1 /*- 2 * BSD LICENSE 3 * 4 * Copyright(c) 2010-2016 Intel Corporation. All rights reserved. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <stdint.h> 35 #include <string.h> 36 #include <stdio.h> 37 #include <errno.h> 38 #include <unistd.h> 39 40 #include <rte_ethdev.h> 41 #include <rte_memcpy.h> 42 #include <rte_string_fns.h> 43 #include <rte_memzone.h> 44 #include <rte_malloc.h> 45 #include <rte_atomic.h> 46 #include <rte_branch_prediction.h> 47 #include <rte_pci.h> 48 #include <rte_ether.h> 49 #include <rte_common.h> 50 #include <rte_errno.h> 51 52 #include <rte_memory.h> 53 #include <rte_eal.h> 54 #include <rte_dev.h> 55 56 #include "virtio_ethdev.h" 57 #include "virtio_pci.h" 58 #include "virtio_logs.h" 59 #include "virtqueue.h" 60 #include "virtio_rxtx.h" 61 62 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev); 63 static int virtio_dev_configure(struct rte_eth_dev *dev); 64 static int virtio_dev_start(struct rte_eth_dev *dev); 65 static void virtio_dev_stop(struct rte_eth_dev *dev); 66 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev); 67 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev); 68 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev); 69 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev); 70 static void virtio_dev_info_get(struct rte_eth_dev *dev, 71 struct rte_eth_dev_info *dev_info); 72 static int virtio_dev_link_update(struct rte_eth_dev *dev, 73 __rte_unused int wait_to_complete); 74 75 static void virtio_set_hwaddr(struct virtio_hw *hw); 76 static void virtio_get_hwaddr(struct virtio_hw *hw); 77 78 static void virtio_dev_stats_get(struct rte_eth_dev *dev, 79 struct rte_eth_stats *stats); 80 static int virtio_dev_xstats_get(struct rte_eth_dev *dev, 81 struct rte_eth_xstat *xstats, unsigned n); 82 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev, 83 struct rte_eth_xstat_name *xstats_names, 84 unsigned limit); 85 static void virtio_dev_stats_reset(struct rte_eth_dev *dev); 86 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev); 87 static int virtio_vlan_filter_set(struct rte_eth_dev *dev, 88 uint16_t vlan_id, int on); 89 static void virtio_mac_addr_add(struct rte_eth_dev *dev, 90 struct ether_addr *mac_addr, 91 uint32_t index, uint32_t vmdq __rte_unused); 92 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index); 93 static void virtio_mac_addr_set(struct rte_eth_dev *dev, 94 struct ether_addr *mac_addr); 95 96 static int virtio_dev_queue_stats_mapping_set( 97 __rte_unused struct rte_eth_dev *eth_dev, 98 __rte_unused uint16_t queue_id, 99 __rte_unused uint8_t stat_idx, 100 __rte_unused uint8_t is_rx); 101 102 /* 103 * The set of PCI devices this driver supports 104 */ 105 static const struct rte_pci_id pci_id_virtio_map[] = { 106 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_DEVICEID_MIN) }, 107 { .vendor_id = 0, /* sentinel */ }, 108 }; 109 110 struct rte_virtio_xstats_name_off { 111 char name[RTE_ETH_XSTATS_NAME_SIZE]; 112 unsigned offset; 113 }; 114 115 /* [rt]x_qX_ is prepended to the name string here */ 116 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = { 117 {"good_packets", offsetof(struct virtnet_rx, stats.packets)}, 118 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)}, 119 {"errors", offsetof(struct virtnet_rx, stats.errors)}, 120 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)}, 121 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)}, 122 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])}, 123 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])}, 124 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])}, 125 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])}, 126 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])}, 127 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])}, 128 {"size_1024_1517_packets", offsetof(struct virtnet_rx, stats.size_bins[6])}, 129 {"size_1518_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])}, 130 }; 131 132 /* [rt]x_qX_ is prepended to the name string here */ 133 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = { 134 {"good_packets", offsetof(struct virtnet_tx, stats.packets)}, 135 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)}, 136 {"errors", offsetof(struct virtnet_tx, stats.errors)}, 137 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)}, 138 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)}, 139 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])}, 140 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])}, 141 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])}, 142 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])}, 143 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])}, 144 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])}, 145 {"size_1024_1517_packets", offsetof(struct virtnet_tx, stats.size_bins[6])}, 146 {"size_1518_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])}, 147 }; 148 149 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \ 150 sizeof(rte_virtio_rxq_stat_strings[0])) 151 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \ 152 sizeof(rte_virtio_txq_stat_strings[0])) 153 154 static int 155 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl, 156 int *dlen, int pkt_num) 157 { 158 uint32_t head, i; 159 int k, sum = 0; 160 virtio_net_ctrl_ack status = ~0; 161 struct virtio_pmd_ctrl result; 162 struct virtqueue *vq; 163 164 ctrl->status = status; 165 166 if (!cvq || !cvq->vq) { 167 PMD_INIT_LOG(ERR, "Control queue is not supported."); 168 return -1; 169 } 170 vq = cvq->vq; 171 head = vq->vq_desc_head_idx; 172 173 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, " 174 "vq->hw->cvq = %p vq = %p", 175 vq->vq_desc_head_idx, status, vq->hw->cvq, vq); 176 177 if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1)) 178 return -1; 179 180 