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_init(int smp_cpus)
 324 {
 325     KVMState *s;
 326     int ret;
 327     int i;
 328
 329     if (smp_cpus > 1)
 330         return -EINVAL;
 331
 332     s = qemu_mallocz(sizeof(KVMState));
 333
 334 #ifdef KVM_CAP_SET_GUEST_DEBUG
 335     TAILQ_INIT(&s->kvm_sw_breakpoints);
 336 #endif
 337     for (i = 0; i < ARRAY_SIZE(s->slots); i++)
 338         s->slots[i].slot = i;
 339
 340     s->vmfd = -1;
 341     s->fd = open("/dev/kvm", O_RDWR);
 342     if (s->fd == -1) {
 343         fprintf(stderr, "Could not access KVM kernel module: %m\n");
 344         ret = -errno;
 345         goto err;
 346     }
 347
 348     ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
 349     if (ret < KVM_API_VERSION) {
 350         if (ret > 0)
 351             ret = -EINVAL;
 352         fprintf(stderr, "kvm version too old\n");
 353         goto err;
 354     }
 355
 356     if (ret > KVM_API_VERSION) {
 357         ret = -EINVAL;
 358         fprintf(stderr, "kvm version not supported\n");
 359         goto err;
 360     }
 361
 362     s->vmfd = kvm_ioctl(s, KVM_CREATE_VM, 0);
 363     if (s->vmfd < 0)
 364         goto err;
 365
 366     /* initially, KVM allocated its own memory and we had to jump through
 367      * hooks to make phys_ram_base point to this.  Modern versions of KVM
 368      * just use a user allocated buffer so we can use regular pages
 369      * unmodified.  Make sure we have a sufficiently modern version of KVM.
 370      */
 371     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_USER_MEMORY);
 372     if (ret <= 0) {
 373         if (ret == 0)
 374             ret = -EINVAL;
 375         fprintf(stderr, "kvm does not support KVM_CAP_USER_MEMORY\n");
 376         goto err;
 377     }
 378
 379     /* There was a nasty bug in < kvm-80 that prevents memory slots from being
 380      * destroyed properly.  Since we rely on this capability, refuse to work
 381      * with any kernel without this capability. */
 382     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION,
 383                     KVM_CAP_DESTROY_MEMORY_REGION_WORKS);
 384     if (ret <= 0) {
 385         if (ret == 0)
 386             ret = -EINVAL;
 387
 388         fprintf(stderr,
 389                 "KVM kernel module broken (DESTROY_MEMORY_REGION)\n"
 390                 "Please upgrade to at least kvm-81.\n");
 391         goto err;
 392     }
 393
 394     s->coalesced_mmio = 0;
 395 #ifdef KVM_CAP_COALESCED_MMIO
 396     ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_COALESCED_MMIO);
 397     if (ret > 0)
 398         s->coalesced_mmio = ret;
 399 #endif
 400
 401     ret = kvm_arch_init(s, smp_cpus);
 402     if (ret < 0)
 403         goto err;
 404
 405     kvm_state = s;
 406
 407     return 0;
 408
 409 err:
 410     if (s) {
 411         if (s->vmfd != -1)
 412             close(s->vmfd);
 413         if (s->fd != -1)
 414             close(s->fd);
 415     }
 416     qemu_free(s);
 417
 418     return ret;
 419 }
 420
 421 static int kvm_handle_io(CPUState *env, uint16_t port, void *data,
 422                          int direction, int size, uint32_t count)
 423 {
 424     int i;
 425     uint8_t *ptr = data;
 426
 427     for (i = 0; i < count; i++) {
 428         if (direction == KVM_EXIT_IO_IN) {
 429             switch (size) {
 430             case 1:
 431                 stb_p(ptr, cpu_inb(env, port));
 432                 break;
 433             case 2:
 434                 stw_p(ptr, cpu_inw(env, port));
 435                 break;
 436             case 4:
 437                 stl_p(ptr, cpu_inl(env, port));
 438                 break;
 439             }
 440         } else {
 441             switch (size) {
 442             case 1:
 443                 cpu_outb(env, port, ldub_p(ptr));
 444                 break;
 445             case 2:
 446                 cpu_outw(env, port, lduw_p(ptr));
 447                 break;
 448             case 4:
 449                 cpu_outl(env, port, ldl_p(ptr));
 450                 break;
 451             }
 452         }
 453
 454         ptr += size;
 455     }
 456
 457     return 1;
 458 }
 459
 460 static void kvm_run_coalesced_mmio(CPUState *env, struct kvm_run *run)
 461 {
 462 #ifdef KVM_CAP_COALESCED_MMIO
 463     KVMState *s = kvm_state;
 464     if (s->coalesced_mmio) {
 465         struct kvm_coalesced_mmio_ring *ring;
 466
 467         ring = (void *)run + (s->coalesced_mmio * TARGET_PAGE_SIZE);
