1 /* 2 * Virtio PCI Bindings 3 * 4 * Copyright IBM, Corp. 2007 5 * Copyright (c) 2009 CodeSourcery 6 * 7 * Authors: 8 * Anthony Liguori <aliguori@us.ibm.com> 9 * Paul Brook <paul@codesourcery.com> 10 * 11 * This work is licensed under the terms of the GNU GPL, version 2. See 12 * the COPYING file in the top-level directory. 13 * 14 * Contributions after 2012-01-13 are licensed under the terms of the 15 * GNU GPL, version 2 or (at your option) any later version. 16 */ 17 18 #include "qemu/osdep.h" 19 20 #include "exec/memop.h" 21 #include "standard-headers/linux/virtio_pci.h" 22 #include "standard-headers/linux/virtio_ids.h" 23 #include "hw/boards.h" 24 #include "hw/virtio/virtio.h" 25 #include "migration/qemu-file-types.h" 26 #include "hw/pci/pci.h" 27 #include "hw/pci/pci_bus.h" 28 #include "hw/qdev-properties.h" 29 #include "qapi/error.h" 30 #include "qemu/error-report.h" 31 #include "qemu/log.h" 32 #include "qemu/module.h" 33 #include "hw/pci/msi.h" 34 #include "hw/pci/msix.h" 35 #include "hw/loader.h" 36 #include "sysemu/kvm.h" 37 #include "hw/virtio/virtio-pci.h" 38 #include "qemu/range.h" 39 #include "hw/virtio/virtio-bus.h" 40 #include "qapi/visitor.h" 41 #include "sysemu/replay.h" 42 #include "trace.h" 43 44 #define VIRTIO_PCI_REGION_SIZE(dev) VIRTIO_PCI_CONFIG_OFF(msix_present(dev)) 45 46 #undef VIRTIO_PCI_CONFIG 47 48 /* The remaining space is defined by each driver as the per-driver 49 * configuration space */ 50 #define VIRTIO_PCI_CONFIG_SIZE(dev) VIRTIO_PCI_CONFIG_OFF(msix_enabled(dev)) 51 52 static void virtio_pci_bus_new(VirtioBusState *bus, size_t bus_size, 53 VirtIOPCIProxy *dev); 54 static void virtio_pci_reset(DeviceState *qdev); 55 56 /* virtio device */ 57 /* DeviceState to VirtIOPCIProxy. For use off data-path. TODO: use QOM. */ 58 static inline VirtIOPCIProxy *to_virtio_pci_proxy(DeviceState *d) 59 { 60 return container_of(d, VirtIOPCIProxy, pci_dev.qdev); 61 } 62 63 /* DeviceState to VirtIOPCIProxy. Note: used on datapath, 64 * be careful and test performance if you change this. 65 */ 66 static inline VirtIOPCIProxy *to_virtio_pci_proxy_fast(DeviceState *d) 67 { 68 return container_of(d, VirtIOPCIProxy, pci_dev.qdev); 69 } 70 71 static void virtio_pci_notify(DeviceState *d, uint16_t vector) 72 { 73 VirtIOPCIProxy *proxy = to_virtio_pci_proxy_fast(d); 74 75 if (msix_enabled(&proxy->pci_dev)) { 76 if (vector != VIRTIO_NO_VECTOR) { 77 msix_notify(&proxy->pci_dev, vector); 78 } 79 } else { 80 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 81 pci_set_irq(&proxy->pci_dev, qatomic_read(&vdev->isr) & 1); 82 } 83 } 84 85 static void virtio_pci_save_config(DeviceState *d, QEMUFile *f) 86 { 87 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 88 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 89 90 pci_device_save(&proxy->pci_dev, f); 91 msix_save(&proxy->pci_dev, f); 92 if (msix_present(&proxy->pci_dev)) 93 qemu_put_be16(f, vdev->config_vector); 94 } 95 96 static const VMStateDescription vmstate_virtio_pci_modern_queue_state = { 97 .name = "virtio_pci/modern_queue_state", 98 .version_id = 1, 99 .minimum_version_id = 1, 100 .fields = (const VMStateField[]) { 101 VMSTATE_UINT16(num, VirtIOPCIQueue), 102 VMSTATE_UNUSED(1), /* enabled was stored as be16 */ 103 VMSTATE_BOOL(enabled, VirtIOPCIQueue), 104 VMSTATE_UINT32_ARRAY(desc, VirtIOPCIQueue, 2), 105 VMSTATE_UINT32_ARRAY(avail, VirtIOPCIQueue, 2), 106 VMSTATE_UINT32_ARRAY(used, VirtIOPCIQueue, 2), 107 VMSTATE_END_OF_LIST() 108 } 109 }; 110 111 static bool virtio_pci_modern_state_needed(void *opaque) 112 { 113 VirtIOPCIProxy *proxy = opaque; 114 115 return virtio_pci_modern(proxy); 116 } 117 118 static const VMStateDescription vmstate_virtio_pci_modern_state_sub = { 119 .name = "virtio_pci/modern_state", 120 .version_id = 1, 121 .minimum_version_id = 1, 122 .needed = &virtio_pci_modern_state_needed, 123 .fields = (const VMStateField[]) { 124 VMSTATE_UINT32(dfselect, VirtIOPCIProxy), 125 VMSTATE_UINT32(gfselect, VirtIOPCIProxy), 126 VMSTATE_UINT32_ARRAY(guest_features, VirtIOPCIProxy, 2), 127 VMSTATE_STRUCT_ARRAY(vqs, VirtIOPCIProxy, VIRTIO_QUEUE_MAX, 0, 128 vmstate_virtio_pci_modern_queue_state, 129 VirtIOPCIQueue), 130 VMSTATE_END_OF_LIST() 131 } 132 }; 133 134 static const VMStateDescription vmstate_virtio_pci = { 135 .name = "virtio_pci", 136 .version_id = 1, 137 .minimum_version_id = 1, 138 .fields = (const VMStateField[]) { 139 VMSTATE_END_OF_LIST() 140 }, 141 .subsections = (const VMStateDescription * const []) { 142 &vmstate_virtio_pci_modern_state_sub, 143 NULL 144 } 145 }; 146 147 static bool virtio_pci_has_extra_state(DeviceState *d) 148 { 149 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 150 151 return proxy->flags & VIRTIO_PCI_FLAG_MIGRATE_EXTRA; 152 } 153 154 static void virtio_pci_save_extra_state(DeviceState *d, QEMUFile *f) 155 { 156 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 157 158 vmstate_save_state(f, &vmstate_virtio_pci, proxy, NULL); 159 } 160 161 static int virtio_pci_load_extra_state(DeviceState *d, QEMUFile *f) 162 { 163 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 164 165 return vmstate_load_state(f, &vmstate_virtio_pci, proxy, 1); 166 } 167 168 static void virtio_pci_save_queue(DeviceState *d, int n, QEMUFile *f) 169 { 170 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 171 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 172 173 if (msix_present(&proxy->pci_dev)) 174 qemu_put_be16(f, virtio_queue_vector(vdev, n)); 175 } 176 177 static int virtio_pci_load_config(DeviceState *d, QEMUFile *f) 178 { 179 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 180 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 181 uint16_t vector; 182 183 int ret; 184 ret = pci_device_load(&proxy->pci_dev, f); 185 if (ret) { 186 return ret; 187 } 188 msix_unuse_all_vectors(&proxy->pci_dev); 189 msix_load(&proxy->pci_dev, f); 190 if (msix_present(&proxy->pci_dev)) { 191 qemu_get_be16s(f, &vector); 192 193 if (vector != VIRTIO_NO_VECTOR && vector >= proxy->nvectors) { 194 return -EINVAL; 195 } 196 } else { 197 vector = VIRTIO_NO_VECTOR; 198 } 199 vdev->config_vector = vector; 200 if (vector != VIRTIO_NO_VECTOR) { 201 msix_vector_use(&proxy->pci_dev, vector); 202 } 203 return 0; 204 } 205 206 static int virtio_pci_load_queue(DeviceState *d, int n, QEMUFile *f) 207 { 208 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 209 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 210 211 uint16_t vector; 212 if (msix_present(&proxy->pci_dev)) { 213 qemu_get_be16s(f, &vector); 214 if (vector != VIRTIO_NO_VECTOR && vector >= proxy->nvectors) { 215 return -EINVAL; 216 } 217 } else { 218 vector = VIRTIO_NO_VECTOR; 219 } 220 virtio_queue_set_vector(vdev, n, vector); 221 if (vector != VIRTIO_NO_VECTOR) { 222 msix_vector_use(&proxy->pci_dev, vector); 223 } 224 225 return 0; 226 } 227 228 typedef struct VirtIOPCIIDInfo { 229 /* virtio id */ 230 uint16_t vdev_id; 231 /* pci device id for the transitional device */ 232 uint16_t trans_devid; 233 uint16_t class_id; 234 } VirtIOPCIIDInfo; 235 236 static const VirtIOPCIIDInfo virtio_pci_id_info[] = { 237 { 238 .vdev_id = VIRTIO_ID_CRYPTO, 239 .class_id = PCI_CLASS_OTHERS, 240 }, { 241 .vdev_id = VIRTIO_ID_FS, 242 .class_id = PCI_CLASS_STORAGE_OTHER, 243 }, { 244 .vdev_id = VIRTIO_ID_NET, 245 .trans_devid = PCI_DEVICE_ID_VIRTIO_NET, 246 .class_id = PCI_CLASS_NETWORK_ETHERNET, 247 }, { 248 .vdev_id = VIRTIO_ID_BLOCK, 249 .trans_devid = PCI_DEVICE_ID_VIRTIO_BLOCK, 250 .class_id = PCI_CLASS_STORAGE_SCSI, 251 }, { 252 .vdev_id = VIRTIO_ID_CONSOLE, 253 .trans_devid = PCI_DEVICE_ID_VIRTIO_CONSOLE, 254 .class_id = PCI_CLASS_COMMUNICATION_OTHER, 255 }, { 256 .vdev_id = VIRTIO_ID_SCSI, 257 .trans_devid = PCI_DEVICE_ID_VIRTIO_SCSI, 258 .class_id = PCI_CLASS_STORAGE_SCSI 259 }, { 260 .vdev_id = VIRTIO_ID_9P, 261 .trans_devid = PCI_DEVICE_ID_VIRTIO_9P, 262 .class_id = PCI_BASE_CLASS_NETWORK, 263 }, { 264 .vdev_id = VIRTIO_ID_BALLOON, 265 .trans_devid = PCI_DEVICE_ID_VIRTIO_BALLOON, 266 .class_id = PCI_CLASS_OTHERS, 267 }, { 268 .vdev_id = VIRTIO_ID_RNG, 269 .trans_devid = PCI_DEVICE_ID_VIRTIO_RNG, 270 .class_id = PCI_CLASS_OTHERS, 271 }, 272 }; 273 274 static const VirtIOPCIIDInfo *virtio_pci_get_id_info(uint16_t vdev_id) 275 { 276 const VirtIOPCIIDInfo *info = NULL; 277 int i; 278 279 for (i = 0; i < ARRAY_SIZE(virtio_pci_id_info); i++) { 280 if (virtio_pci_id_info[i].vdev_id == vdev_id) { 281 info = &virtio_pci_id_info[i]; 282 break; 283 } 284 } 285 286 if (!info) { 287 /* The device id is invalid or not added to the id_info yet. */ 288 error_report("Invalid virtio device(id %u)", vdev_id); 289 abort(); 290 } 291 292 return info; 293 } 294 295 /* 296 * Get the Transitional Device ID for the specific device, return 297 * zero if the device is non-transitional. 298 */ 299 uint16_t virtio_pci_get_trans_devid(uint16_t device_id) 300 { 301 return virtio_pci_get_id_info(device_id)->trans_devid; 302 } 303 304 /* 305 * Get the Class ID for the specific device. 306 */ 307 uint16_t virtio_pci_get_class_id(uint16_t device_id) 308 { 309 return virtio_pci_get_id_info(device_id)->class_id; 310 } 311 312 static bool virtio_pci_ioeventfd_enabled(DeviceState *d) 313 { 314 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 315 316 return (proxy->flags & VIRTIO_PCI_FLAG_USE_IOEVENTFD) != 0; 317 } 318 319 #define QEMU_VIRTIO_PCI_QUEUE_MEM_MULT 0x1000 320 321 static inline int virtio_pci_queue_mem_mult(struct VirtIOPCIProxy *proxy) 322 { 323 return (proxy->flags & VIRTIO_PCI_FLAG_PAGE_PER_VQ) ? 