lib/librte_pmd_virtio/virtio_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 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 #ifdef RTE_EXEC_ENV_LINUXAPP
  40 #include <dirent.h>
  41 #endif
  42
  43 #include <rte_ethdev.h>
  44 #include <rte_memcpy.h>
  45 #include <rte_string_fns.h>
  46 #include <rte_memzone.h>
  47 #include <rte_malloc.h>
  48 #include <rte_atomic.h>
  49 #include <rte_branch_prediction.h>
  50 #include <rte_pci.h>
  51 #include <rte_ether.h>
  52 #include <rte_common.h>
  53
  54 #include <rte_memory.h>
  55 #include <rte_eal.h>
  56 #include <rte_dev.h>
  57
  58 #include "virtio_ethdev.h"
  59 #include "virtio_pci.h"
  60 #include "virtio_logs.h"
  61 #include "virtqueue.h"
  62
  63
  64 static int eth_virtio_dev_init(struct eth_driver *eth_drv,
  65                 struct rte_eth_dev *eth_dev);
  66 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  67 static int  virtio_dev_start(struct rte_eth_dev *dev);
  68 static void virtio_dev_stop(struct rte_eth_dev *dev);
  69 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  70                                 struct rte_eth_dev_info *dev_info);
  71 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  72         __rte_unused int wait_to_complete);
  73
  74 static void virtio_set_hwaddr(struct virtio_hw *hw);
  75 static void virtio_get_hwaddr(struct virtio_hw *hw);
  76
  77 static void virtio_dev_rx_queue_release(__rte_unused void *rxq);
  78 static void virtio_dev_tx_queue_release(__rte_unused void *txq);
  79
  80 static void virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
  81 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  82 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  83
  84 static int virtio_dev_queue_stats_mapping_set(
  85         __rte_unused struct rte_eth_dev *eth_dev,
  86         __rte_unused uint16_t queue_id,
  87         __rte_unused uint8_t stat_idx,
  88         __rte_unused uint8_t is_rx);
  89
  90 /*
  91  * The set of PCI devices this driver supports
  92  */
  93 static struct rte_pci_id pci_id_virtio_map[] = {
  94
  95 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
  96 #include "rte_pci_dev_ids.h"
  97
  98 { .vendor_id = 0, /* sentinel */ },
  99 };
 100
 101 static int
 102 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
 103                 int *dlen, int pkt_num)
 104 {
 105         uint32_t head = vq->vq_desc_head_idx, i;
 106         int k, sum = 0;
 107         virtio_net_ctrl_ack status = ~0;
 108         struct virtio_pmd_ctrl result;
 109
 110         ctrl->status = status;
 111
 112         if (!vq->hw->cvq) {
 113                 PMD_INIT_LOG(ERR,
 114                              "%s(): Control queue is not supported.\n",
 115                              __func__);
 116                 return -1;
 117         }
 118
 119         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 120                 "vq->hw->cvq = %p vq = %p\n",
 121                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 122
 123         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
 124                 return -1;
 125
 126         memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
 127                 sizeof(struct virtio_pmd_ctrl));
 128
 129         /*
 130          * Format is enforced in qemu code:
 131          * One TX packet for header;
 132          * At least one TX packet per argument;
 133          * One RX packet for ACK.
