drivers/net/dpaa/dpaa_rxtx.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  *   Copyright 2016 Freescale Semiconductor, Inc. All rights reserved.
   4  *   Copyright 2017,2019 NXP
   5  *
   6  */
   7
   8 /* System headers */
   9 #include <inttypes.h>
  10 #include <unistd.h>
  11 #include <stdio.h>
  12 #include <limits.h>
  13 #include <sched.h>
  14 #include <pthread.h>
  15
  16 #include <rte_byteorder.h>
  17 #include <rte_common.h>
  18 #include <rte_interrupts.h>
  19 #include <rte_log.h>
  20 #include <rte_debug.h>
  21 #include <rte_pci.h>
  22 #include <rte_atomic.h>
  23 #include <rte_branch_prediction.h>
  24 #include <rte_memory.h>
  25 #include <rte_tailq.h>
  26 #include <rte_eal.h>
  27 #include <rte_alarm.h>
  28 #include <rte_ether.h>
  29 #include <rte_ethdev_driver.h>
  30 #include <rte_malloc.h>
  31 #include <rte_ring.h>
  32 #include <rte_ip.h>
  33 #include <rte_tcp.h>
  34 #include <rte_udp.h>
  35 #include <rte_net.h>
  36 #include <rte_eventdev.h>
  37
  38 #include "dpaa_ethdev.h"
  39 #include "dpaa_rxtx.h"
  40 #include <rte_dpaa_bus.h>
  41 #include <dpaa_mempool.h>
  42
  43 #include <qman.h>
  44 #include <fsl_usd.h>
  45 #include <fsl_qman.h>
  46 #include <fsl_bman.h>
  47 #include <of.h>
  48 #include <netcfg.h>
  49
  50 #define DPAA_MBUF_TO_CONTIG_FD(_mbuf, _fd, _bpid) \
  51         do { \
  52                 (_fd)->cmd = 0; \
  53                 (_fd)->opaque_addr = 0; \
  54                 (_fd)->opaque = QM_FD_CONTIG << DPAA_FD_FORMAT_SHIFT; \
  55                 (_fd)->opaque |= ((_mbuf)->data_off) << DPAA_FD_OFFSET_SHIFT; \
  56                 (_fd)->opaque |= (_mbuf)->pkt_len; \
  57                 (_fd)->addr = (_mbuf)->buf_iova; \
  58                 (_fd)->bpid = _bpid; \
  59         } while (0)
  60
  61 #if (defined RTE_LIBRTE_DPAA_DEBUG_DRIVER)
  62 static void dpaa_display_frame(const struct qm_fd *fd)
  63 {
  64         int ii;
  65         char *ptr;
  66
  67         printf("%s::bpid %x addr %08x%08x, format %d off %d, len %d stat %x\n",
  68                __func__, fd->bpid, fd->addr_hi, fd->addr_lo, fd->format,
  69                 fd->offset, fd->length20, fd->status);
  70
  71         ptr = (char *)rte_dpaa_mem_ptov(fd->addr);
  72         ptr += fd->offset;
  73         printf("%02x ", *ptr);
  74         for (ii = 1; ii < fd->length20; ii++) {
  75                 printf("%02x ", *ptr);
  76                 if ((ii % 16) == 0)
  77                         printf("\n");
  78                 ptr++;
  79         }
  80         printf("\n");
  81 }
  82 #else
  83 #define dpaa_display_frame(a)
  84 #endif
  85
  86 static inline void dpaa_slow_parsing(struct rte_mbuf *m __rte_unused,
  87                                      uint64_t prs __rte_unused)
  88 {
  89         DPAA_DP_LOG(DEBUG, "Slow parsing");
  90         /*TBD:XXX: to be implemented*/
  91 }
  92
  93 static inline void dpaa_eth_packet_info(struct rte_mbuf *m, void *fd_virt_addr)
  94 {
  95         struct annotations_t *annot = GET_ANNOTATIONS(fd_virt_addr);
  96         uint64_t prs = *((uintptr_t *)(&annot->parse)) & DPAA_PARSE_MASK;
  97
  98         DPAA_DP_LOG(DEBUG, " Parsing mbuf: %p with annotations: %p", m, annot);
  99
 100         switch (prs) {
 101         case DPAA_PKT_TYPE_IPV4:
 102                 m->packet_type = RTE_PTYPE_L2_ETHER |
 103                         RTE_PTYPE_L3_IPV4;
 104                 break;
 105         case DPAA_PKT_TYPE_IPV6:
 106                 m->packet_type = RTE_PTYPE_L2_ETHER |
 107                         RTE_PTYPE_L3_IPV6;
 108                 break;
 109         case DPAA_PKT_TYPE_ETHER:
 110                 m->packet_type = RTE_PTYPE_L2_ETHER;
 111                 break;
 112         case DPAA_PKT_TYPE_IPV4_FRAG:
 113         case DPAA_PKT_TYPE_IPV4_FRAG_UDP:
 114         case DPAA_PKT_TYPE_IPV4_FRAG_TCP:
 115         case DPAA_PKT_TYPE_IPV4_FRAG_SCTP:
 116                 m->packet_type = RTE_PTYPE_L2_ETHER |
 117                         RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_FRAG;
 118                 break;
 119         case DPAA_PKT_TYPE_IPV6_FRAG:
 120         case DPAA_PKT_TYPE_IPV6_FRAG_UDP:
 121         case DPAA_PKT_TYPE_IPV6_FRAG_TCP:
 122         case DPAA_PKT_TYPE_IPV6_FRAG_SCTP:
 123                 m->packet_type = RTE_PTYPE_L2_ETHER |
 124                         RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_FRAG;
 125                 break;
 126         case DPAA_PKT_TYPE_IPV4_EXT:
 127                 m->packet_type = RTE_PTYPE_L2_ETHER |
 128                         RTE_PTYPE_L3_IPV4_EXT;
 129                 break;
 130         case DPAA_PKT_TYPE_IPV6_EXT:
 131                 m->packet_type = RTE_PTYPE_L2_ETHER |
 132                         RTE_PTYPE_L3_IPV6_EXT;
 133                 break;
 134         case DPAA_PKT_TYPE_IPV4_TCP:
 135                 m->packet_type = RTE_PTYPE_L2_ETHER |
 136                         RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP;
 137                 break;
 138         case DPAA_PKT_TYPE_IPV6_TCP:
 139                 m->packet_type = RTE_PTYPE_L2_ETHER |
 140                         RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP;
 141                 break;
 142         case DPAA_PKT_TYPE_IPV4_UDP:
 143                 m->packet_type = RTE_PTYPE_L2_ETHER |
 144                         RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP;
