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     if (kvm_arch_put_registers(env)) {
 154         fprintf(stderr, "Fatal: kvm vcpu reset failed\n");
 155         abort();
 156     }
 157 }
 158 #endif
 159
 160 int kvm_irqchip_in_kernel(void)
 161 {
 162     return kvm_state->irqchip_in_kernel;
 163 }
 164
 165 #ifdef KVM_UPSTREAM
 166 int kvm_pit_in_kernel(void)
 167 {
 168     return kvm_state->pit_in_kernel;
 169 }
 170
 171
 172 int kvm_init_vcpu(CPUState *env)
 173 {
 174     KVMState *s = kvm_state;
 175     long mmap_size;
 176     int ret;
 177
 178     dprintf("kvm_init_vcpu\n");
 179
 180     ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, env->cpu_index);
 181     if (ret < 0) {
 182         dprintf("kvm_create_vcpu failed\n");
 183         goto err;
 184     }
 185
 186     env->kvm_fd = ret;
 187     env->kvm_state = s;
 188
 189     mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
 190     if (mmap_size < 0) {
 191         dprintf("KVM_GET_VCPU_MMAP_SIZE failed\n");
 192         goto err;
 193     }
 194
 195     env->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
 196                         env->kvm_fd, 0);
 197     if (env->kvm_run == MAP_FAILED) {
 198         ret = -errno;
 199         dprintf("mmap'ing vcpu state failed\n");
 200         goto err;
 201     }
 202
 203     ret = kvm_arch_init_vcpu(env);
 204     if (ret == 0) {
 205         qemu_register_reset(kvm_reset_vcpu, env);
 206         ret = kvm_arch_put_registers(env);
 207     }
 208 err:
 209     return ret;
 210 }
 211
 212 int kvm_put_mp_state(CPUState *env)
 213 {
 214     struct kvm_mp_state mp_state = { .mp_state = env->mp_state };
 215
 216     return kvm_vcpu_ioctl(env, KVM_SET_MP_STATE, &mp_state);
 217 }
 218
 219 int kvm_get_mp_state(CPUState *env)
 220 {
 221     struct kvm_mp_state mp_state;
 222     int ret;
 223
 224     ret = kvm_vcpu_ioctl(env, KVM_GET_MP_STATE, &mp_state);
 225     if (ret < 0) {
 226         return ret;
 227     }
 228     env->mp_state = mp_state.mp_state;
 229     return 0;
 230 }
 231
 232 /*
 233  * dirty pages logging control
 234  */
 235 static int kvm_dirty_pages_log_change(target_phys_addr_t phys_addr,
 236                                       ram_addr_t size, int flags, int mask)
 237 {
 238     KVMState *s = kvm_state;
 239     KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
 240     int old_flags;
 241
 242     if (mem == NULL)  {
 243             fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
 244                     TARGET_FMT_plx "\n", __func__, phys_addr,
 245                     (target_phys_addr_t)(phys_addr + size - 1));
 246             return -EINVAL;
 247     }
 248
 249     old_flags = mem->flags;
 250
 251     flags = (mem->flags & ~mask) | flags;
 252     mem->flags = flags;
 253
 254     /* If nothing changed effectively, no need to issue ioctl */
 255     if (s->migration_log) {
 256         flags |= KVM_MEM_LOG_DIRTY_PAGES;
 257     }
 258     if (flags == old_flags) {
 259             return 0;
 260     }
 261
 262     return kvm_set_user_memory_region(s, mem);
 263 }
 264
 265 int kvm_log_start(target_phys_addr_t phys_addr, ram_addr_t size)
 266 {
 267         return kvm_dirty_pages_log_change(phys_addr, size,
 268                                           KVM_MEM_LOG_DIRTY_PAGES,
 269                                           KVM_MEM_LOG_DIRTY_PAGES);
 270 }
 271
 272 int kvm_log_stop(target_phys_addr_t phys_addr, ram_addr_t size)
 273 {
 274         return kvm_dirty_pages_log_change(phys_addr, size,
 275                                           0,
 276                                           KVM_MEM_LOG_DIRTY_PAGES);
 277 }
 278
 279 int kvm_set_migration_log(int enable)
 280 {
 281     KVMState *s = kvm_state;
 282     KVMSlot *mem;
 283     int i, err;
 284
 285     s->migration_log = enable;
 286
 287     for (i = 0; i < ARRAY_SIZE(s->slots); i++) {
 288         mem = &s->slots[i];
 289
 290         if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
 291             continue;
 292         }
 293         err = kvm_set_user_memory_region(s, mem);
 294         if (err) {
 295             return err;
 296         }
 297     }
 298     return 0;
 299 }
 300
 301 static int test_le_bit(unsigned long nr, unsigned char *addr)
 302 {
 303     return (addr[nr >> 3] >> (nr & 7)) & 1;
 304 }
 305
 306 /**
 307  * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
 308  * This function updates qemu's dirty bitmap using cpu_physical_memory_set_dirty().