 468         while (ring->first != ring->last) {
 469             struct kvm_coalesced_mmio *ent;
 470
 471             ent = &ring->coalesced_mmio[ring->first];
 472
 473             cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
 474             /* FIXME smp_wmb() */
 475             ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
 476         }
 477     }
 478 #endif
 479 }
 480
 481 int kvm_cpu_exec(CPUState *env)
 482 {
 483     struct kvm_run *run = env->kvm_run;
 484     int ret;
 485
 486     dprintf("kvm_cpu_exec()\n");
 487
 488     do {
 489         kvm_arch_pre_run(env, run);
 490
 491         if (env->exit_request) {
 492             dprintf("interrupt exit requested\n");
 493             ret = 0;
 494             break;
 495         }
 496
 497         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
 498         kvm_arch_post_run(env, run);
 499
 500         if (ret == -EINTR || ret == -EAGAIN) {
 501             dprintf("io window exit\n");
 502             ret = 0;
 503             break;
 504         }
 505
 506         if (ret < 0) {
 507             dprintf("kvm run failed %s\n", strerror(-ret));
 508             abort();
 509         }
 510
 511         kvm_run_coalesced_mmio(env, run);
 512
 513         ret = 0; /* exit loop */
 514         switch (run->exit_reason) {
 515         case KVM_EXIT_IO:
 516             dprintf("handle_io\n");
 517             ret = kvm_handle_io(env, run->io.port,
 518                                 (uint8_t *)run + run->io.data_offset,
 519                                 run->io.direction,
 520                                 run->io.size,
 521                                 run->io.count);
 522             break;
 523         case KVM_EXIT_MMIO:
 524             dprintf("handle_mmio\n");
 525             cpu_physical_memory_rw(run->mmio.phys_addr,
 526                                    run->mmio.data,
 527                                    run->mmio.len,
 528                                    run->mmio.is_write);
 529             ret = 1;
 530             break;
 531         case KVM_EXIT_IRQ_WINDOW_OPEN:
 532             dprintf("irq_window_open\n");
 533             break;
 534         case KVM_EXIT_SHUTDOWN:
 535             dprintf("shutdown\n");
 536             qemu_system_reset_request();
 537             ret = 1;
 538             break;
 539         case KVM_EXIT_UNKNOWN:
 540             dprintf("kvm_exit_unknown\n");
 541             break;
 542         case KVM_EXIT_FAIL_ENTRY:
 543             dprintf("kvm_exit_fail_entry\n");
 544             break;
 545         case KVM_EXIT_EXCEPTION:
 546             dprintf("kvm_exit_exception\n");
 547             break;
 548         case KVM_EXIT_DEBUG:
 549             dprintf("kvm_exit_debug\n");
 550 #ifdef KVM_CAP_SET_GUEST_DEBUG
 551             if (kvm_arch_debug(&run->debug.arch)) {
 552                 gdb_set_stop_cpu(env);
 553                 vm_stop(EXCP_DEBUG);
 554                 env->exception_index = EXCP_DEBUG;
 555                 return 0;
 556             }
 557             /* re-enter, this exception was guest-internal */
 558             ret = 1;
 559 #endif /* KVM_CAP_SET_GUEST_DEBUG */
 560             break;
 561         default:
 562             dprintf("kvm_arch_handle_exit\n");
 563             ret = kvm_arch_handle_exit(env, run);
 564             break;
 565         }
 566     } while (ret > 0);
 567
 568     if (env->exit_request) {
 569         env->exit_request = 0;
 570         env->exception_index = EXCP_INTERRUPT;
 571     }
 572
 573     return ret;
 574 }
 575
 576 void kvm_set_phys_mem(target_phys_addr_t start_addr,
 577                       ram_addr_t size,
 578                       ram_addr_t phys_offset)
 579 {
 580     KVMState *s = kvm_state;
 581     ram_addr_t flags = phys_offset & ~TARGET_PAGE_MASK;
 582     KVMSlot *mem, old;
 583     int err;
 584
 585     if (start_addr & ~TARGET_PAGE_MASK) {
 586         if (flags >= IO_MEM_UNASSIGNED) {
 587             if (!kvm_lookup_overlapping_slot(s, start_addr,
 588                                              start_addr + size)) {
 589                 return;
 590             }
 591             fprintf(stderr, "Unaligned split of a KVM memory slot\n");
 592         } else {
 593             fprintf(stderr, "Only page-aligned memory slots supported\n");
 594         }
 595         abort();
 596     }
 597
 598     /* KVM does not support read-only slots */
