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 "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_IOEVENTFD   0
  38 #define IVSHMEM_MSI     1
  39
  40 #define IVSHMEM_PEER    0
  41 #define IVSHMEM_MASTER  1
  42
  43 #define IVSHMEM_REG_BAR_SIZE 0x100
  44
  45 //#define DEBUG_IVSHMEM
  46 #ifdef DEBUG_IVSHMEM
  47 #define IVSHMEM_DPRINTF(fmt, ...)        \
  48     do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0)
  49 #else
  50 #define IVSHMEM_DPRINTF(fmt, ...)
  51 #endif
  52
  53 #define TYPE_IVSHMEM "ivshmem"
  54 #define IVSHMEM(obj) \
  55     OBJECT_CHECK(IVShmemState, (obj), TYPE_IVSHMEM)
  56
  57 typedef struct Peer {
  58     int nb_eventfds;
  59     EventNotifier *eventfds;
  60 } Peer;
  61
  62 typedef struct EventfdEntry {
  63     PCIDevice *pdev;
  64     int vector;
  65 } EventfdEntry;
  66
  67 typedef struct IVShmemState {
  68     /*< private >*/
  69     PCIDevice parent_obj;
  70     /*< public >*/
  71
  72     uint32_t intrmask;
  73     uint32_t intrstatus;
  74     uint32_t doorbell;
  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_attr;
  89     uint32_t ivshmem_64bit;
  90     int shm_fd; /* shared memory file descriptor */
  91
  92     Peer *peers;
  93     int nb_peers; /* how many guests we have space for */
  94     int max_peer; /* maximum numbered peer */
  95
  96     int vm_id;
  97     uint32_t vectors;
  98     uint32_t features;
  99     EventfdEntry *eventfd_table;
 100
 101     Error *migration_blocker;
 102
 103     char * shmobj;
 104     char * sizearg;
 105     char * role;
 106     int role_val;   /* scalar to avoid multiple string comparisons */
 107 } IVShmemState;
 108
 109 /* registers for the Inter-VM shared memory device */
 110 enum ivshmem_registers {
 111     INTRMASK = 0,
 112     INTRSTATUS = 4,
 113     IVPOSITION = 8,
 114     DOORBELL = 12,
 115 };
 116
 117 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 118                                                     unsigned int feature) {
 119     return (ivs->features & (1 << feature));
 120 }
 121
 122 static inline bool is_power_of_two(uint64_t x) {
 123     return (x & (x - 1)) == 0;
 124 }
 125
 126 /* accessing registers - based on rtl8139 */
 127 static void ivshmem_update_irq(IVShmemState *s, int val)
 128 {
 129     PCIDevice *d = PCI_DEVICE(s);
 130     int isr;
 131     isr = (s->intrstatus & s->intrmask) & 0xffffffff;
 132
 133     /* don't print ISR resets */
 134     if (isr) {
 135         IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
 136            isr ? 1 : 0, s->intrstatus, s->intrmask);
 137     }
 138
 139     pci_set_irq(d, (isr != 0));
 140 }
 141
 142 static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val)
 143 {
 144     IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
 145
 146     s->intrmask = val;
 147
 148     ivshmem_update_irq(s, val);
 149 }
 150
 151 static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
 152 {
 153     uint32_t ret = s->intrmask;
 154
 155     IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
 156
 157     return ret;
 158 }
 159
 160 static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val)
 161 {
 162     IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
 163
 164     s->intrstatus = val;
 165
 166     ivshmem_update_irq(s, val);
 167 }
 168
 169 static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
 170 {
 171     uint32_t ret = s->intrstatus;
 172
 173     /* reading ISR clears all interrupts */
 174     s->intrstatus = 0;
 175
 176     ivshmem_update_irq(s, 0);
 177
 178     return ret;
 179 }
 180
 181 static void ivshmem_io_write(void *opaque, hwaddr addr,
 182                              uint64_t val, unsigned size)
 183 {
 184     IVShmemState *s = opaque;
 185
 186     uint16_t dest = val >> 16;
 187     uint16_t vector = val & 0xff;
 188
 189     addr &= 0xfc;
 190
 191     IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr);
 192     switch (addr)
 193     {
 194         case INTRMASK:
 195             ivshmem_IntrMask_write(s, val);
 196             break;
 197
 198         case INTRSTATUS:
 199             ivshmem_IntrStatus_write(s, val);
 200             break;
 201
 202         case DOORBELL:
 203             /* check that dest VM ID is reasonable */
 204             if (dest > s->max_peer) {
