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 "hw/boards.h" 23 #include "hw/virtio/virtio.h" 24 #include "migration/qemu-file-types.h" 25 #include "hw/pci/pci.h" 26 #include "hw/pci/pci_bus.h" 27 #include "hw/qdev-properties.h" 28 #include "qapi/error.h" 29 #include "qemu/error-report.h" 30 #include "qemu/log.h" 31 #include "qemu/module.h" 32 #include "hw/pci/msi.h" 33 #include "hw/pci/msix.h" 34 #include "hw/loader.h" 35 #include "sysemu/kvm.h" 36 #include "hw/virtio/virtio-pci.h" 37 #include "qemu/range.h" 38 #include "hw/virtio/virtio-bus.h" 39 #include "qapi/visitor.h" 40 #include "sysemu/replay.h" 41 #include "trace.h" 42 43 #define VIRTIO_PCI_REGION_SIZE(dev) VIRTIO_PCI_CONFIG_OFF(msix_present(dev)) 44 45 #undef VIRTIO_PCI_CONFIG 46 47 /* The remaining space is defined by each driver as the per-driver 48 * configuration space */ 49 #define VIRTIO_PCI_CONFIG_SIZE(dev) VIRTIO_PCI_CONFIG_OFF(msix_enabled(dev)) 50 51 static void virtio_pci_bus_new(VirtioBusState *bus, size_t bus_size, 52 VirtIOPCIProxy *dev); 53 static void virtio_pci_reset(DeviceState *qdev); 54 55 /* virtio device */ 56 /* DeviceState to VirtIOPCIProxy. For use off data-path. TODO: use QOM. */ 57 static inline VirtIOPCIProxy *to_virtio_pci_proxy(DeviceState *d) 58 { 59 return container_of(d, VirtIOPCIProxy, pci_dev.qdev); 60 } 61 62 /* DeviceState to VirtIOPCIProxy. Note: used on datapath, 63 * be careful and test performance if you change this. 64 */ 65 static inline VirtIOPCIProxy *to_virtio_pci_proxy_fast(DeviceState *d) 66 { 67 return container_of(d, VirtIOPCIProxy, pci_dev.qdev); 68 } 69 70 static void virtio_pci_notify(DeviceState *d, uint16_t vector) 71 { 72 VirtIOPCIProxy *proxy = to_virtio_pci_proxy_fast(d); 73 74 if (msix_enabled(&proxy->pci_dev)) 75 msix_notify(&proxy->pci_dev, vector); 76 else { 77 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 78 pci_set_irq(&proxy->pci_dev, qatomic_read(&vdev->isr) & 1); 79 } 80 } 81 82 static void virtio_pci_save_config(DeviceState *d, QEMUFile *f) 83 { 84 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 85 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 86 87 pci_device_save(&proxy->pci_dev, f); 88 msix_save(&proxy->pci_dev, f); 89 if (msix_present(&proxy->pci_dev)) 90 qemu_put_be16(f, vdev->config_vector); 91 } 92 93 static const VMStateDescription vmstate_virtio_pci_modern_queue_state = { 94 .name = "virtio_pci/modern_queue_state", 95 .version_id = 1, 96 .minimum_version_id = 1, 97 .fields = (VMStateField[]) { 98 VMSTATE_UINT16(num, VirtIOPCIQueue), 99 VMSTATE_UNUSED(1), /* enabled was stored as be16 */ 100 VMSTATE_BOOL(enabled, VirtIOPCIQueue), 101 VMSTATE_UINT32_ARRAY(desc, VirtIOPCIQueue, 2), 102 VMSTATE_UINT32_ARRAY(avail, VirtIOPCIQueue, 2), 103 VMSTATE_UINT32_ARRAY(used, VirtIOPCIQueue, 2), 104 VMSTATE_END_OF_LIST() 105 } 106 }; 107 108 static bool virtio_pci_modern_state_needed(void *opaque) 109 { 110 VirtIOPCIProxy *proxy = opaque; 111 112 return virtio_pci_modern(proxy); 113 } 114 115 static const VMStateDescription vmstate_virtio_pci_modern_state_sub = { 116 .name = "virtio_pci/modern_state", 117 .version_id = 1, 118 .minimum_version_id = 1, 119 .needed = &virtio_pci_modern_state_needed, 120 .fields = (VMStateField[]) { 121 VMSTATE_UINT32(dfselect, VirtIOPCIProxy), 122 VMSTATE_UINT32(gfselect, VirtIOPCIProxy), 123 VMSTATE_UINT32_ARRAY(guest_features, VirtIOPCIProxy, 2), 124 VMSTATE_STRUCT_ARRAY(vqs, VirtIOPCIProxy, VIRTIO_QUEUE_MAX, 0, 125 vmstate_virtio_pci_modern_queue_state, 126 VirtIOPCIQueue), 127 VMSTATE_END_OF_LIST() 128 } 129 }; 130 131 static const VMStateDescription vmstate_virtio_pci = { 132 .name = "virtio_pci", 133 .version_id = 1, 134 .minimum_version_id = 1, 135 .fields = (VMStateField[]) { 136 VMSTATE_END_OF_LIST() 137 }, 138 .subsections = (const VMStateDescription*[]) { 139 &vmstate_virtio_pci_modern_state_sub, 140 NULL 141 } 142 }; 143 144 static bool virtio_pci_has_extra_state(DeviceState *d) 145 { 146 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 147 148 return proxy->flags & VIRTIO_PCI_FLAG_MIGRATE_EXTRA; 149 } 150 151 static void virtio_pci_save_extra_state(DeviceState *d, QEMUFile *f) 152 { 153 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 154 155 vmstate_save_state(f, &vmstate_virtio_pci, proxy, NULL); 156 } 157 158 static int virtio_pci_load_extra_state(DeviceState *d, QEMUFile *f) 159 { 160 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 161 162 return vmstate_load_state(f, &vmstate_virtio_pci, proxy, 1); 163 } 164 165 static void virtio_pci_save_queue(DeviceState *d, int n, QEMUFile *f) 166 { 167 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 168 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 169 170 if (msix_present(&proxy->pci_dev)) 171 qemu_put_be16(f, virtio_queue_vector(vdev, n)); 172 } 173 174 static int virtio_pci_load_config(DeviceState *d, QEMUFile *f) 175 { 176 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 177 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 178 179 int ret; 180 ret = pci_device_load(&proxy->pci_dev, f); 181 if (ret) { 182 return ret; 183 } 184 msix_unuse_all_vectors(&proxy->pci_dev); 185 msix_load(&proxy->pci_dev, f); 186 if (msix_present(&proxy->pci_dev)) { 187 qemu_get_be16s(f, &vdev->config_vector); 188 } else { 189 vdev->config_vector = VIRTIO_NO_VECTOR; 190 } 191 if (vdev->config_vector != VIRTIO_NO_VECTOR) { 192 return msix_vector_use(&proxy->pci_dev, vdev->config_vector); 193 } 194 return 0; 195 } 196 197 static int virtio_pci_load_queue(DeviceState *d, int n, QEMUFile *f) 198 { 199 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 200 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 201 202 uint16_t vector; 203 if (msix_present(&proxy->pci_dev)) { 204 qemu_get_be16s(f, &vector); 205 } else { 206 vector = VIRTIO_NO_VECTOR; 207 } 208 virtio_queue_set_vector(vdev, n, vector); 209 if (vector != VIRTIO_NO_VECTOR) { 210 return msix_vector_use(&proxy->pci_dev, vector); 211 } 212 213 return 0; 214 } 215 216 static bool virtio_pci_ioeventfd_enabled(DeviceState *d) 217 { 218 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 219 220 return (proxy->flags & VIRTIO_PCI_FLAG_USE_IOEVENTFD) != 0; 221 } 222 223 #define QEMU_VIRTIO_PCI_QUEUE_MEM_MULT 0x1000 224 225 static inline int virtio_pci_queue_mem_mult(struct VirtIOPCIProxy *proxy) 226 { 227 return (proxy->flags & VIRTIO_PCI_FLAG_PAGE_PER_VQ) ? 228 QEMU_VIRTIO_PCI_QUEUE_MEM_MULT : 4; 229 } 230 231 static int virtio_pci_ioeventfd_assign(DeviceState *d, EventNotifier *notifier, 232 int n, bool assign) 233 { 234 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 235 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 236 VirtQueue *vq = virtio_get_queue(vdev, n); 237 bool legacy = virtio_pci_legacy(proxy); 238 bool modern = virtio_pci_modern(proxy); 239 bool fast_mmio = kvm_ioeventfd_any_length_enabled(); 240 bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY; 241 MemoryRegion *modern_mr = &proxy->notify.mr; 242 MemoryRegion *modern_notify_mr = &proxy->notify_pio.mr; 243 MemoryRegion *legacy_mr = &proxy->bar; 244 hwaddr modern_addr = virtio_pci_queue_mem_mult(proxy) * 245 virtio_get_queue_index(vq); 246 hwaddr legacy_addr = VIRTIO_PCI_QUEUE_NOTIFY; 247 248 if (assign) { 249 if (modern) { 250 if (fast_mmio) { 251 memory_region_add_eventfd(modern_mr, modern_addr, 0, 252 false, n, notifier); 253 } else { 254 memory_region_add_eventfd(modern_mr, modern_addr, 2, 255 false, n, notifier); 256 } 257 if (modern_pio) { 258 memory_region_add_eventfd(modern_notify_mr, 0, 2, 259 true, n, notifier); 260 } 261 } 262 if (legacy) { 263 memory_region_add_eventfd(legacy_mr, legacy_addr, 2, 264 true, n, notifier); 265 } 266 } else { 267 if (modern) { 268 if (fast_mmio) { 269 memory_region_del_eventfd(modern_mr, modern_addr, 0, 270 false, n, notifier); 271 } else { 272 memory_region_del_eventfd(modern_mr, modern_addr, 2, 273 false, n, notifier); 274 } 275 if (modern_pio) { 276 memory_region_del_eventfd(modern_notify_mr, 0, 2, 277 true, n, notifier); 278 } 279 } 280 if (legacy) { 281 memory_region_del_eventfd(legacy_mr, legacy_addr, 2, 282 true, n, notifier); 283 } 284 } 285 return 0; 286 } 287 288 static void virtio_pci_start_ioeventfd(VirtIOPCIProxy *proxy) 289 { 290 virtio_bus_start_ioeventfd(&proxy->bus); 291 } 292 293 static void virtio_pci_stop_ioeventfd(VirtIOPCIProxy *proxy) 294 { 295 virtio_bus_stop_ioeventfd(&proxy->bus); 296 } 297 298 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val) 299 { 