hw/ivshmem.c

   1 /*
   2  * Inter-VM Shared Memory PCI device.
   3  *
   4  * Author:
   5  *      Cam Macdonell <cam@cs.ualberta.ca>
   6  *
   7  * Based On: cirrus_vga.c
   8  *          Copyright (c) 2004 Fabrice Bellard
   9  *          Copyright (c) 2004 Makoto Suzuki (suzu)
  10  *
  11  *      and rtl8139.c
  12  *          Copyright (c) 2006 Igor Kovalenko
  13  *
  14  * This code is licensed under the GNU GPL v2.
  15  *
  16  * Contributions after 2012-01-13 are licensed under the terms of the
  17  * GNU GPL, version 2 or (at your option) any later version.
  18  */
  19 #include "hw.h"
  20 #include "pc.h"
  21 #include "pci.h"
  22 #include "msix.h"
  23 #include "kvm.h"
  24 #include "migration.h"
  25 #include "qerror.h"
  26 #include "event_notifier.h"
  27
  28 #include <sys/mman.h>
  29 #include <sys/types.h>
  30
  31 #define IVSHMEM_IOEVENTFD   0
  32 #define IVSHMEM_MSI     1
  33
  34 #define IVSHMEM_PEER    0
  35 #define IVSHMEM_MASTER  1
  36
  37 #define IVSHMEM_REG_BAR_SIZE 0x100
  38
  39 //#define DEBUG_IVSHMEM
  40 #ifdef DEBUG_IVSHMEM
  41 #define IVSHMEM_DPRINTF(fmt, ...)        \
  42     do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0)
  43 #else
  44 #define IVSHMEM_DPRINTF(fmt, ...)
  45 #endif
  46
  47 typedef struct Peer {
  48     int nb_eventfds;
  49     EventNotifier *eventfds;
  50 } Peer;
  51
  52 typedef struct EventfdEntry {
  53     PCIDevice *pdev;
  54     int vector;
  55 } EventfdEntry;
  56
  57 typedef struct IVShmemState {
  58     PCIDevice dev;
  59     uint32_t intrmask;
  60     uint32_t intrstatus;
  61     uint32_t doorbell;
  62
  63     CharDriverState **eventfd_chr;
  64     CharDriverState *server_chr;
  65     MemoryRegion ivshmem_mmio;
  66
  67     /* We might need to register the BAR before we actually have the memory.
  68      * So prepare a container MemoryRegion for the BAR immediately and
  69      * add a subregion when we have the memory.
  70      */
  71     MemoryRegion bar;
  72     MemoryRegion ivshmem;
  73     uint64_t ivshmem_size; /* size of shared memory region */
  74     int shm_fd; /* shared memory file descriptor */
  75
  76     Peer *peers;
  77     int nb_peers; /* how many guests we have space for */
  78     int max_peer; /* maximum numbered peer */
  79
  80     int vm_id;
  81     uint32_t vectors;
  82     uint32_t features;
  83     EventfdEntry *eventfd_table;
  84
  85     Error *migration_blocker;
  86
  87     char * shmobj;
  88     char * sizearg;
  89     char * role;
  90     int role_val;   /* scalar to avoid multiple string comparisons */
  91 } IVShmemState;
  92
  93 /* registers for the Inter-VM shared memory device */
  94 enum ivshmem_registers {
  95     INTRMASK = 0,
  96     INTRSTATUS = 4,
  97     IVPOSITION = 8,
  98     DOORBELL = 12,
  99 };
 100
 101 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 102                                                     unsigned int feature) {
 103     return (ivs->features & (1 << feature));
 104 }
 105
 106 static inline bool is_power_of_two(uint64_t x) {
 107     return (x & (x - 1)) == 0;
 108 }
 109
 110 /* accessing registers - based on rtl8139 */
 111 static void ivshmem_update_irq(IVShmemState *s, int val)
 112 {
 113     int isr;
 114     isr = (s->intrstatus & s->intrmask) & 0xffffffff;
 115
 116     /* don't print ISR resets */
 117     if (isr) {
 118         IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
 119            isr ? 1 : 0, s->intrstatus, s->intrmask);
 120     }
 121
 122     qemu_set_irq(s->dev.irq[0], (isr != 0));
 123 }
 124
 125 static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val)
 126 {
 127     IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
 128
 129     s->intrmask = val;
 130
 131     ivshmem_update_irq(s, val);
 132 }
 133
 134 static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
 135 {
 136     uint32_t ret = s->intrmask;
 137
 138     IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
 139
 140     return ret;
 141 }
 142
 143 static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val)
 144 {
 145     IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
