hw/misc/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/hw.h"
  20 #include "hw/i386/pc.h"
  21 #include "hw/pci/pci.h"
  22 #include "hw/pci/msix.h"
  23 #include "sysemu/kvm.h"
  24 #include "migration/migration.h"
  25 #include "qapi/qmp/qerror.h"
  26 #include "qemu/event_notifier.h"
  27 #include "sysemu/char.h"
  28
  29 #include <sys/mman.h>
  30 #include <sys/types.h>
  31
  32 #define PCI_VENDOR_ID_IVSHMEM   PCI_VENDOR_ID_REDHAT_QUMRANET
  33 #define PCI_DEVICE_ID_IVSHMEM   0x1110
  34
  35 #define IVSHMEM_IOEVENTFD   0
  36 #define IVSHMEM_MSI     1
  37
  38 #define IVSHMEM_PEER    0
  39 #define IVSHMEM_MASTER  1
  40
  41 #define IVSHMEM_REG_BAR_SIZE 0x100
  42
  43 //#define DEBUG_IVSHMEM
  44 #ifdef DEBUG_IVSHMEM
  45 #define IVSHMEM_DPRINTF(fmt, ...)        \
  46     do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0)
  47 #else
  48 #define IVSHMEM_DPRINTF(fmt, ...)
  49 #endif
  50
  51 #define TYPE_IVSHMEM "ivshmem"
  52 #define IVSHMEM(obj) \
  53     OBJECT_CHECK(IVShmemState, (obj), TYPE_IVSHMEM)
  54
  55 typedef struct Peer {
  56     int nb_eventfds;
  57     EventNotifier *eventfds;
  58 } Peer;
  59
  60 typedef struct EventfdEntry {
  61     PCIDevice *pdev;
  62     int vector;
  63 } EventfdEntry;
  64
  65 typedef struct IVShmemState {
  66     /*< private >*/
  67     PCIDevice parent_obj;
  68     /*< public >*/
  69
  70     uint32_t intrmask;
  71     uint32_t intrstatus;
  72     uint32_t doorbell;
  73
  74     CharDriverState **eventfd_chr;
  75     CharDriverState *server_chr;
  76     MemoryRegion ivshmem_mmio;
  77
  78     /* We might need to register the BAR before we actually have the memory.
  79      * So prepare a container MemoryRegion for the BAR immediately and
  80      * add a subregion when we have the memory.
  81      */
  82     MemoryRegion bar;
  83     MemoryRegion ivshmem;
  84     uint64_t ivshmem_size; /* size of shared memory region */
  85     uint32_t ivshmem_attr;
  86     uint32_t ivshmem_64bit;
  87     int shm_fd; /* shared memory file descriptor */
  88
  89     Peer *peers;
  90     int nb_peers; /* how many guests we have space for */
  91     int max_peer; /* maximum numbered peer */
  92
  93     int vm_id;
  94     uint32_t vectors;
  95     uint32_t features;
  96     EventfdEntry *eventfd_table;
  97
  98     Error *migration_blocker;
  99
 100     char * shmobj;
 101     char * sizearg;
 102     char * role;
 103     int role_val;   /* scalar to avoid multiple string comparisons */
 104 } IVShmemState;
 105
 106 /* registers for the Inter-VM shared memory device */
 107 enum ivshmem_registers {
 108     INTRMASK = 0,
 109     INTRSTATUS = 4,
 110     IVPOSITION = 8,
 111     DOORBELL = 12,
 112 };
 113
 114 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 115                                                     unsigned int feature) {
 116     return (ivs->features & (1 << feature));
 117 }
 118
 119 static inline bool is_power_of_two(uint64_t x) {
 120     return (x & (x - 1)) == 0;
 121 }
 122
 123 /* accessing registers - based on rtl8139 */
 124 static void ivshmem_update_irq(IVShmemState *s, int val)
 125 {
 126     PCIDevice *d = PCI_DEVICE(s);
 127     int isr;
 128     isr = (s->intrstatus & s->intrmask) & 0xffffffff;
 129
 130     /* don't print ISR resets */
 131     if (isr) {
 132         IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
 133            isr ? 1 : 0, s->intrstatus, s->intrmask);
 134     }
 135
 136     pci_set_irq(d, (isr != 0));
 137 }
 138
 139 static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val)
 140 {
 141     IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
 142
 143     s->intrmask = val;
 144
 145     ivshmem_update_irq(s, val);
 146 }
 147
 148 static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
 149 {
 150     uint32_t ret = s->intrmask;
 151
 152     IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
 153
 154     return ret;
 155 }
 156
 157 static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val)
 158 {
 159     IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
 160
 161     s->intrstatus = val;
 162
 163     ivshmem_update_irq(s, val);
 164 }
 165
 166 static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
 167 {
 168     uint32_t ret = s->intrstatus;
 169
 170     /* reading ISR clears all interrupts */
 171     s->intrstatus = 0;
 172
 173     ivshmem_update_irq(s, 0);
 174
 175     return ret;
 176 }
 177