 599     phys_offset &= ~IO_MEM_ROM;
 600
 601     while (1) {
 602         mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
 603         if (!mem) {
 604             break;
 605         }
 606
 607         if (flags < IO_MEM_UNASSIGNED && start_addr >= mem->start_addr &&
 608             (start_addr + size <= mem->start_addr + mem->memory_size) &&
 609             (phys_offset - start_addr == mem->phys_offset - mem->start_addr)) {
 610             /* The new slot fits into the existing one and comes with
 611              * identical parameters - nothing to be done. */
 612             return;
 613         }
 614
 615         old = *mem;
 616
 617         /* unregister the overlapping slot */
 618         mem->memory_size = 0;
 619         err = kvm_set_user_memory_region(s, mem);
 620         if (err) {
 621             fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
 622                     __func__, strerror(-err));
 623             abort();
 624         }
 625
 626         /* Workaround for older KVM versions: we can't join slots, even not by
 627          * unregistering the previous ones and then registering the larger
 628          * slot. We have to maintain the existing fragmentation. Sigh.
 629          *
 630          * This workaround assumes that the new slot starts at the same
 631          * address as the first existing one. If not or if some overlapping
 632          * slot comes around later, we will fail (not seen in practice so far)
 633          * - and actually require a recent KVM version. */
 634         if (old.start_addr == start_addr && old.memory_size < size &&
 635             flags < IO_MEM_UNASSIGNED) {
 636             mem = kvm_alloc_slot(s);
 637             mem->memory_size = old.memory_size;
 638             mem->start_addr = old.start_addr;
 639             mem->phys_offset = old.phys_offset;
 640             mem->flags = 0;
 641
 642             err = kvm_set_user_memory_region(s, mem);
 643             if (err) {
 644                 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
 645                         strerror(-err));
 646                 abort();
 647             }
 648
 649             start_addr += old.memory_size;
 650             phys_offset += old.memory_size;
 651             size -= old.memory_size;
 652             continue;
 653         }
 654
 655         /* register prefix slot */
 656         if (old.start_addr < start_addr) {
 657             mem = kvm_alloc_slot(s);
 658             mem->memory_size = start_addr - old.start_addr;
 659             mem->start_addr = old.start_addr;
 660             mem->phys_offset = old.phys_offset;
 661             mem->flags = 0;
 662
 663             err = kvm_set_user_memory_region(s, mem);
 664             if (err) {
 665                 fprintf(stderr, "%s: error registering prefix slot: %s\n",
 666                         __func__, strerror(-err));
 667                 abort();
 668             }
 669         }
 670
 671         /* register suffix slot */
 672         if (old.start_addr + old.memory_size > start_addr + size) {
 673             ram_addr_t size_delta;
 674
 675             mem = kvm_alloc_slot(s);
 676             mem->start_addr = start_addr + size;
 677             size_delta = mem->start_addr - old.start_addr;
 678             mem->memory_size = old.memory_size - size_delta;
 679             mem->phys_offset = old.phys_offset + size_delta;
 680             mem->flags = 0;
 681
 682             err = kvm_set_user_memory_region(s, mem);
 683             if (err) {
 684                 fprintf(stderr, "%s: error registering suffix slot: %s\n",
 685                         __func__, strerror(-err));
 686                 abort();
 687             }
 688         }
 689     }
 690
 691     /* in case the KVM bug workaround already "consumed" the new slot */
 692     if (!size)
 693         return;
 694
 695     /* KVM does not need to know about this memory */
 696     if (flags >= IO_MEM_UNASSIGNED)
 697         return;
 698
 699     mem = kvm_alloc_slot(s);
 700     mem->memory_size = size;
 701     mem->start_addr = start_addr;
 702     mem->phys_offset = phys_offset;
 703     mem->flags = 0;
 704
 705     err = kvm_set_user_memory_region(s, mem);
 706     if (err) {
 707         fprintf(stderr, "%s: error registering slot: %s\n", __func__,
 708                 strerror(-err));
 709         abort();
 710     }
 711 }
 712
 713 int kvm_ioctl(KVMState *s, int type, ...)