 309  * This means all bits are set to dirty.
 310  *
 311  * @start_add: start of logged region.
 312  * @end_addr: end of logged region.
 313  */
 314 int kvm_physical_sync_dirty_bitmap(target_phys_addr_t start_addr,
 315                                    target_phys_addr_t end_addr)
 316 {
 317     KVMState *s = kvm_state;
 318     unsigned long size, allocated_size = 0;
 319     target_phys_addr_t phys_addr;
 320     ram_addr_t addr;
 321     KVMDirtyLog d;
 322     KVMSlot *mem;
 323     int ret = 0;
 324
 325     d.dirty_bitmap = NULL;
 326     while (start_addr < end_addr) {
 327         mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
 328         if (mem == NULL) {
 329             break;
 330         }
 331
 332         size = ((mem->memory_size >> TARGET_PAGE_BITS) + 7) / 8;
 333         if (!d.dirty_bitmap) {
 334             d.dirty_bitmap = qemu_malloc(size);
 335         } else if (size > allocated_size) {
 336             d.dirty_bitmap = qemu_realloc(d.dirty_bitmap, size);
 337         }
 338         allocated_size = size;
 339         memset(d.dirty_bitmap, 0, allocated_size);
 340
 341         d.slot = mem->slot;
 342
 343         if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
 344             dprintf("ioctl failed %d\n", errno);
 345             ret = -1;
 346             break;
 347         }
 348
 349         for (phys_addr = mem->start_addr, addr = mem->phys_offset;
 350              phys_addr < mem->start_addr + mem->memory_size;
 351              phys_addr += TARGET_PAGE_SIZE, addr += TARGET_PAGE_SIZE) {
 352             unsigned char *bitmap = (unsigned char *)d.dirty_bitmap;
 353             unsigned nr = (phys_addr - mem->start_addr) >> TARGET_PAGE_BITS;
 354
 355             if (test_le_bit(nr, bitmap)) {
 356                 cpu_physical_memory_set_dirty(addr);
 357             }
 358         }
 359         start_addr = phys_addr;
 360     }
 361     qemu_free(d.dirty_bitmap);
 362
 363     return ret;
 364 }
 365 #endif
 366
 367 int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 368 {
 369     int ret = -ENOSYS;
 370 #ifdef KVM_CAP_COALESCED_MMIO
 371     KVMState *s = kvm_state;
 372
 373     if (s->coalesced_mmio) {
 374         struct kvm_coalesced_mmio_zone zone;
 375
 376         zone.addr = start;
 377         zone.size = size;
 378
 379         ret = kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
 380     }
 381 #endif
 382
 383     return ret;
 384 }
 385
 386 int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size)
 387 {
 388     int ret = -ENOSYS;
 389 #ifdef KVM_CAP_COALESCED_MMIO
 390     KVMState *s = kvm_state;
 391
 392     if (s->coalesced_mmio) {
 393         struct kvm_coalesced_mmio_zone zone;
 394
 395         zone.addr = start;
 396         zone.size = size;
 397
 398         ret = kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
 399     }
 400 #endif
 401
 402     return ret;
 403 }
 404
 405 int kvm_check_extension(KVMState *s, unsigned int extension)
 406 {
 407     int ret;
 408
 409     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
 410     if (ret < 0) {
 411         ret = 0;
 412     }
 413
 414     return ret;
 415 }
 416 #ifdef KVM_UPSTREAM
 417
 418 int kvm_init(int smp_cpus)
 419 {
 420     static const char upgrade_note[] =
 421         "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
 422         "(see http://sourceforge.net/projects/kvm).\n";
 423     KVMState *s;
 424     int ret;
 425     int i;
 426
 427     if (smp_cpus > 1) {
 428         fprintf(stderr, "No SMP KVM support, use '-smp 1'\n");
 429         return -EINVAL;
 430     }
 431
 432     s = qemu_mallocz(sizeof(KVMState));
 433
 434 #ifdef KVM_CAP_SET_GUEST_DEBUG
 435     QTAILQ_INIT(&s->kvm_sw_breakpoints);
 436 #endif
 437     for (i = 0; i < ARRAY_SIZE(s->slots); i++)
 438         s->slots[i].slot = i;
 439
 440     s->vmfd = -1;
 441     s->fd = open("/dev/kvm", O_RDWR);
 442     if (s->fd == -1) {
 443         fprintf(stderr, "Could not access KVM kernel module: %m\n");
 444         ret = -errno;
 445         goto err;
 446     }
 447
 448     ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
 449     if (ret < KVM_API_VERSION) {
 450         if (ret > 0)
 451             ret = -EINVAL;
 452         fprintf(stderr, "kvm version too old\n");
