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 #include "sysemu/hostmem.h"
  30 #include "qapi/visitor.h"
  31
  32 #include "hw/misc/ivshmem.h"
  33
  34 #include <sys/mman.h>
  35 #include <sys/types.h>
  36 #include <limits.h>
  37
  38 #define PCI_VENDOR_ID_IVSHMEM   PCI_VENDOR_ID_REDHAT_QUMRANET
  39 #define PCI_DEVICE_ID_IVSHMEM   0x1110
  40
  41 #define IVSHMEM_MAX_PEERS G_MAXUINT16
  42 #define IVSHMEM_IOEVENTFD   0
  43 #define IVSHMEM_MSI     1
  44
  45 #define IVSHMEM_PEER    0
  46 #define IVSHMEM_MASTER  1
  47
  48 #define IVSHMEM_REG_BAR_SIZE 0x100
  49
  50 //#define DEBUG_IVSHMEM
  51 #ifdef DEBUG_IVSHMEM
  52 #define IVSHMEM_DPRINTF(fmt, ...)        \
  53     do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0)
  54 #else
  55 #define IVSHMEM_DPRINTF(fmt, ...)
  56 #endif
  57
  58 #define TYPE_IVSHMEM "ivshmem"
  59 #define IVSHMEM(obj) \
  60     OBJECT_CHECK(IVShmemState, (obj), TYPE_IVSHMEM)
  61
  62 #define IVSHMEM_MEMDEV_PROP "memdev"
  63
  64 typedef struct Peer {
  65     int nb_eventfds;
  66     EventNotifier *eventfds;
  67 } Peer;
  68
  69 typedef struct EventfdEntry {
  70     PCIDevice *pdev;
  71     int vector;
  72 } EventfdEntry;
  73
  74 typedef struct IVShmemState {
  75     /*< private >*/
  76     PCIDevice parent_obj;
  77     /*< public >*/
  78
  79     HostMemoryBackend *hostmem;
  80     uint32_t intrmask;
  81     uint32_t intrstatus;
  82
  83     CharDriverState **eventfd_chr;
  84     CharDriverState *server_chr;
  85     Fifo8 incoming_fifo;
  86     MemoryRegion ivshmem_mmio;
  87
  88     /* We might need to register the BAR before we actually have the memory.
  89      * So prepare a container MemoryRegion for the BAR immediately and
  90      * add a subregion when we have the memory.
  91      */
  92     MemoryRegion bar;
  93     MemoryRegion ivshmem;
  94     uint64_t ivshmem_size; /* size of shared memory region */
  95     uint32_t ivshmem_64bit;
  96
  97     Peer *peers;
  98     int nb_peers; /* how many peers we have space for */
  99
 100     int vm_id;
 101     uint32_t vectors;
 102     uint32_t features;
 103     EventfdEntry *eventfd_table;
 104
 105     Error *migration_blocker;
 106
 107     char * shmobj;
 108     char * sizearg;
 109     char * role;
 110     int role_val;   /* scalar to avoid multiple string comparisons */
 111 } IVShmemState;
 112
 113 /* registers for the Inter-VM shared memory device */
 114 enum ivshmem_registers {
 115     INTRMASK = 0,
 116     INTRSTATUS = 4,
 117     IVPOSITION = 8,
 118     DOORBELL = 12,
 119 };
 120
 121 static inline uint32_t ivshmem_has_feature(IVShmemState *ivs,
 122                                                     unsigned int feature) {
 123     return (ivs->features & (1 << feature));
 124 }
 125
 126 /* accessing registers - based on rtl8139 */
 127 static void ivshmem_update_irq(IVShmemState *s)
 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);
 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);
 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);
 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->nb_peers) {
 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             } else {
 214                 IVSHMEM_DPRINTF("Invalid destination vector %d on VM %d\n",
 215                                 vector, dest);
 216             }
 217             break;
 218         default:
 219             IVSHMEM_DPRINTF("Unhandled write " TARGET_FMT_plx "\n", addr);
 220     }
 221 }
 222
 223 static uint64_t ivshmem_io_read(void *opaque, hwaddr addr,
 224                                 unsigned size)
 225 {
 226
 227     IVShmemState *s = opaque;
 228     uint32_t ret;
 229
 230     switch (addr)
 231     {
 232         case INTRMASK:
 233             ret = ivshmem_IntrMask_read(s);
 234             break;
 235
 236         case INTRSTATUS:
 237             ret = ivshmem_IntrStatus_read(s);
 238             break;
 239
 240         case IVPOSITION:
 241             /* return my VM ID if the memory is mapped */
 242             if (memory_region_is_mapped(&s->ivshmem)) {
 243                 ret = s->vm_id;
 244             } else {
 245                 ret = -1;