 146
 147     s->intrstatus = val;
 148
 149     ivshmem_update_irq(s, val);
 150     return;
 151 }
 152
 153 static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
 154 {
 155     uint32_t ret = s->intrstatus;
 156
 157     /* reading ISR clears all interrupts */
 158     s->intrstatus = 0;
 159
 160     ivshmem_update_irq(s, 0);
 161
 162     return ret;
 163 }
 164
 165 static void ivshmem_io_write(void *opaque, target_phys_addr_t addr,
 166                              uint64_t val, unsigned size)
 167 {
 168     IVShmemState *s = opaque;
 169
 170     uint16_t dest = val >> 16;
 171     uint16_t vector = val & 0xff;
 172
 173     addr &= 0xfc;
 174
 175     IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr);
 176     switch (addr)
 177     {
 178         case INTRMASK:
 179             ivshmem_IntrMask_write(s, val);
 180             break;
 181
 182         case INTRSTATUS:
 183             ivshmem_IntrStatus_write(s, val);
 184             break;
 185
 186         case DOORBELL:
 187             /* check that dest VM ID is reasonable */
 188             if (dest > s->max_peer) {
 189                 IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest);
 190                 break;
 191             }
 192
 193             /* check doorbell range */
 194             if (vector < s->peers[dest].nb_eventfds) {
 195                 IVSHMEM_DPRINTF("Notifying VM %d on vector %d\n", dest, vector);
 196                 event_notifier_set(&s->peers[dest].eventfds[vector]);
 197             }
 198             break;
 199         default:
 200             IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest);
 201     }
 202 }
 203
 204 static uint64_t ivshmem_io_read(void *opaque, target_phys_addr_t addr,
 205                                 unsigned size)
 206 {
 207
 208     IVShmemState *s = opaque;
 209     uint32_t ret;
 210
 211     switch (addr)
 212     {
 213         case INTRMASK:
 214             ret = ivshmem_IntrMask_read(s);
 215             break;
 216
 217         case INTRSTATUS:
 218             ret = ivshmem_IntrStatus_read(s);
 219             break;
 220
 221         case IVPOSITION:
 222             /* return my VM ID if the memory is mapped */
 223             if (s->shm_fd > 0) {
 224                 ret = s->vm_id;
 225             } else {
 226                 ret = -1;
 227             }
 228             break;
 229
 230         default:
 231             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 232             ret = 0;
 233     }
 234
 235     return ret;
 236 }
 237
 238 static const MemoryRegionOps ivshmem_mmio_ops = {
 239     .read = ivshmem_io_read,
 240     .write = ivshmem_io_write,
 241     .endianness = DEVICE_NATIVE_ENDIAN,
 242     .impl = {
 243         .min_access_size = 4,
 244         .max_access_size = 4,
 245     },
 246 };
 247
 248 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 249 {
 250     IVShmemState *s = opaque;
 251
 252     ivshmem_IntrStatus_write(s, *buf);
 253
 254     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf);
 255 }
 256
 257 static int ivshmem_can_receive(void * opaque)
 258 {
 259     return 8;
 260 }
 261
 262 static void ivshmem_event(void *opaque, int event)
 263 {
 264     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 265 }
 266
 267 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 268
 269     EventfdEntry *entry = opaque;
 270     PCIDevice *pdev = entry->pdev;
 271
 272     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector);
 273     msix_notify(pdev, entry->vector);
 274 }
 275
 276 static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n,
 277                                                   int vector)
 278 {
 279     /* create a event character device based on the passed eventfd */
 280     IVShmemState *s = opaque;
 281     CharDriverState * chr;
 282     int eventfd = event_notifier_get_fd(n);
 283
 284     chr = qemu_chr_open_eventfd(eventfd);
 285
 286     if (chr == NULL) {
 287         fprintf(stderr, "creating eventfd for eventfd %d failed\n", eventfd);
 288         exit(-1);
 289     }
 290