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl, 181 sizeof(struct virtio_pmd_ctrl)); 182 183 /* 184 * Format is enforced in qemu code: 185 * One TX packet for header; 186 * At least one TX packet per argument; 187 * One RX packet for ACK. 188 */ 189 vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT; 190 vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem; 191 vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr); 192 vq->vq_free_cnt--; 193 i = vq->vq_ring.desc[head].next; 194 195 for (k = 0; k < pkt_num; k++) { 196 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT; 197 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem 198 + sizeof(struct virtio_net_ctrl_hdr) 199 + sizeof(ctrl->status) + sizeof(uint8_t)*sum; 200 vq->vq_ring.desc[i].len = dlen[k]; 201 sum += dlen[k]; 202 vq->vq_free_cnt--; 203 i = vq->vq_ring.desc[i].next; 204 } 205 206 vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE; 207 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem 208 + sizeof(struct virtio_net_ctrl_hdr); 209 vq->vq_ring.desc[i].len = sizeof(ctrl->status); 210 vq->vq_free_cnt--; 211 212 vq->vq_desc_head_idx = vq->vq_ring.desc[i].next; 213 214 vq_update_avail_ring(vq, head); 215 vq_update_avail_idx(vq); 216 217 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index); 218 219 virtqueue_notify(vq); 220 221 rte_rmb(); 222 while (VIRTQUEUE_NUSED(vq) == 0) { 223 rte_rmb(); 224 usleep(100); 225 } 226 227 while (VIRTQUEUE_NUSED(vq)) { 228 uint32_t idx, desc_idx, used_idx; 229 struct vring_used_elem *uep; 230 231 used_idx = (uint32_t)(vq->vq_used_cons_idx 232 & (vq->vq_nentries - 1)); 233 uep = &vq->vq_ring.used->ring[used_idx]; 234 idx = (uint32_t) uep->id; 235 desc_idx = idx; 236 237 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) { 238 desc_idx = vq->vq_ring.desc[desc_idx].next; 239 vq->vq_free_cnt++; 240 } 241 242 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx; 243 vq->vq_desc_head_idx = idx; 244 245 vq->vq_used_cons_idx++; 246 vq->vq_free_cnt++; 247 } 248 249 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d", 250 vq->vq_free_cnt, vq->vq_desc_head_idx); 251 252 memcpy(&result, cvq->virtio_net_hdr_mz->addr, 253 sizeof(struct virtio_pmd_ctrl)); 254 255 return result.status; 256 } 257 258 static int 259 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues) 260 { 261 struct virtio_hw *hw = dev->data->dev_private; 262 struct virtio_pmd_ctrl ctrl; 263 int dlen[1]; 264 int ret; 265 266 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ; 267 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET; 268 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t)); 269 270 dlen[0] = sizeof(uint16_t); 271 272 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1); 273 if (ret) { 274 PMD_INIT_LOG(ERR, "Multiqueue configured but send command " 275 "failed, this is too late now..."); 276 return -EINVAL; 277 } 278 279 return 0; 280 } 281 282 void 283 virtio_dev_queue_release(struct virtqueue *vq) 284 { 285 struct virtio_hw *hw; 286 287 if (vq) { 288 hw = vq->hw; 289 if (vq->configured) 290 hw->vtpci_ops->del_queue(hw, vq); 291 292 rte_free(vq->sw_ring); 293 rte_free(vq); 294 } 295 } 296 297 int virtio_dev_queue_setup(struct rte_eth_dev *dev, 298 int queue_type, 299 uint16_t queue_idx, 300 uint16_t vtpci_queue_idx, 301 uint16_t nb_desc, 302 unsigned int socket_id, 303 void **pvq) 304 { 305 char vq_name[VIRTQUEUE_MAX_NAME_SZ]; 306 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ]; 307 const struct rte_memzone *mz = NULL, *hdr_mz = NULL; 308 unsigned int vq_size, size; 309 struct virtio_hw *hw = dev->data->dev_private; 310 struct virtnet_rx *rxvq = NULL; 311 struct virtnet_tx *txvq = NULL; 312 struct virtnet_ctl *cvq = NULL; 313 struct virtqueue *vq; 314 const char *queue_names[] = {"rvq", "txq", "cvq"}; 315 size_t sz_vq, sz_q = 0, sz_hdr_mz = 0; 316 void *sw_ring = NULL; 317 int ret; 318 319 PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx); 320 321 /* 322 * Read the virtqueue size from the Queue Size field 323 * Always power of 2 and if 0 virtqueue does not exist 324 */ 325 vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx); 326 PMD_INIT_LOG(DEBUG, "vq_size: %u nb_desc:%u", vq_size, nb_desc); 327 if (vq_size == 0) { 328 PMD_INIT_LOG(ERR, "virtqueue does not exist"); 329 return -EINVAL; 330 } 331 332 if (!rte_is_power_of_2(vq_size)) { 333 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2"); 334 return -EINVAL; 335 } 336 337 snprintf(vq_name, sizeof(vq_name), "port%d_%s%d", 338 dev->data->port_id, queue_names[queue_type], queue_idx); 339 340 sz_vq = RTE_ALIGN_CEIL(sizeof(*vq) + 341 vq_size * sizeof(struct vq_desc_extra), 342 RTE_CACHE_LINE_SIZE); 343 if (queue_type == VTNET_RQ) { 344 sz_q = sz_vq + sizeof(*rxvq); 345 } else if (queue_type == VTNET_TQ) { 346 sz_q = sz_vq + sizeof(*txvq); 347 /* 348 * For each xmit packet, allocate a virtio_net_hdr 349 * and indirect ring elements 350 */ 351 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region); 352 } else if (queue_type == VTNET_CQ) { 353 sz_q = sz_vq + sizeof(*cvq); 354 /* Allocate a page for control vq command, data and status */ 355 sz_hdr_mz = PAGE_SIZE; 356 } 357 358 vq = rte_zmalloc_socket(vq_name, sz_q, RTE_CACHE_LINE_SIZE, socket_id); 359 if (vq == NULL) { 360 PMD_INIT_LOG(ERR, "can not allocate vq"); 361 return -ENOMEM; 362 } 363 vq->hw = hw; 364 vq->vq_queue_index = vtpci_queue_idx; 365 vq->vq_nentries = vq_size; 366 367 if (nb_desc == 0 || nb_desc > vq_size) 368 nb_desc = vq_size; 369 vq->vq_free_cnt = nb_desc; 370 371 /* 372 * Reserve a memzone for vring elements 373 */ 374 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN); 375 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN); 376 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", 377 size, vq->vq_ring_size); 378 379 