 453         goto err;
 454     }
 455
 456     if (ret > KVM_API_VERSION) {
 457         ret = -EINVAL;
 458         fprintf(stderr, "kvm version not supported\n");
 459         goto err;
 460     }
 461
 462     s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
 463     if (s->vmfd < 0)
 464         goto err;
 465
 466     /* initially, KVM allocated its own memory and we had to jump through
 467      * hooks to make phys_ram_base point to this.  Modern versions of KVM
 468      * just use a user allocated buffer so we can use regular pages
 469      * unmodified.  Make sure we have a sufficiently modern version of KVM.
 470      */
 471     if (!kvm_check_extension(s, KVM_CAP_USER_MEMORY)) {
 472         ret = -EINVAL;
 473         fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n%s",
 474                 upgrade_note);
 475         goto err;
 476     }
 477
 478     /* There was a nasty bug in < kvm-80 that prevents memory slots from being
 479      * destroyed properly.  Since we rely on this capability, refuse to work
 480      * with any kernel without this capability. */
 481     if (!kvm_check_extension(s, KVM_CAP_DESTROY_MEMORY_REGION_WORKS)) {
 482         ret = -EINVAL;
 483
 484         fprintf(stderr,
 485                 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
 486                 upgrade_note);
 487         goto err;
 488     }
 489
 490 #ifdef KVM_CAP_COALESCED_MMIO
 491     s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
 492 #else
 493     s->coalesced_mmio = 0;
 494 #endif
 495
 496     s->broken_set_mem_region = 1;
 497 #ifdef KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
 498     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
 499     if (ret > 0) {
 500         s->broken_set_mem_region = 0;
 501     }
 502 #endif
 503
 504     ret = kvm_arch_init(s, smp_cpus);
 505     if (ret < 0)
 506         goto err;
 507
 508     kvm_state = s;
 509
 510     return 0;
 511
 512 err:
 513     if (s) {
 514         if (s->vmfd != -1)
 515             close(s->vmfd);
 516         if (s->fd != -1)
 517             close(s->fd);
 518     }
 519     qemu_free(s);
 520
 521     return ret;
 522 }
 523
 524 static int kvm_handle_io(uint16_t port, void *data, int direction, int size,
 525                          uint32_t count)
 526 {
 527     int i;
 528     uint8_t *ptr = data;
 529
 530     for (i = 0; i < count; i++) {
 531         if (direction == KVM_EXIT_IO_IN) {
 532             switch (size) {
 533             case 1:
 534                 stb_p(ptr, cpu_inb(port));
 535                 break;
 536             case 2:
 537                 stw_p(ptr, cpu_inw(port));
 538                 break;
 539             case 4:
 540                 stl_p(ptr, cpu_inl(port));
 541                 break;
 542             }
 543         } else {
 544             switch (size) {
 545             case 1:
 546                 cpu_outb(port, ldub_p(ptr));
 547                 break;
 548             case 2:
 549                 cpu_outw(port, lduw_p(ptr));
 550                 break;
 551             case 4:
 552                 cpu_outl(port, ldl_p(ptr));
 553                 break;
 554             }
 555         }
 556
 557         ptr += size;
 558     }
 559
 560     return 1;
 561 }
 562
 563 static void kvm_run_coalesced_mmio(CPUState *env, struct kvm_run *run)
 564 {
 565 #ifdef KVM_CAP_COALESCED_MMIO
 566     KVMState *s = kvm_state;
 567     if (s->coalesced_mmio) {
 568         struct kvm_coalesced_mmio_ring *ring;
 569
 570         ring = (void *)run + (s->coalesced_mmio * TARGET_PAGE_SIZE);
 571         while (ring->first != ring->last) {
 572             struct kvm_coalesced_mmio *ent;
 573
 574             ent = &ring->coalesced_mmio[ring->first];
 575
 576             cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
 577             /* FIXME smp_wmb() */
 578             ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
 579         }
 580     }
 581 #endif
 582 }
 583
 584 void kvm_cpu_synchronize_state(CPUState *env)
 585 {
 586     if (!env->kvm_state->regs_modified) {
 587         kvm_arch_get_registers(env);
 588         env->kvm_state->regs_modified = 1;
 589     }
 590 }
 591
 592 int kvm_cpu_exec(CPUState *env)
 593 {
 594     struct kvm_run *run = env->kvm_run;
 595     int ret;