 246             }
 247             break;
 248
 249         default:
 250             IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr);
 251             ret = 0;
 252     }
 253
 254     return ret;
 255 }
 256
 257 static const MemoryRegionOps ivshmem_mmio_ops = {
 258     .read = ivshmem_io_read,
 259     .write = ivshmem_io_write,
 260     .endianness = DEVICE_NATIVE_ENDIAN,
 261     .impl = {
 262         .min_access_size = 4,
 263         .max_access_size = 4,
 264     },
 265 };
 266
 267 static void ivshmem_receive(void *opaque, const uint8_t *buf, int size)
 268 {
 269     IVShmemState *s = opaque;
 270
 271     IVSHMEM_DPRINTF("ivshmem_receive 0x%02x size: %d\n", *buf, size);
 272
 273     ivshmem_IntrStatus_write(s, *buf);
 274 }
 275
 276 static int ivshmem_can_receive(void * opaque)
 277 {
 278     return sizeof(long);
 279 }
 280
 281 static void ivshmem_event(void *opaque, int event)
 282 {
 283     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 284 }
 285
 286 static void fake_irqfd(void *opaque, const uint8_t *buf, int size) {
 287
 288     EventfdEntry *entry = opaque;
 289     PCIDevice *pdev = entry->pdev;
 290
 291     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector);
 292     msix_notify(pdev, entry->vector);
 293 }
 294
 295 static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n,
 296                                                   int vector)
 297 {
 298     /* create a event character device based on the passed eventfd */
 299     IVShmemState *s = opaque;
 300     CharDriverState * chr;
 301     int eventfd = event_notifier_get_fd(n);
 302
 303     chr = qemu_chr_open_eventfd(eventfd);
 304
 305     if (chr == NULL) {
 306         error_report("creating chardriver for eventfd %d failed", eventfd);
 307         return NULL;
 308     }
 309     qemu_chr_fe_claim_no_fail(chr);
 310
 311     /* if MSI is supported we need multiple interrupts */
 312     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 313         s->eventfd_table[vector].pdev = PCI_DEVICE(s);
 314         s->eventfd_table[vector].vector = vector;
 315
 316         qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd,
 317                       ivshmem_event, &s->eventfd_table[vector]);
 318     } else {
 319         qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive,
 320                       ivshmem_event, s);
 321     }
 322
 323     return chr;
 324
 325 }
 326
 327 static int check_shm_size(IVShmemState *s, int fd, Error **errp)
 328 {
 329     /* check that the guest isn't going to try and map more memory than the
 330      * the object has allocated return -1 to indicate error */
 331
 332     struct stat buf;
 333
 334     if (fstat(fd, &buf) < 0) {
 335         error_setg(errp, "exiting: fstat on fd %d failed: %s",
 336                    fd, strerror(errno));
 337         return -1;
 338     }
 339
 340     if (s->ivshmem_size > buf.st_size) {
 341         error_setg(errp, "Requested memory size greater"
 342                    " than shared object size (%" PRIu64 " > %" PRIu64")",
 343                    s->ivshmem_size, (uint64_t)buf.st_size);
 344         return -1;
 345     } else {
 346         return 0;
 347     }
 348 }
 349
 350 /* create the shared memory BAR when we are not using the server, so we can
 351  * create the BAR and map the memory immediately */
 352 static int create_shared_memory_BAR(IVShmemState *s, int fd, uint8_t attr,
 353                                     Error **errp)
 354 {
 355     void * ptr;
 356
 357     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 358     if (ptr == MAP_FAILED) {
 359         error_setg_errno(errp, errno, "Failed to mmap shared memory");
 360         return -1;
 361     }
 362
 363     memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s), "ivshmem.bar2",
 364                                s->ivshmem_size, ptr);
 365     vmstate_register_ram(&s->ivshmem, DEVICE(s));
 366     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 367
 368     /* region for shared memory */
 369     pci_register_bar(PCI_DEVICE(s), 2, attr, &s->bar);
 370
 371     return 0;
 372 }
 373
 374 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 375 {
