drivers/net/cxgbe/cxgbe_main.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2014-2018 Chelsio Communications.
   3  * All rights reserved.
   4  */
   5
   6 #include <sys/queue.h>
   7 #include <stdio.h>
   8 #include <errno.h>
   9 #include <stdint.h>
  10 #include <string.h>
  11 #include <unistd.h>
  12 #include <stdarg.h>
  13 #include <inttypes.h>
  14 #include <netinet/in.h>
  15
  16 #include <rte_byteorder.h>
  17 #include <rte_common.h>
  18 #include <rte_cycles.h>
  19 #include <rte_interrupts.h>
  20 #include <rte_log.h>
  21 #include <rte_debug.h>
  22 #include <rte_pci.h>
  23 #include <rte_atomic.h>
  24 #include <rte_branch_prediction.h>
  25 #include <rte_memory.h>
  26 #include <rte_tailq.h>
  27 #include <rte_eal.h>
  28 #include <rte_alarm.h>
  29 #include <rte_ether.h>
  30 #include <rte_ethdev_driver.h>
  31 #include <rte_ethdev_pci.h>
  32 #include <rte_random.h>
  33 #include <rte_dev.h>
  34 #include <rte_kvargs.h>
  35
  36 #include "base/common.h"
  37 #include "base/t4_regs.h"
  38 #include "base/t4_msg.h"
  39 #include "cxgbe.h"
  40 #include "cxgbe_pfvf.h"
  41 #include "clip_tbl.h"
  42 #include "l2t.h"
  43 #include "mps_tcam.h"
  44
  45 /**
  46  * Allocate a chunk of memory. The allocated memory is cleared.
  47  */
  48 void *t4_alloc_mem(size_t size)
  49 {
  50         return rte_zmalloc(NULL, size, 0);
  51 }
  52
  53 /**
  54  * Free memory allocated through t4_alloc_mem().
  55  */
  56 void t4_free_mem(void *addr)
  57 {
  58         rte_free(addr);
  59 }
  60
  61 /*
  62  * Response queue handler for the FW event queue.
  63  */
  64 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
  65                           __rte_unused const struct pkt_gl *gl)
  66 {
  67         u8 opcode = ((const struct rss_header *)rsp)->opcode;
  68
  69         rsp++;                                          /* skip RSS header */
  70
  71         /*
  72          * FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
  73          */
  74         if (unlikely(opcode == CPL_FW4_MSG &&
  75                      ((const struct cpl_fw4_msg *)rsp)->type ==
  76                       FW_TYPE_RSSCPL)) {
  77                 rsp++;
  78                 opcode = ((const struct rss_header *)rsp)->opcode;
  79                 rsp++;
  80                 if (opcode != CPL_SGE_EGR_UPDATE) {
  81                         dev_err(q->adapter, "unexpected FW4/CPL %#x on FW event queue\n",
  82                                 opcode);
  83                         goto out;
  84                 }
  85         }
  86
  87         if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
  88                 /* do nothing */
  89         } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
  90                 const struct cpl_fw6_msg *msg = (const void *)rsp;
  91
  92                 t4_handle_fw_rpl(q->adapter, msg->data);
  93         } else if (opcode == CPL_ABORT_RPL_RSS) {
  94                 const struct cpl_abort_rpl_rss *p = (const void *)rsp;
  95
  96                 cxgbe_hash_del_filter_rpl(q->adapter, p);
  97         } else if (opcode == CPL_SET_TCB_RPL) {
  98                 const struct cpl_set_tcb_rpl *p = (const void *)rsp;
  99
 100                 cxgbe_filter_rpl(q->adapter, p);
 101         } else if (opcode == CPL_ACT_OPEN_RPL) {
 102                 const struct cpl_act_open_rpl *p = (const void *)rsp;
 103
 104                 cxgbe_hash_filter_rpl(q->adapter, p);
 105         } else if (opcode == CPL_L2T_WRITE_RPL) {
 106                 const struct cpl_l2t_write_rpl *p = (const void *)rsp;
 107
 108                 cxgbe_do_l2t_write_rpl(q->adapter, p);
 109         } else {
 110                 dev_err(adapter, "unexpected CPL %#x on FW event queue\n",
 111                         opcode);
 112         }
 113 out:
 114         return 0;
 115 }
 116
 117 /**
 118  * Setup sge control queues to pass control information.
 119  */
 120 int cxgbe_setup_sge_ctrl_txq(struct adapter *adapter)
 121 {
 122         struct sge *s = &adapter->sge;
 123         int err = 0, i = 0;
 124
 125         for_each_port(adapter, i) {
 126                 struct port_info *pi = adap2pinfo(adapter, i);
 127                 char name[RTE_ETH_NAME_MAX_LEN];
 128                 struct sge_ctrl_txq *q = &s->ctrlq[i];
 129
 130                 q->q.size = 1024;
 131                 err = t4_sge_alloc_ctrl_txq(adapter, q,
 132                                             adapter->eth_dev,  i,
 133                                             s->fw_evtq.cntxt_id,
 134                                             rte_socket_id());
 135                 if (err) {
 136                         dev_err(adapter, "Failed to alloc ctrl txq. Err: %d",
 137                                 err);
 138                         goto out;
 139                 }
 140                 snprintf(name, sizeof(name), "%s_ctrl_pool_%d",
 141                          pi->eth_dev->device->driver->name,
 142                          pi->eth_dev->data->port_id);
 143                 q->mb_pool = rte_pktmbuf_pool_create(name, s->ctrlq[i].q.size,
 144                                                      RTE_CACHE_LINE_SIZE,
 145                                                      RTE_MBUF_PRIV_ALIGN,
 146                                                      RTE_MBUF_DEFAULT_BUF_SIZE,
 147                                                      SOCKET_ID_ANY);
 148                 if (!q->mb_pool) {
 149                         err = -rte_errno;
 150                         dev_err(adapter,
 151                                 "Can't create ctrl pool for port %d. Err: %d\n",
 152                                 pi->eth_dev->data->port_id, err);
 153                         goto out;
 154                 }
 155         }
 156         return 0;
 157 out:
 158         t4_free_sge_resources(adapter);
 159         return err;
 160 }
 161
 162 /**
 163  * cxgbe_poll_for_completion: Poll rxq for completion
 164  * @q: rxq to poll
 165  * @ms: milliseconds to delay
 166  * @cnt: number of times to poll
 167  * @c: completion to check for 'done' status
 168  *
 169  * Polls the rxq for reples until completion is done or the count
 170  * expires.
 171  */
 172 int cxgbe_poll_for_completion(struct sge_rspq *q, unsigned int ms,
 173                               unsigned int cnt, struct t4_completion *c)
 174 {
 175         unsigned int i;
 176         unsigned int work_done, budget = 32;
 177
 178         if (!c)
 179                 return -EINVAL;
 180
 181         for (i = 0; i < cnt; i++) {
 182                 cxgbe_poll(q, NULL, budget, &work_done);
 183                 t4_os_lock(&c->lock);
 184                 if (c->done) {
 185                         t4_os_unlock(&c->lock);
 186                         return 0;
 187                 }
 188                 t4_os_unlock(&c->lock);
 189                 rte_delay_ms(ms);
 190         }
 191         return -ETIMEDOUT;
 192 }
 193
 194 int cxgbe_setup_sge_fwevtq(struct adapter *adapter)
 195 {
 196         struct sge *s = &adapter->sge;
 197         int err = 0;
 198         int msi_idx = 0;
 199
 200         err = t4_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->eth_dev,
 201                                msi_idx, NULL, fwevtq_handler, -1, NULL, 0,
 202                                rte_socket_id());
 203         return err;
 204 }
 205
 206 static int closest_timer(const struct sge *s, int time)
 207 {
 208         unsigned int i, match = 0;
 209         int delta, min_delta = INT_MAX;
 210
 211         for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
 212                 delta = time - s->timer_val[i];
 213                 if (delta < 0)
 214                         delta = -delta;
 215                 if (delta < min_delta) {
 216                         min_delta = delta;
 217                         match = i;
 218                 }
 219         }
 220         return match;
 221 }
 222
 223 static int closest_thres(const struct sge *s, int thres)
 224 {
 225         unsigned int i, match = 0;
 226         int delta, min_delta = INT_MAX;
 227
 228         for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
 229                 delta = thres - s->counter_val[i];
 230                 if (delta < 0)
 231                         delta = -delta;
 232                 if (delta < min_delta) {
 233                         min_delta = delta;
 234                         match = i;
 235                 }
 236         }
 237         return match;
 238 }
 239
 240 /**
 241  * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
 242  * @q: the Rx queue
 243  * @us: the hold-off time in us, or 0 to disable timer
 244  * @cnt: the hold-off packet count, or 0 to disable counter
 245  *
 246  * Sets an Rx queue's interrupt hold-off time and packet count.  At least
 247  * one of the two needs to be enabled for the queue to generate interrupts.
