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