1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright(c) 2010-2016 Intel Corporation 3 */ 4 5 #include <stdint.h> 6 #include <string.h> 7 #include <stdio.h> 8 #include <errno.h> 9 #include <unistd.h> 10 11 #include <rte_ethdev_driver.h> 12 #include <rte_ethdev_pci.h> 13 #include <rte_memcpy.h> 14 #include <rte_string_fns.h> 15 #include <rte_memzone.h> 16 #include <rte_malloc.h> 17 #include <rte_branch_prediction.h> 18 #include <rte_pci.h> 19 #include <rte_bus_pci.h> 20 #include <rte_ether.h> 21 #include <rte_ip.h> 22 #include <rte_arp.h> 23 #include <rte_common.h> 24 #include <rte_errno.h> 25 #include <rte_cpuflags.h> 26 27 #include <rte_memory.h> 28 #include <rte_eal.h> 29 #include <rte_dev.h> 30 #include <rte_cycles.h> 31 #include <rte_kvargs.h> 32 33 #include "virtio_ethdev.h" 34 #include "virtio_pci.h" 35 #include "virtio_logs.h" 36 #include "virtqueue.h" 37 #include "virtio_rxtx.h" 38 39 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev); 40 static int virtio_dev_configure(struct rte_eth_dev *dev); 41 static int virtio_dev_start(struct rte_eth_dev *dev); 42 static void virtio_dev_stop(struct rte_eth_dev *dev); 43 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev); 44 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev); 45 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev); 46 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev); 47 static void virtio_dev_info_get(struct rte_eth_dev *dev, 48 struct rte_eth_dev_info *dev_info); 49 static int virtio_dev_link_update(struct rte_eth_dev *dev, 50 int wait_to_complete); 51 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask); 52 53 static void virtio_set_hwaddr(struct virtio_hw *hw); 54 static void virtio_get_hwaddr(struct virtio_hw *hw); 55 56 static int virtio_dev_stats_get(struct rte_eth_dev *dev, 57 struct rte_eth_stats *stats); 58 static int virtio_dev_xstats_get(struct rte_eth_dev *dev, 59 struct rte_eth_xstat *xstats, unsigned n); 60 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev, 61 struct rte_eth_xstat_name *xstats_names, 62 unsigned limit); 63 static void virtio_dev_stats_reset(struct rte_eth_dev *dev); 64 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev); 65 static int virtio_vlan_filter_set(struct rte_eth_dev *dev, 66 uint16_t vlan_id, int on); 67 static int virtio_mac_addr_add(struct rte_eth_dev *dev, 68 struct ether_addr *mac_addr, 69 uint32_t index, uint32_t vmdq); 70 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index); 71 static int virtio_mac_addr_set(struct rte_eth_dev *dev, 72 struct ether_addr *mac_addr); 73 74 static int virtio_intr_enable(struct rte_eth_dev *dev); 75 static int virtio_intr_disable(struct rte_eth_dev *dev); 76 77 static int virtio_dev_queue_stats_mapping_set( 78 struct rte_eth_dev *eth_dev, 79 uint16_t queue_id, 80 uint8_t stat_idx, 81 uint8_t is_rx); 82 83 int virtio_logtype_init; 84 int virtio_logtype_driver; 85 86 static void virtio_notify_peers(struct rte_eth_dev *dev); 87 static void virtio_ack_link_announce(struct rte_eth_dev *dev); 88 89 /* 90 * The set of PCI devices this driver supports 91 */ 92 static const struct rte_pci_id pci_id_virtio_map[] = { 93 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) }, 94 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) }, 95 { .vendor_id = 0, /* sentinel */ }, 96 }; 97 98 struct rte_virtio_xstats_name_off { 99 char name[RTE_ETH_XSTATS_NAME_SIZE]; 100 unsigned offset; 101 }; 102 103 /* [rt]x_qX_ is prepended to the name string here */ 104 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = { 105 {"good_packets", offsetof(struct virtnet_rx, stats.packets)}, 106 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)}, 107 {"errors", offsetof(struct virtnet_rx, stats.errors)}, 108 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)}, 109 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)}, 110 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])}, 111 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])}, 112 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])}, 113 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])}, 114 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])}, 115 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])}, 116 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])}, 117 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])}, 118 }; 119 120 /* [rt]x_qX_ is prepended to the name string here */ 121 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = { 122 {"good_packets", offsetof(struct virtnet_tx, stats.packets)}, 123 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)}, 124 {"errors", offsetof(struct virtnet_tx, stats.errors)}, 125 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)}, 126 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)}, 127 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])}, 128 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])}, 129 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])}, 130 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])}, 131 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])}, 132 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])}, 133 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])}, 134 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])}, 135 }; 136 137 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \ 138 sizeof(rte_virtio_rxq_stat_strings[0])) 139 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \ 140 sizeof(rte_virtio_txq_stat_strings[0])) 141 142 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS]; 143 144 static int 145 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl, 146 int *dlen, int pkt_num) 147 { 148 uint32_t head, i; 149 int k, sum = 0; 150 virtio_net_ctrl_ack status = ~0; 151 struct virtio_pmd_ctrl *result; 152 struct virtqueue *vq; 153 154 ctrl->status = status; 155 156 if (!cvq || !cvq->vq) { 157 PMD_INIT_LOG(ERR, "Control queue is not supported."); 158 return -1; 159 } 160 161 rte_spinlock_lock(&cvq->lock); 162 vq = cvq->vq; 163 head = vq->vq_desc_head_idx; 164 165 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, " 166 "vq->hw->cvq = %p vq = %p", 167 vq->vq_desc_head_idx, status, vq->hw->cvq, vq); 168 169 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) { 170 rte_spinlock_unlock(&cvq->lock); 171 return -1; 172 } 173 174 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl, 175 sizeof(struct virtio_pmd_ctrl)); 176 177 /* 178 * Format is enforced in qemu code: 179 * One TX packet for header; 180 * At least one TX packet per argument; 181 * One RX packet for ACK. 182 */ 183 vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT; 184 vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem; 185 vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr); 186 vq->vq_free_cnt--; 187 i = vq->vq_ring.desc[head].next; 188 189 for (k = 0; k < pkt_num; k++) { 190 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT; 191 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem 192 + sizeof(struct virtio_net_ctrl_hdr) 193 + sizeof(ctrl->status) + sizeof(uint8_t)*sum; 194 vq->vq_ring.desc[i].len = dlen[k]; 195 sum += dlen[k]; 196 vq->vq_free_cnt--; 197 i = vq->vq_ring.desc[i].next; 198 } 199 200 vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE; 201 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem 202 + sizeof(struct virtio_net_ctrl_hdr); 203 vq->vq_ring.desc[i].len = sizeof(ctrl->status); 204 vq->vq_free_cnt--; 205 206 vq->vq_desc_head_idx = vq->vq_ring.desc[i].next; 207 208 vq_update_avail_ring(vq, head); 209 vq_update_avail_idx(vq); 210 211 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index); 212 213 virtqueue_notify(vq); 214 215 rte_rmb(); 216 while (VIRTQUEUE_NUSED(vq) == 0) { 217 rte_rmb(); 218 usleep(100); 219 } 220 221 while (VIRTQUEUE_NUSED(vq)) { 222 uint32_t idx, desc_idx, used_idx; 223 struct vring_used_elem *uep; 224 225 used_idx = (uint32_t)(vq->vq_used_cons_idx 226 & (vq->vq_nentries - 1)); 227 uep = &vq->vq_ring.used->ring[used_idx]; 228 idx = (uint32_t) uep->id; 229 desc_idx = idx; 230 231 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) { 232 desc_idx = vq->vq_ring.desc[desc_idx].next; 233 vq->vq_free_cnt++; 234 } 235 236 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx; 237 