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 "qemu/error-report.h"
  26 #include "qemu/event_notifier.h"
  27 #include "qemu/fifo8.h"
  28 #include "sysemu/char.h"
  29
  30 #include <sys/mman.h>
  31 #include <sys/types.h>
  32 #include <limits.h>
  33
  34 #define PCI_VENDOR_ID_IVSHMEM   PCI_VENDOR_ID_REDHAT_QUMRANET
  35 #define PCI_DEVICE_ID_IVSHMEM   0x1110
  36
  37 #define IVSHMEM_MAX_PEERS G_MAXUINT16
  38 #define IVSHMEM_IOEVENTFD   0
  39 #define IVSHMEM_MSI     1
  40
  41 #define IVSHMEM_PEER    0
  42 #define IVSHMEM_MASTER  1
  43
  44 #define IVSHMEM_REG_BAR_SIZE 0x100
  45
  46 //#define DEBUG_IVSHMEM
  47 #ifdef DEBUG_IVSHMEM
  48 #define IVSHMEM_DPRINTF(fmt, ...)        \
  49     do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0)
  50 #else
  51 #define IVSHMEM_DPRINTF(fmt, ...)
  52 #endif
  53
  54 #define TYPE_IVSHMEM "ivshmem"
  55 #define IVSHMEM(obj) \
  56     OBJECT_CHECK(IVShmemState, (obj), TYPE_IVSHMEM)
  57
  58 typedef struct Peer {
  59     int nb_eventfds;
  60     EventNotifier *eventfds;
  61 } Peer;
  62
  63 typedef struct EventfdEntry {
  64     PCIDevice *pdev;
  65     int vector;
  66 } EventfdEntry;
  67
  68 typedef struct IVShmemState {
  69     /*< private >*/
  70     PCIDevice parent_obj;
  71     /*< public >*/
  72
  73     uint32_t intrmask;
  74     uint32_t intrstatus;
  75
  76     CharDriverState **eventfd_chr;
  77     CharDriverState *server_chr;
  78     Fifo8 incoming_fifo;
  79     MemoryRegion ivshmem_mmio;
  80
  81     /* We might need to register the BAR before we actually have the memory.
  82      * So prepare a container MemoryRegion for the BAR immediately and
  83      * add a subregion when we have the memory.
  84      */
  85     MemoryRegion bar;
  86     MemoryRegion ivshmem;
  87     uint64_t ivshmem_size; /* size of shared memory region */
  88     uint32_t ivshmem_64bit;
  89     int shm_fd; /* shared memory file descriptor */
  90
  91     Peer *peers;
  92     int nb_peers; /* how many guests we have space for */
  93
  94     int vm_id;
  95     uint32_t vectors;
  96     uint32_t features;
  97     EventfdEntry *eventfd_table;
  98
  99     Error *migration_blocker;
 100
 101     char * shmobj;
 102     char * sizearg;
 103     char * role;
 104     int role_val;   /* scalar to avoid multiple string comparisons */
 105 } IVShmemState;
 106
 107 /* registers for the Inter-VM shared memory device */
 108 enum ivshmem_registers {
 109     INTRMASK = 0,
 110     INTRSTATUS = 4,
 111     IVPOSITION = 8,
 112     DOORBELL = 12,
 113 };
 114
 115 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 116                                                     unsigned int feature) {
 117     return (ivs->features & (1 << feature));
 118 }
 119
 120 /* accessing registers - based on rtl8139 */
 121 static void ivshmem_update_irq(IVShmemState *s)
 122 {
 123     PCIDevice *d = PCI_DEVICE(s);
 124     int isr;
 125     isr = (s->intrstatus & s->intrmask) & 0xffffffff;
 126
 127     /* don't print ISR resets */
 128     if (isr) {
 129         IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
 130                         isr ? 1 : 0, s->intrstatus, s->intrmask);
 131     }
 132
 133     pci_set_irq(d, (isr != 0));
 134 }
 135
 136 static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val)
 137 {
 138     IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
 139
 140     s->intrmask = val;
 141
 142     ivshmem_update_irq(s);
 143 }
 144
 145 static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
 146 {
 147     uint32_t ret = s->intrmask;
 148
 149     IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
 150
 151     return ret;
 152 }
 153
 154 static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val)
 155 {