 291     /* if MSI is supported we need multiple interrupts */
 292     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 293         s->eventfd_table[vector].pdev = &s->dev;
 294         s->eventfd_table[vector].vector = vector;
 295
 296         qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd,
 297                       ivshmem_event, &s->eventfd_table[vector]);
 298     } else {
 299         qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
 300                       ivshmem_event, s);
 301     }
 302
 303     return chr;
 304
 305 }
 306
 307 static int check_shm_size(IVShmemState *s, int fd) {
 308     /* check that the guest isn't going to try and map more memory than the
 309      * the object has allocated return -1 to indicate error */
 310
 311     struct stat buf;
 312
 313     fstat(fd, &buf);
 314
 315     if (s->ivshmem_size > buf.st_size) {
 316         fprintf(stderr,
 317                 "IVSHMEM ERROR: Requested memory size greater"
 318                 " than shared object size (%" PRIu64 " > %" PRIu64")\n",
 319                 s->ivshmem_size, (uint64_t)buf.st_size);
 320         return -1;
 321     } else {
 322         return 0;
 323     }
 324 }
 325
 326 /* create the shared memory BAR when we are not using the server, so we can
 327  * create the BAR and map the memory immediately */
 328 static void create_shared_memory_BAR(IVShmemState *s, int fd) {
 329
 330     void * ptr;
 331
 332     s->shm_fd = fd;
 333
 334     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 335
 336     memory_region_init_ram_ptr(&s->ivshmem, "ivshmem.bar2",
 337                                s->ivshmem_size, ptr);
 338     vmstate_register_ram(&s->ivshmem, &s->dev.qdev);
 339     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 340
 341     /* region for shared memory */
 342     pci_register_bar(&s->dev, 2, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->bar);
 343 }
 344
 345 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 346 {
 347     memory_region_add_eventfd(&s->ivshmem_mmio,
 348                               DOORBELL,
 349                               4,
 350                               true,
 351                               (posn << 16) | i,
 352                               &s->peers[posn].eventfds[i]);
 353 }
 354
 355 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 356 {
 357     memory_region_del_eventfd(&s->ivshmem_mmio,
 358                               DOORBELL,
 359                               4,
 360                               true,
 361                               (posn << 16) | i,
 362                               &s->peers[posn].eventfds[i]);
 363 }
 364
 365 static void close_guest_eventfds(IVShmemState *s, int posn)
 366 {
 367     int i, guest_curr_max;
 368
 369     guest_curr_max = s->peers[posn].nb_eventfds;
 370
 371     memory_region_transaction_begin();
 372     for (i = 0; i < guest_curr_max; i++) {
 373         ivshmem_del_eventfd(s, posn, i);
 374     }
 375     memory_region_transaction_commit();
 376     for (i = 0; i < guest_curr_max; i++) {
 377         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 378     }
 379
 380     g_free(s->peers[posn].eventfds);
 381     s->peers[posn].nb_eventfds = 0;
 382 }
 383
 384 static void setup_ioeventfds(IVShmemState *s) {
 385
 386     int i, j;
 387
 388     for (i = 0; i <= s->max_peer; i++) {
 389         for (j = 0; j < s->peers[i].nb_eventfds; j++) {
 390             ivshmem_add_eventfd(s, i, j);
 391         }
 392     }
 393 }
 394
 395 /* this function increase the dynamic storage need to store data about other
 396  * guests */
 397 static void increase_dynamic_storage(IVShmemState *s, int new_min_size) {
 398
 399     int j, old_nb_alloc;
 400
 401     old_nb_alloc = s->nb_peers;
 402
 403     while (new_min_size >= s->nb_peers)
 404         s->nb_peers = s->nb_peers * 2;
 405
 406     IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers);
 407     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 408
 409     /* zero out new pointers */
 410     for (j = old_nb_alloc; j < s->nb_peers; j++) {
 411         s->peers[j].eventfds = NULL;
