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 "qemu/osdep.h"
  20 #include "qapi/error.h"
  21 #include "hw/hw.h"
  22 #include "hw/i386/pc.h"
  23 #include "hw/pci/pci.h"
  24 #include "hw/pci/msi.h"
  25 #include "hw/pci/msix.h"
  26 #include "sysemu/kvm.h"
  27 #include "migration/migration.h"
  28 #include "qemu/error-report.h"
  29 #include "qemu/event_notifier.h"
  30 #include "qemu/fifo8.h"
  31 #include "sysemu/char.h"
  32 #include "sysemu/hostmem.h"
  33 #include "qapi/visitor.h"
  34 #include "exec/ram_addr.h"
  35
  36 #include "hw/misc/ivshmem.h"
  37
  38 #include <sys/mman.h>
  39
  40 #define PCI_VENDOR_ID_IVSHMEM   PCI_VENDOR_ID_REDHAT_QUMRANET
  41 #define PCI_DEVICE_ID_IVSHMEM   0x1110
  42
  43 #define IVSHMEM_MAX_PEERS G_MAXUINT16
  44 #define IVSHMEM_IOEVENTFD   0
  45 #define IVSHMEM_MSI     1
  46
  47 #define IVSHMEM_PEER    0
  48 #define IVSHMEM_MASTER  1
  49
  50 #define IVSHMEM_REG_BAR_SIZE 0x100
  51
  52 //#define DEBUG_IVSHMEM
  53 #ifdef DEBUG_IVSHMEM
  54 #define IVSHMEM_DPRINTF(fmt, ...)        \
  55     do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0)
  56 #else
  57 #define IVSHMEM_DPRINTF(fmt, ...)
  58 #endif
  59
  60 #define TYPE_IVSHMEM "ivshmem"
  61 #define IVSHMEM(obj) \
  62     OBJECT_CHECK(IVShmemState, (obj), TYPE_IVSHMEM)
  63
  64 typedef struct Peer {
  65     int nb_eventfds;
  66     EventNotifier *eventfds;
  67 } Peer;
  68
  69 typedef struct MSIVector {
  70     PCIDevice *pdev;
  71     int virq;
  72 } MSIVector;
  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     MSIVector *msi_vectors;
 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 int ivshmem_can_receive(void * opaque)
 268 {
 269     return sizeof(int64_t);
 270 }
 271
 272 static void ivshmem_event(void *opaque, int event)
 273 {
 274     IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
 275 }
 276
 277 static void ivshmem_vector_notify(void *opaque)
 278 {
 279     MSIVector *entry = opaque;
 280     PCIDevice *pdev = entry->pdev;
 281     IVShmemState *s = IVSHMEM(pdev);
 282     int vector = entry - s->msi_vectors;
 283     EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
 284
 285     if (!event_notifier_test_and_clear(n)) {
 286         return;
 287     }
 288
 289     IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, vector);
 290     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 291         msix_notify(pdev, vector);
 292     } else {
 293         ivshmem_IntrStatus_write(s, 1);
 294     }
 295 }
 296
 297 static int ivshmem_vector_unmask(PCIDevice *dev, unsigned vector,
 298                                  MSIMessage msg)
 299 {
 300     IVShmemState *s = IVSHMEM(dev);
 301     EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
 302     MSIVector *v = &s->msi_vectors[vector];
 303     int ret;
 304
 305     IVSHMEM_DPRINTF("vector unmask %p %d\n", dev, vector);
 306
 307     ret = kvm_irqchip_update_msi_route(kvm_state, v->virq, msg, dev);
 308     if (ret < 0) {
 309         return ret;
 310     }
 311
 312     return kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, n, NULL, v->virq);
 313 }
 314
 315 static void ivshmem_vector_mask(PCIDevice *dev, unsigned vector)
 316 {
 317     IVShmemState *s = IVSHMEM(dev);
 318     EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
 319     int ret;
 320
 321     IVSHMEM_DPRINTF("vector mask %p %d\n", dev, vector);
 322
 323     ret = kvm_irqchip_remove_irqfd_notifier_gsi(kvm_state, n,
 324                                                 s->msi_vectors[vector].virq);
 325     if (ret != 0) {
 326         error_report("remove_irqfd_notifier_gsi failed");
 327     }
 328 }
 329
 330 static void ivshmem_vector_poll(PCIDevice *dev,
 331                                 unsigned int vector_start,
 332                                 unsigned int vector_end)
 333 {
 334     IVShmemState *s = IVSHMEM(dev);
 335     unsigned int vector;
 336
 337     IVSHMEM_DPRINTF("vector poll %p %d-%d\n", dev, vector_start, vector_end);
 338
 339     vector_end = MIN(vector_end, s->vectors);
 340
 341     for (vector = vector_start; vector < vector_end; vector++) {
 342         EventNotifier *notifier = &s->peers[s->vm_id].eventfds[vector];
 343
 344         if (!msix_is_masked(dev, vector)) {
 345             continue;
 346         }
 347
 348         if (event_notifier_test_and_clear(notifier)) {