 714 {
 715     int ret;
 716     void *arg;
 717     va_list ap;
 718
 719     va_start(ap, type);
 720     arg = va_arg(ap, void *);
 721     va_end(ap);
 722
 723     ret = ioctl(s->fd, type, arg);
 724     if (ret == -1)
 725         ret = -errno;
 726
 727     return ret;
 728 }
 729
 730 int kvm_vm_ioctl(KVMState *s, int type, ...)
 731 {
 732     int ret;
 733     void *arg;
 734     va_list ap;
 735
 736     va_start(ap, type);
 737     arg = va_arg(ap, void *);
 738     va_end(ap);
 739
 740     ret = ioctl(s->vmfd, type, arg);
 741     if (ret == -1)
 742         ret = -errno;
 743
 744     return ret;
 745 }
 746
 747 int kvm_vcpu_ioctl(CPUState *env, int type, ...)
 748 {
 749     int ret;
 750     void *arg;
 751     va_list ap;
 752
 753     va_start(ap, type);
 754     arg = va_arg(ap, void *);
 755     va_end(ap);
 756
 757     ret = ioctl(env->kvm_fd, type, arg);
 758     if (ret == -1)
 759         ret = -errno;
 760
 761     return ret;
 762 }
 763
 764 int kvm_has_sync_mmu(void)
 765 {
 766 #ifdef KVM_CAP_SYNC_MMU
 767     KVMState *s = kvm_state;
 768
 769     if (kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_SYNC_MMU) > 0)
 770         return 1;
 771 #endif
 772
 773     return 0;
 774 }
 775
 776 void kvm_setup_guest_memory(void *start, size_t size)
 777 {
 778     if (!kvm_has_sync_mmu()) {
 779 #ifdef MADV_DONTFORK
 780         int ret = madvise(start, size, MADV_DONTFORK);
 781
 782         if (ret) {
 783             perror("madvice");
 784             exit(1);
 785         }
 786 #else
 787         fprintf(stderr,
 788                 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
 789         exit(1);
 790 #endif
 791     }
 792 }
 793
 794 #ifdef KVM_CAP_SET_GUEST_DEBUG
 795 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
 796                                                  target_ulong pc)
 797 {
 798     struct kvm_sw_breakpoint *bp;
 799
 800     TAILQ_FOREACH(bp, &env->kvm_state->kvm_sw_breakpoints, entry) {
 801         if (bp->pc == pc)
 802             return bp;
 803     }
 804     return NULL;
 805 }
 806
 807 int kvm_sw_breakpoints_active(CPUState *env)
 808 {
 809     return !TAILQ_EMPTY(&env->kvm_state->kvm_sw_breakpoints);
 810 }
 811
 812 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 813 {
 814     struct kvm_guest_debug dbg;
 815
 816     dbg.control = 0;
 817     if (env->singlestep_enabled)
 818         dbg.control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
 819
 820     kvm_arch_update_guest_debug(env, &dbg);
 821     dbg.control |= reinject_trap;
 822
 823     return kvm_vcpu_ioctl(env, KVM_SET_GUEST_DEBUG, &dbg);
 824 }
 825
 826 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 827                           target_ulong len, int type)
 828 {
 829     struct kvm_sw_breakpoint *bp;
 830     CPUState *env;
 831     int err;
 832
 833     if (type == GDB_BREAKPOINT_SW) {
 834         bp = kvm_find_sw_breakpoint(current_env, addr);
 835         if (bp) {
 836             bp->use_count++;
 837             return 0;
 838         }
 839
 840         bp = qemu_malloc(sizeof(struct kvm_sw_breakpoint));
 841         if (!bp)
 842             return -ENOMEM;
 843