 145                 break;
 146         case DPAA_PKT_TYPE_IPV6_UDP:
 147                 m->packet_type = RTE_PTYPE_L2_ETHER |
 148                         RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP;
 149                 break;
 150         case DPAA_PKT_TYPE_IPV4_EXT_UDP:
 151                 m->packet_type = RTE_PTYPE_L2_ETHER |
 152                         RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_UDP;
 153                 break;
 154         case DPAA_PKT_TYPE_IPV6_EXT_UDP:
 155                 m->packet_type = RTE_PTYPE_L2_ETHER |
 156                         RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_UDP;
 157                 break;
 158         case DPAA_PKT_TYPE_IPV4_EXT_TCP:
 159                 m->packet_type = RTE_PTYPE_L2_ETHER |
 160                         RTE_PTYPE_L3_IPV4_EXT | RTE_PTYPE_L4_TCP;
 161                 break;
 162         case DPAA_PKT_TYPE_IPV6_EXT_TCP:
 163                 m->packet_type = RTE_PTYPE_L2_ETHER |
 164                         RTE_PTYPE_L3_IPV6_EXT | RTE_PTYPE_L4_TCP;
 165                 break;
 166         case DPAA_PKT_TYPE_IPV4_SCTP:
 167                 m->packet_type = RTE_PTYPE_L2_ETHER |
 168                         RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_SCTP;
 169                 break;
 170         case DPAA_PKT_TYPE_IPV6_SCTP:
 171                 m->packet_type = RTE_PTYPE_L2_ETHER |
 172                         RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_SCTP;
 173                 break;
 174         case DPAA_PKT_TYPE_NONE:
 175                 m->packet_type = 0;
 176                 break;
 177         /* More switch cases can be added */
 178         default:
 179                 dpaa_slow_parsing(m, prs);
 180         }
 181
 182         m->tx_offload = annot->parse.ip_off[0];
 183         m->tx_offload |= (annot->parse.l4_off - annot->parse.ip_off[0])
 184                                         << DPAA_PKT_L3_LEN_SHIFT;
 185
 186         /* Set the hash values */
 187         m->hash.rss = (uint32_t)(annot->hash);
 188         /* All packets with Bad checksum are dropped by interface (and
 189          * corresponding notification issued to RX error queues).
 190          */
 191         m->ol_flags = PKT_RX_RSS_HASH | PKT_RX_IP_CKSUM_GOOD;
 192
 193         /* Check if Vlan is present */
 194         if (prs & DPAA_PARSE_VLAN_MASK)
 195                 m->ol_flags |= PKT_RX_VLAN;
 196         /* Packet received without stripping the vlan */
 197 }
 198
 199 static inline void dpaa_checksum(struct rte_mbuf *mbuf)
 200 {
 201         struct rte_ether_hdr *eth_hdr =
 202                 rte_pktmbuf_mtod(mbuf, struct rte_ether_hdr *);
 203         char *l3_hdr = (char *)eth_hdr + mbuf->l2_len;
 204         struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)l3_hdr;
 205         struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)l3_hdr;
 206
 207         DPAA_DP_LOG(DEBUG, "Calculating checksum for mbuf: %p", mbuf);
 208
 209         if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV4) ||
 210             ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
 211             RTE_PTYPE_L3_IPV4_EXT)) {
 212                 ipv4_hdr = (struct ipv4_hdr *)l3_hdr;
 213                 ipv4_hdr->hdr_checksum = 0;
 214                 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
 215         } else if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
 216                    RTE_PTYPE_L3_IPV6) ||
 217                    ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
 218                    RTE_PTYPE_L3_IPV6_EXT))
 219                 ipv6_hdr = (struct ipv6_hdr *)l3_hdr;
 220
 221         if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_TCP) {
 222                 struct tcp_hdr *tcp_hdr = (struct tcp_hdr *)(l3_hdr +
 223                                           mbuf->l3_len);
 224                 tcp_hdr->cksum = 0;
 225                 if (eth_hdr->ether_type == htons(RTE_ETHER_TYPE_IPv4))
 226                         tcp_hdr->cksum = rte_ipv4_udptcp_cksum(ipv4_hdr,
 227                                                                tcp_hdr);
 228                 else /* assume ethertype == RTE_ETHER_TYPE_IPv6 */
 229                         tcp_hdr->cksum = rte_ipv6_udptcp_cksum(ipv6_hdr,
 230                                                                tcp_hdr);
 231         } else if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) ==
 232                    RTE_PTYPE_L4_UDP) {
 233                 struct udp_hdr *udp_hdr = (struct udp_hdr *)(l3_hdr +
 234                                                              mbuf->l3_len);
 235                 udp_hdr->dgram_cksum = 0;
 236                 if (eth_hdr->ether_type == htons(RTE_ETHER_TYPE_IPv4))
 237                         udp_hdr->dgram_cksum = rte_ipv4_udptcp_cksum(ipv4_hdr,
 238                                                                      udp_hdr);
 239                 else /* assume ethertype == RTE_ETHER_TYPE_IPv6 */
 240                         udp_hdr->dgram_cksum = rte_ipv6_udptcp_cksum(ipv6_hdr,
 241                                                                      udp_hdr);
 242         }
 243 }
 244
 245 static inline void dpaa_checksum_offload(struct rte_mbuf *mbuf,
 246                                          struct qm_fd *fd, char *prs_buf)
 247 {
 248         struct dpaa_eth_parse_results_t *prs;