 349             msix_set_pending(dev, vector);
 350         }
 351     }
 352 }
 353
 354 static void watch_vector_notifier(IVShmemState *s, EventNotifier *n,
 355                                  int vector)
 356 {
 357     int eventfd = event_notifier_get_fd(n);
 358
 359     /* if MSI is supported we need multiple interrupts */
 360     s->msi_vectors[vector].pdev = PCI_DEVICE(s);
 361
 362     qemu_set_fd_handler(eventfd, ivshmem_vector_notify,
 363                         NULL, &s->msi_vectors[vector]);
 364 }
 365
 366 static int check_shm_size(IVShmemState *s, int fd, Error **errp)
 367 {
 368     /* check that the guest isn't going to try and map more memory than the
 369      * the object has allocated return -1 to indicate error */
 370
 371     struct stat buf;
 372
 373     if (fstat(fd, &buf) < 0) {
 374         error_setg(errp, "exiting: fstat on fd %d failed: %s",
 375                    fd, strerror(errno));
 376         return -1;
 377     }
 378
 379     if (s->ivshmem_size > buf.st_size) {
 380         error_setg(errp, "Requested memory size greater"
 381                    " than shared object size (%" PRIu64 " > %" PRIu64")",
 382                    s->ivshmem_size, (uint64_t)buf.st_size);
 383         return -1;
 384     } else {
 385         return 0;
 386     }
 387 }
 388
 389 /* create the shared memory BAR when we are not using the server, so we can
 390  * create the BAR and map the memory immediately */
 391 static int create_shared_memory_BAR(IVShmemState *s, int fd, uint8_t attr,
 392                                     Error **errp)
 393 {
 394     void * ptr;
 395
 396     ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 397     if (ptr == MAP_FAILED) {
 398         error_setg_errno(errp, errno, "Failed to mmap shared memory");
 399         return -1;
 400     }
 401
 402     memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s), "ivshmem.bar2",
 403                                s->ivshmem_size, ptr);
 404     qemu_set_ram_fd(memory_region_get_ram_addr(&s->ivshmem), fd);
 405     vmstate_register_ram(&s->ivshmem, DEVICE(s));
 406     memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 407
 408     /* region for shared memory */
 409     pci_register_bar(PCI_DEVICE(s), 2, attr, &s->bar);
 410
 411     return 0;
 412 }
 413
 414 static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i)
 415 {
 416     memory_region_add_eventfd(&s->ivshmem_mmio,
 417                               DOORBELL,
 418                               4,
 419                               true,
 420                               (posn << 16) | i,
 421                               &s->peers[posn].eventfds[i]);
 422 }
 423
 424 static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i)
 425 {
 426     memory_region_del_eventfd(&s->ivshmem_mmio,
 427                               DOORBELL,
 428                               4,
 429                               true,
 430                               (posn << 16) | i,
 431                               &s->peers[posn].eventfds[i]);
 432 }
 433
 434 static void close_peer_eventfds(IVShmemState *s, int posn)
 435 {
 436     int i, n;
 437
 438     if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 439         return;
 440     }
 441     if (posn < 0 || posn >= s->nb_peers) {
 442         error_report("invalid peer %d", posn);
 443         return;
 444     }
 445
 446     n = s->peers[posn].nb_eventfds;
 447
 448     memory_region_transaction_begin();
 449     for (i = 0; i < n; i++) {
 450         ivshmem_del_eventfd(s, posn, i);
 451     }
 452     memory_region_transaction_commit();
 453     for (i = 0; i < n; i++) {
 454         event_notifier_cleanup(&s->peers[posn].eventfds[i]);
 455     }
 456
 457     g_free(s->peers[posn].eventfds);
 458     s->peers[posn].nb_eventfds = 0;
 459 }
 460
 461 /* this function increase the dynamic storage need to store data about other
 462  * peers */
 463 static int resize_peers(IVShmemState *s, int new_min_size)
 464 {
 465
 466     int j, old_size;
 467
 468     /* limit number of max peers */
 469     if (new_min_size <= 0 || new_min_size > IVSHMEM_MAX_PEERS) {
 470         return -1;
 471     }
 472     if (new_min_size <= s->nb_peers) {
 473         return 0;
 474     }
 475
 476     old_size = s->nb_peers;
 477     s->nb_peers = new_min_size;
 478
