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 "smt.h"
  44 #include "mps_tcam.h"
  45
  46 static const u16 cxgbe_filter_mode_features[] = {
  47         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE |
  48          F_PROTOCOL | F_PORT),
  49         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE |
  50          F_PROTOCOL | F_FCOE),
  51         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_TOS |
  52          F_PORT),
  53         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_TOS |
  54          F_FCOE),
  55         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_PORT |
  56          F_FCOE),
  57         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_TOS |
  58          F_PORT | F_FCOE),
  59         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VLAN |
  60          F_FCOE),
  61         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VNIC_ID |
  62          F_FCOE),
  63         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VLAN |
  64          F_FCOE),
  65         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VNIC_ID |
  66          F_FCOE),
  67         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_VLAN | F_PORT |
  68          F_FCOE),
  69         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_VNIC_ID | F_PORT |
  70          F_FCOE),
  71         (F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_PROTOCOL | F_TOS |
  72          F_PORT | F_FCOE),
  73         (F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VLAN | F_PORT),
  74         (F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VLAN | F_FCOE),
  75         (F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VNIC_ID | F_PORT),
  76         (F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VNIC_ID | F_FCOE),
  77         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VLAN | F_PORT),
  78         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VLAN | F_FCOE),
  79         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VNIC_ID |
  80          F_PORT),
  81         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VNIC_ID |
  82          F_FCOE),
  83         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_VLAN | F_PORT |
  84          F_FCOE),
  85         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_VNIC_ID | F_PORT |
  86          F_FCOE),
  87         (F_FRAGMENTATION | F_MPSHITTYPE | F_TOS | F_VLAN | F_PORT | F_FCOE),
  88         (F_FRAGMENTATION | F_MPSHITTYPE | F_TOS | F_VNIC_ID | F_PORT | F_FCOE),
  89         (F_FRAGMENTATION | F_MPSHITTYPE | F_VLAN | F_VNIC_ID | F_FCOE),
  90         (F_FRAGMENTATION | F_MACMATCH | F_ETHERTYPE | F_PROTOCOL | F_PORT |
  91          F_FCOE),
  92         (F_FRAGMENTATION | F_MACMATCH | F_ETHERTYPE | F_TOS | F_PORT | F_FCOE),
  93         (F_FRAGMENTATION | F_MACMATCH | F_PROTOCOL | F_VLAN | F_PORT | F_FCOE),
  94         (F_FRAGMENTATION | F_MACMATCH | F_PROTOCOL | F_VNIC_ID | F_PORT |
  95          F_FCOE),
  96         (F_FRAGMENTATION | F_MACMATCH | F_TOS | F_VLAN | F_PORT | F_FCOE),
  97         (F_FRAGMENTATION | F_MACMATCH | F_TOS | F_VNIC_ID | F_PORT | F_FCOE),
  98         (F_FRAGMENTATION | F_ETHERTYPE | F_VLAN | F_PORT | F_FCOE),
  99         (F_FRAGMENTATION | F_ETHERTYPE | F_VNIC_ID | F_PORT | F_FCOE),
 100         (F_FRAGMENTATION | F_PROTOCOL | F_TOS | F_VLAN | F_FCOE),
 101         (F_FRAGMENTATION | F_PROTOCOL | F_TOS | F_VNIC_ID | F_FCOE),
 102         (F_FRAGMENTATION | F_VLAN | F_VNIC_ID | F_PORT | F_FCOE),
 103         (F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_PROTOCOL | F_PORT |
 104          F_FCOE),
 105         (F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_TOS | F_PORT | F_FCOE),
 106         (F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VLAN | F_PORT),
 107         (F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VNIC_ID | F_PORT),
 108         (F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VLAN | F_PORT),
 109         (F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VNIC_ID | F_PORT),
 110         (F_MPSHITTYPE | F_ETHERTYPE | F_VLAN | F_PORT | F_FCOE),
 111         (F_MPSHITTYPE | F_ETHERTYPE | F_VNIC_ID | F_PORT | F_FCOE),
 112         (F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VLAN | F_PORT | F_FCOE),
 113         (F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VNIC_ID | F_PORT | F_FCOE),
 114         (F_MPSHITTYPE | F_VLAN | F_VNIC_ID | F_PORT),
 115 };
 116
 117 /**
 118  * Allocate a chunk of memory. The allocated memory is cleared.
 119  */
 120 void *t4_alloc_mem(size_t size)
 121 {
 122         return rte_zmalloc(NULL, size, 0);
 123 }
 124
 125 /**
 126  * Free memory allocated through t4_alloc_mem().
 127  */
 128 void t4_free_mem(void *addr)
 129 {
 130         rte_free(addr);
 131 }
 132
 133 /*
 134  * Response queue handler for the FW event queue.