300 VirtIOPCIProxy *proxy = opaque; 301 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 302 hwaddr pa; 303 304 switch (addr) { 305 case VIRTIO_PCI_GUEST_FEATURES: 306 /* Guest does not negotiate properly? We have to assume nothing. */ 307 if (val & (1 << VIRTIO_F_BAD_FEATURE)) { 308 val = virtio_bus_get_vdev_bad_features(&proxy->bus); 309 } 310 virtio_set_features(vdev, val); 311 break; 312 case VIRTIO_PCI_QUEUE_PFN: 313 pa = (hwaddr)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT; 314 if (pa == 0) { 315 virtio_pci_reset(DEVICE(proxy)); 316 } 317 else 318 virtio_queue_set_addr(vdev, vdev->queue_sel, pa); 319 break; 320 case VIRTIO_PCI_QUEUE_SEL: 321 if (val < VIRTIO_QUEUE_MAX) 322 vdev->queue_sel = val; 323 break; 324 case VIRTIO_PCI_QUEUE_NOTIFY: 325 if (val < VIRTIO_QUEUE_MAX) { 326 virtio_queue_notify(vdev, val); 327 } 328 break; 329 case VIRTIO_PCI_STATUS: 330 if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) { 331 virtio_pci_stop_ioeventfd(proxy); 332 } 333 334 virtio_set_status(vdev, val & 0xFF); 335 336 if (val & VIRTIO_CONFIG_S_DRIVER_OK) { 337 virtio_pci_start_ioeventfd(proxy); 338 } 339 340 if (vdev->status == 0) { 341 virtio_pci_reset(DEVICE(proxy)); 342 } 343 344 /* Linux before 2.6.34 drives the device without enabling 345 the PCI device bus master bit. Enable it automatically 346 for the guest. This is a PCI spec violation but so is 347 initiating DMA with bus master bit clear. */ 348 if (val == (VIRTIO_CONFIG_S_ACKNOWLEDGE | VIRTIO_CONFIG_S_DRIVER)) { 349 pci_default_write_config(&proxy->pci_dev, PCI_COMMAND, 350 proxy->pci_dev.config[PCI_COMMAND] | 351 PCI_COMMAND_MASTER, 1); 352 } 353 break; 354 case VIRTIO_MSI_CONFIG_VECTOR: 355 msix_vector_unuse(&proxy->pci_dev, vdev->config_vector); 356 /* Make it possible for guest to discover an error took place. */ 357 if (msix_vector_use(&proxy->pci_dev, val) < 0) 358 val = VIRTIO_NO_VECTOR; 359 vdev->config_vector = val; 360 break; 361 case VIRTIO_MSI_QUEUE_VECTOR: 362 msix_vector_unuse(&proxy->pci_dev, 363 virtio_queue_vector(vdev, vdev->queue_sel)); 364 /* Make it possible for guest to discover an error took place. */ 365 if (msix_vector_use(&proxy->pci_dev, val) < 0) 366 val = VIRTIO_NO_VECTOR; 367 virtio_queue_set_vector(vdev, vdev->queue_sel, val); 368 break; 369 default: 370 qemu_log_mask(LOG_GUEST_ERROR, 371 "%s: unexpected address 0x%x value 0x%x\n", 372 __func__, addr, val); 373 break; 374 } 375 } 376 377 static uint32_t virtio_ioport_read(VirtIOPCIProxy *proxy, uint32_t addr) 378 { 379 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 380 uint32_t ret = 0xFFFFFFFF; 381 382 switch (addr) { 383 case VIRTIO_PCI_HOST_FEATURES: 384 ret = vdev->host_features; 385 break; 386 case VIRTIO_PCI_GUEST_FEATURES: 387 ret = vdev->guest_features; 388 break; 389 case VIRTIO_PCI_QUEUE_PFN: 390 ret = virtio_queue_get_addr(vdev, vdev->queue_sel) 391 >> VIRTIO_PCI_QUEUE_ADDR_SHIFT; 392 break; 393 case VIRTIO_PCI_QUEUE_NUM: 394 ret = virtio_queue_get_num(vdev, vdev->queue_sel); 395 break; 396 case VIRTIO_PCI_QUEUE_SEL: 397 ret = vdev->queue_sel; 398 break; 399 case VIRTIO_PCI_STATUS: 400 ret = vdev->status; 401 break; 402 case VIRTIO_PCI_ISR: 403 /* reading from the ISR also clears it. */ 404 ret = qatomic_xchg(&vdev->isr, 0); 405 pci_irq_deassert(&proxy->pci_dev); 406 break; 407 case VIRTIO_MSI_CONFIG_VECTOR: 408 ret = vdev->config_vector; 409 break; 410 case VIRTIO_MSI_QUEUE_VECTOR: 411 ret = virtio_queue_vector(vdev, vdev->queue_sel); 412 break; 413 default: 414 break; 415 } 416 417 return ret; 418 } 419 420 static uint64_t virtio_pci_config_read(void *opaque, hwaddr addr, 421 unsigned size) 422 { 423 VirtIOPCIProxy *proxy = opaque; 424 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 425 uint32_t config = VIRTIO_PCI_CONFIG_SIZE(&proxy->pci_dev); 426 uint64_t val = 0; 427 428 if (vdev == NULL) { 429 return UINT64_MAX; 430 } 431 432 if (addr < config) { 433 return virtio_ioport_read(proxy, addr); 434 } 435 addr -= config; 436 437 switch (size) { 438 case 1: 439 val = virtio_config_readb(vdev, addr); 440 break; 441 case 2: 442 val = virtio_config_readw(vdev, addr); 443 if (virtio_is_big_endian(vdev)) { 444 val = bswap16(val); 445 } 446 break; 447 case 4: 448 val = virtio_config_readl(vdev, addr); 449 if (virtio_is_big_endian(vdev)) { 450 val = bswap32(val); 451 } 452 break; 453 } 454 return val; 455 } 456 457 static void virtio_pci_config_write(void *opaque, hwaddr addr, 458 uint64_t val, unsigned size) 459 { 460 VirtIOPCIProxy *proxy = opaque; 461 uint32_t config = VIRTIO_PCI_CONFIG_SIZE(&proxy->pci_dev); 462 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 463 464 if (vdev == NULL) { 465 return; 466 } 467 468 if (addr < config) { 469 virtio_ioport_write(proxy, addr, val); 470 return; 471 } 472 addr -= config; 473 /* 474 * Virtio-PCI is odd. Ioports are LE but config space is target native 475 * endian. 476 */ 477 switch (size) { 478 case 1: 479 virtio_config_writeb(vdev, addr, val); 480 break; 481 case 2: 482 if (virtio_is_big_endian(vdev)) { 483 val = bswap16(val); 484 } 485 virtio_config_writew(vdev, addr, val); 486 break; 487 case 4: 488 if (virtio_is_big_endian(vdev)) { 489 val = bswap32(val); 490 } 491 virtio_config_writel(vdev, addr, val); 492 break; 493 } 494 } 495 496 static const MemoryRegionOps virtio_pci_config_ops = { 497 .read = virtio_pci_config_read, 498 .write = virtio_pci_config_write, 499 .impl = { 500 .min_access_size = 1, 501 .max_access_size = 4, 502 }, 503 .endianness = DEVICE_LITTLE_ENDIAN, 504 }; 505 506 static MemoryRegion *virtio_address_space_lookup(VirtIOPCIProxy *proxy, 507 hwaddr *off, int len) 508 { 509 int i; 510 VirtIOPCIRegion *reg; 511 512 for (i = 0; i < ARRAY_SIZE(proxy->regs); ++i) { 513 reg = &proxy->regs[i]; 514 if (*off >= reg->offset && 515 *off + len <= reg->offset + reg->size) { 516 *off -= reg->offset; 517 return ®->mr; 518 } 519 } 520 521 return NULL; 522 } 523 524 /* Below are generic functions to do memcpy from/to an address space, 525 * without byteswaps, with input validation. 526 * 527 * As regular address_space_* APIs all do some kind of byteswap at least for 528 * some host/target combinations, we are forced to explicitly convert to a 529 * known-endianness integer value. 530 * It doesn't really matter which endian format to go through, so the code 531 * below selects the endian that causes the least amount of work on the given 532 * host. 533 * 534 * Note: host pointer must be aligned. 535 */ 536 static 537 void virtio_address_space_write(VirtIOPCIProxy *proxy, hwaddr addr, 538 const uint8_t *buf, int len) 539 { 540 uint64_t val; 541 MemoryRegion *mr; 542 543 /* address_space_* APIs assume an aligned address. 544 * As address is under guest control, handle illegal values. 545 */ 546 addr &= ~(len - 1); 547 548 mr = virtio_address_space_lookup(proxy, &addr, len); 549 if (!mr) { 550 return; 551 } 552 553 /* Make sure caller aligned buf properly */ 554 assert(!(((uintptr_t)buf) & (len - 1))); 555 556 switch (len) { 557 case 1: 558 val = pci_get_byte(buf); 559 break; 560 case 2: 561 val = pci_get_word(buf); 562 break; 563 case 4: 564 val = pci_get_long(buf); 565 break; 566 default: 567 /* As length is under guest control, handle illegal values. */ 568 return; 569 } 570 memory_region_dispatch_write(mr, addr, val, size_memop(len) | MO_LE, 571 MEMTXATTRS_UNSPECIFIED); 572 } 573 574 static void 575 virtio_address_space_read(VirtIOPCIProxy *proxy, hwaddr addr, 576 uint8_t *buf, int len) 577 { 578 uint64_t val; 579 MemoryRegion *mr; 580 581 /* address_space_* APIs assume an aligned address. 582 * As address is under guest control, handle illegal values. 583 */ 584 addr &= ~(len - 1); 585 586 mr = virtio_address_space_lookup(proxy, &addr, len); 587 if (!mr) { 588 return; 589 } 590 591 /* Make sure caller aligned buf properly */ 592 assert(!(((uintptr_t)buf) & (len - 1))); 593 594 memory_region_dispatch_read(mr, addr, &val, size_memop(len) | MO_LE, 595 MEMTXATTRS_UNSPECIFIED); 596 switch (len) { 597 case 1: 598 pci_set_byte(buf, val); 599 break; 600 case 2: 601 pci_set_word(buf, val); 602 break; 603 case 4: 604 pci_set_long(buf, val); 605 break; 606 default: 607 /* As length is under guest control, handle illegal values. */ 608 break; 609 } 610 } 611 612 static void virtio_write_config(PCIDevice *pci_dev, uint32_t address, 613 uint32_t val, int len) 614 { 615 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); 616 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 617 struct virtio_pci_cfg_cap *cfg; 618 619 pci_default_write_config(pci_dev, address, val, len); 620 621 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_FLR) { 622 pcie_cap_flr_write_config(pci_dev, address, val, len); 623 } 624 625 if (range_covers_byte(address, len, PCI_COMMAND)) { 626 if (!