 479     IVSHMEM_DPRINTF("bumping storage to %d peers\n", s->nb_peers);
 480
 481     s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
 482
 483     for (j = old_size; j < s->nb_peers; j++) {
 484         s->peers[j].eventfds = g_new0(EventNotifier, s->vectors);
 485         s->peers[j].nb_eventfds = 0;
 486     }
 487
 488     return 0;
 489 }
 490
 491 static bool fifo_update_and_get(IVShmemState *s, const uint8_t *buf, int size,
 492                                 void *data, size_t len)
 493 {
 494     const uint8_t *p;
 495     uint32_t num;
 496
 497     assert(len <= sizeof(int64_t)); /* limitation of the fifo */
 498     if (fifo8_is_empty(&s->incoming_fifo) && size == len) {
 499         memcpy(data, buf, size);
 500         return true;
 501     }
 502
 503     IVSHMEM_DPRINTF("short read of %d bytes\n", size);
 504
 505     num = MIN(size, sizeof(int64_t) - fifo8_num_used(&s->incoming_fifo));
 506     fifo8_push_all(&s->incoming_fifo, buf, num);
 507
 508     if (fifo8_num_used(&s->incoming_fifo) < len) {
 509         assert(num == 0);
 510         return false;
 511     }
 512
 513     size -= num;
 514     buf += num;
 515     p = fifo8_pop_buf(&s->incoming_fifo, len, &num);
 516     assert(num == len);
 517
 518     memcpy(data, p, len);
 519
 520     if (size > 0) {
 521         fifo8_push_all(&s->incoming_fifo, buf, size);
 522     }
 523
 524     return true;
 525 }
 526
 527 static bool fifo_update_and_get_i64(IVShmemState *s,
 528                                     const uint8_t *buf, int size, int64_t *i64)
 529 {
 530     if (fifo_update_and_get(s, buf, size, i64, sizeof(*i64))) {
 531         *i64 = GINT64_FROM_LE(*i64);
 532         return true;
 533     }
 534
 535     return false;
 536 }
 537
 538 static int ivshmem_add_kvm_msi_virq(IVShmemState *s, int vector)
 539 {
 540     PCIDevice *pdev = PCI_DEVICE(s);
 541     MSIMessage msg = msix_get_message(pdev, vector);
 542     int ret;
 543
 544     IVSHMEM_DPRINTF("ivshmem_add_kvm_msi_virq vector:%d\n", vector);
 545
 546     if (s->msi_vectors[vector].pdev != NULL) {
 547         return 0;
 548     }
 549
 550     ret = kvm_irqchip_add_msi_route(kvm_state, msg, pdev);
 551     if (ret < 0) {
 552         error_report("ivshmem: kvm_irqchip_add_msi_route failed");
 553         return -1;
 554     }
 555
 556     s->msi_vectors[vector].virq = ret;
 557     s->msi_vectors[vector].pdev = pdev;
 558
 559     return 0;
 560 }
 561
 562 static void setup_interrupt(IVShmemState *s, int vector)
 563 {
 564     EventNotifier *n = &s->peers[s->vm_id].eventfds[vector];
 565     bool with_irqfd = kvm_msi_via_irqfd_enabled() &&
 566         ivshmem_has_feature(s, IVSHMEM_MSI);
 567     PCIDevice *pdev = PCI_DEVICE(s);
 568
 569     IVSHMEM_DPRINTF("setting up interrupt for vector: %d\n", vector);
 570
 571     if (!with_irqfd) {
 572         IVSHMEM_DPRINTF("with eventfd");
 573         watch_vector_notifier(s, n, vector);
 574     } else if (msix_enabled(pdev)) {
 575         IVSHMEM_DPRINTF("with irqfd");
 576         if (ivshmem_add_kvm_msi_virq(s, vector) < 0) {
 577             return;
 578         }
 579
 580         if (!msix_is_masked(pdev, vector)) {
 581             kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, n, NULL,
 582                                                s->msi_vectors[vector].virq);
 583         }
 584     } else {
 585         /* it will be delayed until msix is enabled, in write_config */
 586         IVSHMEM_DPRINTF("with irqfd, delayed until msix enabled");
 587     }
 588 }
 589
 590 static void ivshmem_read(void *opaque, const uint8_t *buf, int size)
 591 {
 592     IVShmemState *s = opaque;
 593     int incoming_fd;
 594     int new_eventfd;
 595     int64_t incoming_posn;
 596     Error *err = NULL;
 597     Peer *peer;
 598
 599     if (!fifo_update_and_get_i64(s, buf, size, &incoming_posn)) {
 600         return;
 601     }
 602
 603     if (incoming_posn < -1) {
 604         IVSHMEM_DPRINTF("invalid incoming_posn %" PRId64 "\n", incoming_posn);
 605         return;
 606     }
 607
 608     /* pick off s->server_chr->msgfd and store it, posn should accompany msg */
 609     incoming_fd = qemu_chr_fe_get_msgfd(s->server_chr);
 610     IVSHMEM_DPRINTF("posn is %" PRId64 ", fd is %d\n",
 611                     incoming_posn, incoming_fd);
 612
 613     /* make sure we have enough space for this peer */
 614     if (incoming_posn >= s->nb_peers) {