 412         s->peers[j].nb_eventfds = 0;
 413     }
 414 }
 415
 416 static void ivshmem_read(void *opaque, const uint8_t * buf, int flags)
 417 {
 418     IVShmemState *s = opaque;
 419     int incoming_fd, tmp_fd;
 420     int guest_max_eventfd;
 421     long incoming_posn;
 422
 423     memcpy(&incoming_posn, buf, sizeof(long));
 424     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 425     tmp_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 426     IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, tmp_fd);
 427
 428     /* make sure we have enough space for this guest */
 429     if (incoming_posn >= s->nb_peers) {
 430         increase_dynamic_storage(s, incoming_posn);
 431     }
 432
 433     if (tmp_fd == -1) {
 434         /* if posn is positive and unseen before then this is our posn*/
 435         if ((incoming_posn >= 0) &&
 436                             (s->peers[incoming_posn].eventfds == NULL)) {
 437             /* receive our posn */
 438             s->vm_id = incoming_posn;
 439             return;
 440         } else {
 441             /* otherwise an fd == -1 means an existing guest has gone away */
 442             IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
 443             close_guest_eventfds(s, incoming_posn);
 444             return;
 445         }
 446     }
 447
 448     /* because of the implementation of get_msgfd, we need a dup */
 449     incoming_fd = dup(tmp_fd);
 450
 451     if (incoming_fd == -1) {
 452         fprintf(stderr, "could not allocate file descriptor %s\n",
 453                                                             strerror(errno));
 454         return;
 455     }
 456
 457     /* if the position is -1, then it's shared memory region fd */
 458     if (incoming_posn == -1) {
 459
 460         void * map_ptr;
 461
 462         s->max_peer = 0;
 463
 464         if (check_shm_size(s, incoming_fd) == -1) {
 465             exit(-1);
 466         }
 467
 468         /* mmap the region and map into the BAR2 */
 469         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 470                                                             incoming_fd, 0);
 471         memory_region_init_ram_ptr(&s->ivshmem,
 472                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 473         vmstate_register_ram(&s->ivshmem, &s->dev.qdev);
 474
 475         IVSHMEM_DPRINTF("guest h/w addr = %" PRIu64 ", size = %" PRIu64 "\n",
 476                          s->ivshmem_offset, s->ivshmem_size);
 477
 478         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 479
 480         /* only store the fd if it is successfully mapped */
 481         s->shm_fd = incoming_fd;
 482
 483         return;
 484     }
 485
 486     /* each guest has an array of eventfds, and we keep track of how many
 487      * guests for each VM */
 488     guest_max_eventfd = s->peers[incoming_posn].nb_eventfds;
 489
 490     if (guest_max_eventfd == 0) {
 491         /* one eventfd per MSI vector */
 492         s->peers[incoming_posn].eventfds = g_new(EventNotifier, s->vectors);
 493     }
 494
 495     /* this is an eventfd for a particular guest VM */
 496     IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
 497                                             guest_max_eventfd, incoming_fd);
 498     event_notifier_init_fd(&s->peers[incoming_posn].eventfds[guest_max_eventfd],
 499                            incoming_fd);
 500
 501     /* increment count for particular guest */
 502     s->peers[incoming_posn].nb_eventfds++;
 503
 504     /* keep track of the maximum VM ID */
 505     if (incoming_posn > s->max_peer) {
 506         s->max_peer = incoming_posn;
 507     }
 508
 509     if (incoming_posn == s->vm_id) {
 510         s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s,
 511                    &s->peers[s->vm_id].eventfds[guest_max_eventfd],
 512                    guest_max_eventfd);
 513     }
 514
 515     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 516         ivshmem_add_eventfd(s, incoming_posn, guest_max_eventfd);
 517     }
 518
 519     return;
 520 }
 521
 522 /* Select the MSI-X vectors used by device.
