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         fprintf(stderr, "No SMP KVM support, use '-smp 1'\n");
 343         return -EINVAL;
 344     }
 345
 346     s = qemu_mallocz(sizeof(KVMState));
 347
 348 #ifdef KVM_CAP_SET_GUEST_DEBUG
 349     TAILQ_INIT(&s->kvm_sw_breakpoints);
 350 #endif
 351     for (i = 0; i < ARRAY_SIZE(s->slots); i++)
 352         s->slots[i].slot = i;
 353
 354     s->vmfd = -1;
 355     s->fd = open("/dev/kvm", O_RDWR);
 356     if (s->fd == -1) {
 357         fprintf(stderr, "Could not access KVM kernel module: %m\n");
 358         ret = -errno;
 359         goto err;
 360     }
 361
 362     ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
 363     if (ret < KVM_API_VERSION) {
 364         if (ret > 0)
 365             ret = -EINVAL;
 366         fprintf(stderr, "kvm version too old\n");
 367         goto err;
 368     }
 369
 370     if (ret > KVM_API_VERSION) {
 371         ret = -EINVAL;
 372         fprintf(stderr, "kvm version not supported\n");
 373         goto err;
 374     }
 375
 376     s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
 377     if (s->vmfd < 0)
 378         goto err;
 379
 380     /* initially, KVM allocated its own memory and we had to jump through
 381      * hooks to make phys_ram_base point to this.  Modern versions of KVM
 382      * just use a user allocated buffer so we can use regular pages
 383      * unmodified.  Make sure we have a sufficiently modern version of KVM.
 384      */
 385     if (!kvm_check_extension(s, KVM_CAP_USER_MEMORY)) {
 386         ret = -EINVAL;
 387         fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n");
 388         goto err;
 389     }
 390
 391     /* There was a nasty bug in < kvm-80 that prevents memory slots from being
 392      * destroyed properly.  Since we rely on this capability, refuse to work
 393      * with any kernel without this capability. */
 394     if (!kvm_check_extension(s, KVM_CAP_DESTROY_MEMORY_REGION_WORKS)) {
 395         ret = -EINVAL;
 396
 397         fprintf(stderr,
 398                 "KVM kernel module broken (DESTROY_MEMORY_REGION)\n"
 399                 "Please upgrade to at least kvm-81.\n");
 400         goto err;
 401     }
 402
 403 #ifdef KVM_CAP_COALESCED_MMIO
 404     s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
 405 #else
 406     s->coalesced_mmio = 0;
 407 #endif
 408
 409     ret = kvm_arch_init(s, smp_cpus);
 410     if (ret < 0)
 411         goto err;
 412
 413     kvm_state = s;
 414
 415     return 0;
 416
 417 err:
 418     if (s) {
 419         if (s->vmfd != -1)
 420             close(s->vmfd);
 421         if (s->fd != -1)
 422             close(s->fd);
 423     }
 424     qemu_free(s);
 425
 426     return ret;
 427 }
 428
 429 static int kvm_handle_io(CPUState *env, uint16_t port, void *data,
 430                          int direction, int size, uint32_t count)
 431 {
 432     int i;
 433     uint8_t *ptr = data;
 434
 435     for (i = 0; i < count; i++) {
 436         if (direction == KVM_EXIT_IO_IN) {
 437             switch (size) {
 438             case 1:
 439                 stb_p(ptr, cpu_inb(env, port));
 440                 break;
 441             case 2:
 442                 stw_p(ptr, cpu_inw(env, port));
 443                 break;
 444             case 4:
 445                 stl_p(ptr, cpu_inl(env, port));
 446                 break;
 447             }
 448         } else {
 449             switch (size) {
 450             case 1:
 451                 cpu_outb(env, port, ldub_p(ptr));
 452                 break;
 453             case 2:
 454                 cpu_outw(env, port, lduw_p(ptr));
 455                 break;
 456             case 4:
 457                 cpu_outl(env, port, ldl_p(ptr));
 458                 break;
 459             }
 460         }
 461
 462         ptr += size;
 463     }
 464
 465     return 1;
 466 }
 467