 249
 250         DPAA_DP_LOG(DEBUG, " Offloading checksum for mbuf: %p", mbuf);
 251
 252         prs = GET_TX_PRS(prs_buf);
 253         prs->l3r = 0;
 254         prs->l4r = 0;
 255         if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV4) ||
 256            ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
 257            RTE_PTYPE_L3_IPV4_EXT))
 258                 prs->l3r = DPAA_L3_PARSE_RESULT_IPV4;
 259         else if (((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
 260                    RTE_PTYPE_L3_IPV6) ||
 261                  ((mbuf->packet_type & RTE_PTYPE_L3_MASK) ==
 262                 RTE_PTYPE_L3_IPV6_EXT))
 263                 prs->l3r = DPAA_L3_PARSE_RESULT_IPV6;
 264
 265         if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_TCP)
 266                 prs->l4r = DPAA_L4_PARSE_RESULT_TCP;
 267         else if ((mbuf->packet_type & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_UDP)
 268                 prs->l4r = DPAA_L4_PARSE_RESULT_UDP;
 269
 270         prs->ip_off[0] = mbuf->l2_len;
 271         prs->l4_off = mbuf->l3_len + mbuf->l2_len;
 272         /* Enable L3 (and L4, if TCP or UDP) HW checksum*/
 273         fd->cmd = DPAA_FD_CMD_RPD | DPAA_FD_CMD_DTC;
 274 }
 275
 276 static inline void
 277 dpaa_unsegmented_checksum(struct rte_mbuf *mbuf, struct qm_fd *fd_arr)
 278 {
 279         if (!mbuf->packet_type) {
 280                 struct rte_net_hdr_lens hdr_lens;
 281
 282                 mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
 283                                 RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
 284                                 | RTE_PTYPE_L4_MASK);
 285                 mbuf->l2_len = hdr_lens.l2_len;
 286                 mbuf->l3_len = hdr_lens.l3_len;
 287         }
 288         if (mbuf->data_off < (DEFAULT_TX_ICEOF +
 289             sizeof(struct dpaa_eth_parse_results_t))) {
 290                 DPAA_DP_LOG(DEBUG, "Checksum offload Err: "
 291                         "Not enough Headroom "
 292                         "space for correct Checksum offload."
 293                         "So Calculating checksum in Software.");
 294                 dpaa_checksum(mbuf);
 295         } else {
 296                 dpaa_checksum_offload(mbuf, fd_arr, mbuf->buf_addr);
 297         }
 298 }
 299
 300 struct rte_mbuf *
 301 dpaa_eth_sg_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
 302 {
 303         struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
 304         struct rte_mbuf *first_seg, *prev_seg, *cur_seg, *temp;
 305         struct qm_sg_entry *sgt, *sg_temp;
 306         void *vaddr, *sg_vaddr;
 307         int i = 0;
 308         uint8_t fd_offset = fd->offset;
 309
 310         vaddr = DPAA_MEMPOOL_PTOV(bp_info, qm_fd_addr(fd));
 311         if (!vaddr) {
 312                 DPAA_PMD_ERR("unable to convert physical address");
 313                 return NULL;
 314         }
 315         sgt = vaddr + fd_offset;
 316         sg_temp = &sgt[i++];
 317         hw_sg_to_cpu(sg_temp);
 318         temp = (struct rte_mbuf *)((char *)vaddr - bp_info->meta_data_size);
 319         sg_vaddr = DPAA_MEMPOOL_PTOV(bp_info, qm_sg_entry_get64(sg_temp));
 320
 321         first_seg = (struct rte_mbuf *)((char *)sg_vaddr -
 322                                                 bp_info->meta_data_size);
 323         first_seg->data_off = sg_temp->offset;
 324         first_seg->data_len = sg_temp->length;
 325         first_seg->pkt_len = sg_temp->length;
 326         rte_mbuf_refcnt_set(first_seg, 1);
 327
 328         first_seg->port = ifid;
 329         first_seg->nb_segs = 1;
 330         first_seg->ol_flags = 0;
 331         prev_seg = first_seg;
 332         while (i < DPAA_SGT_MAX_ENTRIES) {
 333                 sg_temp = &sgt[i++];
 334                 hw_sg_to_cpu(sg_temp);
 335                 sg_vaddr = DPAA_MEMPOOL_PTOV(bp_info,
 336                                              qm_sg_entry_get64(sg_temp));
 337                 cur_seg = (struct rte_mbuf *)((char *)sg_vaddr -
 338                                                       bp_info->meta_data_size);
 339                 cur_seg->data_off = sg_temp->offset;
 340                 cur_seg->data_len = sg_temp->length;
 341                 first_seg->pkt_len += sg_temp->length;
 342                 first_seg->nb_segs += 1;
 343                 rte_mbuf_refcnt_set(cur_seg, 1);
 344                 prev_seg->next = cur_seg;
 345                 if (sg_temp->final) {
 346                         cur_seg->next = NULL;
 347                         break;
 348                 }
 349                 prev_seg = cur_seg;
 350         }
 351         DPAA_DP_LOG(DEBUG, "Received an SG frame len =%d, num_sg =%d",
 352                         first_seg->pkt_len, first_seg->nb_segs);
 353
 354         dpaa_eth_packet_info(first_seg, vaddr);
 355         rte_pktmbuf_free_seg(temp);
 356
 357         return first_seg;
 358 }
 359
 360 static inline struct rte_mbuf *
 361 dpaa_eth_fd_to_mbuf(const struct qm_fd *fd, uint32_t ifid)
 362 {
 363         struct rte_mbuf *mbuf;
 364         struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(fd->bpid);
 365         void *ptr;
 366         uint8_t format =
 367                 (fd->opaque & DPAA_FD_FORMAT_MASK) >> DPAA_FD_FORMAT_SHIFT;
 368         uint16_t offset;
 369         uint32_t length;
 370