 615         if (resize_peers(s, incoming_posn + 1) < 0) {
 616             error_report("failed to resize peers array");
 617             if (incoming_fd != -1) {
 618                 close(incoming_fd);
 619             }
 620             return;
 621         }
 622     }
 623
 624     peer = &s->peers[incoming_posn];
 625
 626     if (incoming_fd == -1) {
 627         /* if posn is positive and unseen before then this is our posn*/
 628         if (incoming_posn >= 0 && s->vm_id == -1) {
 629             /* receive our posn */
 630             s->vm_id = incoming_posn;
 631         } else {
 632             /* otherwise an fd == -1 means an existing peer has gone away */
 633             IVSHMEM_DPRINTF("posn %" PRId64 " has gone away\n", incoming_posn);
 634             close_peer_eventfds(s, incoming_posn);
 635         }
 636         return;
 637     }
 638
 639     /* if the position is -1, then it's shared memory region fd */
 640     if (incoming_posn == -1) {
 641         void * map_ptr;
 642
 643         if (memory_region_is_mapped(&s->ivshmem)) {
 644             error_report("shm already initialized");
 645             close(incoming_fd);
 646             return;
 647         }
 648
 649         if (check_shm_size(s, incoming_fd, &err) == -1) {
 650             error_report_err(err);
 651             close(incoming_fd);
 652             return;
 653         }
 654
 655         /* mmap the region and map into the BAR2 */
 656         map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
 657                                                             incoming_fd, 0);
 658         if (map_ptr == MAP_FAILED) {
 659             error_report("Failed to mmap shared memory %s", strerror(errno));
 660             close(incoming_fd);
 661             return;
 662         }
 663         memory_region_init_ram_ptr(&s->ivshmem, OBJECT(s),
 664                                    "ivshmem.bar2", s->ivshmem_size, map_ptr);
 665         qemu_set_ram_fd(memory_region_get_ram_addr(&s->ivshmem),
 666                         incoming_fd);
 667         vmstate_register_ram(&s->ivshmem, DEVICE(s));
 668
 669         IVSHMEM_DPRINTF("guest h/w addr = %p, size = %" PRIu64 "\n",
 670                         map_ptr, s->ivshmem_size);
 671
 672         memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
 673
 674         return;
 675     }
 676
 677     /* each peer has an associated array of eventfds, and we keep
 678      * track of how many eventfds received so far */
 679     /* get a new eventfd: */
 680     if (peer->nb_eventfds >= s->vectors) {
 681         error_report("Too many eventfd received, device has %d vectors",
 682                      s->vectors);
 683         close(incoming_fd);
 684         return;
 685     }
 686
 687     new_eventfd = peer->nb_eventfds++;
 688
 689     /* this is an eventfd for a particular peer VM */
 690     IVSHMEM_DPRINTF("eventfds[%" PRId64 "][%d] = %d\n", incoming_posn,
 691                     new_eventfd, incoming_fd);
 692     event_notifier_init_fd(&peer->eventfds[new_eventfd], incoming_fd);
 693     fcntl_setfl(incoming_fd, O_NONBLOCK); /* msix/irqfd poll non block */
 694
 695     if (incoming_posn == s->vm_id) {
 696         setup_interrupt(s, new_eventfd);
 697     }
 698
 699     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
 700         ivshmem_add_eventfd(s, incoming_posn, new_eventfd);
 701     }
 702 }
 703
 704 static void ivshmem_check_version(void *opaque, const uint8_t * buf, int size)
 705 {
 706     IVShmemState *s = opaque;
 707     int tmp;
 708     int64_t version;
 709
 710     if (!fifo_update_and_get_i64(s, buf, size, &version)) {
 711         return;
 712     }
 713
 714     tmp = qemu_chr_fe_get_msgfd(s->server_chr);
 715     if (tmp != -1 || version != IVSHMEM_PROTOCOL_VERSION) {
 716         fprintf(stderr, "incompatible version, you are connecting to a ivshmem-"
 717                 "server using a different protocol please check your setup\n");
 718         qemu_chr_delete(s->server_chr);
 719         s->server_chr = NULL;
 720         return;
 721     }
 722
 723     IVSHMEM_DPRINTF("version check ok, switch to real chardev handler\n");
 724     qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read,
 725                           ivshmem_event, s);
 726 }
 727
 728 /* Select the MSI-X vectors used by device.