 156     IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
 157
 158     s->intrstatus = val;
 159
 160     ivshmem_update_irq(s);
 161 }
 162
 163 static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
 164 {
 165     uint32_t ret = s->intrstatus;
 166
 167     /* reading ISR clears all interrupts */
 168     s->intrstatus = 0;
 169
 170     ivshmem_update_irq(s);
 171
 172     return ret;
 173 }
 174
 175 static void ivshmem_io_write(void *opaque, hwaddr addr,
 176                              uint64_t val, unsigned size)
 177 {
 178     IVShmemState *s = opaque;
 179
 180     uint16_t dest = val >> 16;
 181     uint16_t vector = val & 0xff;
 182
 183     addr &= 0xfc;
 184
 185     IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr);
 186     switch (addr)
 187     {
 188         case INTRMASK:
 189             ivshmem_IntrMask_write(s, val);
 190             break;
 191
 192         case INTRSTATUS:
 193             ivshmem_IntrStatus_write(s, val);
 194             break;
 195
 196         case DOORBELL:
 197             /* check that dest VM ID is reasonable */
 198             if (dest >= s->nb_peers) {
 199                 IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest);
 200                 break;
 201             }
 202
 203             /* check doorbell range */
 204             if (vector < s->peers[dest].nb_eventfds) {
 205                 IVSHMEM_DPRINTF("Notifying VM %d on vector %d\n", dest, vector);
 206                 event_notifier_set(&s->peers[dest].eventfds[vector]);
 207             } else {
 208                 IVSHMEM_DPRINTF("Invalid destination vector %d on VM %d\n",
 209                                 vector, dest);
 210             }
 211             break;
 212         default:
 213             IVSHMEM_DPRINTF("Unhandled write " TARGET_FMT_plx "\n", addr);
 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_DPRINTF("ivshmem_receive 0x%02x size: %d\n", *buf, size);
 266
 267     ivshmem_IntrStatus_write(s, *buf);
 268 }
 269
 270 static int ivshmem_can_receive(void * opaque)
 271 {
 272     return sizeof(long);
 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         error_report("creating chardriver for eventfd %d failed", eventfd);
 301         return NULL;
 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, Error **errp)
 322 {
 323     /* check that the guest isn't going to try and map more memory than the
 324      * the object has allocated return -1 to indicate error */
 325
 326     struct stat buf;
 327
 328     if (fstat(fd, &buf) < 0) {
 329         error_setg(errp, "exiting: fstat on fd %d failed: %s",
 330                    fd, strerror(errno));
 331         return -1;
 332     }
 333
 334     if (s->ivshmem_size > buf.st_size) {
 335         error_setg(errp, "Requested memory size greater"
 336                    " than shared object size (%" PRIu64 " > %" PRIu64")",
 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 int create_shared_memory_BAR(IVShmemState *s, int fd, uint8_t attr,
 347                                     Error **errp)
 348 {
 349     void * ptr;
 350
 351     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 352     if (ptr == MAP_FAILED) {
 353         error_setg_errno(errp, errno, "Failed to mmap shared memory");
 354         return -1;
 355     }
 356
 357     s->shm_fd = fd;
 358
 359     memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s), "ivshmem.bar2",
 360                                s->ivshmem_size, ptr);
 361     vmstate_register_ram(&s->ivshmem, DEVICE(s));
 362     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 363
 364     /* region for shared memory */
 365     pci_register_bar(PCI_DEVICE(s), 2, attr, &s->bar);
 366
 367     return 0;
 368 }
 369