 135  */
 136 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
 137                           __rte_unused const struct pkt_gl *gl)
 138 {
 139         u8 opcode = ((const struct rss_header *)rsp)->opcode;
 140
 141         rsp++;                                          /* skip RSS header */
 142
 143         /*
 144          * FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
 145          */
 146         if (unlikely(opcode == CPL_FW4_MSG &&
 147                      ((const struct cpl_fw4_msg *)rsp)->type ==
 148                       FW_TYPE_RSSCPL)) {
 149                 rsp++;
 150                 opcode = ((const struct rss_header *)rsp)->opcode;
 151                 rsp++;
 152                 if (opcode != CPL_SGE_EGR_UPDATE) {
 153                         dev_err(q->adapter, "unexpected FW4/CPL %#x on FW event queue\n",
 154                                 opcode);
 155                         goto out;
 156                 }
 157         }
 158
 159         if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
 160                 /* do nothing */
 161         } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
 162                 const struct cpl_fw6_msg *msg = (const void *)rsp;
 163
 164                 t4_handle_fw_rpl(q->adapter, msg->data);
 165         } else if (opcode == CPL_ABORT_RPL_RSS) {
 166                 const struct cpl_abort_rpl_rss *p = (const void *)rsp;
 167
 168                 cxgbe_hash_del_filter_rpl(q->adapter, p);
 169         } else if (opcode == CPL_SET_TCB_RPL) {
 170                 const struct cpl_set_tcb_rpl *p = (const void *)rsp;
 171
 172                 cxgbe_filter_rpl(q->adapter, p);
 173         } else if (opcode == CPL_ACT_OPEN_RPL) {
 174                 const struct cpl_act_open_rpl *p = (const void *)rsp;
 175
 176                 cxgbe_hash_filter_rpl(q->adapter, p);
 177         } else if (opcode == CPL_L2T_WRITE_RPL) {
 178                 const struct cpl_l2t_write_rpl *p = (const void *)rsp;
 179
 180                 cxgbe_do_l2t_write_rpl(q->adapter, p);
 181         } else if (opcode == CPL_SMT_WRITE_RPL) {
 182                 const struct cpl_smt_write_rpl *p = (const void *)rsp;
 183
 184                 cxgbe_do_smt_write_rpl(q->adapter, p);
 185         } else {
 186                 dev_err(adapter, "unexpected CPL %#x on FW event queue\n",
 187                         opcode);
 188         }
 189 out:
 190         return 0;
 191 }
 192
 193 /**
 194  * Setup sge control queues to pass control information.
 195  */
 196 int cxgbe_setup_sge_ctrl_txq(struct adapter *adapter)
 197 {
 198         struct sge *s = &adapter->sge;
 199         int err = 0, i = 0;
 200
 201         for_each_port(adapter, i) {
 202                 struct port_info *pi = adap2pinfo(adapter, i);
 203                 char name[RTE_ETH_NAME_MAX_LEN];
 204                 struct sge_ctrl_txq *q = &s->ctrlq[i];
 205
 206                 q->q.size = 1024;
 207                 err = t4_sge_alloc_ctrl_txq(adapter, q,
 208                                             adapter->eth_dev,  i,
 209                                             s->fw_evtq.cntxt_id,
 210                                             rte_socket_id());
 211                 if (err) {
 212                         dev_err(adapter, "Failed to alloc ctrl txq. Err: %d",
 213                                 err);
 214                         goto out;
 215                 }
 216                 snprintf(name, sizeof(name), "%s_ctrl_pool_%d",
 217                          pi->eth_dev->device->driver->name,
 218                          pi->eth_dev->data->port_id);
 219                 q->mb_pool = rte_pktmbuf_pool_create(name, s->ctrlq[i].q.size,
 220                                                      RTE_CACHE_LINE_SIZE,
 221                                                      RTE_MBUF_PRIV_ALIGN,
 222                                                      RTE_MBUF_DEFAULT_BUF_SIZE,
 223                                                      SOCKET_ID_ANY);
 224                 if (!q->mb_pool) {
 225                         err = -rte_errno;
 226                         dev_err(adapter,
 227                                 "Can't create ctrl pool for port %d. Err: %d\n",
 228                                 pi->eth_dev->data->port_id, err);
 229                         goto out;
 230                 }
 231         }
 232         return 0;
 233 out:
 234         t4_free_sge_resources(adapter);
 235         return err;
 236 }
 237
 238 /**
 239  * cxgbe_poll_for_completion: Poll rxq for completion
 240  * @q: rxq to poll
 241  * @ms: milliseconds to delay
 242  * @cnt: number of times to poll
 243  * @c: completion to check for 'done' status
 244  *
 245  * Polls the rxq for reples until completion is done or the count
 246  * expires.
 247  */
 248 int cxgbe_poll_for_completion(struct sge_rspq *q, unsigned int ms,
 249                               unsigned int cnt, struct t4_completion *c)
 250 {
 251         unsigned int i;
 252         unsigned int work_done, budget = 32;
 253
 254         if (!c)
 255                 return -EINVAL;
 256
 257         for (i = 0; i < cnt; i++) {
 258                 cxgbe_poll(q, NULL, budget, &work_done);
 259                 t4_os_lock(&c->lock);
 260                 if (c->done) {
 261                         t4_os_unlock(&c->lock);
 262                         return 0;
 263                 }
 264                 t4_os_unlock(&c->lock);
 265                 rte_delay_ms(ms);
 266         }
 267         return -ETIMEDOUT;
 268 }
 269
 270 int cxgbe_setup_sge_fwevtq(struct adapter *adapter)
 271 {
 272         struct sge *s = &adapter->sge;
 273         int err = 0;
 274         int msi_idx = 0;
 275
 276         err = t4_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->eth_dev,
 277                                msi_idx, NULL, fwevtq_handler, -1, NULL, 0,
 278                                rte_socket_id());
 279         return err;
 280 }
 281
 282 static int closest_timer(const struct sge *s, int time)
 283 {
 284         unsigned int i, match = 0;
 285         int delta, min_delta = INT_MAX;
 286
 287         for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
 288                 delta = time - s->timer_val[i];
 289                 if (delta < 0)
 290                         delta = -delta;
 291                 if (delta < min_delta) {
 292                         min_delta = delta;
 293                         match = i;
 294                 }
 295         }
 296         return match;
 297 }
 298
 299 static int closest_thres(const struct sge *s, int thres)
 300 {
 301         unsigned int i, match = 0;
 302         int delta, min_delta = INT_MAX;
 303
 304         for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
 305                 delta = thres - s->counter_val[i];
 306                 if (delta < 0)
 307                         delta = -delta;
 308                 if (delta < min_delta) {
 309                         min_delta = delta;
 310                         match = i;
 311                 }
 312         }
 313         return match;
 314 }
 315
 316 /**
 317  * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
 318  * @q: the Rx queue
 319  * @us: the hold-off time in us, or 0 to disable timer
 320  * @cnt: the hold-off packet count, or 0 to disable counter
 321  *
 322  * Sets an Rx queue's interrupt hold-off time and packet count.  At least
 323  * one of the two needs to be enabled for the queue to generate interrupts.