324 QEMU_VIRTIO_PCI_QUEUE_MEM_MULT : 4; 325 } 326 327 static int virtio_pci_ioeventfd_assign(DeviceState *d, EventNotifier *notifier, 328 int n, bool assign) 329 { 330 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 331 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 332 VirtQueue *vq = virtio_get_queue(vdev, n); 333 bool legacy = virtio_pci_legacy(proxy); 334 bool modern = virtio_pci_modern(proxy); 335 bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY; 336 MemoryRegion *modern_mr = &proxy->notify.mr; 337 MemoryRegion *modern_notify_mr = &proxy->notify_pio.mr; 338 MemoryRegion *legacy_mr = &proxy->bar; 339 hwaddr modern_addr = virtio_pci_queue_mem_mult(proxy) * 340 virtio_get_queue_index(vq); 341 hwaddr legacy_addr = VIRTIO_PCI_QUEUE_NOTIFY; 342 343 if (assign) { 344 if (modern) { 345 memory_region_add_eventfd(modern_mr, modern_addr, 0, 346 false, n, notifier); 347 if (modern_pio) { 348 memory_region_add_eventfd(modern_notify_mr, 0, 2, 349 true, n, notifier); 350 } 351 } 352 if (legacy) { 353 memory_region_add_eventfd(legacy_mr, legacy_addr, 2, 354 true, n, notifier); 355 } 356 } else { 357 if (modern) { 358 memory_region_del_eventfd(modern_mr, modern_addr, 0, 359 false, n, notifier); 360 if (modern_pio) { 361 memory_region_del_eventfd(modern_notify_mr, 0, 2, 362 true, n, notifier); 363 } 364 } 365 if (legacy) { 366 memory_region_del_eventfd(legacy_mr, legacy_addr, 2, 367 true, n, notifier); 368 } 369 } 370 return 0; 371 } 372 373 static void virtio_pci_start_ioeventfd(VirtIOPCIProxy *proxy) 374 { 375 virtio_bus_start_ioeventfd(&proxy->bus); 376 } 377 378 static void virtio_pci_stop_ioeventfd(VirtIOPCIProxy *proxy) 379 { 380 virtio_bus_stop_ioeventfd(&proxy->bus); 381 } 382 383 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val) 384 { 385 VirtIOPCIProxy *proxy = opaque; 386 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 387 uint16_t vector, vq_idx; 388 hwaddr pa; 389 390 switch (addr) { 391 case VIRTIO_PCI_GUEST_FEATURES: 392 /* Guest does not negotiate properly? We have to assume nothing. */ 393 if (val & (1 << VIRTIO_F_BAD_FEATURE)) { 394 val = virtio_bus_get_vdev_bad_features(&proxy->bus); 395 } 396 virtio_set_features(vdev, val); 397 break; 398 case VIRTIO_PCI_QUEUE_PFN: 399 pa = (hwaddr)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT; 400 if (pa == 0) { 401 virtio_pci_reset(DEVICE(proxy)); 402 } 403 else 404 virtio_queue_set_addr(vdev, vdev->queue_sel, pa); 405 break; 406 case VIRTIO_PCI_QUEUE_SEL: 407 if (val < VIRTIO_QUEUE_MAX) 408 vdev->queue_sel = val; 409 break; 410 case VIRTIO_PCI_QUEUE_NOTIFY: 411 vq_idx = val; 412 if (vq_idx < VIRTIO_QUEUE_MAX && virtio_queue_get_num(vdev, vq_idx)) { 413 if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFICATION_DATA)) { 414 VirtQueue *vq = virtio_get_queue(vdev, vq_idx); 415 416 virtio_queue_set_shadow_avail_idx(vq, val >> 16); 417 } 418 virtio_queue_notify(vdev, vq_idx); 419 } 420 break; 421 case VIRTIO_PCI_STATUS: 422 if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) { 423 virtio_pci_stop_ioeventfd(proxy); 424 } 425 426 virtio_set_status(vdev, val & 0xFF); 427 428 if (val & VIRTIO_CONFIG_S_DRIVER_OK) { 429 virtio_pci_start_ioeventfd(proxy); 430 } 431 432 if (vdev->status == 0) { 433 virtio_pci_reset(DEVICE(proxy)); 434 } 435 436 /* Linux before 2.6.34 drives the device without enabling 437 the PCI device bus master bit. Enable it automatically 438 for the guest. This is a PCI spec violation but so is 439 initiating DMA with bus master bit clear. */ 440 if (val == (VIRTIO_CONFIG_S_ACKNOWLEDGE | VIRTIO_CONFIG_S_DRIVER)) { 441 pci_default_write_config(&proxy->pci_dev, PCI_COMMAND, 442 proxy->pci_dev.config[PCI_COMMAND] | 443 PCI_COMMAND_MASTER, 1); 444 } 445 break; 446 case VIRTIO_MSI_CONFIG_VECTOR: 447 if (vdev->config_vector != VIRTIO_NO_VECTOR) { 448 msix_vector_unuse(&proxy->pci_dev, vdev->config_vector); 449 } 450 /* Make it possible for guest to discover an error took place. */ 451 if (val < proxy->nvectors) { 452 msix_vector_use(&proxy->pci_dev, val); 453 } else { 454 val = VIRTIO_NO_VECTOR; 455 } 456 vdev->config_vector = val; 457 break; 458 case VIRTIO_MSI_QUEUE_VECTOR: 459 vector = virtio_queue_vector(vdev, vdev->queue_sel); 460 if (vector != VIRTIO_NO_VECTOR) { 461 msix_vector_unuse(&proxy->pci_dev, vector); 462 } 463 /* Make it possible for guest to discover an error took place. */ 464 if (val < proxy->nvectors) { 465 msix_vector_use(&proxy->pci_dev, val); 466 } else { 467 val = VIRTIO_NO_VECTOR; 468 } 469 virtio_queue_set_vector(vdev, vdev->queue_sel, val); 470 break; 471 default: 472 qemu_log_mask(LOG_GUEST_ERROR, 473 "%s: unexpected address 0x%x value 0x%x\n", 474 __func__, addr, val); 475 break; 476 } 477 } 478 479 static uint32_t virtio_ioport_read(VirtIOPCIProxy *proxy, uint32_t addr) 480 { 481 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 482 uint32_t ret = 0xFFFFFFFF; 483 484 switch (addr) { 485 case VIRTIO_PCI_HOST_FEATURES: 486 ret = vdev->host_features; 487 break; 488 case VIRTIO_PCI_GUEST_FEATURES: 489 ret = vdev->guest_features; 490 break; 491 case VIRTIO_PCI_QUEUE_PFN: 492 ret = virtio_queue_get_addr(vdev, vdev->queue_sel) 493 >> VIRTIO_PCI_QUEUE_ADDR_SHIFT; 494 break; 495 case VIRTIO_PCI_QUEUE_NUM: 496 ret = virtio_queue_get_num(vdev, vdev->queue_sel); 497 break; 498 case VIRTIO_PCI_QUEUE_SEL: 499 ret = vdev->queue_sel; 500 break; 501 case VIRTIO_PCI_STATUS: 502 ret = vdev->status; 503 break; 504 case VIRTIO_PCI_ISR: 505 /* reading from the ISR also clears it. */ 506 ret = qatomic_xchg(&vdev->isr, 0); 507 pci_irq_deassert(&proxy->pci_dev); 508 break; 509 case VIRTIO_MSI_CONFIG_VECTOR: 510 ret = vdev->config_vector; 511 break; 512 case VIRTIO_MSI_QUEUE_VECTOR: 513 ret = virtio_queue_vector(vdev, vdev->queue_sel); 514 break; 515 default: 516 break; 517 } 518 519 return ret; 520 } 521 522 static uint64_t virtio_pci_config_read(void *opaque, hwaddr addr, 523 unsigned size) 524 { 525 VirtIOPCIProxy *proxy = opaque; 526 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 527 uint32_t config = VIRTIO_PCI_CONFIG_SIZE(&proxy->pci_dev); 528 uint64_t val = 0; 529 530 if (vdev == NULL) { 531 return UINT64_MAX; 532 } 533 534 if (addr < config) { 535 return virtio_ioport_read(proxy, addr); 536 } 537 addr -= config; 538 539 switch (size) { 540 case 1: 541 val = virtio_config_readb(vdev, addr); 542 break; 543 case 2: 544 val = virtio_config_readw(vdev, addr); 545 if (virtio_is_big_endian(vdev)) { 546 val = bswap16(val); 547 } 548 break; 549 case 4: 550 val = virtio_config_readl(vdev, addr); 551 if (virtio_is_big_endian(vdev)) { 552 val = bswap32(val); 553 } 554 break; 555 } 556 return val; 557 } 558 559 static void virtio_pci_config_write(void *opaque, hwaddr addr, 560 uint64_t val, unsigned size) 561 { 562 VirtIOPCIProxy *proxy = opaque; 563 uint32_t config = VIRTIO_PCI_CONFIG_SIZE(&proxy->pci_dev); 564 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 565 566 if (vdev == NULL) { 567 return; 568 } 569 570 if (addr < config) { 571 virtio_ioport_write(proxy, addr, val); 572 return; 573 } 574 addr -= config; 575 /* 576 * Virtio-PCI is odd. Ioports are LE but config space is target native 577 * endian. 578 */ 579 switch (size) { 580 case 1: 581 virtio_config_writeb(vdev, addr, val); 582 break; 583 case 2: 584 if (virtio_is_big_endian(vdev)) { 585 val = bswap16(val); 586 } 587 virtio_config_writew(vdev, addr, val); 588 break; 589 case 4: 590 if (virtio_is_big_endian(vdev)) { 591 val = bswap32(val); 592 } 593 virtio_config_writel(vdev, addr, val); 594 break; 595 } 596 } 597 598 static const MemoryRegionOps virtio_pci_config_ops = { 599 .read = virtio_pci_config_read, 600 .write = virtio_pci_config_write, 601 .impl = { 602 .min_access_size = 1, 603 .max_access_size = 4, 604 }, 605 .endianness = DEVICE_LITTLE_ENDIAN, 606 }; 607 608 static MemoryRegion *virtio_address_space_lookup(VirtIOPCIProxy *proxy, 609 hwaddr *off, int len) 610 { 611 int i; 612 VirtIOPCIRegion *reg; 613 614 for (i = 0; i < ARRAY_SIZE(proxy->regs); ++i) { 615 reg = &proxy->regs[i]; 616 if (*off >= reg->offset && 617 *off + len <= reg->offset + reg->size) { 618 *off -= reg->offset; 619 return ®->mr; 620 } 621 } 622 623 return NULL; 624 } 625 626 /* Below are generic functions to do memcpy from/to an address space, 627 * without byteswaps, with input validation. 628 * 629 * As regular address_space_* APIs all do some kind of byteswap at least for 630 * some host/target combinations, we are forced to explicitly convert to a 631 * known-endianness integer value. 632 * It doesn't really matter which endian format to go through, so the code 633 * below selects the endian that causes the least amount of work on the given 634 * host. 635 * 636 * Note: host pointer must be aligned. 637 */ 638 static 639 void virtio_address_space_write(VirtIOPCIProxy *proxy, hwaddr addr, 640 const uint8_t *buf, int len) 641 { 642 uint64_t val; 643 MemoryRegion *mr; 644 645 /* address_space_* APIs assume an aligned address. 646 * As address is under guest control, handle illegal values. 647 */ 648 addr &= ~(len - 1); 649 650 mr = virtio_address_space_lookup(proxy, &addr, len); 651 if (!mr) { 652 return; 653 } 654 655 /* Make sure caller aligned buf properly */ 656 assert(!(((uintptr_t)buf) & (len - 1))); 657 658 switch (len) { 659 case 1: 660 val = pci_get_byte(buf); 661 break; 662 case 2: 663 val = pci_get_word(buf); 664 break; 665 case 4: 666 val = pci_get_long(buf); 667 break; 668 default: 669 /* As length is under guest control, handle illegal values. */ 670 return; 671 } 672 memory_region_dispatch_write(mr, addr, val, size_memop(len) | MO_LE, 673 MEMTXATTRS_UNSPECIFIED); 674 } 675 676 static void 677 virtio_address_space_read(VirtIOPCIProxy *proxy, hwaddr addr, 678 uint8_t *buf, int len) 679 { 680 uint64_t val; 681 MemoryRegion *mr; 682 683 /* address_space_* APIs assume an aligned address. 684 * As address is under guest control, handle illegal values. 685 */ 686 addr &= ~(len - 1); 687 688 mr = virtio_address_space_lookup(proxy, &addr, len); 689 if (!mr) { 690 return; 691 } 692 693 /* Make sure caller aligned buf properly */ 694 assert(!