 178 static void ivshmem_io_write(void *opaque, hwaddr addr,
 179                              uint64_t val, unsigned size)
 180 {
 181     IVShmemState *s = opaque;
 182
 183     uint16_t dest = val >> 16;
 184     uint16_t vector = val & 0xff;
 185
 186     addr &= 0xfc;
 187
 188     IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr);
 189     switch (addr)
 190     {
 191         case INTRMASK:
 192             ivshmem_IntrMask_write(s, val);
 193             break;
 194
 195         case INTRSTATUS:
 196             ivshmem_IntrStatus_write(s, val);
 197             break;
 198
 199         case DOORBELL:
 200             /* check that dest VM ID is reasonable */
 201             if (dest > s->max_peer) {
 202                 IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest);
 203                 break;
 204             }
 205
 206             /* check doorbell range */
 207             if (vector < s->peers[dest].nb_eventfds) {
 208                 IVSHMEM_DPRINTF("Notifying VM %d on vector %d\n", dest, vector);
 209                 event_notifier_set(&s->peers[dest].eventfds[vector]);
 210             }
 211             break;
 212         default:
 213             IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest);
 214     }
 215 }
 216
 217 static uint64_t ivshmem_io_read(void *opaque, hwaddr addr,
 218                                 unsigned size)
 219 {
 220
 221     IVShmemState *s = opaque;
 222     uint32_t ret;
 223
 224     switch (addr)
 225     {
 226         case INTRMASK:
 227             ret = ivshmem_IntrMask_read(s);
 228             break;
 229
 230         case INTRSTATUS:
 231             ret = ivshmem_IntrStatus_read(s);
 232             break;
 233
 234         case IVPOSITION:
 235             /* return my VM ID if the memory is mapped */
 236             if (s->shm_fd > 0) {
 237                 ret = s->vm_id;
 238             } else {
 239                 ret = -1;
 240             }
 241             break;
 242
 243         default:
 244             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 245             ret = 0;
 246     }
 247
 248     return ret;
 249 }
 250
 251 static const MemoryRegionOps ivshmem_mmio_ops = {
 252     .read = ivshmem_io_read,
 253     .write = ivshmem_io_write,
 254     .endianness = DEVICE_NATIVE_ENDIAN,
 255     .impl = {
 256         .min_access_size = 4,
 257         .max_access_size = 4,
 258     },
 259 };
 260
 261 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 262 {
 263     IVShmemState *s = opaque;
 264
 265     ivshmem_IntrStatus_write(s, *buf);
 266
 267     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf);
 268 }
 269
 270 static int ivshmem_can_receive(void * opaque)
 271 {
 272     return 8;
 273 }
 274
 275 static void ivshmem_event(void *opaque, int event)
 276 {
 277     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 278 }
 279
 280 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 281
 282     EventfdEntry *entry = opaque;
 283     PCIDevice *pdev = entry->pdev;
 284
 285     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector);
 286     msix_notify(pdev, entry->vector);
 287 }
 288
 289 static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n,
 290                                                   int vector)
 291 {
 292     /* create a event character device based on the passed eventfd */
 293     IVShmemState *s = opaque;
 294     CharDriverState * chr;
 295     int eventfd = event_notifier_get_fd(n);
 296
 297     chr = qemu_chr_open_eventfd(eventfd);
 298
 299     if (chr == NULL) {
 300         fprintf(stderr, "creating eventfd for eventfd %d failed\n", eventfd);
 301         exit(-1);
 302     }
 303     qemu_chr_fe_claim_no_fail(chr);
 304
 305     /* if MSI is supported we need multiple interrupts */
 306     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 307         s->eventfd_table[vector].pdev = PCI_DEVICE(s);
 308         s->eventfd_table[vector].vector = vector;
 309
 310         qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd,
 311                       ivshmem_event, &s->eventfd_table[vector]);
 312     } else {
 313         qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
 314                       ivshmem_event, s);
 315     }
 316
 317     return chr;
 318
 319 }
 320
 321 static int check_shm_size(IVShmemState *s, int fd) {
 322     /* check that the guest isn't going to try and map more memory than the
 323      * the object has allocated return -1 to indicate error */
