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_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_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     int max_peer; /* maximum numbered peer */
  94
  95     int vm_id;
  96     uint32_t vectors;
  97     uint32_t features;
  98     EventfdEntry *eventfd_table;
  99
 100     Error *migration_blocker;
 101
 102     char * shmobj;
 103     char * sizearg;
 104     char * role;
 105     int role_val;   /* scalar to avoid multiple string comparisons */
 106 } IVShmemState;
 107
 108 /* registers for the Inter-VM shared memory device */
 109 enum ivshmem_registers {
 110     INTRMASK = 0,
 111     INTRSTATUS = 4,
 112     IVPOSITION = 8,
 113     DOORBELL = 12,
 114 };
 115
 116 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 117                                                     unsigned int feature) {
 118     return (ivs->features & (1 << feature));
 119 }
 120
 121 static inline bool is_power_of_two(uint64_t x) {
 122     return (x & (x - 1)) == 0;
 123 }
 124
 125 /* accessing registers - based on rtl8139 */
 126 static void ivshmem_update_irq(IVShmemState *s, int val)
 127 {
 128     PCIDevice *d = PCI_DEVICE(s);
 129     int isr;
 130     isr = (s->intrstatus & s->intrmask) & 0xffffffff;
 131
 132     /* don't print ISR resets */
 133     if (isr) {
 134         IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
 135                         isr ? 1 : 0, s->intrstatus, s->intrmask);
 136     }
 137
 138     pci_set_irq(d, (isr != 0));
 139 }
 140
 141 static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val)
 142 {
 143     IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
 144
 145     s->intrmask = val;
 146
 147     ivshmem_update_irq(s, val);
 148 }
 149
 150 static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
 151 {
 152     uint32_t ret = s->intrmask;
 153
 154     IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
 155
 156     return ret;
 157 }
 158
 159 static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val)
 160 {
 161     IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
 162
 163     s->intrstatus = val;
 164
 165     ivshmem_update_irq(s, val);
 166 }
 167
 168 static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
 169 {
 170     uint32_t ret = s->intrstatus;
 171
 172     /* reading ISR clears all interrupts */
 173     s->intrstatus = 0;
 174
 175     ivshmem_update_irq(s, 0);
 176
 177     return ret;
 178 }
 179
 180 static void ivshmem_io_write(void *opaque, hwaddr addr,
 181                              uint64_t val, unsigned size)
 182 {
 183     IVShmemState *s = opaque;
 184
 185     uint16_t dest = val >> 16;
 186     uint16_t vector = val & 0xff;
 187
 188     addr &= 0xfc;
 189
 190     IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr);
 191     switch (addr)
 192     {
 193         case INTRMASK:
 194             ivshmem_IntrMask_write(s, val);
 195             break;
 196
 197         case INTRSTATUS:
 198             ivshmem_IntrStatus_write(s, val);
 199             break;
 200
 201         case DOORBELL:
 202             /* check that dest VM ID is reasonable */
 203             if (dest > s->max_peer) {
 204                 IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest);
 205                 break;
 206             }
 207
 208             /* check doorbell range */
 209             if (vector < s->peers[dest].nb_eventfds) {
 210                 IVSHMEM_DPRINTF("Notifying VM %d on vector %d\n", dest, vector);
 211                 event_notifier_set(&s->peers[dest].eventfds[vector]);
 212             }
 213             break;
 214         default:
 215             IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest);
 216     }
 217 }
 218
 219 static uint64_t ivshmem_io_read(void *opaque, hwaddr addr,
 220                                 unsigned size)
 221 {
 222
 223     IVShmemState *s = opaque;
 224     uint32_t ret;
 225
 226     switch (addr)
 227     {
 228         case INTRMASK:
 229             ret = ivshmem_IntrMask_read(s);
 230             break;
 231
 232         case INTRSTATUS:
 233             ret = ivshmem_IntrStatus_read(s);
 234             break;
 235
 236         case IVPOSITION:
 237             /* return my VM ID if the memory is mapped */
 238             if (s->shm_fd > 0) {
 239                 ret = s->vm_id;