 468 static void kvm_run_coalesced_mmio(CPUState *env, struct kvm_run *run)
 469 {
 470 #ifdef KVM_CAP_COALESCED_MMIO
 471     KVMState *s = kvm_state;
 472     if (s->coalesced_mmio) {
 473         struct kvm_coalesced_mmio_ring *ring;
 474
 475         ring = (void *)run + (s->coalesced_mmio * TARGET_PAGE_SIZE);
 476         while (ring->first != ring->last) {
 477             struct kvm_coalesced_mmio *ent;
 478
 479             ent = &ring->coalesced_mmio[ring->first];
 480
 481             cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
 482             /* FIXME smp_wmb() */
 483             ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
 484         }
 485     }
 486 #endif
 487 }
 488
 489 int kvm_cpu_exec(CPUState *env)
 490 {
 491     struct kvm_run *run = env->kvm_run;
 492     int ret;
 493
 494     dprintf("kvm_cpu_exec()\n");
 495
 496     do {
 497         kvm_arch_pre_run(env, run);
 498
 499         if (env->exit_request) {
 500             dprintf("interrupt exit requested\n");
 501             ret = 0;
 502             break;
 503         }
 504
 505         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
 506         kvm_arch_post_run(env, run);
 507
 508         if (ret == -EINTR || ret == -EAGAIN) {
 509             dprintf("io window exit\n");
 510             ret = 0;
 511             break;
 512         }
 513
 514         if (ret < 0) {
 515             dprintf("kvm run failed %s\n", strerror(-ret));
 516             abort();
 517         }
 518
 519         kvm_run_coalesced_mmio(env, run);
 520
 521         ret = 0; /* exit loop */
 522         switch (run->exit_reason) {
 523         case KVM_EXIT_IO:
 524             dprintf("handle_io\n");
 525             ret = kvm_handle_io(env, run->io.port,
 526                                 (uint8_t *)run + run->io.data_offset,
 527                                 run->io.direction,
 528                                 run->io.size,
 529                                 run->io.count);
 530             break;
 531         case KVM_EXIT_MMIO:
 532             dprintf("handle_mmio\n");
 533             cpu_physical_memory_rw(run->mmio.phys_addr,
 534                                    run->mmio.data,
 535                                    run->mmio.len,
 536                                    run->mmio.is_write);
 537             ret = 1;
 538             break;
 539         case KVM_EXIT_IRQ_WINDOW_OPEN:
 540             dprintf("irq_window_open\n");
 541             break;
 542         case KVM_EXIT_SHUTDOWN:
 543             dprintf("shutdown\n");
 544             qemu_system_reset_request();
 545             ret = 1;
 546             break;
 547         case KVM_EXIT_UNKNOWN:
 548             dprintf("kvm_exit_unknown\n");
 549             break;
 550         case KVM_EXIT_FAIL_ENTRY:
 551             dprintf("kvm_exit_fail_entry\n");
 552             break;
 553         case KVM_EXIT_EXCEPTION:
 554             dprintf("kvm_exit_exception\n");
 555             break;
 556         case KVM_EXIT_DEBUG:
 557             dprintf("kvm_exit_debug\n");
 558 #ifdef KVM_CAP_SET_GUEST_DEBUG
 559             if (kvm_arch_debug(&run->debug.arch)) {
 560                 gdb_set_stop_cpu(env);
 561                 vm_stop(EXCP_DEBUG);
 562                 env->exception_index = EXCP_DEBUG;
 563                 return 0;
 564             }
 565             /* re-enter, this exception was guest-internal */
 566             ret = 1;
 567 #endif /* KVM_CAP_SET_GUEST_DEBUG */
 568             break;
 569         default:
 570             dprintf("kvm_arch_handle_exit\n");
 571             ret = kvm_arch_handle_exit(env, run);
 572             break;
 573         }
 574     } while (ret > 0);
 575
 576     if (env->exit_request) {
 577         env->exit_request = 0;
 578         env->exception_index = EXCP_INTERRUPT;
 579     }
 580
 581     return ret;
 582 }