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size, socket_id, 380 0, VIRTIO_PCI_VRING_ALIGN); 381 if (mz == NULL) { 382 if (rte_errno == EEXIST) 383 mz = rte_memzone_lookup(vq_name); 384 if (mz == NULL) { 385 ret = -ENOMEM; 386 goto fail_q_alloc; 387 } 388 } 389 390 memset(mz->addr, 0, sizeof(mz->len)); 391 392 vq->vq_ring_mem = mz->phys_addr; 393 vq->vq_ring_virt_mem = mz->addr; 394 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64, 395 (uint64_t)mz->phys_addr); 396 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64, 397 (uint64_t)(uintptr_t)mz->addr); 398 399 if (sz_hdr_mz) { 400 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_%s%d_hdr", 401 dev->data->port_id, queue_names[queue_type], 402 queue_idx); 403 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz, 404 socket_id, 0, 405 RTE_CACHE_LINE_SIZE); 406 if (hdr_mz == NULL) { 407 if (rte_errno == EEXIST) 408 hdr_mz = rte_memzone_lookup(vq_hdr_name); 409 if (hdr_mz == NULL) { 410 ret = -ENOMEM; 411 goto fail_q_alloc; 412 } 413 } 414 } 415 416 if (queue_type == VTNET_RQ) { 417 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) * 418 sizeof(vq->sw_ring[0]); 419 420 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw, 421 RTE_CACHE_LINE_SIZE, socket_id); 422 if (!sw_ring) { 423 PMD_INIT_LOG(ERR, "can not allocate RX soft ring"); 424 ret = -ENOMEM; 425 goto fail_q_alloc; 426 } 427 428 vq->sw_ring = sw_ring; 429 rxvq = (struct virtnet_rx *)RTE_PTR_ADD(vq, sz_vq); 430 rxvq->vq = vq; 431 rxvq->port_id = dev->data->port_id; 432 rxvq->queue_id = queue_idx; 433 rxvq->mz = mz; 434 *pvq = rxvq; 435 } else if (queue_type == VTNET_TQ) { 436 txvq = (struct virtnet_tx *)RTE_PTR_ADD(vq, sz_vq); 437 txvq->vq = vq; 438 txvq->port_id = dev->data->port_id; 439 txvq->queue_id = queue_idx; 440 txvq->mz = mz; 441 txvq->virtio_net_hdr_mz = hdr_mz; 442 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr; 443 444 *pvq = txvq; 445 } else if (queue_type == VTNET_CQ) { 446 cvq = (struct virtnet_ctl *)RTE_PTR_ADD(vq, sz_vq); 447 cvq->vq = vq; 448 cvq->mz = mz; 449 cvq->virtio_net_hdr_mz = hdr_mz; 450 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr; 451 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE); 452 *pvq = cvq; 453 } 454 455 /* For virtio_user case (that is when dev->pci_dev is NULL), we use 456 * virtual address. And we need properly set _offset_, please see 457 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information. 458 */ 459 if (dev->pci_dev) 460 vq->offset = offsetof(struct rte_mbuf, buf_physaddr); 461 else { 462 vq->vq_ring_mem = (uintptr_t)mz->addr; 463 vq->offset = offsetof(struct rte_mbuf, buf_addr); 464 if (queue_type == VTNET_TQ) 465 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr; 466 else if (queue_type == VTNET_CQ) 467 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr; 468 } 469 470 if (queue_type == VTNET_TQ) { 471 struct virtio_tx_region *txr; 472 unsigned int i; 473 474 txr = hdr_mz->addr; 475 memset(txr, 0, vq_size * sizeof(*txr)); 476 for (i = 0; i < vq_size; i++) { 477 struct vring_desc *start_dp = txr[i].tx_indir; 478 479 vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir)); 480 481 /* first indirect descriptor is always the tx header */ 482 start_dp->addr = txvq->virtio_net_hdr_mem 483 + i * sizeof(*txr) 484 + offsetof(struct virtio_tx_region, tx_hdr); 485 486 start_dp->len = hw->vtnet_hdr_size; 487 start_dp->flags = VRING_DESC_F_NEXT; 488 } 489 } 490 491 if (hw->vtpci_ops->setup_queue(hw, vq) < 0) { 492 PMD_INIT_LOG(ERR, "setup_queue failed"); 493 virtio_dev_queue_release(vq); 494 return -EINVAL; 495 } 496 497 vq->configured = 1; 498 return 0; 499 500 fail_q_alloc: 501 rte_free(sw_ring); 502 rte_memzone_free(hdr_mz); 503 rte_memzone_free(mz); 504 rte_free(vq); 505 506 return ret; 507 } 508 509 static int 510 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx, 511 uint32_t socket_id) 512 { 513 struct virtnet_ctl *cvq; 514 int ret; 515 struct virtio_hw *hw = dev->data->dev_private; 516 517 PMD_INIT_FUNC_TRACE(); 518 ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX, 519 vtpci_queue_idx, 0, socket_id, (void **)&cvq); 520 if (ret < 0) { 521 PMD_INIT_LOG(ERR, "control vq initialization failed"); 522 return ret; 523 } 524 525 hw->cvq = cvq; 526 return 0; 527 } 528 529 static void 530 virtio_free_queues(struct rte_eth_dev *dev) 531 { 532 unsigned int i; 533 534 for (i = 0; i < dev->data->nb_rx_queues; i++) 535 virtio_dev_rx_queue_release(dev->data->rx_queues[i]); 536 537 dev->data->nb_rx_queues = 0; 538 539 for (i = 0; i < dev->data->nb_tx_queues; i++) 540 virtio_dev_tx_queue_release(dev->data->tx_queues[i]); 541 542 dev->data->nb_tx_queues = 0; 543 } 544 545 static void 546 virtio_dev_close(struct rte_eth_dev *dev) 547 { 548 struct virtio_hw *hw = dev->data->dev_private; 549 550 PMD_INIT_LOG(DEBUG, "virtio_dev_close"); 551 552 if (hw->started == 1) 553 virtio_dev_stop(dev); 554 555 /* reset the NIC */ 556 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) 557 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR); 558 vtpci_reset(hw); 559 virtio_dev_free_mbufs(dev); 560 virtio_free_queues(dev); 561 } 562 563 static void 564 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev) 565 { 566 struct virtio_hw *hw = dev->data->dev_private; 567 struct virtio_pmd_ctrl ctrl; 568 int dlen[1]; 569 int ret; 570 571 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) { 572 PMD_INIT_LOG(INFO, "host does not support rx control\n"); 573 return; 574 } 575 576 ctrl.hdr.class = VIRTIO_NET_CTRL_RX; 577 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC; 578 ctrl.data[0] = 1; 579 dlen[0] = 1; 580 581 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1); 582 if (ret) 583 PMD_INIT_LOG(ERR, "Failed to enable promisc"); 584 } 585 586 static void 587 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev) 