 134          */
 135         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 136         vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
 137         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 138         vq->vq_free_cnt--;
 139         i = vq->vq_ring.desc[head].next;
 140
 141         for (k = 0; k < pkt_num; k++) {
 142                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 143                 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 144                         + sizeof(struct virtio_net_ctrl_hdr)
 145                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 146                 vq->vq_ring.desc[i].len = dlen[k];
 147                 sum += dlen[k];
 148                 vq->vq_free_cnt--;
 149                 i = vq->vq_ring.desc[i].next;
 150         }
 151
 152         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 153         vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 154                         + sizeof(struct virtio_net_ctrl_hdr);
 155         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 156         vq->vq_free_cnt--;
 157
 158         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 159
 160         vq_update_avail_ring(vq, head);
 161         vq_update_avail_idx(vq);
 162
 163         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d\n", vq->vq_queue_index);
 164
 165         virtqueue_notify(vq);
 166
 167         while (vq->vq_used_cons_idx == vq->vq_ring.used->idx)
 168                 usleep(100);
 169
 170         while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
 171                 uint32_t idx, desc_idx, used_idx;
 172                 struct vring_used_elem *uep;
 173
 174                 rmb();
 175
 176                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 177                                 & (vq->vq_nentries - 1));
 178                 uep = &vq->vq_ring.used->ring[used_idx];
 179                 idx = (uint32_t) uep->id;
 180                 desc_idx = idx;
 181
 182                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 183                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 184                         vq->vq_free_cnt++;
 185                 }
 186
 187                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 188                 vq->vq_desc_head_idx = idx;
 189
 190                 vq->vq_used_cons_idx++;
 191                 vq->vq_free_cnt++;
 192         }
 193
 194         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d\n",
 195                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 196
 197         memcpy(&result, vq->virtio_net_hdr_mz->addr,
 198                         sizeof(struct virtio_pmd_ctrl));
 199
 200         return result.status;
 201 }
 202
 203 static int
 204 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 205 {
 206         struct virtio_hw *hw
 207                 = VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 208         struct virtio_pmd_ctrl ctrl;
 209         int dlen[1];
 210         int ret;
 211
 212         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 213         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 214         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 215
 216         PMD_INIT_LOG(DEBUG, "ctrl.data=%d\n", *(int *)ctrl.data);
 217
 218         dlen[0] = sizeof(uint16_t);
 219
 220         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 221
 222         if (ret) {
 223                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 224                           "failed, this is too late now...\n");
 225                 return -EINVAL;
 226         }
 227
 228         return 0;
 229 }
 230
 231 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 232                         int queue_type,
 233                         uint16_t queue_idx,
 234                         uint8_t  vtpci_queue_idx,
 235                         uint16_t nb_desc,
 236                         unsigned int socket_id,
 237                         struct virtqueue **pvq)
 238 {
 239         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 240         const struct rte_memzone *mz;
 241         uint16_t vq_size;
 242         int size;
 243         struct virtio_hw *hw =
 244                 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 245         struct virtqueue  *vq = NULL;
 246
 247         /* Write the virtqueue index to the Queue Select Field */
 248         VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vtpci_queue_idx);
 249         PMD_INIT_LOG(DEBUG, "selecting queue: %d\n", vtpci_queue_idx);
 250
 251         /*
 252          * Read the virtqueue size from the Queue Size field
 253          * Always power of 2 and if 0 virtqueue does not exist
 254          */
 255         vq_size = VIRTIO_READ_REG_2(hw, VIRTIO_PCI_QUEUE_NUM);
 256         PMD_INIT_LOG(DEBUG, "vq_size: %d nb_desc:%d\n", vq_size, nb_desc);
 257         if (nb_desc == 0)
 258                 nb_desc = vq_size;
 259         if (vq_size == 0) {
 260                 PMD_INIT_LOG(ERR, "%s: virtqueue does not exist\n", __func__);
 261                 return -EINVAL;
 262         } else if (!rte_is_power_of_2(vq_size)) {
 263                 PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2\n", __func__);
 264                 return -EINVAL;
 265         } else if (nb_desc != vq_size) {
 266                 PMD_INIT_LOG(ERR, "Warning: nb_desc(%d) is not equal to vq size (%d), fall to vq size\n",
 267                         nb_desc, vq_size);
 268                 nb_desc = vq_size;
 269         }
 270
 271         if (queue_type == VTNET_RQ) {
 272                 rte_snprintf(vq_name, sizeof(vq_name), "port%d_rvq%d",
 273                         dev->data->port_id, queue_idx);
 274                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 275                         vq_size * sizeof(struct vq_desc_extra), CACHE_LINE_SIZE);
 276                 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
 277         } else if (queue_type == VTNET_TQ) {
 278                 rte_snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d",
 279                         dev->data->port_id, queue_idx);
 280                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 281                         vq_size * sizeof(struct vq_desc_extra), CACHE_LINE_SIZE);
 282                 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
 283         } else if (queue_type == VTNET_CQ) {
 284                 rte_snprintf(vq_name, sizeof(vq_name), "port%d_cvq",
 285                         dev->data->port_id);
 286                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 287                         vq_size * sizeof(struct vq_desc_extra),
 288                         CACHE_LINE_SIZE);
 289                 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
 290         }
 291         if (vq == NULL) {
 292                 PMD_INIT_LOG(ERR, "%s: Can not allocate virtqueue\n", __func__);
 293                 return (-ENOMEM);
 294         }
 295
 296         vq->hw = hw;
 297         vq->port_id = dev->data->port_id;
 298         vq->queue_id = queue_idx;
 299         vq->vq_queue_index = vtpci_queue_idx;
 300         vq->vq_alignment = VIRTIO_PCI_VRING_ALIGN;
 301         vq->vq_nentries = vq_size;
 302         vq->vq_free_cnt = vq_size;
 303
 304         /*
 305          * Reserve a memzone for vring elements
 306          */
 307         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 308         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 309         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d\n", size, vq->vq_ring_size);
 310
 311         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 312                 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
 313         if (mz == NULL) {
 314                 rte_free(vq);
 315                 return -ENOMEM;
 316         }
 317
 318         /*
 319          * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
 320          * and only accepts 32 bit page frame number.
