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