588 { 589 struct virtio_hw *hw = dev->data->dev_private; 590 struct virtio_pmd_ctrl ctrl; 591 int dlen[1]; 592 int ret; 593 594 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) { 595 PMD_INIT_LOG(INFO, "host does not support rx control\n"); 596 return; 597 } 598 599 ctrl.hdr.class = VIRTIO_NET_CTRL_RX; 600 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC; 601 ctrl.data[0] = 0; 602 dlen[0] = 1; 603 604 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1); 605 if (ret) 606 PMD_INIT_LOG(ERR, "Failed to disable promisc"); 607 } 608 609 static void 610 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev) 611 { 612 struct virtio_hw *hw = dev->data->dev_private; 613 struct virtio_pmd_ctrl ctrl; 614 int dlen[1]; 615 int ret; 616 617 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) { 618 PMD_INIT_LOG(INFO, "host does not support rx control\n"); 619 return; 620 } 621 622 ctrl.hdr.class = VIRTIO_NET_CTRL_RX; 623 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI; 624 ctrl.data[0] = 1; 625 dlen[0] = 1; 626 627 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1); 628 if (ret) 629 PMD_INIT_LOG(ERR, "Failed to enable allmulticast"); 630 } 631 632 static void 633 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev) 634 { 635 struct virtio_hw *hw = dev->data->dev_private; 636 struct virtio_pmd_ctrl ctrl; 637 int dlen[1]; 638 int ret; 639 640 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) { 641 PMD_INIT_LOG(INFO, "host does not support rx control\n"); 642 return; 643 } 644 645 ctrl.hdr.class = VIRTIO_NET_CTRL_RX; 646 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI; 647 ctrl.data[0] = 0; 648 dlen[0] = 1; 649 650 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1); 651 if (ret) 652 PMD_INIT_LOG(ERR, "Failed to disable allmulticast"); 653 } 654 655 /* 656 * dev_ops for virtio, bare necessities for basic operation 657 */ 658 static const struct eth_dev_ops virtio_eth_dev_ops = { 659 .dev_configure = virtio_dev_configure, 660 .dev_start = virtio_dev_start, 661 .dev_stop = virtio_dev_stop, 662 .dev_close = virtio_dev_close, 663 .promiscuous_enable = virtio_dev_promiscuous_enable, 664 .promiscuous_disable = virtio_dev_promiscuous_disable, 665 .allmulticast_enable = virtio_dev_allmulticast_enable, 666 .allmulticast_disable = virtio_dev_allmulticast_disable, 667 668 .dev_infos_get = virtio_dev_info_get, 669 .stats_get = virtio_dev_stats_get, 670 .xstats_get = virtio_dev_xstats_get, 671 .xstats_get_names = virtio_dev_xstats_get_names, 672 .stats_reset = virtio_dev_stats_reset, 673 .xstats_reset = virtio_dev_stats_reset, 674 .link_update = virtio_dev_link_update, 675 .rx_queue_setup = virtio_dev_rx_queue_setup, 676 .rx_queue_release = virtio_dev_rx_queue_release, 677 .tx_queue_setup = virtio_dev_tx_queue_setup, 678 .tx_queue_release = virtio_dev_tx_queue_release, 679 /* collect stats per queue */ 680 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set, 681 .vlan_filter_set = virtio_vlan_filter_set, 682 .mac_addr_add = virtio_mac_addr_add, 683 .mac_addr_remove = virtio_mac_addr_remove, 684 .mac_addr_set = virtio_mac_addr_set, 685 }; 686 687 static inline int 688 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev, 689 struct rte_eth_link *link) 690 { 691 struct rte_eth_link *dst = link; 692 struct rte_eth_link *src = &(dev->data->dev_link); 693 694 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst, 695 *(uint64_t *)src) == 0) 696 return -1; 697 698 return 0; 699 } 700 701 /** 702 * Atomically writes the link status information into global 703 * structure rte_eth_dev. 704 * 705 * @param dev 706 * - Pointer to the structure rte_eth_dev to read from. 707 * - Pointer to the buffer to be saved with the link status. 708 * 709 * @return 710 * - On success, zero. 711 * - On failure, negative value. 712 */ 713 static inline int 714 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev, 715 struct rte_eth_link *link) 716 { 717 struct rte_eth_link *dst = &(dev->data->dev_link); 718 struct rte_eth_link *src = link; 719 720 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst, 721 *(uint64_t *)src) == 0) 722 return -1; 723 724 return 0; 725 } 726 727 static void 728 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats) 729 { 730 unsigned i; 731 732 for (i = 0; i < dev->data->nb_tx_queues; i++) { 733 const struct virtnet_tx *txvq = dev->data->tx_queues[i]; 734 if (txvq == NULL) 735 continue; 736 737 stats->opackets += txvq->stats.packets; 738 stats->obytes += txvq->stats.bytes; 739 stats->oerrors += txvq->stats.errors; 740 741 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) { 742 stats->q_opackets[i] = txvq->stats.packets; 743 stats->q_obytes[i] = txvq->stats.bytes; 744 } 745 } 746 747 for (i = 0; i < dev->data->nb_rx_queues; i++) { 748 const struct virtnet_rx *rxvq = dev->data->rx_queues[i]; 749 if (rxvq == NULL) 750 continue; 751 752 stats->ipackets += rxvq->stats.packets; 753 stats->ibytes += rxvq->stats.bytes; 754 stats->ierrors += rxvq->stats.errors; 755 756 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) { 757 stats->q_ipackets[i] = rxvq->stats.packets; 758 stats->q_ibytes[i] = rxvq->stats.bytes; 759 } 760 } 761 762 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed; 763 } 764 765 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev, 766 struct rte_eth_xstat_name *xstats_names, 767 __rte_unused unsigned limit) 768 { 769 unsigned i; 770 unsigned count = 0; 771 unsigned t; 772 773 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS + 774 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS; 775 776 if (xstats_names != NULL) { 777 /* Note: limit checked in rte_eth_xstats_names() */ 778 779 for (i = 0; i < dev->data->nb_rx_queues; i++) { 780 struct virtqueue *rxvq = dev->data->rx_queues[i]; 781 if (rxvq == NULL) 782 continue; 783 