 321          * Check if the allocated physical memory exceeds 16TB.
 322          */
 323         if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
 324                 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!\n");
 325                 rte_free(vq);
 326                 return -ENOMEM;
 327         }
 328
 329         memset(mz->addr, 0, sizeof(mz->len));
 330         vq->mz = mz;
 331         vq->vq_ring_mem = mz->phys_addr;
 332         vq->vq_ring_virt_mem = mz->addr;
 333         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%"PRIx64"\n", (uint64_t)mz->phys_addr);
 334         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64"\n", (uint64_t)mz->addr);
 335         vq->virtio_net_hdr_mz  = NULL;
 336         vq->virtio_net_hdr_mem = (void *)NULL;
 337
 338         if (queue_type == VTNET_TQ) {
 339                 /*
 340                  * For each xmit packet, allocate a virtio_net_hdr
 341                  */
 342                 rte_snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
 343                         dev->data->port_id, queue_idx);
 344                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 345                         vq_size * sizeof(struct virtio_net_hdr),
 346                         socket_id, 0, CACHE_LINE_SIZE);
 347                 if (vq->virtio_net_hdr_mz == NULL) {
 348                         rte_free(vq);
 349                         return -ENOMEM;
 350                 }
 351                 vq->virtio_net_hdr_mem =
 352                         (void *)(uintptr_t)vq->virtio_net_hdr_mz->phys_addr;
 353                 memset(vq->virtio_net_hdr_mz->addr, 0,
 354                         vq_size * sizeof(struct virtio_net_hdr));
 355         } else if (queue_type == VTNET_CQ) {
 356                 /* Allocate a page for control vq command, data and status */
 357                 rte_snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
 358                         dev->data->port_id);
 359                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 360                         PAGE_SIZE, socket_id, 0, CACHE_LINE_SIZE);
 361                 if (vq->virtio_net_hdr_mz == NULL) {
 362                         rte_free(vq);
 363                         return -ENOMEM;
 364                 }
 365                 vq->virtio_net_hdr_mem =
 366                         (void *)(uintptr_t)vq->virtio_net_hdr_mz->phys_addr;
 367                 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 368         }
 369
 370         /*
 371          * Set guest physical address of the virtqueue
 372          * in VIRTIO_PCI_QUEUE_PFN config register of device
 373          */
 374         VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
 375                         mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
 376         *pvq = vq;
 377         return 0;
 378 }
 379
 380 static int
 381 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 382                 uint32_t socket_id)
 383 {
 384         struct virtqueue *vq;
 385         uint16_t nb_desc = 0;
 386         int ret;
 387         struct virtio_hw *hw =
 388                 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 389
 390         PMD_INIT_FUNC_TRACE();
 391         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 392                         vtpci_queue_idx, nb_desc, socket_id, &vq);
 393
 394         if (ret < 0) {
 395                 PMD_INIT_LOG(ERR, "control vq initialization failed\n");
 396                 return ret;
 397         }
 398
 399         hw->cvq = vq;
 400         return 0;
 401 }
 402
 403 static void
 404 virtio_dev_close(struct rte_eth_dev *dev)
 405 {
 406         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 407
 408         virtio_dev_stop(dev);
 409 }
 410
 411 /*
 412  * dev_ops for virtio, bare necessities for basic operation
 413  */
 414 static struct eth_dev_ops virtio_eth_dev_ops = {
 415         .dev_configure           = virtio_dev_configure,
 416         .dev_start               = virtio_dev_start,
 417         .dev_stop                = virtio_dev_stop,
 418         .dev_close               = virtio_dev_close,
 419
 420         .dev_infos_get           = virtio_dev_info_get,
 421         .stats_get               = virtio_dev_stats_get,
 422         .stats_reset             = virtio_dev_stats_reset,
 423         .link_update             = virtio_dev_link_update,
 424         .mac_addr_add            = NULL,
 425         .mac_addr_remove         = NULL,