 240             } else {
 241                 ret = -1;
 242             }
 243             break;
 244
 245         default:
 246             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 247             ret = 0;
 248     }
 249
 250     return ret;
 251 }
 252
 253 static const MemoryRegionOps ivshmem_mmio_ops = {
 254     .read = ivshmem_io_read,
 255     .write = ivshmem_io_write,
 256     .endianness = DEVICE_NATIVE_ENDIAN,
 257     .impl = {
 258         .min_access_size = 4,
 259         .max_access_size = 4,
 260     },
 261 };
 262
 263 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 264 {
 265     IVShmemState *s = opaque;
 266
 267     ivshmem_IntrStatus_write(s, *buf);
 268
 269     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf);
 270 }
 271
 272 static int ivshmem_can_receive(void * opaque)
 273 {
 274     return sizeof(long);
 275 }
 276
 277 static void ivshmem_event(void *opaque, int event)
 278 {
 279     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 280 }
 281
 282 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 283
 284     EventfdEntry *entry = opaque;
 285     PCIDevice *pdev = entry->pdev;
 286
 287     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector);
 288     msix_notify(pdev, entry->vector);
 289 }
 290
 291 static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n,
 292                                                   int vector)
 293 {
 294     /* create a event character device based on the passed eventfd */
 295     IVShmemState *s = opaque;
 296     CharDriverState * chr;
 297     int eventfd = event_notifier_get_fd(n);
 298
 299     chr = qemu_chr_open_eventfd(eventfd);
 300
 301     if (chr == NULL) {
 302         error_report("creating eventfd for eventfd %d failed", eventfd);
 303         exit(1);
 304     }
 305     qemu_chr_fe_claim_no_fail(chr);
 306
 307     /* if MSI is supported we need multiple interrupts */
 308     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 309         s->eventfd_table[vector].pdev = PCI_DEVICE(s);
 310         s->eventfd_table[vector].vector = vector;
 311
 312         qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd,
 313                       ivshmem_event, &s->eventfd_table[vector]);
 314     } else {
 315         qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
 316                       ivshmem_event, s);
 317     }
 318
 319     return chr;
 320
 321 }
 322
 323 static int check_shm_size(IVShmemState *s, int fd) {
 324     /* check that the guest isn't going to try and map more memory than the
 325      * the object has allocated return -1 to indicate error */
 326
 327     struct stat buf;
 328
 329     if (fstat(fd, &buf) < 0) {
 330         error_report("exiting: fstat on fd %d failed: %s",
 331                      fd, strerror(errno));
 332         return -1;
 333     }
 334
 335     if (s->ivshmem_size > buf.st_size) {
 336         error_report("Requested memory size greater"
 337                      " than shared object size (%" PRIu64 " > %" PRIu64")",
 338                      s->ivshmem_size, (uint64_t)buf.st_size);
 339         return -1;
 340     } else {
 341         return 0;
 342     }
 343 }
 344
 345 /* create the shared memory BAR when we are not using the server, so we can
 346  * create the BAR and map the memory immediately */
 347 static void create_shared_memory_BAR(IVShmemState *s, int fd, uint8_t attr) {
 348
 349     void * ptr;
 350
 351     s->shm_fd = fd;
 352
 353     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 354
 355     memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s), "ivshmem.bar2",
 356                                s->ivshmem_size, ptr);
 357     vmstate_register_ram(&s->ivshmem, DEVICE(s));
 358     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 359
 360     /* region for shared memory */
 361     pci_register_bar(PCI_DEVICE(s), 2, attr, &s->bar);
 362 }
 363
 364 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 365 {
 366     memory_region_add_eventfd(&s->ivshmem_mmio,
 367                               DOORBELL,
 368                               4,
 369                               true,
 370                               (posn << 16) | i,
 371                               &s->peers[posn].eventfds[i]);
 372 }
 373
 374 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 375 {
 376     memory_region_del_eventfd(&s->ivshmem_mmio,