 370 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 371 {
 372     memory_region_add_eventfd(&s->ivshmem_mmio,
 373                               DOORBELL,
 374                               4,
 375                               true,
 376                               (posn << 16) | i,
 377                               &s->peers[posn].eventfds[i]);
 378 }
 379
 380 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 381 {
 382     memory_region_del_eventfd(&s->ivshmem_mmio,
 383                               DOORBELL,
 384                               4,
 385                               true,
 386                               (posn << 16) | i,
 387                               &s->peers[posn].eventfds[i]);
 388 }
 389
 390 static void close_guest_eventfds(IVShmemState *s, int posn)
 391 {
 392     int i, guest_curr_max;
 393
 394     if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 395         return;
 396     }
 397     if (posn < 0 || posn >= s->nb_peers) {
 398         error_report("invalid peer %d", posn);
 399         return;
 400     }
 401
 402     guest_curr_max = s->peers[posn].nb_eventfds;
 403
 404     memory_region_transaction_begin();
 405     for (i = 0; i < guest_curr_max; i++) {
 406         ivshmem_del_eventfd(s, posn, i);
 407     }
 408     memory_region_transaction_commit();
 409     for (i = 0; i < guest_curr_max; i++) {
 410         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 411     }
 412
 413     g_free(s->peers[posn].eventfds);
 414     s->peers[posn].nb_eventfds = 0;
 415 }
 416
 417 /* this function increase the dynamic storage need to store data about other
 418  * guests */
 419 static int resize_peers(IVShmemState *s, int new_min_size)
 420 {
 421
 422     int j, old_size;
 423
 424     /* limit number of max peers */
 425     if (new_min_size <= 0 || new_min_size > IVSHMEM_MAX_PEERS) {
 426         return -1;
 427     }
 428     if (new_min_size <= s->nb_peers) {
 429         return 0;
 430     }
 431
 432     old_size = s->nb_peers;
 433     s->nb_peers = new_min_size;
 434
 435     IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers);
 436
 437     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 438
 439     for (j = old_size; j < s->nb_peers; j++) {
 440         s->peers[j].eventfds = g_new0(EventNotifier, s->vectors);
 441         s->peers[j].nb_eventfds = 0;
 442     }
 443
 444     return 0;
 445 }
 446
 447 static bool fifo_update_and_get(IVShmemState *s, const uint8_t *buf, int size,
 448                                 void *data, size_t len)
 449 {
 450     const uint8_t *p;
 451     uint32_t num;
 452
 453     assert(len <= sizeof(long)); /* limitation of the fifo */
 454     if (fifo8_is_empty(&s->incoming_fifo) && size == len) {
 455         memcpy(data, buf, size);
 456         return true;
 457     }
 458
 459     IVSHMEM_DPRINTF("short read of %d bytes\n", size);
 460
 461     num = MIN(size, sizeof(long) - fifo8_num_used(&s->incoming_fifo));
 462     fifo8_push_all(&s->incoming_fifo, buf, num);
 463
 464     if (fifo8_num_used(&s->incoming_fifo) < len) {
 465         assert(num == 0);
 466         return false;
 467     }
 468
 469     size -= num;
 470     buf += num;
 471     p = fifo8_pop_buf(&s->incoming_fifo, len, &num);
 472     assert(num == len);
 473
 474     memcpy(data, p, len);
 475
 476     if (size > 0) {
 477         fifo8_push_all(&s->incoming_fifo, buf, size);
 478     }
 479
 480     return true;
 481 }
 482
 483 static void ivshmem_read(void *opaque, const uint8_t *buf, int size)
 484 {
 485     IVShmemState *s = opaque;
 486     int incoming_fd;
 487     int new_eventfd;
 488     long incoming_posn;
 489     Error *err = NULL;
 490     Peer *peer;
 491
 492     if (!fifo_update_and_get(s, buf, size,