 205                 IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest);
 206                 break;
 207             }
 208
 209             /* check doorbell range */
 210             if (vector < s->peers[dest].nb_eventfds) {
 211                 IVSHMEM_DPRINTF("Notifying VM %d on vector %d\n", dest, vector);
 212                 event_notifier_set(&s->peers[dest].eventfds[vector]);
 213             }
 214             break;
 215         default:
 216             IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest);
 217     }
 218 }
 219
 220 static uint64_t ivshmem_io_read(void *opaque, hwaddr addr,
 221                                 unsigned size)
 222 {
 223
 224     IVShmemState *s = opaque;
 225     uint32_t ret;
 226
 227     switch (addr)
 228     {
 229         case INTRMASK:
 230             ret = ivshmem_IntrMask_read(s);
 231             break;
 232
 233         case INTRSTATUS:
 234             ret = ivshmem_IntrStatus_read(s);
 235             break;
 236
 237         case IVPOSITION:
 238             /* return my VM ID if the memory is mapped */
 239             if (s->shm_fd > 0) {
 240                 ret = s->vm_id;
 241             } else {
 242                 ret = -1;
 243             }
 244             break;
 245
 246         default:
 247             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 248             ret = 0;
 249     }
 250
 251     return ret;
 252 }
 253
 254 static const MemoryRegionOps ivshmem_mmio_ops = {
 255     .read = ivshmem_io_read,
 256     .write = ivshmem_io_write,
 257     .endianness = DEVICE_NATIVE_ENDIAN,
 258     .impl = {
 259         .min_access_size = 4,
 260         .max_access_size = 4,
 261     },
 262 };
 263
 264 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 265 {
 266     IVShmemState *s = opaque;
 267
 268     ivshmem_IntrStatus_write(s, *buf);
 269
 270     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf);
 271 }
 272
 273 static int ivshmem_can_receive(void * opaque)
 274 {
 275     return 8;
 276 }
 277
 278 static void ivshmem_event(void *opaque, int event)
 279 {
 280     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 281 }
 282
 283 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 284
 285     EventfdEntry *entry = opaque;
 286     PCIDevice *pdev = entry->pdev;
 287
 288     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector);
 289     msix_notify(pdev, entry->vector);
 290 }
 291
 292 static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n,
 293                                                   int vector)
 294 {
 295     /* create a event character device based on the passed eventfd */
 296     IVShmemState *s = opaque;
 297     CharDriverState * chr;
 298     int eventfd = event_notifier_get_fd(n);
 299
 300     chr = qemu_chr_open_eventfd(eventfd);
 301
 302     if (chr == NULL) {
 303         fprintf(stderr, "creating eventfd for eventfd %d failed\n", eventfd);
 304         exit(-1);
 305     }
 306     qemu_chr_fe_claim_no_fail(chr);
 307
 308     /* if MSI is supported we need multiple interrupts */
 309     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 310         s->eventfd_table[vector].pdev = PCI_DEVICE(s);
 311         s->eventfd_table[vector].vector = vector;
 312
 313         qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd,
 314                       ivshmem_event, &s->eventfd_table[vector]);
 315     } else {
 316         qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
 317                       ivshmem_event, s);
 318     }
 319
 320     return chr;
 321
 322 }
 323
 324 static int check_shm_size(IVShmemState *s, int fd) {
 325     /* check that the guest isn't going to try and map more memory than the
 326      * the object has allocated return -1 to indicate error */
 327
 328     struct stat buf;
 329
 330     if (fstat(fd, &buf) < 0) {
 331         fprintf(stderr, "ivshmem: exiting: fstat on fd %d failed: %s\n",
 332                 fd, strerror(errno));
 333         return -1;
 334     }
 335
 336     if (s->ivshmem_size > buf.st_size) {
 337         fprintf(stderr,
 338                 "IVSHMEM ERROR: Requested memory size greater"
 339                 " than shared object size (%" PRIu64 " > %" PRIu64")\n",
 340                 s->ivshmem_size, (uint64_t)buf.st_size);
