lib/librte_pmd_enic/enic_main.c

   1 /*
   2  * Copyright 2008-2014 Cisco Systems, Inc.  All rights reserved.
   3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
   4  *
   5  * Copyright (c) 2014, Cisco Systems, Inc.
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  *
  12  * 1. Redistributions of source code must retain the above copyright
  13  * notice, this list of conditions and the following disclaimer.
  14  *
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  * notice, this list of conditions and the following disclaimer in
  17  * the documentation and/or other materials provided with the
  18  * distribution.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
  30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  31  * POSSIBILITY OF SUCH DAMAGE.
  32  *
  33  */
  34 #ident "$Id$"
  35
  36 #include <stdio.h>
  37
  38 #include <sys/stat.h>
  39 #include <sys/mman.h>
  40 #include <fcntl.h>
  41 #include <libgen.h>
  42
  43 #include <rte_pci.h>
  44 #include <rte_memzone.h>
  45 #include <rte_malloc.h>
  46 #include <rte_mbuf.h>
  47 #include <rte_string_fns.h>
  48 #include <rte_ethdev.h>
  49
  50 #include "enic_compat.h"
  51 #include "enic.h"
  52 #include "wq_enet_desc.h"
  53 #include "rq_enet_desc.h"
  54 #include "cq_enet_desc.h"
  55 #include "vnic_enet.h"
  56 #include "vnic_dev.h"
  57 #include "vnic_wq.h"
  58 #include "vnic_rq.h"
  59 #include "vnic_cq.h"
  60 #include "vnic_intr.h"
  61 #include "vnic_nic.h"
  62
  63 static inline int enic_is_sriov_vf(struct enic *enic)
  64 {
  65         return enic->pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
  66 }
  67
  68 static int is_zero_addr(uint8_t *addr)
  69 {
  70         return !(addr[0] |  addr[1] | addr[2] | addr[3] | addr[4] | addr[5]);
  71 }
  72
  73 static int is_mcast_addr(uint8_t *addr)
  74 {
  75         return addr[0] & 1;
  76 }
  77
  78 static int is_eth_addr_valid(uint8_t *addr)
  79 {
  80         return !is_mcast_addr(addr) && !is_zero_addr(addr);
  81 }
  82
  83 static inline struct rte_mbuf *
  84 enic_rxmbuf_alloc(struct rte_mempool *mp)
  85 {
  86         struct rte_mbuf *m;
  87
  88         m = __rte_mbuf_raw_alloc(mp);
  89         __rte_mbuf_sanity_check_raw(m, 0);
  90         return m;
  91 }
  92
  93 void enic_set_hdr_split_size(struct enic *enic, u16 split_hdr_size)
  94 {
  95         vnic_set_hdr_split_size(enic->vdev, split_hdr_size);
  96 }
  97
  98 static void enic_free_wq_buf(__rte_unused struct vnic_wq *wq, struct vnic_wq_buf *buf)
  99 {
 100         struct rte_mbuf *mbuf = (struct rte_mbuf *)buf->os_buf;
 101
 102         rte_mempool_put(mbuf->pool, mbuf);
 103         buf->os_buf = NULL;
 104 }
 105
 106 static void enic_wq_free_buf(struct vnic_wq *wq,
 107         __rte_unused struct cq_desc *cq_desc,
 108         struct vnic_wq_buf *buf,
 109         __rte_unused void *opaque)
 110 {
 111         enic_free_wq_buf(wq, buf);
 112 }
 113
 114 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
 115         __rte_unused u8 type, u16 q_number, u16 completed_index, void *opaque)
 116 {
 117         struct enic *enic = vnic_dev_priv(vdev);
 118
 119         vnic_wq_service(&enic->wq[q_number], cq_desc,
 120                 completed_index, enic_wq_free_buf,
 121                 opaque);
 122
 123         return 0;
 124 }
 125
 126 static void enic_log_q_error(struct enic *enic)
 127 {
 128         unsigned int i;
 129         u32 error_status;
 130
 131         for (i = 0; i < enic->wq_count; i++) {
 132                 error_status = vnic_wq_error_status(&enic->wq[i]);
 133                 if (error_status)
 134                         dev_err(enic, "WQ[%d] error_status %d\n", i,
 135                                 error_status);
 136         }
 137
 138         for (i = 0; i < enic->rq_count; i++) {
 139                 error_status = vnic_rq_error_status(&enic->rq[i]);
 140                 if (error_status)
 141                         dev_err(enic, "RQ[%d] error_status %d\n", i,
 142                                 error_status);
 143         }
 144 }
 145
 146 unsigned int enic_cleanup_wq(struct enic *enic, struct vnic_wq *wq)
 147 {
 148         unsigned int cq = enic_cq_wq(enic, wq->index);
 149
 150         /* Return the work done */
 151         return vnic_cq_service(&enic->cq[cq],
 152                 -1 /*wq_work_to_do*/, enic_wq_service, NULL);
 153 }
 154
 155
 156 int enic_send_pkt(struct enic *enic, struct vnic_wq *wq,
 157         struct rte_mbuf *tx_pkt, unsigned short len,
 158         uint8_t sop, uint8_t eop,
 159         uint16_t ol_flags, uint16_t vlan_tag)
 160 {
 161         struct wq_enet_desc *desc = vnic_wq_next_desc(wq);
 162         uint16_t mss = 0;
 163         uint8_t cq_entry = eop;
 164         uint8_t vlan_tag_insert = 0;
 165         uint64_t bus_addr = (dma_addr_t)
 166             (tx_pkt->buf_physaddr + RTE_PKTMBUF_HEADROOM);
 167
 168         if (sop) {
 169                 if (ol_flags & PKT_TX_VLAN_PKT)
 170                         vlan_tag_insert = 1;
 171
 172                 if (enic->hw_ip_checksum) {
 173                         if (ol_flags & PKT_TX_IP_CKSUM)
 174                                 mss |= ENIC_CALC_IP_CKSUM;
 175
 176                         if (ol_flags & PKT_TX_TCP_UDP_CKSUM)
 177                                 mss |= ENIC_CALC_TCP_UDP_CKSUM;
 178                 }
 179         }
 180
 181         wq_enet_desc_enc(desc,
 182                 bus_addr,
 183                 len,
 184                 mss,