 324
 325     struct stat buf;
 326
 327     if (fstat(fd, &buf) < 0) {
 328         fprintf(stderr, "ivshmem: exiting: fstat on fd %d failed: %s\n",
 329                 fd, strerror(errno));
 330         return -1;
 331     }
 332
 333     if (s->ivshmem_size > buf.st_size) {
 334         fprintf(stderr,
 335                 "IVSHMEM ERROR: Requested memory size greater"
 336                 " than shared object size (%" PRIu64 " > %" PRIu64")\n",
 337                 s->ivshmem_size, (uint64_t)buf.st_size);
 338         return -1;
 339     } else {
 340         return 0;
 341     }
 342 }
 343
 344 /* create the shared memory BAR when we are not using the server, so we can
 345  * create the BAR and map the memory immediately */
 346 static void create_shared_memory_BAR(IVShmemState *s, int fd) {
 347
 348     void * ptr;
 349
 350     s->shm_fd = fd;
 351
 352     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 353
 354     memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s), "ivshmem.bar2",
 355                                s->ivshmem_size, ptr);
 356     vmstate_register_ram(&s->ivshmem, DEVICE(s));
 357     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 358
 359     /* region for shared memory */
 360     pci_register_bar(PCI_DEVICE(s), 2, s->ivshmem_attr, &s->bar);
 361 }
 362
 363 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 364 {
 365     memory_region_add_eventfd(&s->ivshmem_mmio,
 366                               DOORBELL,
 367                               4,
 368                               true,
 369                               (posn << 16) | i,
 370                               &s->peers[posn].eventfds[i]);
 371 }
 372
 373 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 374 {
 375     memory_region_del_eventfd(&s->ivshmem_mmio,
 376                               DOORBELL,
 377                               4,
 378                               true,
 379                               (posn << 16) | i,
 380                               &s->peers[posn].eventfds[i]);
 381 }
 382
 383 static void close_guest_eventfds(IVShmemState *s, int posn)
 384 {
 385     int i, guest_curr_max;
 386
 387     if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 388         return;
 389     }
 390
 391     guest_curr_max = s->peers[posn].nb_eventfds;
 392
 393     memory_region_transaction_begin();
 394     for (i = 0; i < guest_curr_max; i++) {
 395         ivshmem_del_eventfd(s, posn, i);
 396     }
 397     memory_region_transaction_commit();
 398     for (i = 0; i < guest_curr_max; i++) {
 399         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 400     }
 401
 402     g_free(s->peers[posn].eventfds);
 403     s->peers[posn].nb_eventfds = 0;
 404 }
 405
 406 /* this function increase the dynamic storage need to store data about other
 407  * guests */
 408 static void increase_dynamic_storage(IVShmemState *s, int new_min_size) {
 409
 410     int j, old_nb_alloc;
 411
 412     old_nb_alloc = s->nb_peers;
 413
 414     while (new_min_size >= s->nb_peers)
 415         s->nb_peers = s->nb_peers * 2;
 416
 417     IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers);
 418     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 419
 420     /* zero out new pointers */
 421     for (j = old_nb_alloc; j < s->nb_peers; j++) {
 422         s->peers[j].eventfds = NULL;
 423         s->peers[j].nb_eventfds = 0;
 424     }
 425 }
 426
 427 static void ivshmem_read(void *opaque, const uint8_t * buf, int flags)
 428 {
 429     IVShmemState *s = opaque;
 430     int incoming_fd, tmp_fd;
 431     int guest_max_eventfd;
 432     long incoming_posn;
 433
 434     memcpy(&incoming_posn, buf, sizeof(long));
 435     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 436     tmp_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 437     IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, tmp_fd);
 438
 439     /* make sure we have enough space for this guest */
 440     if (incoming_posn >= s->nb_peers) {
 441         increase_dynamic_storage(s, incoming_posn);
 442     }
 443
 444     if (tmp_fd == -1) {
 445         /* if posn is positive and unseen before then this is our posn*/
 446         if ((incoming_posn >= 0) &&
 447                             (s->peers[incoming_posn].eventfds == NULL)) {
 448             /* receive our posn */
 449             s->vm_id = incoming_posn;
 450             return;
 451         } else {