 341         return -1;
 342     } else {
 343         return 0;
 344     }
 345 }
 346
 347 /* create the shared memory BAR when we are not using the server, so we can
 348  * create the BAR and map the memory immediately */
 349 static void create_shared_memory_BAR(IVShmemState *s, int fd) {
 350
 351     void * ptr;
 352
 353     s->shm_fd = fd;
 354
 355     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 356
 357     memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s), "ivshmem.bar2",
 358                                s->ivshmem_size, ptr);
 359     vmstate_register_ram(&s->ivshmem, DEVICE(s));
 360     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 361
 362     /* region for shared memory */
 363     pci_register_bar(PCI_DEVICE(s), 2, s->ivshmem_attr, &s->bar);
 364 }
 365
 366 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 367 {
 368     memory_region_add_eventfd(&s->ivshmem_mmio,
 369                               DOORBELL,
 370                               4,
 371                               true,
 372                               (posn << 16) | i,
 373                               &s->peers[posn].eventfds[i]);
 374 }
 375
 376 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 377 {
 378     memory_region_del_eventfd(&s->ivshmem_mmio,
 379                               DOORBELL,
 380                               4,
 381                               true,
 382                               (posn << 16) | i,
 383                               &s->peers[posn].eventfds[i]);
 384 }
 385
 386 static void close_guest_eventfds(IVShmemState *s, int posn)
 387 {
 388     int i, guest_curr_max;
 389
 390     if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 391         return;
 392     }
 393     if (posn < 0 || posn >= s->nb_peers) {
 394         return;
 395     }
 396
 397     guest_curr_max = s->peers[posn].nb_eventfds;
 398
 399     memory_region_transaction_begin();
 400     for (i = 0; i < guest_curr_max; i++) {
 401         ivshmem_del_eventfd(s, posn, i);
 402     }
 403     memory_region_transaction_commit();
 404     for (i = 0; i < guest_curr_max; i++) {
 405         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 406     }
 407
 408     g_free(s->peers[posn].eventfds);
 409     s->peers[posn].nb_eventfds = 0;
 410 }
 411
 412 /* this function increase the dynamic storage need to store data about other
 413  * guests */
 414 static int increase_dynamic_storage(IVShmemState *s, int new_min_size)
 415 {
 416
 417     int j, old_nb_alloc;
 418
 419     /* check for integer overflow */
 420     if (new_min_size >= INT_MAX / sizeof(Peer) - 1 || new_min_size <= 0) {
 421         return -1;
 422     }
 423
 424     old_nb_alloc = s->nb_peers;
 425
 426     if (new_min_size >= s->nb_peers) {
 427         /* +1 because #new_min_size is used as last array index */
 428         s->nb_peers = new_min_size + 1;
 429     } else {
 430         return 0;
 431     }
 432
 433     IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers);
 434     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 435
 436     /* zero out new pointers */
 437     for (j = old_nb_alloc; j < s->nb_peers; j++) {
 438         s->peers[j].eventfds = NULL;
 439         s->peers[j].nb_eventfds = 0;
 440     }
 441
 442     return 0;
 443 }
 444
 445 static void ivshmem_read(void *opaque, const uint8_t *buf, int size)
 446 {
 447     IVShmemState *s = opaque;
 448     int incoming_fd, tmp_fd;
 449     int guest_max_eventfd;
 450     long incoming_posn;
 451
 452     if (fifo8_is_empty(&s->incoming_fifo) && size == sizeof(incoming_posn)) {
 453         memcpy(&incoming_posn, buf, size);
 454     } else {
 455         const uint8_t *p;
 456         uint32_t num;
 457
 458         IVSHMEM_DPRINTF("short read of %d bytes\n", size);
 459         num = MAX(size, sizeof(long) - fifo8_num_used(&s->incoming_fifo));
 460         fifo8_push_all(&s->incoming_fifo, buf, num);
 461         if (fifo8_num_used(&s->incoming_fifo) < sizeof(incoming_posn)) {
 462             return;
 463         }
 464         size -= num;
 465         buf += num;
 466         p = fifo8_pop_buf(&s->incoming_fifo, sizeof(incoming_posn), &num);
 467         g_assert(num == sizeof(incoming_posn));
 468         memcpy(&incoming_posn, p, sizeof(incoming_posn));