 523  * ivshmem maps events to vectors statically, so
 524  * we just enable all vectors on init and after reset. */
 525 static void ivshmem_use_msix(IVShmemState * s)
 526 {
 527     int i;
 528
 529     if (!msix_present(&s->dev)) {
 530         return;
 531     }
 532
 533     for (i = 0; i < s->vectors; i++) {
 534         msix_vector_use(&s->dev, i);
 535     }
 536 }
 537
 538 static void ivshmem_reset(DeviceState *d)
 539 {
 540     IVShmemState *s = DO_UPCAST(IVShmemState, dev.qdev, d);
 541
 542     s->intrstatus = 0;
 543     ivshmem_use_msix(s);
 544     return;
 545 }
 546
 547 static uint64_t ivshmem_get_size(IVShmemState * s) {
 548
 549     uint64_t value;
 550     char *ptr;
 551
 552     value = strtoull(s->sizearg, &ptr, 10);
 553     switch (*ptr) {
 554         case 0: case 'M': case 'm':
 555             value <<= 20;
 556             break;
 557         case 'G': case 'g':
 558             value <<= 30;
 559             break;
 560         default:
 561             fprintf(stderr, "qemu: invalid ram size: %s\n", s->sizearg);
 562             exit(1);
 563     }
 564
 565     /* BARs must be a power of 2 */
 566     if (!is_power_of_two(value)) {
 567         fprintf(stderr, "ivshmem: size must be power of 2\n");
 568         exit(1);
 569     }
 570
 571     return value;
 572 }
 573
 574 static void ivshmem_setup_msi(IVShmemState * s)
 575 {
 576     if (msix_init_exclusive_bar(&s->dev, s->vectors, 1)) {
 577         IVSHMEM_DPRINTF("msix initialization failed\n");
 578         exit(1);
 579     }
 580
 581     IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 582
 583     /* allocate QEMU char devices for receiving interrupts */
 584     s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
 585
 586     ivshmem_use_msix(s);
 587 }
 588
 589 static void ivshmem_save(QEMUFile* f, void *opaque)
 590 {
 591     IVShmemState *proxy = opaque;
 592
 593     IVSHMEM_DPRINTF("ivshmem_save\n");
 594     pci_device_save(&proxy->dev, f);
 595
 596     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 597         msix_save(&proxy->dev, f);
 598     } else {
 599         qemu_put_be32(f, proxy->intrstatus);
 600         qemu_put_be32(f, proxy->intrmask);
 601     }
 602
 603 }
 604
 605 static int ivshmem_load(QEMUFile* f, void *opaque, int version_id)
 606 {
 607     IVSHMEM_DPRINTF("ivshmem_load\n");
 608
 609     IVShmemState *proxy = opaque;
 610     int ret;
 611
 612     if (version_id > 0) {
 613         return -EINVAL;
 614     }
 615
 616     if (proxy->role_val == IVSHMEM_PEER) {
 617         fprintf(stderr, "ivshmem: 'peer' devices are not migratable\n");
 618         return -EINVAL;
 619     }
 620
 621     ret = pci_device_load(&proxy->dev, f);
 622     if (ret) {
 623         return ret;
 624     }
 625
 626     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 627         msix_load(&proxy->dev, f);
 628         ivshmem_use_msix(proxy);
 629     } else {
 630         proxy->intrstatus = qemu_get_be32(f);
 631         proxy->intrmask = qemu_get_be32(f);
 632     }
 633
 634     return 0;
 635 }
 636
 637 static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address,
 638                                  uint32_t val, int len)
 639 {
 640     pci_default_write_config(pci_dev, address, val, len);
 641     msix_write_config(pci_dev, address, val, len);
 642 }
 643
 644 static int pci_ivshmem_init(PCIDevice *dev)
 645 {
 646     IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
 647     uint8_t *pci_conf;
 648
 649     if (s->sizearg == NULL)
 650         s->ivshmem_size = 4 << 20; /* 4 MB default */
 651     else {
 652         s->ivshmem_size = ivshmem_get_size(s);
 653     }
 654
 655     register_savevm(&s->dev.qdev, "ivshmem", 0, 0, ivshmem_save, ivshmem_load,
 656                                                                         dev);
 657
 658     /* IRQFD requires MSI */
 659     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 660         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 661         fprintf(stderr, "ivshmem: ioeventfd/irqfd requires MSI\n");
 662         exit(1);
 663     }
 664
 665     /* check that role is reasonable */
 666     if (s->role) {
 667         if (strncmp(s->role, "peer", 5) == 0) {
 668             s->role_val = IVSHMEM_PEER;
 669         } else if (strncmp(s->role, "master", 7) == 0) {
 670             s->role_val = IVSHMEM_MASTER;
 671         } else {
 672             fprintf(stderr, "ivshmem: 'role' must be 'peer' or 'master'\n");
 673             exit(1);
 674         }
 675     } else {
 676         s->role_val = IVSHMEM_MASTER; /* default */
 677     }
 678
 679     if (s->role_val == IVSHMEM_PEER) {
 680         error_set(&s->migration_blocker, QERR_DEVICE_FEATURE_BLOCKS_MIGRATION, "ivshmem", "peer mode");
 681         migrate_add_blocker(s->migration_blocker);
 682     }
 683
 684     pci_conf = s->dev.config;
 685     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 686
 687     pci_config_set_interrupt_pin(pci_conf, 1);
 688
 689     s->shm_fd = 0;
 690
 691     memory_region_init_io(&s->ivshmem_mmio, &ivshmem_mmio_ops, s,