vq->vq_desc_head_idx = idx; 238 239 vq->vq_used_cons_idx++; 240 vq->vq_free_cnt++; 241 } 242 243 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d", 244 vq->vq_free_cnt, vq->vq_desc_head_idx); 245 246 result = cvq->virtio_net_hdr_mz->addr; 247 248 rte_spinlock_unlock(&cvq->lock); 249 return result->status; 250 } 251 252 static int 253 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues) 254 { 255 struct virtio_hw *hw = dev->data->dev_private; 256 struct virtio_pmd_ctrl ctrl; 257 int dlen[1]; 258 int ret; 259 260 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ; 261 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET; 262 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t)); 263 264 dlen[0] = sizeof(uint16_t); 265 266 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1); 267 if (ret) { 268 PMD_INIT_LOG(ERR, "Multiqueue configured but send command " 269 "failed, this is too late now..."); 270 return -EINVAL; 271 } 272 273 return 0; 274 } 275 276 static void 277 virtio_dev_queue_release(void *queue __rte_unused) 278 { 279 /* do nothing */ 280 } 281 282 static uint16_t 283 virtio_get_nr_vq(struct virtio_hw *hw) 284 { 285 uint16_t nr_vq = hw->max_queue_pairs * 2; 286 287 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) 288 nr_vq += 1; 289 290 return nr_vq; 291 } 292 293 static void 294 virtio_init_vring(struct virtqueue *vq) 295 { 296 int size = vq->vq_nentries; 297 struct vring *vr = &vq->vq_ring; 298 uint8_t *ring_mem = vq->vq_ring_virt_mem; 299 300 PMD_INIT_FUNC_TRACE(); 301 302 /* 303 * Reinitialise since virtio port might have been stopped and restarted 304 */ 305 memset(ring_mem, 0, vq->vq_ring_size); 306 vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN); 307 vq->vq_used_cons_idx = 0; 308 vq->vq_desc_head_idx = 0; 309 vq->vq_avail_idx = 0; 310 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1); 311 vq->vq_free_cnt = vq->vq_nentries; 312 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries); 313 314 vring_desc_init(vr->desc, size); 315 316 /* 317 * Disable device(host) interrupting guest 318 */ 319 virtqueue_disable_intr(vq); 320 } 321 322 static int 323 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx) 324 { 325 char vq_name[VIRTQUEUE_MAX_NAME_SZ]; 326 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ]; 327 const struct rte_memzone *mz = NULL, *hdr_mz = NULL; 328 unsigned int vq_size, size; 329 struct virtio_hw *hw = dev->data->dev_private; 330 struct virtnet_rx *rxvq = NULL; 331 struct virtnet_tx *txvq = NULL; 332 struct virtnet_ctl *cvq = NULL; 333 struct virtqueue *vq; 334 size_t sz_hdr_mz = 0; 335 void *sw_ring = NULL; 336 int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx); 337 int ret; 338 int numa_node = dev->device->numa_node; 339 340 PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d", 341 vtpci_queue_idx, numa_node); 342 343 /* 344 * Read the virtqueue size from the Queue Size field 345 * Always power of 2 and if 0 virtqueue does not exist 346 */ 347 vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx); 348 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size); 349 if (vq_size == 0) { 350 PMD_INIT_LOG(ERR, "virtqueue does not exist"); 351 return -EINVAL; 352 } 353 354 if (!rte_is_power_of_2(vq_size)) { 355 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2"); 356 return -EINVAL; 357 } 358 359 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d", 360 dev->data->port_id, vtpci_queue_idx); 361 362 size = RTE_ALIGN_CEIL(sizeof(*vq) + 363 vq_size * sizeof(struct vq_desc_extra), 364 RTE_CACHE_LINE_SIZE); 365 if (queue_type == VTNET_TQ) { 366 /* 367 * For each xmit packet, allocate a virtio_net_hdr 368 * and indirect ring elements 369 */ 370 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region); 371 } else if (queue_type == VTNET_CQ) { 372 /* Allocate a page for control vq command, data and status */ 373 sz_hdr_mz = PAGE_SIZE; 374 } 375 376 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE, 377 numa_node); 378 if (vq == NULL) { 379 PMD_INIT_LOG(ERR, "can not allocate vq"); 380 return -ENOMEM; 381 } 382 hw->vqs[vtpci_queue_idx] = vq; 383 384 vq->hw = hw; 385 vq->vq_queue_index = vtpci_queue_idx; 386 vq->vq_nentries = vq_size; 387 388 /* 389 * Reserve a memzone for vring elements 390 */ 391 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN); 392 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN); 393 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", 394 size, vq->vq_ring_size); 395 396 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size, 397 numa_node, RTE_MEMZONE_IOVA_CONTIG, 398 VIRTIO_PCI_VRING_ALIGN); 399 if (mz == NULL) { 400 if (rte_errno == EEXIST) 401 mz = rte_memzone_lookup(vq_name); 402 if (mz == NULL) { 403 ret = -ENOMEM; 404 goto fail_q_alloc; 405 } 406 } 407 408 memset(mz->addr, 0, mz->len); 409 410 vq->vq_ring_mem = mz->iova; 411 vq->vq_ring_virt_mem = mz->addr; 412 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64, 413 (uint64_t)mz->iova); 414 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64, 415 (uint64_t)(uintptr_t)mz->addr); 416 417 virtio_init_vring(vq); 418 419 if (sz_hdr_mz) { 420 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr", 421 dev->data->port_id, vtpci_queue_idx); 422 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz, 423 numa_node, RTE_MEMZONE_IOVA_CONTIG, 424 RTE_CACHE_LINE_SIZE); 425 if (hdr_mz == NULL) { 426 if (rte_errno == EEXIST) 427 hdr_mz = rte_memzone_lookup(vq_hdr_name); 428 if (hdr_mz == NULL) { 429 ret = -ENOMEM; 430 goto fail_q_alloc; 431 } 432 } 433 } 434 435 if (queue_type == VTNET_RQ) { 436 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) * 437 sizeof(vq->sw_ring[0]); 438 439 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw, 440 RTE_CACHE_LINE_SIZE, numa_node); 441 if (!sw_ring) { 442 PMD_INIT_LOG(ERR, "can not allocate RX soft ring"); 443 ret = -ENOMEM; 444 goto fail_q_alloc; 445 } 446 447 vq->sw_ring = sw_ring; 448 rxvq = &vq->rxq; 449 rxvq->vq = vq; 450 rxvq->port_id = dev->data->port_id; 451 rxvq->mz = mz; 452 } else if (queue_type == VTNET_TQ) { 453 txvq = &vq->txq; 454 txvq->vq = vq; 455 txvq->port_id = dev->data->port_id; 456 txvq->mz = mz; 457 txvq->virtio_net_hdr_mz = hdr_mz; 458 txvq->virtio_net_hdr_mem = hdr_mz->iova; 459 } else if (queue_type == VTNET_CQ) { 460 cvq = &vq->cq; 461 cvq->vq = vq; 462 cvq->mz = mz; 463 cvq->virtio_net_hdr_mz = hdr_mz; 464 cvq->virtio_net_hdr_mem = hdr_mz->iova; 465 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE); 466 467 hw->cvq = cvq; 468 } 469 470 /* For virtio_user case (that is when hw->dev is NULL), we use 471 * virtual address. And we need properly set _offset_, please see 472 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information. 473 */ 474 if (!hw->virtio_user_dev) 475 vq->offset = offsetof(struct rte_mbuf, buf_iova); 476 else { 477 vq->vq_ring_mem = (uintptr_t)mz->addr; 478 vq->offset = offsetof(struct rte_mbuf, buf_addr); 479 if (queue_type == VTNET_TQ) 480 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr; 481 else if (queue_type == VTNET_CQ) 482 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr; 483 } 484 485 if (queue_type == VTNET_TQ) { 486 struct virtio_tx_region *txr; 487 unsigned int i; 488 489 txr = hdr_mz->addr; 490 memset(txr, 0, vq_size * sizeof(*txr)); 491 for (i = 0; i < vq_size; i++) { 492 struct vring_desc *start_dp = txr[i].tx_indir; 493 494 vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir)); 495 496 /* first indirect descriptor is always the tx header */ 497 start_dp->addr = txvq->virtio_net_hdr_mem 498 + i * sizeof(*txr) 499 + offsetof(struct virtio_tx_region, tx_hdr); 500 501 start_dp->len = hw->vtnet_hdr_size; 502 start_dp->flags = VRING_DESC_F_NEXT; 503 } 504 } 505 506 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) { 507 PMD_INIT_LOG(ERR, "setup_queue failed"); 508 return -EINVAL; 509 } 510 511 return 0; 512 513 fail_q_alloc: 514 rte_free(sw_ring); 515 rte_memzone_free(hdr_mz); 516 rte_memzone_free(mz); 517 rte_free(vq); 518 519 return ret; 520 } 521 522 static void 523 virtio_free_queues(struct virtio_hw *hw) 524 { 525 uint16_t nr_vq = virtio_get_nr_vq(hw); 526 struct virtqueue *vq; 527 int queue_type; 528 uint16_t i; 529 530 if (hw->vqs == NULL) 531 return; 532 533 for (i = 0; i < nr_vq; i++) { 534 vq = hw->vqs[i]; 535 if (!vq) 536 continue; 537 538 queue_type = virtio_get_queue_type(hw, i); 539 if (queue_type == VTNET_RQ) { 540 rte_free(vq->sw_ring); 541 rte_memzone_free(vq->rxq.mz); 542 } else if (queue_type == VTNET_TQ) { 543 rte_memzone_free(vq->txq.mz); 544 rte_memzone_free(vq->txq.virtio_net_hdr_mz); 545 } else { 546 rte_memzone_free(vq->cq.mz); 547 rte_memzone_free(vq->cq.virtio_net_hdr_mz); 548 } 549 550 rte_free(vq); 551 hw->vqs[i] = NULL; 552 } 553 554 rte_free(hw->vqs); 555 hw->vqs = NULL; 556 } 557 558 static int 559 virtio_alloc_queues(struct rte_eth_dev *dev) 560 { 561 struct virtio_hw *hw = dev->data->dev_private; 562 uint16_t nr_vq = virtio_get_nr_vq(hw); 563 uint16_t i; 564 int ret; 565 566 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0); 567 if (!hw->vqs) { 568 PMD_INIT_LOG(ERR, "failed to allocate vqs"); 569 return -ENOMEM; 570 } 571 572 for (i = 0; i < nr_vq; i++) { 573 ret = virtio_init_queue(dev, i); 574 if (ret < 0) { 575 virtio_free_queues(hw); 576 return ret; 577 } 578 } 579 580 return 0; 581 } 582 583 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev); 584 585 static void 586 virtio_dev_close(struct rte_eth_dev *dev) 587 { 588 struct virtio_hw *hw = dev->data->dev_private; 589 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf; 590 591 PMD_INIT_LOG(DEBUG, "virtio_dev_close"); 592 593 if (!hw->opened) 594 return; 595 hw->opened = false; 596 597 /* reset the NIC */ 598 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) 599 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR); 600 if (intr_conf->rxq) 601 virtio_queues_unbind_intr(dev); 602 603 if (intr_conf->lsc || intr_conf->rxq) { 604 virtio_intr_disable(dev); 605 rte_intr_efd_disable(dev->intr_handle); 606 rte_free(dev->intr_handle->intr_vec); 607 dev->intr_handle->intr_vec = NULL; 608 } 609 610 vtpci_reset(hw); 611 virtio_dev_free_mbufs(dev); 612 virtio_free_queues(hw); 613 } 614 615 static void 616 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev) 617 { 618 struct virtio_hw *hw = dev->data->dev_private; 619 struct virtio_pmd_ctrl ctrl; 620 int dlen[1]; 621 int ret; 622 623 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) { 624 PMD_INIT_LOG(INFO, "host does not support rx control"); 625 return; 626 } 627 628 ctrl.hdr.class = VIRTIO_NET_CTRL_RX; 629 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC; 630 ctrl.data[0] = 1; 631 dlen[0] = 1; 632 633 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1); 634 if (ret) 635 PMD_INIT_LOG(ERR, "Failed to enable promisc"); 636 } 637 638 static void 639 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev) 640 { 641 struct virtio_hw *hw = dev->data->dev_private; 642 struct virtio_pmd_ctrl ctrl; 643 int dlen[1]; 644 int ret; 645 646 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) { 647 PMD_INIT_LOG(INFO, "host does not support rx control"); 648 return; 649 } 650 651 ctrl.hdr.class = VIRTIO_NET_CTRL_RX; 652 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC; 653 ctrl.data[0] = 0; 654 dlen[0] = 1; 655 656 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1); 657 if (ret) 658 PMD_INIT_LOG(ERR, "Failed to disable promisc"); 659 } 660 661 static void 662 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev) 663 { 664 struct virtio_hw *hw = dev->data->dev_private; 665 struct virtio_pmd_ctrl ctrl; 666 int dlen[1]; 667 int ret; 668 669 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) { 670 PMD_INIT_LOG(INFO, "host does not support rx control"); 671 return; 672 } 673 674 ctrl.hdr.class = VIRTIO_NET_CTRL_RX; 675 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI; 676 ctrl.data[0] = 1; 677 dlen[0] = 1; 678 679 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1); 680 if (ret) 681 PMD_INIT_LOG(ERR, "Failed to enable allmulticast"); 682 } 683 684 static void 685 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev) 686 { 687 struct virtio_hw *hw = dev->data->dev_private; 688 struct virtio_pmd_ctrl ctrl; 689 int dlen[1]; 690 int ret; 691 692 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) { 693 PMD_INIT_LOG(INFO, "host does not support rx control"); 694 return; 695 } 696 697 ctrl.hdr.class = VIRTIO_NET_CTRL_RX; 698 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI; 699 ctrl.data[0] = 0; 700 dlen[0] = 1; 701 702 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1); 703 if (ret) 704 PMD_INIT_LOG(ERR, "Failed to disable allmulticast"); 705 } 706 707 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */ 708 static int 709 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu) 710 { 711 struct virtio_hw *hw = dev->data->dev_private; 712 uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN + 713 hw->vtnet_hdr_size; 714 uint32_t frame_size = mtu + ether_hdr_len; 715 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len; 716 717 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN); 718 719 if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) { 720 PMD_INIT_LOG(ERR, "MTU should be between %d and %d", 721 ETHER_MIN_MTU, max_frame_size - ether_hdr_len); 722 return -EINVAL; 723 } 724 return 0; 725 } 726 727 static int 728 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id) 729 { 730 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id]; 731 struct virtqueue *vq = rxvq->vq; 732 733 virtqueue_enable_intr(vq); 734 return 0; 735 } 736 737 static int 738 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id) 739 { 740 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id]; 741 struct virtqueue *vq = rxvq->vq; 742 743 virtqueue_disable_intr(vq); 744 return 0; 745 } 746 747 /* 748 * dev_ops for virtio, bare necessities for basic operation 749 */ 750 static const struct eth_dev_ops virtio_eth_dev_ops = { 751 .dev_configure = virtio_dev_configure, 752 .dev_start = virtio_dev_start, 753 .dev_stop = virtio_dev_stop, 754 .dev_close = virtio_dev_close, 755 .promiscuous_enable = virtio_dev_promiscuous_enable, 756 .promiscuous_disable = virtio_dev_promiscuous_disable, 757 .allmulticast_enable = virtio_dev_allmulticast_enable, 758 .allmulticast_disable = virtio_dev_allmulticast_disable, 759 .mtu_set = virtio_mtu_set, 760 .dev_infos_get = virtio_dev_info_get, 761 .stats_get = virtio_dev_stats_get, 762 .xstats_get = virtio_dev_xstats_get, 763 .xstats_get_names = virtio_dev_xstats_get_names, 764 .stats_reset = virtio_dev_stats_reset, 765 .xstats_reset = virtio_dev_stats_reset, 766 .link_update = virtio_dev_link_update, 767 .vlan_offload_set = virtio_dev_vlan_offload_set, 768 .rx_queue_setup = virtio_dev_rx_queue_setup, 769 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable, 770 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable, 771 .rx_queue_release = virtio_dev_queue_release, 772 .rx_descriptor_done = virtio_dev_rx_queue_done, 773 .tx_queue_setup = virtio_dev_tx_queue_setup, 774 .tx_queue_release = virtio_dev_queue_release, 775 /* collect stats per queue */ 776 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set, 777 .vlan_filter_set = virtio_vlan_filter_set, 778 .mac_addr_add = virtio_mac_addr_add, 779 .mac_addr_remove = virtio_mac_addr_remove, 780 .mac_addr_set = virtio_mac_addr_set, 781 }; 782 783 static void 784 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats) 785 { 786 unsigned i; 787 788 for (i = 0; i < dev->data->nb_tx_queues; i++) { 789 const struct virtnet_tx *txvq = dev->data->tx_queues[i]; 790 if (txvq == NULL) 791 continue; 792 793 stats->opackets += txvq->stats.packets; 794 stats->obytes += txvq->stats.bytes; 795 stats->oerrors += txvq->stats.errors; 796 797 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) { 798 stats->q_opackets[i] = txvq->stats.packets; 799 stats->q_obytes[i] = txvq->stats.bytes; 800 } 801 } 802 803 for (i = 0; i < dev->data->nb_rx_queues; i++) { 804 const struct virtnet_rx *rxvq = dev->data->rx_queues[i]; 805 if (rxvq == NULL) 806 continue; 807 808 stats->ipackets += rxvq->stats.packets; 809 stats->ibytes += rxvq->stats.bytes; 810 stats->ierrors += rxvq->stats.errors; 811 812 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) { 813 stats->q_ipackets[i] = rxvq->stats.packets; 814 stats->q_ibytes[i] = rxvq->stats.bytes; 815 } 816 } 817 818 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed; 819 } 820 821 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev, 