 185                 0 /* header_length */,
 186                 0 /* offload_mode WQ_ENET_OFFLOAD_MODE_CSUM */,
 187                 eop,
 188                 cq_entry,
 189                 0 /* fcoe_encap */,
 190                 vlan_tag_insert,
 191                 vlan_tag,
 192                 0 /* loopback */);
 193
 194         vnic_wq_post(wq, (void *)tx_pkt, bus_addr, len,
 195                 sop, eop,
 196                 1 /*desc_skip_cnt*/,
 197                 cq_entry,
 198                 0 /*compressed send*/,
 199                 0 /*wrid*/);
 200
 201         return 0;
 202 }
 203
 204 void enic_dev_stats_clear(struct enic *enic)
 205 {
 206         if (vnic_dev_stats_clear(enic->vdev))
 207                 dev_err(enic, "Error in clearing stats\n");
 208 }
 209
 210 void enic_dev_stats_get(struct enic *enic, struct rte_eth_stats *r_stats)
 211 {
 212         struct vnic_stats *stats;
 213
 214         memset(r_stats, 0, sizeof(*r_stats));
 215         if (vnic_dev_stats_dump(enic->vdev, &stats)) {
 216                 dev_err(enic, "Error in getting stats\n");
 217                 return;
 218         }
 219
 220         r_stats->ipackets = stats->rx.rx_frames_ok;
 221         r_stats->opackets = stats->tx.tx_frames_ok;
 222
 223         r_stats->ibytes = stats->rx.rx_bytes_ok;
 224         r_stats->obytes = stats->tx.tx_bytes_ok;
 225
 226         r_stats->ierrors = stats->rx.rx_errors;
 227         r_stats->oerrors = stats->tx.tx_errors;
 228
 229         r_stats->imcasts = stats->rx.rx_multicast_frames_ok;
 230         r_stats->rx_nombuf = stats->rx.rx_no_bufs;
 231 }
 232
 233 void enic_del_mac_address(struct enic *enic)
 234 {
 235         if (vnic_dev_del_addr(enic->vdev, enic->mac_addr))
 236                 dev_err(enic, "del mac addr failed\n");
 237 }
 238
 239 void enic_set_mac_address(struct enic *enic, uint8_t *mac_addr)
 240 {
 241         int err;
 242
 243         if (!is_eth_addr_valid(mac_addr)) {
 244                 dev_err(enic, "invalid mac address\n");
 245                 return;
 246         }
 247
 248         err = vnic_dev_del_addr(enic->vdev, mac_addr);
 249         if (err) {
 250                 dev_err(enic, "del mac addr failed\n");
 251                 return;
 252         }
 253
 254         ether_addr_copy((struct ether_addr *)mac_addr,
 255                 (struct ether_addr *)enic->mac_addr);
 256
 257         err = vnic_dev_add_addr(enic->vdev, mac_addr);
 258         if (err) {
 259                 dev_err(enic, "add mac addr failed\n");
 260                 return;
 261         }
 262 }
 263
 264 static void
 265 enic_free_rq_buf(__rte_unused struct vnic_rq *rq, struct vnic_rq_buf *buf)
 266 {
 267         if (!buf->os_buf)
 268                 return;
 269
 270         rte_pktmbuf_free((struct rte_mbuf *)buf->os_buf);
 271         buf->os_buf = NULL;
 272 }
 273
 274 void enic_init_vnic_resources(struct enic *enic)
 275 {
 276         unsigned int error_interrupt_enable = 1;
 277         unsigned int error_interrupt_offset = 0;
 278         unsigned int index = 0;
 279
 280         for (index = 0; index < enic->rq_count; index++) {
 281                 vnic_rq_init(&enic->rq[index],
 282                         enic_cq_rq(enic, index),
 283                         error_interrupt_enable,
 284                         error_interrupt_offset);
 285         }
 286
 287         for (index = 0; index < enic->wq_count; index++) {
 288                 vnic_wq_init(&enic->wq[index],
 289                         enic_cq_wq(enic, index),
 290                         error_interrupt_enable,
 291                         error_interrupt_offset);
 292         }
 293
 294         vnic_dev_stats_clear(enic->vdev);
 295
 296         for (index = 0; index < enic->cq_count; index++) {
 297                 vnic_cq_init(&enic->cq[index],
 298                         0 /* flow_control_enable */,
 299                         1 /* color_enable */,
 300                         0 /* cq_head */,
 301                         0 /* cq_tail */,
 302                         1 /* cq_tail_color */,
 303                         0 /* interrupt_enable */,
 304                         1 /* cq_entry_enable */,
 305                         0 /* cq_message_enable */,
 306                         0 /* interrupt offset */,
 307                         0 /* cq_message_addr */);
 308         }
 309
 310         vnic_intr_init(&enic->intr,
 311                 enic->config.intr_timer_usec,
 312                 enic->config.intr_timer_type,
 313                 /*mask_on_assertion*/1);
 314 }
 315
 316
 317 static int enic_rq_alloc_buf(struct vnic_rq *rq)
 318 {
 319         struct enic *enic = vnic_dev_priv(rq->vdev);
 320         dma_addr_t dma_addr;
 321         struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
 322         uint8_t type = RQ_ENET_TYPE_ONLY_SOP;
 323         u16 split_hdr_size = vnic_get_hdr_split_size(enic->vdev);
 324         struct rte_mbuf *mbuf = enic_rxmbuf_alloc(rq->mp);
 325         struct rte_mbuf *hdr_mbuf = NULL;
 326
 327         if (!mbuf) {
 328                 dev_err(enic, "mbuf alloc in enic_rq_alloc_buf failed\n");
 329                 return -1;
 330         }
 331
 332         if (unlikely(split_hdr_size)) {
 333                 if (vnic_rq_desc_avail(rq) < 2) {
 334                         rte_mempool_put(mbuf->pool, mbuf);
 335                         return -1;
 336                 }
 337                 hdr_mbuf = enic_rxmbuf_alloc(rq->mp);