 583
 584 void kvm_set_phys_mem(target_phys_addr_t start_addr,
 585                       ram_addr_t size,
 586                       ram_addr_t phys_offset)
 587 {
 588     KVMState *s = kvm_state;
 589     ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK;
 590     KVMSlot *mem, old;
 591     int err;
 592
 593     if (start_addr & ~TARGET_PAGE_MASK) {
 594         if (flags >= IO_MEM_UNASSIGNED) {
 595             if (!kvm_lookup_overlapping_slot(s, start_addr,
 596                                              start_addr + size)) {
 597                 return;
 598             }
 599             fprintf(stderr, "Unaligned split of a KVM memory slot\n");
 600         } else {
 601             fprintf(stderr, "Only page-aligned memory slots supported\n");
 602         }
 603         abort();
 604     }
 605
 606     /* KVM does not support read-only slots */
 607     phys_offset &= ~IO_MEM_ROM;
 608
 609     while (1) {
 610         mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
 611         if (!mem) {
 612             break;
 613         }
 614
 615         if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr &&
 616             (start_addr + size <= mem->start_addr + mem->memory_size) &&
 617             (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) {
 618             /* The new slot fits into the existing one and comes with
 619              * identical parameters - nothing to be done. */
 620             return;
 621         }
 622
 623         old = *mem;
 624
 625         /* unregister the overlapping slot */
 626         mem->memory_size = 0;
 627         err = kvm_set_user_memory_region(s, mem);
 628         if (err) {
 629             fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
 630                     __func__, strerror(-err));
 631             abort();
 632         }
 633
 634         /* Workaround for older KVM versions: we can't join slots, even not by
 635          * unregistering the previous ones and then registering the larger
 636          * slot. We have to maintain the existing fragmentation. Sigh.
 637          *
 638          * This workaround assumes that the new slot starts at the same
 639          * address as the first existing one. If not or if some overlapping
 640          * slot comes around later, we will fail (not seen in practice so far)
 641          * - and actually require a recent KVM version. */
 642         if (old.start_addr == start_addr && old.memory_size < size &&
 643             flags < IO_MEM_UNASSIGNED) {
 644             mem = kvm_alloc_slot(s);
 645             mem->memory_size = old.memory_size;
 646             mem->start_addr = old.start_addr;
 647             mem->phys_offset = old.phys_offset;
 648             mem->flags = 0;
 649
 650             err = kvm_set_user_memory_region(s, mem);
 651             if (err) {
 652                 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
 653                         strerror(-err));
 654                 abort();
 655             }
 656
 657             start_addr += old.memory_size;
 658             phys_offset += old.memory_size;
 659             size -= old.memory_size;
 660             continue;
 661         }
 662
 663         /* register prefix slot */
 664         if (old.start_addr < start_addr) {
 665             mem = kvm_alloc_slot(s);
 666             mem->memory_size = start_addr - old.start_addr;
 667             mem->start_addr = old.start_addr;
 668             mem->phys_offset = old.phys_offset;
 669             mem->flags = 0;
 670
 671             err = kvm_set_user_memory_region(s, mem);
 672             if (err) {
 673                 fprintf(stderr, "%s: error registering prefix slot: %s\n",
 674                         __func__, strerror(-err));
 675                 abort();
 676             }
 677         }
 678
 679         /* register suffix slot */