822 struct rte_eth_xstat_name *xstats_names, 823 __rte_unused unsigned limit) 824 { 825 unsigned i; 826 unsigned count = 0; 827 unsigned t; 828 829 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS + 830 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS; 831 832 if (xstats_names != NULL) { 833 /* Note: limit checked in rte_eth_xstats_names() */ 834 835 for (i = 0; i < dev->data->nb_rx_queues; i++) { 836 struct virtnet_rx *rxvq = dev->data->rx_queues[i]; 837 if (rxvq == NULL) 838 continue; 839 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) { 840 snprintf(xstats_names[count].name, 841 sizeof(xstats_names[count].name), 842 "rx_q%u_%s", i, 843 rte_virtio_rxq_stat_strings[t].name); 844 count++; 845 } 846 } 847 848 for (i = 0; i < dev->data->nb_tx_queues; i++) { 849 struct virtnet_tx *txvq = dev->data->tx_queues[i]; 850 if (txvq == NULL) 851 continue; 852 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) { 853 snprintf(xstats_names[count].name, 854 sizeof(xstats_names[count].name), 855 "tx_q%u_%s", i, 856 rte_virtio_txq_stat_strings[t].name); 857 count++; 858 } 859 } 860 return count; 861 } 862 return nstats; 863 } 864 865 static int 866 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats, 867 unsigned n) 868 { 869 unsigned i; 870 unsigned count = 0; 871 872 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS + 873 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS; 874 875 if (n < nstats) 876 return nstats; 877 878 for (i = 0; i < dev->data->nb_rx_queues; i++) { 879 struct virtnet_rx *rxvq = dev->data->rx_queues[i]; 880 881 if (rxvq == NULL) 882 continue; 883 884 unsigned t; 885 886 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) { 887 xstats[count].value = *(uint64_t *)(((char *)rxvq) + 888 rte_virtio_rxq_stat_strings[t].offset); 889 xstats[count].id = count; 890 count++; 891 } 892 } 893 894 for (i = 0; i < dev->data->nb_tx_queues; i++) { 895 struct virtnet_tx *txvq = dev->data->tx_queues[i]; 896 897 if (txvq == NULL) 898 continue; 899 900 unsigned t; 901 902 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) { 903 xstats[count].value = *(uint64_t *)(((char *)txvq) + 904 rte_virtio_txq_stat_strings[t].offset); 905 xstats[count].id = count; 906 count++; 907 } 908 } 909 910 return count; 911 } 912 913 static int 914 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) 915 { 916 virtio_update_stats(dev, stats); 917 918 return 0; 919 } 920 921 static void 922 virtio_dev_stats_reset(struct rte_eth_dev *dev) 923 { 924 unsigned int i; 925 926 for (i = 0; i < dev->data->nb_tx_queues; i++) { 927 struct virtnet_tx *txvq = dev->data->tx_queues[i]; 928 if (txvq == NULL) 929 continue; 930 931 txvq->stats.packets = 0; 932 txvq->stats.bytes = 0; 933 txvq->stats.errors = 0; 934 txvq->stats.multicast = 0; 935 txvq->stats.broadcast = 0; 936 memset(txvq->stats.size_bins, 0, 937 sizeof(txvq->stats.size_bins[0]) * 8); 938 } 939 940 for (i = 0; i < dev->data->nb_rx_queues; i++) { 941 struct virtnet_rx *rxvq = dev->data->rx_queues[i]; 942 if (rxvq == NULL) 943 continue; 944 945 rxvq->stats.packets = 0; 946 rxvq->stats.bytes = 0; 947 rxvq->stats.errors = 0; 948 rxvq->stats.multicast = 0; 949 rxvq->stats.broadcast = 0; 950 memset(rxvq->stats.size_bins, 0, 951 sizeof(rxvq->stats.size_bins[0]) * 8); 952 } 953 } 954 955 static void 956 virtio_set_hwaddr(struct virtio_hw *hw) 957 { 958 vtpci_write_dev_config(hw, 959 offsetof(struct virtio_net_config, mac), 960 &hw->mac_addr, ETHER_ADDR_LEN); 961 } 962 963 static void 964 virtio_get_hwaddr(struct virtio_hw *hw) 965 { 966 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) { 967 vtpci_read_dev_config(hw, 968 offsetof(struct virtio_net_config, mac), 969 &hw->mac_addr, ETHER_ADDR_LEN); 970 } else { 971 eth_random_addr(&hw->mac_addr[0]); 972 virtio_set_hwaddr(hw); 973 } 974 } 975 976 static int 977 virtio_mac_table_set(struct virtio_hw *hw, 978 const struct virtio_net_ctrl_mac *uc, 979 const struct virtio_net_ctrl_mac *mc) 980 { 981 struct virtio_pmd_ctrl ctrl; 982 int err, len[2]; 983 984 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) { 985 PMD_DRV_LOG(INFO, "host does not support mac table"); 986 return -1; 987 } 988 989 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC; 990 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; 991 992 len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries); 993 memcpy(ctrl.data, uc, len[0]); 994 995 len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries); 996 memcpy(ctrl.data + len[0], mc, len[1]); 997 998 err = virtio_send_command(hw->cvq, &ctrl, len, 2); 999 if (err != 0) 1000 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err); 1001 return err; 1002 } 1003 1004 static int 1005 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr, 1006 uint32_t index, uint32_t vmdq __rte_unused) 1007 { 1008 struct virtio_hw *hw = dev->data->dev_private; 1009 const struct ether_addr *addrs = dev->data->mac_addrs; 1010 unsigned int i; 1011 struct virtio_net_ctrl_mac *uc, *mc; 1012 1013 if (index >= VIRTIO_MAX_MAC_ADDRS) { 1014 PMD_DRV_LOG(ERR, "mac address index %u out of range", index); 1015 return -EINVAL; 1016 } 1017 1018 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries)); 1019 uc->entries = 0; 1020 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries)); 1021 mc->entries = 0; 1022 1023 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) { 1024 const struct ether_addr *addr 1025 = (i == index) ? mac_addr : addrs + i; 1026 struct virtio_net_ctrl_mac *tbl 1027 = is_multicast_ether_addr(addr) ? mc : uc; 1028 1029 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN); 1030 } 1031 1032 return virtio_mac_table_set(hw, uc, mc); 1033 } 1034 1035 static void 1036 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index) 1037 { 1038 struct virtio_hw *hw = dev->data->dev_private; 1039 struct ether_addr *addrs = dev->data->mac_addrs; 1040 struct virtio_net_ctrl_mac *uc, *mc; 1041 unsigned int i; 1042 1043 if (index >= VIRTIO_MAX_MAC_ADDRS) { 1044 PMD_DRV_LOG(ERR, "mac address index %u out of range", index); 1045 return; 1046 } 1047 1048 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries)); 1049 uc->entries = 0; 1050 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries)); 1051 mc->entries = 0; 1052 1053 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) { 1054 struct virtio_net_ctrl_mac *tbl; 1055 1056 if (i == index || is_zero_ether_addr(addrs + i)) 1057 continue; 1058 1059 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc; 1060 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN); 1061 } 1062 1063 virtio_mac_table_set(hw, uc, mc); 1064 } 1065 1066 static int 1067 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr) 1068 { 1069 struct virtio_hw *hw = dev->data->dev_private; 1070 1071 memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN); 1072 1073 /* Use atomic update if available */ 1074 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) { 1075 struct virtio_pmd_ctrl ctrl; 1076 int len = ETHER_ADDR_LEN; 1077 1078 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC; 1079 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET; 1080 1081 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN); 1082 return virtio_send_command(hw->cvq, &ctrl, &len, 1); 1083 } 1084 1085 if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) 1086 return -ENOTSUP; 1087 1088 virtio_set_hwaddr(hw); 1089 return 0; 1090 } 1091 1092 static int 1093 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on) 1094 { 1095 struct virtio_hw *hw = dev->data->dev_private; 1096 struct virtio_pmd_ctrl ctrl; 1097 int len; 1098 1099 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) 1100 return -ENOTSUP; 1101 1102 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN; 1103 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL; 1104 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id)); 1105 len = sizeof(vlan_id); 1106 1107 return virtio_send_command(hw->cvq, &ctrl, &len, 1); 1108 } 1109 1110 static int 1111 virtio_intr_enable(struct rte_eth_dev *dev) 1112 { 1113 struct virtio_hw *hw = dev->data->dev_private; 1114 1115 if (rte_intr_enable(dev->intr_handle) < 0) 1116 return -1; 1117 1118 if (!hw->virtio_user_dev) 1119 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev)); 1120 1121 