 729  * ivshmem maps events to vectors statically, so
 730  * we just enable all vectors on init and after reset. */
 731 static void ivshmem_use_msix(IVShmemState * s)
 732 {
 733     PCIDevice *d = PCI_DEVICE(s);
 734     int i;
 735
 736     IVSHMEM_DPRINTF("%s, msix present: %d\n", __func__, msix_present(d));
 737     if (!msix_present(d)) {
 738         return;
 739     }
 740
 741     for (i = 0; i < s->vectors; i++) {
 742         msix_vector_use(d, i);
 743     }
 744 }
 745
 746 static void ivshmem_reset(DeviceState *d)
 747 {
 748     IVShmemState *s = IVSHMEM(d);
 749
 750     s->intrstatus = 0;
 751     s->intrmask = 0;
 752     ivshmem_use_msix(s);
 753 }
 754
 755 static int ivshmem_setup_interrupts(IVShmemState *s)
 756 {
 757     /* allocate QEMU callback data for receiving interrupts */
 758     s->msi_vectors = g_malloc0(s->vectors * sizeof(MSIVector));
 759
 760     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
 761         if (msix_init_exclusive_bar(PCI_DEVICE(s), s->vectors, 1)) {
 762             return -1;
 763         }
 764
 765         IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
 766         ivshmem_use_msix(s);
 767     }
 768
 769     return 0;
 770 }
 771
 772 static void ivshmem_enable_irqfd(IVShmemState *s)
 773 {
 774     PCIDevice *pdev = PCI_DEVICE(s);
 775     int i;
 776
 777     for (i = 0; i < s->peers[s->vm_id].nb_eventfds; i++) {
 778         ivshmem_add_kvm_msi_virq(s, i);
 779     }
 780
 781     if (msix_set_vector_notifiers(pdev,
 782                                   ivshmem_vector_unmask,
 783                                   ivshmem_vector_mask,
 784                                   ivshmem_vector_poll)) {
 785         error_report("ivshmem: msix_set_vector_notifiers failed");
 786     }
 787 }
 788
 789 static void ivshmem_remove_kvm_msi_virq(IVShmemState *s, int vector)
 790 {
 791     IVSHMEM_DPRINTF("ivshmem_remove_kvm_msi_virq vector:%d\n", vector);
 792
 793     if (s->msi_vectors[vector].pdev == NULL) {
 794         return;
 795     }
 796
 797     /* it was cleaned when masked in the frontend. */
 798     kvm_irqchip_release_virq(kvm_state, s->msi_vectors[vector].virq);
 799
 800     s->msi_vectors[vector].pdev = NULL;
 801 }
 802
 803 static void ivshmem_disable_irqfd(IVShmemState *s)
 804 {
 805     PCIDevice *pdev = PCI_DEVICE(s);
 806     int i;
 807
 808     for (i = 0; i < s->peers[s->vm_id].nb_eventfds; i++) {
 809         ivshmem_remove_kvm_msi_virq(s, i);
 810     }
 811
 812     msix_unset_vector_notifiers(pdev);
 813 }
 814
 815 static void ivshmem_write_config(PCIDevice *pdev, uint32_t address,
 816                                  uint32_t val, int len)
 817 {
 818     IVShmemState *s = IVSHMEM(pdev);
 819     int is_enabled, was_enabled = msix_enabled(pdev);
 820
 821     pci_default_write_config(pdev, address, val, len);
 822     is_enabled = msix_enabled(pdev);
 823
 824     if (kvm_msi_via_irqfd_enabled() && s->vm_id != -1) {
 825         if (!was_enabled && is_enabled) {
 826             ivshmem_enable_irqfd(s);
 827         } else if (was_enabled && !is_enabled) {
 828             ivshmem_disable_irqfd(s);
 829         }
 830     }
 831 }
 832
 833 static void pci_ivshmem_realize(PCIDevice *dev, Error **errp)
 834 {
 835     IVShmemState *s = IVSHMEM(dev);
 836     uint8_t *pci_conf;
 837     uint8_t attr = PCI_BASE_ADDRESS_SPACE_MEMORY |
 838         PCI_BASE_ADDRESS_MEM_PREFETCH;
 839
 840     if (!!s->server_chr + !!s->shmobj + !!s->hostmem != 1) {
 841         error_setg(errp,
 842                    "You must specify either 'shm', 'chardev' or 'x-memdev'");
 843         return;
 844     }
 845
 846     if (s->hostmem) {
 847         MemoryRegion *mr;
 848
 849         if (s->sizearg) {
 850             g_warning("size argument ignored with hostmem");
 851         }
 852
 853         mr = host_memory_backend_get_memory(s->hostmem, errp);