 248  */
 249 int cxgb4_set_rspq_intr_params(struct sge_rspq *q, unsigned int us,
 250                                unsigned int cnt)
 251 {
 252         struct adapter *adap = q->adapter;
 253         unsigned int timer_val;
 254
 255         if (cnt) {
 256                 int err;
 257                 u32 v, new_idx;
 258
 259                 new_idx = closest_thres(&adap->sge, cnt);
 260                 if (q->desc && q->pktcnt_idx != new_idx) {
 261                         /* the queue has already been created, update it */
 262                         v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
 263                             V_FW_PARAMS_PARAM_X(
 264                             FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
 265                             V_FW_PARAMS_PARAM_YZ(q->cntxt_id);
 266                         err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
 267                                             &v, &new_idx);
 268                         if (err)
 269                                 return err;
 270                 }
 271                 q->pktcnt_idx = new_idx;
 272         }
 273
 274         timer_val = (us == 0) ? X_TIMERREG_RESTART_COUNTER :
 275                                 closest_timer(&adap->sge, us);
 276
 277         if ((us | cnt) == 0)
 278                 q->intr_params = V_QINTR_TIMER_IDX(X_TIMERREG_UPDATE_CIDX);
 279         else
 280                 q->intr_params = V_QINTR_TIMER_IDX(timer_val) |
 281                                  V_QINTR_CNT_EN(cnt > 0);
 282         return 0;
 283 }
 284
 285 /**
 286  * Allocate an active-open TID and set it to the supplied value.
 287  */
 288 int cxgbe_alloc_atid(struct tid_info *t, void *data)
 289 {
 290         int atid = -1;
 291
 292         t4_os_lock(&t->atid_lock);
 293         if (t->afree) {
 294                 union aopen_entry *p = t->afree;
 295
 296                 atid = p - t->atid_tab;
 297                 t->afree = p->next;
 298                 p->data = data;
 299                 t->atids_in_use++;
 300         }
 301         t4_os_unlock(&t->atid_lock);
 302         return atid;
 303 }
 304
 305 /**
 306  * Release an active-open TID.
 307  */
 308 void cxgbe_free_atid(struct tid_info *t, unsigned int atid)
 309 {
 310         union aopen_entry *p = &t->atid_tab[atid];
 311
 312         t4_os_lock(&t->atid_lock);
 313         p->next = t->afree;
 314         t->afree = p;
 315         t->atids_in_use--;
 316         t4_os_unlock(&t->atid_lock);
 317 }
 318
 319 /**
 320  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
 321  */
 322 static void mk_tid_release(struct rte_mbuf *mbuf, unsigned int tid)
 323 {
 324         struct cpl_tid_release *req;
 325
 326         req = rte_pktmbuf_mtod(mbuf, struct cpl_tid_release *);
 327         INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
 328 }
 329
 330 /**
 331  * Release a TID and inform HW.  If we are unable to allocate the release
 332  * message we defer to a work queue.
 333  */
 334 void cxgbe_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
 335                       unsigned short family)
 336 {
 337         struct rte_mbuf *mbuf;
 338         struct adapter *adap = container_of(t, struct adapter, tids);
 339
 340         WARN_ON(tid >= t->ntids);
 341
 342         if (t->tid_tab[tid]) {
 343                 t->tid_tab[tid] = NULL;
 344                 rte_atomic32_dec(&t->conns_in_use);
 345                 if (t->hash_base && tid >= t->hash_base) {
 346                         if (family == FILTER_TYPE_IPV4)
 347                                 rte_atomic32_dec(&t->hash_tids_in_use);
 348                 } else {
 349                         if (family == FILTER_TYPE_IPV4)
 350                                 rte_atomic32_dec(&t->tids_in_use);
 351                 }
 352         }
 353
 354         mbuf = rte_pktmbuf_alloc((&adap->sge.ctrlq[chan])->mb_pool);
 355         if (mbuf) {
 356                 mbuf->data_len = sizeof(struct cpl_tid_release);
 357                 mbuf->pkt_len = mbuf->data_len;
 358                 mk_tid_release(mbuf, tid);
 359                 t4_mgmt_tx(&adap->sge.ctrlq[chan], mbuf);
 360         }
 361 }
 362
 363 /**
 364  * Insert a TID.
 365  */
 366 void cxgbe_insert_tid(struct tid_info *t, void *data, unsigned int tid,
 367                       unsigned short family)
 368 {
 369         t->tid_tab[tid] = data;
 370         if (t->hash_base && tid >= t->hash_base) {
 371                 if (family == FILTER_TYPE_IPV4)
 372                         rte_atomic32_inc(&t->hash_tids_in_use);
 373         } else {
 374                 if (family == FILTER_TYPE_IPV4)
 375                         rte_atomic32_inc(&t->tids_in_use);
 376         }
 377
 378         rte_atomic32_inc(&t->conns_in_use);
 379 }
 380
 381 /**
 382  * Free TID tables.
 383  */
 384 static void tid_free(struct tid_info *t)
 385 {
 386         if (t->tid_tab) {
 387                 if (t->ftid_bmap)
 388                         rte_bitmap_free(t->ftid_bmap);
 389
 390                 if (t->ftid_bmap_array)
 391                         t4_os_free(t->ftid_bmap_array);
 392
 393                 t4_os_free(t->tid_tab);
 394         }
 395
 396         memset(t, 0, sizeof(struct tid_info));
 397 }
 398
 399 /**
 400  * Allocate and initialize the TID tables.  Returns 0 on success.
 401  */
 402 static int tid_init(struct tid_info *t)
 403 {
 404         size_t size;
 405         unsigned int ftid_bmap_size;
 406         unsigned int natids = t->natids;
 407         unsigned int max_ftids = t->nftids;
 408
 409         ftid_bmap_size = rte_bitmap_get_memory_footprint(t->nftids);
 410         size = t->ntids * sizeof(*t->tid_tab) +
 411                 max_ftids * sizeof(*t->ftid_tab) +
 412                 natids * sizeof(*t->atid_tab);
 413
 414         t->tid_tab = t4_os_alloc(size);
 415         if (!t->tid_tab)
 416                 return -ENOMEM;
 417
 418         t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
 419         t->ftid_tab = (struct filter_entry *)&t->atid_tab[t->natids];
 420         t->ftid_bmap_array = t4_os_alloc(ftid_bmap_size);
 421         if (!t->ftid_bmap_array) {
 422                 tid_free(t);
 423                 return -ENOMEM;
 424         }
 425
 426         t4_os_lock_init(&t->atid_lock);
 427         t4_os_lock_init(&t->ftid_lock);
 428
 429         t->afree = NULL;
 430         t->atids_in_use = 0;
 431         rte_atomic32_init(&t->tids_in_use);
 432         rte_atomic32_set(&t->tids_in_use, 0);
 433         rte_atomic32_init(&t->conns_in_use);
 434         rte_atomic32_set(&t->conns_in_use, 0);
 435
 436         /* Setup the free list for atid_tab and clear the stid bitmap. */
 437         if (natids) {
 438                 while (--natids)
 439                         t->atid_tab[natids - 1].next = &t->atid_tab[natids];
 440                 t->afree = t->atid_tab;
 441         }
 442
 443         t->ftid_bmap = rte_bitmap_init(t->nftids, t->ftid_bmap_array,
 444                                        ftid_bmap_size);
 445         if (!t->ftid_bmap) {
 446                 tid_free(t);
 447                 return -ENOMEM;
 448         }
 449
 450         return 0;
 451 }
 452
 453 static inline bool is_x_1g_port(const struct link_config *lc)
 454 {
 455         return (lc->pcaps & FW_PORT_CAP32_SPEED_1G) != 0;
 456 }
 457
 458 static inline bool is_x_10g_port(const struct link_config *lc)
 459 {
 460         unsigned int speeds, high_speeds;
 461
 462         speeds = V_FW_PORT_CAP32_SPEED(G_FW_PORT_CAP32_SPEED(lc->pcaps));
 463         high_speeds = speeds &
 464                       ~(FW_PORT_CAP32_SPEED_100M | FW_PORT_CAP32_SPEED_1G);
 465
 466         return high_speeds != 0;
 467 }
 468
 469 static inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
 470                       unsigned int us, unsigned int cnt,
 471                       unsigned int size, unsigned int iqe_size)
 472 {
 473         q->adapter = adap;
 474         cxgb4_set_rspq_intr_params(q, us, cnt);
 475         q->iqe_len = iqe_size;
 476         q->size = size;
 477 }
 478
 479 int cxgbe_cfg_queue_count(struct rte_eth_dev *eth_dev)
 480 {
 481         struct port_info *pi = eth_dev->data->dev_private;
 482         struct adapter *adap = pi->adapter;
 483         struct sge *s = &adap->sge;
 484         unsigned int max_queues = s->max_ethqsets / adap->params.nports;
 485
 486         if ((eth_dev->data->nb_rx_queues < 1) ||
 487             (eth_dev->data->nb_tx_queues < 1))
 488                 return -EINVAL;
 489
 490         if ((eth_dev->data->nb_rx_queues > max_queues) ||
 491             (eth_dev->data->nb_tx_queues > max_queues))
 492                 return -EINVAL;
 493
 494         if (eth_dev->data->nb_rx_queues > pi->rss_size)
 495                 return -EINVAL;
 496
 497         /* We must configure RSS, since config has changed*/
 498         pi->flags &= ~PORT_RSS_DONE;
 499
 500         pi->n_rx_qsets = eth_dev->data->nb_rx_queues;
 501         pi->n_tx_qsets = eth_dev->data->nb_tx_queues;
 502
 503         return 0;
 504 }
 505
 506 void cxgbe_cfg_queues(struct rte_eth_dev *eth_dev)
 507 {
 508         struct port_info *pi = eth_dev->data->dev_private;
 509         struct adapter *adap = pi->adapter;
 510         struct sge *s = &adap->sge;
 511         unsigned int i, nb_ports = 0, qidx = 0;
 512         unsigned int q_per_port = 0;
 513
 514         if (!(adap->flags & CFG_QUEUES)) {
 515                 for_each_port(adap, i) {
 516                         struct port_info *tpi = adap2pinfo(adap, i);
 517
 518                         nb_ports += (is_x_10g_port(&tpi->link_cfg)) ||
 519                                      is_x_1g_port(&tpi->link_cfg) ? 1 : 0;
 520                 }
 521
 522                 /*
 523                  * We default up to # of cores queues per 1G/10G port.