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) { 784 snprintf(xstats_names[count].name, 785 sizeof(xstats_names[count].name), 786 "rx_q%u_%s", i, 787 rte_virtio_rxq_stat_strings[t].name); 788 count++; 789 } 790 } 791 792 for (i = 0; i < dev->data->nb_tx_queues; i++) { 793 struct virtqueue *txvq = dev->data->tx_queues[i]; 794 if (txvq == NULL) 795 continue; 796 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) { 797 snprintf(xstats_names[count].name, 798 sizeof(xstats_names[count].name), 799 "tx_q%u_%s", i, 800 rte_virtio_txq_stat_strings[t].name); 801 count++; 802 } 803 } 804 return count; 805 } 806 return nstats; 807 } 808 809 static int 810 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats, 811 unsigned n) 812 { 813 unsigned i; 814 unsigned count = 0; 815 816 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS + 817 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS; 818 819 if (n < nstats) 820 return nstats; 821 822 for (i = 0; i < dev->data->nb_rx_queues; i++) { 823 struct virtnet_rx *rxvq = dev->data->rx_queues[i]; 824 825 if (rxvq == NULL) 826 continue; 827 828 unsigned t; 829 830 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) { 831 xstats[count].value = *(uint64_t *)(((char *)rxvq) + 832 rte_virtio_rxq_stat_strings[t].offset); 833 count++; 834 } 835 } 836 837 for (i = 0; i < dev->data->nb_tx_queues; i++) { 838 struct virtnet_tx *txvq = dev->data->tx_queues[i]; 839 840 if (txvq == NULL) 841 continue; 842 843 unsigned t; 844 845 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) { 846 xstats[count].value = *(uint64_t *)(((char *)txvq) + 847 rte_virtio_txq_stat_strings[t].offset); 848 count++; 849 } 850 } 851 852 return count; 853 } 854 855 static void 856 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) 857 { 858 virtio_update_stats(dev, stats); 859 } 860 861 static void 862 virtio_dev_stats_reset(struct rte_eth_dev *dev) 863 { 864 unsigned int i; 865 866 for (i = 0; i < dev->data->nb_tx_queues; i++) { 867 struct virtnet_tx *txvq = dev->data->tx_queues[i]; 868 if (txvq == NULL) 869 continue; 870 871 txvq->stats.packets = 0; 872 txvq->stats.bytes = 0; 873 txvq->stats.errors = 0; 874 txvq->stats.multicast = 0; 875 txvq->stats.broadcast = 0; 876 memset(txvq->stats.size_bins, 0, 877 sizeof(txvq->stats.size_bins[0]) * 8); 878 } 879 880 for (i = 0; i < dev->data->nb_rx_queues; i++) { 881 struct virtnet_rx *rxvq = dev->data->rx_queues[i]; 882 if (rxvq == NULL) 883 continue; 884 885 rxvq->stats.packets = 0; 886 rxvq->stats.bytes = 0; 887 rxvq->stats.errors = 0; 888 rxvq->stats.multicast = 0; 889 rxvq->stats.broadcast = 0; 890 memset(rxvq->stats.size_bins, 0, 891 sizeof(rxvq->stats.size_bins[0]) * 8); 892 } 893 } 894 895 static void 896 virtio_set_hwaddr(struct virtio_hw *hw) 897 { 898 vtpci_write_dev_config(hw, 899 offsetof(struct virtio_net_config, mac), 900 &hw->mac_addr, ETHER_ADDR_LEN); 901 } 902 903 static void 904 virtio_get_hwaddr(struct virtio_hw *hw) 905 { 906 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) { 907 vtpci_read_dev_config(hw, 908 offsetof(struct virtio_net_config, mac), 909 &hw->mac_addr, ETHER_ADDR_LEN); 910 } else { 911 eth_random_addr(&hw->mac_addr[0]); 912 virtio_set_hwaddr(hw); 913 } 914 } 915 916 static void 917 virtio_mac_table_set(struct virtio_hw *hw, 918 const struct virtio_net_ctrl_mac *uc, 919 const struct virtio_net_ctrl_mac *mc) 920 { 921 struct virtio_pmd_ctrl ctrl; 922 int err, len[2]; 923 924 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) { 925 PMD_DRV_LOG(INFO, "host does not support mac table"); 926 return; 927 } 928 929 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC; 930 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; 931 932 len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries); 933 memcpy(ctrl.data, uc, len[0]); 934 935 len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries); 936 memcpy(ctrl.data + len[0], mc, len[1]); 937 938 err = virtio_send_command(hw->cvq, &ctrl, len, 2); 939 if (err != 0) 940 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err); 941 } 942 943 static void 944 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr, 945 uint32_t index, uint32_t vmdq __rte_unused) 946 { 947 struct virtio_hw *hw = dev->data->dev_private; 948 const struct ether_addr *addrs = dev->data->mac_addrs; 949 unsigned int i; 950 struct virtio_net_ctrl_mac *uc, *mc; 951 952 if (index >= VIRTIO_MAX_MAC_ADDRS) { 953 PMD_DRV_LOG(ERR, "mac address index %u out of range", index); 954 return; 955 } 956 957 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries)); 958 uc->entries = 0; 959 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries)); 960 mc->entries = 0; 961 962 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) { 963 const struct ether_addr *addr 964 = (i == index) ? mac_addr : addrs + i; 965 struct virtio_net_ctrl_mac *tbl 966 = is_multicast_ether_addr(addr) ? mc : uc; 967 968 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN); 969 } 970 971 virtio_mac_table_set(hw, uc, mc); 972 } 973 974 static void 975 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index) 976 { 977 struct virtio_hw *hw = dev->data->dev_private; 978 struct ether_addr *addrs = dev->data->mac_addrs; 979 struct virtio_net_ctrl_mac *uc, *mc; 980 unsigned int i; 981 982 if (index >= VIRTIO_MAX_MAC_ADDRS) { 983 PMD_DRV_LOG(ERR, "mac address index %u out of range", index); 984 return; 985 } 986 987 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries)); 988 uc->entries = 0; 989 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries)); 990 mc->entries = 0; 991 992 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) { 993 struct virtio_net_ctrl_mac *tbl; 994 995 if (i == index || is_zero_ether_addr(addrs + i)) 996 continue; 997 998 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc; 