(pci_dev->config[PCI_COMMAND] & PCI_COMMAND_MASTER)) { 627 virtio_set_disabled(vdev, true); 628 virtio_pci_stop_ioeventfd(proxy); 629 virtio_set_status(vdev, vdev->status & ~VIRTIO_CONFIG_S_DRIVER_OK); 630 } else { 631 virtio_set_disabled(vdev, false); 632 } 633 } 634 635 if (proxy->config_cap && 636 ranges_overlap(address, len, proxy->config_cap + offsetof(struct virtio_pci_cfg_cap, 637 pci_cfg_data), 638 sizeof cfg->pci_cfg_data)) { 639 uint32_t off; 640 uint32_t len; 641 642 cfg = (void *)(proxy->pci_dev.config + proxy->config_cap); 643 off = le32_to_cpu(cfg->cap.offset); 644 len = le32_to_cpu(cfg->cap.length); 645 646 if (len == 1 || len == 2 || len == 4) { 647 assert(len <= sizeof cfg->pci_cfg_data); 648 virtio_address_space_write(proxy, off, cfg->pci_cfg_data, len); 649 } 650 } 651 } 652 653 static uint32_t virtio_read_config(PCIDevice *pci_dev, 654 uint32_t address, int len) 655 { 656 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); 657 struct virtio_pci_cfg_cap *cfg; 658 659 if (proxy->config_cap && 660 ranges_overlap(address, len, proxy->config_cap + offsetof(struct virtio_pci_cfg_cap, 661 pci_cfg_data), 662 sizeof cfg->pci_cfg_data)) { 663 uint32_t off; 664 uint32_t len; 665 666 cfg = (void *)(proxy->pci_dev.config + proxy->config_cap); 667 off = le32_to_cpu(cfg->cap.offset); 668 len = le32_to_cpu(cfg->cap.length); 669 670 if (len == 1 || len == 2 || len == 4) { 671 assert(len <= sizeof cfg->pci_cfg_data); 672 virtio_address_space_read(proxy, off, cfg->pci_cfg_data, len); 673 } 674 } 675 676 return pci_default_read_config(pci_dev, address, len); 677 } 678 679 static int kvm_virtio_pci_vq_vector_use(VirtIOPCIProxy *proxy, 680 unsigned int queue_no, 681 unsigned int vector) 682 { 683 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector]; 684 int ret; 685 686 if (irqfd->users == 0) { 687 KVMRouteChange c = kvm_irqchip_begin_route_changes(kvm_state); 688 ret = kvm_irqchip_add_msi_route(&c, vector, &proxy->pci_dev); 689 if (ret < 0) { 690 return ret; 691 } 692 kvm_irqchip_commit_route_changes(&c); 693 irqfd->virq = ret; 694 } 695 irqfd->users++; 696 return 0; 697 } 698 699 static void kvm_virtio_pci_vq_vector_release(VirtIOPCIProxy *proxy, 700 unsigned int vector) 701 { 702 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector]; 703 if (--irqfd->users == 0) { 704 kvm_irqchip_release_virq(kvm_state, irqfd->virq); 705 } 706 } 707 708 static int kvm_virtio_pci_irqfd_use(VirtIOPCIProxy *proxy, 709 unsigned int queue_no, 710 unsigned int vector) 711 { 712 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector]; 713 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 714 VirtQueue *vq = virtio_get_queue(vdev, queue_no); 715 EventNotifier *n = virtio_queue_get_guest_notifier(vq); 716 return kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, n, NULL, irqfd->virq); 717 } 718 719 static void kvm_virtio_pci_irqfd_release(VirtIOPCIProxy *proxy, 720 unsigned int queue_no, 721 unsigned int vector) 722 { 723 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 724 VirtQueue *vq = virtio_get_queue(vdev, queue_no); 725 EventNotifier *n = virtio_queue_get_guest_notifier(vq); 726 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector]; 727 int ret; 728 729 ret = kvm_irqchip_remove_irqfd_notifier_gsi(kvm_state, n, irqfd->virq); 730 assert(ret == 0); 731 } 732 733 static int kvm_virtio_pci_vector_use(VirtIOPCIProxy *proxy, int nvqs) 734 { 735 PCIDevice *dev = &proxy->pci_dev; 736 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 737 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 738 unsigned int vector; 739 int ret, queue_no; 740 741 for (queue_no = 0; queue_no < nvqs; queue_no++) { 742 if (!virtio_queue_get_num(vdev, queue_no)) { 743 break; 744 } 745 vector = virtio_queue_vector(vdev, queue_no); 746 if (vector >= msix_nr_vectors_allocated(dev)) { 747 continue; 748 } 749 ret = kvm_virtio_pci_vq_vector_use(proxy, queue_no, vector); 750 if (ret < 0) { 751 goto undo; 752 } 753 /* If guest supports masking, set up irqfd now. 754 * Otherwise, delay until unmasked in the frontend. 755 */ 756 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) { 757 ret = kvm_virtio_pci_irqfd_use(proxy, queue_no, vector); 758 if (ret < 0) { 759 kvm_virtio_pci_vq_vector_release(proxy, vector); 760 goto undo; 761 } 762 } 763 } 764 return 0; 765 766 undo: 767 while (--queue_no >= 0) { 768 vector = virtio_queue_vector(vdev, queue_no); 769 if (vector >= msix_nr_vectors_allocated(dev)) { 770 continue; 771 } 772 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) { 773 kvm_virtio_pci_irqfd_release(proxy, queue_no, vector); 774 } 775 kvm_virtio_pci_vq_vector_release(proxy, vector); 776 } 777 return ret; 778 } 779 780 static void kvm_virtio_pci_vector_release(VirtIOPCIProxy *proxy, int nvqs) 781 { 782 PCIDevice *dev = &proxy->pci_dev; 783 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 784 unsigned int vector; 785 int queue_no; 786 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 787 788 for (queue_no = 0; queue_no < nvqs; queue_no++) { 789 if (!virtio_queue_get_num(vdev, queue_no)) { 790 break; 791 } 792 vector = virtio_queue_vector(vdev, queue_no); 793 if (vector >= msix_nr_vectors_allocated(dev)) { 794 continue; 795 } 796 /* If guest supports masking, clean up irqfd now. 797 * Otherwise, it was cleaned when masked in the frontend. 798 */ 799 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) { 800 kvm_virtio_pci_irqfd_release(proxy, queue_no, vector); 801 } 802 kvm_virtio_pci_vq_vector_release(proxy, vector); 803 } 804 } 805 806 static int virtio_pci_vq_vector_unmask(VirtIOPCIProxy *proxy, 807 unsigned int queue_no, 808 unsigned int vector, 809 MSIMessage msg) 810 { 811 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 812 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 813 VirtQueue *vq = virtio_get_queue(vdev, queue_no); 814 EventNotifier *n = virtio_queue_get_guest_notifier(vq); 815 VirtIOIRQFD *irqfd; 816 int ret = 0; 817 818 if (proxy->vector_irqfd) { 819 irqfd = &proxy->vector_irqfd[vector]; 820 if (irqfd->msg.data != msg.data || irqfd->msg.address != msg.address) { 821 ret = kvm_irqchip_update_msi_route(kvm_state, irqfd->virq, msg, 822 &proxy->pci_dev); 823 if (ret < 0) { 824 return ret; 825 } 826 kvm_irqchip_commit_routes(kvm_state); 827 } 828 } 829 830 /* If guest supports masking, irqfd is already setup, unmask it. 831 * Otherwise, set it up now. 832 */ 833 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) { 834 k->guest_notifier_mask(vdev, queue_no, false); 835 /* Test after unmasking to avoid losing events. */ 836 if (k->guest_notifier_pending && 837 k->guest_notifier_pending(vdev, queue_no)) { 838 event_notifier_set(n); 839 } 840 } else { 841 ret = kvm_virtio_pci_irqfd_use(proxy, queue_no, vector); 842 } 843 return ret; 844 } 845 846 static void virtio_pci_vq_vector_mask(VirtIOPCIProxy *proxy, 847 unsigned int queue_no, 848 unsigned int vector) 849 { 850 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 851 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 852 853 /* If guest supports masking, keep irqfd but mask it. 854 * Otherwise, clean it up now. 855 */ 856 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) { 857 k->guest_notifier_mask(vdev, queue_no, true); 858 } else { 859 kvm_virtio_pci_irqfd_release(proxy, queue_no, vector); 860 } 861 } 862 863 static int virtio_pci_vector_unmask(PCIDevice *dev, unsigned vector, 864 MSIMessage msg) 865 { 866 VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev); 867 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 868 VirtQueue *vq = virtio_vector_first_queue(vdev, vector); 869 int ret, index, unmasked = 0; 870 871 while (vq) { 872 index = virtio_get_queue_index(vq); 873 if (!virtio_queue_get_num(vdev, index)) { 874 break; 875 } 876 if (index < proxy->nvqs_with_notifiers) { 877 ret = virtio_pci_vq_vector_unmask(proxy, index, vector, msg); 878 if (ret < 0) { 879 goto undo; 880 } 881 ++unmasked; 882 } 883 vq = virtio_vector_next_queue(vq); 884 } 885 886 return 0; 887 888 undo: 889 vq = virtio_vector_first_queue(vdev, vector); 890 while (vq && unmasked >= 0) { 891 index = virtio_get_queue_index(vq); 892 if (index < proxy->nvqs_with_notifiers) { 