(((uintptr_t)buf) & (len - 1))); 695 696 memory_region_dispatch_read(mr, addr, &val, size_memop(len) | MO_LE, 697 MEMTXATTRS_UNSPECIFIED); 698 switch (len) { 699 case 1: 700 pci_set_byte(buf, val); 701 break; 702 case 2: 703 pci_set_word(buf, val); 704 break; 705 case 4: 706 pci_set_long(buf, val); 707 break; 708 default: 709 /* As length is under guest control, handle illegal values. */ 710 break; 711 } 712 } 713 714 static void virtio_pci_ats_ctrl_trigger(PCIDevice *pci_dev, bool enable) 715 { 716 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); 717 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 718 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 719 720 vdev->device_iotlb_enabled = enable; 721 722 if (k->toggle_device_iotlb) { 723 k->toggle_device_iotlb(vdev); 724 } 725 } 726 727 static void pcie_ats_config_write(PCIDevice *dev, uint32_t address, 728 uint32_t val, int len) 729 { 730 uint32_t off; 731 uint16_t ats_cap = dev->exp.ats_cap; 732 733 if (!ats_cap || address < ats_cap) { 734 return; 735 } 736 off = address - ats_cap; 737 if (off >= PCI_EXT_CAP_ATS_SIZEOF) { 738 return; 739 } 740 741 if (range_covers_byte(off, len, PCI_ATS_CTRL + 1)) { 742 virtio_pci_ats_ctrl_trigger(dev, !!(val & PCI_ATS_CTRL_ENABLE)); 743 } 744 } 745 746 static void virtio_write_config(PCIDevice *pci_dev, uint32_t address, 747 uint32_t val, int len) 748 { 749 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); 750 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 751 struct virtio_pci_cfg_cap *cfg; 752 753 pci_default_write_config(pci_dev, address, val, len); 754 755 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_FLR) { 756 pcie_cap_flr_write_config(pci_dev, address, val, len); 757 } 758 759 if (proxy->flags & VIRTIO_PCI_FLAG_ATS) { 760 pcie_ats_config_write(pci_dev, address, val, len); 761 } 762 763 if (range_covers_byte(address, len, PCI_COMMAND)) { 764 if (!(pci_dev->config[PCI_COMMAND] & PCI_COMMAND_MASTER)) { 765 virtio_set_disabled(vdev, true); 766 virtio_pci_stop_ioeventfd(proxy); 767 virtio_set_status(vdev, vdev->status & ~VIRTIO_CONFIG_S_DRIVER_OK); 768 } else { 769 virtio_set_disabled(vdev, false); 770 } 771 } 772 773 if (proxy->config_cap && 774 ranges_overlap(address, len, proxy->config_cap + offsetof(struct virtio_pci_cfg_cap, 775 pci_cfg_data), 776 sizeof cfg->pci_cfg_data)) { 777 uint32_t off; 778 uint32_t caplen; 779 780 cfg = (void *)(proxy->pci_dev.config + proxy->config_cap); 781 off = le32_to_cpu(cfg->cap.offset); 782 caplen = le32_to_cpu(cfg->cap.length); 783 784 if (caplen == 1 || caplen == 2 || caplen == 4) { 785 assert(caplen <= sizeof cfg->pci_cfg_data); 786 virtio_address_space_write(proxy, off, cfg->pci_cfg_data, caplen); 787 } 788 } 789 } 790 791 static uint32_t virtio_read_config(PCIDevice *pci_dev, 792 uint32_t address, int len) 793 { 794 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); 795 struct virtio_pci_cfg_cap *cfg; 796 797 if (proxy->config_cap && 798 ranges_overlap(address, len, proxy->config_cap + offsetof(struct virtio_pci_cfg_cap, 799 pci_cfg_data), 800 sizeof cfg->pci_cfg_data)) { 801 uint32_t off; 802 uint32_t caplen; 803 804 cfg = (void *)(proxy->pci_dev.config + proxy->config_cap); 805 off = le32_to_cpu(cfg->cap.offset); 806 caplen = le32_to_cpu(cfg->cap.length); 807 808 if (caplen == 1 || caplen == 2 || caplen == 4) { 809 assert(caplen <= sizeof cfg->pci_cfg_data); 810 virtio_address_space_read(proxy, off, cfg->pci_cfg_data, caplen); 811 } 812 } 813 814 return pci_default_read_config(pci_dev, address, len); 815 } 816 817 static int kvm_virtio_pci_vq_vector_use(VirtIOPCIProxy *proxy, 818 unsigned int vector) 819 { 820 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector]; 821 int ret; 822 823 if (irqfd->users == 0) { 824 KVMRouteChange c = kvm_irqchip_begin_route_changes(kvm_state); 825 ret = kvm_irqchip_add_msi_route(&c, vector, &proxy->pci_dev); 826 if (ret < 0) { 827 return ret; 828 } 829 kvm_irqchip_commit_route_changes(&c); 830 irqfd->virq = ret; 831 } 832 irqfd->users++; 833 return 0; 834 } 835 836 static void kvm_virtio_pci_vq_vector_release(VirtIOPCIProxy *proxy, 837 unsigned int vector) 838 { 839 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector]; 840 if (--irqfd->users == 0) { 841 kvm_irqchip_release_virq(kvm_state, irqfd->virq); 842 } 843 } 844 845 static int kvm_virtio_pci_irqfd_use(VirtIOPCIProxy *proxy, 846 EventNotifier *n, 847 unsigned int vector) 848 { 849 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector]; 850 return kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, n, NULL, irqfd->virq); 851 } 852 853 static void kvm_virtio_pci_irqfd_release(VirtIOPCIProxy *proxy, 854 EventNotifier *n , 855 unsigned int vector) 856 { 857 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector]; 858 int ret; 859 860 ret = kvm_irqchip_remove_irqfd_notifier_gsi(kvm_state, n, irqfd->virq); 861 assert(ret == 0); 862 } 863 static int virtio_pci_get_notifier(VirtIOPCIProxy *proxy, int queue_no, 864 EventNotifier **n, unsigned int *vector) 865 { 866 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 867 VirtQueue *vq; 868 869 if (queue_no == VIRTIO_CONFIG_IRQ_IDX) { 870 *n = virtio_config_get_guest_notifier(vdev); 871 *vector = vdev->config_vector; 872 } else { 873 if (!virtio_queue_get_num(vdev, queue_no)) { 874 return -1; 875 } 876 *vector = virtio_queue_vector(vdev, queue_no); 877 vq = virtio_get_queue(vdev, queue_no); 878 *n = virtio_queue_get_guest_notifier(vq); 879 } 880 return 0; 881 } 882 883 static int kvm_virtio_pci_vector_use_one(VirtIOPCIProxy *proxy, int queue_no) 884 { 885 unsigned int vector; 886 int ret; 887 EventNotifier *n; 888 PCIDevice *dev = &proxy->pci_dev; 889 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 890 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 891 892 ret = virtio_pci_get_notifier(proxy, queue_no, &n, &vector); 893 if (ret < 0) { 894 return ret; 895 } 896 if (vector >= msix_nr_vectors_allocated(dev)) { 897 return 0; 898 } 899 ret = kvm_virtio_pci_vq_vector_use(proxy, vector); 900 if (ret < 0) { 901 goto undo; 902 } 903 /* 904 * If guest supports masking, set up irqfd now. 905 * Otherwise, delay until unmasked in the frontend. 906 */ 907 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) { 908 ret = kvm_virtio_pci_irqfd_use(proxy, n, vector); 909 if (ret < 0) { 910 kvm_virtio_pci_vq_vector_release(proxy, vector); 911 goto undo; 912 } 913 } 914 915 return 0; 916 undo: 917 918 vector = virtio_queue_vector(vdev, queue_no); 919 if (vector >= msix_nr_vectors_allocated(dev)) { 920 return ret; 921 } 922 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) { 923 ret = virtio_pci_get_notifier(proxy, queue_no, &n, &vector); 924 if (ret < 0) { 925 return ret; 926 } 927 kvm_virtio_pci_irqfd_release(proxy, n, vector); 928 } 929 return ret; 930 } 931 static int kvm_virtio_pci_vector_vq_use(VirtIOPCIProxy *proxy, int nvqs) 932 { 933 int queue_no; 934 int ret = 0; 935 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 936 937 for (queue_no = 0; queue_no < nvqs; queue_no++) { 938 if (!virtio_queue_get_num(vdev, queue_no)) { 939 return -1; 940 } 941 ret = kvm_virtio_pci_vector_use_one(proxy, queue_no); 942 } 943 return ret; 944 } 945 946 static int kvm_virtio_pci_vector_config_use(VirtIOPCIProxy *proxy) 947 { 948 return kvm_virtio_pci_vector_use_one(proxy, VIRTIO_CONFIG_IRQ_IDX); 949 } 950 951 static void kvm_virtio_pci_vector_release_one(VirtIOPCIProxy *proxy, 952 int queue_no) 953 { 954 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 955 unsigned int vector; 956 EventNotifier *n; 957 int ret; 958 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 959 PCIDevice *dev = &proxy->pci_dev; 960 961 ret = virtio_pci_get_notifier(proxy, queue_no, &n, &vector); 962 if (ret < 0) { 963 return; 964 } 965 if (vector >= msix_nr_vectors_allocated(dev)) { 966 return; 967 } 968 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) { 969 kvm_virtio_pci_irqfd_release(proxy, n, vector); 970 } 971 kvm_virtio_pci_vq_vector_release(proxy, vector); 972 } 973 974 static void kvm_virtio_pci_vector_vq_release(VirtIOPCIProxy *proxy, int nvqs) 975 { 976 int queue_no; 977 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 978 979 for (queue_no = 0; queue_no < nvqs; queue_no++) { 980 if (!virtio_queue_get_num(vdev, queue_no)) { 981 break; 982 } 983 kvm_virtio_pci_vector_release_one(proxy, queue_no); 984 } 985 } 986 987 static void kvm_virtio_pci_vector_config_release(VirtIOPCIProxy *proxy) 988 { 989 kvm_virtio_pci_vector_release_one(proxy, VIRTIO_CONFIG_IRQ_IDX); 990 } 991 992 static int virtio_pci_one_vector_unmask(VirtIOPCIProxy *proxy, 993 unsigned int queue_no, 994 unsigned int vector, 995 MSIMessage msg, 996 EventNotifier *n) 997 { 998 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 999 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1000 VirtIOIRQFD *irqfd; 1001 int ret = 0; 1002 1003 if (proxy->vector_irqfd) { 1004 irqfd = &proxy->vector_irqfd[vector]; 1005 if (irqfd->msg.data != msg.data || irqfd->msg.address != msg.address) { 1006 ret = kvm_irqchip_update_msi_route(kvm_state, irqfd->virq, msg, 1007 &proxy->pci_dev); 1008 if (ret < 0) { 1009 return ret; 1010 } 1011 kvm_irqchip_commit_routes(kvm_state); 1012 } 1013 } 1014 1015 /* If guest supports masking, irqfd is already setup, unmask it. 1016 * Otherwise, set it up now. 1017 */ 1018 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) { 1019 k->guest_notifier_mask(vdev, queue_no, false); 1020 /* Test after unmasking to avoid losing events. */ 1021 if (k->guest_notifier_pending && 1022 k->guest_notifier_pending(vdev, queue_no)) { 1023 event_notifier_set(n); 1024 } 1025 } else { 1026 ret = kvm_virtio_pci_irqfd_use(proxy, n, vector); 1027 } 1028 return ret; 1029 } 1030 1031 static void virtio_pci_one_vector_mask(VirtIOPCIProxy *proxy, 1032 unsigned int queue_no, 1033 unsigned int vector, 