 596
 597     dprintf("kvm_cpu_exec()\n");
 598
 599     do {
 600         if (env->exit_request) {
 601             dprintf("interrupt exit requested\n");
 602             ret = 0;
 603             break;
 604         }
 605
 606         if (env->kvm_state->regs_modified) {
 607             kvm_arch_put_registers(env);
 608             env->kvm_state->regs_modified = 0;
 609         }
 610
 611         kvm_arch_pre_run(env, run);
 612         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
 613         kvm_arch_post_run(env, run);
 614
 615         if (ret == -EINTR || ret == -EAGAIN) {
 616             dprintf("io window exit\n");
 617             ret = 0;
 618             break;
 619         }
 620
 621         if (ret < 0) {
 622             dprintf("kvm run failed %s\n", strerror(-ret));
 623             abort();
 624         }
 625
 626         kvm_run_coalesced_mmio(env, run);
 627
 628         ret = 0; /* exit loop */
 629         switch (run->exit_reason) {
 630         case KVM_EXIT_IO:
 631             dprintf("handle_io\n");
 632             ret = kvm_handle_io(run->io.port,
 633                                 (uint8_t *)run + run->io.data_offset,
 634                                 run->io.direction,
 635                                 run->io.size,
 636                                 run->io.count);
 637             break;
 638         case KVM_EXIT_MMIO:
 639             dprintf("handle_mmio\n");
 640             cpu_physical_memory_rw(run->mmio.phys_addr,
 641                                    run->mmio.data,
 642                                    run->mmio.len,
 643                                    run->mmio.is_write);
 644             ret = 1;
 645             break;
 646         case KVM_EXIT_IRQ_WINDOW_OPEN:
 647             dprintf("irq_window_open\n");
 648             break;
 649         case KVM_EXIT_SHUTDOWN:
 650             dprintf("shutdown\n");
 651             qemu_system_reset_request();
 652             ret = 1;
 653             break;
 654         case KVM_EXIT_UNKNOWN:
 655             dprintf("kvm_exit_unknown\n");
 656             break;
 657         case KVM_EXIT_FAIL_ENTRY:
 658             dprintf("kvm_exit_fail_entry\n");
 659             break;
 660         case KVM_EXIT_EXCEPTION:
 661             dprintf("kvm_exit_exception\n");
 662             break;
 663         case KVM_EXIT_DEBUG:
 664             dprintf("kvm_exit_debug\n");
 665 #ifdef KVM_CAP_SET_GUEST_DEBUG
 666             if (kvm_arch_debug(&run->debug.arch)) {
 667                 gdb_set_stop_cpu(env);
 668                 vm_stop(EXCP_DEBUG);
 669                 env->exception_index = EXCP_DEBUG;
 670                 return 0;
 671             }
 672             /* re-enter, this exception was guest-internal */
 673             ret = 1;
 674 #endif /* KVM_CAP_SET_GUEST_DEBUG */
 675             break;
 676         default:
 677             dprintf("kvm_arch_handle_exit\n");
 678             ret = kvm_arch_handle_exit(env, run);
 679             break;
 680         }
 681     } while (ret > 0);
 682
 683     if (env->exit_request) {
 684         env->exit_request = 0;
 685         env->exception_index = EXCP_INTERRUPT;
 686     }
 687
 688     return ret;
 689 }
 690
 691 void kvm_set_phys_mem(target_phys_addr_t start_addr,
 692                       ram_addr_t size,
 693                       ram_addr_t phys_offset)
 694 {
 695     KVMState *s = kvm_state;
 696     ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK;
 697     KVMSlot *mem, old;
 698     int err;
 699
 700     if (start_addr & ~TARGET_PAGE_MASK) {
 701         if (flags >= IO_MEM_UNASSIGNED) {
 702             if (!kvm_lookup_overlapping_slot(s, start_addr,
 703                                              start_addr + size)) {
 704                 return;
 705             }
 706             fprintf(stderr, "Unaligned split of a KVM memory slot\n");
 707         } else {
 708             fprintf(stderr, "Only page-aligned memory slots supported\n");
 709         }
 710         abort();
 711     }
 712
 713     /* KVM does not support read-only slots */
 714     phys_offset &= ~IO_MEM_ROM;
 715
 716     while (1) {
 717         mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
 718         if (!mem) {
 719             break;
 720         }
 721