 338                 if (!hdr_mbuf) {
 339                         rte_mempool_put(mbuf->pool, mbuf);
 340                         dev_err(enic,
 341                                 "hdr_mbuf alloc in enic_rq_alloc_buf failed\n");
 342                         return -1;
 343                 }
 344
 345                 hdr_mbuf->data_off = RTE_PKTMBUF_HEADROOM;
 346
 347                 hdr_mbuf->nb_segs = 2;
 348                 hdr_mbuf->port = rq->index;
 349                 hdr_mbuf->next = mbuf;
 350
 351                 dma_addr = (dma_addr_t)
 352                     (hdr_mbuf->buf_physaddr + hdr_mbuf->data_off);
 353
 354                 rq_enet_desc_enc(desc, dma_addr, type, split_hdr_size);
 355
 356                 vnic_rq_post(rq, (void *)hdr_mbuf, 0 /*os_buf_index*/, dma_addr,
 357                         (unsigned int)split_hdr_size, 0 /*wrid*/);
 358
 359                 desc = vnic_rq_next_desc(rq);
 360                 type = RQ_ENET_TYPE_NOT_SOP;
 361         } else {
 362                 mbuf->nb_segs = 1;
 363                 mbuf->port = rq->index;
 364         }
 365
 366         mbuf->data_off = RTE_PKTMBUF_HEADROOM;
 367         mbuf->next = NULL;
 368
 369         dma_addr = (dma_addr_t)
 370             (mbuf->buf_physaddr + mbuf->data_off);
 371
 372         rq_enet_desc_enc(desc, dma_addr, type, mbuf->buf_len);
 373
 374         vnic_rq_post(rq, (void *)mbuf, 0 /*os_buf_index*/, dma_addr,
 375                 (unsigned int)mbuf->buf_len, 0 /*wrid*/);
 376
 377         return 0;
 378 }
 379
 380 static int enic_rq_indicate_buf(struct vnic_rq *rq,
 381         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
 382         int skipped, void *opaque)
 383 {
 384         struct enic *enic = vnic_dev_priv(rq->vdev);
 385         struct rte_mbuf **rx_pkt_bucket = (struct rte_mbuf **)opaque;
 386         struct rte_mbuf *rx_pkt = NULL;
 387         struct rte_mbuf *hdr_rx_pkt = NULL;
 388
 389         u8 type, color, eop, sop, ingress_port, vlan_stripped;
 390         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
 391         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
 392         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
 393         u8 packet_error;
 394         u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
 395         u32 rss_hash;
 396
 397         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
 398                 &type, &color, &q_number, &completed_index,
 399                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
 400                 &csum_not_calc, &rss_hash, &bytes_written,
 401                 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
 402                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
 403                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
 404                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
 405                 &fcs_ok);
 406
 407         rx_pkt = (struct rte_mbuf *)buf->os_buf;
 408         buf->os_buf = NULL;
 409
 410         if (unlikely(packet_error)) {
 411                 dev_err(enic, "packet error\n");
 412                 rx_pkt->data_len = 0;
 413                 return 0;
 414         }
 415
 416         if (unlikely(skipped)) {
 417                 rx_pkt->data_len = 0;
 418                 return 0;
 419         }
 420
 421         if (likely(!vnic_get_hdr_split_size(enic->vdev))) {
 422                 /* No header split configured */
 423                 *rx_pkt_bucket = rx_pkt;
 424                 rx_pkt->pkt_len = bytes_written;
 425
 426                 if (ipv4) {
 427                         rx_pkt->ol_flags |= PKT_RX_IPV4_HDR;
 428                         if (!csum_not_calc) {
 429                                 if (unlikely(!ipv4_csum_ok))
 430                                         rx_pkt->ol_flags |= PKT_RX_IP_CKSUM_BAD;
 431
 432                                 if ((tcp || udp) && (!tcp_udp_csum_ok))
 433                                         rx_pkt->ol_flags |= PKT_RX_L4_CKSUM_BAD;
 434                         }
 435                 } else if (ipv6)
 436                         rx_pkt->ol_flags |= PKT_RX_IPV6_HDR;
 437         } else {
 438                 /* Header split */
 439                 if (sop && !eop) {
 440                         /* This piece is header */
 441                         *rx_pkt_bucket = rx_pkt;
 442                         rx_pkt->pkt_len = bytes_written;
 443                 } else {
 444                         if (sop && eop) {
 445                                 /* The packet is smaller than split_hdr_size */
 446                                 *rx_pkt_bucket = rx_pkt;
 447                                 rx_pkt->pkt_len = bytes_written;
 448                                 if (ipv4) {
 449                                         rx_pkt->ol_flags |= PKT_RX_IPV4_HDR;
 450                                         if (!csum_not_calc) {
 451                                                 if (unlikely(!ipv4_csum_ok))
 452                                                         rx_pkt->ol_flags |=
 453                                                             PKT_RX_IP_CKSUM_BAD;
 454
 455                                                 if ((tcp || udp) &&
 456                                                     (!tcp_udp_csum_ok))
 457                                                         rx_pkt->ol_flags |=
 458                                                             PKT_RX_L4_CKSUM_BAD;