1034 EventNotifier *n) 1035 { 1036 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1037 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1038 1039 /* If guest supports masking, keep irqfd but mask it. 1040 * Otherwise, clean it up now. 1041 */ 1042 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) { 1043 k->guest_notifier_mask(vdev, queue_no, true); 1044 } else { 1045 kvm_virtio_pci_irqfd_release(proxy, n, vector); 1046 } 1047 } 1048 1049 static int virtio_pci_vector_unmask(PCIDevice *dev, unsigned vector, 1050 MSIMessage msg) 1051 { 1052 VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev); 1053 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1054 VirtQueue *vq = virtio_vector_first_queue(vdev, vector); 1055 EventNotifier *n; 1056 int ret, index, unmasked = 0; 1057 1058 while (vq) { 1059 index = virtio_get_queue_index(vq); 1060 if (!virtio_queue_get_num(vdev, index)) { 1061 break; 1062 } 1063 if (index < proxy->nvqs_with_notifiers) { 1064 n = virtio_queue_get_guest_notifier(vq); 1065 ret = virtio_pci_one_vector_unmask(proxy, index, vector, msg, n); 1066 if (ret < 0) { 1067 goto undo; 1068 } 1069 ++unmasked; 1070 } 1071 vq = virtio_vector_next_queue(vq); 1072 } 1073 /* unmask config intr */ 1074 if (vector == vdev->config_vector) { 1075 n = virtio_config_get_guest_notifier(vdev); 1076 ret = virtio_pci_one_vector_unmask(proxy, VIRTIO_CONFIG_IRQ_IDX, vector, 1077 msg, n); 1078 if (ret < 0) { 1079 goto undo_config; 1080 } 1081 } 1082 return 0; 1083 undo_config: 1084 n = virtio_config_get_guest_notifier(vdev); 1085 virtio_pci_one_vector_mask(proxy, VIRTIO_CONFIG_IRQ_IDX, vector, n); 1086 undo: 1087 vq = virtio_vector_first_queue(vdev, vector); 1088 while (vq && unmasked >= 0) { 1089 index = virtio_get_queue_index(vq); 1090 if (index < proxy->nvqs_with_notifiers) { 1091 n = virtio_queue_get_guest_notifier(vq); 1092 virtio_pci_one_vector_mask(proxy, index, vector, n); 1093 --unmasked; 1094 } 1095 vq = virtio_vector_next_queue(vq); 1096 } 1097 return ret; 1098 } 1099 1100 static void virtio_pci_vector_mask(PCIDevice *dev, unsigned vector) 1101 { 1102 VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev); 1103 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1104 VirtQueue *vq = virtio_vector_first_queue(vdev, vector); 1105 EventNotifier *n; 1106 int index; 1107 1108 while (vq) { 1109 index = virtio_get_queue_index(vq); 1110 n = virtio_queue_get_guest_notifier(vq); 1111 if (!virtio_queue_get_num(vdev, index)) { 1112 break; 1113 } 1114 if (index < proxy->nvqs_with_notifiers) { 1115 virtio_pci_one_vector_mask(proxy, index, vector, n); 1116 } 1117 vq = virtio_vector_next_queue(vq); 1118 } 1119 1120 if (vector == vdev->config_vector) { 1121 n = virtio_config_get_guest_notifier(vdev); 1122 virtio_pci_one_vector_mask(proxy, VIRTIO_CONFIG_IRQ_IDX, vector, n); 1123 } 1124 } 1125 1126 static void virtio_pci_vector_poll(PCIDevice *dev, 1127 unsigned int vector_start, 1128 unsigned int vector_end) 1129 { 1130 VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev); 1131 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1132 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1133 int queue_no; 1134 unsigned int vector; 1135 EventNotifier *notifier; 1136 int ret; 1137 1138 for (queue_no = 0; queue_no < proxy->nvqs_with_notifiers; queue_no++) { 1139 ret = virtio_pci_get_notifier(proxy, queue_no, ¬ifier, &vector); 1140 if (ret < 0) { 1141 break; 1142 } 1143 if (vector < vector_start || vector >= vector_end || 1144 !msix_is_masked(dev, vector)) { 1145 continue; 1146 } 1147 if (k->guest_notifier_pending) { 1148 if (k->guest_notifier_pending(vdev, queue_no)) { 1149 msix_set_pending(dev, vector); 1150 } 1151 } else if (event_notifier_test_and_clear(notifier)) { 1152 msix_set_pending(dev, vector); 1153 } 1154 } 1155 /* poll the config intr */ 1156 ret = virtio_pci_get_notifier(proxy, VIRTIO_CONFIG_IRQ_IDX, ¬ifier, 1157 &vector); 1158 if (ret < 0) { 1159 return; 1160 } 1161 if (vector < vector_start || vector >= vector_end || 1162 !msix_is_masked(dev, vector)) { 1163 return; 1164 } 1165 if (k->guest_notifier_pending) { 1166 if (k->guest_notifier_pending(vdev, VIRTIO_CONFIG_IRQ_IDX)) { 1167 msix_set_pending(dev, vector); 1168 } 1169 } else if (event_notifier_test_and_clear(notifier)) { 1170 msix_set_pending(dev, vector); 1171 } 1172 } 1173 1174 void virtio_pci_set_guest_notifier_fd_handler(VirtIODevice *vdev, VirtQueue *vq, 1175 int n, bool assign, 1176 bool with_irqfd) 1177 { 1178 if (n == VIRTIO_CONFIG_IRQ_IDX) { 1179 virtio_config_set_guest_notifier_fd_handler(vdev, assign, with_irqfd); 1180 } else { 1181 virtio_queue_set_guest_notifier_fd_handler(vq, assign, with_irqfd); 1182 } 1183 } 1184 1185 static int virtio_pci_set_guest_notifier(DeviceState *d, int n, bool assign, 1186 bool with_irqfd) 1187 { 1188 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 1189 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1190 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev); 1191 VirtQueue *vq = NULL; 1192 EventNotifier *notifier = NULL; 1193 1194 if (n == VIRTIO_CONFIG_IRQ_IDX) { 1195 notifier = virtio_config_get_guest_notifier(vdev); 1196 } else { 1197 vq = virtio_get_queue(vdev, n); 1198 notifier = virtio_queue_get_guest_notifier(vq); 1199 } 1200 1201 if (assign) { 1202 int r = event_notifier_init(notifier, 0); 1203 if (r < 0) { 1204 return r; 1205 } 1206 virtio_pci_set_guest_notifier_fd_handler(vdev, vq, n, true, with_irqfd); 1207 } else { 1208 virtio_pci_set_guest_notifier_fd_handler(vdev, vq, n, false, 1209 with_irqfd); 1210 event_notifier_cleanup(notifier); 1211 } 1212 1213 if (!msix_enabled(&proxy->pci_dev) && 1214 vdev->use_guest_notifier_mask && 1215 vdc->guest_notifier_mask) { 1216 vdc->guest_notifier_mask(vdev, n, !assign); 1217 } 1218 1219 return 0; 1220 } 1221 1222 static bool virtio_pci_query_guest_notifiers(DeviceState *d) 1223 { 1224 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 1225 return msix_enabled(&proxy->pci_dev); 1226 } 1227 1228 static int virtio_pci_set_guest_notifiers(DeviceState *d, int nvqs, bool assign) 1229 { 1230 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 1231 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1232 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 1233 int r, n; 1234 bool with_irqfd = msix_enabled(&proxy->pci_dev) && 1235 kvm_msi_via_irqfd_enabled(); 1236 1237 nvqs = MIN(nvqs, VIRTIO_QUEUE_MAX); 1238 1239 /* 1240 * When deassigning, pass a consistent nvqs value to avoid leaking 1241 * notifiers. But first check we've actually been configured, exit 1242 * early if we haven't. 1243 */ 1244 if (!assign && !proxy->nvqs_with_notifiers) { 1245 return 0; 1246 } 1247 assert(assign || nvqs == proxy->nvqs_with_notifiers); 1248 1249 proxy->nvqs_with_notifiers = nvqs; 1250 1251 /* Must unset vector notifier while guest notifier is still assigned */ 1252 if ((proxy->vector_irqfd || 1253 (vdev->use_guest_notifier_mask && k->guest_notifier_mask)) && 1254 !assign) { 1255 msix_unset_vector_notifiers(&proxy->pci_dev); 1256 if (proxy->vector_irqfd) { 1257 kvm_virtio_pci_vector_vq_release(proxy, nvqs); 1258 kvm_virtio_pci_vector_config_release(proxy); 1259 g_free(proxy->vector_irqfd); 1260 proxy->vector_irqfd = NULL; 1261 } 1262 } 1263 1264 for (n = 0; n < nvqs; n++) { 1265 if (!virtio_queue_get_num(vdev, n)) { 1266 break; 1267 } 1268 1269 r = virtio_pci_set_guest_notifier(d, n, assign, with_irqfd); 1270 if (r < 0) { 1271 goto assign_error; 1272 } 1273 } 1274 r = virtio_pci_set_guest_notifier(d, VIRTIO_CONFIG_IRQ_IDX, assign, 1275 with_irqfd); 1276 if (r < 0) { 1277 goto config_assign_error; 1278 } 1279 /* Must set vector notifier after guest notifier has been assigned */ 1280 if ((with_irqfd || 1281 (vdev->use_guest_notifier_mask && k->guest_notifier_mask)) && 1282 assign) { 1283 if (with_irqfd) { 1284 proxy->vector_irqfd = 1285 g_malloc0(sizeof(*proxy->vector_irqfd) * 1286 msix_nr_vectors_allocated(&proxy->pci_dev)); 1287 r = kvm_virtio_pci_vector_vq_use(proxy, nvqs); 1288 if (r < 0) { 1289 goto config_assign_error; 1290 } 1291 r = kvm_virtio_pci_vector_config_use(proxy); 1292 if (r < 0) { 1293 goto config_error; 1294 } 1295 } 1296 1297 r = msix_set_vector_notifiers(&proxy->pci_dev, virtio_pci_vector_unmask, 1298 virtio_pci_vector_mask, 1299 virtio_pci_vector_poll); 1300 if (r < 0) { 1301 goto notifiers_error; 1302 } 1303 } 1304 1305 return 0; 1306 1307 notifiers_error: 1308 if (with_irqfd) { 1309 assert(assign); 1310 kvm_virtio_pci_vector_vq_release(proxy, nvqs); 1311 } 1312 config_error: 1313 if (with_irqfd) { 1314 kvm_virtio_pci_vector_config_release(proxy); 1315 } 1316 config_assign_error: 1317 virtio_pci_set_guest_notifier(d, VIRTIO_CONFIG_IRQ_IDX, !assign, 1318 with_irqfd); 1319 assign_error: 1320 /* We get here on assignment failure. Recover by undoing for VQs 0 .. n. */ 1321 assert(assign); 1322 while (--n >= 0) { 1323 virtio_pci_set_guest_notifier(d, n, !assign, with_irqfd); 1324 } 1325 g_free(proxy->vector_irqfd); 1326 proxy->vector_irqfd = NULL; 1327 return r; 1328 } 1329 1330 static int virtio_pci_set_host_notifier_mr(DeviceState *d, int n, 1331 MemoryRegion *mr, bool assign) 1332 { 1333 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 1334 int offset; 1335 1336 if (n >= VIRTIO_QUEUE_MAX || !virtio_pci_modern(proxy) || 1337 virtio_pci_queue_mem_mult(proxy) != memory_region_size(mr)) { 1338 return -1; 1339 } 1340 1341 if (assign) { 1342 offset = virtio_pci_queue_mem_mult(proxy) * n; 1343 memory_region_add_subregion_overlap(&proxy->notify.mr, offset, mr, 1); 1344 } else { 1345 memory_region_del_subregion(&proxy->notify.mr, mr); 1346 } 1347 1348 return 0; 1349 } 1350 1351 static void virtio_pci_vmstate_change(DeviceState *d, bool running) 1352 { 1353 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 1354 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1355 1356 if (running) { 1357 /* Old QEMU versions did not set bus master enable on status write. 