 371         if (unlikely(format == qm_fd_sg))
 372                 return dpaa_eth_sg_to_mbuf(fd, ifid);
 373
 374         offset = (fd->opaque & DPAA_FD_OFFSET_MASK) >> DPAA_FD_OFFSET_SHIFT;
 375         length = fd->opaque & DPAA_FD_LENGTH_MASK;
 376
 377         DPAA_DP_LOG(DEBUG, " FD--->MBUF off %d len = %d", offset, length);
 378
 379         /* Ignoring case when format != qm_fd_contig */
 380         dpaa_display_frame(fd);
 381         ptr = DPAA_MEMPOOL_PTOV(bp_info, qm_fd_addr(fd));
 382
 383         mbuf = (struct rte_mbuf *)((char *)ptr - bp_info->meta_data_size);
 384         /* Prefetch the Parse results and packet data to L1 */
 385         rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
 386
 387         mbuf->data_off = offset;
 388         mbuf->data_len = length;
 389         mbuf->pkt_len = length;
 390
 391         mbuf->port = ifid;
 392         mbuf->nb_segs = 1;
 393         mbuf->ol_flags = 0;
 394         mbuf->next = NULL;
 395         rte_mbuf_refcnt_set(mbuf, 1);
 396         dpaa_eth_packet_info(mbuf, mbuf->buf_addr);
 397
 398         return mbuf;
 399 }
 400
 401 /* Specific for LS1043 */
 402 void
 403 dpaa_rx_cb_no_prefetch(struct qman_fq **fq, struct qm_dqrr_entry **dqrr,
 404            void **bufs, int num_bufs)
 405 {
 406         struct rte_mbuf *mbuf;
 407         struct dpaa_bp_info *bp_info;
 408         const struct qm_fd *fd;
 409         void *ptr;
 410         struct dpaa_if *dpaa_intf;
 411         uint16_t offset, i;
 412         uint32_t length;
 413         uint8_t format;
 414
 415         bp_info = DPAA_BPID_TO_POOL_INFO(dqrr[0]->fd.bpid);
 416         ptr = rte_dpaa_mem_ptov(qm_fd_addr(&dqrr[0]->fd));
 417         rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
 418         bufs[0] = (struct rte_mbuf *)((char *)ptr - bp_info->meta_data_size);
 419
 420         for (i = 0; i < num_bufs; i++) {
 421                 if (i < num_bufs - 1) {
 422                         bp_info = DPAA_BPID_TO_POOL_INFO(dqrr[i + 1]->fd.bpid);
 423                         ptr = rte_dpaa_mem_ptov(qm_fd_addr(&dqrr[i + 1]->fd));
 424                         rte_prefetch0((void *)((uint8_t *)ptr +
 425                                         DEFAULT_RX_ICEOF));
 426                         bufs[i + 1] = (struct rte_mbuf *)((char *)ptr -
 427                                         bp_info->meta_data_size);
 428                 }
 429
 430                 fd = &dqrr[i]->fd;
 431                 dpaa_intf = fq[0]->dpaa_intf;
 432
 433                 format = (fd->opaque & DPAA_FD_FORMAT_MASK) >>
 434                                 DPAA_FD_FORMAT_SHIFT;
 435                 if (unlikely(format == qm_fd_sg)) {
 436                         bufs[i] = dpaa_eth_sg_to_mbuf(fd, dpaa_intf->ifid);
 437                         continue;
 438                 }
 439
 440                 offset = (fd->opaque & DPAA_FD_OFFSET_MASK) >>
 441                                 DPAA_FD_OFFSET_SHIFT;
 442                 length = fd->opaque & DPAA_FD_LENGTH_MASK;
 443
 444                 mbuf = bufs[i];
 445                 mbuf->data_off = offset;
 446                 mbuf->data_len = length;
 447                 mbuf->pkt_len = length;
 448                 mbuf->port = dpaa_intf->ifid;
 449
 450                 mbuf->nb_segs = 1;
 451                 mbuf->ol_flags = 0;
 452                 mbuf->next = NULL;
 453                 rte_mbuf_refcnt_set(mbuf, 1);
 454                 dpaa_eth_packet_info(mbuf, mbuf->buf_addr);
 455         }
 456 }
 457
 458 void
 459 dpaa_rx_cb(struct qman_fq **fq, struct qm_dqrr_entry **dqrr,
 460            void **bufs, int num_bufs)
 461 {
 462         struct rte_mbuf *mbuf;
 463         const struct qm_fd *fd;
 464         struct dpaa_if *dpaa_intf;
 465         uint16_t offset, i;
 466         uint32_t length;
 467         uint8_t format;
 468
 469         for (i = 0; i < num_bufs; i++) {
 470                 fd = &dqrr[i]->fd;
 471                 dpaa_intf = fq[0]->dpaa_intf;
 472
 473                 format = (fd->opaque & DPAA_FD_FORMAT_MASK) >>
 474                                 DPAA_FD_FORMAT_SHIFT;
 475                 if (unlikely(format == qm_fd_sg)) {
 476                         bufs[i] = dpaa_eth_sg_to_mbuf(fd, dpaa_intf->ifid);
 477                         continue;
 478                 }
 479
 480                 offset = (fd->opaque & DPAA_FD_OFFSET_MASK) >>
 481                                 DPAA_FD_OFFSET_SHIFT;
 482                 length = fd->opaque & DPAA_FD_LENGTH_MASK;
 483
 484                 mbuf = bufs[i];
 485                 mbuf->data_off = offset;
 486                 mbuf->data_len = length;
 487                 mbuf->pkt_len = length;
 488                 mbuf->port = dpaa_intf->ifid;
 489
 490                 mbuf->nb_segs = 1;
 491                 mbuf->ol_flags = 0;
 492                 mbuf->next = NULL;
 493                 rte_mbuf_refcnt_set(mbuf, 1);
 494                 dpaa_eth_packet_info(mbuf, mbuf->buf_addr);
 495         }
 496 }
 497
 498 void dpaa_rx_cb_prepare(struct qm_dqrr_entry *dq, void **bufs)
 499 {
 500         struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(dq->fd.bpid);
 501         void *ptr = rte_dpaa_mem_ptov(qm_fd_addr(&dq->fd));
 502
 503         /* In case of LS1046, annotation stashing is disabled due to L2 cache
 504          * being bottleneck in case of multicore scanario for this platform.