 493                              &incoming_posn, sizeof(incoming_posn))) {
 494         return;
 495     }
 496
 497     if (incoming_posn < -1) {
 498         IVSHMEM_DPRINTF("invalid incoming_posn %ld\n", incoming_posn);
 499         return;
 500     }
 501
 502     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 503     incoming_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 504     IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, incoming_fd);
 505
 506     /* make sure we have enough space for this guest */
 507     if (incoming_posn >= s->nb_peers) {
 508         if (resize_peers(s, incoming_posn + 1) < 0) {
 509             error_report("failed to resize peers array");
 510             if (incoming_fd != -1) {
 511                 close(incoming_fd);
 512             }
 513             return;
 514         }
 515     }
 516
 517     peer = &s->peers[incoming_posn];
 518
 519     if (incoming_fd == -1) {
 520         /* if posn is positive and unseen before then this is our posn*/
 521         if (incoming_posn >= 0 && s->vm_id == -1) {
 522             /* receive our posn */
 523             s->vm_id = incoming_posn;
 524         } else {
 525             /* otherwise an fd == -1 means an existing guest has gone away */
 526             IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
 527             close_guest_eventfds(s, incoming_posn);
 528         }
 529         return;
 530     }
 531
 532     /* if the position is -1, then it's shared memory region fd */
 533     if (incoming_posn == -1) {
 534         void * map_ptr;
 535
 536         if (s->shm_fd >= 0) {
 537             error_report("shm already initialized");
 538             close(incoming_fd);
 539             return;
 540         }
 541
 542         if (check_shm_size(s, incoming_fd, &err) == -1) {
 543             error_report_err(err);
 544             close(incoming_fd);
 545             return;
 546         }
 547
 548         /* mmap the region and map into the BAR2 */
 549         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 550                                                             incoming_fd, 0);
 551         if (map_ptr == MAP_FAILED) {
 552             error_report("Failed to mmap shared memory %s", strerror(errno));
 553             close(incoming_fd);
 554             return;
 555         }
 556         memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s),
 557                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 558         vmstate_register_ram(&s->ivshmem, DEVICE(s));
 559
 560         IVSHMEM_DPRINTF("guest h/w addr = %p, size = %" PRIu64 "\n",
 561                         map_ptr, s->ivshmem_size);
 562
 563         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 564
 565         /* only store the fd if it is successfully mapped */
 566         s->shm_fd = incoming_fd;
 567
 568         return;
 569     }
 570
 571     /* each peer has an associated array of eventfds, and we keep
 572      * track of how many eventfds received so far */
 573     /* get a new eventfd: */
 574     new_eventfd = peer->nb_eventfds++;
 575
 576     /* this is an eventfd for a particular guest VM */
 577     IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
 578                     new_eventfd, incoming_fd);
 579     event_notifier_init_fd(&peer->eventfds[new_eventfd], incoming_fd);
 580
 581     if (incoming_posn == s->vm_id) {
 582         s->eventfd_chr[new_eventfd] = create_eventfd_chr_device(s,
 583                    &s->peers[s->vm_id].eventfds[new_eventfd],
 584                    new_eventfd);
 585     }
 586
 587     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 588         ivshmem_add_eventfd(s, incoming_posn, new_eventfd);
 589     }
 590 }
 591
 592 /* Select the MSI-X vectors used by device.