 426         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 427         /* meaningfull only to multiple queue */
 428         .rx_queue_release        = virtio_dev_rx_queue_release,
 429         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 430         /* meaningfull only to multiple queue */
 431         .tx_queue_release        = virtio_dev_tx_queue_release,
 432         /* collect stats per queue */
 433         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 434 };
 435
 436 static inline int
 437 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 438                                 struct rte_eth_link *link)
 439 {
 440         struct rte_eth_link *dst = link;
 441         struct rte_eth_link *src = &(dev->data->dev_link);
 442
 443         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 444                         *(uint64_t *)src) == 0)
 445                 return -1;
 446
 447         return 0;
 448 }
 449
 450 /**
 451  * Atomically writes the link status information into global
 452  * structure rte_eth_dev.
 453  *
 454  * @param dev
 455  *   - Pointer to the structure rte_eth_dev to read from.
 456  *   - Pointer to the buffer to be saved with the link status.
 457  *
 458  * @return
 459  *   - On success, zero.
 460  *   - On failure, negative value.
 461  */
 462 static inline int
 463 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 464                 struct rte_eth_link *link)
 465 {
 466         struct rte_eth_link *dst = &(dev->data->dev_link);
 467         struct rte_eth_link *src = link;
 468
 469         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 470                                         *(uint64_t *)src) == 0)
 471                 return -1;
 472
 473         return 0;
 474 }
 475
 476 static void
 477 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 478 {
 479         struct virtio_hw *hw =
 480                 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 481         if (stats)
 482                 memcpy(stats, &hw->eth_stats, sizeof(*stats));
 483 }
 484
 485 static void
 486 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 487 {
 488         struct virtio_hw *hw =
 489                 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 490         /* Reset software totals */
 491         memset(&hw->eth_stats, 0, sizeof(hw->eth_stats));
 492 }
 493
 494 static void
 495 virtio_set_hwaddr(struct virtio_hw *hw)
 496 {
 497         vtpci_write_dev_config(hw,
 498                         offsetof(struct virtio_net_config, mac),
 499                         &hw->mac_addr, ETHER_ADDR_LEN);
 500 }
 501
 502 static void
 503 virtio_get_hwaddr(struct virtio_hw *hw)
 504 {
 505         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 506                 vtpci_read_dev_config(hw,
 507                         offsetof(struct virtio_net_config, mac),
 508                         &hw->mac_addr, ETHER_ADDR_LEN);
 509         } else {
 510                 eth_random_addr(&hw->mac_addr[0]);
 511                 virtio_set_hwaddr(hw);
 512         }
 513 }
 514
 515
 516 static void
 517 virtio_negotiate_features(struct virtio_hw *hw)
 518 {
 519         uint32_t guest_features, mask;
 520
 521         mask = VIRTIO_NET_F_CTRL_RX | VIRTIO_NET_F_CTRL_VLAN;
 522         mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
 523
 524         /* TSO and LRO are only available when their corresponding
 525          * checksum offload feature is also negotiated.
 526          */
 527         mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN;
 528         mask |= VIRTIO_NET_F_GUEST_TSO4 | VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN;
 529         mask |= VTNET_LRO_FEATURES;
 530
 531         /* rx_mbuf should not be in multiple merged segments */
 532         mask |= VIRTIO_NET_F_MRG_RXBUF;
 533
 534         /* not negotiating INDIRECT descriptor table support */
 535         mask |= VIRTIO_RING_F_INDIRECT_DESC;
 536
 537         /* Prepare guest_features: feature that driver wants to support */
 538         guest_features = VTNET_FEATURES & ~mask;
 539         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x\n",
 540                 guest_features);
 541
 542         /* Read device(host) feature bits */
 543         hw->host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
 544         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x\n",
 545                 hw->host_features);
 546
 547         /*
 548          * Negotiate features: Subset of device feature bits are written back
 549          * guest feature bits.