return 0; 1122 } 1123 1124 static int 1125 virtio_intr_disable(struct rte_eth_dev *dev) 1126 { 1127 struct virtio_hw *hw = dev->data->dev_private; 1128 1129 if (rte_intr_disable(dev->intr_handle) < 0) 1130 return -1; 1131 1132 if (!hw->virtio_user_dev) 1133 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev)); 1134 1135 return 0; 1136 } 1137 1138 static int 1139 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features) 1140 { 1141 uint64_t host_features; 1142 1143 /* Prepare guest_features: feature that driver wants to support */ 1144 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64, 1145 req_features); 1146 1147 /* Read device(host) feature bits */ 1148 host_features = VTPCI_OPS(hw)->get_features(hw); 1149 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64, 1150 host_features); 1151 1152 /* If supported, ensure MTU value is valid before acknowledging it. */ 1153 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) { 1154 struct virtio_net_config config; 1155 1156 vtpci_read_dev_config(hw, 1157 offsetof(struct virtio_net_config, mtu), 1158 &config.mtu, sizeof(config.mtu)); 1159 1160 if (config.mtu < ETHER_MIN_MTU) 1161 req_features &= ~(1ULL << VIRTIO_NET_F_MTU); 1162 } 1163 1164 /* 1165 * Negotiate features: Subset of device feature bits are written back 1166 * guest feature bits. 1167 */ 1168 hw->guest_features = req_features; 1169 hw->guest_features = vtpci_negotiate_features(hw, host_features); 1170 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64, 1171 hw->guest_features); 1172 1173 if (hw->modern) { 1174 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) { 1175 PMD_INIT_LOG(ERR, 1176 "VIRTIO_F_VERSION_1 features is not enabled."); 1177 return -1; 1178 } 1179 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK); 1180 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) { 1181 PMD_INIT_LOG(ERR, 1182 "failed to set FEATURES_OK status!"); 1183 return -1; 1184 } 1185 } 1186 1187 hw->req_guest_features = req_features; 1188 1189 return 0; 1190 } 1191 1192 int 1193 virtio_dev_pause(struct rte_eth_dev *dev) 1194 { 1195 struct virtio_hw *hw = dev->data->dev_private; 1196 1197 rte_spinlock_lock(&hw->state_lock); 1198 1199 if (hw->started == 0) { 1200 /* Device is just stopped. */ 1201 rte_spinlock_unlock(&hw->state_lock); 1202 return -1; 1203 } 1204 hw->started = 0; 1205 /* 1206 * Prevent the worker threads from touching queues to avoid contention, 1207 * 1 ms should be enough for the ongoing Tx function to finish. 1208 */ 1209 rte_delay_ms(1); 1210 return 0; 1211 } 1212 1213 /* 1214 * Recover hw state to let the worker threads continue. 1215 */ 1216 void 1217 virtio_dev_resume(struct rte_eth_dev *dev) 1218 { 1219 struct virtio_hw *hw = dev->data->dev_private; 1220 1221 hw->started = 1; 1222 rte_spinlock_unlock(&hw->state_lock); 1223 } 1224 1225 /* 1226 * Should be called only after device is paused. 1227 */ 1228 int 1229 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts, 1230 int nb_pkts) 1231 { 1232 struct virtio_hw *hw = dev->data->dev_private; 1233 struct virtnet_tx *txvq = dev->data->tx_queues[0]; 1234 int ret; 1235 1236 hw->inject_pkts = tx_pkts; 1237 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts); 1238 hw->inject_pkts = NULL; 1239 1240 return ret; 1241 } 1242 1243 static void 1244 virtio_notify_peers(struct rte_eth_dev *dev) 1245 { 1246 struct virtio_hw *hw = dev->data->dev_private; 1247 struct virtnet_rx *rxvq; 1248 struct rte_mbuf *rarp_mbuf; 1249 1250 if (!dev->data->rx_queues) 1251 return; 1252 1253 rxvq = dev->data->rx_queues[0]; 1254 if (!rxvq) 1255 return; 1256 1257 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool, 1258 (struct ether_addr *)hw->mac_addr); 1259 if (rarp_mbuf == NULL) { 1260 PMD_DRV_LOG(ERR, "failed to make RARP packet."); 1261 return; 1262 } 1263 1264 /* If virtio port just stopped, no need to send RARP */ 1265 if (virtio_dev_pause(dev) < 0) { 1266 rte_pktmbuf_free(rarp_mbuf); 1267 return; 1268 } 1269 1270 virtio_inject_pkts(dev, &rarp_mbuf, 1); 1271 virtio_dev_resume(dev); 1272 } 1273 1274 static void 1275 virtio_ack_link_announce(struct rte_eth_dev *dev) 1276 { 1277 struct virtio_hw *hw = dev->data->dev_private; 1278 struct virtio_pmd_ctrl ctrl; 1279 1280 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE; 1281 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK; 1282 1283 virtio_send_command(hw->cvq, &ctrl, NULL, 0); 1284 } 1285 1286 /* 1287 * Process virtio config changed interrupt. Call the callback 1288 * if link state changed, generate gratuitous RARP packet if 1289 * the status indicates an ANNOUNCE. 1290 */ 1291 void 1292 virtio_interrupt_handler(void *param) 1293 { 1294 struct rte_eth_dev *dev = param; 1295 struct virtio_hw *hw = dev->data->dev_private; 1296 uint8_t isr; 1297 uint16_t status; 1298 1299 /* Read interrupt status which clears interrupt */ 1300 isr = vtpci_isr(hw); 1301 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr); 1302 1303 if (virtio_intr_enable(dev) < 0) 1304 PMD_DRV_LOG(ERR, "interrupt enable failed"); 1305 1306 if (isr & VIRTIO_PCI_ISR_CONFIG) { 1307 if (virtio_dev_link_update(dev, 0) == 0) 1308 _rte_eth_dev_callback_process(dev, 1309 RTE_ETH_EVENT_INTR_LSC, 1310 NULL); 1311 1312 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) { 1313 vtpci_read_dev_config(hw, 1314 offsetof(struct virtio_net_config, status), 1315 &status, sizeof(status)); 1316 if (status & VIRTIO_NET_S_ANNOUNCE) { 1317 virtio_notify_peers(dev); 1318 if (hw->cvq) 1319 virtio_ack_link_announce(dev); 1320 } 1321 } 1322 } 1323 } 1324 1325 /* set rx and tx handlers according to what is supported */ 1326 static void 1327 set_rxtx_funcs(struct rte_eth_dev *eth_dev) 1328 { 1329 struct virtio_hw *hw = eth_dev->data->dev_private; 1330 1331 if (hw->use_simple_rx) { 1332 PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u", 1333 eth_dev->data->port_id); 1334 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec; 1335 } else if (hw->use_inorder_rx) { 1336 PMD_INIT_LOG(INFO, 1337 "virtio: using inorder mergeable buffer Rx path on port %u", 1338 eth_dev->data->port_id); 1339 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts_inorder; 1340 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) { 1341 PMD_INIT_LOG(INFO, 1342 "virtio: using mergeable buffer Rx path on port %u", 1343 eth_dev->data->port_id); 1344 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts; 1345 } else { 1346 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u", 1347 eth_dev->data->port_id); 1348 eth_dev->rx_pkt_burst = &virtio_recv_pkts; 1349 } 1350 1351 if (hw->use_inorder_tx) { 1352 PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u", 1353 eth_dev->data->port_id); 1354 eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder; 1355 } else { 1356 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u", 1357 eth_dev->data->port_id); 1358 eth_dev->tx_pkt_burst = virtio_xmit_pkts; 1359 } 1360 } 1361 1362 /* Only support 1:1 queue/interrupt mapping so far. 1363 * TODO: support n:1 queue/interrupt mapping when there are limited number of 1364 * interrupt vectors (<N+1). 1365 */ 1366 static int 1367 virtio_queues_bind_intr(struct rte_eth_dev *dev) 1368 { 1369 uint32_t i; 1370 struct virtio_hw *hw = dev->data->dev_private; 1371 1372 PMD_INIT_LOG(INFO, "queue/interrupt binding"); 1373 for (i = 0; i < dev->data->nb_rx_queues; ++i) { 1374 dev->intr_handle->intr_vec[i] = i + 1; 1375 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) == 1376 VIRTIO_MSI_NO_VECTOR) { 1377 PMD_DRV_LOG(ERR, "failed to set queue vector"); 1378 return -EBUSY; 1379 } 1380 } 1381 1382 return 0; 1383 } 1384 1385 static void 1386 virtio_queues_unbind_intr(struct rte_eth_dev *dev) 1387 { 1388 uint32_t i; 1389 struct virtio_hw *hw = dev->data->dev_private; 1390 1391 PMD_INIT_LOG(INFO, "queue/interrupt unbinding"); 1392 for (i = 0; i < dev->data->nb_rx_queues; ++i) 1393 VTPCI_OPS(hw)->set_queue_irq(hw, 1394 hw->vqs[i * VTNET_CQ], 1395 VIRTIO_MSI_NO_VECTOR); 1396 } 1397 1398 static int 1399 virtio_configure_intr(struct rte_eth_dev *dev) 1400 { 1401 struct virtio_hw *hw = dev->data->dev_private; 1402 1403 if (!rte_intr_cap_multiple(dev->intr_handle)) { 1404 PMD_INIT_LOG(ERR, "Multiple intr vector not supported"); 1405 return -ENOTSUP; 1406 } 1407 1408 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) { 1409 PMD_INIT_LOG(ERR, "Fail to create eventfd"); 1410 return -1; 1411 } 1412 1413 if (!dev->intr_handle->intr_vec) { 1414 dev->intr_handle->intr_vec = 1415 rte_zmalloc("intr_vec", 1416 hw->max_queue_pairs * sizeof(int), 0); 1417 if (!dev->intr_handle->intr_vec) { 1418 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors", 1419 hw->max_queue_pairs); 1420 return -ENOMEM; 1421 } 1422 } 1423 1424 /* Re-register callback to update max_intr */ 1425 rte_intr_callback_unregister(dev->intr_handle, 1426 virtio_interrupt_handler, 1427 dev); 1428 rte_intr_callback_register(dev->intr_handle, 1429 virtio_interrupt_handler, 1430 dev); 1431 1432 /* DO NOT try to remove this! This function will enable msix, or QEMU 1433 * will encounter SIGSEGV when DRIVER_OK is sent. 1434 * And for legacy devices, this should be done before queue/vec binding 1435 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR 1436 * (22) will be ignored. 