 505          * So we prefetch the annoation beforehand, so that it is available
 506          * in cache when accessed.
 507          */
 508         rte_prefetch0((void *)((uint8_t *)ptr + DEFAULT_RX_ICEOF));
 509
 510         *bufs = (struct rte_mbuf *)((char *)ptr - bp_info->meta_data_size);
 511 }
 512
 513 static uint16_t
 514 dpaa_eth_queue_portal_rx(struct qman_fq *fq,
 515                          struct rte_mbuf **bufs,
 516                          uint16_t nb_bufs)
 517 {
 518         int ret;
 519
 520         if (unlikely(fq->qp == NULL)) {
 521                 ret = rte_dpaa_portal_fq_init((void *)0, fq);
 522                 if (ret) {
 523                         DPAA_PMD_ERR("Failure in affining portal %d", ret);
 524                         return 0;
 525                 }
 526         }
 527
 528         return qman_portal_poll_rx(nb_bufs, (void **)bufs, fq->qp);
 529 }
 530
 531 enum qman_cb_dqrr_result
 532 dpaa_rx_cb_parallel(void *event,
 533                     struct qman_portal *qm __always_unused,
 534                     struct qman_fq *fq,
 535                     const struct qm_dqrr_entry *dqrr,
 536                     void **bufs)
 537 {
 538         u32 ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
 539         struct rte_mbuf *mbuf;
 540         struct rte_event *ev = (struct rte_event *)event;
 541
 542         mbuf = dpaa_eth_fd_to_mbuf(&dqrr->fd, ifid);
 543         ev->event_ptr = (void *)mbuf;
 544         ev->flow_id = fq->ev.flow_id;
 545         ev->sub_event_type = fq->ev.sub_event_type;
 546         ev->event_type = RTE_EVENT_TYPE_ETHDEV;
 547         ev->op = RTE_EVENT_OP_NEW;
 548         ev->sched_type = fq->ev.sched_type;
 549         ev->queue_id = fq->ev.queue_id;
 550         ev->priority = fq->ev.priority;
 551         ev->impl_opaque = (uint8_t)DPAA_INVALID_MBUF_SEQN;
 552         mbuf->seqn = DPAA_INVALID_MBUF_SEQN;
 553         *bufs = mbuf;
 554
 555         return qman_cb_dqrr_consume;
 556 }
 557
 558 enum qman_cb_dqrr_result
 559 dpaa_rx_cb_atomic(void *event,
 560                   struct qman_portal *qm __always_unused,
 561                   struct qman_fq *fq,
 562                   const struct qm_dqrr_entry *dqrr,
 563                   void **bufs)
 564 {
 565         u8 index;
 566         u32 ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
 567         struct rte_mbuf *mbuf;
 568         struct rte_event *ev = (struct rte_event *)event;
 569
 570         mbuf = dpaa_eth_fd_to_mbuf(&dqrr->fd, ifid);
 571         ev->event_ptr = (void *)mbuf;
 572         ev->flow_id = fq->ev.flow_id;
 573         ev->sub_event_type = fq->ev.sub_event_type;
 574         ev->event_type = RTE_EVENT_TYPE_ETHDEV;
 575         ev->op = RTE_EVENT_OP_NEW;
 576         ev->sched_type = fq->ev.sched_type;
 577         ev->queue_id = fq->ev.queue_id;
 578         ev->priority = fq->ev.priority;
 579
 580         /* Save active dqrr entries */
 581         index = DQRR_PTR2IDX(dqrr);
 582         DPAA_PER_LCORE_DQRR_SIZE++;
 583         DPAA_PER_LCORE_DQRR_HELD |= 1 << index;
 584         DPAA_PER_LCORE_DQRR_MBUF(index) = mbuf;
 585         ev->impl_opaque = index + 1;
 586         mbuf->seqn = (uint32_t)index + 1;
 587         *bufs = mbuf;
 588
 589         return qman_cb_dqrr_defer;
 590 }
 591
 592 uint16_t dpaa_eth_queue_rx(void *q,
 593                            struct rte_mbuf **bufs,
 594                            uint16_t nb_bufs)
 595 {
 596         struct qman_fq *fq = q;
 597         struct qm_dqrr_entry *dq;
 598         uint32_t num_rx = 0, ifid = ((struct dpaa_if *)fq->dpaa_intf)->ifid;
 599         int num_rx_bufs, ret;
 600         uint32_t vdqcr_flags = 0;
 601
 602         if (unlikely(rte_dpaa_bpid_info == NULL &&
 603                                 rte_eal_process_type() == RTE_PROC_SECONDARY))
 604                 rte_dpaa_bpid_info = fq->bp_array;
 605
 606         if (likely(fq->is_static))
 607                 return dpaa_eth_queue_portal_rx(fq, bufs, nb_bufs);
 608
 609         if (unlikely(!RTE_PER_LCORE(dpaa_io))) {
 610                 ret = rte_dpaa_portal_init((void *)0);
 611                 if (ret) {
 612                         DPAA_PMD_ERR("Failure in affining portal");
 613                         return 0;
 614                 }
 615         }
 616
 617         /* Until request for four buffers, we provide exact number of buffers.