 376     memory_region_add_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 ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 385 {
 386     memory_region_del_eventfd(&s->ivshmem_mmio,
 387                               DOORBELL,
 388                               4,
 389                               true,
 390                               (posn << 16) | i,
 391                               &s->peers[posn].eventfds[i]);
 392 }
 393
 394 static void close_peer_eventfds(IVShmemState *s, int posn)
 395 {
 396     int i, n;
 397
 398     if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 399         return;
 400     }
 401     if (posn < 0 || posn >= s->nb_peers) {
 402         error_report("invalid peer %d", posn);
 403         return;
 404     }
 405
 406     n = s->peers[posn].nb_eventfds;
 407
 408     memory_region_transaction_begin();
 409     for (i = 0; i < n; i++) {
 410         ivshmem_del_eventfd(s, posn, i);
 411     }
 412     memory_region_transaction_commit();
 413     for (i = 0; i < n; i++) {
 414         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 415     }
 416
 417     g_free(s->peers[posn].eventfds);
 418     s->peers[posn].nb_eventfds = 0;
 419 }
 420
 421 /* this function increase the dynamic storage need to store data about other
 422  * peers */
 423 static int resize_peers(IVShmemState *s, int new_min_size)
 424 {
 425
 426     int j, old_size;
 427
 428     /* limit number of max peers */
 429     if (new_min_size <= 0 || new_min_size > IVSHMEM_MAX_PEERS) {
 430         return -1;
 431     }
 432     if (new_min_size <= s->nb_peers) {
 433         return 0;
 434     }
 435
 436     old_size = s->nb_peers;
 437     s->nb_peers = new_min_size;
 438
 439     IVSHMEM_DPRINTF("bumping storage to %d peers\n", s->nb_peers);
 440
 441     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 442
 443     for (j = old_size; j < s->nb_peers; j++) {
 444         s->peers[j].eventfds = g_new0(EventNotifier, s->vectors);
 445         s->peers[j].nb_eventfds = 0;
 446     }
 447
 448     return 0;
 449 }
 450
 451 static bool fifo_update_and_get(IVShmemState *s, const uint8_t *buf, int size,
 452                                 void *data, size_t len)
 453 {
 454     const uint8_t *p;
 455     uint32_t num;
 456
 457     assert(len <= sizeof(long)); /* limitation of the fifo */
 458     if (fifo8_is_empty(&s->incoming_fifo) && size == len) {
 459         memcpy(data, buf, size);
 460         return true;
 461     }
 462
 463     IVSHMEM_DPRINTF("short read of %d bytes\n", size);
 464
 465     num = MIN(size, sizeof(long) - fifo8_num_used(&s->incoming_fifo));
 466     fifo8_push_all(&s->incoming_fifo, buf, num);
 467
 468     if (fifo8_num_used(&s->incoming_fifo) < len) {
 469         assert(num == 0);
 470         return false;
 471     }
 472
 473     size -= num;
 474     buf += num;
 475     p = fifo8_pop_buf(&s->incoming_fifo, len, &num);
 476     assert(num == len);
 477
 478     memcpy(data, p, len);
 479
 480     if (size > 0) {
 481         fifo8_push_all(&s->incoming_fifo, buf, size);
 482     }
 483
 484     return true;
 485 }
 486
 487 static void ivshmem_read(void *opaque, const uint8_t *buf, int size)
 488 {
 489     IVShmemState *s = opaque;
 490     int incoming_fd;
 491     int new_eventfd;
 492     long incoming_posn;
 493     Error *err = NULL;
 494     Peer *peer;
 495
 496     if (!fifo_update_and_get(s, buf, size,
 497                              &incoming_posn, sizeof(incoming_posn))) {
 498         return;
 499     }
 500
 501     if (incoming_posn < -1) {
 502         IVSHMEM_DPRINTF("invalid incoming_posn %ld\n", incoming_posn);
 503         return;
 504     }
 505
 506     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 507     incoming_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 508     IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, incoming_fd);
 509
 510     /* make sure we have enough space for this peer */
 511     if (incoming_posn >= s->nb_peers) {
 512         if (resize_peers(s, incoming_posn + 1) < 0) {
 513             error_report("failed to resize peers array");
 514             if (incoming_fd != -1) {