 524                  */
 525                 if (nb_ports)
 526                         q_per_port = (s->max_ethqsets -
 527                                      (adap->params.nports - nb_ports)) /
 528                                      nb_ports;
 529
 530                 if (q_per_port > rte_lcore_count())
 531                         q_per_port = rte_lcore_count();
 532
 533                 for_each_port(adap, i) {
 534                         struct port_info *pi = adap2pinfo(adap, i);
 535
 536                         pi->first_qset = qidx;
 537
 538                         /* Initially n_rx_qsets == n_tx_qsets */
 539                         pi->n_rx_qsets = (is_x_10g_port(&pi->link_cfg) ||
 540                                           is_x_1g_port(&pi->link_cfg)) ?
 541                                           q_per_port : 1;
 542                         pi->n_tx_qsets = pi->n_rx_qsets;
 543
 544                         if (pi->n_rx_qsets > pi->rss_size)
 545                                 pi->n_rx_qsets = pi->rss_size;
 546
 547                         qidx += pi->n_rx_qsets;
 548                 }
 549
 550                 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
 551                         struct sge_eth_rxq *r = &s->ethrxq[i];
 552
 553                         init_rspq(adap, &r->rspq, 5, 32, 1024, 64);
 554                         r->usembufs = 1;
 555                         r->fl.size = (r->usembufs ? 1024 : 72);
 556                 }
 557
 558                 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
 559                         s->ethtxq[i].q.size = 1024;
 560
 561                 init_rspq(adap, &adap->sge.fw_evtq, 0, 0, 1024, 64);
 562                 adap->flags |= CFG_QUEUES;
 563         }
 564 }
 565
 566 void cxgbe_stats_get(struct port_info *pi, struct port_stats *stats)
 567 {
 568         t4_get_port_stats_offset(pi->adapter, pi->tx_chan, stats,
 569                                  &pi->stats_base);
 570 }
 571
 572 void cxgbe_stats_reset(struct port_info *pi)
 573 {
 574         t4_clr_port_stats(pi->adapter, pi->tx_chan);
 575 }
 576
 577 static void setup_memwin(struct adapter *adap)
 578 {
 579         u32 mem_win0_base;
 580
 581         /* For T5, only relative offset inside the PCIe BAR is passed */
 582         mem_win0_base = MEMWIN0_BASE;
 583
 584         /*
 585          * Set up memory window for accessing adapter memory ranges.  (Read
 586          * back MA register to ensure that changes propagate before we attempt
 587          * to use the new values.)
 588          */
 589         t4_write_reg(adap,
 590                      PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
 591                                          MEMWIN_NIC),
 592                      mem_win0_base | V_BIR(0) |
 593                      V_WINDOW(ilog2(MEMWIN0_APERTURE) - X_WINDOW_SHIFT));
 594         t4_read_reg(adap,
 595                     PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
 596                                         MEMWIN_NIC));
 597 }
 598
 599 int cxgbe_init_rss(struct adapter *adap)
 600 {
 601         unsigned int i;
 602
 603         if (is_pf4(adap)) {
 604                 int err;
 605
 606                 err = t4_init_rss_mode(adap, adap->mbox);
 607                 if (err)
 608                         return err;
 609         }
 610
 611         for_each_port(adap, i) {
 612                 struct port_info *pi = adap2pinfo(adap, i);
 613
 614                 pi->rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
 615                 if (!pi->rss)
 616                         return -ENOMEM;
 617
 618                 pi->rss_hf = CXGBE_RSS_HF_ALL;
 619         }
 620         return 0;
 621 }
 622
 623 /**
 624  * Dump basic information about the adapter.
 625  */
 626 void cxgbe_print_adapter_info(struct adapter *adap)
 627 {
 628         /**
 629          * Hardware/Firmware/etc. Version/Revision IDs.
 630          */
 631         t4_dump_version_info(adap);
 632 }
 633
 634 void cxgbe_print_port_info(struct adapter *adap)
 635 {
 636         int i;
 637         char buf[80];
 638         struct rte_pci_addr *loc = &adap->pdev->addr;
 639
 640         for_each_port(adap, i) {
 641                 const struct port_info *pi = adap2pinfo(adap, i);
 642                 char *bufp = buf;
 643
 644                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
 645                         bufp += sprintf(bufp, "100M/");
 646                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
 647                         bufp += sprintf(bufp, "1G/");
 648                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
 649                         bufp += sprintf(bufp, "10G/");
 650                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
 651                         bufp += sprintf(bufp, "25G/");
 652                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
 653                         bufp += sprintf(bufp, "40G/");
 654                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
 655                         bufp += sprintf(bufp, "50G/");
 656                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
 657                         bufp += sprintf(bufp, "100G/");
 658                 if (bufp != buf)
 659                         --bufp;
 660                 sprintf(bufp, "BASE-%s",
 661                         t4_get_port_type_description(
 662                                         (enum fw_port_type)pi->port_type));
 663
 664                 dev_info(adap,
 665                          " " PCI_PRI_FMT " Chelsio rev %d %s %s\n",
 666                          loc->domain, loc->bus, loc->devid, loc->function,
 667                          CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
 668                          (adap->flags & USING_MSIX) ? " MSI-X" :
 669                          (adap->flags & USING_MSI) ? " MSI" : "");
 670         }
 671 }
 672
 673 static int check_devargs_handler(const char *key, const char *value, void *p)
 674 {
 675         if (!strncmp(key, CXGBE_DEVARG_CMN_KEEP_OVLAN, strlen(key)) ||
 676             !strncmp(key, CXGBE_DEVARG_CMN_TX_MODE_LATENCY, strlen(key)) ||
 677             !strncmp(key, CXGBE_DEVARG_VF_FORCE_LINK_UP, strlen(key))) {
 678                 if (!strncmp(value, "1", 1)) {
 679                         bool *dst_val = (bool *)p;
 680
 681                         *dst_val = true;
 682                 }
 683         }
 684
 685         return 0;
 686 }
 687
 688 static int cxgbe_get_devargs(struct rte_devargs *devargs, const char *key,
 689                              void *p)
 690 {
 691         struct rte_kvargs *kvlist;
 692         int ret = 0;
 693
 694         if (!devargs)
 695                 return 0;
 696
 697         kvlist = rte_kvargs_parse(devargs->args, NULL);
 698         if (!kvlist)
 699                 return 0;
 700
 701         if (!rte_kvargs_count(kvlist, key))
 702                 goto out;
 703
 704         ret = rte_kvargs_process(kvlist, key, check_devargs_handler, p);
 705
 706 out:
 707         rte_kvargs_free(kvlist);
 708
 709         return ret;
 710 }
 711
 712 static void cxgbe_get_devargs_int(struct adapter *adap, bool *dst,
 713                                   const char *key, bool default_value)
 714 {
 715         struct rte_pci_device *pdev = adap->pdev;
 716         int ret;
 717         bool devarg_value = default_value;
 718
 719         *dst = default_value;
 720         if (!pdev)
 721                 return;
 722
 723         ret = cxgbe_get_devargs(pdev->device.devargs, key, &devarg_value);
 724         if (ret)
 725                 return;
 726
 727         *dst = devarg_value;
 728 }
 729
 730 void cxgbe_process_devargs(struct adapter *adap)
 731 {
 732         cxgbe_get_devargs_int(adap, &adap->devargs.keep_ovlan,
 733                               CXGBE_DEVARG_CMN_KEEP_OVLAN, false);
 734         cxgbe_get_devargs_int(adap, &adap->devargs.tx_mode_latency,
 735                               CXGBE_DEVARG_CMN_TX_MODE_LATENCY, false);
 736         cxgbe_get_devargs_int(adap, &adap->devargs.force_link_up,
 737                               CXGBE_DEVARG_VF_FORCE_LINK_UP, false);
 738 }
 739
 740 static void configure_vlan_types(struct adapter *adapter)
 741 {
 742         int i;
 743
 744         for_each_port(adapter, i) {
 745                 /* OVLAN Type 0x88a8 */
 746                 t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN0),
 747                                  V_OVLAN_MASK(M_OVLAN_MASK) |
 748                                  V_OVLAN_ETYPE(M_OVLAN_ETYPE),
 749                                  V_OVLAN_MASK(M_OVLAN_MASK) |
 750                                  V_OVLAN_ETYPE(0x88a8));
 751                 /* OVLAN Type 0x9100 */
 752                 t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN1),
 753                                  V_OVLAN_MASK(M_OVLAN_MASK) |
 754                                  V_OVLAN_ETYPE(M_OVLAN_ETYPE),
 755                                  V_OVLAN_MASK(M_OVLAN_MASK) |
 756                                  V_OVLAN_ETYPE(0x9100));
 757
 758                 /* IVLAN 0X8100 */
 759                 t4_set_reg_field(adapter, MPS_PORT_RX_IVLAN(i),
 760                                  V_IVLAN_ETYPE(M_IVLAN_ETYPE),
 761                                  V_IVLAN_ETYPE(0x8100));
 762
 763                 t4_set_reg_field(adapter, MPS_PORT_RX_CTL(i),
 764                                  F_OVLAN_EN0 | F_OVLAN_EN1 |
 765                                  F_IVLAN_EN,
 766                                  F_OVLAN_EN0 | F_OVLAN_EN1 |
 767                                  F_IVLAN_EN);
 768         }
 769
 770         t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG, V_RM_OVLAN(1),
 771                                V_RM_OVLAN(!adapter->devargs.keep_ovlan));
 772 }
 773
 774 static void configure_pcie_ext_tag(struct adapter *adapter)
 775 {
 776         u16 v;
 777         int pos = t4_os_find_pci_capability(adapter, PCI_CAP_ID_EXP);
 778
 779         if (!pos)
 780                 return;
 781
 782         if (pos > 0) {
 783                 t4_os_pci_read_cfg2(adapter, pos + PCI_EXP_DEVCTL, &v);
 784                 v |= PCI_EXP_DEVCTL_EXT_TAG;
 785                 t4_os_pci_write_cfg2(adapter, pos + PCI_EXP_DEVCTL, v);
 786                 if (is_t6(adapter->params.chip)) {
 787                         t4_set_reg_field(adapter, A_PCIE_CFG2,
 788                                          V_T6_TOTMAXTAG(M_T6_TOTMAXTAG),
 789                                          V_T6_TOTMAXTAG(7));
 790                         t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
 791                                          V_T6_MINTAG(M_T6_MINTAG),
 792                                          V_T6_MINTAG(8));
 793                 } else {
 794                         t4_set_reg_field(adapter, A_PCIE_CFG2,
 795                                          V_TOTMAXTAG(M_TOTMAXTAG),
 796                                          V_TOTMAXTAG(3));
 797                         t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
 798                                          V_MINTAG(M_MINTAG),
 799                                          V_MINTAG(8));
 800                 }
 801         }
 802 }
 803
 804 /* Figure out how many Queue Sets we can support */
 805 void cxgbe_configure_max_ethqsets(struct adapter *adapter)
 806 {
 807         unsigned int ethqsets;
 808
 809         /*
 810          * We need to reserve an Ingress Queue for the Asynchronous Firmware
 811          * Event Queue.