1437 */ 1438 if (virtio_intr_enable(dev) < 0) { 1439 PMD_DRV_LOG(ERR, "interrupt enable failed"); 1440 return -1; 1441 } 1442 1443 if (virtio_queues_bind_intr(dev) < 0) { 1444 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt"); 1445 return -1; 1446 } 1447 1448 return 0; 1449 } 1450 1451 /* reset device and renegotiate features if needed */ 1452 static int 1453 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features) 1454 { 1455 struct virtio_hw *hw = eth_dev->data->dev_private; 1456 struct virtio_net_config *config; 1457 struct virtio_net_config local_config; 1458 struct rte_pci_device *pci_dev = NULL; 1459 int ret; 1460 1461 /* Reset the device although not necessary at startup */ 1462 vtpci_reset(hw); 1463 1464 if (hw->vqs) { 1465 virtio_dev_free_mbufs(eth_dev); 1466 virtio_free_queues(hw); 1467 } 1468 1469 /* Tell the host we've noticed this device. */ 1470 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK); 1471 1472 /* Tell the host we've known how to drive the device. */ 1473 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER); 1474 if (virtio_negotiate_features(hw, req_features) < 0) 1475 return -1; 1476 1477 if (!hw->virtio_user_dev) { 1478 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev); 1479 rte_eth_copy_pci_info(eth_dev, pci_dev); 1480 } 1481 1482 /* If host does not support both status and MSI-X then disable LSC */ 1483 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) && 1484 hw->use_msix != VIRTIO_MSIX_NONE) 1485 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC; 1486 else 1487 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC; 1488 1489 /* Setting up rx_header size for the device */ 1490 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) || 1491 vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) 1492 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf); 1493 else 1494 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr); 1495 1496 /* Copy the permanent MAC address to: virtio_hw */ 1497 virtio_get_hwaddr(hw); 1498 ether_addr_copy((struct ether_addr *) hw->mac_addr, 1499 ð_dev->data->mac_addrs[0]); 1500 PMD_INIT_LOG(DEBUG, 1501 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X", 1502 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2], 1503 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]); 1504 1505 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) { 1506 config = &local_config; 1507 1508 vtpci_read_dev_config(hw, 1509 offsetof(struct virtio_net_config, mac), 1510 &config->mac, sizeof(config->mac)); 1511 1512 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) { 1513 vtpci_read_dev_config(hw, 1514 offsetof(struct virtio_net_config, status), 1515 &config->status, sizeof(config->status)); 1516 } else { 1517 PMD_INIT_LOG(DEBUG, 1518 "VIRTIO_NET_F_STATUS is not supported"); 1519 config->status = 0; 1520 } 1521 1522 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) { 1523 vtpci_read_dev_config(hw, 1524 offsetof(struct virtio_net_config, max_virtqueue_pairs), 1525 &config->max_virtqueue_pairs, 1526 sizeof(config->max_virtqueue_pairs)); 1527 } else { 1528 PMD_INIT_LOG(DEBUG, 1529 "VIRTIO_NET_F_MQ is not supported"); 1530 config->max_virtqueue_pairs = 1; 1531 } 1532 1533 hw->max_queue_pairs = config->max_virtqueue_pairs; 1534 1535 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) { 1536 vtpci_read_dev_config(hw, 1537 offsetof(struct virtio_net_config, mtu), 1538 &config->mtu, 1539 sizeof(config->mtu)); 1540 1541 /* 1542 * MTU value has already been checked at negotiation 1543 * time, but check again in case it has changed since 1544 * then, which should not happen. 1545 */ 1546 if (config->mtu < ETHER_MIN_MTU) { 1547 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)", 1548 config->mtu); 1549 return -1; 1550 } 1551 1552 hw->max_mtu = config->mtu; 1553 /* Set initial MTU to maximum one supported by vhost */ 1554 eth_dev->data->mtu = config->mtu; 1555 1556 } else { 1557 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN - 1558 VLAN_TAG_LEN - hw->vtnet_hdr_size; 1559 } 1560 1561 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d", 1562 config->max_virtqueue_pairs); 1563 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status); 1564 PMD_INIT_LOG(DEBUG, 1565 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X", 1566 config->mac[0], config->mac[1], 1567 config->mac[2], config->mac[3], 1568 config->mac[4], config->mac[5]); 1569 } else { 1570 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1"); 1571 hw->max_queue_pairs = 1; 1572 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN - 1573 VLAN_TAG_LEN - hw->vtnet_hdr_size; 1574 } 1575 1576 ret = virtio_alloc_queues(eth_dev); 1577 if (ret < 0) 1578 return ret; 1579 1580 if (eth_dev->data->dev_conf.intr_conf.rxq) { 1581 if (virtio_configure_intr(eth_dev) < 0) { 1582 PMD_INIT_LOG(ERR, "failed to configure interrupt"); 1583 return -1; 1584 } 1585 } 1586 1587 vtpci_reinit_complete(hw); 1588 1589 if (pci_dev) 1590 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x", 1591 eth_dev->data->port_id, pci_dev->id.vendor_id, 1592 pci_dev->id.device_id); 1593 1594 return 0; 1595 } 1596 1597 /* 1598 * Remap the PCI device again (IO port map for legacy device and 1599 * memory map for modern device), so that the secondary process 1600 * could have the PCI initiated correctly. 1601 */ 1602 static int 1603 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw) 1604 { 1605 if (hw->modern) { 1606 /* 1607 * We don't have to re-parse the PCI config space, since 1608 * rte_pci_map_device() makes sure the mapped address 1609 * in secondary process would equal to the one mapped in 1610 * the primary process: error will be returned if that 1611 * requirement is not met. 1612 * 1613 * That said, we could simply reuse all cap pointers 1614 * (such as dev_cfg, common_cfg, etc.) parsed from the 1615 * primary process, which is stored in shared memory. 1616 */ 1617 if (rte_pci_map_device(pci_dev)) { 1618 PMD_INIT_LOG(DEBUG, "failed to map pci device!"); 1619 return -1; 1620 } 1621 } else { 1622 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0) 1623 return -1; 1624 } 1625 1626 return 0; 1627 } 1628 1629 static void 1630 virtio_set_vtpci_ops(struct virtio_hw *hw) 1631 { 1632 #ifdef RTE_VIRTIO_USER 1633 if (hw->virtio_user_dev) 1634 VTPCI_OPS(hw) = &virtio_user_ops; 1635 else 1636 #endif 1637 if (hw->modern) 1638 VTPCI_OPS(hw) = &modern_ops; 1639 else 1640 VTPCI_OPS(hw) = &legacy_ops; 1641 } 1642 1643 /* 1644 * This function is based on probe() function in virtio_pci.c 1645 * It returns 0 on success. 1646 */ 1647 int 1648 eth_virtio_dev_init(struct rte_eth_dev *eth_dev) 1649 { 1650 struct virtio_hw *hw = eth_dev->data->dev_private; 1651 int ret; 1652 1653 RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf)); 1654 1655 eth_dev->dev_ops = &virtio_eth_dev_ops; 1656 1657 if (rte_eal_process_type() == RTE_PROC_SECONDARY) { 1658 if (!hw->virtio_user_dev) { 1659 ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw); 1660 if (ret) 1661 return ret; 1662 } 1663 1664 virtio_set_vtpci_ops(hw); 1665 set_rxtx_funcs(eth_dev); 1666 1667 return 0; 1668 } 1669 1670 /* Allocate memory for storing MAC addresses */ 1671 eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0); 1672 if (eth_dev->data->mac_addrs == NULL) { 1673 PMD_INIT_LOG(ERR, 1674 "Failed to allocate %d bytes needed to store MAC addresses", 1675 VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN); 1676 return -ENOMEM; 1677 } 1678 1679 hw->port_id = eth_dev->data->port_id; 1680 /* For virtio_user case the hw->virtio_user_dev is populated by 1681 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called. 