 324  */
 325 int cxgb4_set_rspq_intr_params(struct sge_rspq *q, unsigned int us,
 326                                unsigned int cnt)
 327 {
 328         struct adapter *adap = q->adapter;
 329         unsigned int timer_val;
 330
 331         if (cnt) {
 332                 int err;
 333                 u32 v, new_idx;
 334
 335                 new_idx = closest_thres(&adap->sge, cnt);
 336                 if (q->desc && q->pktcnt_idx != new_idx) {
 337                         /* the queue has already been created, update it */
 338                         v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
 339                             V_FW_PARAMS_PARAM_X(
 340                             FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
 341                             V_FW_PARAMS_PARAM_YZ(q->cntxt_id);
 342                         err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
 343                                             &v, &new_idx);
 344                         if (err)
 345                                 return err;
 346                 }
 347                 q->pktcnt_idx = new_idx;
 348         }
 349
 350         timer_val = (us == 0) ? X_TIMERREG_RESTART_COUNTER :
 351                                 closest_timer(&adap->sge, us);
 352
 353         if ((us | cnt) == 0)
 354                 q->intr_params = V_QINTR_TIMER_IDX(X_TIMERREG_UPDATE_CIDX);
 355         else
 356                 q->intr_params = V_QINTR_TIMER_IDX(timer_val) |
 357                                  V_QINTR_CNT_EN(cnt > 0);
 358         return 0;
 359 }
 360
 361 /**
 362  * Allocate an active-open TID and set it to the supplied value.
 363  */
 364 int cxgbe_alloc_atid(struct tid_info *t, void *data)
 365 {
 366         int atid = -1;
 367
 368         t4_os_lock(&t->atid_lock);
 369         if (t->afree) {
 370                 union aopen_entry *p = t->afree;
 371
 372                 atid = p - t->atid_tab;
 373                 t->afree = p->next;
 374                 p->data = data;
 375                 t->atids_in_use++;
 376         }
 377         t4_os_unlock(&t->atid_lock);
 378         return atid;
 379 }
 380
 381 /**
 382  * Release an active-open TID.
 383  */
 384 void cxgbe_free_atid(struct tid_info *t, unsigned int atid)
 385 {
 386         union aopen_entry *p = &t->atid_tab[atid];
 387
 388         t4_os_lock(&t->atid_lock);
 389         p->next = t->afree;
 390         t->afree = p;
 391         t->atids_in_use--;
 392         t4_os_unlock(&t->atid_lock);
 393 }
 394
 395 /**
 396  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
 397  */
 398 static void mk_tid_release(struct rte_mbuf *mbuf, unsigned int tid)
 399 {
 400         struct cpl_tid_release *req;
 401
 402         req = rte_pktmbuf_mtod(mbuf, struct cpl_tid_release *);
 403         INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
 404 }
 405
 406 /**
 407  * Release a TID and inform HW.  If we are unable to allocate the release
 408  * message we defer to a work queue.
 409  */
 410 void cxgbe_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
 411                       unsigned short family)
 412 {
 413         struct rte_mbuf *mbuf;
 414         struct adapter *adap = container_of(t, struct adapter, tids);
 415
 416         WARN_ON(tid >= t->ntids);
 417
 418         if (t->tid_tab[tid]) {
 419                 t->tid_tab[tid] = NULL;
 420                 rte_atomic32_dec(&t->conns_in_use);
 421                 if (t->hash_base && tid >= t->hash_base) {
 422                         if (family == FILTER_TYPE_IPV4)
 423                                 rte_atomic32_dec(&t->hash_tids_in_use);
 424                 } else {
 425                         if (family == FILTER_TYPE_IPV4)
 426                                 rte_atomic32_dec(&t->tids_in_use);
 427                 }
 428         }
 429
 430         mbuf = rte_pktmbuf_alloc((&adap->sge.ctrlq[chan])->mb_pool);
 431         if (mbuf) {
 432                 mbuf->data_len = sizeof(struct cpl_tid_release);
 433                 mbuf->pkt_len = mbuf->data_len;
 434                 mk_tid_release(mbuf, tid);
 435                 t4_mgmt_tx(&adap->sge.ctrlq[chan], mbuf);
 436         }
 437 }
 438
 439 /**
 440  * Insert a TID.