1358 * Detect DRIVER set and enable it. 1359 */ 1360 if ((proxy->flags & VIRTIO_PCI_FLAG_BUS_MASTER_BUG_MIGRATION) && 1361 (vdev->status & VIRTIO_CONFIG_S_DRIVER) && 1362 !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) { 1363 pci_default_write_config(&proxy->pci_dev, PCI_COMMAND, 1364 proxy->pci_dev.config[PCI_COMMAND] | 1365 PCI_COMMAND_MASTER, 1); 1366 } 1367 virtio_pci_start_ioeventfd(proxy); 1368 } else { 1369 virtio_pci_stop_ioeventfd(proxy); 1370 } 1371 } 1372 1373 /* 1374 * virtio-pci: This is the PCIDevice which has a virtio-pci-bus. 1375 */ 1376 1377 static int virtio_pci_query_nvectors(DeviceState *d) 1378 { 1379 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1380 1381 return proxy->nvectors; 1382 } 1383 1384 static AddressSpace *virtio_pci_get_dma_as(DeviceState *d) 1385 { 1386 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1387 PCIDevice *dev = &proxy->pci_dev; 1388 1389 return pci_get_address_space(dev); 1390 } 1391 1392 static bool virtio_pci_iommu_enabled(DeviceState *d) 1393 { 1394 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1395 PCIDevice *dev = &proxy->pci_dev; 1396 AddressSpace *dma_as = pci_device_iommu_address_space(dev); 1397 1398 if (dma_as == &address_space_memory) { 1399 return false; 1400 } 1401 1402 return true; 1403 } 1404 1405 static bool virtio_pci_queue_enabled(DeviceState *d, int n) 1406 { 1407 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1408 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1409 1410 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 1411 return proxy->vqs[n].enabled; 1412 } 1413 1414 return virtio_queue_enabled_legacy(vdev, n); 1415 } 1416 1417 static int virtio_pci_add_mem_cap(VirtIOPCIProxy *proxy, 1418 struct virtio_pci_cap *cap) 1419 { 1420 PCIDevice *dev = &proxy->pci_dev; 1421 int offset; 1422 1423 offset = pci_add_capability(dev, PCI_CAP_ID_VNDR, 0, 1424 cap->cap_len, &error_abort); 1425 1426 assert(cap->cap_len >= sizeof *cap); 1427 memcpy(dev->config + offset + PCI_CAP_FLAGS, &cap->cap_len, 1428 cap->cap_len - PCI_CAP_FLAGS); 1429 1430 return offset; 1431 } 1432 1433 static void virtio_pci_set_vector(VirtIODevice *vdev, 1434 VirtIOPCIProxy *proxy, 1435 int queue_no, uint16_t old_vector, 1436 uint16_t new_vector) 1437 { 1438 bool kvm_irqfd = (vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) && 1439 msix_enabled(&proxy->pci_dev) && kvm_msi_via_irqfd_enabled(); 1440 1441 if (new_vector == old_vector) { 1442 return; 1443 } 1444 1445 /* 1446 * If the device uses irqfd and the vector changes after DRIVER_OK is 1447 * set, we need to release the old vector and set up the new one. 1448 * Otherwise just need to set the new vector on the device. 1449 */ 1450 if (kvm_irqfd && old_vector != VIRTIO_NO_VECTOR) { 1451 kvm_virtio_pci_vector_release_one(proxy, queue_no); 1452 } 1453 /* Set the new vector on the device. */ 1454 if (queue_no == VIRTIO_CONFIG_IRQ_IDX) { 1455 vdev->config_vector = new_vector; 1456 } else { 1457 virtio_queue_set_vector(vdev, queue_no, new_vector); 1458 } 1459 /* If the new vector changed need to set it up. */ 1460 if (kvm_irqfd && new_vector != VIRTIO_NO_VECTOR) { 1461 kvm_virtio_pci_vector_use_one(proxy, queue_no); 1462 } 1463 } 1464 1465 int virtio_pci_add_shm_cap(VirtIOPCIProxy *proxy, 1466 uint8_t bar, uint64_t offset, uint64_t length, 1467 uint8_t id) 1468 { 1469 struct virtio_pci_cap64 cap = { 1470 .cap.cap_len = sizeof cap, 1471 .cap.cfg_type = VIRTIO_PCI_CAP_SHARED_MEMORY_CFG, 1472 }; 1473 1474 cap.cap.bar = bar; 1475 cap.cap.length = cpu_to_le32(length); 1476 cap.length_hi = cpu_to_le32(length >> 32); 1477 cap.cap.offset = cpu_to_le32(offset); 1478 cap.offset_hi = cpu_to_le32(offset >> 32); 1479 cap.cap.id = id; 1480 return virtio_pci_add_mem_cap(proxy, &cap.cap); 1481 } 1482 1483 static uint64_t virtio_pci_common_read(void *opaque, hwaddr addr, 1484 unsigned size) 1485 { 1486 VirtIOPCIProxy *proxy = opaque; 1487 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1488 uint32_t val = 0; 1489 int i; 1490 1491 if (vdev == NULL) { 1492 return UINT64_MAX; 1493 } 1494 1495 switch (addr) { 1496 case VIRTIO_PCI_COMMON_DFSELECT: 1497 val = proxy->dfselect; 1498 break; 1499 case VIRTIO_PCI_COMMON_DF: 1500 if (proxy->dfselect <= 1) { 1501 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev); 1502 1503 val = (vdev->host_features & ~vdc->legacy_features) >> 1504 (32 * proxy->dfselect); 1505 } 1506 break; 1507 case VIRTIO_PCI_COMMON_GFSELECT: 1508 val = proxy->gfselect; 1509 break; 1510 case VIRTIO_PCI_COMMON_GF: 1511 if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) { 1512 val = proxy->guest_features[proxy->gfselect]; 1513 } 1514 break; 1515 case VIRTIO_PCI_COMMON_MSIX: 1516 val = vdev->config_vector; 1517 break; 1518 case VIRTIO_PCI_COMMON_NUMQ: 1519 for (i = 0; i < VIRTIO_QUEUE_MAX; ++i) { 1520 if (virtio_queue_get_num(vdev, i)) { 1521 val = i + 1; 1522 } 1523 } 1524 break; 1525 case VIRTIO_PCI_COMMON_STATUS: 1526 val = vdev->status; 1527 break; 1528 case VIRTIO_PCI_COMMON_CFGGENERATION: 1529 val = vdev->generation; 1530 break; 1531 case VIRTIO_PCI_COMMON_Q_SELECT: 1532 val = vdev->queue_sel; 1533 break; 1534 case VIRTIO_PCI_COMMON_Q_SIZE: 1535 val = virtio_queue_get_num(vdev, vdev->queue_sel); 1536 break; 1537 case VIRTIO_PCI_COMMON_Q_MSIX: 1538 val = virtio_queue_vector(vdev, vdev->queue_sel); 1539 break; 1540 case VIRTIO_PCI_COMMON_Q_ENABLE: 1541 val = proxy->vqs[vdev->queue_sel].enabled; 1542 break; 1543 case VIRTIO_PCI_COMMON_Q_NOFF: 1544 /* Simply map queues in order */ 1545 val = vdev->queue_sel; 1546 break; 1547 case VIRTIO_PCI_COMMON_Q_DESCLO: 1548 val = proxy->vqs[vdev->queue_sel].desc[0]; 1549 break; 1550 case VIRTIO_PCI_COMMON_Q_DESCHI: 1551 val = proxy->vqs[vdev->queue_sel].desc[1]; 1552 break; 1553 case VIRTIO_PCI_COMMON_Q_AVAILLO: 1554 val = proxy->vqs[vdev->queue_sel].avail[0]; 1555 break; 1556 case VIRTIO_PCI_COMMON_Q_AVAILHI: 1557 val = proxy->vqs[vdev->queue_sel].avail[1]; 1558 break; 1559 case VIRTIO_PCI_COMMON_Q_USEDLO: 1560 val = proxy->vqs[vdev->queue_sel].used[0]; 1561 break; 1562 case VIRTIO_PCI_COMMON_Q_USEDHI: 1563 val = proxy->vqs[vdev->queue_sel].used[1]; 1564 break; 1565 case VIRTIO_PCI_COMMON_Q_RESET: 1566 val = proxy->vqs[vdev->queue_sel].reset; 1567 break; 1568 default: 1569 val = 0; 1570 } 1571 1572 return val; 1573 } 1574 1575 static void virtio_pci_common_write(void *opaque, hwaddr addr, 1576 uint64_t val, unsigned size) 1577 { 1578 VirtIOPCIProxy *proxy = opaque; 1579 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1580 uint16_t vector; 1581 1582 if (vdev == NULL) { 1583 return; 1584 } 1585 1586 switch (addr) { 1587 case VIRTIO_PCI_COMMON_DFSELECT: 1588 proxy->dfselect = val; 1589 break; 1590 case VIRTIO_PCI_COMMON_GFSELECT: 1591 proxy->gfselect = val; 1592 break; 1593 case VIRTIO_PCI_COMMON_GF: 1594 if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) { 1595 proxy->guest_features[proxy->gfselect] = val; 1596 virtio_set_features(vdev, 1597 (((uint64_t)proxy->guest_features[1]) << 32) | 1598 proxy->guest_features[0]); 1599 } 1600 break; 1601 case VIRTIO_PCI_COMMON_MSIX: 1602 if (vdev->config_vector != VIRTIO_NO_VECTOR) { 1603 msix_vector_unuse(&proxy->pci_dev, vdev->config_vector); 1604 } 1605 /* Make it possible for guest to discover an error took place. */ 1606 if (val < proxy->nvectors) { 1607 msix_vector_use(&proxy->pci_dev, val); 1608 } else { 1609 val = VIRTIO_NO_VECTOR; 1610 } 1611 virtio_pci_set_vector(vdev, proxy, VIRTIO_CONFIG_IRQ_IDX, 1612 vdev->config_vector, val); 1613 break; 1614 case VIRTIO_PCI_COMMON_STATUS: 1615 if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) { 1616 virtio_pci_stop_ioeventfd(proxy); 1617 } 1618 1619 virtio_set_status(vdev, val & 0xFF); 1620 1621 if (val & VIRTIO_CONFIG_S_DRIVER_OK) { 1622 virtio_pci_start_ioeventfd(proxy); 1623 } 1624 1625 if (vdev->status == 0) { 1626 virtio_pci_reset(DEVICE(proxy)); 1627 } 1628 1629 break; 1630 case VIRTIO_PCI_COMMON_Q_SELECT: 1631 if (val < VIRTIO_QUEUE_MAX) { 1632 vdev->queue_sel = val; 1633 } 1634 break; 1635 case VIRTIO_PCI_COMMON_Q_SIZE: 1636 proxy->vqs[vdev->queue_sel].num = val; 1637 virtio_queue_set_num(vdev, vdev->queue_sel, 1638 proxy->vqs[vdev->queue_sel].num); 1639 virtio_init_region_cache(vdev, vdev->queue_sel); 1640 break; 1641 case VIRTIO_PCI_COMMON_Q_MSIX: 1642 vector = virtio_queue_vector(vdev, vdev->queue_sel); 1643 if (vector != VIRTIO_NO_VECTOR) { 1644 msix_vector_unuse(&proxy->pci_dev, vector); 1645 } 1646 /* Make it possible for guest to discover an error took place. */ 1647 if (val < proxy->nvectors) { 1648 msix_vector_use(&proxy->pci_dev, val); 1649 } else { 1650 val = VIRTIO_NO_VECTOR; 1651 } 1652 virtio_pci_set_vector(vdev, proxy, vdev->queue_sel, vector, val); 1653 break; 1654 case VIRTIO_PCI_COMMON_Q_ENABLE: 1655 if (val == 1) { 1656 virtio_queue_set_num(vdev, vdev->queue_sel, 1657 proxy->vqs[vdev->queue_sel].num); 1658 virtio_queue_set_rings(vdev, vdev->queue_sel, 1659 ((uint64_t)proxy->vqs[vdev->queue_sel].desc[1]) << 32 | 1660 proxy->vqs[vdev->queue_sel].desc[0], 1661 ((uint64_t)proxy->vqs[vdev->queue_sel].avail[1]) << 32 | 1662 proxy->vqs[vdev->queue_sel].avail[0], 1663 ((uint64_t)proxy->vqs[vdev->queue_sel].used[1]) << 32 | 1664 proxy->vqs[vdev->queue_sel].used[0]); 1665 proxy->vqs[vdev->queue_sel].enabled = 1; 1666 proxy->vqs[vdev->queue_sel].reset = 0; 1667 virtio_queue_enable(vdev, vdev->queue_sel); 1668 } else { 1669 virtio_error(vdev, "wrong value for queue_enable %"PRIx64, val); 1670 } 1671 break; 1672 case VIRTIO_PCI_COMMON_Q_DESCLO: 1673 proxy->vqs[vdev->queue_sel].desc[0] = val; 1674 break; 1675 case VIRTIO_PCI_COMMON_Q_DESCHI: 1676 proxy->vqs[vdev->queue_sel].desc[1] = val; 1677 break; 1678 case VIRTIO_PCI_COMMON_Q_AVAILLO: 1679 