1682 */ 1683 if (!hw->virtio_user_dev) { 1684 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw); 1685 if (ret) 1686 goto out; 1687 } 1688 1689 /* reset device and negotiate default features */ 1690 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES); 1691 if (ret < 0) 1692 goto out; 1693 1694 return 0; 1695 1696 out: 1697 rte_free(eth_dev->data->mac_addrs); 1698 return ret; 1699 } 1700 1701 static int 1702 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev) 1703 { 1704 PMD_INIT_FUNC_TRACE(); 1705 1706 if (rte_eal_process_type() == RTE_PROC_SECONDARY) 1707 return 0; 1708 1709 virtio_dev_stop(eth_dev); 1710 virtio_dev_close(eth_dev); 1711 1712 eth_dev->dev_ops = NULL; 1713 eth_dev->tx_pkt_burst = NULL; 1714 eth_dev->rx_pkt_burst = NULL; 1715 1716 if (eth_dev->device) 1717 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev)); 1718 1719 PMD_INIT_LOG(DEBUG, "dev_uninit completed"); 1720 1721 return 0; 1722 } 1723 1724 static int vdpa_check_handler(__rte_unused const char *key, 1725 const char *value, __rte_unused void *opaque) 1726 { 1727 if (strcmp(value, "1")) 1728 return -1; 1729 1730 return 0; 1731 } 1732 1733 static int 1734 vdpa_mode_selected(struct rte_devargs *devargs) 1735 { 1736 struct rte_kvargs *kvlist; 1737 const char *key = "vdpa"; 1738 int ret = 0; 1739 1740 if (devargs == NULL) 1741 return 0; 1742 1743 kvlist = rte_kvargs_parse(devargs->args, NULL); 1744 if (kvlist == NULL) 1745 return 0; 1746 1747 if (!rte_kvargs_count(kvlist, key)) 1748 goto exit; 1749 1750 /* vdpa mode selected when there's a key-value pair: vdpa=1 */ 1751 if (rte_kvargs_process(kvlist, key, 1752 vdpa_check_handler, NULL) < 0) { 1753 goto exit; 1754 } 1755 ret = 1; 1756 1757 exit: 1758 rte_kvargs_free(kvlist); 1759 return ret; 1760 } 1761 1762 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, 1763 struct rte_pci_device *pci_dev) 1764 { 1765 if (rte_eal_iopl_init() != 0) { 1766 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD"); 1767 return 1; 1768 } 1769 1770 /* virtio pmd skips probe if device needs to work in vdpa mode */ 1771 if (vdpa_mode_selected(pci_dev->device.devargs)) 1772 return 1; 1773 1774 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw), 1775 eth_virtio_dev_init); 1776 } 1777 1778 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev) 1779 { 1780 return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit); 1781 } 1782 1783 static struct rte_pci_driver rte_virtio_pmd = { 1784 .driver = { 1785 .name = "net_virtio", 1786 }, 1787 .id_table = pci_id_virtio_map, 1788 .drv_flags = 0, 1789 .probe = eth_virtio_pci_probe, 1790 .remove = eth_virtio_pci_remove, 1791 }; 1792 1793 RTE_INIT(rte_virtio_pmd_init) 1794 { 1795 rte_eal_iopl_init(); 1796 rte_pci_register(&rte_virtio_pmd); 1797 } 1798 1799 static bool 1800 rx_offload_enabled(struct virtio_hw *hw) 1801 { 1802 return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) || 1803 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) || 1804 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6); 1805 } 1806 1807 static bool 1808 tx_offload_enabled(struct virtio_hw *hw) 1809 { 1810 return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) || 1811 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) || 1812 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6); 1813 } 1814 1815 /* 1816 * Configure virtio device 1817 * It returns 0 on success. 1818 */ 1819 static int 1820 virtio_dev_configure(struct rte_eth_dev *dev) 1821 { 1822 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode; 1823 const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode; 1824 struct virtio_hw *hw = dev->data->dev_private; 1825 uint64_t rx_offloads = rxmode->offloads; 1826 uint64_t tx_offloads = txmode->offloads; 1827 uint64_t req_features; 1828 int ret; 1829 1830 PMD_INIT_LOG(DEBUG, "configure"); 1831 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES; 1832 1833 if (dev->data->dev_conf.intr_conf.rxq) { 1834 ret = virtio_init_device(dev, hw->req_guest_features); 1835 if (ret < 0) 1836 return ret; 1837 } 1838 1839 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM | 1840 DEV_RX_OFFLOAD_TCP_CKSUM)) 1841 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM); 1842 1843 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) 1844 req_features |= 1845 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | 1846 (1ULL << VIRTIO_NET_F_GUEST_TSO6); 1847 1848 if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM | 1849 DEV_TX_OFFLOAD_TCP_CKSUM)) 1850 req_features |= (1ULL << VIRTIO_NET_F_CSUM); 1851 1852 if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO) 1853 req_features |= 1854 (1ULL << VIRTIO_NET_F_HOST_TSO4) | 1855 (1ULL << VIRTIO_NET_F_HOST_TSO6); 1856 1857 /* if request features changed, reinit the device */ 1858 if (req_features != hw->req_guest_features) { 1859 ret = virtio_init_device(dev, req_features); 1860 if (ret < 0) 1861 return ret; 1862 } 1863 1864 if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM | 1865 DEV_RX_OFFLOAD_TCP_CKSUM)) && 1866 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) { 1867 PMD_DRV_LOG(ERR, 1868 "rx checksum not available on this host"); 1869 return -ENOTSUP; 1870 } 1871 1872 if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) && 1873 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) || 1874 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) { 1875 PMD_DRV_LOG(ERR, 1876 "Large Receive Offload not available on this host"); 1877 return -ENOTSUP; 1878 } 1879 1880 /* start control queue */ 1881 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) 1882 virtio_dev_cq_start(dev); 1883 1884 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP) 1885 hw->vlan_strip = 1; 1886 1887 if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER) 1888 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) { 1889 PMD_DRV_LOG(ERR, 1890 "vlan filtering not available on this host"); 1891 return -ENOTSUP; 1892 } 1893 1894 hw->has_tx_offload = tx_offload_enabled(hw); 1895 hw->has_rx_offload = rx_offload_enabled(hw); 1896 1897 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) 1898 /* Enable vector (0) for Link State Intrerrupt */ 1899 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) == 1900 VIRTIO_MSI_NO_VECTOR) { 1901 PMD_DRV_LOG(ERR, "failed to set config vector"); 1902 return -EBUSY; 1903 } 1904 1905 rte_spinlock_init(&hw->state_lock); 1906 1907 hw->use_simple_rx = 1; 1908 1909 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) { 1910 hw->use_inorder_tx = 1; 1911 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) { 1912 hw->use_inorder_rx = 1; 1913 hw->use_simple_rx = 0; 1914 } else { 1915 hw->use_inorder_rx = 0; 1916 } 1917 } 1918 1919 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM 1920 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) { 1921 hw->use_simple_rx = 0; 1922 } 1923 #endif 1924 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) { 1925 hw->use_simple_rx = 0; 1926 } 1927 1928 if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM | 1929 DEV_RX_OFFLOAD_TCP_CKSUM | 1930 DEV_RX_OFFLOAD_TCP_LRO | 1931 DEV_RX_OFFLOAD_VLAN_STRIP)) 1932 hw->use_simple_rx = 0; 1933 1934 hw->opened = true; 1935 1936 return 0; 1937 } 1938 1939 1940 static int 1941 virtio_dev_start(struct rte_eth_dev *dev) 1942 { 1943 uint16_t nb_queues, i; 1944 struct virtnet_rx *rxvq; 1945 struct virtnet_tx *txvq __rte_unused; 1946 struct virtio_hw *hw = dev->data->dev_private; 1947 int ret; 1948 1949 /* Finish the initialization of the queues */ 1950 for (i = 0; i < dev->data->nb_rx_queues; i++) { 1951 ret = virtio_dev_rx_queue_setup_finish(dev, i); 1952 if (ret < 0) 1953 return ret; 1954 } 1955 for (i = 0; i < dev->data->nb_tx_queues; i++) { 1956 ret = virtio_dev_tx_queue_setup_finish(dev, i); 1957 if (ret < 0) 1958 return ret; 1959 } 1960 1961 /* check if lsc interrupt feature is enabled */ 1962 if (dev->data->dev_conf.intr_conf.lsc) { 1963 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) { 1964 PMD_DRV_LOG(ERR, "link status not supported by host"); 1965 return -ENOTSUP; 1966 } 1967 } 1968 1969 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that 1970 * in device configure, but it could be unmapped when device is 1971 * stopped. 