 844         bp->pc = addr;
 845         bp->use_count = 1;
 846         err = kvm_arch_insert_sw_breakpoint(current_env, bp);
 847         if (err) {
 848             free(bp);
 849             return err;
 850         }
 851
 852         TAILQ_INSERT_HEAD(&current_env->kvm_state->kvm_sw_breakpoints,
 853                           bp, entry);
 854     } else {
 855         err = kvm_arch_insert_hw_breakpoint(addr, len, type);
 856         if (err)
 857             return err;
 858     }
 859
 860     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 861         err = kvm_update_guest_debug(env, 0);
 862         if (err)
 863             return err;
 864     }
 865     return 0;
 866 }
 867
 868 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
 869                           target_ulong len, int type)
 870 {
 871     struct kvm_sw_breakpoint *bp;
 872     CPUState *env;
 873     int err;
 874
 875     if (type == GDB_BREAKPOINT_SW) {
 876         bp = kvm_find_sw_breakpoint(current_env, addr);
 877         if (!bp)
 878             return -ENOENT;
 879
 880         if (bp->use_count > 1) {
 881             bp->use_count--;
 882             return 0;
 883         }
 884
 885         err = kvm_arch_remove_sw_breakpoint(current_env, bp);
 886         if (err)
 887             return err;
 888
 889         TAILQ_REMOVE(&current_env->kvm_state->kvm_sw_breakpoints, bp, entry);
 890         qemu_free(bp);
 891     } else {
 892         err = kvm_arch_remove_hw_breakpoint(addr, len, type);
 893         if (err)
 894             return err;
 895     }
 896
 897     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 898         err = kvm_update_guest_debug(env, 0);
 899         if (err)
 900             return err;
 901     }
 902     return 0;
 903 }
 904
 905 void kvm_remove_all_breakpoints(CPUState *current_env)
 906 {
 907     struct kvm_sw_breakpoint *bp, *next;
 908     KVMState *s = current_env->kvm_state;
 909     CPUState *env;
 910
 911     TAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
 912         if (kvm_arch_remove_sw_breakpoint(current_env, bp) != 0) {
 913             /* Try harder to find a CPU that currently sees the breakpoint. */
 914             for (env = first_cpu; env != NULL; env = env->next_cpu) {
 915                 if (kvm_arch_remove_sw_breakpoint(env, bp) == 0)
 916                     break;
 917             }
 918         }
 919     }
 920     kvm_arch_remove_all_hw_breakpoints();
 921
 922     for (env = first_cpu; env != NULL; env = env->next_cpu)
 923         kvm_update_guest_debug(env, 0);
 924 }
 925
 926 #else /* !KVM_CAP_SET_GUEST_DEBUG */
 927
 928 int kvm_update_guest_debug(CPUState *env, unsigned long reinject_trap)
 929 {
 930     return -EINVAL;
 931 }
 932
 933 int kvm_insert_breakpoint(CPUState *current_env, target_ulong addr,
 934                           target_ulong len, int type)
 935 {
 936     return -EINVAL;
 937 }
 938
 939 int kvm_remove_breakpoint(CPUState *current_env, target_ulong addr,
 940                           target_ulong len, int type)
 941 {
 942     return -EINVAL;
 943 }
 944
 945 void kvm_remove_all_breakpoints(CPUState *current_env)
 946 {
 947 }
 948 #endif /* !KVM_CAP_SET_GUEST_DEBUG */