 515                 close(incoming_fd);
 516             }
 517             return;
 518         }
 519     }
 520
 521     peer = &s->peers[incoming_posn];
 522
 523     if (incoming_fd == -1) {
 524         /* if posn is positive and unseen before then this is our posn*/
 525         if (incoming_posn >= 0 && s->vm_id == -1) {
 526             /* receive our posn */
 527             s->vm_id = incoming_posn;
 528         } else {
 529             /* otherwise an fd == -1 means an existing peer has gone away */
 530             IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
 531             close_peer_eventfds(s, incoming_posn);
 532         }
 533         return;
 534     }
 535
 536     /* if the position is -1, then it's shared memory region fd */
 537     if (incoming_posn == -1) {
 538         void * map_ptr;
 539
 540         if (memory_region_is_mapped(&s->ivshmem)) {
 541             error_report("shm already initialized");
 542             close(incoming_fd);
 543             return;
 544         }
 545
 546         if (check_shm_size(s, incoming_fd, &err) == -1) {
 547             error_report_err(err);
 548             close(incoming_fd);
 549             return;
 550         }
 551
 552         /* mmap the region and map into the BAR2 */
 553         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 554                                                             incoming_fd, 0);
 555         if (map_ptr == MAP_FAILED) {
 556             error_report("Failed to mmap shared memory %s", strerror(errno));
 557             close(incoming_fd);
 558             return;
 559         }
 560         memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s),
 561                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 562         vmstate_register_ram(&s->ivshmem, DEVICE(s));
 563
 564         IVSHMEM_DPRINTF("guest h/w addr = %p, size = %" PRIu64 "\n",
 565                         map_ptr, s->ivshmem_size);
 566
 567         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 568
 569         close(incoming_fd);
 570         return;
 571     }
 572
 573     /* each peer has an associated array of eventfds, and we keep
 574      * track of how many eventfds received so far */
 575     /* get a new eventfd: */
 576     if (peer->nb_eventfds >= s->vectors) {
 577         error_report("Too many eventfd received, device has %d vectors",
 578                      s->vectors);
 579         close(incoming_fd);
 580         return;
 581     }
 582
 583     new_eventfd = peer->nb_eventfds++;
 584
 585     /* this is an eventfd for a particular peer VM */
 586     IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
 587                     new_eventfd, incoming_fd);
 588     event_notifier_init_fd(&peer->eventfds[new_eventfd], incoming_fd);
 589
 590     if (incoming_posn == s->vm_id) {
 591         s->eventfd_chr[new_eventfd] = create_eventfd_chr_device(s,
 592                    &s->peers[s->vm_id].eventfds[new_eventfd],
 593                    new_eventfd);
 594     }
 595
 596     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 597         ivshmem_add_eventfd(s, incoming_posn, new_eventfd);
 598     }
 599 }
 600
 601 static void ivshmem_check_version(void *opaque, const uint8_t * buf, int size)
 602 {
 603     IVShmemState *s = opaque;
 604     int tmp;
 605     long version;
 606
 607     if (!fifo_update_and_get(s, buf, size,
 608                              &version, sizeof(version))) {
 609         return;
 610     }
 611
 612     tmp = qemu_chr_fe_get_msgfd(s->server_chr);
 613     if (tmp != -1 || version != IVSHMEM_PROTOCOL_VERSION) {
 614         fprintf(stderr, "incompatible version, you are connecting to a ivshmem-"
 615                 "server using a different protocol please check your setup\n");
 616         qemu_chr_delete(s->server_chr);
 617         s->server_chr = NULL;
 618         return;
 619     }
 620
 621     IVSHMEM_DPRINTF("version check ok, switch to real chardev handler\n");
 622     qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 623                           ivshmem_event, s);
 624 }
 625
 626 /* Select the MSI-X vectors used by device.