 469         if (size > 0) {
 470             fifo8_push_all(&s->incoming_fifo, buf, size);
 471         }
 472     }
 473
 474     if (incoming_posn < -1) {
 475         IVSHMEM_DPRINTF("invalid incoming_posn %ld\n", incoming_posn);
 476         return;
 477     }
 478
 479     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 480     tmp_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 481     IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, tmp_fd);
 482
 483     /* make sure we have enough space for this guest */
 484     if (incoming_posn >= s->nb_peers) {
 485         if (increase_dynamic_storage(s, incoming_posn) < 0) {
 486             error_report("increase_dynamic_storage() failed");
 487             if (tmp_fd != -1) {
 488                 close(tmp_fd);
 489             }
 490             return;
 491         }
 492     }
 493
 494     if (tmp_fd == -1) {
 495         /* if posn is positive and unseen before then this is our posn*/
 496         if ((incoming_posn >= 0) &&
 497                             (s->peers[incoming_posn].eventfds == NULL)) {
 498             /* receive our posn */
 499             s->vm_id = incoming_posn;
 500             return;
 501         } else {
 502             /* otherwise an fd == -1 means an existing guest has gone away */
 503             IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
 504             close_guest_eventfds(s, incoming_posn);
 505             return;
 506         }
 507     }
 508
 509     /* because of the implementation of get_msgfd, we need a dup */
 510     incoming_fd = dup(tmp_fd);
 511
 512     if (incoming_fd == -1) {
 513         fprintf(stderr, "could not allocate file descriptor %s\n",
 514                                                             strerror(errno));
 515         close(tmp_fd);
 516         return;
 517     }
 518
 519     /* if the position is -1, then it's shared memory region fd */
 520     if (incoming_posn == -1) {
 521
 522         void * map_ptr;
 523
 524         s->max_peer = 0;
 525
 526         if (check_shm_size(s, incoming_fd) == -1) {
 527             exit(-1);
 528         }
 529
 530         /* mmap the region and map into the BAR2 */
 531         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 532                                                             incoming_fd, 0);
 533         memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s),
 534                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 535         vmstate_register_ram(&s->ivshmem, DEVICE(s));
 536
 537         IVSHMEM_DPRINTF("guest h/w addr = %p, size = %" PRIu64 "\n",
 538                          map_ptr, s->ivshmem_size);
 539
 540         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 541
 542         /* only store the fd if it is successfully mapped */
 543         s->shm_fd = incoming_fd;
 544
 545         return;
 546     }
 547
 548     /* each guest has an array of eventfds, and we keep track of how many
 549      * guests for each VM */
 550     guest_max_eventfd = s->peers[incoming_posn].nb_eventfds;
 551
 552     if (guest_max_eventfd == 0) {
 553         /* one eventfd per MSI vector */
 554         s->peers[incoming_posn].eventfds = g_new(EventNotifier, s->vectors);
 555     }
 556
 557     /* this is an eventfd for a particular guest VM */
 558     IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
 559                                             guest_max_eventfd, incoming_fd);
 560     event_notifier_init_fd(&s->peers[incoming_posn].eventfds[guest_max_eventfd],
 561                            incoming_fd);
 562
 563     /* increment count for particular guest */
 564     s->peers[incoming_posn].nb_eventfds++;
 565
 566     /* keep track of the maximum VM ID */
 567     if (incoming_posn > s->max_peer) {
 568         s->max_peer = incoming_posn;
 569     }
 570
 571     if (incoming_posn == s->vm_id) {
 572         s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s,
 573                    &s->peers[s->vm_id].eventfds[guest_max_eventfd],
 574                    guest_max_eventfd);
 575     }
 576
 577     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 578         ivshmem_add_eventfd(s, incoming_posn, guest_max_eventfd);
 579     }
 580 }
 581
 582 /* Select the MSI-X vectors used by device.