 452             /* otherwise an fd == -1 means an existing guest has gone away */
 453             IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
 454             close_guest_eventfds(s, incoming_posn);
 455             return;
 456         }
 457     }
 458
 459     /* because of the implementation of get_msgfd, we need a dup */
 460     incoming_fd = dup(tmp_fd);
 461
 462     if (incoming_fd == -1) {
 463         fprintf(stderr, "could not allocate file descriptor %s\n",
 464                                                             strerror(errno));
 465         return;
 466     }
 467
 468     /* if the position is -1, then it's shared memory region fd */
 469     if (incoming_posn == -1) {
 470
 471         void * map_ptr;
 472
 473         s->max_peer = 0;
 474
 475         if (check_shm_size(s, incoming_fd) == -1) {
 476             exit(-1);
 477         }
 478
 479         /* mmap the region and map into the BAR2 */
 480         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 481                                                             incoming_fd, 0);
 482         memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s),
 483                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 484         vmstate_register_ram(&s->ivshmem, DEVICE(s));
 485
 486         IVSHMEM_DPRINTF("guest h/w addr = %p, size = %" PRIu64 "\n",
 487                          map_ptr, s->ivshmem_size);
 488
 489         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 490
 491         /* only store the fd if it is successfully mapped */
 492         s->shm_fd = incoming_fd;
 493
 494         return;
 495     }
 496
 497     /* each guest has an array of eventfds, and we keep track of how many
 498      * guests for each VM */
 499     guest_max_eventfd = s->peers[incoming_posn].nb_eventfds;
 500
 501     if (guest_max_eventfd == 0) {
 502         /* one eventfd per MSI vector */
 503         s->peers[incoming_posn].eventfds = g_new(EventNotifier, s->vectors);
 504     }
 505
 506     /* this is an eventfd for a particular guest VM */
 507     IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
 508                                             guest_max_eventfd, incoming_fd);
 509     event_notifier_init_fd(&s->peers[incoming_posn].eventfds[guest_max_eventfd],
 510                            incoming_fd);
 511
 512     /* increment count for particular guest */
 513     s->peers[incoming_posn].nb_eventfds++;
 514
 515     /* keep track of the maximum VM ID */
 516     if (incoming_posn > s->max_peer) {
 517         s->max_peer = incoming_posn;
 518     }
 519
 520     if (incoming_posn == s->vm_id) {
 521         s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s,
 522                    &s->peers[s->vm_id].eventfds[guest_max_eventfd],
 523                    guest_max_eventfd);
 524     }
 525
 526     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 527         ivshmem_add_eventfd(s, incoming_posn, guest_max_eventfd);
 528     }
 529 }
 530
 531 /* Select the MSI-X vectors used by device.
 532  * ivshmem maps events to vectors statically, so
 533  * we just enable all vectors on init and after reset. */
 534 static void ivshmem_use_msix(IVShmemState * s)
 535 {
 536     PCIDevice *d = PCI_DEVICE(s);
 537     int i;
 538
 539     if (!msix_present(d)) {
 540         return;
 541     }
 542
 543     for (i = 0; i < s->vectors; i++) {
 544         msix_vector_use(d, i);
 545     }
 546 }
 547
 548 static void ivshmem_reset(DeviceState *d)
 549 {
 550     IVShmemState *s = IVSHMEM(d);
 551
 552     s->intrstatus = 0;
 553     ivshmem_use_msix(s);
 554 }
 555
 556 static uint64_t ivshmem_get_size(IVShmemState * s) {
 557
 558     uint64_t value;
 559     char *ptr;
 560
 561     value = strtoull(s->sizearg, &ptr, 10);
 562     switch (*ptr) {
 563         case 0: case 'M': case 'm':
 564             value <<= 20;
 565             break;
 566         case 'G': case 'g':
 567             value <<= 30;
 568             break;
 569         default:
 570             fprintf(stderr, "qemu: invalid ram size: %s\n", s->sizearg);
 571             exit(1);
 572     }
 573
 574     /* BARs must be a power of 2 */
 575     if (!is_power_of_two(value)) {
 576         fprintf(stderr, "ivshmem: size must be power of 2\n");
 577         exit(1);
 578     }
 579
 580     return value;
 581 }
 582
 583 static void ivshmem_setup_msi(IVShmemState * s)
 584 {
 585     if (msix_init_exclusive_bar(PCI_DEVICE(s), s->vectors, 1)) {