 627  * ivshmem maps events to vectors statically, so
 628  * we just enable all vectors on init and after reset. */
 629 static void ivshmem_use_msix(IVShmemState * s)
 630 {
 631     PCIDevice *d = PCI_DEVICE(s);
 632     int i;
 633
 634     IVSHMEM_DPRINTF("%s, msix present: %d\n", __func__, msix_present(d));
 635     if (!msix_present(d)) {
 636         return;
 637     }
 638
 639     for (i = 0; i < s->vectors; i++) {
 640         msix_vector_use(d, i);
 641     }
 642 }
 643
 644 static void ivshmem_reset(DeviceState *d)
 645 {
 646     IVShmemState *s = IVSHMEM(d);
 647
 648     s->intrstatus = 0;
 649     s->intrmask = 0;
 650     ivshmem_use_msix(s);
 651 }
 652
 653 static int ivshmem_setup_msi(IVShmemState * s)
 654 {
 655     if (msix_init_exclusive_bar(PCI_DEVICE(s), s->vectors, 1)) {
 656         return -1;
 657     }
 658
 659     IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 660
 661     /* allocate QEMU char devices for receiving interrupts */
 662     s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
 663
 664     ivshmem_use_msix(s);
 665     return 0;
 666 }
 667
 668 static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address,
 669                                  uint32_t val, int len)
 670 {
 671     pci_default_write_config(pci_dev, address, val, len);
 672 }
 673
 674 static void pci_ivshmem_realize(PCIDevice *dev, Error **errp)
 675 {
 676     IVShmemState *s = IVSHMEM(dev);
 677     uint8_t *pci_conf;
 678     uint8_t attr = PCI_BASE_ADDRESS_SPACE_MEMORY |
 679         PCI_BASE_ADDRESS_MEM_PREFETCH;
 680
 681     if (!!s->server_chr + !!s->shmobj + !!s->hostmem != 1) {
 682         error_setg(errp, "You must specify either a shmobj, a chardev"
 683                    " or a hostmem");
 684         return;
 685     }
 686
 687     if (s->hostmem) {
 688         MemoryRegion *mr;
 689
 690         if (s->sizearg) {
 691             g_warning("size argument ignored with hostmem");
 692         }
 693
 694         mr = host_memory_backend_get_memory(s->hostmem, errp);
 695         s->ivshmem_size = memory_region_size(mr);
 696     } else if (s->sizearg == NULL) {
 697         s->ivshmem_size = 4 << 20; /* 4 MB default */
 698     } else {
 699         char *end;
 700         int64_t size = qemu_strtosz(s->sizearg, &end);
 701         if (size < 0 || *end != '\0' || !is_power_of_2(size)) {
 702             error_setg(errp, "Invalid size %s", s->sizearg);
 703             return;
 704         }
 705         s->ivshmem_size = size;
 706     }
 707
 708     fifo8_create(&s->incoming_fifo, sizeof(long));
 709
 710     /* IRQFD requires MSI */
 711     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 712         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 713         error_setg(errp, "ioeventfd/irqfd requires MSI");
 714         return;
 715     }
 716
 717     /* check that role is reasonable */
 718     if (s->role) {
 719         if (strncmp(s->role, "peer", 5) == 0) {
 720             s->role_val = IVSHMEM_PEER;
 721         } else if (strncmp(s->role, "master", 7) == 0) {
 722             s->role_val = IVSHMEM_MASTER;
 723         } else {
 724             error_setg(errp, "'role' must be 'peer' or 'master'");
 725             return;
 726         }
 727     } else {
 728         s->role_val = IVSHMEM_MASTER; /* default */
 729     }
 730
 731     if (s->role_val == IVSHMEM_PEER) {
 732         error_setg(&s->migration_blocker,
 733                    "Migration is disabled when using feature 'peer mode' in device 'ivshmem'");
 734         migrate_add_blocker(s->migration_blocker);
 735     }
 736
 737     pci_conf = dev->config;
 738     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 739
 740     pci_config_set_interrupt_pin(pci_conf, 1);
 741
 742     memory_region_init_io(&s->ivshmem_mmio, OBJECT(s), &ivshmem_mmio_ops, s,
 743                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 744
 745     /* region for registers*/
 746     pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 747                      &s->ivshmem_mmio);
 748
 749     memory_region_init(&s->bar, OBJECT(s), "ivshmem-bar2-container", s->ivshmem_size);
 750     if (s->ivshmem_64bit) {
 751         attr |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 752     }
 753
 754     if (s->hostmem != NULL) {
 755         MemoryRegion *mr;
 756
 757         IVSHMEM_DPRINTF("using hostmem\n");
 758
 759         mr = host_memory_backend_get_memory(MEMORY_BACKEND(s->hostmem), errp);