893 virtio_pci_vq_vector_mask(proxy, index, vector); 894 --unmasked; 895 } 896 vq = virtio_vector_next_queue(vq); 897 } 898 return ret; 899 } 900 901 static void virtio_pci_vector_mask(PCIDevice *dev, unsigned vector) 902 { 903 VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev); 904 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 905 VirtQueue *vq = virtio_vector_first_queue(vdev, vector); 906 int index; 907 908 while (vq) { 909 index = virtio_get_queue_index(vq); 910 if (!virtio_queue_get_num(vdev, index)) { 911 break; 912 } 913 if (index < proxy->nvqs_with_notifiers) { 914 virtio_pci_vq_vector_mask(proxy, index, vector); 915 } 916 vq = virtio_vector_next_queue(vq); 917 } 918 } 919 920 static void virtio_pci_vector_poll(PCIDevice *dev, 921 unsigned int vector_start, 922 unsigned int vector_end) 923 { 924 VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev); 925 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 926 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 927 int queue_no; 928 unsigned int vector; 929 EventNotifier *notifier; 930 VirtQueue *vq; 931 932 for (queue_no = 0; queue_no < proxy->nvqs_with_notifiers; queue_no++) { 933 if (!virtio_queue_get_num(vdev, queue_no)) { 934 break; 935 } 936 vector = virtio_queue_vector(vdev, queue_no); 937 if (vector < vector_start || vector >= vector_end || 938 !msix_is_masked(dev, vector)) { 939 continue; 940 } 941 vq = virtio_get_queue(vdev, queue_no); 942 notifier = virtio_queue_get_guest_notifier(vq); 943 if (k->guest_notifier_pending) { 944 if (k->guest_notifier_pending(vdev, queue_no)) { 945 msix_set_pending(dev, vector); 946 } 947 } else if (event_notifier_test_and_clear(notifier)) { 948 msix_set_pending(dev, vector); 949 } 950 } 951 } 952 953 static int virtio_pci_set_guest_notifier(DeviceState *d, int n, bool assign, 954 bool with_irqfd) 955 { 956 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 957 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 958 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev); 959 VirtQueue *vq = virtio_get_queue(vdev, n); 960 EventNotifier *notifier = virtio_queue_get_guest_notifier(vq); 961 962 if (assign) { 963 int r = event_notifier_init(notifier, 0); 964 if (r < 0) { 965 return r; 966 } 967 virtio_queue_set_guest_notifier_fd_handler(vq, true, with_irqfd); 968 } else { 969 virtio_queue_set_guest_notifier_fd_handler(vq, false, with_irqfd); 970 event_notifier_cleanup(notifier); 971 } 972 973 if (!msix_enabled(&proxy->pci_dev) && 974 vdev->use_guest_notifier_mask && 975 vdc->guest_notifier_mask) { 976 vdc->guest_notifier_mask(vdev, n, !assign); 977 } 978 979 return 0; 980 } 981 982 static bool virtio_pci_query_guest_notifiers(DeviceState *d) 983 { 984 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 985 return msix_enabled(&proxy->pci_dev); 986 } 987 988 static int virtio_pci_set_guest_notifiers(DeviceState *d, int nvqs, bool assign) 989 { 990 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 991 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 992 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev); 993 int r, n; 994 bool with_irqfd = msix_enabled(&proxy->pci_dev) && 995 kvm_msi_via_irqfd_enabled(); 996 997 nvqs = MIN(nvqs, VIRTIO_QUEUE_MAX); 998 999 /* When deassigning, pass a consistent nvqs value 1000 * to avoid leaking notifiers. 1001 */ 1002 assert(assign || nvqs == proxy->nvqs_with_notifiers); 1003 1004 proxy->nvqs_with_notifiers = nvqs; 1005 1006 /* Must unset vector notifier while guest notifier is still assigned */ 1007 if ((proxy->vector_irqfd || k->guest_notifier_mask) && !assign) { 1008 msix_unset_vector_notifiers(&proxy->pci_dev); 1009 if (proxy->vector_irqfd) { 1010 kvm_virtio_pci_vector_release(proxy, nvqs); 1011 g_free(proxy->vector_irqfd); 1012 proxy->vector_irqfd = NULL; 1013 } 1014 } 1015 1016 for (n = 0; n < nvqs; n++) { 1017 if (!virtio_queue_get_num(vdev, n)) { 1018 break; 1019 } 1020 1021 r = virtio_pci_set_guest_notifier(d, n, assign, with_irqfd); 1022 if (r < 0) { 1023 goto assign_error; 1024 } 1025 } 1026 1027 /* Must set vector notifier after guest notifier has been assigned */ 1028 if ((with_irqfd || k->guest_notifier_mask) && assign) { 1029 if (with_irqfd) { 1030 proxy->vector_irqfd = 1031 g_malloc0(sizeof(*proxy->vector_irqfd) * 1032 msix_nr_vectors_allocated(&proxy->pci_dev)); 1033 r = kvm_virtio_pci_vector_use(proxy, nvqs); 1034 if (r < 0) { 1035 goto assign_error; 1036 } 1037 } 1038 r = msix_set_vector_notifiers(&proxy->pci_dev, 1039 virtio_pci_vector_unmask, 1040 virtio_pci_vector_mask, 1041 virtio_pci_vector_poll); 1042 if (r < 0) { 1043 goto notifiers_error; 1044 } 1045 } 1046 1047 return 0; 1048 1049 notifiers_error: 1050 if (with_irqfd) { 1051 assert(assign); 1052 kvm_virtio_pci_vector_release(proxy, nvqs); 1053 } 1054 1055 assign_error: 1056 /* We get here on assignment failure. Recover by undoing for VQs 0 .. n. */ 1057 assert(assign); 1058 while (--n >= 0) { 1059 virtio_pci_set_guest_notifier(d, n, !assign, with_irqfd); 1060 } 1061 return r; 1062 } 1063 1064 static int virtio_pci_set_host_notifier_mr(DeviceState *d, int n, 1065 MemoryRegion *mr, bool assign) 1066 { 1067 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 1068 int offset; 1069 1070 if (n >= VIRTIO_QUEUE_MAX || !virtio_pci_modern(proxy) || 1071 virtio_pci_queue_mem_mult(proxy) != memory_region_size(mr)) { 1072 return -1; 1073 } 1074 1075 if (assign) { 1076 offset = virtio_pci_queue_mem_mult(proxy) * n; 1077 memory_region_add_subregion_overlap(&proxy->notify.mr, offset, mr, 1); 1078 } else { 1079 memory_region_del_subregion(&proxy->notify.mr, mr); 1080 } 1081 1082 return 0; 1083 } 1084 1085 static void virtio_pci_vmstate_change(DeviceState *d, bool running) 1086 { 1087 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d); 1088 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1089 1090 if (running) { 1091 /* Old QEMU versions did not set bus master enable on status write. 1092 * Detect DRIVER set and enable it. 1093 */ 1094 if ((proxy->flags & VIRTIO_PCI_FLAG_BUS_MASTER_BUG_MIGRATION) && 1095 (vdev->status & VIRTIO_CONFIG_S_DRIVER) && 1096 !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) { 1097 pci_default_write_config(&proxy->pci_dev, PCI_COMMAND, 1098 proxy->pci_dev.config[PCI_COMMAND] | 1099 PCI_COMMAND_MASTER, 1); 1100 } 1101 virtio_pci_start_ioeventfd(proxy); 1102 } else { 1103 virtio_pci_stop_ioeventfd(proxy); 1104 } 1105 } 1106 1107 /* 1108 * virtio-pci: This is the PCIDevice which has a virtio-pci-bus. 1109 */ 1110 1111 static int virtio_pci_query_nvectors(DeviceState *d) 1112 { 1113 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1114 1115 return proxy->nvectors; 1116 } 1117 1118 static AddressSpace *virtio_pci_get_dma_as(DeviceState *d) 1119 { 1120 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1121 PCIDevice *dev = &proxy->pci_dev; 1122 1123 return pci_get_address_space(dev); 1124 } 1125 1126 static bool virtio_pci_iommu_enabled(DeviceState *d) 1127 { 1128 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1129 PCIDevice *dev = &proxy->pci_dev; 1130 AddressSpace *dma_as = pci_device_iommu_address_space(dev); 1131 1132 if (dma_as == &address_space_memory) { 1133 return false; 1134 } 1135 1136 return true; 1137 } 1138 1139 static bool virtio_pci_queue_enabled(DeviceState *d, int n) 1140 { 1141 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1142 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1143 1144 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) { 1145 return proxy->vqs[n].enabled; 1146 } 1147 1148 return virtio_queue_enabled_legacy(vdev, n); 1149 } 1150 1151 static int virtio_pci_add_mem_cap(VirtIOPCIProxy *proxy, 1152 struct virtio_pci_cap *cap) 1153 { 1154 PCIDevice *dev = &proxy->pci_dev; 1155 int offset; 1156 1157 offset = pci_add_capability(dev, PCI_CAP_ID_VNDR, 0, 1158 cap->cap_len, &error_abort); 1159 1160 assert(cap->cap_len >= sizeof *cap); 1161 memcpy(dev->config + offset + PCI_CAP_FLAGS, &cap->cap_len, 1162 cap->cap_len - PCI_CAP_FLAGS); 1163 1164 return offset; 1165 } 1166 1167 static uint64_t virtio_pci_common_read(void *opaque, hwaddr addr, 1168 unsigned size) 1169 { 1170 VirtIOPCIProxy *proxy = opaque; 1171 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1172 uint32_t val = 0; 1173 int i; 1174 1175 if (vdev == NULL) { 1176 return UINT64_MAX; 1177 } 1178 1179 switch (addr) { 1180 case VIRTIO_PCI_COMMON_DFSELECT: 1181 val = proxy->dfselect; 1182 break; 1183 case VIRTIO_PCI_COMMON_DF: 