 550          */
 551         hw->guest_features = vtpci_negotiate_features(hw, guest_features);
 552         PMD_INIT_LOG(DEBUG, "features after negotiate = %x\n",
 553                 hw->guest_features);
 554 }
 555
 556 #ifdef RTE_EXEC_ENV_LINUXAPP
 557 static int
 558 parse_sysfs_value(const char *filename, unsigned long *val)
 559 {
 560         FILE *f;
 561         char buf[BUFSIZ];
 562         char *end = NULL;
 563
 564         f = fopen(filename, "r");
 565         if (f == NULL) {
 566                 PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s\n",
 567                              __func__, filename);
 568                 return -1;
 569         }
 570
 571         if (fgets(buf, sizeof(buf), f) == NULL) {
 572                 PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s\n",
 573                              __func__, filename);
 574                 fclose(f);
 575                 return -1;
 576         }
 577         *val = strtoul(buf, &end, 0);
 578         if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
 579                 PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s\n",
 580                              __func__, filename);
 581                 fclose(f);
 582                 return -1;
 583         }
 584         fclose(f);
 585         return 0;
 586 }
 587
 588 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen)
 589 {
 590         unsigned int uio_num;
 591         struct dirent *e;
 592         DIR *dir;
 593         char dirname[PATH_MAX];
 594
 595         /* depending on kernel version, uio can be located in uio/uioX
 596          * or uio:uioX */
 597         rte_snprintf(dirname, sizeof(dirname),
 598                      SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
 599                      loc->domain, loc->bus, loc->devid, loc->function);
 600         dir = opendir(dirname);
 601         if (dir == NULL) {
 602                 /* retry with the parent directory */
 603                 rte_snprintf(dirname, sizeof(dirname),
 604                              SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
 605                              loc->domain, loc->bus, loc->devid, loc->function);
 606                 dir = opendir(dirname);
 607
 608                 if (dir == NULL) {
 609                         PMD_INIT_LOG(ERR, "Cannot opendir %s\n", dirname);
 610                         return -1;
 611                 }
 612         }
 613
 614         /* take the first file starting with "uio" */
 615         while ((e = readdir(dir)) != NULL) {
 616                 /* format could be uio%d ...*/
 617                 int shortprefix_len = sizeof("uio") - 1;
 618                 /* ... or uio:uio%d */
 619                 int longprefix_len = sizeof("uio:uio") - 1;
 620                 char *endptr;
 621
 622                 if (strncmp(e->d_name, "uio", 3) != 0)
 623                         continue;
 624
 625                 /* first try uio%d */
 626                 errno = 0;
 627                 uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
 628                 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
 629                         rte_snprintf(buf, buflen, "%s/uio%u", dirname, uio_num);
 630                         break;
 631                 }
 632
 633                 /* then try uio:uio%d */
 634                 errno = 0;
 635                 uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
 636                 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
 637                         rte_snprintf(buf, buflen, "%s/uio:uio%u", dirname,
 638                                      uio_num);
 639                         break;
 640                 }
 641         }
 642         closedir(dir);
 643
 644         /* No uio resource found */
 645         if (e == NULL) {
 646                 PMD_INIT_LOG(ERR, "Could not find uio resource\n");
 647                 return -1;
 648         }
 649
 650         return 0;
 651 }
 652 #endif
 653
 654 /*
 655  * This function is based on probe() function in virtio_pci.c
 656  * It returns 0 on success.