 459                                         }
 460                                 } else if (ipv6)
 461                                         rx_pkt->ol_flags |= PKT_RX_IPV6_HDR;
 462                         } else {
 463                                 /* Payload */
 464                                 hdr_rx_pkt = *rx_pkt_bucket;
 465                                 hdr_rx_pkt->pkt_len += bytes_written;
 466                                 if (ipv4) {
 467                                         hdr_rx_pkt->ol_flags |= PKT_RX_IPV4_HDR;
 468                                         if (!csum_not_calc) {
 469                                                 if (unlikely(!ipv4_csum_ok))
 470                                                         hdr_rx_pkt->ol_flags |=
 471                                                             PKT_RX_IP_CKSUM_BAD;
 472
 473                                                 if ((tcp || udp) &&
 474                                                     (!tcp_udp_csum_ok))
 475                                                         hdr_rx_pkt->ol_flags |=
 476                                                             PKT_RX_L4_CKSUM_BAD;
 477                                         }
 478                                 } else if (ipv6)
 479                                         hdr_rx_pkt->ol_flags |= PKT_RX_IPV6_HDR;
 480
 481                         }
 482                 }
 483         }
 484
 485         rx_pkt->data_len = bytes_written;
 486
 487         if (rss_hash) {
 488                 rx_pkt->ol_flags |= PKT_RX_RSS_HASH;
 489                 rx_pkt->hash.rss = rss_hash;
 490         }
 491
 492         if (vlan_tci) {
 493                 rx_pkt->ol_flags |= PKT_RX_VLAN_PKT;
 494                 rx_pkt->vlan_tci = vlan_tci;
 495         }
 496
 497         return eop;
 498 }
 499
 500 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
 501         __rte_unused u8 type, u16 q_number, u16 completed_index, void *opaque)
 502 {
 503         struct enic *enic = vnic_dev_priv(vdev);
 504
 505         return vnic_rq_service(&enic->rq[q_number], cq_desc,
 506                 completed_index, VNIC_RQ_RETURN_DESC,
 507                 enic_rq_indicate_buf, opaque);
 508
 509 }
 510
 511 int enic_poll(struct vnic_rq *rq, struct rte_mbuf **rx_pkts,
 512         unsigned int budget, unsigned int *work_done)
 513 {
 514         struct enic *enic = vnic_dev_priv(rq->vdev);
 515         unsigned int cq = enic_cq_rq(enic, rq->index);
 516         int err = 0;
 517
 518         *work_done = vnic_cq_service(&enic->cq[cq],
 519                 budget, enic_rq_service, (void *)rx_pkts);
 520
 521         if (*work_done) {
 522                 vnic_rq_fill(rq, enic_rq_alloc_buf);
 523
 524                 /* Need at least one buffer on ring to get going */
 525                 if (vnic_rq_desc_used(rq) == 0) {
 526                         dev_err(enic, "Unable to alloc receive buffers\n");
 527                         err = -1;
 528                 }
 529         }
 530         return err;
 531 }
 532
 533 static void *
 534 enic_alloc_consistent(__rte_unused void *priv, size_t size,
 535         dma_addr_t *dma_handle, u8 *name)
 536 {
 537         void *vaddr;
 538         const struct rte_memzone *rz;
 539         *dma_handle = 0;
 540
 541         rz = rte_memzone_reserve_aligned((const char *)name,
 542                 size, 0, 0, ENIC_ALIGN);
 543         if (!rz) {
 544                 pr_err("%s : Failed to allocate memory requested for %s",
 545                         __func__, name);
 546                 return NULL;
 547         }
 548
 549         vaddr = rz->addr;
 550         *dma_handle = (dma_addr_t)rz->phys_addr;
 551
 552         return vaddr;
 553 }
 554
 555 static void
 556 enic_free_consistent(__rte_unused struct rte_pci_device *hwdev,
 557         __rte_unused size_t size,
 558         __rte_unused void *vaddr,
 559         __rte_unused dma_addr_t dma_handle)
 560 {
 561         /* Nothing to be done */
 562 }
 563
 564 static void
 565 enic_intr_handler(__rte_unused struct rte_intr_handle *handle,
 566         void *arg)
 567 {
 568         struct enic *enic = pmd_priv((struct rte_eth_dev *)arg);
 569
 570         dev_err(enic, "Err intr.\n");
 571         vnic_intr_return_all_credits(&enic->intr);
 572
 573         enic_log_q_error(enic);
 574 }
 575
 576 int enic_enable(struct enic *enic)
 577 {
 578         unsigned int index;
 579         struct rte_eth_dev *eth_dev = enic->rte_dev;
 580
 581         eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
 582         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
 583         vnic_dev_notify_set(enic->vdev, -1); /* No Intr for notify */
 584
 585         if (enic_clsf_init(enic))
 586                 dev_warning(enic, "Init of hash table for clsf failed."\
 587                         "Flow director feature will not work\n");
 588
 589         /* Fill RQ bufs */
 590         for (index = 0; index < enic->rq_count; index++) {
 591                 vnic_rq_fill(&enic->rq[index], enic_rq_alloc_buf);
 592
 593                 /* Need at least one buffer on ring to get going
 594                 */
 595                 if (vnic_rq_desc_used(&enic->rq[index]) == 0) {
 596                         dev_err(enic, "Unable to alloc receive buffers\n");
 597                         return -1;
 598                 }
 599         }
 600
 601         for (index = 0; index < enic->wq_count; index++)