999 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN); 1000 } 1001 1002 virtio_mac_table_set(hw, uc, mc); 1003 } 1004 1005 static void 1006 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr) 1007 { 1008 struct virtio_hw *hw = dev->data->dev_private; 1009 1010 memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN); 1011 1012 /* Use atomic update if available */ 1013 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) { 1014 struct virtio_pmd_ctrl ctrl; 1015 int len = ETHER_ADDR_LEN; 1016 1017 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC; 1018 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET; 1019 1020 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN); 1021 virtio_send_command(hw->cvq, &ctrl, &len, 1); 1022 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) 1023 virtio_set_hwaddr(hw); 1024 } 1025 1026 static int 1027 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on) 1028 { 1029 struct virtio_hw *hw = dev->data->dev_private; 1030 struct virtio_pmd_ctrl ctrl; 1031 int len; 1032 1033 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) 1034 return -ENOTSUP; 1035 1036 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN; 1037 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL; 1038 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id)); 1039 len = sizeof(vlan_id); 1040 1041 return virtio_send_command(hw->cvq, &ctrl, &len, 1); 1042 } 1043 1044 static int 1045 virtio_negotiate_features(struct virtio_hw *hw) 1046 { 1047 uint64_t host_features; 1048 1049 /* Prepare guest_features: feature that driver wants to support */ 1050 hw->guest_features = VIRTIO_PMD_GUEST_FEATURES; 1051 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64, 1052 hw->guest_features); 1053 1054 /* Read device(host) feature bits */ 1055 host_features = hw->vtpci_ops->get_features(hw); 1056 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64, 1057 host_features); 1058 1059 /* 1060 * Negotiate features: Subset of device feature bits are written back 1061 * guest feature bits. 1062 */ 1063 hw->guest_features = vtpci_negotiate_features(hw, host_features); 1064 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64, 1065 hw->guest_features); 1066 1067 if (hw->modern) { 1068 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) { 1069 PMD_INIT_LOG(ERR, 1070 "VIRTIO_F_VERSION_1 features is not enabled."); 1071 return -1; 1072 } 1073 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK); 1074 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) { 1075 PMD_INIT_LOG(ERR, 1076 "failed to set FEATURES_OK status!"); 1077 return -1; 1078 } 1079 } 1080 1081 return 0; 1082 } 1083 1084 /* 1085 * Process Virtio Config changed interrupt and call the callback 1086 * if link state changed. 1087 */ 1088 static void 1089 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle, 1090 void *param) 1091 { 1092 struct rte_eth_dev *dev = param; 1093 struct virtio_hw *hw = dev->data->dev_private; 1094 uint8_t isr; 1095 1096 /* Read interrupt status which clears interrupt */ 1097 isr = vtpci_isr(hw); 1098 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr); 1099 1100 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) 1101 PMD_DRV_LOG(ERR, "interrupt enable failed"); 1102 1103 if (isr & VIRTIO_PCI_ISR_CONFIG) { 1104 if (virtio_dev_link_update(dev, 0) == 0) 1105 _rte_eth_dev_callback_process(dev, 1106 RTE_ETH_EVENT_INTR_LSC); 1107 } 1108 1109 } 1110 1111 static void 1112 rx_func_get(struct rte_eth_dev *eth_dev) 1113 { 1114 struct virtio_hw *hw = eth_dev->data->dev_private; 1115 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) 1116 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts; 1117 else 1118 eth_dev->rx_pkt_burst = &virtio_recv_pkts; 1119 } 1120 1121 /* 1122 * This function is based on probe() function in virtio_pci.c 1123 * It returns 0 on success. 1124 */ 1125 int 1126 eth_virtio_dev_init(struct rte_eth_dev *eth_dev) 1127 { 1128 struct virtio_hw *hw = eth_dev->data->dev_private; 1129 struct virtio_net_config *config; 1130 struct virtio_net_config local_config; 1131 struct rte_pci_device *pci_dev; 1132 uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE; 1133 int ret; 1134 1135 RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf)); 1136 1137 eth_dev->dev_ops = &virtio_eth_dev_ops; 1138 eth_dev->tx_pkt_burst = &virtio_xmit_pkts; 1139 1140 if (rte_eal_process_type() == RTE_PROC_SECONDARY) { 1141 rx_func_get(eth_dev); 1142 return 0; 1143 } 1144 1145 /* Allocate memory for storing MAC addresses */ 1146 eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0); 1147 if (eth_dev->data->mac_addrs == NULL) { 1148 PMD_INIT_LOG(ERR, 1149 "Failed to allocate %d bytes needed to store MAC addresses", 1150 VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN); 1151 return -ENOMEM; 1152 } 1153 1154 pci_dev = eth_dev->pci_dev; 1155 1156 if (pci_dev) { 1157 ret = vtpci_init(pci_dev, hw, &dev_flags); 1158 if (ret) 1159 return ret; 1160 } 1161 1162 /* Reset the device although not necessary at startup */ 1163 vtpci_reset(hw); 1164 1165 /* Tell the host we've noticed this device. */ 1166 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK); 1167 1168 /* Tell the host we've known how to drive the device. */ 1169 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER); 1170 if (virtio_negotiate_features(hw) < 0) 1171 return -1; 1172 1173 /* If host does not support status then disable LSC */ 1174 if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) 1175 dev_flags &= ~RTE_ETH_DEV_INTR_LSC; 1176 1177 rte_eth_copy_pci_info(eth_dev, pci_dev); 1178 eth_dev->data->dev_flags = dev_flags; 1179 1180 rx_func_get(eth_dev); 1181 1182 /* Setting up rx_header size for the device */ 1183 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) || 1184 vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) 