 441  */
 442 void cxgbe_insert_tid(struct tid_info *t, void *data, unsigned int tid,
 443                       unsigned short family)
 444 {
 445         t->tid_tab[tid] = data;
 446         if (t->hash_base && tid >= t->hash_base) {
 447                 if (family == FILTER_TYPE_IPV4)
 448                         rte_atomic32_inc(&t->hash_tids_in_use);
 449         } else {
 450                 if (family == FILTER_TYPE_IPV4)
 451                         rte_atomic32_inc(&t->tids_in_use);
 452         }
 453
 454         rte_atomic32_inc(&t->conns_in_use);
 455 }
 456
 457 /**
 458  * Free TID tables.
 459  */
 460 static void tid_free(struct tid_info *t)
 461 {
 462         if (t->tid_tab) {
 463                 if (t->ftid_bmap)
 464                         rte_bitmap_free(t->ftid_bmap);
 465
 466                 if (t->ftid_bmap_array)
 467                         t4_os_free(t->ftid_bmap_array);
 468
 469                 t4_os_free(t->tid_tab);
 470         }
 471
 472         memset(t, 0, sizeof(struct tid_info));
 473 }
 474
 475 /**
 476  * Allocate and initialize the TID tables.  Returns 0 on success.
 477  */
 478 static int tid_init(struct tid_info *t)
 479 {
 480         size_t size;
 481         unsigned int ftid_bmap_size;
 482         unsigned int natids = t->natids;
 483         unsigned int max_ftids = t->nftids;
 484
 485         ftid_bmap_size = rte_bitmap_get_memory_footprint(t->nftids);
 486         size = t->ntids * sizeof(*t->tid_tab) +
 487                 max_ftids * sizeof(*t->ftid_tab) +
 488                 natids * sizeof(*t->atid_tab);
 489
 490         t->tid_tab = t4_os_alloc(size);
 491         if (!t->tid_tab)
 492                 return -ENOMEM;
 493
 494         t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
 495         t->ftid_tab = (struct filter_entry *)&t->atid_tab[t->natids];
 496         t->ftid_bmap_array = t4_os_alloc(ftid_bmap_size);
 497         if (!t->ftid_bmap_array) {
 498                 tid_free(t);
 499                 return -ENOMEM;
 500         }
 501
 502         t4_os_lock_init(&t->atid_lock);
 503         t4_os_lock_init(&t->ftid_lock);
 504
 505         t->afree = NULL;
 506         t->atids_in_use = 0;
 507         rte_atomic32_init(&t->tids_in_use);
 508         rte_atomic32_set(&t->tids_in_use, 0);
 509         rte_atomic32_init(&t->conns_in_use);
 510         rte_atomic32_set(&t->conns_in_use, 0);
 511
 512         /* Setup the free list for atid_tab and clear the stid bitmap. */
 513         if (natids) {
 514                 while (--natids)
 515                         t->atid_tab[natids - 1].next = &t->atid_tab[natids];
 516                 t->afree = t->atid_tab;
 517         }
 518
 519         t->ftid_bmap = rte_bitmap_init(t->nftids, t->ftid_bmap_array,
 520                                        ftid_bmap_size);
 521         if (!t->ftid_bmap) {
 522                 tid_free(t);
 523                 return -ENOMEM;
 524         }
 525
 526         return 0;
 527 }
 528
 529 static inline bool is_x_1g_port(const struct link_config *lc)
 530 {
 531         return (lc->pcaps & FW_PORT_CAP32_SPEED_1G) != 0;
 532 }
 533
 534 static inline bool is_x_10g_port(const struct link_config *lc)
 535 {
 536         unsigned int speeds, high_speeds;
 537
 538         speeds = V_FW_PORT_CAP32_SPEED(G_FW_PORT_CAP32_SPEED(lc->pcaps));
 539         high_speeds = speeds &
 540                       ~(FW_PORT_CAP32_SPEED_100M | FW_PORT_CAP32_SPEED_1G);
 541
 542         return high_speeds != 0;
 543 }
 544
 545 static inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
 546                       unsigned int us, unsigned int cnt,
 547                       unsigned int size, unsigned int iqe_size)
 548 {
 549         q->adapter = adap;
 550         cxgb4_set_rspq_intr_params(q, us, cnt);
 551         q->iqe_len = iqe_size;
 552         q->size = size;
 553 }
 554
 555 int cxgbe_cfg_queue_count(struct rte_eth_dev *eth_dev)
 556 {
 557         struct port_info *pi = eth_dev->data->dev_private;
 558         struct adapter *adap = pi->adapter;
 559         struct sge *s = &adap->sge;
 560         unsigned int max_queues = s->max_ethqsets / adap->params.nports;
 561
 562         if ((eth_dev->data->nb_rx_queues < 1) ||
 563             (eth_dev->data->nb_tx_queues < 1))
 564                 return -EINVAL;
 565
 566         if ((eth_dev->data->nb_rx_queues > max_queues) ||
 567             (eth_dev->data->nb_tx_queues > max_queues))
 568                 return -EINVAL;
 569
 570         if (eth_dev->data->nb_rx_queues > pi->rss_size)
 571                 return -EINVAL;
 572
 573         /* We must configure RSS, since config has changed*/
 574         pi->flags &= ~PORT_RSS_DONE;
 575
 576         pi->n_rx_qsets = eth_dev->data->nb_rx_queues;
 577         pi->n_tx_qsets = eth_dev->data->nb_tx_queues;
 578
 579         return 0;
 580 }
 581
 582 void cxgbe_cfg_queues(struct rte_eth_dev *eth_dev)
 583 {
 584         struct port_info *pi = eth_dev->data->dev_private;
 585         struct adapter *adap = pi->adapter;
 586         struct sge *s = &adap->sge;
 587         unsigned int i, nb_ports = 0, qidx = 0;
 588         unsigned int q_per_port = 0;
 589
 590         if (!(adap->flags & CFG_QUEUES)) {
 591                 for_each_port(adap, i) {
 592                         struct port_info *tpi = adap2pinfo(adap, i);
 593
 594                         nb_ports += (is_x_10g_port(&tpi->link_cfg)) ||
 595                                      is_x_1g_port(&tpi->link_cfg) ? 1 : 0;
 596                 }
 597
 598                 /*
 599                  * We default up to # of cores queues per 1G/10G port.