 618          * Otherwise we do not set the QM_VDQCR_EXACT flag.
 619          * Not setting QM_VDQCR_EXACT flag can provide two more buffers than
 620          * requested, so we request two less in this case.
 621          */
 622         if (nb_bufs < 4) {
 623                 vdqcr_flags = QM_VDQCR_EXACT;
 624                 num_rx_bufs = nb_bufs;
 625         } else {
 626                 num_rx_bufs = nb_bufs > DPAA_MAX_DEQUEUE_NUM_FRAMES ?
 627                         (DPAA_MAX_DEQUEUE_NUM_FRAMES - 2) : (nb_bufs - 2);
 628         }
 629         ret = qman_set_vdq(fq, num_rx_bufs, vdqcr_flags);
 630         if (ret)
 631                 return 0;
 632
 633         do {
 634                 dq = qman_dequeue(fq);
 635                 if (!dq)
 636                         continue;
 637                 bufs[num_rx++] = dpaa_eth_fd_to_mbuf(&dq->fd, ifid);
 638                 qman_dqrr_consume(fq, dq);
 639         } while (fq->flags & QMAN_FQ_STATE_VDQCR);
 640
 641         return num_rx;
 642 }
 643
 644 static void *dpaa_get_pktbuf(struct dpaa_bp_info *bp_info)
 645 {
 646         int ret;
 647         size_t buf = 0;
 648         struct bm_buffer bufs;
 649
 650         ret = bman_acquire(bp_info->bp, &bufs, 1, 0);
 651         if (ret <= 0) {
 652                 DPAA_PMD_WARN("Failed to allocate buffers %d", ret);
 653                 return (void *)buf;
 654         }
 655
 656         DPAA_DP_LOG(DEBUG, "got buffer 0x%" PRIx64 " from pool %d",
 657                     (uint64_t)bufs.addr, bufs.bpid);
 658
 659         buf = (size_t)DPAA_MEMPOOL_PTOV(bp_info, bufs.addr)
 660                                 - bp_info->meta_data_size;
 661         if (!buf)
 662                 goto out;
 663
 664 out:
 665         return (void *)buf;
 666 }
 667
 668 static struct rte_mbuf *dpaa_get_dmable_mbuf(struct rte_mbuf *mbuf,
 669                                              struct dpaa_if *dpaa_intf)
 670 {
 671         struct rte_mbuf *dpaa_mbuf;
 672
 673         /* allocate pktbuffer on bpid for dpaa port */
 674         dpaa_mbuf = dpaa_get_pktbuf(dpaa_intf->bp_info);
 675         if (!dpaa_mbuf)
 676                 return NULL;
 677
 678         memcpy((uint8_t *)(dpaa_mbuf->buf_addr) + RTE_PKTMBUF_HEADROOM, (void *)
 679                 ((uint8_t *)(mbuf->buf_addr) + mbuf->data_off), mbuf->pkt_len);
 680
 681         /* Copy only the required fields */
 682         dpaa_mbuf->data_off = RTE_PKTMBUF_HEADROOM;
 683         dpaa_mbuf->pkt_len = mbuf->pkt_len;
 684         dpaa_mbuf->ol_flags = mbuf->ol_flags;
 685         dpaa_mbuf->packet_type = mbuf->packet_type;
 686         dpaa_mbuf->tx_offload = mbuf->tx_offload;
 687         rte_pktmbuf_free(mbuf);
 688         return dpaa_mbuf;
 689 }
 690
 691 int
 692 dpaa_eth_mbuf_to_sg_fd(struct rte_mbuf *mbuf,
 693                 struct qm_fd *fd,
 694                 uint32_t bpid)
 695 {
 696         struct rte_mbuf *cur_seg = mbuf, *prev_seg = NULL;
 697         struct dpaa_bp_info *bp_info = DPAA_BPID_TO_POOL_INFO(bpid);
 698         struct rte_mbuf *temp, *mi;
 699         struct qm_sg_entry *sg_temp, *sgt;
 700         int i = 0;
 701
 702         DPAA_DP_LOG(DEBUG, "Creating SG FD to transmit");
 703
 704         temp = rte_pktmbuf_alloc(bp_info->mp);
 705         if (!temp) {
 706                 DPAA_PMD_ERR("Failure in allocation of mbuf");
 707                 return -1;
 708         }
 709         if (temp->buf_len < ((mbuf->nb_segs * sizeof(struct qm_sg_entry))
 710                                 + temp->data_off)) {
 711                 DPAA_PMD_ERR("Insufficient space in mbuf for SG entries");
 712                 return -1;
 713         }
 714
 715         fd->cmd = 0;
 716         fd->opaque_addr = 0;
 717
 718         if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK) {
 719                 if (!mbuf->packet_type) {
 720                         struct rte_net_hdr_lens hdr_lens;
 721
 722                         mbuf->packet_type = rte_net_get_ptype(mbuf, &hdr_lens,
 723                                         RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK
 724                                         | RTE_PTYPE_L4_MASK);
 725                         mbuf->l2_len = hdr_lens.l2_len;
 726                         mbuf->l3_len = hdr_lens.l3_len;
 727                 }
 728                 if (temp->data_off < DEFAULT_TX_ICEOF
 729                         + sizeof(struct dpaa_eth_parse_results_t))
 730                         temp->data_off = DEFAULT_TX_ICEOF
 731                                 + sizeof(struct dpaa_eth_parse_results_t);
 732                 dcbz_64(temp->buf_addr);
 733                 dpaa_checksum_offload(mbuf, fd, temp->buf_addr);
 734         }
 735
 736         sgt = temp->buf_addr + temp->data_off;
 737         fd->format = QM_FD_SG;
 738         fd->addr = temp->buf_iova;
 739         fd->offset = temp->data_off;
 740         fd->bpid = bpid;
 741         fd->length20 = mbuf->pkt_len;
 742
 743         while (i < DPAA_SGT_MAX_ENTRIES) {
 744                 sg_temp = &sgt[i++];
 745                 sg_temp->opaque = 0;
 746                 sg_temp->val = 0;
 747                 sg_temp->addr = cur_seg->buf_iova;
 748                 sg_temp->offset = cur_seg->data_off;
 749                 sg_temp->length = cur_seg->data_len;
 750                 if (RTE_MBUF_DIRECT(cur_seg)) {
 751                         if (rte_mbuf_refcnt_read(cur_seg) > 1) {
 752                                 /*If refcnt > 1, invalid bpid is set to ensure
 753                                  * buffer is not freed by HW.