 583  * ivshmem maps events to vectors statically, so
 584  * we just enable all vectors on init and after reset. */
 585 static void ivshmem_use_msix(IVShmemState * s)
 586 {
 587     PCIDevice *d = PCI_DEVICE(s);
 588     int i;
 589
 590     if (!msix_present(d)) {
 591         return;
 592     }
 593
 594     for (i = 0; i < s->vectors; i++) {
 595         msix_vector_use(d, i);
 596     }
 597 }
 598
 599 static void ivshmem_reset(DeviceState *d)
 600 {
 601     IVShmemState *s = IVSHMEM(d);
 602
 603     s->intrstatus = 0;
 604     ivshmem_use_msix(s);
 605 }
 606
 607 static uint64_t ivshmem_get_size(IVShmemState * s) {
 608
 609     uint64_t value;
 610     char *ptr;
 611
 612     value = strtoull(s->sizearg, &ptr, 10);
 613     switch (*ptr) {
 614         case 0: case 'M': case 'm':
 615             value <<= 20;
 616             break;
 617         case 'G': case 'g':
 618             value <<= 30;
 619             break;
 620         default:
 621             fprintf(stderr, "qemu: invalid ram size: %s\n", s->sizearg);
 622             exit(1);
 623     }
 624
 625     /* BARs must be a power of 2 */
 626     if (!is_power_of_two(value)) {
 627         fprintf(stderr, "ivshmem: size must be power of 2\n");
 628         exit(1);
 629     }
 630
 631     return value;
 632 }
 633
 634 static void ivshmem_setup_msi(IVShmemState * s)
 635 {
 636     if (msix_init_exclusive_bar(PCI_DEVICE(s), s->vectors, 1)) {
 637         IVSHMEM_DPRINTF("msix initialization failed\n");
 638         exit(1);
 639     }
 640
 641     IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 642
 643     /* allocate QEMU char devices for receiving interrupts */
 644     s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
 645
 646     ivshmem_use_msix(s);
 647 }
 648
 649 static void ivshmem_save(QEMUFile* f, void *opaque)
 650 {
 651     IVShmemState *proxy = opaque;
 652     PCIDevice *pci_dev = PCI_DEVICE(proxy);
 653
 654     IVSHMEM_DPRINTF("ivshmem_save\n");
 655     pci_device_save(pci_dev, f);
 656
 657     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 658         msix_save(pci_dev, f);
 659     } else {
 660         qemu_put_be32(f, proxy->intrstatus);
 661         qemu_put_be32(f, proxy->intrmask);
 662     }
 663
 664 }
 665
 666 static int ivshmem_load(QEMUFile* f, void *opaque, int version_id)
 667 {
 668     IVSHMEM_DPRINTF("ivshmem_load\n");
 669
 670     IVShmemState *proxy = opaque;
 671     PCIDevice *pci_dev = PCI_DEVICE(proxy);
 672     int ret;
 673
 674     if (version_id > 0) {
 675         return -EINVAL;
 676     }
 677
 678     if (proxy->role_val == IVSHMEM_PEER) {
 679         fprintf(stderr, "ivshmem: 'peer' devices are not migratable\n");
 680         return -EINVAL;
 681     }
 682
 683     ret = pci_device_load(pci_dev, f);
 684     if (ret) {
 685         return ret;
 686     }
 687
 688     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 689         msix_load(pci_dev, f);
 690         ivshmem_use_msix(proxy);
 691     } else {
 692         proxy->intrstatus = qemu_get_be32(f);
 693         proxy->intrmask = qemu_get_be32(f);
 694     }
 695
 696     return 0;
 697 }
 698
 699 static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address,
 700                                  uint32_t val, int len)
 701 {
 702     pci_default_write_config(pci_dev, address, val, len);
 703     msix_write_config(pci_dev, address, val, len);
 704 }
 705
 706 static int pci_ivshmem_init(PCIDevice *dev)
 707 {