1185 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf); 1186 else 1187 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr); 1188 1189 /* Copy the permanent MAC address to: virtio_hw */ 1190 virtio_get_hwaddr(hw); 1191 ether_addr_copy((struct ether_addr *) hw->mac_addr, 1192 ð_dev->data->mac_addrs[0]); 1193 PMD_INIT_LOG(DEBUG, 1194 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X", 1195 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2], 1196 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]); 1197 1198 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) { 1199 config = &local_config; 1200 1201 vtpci_read_dev_config(hw, 1202 offsetof(struct virtio_net_config, mac), 1203 &config->mac, sizeof(config->mac)); 1204 1205 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) { 1206 vtpci_read_dev_config(hw, 1207 offsetof(struct virtio_net_config, status), 1208 &config->status, sizeof(config->status)); 1209 } else { 1210 PMD_INIT_LOG(DEBUG, 1211 "VIRTIO_NET_F_STATUS is not supported"); 1212 config->status = 0; 1213 } 1214 1215 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) { 1216 vtpci_read_dev_config(hw, 1217 offsetof(struct virtio_net_config, max_virtqueue_pairs), 1218 &config->max_virtqueue_pairs, 1219 sizeof(config->max_virtqueue_pairs)); 1220 } else { 1221 PMD_INIT_LOG(DEBUG, 1222 "VIRTIO_NET_F_MQ is not supported"); 1223 config->max_virtqueue_pairs = 1; 1224 } 1225 1226 hw->max_rx_queues = 1227 (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ? 1228 VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs; 1229 hw->max_tx_queues = 1230 (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ? 1231 VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs; 1232 1233 virtio_dev_cq_queue_setup(eth_dev, 1234 config->max_virtqueue_pairs * 2, 1235 SOCKET_ID_ANY); 1236 1237 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d", 1238 config->max_virtqueue_pairs); 1239 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status); 1240 PMD_INIT_LOG(DEBUG, 1241 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X", 1242 config->mac[0], config->mac[1], 1243 config->mac[2], config->mac[3], 1244 config->mac[4], config->mac[5]); 1245 } else { 1246 hw->max_rx_queues = 1; 1247 hw->max_tx_queues = 1; 1248 } 1249 1250 PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d hw->max_tx_queues=%d", 1251 hw->max_rx_queues, hw->max_tx_queues); 1252 if (pci_dev) 1253 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x", 1254 eth_dev->data->port_id, pci_dev->id.vendor_id, 1255 pci_dev->id.device_id); 1256 1257 /* Setup interrupt callback */ 1258 if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) 1259 rte_intr_callback_register(&pci_dev->intr_handle, 1260 virtio_interrupt_handler, eth_dev); 1261 1262 virtio_dev_cq_start(eth_dev); 1263 1264 return 0; 1265 } 1266 1267 static int 1268 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev) 1269 { 1270 struct rte_pci_device *pci_dev; 1271 struct virtio_hw *hw = eth_dev->data->dev_private; 1272 1273 PMD_INIT_FUNC_TRACE(); 1274 1275 if (rte_eal_process_type() == RTE_PROC_SECONDARY) 1276 return -EPERM; 1277 1278 /* Close it anyway since there's no way to know if closed */ 1279 virtio_dev_close(eth_dev); 1280 1281 pci_dev = eth_dev->pci_dev; 1282 1283 eth_dev->dev_ops = NULL; 1284 eth_dev->tx_pkt_burst = NULL; 1285 eth_dev->rx_pkt_burst = NULL; 1286 1287 if (hw->cvq) 1288 virtio_dev_queue_release(hw->cvq->vq); 1289 1290 rte_free(eth_dev->data->mac_addrs); 1291 eth_dev->data->mac_addrs = NULL; 1292 1293 /* reset interrupt callback */ 1294 if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) 1295 rte_intr_callback_unregister(&pci_dev->intr_handle, 1296 virtio_interrupt_handler, 1297 eth_dev); 1298 rte_eal_pci_unmap_device(pci_dev); 1299 1300 PMD_INIT_LOG(DEBUG, "dev_uninit completed"); 1301 1302 return 0; 1303 } 1304 1305 static struct eth_driver rte_virtio_pmd = { 1306 .pci_drv = { 1307 .name = "rte_virtio_pmd", 1308 .id_table = pci_id_virtio_map, 1309 .drv_flags = RTE_PCI_DRV_DETACHABLE, 1310 }, 1311 .eth_dev_init = eth_virtio_dev_init, 1312 .eth_dev_uninit = eth_virtio_dev_uninit, 1313 .dev_private_size = sizeof(struct virtio_hw), 1314 }; 1315 1316 /* 1317 * Driver initialization routine. 1318 * Invoked once at EAL init time. 1319 * Register itself as the [Poll Mode] Driver of PCI virtio devices. 1320 * Returns 0 on success. 1321 */ 1322 static int 1323 rte_virtio_pmd_init(const char *name __rte_unused, 1324 const char *param __rte_unused) 1325 { 1326 if (rte_eal_iopl_init() != 0) { 1327 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD"); 1328 return -1; 1329 } 1330 1331 rte_eth_driver_register(&rte_virtio_pmd); 1332 return 0; 1333 } 1334 1335 /* 1336 * Configure virtio device 1337 * It returns 0 on success. 1338 */ 1339 static int 1340 virtio_dev_configure(struct rte_eth_dev *dev) 1341 { 1342 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode; 1343 struct virtio_hw *hw = dev->data->dev_private; 1344 1345 PMD_INIT_LOG(DEBUG, "configure"); 1346 1347 if (rxmode->hw_ip_checksum) { 1348 PMD_DRV_LOG(ERR, "HW IP checksum not supported"); 1349 return -EINVAL; 1350 } 1351 1352 hw->vlan_strip = rxmode->hw_vlan_strip; 1353 1354 if (rxmode->hw_vlan_filter 1355 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) { 1356 PMD_DRV_LOG(NOTICE, 1357 "vlan filtering not available on this host"); 1358 return -ENOTSUP; 1359 } 1360 1361 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) 1362 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) { 1363 PMD_DRV_LOG(ERR, "failed to set config vector"); 1364 return -EBUSY; 1365 } 1366 1367 return 0; 1368 } 1369 1370 1371 static int 1372 virtio_dev_start(struct rte_eth_dev *dev) 1373 { 1374 uint16_t nb_queues, i; 1375 struct virtio_hw *hw = dev->data->dev_private; 1376 struct