proxy->vqs[vdev->queue_sel].avail[0] = val; 1680 break; 1681 case VIRTIO_PCI_COMMON_Q_AVAILHI: 1682 proxy->vqs[vdev->queue_sel].avail[1] = val; 1683 break; 1684 case VIRTIO_PCI_COMMON_Q_USEDLO: 1685 proxy->vqs[vdev->queue_sel].used[0] = val; 1686 break; 1687 case VIRTIO_PCI_COMMON_Q_USEDHI: 1688 proxy->vqs[vdev->queue_sel].used[1] = val; 1689 break; 1690 case VIRTIO_PCI_COMMON_Q_RESET: 1691 if (val == 1) { 1692 proxy->vqs[vdev->queue_sel].reset = 1; 1693 1694 virtio_queue_reset(vdev, vdev->queue_sel); 1695 1696 proxy->vqs[vdev->queue_sel].reset = 0; 1697 proxy->vqs[vdev->queue_sel].enabled = 0; 1698 } 1699 break; 1700 default: 1701 break; 1702 } 1703 } 1704 1705 1706 static uint64_t virtio_pci_notify_read(void *opaque, hwaddr addr, 1707 unsigned size) 1708 { 1709 VirtIOPCIProxy *proxy = opaque; 1710 if (virtio_bus_get_device(&proxy->bus) == NULL) { 1711 return UINT64_MAX; 1712 } 1713 1714 return 0; 1715 } 1716 1717 static void virtio_pci_notify_write(void *opaque, hwaddr addr, 1718 uint64_t val, unsigned size) 1719 { 1720 VirtIOPCIProxy *proxy = opaque; 1721 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1722 1723 unsigned queue = addr / virtio_pci_queue_mem_mult(proxy); 1724 1725 if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) { 1726 trace_virtio_pci_notify_write(addr, val, size); 1727 virtio_queue_notify(vdev, queue); 1728 } 1729 } 1730 1731 static void virtio_pci_notify_write_pio(void *opaque, hwaddr addr, 1732 uint64_t val, unsigned size) 1733 { 1734 VirtIOPCIProxy *proxy = opaque; 1735 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1736 1737 unsigned queue = val; 1738 1739 if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) { 1740 trace_virtio_pci_notify_write_pio(addr, val, size); 1741 virtio_queue_notify(vdev, queue); 1742 } 1743 } 1744 1745 static uint64_t virtio_pci_isr_read(void *opaque, hwaddr addr, 1746 unsigned size) 1747 { 1748 VirtIOPCIProxy *proxy = opaque; 1749 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1750 uint64_t val; 1751 1752 if (vdev == NULL) { 1753 return UINT64_MAX; 1754 } 1755 1756 val = qatomic_xchg(&vdev->isr, 0); 1757 pci_irq_deassert(&proxy->pci_dev); 1758 return val; 1759 } 1760 1761 static void virtio_pci_isr_write(void *opaque, hwaddr addr, 1762 uint64_t val, unsigned size) 1763 { 1764 } 1765 1766 static uint64_t virtio_pci_device_read(void *opaque, hwaddr addr, 1767 unsigned size) 1768 { 1769 VirtIOPCIProxy *proxy = opaque; 1770 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1771 uint64_t val; 1772 1773 if (vdev == NULL) { 1774 return UINT64_MAX; 1775 } 1776 1777 switch (size) { 1778 case 1: 1779 val = virtio_config_modern_readb(vdev, addr); 1780 break; 1781 case 2: 1782 val = virtio_config_modern_readw(vdev, addr); 1783 break; 1784 case 4: 1785 val = virtio_config_modern_readl(vdev, addr); 1786 break; 1787 default: 1788 val = 0; 1789 break; 1790 } 1791 return val; 1792 } 1793 1794 static void virtio_pci_device_write(void *opaque, hwaddr addr, 1795 uint64_t val, unsigned size) 1796 { 1797 VirtIOPCIProxy *proxy = opaque; 1798 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1799 1800 if (vdev == NULL) { 1801 return; 1802 } 1803 1804 switch (size) { 1805 case 1: 1806 virtio_config_modern_writeb(vdev, addr, val); 1807 break; 1808 case 2: 1809 virtio_config_modern_writew(vdev, addr, val); 1810 break; 1811 case 4: 1812 virtio_config_modern_writel(vdev, addr, val); 1813 break; 1814 } 1815 } 1816 1817 static void virtio_pci_modern_regions_init(VirtIOPCIProxy *proxy, 1818 const char *vdev_name) 1819 { 1820 static const MemoryRegionOps common_ops = { 1821 .read = virtio_pci_common_read, 1822 .write = virtio_pci_common_write, 1823 .impl = { 1824 .min_access_size = 1, 1825 .max_access_size = 4, 1826 }, 1827 .endianness = DEVICE_LITTLE_ENDIAN, 1828 }; 1829 static const MemoryRegionOps isr_ops = { 1830 .read = virtio_pci_isr_read, 1831 .write = virtio_pci_isr_write, 1832 .impl = { 1833 .min_access_size = 1, 1834 .max_access_size = 4, 1835 }, 1836 .endianness = DEVICE_LITTLE_ENDIAN, 1837 }; 1838 static const MemoryRegionOps device_ops = { 1839 .read = virtio_pci_device_read, 1840 .write = virtio_pci_device_write, 1841 .impl = { 1842 .min_access_size = 1, 1843 .max_access_size = 4, 1844 }, 1845 .endianness = DEVICE_LITTLE_ENDIAN, 1846 }; 1847 static const MemoryRegionOps notify_ops = { 1848 .read = virtio_pci_notify_read, 1849 .write = virtio_pci_notify_write, 1850 .impl = { 1851 .min_access_size = 1, 1852 .max_access_size = 4, 1853 }, 1854 .endianness = DEVICE_LITTLE_ENDIAN, 1855 }; 1856 static const MemoryRegionOps notify_pio_ops = { 1857 .read = virtio_pci_notify_read, 1858 .write = virtio_pci_notify_write_pio, 1859 .impl = { 1860 .min_access_size = 1, 1861 .max_access_size = 4, 1862 }, 1863 .endianness = DEVICE_LITTLE_ENDIAN, 1864 }; 1865 g_autoptr(GString) name = g_string_new(NULL); 1866 1867 g_string_printf(name, "virtio-pci-common-%s", vdev_name); 1868 memory_region_init_io(&proxy->common.mr, OBJECT(proxy), 1869 &common_ops, 1870 proxy, 1871 name->str, 1872 proxy->common.size); 1873 1874 g_string_printf(name, "virtio-pci-isr-%s", vdev_name); 1875 memory_region_init_io(&proxy->isr.mr, OBJECT(proxy), 1876 &isr_ops, 1877 proxy, 1878 name->str, 1879 proxy->isr.size); 1880 1881 g_string_printf(name, "virtio-pci-device-%s", vdev_name); 1882 memory_region_init_io(&proxy->device.mr, OBJECT(proxy), 1883 &device_ops, 1884 proxy, 1885 name->str, 1886 proxy->device.size); 1887 1888 g_string_printf(name, "virtio-pci-notify-%s", vdev_name); 1889 memory_region_init_io(&proxy->notify.mr, OBJECT(proxy), 1890 ¬ify_ops, 1891 proxy, 1892 name->str, 1893 proxy->notify.size); 1894 1895 g_string_printf(name, "virtio-pci-notify-pio-%s", vdev_name); 1896 memory_region_init_io(&proxy->notify_pio.mr, OBJECT(proxy), 1897 ¬ify_pio_ops, 1898 proxy, 1899 name->str, 1900 proxy->notify_pio.size); 1901 } 1902 1903 static void virtio_pci_modern_region_map(VirtIOPCIProxy *proxy, 1904 VirtIOPCIRegion *region, 1905 struct virtio_pci_cap *cap, 1906 MemoryRegion *mr, 1907 uint8_t bar) 1908 { 1909 memory_region_add_subregion(mr, region->offset, ®ion->mr); 1910 1911 cap->cfg_type = region->type; 1912 cap->bar = bar; 1913 cap->offset = cpu_to_le32(region->offset); 1914 cap->length = cpu_to_le32(region->size); 1915 virtio_pci_add_mem_cap(proxy, cap); 1916 1917 } 1918 1919 static void virtio_pci_modern_mem_region_map(VirtIOPCIProxy *proxy, 1920 VirtIOPCIRegion *region, 1921 struct virtio_pci_cap *cap) 1922 { 1923 virtio_pci_modern_region_map(proxy, region, cap, 1924 &proxy->modern_bar, proxy->modern_mem_bar_idx); 1925 } 1926 1927 static void virtio_pci_modern_io_region_map(VirtIOPCIProxy *proxy, 1928 VirtIOPCIRegion *region, 1929 struct virtio_pci_cap *cap) 1930 { 1931 virtio_pci_modern_region_map(proxy, region, cap, 1932 &proxy->io_bar, proxy->modern_io_bar_idx); 1933 } 1934 1935 static void virtio_pci_modern_mem_region_unmap(VirtIOPCIProxy *proxy, 1936 VirtIOPCIRegion *region) 1937 { 1938 memory_region_del_subregion(&proxy->modern_bar, 1939 ®ion->mr); 1940 } 1941 1942 static void virtio_pci_modern_io_region_unmap(VirtIOPCIProxy *proxy, 1943 VirtIOPCIRegion *region) 1944 { 1945 memory_region_del_subregion(&proxy->io_bar, 1946 ®ion->mr); 1947 } 1948 1949 static void virtio_pci_pre_plugged(DeviceState *d, Error **errp) 1950 { 1951 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1952 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1953 1954 if (virtio_pci_modern(proxy)) { 1955 virtio_add_feature(&vdev->host_features, VIRTIO_F_VERSION_1); 1956 } 1957 1958 virtio_add_feature(&vdev->host_features, VIRTIO_F_BAD_FEATURE); 1959 } 1960 1961 /* This is called by virtio-bus just after the device is plugged. */ 1962 static void virtio_pci_device_plugged(DeviceState *d, Error **errp) 1963 { 1964 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1965 VirtioBusState *bus = &proxy->bus; 1966 bool legacy = virtio_pci_legacy(proxy); 1967 bool modern; 1968 bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY; 1969 uint8_t *config; 1970 uint32_t size; 1971 VirtIODevice *vdev = virtio_bus_get_device(bus); 1972 1973 /* 1974 * Virtio capabilities present without 1975 * VIRTIO_F_VERSION_1 confuses guests 1976 */ 1977 if (!proxy->ignore_backend_features && 1978 !virtio_has_feature(vdev->host_features, VIRTIO_F_VERSION_1)) { 1979 virtio_pci_disable_modern(proxy); 1980 1981 if (!legacy) { 1982 error_setg(errp, "Device doesn't support modern mode, and legacy" 1983 " mode is disabled"); 1984 error_append_hint(errp, "Set disable-legacy to off\n"); 1985 1986 return; 1987 } 1988 } 1989 1990 modern = virtio_pci_modern(proxy); 1991 1992 config = proxy->pci_dev.config; 1993 if (proxy->class_code) { 1994 pci_config_set_class(config, proxy->class_code); 1995 } 1996 1997 if (legacy) { 1998 if (!virtio_legacy_allowed(vdev)) { 1999 /* 2000 * To avoid migration issues, we allow legacy mode when legacy 2001 * check is disabled in the old machine types (< 5.1). 2002 */ 2003 if (virtio_legacy_check_disabled(vdev)) { 2004 warn_report("device is modern-only, but for backward " 2005 "compatibility legacy is allowed"); 2006 } else { 2007 error_setg(errp, 2008 "device is modern-only, use disable-legacy=on"); 2009 return; 2010 } 2011 } 2012 if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) { 2013 error_setg(errp, "VIRTIO_F_IOMMU_PLATFORM was supported by" 2014 " neither legacy nor transitional device"); 2015 return; 2016 } 2017 /* 2018 * Legacy and transitional devices use specific subsystem IDs. 2019 * Note that the subsystem vendor ID (config + PCI_SUBSYSTEM_VENDOR_ID) 2020 * is set to PCI_SUBVENDOR_ID_REDHAT_QUMRANET by default. 