 600                  */
 601                 if (nb_ports)
 602                         q_per_port = (s->max_ethqsets -
 603                                      (adap->params.nports - nb_ports)) /
 604                                      nb_ports;
 605
 606                 if (q_per_port > rte_lcore_count())
 607                         q_per_port = rte_lcore_count();
 608
 609                 for_each_port(adap, i) {
 610                         struct port_info *pi = adap2pinfo(adap, i);
 611
 612                         pi->first_qset = qidx;
 613
 614                         /* Initially n_rx_qsets == n_tx_qsets */
 615                         pi->n_rx_qsets = (is_x_10g_port(&pi->link_cfg) ||
 616                                           is_x_1g_port(&pi->link_cfg)) ?
 617                                           q_per_port : 1;
 618                         pi->n_tx_qsets = pi->n_rx_qsets;
 619
 620                         if (pi->n_rx_qsets > pi->rss_size)
 621                                 pi->n_rx_qsets = pi->rss_size;
 622
 623                         qidx += pi->n_rx_qsets;
 624                 }
 625
 626                 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
 627                         struct sge_eth_rxq *r = &s->ethrxq[i];
 628
 629                         init_rspq(adap, &r->rspq, 5, 32, 1024, 64);
 630                         r->usembufs = 1;
 631                         r->fl.size = (r->usembufs ? 1024 : 72);
 632                 }
 633
 634                 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
 635                         s->ethtxq[i].q.size = 1024;
 636
 637                 init_rspq(adap, &adap->sge.fw_evtq, 0, 0, 1024, 64);
 638                 adap->flags |= CFG_QUEUES;
 639         }
 640 }
 641
 642 void cxgbe_stats_get(struct port_info *pi, struct port_stats *stats)
 643 {
 644         t4_get_port_stats_offset(pi->adapter, pi->tx_chan, stats,
 645                                  &pi->stats_base);
 646 }
 647
 648 void cxgbe_stats_reset(struct port_info *pi)
 649 {
 650         t4_clr_port_stats(pi->adapter, pi->tx_chan);
 651 }
 652
 653 static void setup_memwin(struct adapter *adap)
 654 {
 655         u32 mem_win0_base;
 656
 657         /* For T5, only relative offset inside the PCIe BAR is passed */
 658         mem_win0_base = MEMWIN0_BASE;
 659
 660         /*
 661          * Set up memory window for accessing adapter memory ranges.  (Read
 662          * back MA register to ensure that changes propagate before we attempt
 663          * to use the new values.)
 664          */
 665         t4_write_reg(adap,
 666                      PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
 667                                          MEMWIN_NIC),
 668                      mem_win0_base | V_BIR(0) |
 669                      V_WINDOW(ilog2(MEMWIN0_APERTURE) - X_WINDOW_SHIFT));
 670         t4_read_reg(adap,
 671                     PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
 672                                         MEMWIN_NIC));
 673 }
 674
 675 int cxgbe_init_rss(struct adapter *adap)
 676 {
 677         unsigned int i;
 678
 679         if (is_pf4(adap)) {
 680                 int err;
 681
 682                 err = t4_init_rss_mode(adap, adap->mbox);
 683                 if (err)
 684                         return err;
 685         }
 686
 687         for_each_port(adap, i) {
 688                 struct port_info *pi = adap2pinfo(adap, i);
 689
 690                 pi->rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
 691                 if (!pi->rss)
 692                         return -ENOMEM;
 693
 694                 pi->rss_hf = CXGBE_RSS_HF_ALL;
 695         }
 696         return 0;
 697 }
 698
 699 /**
 700  * Dump basic information about the adapter.
 701  */
 702 void cxgbe_print_adapter_info(struct adapter *adap)
 703 {
 704         /**
 705          * Hardware/Firmware/etc. Version/Revision IDs.