 657  */
 658 static int
 659 eth_virtio_dev_init(__rte_unused struct eth_driver *eth_drv,
 660                 struct rte_eth_dev *eth_dev)
 661 {
 662         struct virtio_net_config *config;
 663         struct virtio_net_config local_config;
 664         uint32_t offset_conf = sizeof(config->mac);
 665         struct rte_pci_device *pci_dev;
 666         struct virtio_hw *hw =
 667                 VIRTIO_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 668
 669         if (RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr)) {
 670                 PMD_INIT_LOG(ERR,
 671                         "MBUF HEADROOM should be enough to hold virtio net hdr\n");
 672                 return -1;
 673         }
 674
 675         if (!(rte_eal_get_configuration()->flags & EAL_FLG_HIGH_IOPL)) {
 676                 PMD_INIT_LOG(ERR,
 677                         "IOPL call failed in EAL init - cannot use virtio PMD driver\n");
 678                 return -1;
 679         }
 680
 681         eth_dev->dev_ops = &virtio_eth_dev_ops;
 682         eth_dev->rx_pkt_burst = &virtio_recv_pkts;
 683         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
 684
 685         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
 686                 return 0;
 687
 688         pci_dev = eth_dev->pci_dev;
 689
 690         hw->device_id = pci_dev->id.device_id;
 691         hw->vendor_id = pci_dev->id.vendor_id;
 692 #ifdef RTE_EXEC_ENV_LINUXAPP
 693         {
 694                 char dirname[PATH_MAX];
 695                 char filename[PATH_MAX];
 696                 unsigned long start, size;
 697
 698                 if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname)) < 0)
 699                         return -1;
 700
 701                 /* get portio size */
 702                 rte_snprintf(filename, sizeof(filename),
 703                              "%s/portio/port0/size", dirname);
 704                 if (parse_sysfs_value(filename, &size) < 0) {
 705                         PMD_INIT_LOG(ERR, "%s(): cannot parse size\n",
 706                                      __func__);
 707                         return -1;
 708                 }
 709
 710                 /* get portio start */
 711                 rte_snprintf(filename, sizeof(filename),
 712                              "%s/portio/port0/start", dirname);
 713                 if (parse_sysfs_value(filename, &start) < 0) {
 714                         PMD_INIT_LOG(ERR, "%s(): cannot parse portio start\n",
 715                                      __func__);
 716                         return -1;
 717                 }
 718                 pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
 719                 pci_dev->mem_resource[0].len =  (uint64_t)size;
 720                 PMD_INIT_LOG(DEBUG,
 721                              "PCI Port IO found start=0x%lx with size=0x%lx\n",
 722                              start, size);
 723         }
 724 #endif
 725         hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
 726
 727         /* Reset the device although not necessary at startup */
 728         vtpci_reset(hw);
 729
 730         /* Tell the host we've noticed this device. */
 731         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
 732
 733         /* Tell the host we've known how to drive the device. */
 734         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
 735         virtio_negotiate_features(hw);
 736
 737         /* Setting up rx_header size for the device */
 738         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
 739                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 740         else
 741                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
 742
 743         /* Allocate memory for storing MAC addresses */
 744         eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
 745         if (eth_dev->data->mac_addrs == NULL) {
 746                 PMD_INIT_LOG(ERR,
 747                         "Failed to allocate %d bytes needed to store MAC addresses",
 748                         ETHER_ADDR_LEN);
 749                 return -ENOMEM;
 750         }
 751
 752         /* Copy the permanent MAC address to: virtio_hw */
 753         virtio_get_hwaddr(hw);
 754         ether_addr_copy((struct ether_addr *) hw->mac_addr,
 755                         &eth_dev->data->mac_addrs[0]);
 756         PMD_INIT_LOG(DEBUG,
 757                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X\n",
 758                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
 759                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
 760
 761         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
 762                 config = &local_config;
 763
 764                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
 765                         offset_conf += sizeof(config->status);
 766                 } else {
 767                         PMD_INIT_LOG(DEBUG,
 768                                      "VIRTIO_NET_F_STATUS is not supported\n");
 769                         config->status = 0;