 602                 vnic_wq_enable(&enic->wq[index]);
 603         for (index = 0; index < enic->rq_count; index++)
 604                 vnic_rq_enable(&enic->rq[index]);
 605
 606         vnic_dev_enable_wait(enic->vdev);
 607
 608 #ifndef VFIO_PRESENT
 609         /* Register and enable error interrupt */
 610         rte_intr_callback_register(&(enic->pdev->intr_handle),
 611                 enic_intr_handler, (void *)enic->rte_dev);
 612
 613         rte_intr_enable(&(enic->pdev->intr_handle));
 614 #endif
 615         vnic_intr_unmask(&enic->intr);
 616
 617         return 0;
 618 }
 619
 620 int enic_alloc_intr_resources(struct enic *enic)
 621 {
 622         int err;
 623
 624         dev_info(enic, "vNIC resources used:  "\
 625                 "wq %d rq %d cq %d intr %d\n",
 626                 enic->wq_count, enic->rq_count,
 627                 enic->cq_count, enic->intr_count);
 628
 629         err = vnic_intr_alloc(enic->vdev, &enic->intr, 0);
 630         if (err)
 631                 enic_free_vnic_resources(enic);
 632
 633         return err;
 634 }
 635
 636 void enic_free_rq(void *rxq)
 637 {
 638         struct vnic_rq *rq = (struct vnic_rq *)rxq;
 639         struct enic *enic = vnic_dev_priv(rq->vdev);
 640
 641         vnic_rq_free(rq);
 642         vnic_cq_free(&enic->cq[rq->index]);
 643 }
 644
 645 void enic_start_wq(struct enic *enic, uint16_t queue_idx)
 646 {
 647         vnic_wq_enable(&enic->wq[queue_idx]);
 648 }
 649
 650 int enic_stop_wq(struct enic *enic, uint16_t queue_idx)
 651 {
 652         return vnic_wq_disable(&enic->wq[queue_idx]);
 653 }
 654
 655 void enic_start_rq(struct enic *enic, uint16_t queue_idx)
 656 {
 657         vnic_rq_enable(&enic->rq[queue_idx]);
 658 }
 659
 660 int enic_stop_rq(struct enic *enic, uint16_t queue_idx)
 661 {
 662         return vnic_rq_disable(&enic->rq[queue_idx]);
 663 }
 664
 665 int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
 666         unsigned int socket_id, struct rte_mempool *mp,
 667         uint16_t nb_desc)
 668 {
 669         int err;
 670         struct vnic_rq *rq = &enic->rq[queue_idx];
 671
 672         rq->socket_id = socket_id;
 673         rq->mp = mp;
 674
 675         if (nb_desc) {
 676                 if (nb_desc > enic->config.rq_desc_count) {
 677                         dev_warning(enic,
 678                                 "RQ %d - number of rx desc in cmd line (%d)"\
 679                                 "is greater than that in the UCSM/CIMC adapter"\
 680                                 "policy.  Applying the value in the adapter "\
 681                                 "policy (%d).\n",
 682                                 queue_idx, nb_desc, enic->config.rq_desc_count);
 683                 } else if (nb_desc != enic->config.rq_desc_count) {
 684                         enic->config.rq_desc_count = nb_desc;
 685                         dev_info(enic,
 686                                 "RX Queues - effective number of descs:%d\n",
 687                                 nb_desc);
 688                 }
 689         }
 690
 691         /* Allocate queue resources */
 692         err = vnic_rq_alloc(enic->vdev, &enic->rq[queue_idx], queue_idx,
 693                 enic->config.rq_desc_count,
 694                 sizeof(struct rq_enet_desc));
 695         if (err) {
 696                 dev_err(enic, "error in allocation of rq\n");
 697                 return err;
 698         }
 699
 700         err = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
 701                 socket_id, enic->config.rq_desc_count,
 702                 sizeof(struct cq_enet_rq_desc));
 703         if (err) {
 704                 vnic_rq_free(rq);
 705                 dev_err(enic, "error in allocation of cq for rq\n");
 706         }
 707
 708         return err;
 709 }
 710
 711 void enic_free_wq(void *txq)
 712 {
 713         struct vnic_wq *wq = (struct vnic_wq *)txq;
 714         struct enic *enic = vnic_dev_priv(wq->vdev);
 715
 716         vnic_wq_free(wq);
 717         vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
 718 }
 719
 720 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
 721         unsigned int socket_id, uint16_t nb_desc)
 722 {
 723         int err;
 724         struct vnic_wq *wq = &enic->wq[queue_idx];
 725         unsigned int cq_index = enic_cq_wq(enic, queue_idx);
 726
 727         wq->socket_id = socket_id;
 728         if (nb_desc) {
 729                 if (nb_desc > enic->config.wq_desc_count) {
 730                         dev_warning(enic,
 731                                 "WQ %d - number of tx desc in cmd line (%d)"\
 732                                 "is greater than that in the UCSM/CIMC adapter"\
 733                                 "policy.  Applying the value in the adapter "\
 734                                 "policy (%d)\n",
 735                                 queue_idx, nb_desc, enic->config.wq_desc_count);
 736                 } else if (nb_desc != enic->config.wq_desc_count) {
 737                         enic->config.wq_desc_count = nb_desc;
 738                         dev_info(enic,
 739                                 "TX Queues - effective number of descs:%d\n",
 740                                 nb_desc);
 741                 }
 742         }
 743
 744         /* Allocate queue resources */
 745         err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
 746                 enic->config.wq_desc_count,
 747                 sizeof(struct wq_enet_desc));