 680         if (old.start_addr + old.memory_size > start_addr + size) {
 681             ram_addr_t size_delta;
 682
 683             mem = kvm_alloc_slot(s);
 684             mem->start_addr = start_addr + size;
 685             size_delta = mem->start_addr - old.start_addr;
 686             mem->memory_size = old.memory_size - size_delta;
 687             mem->phys_offset = old.phys_offset + size_delta;
 688             mem->flags = 0;
 689
 690             err = kvm_set_user_memory_region(s, mem);
 691             if (err) {
 692                 fprintf(stderr, "%s: error registering suffix slot: %s\n",
 693                         __func__, strerror(-err));
 694                 abort();
 695             }
 696         }
 697     }
 698
 699     /* in case the KVM bug workaround already "consumed" the new slot */
 700     if (!size)
 701         return;
 702
 703     /* KVM does not need to know about this memory */
 704     if (flags >= IO_MEM_UNASSIGNED)
 705         return;
 706
 707     mem = kvm_alloc_slot(s);
 708     mem->memory_size = size;
 709     mem->start_addr = start_addr;
 710     mem->phys_offset = phys_offset;
 711     mem->flags = 0;
 712
 713     err = kvm_set_user_memory_region(s, mem);
 714     if (err) {
 715         fprintf(stderr, "%s: error registering slot: %s\n", __func__,
 716                 strerror(-err));
 717         abort();
 718     }
 719 }
 720
 721 int kvm_ioctl(KVMState *s, int type, ...)
 722 {
 723     int ret;
 724     void *arg;
 725     va_list ap;
 726
 727     va_start(ap, type);
 728     arg = va_arg(ap, void *);
 729     va_end(ap);
 730
 731     ret = ioctl(s->fd, type, arg);
 732     if (ret == -1)
 733         ret = -errno;
 734
 735     return ret;
 736 }
 737
 738 int kvm_vm_ioctl(KVMState *s, int type, ...)
 739 {
 740     int ret;
 741     void *arg;
 742     va_list ap;
 743
 744     va_start(ap, type);
 745     arg = va_arg(ap, void *);
 746     va_end(ap);
 747
 748     ret = ioctl(s->vmfd, type, arg);
 749     if (ret == -1)
 750         ret = -errno;
 751
 752     return ret;
 753 }
 754
 755 int kvm_vcpu_ioctl(CPUState *env, int type, ...)
 756 {
 757     int ret;
 758     void *arg;
 759     va_list ap;
 760
 761     va_start(ap, type);
 762     arg = va_arg(ap, void *);
 763     va_end(ap);
 764
 765     ret = ioctl(env->kvm_fd, type, arg);
 766     if (ret == -1)
 767         ret = -errno;
 768
 769     return ret;
 770 }
 771
 772 int kvm_has_sync_mmu(void)
 773 {
 774 #ifdef KVM_CAP_SYNC_MMU
 775     KVMState *s = kvm_state;
 776
 777     return kvm_check_extension(s, KVM_CAP_SYNC_MMU);
 778 #else
 779     return 0;
 780 #endif
 781 }
 782
 783 void kvm_setup_guest_memory(void *start, size_t size)
 784 {
 785     if (!kvm_has_sync_mmu()) {
 786 #ifdef MADV_DONTFORK
 787         int ret = madvise(start, size, MADV_DONTFORK);
 788
 789         if (ret) {
 790             perror("madvice");
 791             exit(1);
 792         }
 793 #else
 794         fprintf(stderr,
 795                 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
 796         exit(1);
 797 #endif
 798     }
 799 }
 800
 801 #ifdef KVM_CAP_SET_GUEST_DEBUG
 802 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
 803                                                  target_ulong pc)
 804 {
 805     struct kvm_sw_breakpoint *bp;
 806
 807     TAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) {
 808         if (bp->pc == pc)
 809             return bp;
 810     }
 811     return NULL;
 812 }
 813
 814 int kvm_sw_breakpoints_active(CPUState *env)
 815 {
 816     return !TAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints);
 817 }
 818
 819 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 820 {
 821     struct kvm_guest_debug dbg;
 822
 823     dbg.control = 0;
 824     if (env->singlestep_enabled)
 825         dbg.control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
 826
