1 #include "kvm/virtio-pci.h" 2 3 #include "kvm/ioport.h" 4 #include "kvm/kvm.h" 5 #include "kvm/virtio-pci-dev.h" 6 #include "kvm/irq.h" 7 #include "kvm/virtio.h" 8 #include "kvm/ioeventfd.h" 9 10 #include <sys/ioctl.h> 11 #include <linux/virtio_pci.h> 12 #include <linux/byteorder.h> 13 #include <string.h> 14 15 static void virtio_pci__ioevent_callback(struct kvm *kvm, void *param) 16 { 17 struct virtio_pci_ioevent_param *ioeventfd = param; 18 struct virtio_pci *vpci = ioeventfd->vdev->virtio; 19 20 ioeventfd->vdev->ops->notify_vq(kvm, vpci->dev, ioeventfd->vq); 21 } 22 23 static int virtio_pci__init_ioeventfd(struct kvm *kvm, struct virtio_device *vdev, u32 vq) 24 { 25 struct ioevent ioevent; 26 struct virtio_pci *vpci = vdev->virtio; 27 int i, r, flags = IOEVENTFD_FLAG_PIO; 28 int fds[2]; 29 30 vpci->ioeventfds[vq] = (struct virtio_pci_ioevent_param) { 31 .vdev = vdev, 32 .vq = vq, 33 }; 34 35 ioevent = (struct ioevent) { 36 .fn = virtio_pci__ioevent_callback, 37 .fn_ptr = &vpci->ioeventfds[vq], 38 .datamatch = vq, 39 .fn_kvm = kvm, 40 }; 41 42 /* 43 * Vhost will poll the eventfd in host kernel side, otherwise we 44 * need to poll in userspace. 45 */ 46 if (!vdev->use_vhost) 47 flags |= IOEVENTFD_FLAG_USER_POLL; 48 49 /* ioport */ 50 ioevent.io_addr = vpci->port_addr + VIRTIO_PCI_QUEUE_NOTIFY; 51 ioevent.io_len = sizeof(u16); 52 ioevent.fd = fds[0] = eventfd(0, 0); 53 r = ioeventfd__add_event(&ioevent, flags); 54 if (r) 55 return r; 56 57 /* mmio */ 58 ioevent.io_addr = vpci->mmio_addr + VIRTIO_PCI_QUEUE_NOTIFY; 59 ioevent.io_len = sizeof(u32); 60 ioevent.fd = fds[1] = eventfd(0, 0); 61 r = ioeventfd__add_event(&ioevent, flags); 62 if (r) 63 goto free_ioport_evt; 64 65 if (vdev->ops->notify_vq_eventfd) 66 for (i = 0; i < 2; ++i) 67 vdev->ops->notify_vq_eventfd(kvm, vpci->dev, vq, 68 fds[i]); 69 return 0; 70 71 free_ioport_evt: 72 ioeventfd__del_event(vpci->port_addr + VIRTIO_PCI_QUEUE_NOTIFY, vq); 73 return r; 74 } 75 76 static inline bool virtio_pci__msix_enabled(struct virtio_pci *vpci) 77 { 78 return vpci->pci_hdr.msix.ctrl & cpu_to_le16(PCI_MSIX_FLAGS_ENABLE); 79 } 80 81 static bool virtio_pci__specific_io_in(struct kvm *kvm, struct virtio_device *vdev, u16 port, 82 void *data, int size, int offset) 83 { 84 u32 config_offset; 85 struct virtio_pci *vpci = vdev->virtio; 86 int type = virtio__get_dev_specific_field(offset - 20, 87 virtio_pci__msix_enabled(vpci), 88 &config_offset); 89 if (type == VIRTIO_PCI_O_MSIX) { 90 switch (offset) { 91 case VIRTIO_MSI_CONFIG_VECTOR: 92 ioport__write16(data, vpci->config_vector); 93 break; 94 case VIRTIO_MSI_QUEUE_VECTOR: 95 