 812          *
 813          * For each Queue Set, we'll need the ability to allocate two Egress
 814          * Contexts -- one for the Ingress Queue Free List and one for the TX
 815          * Ethernet Queue.
 816          */
 817         if (is_pf4(adapter)) {
 818                 struct pf_resources *pfres = &adapter->params.pfres;
 819
 820                 ethqsets = pfres->niqflint - 1;
 821                 if (pfres->neq < ethqsets * 2)
 822                         ethqsets = pfres->neq / 2;
 823         } else {
 824                 struct vf_resources *vfres = &adapter->params.vfres;
 825
 826                 ethqsets = vfres->niqflint - 1;
 827                 if (vfres->nethctrl != ethqsets)
 828                         ethqsets = min(vfres->nethctrl, ethqsets);
 829                 if (vfres->neq < ethqsets * 2)
 830                         ethqsets = vfres->neq / 2;
 831         }
 832
 833         if (ethqsets > MAX_ETH_QSETS)
 834                 ethqsets = MAX_ETH_QSETS;
 835         adapter->sge.max_ethqsets = ethqsets;
 836 }
 837
 838 /*
 839  * Tweak configuration based on system architecture, etc.  Most of these have
 840  * defaults assigned to them by Firmware Configuration Files (if we're using
 841  * them) but need to be explicitly set if we're using hard-coded
 842  * initialization. So these are essentially common tweaks/settings for
 843  * Configuration Files and hard-coded initialization ...
 844  */
 845 static int adap_init0_tweaks(struct adapter *adapter)
 846 {
 847         u8 rx_dma_offset;
 848
 849         /*
 850          * Fix up various Host-Dependent Parameters like Page Size, Cache
 851          * Line Size, etc.  The firmware default is for a 4KB Page Size and
 852          * 64B Cache Line Size ...
 853          */
 854         t4_fixup_host_params_compat(adapter, CXGBE_PAGE_SIZE, L1_CACHE_BYTES,
 855                                     T5_LAST_REV);
 856
 857         /*
 858          * Keep the chip default offset to deliver Ingress packets into our
 859          * DMA buffers to zero
 860          */
 861         rx_dma_offset = 0;
 862         t4_set_reg_field(adapter, A_SGE_CONTROL, V_PKTSHIFT(M_PKTSHIFT),
 863                          V_PKTSHIFT(rx_dma_offset));
 864
 865         t4_set_reg_field(adapter, A_SGE_FLM_CFG,
 866                          V_CREDITCNT(M_CREDITCNT) | M_CREDITCNTPACKING,
 867                          V_CREDITCNT(3) | V_CREDITCNTPACKING(1));
 868
 869         t4_set_reg_field(adapter, A_SGE_INGRESS_RX_THRESHOLD,
 870                          V_THRESHOLD_3(M_THRESHOLD_3), V_THRESHOLD_3(32U));
 871
 872         t4_set_reg_field(adapter, A_SGE_CONTROL2, V_IDMAARBROUNDROBIN(1U),
 873                          V_IDMAARBROUNDROBIN(1U));
 874
 875         /*
 876          * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
 877          * adds the pseudo header itself.
 878          */
 879         t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG,
 880                                F_CSUM_HAS_PSEUDO_HDR, 0);
 881
 882         return 0;
 883 }
 884
 885 /*
 886  * Attempt to initialize the adapter via a Firmware Configuration File.
 887  */
 888 static int adap_init0_config(struct adapter *adapter, int reset)
 889 {
 890         struct fw_caps_config_cmd caps_cmd;
 891         unsigned long mtype = 0, maddr = 0;
 892         u32 finiver, finicsum, cfcsum;
 893         int ret;
 894         int config_issued = 0;
 895         int cfg_addr;
 896         char config_name[20];
 897
 898         /*
 899          * Reset device if necessary.
 900          */
 901         if (reset) {
 902                 ret = t4_fw_reset(adapter, adapter->mbox,
 903                                   F_PIORSTMODE | F_PIORST);
 904                 if (ret < 0) {
 905                         dev_warn(adapter, "Firmware reset failed, error %d\n",
 906                                  -ret);
 907                         goto bye;
 908                 }
 909         }
 910
 911         cfg_addr = t4_flash_cfg_addr(adapter);
 912         if (cfg_addr < 0) {
 913                 ret = cfg_addr;
 914                 dev_warn(adapter, "Finding address for firmware config file in flash failed, error %d\n",
 915                          -ret);
 916                 goto bye;
 917         }
 918
 919         strcpy(config_name, "On Flash");
 920         mtype = FW_MEMTYPE_CF_FLASH;
 921         maddr = cfg_addr;
 922
 923         /*
 924          * Issue a Capability Configuration command to the firmware to get it
 925          * to parse the Configuration File.  We don't use t4_fw_config_file()
 926          * because we want the ability to modify various features after we've
 927          * processed the configuration file ...
 928          */
 929         memset(&caps_cmd, 0, sizeof(caps_cmd));
 930         caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
 931                                            F_FW_CMD_REQUEST | F_FW_CMD_READ);
 932         caps_cmd.cfvalid_to_len16 =
 933                 cpu_to_be32(F_FW_CAPS_CONFIG_CMD_CFVALID |
 934                             V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
 935                             V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
 936                             FW_LEN16(caps_cmd));
 937         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
 938                          &caps_cmd);
 939         /*
 940          * If the CAPS_CONFIG failed with an ENOENT (for a Firmware
 941          * Configuration File in FLASH), our last gasp effort is to use the
 942          * Firmware Configuration File which is embedded in the firmware.  A
 943          * very few early versions of the firmware didn't have one embedded
 944          * but we can ignore those.
 945          */
 946         if (ret == -ENOENT) {
 947                 dev_info(adapter, "%s: Going for embedded config in firmware..\n",
 948                          __func__);
 949
 950                 memset(&caps_cmd, 0, sizeof(caps_cmd));
 951                 caps_cmd.op_to_write =
 952                         cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
 953                                     F_FW_CMD_REQUEST | F_FW_CMD_READ);
 954                 caps_cmd.cfvalid_to_len16 = cpu_to_be32(FW_LEN16(caps_cmd));
 955                 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
 956                                  sizeof(caps_cmd), &caps_cmd);
 957                 strcpy(config_name, "Firmware Default");
 958         }
 959
 960         config_issued = 1;
 961         if (ret < 0)
 962                 goto bye;
 963
 964         finiver = be32_to_cpu(caps_cmd.finiver);
 965         finicsum = be32_to_cpu(caps_cmd.finicsum);
 966         cfcsum = be32_to_cpu(caps_cmd.cfcsum);
 967         if (finicsum != cfcsum)
 968                 dev_warn(adapter, "Configuration File checksum mismatch: [fini] csum=%#x, computed csum=%#x\n",
 969                          finicsum, cfcsum);
 970
 971         /*
 972          * If we're a pure NIC driver then disable all offloading facilities.
 973          * This will allow the firmware to optimize aspects of the hardware
 974          * configuration which will result in improved performance.