 827     kvm_arch_update_guest_debug(env, &dbg);
 828     dbg.control |= reinject_trap;
 829
 830     return kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg);
 831 }
 832
 833 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 834                           target_ulong len, int type)
 835 {
 836     struct kvm_sw_breakpoint *bp;
 837     CPUState *env;
 838     int err;
 839
 840     if (type == GDB_BREAKPOINT_SW) {
 841         bp = kvm_find_sw_breakpoint(current_env, addr);
 842         if (bp) {
 843             bp->use_count++;
 844             return 0;
 845         }
 846
 847         bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint));
 848         if (!bp)
 849             return -ENOMEM;
 850
 851         bp->pc = addr;
 852         bp->use_count = 1;
 853         err = kvm_arch_insert_sw_breakpoint(current_env, bp);
 854         if (err) {
 855             free(bp);
 856             return err;
 857         }
 858
 859         TAILQ_INSERT_HEAD(&current_env->kvm_state->kvm_sw_breakpoints,
 860                           bp, entry);
 861     } else {
 862         err = kvm_arch_insert_hw_breakpoint(addr, len, type);
 863         if (err)
 864             return err;
 865     }
 866
 867     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 868         err = kvm_update_guest_debug(env, 0);
 869         if (err)
 870             return err;
 871     }
 872     return 0;
 873 }
 874
 875 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
 876                           target_ulong len, int type)
 877 {
 878     struct kvm_sw_breakpoint *bp;
 879     CPUState *env;
 880     int err;
 881
 882     if (type == GDB_BREAKPOINT_SW) {
 883         bp = kvm_find_sw_breakpoint(current_env, addr);
 884         if (!bp)
 885             return -ENOENT;
 886
 887         if (bp->use_count > 1) {
 888             bp->use_count--;
 889             return 0;
 890         }
 891
 892         err = kvm_arch_remove_sw_breakpoint(current_env, bp);
 893         if (err)
 894             return err;
 895
 896         TAILQ_REMOVE(&current_env->kvm_state->kvm_sw_breakpoints, bp, entry);
 897         qemu_free(bp);
 898     } else {
 899         err = kvm_arch_remove_hw_breakpoint(addr, len, type);
 900         if (err)
 901             return err;
 902     }
 903
 904     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 905         err = kvm_update_guest_debug(env, 0);
 906         if (err)
 907             return err;
 908     }
 909     return 0;
 910 }
 911
 912 void kvm_remove_all_breakpoints(CPUState *current_env)
 913 {
 914     struct kvm_sw_breakpoint *bp, *next;
 915     KVMState *s = current_env->kvm_state;
 916     CPUState *env;
 917
 918     TAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
 919         if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) {
 920             /* Try harder to find a CPU that currently sees the breakpoint. */
 921             for (env = first_cpu; env != NULL; env = env->next_cpu) {
 922                 if (kvm_arch_remove_sw_breakpoint(env, bp) == 0)
 923                     break;
 924             }
 925         }
 926     }
 927     kvm_arch_remove_all_hw_breakpoints();
 928
 929     for (env = first_cpu; env != NULL; env = env->next_cpu)
 930         kvm_update_guest_debug(env, 0);
 931 }
 932
 933 #else /* !KVM_CAP_SET_GUEST_DEBUG */
 934
 935 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 936 {
 937     return -EINVAL;
 938 }
 939
 940 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 941                           target_ulong len, int type)
 942 {
 943     return -EINVAL;
 944 }
 945
 946 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
 947                           target_ulong len, int type)
 948 {
 949     return -EINVAL;
 950 }
 951
 952 void kvm_remove_all_breakpoints(CPUState *current_env)
 953 {
 954 }
 955 #endif /* !KVM_CAP_SET_GUEST_DEBUG */