ioport__write16(data, vpci->vq_vector[vpci->queue_selector]); 96 break; 97 }; 98 99 return true; 100 } else if (type == VIRTIO_PCI_O_CONFIG) { 101 u8 cfg; 102 103 cfg = vdev->ops->get_config(kvm, vpci->dev)[config_offset]; 104 ioport__write8(data, cfg); 105 return true; 106 } 107 108 return false; 109 } 110 111 static bool virtio_pci__io_in(struct ioport *ioport, struct kvm *kvm, u16 port, void *data, int size) 112 { 113 unsigned long offset; 114 bool ret = true; 115 struct virtio_device *vdev; 116 struct virtio_pci *vpci; 117 u32 val; 118 119 vdev = ioport->priv; 120 vpci = vdev->virtio; 121 offset = port - vpci->port_addr; 122 123 switch (offset) { 124 case VIRTIO_PCI_HOST_FEATURES: 125 val = vdev->ops->get_host_features(kvm, vpci->dev); 126 ioport__write32(data, val); 127 break; 128 case VIRTIO_PCI_QUEUE_PFN: 129 val = vdev->ops->get_pfn_vq(kvm, vpci->dev, vpci->queue_selector); 130 ioport__write32(data, val); 131 break; 132 case VIRTIO_PCI_QUEUE_NUM: 133 val = vdev->ops->get_size_vq(kvm, vpci->dev, vpci->queue_selector); 134 ioport__write16(data, val); 135 break; 136 case VIRTIO_PCI_STATUS: 137 ioport__write8(data, vpci->status); 138 break; 139 case VIRTIO_PCI_ISR: 140 ioport__write8(data, vpci->isr); 141 kvm__irq_line(kvm, vpci->pci_hdr.irq_line, VIRTIO_IRQ_LOW); 142 vpci->isr = VIRTIO_IRQ_LOW; 143 break; 144 default: 145 ret = virtio_pci__specific_io_in(kvm, vdev, port, data, size, offset); 146 break; 147 }; 148 149 return ret; 150 } 151 152 static bool virtio_pci__specific_io_out(struct kvm *kvm, struct virtio_device *vdev, u16 port, 153 void *data, int size, int offset) 154 { 155 struct virtio_pci *vpci = vdev->virtio; 156 u32 config_offset, gsi, vec; 157 int type = virtio__get_dev_specific_field(offset - 20, virtio_pci__msix_enabled(vpci), 158 &config_offset); 159 if (type == VIRTIO_PCI_O_MSIX) { 160 switch (offset) { 161 case VIRTIO_MSI_CONFIG_VECTOR: 162 vec = vpci->config_vector = ioport__read16(data); 163 if (vec == VIRTIO_MSI_NO_VECTOR) 164 break; 165 166 gsi = irq__add_msix_route(kvm, &vpci->msix_table[vec].msg); 167 168 vpci->config_gsi = gsi; 169 break; 170 case VIRTIO_MSI_QUEUE_VECTOR: 171 vec = vpci->vq_vector[vpci->queue_selector] = ioport__read16(data); 172 173 if (vec == VIRTIO_MSI_NO_VECTOR) 174 break; 175 176 gsi = irq__add_msix_route(kvm, &vpci->msix_table[vec].msg); 177 vpci->gsis[vpci->queue_selector] = gsi; 178 if (vdev->ops->notify_vq_gsi) 179 vdev->ops->notify_vq_gsi(kvm, vpci->dev, 180 vpci->queue_selector, gsi); 181 break; 182 }; 183 184 return true; 185 } else if (type == VIRTIO_PCI_O_CONFIG) { 186 vdev->ops->get_config(kvm, vpci->dev)[config_offset] = *(u8 *)data; 187 