 748         if (err) {
 749                 dev_err(enic, "error in allocation of wq\n");
 750                 return err;
 751         }
 752
 753         err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
 754                 socket_id, enic->config.wq_desc_count,
 755                 sizeof(struct cq_enet_wq_desc));
 756         if (err) {
 757                 vnic_wq_free(wq);
 758                 dev_err(enic, "error in allocation of cq for wq\n");
 759         }
 760
 761         return err;
 762 }
 763
 764 int enic_disable(struct enic *enic)
 765 {
 766         unsigned int i;
 767         int err;
 768
 769         vnic_intr_mask(&enic->intr);
 770         (void)vnic_intr_masked(&enic->intr); /* flush write */
 771
 772         vnic_dev_disable(enic->vdev);
 773
 774         enic_clsf_destroy(enic);
 775
 776         if (!enic_is_sriov_vf(enic))
 777                 vnic_dev_del_addr(enic->vdev, enic->mac_addr);
 778
 779         for (i = 0; i < enic->wq_count; i++) {
 780                 err = vnic_wq_disable(&enic->wq[i]);
 781                 if (err)
 782                         return err;
 783         }
 784         for (i = 0; i < enic->rq_count; i++) {
 785                 err = vnic_rq_disable(&enic->rq[i]);
 786                 if (err)
 787                         return err;
 788         }
 789
 790         vnic_dev_set_reset_flag(enic->vdev, 1);
 791         vnic_dev_notify_unset(enic->vdev);
 792
 793         for (i = 0; i < enic->wq_count; i++)
 794                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
 795         for (i = 0; i < enic->rq_count; i++)
 796                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
 797         for (i = 0; i < enic->cq_count; i++)
 798                 vnic_cq_clean(&enic->cq[i]);
 799         vnic_intr_clean(&enic->intr);
 800
 801         return 0;
 802 }
 803
 804 static int enic_dev_wait(struct vnic_dev *vdev,
 805         int (*start)(struct vnic_dev *, int),
 806         int (*finished)(struct vnic_dev *, int *),
 807         int arg)
 808 {
 809         int done;
 810         int err;
 811         int i;
 812
 813         err = start(vdev, arg);
 814         if (err)
 815                 return err;
 816
 817         /* Wait for func to complete...2 seconds max */
 818         for (i = 0; i < 2000; i++) {
 819                 err = finished(vdev, &done);
 820                 if (err)
 821                         return err;
 822                 if (done)
 823                         return 0;
 824                 usleep(1000);
 825         }
 826         return -ETIMEDOUT;
 827 }
 828
 829 static int enic_dev_open(struct enic *enic)
 830 {
 831         int err;
 832
 833         err = enic_dev_wait(enic->vdev, vnic_dev_open,
 834                 vnic_dev_open_done, 0);
 835         if (err)
 836                 dev_err(enic_get_dev(enic),
 837                         "vNIC device open failed, err %d\n", err);
 838
 839         return err;
 840 }
 841
 842 static int enic_set_rsskey(struct enic *enic)
 843 {
 844         dma_addr_t rss_key_buf_pa;
 845         union vnic_rss_key *rss_key_buf_va = NULL;
 846         static union vnic_rss_key rss_key = {
 847                 .key[0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
 848                 .key[1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
 849                 .key[2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
 850                 .key[3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
 851         };
 852         int err;
 853         u8 name[NAME_MAX];
 854
 855         snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
 856         rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
 857                 &rss_key_buf_pa, name);
 858         if (!rss_key_buf_va)
 859                 return -ENOMEM;
 860
 861         rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
 862
 863         err = enic_set_rss_key(enic,
 864                 rss_key_buf_pa,
 865                 sizeof(union vnic_rss_key));
 866
 867         enic_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
 868                 rss_key_buf_va, rss_key_buf_pa);
 869
 870         return err;
 871 }
 872
 873 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
 874 {
 875         dma_addr_t rss_cpu_buf_pa;
 876         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
 877         int i;
 878         int err;
 879         u8 name[NAME_MAX];
 880
 881         snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
 882         rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
 883                 &rss_cpu_buf_pa, name);
 884         if (!rss_cpu_buf_va)
 885                 return -ENOMEM;
 886
 887         for (i = 0; i < (1 << rss_hash_bits); i++)
 888                 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
 889
 890         err = enic_set_rss_cpu(enic,
 891                 rss_cpu_buf_pa,
 892                 sizeof(union vnic_rss_cpu));
 893
 894         enic_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
 895                 rss_cpu_buf_va, rss_cpu_buf_pa);
 896
 897         return err;
 898 }
 899
 900 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
 901         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
 902 {
 903         const u8 tso_ipid_split_en = 0;
 904         int err;
 905
 906         /* Enable VLAN tag stripping */
 907
 908         err = enic_set_nic_cfg(enic,