 975          */
 976         caps_cmd.niccaps &= cpu_to_be16(~FW_CAPS_CONFIG_NIC_ETHOFLD);
 977         caps_cmd.toecaps = 0;
 978         caps_cmd.iscsicaps = 0;
 979         caps_cmd.rdmacaps = 0;
 980         caps_cmd.fcoecaps = 0;
 981
 982         /*
 983          * And now tell the firmware to use the configuration we just loaded.
 984          */
 985         caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
 986                                            F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
 987         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
 988         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
 989                          NULL);
 990         if (ret < 0) {
 991                 dev_warn(adapter, "Unable to finalize Firmware Capabilities %d\n",
 992                          -ret);
 993                 goto bye;
 994         }
 995
 996         /*
 997          * Tweak configuration based on system architecture, etc.
 998          */
 999         ret = adap_init0_tweaks(adapter);
1000         if (ret < 0) {
1001                 dev_warn(adapter, "Unable to do init0-tweaks %d\n", -ret);
1002                 goto bye;
1003         }
1004
1005         /*
1006          * And finally tell the firmware to initialize itself using the
1007          * parameters from the Configuration File.
1008          */
1009         ret = t4_fw_initialize(adapter, adapter->mbox);
1010         if (ret < 0) {
1011                 dev_warn(adapter, "Initializing Firmware failed, error %d\n",
1012                          -ret);
1013                 goto bye;
1014         }
1015
1016         /*
1017          * Return successfully and note that we're operating with parameters
1018          * not supplied by the driver, rather than from hard-wired
1019          * initialization constants buried in the driver.
1020          */
1021         dev_info(adapter,
1022                  "Successfully configured using Firmware Configuration File \"%s\", version %#x, computed checksum %#x\n",
1023                  config_name, finiver, cfcsum);
1024
1025         return 0;
1026
1027         /*
1028          * Something bad happened.  Return the error ...  (If the "error"
1029          * is that there's no Configuration File on the adapter we don't
1030          * want to issue a warning since this is fairly common.)
1031          */
1032 bye:
1033         if (config_issued && ret != -ENOENT)
1034                 dev_warn(adapter, "\"%s\" configuration file error %d\n",
1035                          config_name, -ret);
1036
1037         dev_debug(adapter, "%s: returning ret = %d ..\n", __func__, ret);
1038         return ret;
1039 }
1040
1041 static int adap_init0(struct adapter *adap)
1042 {
1043         struct fw_caps_config_cmd caps_cmd;
1044         int ret = 0;
1045         u32 v, port_vec;
1046         enum dev_state state;
1047         u32 params[7], val[7];
1048         int reset = 1;
1049         int mbox = adap->mbox;
1050
1051         /*
1052          * Contact FW, advertising Master capability.
1053          */
1054         ret = t4_fw_hello(adap, adap->mbox, adap->mbox, MASTER_MAY, &state);
1055         if (ret < 0) {
1056                 dev_err(adap, "%s: could not connect to FW, error %d\n",
1057                         __func__, -ret);
1058                 goto bye;
1059         }
1060
1061         CXGBE_DEBUG_MBOX(adap, "%s: adap->mbox = %d; ret = %d\n", __func__,
1062                          adap->mbox, ret);
1063
1064         if (ret == mbox)
1065                 adap->flags |= MASTER_PF;
1066
1067         if (state == DEV_STATE_INIT) {
1068                 /*
1069                  * Force halt and reset FW because a previous instance may have
1070                  * exited abnormally without properly shutting down
1071                  */
1072                 ret = t4_fw_halt(adap, adap->mbox, reset);
1073                 if (ret < 0) {
1074                         dev_err(adap, "Failed to halt. Exit.\n");
1075                         goto bye;
1076                 }
1077
1078                 ret = t4_fw_restart(adap, adap->mbox, reset);
1079                 if (ret < 0) {
1080                         dev_err(adap, "Failed to restart. Exit.\n");
1081                         goto bye;
1082                 }
1083                 state = (enum dev_state)((unsigned)state & ~DEV_STATE_INIT);
1084         }
1085
1086         t4_get_version_info(adap);
1087
1088         ret = t4_get_core_clock(adap, &adap->params.vpd);
1089         if (ret < 0) {
1090                 dev_err(adap, "%s: could not get core clock, error %d\n",
1091                         __func__, -ret);
1092                 goto bye;
1093         }
1094
1095         /*
1096          * If the firmware is initialized already (and we're not forcing a
1097          * master initialization), note that we're living with existing
1098          * adapter parameters.  Otherwise, it's time to try initializing the
1099          * adapter ...
1100          */
1101         if (state == DEV_STATE_INIT) {
1102                 dev_info(adap, "Coming up as %s: Adapter already initialized\n",
1103                          adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
1104         } else {
1105                 dev_info(adap, "Coming up as MASTER: Initializing adapter\n");
1106
1107                 ret = adap_init0_config(adap, reset);
1108                 if (ret == -ENOENT) {
1109                         dev_err(adap,
1110                                 "No Configuration File present on adapter. Using hard-wired configuration parameters.\n");
1111                         goto bye;
1112                 }
1113         }
1114         if (ret < 0) {
1115                 dev_err(adap, "could not initialize adapter, error %d\n", -ret);
1116                 goto bye;
1117         }
1118
1119         /* Now that we've successfully configured and initialized the adapter
1120          * (or found it already initialized), we can ask the Firmware what
1121          * resources it has provisioned for us.
1122          */
1123         ret = t4_get_pfres(adap);
1124         if (ret) {
1125                 dev_err(adap->pdev_dev,
1126                         "Unable to retrieve resource provisioning info\n");
1127                 goto bye;
1128         }
1129
1130         /* Find out what ports are available to us. */
1131         v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
1132             V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
1133         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
1134         if (ret < 0) {
1135                 dev_err(adap, "%s: failure in t4_query_params; error = %d\n",
1136                         __func__, ret);
1137                 goto bye;
1138         }
1139
1140         adap->params.nports = hweight32(port_vec);
1141         adap->params.portvec = port_vec;
1142
1143         dev_debug(adap, "%s: adap->params.nports = %u\n", __func__,
1144                   adap->params.nports);
1145
1146         /*
1147          * Give the SGE code a chance to pull in anything that it needs ...
1148          * Note that this must be called after we retrieve our VPD parameters
1149          * in order to know how to convert core ticks to seconds, etc.
1150          */
1151         ret = t4_sge_init(adap);
1152         if (ret < 0) {
1153                 dev_err(adap, "t4_sge_init failed with error %d\n",
1154                         -ret);
1155                 goto bye;
1156         }
1157
1158         /*
1159          * Grab some of our basic fundamental operating parameters.
1160          */
1161         params[0] = CXGBE_FW_PARAM_PFVF(L2T_START);
1162         params[1] = CXGBE_FW_PARAM_PFVF(L2T_END);
1163         params[2] = CXGBE_FW_PARAM_PFVF(FILTER_START);
1164         params[3] = CXGBE_FW_PARAM_PFVF(FILTER_END);
1165         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 4, params, val);
1166         if (ret < 0)
1167                 goto bye;
1168         adap->l2t_start = val[0];
1169         adap->l2t_end = val[1];
1170         adap->tids.ftid_base = val[2];
1171         adap->tids.nftids = val[3] - val[2] + 1;
1172
1173         params[0] = CXGBE_FW_PARAM_PFVF(CLIP_START);
1174         params[1] = CXGBE_FW_PARAM_PFVF(CLIP_END);
1175         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
1176         if (ret < 0)
1177                 goto bye;
1178         adap->clipt_start = val[0];
1179         adap->clipt_end = val[1];
1180
1181         /*
1182          * Get device capabilities so we can determine what resources we need
1183          * to manage.
1184          */
1185         memset(&caps_cmd, 0, sizeof(caps_cmd));
1186         caps_cmd.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1187                                      F_FW_CMD_REQUEST | F_FW_CMD_READ);
1188         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
1189         ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
1190                          &caps_cmd);
1191         if (ret < 0)
1192                 goto bye;
1193
1194         if ((caps_cmd.niccaps & cpu_to_be16(FW_CAPS_CONFIG_NIC_HASHFILTER)) &&
1195             is_t6(adap->params.chip)) {
1196                 if (cxgbe_init_hash_filter(adap) < 0)
1197                         goto bye;
1198         }
1199
1200         /* See if FW supports FW_FILTER2 work request */
1201         if (is_t4(adap->params.chip)) {
1202                 adap->params.filter2_wr_support = 0;
1203         } else {
1204                 params[0] = CXGBE_FW_PARAM_DEV(FILTER2_WR);
1205                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1206                                       1, params, val);
1207                 adap->params.filter2_wr_support = (ret == 0 && val[0] != 0);
1208         }
1209
1210         /* query tid-related parameters */
1211         params[0] = CXGBE_FW_PARAM_DEV(NTID);
1212         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
1213                               params, val);
1214         if (ret < 0)
1215                 goto bye;
1216         adap->tids.ntids = val[0];
1217         adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
1218
1219         /* If we're running on newer firmware, let it know that we're
1220          * prepared to deal with encapsulated CPL messages.  Older
1221          * firmware won't understand this and we'll just get
1222          * unencapsulated messages ...
1223          */
1224         params[0] = CXGBE_FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
1225         val[0] = 1;
1226         (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1227
1228         /*
1229          * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
1230          * capability.  Earlier versions of the firmware didn't have the
1231          * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
1232          * permission to use ULPTX MEMWRITE DSGL.