1972 */ 1973 if (dev->data->dev_conf.intr_conf.lsc || 1974 dev->data->dev_conf.intr_conf.rxq) { 1975 virtio_intr_disable(dev); 1976 1977 /* Setup interrupt callback */ 1978 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) 1979 rte_intr_callback_register(dev->intr_handle, 1980 virtio_interrupt_handler, 1981 dev); 1982 1983 if (virtio_intr_enable(dev) < 0) { 1984 PMD_DRV_LOG(ERR, "interrupt enable failed"); 1985 return -EIO; 1986 } 1987 } 1988 1989 /*Notify the backend 1990 *Otherwise the tap backend might already stop its queue due to fullness. 1991 *vhost backend will have no chance to be waked up 1992 */ 1993 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues); 1994 if (hw->max_queue_pairs > 1) { 1995 if (virtio_set_multiple_queues(dev, nb_queues) != 0) 1996 return -EINVAL; 1997 } 1998 1999 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues); 2000 2001 for (i = 0; i < dev->data->nb_rx_queues; i++) { 2002 rxvq = dev->data->rx_queues[i]; 2003 /* Flush the old packets */ 2004 virtqueue_rxvq_flush(rxvq->vq); 2005 virtqueue_notify(rxvq->vq); 2006 } 2007 2008 for (i = 0; i < dev->data->nb_tx_queues; i++) { 2009 txvq = dev->data->tx_queues[i]; 2010 virtqueue_notify(txvq->vq); 2011 } 2012 2013 PMD_INIT_LOG(DEBUG, "Notified backend at initialization"); 2014 2015 for (i = 0; i < dev->data->nb_rx_queues; i++) { 2016 rxvq = dev->data->rx_queues[i]; 2017 VIRTQUEUE_DUMP(rxvq->vq); 2018 } 2019 2020 for (i = 0; i < dev->data->nb_tx_queues; i++) { 2021 txvq = dev->data->tx_queues[i]; 2022 VIRTQUEUE_DUMP(txvq->vq); 2023 } 2024 2025 set_rxtx_funcs(dev); 2026 hw->started = true; 2027 2028 /* Initialize Link state */ 2029 virtio_dev_link_update(dev, 0); 2030 2031 return 0; 2032 } 2033 2034 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev) 2035 { 2036 struct virtio_hw *hw = dev->data->dev_private; 2037 uint16_t nr_vq = virtio_get_nr_vq(hw); 2038 const char *type __rte_unused; 2039 unsigned int i, mbuf_num = 0; 2040 struct virtqueue *vq; 2041 struct rte_mbuf *buf; 2042 int queue_type; 2043 2044 if (hw->vqs == NULL) 2045 return; 2046 2047 for (i = 0; i < nr_vq; i++) { 2048 vq = hw->vqs[i]; 2049 if (!vq) 2050 continue; 2051 2052 queue_type = virtio_get_queue_type(hw, i); 2053 if (queue_type == VTNET_RQ) 2054 type = "rxq"; 2055 else if (queue_type == VTNET_TQ) 2056 type = "txq"; 2057 else 2058 continue; 2059 2060 PMD_INIT_LOG(DEBUG, 2061 "Before freeing %s[%d] used and unused buf", 2062 type, i); 2063 VIRTQUEUE_DUMP(vq); 2064 2065 while ((buf = virtqueue_detach_unused(vq)) != NULL) { 2066 rte_pktmbuf_free(buf); 2067 mbuf_num++; 2068 } 2069 2070 PMD_INIT_LOG(DEBUG, 2071 "After freeing %s[%d] used and unused buf", 2072 type, i); 2073 VIRTQUEUE_DUMP(vq); 2074 } 2075 2076 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num); 2077 } 2078 2079 /* 2080 * Stop device: disable interrupt and mark link down 2081 */ 2082 static void 2083 virtio_dev_stop(struct rte_eth_dev *dev) 2084 { 2085 struct virtio_hw *hw = dev->data->dev_private; 2086 struct rte_eth_link link; 2087 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf; 2088 2089 PMD_INIT_LOG(DEBUG, "stop"); 2090 2091 rte_spinlock_lock(&hw->state_lock); 2092 if (!hw->started) 2093 goto out_unlock; 2094 hw->started = false; 2095 2096 if (intr_conf->lsc || intr_conf->rxq) { 2097 virtio_intr_disable(dev); 2098 2099 /* Reset interrupt callback */ 2100 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) { 2101 rte_intr_callback_unregister(dev->intr_handle, 2102 virtio_interrupt_handler, 2103 dev); 2104 } 2105 } 2106 2107 memset(&link, 0, sizeof(link)); 2108 rte_eth_linkstatus_set(dev, &link); 2109 out_unlock: 2110 rte_spinlock_unlock(&hw->state_lock); 2111 } 2112 2113 static int 2114 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete) 2115 { 2116 struct rte_eth_link link; 2117 uint16_t status; 2118 struct virtio_hw *hw = dev->data->dev_private; 2119 2120 memset(&link, 0, sizeof(link)); 2121 link.link_duplex = ETH_LINK_FULL_DUPLEX; 2122 link.link_speed = ETH_SPEED_NUM_10G; 2123 link.link_autoneg = ETH_LINK_FIXED; 2124 2125 if (!hw->started) { 2126 link.link_status = ETH_LINK_DOWN; 2127 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) { 2128 PMD_INIT_LOG(DEBUG, "Get link status from hw"); 2129 vtpci_read_dev_config(hw, 2130 offsetof(struct virtio_net_config, status), 2131 &status, sizeof(status)); 2132 if ((status & VIRTIO_NET_S_LINK_UP) == 0) { 2133 link.link_status = ETH_LINK_DOWN; 2134 PMD_INIT_LOG(DEBUG, "Port %d is down", 2135 dev->data->port_id); 2136 } else { 2137 link.link_status = ETH_LINK_UP; 2138 PMD_INIT_LOG(DEBUG, "Port %d is up", 2139 dev->data->port_id); 2140 } 2141 } else { 2142 link.link_status = ETH_LINK_UP; 2143 } 2144 2145 return rte_eth_linkstatus_set(dev, &link); 2146 } 2147 2148 static int 2149 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask) 2150 { 2151 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode; 2152 struct virtio_hw *hw = dev->data->dev_private; 2153 uint64_t offloads = rxmode->offloads; 2154 2155 if (mask & ETH_VLAN_FILTER_MASK) { 2156 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) && 2157 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) { 2158 2159 PMD_DRV_LOG(NOTICE, 2160 "vlan filtering not available on this host"); 2161 2162 return -ENOTSUP; 2163 } 2164 } 2165 2166 if (mask & ETH_VLAN_STRIP_MASK) 2167 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP); 2168 2169 return 0; 2170 } 2171 2172 static void 2173 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info) 2174 { 2175 uint64_t tso_mask, host_features; 2176 struct virtio_hw *hw = dev->data->dev_private; 2177 2178 dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */ 2179 2180 dev_info->max_rx_queues = 2181 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES); 2182 dev_info->max_tx_queues = 2183 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES); 2184 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE; 2185 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN; 2186 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS; 2187 2188 host_features = VTPCI_OPS(hw)->get_features(hw); 2189 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP; 2190 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) { 2191 dev_info->rx_offload_capa |= 2192 DEV_RX_OFFLOAD_TCP_CKSUM | 2193 DEV_RX_OFFLOAD_UDP_CKSUM; 2194 } 2195 if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN)) 2196 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER; 2197 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) | 2198 (1ULL << VIRTIO_NET_F_GUEST_TSO6); 2199 if ((host_features & tso_mask) == tso_mask) 2200 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO; 2201 2202 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS | 2203 DEV_TX_OFFLOAD_VLAN_INSERT; 2204 if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) { 2205 dev_info->tx_offload_capa |= 2206 DEV_TX_OFFLOAD_UDP_CKSUM | 2207 DEV_TX_OFFLOAD_TCP_CKSUM; 2208 } 2209 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) | 2210 (1ULL << VIRTIO_NET_F_HOST_TSO6); 2211 if ((host_features & tso_mask) == tso_mask) 2212 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO; 2213 } 2214 2215 /* 2216 * It enables testpmd to collect per queue stats. 2217 */ 2218 static int 2219 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev, 2220 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx, 2221 __rte_unused uint8_t is_rx) 2222 { 2223 return 0; 2224 } 2225 2226 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__); 2227 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map); 2228 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci"); 2229 2230 RTE_INIT(virtio_init_log) 2231 { 2232 virtio_logtype_init = rte_log_register("pmd.net.virtio.init"); 2233 if (virtio_logtype_init >= 0) 2234 rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE); 2235 virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver"); 2236 if (virtio_logtype_driver >= 0) 2237 rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE); 2238 } 2239