 706          */
 707         t4_dump_version_info(adap);
 708 }
 709
 710 void cxgbe_print_port_info(struct adapter *adap)
 711 {
 712         int i;
 713         char buf[80];
 714         struct rte_pci_addr *loc = &adap->pdev->addr;
 715
 716         for_each_port(adap, i) {
 717                 const struct port_info *pi = adap2pinfo(adap, i);
 718                 char *bufp = buf;
 719
 720                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
 721                         bufp += sprintf(bufp, "100M/");
 722                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
 723                         bufp += sprintf(bufp, "1G/");
 724                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
 725                         bufp += sprintf(bufp, "10G/");
 726                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
 727                         bufp += sprintf(bufp, "25G/");
 728                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
 729                         bufp += sprintf(bufp, "40G/");
 730                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
 731                         bufp += sprintf(bufp, "50G/");
 732                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
 733                         bufp += sprintf(bufp, "100G/");
 734                 if (bufp != buf)
 735                         --bufp;
 736                 sprintf(bufp, "BASE-%s",
 737                         t4_get_port_type_description(
 738                                         (enum fw_port_type)pi->port_type));
 739
 740                 dev_info(adap,
 741                          " " PCI_PRI_FMT " Chelsio rev %d %s %s\n",
 742                          loc->domain, loc->bus, loc->devid, loc->function,
 743                          CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
 744                          (adap->flags & USING_MSIX) ? " MSI-X" :
 745                          (adap->flags & USING_MSI) ? " MSI" : "");
 746         }
 747 }
 748
 749 static int check_devargs_handler(const char *key, const char *value, void *p)
 750 {
 751         if (!strncmp(key, CXGBE_DEVARG_CMN_KEEP_OVLAN, strlen(key)) ||
 752             !strncmp(key, CXGBE_DEVARG_CMN_TX_MODE_LATENCY, strlen(key)) ||
 753             !strncmp(key, CXGBE_DEVARG_VF_FORCE_LINK_UP, strlen(key))) {
 754                 if (!strncmp(value, "1", 1)) {
 755                         bool *dst_val = (bool *)p;
 756
 757                         *dst_val = true;
 758                 }
 759         }
 760
 761         if (!strncmp(key, CXGBE_DEVARG_PF_FILTER_MODE, strlen(key)) ||
 762             !strncmp(key, CXGBE_DEVARG_PF_FILTER_MASK, strlen(key))) {
 763                 u32 *dst_val = (u32 *)p;
 764                 char *endptr = NULL;
 765                 u32 arg_val;
 766
 767                 arg_val = strtoul(value, &endptr, 16);
 768                 if (errno || endptr == value)
 769                         return -EINVAL;
 770
 771                 *dst_val = arg_val;
 772         }
 773
 774         return 0;
 775 }
 776
 777 static int cxgbe_get_devargs(struct rte_devargs *devargs, const char *key,
 778                              void *p)
 779 {
 780         struct rte_kvargs *kvlist;
 781         int ret = 0;
 782
 783         if (!devargs)
 784                 return 0;
 785
 786         kvlist = rte_kvargs_parse(devargs->args, NULL);
 787         if (!kvlist)
 788                 return 0;
 789
 790         if (!rte_kvargs_count(kvlist, key))
 791                 goto out;
 792
 793         ret = rte_kvargs_process(kvlist, key, check_devargs_handler, p);
 794
 795 out:
 796         rte_kvargs_free(kvlist);
 797
 798         return ret;
 799 }
 800
 801 static void cxgbe_get_devargs_int(struct adapter *adap, bool *dst,
 802                                   const char *key, bool default_value)
 803 {
 804         struct rte_pci_device *pdev = adap->pdev;
 805         int ret;
 806         bool devarg_value = default_value;
 807
 808         *dst = default_value;
 809         if (!pdev)
 810                 return;
 811
 812         ret = cxgbe_get_devargs(pdev->device.devargs, key, &devarg_value);
 813         if (ret)
 814                 return;
 815
 816         *dst = devarg_value;
 817 }
 818
 819 static void cxgbe_get_devargs_u32(struct adapter *adap, u32 *dst,
 820                                   const char *key, u32 default_value)
 821 {
 822         struct rte_pci_device *pdev = adap->pdev;
 823         u32 devarg_value = default_value;
 824         int ret;
 825
 826         *dst = default_value;
 827         if (!pdev)
 828                 return;
 829
 830         ret = cxgbe_get_devargs(pdev->device.devargs, key, &devarg_value);
 831         if (ret)
 832                 return;
 833
 834         *dst = devarg_value;
 835 }
 836
 837 void cxgbe_process_devargs(struct adapter *adap)
 838 {
 839         cxgbe_get_devargs_int(adap, &adap->devargs.keep_ovlan,
 840                               CXGBE_DEVARG_CMN_KEEP_OVLAN, false);
 841         cxgbe_get_devargs_int(adap, &adap->devargs.tx_mode_latency,
 842                               CXGBE_DEVARG_CMN_TX_MODE_LATENCY, false);
 843         cxgbe_get_devargs_int(adap, &adap->devargs.force_link_up,
 844                               CXGBE_DEVARG_VF_FORCE_LINK_UP, false);
 845         cxgbe_get_devargs_u32(adap, &adap->devargs.filtermode,
 846                               CXGBE_DEVARG_PF_FILTER_MODE, 0);