1233          */
1234         if (is_t4(adap->params.chip)) {
1235                 adap->params.ulptx_memwrite_dsgl = false;
1236         } else {
1237                 params[0] = CXGBE_FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
1238                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1239                                       1, params, val);
1240                 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
1241         }
1242
1243         /* Query for max number of packets that can be coalesced for Tx */
1244         params[0] = CXGBE_FW_PARAM_PFVF(MAX_PKTS_PER_ETH_TX_PKTS_WR);
1245         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1246         if (!ret && val[0] > 0)
1247                 adap->params.max_tx_coalesce_num = val[0];
1248         else
1249                 adap->params.max_tx_coalesce_num = ETH_COALESCE_PKT_NUM;
1250
1251         /*
1252          * The MTU/MSS Table is initialized by now, so load their values.  If
1253          * we're initializing the adapter, then we'll make any modifications
1254          * we want to the MTU/MSS Table and also initialize the congestion
1255          * parameters.
1256          */
1257         t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
1258         if (state != DEV_STATE_INIT) {
1259                 int i;
1260
1261                 /*
1262                  * The default MTU Table contains values 1492 and 1500.
1263                  * However, for TCP, it's better to have two values which are
1264                  * a multiple of 8 +/- 4 bytes apart near this popular MTU.
1265                  * This allows us to have a TCP Data Payload which is a
1266                  * multiple of 8 regardless of what combination of TCP Options
1267                  * are in use (always a multiple of 4 bytes) which is
1268                  * important for performance reasons.  For instance, if no
1269                  * options are in use, then we have a 20-byte IP header and a
1270                  * 20-byte TCP header.  In this case, a 1500-byte MSS would
1271                  * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
1272                  * which is not a multiple of 8.  So using an MSS of 1488 in
1273                  * this case results in a TCP Data Payload of 1448 bytes which
1274                  * is a multiple of 8.  On the other hand, if 12-byte TCP Time
1275                  * Stamps have been negotiated, then an MTU of 1500 bytes
1276                  * results in a TCP Data Payload of 1448 bytes which, as
1277                  * above, is a multiple of 8 bytes ...
1278                  */
1279                 for (i = 0; i < NMTUS; i++)
1280                         if (adap->params.mtus[i] == 1492) {
1281                                 adap->params.mtus[i] = 1488;
1282                                 break;
1283                         }
1284
1285                 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
1286                              adap->params.b_wnd);
1287         }
1288         t4_init_sge_params(adap);
1289         t4_init_tp_params(adap);
1290         configure_pcie_ext_tag(adap);
1291         configure_vlan_types(adap);
1292         cxgbe_configure_max_ethqsets(adap);
1293
1294         adap->params.drv_memwin = MEMWIN_NIC;
1295         adap->flags |= FW_OK;
1296         dev_debug(adap, "%s: returning zero..\n", __func__);
1297         return 0;
1298
1299         /*
1300          * Something bad happened.  If a command timed out or failed with EIO
1301          * FW does not operate within its spec or something catastrophic
1302          * happened to HW/FW, stop issuing commands.
1303          */
1304 bye:
1305         if (ret != -ETIMEDOUT && ret != -EIO)
1306                 t4_fw_bye(adap, adap->mbox);
1307         return ret;
1308 }
1309
1310 /**
1311  * t4_os_portmod_changed - handle port module changes
1312  * @adap: the adapter associated with the module change
1313  * @port_id: the port index whose module status has changed
1314  *
1315  * This is the OS-dependent handler for port module changes.  It is
1316  * invoked when a port module is removed or inserted for any OS-specific
1317  * processing.
1318  */
1319 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
1320 {
1321         static const char * const mod_str[] = {
1322                 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
1323         };
1324
1325         const struct port_info *pi = adap2pinfo(adap, port_id);
1326
1327         if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
1328                 dev_info(adap, "Port%d: port module unplugged\n", pi->port_id);
1329         else if (pi->mod_type < ARRAY_SIZE(mod_str))
1330                 dev_info(adap, "Port%d: %s port module inserted\n", pi->port_id,
1331                          mod_str[pi->mod_type]);
1332         else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
1333                 dev_info(adap, "Port%d: unsupported port module inserted\n",
1334                          pi->port_id);
1335         else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
1336                 dev_info(adap, "Port%d: unknown port module inserted\n",
1337                          pi->port_id);
1338         else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
1339                 dev_info(adap, "Port%d: transceiver module error\n",
1340                          pi->port_id);
1341         else
1342                 dev_info(adap, "Port%d: unknown module type %d inserted\n",
1343                          pi->port_id, pi->mod_type);
1344 }
1345
1346 bool cxgbe_force_linkup(struct adapter *adap)
1347 {
1348         if (is_pf4(adap))
1349                 return false;   /* force_linkup not required for pf driver */
1350
1351         return adap->devargs.force_link_up;
1352 }
1353
1354 /**
1355  * link_start - enable a port
1356  * @dev: the port to enable
1357  *
1358  * Performs the MAC and PHY actions needed to enable a port.
1359  */
1360 int cxgbe_link_start(struct port_info *pi)
1361 {
1362         struct adapter *adapter = pi->adapter;
1363         u64 conf_offloads;
1364         unsigned int mtu;
1365         int ret;
1366
1367         mtu = pi->eth_dev->data->dev_conf.rxmode.max_rx_pkt_len -
1368               (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN);
1369
1370         conf_offloads = pi->eth_dev->data->dev_conf.rxmode.offloads;
1371
1372         /*
1373          * We do not set address filters and promiscuity here, the stack does
1374          * that step explicitly.
1375          */
1376         ret = t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu, -1, -1, -1,
1377                             !!(conf_offloads & DEV_RX_OFFLOAD_VLAN_STRIP),
1378                             true);
1379         if (ret == 0) {
1380                 ret = cxgbe_mpstcam_modify(pi, (int)pi->xact_addr_filt,
1381                                 (u8 *)&pi->eth_dev->data->mac_addrs[0]);
1382                 if (ret >= 0) {
1383                         pi->xact_addr_filt = ret;
1384                         ret = 0;
1385                 }
1386         }
1387         if (ret == 0 && is_pf4(adapter))
1388                 ret = t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
1389                                     &pi->link_cfg);
1390         if (ret == 0) {
1391                 /*
1392                  * Enabling a Virtual Interface can result in an interrupt
1393                  * during the processing of the VI Enable command and, in some
1394                  * paths, result in an attempt to issue another command in the
1395                  * interrupt context.  Thus, we disable interrupts during the
1396                  * course of the VI Enable command ...
1397                  */
1398                 ret = t4_enable_vi_params(adapter, adapter->mbox, pi->viid,
1399                                           true, true, false);
1400         }
1401
1402         if (ret == 0 && cxgbe_force_linkup(adapter))
1403                 pi->eth_dev->data->dev_link.link_status = ETH_LINK_UP;
1404         return ret;
1405 }
1406
1407 /**
1408  * cxgbe_write_rss_conf - flash the RSS configuration for a given port
1409  * @pi: the port
1410  * @rss_hf: Hash configuration to apply
1411  */
1412 int cxgbe_write_rss_conf(const struct port_info *pi, uint64_t rss_hf)
1413 {
1414         struct adapter *adapter = pi->adapter;
1415         const struct sge_eth_rxq *rxq;
1416         u64 flags = 0;
1417         u16 rss;
1418         int err;
1419
1420         /*  Should never be called before setting up sge eth rx queues */
1421         if (!(adapter->flags & FULL_INIT_DONE)) {
1422                 dev_err(adap, "%s No RXQs available on port %d\n",
1423                         __func__, pi->port_id);
1424                 return -EINVAL;
1425         }
1426
1427         /* Don't allow unsupported hash functions */
1428         if (rss_hf & ~CXGBE_RSS_HF_ALL)
1429                 return -EINVAL;
1430
1431         if (rss_hf & CXGBE_RSS_HF_IPV4_MASK)
1432                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
1433
1434         if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
1435                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
1436
1437         if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
1438                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
1439                          F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1440
1441         if (rss_hf & CXGBE_RSS_HF_IPV6_MASK)
1442                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
1443
1444         if (rss_hf & CXGBE_RSS_HF_TCP_IPV6_MASK)
1445                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1446                          F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
1447
1448         if (rss_hf & CXGBE_RSS_HF_UDP_IPV6_MASK)
1449                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1450                          F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
1451                          F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1452
1453         rxq = &adapter->sge.ethrxq[pi->first_qset];
1454         rss = rxq[0].rspq.abs_id;
1455
1456         /* If Tunnel All Lookup isn't specified in the global RSS
1457          * Configuration, then we need to specify a default Ingress
1458          * Queue for any ingress packets which aren't hashed.  We'll
1459          * use our first ingress queue ...
1460          */
1461         err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
1462                                flags, rss);
1463         return err;
1464 }
1465
1466 /**
1467  * cxgbe_write_rss - write the RSS table for a given port
1468  * @pi: the port
1469  * @queues: array of queue indices for RSS
1470  *
1471  * Sets up the portion of the HW RSS table for the port's VI to distribute
1472  * packets to the Rx queues in @queues.