 586         IVSHMEM_DPRINTF("msix initialization failed\n");
 587         exit(1);
 588     }
 589
 590     IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 591
 592     /* allocate QEMU char devices for receiving interrupts */
 593     s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
 594
 595     ivshmem_use_msix(s);
 596 }
 597
 598 static void ivshmem_save(QEMUFile* f, void *opaque)
 599 {
 600     IVShmemState *proxy = opaque;
 601     PCIDevice *pci_dev = PCI_DEVICE(proxy);
 602
 603     IVSHMEM_DPRINTF("ivshmem_save\n");
 604     pci_device_save(pci_dev, f);
 605
 606     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 607         msix_save(pci_dev, f);
 608     } else {
 609         qemu_put_be32(f, proxy->intrstatus);
 610         qemu_put_be32(f, proxy->intrmask);
 611     }
 612
 613 }
 614
 615 static int ivshmem_load(QEMUFile* f, void *opaque, int version_id)
 616 {
 617     IVSHMEM_DPRINTF("ivshmem_load\n");
 618
 619     IVShmemState *proxy = opaque;
 620     PCIDevice *pci_dev = PCI_DEVICE(proxy);
 621     int ret;
 622
 623     if (version_id > 0) {
 624         return -EINVAL;
 625     }
 626
 627     if (proxy->role_val == IVSHMEM_PEER) {
 628         fprintf(stderr, "ivshmem: 'peer' devices are not migratable\n");
 629         return -EINVAL;
 630     }
 631
 632     ret = pci_device_load(pci_dev, f);
 633     if (ret) {
 634         return ret;
 635     }
 636
 637     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 638         msix_load(pci_dev, f);
 639         ivshmem_use_msix(proxy);
 640     } else {
 641         proxy->intrstatus = qemu_get_be32(f);
 642         proxy->intrmask = qemu_get_be32(f);
 643     }
 644
 645     return 0;
 646 }
 647
 648 static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address,
 649                                  uint32_t val, int len)
 650 {
 651     pci_default_write_config(pci_dev, address, val, len);
 652     msix_write_config(pci_dev, address, val, len);
 653 }
 654
 655 static int pci_ivshmem_init(PCIDevice *dev)
 656 {
 657     IVShmemState *s = IVSHMEM(dev);
 658     uint8_t *pci_conf;
 659
 660     if (s->sizearg == NULL)
 661         s->ivshmem_size = 4 << 20; /* 4 MB default */
 662     else {
 663         s->ivshmem_size = ivshmem_get_size(s);
 664     }
 665
 666     register_savevm(DEVICE(dev), "ivshmem", 0, 0, ivshmem_save, ivshmem_load,
 667                                                                         dev);
 668
 669     /* IRQFD requires MSI */
 670     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 671         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 672         fprintf(stderr, "ivshmem: ioeventfd/irqfd requires MSI\n");
 673         exit(1);
 674     }
 675
 676     /* check that role is reasonable */
 677     if (s->role) {
 678         if (strncmp(s->role, "peer", 5) == 0) {
 679             s->role_val = IVSHMEM_PEER;
 680         } else if (strncmp(s->role, "master", 7) == 0) {
 681             s->role_val = IVSHMEM_MASTER;
 682         } else {
 683             fprintf(stderr, "ivshmem: 'role' must be 'peer' or 'master'\n");
 684             exit(1);
 685         }
 686     } else {
 687         s->role_val = IVSHMEM_MASTER; /* default */
 688     }
 689
 690     if (s->role_val == IVSHMEM_PEER) {
 691         error_setg(&s->migration_blocker,
 692                    "Migration is disabled when using feature 'peer mode' in device 'ivshmem'");
 693         migrate_add_blocker(s->migration_blocker);
 694     }
 695
 696     pci_conf = dev->config;
 697     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 698
 699     pci_config_set_interrupt_pin(pci_conf, 1);
 700
 701     s->shm_fd = 0;
 702
 703     memory_region_init_io(&s->ivshmem_mmio, OBJECT(s), &ivshmem_mmio_ops, s,
 704                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 705
 706     /* region for registers*/
 707     pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 708                      &s->ivshmem_mmio);
 709
 710     memory_region_init(&s->bar, OBJECT(s), "ivshmem-bar2-container", s->ivshmem_size);
 711     s->ivshmem_attr = PCI_BASE_ADDRESS_SPACE_MEMORY |
 712         PCI_BASE_ADDRESS_MEM_PREFETCH;
 713     if (s->ivshmem_64bit) {
 714         s->ivshmem_attr |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 715     }
 716
 717     if ((s->server_chr != NULL) &&
 718                         (strncmp(s->server_chr->filename, "unix:", 5) == 0)) {