 760         vmstate_register_ram(mr, DEVICE(s));
 761         memory_region_add_subregion(&s->bar, 0, mr);
 762         pci_register_bar(PCI_DEVICE(s), 2, attr, &s->bar);
 763     } else if (s->server_chr != NULL) {
 764         if (strncmp(s->server_chr->filename, "unix:", 5)) {
 765             error_setg(errp, "chardev is not a unix client socket");
 766             return;
 767         }
 768
 769         /* if we get a UNIX socket as the parameter we will talk
 770          * to the ivshmem server to receive the memory region */
 771
 772         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 773                         s->server_chr->filename);
 774
 775         if (ivshmem_has_feature(s, IVSHMEM_MSI) &&
 776             ivshmem_setup_msi(s)) {
 777             error_setg(errp, "msix initialization failed");
 778             return;
 779         }
 780
 781         /* we allocate enough space for 16 peers and grow as needed */
 782         resize_peers(s, 16);
 783         s->vm_id = -1;
 784
 785         pci_register_bar(dev, 2, attr, &s->bar);
 786
 787         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 788
 789         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive,
 790                               ivshmem_check_version, ivshmem_event, s);
 791     } else {
 792         /* just map the file immediately, we're not using a server */
 793         int fd;
 794
 795         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 796
 797         /* try opening with O_EXCL and if it succeeds zero the memory
 798          * by truncating to 0 */
 799         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 800                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 801            /* truncate file to length PCI device's memory */
 802             if (ftruncate(fd, s->ivshmem_size) != 0) {
 803                 error_report("could not truncate shared file");
 804             }
 805
 806         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 807                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 808             error_setg(errp, "could not open shared file");
 809             return;
 810         }
 811
 812         if (check_shm_size(s, fd, errp) == -1) {
 813             return;
 814         }
 815
 816         create_shared_memory_BAR(s, fd, attr, errp);
 817         close(fd);
 818     }
 819 }
 820
 821 static void pci_ivshmem_exit(PCIDevice *dev)
 822 {
 823     IVShmemState *s = IVSHMEM(dev);
 824     int i;
 825
 826     fifo8_destroy(&s->incoming_fifo);
 827
 828     if (s->migration_blocker) {
 829         migrate_del_blocker(s->migration_blocker);
 830         error_free(s->migration_blocker);
 831     }
 832
 833     if (memory_region_is_mapped(&s->ivshmem)) {
 834         if (!s->hostmem) {
 835             void *addr = memory_region_get_ram_ptr(&s->ivshmem);
 836
 837             if (munmap(addr, s->ivshmem_size) == -1) {
 838                 error_report("Failed to munmap shared memory %s",
 839                              strerror(errno));
 840             }
 841         }
 842
 843         vmstate_unregister_ram(&s->ivshmem, DEVICE(dev));
 844         memory_region_del_subregion(&s->bar, &s->ivshmem);
 845     }
 846
 847     if (s->eventfd_chr) {
 848         for (i = 0; i < s->vectors; i++) {
 849             if (s->eventfd_chr[i]) {
 850                 qemu_chr_free(s->eventfd_chr[i]);
 851             }
 852         }
 853         g_free(s->eventfd_chr);
 854     }
 855
 856     if (s->peers) {
 857         for (i = 0; i < s->nb_peers; i++) {
 858             close_peer_eventfds(s, i);
 859         }
 860         g_free(s->peers);
 861     }
 862
 863     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 864         msix_uninit_exclusive_bar(dev);
 865     }
 866
 867     g_free(s->eventfd_table);
 868 }
 869
 870 static bool test_msix(void *opaque, int version_id)
 871 {
 872     IVShmemState *s = opaque;
 873
 874     return ivshmem_has_feature(s, IVSHMEM_MSI);
 875 }
 876
 877 static bool test_no_msix(void *opaque, int version_id)
 878 {
 879     return !test_msix(opaque, version_id);
 880 }
 881
 882 static int ivshmem_pre_load(void *opaque)
 883 {
 884     IVShmemState *s = opaque;
 885
 886     if (s->role_val == IVSHMEM_PEER) {
 887         error_report("'peer' devices are not migratable");
 888         return -EINVAL;
 889     }
 890
 891     return 0;
 892 }
 893