 593  * ivshmem maps events to vectors statically, so
 594  * we just enable all vectors on init and after reset. */
 595 static void ivshmem_use_msix(IVShmemState * s)
 596 {
 597     PCIDevice *d = PCI_DEVICE(s);
 598     int i;
 599
 600     IVSHMEM_DPRINTF("%s, msix present: %d\n", __func__, msix_present(d));
 601     if (!msix_present(d)) {
 602         return;
 603     }
 604
 605     for (i = 0; i < s->vectors; i++) {
 606         msix_vector_use(d, i);
 607     }
 608 }
 609
 610 static void ivshmem_reset(DeviceState *d)
 611 {
 612     IVShmemState *s = IVSHMEM(d);
 613
 614     s->intrstatus = 0;
 615     ivshmem_use_msix(s);
 616 }
 617
 618 static uint64_t ivshmem_get_size(IVShmemState * s, Error **errp) {
 619
 620     uint64_t value;
 621     char *ptr;
 622
 623     value = strtoull(s->sizearg, &ptr, 10);
 624     switch (*ptr) {
 625         case 0: case 'M': case 'm':
 626             value <<= 20;
 627             break;
 628         case 'G': case 'g':
 629             value <<= 30;
 630             break;
 631         default:
 632             error_setg(errp, "invalid ram size: %s", s->sizearg);
 633             return 0;
 634     }
 635
 636     /* BARs must be a power of 2 */
 637     if (!is_power_of_2(value)) {
 638         error_setg(errp, "size must be power of 2");
 639         return 0;
 640     }
 641
 642     return value;
 643 }
 644
 645 static int ivshmem_setup_msi(IVShmemState * s)
 646 {
 647     if (msix_init_exclusive_bar(PCI_DEVICE(s), s->vectors, 1)) {
 648         return -1;
 649     }
 650
 651     IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 652
 653     /* allocate QEMU char devices for receiving interrupts */
 654     s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
 655
 656     ivshmem_use_msix(s);
 657     return 0;
 658 }
 659
 660 static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address,
 661                                  uint32_t val, int len)
 662 {
 663     pci_default_write_config(pci_dev, address, val, len);
 664 }
 665
 666 static void pci_ivshmem_realize(PCIDevice *dev, Error **errp)
 667 {
 668     IVShmemState *s = IVSHMEM(dev);
 669     uint8_t *pci_conf;
 670     uint8_t attr = PCI_BASE_ADDRESS_SPACE_MEMORY |
 671         PCI_BASE_ADDRESS_MEM_PREFETCH;
 672     Error *local_err = NULL;
 673
 674     s->shm_fd = -1;
 675
 676     if (s->sizearg == NULL) {
 677         s->ivshmem_size = 4 << 20; /* 4 MB default */
 678     } else {
 679         s->ivshmem_size = ivshmem_get_size(s, &local_err);
 680         if (local_err) {
 681             error_propagate(errp, local_err);
 682             return;
 683         }
 684     }
 685
 686     fifo8_create(&s->incoming_fifo, sizeof(long));
 687
 688     /* IRQFD requires MSI */
 689     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 690         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 691         error_setg(errp, "ioeventfd/irqfd requires MSI");
 692         return;
 693     }
 694
 695     /* check that role is reasonable */
 696     if (s->role) {
 697         if (strncmp(s->role, "peer", 5) == 0) {
 698             s->role_val = IVSHMEM_PEER;
 699         } else if (strncmp(s->role, "master", 7) == 0) {
 700             s->role_val = IVSHMEM_MASTER;
 701         } else {
 702             error_setg(errp, "'role' must be 'peer' or 'master'");
 703             return;
 704         }
 705     } else {
 706         s->role_val = IVSHMEM_MASTER; /* default */
 707     }
 708
 709     if (s->role_val == IVSHMEM_PEER) {
 710         error_setg(&s->migration_blocker,
 711                    "Migration is disabled when using feature 'peer mode' in device 'ivshmem'");
 712         migrate_add_blocker(s->migration_blocker);
 713     }
 714
 715     pci_conf = dev->config;
 716     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 717
 718     pci_config_set_interrupt_pin(pci_conf, 1);
 719
 720     memory_region_init_io(&s->ivshmem_mmio, OBJECT(s), &ivshmem_mmio_ops, s,
 721                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 722
 723     /* region for registers*/