 719         /* if we get a UNIX socket as the parameter we will talk
 720          * to the ivshmem server to receive the memory region */
 721
 722         if (s->shmobj != NULL) {
 723             fprintf(stderr, "WARNING: do not specify both 'chardev' "
 724                                                 "and 'shm' with ivshmem\n");
 725         }
 726
 727         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 728                                                     s->server_chr->filename);
 729
 730         if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 731             ivshmem_setup_msi(s);
 732         }
 733
 734         /* we allocate enough space for 16 guests and grow as needed */
 735         s->nb_peers = 16;
 736         s->vm_id = -1;
 737
 738         /* allocate/initialize space for interrupt handling */
 739         s->peers = g_malloc0(s->nb_peers * sizeof(Peer));
 740
 741         pci_register_bar(dev, 2, s->ivshmem_attr, &s->bar);
 742
 743         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 744
 745         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 746                      ivshmem_event, s);
 747     } else {
 748         /* just map the file immediately, we're not using a server */
 749         int fd;
 750
 751         if (s->shmobj == NULL) {
 752             fprintf(stderr, "Must specify 'chardev' or 'shm' to ivshmem\n");
 753             exit(1);
 754         }
 755
 756         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 757
 758         /* try opening with O_EXCL and if it succeeds zero the memory
 759          * by truncating to 0 */
 760         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 761                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 762            /* truncate file to length PCI device's memory */
 763             if (ftruncate(fd, s->ivshmem_size) != 0) {
 764                 fprintf(stderr, "ivshmem: could not truncate shared file\n");
 765             }
 766
 767         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 768                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 769             fprintf(stderr, "ivshmem: could not open shared file\n");
 770             exit(-1);
 771
 772         }
 773
 774         if (check_shm_size(s, fd) == -1) {
 775             exit(-1);
 776         }
 777
 778         create_shared_memory_BAR(s, fd);
 779
 780     }
 781
 782     dev->config_write = ivshmem_write_config;
 783
 784     return 0;
 785 }
 786
 787 static void pci_ivshmem_uninit(PCIDevice *dev)
 788 {
 789     IVShmemState *s = IVSHMEM(dev);
 790
 791     if (s->migration_blocker) {
 792         migrate_del_blocker(s->migration_blocker);
 793         error_free(s->migration_blocker);
 794     }
 795
 796     memory_region_del_subregion(&s->bar, &s->ivshmem);
 797     vmstate_unregister_ram(&s->ivshmem, DEVICE(dev));
 798     unregister_savevm(DEVICE(dev), "ivshmem", s);
 799 }
 800
 801 static Property ivshmem_properties[] = {
 802     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
 803     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
 804     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
 805     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
 806     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
 807     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
 808     DEFINE_PROP_STRING("role", IVShmemState, role),
 809     DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1),
 810     DEFINE_PROP_END_OF_LIST(),
 811 };
 812
 813 static void ivshmem_class_init(ObjectClass *klass, void *data)
 814 {
 815     DeviceClass *dc = DEVICE_CLASS(klass);
 816     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 817
 818     k->init = pci_ivshmem_init;
 819     k->exit = pci_ivshmem_uninit;
 820     k->vendor_id = PCI_VENDOR_ID_IVSHMEM;
 821     k->device_id = PCI_DEVICE_ID_IVSHMEM;
 822     k->class_id = PCI_CLASS_MEMORY_RAM;
 823     dc->reset = ivshmem_reset;
 824     dc->props = ivshmem_properties;
 825     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 826 }
 827
 828 static const TypeInfo ivshmem_info = {
 829     .name          = TYPE_IVSHMEM,
 830     .parent        = TYPE_PCI_DEVICE,
 831     .instance_size = sizeof(IVShmemState),
 832     .class_init    = ivshmem_class_init,
 833 };
 834
 835 static void ivshmem_register_types(void)
 836 {
 837     type_register_static(&ivshmem_info);
 838 }
 839
 840 type_init(ivshmem_register_types)