1473  */
1474 int cxgbe_write_rss(const struct port_info *pi, const u16 *queues)
1475 {
1476         u16 *rss;
1477         int i, err;
1478         struct adapter *adapter = pi->adapter;
1479         const struct sge_eth_rxq *rxq;
1480
1481         /*  Should never be called before setting up sge eth rx queues */
1482         BUG_ON(!(adapter->flags & FULL_INIT_DONE));
1483
1484         rxq = &adapter->sge.ethrxq[pi->first_qset];
1485         rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
1486         if (!rss)
1487                 return -ENOMEM;
1488
1489         /* map the queue indices to queue ids */
1490         for (i = 0; i < pi->rss_size; i++, queues++)
1491                 rss[i] = rxq[*queues].rspq.abs_id;
1492
1493         err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
1494                                   pi->rss_size, rss, pi->rss_size);
1495         rte_free(rss);
1496         return err;
1497 }
1498
1499 /**
1500  * setup_rss - configure RSS
1501  * @adapter: the adapter
1502  *
1503  * Sets up RSS to distribute packets to multiple receive queues.  We
1504  * configure the RSS CPU lookup table to distribute to the number of HW
1505  * receive queues, and the response queue lookup table to narrow that
1506  * down to the response queues actually configured for each port.
1507  * We always configure the RSS mapping for all ports since the mapping
1508  * table has plenty of entries.
1509  */
1510 int cxgbe_setup_rss(struct port_info *pi)
1511 {
1512         int j, err;
1513         struct adapter *adapter = pi->adapter;
1514
1515         dev_debug(adapter, "%s:  pi->rss_size = %u; pi->n_rx_qsets = %u\n",
1516                   __func__, pi->rss_size, pi->n_rx_qsets);
1517
1518         if (!(pi->flags & PORT_RSS_DONE)) {
1519                 if (adapter->flags & FULL_INIT_DONE) {
1520                         /* Fill default values with equal distribution */
1521                         for (j = 0; j < pi->rss_size; j++)
1522                                 pi->rss[j] = j % pi->n_rx_qsets;
1523
1524                         err = cxgbe_write_rss(pi, pi->rss);
1525                         if (err)
1526                                 return err;
1527
1528                         err = cxgbe_write_rss_conf(pi, pi->rss_hf);
1529                         if (err)
1530                                 return err;
1531                         pi->flags |= PORT_RSS_DONE;
1532                 }
1533         }
1534         return 0;
1535 }
1536
1537 /*
1538  * Enable NAPI scheduling and interrupt generation for all Rx queues.
1539  */
1540 static void enable_rx(struct adapter *adap, struct sge_rspq *q)
1541 {
1542         /* 0-increment GTS to start the timer and enable interrupts */
1543         t4_write_reg(adap, is_pf4(adap) ? MYPF_REG(A_SGE_PF_GTS) :
1544                                           T4VF_SGE_BASE_ADDR + A_SGE_VF_GTS,
1545                      V_SEINTARM(q->intr_params) |
1546                      V_INGRESSQID(q->cntxt_id));
1547 }
1548
1549 void cxgbe_enable_rx_queues(struct port_info *pi)
1550 {
1551         struct adapter *adap = pi->adapter;
1552         struct sge *s = &adap->sge;
1553         unsigned int i;
1554
1555         for (i = 0; i < pi->n_rx_qsets; i++)
1556                 enable_rx(adap, &s->ethrxq[pi->first_qset + i].rspq);
1557 }
1558
1559 /**
1560  * fw_caps_to_speed_caps - translate Firmware Port Caps to Speed Caps.
1561  * @port_type: Firmware Port Type
1562  * @fw_caps: Firmware Port Capabilities
1563  * @speed_caps: Device Info Speed Capabilities
1564  *
1565  * Translate a Firmware Port Capabilities specification to Device Info
1566  * Speed Capabilities.
1567  */
1568 static void fw_caps_to_speed_caps(enum fw_port_type port_type,
1569                                   unsigned int fw_caps,
1570                                   u32 *speed_caps)
1571 {
1572 #define SET_SPEED(__speed_name) \
1573         do { \
1574                 *speed_caps |= ETH_LINK_ ## __speed_name; \
1575         } while (0)
1576
1577 #define FW_CAPS_TO_SPEED(__fw_name) \
1578         do { \
1579                 if (fw_caps & FW_PORT_CAP32_ ## __fw_name) \
1580                         SET_SPEED(__fw_name); \
1581         } while (0)
1582
1583         switch (port_type) {
1584         case FW_PORT_TYPE_BT_SGMII:
1585         case FW_PORT_TYPE_BT_XFI:
1586         case FW_PORT_TYPE_BT_XAUI:
1587                 FW_CAPS_TO_SPEED(SPEED_100M);
1588                 FW_CAPS_TO_SPEED(SPEED_1G);
1589                 FW_CAPS_TO_SPEED(SPEED_10G);
1590                 break;
1591
1592         case FW_PORT_TYPE_KX4:
1593         case FW_PORT_TYPE_KX:
1594         case FW_PORT_TYPE_FIBER_XFI:
1595         case FW_PORT_TYPE_FIBER_XAUI:
1596         case FW_PORT_TYPE_SFP:
1597         case FW_PORT_TYPE_QSFP_10G:
1598         case FW_PORT_TYPE_QSA:
1599                 FW_CAPS_TO_SPEED(SPEED_1G);
1600                 FW_CAPS_TO_SPEED(SPEED_10G);
1601                 break;
1602
1603         case FW_PORT_TYPE_KR:
1604                 SET_SPEED(SPEED_10G);
1605                 break;
1606
1607         case FW_PORT_TYPE_BP_AP:
1608         case FW_PORT_TYPE_BP4_AP:
1609                 SET_SPEED(SPEED_1G);
1610                 SET_SPEED(SPEED_10G);
1611                 break;
1612
1613         case FW_PORT_TYPE_BP40_BA:
1614         case FW_PORT_TYPE_QSFP:
1615                 SET_SPEED(SPEED_40G);
1616                 break;
1617
1618         case FW_PORT_TYPE_CR_QSFP:
1619         case FW_PORT_TYPE_SFP28:
1620         case FW_PORT_TYPE_KR_SFP28:
1621                 FW_CAPS_TO_SPEED(SPEED_1G);
1622                 FW_CAPS_TO_SPEED(SPEED_10G);
1623                 FW_CAPS_TO_SPEED(SPEED_25G);
1624                 break;
1625
1626         case FW_PORT_TYPE_CR2_QSFP:
1627                 SET_SPEED(SPEED_50G);
1628                 break;
1629
1630         case FW_PORT_TYPE_KR4_100G:
1631         case FW_PORT_TYPE_CR4_QSFP:
1632                 FW_CAPS_TO_SPEED(SPEED_25G);
1633                 FW_CAPS_TO_SPEED(SPEED_40G);
1634                 FW_CAPS_TO_SPEED(SPEED_50G);
1635                 FW_CAPS_TO_SPEED(SPEED_100G);
1636                 break;
1637
1638         default:
1639                 break;
1640         }
1641
1642 #undef FW_CAPS_TO_SPEED
1643 #undef SET_SPEED
1644 }
1645
1646 /**
1647  * cxgbe_get_speed_caps - Fetch supported speed capabilities
1648  * @pi: Underlying port's info
1649  * @speed_caps: Device Info speed capabilities
1650  *
1651  * Fetch supported speed capabilities of the underlying port.
1652  */
1653 void cxgbe_get_speed_caps(struct port_info *pi, u32 *speed_caps)
1654 {
1655         *speed_caps = 0;
1656
1657         fw_caps_to_speed_caps(pi->port_type, pi->link_cfg.pcaps,
1658                               speed_caps);
1659
1660         if (!(pi->link_cfg.pcaps & FW_PORT_CAP32_ANEG))
1661                 *speed_caps |= ETH_LINK_SPEED_FIXED;
1662 }
1663
1664 /**
1665  * cxgbe_set_link_status - Set device link up or down.
1666  * @pi: Underlying port's info
1667  * @status: 0 - down, 1 - up
1668  *
1669  * Set the device link up or down.
1670  */
1671 int cxgbe_set_link_status(struct port_info *pi, bool status)
1672 {
1673         struct adapter *adapter = pi->adapter;
1674         int err = 0;
1675
1676         err = t4_enable_vi(adapter, adapter->mbox, pi->viid, status, status);
1677         if (err) {
1678                 dev_err(adapter, "%s: disable_vi failed: %d\n", __func__, err);
1679                 return err;
1680         }
1681
1682         if (!status)
1683                 t4_reset_link_config(adapter, pi->pidx);
1684
1685         return 0;
1686 }
1687
1688 /**
1689  * cxgb_up - enable the adapter
1690  * @adap: adapter being enabled
1691  *
1692  * Called when the first port is enabled, this function performs the
1693  * actions necessary to make an adapter operational, such as completing
1694  * the initialization of HW modules, and enabling interrupts.
1695  */
1696 int cxgbe_up(struct adapter *adap)
1697 {
1698         enable_rx(adap, &adap->sge.fw_evtq);
1699         t4_sge_tx_monitor_start(adap);
1700         if (is_pf4(adap))
1701                 t4_intr_enable(adap);
1702         adap->flags |= FULL_INIT_DONE;
1703
1704         /* TODO: deadman watchdog ?? */
1705         return 0;
1706 }
1707
1708 /*
1709  * Close the port
1710  */
1711 int cxgbe_down(struct port_info *pi)
1712 {
1713         return cxgbe_set_link_status(pi, false);
1714 }
1715
1716 /*
1717  * Release resources when all the ports have been stopped.