 724     pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 725                      &s->ivshmem_mmio);
 726
 727     memory_region_init(&s->bar, OBJECT(s), "ivshmem-bar2-container", s->ivshmem_size);
 728     if (s->ivshmem_64bit) {
 729         attr |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 730     }
 731
 732     if (s->server_chr != NULL) {
 733         if (strncmp(s->server_chr->filename, "unix:", 5)) {
 734             error_setg(errp, "chardev is not a unix client socket");
 735             return;
 736         }
 737
 738         /* if we get a UNIX socket as the parameter we will talk
 739          * to the ivshmem server to receive the memory region */
 740
 741         if (s->shmobj != NULL) {
 742             error_setg(errp, "do not specify both 'chardev' "
 743                        "and 'shm' with ivshmem");
 744             return;
 745         }
 746
 747         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 748                         s->server_chr->filename);
 749
 750         if (ivshmem_has_feature(s, IVSHMEM_MSI) &&
 751             ivshmem_setup_msi(s)) {
 752             error_setg(errp, "msix initialization failed");
 753             return;
 754         }
 755
 756         /* we allocate enough space for 16 guests and grow as needed */
 757         resize_peers(s, 16);
 758         s->vm_id = -1;
 759
 760         pci_register_bar(dev, 2, attr, &s->bar);
 761
 762         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 763
 764         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 765                      ivshmem_event, s);
 766     } else {
 767         /* just map the file immediately, we're not using a server */
 768         int fd;
 769
 770         if (s->shmobj == NULL) {
 771             error_setg(errp, "Must specify 'chardev' or 'shm' to ivshmem");
 772             return;
 773         }
 774
 775         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 776
 777         /* try opening with O_EXCL and if it succeeds zero the memory
 778          * by truncating to 0 */
 779         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 780                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 781            /* truncate file to length PCI device's memory */
 782             if (ftruncate(fd, s->ivshmem_size) != 0) {
 783                 error_report("could not truncate shared file");
 784             }
 785
 786         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 787                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 788             error_setg(errp, "could not open shared file");
 789             return;
 790         }
 791
 792         if (check_shm_size(s, fd, errp) == -1) {
 793             return;
 794         }
 795
 796         create_shared_memory_BAR(s, fd, attr, errp);
 797     }
 798 }
 799
 800 static void pci_ivshmem_exit(PCIDevice *dev)
 801 {
 802     IVShmemState *s = IVSHMEM(dev);
 803     int i;
 804
 805     fifo8_destroy(&s->incoming_fifo);
 806
 807     if (s->migration_blocker) {
 808         migrate_del_blocker(s->migration_blocker);
 809         error_free(s->migration_blocker);
 810     }
 811
 812     if (s->shm_fd >= 0) {
 813         void *addr = memory_region_get_ram_ptr(&s->ivshmem);
 814
 815         vmstate_unregister_ram(&s->ivshmem, DEVICE(dev));
 816         memory_region_del_subregion(&s->bar, &s->ivshmem);
 817
 818         if (munmap(addr, s->ivshmem_size) == -1) {
 819             error_report("Failed to munmap shared memory %s", strerror(errno));
 820         }
 821     }
 822
 823     if (s->eventfd_chr) {
 824         for (i = 0; i < s->vectors; i++) {
 825             if (s->eventfd_chr[i]) {
 826                 qemu_chr_free(s->eventfd_chr[i]);
 827             }
 828         }
 829         g_free(s->eventfd_chr);
 830     }
 831
 832     if (s->peers) {
 833         for (i = 0; i < s->nb_peers; i++) {
 834             close_guest_eventfds(s, i);
 835         }
 836         g_free(s->peers);
 837     }
 838
 839     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 840         msix_uninit_exclusive_bar(dev);