 377                               DOORBELL,
 378                               4,
 379                               true,
 380                               (posn << 16) | i,
 381                               &s->peers[posn].eventfds[i]);
 382 }
 383
 384 static void close_guest_eventfds(IVShmemState *s, int posn)
 385 {
 386     int i, guest_curr_max;
 387
 388     if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 389         return;
 390     }
 391     if (posn < 0 || posn >= s->nb_peers) {
 392         return;
 393     }
 394
 395     guest_curr_max = s->peers[posn].nb_eventfds;
 396
 397     memory_region_transaction_begin();
 398     for (i = 0; i < guest_curr_max; i++) {
 399         ivshmem_del_eventfd(s, posn, i);
 400     }
 401     memory_region_transaction_commit();
 402     for (i = 0; i < guest_curr_max; i++) {
 403         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 404     }
 405
 406     g_free(s->peers[posn].eventfds);
 407     s->peers[posn].nb_eventfds = 0;
 408 }
 409
 410 /* this function increase the dynamic storage need to store data about other
 411  * guests */
 412 static int increase_dynamic_storage(IVShmemState *s, int new_min_size)
 413 {
 414
 415     int j, old_nb_alloc;
 416
 417     /* check for integer overflow */
 418     if (new_min_size >= INT_MAX / sizeof(Peer) - 1 || new_min_size <= 0) {
 419         return -1;
 420     }
 421
 422     old_nb_alloc = s->nb_peers;
 423
 424     if (new_min_size >= s->nb_peers) {
 425         /* +1 because #new_min_size is used as last array index */
 426         s->nb_peers = new_min_size + 1;
 427     } else {
 428         return 0;
 429     }
 430
 431     IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers);
 432     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 433
 434     /* zero out new pointers */
 435     for (j = old_nb_alloc; j < s->nb_peers; j++) {
 436         s->peers[j].eventfds = NULL;
 437         s->peers[j].nb_eventfds = 0;
 438     }
 439
 440     return 0;
 441 }
 442
 443 static bool fifo_update_and_get(IVShmemState *s, const uint8_t *buf, int size,
 444                                 void *data, size_t len)
 445 {
 446     const uint8_t *p;
 447     uint32_t num;
 448
 449     assert(len <= sizeof(long)); /* limitation of the fifo */
 450     if (fifo8_is_empty(&s->incoming_fifo) && size == len) {
 451         memcpy(data, buf, size);
 452         return true;
 453     }
 454
 455     IVSHMEM_DPRINTF("short read of %d bytes\n", size);
 456
 457     num = MIN(size, sizeof(long) - fifo8_num_used(&s->incoming_fifo));
 458     fifo8_push_all(&s->incoming_fifo, buf, num);
 459
 460     if (fifo8_num_used(&s->incoming_fifo) < len) {
 461         assert(num == 0);
 462         return false;
 463     }
 464
 465     size -= num;
 466     buf += num;
 467     p = fifo8_pop_buf(&s->incoming_fifo, len, &num);
 468     assert(num == len);
 469
 470     memcpy(data, p, len);
 471
 472     if (size > 0) {
 473         fifo8_push_all(&s->incoming_fifo, buf, size);
 474     }
 475
 476     return true;
 477 }
 478
 479 static void ivshmem_read(void *opaque, const uint8_t *buf, int size)
 480 {
 481     IVShmemState *s = opaque;
 482     int incoming_fd;
 483     int guest_max_eventfd;
 484     long incoming_posn;
 485
 486     if (!fifo_update_and_get(s, buf, size,
 487                              &incoming_posn, sizeof(incoming_posn))) {
 488         return;
 489     }
 490
 491     if (incoming_posn < -1) {
 492         IVSHMEM_DPRINTF("invalid incoming_posn %ld\n", incoming_posn);
 493         return;
 494     }
 495
 496     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 497     incoming_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 498     IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, incoming_fd);
 499
 500     /* make sure we have enough space for this guest */
 501     if (incoming_posn >= s->nb_peers) {
 502         if (increase_dynamic_storage(s, incoming_posn) < 0) {
 503             error_report("increase_dynamic_storage() failed");
 504             if (incoming_fd != -1) {
 505                 close(incoming_fd);
 506             }
 507             return;
 508         }
 509     }
 510
 511     if (incoming_fd == -1) {
 512         /* if posn is positive and unseen before then this is our posn*/
 513         if ((incoming_posn >= 0) &&