 708     IVShmemState *s = IVSHMEM(dev);
 709     uint8_t *pci_conf;
 710
 711     if (s->sizearg == NULL)
 712         s->ivshmem_size = 4 << 20; /* 4 MB default */
 713     else {
 714         s->ivshmem_size = ivshmem_get_size(s);
 715     }
 716
 717     fifo8_create(&s->incoming_fifo, sizeof(long));
 718
 719     register_savevm(DEVICE(dev), "ivshmem", 0, 0, ivshmem_save, ivshmem_load,
 720                                                                         dev);
 721
 722     /* IRQFD requires MSI */
 723     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 724         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 725         fprintf(stderr, "ivshmem: ioeventfd/irqfd requires MSI\n");
 726         exit(1);
 727     }
 728
 729     /* check that role is reasonable */
 730     if (s->role) {
 731         if (strncmp(s->role, "peer", 5) == 0) {
 732             s->role_val = IVSHMEM_PEER;
 733         } else if (strncmp(s->role, "master", 7) == 0) {
 734             s->role_val = IVSHMEM_MASTER;
 735         } else {
 736             fprintf(stderr, "ivshmem: 'role' must be 'peer' or 'master'\n");
 737             exit(1);
 738         }
 739     } else {
 740         s->role_val = IVSHMEM_MASTER; /* default */
 741     }
 742
 743     if (s->role_val == IVSHMEM_PEER) {
 744         error_setg(&s->migration_blocker,
 745                    "Migration is disabled when using feature 'peer mode' in device 'ivshmem'");
 746         migrate_add_blocker(s->migration_blocker);
 747     }
 748
 749     pci_conf = dev->config;
 750     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 751
 752     pci_config_set_interrupt_pin(pci_conf, 1);
 753
 754     s->shm_fd = 0;
 755
 756     memory_region_init_io(&s->ivshmem_mmio, OBJECT(s), &ivshmem_mmio_ops, s,
 757                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 758
 759     /* region for registers*/
 760     pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 761                      &s->ivshmem_mmio);
 762
 763     memory_region_init(&s->bar, OBJECT(s), "ivshmem-bar2-container", s->ivshmem_size);
 764     s->ivshmem_attr = PCI_BASE_ADDRESS_SPACE_MEMORY |
 765         PCI_BASE_ADDRESS_MEM_PREFETCH;
 766     if (s->ivshmem_64bit) {
 767         s->ivshmem_attr |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 768     }
 769
 770     if ((s->server_chr != NULL) &&
 771                         (strncmp(s->server_chr->filename, "unix:", 5) == 0)) {
 772         /* if we get a UNIX socket as the parameter we will talk
 773          * to the ivshmem server to receive the memory region */
 774
 775         if (s->shmobj != NULL) {
 776             fprintf(stderr, "WARNING: do not specify both 'chardev' "
 777                                                 "and 'shm' with ivshmem\n");
 778         }
 779
 780         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 781                                                     s->server_chr->filename);
 782
 783         if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 784             ivshmem_setup_msi(s);
 785         }
 786
 787         /* we allocate enough space for 16 guests and grow as needed */
 788         s->nb_peers = 16;
 789         s->vm_id = -1;
 790
 791         /* allocate/initialize space for interrupt handling */
 792         s->peers = g_malloc0(s->nb_peers * sizeof(Peer));
 793
 794         pci_register_bar(dev, 2, s->ivshmem_attr, &s->bar);
 795
 796         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 797