 841     }
 842
 843     g_free(s->eventfd_table);
 844 }
 845
 846 static bool test_msix(void *opaque, int version_id)
 847 {
 848     IVShmemState *s = opaque;
 849
 850     return ivshmem_has_feature(s, IVSHMEM_MSI);
 851 }
 852
 853 static bool test_no_msix(void *opaque, int version_id)
 854 {
 855     return !test_msix(opaque, version_id);
 856 }
 857
 858 static int ivshmem_pre_load(void *opaque)
 859 {
 860     IVShmemState *s = opaque;
 861
 862     if (s->role_val == IVSHMEM_PEER) {
 863         error_report("'peer' devices are not migratable");
 864         return -EINVAL;
 865     }
 866
 867     return 0;
 868 }
 869
 870 static int ivshmem_post_load(void *opaque, int version_id)
 871 {
 872     IVShmemState *s = opaque;
 873
 874     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 875         ivshmem_use_msix(s);
 876     }
 877
 878     return 0;
 879 }
 880
 881 static int ivshmem_load_old(QEMUFile *f, void *opaque, int version_id)
 882 {
 883     IVShmemState *s = opaque;
 884     PCIDevice *pdev = PCI_DEVICE(s);
 885     int ret;
 886
 887     IVSHMEM_DPRINTF("ivshmem_load_old\n");
 888
 889     if (version_id != 0) {
 890         return -EINVAL;
 891     }
 892
 893     if (s->role_val == IVSHMEM_PEER) {
 894         error_report("'peer' devices are not migratable");
 895         return -EINVAL;
 896     }
 897
 898     ret = pci_device_load(pdev, f);
 899     if (ret) {
 900         return ret;
 901     }
 902
 903     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 904         msix_load(pdev, f);
 905         ivshmem_use_msix(s);
 906     } else {
 907         s->intrstatus = qemu_get_be32(f);
 908         s->intrmask = qemu_get_be32(f);
 909     }
 910
 911     return 0;
 912 }
 913
 914 static const VMStateDescription ivshmem_vmsd = {
 915     .name = "ivshmem",
 916     .version_id = 1,
 917     .minimum_version_id = 1,
 918     .pre_load = ivshmem_pre_load,
 919     .post_load = ivshmem_post_load,
 920     .fields = (VMStateField[]) {
 921         VMSTATE_PCI_DEVICE(parent_obj, IVShmemState),
 922
 923         VMSTATE_MSIX_TEST(parent_obj, IVShmemState, test_msix),
 924         VMSTATE_UINT32_TEST(intrstatus, IVShmemState, test_no_msix),
 925         VMSTATE_UINT32_TEST(intrmask, IVShmemState, test_no_msix),
 926
 927         VMSTATE_END_OF_LIST()
 928     },
 929     .load_state_old = ivshmem_load_old,
 930     .minimum_version_id_old = 0
 931 };
 932
 933 static Property ivshmem_properties[] = {
 934     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
 935     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
 936     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
 937     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
 938     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
 939     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
 940     DEFINE_PROP_STRING("role", IVShmemState, role),
 941     DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1),
 942     DEFINE_PROP_END_OF_LIST(),
 943 };
 944
 945 static void ivshmem_class_init(ObjectClass *klass, void *data)
 946 {
 947     DeviceClass *dc = DEVICE_CLASS(klass);
 948     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 949
 950     k->realize = pci_ivshmem_realize;
 951     k->exit = pci_ivshmem_exit;
 952     k->config_write = ivshmem_write_config;
 953     k->vendor_id = PCI_VENDOR_ID_IVSHMEM;
 954     k->device_id = PCI_DEVICE_ID_IVSHMEM;
 955     k->class_id = PCI_CLASS_MEMORY_RAM;
 956     dc->reset = ivshmem_reset;
 957     dc->props = ivshmem_properties;
 958     dc->vmsd = &ivshmem_vmsd;
 959     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 960     dc->desc = "Inter-VM shared memory";
 961 }
 962
 963 static const TypeInfo ivshmem_info = {
 964     .name          = TYPE_IVSHMEM,
 965     .parent        = TYPE_PCI_DEVICE,
 966     .instance_size = sizeof(IVShmemState),
 967     .class_init    = ivshmem_class_init,
 968 };
 969
 970 static void ivshmem_register_types(void)
 971 {
 972     type_register_static(&ivshmem_info);
 973 }
 974
 975 type_init(ivshmem_register_types)