 770                 }
 771
 772                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
 773                         offset_conf += sizeof(config->max_virtqueue_pairs);
 774                 } else {
 775                         PMD_INIT_LOG(DEBUG,
 776                                      "VIRTIO_NET_F_MQ is not supported\n");
 777                         config->max_virtqueue_pairs = 1;
 778                 }
 779
 780                 vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
 781
 782                 hw->max_rx_queues =
 783                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
 784                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
 785                 hw->max_tx_queues =
 786                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
 787                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
 788
 789                 virtio_dev_cq_queue_setup(eth_dev,
 790                                         config->max_virtqueue_pairs * 2,
 791                                         SOCKET_ID_ANY);
 792
 793                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d\n",
 794                                 config->max_virtqueue_pairs);
 795                 PMD_INIT_LOG(DEBUG, "config->status=%d\n", config->status);
 796                 PMD_INIT_LOG(DEBUG,
 797                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X\n",
 798                                 config->mac[0], config->mac[1],
 799                                 config->mac[2], config->mac[3],
 800                                 config->mac[4], config->mac[5]);
 801         } else {
 802                 hw->max_rx_queues = 1;
 803                 hw->max_tx_queues = 1;
 804         }
 805
 806         eth_dev->data->nb_rx_queues = hw->max_rx_queues;
 807         eth_dev->data->nb_tx_queues = hw->max_tx_queues;
 808
 809         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d\n",
 810                         hw->max_rx_queues, hw->max_tx_queues);
 811         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
 812                         eth_dev->data->port_id, pci_dev->id.vendor_id,
 813                         pci_dev->id.device_id);
 814         return 0;
 815 }
 816
 817 static struct eth_driver rte_virtio_pmd = {
 818         {
 819                 .name = "rte_virtio_pmd",
 820                 .id_table = pci_id_virtio_map,
 821                 .drv_flags = RTE_PCI_DRV_NEED_IGB_UIO,
 822         },
 823         .eth_dev_init = eth_virtio_dev_init,
 824         .dev_private_size = sizeof(struct virtio_adapter),
 825 };
 826
 827 /*
 828  * Driver initialization routine.
 829  * Invoked once at EAL init time.
 830  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
 831  * Returns 0 on success.
 832  */
 833 static int
 834 rte_virtio_pmd_init(const char *name __rte_unused, const char *param __rte_unused)
 835 {
 836         rte_eth_driver_register(&rte_virtio_pmd);
 837         return 0;
 838 }
 839
 840 /*
 841  * Only 1 queue is supported, no queue release related operation
 842  */
 843 static void
 844 virtio_dev_rx_queue_release(__rte_unused void *rxq)
 845 {
 846 }
 847
 848 static void
 849 virtio_dev_tx_queue_release(__rte_unused void *txq)
 850 {
 851 }
 852
 853 /*
 854  * Configure virtio device
 855  * It returns 0 on success.
 856  */
 857 static int
 858 virtio_dev_configure(__rte_unused struct rte_eth_dev *dev)
 859 {
 860         return 0;
 861 }
 862
 863
 864 static int
 865 virtio_dev_start(struct rte_eth_dev *dev)
 866 {
 867         uint16_t nb_queues, i;
 868         uint16_t status;
 869         struct virtio_hw *hw =
 870                 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 871
 872         /* Tell the host we've noticed this device. */
 873         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
 874
 875         /* Tell the host we've known how to drive the device. */
 876         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
 877
 878         hw->adapter_stopped = 0;
 879
 880         virtio_dev_cq_start(dev);
 881
 882         /* Do final configuration before rx/tx engine starts */
 883         virtio_dev_rxtx_start(dev);
 884
 885         /* Check VIRTIO_NET_F_STATUS for link status*/
 886         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
 887                 vtpci_read_dev_config(hw,
 888                                 offsetof(struct virtio_net_config, status),
 889                                 &status, sizeof(status));
 890                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
 891                         PMD_INIT_LOG(ERR,     "Port: %d Link is DOWN\n", dev->data->port_id);
 892                         return -EIO;
 893                 } else {
 894                         PMD_INIT_LOG(DEBUG, "Port: %d Link is UP\n",  dev->data->port_id);
 895                 }
 896         }
 897         vtpci_reinit_complete(hw);
 898
 899         /*Notify the backend
 900          *Otherwise the tap backend might already stop its queue due to fullness.