 847         cxgbe_get_devargs_u32(adap, &adap->devargs.filtermask,
 848                               CXGBE_DEVARG_PF_FILTER_MASK, 0);
 849 }
 850
 851 static void configure_vlan_types(struct adapter *adapter)
 852 {
 853         int i;
 854
 855         for_each_port(adapter, i) {
 856                 /* OVLAN Type 0x88a8 */
 857                 t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN0),
 858                                  V_OVLAN_MASK(M_OVLAN_MASK) |
 859                                  V_OVLAN_ETYPE(M_OVLAN_ETYPE),
 860                                  V_OVLAN_MASK(M_OVLAN_MASK) |
 861                                  V_OVLAN_ETYPE(0x88a8));
 862                 /* OVLAN Type 0x9100 */
 863                 t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN1),
 864                                  V_OVLAN_MASK(M_OVLAN_MASK) |
 865                                  V_OVLAN_ETYPE(M_OVLAN_ETYPE),
 866                                  V_OVLAN_MASK(M_OVLAN_MASK) |
 867                                  V_OVLAN_ETYPE(0x9100));
 868
 869                 /* IVLAN 0X8100 */
 870                 t4_set_reg_field(adapter, MPS_PORT_RX_IVLAN(i),
 871                                  V_IVLAN_ETYPE(M_IVLAN_ETYPE),
 872                                  V_IVLAN_ETYPE(0x8100));
 873
 874                 t4_set_reg_field(adapter, MPS_PORT_RX_CTL(i),
 875                                  F_OVLAN_EN0 | F_OVLAN_EN1 |
 876                                  F_IVLAN_EN,
 877                                  F_OVLAN_EN0 | F_OVLAN_EN1 |
 878                                  F_IVLAN_EN);
 879         }
 880
 881         t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG, V_RM_OVLAN(1),
 882                                V_RM_OVLAN(!adapter->devargs.keep_ovlan));
 883 }
 884
 885 static int cxgbe_get_filter_vnic_mode_from_devargs(u32 val)
 886 {
 887         u32 vnic_mode;
 888
 889         vnic_mode = val & (CXGBE_DEVARGS_FILTER_MODE_PF_VF |
 890                            CXGBE_DEVARGS_FILTER_MODE_VLAN_OUTER);
 891         if (vnic_mode) {
 892                 switch (vnic_mode) {
 893                 case CXGBE_DEVARGS_FILTER_MODE_VLAN_OUTER:
 894                         return CXGBE_FILTER_VNIC_MODE_OVLAN;
 895                 case CXGBE_DEVARGS_FILTER_MODE_PF_VF:
 896                         return CXGBE_FILTER_VNIC_MODE_PFVF;
 897                 default:
 898                         return -EINVAL;
 899                 }
 900         }
 901
 902         return CXGBE_FILTER_VNIC_MODE_NONE;
 903 }
 904
 905 static int cxgbe_get_filter_mode_from_devargs(u32 val, bool closest_match)
 906 {
 907         int vnic_mode, fmode = 0;
 908         bool found = false;
 909         u8 i;
 910
 911         if (val >= CXGBE_DEVARGS_FILTER_MODE_MAX) {
 912                 pr_err("Unsupported flags set in filter mode. Must be < 0x%x\n",
 913                        CXGBE_DEVARGS_FILTER_MODE_MAX);
 914                 return -ERANGE;
 915         }
 916
 917         vnic_mode = cxgbe_get_filter_vnic_mode_from_devargs(val);
 918         if (vnic_mode < 0) {
 919                 pr_err("Unsupported Vnic-mode, more than 1 Vnic-mode selected\n");
 920                 return vnic_mode;
 921         }
 922
 923         if (vnic_mode)
 924                 fmode |= F_VNIC_ID;
 925         if (val & CXGBE_DEVARGS_FILTER_MODE_PHYSICAL_PORT)
 926                 fmode |= F_PORT;
 927         if (val & CXGBE_DEVARGS_FILTER_MODE_ETHERNET_DSTMAC)
 928                 fmode |= F_MACMATCH;
 929         if (val & CXGBE_DEVARGS_FILTER_MODE_ETHERNET_ETHTYPE)
 930                 fmode |= F_ETHERTYPE;
 931         if (val & CXGBE_DEVARGS_FILTER_MODE_VLAN_INNER)
 932                 fmode |= F_VLAN;
 933         if (val & CXGBE_DEVARGS_FILTER_MODE_IP_TOS)
 934                 fmode |= F_TOS;
 935         if (val & CXGBE_DEVARGS_FILTER_MODE_IP_PROTOCOL)
 936                 fmode |= F_PROTOCOL;
 937
 938         for (i = 0; i < ARRAY_SIZE(cxgbe_filter_mode_features); i++) {
 939                 if ((cxgbe_filter_mode_features[i] & fmode) == fmode) {
 940                         found = true;
 941                         break;
 942                 }
 943         }
 944
 945         if (!found)
 946                 return -EINVAL;
 947
 948         return closest_match ? cxgbe_filter_mode_features[i] : fmode;
 949 }
 950
 951 static int configure_filter_mode_mask(struct adapter *adap)
 952 {
 953         u32 params[2], val[2], nparams = 0;
 954         int ret;
 955
 956         if (!adap->devargs.filtermode && !adap->devargs.filtermask)
 957                 return 0;
 958
 959         if (!adap->devargs.filtermode || !adap->devargs.filtermask) {
 960                 pr_err("Unsupported, Provide both filtermode and filtermask devargs\n");
 961                 return -EINVAL;
 962         }
 963
 964         if (adap->devargs.filtermask & ~adap->devargs.filtermode) {
 965                 pr_err("Unsupported, filtermask (0x%x) must be subset of filtermode (0x%x)\n",
 966                        adap->devargs.filtermask, adap->devargs.filtermode);
 967
 968                 return -EINVAL;
 969         }
 970
 971         params[0] = CXGBE_FW_PARAM_DEV(FILTER) |
 972                     V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_FILTER_MODE_MASK);
 973
 974         ret = cxgbe_get_filter_mode_from_devargs(adap->devargs.filtermode,
 975                                                  true);
 976         if (ret < 0) {
 977                 pr_err("Unsupported filtermode devargs combination:0x%x\n",