 722         if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr &&
 723             (start_addr + size <= mem->start_addr + mem->memory_size) &&
 724             (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) {
 725             /* The new slot fits into the existing one and comes with
 726              * identical parameters - nothing to be done. */
 727             return;
 728         }
 729
 730         old = *mem;
 731
 732         /* unregister the overlapping slot */
 733         mem->memory_size = 0;
 734         err = kvm_set_user_memory_region(s, mem);
 735         if (err) {
 736             fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
 737                     __func__, strerror(-err));
 738             abort();
 739         }
 740
 741         /* Workaround for older KVM versions: we can't join slots, even not by
 742          * unregistering the previous ones and then registering the larger
 743          * slot. We have to maintain the existing fragmentation. Sigh.
 744          *
 745          * This workaround assumes that the new slot starts at the same
 746          * address as the first existing one. If not or if some overlapping
 747          * slot comes around later, we will fail (not seen in practice so far)
 748          * - and actually require a recent KVM version. */
 749         if (s->broken_set_mem_region &&
 750             old.start_addr == start_addr && old.memory_size < size &&
 751             flags < IO_MEM_UNASSIGNED) {
 752             mem = kvm_alloc_slot(s);
 753             mem->memory_size = old.memory_size;
 754             mem->start_addr = old.start_addr;
 755             mem->phys_offset = old.phys_offset;
 756             mem->flags = 0;
 757
 758             err = kvm_set_user_memory_region(s, mem);
 759             if (err) {
 760                 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
 761                         strerror(-err));
 762                 abort();
 763             }
 764
 765             start_addr += old.memory_size;
 766             phys_offset += old.memory_size;
 767             size -= old.memory_size;
 768             continue;
 769         }
 770
 771         /* register prefix slot */
 772         if (old.start_addr < start_addr) {
 773             mem = kvm_alloc_slot(s);
 774             mem->memory_size = start_addr - old.start_addr;
 775             mem->start_addr = old.start_addr;
 776             mem->phys_offset = old.phys_offset;
 777             mem->flags = 0;
 778
 779             err = kvm_set_user_memory_region(s, mem);
 780             if (err) {
 781                 fprintf(stderr, "%s: error registering prefix slot: %s\n",
 782                         __func__, strerror(-err));
 783                 abort();
 784             }
 785         }
 786
 787         /* register suffix slot */
 788         if (old.start_addr + old.memory_size > start_addr + size) {
 789             ram_addr_t size_delta;
 790
 791             mem = kvm_alloc_slot(s);
 792             mem->start_addr = start_addr + size;
 793             size_delta = mem->start_addr - old.start_addr;
 794             mem->memory_size = old.memory_size - size_delta;
 795             mem->phys_offset = old.phys_offset + size_delta;
 796             mem->flags = 0;
 797
 798             err = kvm_set_user_memory_region(s, mem);
 799             if (err) {
 800                 fprintf(stderr, "%s: error registering suffix slot: %s\n",
 801                         __func__, strerror(-err));
 802                 abort();
 803             }
 804         }
 805     }
 806
 807     /* in case the KVM bug workaround already "consumed" the new slot */
 808     if (!size)
 809         return;
 810
 811     /* KVM does not need to know about this memory */
 812     if (flags >= IO_MEM_UNASSIGNED)
 813         return;
 814
 815     mem = kvm_alloc_slot(s);
 816     mem->memory_size = size;
 817     mem->start_addr = start_addr;
 818     mem->phys_offset = phys_offset;
 819     mem->flags = 0;
 820
 821     err = kvm_set_user_memory_region(s, mem);
 822     if (err) {
 823         fprintf(stderr, "%s: error registering slot: %s\n", __func__,
 824                 strerror(-err));
 825         abort();
 826     }
 827 }
 828
 829 #endif
 830 int kvm_ioctl(KVMState *s, int type, ...)