 854         s->ivshmem_size = memory_region_size(mr);
 855     } else if (s->sizearg == NULL) {
 856         s->ivshmem_size = 4 << 20; /* 4 MB default */
 857     } else {
 858         char *end;
 859         int64_t size = qemu_strtosz(s->sizearg, &end);
 860         if (size < 0 || *end != '\0' || !is_power_of_2(size)) {
 861             error_setg(errp, "Invalid size %s", s->sizearg);
 862             return;
 863         }
 864         s->ivshmem_size = size;
 865     }
 866
 867     fifo8_create(&s->incoming_fifo, sizeof(int64_t));
 868
 869     /* IRQFD requires MSI */
 870     if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
 871         !ivshmem_has_feature(s, IVSHMEM_MSI)) {
 872         error_setg(errp, "ioeventfd/irqfd requires MSI");
 873         return;
 874     }
 875
 876     /* check that role is reasonable */
 877     if (s->role) {
 878         if (strncmp(s->role, "peer", 5) == 0) {
 879             s->role_val = IVSHMEM_PEER;
 880         } else if (strncmp(s->role, "master", 7) == 0) {
 881             s->role_val = IVSHMEM_MASTER;
 882         } else {
 883             error_setg(errp, "'role' must be 'peer' or 'master'");
 884             return;
 885         }
 886     } else {
 887         s->role_val = IVSHMEM_MASTER; /* default */
 888     }
 889
 890     if (s->role_val == IVSHMEM_PEER) {
 891         error_setg(&s->migration_blocker,
 892                    "Migration is disabled when using feature 'peer mode' in device 'ivshmem'");
 893         migrate_add_blocker(s->migration_blocker);
 894     }
 895
 896     pci_conf = dev->config;
 897     pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY;
 898
 899     pci_config_set_interrupt_pin(pci_conf, 1);
 900
 901     memory_region_init_io(&s->ivshmem_mmio, OBJECT(s), &ivshmem_mmio_ops, s,
 902                           "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
 903
 904     /* region for registers*/
 905     pci_register_bar(dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
 906                      &s->ivshmem_mmio);
 907
 908     memory_region_init(&s->bar, OBJECT(s), "ivshmem-bar2-container", s->ivshmem_size);
 909     if (s->ivshmem_64bit) {
 910         attr |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 911     }
 912
 913     if (s->hostmem != NULL) {
 914         MemoryRegion *mr;
 915
 916         IVSHMEM_DPRINTF("using hostmem\n");
 917
 918         mr = host_memory_backend_get_memory(MEMORY_BACKEND(s->hostmem), errp);
 919         vmstate_register_ram(mr, DEVICE(s));
 920         memory_region_add_subregion(&s->bar, 0, mr);
 921         pci_register_bar(PCI_DEVICE(s), 2, attr, &s->bar);
 922     } else if (s->server_chr != NULL) {
 923         /* FIXME do not rely on what chr drivers put into filename */
 924         if (strncmp(s->server_chr->filename, "unix:", 5)) {
 925             error_setg(errp, "chardev is not a unix client socket");
 926             return;
 927         }
 928
 929         /* if we get a UNIX socket as the parameter we will talk
 930          * to the ivshmem server to receive the memory region */
 931
 932         IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
 933                         s->server_chr->filename);
 934
 935         if (ivshmem_setup_interrupts(s) < 0) {
 936             error_setg(errp, "failed to initialize interrupts");
 937             return;
 938         }
 939
 940         /* we allocate enough space for 16 peers and grow as needed */
 941         resize_peers(s, 16);
 942         s->vm_id = -1;
 943
 944         pci_register_bar(dev, 2, attr, &s->bar);
 945
 946         s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
 947
 948         qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive,
 949                               ivshmem_check_version, ivshmem_event, s);
 950     } else {
 951         /* just map the file immediately, we're not using a server */