1184 if (proxy->dfselect <= 1) { 1185 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev); 1186 1187 val = (vdev->host_features & ~vdc->legacy_features) >> 1188 (32 * proxy->dfselect); 1189 } 1190 break; 1191 case VIRTIO_PCI_COMMON_GFSELECT: 1192 val = proxy->gfselect; 1193 break; 1194 case VIRTIO_PCI_COMMON_GF: 1195 if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) { 1196 val = proxy->guest_features[proxy->gfselect]; 1197 } 1198 break; 1199 case VIRTIO_PCI_COMMON_MSIX: 1200 val = vdev->config_vector; 1201 break; 1202 case VIRTIO_PCI_COMMON_NUMQ: 1203 for (i = 0; i < VIRTIO_QUEUE_MAX; ++i) { 1204 if (virtio_queue_get_num(vdev, i)) { 1205 val = i + 1; 1206 } 1207 } 1208 break; 1209 case VIRTIO_PCI_COMMON_STATUS: 1210 val = vdev->status; 1211 break; 1212 case VIRTIO_PCI_COMMON_CFGGENERATION: 1213 val = vdev->generation; 1214 break; 1215 case VIRTIO_PCI_COMMON_Q_SELECT: 1216 val = vdev->queue_sel; 1217 break; 1218 case VIRTIO_PCI_COMMON_Q_SIZE: 1219 val = virtio_queue_get_num(vdev, vdev->queue_sel); 1220 break; 1221 case VIRTIO_PCI_COMMON_Q_MSIX: 1222 val = virtio_queue_vector(vdev, vdev->queue_sel); 1223 break; 1224 case VIRTIO_PCI_COMMON_Q_ENABLE: 1225 val = proxy->vqs[vdev->queue_sel].enabled; 1226 break; 1227 case VIRTIO_PCI_COMMON_Q_NOFF: 1228 /* Simply map queues in order */ 1229 val = vdev->queue_sel; 1230 break; 1231 case VIRTIO_PCI_COMMON_Q_DESCLO: 1232 val = proxy->vqs[vdev->queue_sel].desc[0]; 1233 break; 1234 case VIRTIO_PCI_COMMON_Q_DESCHI: 1235 val = proxy->vqs[vdev->queue_sel].desc[1]; 1236 break; 1237 case VIRTIO_PCI_COMMON_Q_AVAILLO: 1238 val = proxy->vqs[vdev->queue_sel].avail[0]; 1239 break; 1240 case VIRTIO_PCI_COMMON_Q_AVAILHI: 1241 val = proxy->vqs[vdev->queue_sel].avail[1]; 1242 break; 1243 case VIRTIO_PCI_COMMON_Q_USEDLO: 1244 val = proxy->vqs[vdev->queue_sel].used[0]; 1245 break; 1246 case VIRTIO_PCI_COMMON_Q_USEDHI: 1247 val = proxy->vqs[vdev->queue_sel].used[1]; 1248 break; 1249 default: 1250 val = 0; 1251 } 1252 1253 return val; 1254 } 1255 1256 static void virtio_pci_common_write(void *opaque, hwaddr addr, 1257 uint64_t val, unsigned size) 1258 { 1259 VirtIOPCIProxy *proxy = opaque; 1260 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1261 1262 if (vdev == NULL) { 1263 return; 1264 } 1265 1266 switch (addr) { 1267 case VIRTIO_PCI_COMMON_DFSELECT: 1268 proxy->dfselect = val; 1269 break; 1270 case VIRTIO_PCI_COMMON_GFSELECT: 1271 proxy->gfselect = val; 1272 break; 1273 case VIRTIO_PCI_COMMON_GF: 1274 if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) { 1275 proxy->guest_features[proxy->gfselect] = val; 1276 virtio_set_features(vdev, 1277 (((uint64_t)proxy->guest_features[1]) << 32) | 1278 proxy->guest_features[0]); 1279 } 1280 break; 1281 case VIRTIO_PCI_COMMON_MSIX: 1282 msix_vector_unuse(&proxy->pci_dev, vdev->config_vector); 1283 /* Make it possible for guest to discover an error took place. */ 1284 if (msix_vector_use(&proxy->pci_dev, val) < 0) { 1285 val = VIRTIO_NO_VECTOR; 1286 } 1287 vdev->config_vector = val; 1288 break; 1289 case VIRTIO_PCI_COMMON_STATUS: 1290 if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) { 1291 virtio_pci_stop_ioeventfd(proxy); 1292 } 1293 1294 virtio_set_status(vdev, val & 0xFF); 1295 1296 if (val & VIRTIO_CONFIG_S_DRIVER_OK) { 1297 virtio_pci_start_ioeventfd(proxy); 1298 } 1299 1300 if (vdev->status == 0) { 1301 virtio_pci_reset(DEVICE(proxy)); 1302 } 1303 1304 break; 1305 case VIRTIO_PCI_COMMON_Q_SELECT: 1306 if (val < VIRTIO_QUEUE_MAX) { 1307 vdev->queue_sel = val; 1308 } 1309 break; 1310 case VIRTIO_PCI_COMMON_Q_SIZE: 1311 proxy->vqs[vdev->queue_sel].num = val; 1312 virtio_queue_set_num(vdev, vdev->queue_sel, 1313 proxy->vqs[vdev->queue_sel].num); 1314 break; 1315 case VIRTIO_PCI_COMMON_Q_MSIX: 1316 msix_vector_unuse(&proxy->pci_dev, 1317 virtio_queue_vector(vdev, vdev->queue_sel)); 1318 /* Make it possible for guest to discover an error took place. */ 1319 if (msix_vector_use(&proxy->pci_dev, val) < 0) { 1320 val = VIRTIO_NO_VECTOR; 1321 } 1322 virtio_queue_set_vector(vdev, vdev->queue_sel, val); 1323 break; 1324 case VIRTIO_PCI_COMMON_Q_ENABLE: 1325 if (val == 1) { 1326 virtio_queue_set_num(vdev, vdev->queue_sel, 1327 proxy->vqs[vdev->queue_sel].num); 1328 virtio_queue_set_rings(vdev, vdev->queue_sel, 1329 ((uint64_t)proxy->vqs[vdev->queue_sel].desc[1]) << 32 | 1330 proxy->vqs[vdev->queue_sel].desc[0], 1331 ((uint64_t)proxy->vqs[vdev->queue_sel].avail[1]) << 32 | 1332 proxy->vqs[vdev->queue_sel].avail[0], 1333 ((uint64_t)proxy->vqs[vdev->queue_sel].used[1]) << 32 | 1334 proxy->vqs[vdev->queue_sel].used[0]); 1335 proxy->vqs[vdev->queue_sel].enabled = 1; 1336 } else { 1337 virtio_error(vdev, "wrong value for queue_enable %"PRIx64, val); 1338 } 1339 break; 1340 case VIRTIO_PCI_COMMON_Q_DESCLO: 1341 proxy->vqs[vdev->queue_sel].desc[0] = val; 1342 break; 1343 case VIRTIO_PCI_COMMON_Q_DESCHI: 1344 proxy->vqs[vdev->queue_sel].desc[1] = val; 1345 break; 1346 case VIRTIO_PCI_COMMON_Q_AVAILLO: 1347 proxy->vqs[vdev->queue_sel].avail[0] = val; 1348 break; 1349 case VIRTIO_PCI_COMMON_Q_AVAILHI: 1350 proxy->vqs[vdev->queue_sel].avail[1] = val; 1351 break; 1352 case VIRTIO_PCI_COMMON_Q_USEDLO: 1353 proxy->vqs[vdev->queue_sel].used[0] = val; 1354 break; 1355 case VIRTIO_PCI_COMMON_Q_USEDHI: 1356 proxy->vqs[vdev->queue_sel].used[1] = val; 1357 break; 1358 default: 1359 break; 1360 } 1361 } 1362 1363 1364 static uint64_t virtio_pci_notify_read(void *opaque, hwaddr addr, 1365 unsigned size) 1366 { 1367 VirtIOPCIProxy *proxy = opaque; 1368 if (virtio_bus_get_device(&proxy->bus) == NULL) { 1369 return UINT64_MAX; 1370 } 1371 1372 return 0; 1373 } 1374 1375 static void virtio_pci_notify_write(void *opaque, hwaddr addr, 1376 uint64_t val, unsigned size) 1377 { 1378 VirtIOPCIProxy *proxy = opaque; 1379 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1380 1381 unsigned queue = addr / virtio_pci_queue_mem_mult(proxy); 1382 1383 if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) { 1384 trace_virtio_pci_notify_write(addr, val, size); 1385 virtio_queue_notify(vdev, queue); 1386 } 1387 } 1388 1389 static void virtio_pci_notify_write_pio(void *opaque, hwaddr addr, 1390 uint64_t val, unsigned size) 1391 { 1392 VirtIOPCIProxy *proxy = opaque; 1393 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1394 1395 unsigned queue = val; 1396 1397 if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) { 1398 trace_virtio_pci_notify_write_pio(addr, val, size); 1399 virtio_queue_notify(vdev, queue); 1400 } 1401 } 1402 1403 static uint64_t virtio_pci_isr_read(void *opaque, hwaddr addr, 1404 unsigned size) 1405 { 1406 VirtIOPCIProxy *proxy = opaque; 1407 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1408 uint64_t val; 1409 1410 if (vdev == NULL) { 1411 return UINT64_MAX; 1412 } 1413 1414 val = qatomic_xchg(&vdev->isr, 0); 1415 pci_irq_deassert(&proxy->pci_dev); 1416 return val; 1417 } 1418 1419 static void virtio_pci_isr_write(void *opaque, hwaddr addr, 1420 uint64_t val, unsigned size) 1421 { 1422 } 1423 1424 static uint64_t virtio_pci_device_read(void *opaque, hwaddr addr, 1425 unsigned size) 1426 { 1427 VirtIOPCIProxy *proxy = opaque; 1428 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1429 uint64_t val; 1430 1431 if (vdev == NULL) { 1432 return UINT64_MAX; 1433 } 1434 1435 switch (size) { 1436 case 1: 1437 val = virtio_config_modern_readb(vdev, addr); 1438 break; 1439 case 2: 1440 val = virtio_config_modern_readw(vdev, addr); 1441 break; 1442 case 4: 1443 val = virtio_config_modern_readl(vdev, addr); 1444 break; 1445 default: 1446 val = 0; 1447 break; 1448 } 1449 return val; 1450 } 1451 1452 static void virtio_pci_device_write(void *opaque, hwaddr addr, 1453 uint64_t val, unsigned size) 1454 { 1455 VirtIOPCIProxy *proxy = opaque; 1456 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1457 1458 if (vdev == NULL) { 1459 return; 1460 } 1461 1462 switch (size) { 1463 case 1: 1464 virtio_config_modern_writeb(vdev, addr, val); 1465 break; 1466 case 2: 1467 virtio_config_modern_writew(vdev, addr, val); 1468 break; 1469 case 4: 1470 virtio_config_modern_writel(vdev, addr, val); 1471 break; 1472 } 1473 } 1474 1475 static void virtio_pci_modern_regions_init(VirtIOPCIProxy *proxy, 1476 const char *vdev_name) 1477 { 1478 static const MemoryRegionOps common_ops = { 1479 .read = virtio_pci_common_read, 1480 .write = virtio_pci_common_write, 1481 .impl = { 1482 .min_access_size = 1, 1483 .max_access_size = 4, 1484 }, 1485 .endianness = DEVICE_LITTLE_ENDIAN, 