 514                             (s->peers[incoming_posn].eventfds == NULL)) {
 515             /* receive our posn */
 516             s->vm_id = incoming_posn;
 517             return;
 518         } else {
 519             /* otherwise an fd == -1 means an existing guest has gone away */
 520             IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
 521             close_guest_eventfds(s, incoming_posn);
 522             return;
 523         }
 524     }
 525
 526     /* if the position is -1, then it's shared memory region fd */
 527     if (incoming_posn == -1) {
 528
 529         void * map_ptr;
 530
 531         s->max_peer = 0;
 532
 533         if (check_shm_size(s, incoming_fd) == -1) {
 534             exit(1);
 535         }
 536
 537         /* mmap the region and map into the BAR2 */
 538         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 539                                                             incoming_fd, 0);
 540         memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s),
 541                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 542         vmstate_register_ram(&s->ivshmem, DEVICE(s));
 543
 544         IVSHMEM_DPRINTF("guest h/w addr = %p, size = %" PRIu64 "\n",
 545                         map_ptr, s->ivshmem_size);
 546
 547         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 548
 549         /* only store the fd if it is successfully mapped */
 550         s->shm_fd = incoming_fd;
 551
 552         return;
 553     }
 554
 555     /* each guest has an array of eventfds, and we keep track of how many
 556      * guests for each VM */
 557     guest_max_eventfd = s->peers[incoming_posn].nb_eventfds;
 558
 559     if (guest_max_eventfd == 0) {
 560         /* one eventfd per MSI vector */
 561         s->peers[incoming_posn].eventfds = g_new(EventNotifier, s->vectors);
 562     }
 563
 564     /* this is an eventfd for a particular guest VM */
 565     IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
 566                     guest_max_eventfd, incoming_fd);
 567     event_notifier_init_fd(&s->peers[incoming_posn].eventfds[guest_max_eventfd],
 568                            incoming_fd);
 569
 570     /* increment count for particular guest */
 571     s->peers[incoming_posn].nb_eventfds++;
 572
 573     /* keep track of the maximum VM ID */
 574     if (incoming_posn > s->max_peer) {
 575         s->max_peer = incoming_posn;
 576     }
 577
 578     if (incoming_posn == s->vm_id) {
 579         s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s,
 580                    &s->peers[s->vm_id].eventfds[guest_max_eventfd],
 581                    guest_max_eventfd);
 582     }
 583
 584     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 585         ivshmem_add_eventfd(s, incoming_posn, guest_max_eventfd);
 586     }
 587 }
 588
 589 /* Select the MSI-X vectors used by device.
 590  * ivshmem maps events to vectors statically, so
 591  * we just enable all vectors on init and after reset. */
 592 static void ivshmem_use_msix(IVShmemState * s)
 593 {
 594     PCIDevice *d = PCI_DEVICE(s);
 595     int i;
 596
 597     if (!msix_present(d)) {
 598         return;
 599     }
 600
 601     for (i = 0; i < s->vectors; i++) {
 602         msix_vector_use(d, i);
 603     }
 604 }
 605
 606 static void ivshmem_reset(DeviceState *d)
 607 {
 608     IVShmemState *s = IVSHMEM(d);
 609
 610     s->intrstatus = 0;
 611     ivshmem_use_msix(s);
 612 }
 613
 614 static uint64_t ivshmem_get_size(IVShmemState * s) {
 615
 616     uint64_t value;
 617     char *ptr;
 618
 619     value = strtoull(s->sizearg, &ptr, 10);
 620     switch (*ptr) {
 621         case 0: case 'M': case 'm':
 622             value <<= 20;
 623             break;
 624         case 'G': case 'g':
 625             value <<= 30;
 626             break;
 627         default:
 628             error_report("invalid ram size: %s", s->sizearg);
 629             exit(1);
 630     }
 631
 632     /* BARs must be a power of 2 */
 633     if (!is_power_of_two(value)) {
 634         error_report("size must be power of 2");
 635         exit(1);
 636     }
 637
 638     return value;
 639 }
 640
 641 static void ivshmem_setup_msi(IVShmemState * s)
 642 {
 643     if (msix_init_exclusive_bar(PCI_DEVICE(s), s->vectors, 1)) {
 644         IVSHMEM_DPRINTF("msix initialization failed\n");