 798         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 799                      ivshmem_event, s);
 800     } else {
 801         /* just map the file immediately, we're not using a server */
 802         int fd;
 803
 804         if (s->shmobj == NULL) {
 805             fprintf(stderr, "Must specify 'chardev' or 'shm' to ivshmem\n");
 806             exit(1);
 807         }
 808
 809         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 810
 811         /* try opening with O_EXCL and if it succeeds zero the memory
 812          * by truncating to 0 */
 813         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 814                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 815            /* truncate file to length PCI device's memory */
 816             if (ftruncate(fd, s->ivshmem_size) != 0) {
 817                 fprintf(stderr, "ivshmem: could not truncate shared file\n");
 818             }
 819
 820         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 821                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 822             fprintf(stderr, "ivshmem: could not open shared file\n");
 823             exit(-1);
 824
 825         }
 826
 827         if (check_shm_size(s, fd) == -1) {
 828             exit(-1);
 829         }
 830
 831         create_shared_memory_BAR(s, fd);
 832
 833     }
 834
 835     dev->config_write = ivshmem_write_config;
 836
 837     return 0;
 838 }
 839
 840 static void pci_ivshmem_uninit(PCIDevice *dev)
 841 {
 842     IVShmemState *s = IVSHMEM(dev);
 843
 844     if (s->migration_blocker) {
 845         migrate_del_blocker(s->migration_blocker);
 846         error_free(s->migration_blocker);
 847     }
 848
 849     memory_region_del_subregion(&s->bar, &s->ivshmem);
 850     vmstate_unregister_ram(&s->ivshmem, DEVICE(dev));
 851     unregister_savevm(DEVICE(dev), "ivshmem", s);
 852     fifo8_destroy(&s->incoming_fifo);
 853 }
 854
 855 static Property ivshmem_properties[] = {
 856     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
 857     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
 858     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
 859     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
 860     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
 861     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
 862     DEFINE_PROP_STRING("role", IVShmemState, role),
 863     DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1),
 864     DEFINE_PROP_END_OF_LIST(),
 865 };
 866
 867 static void ivshmem_class_init(ObjectClass *klass, void *data)
 868 {
 869     DeviceClass *dc = DEVICE_CLASS(klass);
 870     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 871
 872     k->init = pci_ivshmem_init;
 873     k->exit = pci_ivshmem_uninit;
 874     k->vendor_id = PCI_VENDOR_ID_IVSHMEM;
 875     k->device_id = PCI_DEVICE_ID_IVSHMEM;
 876     k->class_id = PCI_CLASS_MEMORY_RAM;
 877     dc->reset = ivshmem_reset;
 878     dc->props = ivshmem_properties;
 879     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 880 }
 881
 882 static const TypeInfo ivshmem_info = {
 883     .name          = TYPE_IVSHMEM,
 884     .parent        = TYPE_PCI_DEVICE,
 885     .instance_size = sizeof(IVShmemState),
 886     .class_init    = ivshmem_class_init,
 887 };
 888
 889 static void ivshmem_register_types(void)
 890 {
 891     type_register_static(&ivshmem_info);
 892 }
 893
 894 type_init(ivshmem_register_types)