 952         int fd;
 953
 954         IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
 955
 956         /* try opening with O_EXCL and if it succeeds zero the memory
 957          * by truncating to 0 */
 958         if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
 959                         S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
 960            /* truncate file to length PCI device's memory */
 961             if (ftruncate(fd, s->ivshmem_size) != 0) {
 962                 error_report("could not truncate shared file");
 963             }
 964
 965         } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR,
 966                         S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
 967             error_setg(errp, "could not open shared file");
 968             return;
 969         }
 970
 971         if (check_shm_size(s, fd, errp) == -1) {
 972             return;
 973         }
 974
 975         create_shared_memory_BAR(s, fd, attr, errp);
 976     }
 977 }
 978
 979 static void pci_ivshmem_exit(PCIDevice *dev)
 980 {
 981     IVShmemState *s = IVSHMEM(dev);
 982     int i;
 983
 984     fifo8_destroy(&s->incoming_fifo);
 985
 986     if (s->migration_blocker) {
 987         migrate_del_blocker(s->migration_blocker);
 988         error_free(s->migration_blocker);
 989     }
 990
 991     if (memory_region_is_mapped(&s->ivshmem)) {
 992         if (!s->hostmem) {
 993             void *addr = memory_region_get_ram_ptr(&s->ivshmem);
 994             int fd;
 995
 996             if (munmap(addr, s->ivshmem_size) == -1) {
 997                 error_report("Failed to munmap shared memory %s",
 998                              strerror(errno));
 999             }
1000
1001             fd = qemu_get_ram_fd(memory_region_get_ram_addr(&s->ivshmem));
1002             if (fd != -1) {
1003                 close(fd);
1004             }
1005         }
1006
1007         vmstate_unregister_ram(&s->ivshmem, DEVICE(dev));
1008         memory_region_del_subregion(&s->bar, &s->ivshmem);
1009     }
1010
1011     if (s->eventfd_chr) {
1012         for (i = 0; i < s->vectors; i++) {
1013             if (s->eventfd_chr[i]) {
1014                 qemu_chr_free(s->eventfd_chr[i]);
1015             }
1016         }
1017         g_free(s->eventfd_chr);
1018     }
1019
1020     if (s->peers) {
1021         for (i = 0; i < s->nb_peers; i++) {
1022             close_peer_eventfds(s, i);
1023         }
1024         g_free(s->peers);
1025     }
1026
1027     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
1028         msix_uninit_exclusive_bar(dev);
1029     }
1030
1031     g_free(s->msi_vectors);
1032 }
1033
1034 static bool test_msix(void *opaque, int version_id)
1035 {
1036     IVShmemState *s = opaque;
1037
1038     return ivshmem_has_feature(s, IVSHMEM_MSI);
1039 }
1040
1041 static bool test_no_msix(void *opaque, int version_id)
1042 {
1043     return !test_msix(opaque, version_id);
1044 }
1045
1046 static int ivshmem_pre_load(void *opaque)
1047 {
1048     IVShmemState *s = opaque;
1049
1050     if (s->role_val == IVSHMEM_PEER) {
1051         error_report("'peer' devices are not migratable");
1052         return -EINVAL;
1053     }
1054
1055     return 0;
1056 }
1057
1058 static int ivshmem_post_load(void *opaque, int version_id)
1059 {
1060     IVShmemState *s = opaque;
1061
1062     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
1063         ivshmem_use_msix(s);
1064     }
1065
1066     return 0;
1067 }
1068
1069 static int ivshmem_load_old(QEMUFile *f, void *opaque, int version_id)
1070 {
1071     IVShmemState *s = opaque;
1072     PCIDevice *pdev = PCI_DEVICE(s);
1073     int ret;
1074
1075     IVSHMEM_DPRINTF("ivshmem_load_old\n");
1076
1077     if (version_id != 0) {
1078         return -EINVAL;
1079     }
1080
1081     if (s->role_val == IVSHMEM_PEER) {