188 return true; 189 } 190 191 return false; 192 } 193 194 static bool virtio_pci__io_out(struct ioport *ioport, struct kvm *kvm, u16 port, void *data, int size) 195 { 196 unsigned long offset; 197 bool ret = true; 198 struct virtio_device *vdev; 199 struct virtio_pci *vpci; 200 u32 val; 201 202 vdev = ioport->priv; 203 vpci = vdev->virtio; 204 offset = port - vpci->port_addr; 205 206 switch (offset) { 207 case VIRTIO_PCI_GUEST_FEATURES: 208 val = ioport__read32(data); 209 vdev->ops->set_guest_features(kvm, vpci->dev, val); 210 break; 211 case VIRTIO_PCI_QUEUE_PFN: 212 val = ioport__read32(data); 213 virtio_pci__init_ioeventfd(kvm, vdev, vpci->queue_selector); 214 vdev->ops->init_vq(kvm, vpci->dev, vpci->queue_selector, 215 1 << VIRTIO_PCI_QUEUE_ADDR_SHIFT, 216 VIRTIO_PCI_VRING_ALIGN, val); 217 break; 218 case VIRTIO_PCI_QUEUE_SEL: 219 vpci->queue_selector = ioport__read16(data); 220 break; 221 case VIRTIO_PCI_QUEUE_NOTIFY: 222 val = ioport__read16(data); 223 vdev->ops->notify_vq(kvm, vpci->dev, val); 224 break; 225 case VIRTIO_PCI_STATUS: 226 vpci->status = ioport__read8(data); 227 if (vdev->ops->notify_status) 228 vdev->ops->notify_status(kvm, vpci->dev, vpci->status); 229 break; 230 default: 231 ret = virtio_pci__specific_io_out(kvm, vdev, port, data, size, offset); 232 break; 233 }; 234 235 return ret; 236 } 237 238 static struct ioport_operations virtio_pci__io_ops = { 239 .io_in = virtio_pci__io_in, 240 .io_out = virtio_pci__io_out, 241 }; 242 243 static void virtio_pci__msix_mmio_callback(u64 addr, u8 *data, u32 len, 244 u8 is_write, void *ptr) 245 { 246 struct virtio_pci *vpci = ptr; 247 void *table; 248 u32 offset; 249 250 if (addr > vpci->msix_io_block + PCI_IO_SIZE) { 251 table = &vpci->msix_pba; 252 offset = vpci->msix_io_block + PCI_IO_SIZE; 253 } else { 254 table = &vpci->msix_table; 255 offset = vpci->msix_io_block; 256 } 257 258 if (is_write) 259 memcpy(table + addr - offset, data, len); 260 else 261 memcpy(data, table + addr - offset, len); 262 } 263 264 static void virtio_pci__signal_msi(struct kvm *kvm, struct virtio_pci *vpci, int vec) 265 { 266 struct kvm_msi msi = { 267 .address_lo = vpci->msix_table[vec].msg.address_lo, 268 .address_hi = vpci->msix_table[vec].msg.address_hi, 269 .data = vpci->msix_table[vec].msg.data, 270 }; 271 272 ioctl(kvm->vm_fd, KVM_SIGNAL_MSI, &msi); 273 } 274 275 int virtio_pci__signal_vq(struct kvm *kvm, struct virtio_device *vdev, u32 vq) 276 { 277 struct virtio_pci *vpci = vdev->virtio; 278 int tbl = vpci->vq_vector[vq]; 279 280 if (virtio_pci__msix_enabled(vpci) && tbl != VIRTIO_MSI_NO_VECTOR) { 281 if (vpci->pci_hdr.msix.ctrl & cpu_to_le16(PCI_MSIX_FLAGS_MASKALL) || 282 vpci->msix_table[tbl].ctrl & cpu_to_le16(PCI_MSIX_ENTRY_CTRL_MASKBIT)) { 283 284 vpci->msix_pba |= 1 << tbl; 285 return 0; 286 } 287 288 if (vpci->features & VIRTIO_PCI_F_SIGNAL_MSI) 289 virtio_pci__signal_msi(kvm, vpci, vpci->vq_vector[vq]); 290 else 291 kvm__irq_trigger(kvm, vpci->gsis[vq]); 292 } else { 293 vpci->isr = VIRTIO_IRQ_HIGH; 294 kvm__irq_trigger(kvm, vpci->pci_hdr.irq_line); 295 } 296 return 0; 297 } 298 299 int virtio_pci__signal_config(struct kvm *kvm, struct virtio_device *vdev) 300 { 301 struct virtio_pci *vpci = vdev->virtio; 302 int tbl = vpci->config_vector; 303 304 if (virtio_pci__msix_enabled(vpci) && tbl != VIRTIO_MSI_NO_VECTOR) { 305 if (vpci->pci_hdr.msix.ctrl & cpu_to_le16(PCI_MSIX_FLAGS_MASKALL) || 306 vpci->msix_table[tbl].ctrl & cpu_to_le16(PCI_MSIX_ENTRY_CTRL_MASKBIT)) { 307 308 vpci->msix_pba |= 1 << tbl; 309 return 0; 310 } 311 312 if (vpci->features & VIRTIO_PCI_F_SIGNAL_MSI) 313 virtio_pci__signal_msi(kvm, vpci, tbl); 314 else 315 kvm__irq_trigger(kvm, vpci->config_gsi); 316 } else { 317 vpci->isr = VIRTIO_PCI_ISR_CONFIG; 318 kvm__irq_trigger(kvm, vpci->pci_hdr.irq_line); 319 } 320 321 return 0; 322 } 323 324 static void virtio_pci__io_mmio_callback(u64 addr, u8 *data, u32 len, 325 u8 is_write, void *ptr) 326 { 327 struct virtio_pci *vpci = ptr; 328 int direction = is_write ? KVM_EXIT_IO_OUT : KVM_EXIT_IO_IN; 329 u16 port = vpci->port_addr + (addr & (IOPORT_SIZE - 1)); 330 331 kvm__emulate_io(vpci->kvm, port, data, direction, len, 1); 332 } 333 334 int virtio_pci__init(struct kvm *kvm, void *dev, struct virtio_device *vdev, 335 int device_id, int subsys_id, int class) 336 { 337 struct virtio_pci *vpci = vdev->virtio; 338 u8 pin, line; 339 int r; 340 341 vpci->kvm = kvm; 342 vpci->dev = dev; 343 344 r = ioport__register(kvm, IOPORT_EMPTY, &virtio_pci__io_ops, IOPORT_SIZE, vdev); 345 if (r < 0) 346 return r; 347 vpci->port_addr = (u16)r; 348 349 vpci->mmio_addr = pci_get_io_space_block(IOPORT_SIZE); 350 r = kvm__register_mmio(kvm, vpci->mmio_addr, IOPORT_SIZE, false, 351 virtio_pci__io_mmio_callback, vpci); 352 if (r < 0) 353 goto free_ioport; 354 355 vpci->msix_io_block = pci_get_io_space_block(PCI_IO_SIZE * 2); 356 r = kvm__register_mmio(kvm, vpci->msix_io_block, PCI_IO_SIZE * 2, false, 357 virtio_pci__msix_mmio_callback, vpci); 358 if (r < 0) 359 goto free_mmio; 360 361 vpci->pci_hdr = (struct pci_device_header) { 362 .vendor_id = cpu_to_le16(PCI_VENDOR_ID_REDHAT_QUMRANET), 363 .device_id = cpu_to_le16(device_id), 364 .header_type = PCI_HEADER_TYPE_NORMAL, 365 .revision_id = 0, 366 .class[0] = class & 0xff, 367 .class[1] = (class >> 8) & 0xff, 368 .class[2] = (class >> 16) & 0xff, 369 .subsys_vendor_id = cpu_to_le16(PCI_SUBSYSTEM_VENDOR_ID_REDHAT_QUMRANET), 370 .subsys_id = cpu_to_le16(subsys_id), 371 .bar[0] = cpu_to_le32(vpci->mmio_addr 372 | PCI_BASE_ADDRESS_SPACE_MEMORY), 373 .bar[1] = cpu_to_le32(vpci->port_addr 374 | PCI_BASE_ADDRESS_SPACE_IO), 375 .bar[2] = cpu_to_le32(vpci->msix_io_block 376 | PCI_BASE_ADDRESS_SPACE_MEMORY), 377 .status = cpu_to_le16(PCI_STATUS_CAP_LIST), 378 .capabilities = (void *)&vpci->pci_hdr.msix - (void *)&vpci->pci_hdr, 379 .bar_size[0] = IOPORT_SIZE, 380 .bar_size[1] = IOPORT_SIZE, 381 .bar_size[2] = PCI_IO_SIZE * 2, 382 }; 383 384 vpci->dev_hdr = (struct device_header) { 385 .bus_type = DEVICE_BUS_PCI, 386 .data = &vpci->pci_hdr, 387 }; 388 389 vpci->pci_hdr.msix.cap = PCI_CAP_ID_MSIX; 390 vpci->pci_hdr.msix.next = 0; 391 /* 392 * We at most have VIRTIO_PCI_MAX_VQ entries for virt queue, 393 * VIRTIO_PCI_MAX_CONFIG entries for config. 394 * 395 * To quote the PCI spec: 396 * 397 * System software reads this field to determine the 398 * MSI-X Table Size N, which is encoded as N-1. 399 * For example, a returned value of "00000000011" 400 * indicates a table size of 4. 401 */ 402 vpci->pci_hdr.msix.ctrl = cpu_to_le16(VIRTIO_PCI_MAX_VQ + VIRTIO_PCI_MAX_CONFIG - 1); 403 404 /* Both table and PBA are mapped to the same BAR (2) */ 405 vpci->pci_hdr.msix.table_offset = cpu_to_le32(2); 406 vpci->pci_hdr.msix.pba_offset = cpu_to_le32(2 | PCI_IO_SIZE); 407 vpci->config_vector = 0; 408 409 r = irq__register_device(subsys_id, &pin, &line); 410 if (r < 0) 411 goto free_msix_mmio; 412 413 if (kvm__supports_extension(kvm, KVM_CAP_SIGNAL_MSI)) 414 vpci->features |= VIRTIO_PCI_F_SIGNAL_MSI; 415 416 vpci->pci_hdr.irq_pin = pin; 417 vpci->pci_hdr.irq_line = line; 418 r = device__register(&vpci->dev_hdr); 419 if (r < 0) 420 goto free_msix_mmio; 421 422 return 0; 423 424 free_msix_mmio: 425 kvm__deregister_mmio(kvm, vpci->msix_io_block); 426 free_mmio: 427 kvm__deregister_mmio(kvm, vpci->mmio_addr); 428 free_ioport: 429 ioport__unregister(kvm, vpci->port_addr); 430 return r; 431 } 432 433 int virtio_pci__exit(struct kvm *kvm, struct virtio_device *vdev) 434 { 435 struct virtio_pci *vpci = vdev->virtio; 436 int i; 437 438 kvm__deregister_mmio(kvm, vpci->mmio_addr); 439 kvm__deregister_mmio(kvm, vpci->msix_io_block); 440 ioport__unregister(kvm, vpci->port_addr); 441 442 for (i = 0; i < VIRTIO_PCI_MAX_VQ; i++) { 443 ioeventfd__del_event(vpci->port_addr + VIRTIO_PCI_QUEUE_NOTIFY, i); 444 ioeventfd__del_event(vpci->mmio_addr + VIRTIO_PCI_QUEUE_NOTIFY, i); 445 } 446 447 return 0; 448 } 449