1718  */
1719 void cxgbe_close(struct adapter *adapter)
1720 {
1721         struct port_info *pi;
1722         int i;
1723
1724         if (adapter->flags & FULL_INIT_DONE) {
1725                 tid_free(&adapter->tids);
1726                 t4_cleanup_mpstcam(adapter);
1727                 t4_cleanup_clip_tbl(adapter);
1728                 t4_cleanup_l2t(adapter);
1729                 if (is_pf4(adapter))
1730                         t4_intr_disable(adapter);
1731                 t4_sge_tx_monitor_stop(adapter);
1732                 t4_free_sge_resources(adapter);
1733                 for_each_port(adapter, i) {
1734                         pi = adap2pinfo(adapter, i);
1735                         if (pi->viid != 0)
1736                                 t4_free_vi(adapter, adapter->mbox,
1737                                            adapter->pf, 0, pi->viid);
1738                         rte_eth_dev_release_port(pi->eth_dev);
1739                 }
1740                 adapter->flags &= ~FULL_INIT_DONE;
1741         }
1742
1743         if (is_pf4(adapter) && (adapter->flags & FW_OK))
1744                 t4_fw_bye(adapter, adapter->mbox);
1745 }
1746
1747 int cxgbe_probe(struct adapter *adapter)
1748 {
1749         struct port_info *pi;
1750         int chip;
1751         int func, i;
1752         int err = 0;
1753         u32 whoami;
1754
1755         whoami = t4_read_reg(adapter, A_PL_WHOAMI);
1756         chip = t4_get_chip_type(adapter,
1757                         CHELSIO_PCI_ID_VER(adapter->pdev->id.device_id));
1758         if (chip < 0)
1759                 return chip;
1760
1761         func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
1762                G_SOURCEPF(whoami) : G_T6_SOURCEPF(whoami);
1763
1764         adapter->mbox = func;
1765         adapter->pf = func;
1766
1767         t4_os_lock_init(&adapter->mbox_lock);
1768         TAILQ_INIT(&adapter->mbox_list);
1769         t4_os_lock_init(&adapter->win0_lock);
1770
1771         err = t4_prep_adapter(adapter);
1772         if (err)
1773                 return err;
1774
1775         setup_memwin(adapter);
1776         err = adap_init0(adapter);
1777         if (err) {
1778                 dev_err(adapter, "%s: Adapter initialization failed, error %d\n",
1779                         __func__, err);
1780                 goto out_free;
1781         }
1782
1783         if (!is_t4(adapter->params.chip)) {
1784                 /*
1785                  * The userspace doorbell BAR is split evenly into doorbell
1786                  * regions, each associated with an egress queue.  If this
1787                  * per-queue region is large enough (at least UDBS_SEG_SIZE)
1788                  * then it can be used to submit a tx work request with an
1789                  * implied doorbell.  Enable write combining on the BAR if
1790                  * there is room for such work requests.
1791                  */
1792                 int s_qpp, qpp, num_seg;
1793
1794                 s_qpp = (S_QUEUESPERPAGEPF0 +
1795                         (S_QUEUESPERPAGEPF1 - S_QUEUESPERPAGEPF0) *
1796                         adapter->pf);
1797                 qpp = 1 << ((t4_read_reg(adapter,
1798                                 A_SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp)
1799                                 & M_QUEUESPERPAGEPF0);
1800                 num_seg = CXGBE_PAGE_SIZE / UDBS_SEG_SIZE;
1801                 if (qpp > num_seg)
1802                         dev_warn(adapter, "Incorrect SGE EGRESS QUEUES_PER_PAGE configuration, continuing in debug mode\n");
1803
1804                 adapter->bar2 = (void *)adapter->pdev->mem_resource[2].addr;
1805                 if (!adapter->bar2) {
1806                         dev_err(adapter, "cannot map device bar2 region\n");
1807                         err = -ENOMEM;
1808                         goto out_free;
1809                 }
1810                 t4_write_reg(adapter, A_SGE_STAT_CFG, V_STATSOURCE_T5(7) |
1811                              V_STATMODE(0));
1812         }
1813
1814         for_each_port(adapter, i) {
1815                 const unsigned int numa_node = rte_socket_id();
1816                 char name[RTE_ETH_NAME_MAX_LEN];
1817                 struct rte_eth_dev *eth_dev;
1818
1819                 snprintf(name, sizeof(name), "%s_%d",
1820                          adapter->pdev->device.name, i);
1821
1822                 if (i == 0) {
1823                         /* First port is already allocated by DPDK */
1824                         eth_dev = adapter->eth_dev;
1825                         goto allocate_mac;
1826                 }
1827
1828                 /*
1829                  * now do all data allocation - for eth_dev structure,
1830                  * and internal (private) data for the remaining ports
1831                  */
1832
1833                 /* reserve an ethdev entry */
1834                 eth_dev = rte_eth_dev_allocate(name);
1835                 if (!eth_dev)
1836                         goto out_free;
1837
1838                 eth_dev->data->dev_private =
1839                         rte_zmalloc_socket(name, sizeof(struct port_info),
1840                                            RTE_CACHE_LINE_SIZE, numa_node);
1841                 if (!eth_dev->data->dev_private)
1842                         goto out_free;
1843
1844 allocate_mac:
1845                 pi = eth_dev->data->dev_private;
1846                 adapter->port[i] = pi;
1847                 pi->eth_dev = eth_dev;
1848                 pi->adapter = adapter;
1849                 pi->xact_addr_filt = -1;
1850                 pi->port_id = i;
1851                 pi->pidx = i;
1852
1853                 pi->eth_dev->device = &adapter->pdev->device;
1854                 pi->eth_dev->dev_ops = adapter->eth_dev->dev_ops;
1855                 pi->eth_dev->tx_pkt_burst = adapter->eth_dev->tx_pkt_burst;
1856                 pi->eth_dev->rx_pkt_burst = adapter->eth_dev->rx_pkt_burst;
1857
1858                 rte_eth_copy_pci_info(pi->eth_dev, adapter->pdev);
1859
1860                 pi->eth_dev->data->mac_addrs = rte_zmalloc(name,
1861                                                         RTE_ETHER_ADDR_LEN, 0);
1862                 if (!pi->eth_dev->data->mac_addrs) {
1863                         dev_err(adapter, "%s: Mem allocation failed for storing mac addr, aborting\n",
1864                                 __func__);
1865                         err = -1;
1866                         goto out_free;
1867                 }
1868
1869                 if (i > 0) {
1870                         /* First port will be notified by upper layer */
1871                         rte_eth_dev_probing_finish(eth_dev);
1872                 }
1873         }
1874
1875         if (adapter->flags & FW_OK) {
1876                 err = t4_port_init(adapter, adapter->mbox, adapter->pf, 0);
1877                 if (err) {
1878                         dev_err(adapter, "%s: t4_port_init failed with err %d\n",
1879                                 __func__, err);
1880                         goto out_free;
1881                 }
1882         }
1883
1884         cxgbe_cfg_queues(adapter->eth_dev);
1885
1886         cxgbe_print_adapter_info(adapter);
1887         cxgbe_print_port_info(adapter);
1888
1889         adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
1890                                           adapter->clipt_end);
1891         if (!adapter->clipt) {
1892                 /* We tolerate a lack of clip_table, giving up some
1893                  * functionality
1894                  */
1895                 dev_warn(adapter, "could not allocate CLIP. Continuing\n");
1896         }
1897
1898         adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
1899         if (!adapter->l2t) {
1900                 /* We tolerate a lack of L2T, giving up some functionality */
1901                 dev_warn(adapter, "could not allocate L2T. Continuing\n");
1902         }
1903
1904         if (tid_init(&adapter->tids) < 0) {
1905                 /* Disable filtering support */
1906                 dev_warn(adapter, "could not allocate TID table, "
1907                          "filter support disabled. Continuing\n");
1908         }
1909
1910         t4_os_lock_init(&adapter->flow_lock);
1911
1912         adapter->mpstcam = t4_init_mpstcam(adapter);
1913         if (!adapter->mpstcam)
1914                 dev_warn(adapter, "could not allocate mps tcam table."
1915                          " Continuing\n");
1916
1917         if (is_hashfilter(adapter)) {
1918                 if (t4_read_reg(adapter, A_LE_DB_CONFIG) & F_HASHEN) {
1919                         u32 hash_base, hash_reg;
1920
1921                         hash_reg = A_LE_DB_TID_HASHBASE;
1922                         hash_base = t4_read_reg(adapter, hash_reg);
1923                         adapter->tids.hash_base = hash_base / 4;
1924                 }
1925         } else {
1926                 /* Disable hash filtering support */
1927                 dev_warn(adapter,
1928                          "Maskless filter support disabled. Continuing\n");
1929         }
1930
1931         err = cxgbe_init_rss(adapter);
1932         if (err)
1933                 goto out_free;
1934
1935         return 0;
1936
1937 out_free:
1938         for_each_port(adapter, i) {
1939                 pi = adap2pinfo(adapter, i);
1940                 if (pi->viid != 0)
1941                         t4_free_vi(adapter, adapter->mbox, adapter->pf,
1942                                    0, pi->viid);
1943                 rte_eth_dev_release_port(pi->eth_dev);
1944         }
1945
1946         if (adapter->flags & FW_OK)
1947                 t4_fw_bye(adapter, adapter->mbox);
1948         return -err;
1949 }