 894 static int ivshmem_post_load(void *opaque, int version_id)
 895 {
 896     IVShmemState *s = opaque;
 897
 898     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 899         ivshmem_use_msix(s);
 900     }
 901
 902     return 0;
 903 }
 904
 905 static int ivshmem_load_old(QEMUFile *f, void *opaque, int version_id)
 906 {
 907     IVShmemState *s = opaque;
 908     PCIDevice *pdev = PCI_DEVICE(s);
 909     int ret;
 910
 911     IVSHMEM_DPRINTF("ivshmem_load_old\n");
 912
 913     if (version_id != 0) {
 914         return -EINVAL;
 915     }
 916
 917     if (s->role_val == IVSHMEM_PEER) {
 918         error_report("'peer' devices are not migratable");
 919         return -EINVAL;
 920     }
 921
 922     ret = pci_device_load(pdev, f);
 923     if (ret) {
 924         return ret;
 925     }
 926
 927     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 928         msix_load(pdev, f);
 929         ivshmem_use_msix(s);
 930     } else {
 931         s->intrstatus = qemu_get_be32(f);
 932         s->intrmask = qemu_get_be32(f);
 933     }
 934
 935     return 0;
 936 }
 937
 938 static const VMStateDescription ivshmem_vmsd = {
 939     .name = "ivshmem",
 940     .version_id = 1,
 941     .minimum_version_id = 1,
 942     .pre_load = ivshmem_pre_load,
 943     .post_load = ivshmem_post_load,
 944     .fields = (VMStateField[]) {
 945         VMSTATE_PCI_DEVICE(parent_obj, IVShmemState),
 946
 947         VMSTATE_MSIX_TEST(parent_obj, IVShmemState, test_msix),
 948         VMSTATE_UINT32_TEST(intrstatus, IVShmemState, test_no_msix),
 949         VMSTATE_UINT32_TEST(intrmask, IVShmemState, test_no_msix),
 950
 951         VMSTATE_END_OF_LIST()
 952     },
 953     .load_state_old = ivshmem_load_old,
 954     .minimum_version_id_old = 0
 955 };
 956
 957 static Property ivshmem_properties[] = {
 958     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
 959     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
 960     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
 961     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
 962     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
 963     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
 964     DEFINE_PROP_STRING("role", IVShmemState, role),
 965     DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1),
 966     DEFINE_PROP_END_OF_LIST(),
 967 };
 968
 969 static void ivshmem_class_init(ObjectClass *klass, void *data)
 970 {
 971     DeviceClass *dc = DEVICE_CLASS(klass);
 972     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 973
 974     k->realize = pci_ivshmem_realize;
 975     k->exit = pci_ivshmem_exit;
 976     k->config_write = ivshmem_write_config;
 977     k->vendor_id = PCI_VENDOR_ID_IVSHMEM;
 978     k->device_id = PCI_DEVICE_ID_IVSHMEM;
 979     k->class_id = PCI_CLASS_MEMORY_RAM;
 980     dc->reset = ivshmem_reset;
 981     dc->props = ivshmem_properties;
 982     dc->vmsd = &ivshmem_vmsd;
 983     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 984     dc->desc = "Inter-VM shared memory";
 985 }
 986
 987 static void ivshmem_check_memdev_is_busy(Object *obj, const char *name,
 988                                          Object *val, Error **errp)
 989 {
 990     MemoryRegion *mr;
 991
 992     mr = host_memory_backend_get_memory(MEMORY_BACKEND(val), errp);
 993     if (memory_region_is_mapped(mr)) {
 994         char *path = object_get_canonical_path_component(val);
 995         error_setg(errp, "can't use already busy memdev: %s", path);
 996         g_free(path);
 997     } else {
 998         qdev_prop_allow_set_link_before_realize(obj, name, val, errp);
 999     }
1000 }
1001
1002 static void ivshmem_init(Object *obj)
1003 {
1004     IVShmemState *s = IVSHMEM(obj);
1005
1006     object_property_add_link(obj, IVSHMEM_MEMDEV_PROP, TYPE_MEMORY_BACKEND,
1007                              (Object **)&s->hostmem,
1008                              ivshmem_check_memdev_is_busy,
1009                              OBJ_PROP_LINK_UNREF_ON_RELEASE,
1010                              &error_abort);
1011 }
1012
1013 static const TypeInfo ivshmem_info = {
1014     .name          = TYPE_IVSHMEM,
1015     .parent        = TYPE_PCI_DEVICE,
1016     .instance_size = sizeof(IVShmemState),
1017     .instance_init = ivshmem_init,
1018     .class_init    = ivshmem_class_init,
1019 };
1020
1021 static void ivshmem_register_types(void)
1022 {
1023     type_register_static(&ivshmem_info);
1024 }
1025
1026 type_init(ivshmem_register_types)