 831 {
 832     int ret;
 833     void *arg;
 834     va_list ap;
 835
 836     va_start(ap, type);
 837     arg = va_arg(ap, void *);
 838     va_end(ap);
 839
 840     ret = ioctl(s->fd, type, arg);
 841     if (ret == -1)
 842         ret = -errno;
 843
 844     return ret;
 845 }
 846
 847 int kvm_vm_ioctl(KVMState *s, int type, ...)
 848 {
 849     int ret;
 850     void *arg;
 851     va_list ap;
 852
 853     va_start(ap, type);
 854     arg = va_arg(ap, void *);
 855     va_end(ap);
 856
 857     ret = ioctl(s->vmfd, type, arg);
 858     if (ret == -1)
 859         ret = -errno;
 860
 861     return ret;
 862 }
 863
 864 #ifdef KVM_UPSTREAM
 865 int kvm_vcpu_ioctl(CPUState *env, int type, ...)
 866 {
 867     int ret;
 868     void *arg;
 869     va_list ap;
 870
 871     va_start(ap, type);
 872     arg = va_arg(ap, void *);
 873     va_end(ap);
 874
 875     ret = ioctl(env->kvm_fd, type, arg);
 876     if (ret == -1)
 877         ret = -errno;
 878
 879     return ret;
 880 }
 881
 882 #endif
 883
 884 int kvm_has_sync_mmu(void)
 885 {
 886 #ifdef KVM_CAP_SYNC_MMU
 887     KVMState *s = kvm_state;
 888
 889     return kvm_check_extension(s, KVM_CAP_SYNC_MMU);
 890 #else
 891     return 0;
 892 #endif
 893 }
 894
 895 #ifdef KVM_UPSTREAM
 896 void kvm_setup_guest_memory(void *start, size_t size)
 897 {
 898     if (!kvm_has_sync_mmu()) {
 899 #ifdef MADV_DONTFORK
 900         int ret = madvise(start, size, MADV_DONTFORK);
 901
 902         if (ret) {
 903             perror("madvice");
 904             exit(1);
 905         }
 906 #else
 907         fprintf(stderr,
 908                 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
 909         exit(1);
 910 #endif
 911     }
 912 }
 913
 914 #endif /* KVM_UPSTREAM */
 915
 916 #ifdef KVM_CAP_SET_GUEST_DEBUG
 917
 918 #ifdef KVM_UPSTREAM
 919 static void on_vcpu(CPUState *env, void (*func)(void *data), void *data)
 920 {
 921 #ifdef CONFIG_IOTHREAD
 922     if (env == cpu_single_env) {
 923         func(data);
 924         return;
 925     }
 926     abort();
 927 #else
 928     func(data);
 929 #endif
 930 }
 931 #endif /* KVM_UPSTREAM */
 932
 933 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
 934                                                  target_ulong pc)
 935 {
 936     struct kvm_sw_breakpoint *bp;
 937
 938     QTAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) {
 939         if (bp->pc == pc)
 940             return bp;
 941     }
 942     return NULL;
 943 }
 944
 945 int kvm_sw_breakpoints_active(CPUState *env)
 946 {
 947     return !QTAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints);
 948 }
 949
 950 #ifdef KVM_UPSTREAM
 951
 952 struct kvm_set_guest_debug_data {
 953     struct kvm_guest_debug dbg;
 954     CPUState *env;
 955     int err;
 956 };
 957
 958 static void kvm_invoke_set_guest_debug(void *data)
 959 {
 960     struct kvm_set_guest_debug_data *dbg_data = data;
 961     CPUState *env = dbg_data->env;
 962
 963     if (env->kvm_state->regs_modified) {
 964         kvm_arch_put_registers(env);
 965         env->kvm_state->regs_modified = 0;
 966     }
 967     dbg_data->err = kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg_data->dbg);
 968 }
 969
 970 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 971 {
 972     struct kvm_set_guest_debug_data data;
 973
 974     data.dbg.control = 0;
 975     if (env->singlestep_enabled)
 976         data.dbg.control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
 977
 978     kvm_arch_update_guest_debug(env, &data.dbg);
 979     data.dbg.control |= reinject_trap;
 980     data.env = env;
 981
 982     on_vcpu(env, kvm_invoke_set_guest_debug, &data);
 983     return data.err;
 984 }
 985 #endif
 986