virtnet_rx *rxvq; 1377 struct virtnet_tx *txvq __rte_unused; 1378 1379 /* check if lsc interrupt feature is enabled */ 1380 if (dev->data->dev_conf.intr_conf.lsc) { 1381 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) { 1382 PMD_DRV_LOG(ERR, "link status not supported by host"); 1383 return -ENOTSUP; 1384 } 1385 1386 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) { 1387 PMD_DRV_LOG(ERR, "interrupt enable failed"); 1388 return -EIO; 1389 } 1390 } 1391 1392 /* Initialize Link state */ 1393 virtio_dev_link_update(dev, 0); 1394 1395 /* On restart after stop do not touch queues */ 1396 if (hw->started) 1397 return 0; 1398 1399 /* Do final configuration before rx/tx engine starts */ 1400 virtio_dev_rxtx_start(dev); 1401 vtpci_reinit_complete(hw); 1402 1403 hw->started = 1; 1404 1405 /*Notify the backend 1406 *Otherwise the tap backend might already stop its queue due to fullness. 1407 *vhost backend will have no chance to be waked up 1408 */ 1409 nb_queues = dev->data->nb_rx_queues; 1410 if (nb_queues > 1) { 1411 if (virtio_set_multiple_queues(dev, nb_queues) != 0) 1412 return -EINVAL; 1413 } 1414 1415 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues); 1416 1417 for (i = 0; i < nb_queues; i++) { 1418 rxvq = dev->data->rx_queues[i]; 1419 virtqueue_notify(rxvq->vq); 1420 } 1421 1422 PMD_INIT_LOG(DEBUG, "Notified backend at initialization"); 1423 1424 for (i = 0; i < dev->data->nb_rx_queues; i++) { 1425 rxvq = dev->data->rx_queues[i]; 1426 VIRTQUEUE_DUMP(rxvq->vq); 1427 } 1428 1429 for (i = 0; i < dev->data->nb_tx_queues; i++) { 1430 txvq = dev->data->tx_queues[i]; 1431 VIRTQUEUE_DUMP(txvq->vq); 1432 } 1433 1434 return 0; 1435 } 1436 1437 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev) 1438 { 1439 struct rte_mbuf *buf; 1440 int i, mbuf_num = 0; 1441 1442 for (i = 0; i < dev->data->nb_rx_queues; i++) { 1443 struct virtnet_rx *rxvq = dev->data->rx_queues[i]; 1444 1445 PMD_INIT_LOG(DEBUG, 1446 "Before freeing rxq[%d] used and unused buf", i); 1447 VIRTQUEUE_DUMP(rxvq->vq); 1448 1449 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq); 1450 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) { 1451 rte_pktmbuf_free(buf); 1452 mbuf_num++; 1453 } 1454 1455 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num); 1456 PMD_INIT_LOG(DEBUG, 1457 "After freeing rxq[%d] used and unused buf", i); 1458 VIRTQUEUE_DUMP(rxvq->vq); 1459 } 1460 1461 for (i = 0; i < dev->data->nb_tx_queues; i++) { 1462 struct virtnet_tx *txvq = dev->data->tx_queues[i]; 1463 1464 PMD_INIT_LOG(DEBUG, 1465 "Before freeing txq[%d] used and unused bufs", 1466 i); 1467 VIRTQUEUE_DUMP(txvq->vq); 1468 1469 mbuf_num = 0; 1470 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) { 1471 rte_pktmbuf_free(buf); 1472 mbuf_num++; 1473 } 1474 1475 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num); 1476 PMD_INIT_LOG(DEBUG, 1477 "After freeing txq[%d] used and unused buf", i); 1478 VIRTQUEUE_DUMP(txvq->vq); 1479 } 1480 } 1481 1482 /* 1483 * Stop device: disable interrupt and mark link down 1484 */ 1485 static void 1486 virtio_dev_stop(struct rte_eth_dev *dev) 1487 { 1488 struct rte_eth_link link; 1489 struct virtio_hw *hw = dev->data->dev_private; 1490 1491 PMD_INIT_LOG(DEBUG, "stop"); 1492 1493 hw->started = 0; 1494 1495 if (dev->data->dev_conf.intr_conf.lsc) 1496 rte_intr_disable(&dev->pci_dev->intr_handle); 1497 1498 memset(&link, 0, sizeof(link)); 1499 virtio_dev_atomic_write_link_status(dev, &link); 1500 } 1501 1502 static int 1503 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete) 1504 { 1505 struct rte_eth_link link, old; 1506 uint16_t status; 1507 struct virtio_hw *hw = dev->data->dev_private; 1508 memset(&link, 0, sizeof(link)); 1509 virtio_dev_atomic_read_link_status(dev, &link); 1510 old = link; 1511 link.link_duplex = ETH_LINK_FULL_DUPLEX; 1512 link.link_speed = SPEED_10G; 1513 1514 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) { 1515 PMD_INIT_LOG(DEBUG, "Get link status from hw"); 1516 vtpci_read_dev_config(hw, 1517 offsetof(struct virtio_net_config, status), 1518 &status, sizeof(status)); 1519 if ((status & VIRTIO_NET_S_LINK_UP) == 0) { 1520 link.link_status = ETH_LINK_DOWN; 1521 PMD_INIT_LOG(DEBUG, "Port %d is down", 1522 dev->data->port_id); 1523 } else { 1524 link.link_status = ETH_LINK_UP; 1525 PMD_INIT_LOG(DEBUG, "Port %d is up", 1526 dev->data->port_id); 1527 } 1528 } else { 1529 link.link_status = ETH_LINK_UP; 1530 } 1531 virtio_dev_atomic_write_link_status(dev, &link); 1532 1533 return (old.link_status == link.link_status) ? -1 : 0; 1534 } 1535 1536 static void 1537 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info) 1538 { 1539 struct virtio_hw *hw = dev->data->dev_private; 1540 1541 if (dev->pci_dev) 1542 dev_info->driver_name = dev->driver->pci_drv.name; 1543 else 1544 dev_info->driver_name = "virtio_user PMD"; 1545 dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues; 1546 dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues; 1547 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE; 1548 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN; 1549 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS; 1550 dev_info->default_txconf = (struct rte_eth_txconf) { 1551 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS 1552 }; 1553 } 1554 1555 /* 1556 * It enables testpmd to collect per queue stats. 1557 */ 1558 static int 1559 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev, 1560 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx, 1561 __rte_unused uint8_t is_rx) 1562 { 1563 return 0; 1564 } 1565 1566 static struct rte_driver rte_virtio_driver = { 1567 .type = PMD_PDEV, 1568 .init = rte_virtio_pmd_init, 1569 }; 1570 1571 PMD_REGISTER_DRIVER(rte_virtio_driver, net_virtio); 1572 DRIVER_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map); 1573