1486 }; 1487 static const MemoryRegionOps isr_ops = { 1488 .read = virtio_pci_isr_read, 1489 .write = virtio_pci_isr_write, 1490 .impl = { 1491 .min_access_size = 1, 1492 .max_access_size = 4, 1493 }, 1494 .endianness = DEVICE_LITTLE_ENDIAN, 1495 }; 1496 static const MemoryRegionOps device_ops = { 1497 .read = virtio_pci_device_read, 1498 .write = virtio_pci_device_write, 1499 .impl = { 1500 .min_access_size = 1, 1501 .max_access_size = 4, 1502 }, 1503 .endianness = DEVICE_LITTLE_ENDIAN, 1504 }; 1505 static const MemoryRegionOps notify_ops = { 1506 .read = virtio_pci_notify_read, 1507 .write = virtio_pci_notify_write, 1508 .impl = { 1509 .min_access_size = 1, 1510 .max_access_size = 4, 1511 }, 1512 .endianness = DEVICE_LITTLE_ENDIAN, 1513 }; 1514 static const MemoryRegionOps notify_pio_ops = { 1515 .read = virtio_pci_notify_read, 1516 .write = virtio_pci_notify_write_pio, 1517 .impl = { 1518 .min_access_size = 1, 1519 .max_access_size = 4, 1520 }, 1521 .endianness = DEVICE_LITTLE_ENDIAN, 1522 }; 1523 g_autoptr(GString) name = g_string_new(NULL); 1524 1525 g_string_printf(name, "virtio-pci-common-%s", vdev_name); 1526 memory_region_init_io(&proxy->common.mr, OBJECT(proxy), 1527 &common_ops, 1528 proxy, 1529 name->str, 1530 proxy->common.size); 1531 1532 g_string_printf(name, "virtio-pci-isr-%s", vdev_name); 1533 memory_region_init_io(&proxy->isr.mr, OBJECT(proxy), 1534 &isr_ops, 1535 proxy, 1536 name->str, 1537 proxy->isr.size); 1538 1539 g_string_printf(name, "virtio-pci-device-%s", vdev_name); 1540 memory_region_init_io(&proxy->device.mr, OBJECT(proxy), 1541 &device_ops, 1542 proxy, 1543 name->str, 1544 proxy->device.size); 1545 1546 g_string_printf(name, "virtio-pci-notify-%s", vdev_name); 1547 memory_region_init_io(&proxy->notify.mr, OBJECT(proxy), 1548 ¬ify_ops, 1549 proxy, 1550 name->str, 1551 proxy->notify.size); 1552 1553 g_string_printf(name, "virtio-pci-notify-pio-%s", vdev_name); 1554 memory_region_init_io(&proxy->notify_pio.mr, OBJECT(proxy), 1555 ¬ify_pio_ops, 1556 proxy, 1557 name->str, 1558 proxy->notify_pio.size); 1559 } 1560 1561 static void virtio_pci_modern_region_map(VirtIOPCIProxy *proxy, 1562 VirtIOPCIRegion *region, 1563 struct virtio_pci_cap *cap, 1564 MemoryRegion *mr, 1565 uint8_t bar) 1566 { 1567 memory_region_add_subregion(mr, region->offset, ®ion->mr); 1568 1569 cap->cfg_type = region->type; 1570 cap->bar = bar; 1571 cap->offset = cpu_to_le32(region->offset); 1572 cap->length = cpu_to_le32(region->size); 1573 virtio_pci_add_mem_cap(proxy, cap); 1574 1575 } 1576 1577 static void virtio_pci_modern_mem_region_map(VirtIOPCIProxy *proxy, 1578 VirtIOPCIRegion *region, 1579 struct virtio_pci_cap *cap) 1580 { 1581 virtio_pci_modern_region_map(proxy, region, cap, 1582 &proxy->modern_bar, proxy->modern_mem_bar_idx); 1583 } 1584 1585 static void virtio_pci_modern_io_region_map(VirtIOPCIProxy *proxy, 1586 VirtIOPCIRegion *region, 1587 struct virtio_pci_cap *cap) 1588 { 1589 virtio_pci_modern_region_map(proxy, region, cap, 1590 &proxy->io_bar, proxy->modern_io_bar_idx); 1591 } 1592 1593 static void virtio_pci_modern_mem_region_unmap(VirtIOPCIProxy *proxy, 1594 VirtIOPCIRegion *region) 1595 { 1596 memory_region_del_subregion(&proxy->modern_bar, 1597 ®ion->mr); 1598 } 1599 1600 static void virtio_pci_modern_io_region_unmap(VirtIOPCIProxy *proxy, 1601 VirtIOPCIRegion *region) 1602 { 1603 memory_region_del_subregion(&proxy->io_bar, 1604 ®ion->mr); 1605 } 1606 1607 static void virtio_pci_pre_plugged(DeviceState *d, Error **errp) 1608 { 1609 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1610 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1611 1612 if (virtio_pci_modern(proxy)) { 1613 virtio_add_feature(&vdev->host_features, VIRTIO_F_VERSION_1); 1614 } 1615 1616 virtio_add_feature(&vdev->host_features, VIRTIO_F_BAD_FEATURE); 1617 } 1618 1619 /* This is called by virtio-bus just after the device is plugged. */ 1620 static void virtio_pci_device_plugged(DeviceState *d, Error **errp) 1621 { 1622 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1623 VirtioBusState *bus = &proxy->bus; 1624 bool legacy = virtio_pci_legacy(proxy); 1625 bool modern; 1626 bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY; 1627 uint8_t *config; 1628 uint32_t size; 1629 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus); 1630 1631 /* 1632 * Virtio capabilities present without 1633 * VIRTIO_F_VERSION_1 confuses guests 1634 */ 1635 if (!proxy->ignore_backend_features && 1636 !virtio_has_feature(vdev->host_features, VIRTIO_F_VERSION_1)) { 1637 virtio_pci_disable_modern(proxy); 1638 1639 if (!legacy) { 1640 error_setg(errp, "Device doesn't support modern mode, and legacy" 1641 " mode is disabled"); 1642 error_append_hint(errp, "Set disable-legacy to off\n"); 1643 1644 return; 1645 } 1646 } 1647 1648 modern = virtio_pci_modern(proxy); 1649 1650 config = proxy->pci_dev.config; 1651 if (proxy->class_code) { 1652 pci_config_set_class(config, proxy->class_code); 1653 } 1654 1655 if (legacy) { 1656 if (!virtio_legacy_allowed(vdev)) { 1657 /* 1658 * To avoid migration issues, we allow legacy mode when legacy 1659 * check is disabled in the old machine types (< 5.1). 1660 */ 1661 if (virtio_legacy_check_disabled(vdev)) { 1662 warn_report("device is modern-only, but for backward " 1663 "compatibility legacy is allowed"); 1664 } else { 1665 error_setg(errp, 1666 "device is modern-only, use disable-legacy=on"); 1667 return; 1668 } 1669 } 1670 if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) { 1671 error_setg(errp, "VIRTIO_F_IOMMU_PLATFORM was supported by" 1672 " neither legacy nor transitional device"); 1673 return ; 1674 } 1675 /* 1676 * Legacy and transitional devices use specific subsystem IDs. 1677 * Note that the subsystem vendor ID (config + PCI_SUBSYSTEM_VENDOR_ID) 1678 * is set to PCI_SUBVENDOR_ID_REDHAT_QUMRANET by default. 1679 */ 1680 pci_set_word(config + PCI_SUBSYSTEM_ID, virtio_bus_get_vdev_id(bus)); 1681 } else { 1682 /* pure virtio-1.0 */ 1683 pci_set_word(config + PCI_VENDOR_ID, 1684 PCI_VENDOR_ID_REDHAT_QUMRANET); 1685 pci_set_word(config + PCI_DEVICE_ID, 1686 0x1040 + virtio_bus_get_vdev_id(bus)); 1687 pci_config_set_revision(config, 1); 1688 } 1689 config[PCI_INTERRUPT_PIN] = 1; 1690 1691 1692 if (modern) { 1693 struct virtio_pci_cap cap = { 1694 .cap_len = sizeof cap, 1695 }; 1696 struct virtio_pci_notify_cap notify = { 1697 .cap.cap_len = sizeof notify, 1698 .notify_off_multiplier = 1699 cpu_to_le32(virtio_pci_queue_mem_mult(proxy)), 1700 }; 1701 struct virtio_pci_cfg_cap cfg = { 1702 .cap.cap_len = sizeof cfg, 1703 .cap.cfg_type = VIRTIO_PCI_CAP_PCI_CFG, 1704 }; 1705 struct virtio_pci_notify_cap notify_pio = { 1706 .cap.cap_len = sizeof notify, 1707 .notify_off_multiplier = cpu_to_le32(0x0), 1708 }; 1709 1710 struct virtio_pci_cfg_cap *cfg_mask; 1711 1712 virtio_pci_modern_regions_init(proxy, vdev->name); 1713 1714 virtio_pci_modern_mem_region_map(proxy, &proxy->common, &cap); 1715 virtio_pci_modern_mem_region_map(proxy, &proxy->isr, &cap); 1716 virtio_pci_modern_mem_region_map(proxy, &proxy->device, &cap); 1717 virtio_pci_modern_mem_region_map(proxy, &proxy->notify, ¬ify.cap); 1718 1719 if (modern_pio) { 1720 memory_region_init(&proxy->io_bar, OBJECT(proxy), 1721 "virtio-pci-io", 0x4); 1722 1723 pci_register_bar(&proxy->pci_dev, proxy->modern_io_bar_idx, 1724 PCI_BASE_ADDRESS_SPACE_IO, &proxy->io_bar); 1725 1726 virtio_pci_modern_io_region_map(proxy, &proxy->notify_pio, 1727 ¬ify_pio.cap); 1728 } 1729 1730 pci_register_bar(&proxy->pci_dev, proxy->modern_mem_bar_idx, 1731 PCI_BASE_ADDRESS_SPACE_MEMORY | 1732 PCI_BASE_ADDRESS_MEM_PREFETCH | 1733 PCI_BASE_ADDRESS_MEM_TYPE_64, 1734 &proxy->modern_bar); 1735 1736 proxy->config_cap = virtio_pci_add_mem_cap(proxy, &cfg.cap); 1737 cfg_mask = (void *)(proxy->pci_dev.wmask + proxy->config_cap); 1738 pci_set_byte(&cfg_mask->cap.bar, ~0x0); 1739 pci_set_long((uint8_t *)&cfg_mask->cap.offset, ~0x0); 1740 pci_set_long((uint8_t *)&cfg_mask->cap.length, ~0x0); 1741 pci_set_long(cfg_mask->pci_cfg_data, ~0x0); 1742 } 1743 1744 if (proxy->nvectors) { 1745 int err = msix_init_exclusive_bar(&proxy->pci_dev, proxy->nvectors, 1746 proxy->msix_bar_idx, NULL); 1747 if (err) { 1748 /* Notice when a system that supports MSIx can't initialize it */ 1749 if (err != -ENOTSUP) { 1750 warn_report("unable to init msix vectors to %" PRIu32, 1751 proxy->nvectors); 1752 } 1753 proxy->nvectors = 0; 1754 } 1755 } 1756 1757 proxy->pci_dev.config_write = virtio_write_config; 