 754                                  */
 755                                 sg_temp->bpid = 0xff;
 756                                 rte_mbuf_refcnt_update(cur_seg, -1);
 757                         } else {
 758                                 sg_temp->bpid =
 759                                         DPAA_MEMPOOL_TO_BPID(cur_seg->pool);
 760                         }
 761                         cur_seg = cur_seg->next;
 762                 } else {
 763                         /* Get owner MBUF from indirect buffer */
 764                         mi = rte_mbuf_from_indirect(cur_seg);
 765                         if (rte_mbuf_refcnt_read(mi) > 1) {
 766                                 /*If refcnt > 1, invalid bpid is set to ensure
 767                                  * owner buffer is not freed by HW.
 768                                  */
 769                                 sg_temp->bpid = 0xff;
 770                         } else {
 771                                 sg_temp->bpid = DPAA_MEMPOOL_TO_BPID(mi->pool);
 772                                 rte_mbuf_refcnt_update(mi, 1);
 773                         }
 774                         prev_seg = cur_seg;
 775                         cur_seg = cur_seg->next;
 776                         prev_seg->next = NULL;
 777                         rte_pktmbuf_free(prev_seg);
 778                 }
 779                 if (cur_seg == NULL) {
 780                         sg_temp->final = 1;
 781                         cpu_to_hw_sg(sg_temp);
 782                         break;
 783                 }
 784                 cpu_to_hw_sg(sg_temp);
 785         }
 786         return 0;
 787 }
 788
 789 /* Handle mbufs which are not segmented (non SG) */
 790 static inline void
 791 tx_on_dpaa_pool_unsegmented(struct rte_mbuf *mbuf,
 792                             struct dpaa_bp_info *bp_info,
 793                             struct qm_fd *fd_arr)
 794 {
 795         struct rte_mbuf *mi = NULL;
 796
 797         if (RTE_MBUF_DIRECT(mbuf)) {
 798                 if (rte_mbuf_refcnt_read(mbuf) > 1) {
 799                         /* In case of direct mbuf and mbuf being cloned,
 800                          * BMAN should _not_ release buffer.
 801                          */
 802                         DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, 0xff);
 803                         /* Buffer should be releasd by EAL */
 804                         rte_mbuf_refcnt_update(mbuf, -1);
 805                 } else {
 806                         /* In case of direct mbuf and no cloning, mbuf can be
 807                          * released by BMAN.
 808                          */
 809                         DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, bp_info->bpid);
 810                 }
 811         } else {
 812                 /* This is data-containing core mbuf: 'mi' */
 813                 mi = rte_mbuf_from_indirect(mbuf);
 814                 if (rte_mbuf_refcnt_read(mi) > 1) {
 815                         /* In case of indirect mbuf, and mbuf being cloned,
 816                          * BMAN should _not_ release it and let EAL release
 817                          * it through pktmbuf_free below.
 818                          */
 819                         DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, 0xff);
 820                 } else {
 821                         /* In case of indirect mbuf, and no cloning, core mbuf
 822                          * should be released by BMAN.
 823                          * Increate refcnt of core mbuf so that when
 824                          * pktmbuf_free is called and mbuf is released, EAL
 825                          * doesn't try to release core mbuf which would have
 826                          * been released by BMAN.
 827                          */
 828                         rte_mbuf_refcnt_update(mi, 1);
 829                         DPAA_MBUF_TO_CONTIG_FD(mbuf, fd_arr, bp_info->bpid);
 830                 }
 831                 rte_pktmbuf_free(mbuf);
 832         }
 833
 834         if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK)
 835                 dpaa_unsegmented_checksum(mbuf, fd_arr);
 836 }
 837
 838 /* Handle all mbufs on dpaa BMAN managed pool */
 839 static inline uint16_t
 840 tx_on_dpaa_pool(struct rte_mbuf *mbuf,
 841                 struct dpaa_bp_info *bp_info,
 842                 struct qm_fd *fd_arr)
 843 {
 844         DPAA_DP_LOG(DEBUG, "BMAN offloaded buffer, mbuf: %p", mbuf);
 845
 846         if (mbuf->nb_segs == 1) {
 847                 /* Case for non-segmented buffers */
 848                 tx_on_dpaa_pool_unsegmented(mbuf, bp_info, fd_arr);
 849         } else if (mbuf->nb_segs > 1 &&
 850                    mbuf->nb_segs <= DPAA_SGT_MAX_ENTRIES) {
 851                 if (dpaa_eth_mbuf_to_sg_fd(mbuf, fd_arr, bp_info->bpid)) {
 852                         DPAA_PMD_DEBUG("Unable to create Scatter Gather FD");
 853                         return 1;
 854                 }
 855         } else {
 856                 DPAA_PMD_DEBUG("Number of Segments not supported");
 857                 return 1;
 858         }
 859
 860         return 0;
 861 }
 862
 863 /* Handle all mbufs on an external pool (non-dpaa) */
 864 static inline uint16_t
 865 tx_on_external_pool(struct qman_fq *txq, struct rte_mbuf *mbuf,
 866                     struct qm_fd *fd_arr)
 867 {
 868         struct dpaa_if *dpaa_intf = txq->dpaa_intf;
 869         struct rte_mbuf *dmable_mbuf;