 909                 rss_default_cpu, rss_hash_type,
 910                 rss_hash_bits, rss_base_cpu,
 911                 rss_enable, tso_ipid_split_en,
 912                 enic->ig_vlan_strip_en);
 913
 914         return err;
 915 }
 916
 917 int enic_set_rss_nic_cfg(struct enic *enic)
 918 {
 919         const u8 rss_default_cpu = 0;
 920         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
 921             NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
 922             NIC_CFG_RSS_HASH_TYPE_IPV6 |
 923             NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
 924         const u8 rss_hash_bits = 7;
 925         const u8 rss_base_cpu = 0;
 926         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
 927
 928         if (rss_enable) {
 929                 if (!enic_set_rsskey(enic)) {
 930                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
 931                                 rss_enable = 0;
 932                                 dev_warning(enic, "RSS disabled, "\
 933                                         "Failed to set RSS cpu indirection table.");
 934                         }
 935                 } else {
 936                         rss_enable = 0;
 937                         dev_warning(enic,
 938                                 "RSS disabled, Failed to set RSS key.\n");
 939                 }
 940         }
 941
 942         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
 943                 rss_hash_bits, rss_base_cpu, rss_enable);
 944 }
 945
 946 int enic_setup_finish(struct enic *enic)
 947 {
 948         int ret;
 949
 950         ret = enic_set_rss_nic_cfg(enic);
 951         if (ret) {
 952                 dev_err(enic, "Failed to config nic, aborting.\n");
 953                 return -1;
 954         }
 955
 956         vnic_dev_add_addr(enic->vdev, enic->mac_addr);
 957
 958         /* Default conf */
 959         vnic_dev_packet_filter(enic->vdev,
 960                 1 /* directed  */,
 961                 1 /* multicast */,
 962                 1 /* broadcast */,
 963                 0 /* promisc   */,
 964                 1 /* allmulti  */);
 965
 966         enic->promisc = 0;
 967         enic->allmulti = 1;
 968
 969         return 0;
 970 }
 971
 972 #ifdef VFIO_PRESENT
 973 static void enic_eventfd_init(struct enic *enic)
 974 {
 975         enic->eventfd = enic->pdev->intr_handle.fd;
 976 }
 977
 978 void *enic_err_intr_handler(void *arg)
 979 {
 980         struct enic *enic = (struct enic *)arg;
 981         unsigned int intr = enic_msix_err_intr(enic);
 982         ssize_t size;
 983         uint64_t data;
 984
 985         while (1) {
 986                 size = read(enic->eventfd, &data, sizeof(data));
 987                 dev_err(enic, "Err intr.\n");
 988                 vnic_intr_return_all_credits(&enic->intr);
 989
 990                 enic_log_q_error(enic);
 991         }
 992
 993         return NULL;
 994 }
 995 #endif
 996
 997 void enic_add_packet_filter(struct enic *enic)
 998 {
 999         /* Args -> directed, multicast, broadcast, promisc, allmulti */
1000         vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
1001                 enic->promisc, enic->allmulti);
1002 }
1003
1004 int enic_get_link_status(struct enic *enic)
1005 {
1006         return vnic_dev_link_status(enic->vdev);
1007 }
1008
1009
1010 #ifdef VFIO_PRESENT
1011 static int enic_create_err_intr_thread(struct enic *enic)
1012 {
1013         pthread_attr_t intr_attr;
1014
1015         /* create threads for error interrupt handling */
1016         pthread_attr_init(&intr_attr);
1017         pthread_attr_setstacksize(&intr_attr, 0x100000);
1018
1019         /* ERR */
1020         if (pthread_create(&enic->err_intr_thread, &intr_attr,
1021                     enic_err_intr_handler, (void *)enic)) {
1022                 dev_err(enic, "Failed to create err interrupt handler threads\n");
1023                 return -1;
1024         }
1025
1026         pthread_attr_destroy(&intr_attr);
1027
1028         return 0;
1029 }
1030
1031
1032 static int enic_set_intr_mode(struct enic *enic)
1033 {
1034         struct vfio_irq_set *irq_set;
1035         int *fds;
1036         int size;
1037         int ret = -1;
1038         int index;
1039
1040         if (enic->intr_count < 1) {
1041                 dev_err(enic, "Unsupported resource conf.\n");
1042                 return -1;
1043         }
1044         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1045
1046         enic->intr_count = 1;
1047
1048         enic_eventfd_init(enic);
1049         size = sizeof(*irq_set) + (sizeof(int));
1050
1051         irq_set = rte_zmalloc("enic_vfio_irq", size, 0);
1052         irq_set->argsz = size;
1053         irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
1054         irq_set->start = 0;
1055         irq_set->count = 1; /* For error interrupt only */
1056         irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD |
1057             VFIO_IRQ_SET_ACTION_TRIGGER;
1058         fds = (int *)&irq_set->data;
1059
1060         fds[0] = enic->eventfd;
1061
1062         ret = ioctl(enic->pdev->intr_handle.vfio_dev_fd,
1063                 VFIO_DEVICE_SET_IRQS, irq_set);
1064         rte_free(irq_set);
1065         if (ret) {
1066                 dev_err(enic, "Failed to set eventfds for interrupts\n");
1067                 return -1;
1068         }
1069
1070         enic_create_err_intr_thread(enic);
1071         return 0;
1072 }
1073
1074 static void enic_clear_intr_mode(struct enic *enic)
