove_sw_breakpoint(env, bp) == 0)
1060                     break;
1061             }
1062         }
1063     }
1064     kvm_arch_remove_all_hw_breakpoints();
1065
1066     for (env = first_cpu; env != NULL; env = env->next_cpu)
1067         kvm_update_guest_debug(env, 0);
1068 }
1069
1070 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1071
1072 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
1073 {
1074     return -EINVAL;
1075 }
1076
1077 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
1078                           target_ulong len, int type)
1079 {
1080     return -EINVAL;
1081 }
1082
1083 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
1084                           target_ulong len, int type)
1085 {
1086     return -EINVAL;
1087 }
1088
1089 void kvm_remove_all_breakpoints(CPUState *current_env)
1090 {
1091 }
1092 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1093
1094 #include "qemu-kvm.c"