 870
 871         DPAA_DP_LOG(DEBUG, "Non-BMAN offloaded buffer."
 872                     "Allocating an offloaded buffer");
 873         dmable_mbuf = dpaa_get_dmable_mbuf(mbuf, dpaa_intf);
 874         if (!dmable_mbuf) {
 875                 DPAA_DP_LOG(DEBUG, "no dpaa buffers.");
 876                 return 1;
 877         }
 878
 879         DPAA_MBUF_TO_CONTIG_FD(dmable_mbuf, fd_arr, dpaa_intf->bp_info->bpid);
 880         if (mbuf->ol_flags & DPAA_TX_CKSUM_OFFLOAD_MASK)
 881                 dpaa_unsegmented_checksum(mbuf, fd_arr);
 882
 883         return 0;
 884 }
 885
 886 uint16_t
 887 dpaa_eth_queue_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
 888 {
 889         struct rte_mbuf *mbuf, *mi = NULL;
 890         struct rte_mempool *mp;
 891         struct dpaa_bp_info *bp_info;
 892         struct qm_fd fd_arr[DPAA_TX_BURST_SIZE];
 893         uint32_t frames_to_send, loop, sent = 0;
 894         uint16_t state;
 895         int ret;
 896         uint32_t seqn, index, flags[DPAA_TX_BURST_SIZE] = {0};
 897
 898         if (unlikely(!RTE_PER_LCORE(dpaa_io))) {
 899                 ret = rte_dpaa_portal_init((void *)0);
 900                 if (ret) {
 901                         DPAA_PMD_ERR("Failure in affining portal");
 902                         return 0;
 903                 }
 904         }
 905
 906         DPAA_DP_LOG(DEBUG, "Transmitting %d buffers on queue: %p", nb_bufs, q);
 907
 908         while (nb_bufs) {
 909                 frames_to_send = (nb_bufs > DPAA_TX_BURST_SIZE) ?
 910                                 DPAA_TX_BURST_SIZE : nb_bufs;
 911                 for (loop = 0; loop < frames_to_send; loop++) {
 912                         mbuf = *(bufs++);
 913                         seqn = mbuf->seqn;
 914                         if (seqn != DPAA_INVALID_MBUF_SEQN) {
 915                                 index = seqn - 1;
 916                                 if (DPAA_PER_LCORE_DQRR_HELD & (1 << index)) {
 917                                         flags[loop] =
 918                                            ((index & QM_EQCR_DCA_IDXMASK) << 8);
 919                                         flags[loop] |= QMAN_ENQUEUE_FLAG_DCA;
 920                                         DPAA_PER_LCORE_DQRR_SIZE--;
 921                                         DPAA_PER_LCORE_DQRR_HELD &=
 922                                                                 ~(1 << index);
 923                                 }
 924                         }
 925
 926                         if (likely(RTE_MBUF_DIRECT(mbuf))) {
 927                                 mp = mbuf->pool;
 928                                 bp_info = DPAA_MEMPOOL_TO_POOL_INFO(mp);
 929                                 if (likely(mp->ops_index ==
 930                                                 bp_info->dpaa_ops_index &&
 931                                         mbuf->nb_segs == 1 &&
 932                                         rte_mbuf_refcnt_read(mbuf) == 1)) {
 933                                         DPAA_MBUF_TO_CONTIG_FD(mbuf,
 934                                                 &fd_arr[loop], bp_info->bpid);
 935                                         if (mbuf->ol_flags &
 936                                                 DPAA_TX_CKSUM_OFFLOAD_MASK)
 937                                                 dpaa_unsegmented_checksum(mbuf,
 938                                                         &fd_arr[loop]);
 939                                         continue;
 940                                 }
 941                         } else {
 942                                 mi = rte_mbuf_from_indirect(mbuf);
 943                                 mp = mi->pool;
 944                         }
 945
 946                         bp_info = DPAA_MEMPOOL_TO_POOL_INFO(mp);
 947                         if (likely(mp->ops_index == bp_info->dpaa_ops_index)) {
 948                                 state = tx_on_dpaa_pool(mbuf, bp_info,
 949                                                         &fd_arr[loop]);
 950                                 if (unlikely(state)) {
 951                                         /* Set frames_to_send & nb_bufs so
 952                                          * that packets are transmitted till
 953                                          * previous frame.
 954                                          */
 955                                         frames_to_send = loop;
 956                                         nb_bufs = loop;
 957                                         goto send_pkts;
 958                                 }
 959                         } else {
 960                                 state = tx_on_external_pool(q, mbuf,
 961                                                             &fd_arr[loop]);
 962                                 if (unlikely(state)) {
 963                                         /* Set frames_to_send & nb_bufs so
 964                                          * that packets are transmitted till
 965                                          * previous frame.
 966                                          */
 967                                         frames_to_send = loop;
 968                                         nb_bufs = loop;
 969                                         goto send_pkts;
 970                                 }
 971                         }
 972                 }
 973
 974 send_pkts:
 975                 loop = 0;
 976                 while (loop < frames_to_send) {
 977                         loop += qman_enqueue_multi(q, &fd_arr[loop],
 978                                                    &flags[loop],
 979                                                    frames_to_send - loop);
 980                 }
 981                 nb_bufs -= frames_to_send;
 982                 sent += frames_to_send;
 983         }
 984
 985         DPAA_DP_LOG(DEBUG, "Transmitted %d buffers on queue: %p", sent, q);
 986
 987         return sent;
 988 }
 989
 990 uint16_t dpaa_eth_tx_drop_all(void *q  __rte_unused,
 991                               struct rte_mbuf **bufs __rte_unused,
 992                 uint16_t nb_bufs __rte_unused)
 993 {
 994         DPAA_DP_LOG(DEBUG, "Drop all packets");
 995
 996         /* Drop all incoming packets. No need to free packets here
 997          * because the rte_eth f/w frees up the packets through tx_buffer
 998          * callback in case this functions returns count less than nb_bufs
 999          */
1000         return 0;
1001 }