1758 proxy->pci_dev.config_read = virtio_read_config; 1759 1760 if (legacy) { 1761 size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev) 1762 + virtio_bus_get_vdev_config_len(bus); 1763 size = pow2ceil(size); 1764 1765 memory_region_init_io(&proxy->bar, OBJECT(proxy), 1766 &virtio_pci_config_ops, 1767 proxy, "virtio-pci", size); 1768 1769 pci_register_bar(&proxy->pci_dev, proxy->legacy_io_bar_idx, 1770 PCI_BASE_ADDRESS_SPACE_IO, &proxy->bar); 1771 } 1772 } 1773 1774 static void virtio_pci_device_unplugged(DeviceState *d) 1775 { 1776 VirtIOPCIProxy *proxy = VIRTIO_PCI(d); 1777 bool modern = virtio_pci_modern(proxy); 1778 bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY; 1779 1780 virtio_pci_stop_ioeventfd(proxy); 1781 1782 if (modern) { 1783 virtio_pci_modern_mem_region_unmap(proxy, &proxy->common); 1784 virtio_pci_modern_mem_region_unmap(proxy, &proxy->isr); 1785 virtio_pci_modern_mem_region_unmap(proxy, &proxy->device); 1786 virtio_pci_modern_mem_region_unmap(proxy, &proxy->notify); 1787 if (modern_pio) { 1788 virtio_pci_modern_io_region_unmap(proxy, &proxy->notify_pio); 1789 } 1790 } 1791 } 1792 1793 static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp) 1794 { 1795 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); 1796 VirtioPCIClass *k = VIRTIO_PCI_GET_CLASS(pci_dev); 1797 bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) && 1798 !pci_bus_is_root(pci_get_bus(pci_dev)); 1799 1800 if (kvm_enabled() && !kvm_has_many_ioeventfds()) { 1801 proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD; 1802 } 1803 1804 /* fd-based ioevents can't be synchronized in record/replay */ 1805 if (replay_mode != REPLAY_MODE_NONE) { 1806 proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD; 1807 } 1808 1809 /* 1810 * virtio pci bar layout used by default. 1811 * subclasses can re-arrange things if needed. 1812 * 1813 * region 0 -- virtio legacy io bar 1814 * region 1 -- msi-x bar 1815 * region 2 -- virtio modern io bar (off by default) 1816 * region 4+5 -- virtio modern memory (64bit) bar 1817 * 1818 */ 1819 proxy->legacy_io_bar_idx = 0; 1820 proxy->msix_bar_idx = 1; 1821 proxy->modern_io_bar_idx = 2; 1822 proxy->modern_mem_bar_idx = 4; 1823 1824 proxy->common.offset = 0x0; 1825 proxy->common.size = 0x1000; 1826 proxy->common.type = VIRTIO_PCI_CAP_COMMON_CFG; 1827 1828 proxy->isr.offset = 0x1000; 1829 proxy->isr.size = 0x1000; 1830 proxy->isr.type = VIRTIO_PCI_CAP_ISR_CFG; 1831 1832 proxy->device.offset = 0x2000; 1833 proxy->device.size = 0x1000; 1834 proxy->device.type = VIRTIO_PCI_CAP_DEVICE_CFG; 1835 1836 proxy->notify.offset = 0x3000; 1837 proxy->notify.size = virtio_pci_queue_mem_mult(proxy) * VIRTIO_QUEUE_MAX; 1838 proxy->notify.type = VIRTIO_PCI_CAP_NOTIFY_CFG; 1839 1840 proxy->notify_pio.offset = 0x0; 1841 proxy->notify_pio.size = 0x4; 1842 proxy->notify_pio.type = VIRTIO_PCI_CAP_NOTIFY_CFG; 1843 1844 /* subclasses can enforce modern, so do this unconditionally */ 1845 memory_region_init(&proxy->modern_bar, OBJECT(proxy), "virtio-pci", 1846 /* PCI BAR regions must be powers of 2 */ 1847 pow2ceil(proxy->notify.offset + proxy->notify.size)); 1848 1849 if (proxy->disable_legacy == ON_OFF_AUTO_AUTO) { 1850 proxy->disable_legacy = pcie_port ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF; 1851 } 1852 1853 if (!virtio_pci_modern(proxy) && !virtio_pci_legacy(proxy)) { 1854 error_setg(errp, "device cannot work as neither modern nor legacy mode" 1855 " is enabled"); 1856 error_append_hint(errp, "Set either disable-modern or disable-legacy" 1857 " to off\n"); 1858 return; 1859 } 1860 1861 if (pcie_port && pci_is_express(pci_dev)) { 1862 int pos; 1863 uint16_t last_pcie_cap_offset = PCI_CONFIG_SPACE_SIZE; 1864 1865 pos = pcie_endpoint_cap_init(pci_dev, 0); 1866 assert(pos > 0); 1867 1868 pos = pci_add_capability(pci_dev, PCI_CAP_ID_PM, 0, 1869 PCI_PM_SIZEOF, errp); 1870 if (pos < 0) { 1871 return; 1872 } 1873 1874 pci_dev->exp.pm_cap = pos; 1875 1876 /* 1877 * Indicates that this function complies with revision 1.2 of the 1878 * PCI Power Management Interface Specification. 1879 */ 1880 pci_set_word(pci_dev->config + pos + PCI_PM_PMC, 0x3); 1881 1882 if (proxy->flags & VIRTIO_PCI_FLAG_AER) { 1883 pcie_aer_init(pci_dev, PCI_ERR_VER, last_pcie_cap_offset, 1884 PCI_ERR_SIZEOF, NULL); 1885 last_pcie_cap_offset += PCI_ERR_SIZEOF; 1886 } 1887 1888 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_DEVERR) { 1889 /* Init error enabling flags */ 1890 pcie_cap_deverr_init(pci_dev); 1891 } 1892 1893 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_LNKCTL) { 1894 /* Init Link Control Register */ 1895 pcie_cap_lnkctl_init(pci_dev); 1896 } 1897 1898 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_PM) { 1899 /* Init Power Management Control Register */ 1900 pci_set_word(pci_dev->wmask + pos + PCI_PM_CTRL, 1901 PCI_PM_CTRL_STATE_MASK); 1902 } 1903 1904 if (proxy->flags & VIRTIO_PCI_FLAG_ATS) { 1905 pcie_ats_init(pci_dev, last_pcie_cap_offset, 1906 proxy->flags & VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED); 1907 last_pcie_cap_offset += PCI_EXT_CAP_ATS_SIZEOF; 1908 } 1909 1910 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_FLR) { 1911 /* Set Function Level Reset capability bit */ 1912 pcie_cap_flr_init(pci_dev); 1913 } 1914 } else { 1915 /* 1916 * make future invocations of pci_is_express() return false 1917 * and pci_config_size() return PCI_CONFIG_SPACE_SIZE. 1918 */ 1919 pci_dev->cap_present &= ~QEMU_PCI_CAP_EXPRESS; 1920 } 1921 1922 virtio_pci_bus_new(&proxy->bus, sizeof(proxy->bus), proxy); 1923 if (k->realize) { 1924 k->realize(proxy, errp); 1925 } 1926 } 1927 1928 static void virtio_pci_exit(PCIDevice *pci_dev) 1929 { 1930 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); 1931 bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) && 1932 !pci_bus_is_root(pci_get_bus(pci_dev)); 1933 1934 msix_uninit_exclusive_bar(pci_dev); 1935 if (proxy->flags & VIRTIO_PCI_FLAG_AER && pcie_port && 1936 pci_is_express(pci_dev)) { 1937 pcie_aer_exit(pci_dev); 1938 } 1939 } 1940 1941 static void virtio_pci_reset(DeviceState *qdev) 1942 { 1943 VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev); 1944 VirtioBusState *bus = VIRTIO_BUS(&proxy->bus); 1945 PCIDevice *dev = PCI_DEVICE(qdev); 1946 int i; 1947 1948 virtio_pci_stop_ioeventfd(proxy); 1949 virtio_bus_reset(bus); 1950 msix_unuse_all_vectors(&proxy->pci_dev); 1951 1952 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) { 1953 proxy->vqs[i].enabled = 0; 1954 proxy->vqs[i].num = 0; 1955 proxy->vqs[i].desc[0] = proxy->vqs[i].desc[1] = 0; 1956 proxy->vqs[i].avail[0] = proxy->vqs[i].avail[1] = 0; 1957 proxy->vqs[i].used[0] = proxy->vqs[i].used[1] = 0; 1958 } 1959 1960 if (pci_is_express(dev)) { 1961 pcie_cap_deverr_reset(dev); 1962 pcie_cap_lnkctl_reset(dev); 1963 1964 pci_set_word(dev->config + dev->exp.pm_cap + PCI_PM_CTRL, 0); 1965 } 1966 } 1967 1968 static Property virtio_pci_properties[] = { 1969 DEFINE_PROP_BIT("virtio-pci-bus-master-bug-migration", VirtIOPCIProxy, flags, 1970 VIRTIO_PCI_FLAG_BUS_MASTER_BUG_MIGRATION_BIT, false), 1971 DEFINE_PROP_BIT("migrate-extra", VirtIOPCIProxy, flags, 1972 VIRTIO_PCI_FLAG_MIGRATE_EXTRA_BIT, true), 1973 DEFINE_PROP_BIT("modern-pio-notify", VirtIOPCIProxy, flags, 1974 VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY_BIT, false), 1975 DEFINE_PROP_BIT("x-disable-pcie", VirtIOPCIProxy, flags, 1976 VIRTIO_PCI_FLAG_DISABLE_PCIE_BIT, false), 1977 DEFINE_PROP_BIT("page-per-vq", VirtIOPCIProxy, flags, 1978 VIRTIO_PCI_FLAG_PAGE_PER_VQ_BIT, false), 1979 DEFINE_PROP_BOOL("x-ignore-backend-features", VirtIOPCIProxy, 1980 ignore_backend_features, false), 1981 DEFINE_PROP_BIT("ats", VirtIOPCIProxy, flags, 1982 VIRTIO_PCI_FLAG_ATS_BIT, false), 1983 DEFINE_PROP_BIT("x-ats-page-aligned", VirtIOPCIProxy, flags, 1984 VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED_BIT, true), 1985 DEFINE_PROP_BIT("x-pcie-deverr-init", VirtIOPCIProxy, flags, 1986 VIRTIO_PCI_FLAG_INIT_DEVERR_BIT, true), 1987 DEFINE_PROP_BIT("x-pcie-lnkctl-init", VirtIOPCIProxy, flags, 1988 VIRTIO_PCI_FLAG_INIT_LNKCTL_BIT, true), 1989 DEFINE_PROP_BIT("x-pcie-pm-init", VirtIOPCIProxy, flags, 1990 VIRTIO_PCI_FLAG_INIT_PM_BIT, true), 1991 DEFINE_PROP_BIT("x-pcie-flr-init", VirtIOPCIProxy, flags, 1992 VIRTIO_PCI_FLAG_INIT_FLR_BIT, true), 1993 DEFINE_PROP_BIT("aer", VirtIOPCIProxy, flags, 1994 VIRTIO_PCI_FLAG_AER_BIT, false), 1995 DEFINE_PROP_END_OF_LIST(), 1996 }; 1997 1998 static void virtio_pci_dc_realize(DeviceState *qdev, Error **errp) 1999 { 2000 VirtioPCIClass *vpciklass = VIRTIO_PCI_GET_CLASS(qdev); 2001 VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev); 2002 PCIDevice *pci_dev = &proxy->pci_dev; 2003 2004 if (!