2021 */ 2022 pci_set_word(config + PCI_SUBSYSTEM_ID, virtio_bus_get_vdev_id(bus)); 2023 if (proxy->trans_devid) { 2024 pci_config_set_device_id(config, proxy->trans_devid); 2025 } 2026 } else { 2027 /* pure virtio-1.0 */ 2028 pci_set_word(config + PCI_VENDOR_ID, 2029 PCI_VENDOR_ID_REDHAT_QUMRANET); 2030 pci_set_word(config + PCI_DEVICE_ID, 2031 PCI_DEVICE_ID_VIRTIO_10_BASE + virtio_bus_get_vdev_id(bus)); 2032 pci_config_set_revision(config, 1); 2033 } 2034 config[PCI_INTERRUPT_PIN] = 1; 2035 2036 2037 if (modern) { 2038 struct virtio_pci_cap cap = { 2039 .cap_len = sizeof cap, 2040 }; 2041 struct virtio_pci_notify_cap notify = { 2042 .cap.cap_len = sizeof notify, 2043 .notify_off_multiplier = 2044 cpu_to_le32(virtio_pci_queue_mem_mult(proxy)), 2045 }; 2046 struct virtio_pci_cfg_cap cfg = { 2047 .cap.cap_len = sizeof cfg, 2048 .cap.cfg_type = VIRTIO_PCI_CAP_PCI_CFG, 2049 }; 2050 struct virtio_pci_notify_cap notify_pio = { 2051 .cap.cap_len = sizeof notify, 2052 .notify_off_multiplier = cpu_to_le32(0x0), 2053 }; 2054 2055 struct virtio_pci_cfg_cap *cfg_mask; 2056 2057 virtio_pci_modern_regions_init(proxy, vdev->name); 2058 2059 virtio_pci_modern_mem_region_map(proxy, &proxy->common, &cap); 2060 virtio_pci_modern_mem_region_map(proxy, &proxy->isr, &cap); 2061 virtio_pci_modern_mem_region_map(proxy, &proxy->device, &cap); 2062 virtio_pci_modern_mem_region_map(proxy, &proxy->notify, ¬ify.cap); 2063 2064 if (modern_pio) { 2065 memory_region_init(&proxy->io_bar, OBJECT(proxy), 2066 "virtio-pci-io", 0x4); 2067 2068 pci_register_bar(&proxy->pci_dev, proxy->modern_io_bar_idx, 2069 PCI_BASE_ADDRESS_SPACE_IO, &proxy->io_bar); 2070 2071 virtio_pci_modern_io_region_map(proxy, &proxy->notify_pio, 2072 ¬ify_pio.cap); 2073 } 2074 2075 pci_register_bar(&proxy->pci_dev, proxy->modern_mem_bar_idx, 2076 PCI_BASE_ADDRESS_SPACE_MEMORY | 2077 PCI_BASE_ADDRESS_MEM_PREFETCH | 2078 PCI_BASE_ADDRESS_MEM_TYPE_64, 2079 &proxy->modern_bar); 2080 2081 proxy->config_cap = virtio_pci_add_mem_cap(proxy, &cfg.cap); 2082 cfg_mask = (void *)(proxy->pci_dev.wmask + proxy->config_cap); 2083 pci_set_byte(&cfg_mask->cap.bar, ~0x0); 2084 pci_set_long((uint8_t *)&cfg_mask->cap.offset, ~0x0); 2085 pci_set_long((uint8_t *)&cfg_mask->cap.length, ~0x0); 2086 pci_set_long(cfg_mask->pci_cfg_data, ~0x0); 2087 } 2088 2089 if (proxy->nvectors) { 2090 int err = msix_init_exclusive_bar(&proxy->pci_dev, proxy->nvectors, 2091 proxy->msix_bar_idx, NULL); 2092 if (err) { 2093 /* Notice when a system that supports MSIx can't initialize it */ 2094 if (err != -ENOTSUP) { 2095 warn_report("unable to init msix vectors to %" PRIu32, 2096 proxy->nvectors); 2097 } 2098 proxy->nvectors = 0; 2099 } 2100 } 2101 2102 proxy->pci_dev.config_write = virtio_write_config; 2103 proxy->pci_dev.config_read = virtio_read_config; 2104 2105 if (legacy) { 2106 size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev) 2107 + virtio_bus_get_vdev_config_len(bus); 2108 size = pow2ceil(size); 2109 2110 memory_region_init_io(&proxy->bar, OBJECT(proxy), 2111 &virtio_pci_config_ops, 2112 proxy, "virtio-pci", size); 2113 2114 pci_register_bar(&proxy->pci_dev, proxy->legacy_io_bar_idx, 2115 PCI_BASE_ADDRESS_SPACE_IO, &proxy->bar); 2116 } 2117 } 2118 2119 static void virtio_pci_device_unplugged(DeviceState *d) 2120 { 2121 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 2122 bool modern = virtio_pci_modern(proxy); 2123 bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY; 2124 2125 virtio_pci_stop_ioeventfd(proxy); 2126 2127 if (modern) { 2128 virtio_pci_modern_mem_region_unmap(proxy, &proxy->common); 2129 virtio_pci_modern_mem_region_unmap(proxy, &proxy->isr); 2130 virtio_pci_modern_mem_region_unmap(proxy, &proxy->device); 2131 virtio_pci_modern_mem_region_unmap(proxy, &proxy->notify); 2132 if (modern_pio) { 2133 virtio_pci_modern_io_region_unmap(proxy, &proxy->notify_pio); 2134 } 2135 } 2136 } 2137 2138 static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp) 2139 { 2140 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); 2141 VirtioPCIClass *k = VIRTIO_PCI_GET_CLASS(pci_dev); 2142 bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) && 2143 !pci_bus_is_root(pci_get_bus(pci_dev)); 2144 2145 /* fd-based ioevents can't be synchronized in record/replay */ 2146 if (replay_mode != REPLAY_MODE_NONE) { 2147 proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD; 2148 } 2149 2150 /* 2151 * virtio pci bar layout used by default. 2152 * subclasses can re-arrange things if needed. 2153 * 2154 * region 0 -- virtio legacy io bar 2155 * region 1 -- msi-x bar 2156 * region 2 -- virtio modern io bar (off by default) 2157 * region 4+5 -- virtio modern memory (64bit) bar 2158 * 2159 */ 2160 proxy->legacy_io_bar_idx = 0; 2161 proxy->msix_bar_idx = 1; 2162 proxy->modern_io_bar_idx = 2; 2163 proxy->modern_mem_bar_idx = 4; 2164 2165 proxy->common.offset = 0x0; 2166 proxy->common.size = 0x1000; 2167 proxy->common.type = VIRTIO_PCI_CAP_COMMON_CFG; 2168 2169 proxy->isr.offset = 0x1000; 2170 proxy->isr.size = 0x1000; 2171 proxy->isr.type = VIRTIO_PCI_CAP_ISR_CFG; 2172 2173 proxy->device.offset = 0x2000; 2174 proxy->device.size = 0x1000; 2175 proxy->device.type = VIRTIO_PCI_CAP_DEVICE_CFG; 2176 2177 proxy->notify.offset = 0x3000; 2178 proxy->notify.size = virtio_pci_queue_mem_mult(proxy) * VIRTIO_QUEUE_MAX; 2179 proxy->notify.type = VIRTIO_PCI_CAP_NOTIFY_CFG; 2180 2181 proxy->notify_pio.offset = 0x0; 2182 proxy->notify_pio.size = 0x4; 2183 proxy->notify_pio.type = VIRTIO_PCI_CAP_NOTIFY_CFG; 2184 2185 /* subclasses can enforce modern, so do this unconditionally */ 2186 memory_region_init(&proxy->modern_bar, OBJECT(proxy), "virtio-pci", 2187 /* PCI BAR regions must be powers of 2 */ 2188 pow2ceil(proxy->notify.offset + proxy->notify.size)); 2189 2190 if (proxy->disable_legacy == ON_OFF_AUTO_AUTO) { 2191 proxy->disable_legacy = pcie_port ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF; 2192 } 2193 2194 if (!virtio_pci_modern(proxy) && !virtio_pci_legacy(proxy)) { 2195 error_setg(errp, "device cannot work as neither modern nor legacy mode" 2196 " is enabled"); 2197 error_append_hint(errp, "Set either disable-modern or disable-legacy" 2198 " to off\n"); 2199 return; 2200 } 2201 2202 if (pcie_port && pci_is_express(pci_dev)) { 2203 int pos; 2204 uint16_t last_pcie_cap_offset = PCI_CONFIG_SPACE_SIZE; 2205 2206 pos = pcie_endpoint_cap_init(pci_dev, 0); 2207 assert(pos > 0); 2208 2209 pos = pci_add_capability(pci_dev, PCI_CAP_ID_PM, 0, 2210 PCI_PM_SIZEOF, errp); 2211 if (pos < 0) { 2212 return; 2213 } 2214 2215 pci_dev->exp.pm_cap = pos; 2216 2217 /* 2218 * Indicates that this function complies with revision 1.2 of the 2219 * PCI Power Management Interface Specification. 2220 */ 2221 pci_set_word(pci_dev->config + pos + PCI_PM_PMC, 0x3); 2222 2223 if (proxy->flags & VIRTIO_PCI_FLAG_AER) { 2224 pcie_aer_init(pci_dev, PCI_ERR_VER, last_pcie_cap_offset, 2225 PCI_ERR_SIZEOF, NULL); 2226 last_pcie_cap_offset += PCI_ERR_SIZEOF; 2227 } 2228 2229 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_DEVERR) { 2230 /* Init error enabling flags */ 2231 pcie_cap_deverr_init(pci_dev); 2232 } 2233 2234 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_LNKCTL) { 2235 /* Init Link Control Register */ 2236 pcie_cap_lnkctl_init(pci_dev); 2237 } 2238 2239 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_PM) { 2240 /* Init Power Management Control Register */ 2241 pci_set_word(pci_dev->wmask + pos + PCI_PM_CTRL, 2242 PCI_PM_CTRL_STATE_MASK); 2243 } 2244 2245 if (proxy->flags & VIRTIO_PCI_FLAG_ATS) { 2246 pcie_ats_init(pci_dev, last_pcie_cap_offset, 2247 proxy->flags & VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED); 2248 last_pcie_cap_offset += PCI_EXT_CAP_ATS_SIZEOF; 2249 } 2250 2251 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_FLR) { 2252 /* Set Function Level Reset capability bit */ 2253 pcie_cap_flr_init(pci_dev); 2254 } 2255 } else { 2256 /* 2257 * make future invocations of pci_is_express() return false 2258 * and pci_config_size() return PCI_CONFIG_SPACE_SIZE. 2259 */ 2260 pci_dev->cap_present &= ~QEMU_PCI_CAP_EXPRESS; 2261 } 2262 2263 virtio_pci_bus_new(&proxy->bus, sizeof(proxy->bus), proxy); 2264 if (k->realize) { 2265 k->realize(proxy, errp); 2266 } 2267 } 2268 2269 static void virtio_pci_exit(PCIDevice *pci_dev) 2270 { 2271 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); 2272 bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) && 2273 !pci_bus_is_root(pci_get_bus(pci_dev)); 2274 2275 msix_uninit_exclusive_bar(pci_dev); 2276 if (proxy->flags & VIRTIO_PCI_FLAG_AER && pcie_port && 2277 pci_is_express(pci_dev)) { 2278 pcie_aer_exit(pci_dev); 2279 } 2280 } 2281 2282 static void virtio_pci_reset(DeviceState *qdev) 2283 { 2284 VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev); 2285 VirtioBusState *bus = VIRTIO_BUS(&proxy->bus); 2286 int i; 2287 2288 virtio_bus_reset(bus); 2289 msix_unuse_all_vectors(&proxy->pci_dev); 2290 2291 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 2292 proxy->vqs[i].enabled = 0; 2293 proxy->vqs[i].reset = 0; 2294 proxy->vqs[i].num = 0; 2295 proxy->vqs[i].desc[0] = proxy->vqs[i].desc[1] = 0; 2296 proxy->vqs[i].avail[0] = proxy->vqs[i].avail[1] = 0; 2297 proxy->vqs[i].used[0] = proxy->vqs[i].used[1] = 0; 2298 } 2299 } 2300 2301 static void virtio_pci_bus_reset_hold(Object *obj, ResetType type) 2302 { 2303 PCIDevice *dev = PCI_DEVICE(obj); 2304 DeviceState *qdev = DEVICE(obj); 2305 2306 virtio_pci_reset(qdev); 2307 2308 if (pci_is_express(dev)) { 2309 pcie_cap_deverr_reset(dev); 2310 pcie_cap_lnkctl_reset(dev); 2311 2312 pci_set_word(dev->config + dev->exp.pm_cap + PCI_PM_CTRL, 0); 2313 } 2314 } 2315 2316 static Property virtio_pci_properties[] = { 2317 DEFINE_PROP_BIT("virtio-pci-bus-master-bug-migration", VirtIOPCIProxy, flags, 2318 VIRTIO_PCI_FLAG_BUS_MASTER_BUG_MIGRATION_BIT, false), 2319 DEFINE_PROP_BIT("migrate-extra", VirtIOPCIProxy, flags, 2320 VIRTIO_PCI_FLAG_MIGRATE_EXTRA_BIT, true), 2321 DEFINE_PROP_BIT("modern-pio-notify", VirtIOPCIProxy, flags, 2322 VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY_BIT, false), 2323 DEFINE_PROP_BIT("x-disable-pcie", VirtIOPCIProxy, flags, 2324 VIRTIO_PCI_FLAG_DISABLE_PCIE_BIT, false), 2325 DEFINE_PROP_BIT("page-per-vq", VirtIOPCIProxy, flags, 2326 VIRTIO_PCI_FLAG_PAGE_PER_VQ_BIT, false), 2327 DEFINE_PROP_BOOL("x-ignore-backend-features", VirtIOPCIProxy, 2328 ignore_backend_features, false), 2329 DEFINE_PROP_BIT("ats", VirtIOPCIProxy, flags, 2330 VIRTIO_PCI_FLAG_ATS_BIT, false), 2331 DEFINE_PROP_BIT("x-ats-page-aligned", VirtIOPCIProxy, flags, 2332 VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED_BIT, true), 2333 DEFINE_PROP_BIT("x-pcie-deverr-init", VirtIOPCIProxy, flags, 2334 VIRTIO_PCI_FLAG_INIT_DEVERR_BIT, true), 2335 DEFINE_PROP_BIT("x-pcie-lnkctl-init", VirtIOPCIProxy, flags, 2336 VIRTIO_PCI_FLAG_INIT_LNKCTL_BIT, true), 2337 DEFINE_PROP_BIT("x-pcie-pm-init", VirtIOPCIProxy, flags, 2338 VIRTIO_PCI_FLAG_INIT_PM_BIT, true), 2339 DEFINE_PROP_BIT("x-pcie-flr-init", VirtIOPCIProxy, flags, 2340 VIRTIO_PCI_FLAG_INIT_FLR_BIT, true), 2341 DEFINE_PROP_BIT("aer", VirtIOPCIProxy, flags, 2342 VIRTIO_PCI_FLAG_AER_BIT, false), 2343 DEFINE_PROP_END_OF_LIST(), 2344 }; 2345 2346 static void virtio_pci_dc_realize(DeviceState *qdev, Error **errp) 2347 { 2348 VirtioPCIClass *vpciklass = VIRTIO_PCI_GET_CLASS(qdev); 2349 VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev); 2350 PCIDevice *pci_dev = &proxy->pci_dev; 2351 2352 if (!