1082         error_report("'peer' devices are not migratable");
1083         return -EINVAL;
1084     }
1085
1086     ret = pci_device_load(pdev, f);
1087     if (ret) {
1088         return ret;
1089     }
1090
1091     if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
1092         msix_load(pdev, f);
1093         ivshmem_use_msix(s);
1094     } else {
1095         s->intrstatus = qemu_get_be32(f);
1096         s->intrmask = qemu_get_be32(f);
1097     }
1098
1099     return 0;
1100 }
1101
1102 static const VMStateDescription ivshmem_vmsd = {
1103     .name = "ivshmem",
1104     .version_id = 1,
1105     .minimum_version_id = 1,
1106     .pre_load = ivshmem_pre_load,
1107     .post_load = ivshmem_post_load,
1108     .fields = (VMStateField[]) {
1109         VMSTATE_PCI_DEVICE(parent_obj, IVShmemState),
1110
1111         VMSTATE_MSIX_TEST(parent_obj, IVShmemState, test_msix),
1112         VMSTATE_UINT32_TEST(intrstatus, IVShmemState, test_no_msix),
1113         VMSTATE_UINT32_TEST(intrmask, IVShmemState, test_no_msix),
1114
1115         VMSTATE_END_OF_LIST()
1116     },
1117     .load_state_old = ivshmem_load_old,
1118     .minimum_version_id_old = 0
1119 };
1120
1121 static Property ivshmem_properties[] = {
1122     DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
1123     DEFINE_PROP_STRING("size", IVShmemState, sizearg),
1124     DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1),
1125     DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false),
1126     DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true),
1127     DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
1128     DEFINE_PROP_STRING("role", IVShmemState, role),
1129     DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1),
1130     DEFINE_PROP_END_OF_LIST(),
1131 };
1132
1133 static void ivshmem_class_init(ObjectClass *klass, void *data)
1134 {
1135     DeviceClass *dc = DEVICE_CLASS(klass);
1136     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1137
1138     k->realize = pci_ivshmem_realize;
1139     k->exit = pci_ivshmem_exit;
1140     k->config_write = ivshmem_write_config;
1141     k->vendor_id = PCI_VENDOR_ID_IVSHMEM;
1142     k->device_id = PCI_DEVICE_ID_IVSHMEM;
1143     k->class_id = PCI_CLASS_MEMORY_RAM;
1144     dc->reset = ivshmem_reset;
1145     dc->props = ivshmem_properties;
1146     dc->vmsd = &ivshmem_vmsd;
1147     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
1148     dc->desc = "Inter-VM shared memory";
1149 }
1150
1151 static void ivshmem_check_memdev_is_busy(Object *obj, const char *name,
1152                                          Object *val, Error **errp)
1153 {
1154     MemoryRegion *mr;
1155
1156     mr = host_memory_backend_get_memory(MEMORY_BACKEND(val), errp);
1157     if (memory_region_is_mapped(mr)) {
1158         char *path = object_get_canonical_path_component(val);
1159         error_setg(errp, "can't use already busy memdev: %s", path);
1160         g_free(path);
1161     } else {
1162         qdev_prop_allow_set_link_before_realize(obj, name, val, errp);
1163     }
1164 }
1165
1166 static void ivshmem_init(Object *obj)
1167 {
1168     IVShmemState *s = IVSHMEM(obj);
1169
1170     object_property_add_link(obj, "x-memdev", TYPE_MEMORY_BACKEND,
1171                              (Object **)&s->hostmem,
1172                              ivshmem_check_memdev_is_busy,
1173                              OBJ_PROP_LINK_UNREF_ON_RELEASE,
1174                              &error_abort);
1175 }
1176
1177 static const TypeInfo ivshmem_info = {
1178     .name          = TYPE_IVSHMEM,
1179     .parent        = TYPE_PCI_DEVICE,
1180     .instance_size = sizeof(IVShmemState),
1181     .instance_init = ivshmem_init,
1182     .class_init    = ivshmem_class_init,
1183 };
1184
1185 static void ivshmem_register_types(void)
1186 {
1187     type_register_static(&ivshmem_info);
1188 }
1189
1190 type_init(ivshmem_register_types)