 901          *vhost backend will have no chance to be waked up
 902          */
 903         nb_queues = dev->data->nb_rx_queues;
 904         if (nb_queues > 1) {
 905                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
 906                         return -EINVAL;
 907         }
 908
 909         PMD_INIT_LOG(DEBUG, "nb_queues=%d\n", nb_queues);
 910
 911         for (i = 0; i < nb_queues; i++)
 912                 virtqueue_notify(dev->data->rx_queues[i]);
 913
 914         PMD_INIT_LOG(DEBUG, "Notified backend at initialization\n");
 915
 916         for (i = 0; i < dev->data->nb_rx_queues; i++)
 917                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
 918
 919         for (i = 0; i < dev->data->nb_tx_queues; i++)
 920                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
 921
 922         return 0;
 923 }
 924
 925 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
 926 {
 927         struct rte_mbuf *buf;
 928         int i, mbuf_num = 0;
 929
 930         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 931                 PMD_INIT_LOG(DEBUG,
 932                              "Before freeing rxq[%d] used and unused buf\n", i);
 933                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
 934
 935                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
 936                                         dev->data->rx_queues[i])) != NULL) {
 937                         rte_pktmbuf_free_seg(buf);
 938                         mbuf_num++;
 939                 }
 940
 941                 PMD_INIT_LOG(DEBUG, "free %d mbufs\n", mbuf_num);
 942                 PMD_INIT_LOG(DEBUG,
 943                              "After freeing rxq[%d] used and unused buf\n", i);
 944                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
 945         }
 946
 947         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 948                 PMD_INIT_LOG(DEBUG,
 949                              "Before freeing txq[%d] used and unused bufs\n",
 950                              i);
 951                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
 952
 953                 mbuf_num = 0;
 954                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
 955                                         dev->data->tx_queues[i])) != NULL) {
 956                         rte_pktmbuf_free_seg(buf);
 957                         mbuf_num++;
 958                 }
 959
 960                 PMD_INIT_LOG(DEBUG, "free %d mbufs\n", mbuf_num);
 961                 PMD_INIT_LOG(DEBUG, "After freeing txq[%d] used and "
 962                         "unused buf\n", i);
 963                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
 964         }
 965 }
 966
 967 /*
 968  * Stop device: disable rx and tx functions to allow for reconfiguring.
 969  */
 970 static void
 971 virtio_dev_stop(struct rte_eth_dev *dev)
 972 {
 973         struct virtio_hw *hw =
 974                 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 975
 976         /* reset the NIC */
 977         vtpci_reset(hw);
 978         virtio_dev_free_mbufs(dev);
 979 }
 980
 981 static int
 982 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
 983 {
 984         struct rte_eth_link link, old;
 985         uint16_t status;
 986         struct virtio_hw *hw =
 987                 VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 988         memset(&link, 0, sizeof(link));
 989         virtio_dev_atomic_read_link_status(dev, &link);
 990         old = link;
 991         link.link_duplex = FULL_DUPLEX;
 992         link.link_speed  = SPEED_10G;
 993         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
 994                 PMD_INIT_LOG(DEBUG, "Get link status from hw\n");
 995                 vtpci_read_dev_config(hw,
 996                                 offsetof(struct virtio_net_config, status),
 997                                 &status, sizeof(status));
 998                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
 999                         link.link_status = 0;
1000                         PMD_INIT_LOG(DEBUG, "Port %d is down\n",
1001                                      dev->data->port_id);
1002                 } else {
1003                         link.link_status = 1;
1004                         PMD_INIT_LOG(DEBUG, "Port %d is up\n",
1005                                      dev->data->port_id);
1006                 }
1007         } else {
1008                 link.link_status = 1;   /* Link up */
1009         }
1010         virtio_dev_atomic_write_link_status(dev, &link);
1011         if (old.link_status == link.link_status)
1012                 return -1;
1013         /*changed*/
1014         return 0;
1015 }
1016
1017 static void
1018 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1019 {
1020         struct virtio_hw *hw = VIRTIO_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1021
1022         dev_info->driver_name = dev->driver->pci_drv.name;
1023         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1024         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1025         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1026         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1027         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1028 }
1029
1030 /*
1031  * It enables testpmd to collect per queue stats.
1032  */
1033 static int
1034 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1035 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1036 __rte_unused uint8_t is_rx)
1037 {
1038         return 0;
1039 }
1040
1041 static struct rte_driver rte_virtio_driver = {
1042         .type = PMD_PDEV,
1043         .init = rte_virtio_pmd_init,
1044 };
1045
1046 PMD_REGISTER_DRIVER(rte_virtio_driver);