(proxy->flags & VIRTIO_PCI_FLAG_DISABLE_PCIE) && 2353 virtio_pci_modern(proxy)) { 2354 pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS; 2355 } 2356 2357 vpciklass->parent_dc_realize(qdev, errp); 2358 } 2359 2360 static void virtio_pci_class_init(ObjectClass *klass, void *data) 2361 { 2362 DeviceClass *dc = DEVICE_CLASS(klass); 2363 PCIDeviceClass *k = PCI_DEVICE_CLASS(klass); 2364 VirtioPCIClass *vpciklass = VIRTIO_PCI_CLASS(klass); 2365 ResettableClass *rc = RESETTABLE_CLASS(klass); 2366 2367 device_class_set_props(dc, virtio_pci_properties); 2368 k->realize = virtio_pci_realize; 2369 k->exit = virtio_pci_exit; 2370 k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET; 2371 k->revision = VIRTIO_PCI_ABI_VERSION; 2372 k->class_id = PCI_CLASS_OTHERS; 2373 device_class_set_parent_realize(dc, virtio_pci_dc_realize, 2374 &vpciklass->parent_dc_realize); 2375 rc->phases.hold = virtio_pci_bus_reset_hold; 2376 } 2377 2378 static const TypeInfo virtio_pci_info = { 2379 .name = TYPE_VIRTIO_PCI, 2380 .parent = TYPE_PCI_DEVICE, 2381 .instance_size = sizeof(VirtIOPCIProxy), 2382 .class_init = virtio_pci_class_init, 2383 .class_size = sizeof(VirtioPCIClass), 2384 .abstract = true, 2385 }; 2386 2387 static Property virtio_pci_generic_properties[] = { 2388 DEFINE_PROP_ON_OFF_AUTO("disable-legacy", VirtIOPCIProxy, disable_legacy, 2389 ON_OFF_AUTO_AUTO), 2390 DEFINE_PROP_BOOL("disable-modern", VirtIOPCIProxy, disable_modern, false), 2391 DEFINE_PROP_END_OF_LIST(), 2392 }; 2393 2394 static void virtio_pci_base_class_init(ObjectClass *klass, void *data) 2395 { 2396 const VirtioPCIDeviceTypeInfo *t = data; 2397 if (t->class_init) { 2398 t->class_init(klass, NULL); 2399 } 2400 } 2401 2402 static void virtio_pci_generic_class_init(ObjectClass *klass, void *data) 2403 { 2404 DeviceClass *dc = DEVICE_CLASS(klass); 2405 2406 device_class_set_props(dc, virtio_pci_generic_properties); 2407 } 2408 2409 static void virtio_pci_transitional_instance_init(Object *obj) 2410 { 2411 VirtIOPCIProxy *proxy = VIRTIO_PCI(obj); 2412 2413 proxy->disable_legacy = ON_OFF_AUTO_OFF; 2414 proxy->disable_modern = false; 2415 } 2416 2417 static void virtio_pci_non_transitional_instance_init(Object *obj) 2418 { 2419 VirtIOPCIProxy *proxy = VIRTIO_PCI(obj); 2420 2421 proxy->disable_legacy = ON_OFF_AUTO_ON; 2422 proxy->disable_modern = false; 2423 } 2424 2425 void virtio_pci_types_register(const VirtioPCIDeviceTypeInfo *t) 2426 { 2427 char *base_name = NULL; 2428 TypeInfo base_type_info = { 2429 .name = t->base_name, 2430 .parent = t->parent ? t->parent : TYPE_VIRTIO_PCI, 2431 .instance_size = t->instance_size, 2432 .instance_init = t->instance_init, 2433 .instance_finalize = t->instance_finalize, 2434 .class_size = t->class_size, 2435 .abstract = true, 2436 .interfaces = t->interfaces, 2437 }; 2438 TypeInfo generic_type_info = { 2439 .name = t->generic_name, 2440 .parent = base_type_info.name, 2441 .class_init = virtio_pci_generic_class_init, 2442 .interfaces = (InterfaceInfo[]) { 2443 { INTERFACE_PCIE_DEVICE }, 2444 { INTERFACE_CONVENTIONAL_PCI_DEVICE }, 2445 { } 2446 }, 2447 }; 2448 2449 if (!base_type_info.name) { 2450 /* No base type -> register a single generic device type */ 2451 /* use intermediate %s-base-type to add generic device props */ 2452 base_name = g_strdup_printf("%s-base-type", t->generic_name); 2453 base_type_info.name = base_name; 2454 base_type_info.class_init = virtio_pci_generic_class_init; 2455 2456 generic_type_info.parent = base_name; 2457 generic_type_info.class_init = virtio_pci_base_class_init; 2458 generic_type_info.class_data = (void *)t; 2459 2460 assert(!t->non_transitional_name); 2461 assert(!t->transitional_name); 2462 } else { 2463 base_type_info.class_init = virtio_pci_base_class_init; 2464 base_type_info.class_data = (void *)t; 2465 } 2466 2467 type_register(&base_type_info); 2468 if (generic_type_info.name) { 2469 type_register(&generic_type_info); 2470 } 2471 2472 if (t->non_transitional_name) { 2473 const TypeInfo non_transitional_type_info = { 2474 .name = t->non_transitional_name, 2475 .parent = base_type_info.name, 2476 .instance_init = virtio_pci_non_transitional_instance_init, 2477 .interfaces = (InterfaceInfo[]) { 2478 { INTERFACE_PCIE_DEVICE }, 2479 { INTERFACE_CONVENTIONAL_PCI_DEVICE }, 2480 { } 2481 }, 2482 }; 2483 type_register(&non_transitional_type_info); 2484 } 2485 2486 if (t->transitional_name) { 2487 const TypeInfo transitional_type_info = { 2488 .name = t->transitional_name, 2489 .parent = base_type_info.name, 2490 .instance_init = virtio_pci_transitional_instance_init, 2491 .interfaces = (InterfaceInfo[]) { 2492 /* 2493 * Transitional virtio devices work only as Conventional PCI 2494 * devices because they require PIO ports. 2495 */ 2496 { INTERFACE_CONVENTIONAL_PCI_DEVICE }, 2497 { } 2498 }, 2499 }; 2500 type_register(&transitional_type_info); 2501 } 2502 g_free(base_name); 2503 } 2504 2505 unsigned virtio_pci_optimal_num_queues(unsigned fixed_queues) 2506 { 2507 /* 2508 * 1:1 vq to vCPU mapping is ideal because the same vCPU that submitted 2509 * virtqueue buffers can handle their completion. When a different vCPU 2510 * handles completion it may need to IPI the vCPU that submitted the 2511 * request and this adds overhead. 2512 * 2513 * Virtqueues consume guest RAM and MSI-X vectors. This is wasteful in 2514 * guests with very many vCPUs and a device that is only used by a few 2515 * vCPUs. Unfortunately optimizing that case requires manual pinning inside 2516 * the guest, so those users might as well manually set the number of 2517 * queues. There is no upper limit that can be applied automatically and 2518 * doing so arbitrarily would result in a sudden performance drop once the 2519 * threshold number of vCPUs is exceeded. 2520 */ 2521 unsigned num_queues = current_machine->smp.cpus; 2522 2523 /* 2524 * The maximum number of MSI-X vectors is PCI_MSIX_FLAGS_QSIZE + 1, but the 2525 * config change interrupt and the fixed virtqueues must be taken into 2526 * account too. 2527 */ 2528 num_queues = MIN(num_queues, PCI_MSIX_FLAGS_QSIZE - fixed_queues); 2529 2530 /* 2531 * There is a limit to how many virtqueues a device can have. 2532 */ 2533 return MIN(num_queues, VIRTIO_QUEUE_MAX - fixed_queues); 2534 } 2535 2536 /* virtio-pci-bus */ 2537 2538 static void virtio_pci_bus_new(VirtioBusState *bus, size_t bus_size, 2539 VirtIOPCIProxy *dev) 2540 { 2541 DeviceState *qdev = DEVICE(dev); 2542 char virtio_bus_name[] = "virtio-bus"; 2543 2544 qbus_init(bus, bus_size, TYPE_VIRTIO_PCI_BUS, qdev, virtio_bus_name); 2545 } 2546 2547 static void virtio_pci_bus_class_init(ObjectClass *klass, void *data) 2548 { 2549 BusClass *bus_class = BUS_CLASS(klass); 2550 VirtioBusClass *k = VIRTIO_BUS_CLASS(klass); 2551 bus_class->max_dev = 1; 2552 k->notify = virtio_pci_notify; 2553 k->save_config = virtio_pci_save_config; 2554 k->load_config = virtio_pci_load_config; 2555 k->save_queue = virtio_pci_save_queue; 2556 k->load_queue = virtio_pci_load_queue; 2557 k->save_extra_state = virtio_pci_save_extra_state; 2558 k->load_extra_state = virtio_pci_load_extra_state; 2559 k->has_extra_state = virtio_pci_has_extra_state; 2560 k->query_guest_notifiers = virtio_pci_query_guest_notifiers; 2561 k->set_guest_notifiers = virtio_pci_set_guest_notifiers; 2562 k->set_host_notifier_mr = virtio_pci_set_host_notifier_mr; 2563 k->vmstate_change = virtio_pci_vmstate_change; 2564 k->pre_plugged = virtio_pci_pre_plugged; 2565 k->device_plugged = virtio_pci_device_plugged; 2566 k->device_unplugged = virtio_pci_device_unplugged; 2567 k->query_nvectors = virtio_pci_query_nvectors; 2568 k->ioeventfd_enabled = virtio_pci_ioeventfd_enabled; 2569 k->ioeventfd_assign = virtio_pci_ioeventfd_assign; 2570 k->get_dma_as = virtio_pci_get_dma_as; 2571 k->iommu_enabled = virtio_pci_iommu_enabled; 2572 k->queue_enabled = virtio_pci_queue_enabled; 2573 } 2574 2575 static const TypeInfo virtio_pci_bus_info = { 2576 .name = TYPE_VIRTIO_PCI_BUS, 2577 .parent = TYPE_VIRTIO_BUS, 2578 .instance_size = sizeof(VirtioPCIBusState), 2579 .class_size = sizeof(VirtioPCIBusClass), 2580 .class_init = virtio_pci_bus_class_init, 2581 }; 2582 2583 static void virtio_pci_register_types(void) 2584 { 2585 /* Base types: */ 2586 type_register_static(&virtio_pci_bus_info); 2587 type_register_static(&virtio_pci_info); 2588 } 2589 2590 type_init(virtio_pci_register_types) 2591 2592