 645         exit(1);
 646     }
 647
 648     IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 649
 650     /* allocate QEMU char devices for receiving interrupts */
 651     s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
 652
 653     ivshmem_use_msix(s);
 654 }
 655
 656 static void ivshmem_save(QEMUFile* f, void *opaque)
 657 {
 658     IVShmemState *proxy = opaque;
 659     PCIDevice *pci_dev = PCI_DEVICE(proxy);
 660
 661     IVSHMEM_DPRINTF("ivshmem_save\n");
 662     pci_device_save(pci_dev, f);
 663
 664     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 665         msix_save(pci_dev, f);
 666     } else {
 667         qemu_put_be32(f, proxy->intrstatus);
 668         qemu_put_be32(f, proxy->intrmask);
 669     }
 670
 671 }
 672
 673 static int ivshmem_load(QEMUFile* f, void *opaque, int version_id)
 674 {
 675     IVSHMEM_DPRINTF("ivshmem_load\n");
 676
 677     IVShmemState *proxy = opaque;
 678     PCIDevice *pci_dev = PCI_DEVICE(proxy);
 679     int ret;
 680
 681     if (version_id > 0) {
 682         return -EINVAL;
 683     }
 684
 685     if (proxy->role_val == IVSHMEM_PEER) {
 686         error_report("'peer' devices are not migratable");
 687         return -EINVAL;
 688     }
 689
 690     ret = pci_device_load(pci_dev, f);
 691     if (ret) {
 692         return ret;
 693     }
 694
 695     if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
 696         msix_load(pci_dev, f);
 697         ivshmem_use_msix(proxy);
 698     } else {
 699         proxy->intrstatus = qemu_get_be32(f);
 700         proxy->intrmask = qemu_get_be32(f);
 701     }
 702
 703     return 0;
 704 }
 705
 706 static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address,
 707                                  uint32_t val, int len)
 708 {
 709     pci_default_write_config(pci_dev, address, val, len);
 710 }
 711
 712 static int pci_ivshmem_init(PCIDevice *dev)
 713 {
 714     IVShmemState *s = IVSHMEM(dev);
 715     uint8_t *pci_conf;
 716     uint8_t attr = PCI_BASE_ADDRESS_SPACE_MEMORY |
 717         PCI_BASE_ADDRESS_MEM_PREFETCH;
 718
 719     if (s->sizearg == NULL)
 720         s->ivshmem_size = 4 << 20; /* 4 MB default */
 721     else {
 722         s->ivshmem_size = ivshmem_get_size(s);
 723     }
 724
 725     fifo8_create(&s->incoming_fifo, sizeof(long));
 726
 727     register_savevm(DEVICE(dev), "ivshmem", 0, 0, ivshmem_save, ivshmem_load,
 728                                                                         dev);
 729
 730     /* IRQFD requires MSI */
 731     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 732         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 733         error_report("ioeventfd/irqfd requires MSI");
 734         exit(1);
 735     }
 736
 737     /* check that role is reasonable */
 738     if (s->role) {
 739         if (strncmp(s->role, "peer", 5) == 0) {
 740             s->role_val = IVSHMEM_PEER;
 741         } else if (strncmp(s->role, "master", 7) == 0) {
 742             s->role_val = IVSHMEM_MASTER;
 743         } else {
 744             error_report("'role' must be 'peer' or 'master'");
 745             exit(1);
 746         }
 747     } else {
 748         s->role_val = IVSHMEM_MASTER; /* default */
 749     }
 750
 751     if (s->role_val == IVSHMEM_PEER) {
 752         error_setg(&s->migration_blocker,
 753                    "Migration is disabled when using feature 'peer mode' in device 'ivshmem'");
 754         migrate_add_blocker(s->migration_blocker);
 755     }
 756
 757     pci_conf = dev->config;
 758     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 759
 760     pci_config_set_interrupt_pin(pci_conf, 1);
 761
 762     s->shm_fd = 0;
 763
 764     memory_region_init_io(&s->ivshmem_mmio, OBJECT(s), &ivshmem_mmio_ops, s,
 765                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 766
 767     /* region for registers*/
 768     pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 769                      &s->ivshmem_mmio);
 770
 771     memory_region_init(&s->bar, OBJECT(s), "ivshmem-bar2-container", s->ivshmem_size);
 772     if (s->ivshmem_64bit) {
 773         attr |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 774     }
 775
 776     if ((s->server_chr != NULL) &&
 777                         (strncmp(s->server_chr->filename, "unix:", 5) == 0)) {
 778         /* if we get a UNIX socket as the parameter we will talk