 692                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 693
 694     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 695         setup_ioeventfds(s);
 696     }
 697
 698     /* region for registers*/
 699     pci_register_bar(&s->dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 700                      &s->ivshmem_mmio);
 701
 702     memory_region_init(&s->bar, "ivshmem-bar2-container", s->ivshmem_size);
 703
 704     if ((s->server_chr != NULL) &&
 705                         (strncmp(s->server_chr->filename, "unix:", 5) == 0)) {
 706         /* if we get a UNIX socket as the parameter we will talk
 707          * to the ivshmem server to receive the memory region */
 708
 709         if (s->shmobj != NULL) {
 710             fprintf(stderr, "WARNING: do not specify both 'chardev' "
 711                                                 "and 'shm' with ivshmem\n");
 712         }
 713
 714         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 715                                                     s->server_chr->filename);
 716
 717         if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 718             ivshmem_setup_msi(s);
 719         }
 720
 721         /* we allocate enough space for 16 guests and grow as needed */
 722         s->nb_peers = 16;
 723         s->vm_id = -1;
 724
 725         /* allocate/initialize space for interrupt handling */
 726         s->peers = g_malloc0(s->nb_peers * sizeof(Peer));
 727
 728         pci_register_bar(&s->dev, 2,
 729                          PCI_BASE_ADDRESS_SPACE_MEMORY, &s->bar);
 730
 731         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 732
 733         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 734                      ivshmem_event, s);
 735     } else {
 736         /* just map the file immediately, we're not using a server */
 737         int fd;
 738
 739         if (s->shmobj == NULL) {
 740             fprintf(stderr, "Must specify 'chardev' or 'shm' to ivshmem\n");
 741         }
 742
 743         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 744
 745         /* try opening with O_EXCL and if it succeeds zero the memory
 746          * by truncating to 0 */
 747         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 748                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 749            /* truncate file to length PCI device's memory */
 750             if (ftruncate(fd, s->ivshmem_size) != 0) {
 751                 fprintf(stderr, "ivshmem: could not truncate shared file\n");
 752             }
 753
 754         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 755                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 756             fprintf(stderr, "ivshmem: could not open shared file\n");
 757             exit(-1);
 758
 759         }
 760
 761         if (check_shm_size(s, fd) == -1) {
 762             exit(-1);
 763         }
 764
 765         create_shared_memory_BAR(s, fd);
 766
 767     }
 768
 769     s->dev.config_write = ivshmem_write_config;
 770
 771     return 0;
 772 }
 773
 774 static void pci_ivshmem_uninit(PCIDevice *dev)
 775 {
 776     IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev);
 777
 778     if (s->migration_blocker) {
 779         migrate_del_blocker(s->migration_blocker);
 780         error_free(s->migration_blocker);
 781     }
 782
 783     memory_region_destroy(&s->ivshmem_mmio);
 784     memory_region_del_subregion(&s->bar, &s->ivshmem);
 785     vmstate_unregister_ram(&s->ivshmem, &s->dev.qdev);
 786     memory_region_destroy(&s->ivshmem);
 787     memory_region_destroy(&s->bar);
 788     unregister_savevm(&dev->qdev, "ivshmem", s);
 789 }
 790
 791 static Property ivshmem_properties[] = {
 792     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
 793     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
 794     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
 795     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
 796     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
 797     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
 798     DEFINE_PROP_STRING("role", IVShmemState, role),
 799     DEFINE_PROP_END_OF_LIST(),
 800 };
 801
 802 static void ivshmem_class_init(ObjectClass *klass, void *data)
 803 {
 804     DeviceClass *dc = DEVICE_CLASS(klass);
 805     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 806
 807     k->init = pci_ivshmem_init;
 808     k->exit = pci_ivshmem_uninit;
 809     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
 810     k->device_id = 0x1110;
 811     k->class_id = PCI_CLASS_MEMORY_RAM;
 812     dc->reset = ivshmem_reset;
 813     dc->props = ivshmem_properties;
 814 }
 815
 816 static TypeInfo ivshmem_info = {
 817     .name          = "ivshmem",
 818     .parent        = TYPE_PCI_DEVICE,
 819     .instance_size = sizeof(IVShmemState),
 820     .class_init    = ivshmem_class_init,
 821 };
 822
 823 static void ivshmem_register_types(void)
 824 {
 825     type_register_static(&ivshmem_info);
 826 }
 827
 828 type_init(ivshmem_register_types)