(proxy->flags & VIRTIO_PCI_FLAG_DISABLE_PCIE) && 2005 virtio_pci_modern(proxy)) { 2006 pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS; 2007 } 2008 2009 vpciklass->parent_dc_realize(qdev, errp); 2010 } 2011 2012 static void virtio_pci_class_init(ObjectClass *klass, void *data) 2013 { 2014 DeviceClass *dc = DEVICE_CLASS(klass); 2015 PCIDeviceClass *k = PCI_DEVICE_CLASS(klass); 2016 VirtioPCIClass *vpciklass = VIRTIO_PCI_CLASS(klass); 2017 2018 device_class_set_props(dc, virtio_pci_properties); 2019 k->realize = virtio_pci_realize; 2020 k->exit = virtio_pci_exit; 2021 k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET; 2022 k->revision = VIRTIO_PCI_ABI_VERSION; 2023 k->class_id = PCI_CLASS_OTHERS; 2024 device_class_set_parent_realize(dc, virtio_pci_dc_realize, 2025 &vpciklass->parent_dc_realize); 2026 dc->reset = virtio_pci_reset; 2027 } 2028 2029 static const TypeInfo virtio_pci_info = { 2030 .name = TYPE_VIRTIO_PCI, 2031 .parent = TYPE_PCI_DEVICE, 2032 .instance_size = sizeof(VirtIOPCIProxy), 2033 .class_init = virtio_pci_class_init, 2034 .class_size = sizeof(VirtioPCIClass), 2035 .abstract = true, 2036 }; 2037 2038 static Property virtio_pci_generic_properties[] = { 2039 DEFINE_PROP_ON_OFF_AUTO("disable-legacy", VirtIOPCIProxy, disable_legacy, 2040 ON_OFF_AUTO_AUTO), 2041 DEFINE_PROP_BOOL("disable-modern", VirtIOPCIProxy, disable_modern, false), 2042 DEFINE_PROP_END_OF_LIST(), 2043 }; 2044 2045 static void virtio_pci_base_class_init(ObjectClass *klass, void *data) 2046 { 2047 const VirtioPCIDeviceTypeInfo *t = data; 2048 if (t->class_init) { 2049 t->class_init(klass, NULL); 2050 } 2051 } 2052 2053 static void virtio_pci_generic_class_init(ObjectClass *klass, void *data) 2054 { 2055 DeviceClass *dc = DEVICE_CLASS(klass); 2056 2057 device_class_set_props(dc, virtio_pci_generic_properties); 2058 } 2059 2060 static void virtio_pci_transitional_instance_init(Object *obj) 2061 { 2062 VirtIOPCIProxy *proxy = VIRTIO_PCI(obj); 2063 2064 proxy->disable_legacy = ON_OFF_AUTO_OFF; 2065 proxy->disable_modern = false; 2066 } 2067 2068 static void virtio_pci_non_transitional_instance_init(Object *obj) 2069 { 2070 VirtIOPCIProxy *proxy = VIRTIO_PCI(obj); 2071 2072 proxy->disable_legacy = ON_OFF_AUTO_ON; 2073 proxy->disable_modern = false; 2074 } 2075 2076 void virtio_pci_types_register(const VirtioPCIDeviceTypeInfo *t) 2077 { 2078 char *base_name = NULL; 2079 TypeInfo base_type_info = { 2080 .name = t->base_name, 2081 .parent = t->parent ? t->parent : TYPE_VIRTIO_PCI, 2082 .instance_size = t->instance_size, 2083 .instance_init = t->instance_init, 2084 .class_size = t->class_size, 2085 .abstract = true, 2086 .interfaces = t->interfaces, 2087 }; 2088 TypeInfo generic_type_info = { 2089 .name = t->generic_name, 2090 .parent = base_type_info.name, 2091 .class_init = virtio_pci_generic_class_init, 2092 .interfaces = (InterfaceInfo[]) { 2093 { INTERFACE_PCIE_DEVICE }, 2094 { INTERFACE_CONVENTIONAL_PCI_DEVICE }, 2095 { } 2096 }, 2097 }; 2098 2099 if (!base_type_info.name) { 2100 /* No base type -> register a single generic device type */ 2101 /* use intermediate %s-base-type to add generic device props */ 2102 base_name = g_strdup_printf("%s-base-type", t->generic_name); 2103 base_type_info.name = base_name; 2104 base_type_info.class_init = virtio_pci_generic_class_init; 2105 2106 generic_type_info.parent = base_name; 2107 generic_type_info.class_init = virtio_pci_base_class_init; 2108 generic_type_info.class_data = (void *)t; 2109 2110 assert(!t->non_transitional_name); 2111 assert(!t->transitional_name); 2112 } else { 2113 base_type_info.class_init = virtio_pci_base_class_init; 2114 base_type_info.class_data = (void *)t; 2115 } 2116 2117 type_register(&base_type_info); 2118 if (generic_type_info.name) { 2119 type_register(&generic_type_info); 2120 } 2121 2122 if (t->non_transitional_name) { 2123 const TypeInfo non_transitional_type_info = { 2124 .name = t->non_transitional_name, 2125 .parent = base_type_info.name, 2126 .instance_init = virtio_pci_non_transitional_instance_init, 2127 .interfaces = (InterfaceInfo[]) { 2128 { INTERFACE_PCIE_DEVICE }, 2129 { INTERFACE_CONVENTIONAL_PCI_DEVICE }, 2130 { } 2131 }, 2132 }; 2133 type_register(&non_transitional_type_info); 2134 } 2135 2136 if (t->transitional_name) { 2137 const TypeInfo transitional_type_info = { 2138 .name = t->transitional_name, 2139 .parent = base_type_info.name, 2140 .instance_init = virtio_pci_transitional_instance_init, 2141 .interfaces = (InterfaceInfo[]) { 2142 /* 2143 * Transitional virtio devices work only as Conventional PCI 2144 * devices because they require PIO ports. 2145 */ 2146 { INTERFACE_CONVENTIONAL_PCI_DEVICE }, 2147 { } 2148 }, 2149 }; 2150 type_register(&transitional_type_info); 2151 } 2152 g_free(base_name); 2153 } 2154 2155 unsigned virtio_pci_optimal_num_queues(unsigned fixed_queues) 2156 { 2157 /* 2158 * 1:1 vq to vCPU mapping is ideal because the same vCPU that submitted 2159 * virtqueue buffers can handle their completion. When a different vCPU 2160 * handles completion it may need to IPI the vCPU that submitted the 2161 * request and this adds overhead. 2162 * 2163 * Virtqueues consume guest RAM and MSI-X vectors. This is wasteful in 2164 * guests with very many vCPUs and a device that is only used by a few 2165 * vCPUs. Unfortunately optimizing that case requires manual pinning inside 2166 * the guest, so those users might as well manually set the number of 2167 * queues. There is no upper limit that can be applied automatically and 2168 * doing so arbitrarily would result in a sudden performance drop once the 2169 * threshold number of vCPUs is exceeded. 2170 */ 2171 unsigned num_queues = current_machine->smp.cpus; 2172 2173 /* 2174 * The maximum number of MSI-X vectors is PCI_MSIX_FLAGS_QSIZE + 1, but the 2175 * config change interrupt and the fixed virtqueues must be taken into 2176 * account too. 2177 */ 2178 num_queues = MIN(num_queues, PCI_MSIX_FLAGS_QSIZE - fixed_queues); 2179 2180 /* 2181 * There is a limit to how many virtqueues a device can have. 2182 */ 2183 return MIN(num_queues, VIRTIO_QUEUE_MAX - fixed_queues); 2184 } 2185 2186 /* virtio-pci-bus */ 2187 2188 static void virtio_pci_bus_new(VirtioBusState *bus, size_t bus_size, 2189 VirtIOPCIProxy *dev) 2190 { 2191 DeviceState *qdev = DEVICE(dev); 2192 char virtio_bus_name[] = "virtio-bus"; 2193 2194 qbus_init(bus, bus_size, TYPE_VIRTIO_PCI_BUS, qdev, virtio_bus_name); 2195 } 2196 2197 static void virtio_pci_bus_class_init(ObjectClass *klass, void *data) 2198 { 2199 BusClass *bus_class = BUS_CLASS(klass); 2200 VirtioBusClass *k = VIRTIO_BUS_CLASS(klass); 2201 bus_class->max_dev = 1; 2202 k->notify = virtio_pci_notify; 2203 k->save_config = virtio_pci_save_config; 2204 k->load_config = virtio_pci_load_config; 2205 k->save_queue = virtio_pci_save_queue; 2206 k->load_queue = virtio_pci_load_queue; 2207 k->save_extra_state = virtio_pci_save_extra_state; 2208 k->load_extra_state = virtio_pci_load_extra_state; 2209 k->has_extra_state = virtio_pci_has_extra_state; 2210 k->query_guest_notifiers = virtio_pci_query_guest_notifiers; 2211 k->set_guest_notifiers = virtio_pci_set_guest_notifiers; 2212 k->set_host_notifier_mr = virtio_pci_set_host_notifier_mr; 2213 k->vmstate_change = virtio_pci_vmstate_change; 2214 k->pre_plugged = virtio_pci_pre_plugged; 2215 k->device_plugged = virtio_pci_device_plugged; 2216 k->device_unplugged = virtio_pci_device_unplugged; 2217 k->query_nvectors = virtio_pci_query_nvectors; 2218 k->ioeventfd_enabled = virtio_pci_ioeventfd_enabled; 2219 k->ioeventfd_assign = virtio_pci_ioeventfd_assign; 2220 k->get_dma_as = virtio_pci_get_dma_as; 2221 k->iommu_enabled = virtio_pci_iommu_enabled; 2222 k->queue_enabled = virtio_pci_queue_enabled; 2223 } 2224 2225 static const TypeInfo virtio_pci_bus_info = { 2226 .name = TYPE_VIRTIO_PCI_BUS, 2227 .parent = TYPE_VIRTIO_BUS, 2228 .instance_size = sizeof(VirtioPCIBusState), 2229 .class_size = sizeof(VirtioPCIBusClass), 2230 .class_init = virtio_pci_bus_class_init, 2231 }; 2232 2233 static void virtio_pci_register_types(void) 2234 { 2235 /* Base types: */ 2236 type_register_static(&virtio_pci_bus_info); 2237 type_register_static(&virtio_pci_info); 2238 } 2239 2240 type_init(virtio_pci_register_types) 2241 2242