 779          * to the ivshmem server to receive the memory region */
 780
 781         if (s->shmobj != NULL) {
 782             error_report("WARNING: do not specify both 'chardev' "
 783                          "and 'shm' with ivshmem");
 784         }
 785
 786         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 787                         s->server_chr->filename);
 788
 789         if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 790             ivshmem_setup_msi(s);
 791         }
 792
 793         /* we allocate enough space for 16 guests and grow as needed */
 794         s->nb_peers = 16;
 795         s->vm_id = -1;
 796
 797         /* allocate/initialize space for interrupt handling */
 798         s->peers = g_malloc0(s->nb_peers * sizeof(Peer));
 799
 800         pci_register_bar(dev, 2, attr, &s->bar);
 801
 802         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 803
 804         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 805                      ivshmem_event, s);
 806     } else {
 807         /* just map the file immediately, we're not using a server */
 808         int fd;
 809
 810         if (s->shmobj == NULL) {
 811             error_report("Must specify 'chardev' or 'shm' to ivshmem");
 812             exit(1);
 813         }
 814
 815         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 816
 817         /* try opening with O_EXCL and if it succeeds zero the memory
 818          * by truncating to 0 */
 819         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 820                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 821            /* truncate file to length PCI device's memory */
 822             if (ftruncate(fd, s->ivshmem_size) != 0) {
 823                 error_report("could not truncate shared file");
 824             }
 825
 826         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 827                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 828             error_report("could not open shared file");
 829             exit(1);
 830
 831         }
 832
 833         if (check_shm_size(s, fd) == -1) {
 834             exit(1);
 835         }
 836
 837         create_shared_memory_BAR(s, fd, attr);
 838     }
 839
 840     dev->config_write = ivshmem_write_config;
 841
 842     return 0;
 843 }
 844
 845 static void pci_ivshmem_uninit(PCIDevice *dev)
 846 {
 847     IVShmemState *s = IVSHMEM(dev);
 848
 849     if (s->migration_blocker) {
 850         migrate_del_blocker(s->migration_blocker);
 851         error_free(s->migration_blocker);
 852     }
 853
 854     memory_region_del_subregion(&s->bar, &s->ivshmem);
 855     vmstate_unregister_ram(&s->ivshmem, DEVICE(dev));
 856     unregister_savevm(DEVICE(dev), "ivshmem", s);
 857     fifo8_destroy(&s->incoming_fifo);
 858 }
 859
 860 static Property ivshmem_properties[] = {
 861     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
 862     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
 863     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
 864     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
 865     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
 866     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
 867     DEFINE_PROP_STRING("role", IVShmemState, role),
 868     DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1),
 869     DEFINE_PROP_END_OF_LIST(),
 870 };
 871
 872 static void ivshmem_class_init(ObjectClass *klass, void *data)
 873 {
 874     DeviceClass *dc = DEVICE_CLASS(klass);
 875     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 876
 877     k->init = pci_ivshmem_init;
 878     k->exit = pci_ivshmem_uninit;
 879     k->vendor_id = PCI_VENDOR_ID_IVSHMEM;
 880     k->device_id = PCI_DEVICE_ID_IVSHMEM;
 881     k->class_id = PCI_CLASS_MEMORY_RAM;
 882     dc->reset = ivshmem_reset;
 883     dc->props = ivshmem_properties;
 884     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 885 }
 886
 887 static const TypeInfo ivshmem_info = {
 888     .name          = TYPE_IVSHMEM,
 889     .parent        = TYPE_PCI_DEVICE,
 890     .instance_size = sizeof(IVShmemState),
 891     .class_init    = ivshmem_class_init,
 892 };
 893
 894 static void ivshmem_register_types(void)
 895 {
 896     type_register_static(&ivshmem_info);
 897 }
 898
 899 type_init(ivshmem_register_types)