1075 {
1076         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1077 }
1078 #endif
1079
1080 static void enic_dev_deinit(struct enic *enic)
1081 {
1082         struct rte_eth_dev *eth_dev = enic->rte_dev;
1083
1084         if (eth_dev->data->mac_addrs)
1085                 rte_free(eth_dev->data->mac_addrs);
1086
1087 #ifdef VFIO_PRESENT
1088         enic_clear_intr_mode(enic);
1089 #endif
1090 }
1091
1092
1093 int enic_set_vnic_res(struct enic *enic)
1094 {
1095         struct rte_eth_dev *eth_dev = enic->rte_dev;
1096
1097         if ((enic->rq_count < eth_dev->data->nb_rx_queues) ||
1098                 (enic->wq_count < eth_dev->data->nb_tx_queues)) {
1099                 dev_err(dev, "Not enough resources configured, aborting\n");
1100                 return -1;
1101         }
1102
1103         enic->rq_count = eth_dev->data->nb_rx_queues;
1104         enic->wq_count = eth_dev->data->nb_tx_queues;
1105         if (enic->cq_count < (enic->rq_count + enic->wq_count)) {
1106                 dev_err(dev, "Not enough resources configured, aborting\n");
1107                 return -1;
1108         }
1109
1110         enic->cq_count = enic->rq_count + enic->wq_count;
1111         return 0;
1112 }
1113
1114 static int enic_dev_init(struct enic *enic)
1115 {
1116         int err;
1117         struct rte_eth_dev *eth_dev = enic->rte_dev;
1118
1119         vnic_dev_intr_coal_timer_info_default(enic->vdev);
1120
1121         /* Get vNIC configuration
1122         */
1123         err = enic_get_vnic_config(enic);
1124         if (err) {
1125                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
1126                 return err;
1127         }
1128
1129         eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN, 0);
1130         if (!eth_dev->data->mac_addrs) {
1131                 dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1132                 return -1;
1133         }
1134         ether_addr_copy((struct ether_addr *) enic->mac_addr,
1135                 &eth_dev->data->mac_addrs[0]);
1136
1137
1138         /* Get available resource counts
1139         */
1140         enic_get_res_counts(enic);
1141
1142 #ifdef VFIO_PRESENT
1143         /* Set interrupt mode based on resource counts and system
1144          * capabilities
1145          */
1146         err = enic_set_intr_mode(enic);
1147         if (err) {
1148                 rte_free(eth_dev->data->mac_addrs);
1149                 enic_clear_intr_mode(enic);
1150                 dev_err(dev, "Failed to set intr mode based on resource "\
1151                         "counts and system capabilities, aborting\n");
1152                 return err;
1153         }
1154 #endif
1155
1156         vnic_dev_set_reset_flag(enic->vdev, 0);
1157
1158         return 0;
1159
1160 }
1161
1162 int enic_probe(struct enic *enic)
1163 {
1164         struct rte_pci_device *pdev = enic->pdev;
1165         int err = -1;
1166
1167         dev_info(enic, " Initializing ENIC PMD version %s\n", DRV_VERSION);
1168
1169         enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1170         enic->bar0.len = pdev->mem_resource[0].len;
1171
1172         /* Register vNIC device */
1173         enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1174         if (!enic->vdev) {
1175                 dev_err(enic, "vNIC registration failed, aborting\n");
1176                 goto err_out;
1177         }
1178
1179         vnic_register_cbacks(enic->vdev,
1180                 enic_alloc_consistent,
1181                 enic_free_consistent);
1182
1183         /* Issue device open to get device in known state */
1184         err = enic_dev_open(enic);
1185         if (err) {
1186                 dev_err(enic, "vNIC dev open failed, aborting\n");
1187                 goto err_out_unregister;
1188         }
1189
1190         /* Set ingress vlan rewrite mode before vnic initialization */
1191         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1192                 IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN);
1193         if (err) {
1194                 dev_err(enic,
1195                         "Failed to set ingress vlan rewrite mode, aborting.\n");
1196                 goto err_out_dev_close;
1197         }
1198
1199         /* Issue device init to initialize the vnic-to-switch link.
1200          * We'll start with carrier off and wait for link UP
1201          * notification later to turn on carrier.  We don't need
1202          * to wait here for the vnic-to-switch link initialization
1203          * to complete; link UP notification is the indication that
1204          * the process is complete.
1205          */
1206
1207         err = vnic_dev_init(enic->vdev, 0);
1208         if (err) {
1209                 dev_err(enic, "vNIC dev init failed, aborting\n");
1210                 goto err_out_dev_close;
1211         }
1212
1213         err = enic_dev_init(enic);
1214         if (err) {
1215                 dev_err(enic, "Device initialization failed, aborting\n");
1216                 goto err_out_dev_close;
1217         }
1218
1219         return 0;
1220
1221 err_out_dev_close:
1222         vnic_dev_close(enic->vdev);
1223 err_out_unregister:
1224         vnic_dev_unregister(enic->vdev);
1225 err_out:
1226         return err;
1227 }
1228
1229 void enic_remove(struct enic *enic)
1230 {
1231         enic_dev_deinit(enic);
1232         vnic_dev_close(enic->vdev);
1233         vnic_dev_unregister(enic->vdev);
1234 }