 987 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 988                           target_ulong len, int type)
 989 {
 990     struct kvm_sw_breakpoint *bp;
 991     CPUState *env;
 992     int err;
 993
 994     if (type == GDB_BREAKPOINT_SW) {
 995         bp = kvm_find_sw_breakpoint(current_env, addr);
 996         if (bp) {
 997             bp->use_count++;
 998             return 0;
 999         }
1000
1001         bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint));
1002         if (!bp)
1003             return -ENOMEM;
1004
1005         bp->pc = addr;
1006         bp->use_count = 1;
1007         err = kvm_arch_insert_sw_breakpoint(current_env, bp);
1008         if (err) {
1009             free(bp);
1010             return err;
1011         }
1012
1013         QTAILQ_INSERT_HEAD(&current_env->kvm_state->kvm_sw_breakpoints,
1014                           bp, entry);
1015     } else {
1016         err = kvm_arch_insert_hw_breakpoint(addr, len, type);
1017         if (err)
1018             return err;
1019     }
1020
1021     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1022         err = kvm_update_guest_debug(env, 0);
1023         if (err)
1024             return err;
1025     }
1026     return 0;
1027 }
1028
1029 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
1030                           target_ulong len, int type)
1031 {
1032     struct kvm_sw_breakpoint *bp;
1033     CPUState *env;
1034     int err;
1035
1036     if (type == GDB_BREAKPOINT_SW) {
1037         bp = kvm_find_sw_breakpoint(current_env, addr);
1038         if (!bp)
1039             return -ENOENT;
1040
1041         if (bp->use_count > 1) {
1042             bp->use_count--;
1043             return 0;
1044         }
1045
1046         err = kvm_arch_remove_sw_breakpoint(current_env, bp);
1047         if (err)
1048             return err;
1049
1050         QTAILQ_REMOVE(&current_env->kvm_state->kvm_sw_breakpoints, bp, entry);
1051         qemu_free(bp);
1052     } else {
1053         err = kvm_arch_remove_hw_breakpoint(addr, len, type);
1054         if (err)
1055             return err;
1056     }
1057
1058     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1059         err = kvm_update_guest_debug(env, 0);
1060         if (err)
1061             return err;
1062     }
1063     return 0;
1064 }
1065
1066 void kvm_remove_all_breakpoints(CPUState *current_env)
1067 {
1068     struct kvm_sw_breakpoint *bp, *next;
1069     KVMState *s = current_env->kvm_state;
1070     CPUState *env;
1071
1072     QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
1073         if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) {
1074             /* Try harder to find a CPU that currently sees the breakpoint. */
1075             for (env = first_cpu; env != NULL; env = env->next_cpu) {
1076                 if (kvm_arch_remove_sw_breakpoint(env, bp) == 0)
1077                     break;
1078             }
1079         }
1080     }
1081     kvm_arch_remove_all_hw_breakpoints();
1082
1083     for (env = first_cpu; env != NULL; env = env->next_cpu)
1084         kvm_update_guest_debug(env, 0);
1085 }
1086
1087 #else /* !KVM_CAP_SET_GUEST_DEBUG */
1088
1089 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
1090 {
1091     return -EINVAL;
1092 }
1093
1094 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
1095                           target_ulong len, int type)
1096 {
1097     return -EINVAL;
1098 }
1099
1100 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
1101                           target_ulong len, int type)
1102 {
1103     return -EINVAL;
1104 }
1105
1106 void kvm_remove_all_breakpoints(CPUState *current_env)
1107 {
1108 }
1109 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
1110
1111 #include "qemu-kvm.c"