 978                        adap->devargs.filtermode);
 979                 return ret;
 980         }
 981
 982         val[0] = V_FW_PARAMS_PARAM_FILTER_MODE(ret);
 983
 984         ret = cxgbe_get_filter_mode_from_devargs(adap->devargs.filtermask,
 985                                                  false);
 986         if (ret < 0) {
 987                 pr_err("Unsupported filtermask devargs combination:0x%x\n",
 988                        adap->devargs.filtermask);
 989                 return ret;
 990         }
 991
 992         val[0] |= V_FW_PARAMS_PARAM_FILTER_MASK(ret);
 993
 994         nparams++;
 995
 996         ret = cxgbe_get_filter_vnic_mode_from_devargs(adap->devargs.filtermode);
 997         if (ret < 0)
 998                 return ret;
 999
1000         if (ret) {
1001                 params[1] = CXGBE_FW_PARAM_DEV(FILTER) |
1002                             V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_FILTER_VNIC_MODE);
1003
1004                 val[1] = ret - 1;
1005
1006                 nparams++;
1007         }
1008
1009         return t4_set_params(adap, adap->mbox, adap->pf, 0, nparams,
1010                              params, val);
1011 }
1012
1013 static void configure_pcie_ext_tag(struct adapter *adapter)
1014 {
1015         u16 v;
1016         int pos = t4_os_find_pci_capability(adapter, PCI_CAP_ID_EXP);
1017
1018         if (!pos)
1019                 return;
1020
1021         if (pos > 0) {
1022                 t4_os_pci_read_cfg2(adapter, pos + PCI_EXP_DEVCTL, &v);
1023                 v |= PCI_EXP_DEVCTL_EXT_TAG;
1024                 t4_os_pci_write_cfg2(adapter, pos + PCI_EXP_DEVCTL, v);
1025                 if (is_t6(adapter->params.chip)) {
1026                         t4_set_reg_field(adapter, A_PCIE_CFG2,
1027                                          V_T6_TOTMAXTAG(M_T6_TOTMAXTAG),
1028                                          V_T6_TOTMAXTAG(7));
1029                         t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
1030                                          V_T6_MINTAG(M_T6_MINTAG),
1031                                          V_T6_MINTAG(8));
1032                 } else {
1033                         t4_set_reg_field(adapter, A_PCIE_CFG2,
1034                                          V_TOTMAXTAG(M_TOTMAXTAG),
1035                                          V_TOTMAXTAG(3));
1036                         t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
1037                                          V_MINTAG(M_MINTAG),
1038                                          V_MINTAG(8));
1039                 }
1040         }
1041 }
1042
1043 /* Figure out how many Queue Sets we can support */
1044 void cxgbe_configure_max_ethqsets(struct adapter *adapter)
1045 {
1046         unsigned int ethqsets;
1047
1048         /*
1049          * We need to reserve an Ingress Queue for the Asynchronous Firmware
1050          * Event Queue.
1051          *
1052          * For each Queue Set, we'll need the ability to allocate two Egress
1053          * Contexts -- one for the Ingress Queue Free List and one for the TX
1054          * Ethernet Queue.
1055          */
1056         if (is_pf4(adapter)) {
1057                 struct pf_resources *pfres = &adapter->params.pfres;
1058
1059                 ethqsets = pfres->niqflint - 1;
1060                 if (pfres->neq < ethqsets * 2)
1061                         ethqsets = pfres->neq / 2;
1062         } else {
1063                 struct vf_resources *vfres = &adapter->params.vfres;
1064
1065                 ethqsets = vfres->niqflint - 1;
1066                 if (vfres->nethctrl != ethqsets)
1067                         ethqsets = min(vfres->nethctrl, ethqsets);
1068                 if (vfres->neq < ethqsets * 2)
1069                         ethqsets = vfres->neq / 2;
1070         }
1071
1072         if (ethqsets > MAX_ETH_QSETS)
1073                 ethqsets = MAX_ETH_QSETS;
1074         adapter->sge.max_ethqsets = ethqsets;
1075 }
1076
1077 /*
1078  * Tweak configuration based on system architecture, etc.  Most of these have
1079  * defaults assigned to them by Firmware Configuration Files (if we're using
1080  * them) but need to be explicitly set if we're using hard-coded
1081  * initialization. So these are essentially common tweaks/settings for
1082  * Configuration Files and hard-coded initialization ...
1083  */
1084 static int adap_init0_tweaks(struct adapter *adapter)
1085 {
1086         u8 rx_dma_offset;
1087
1088         /*
1089          * Fix up various Host-Dependent Parameters like Page Size, Cache
1090          * Line Size, etc.  The firmware default is for a 4KB Page Size and
1091          * 64B Cache Line Size ...
1092          */
1093         t4_fixup_host_params_compat(adapter, CXGBE_PAGE_SIZE, L1_CACHE_BYTES,
1094                                     T5_LAST_REV);
1095
1096         /*
1097          * Keep the chip default offset to deliver Ingress packets into our
1098          * DMA buffers to zero
1099          */
1100         rx_dma_offset = 0;
1101         t4_set_reg_field(adapter, A_SGE_CONTROL, V_PKTSHIFT(M_PKTSHIFT),
1102                          V_PKTSHIFT(rx_dma_offset));
1103
1104         t4_set_reg_field(adapter, A_SGE_FLM_CFG,
1105                          V_CREDITCNT(M_CREDITCNT) | M_CREDITCNTPACKING,
1106                          V_CREDITCNT(3) | V_CREDITCNTPACKING(1));
1107
1108         t4_set_reg_field(adapter, A_SGE_INGRESS_RX_THRESHOLD,
1109                          V_THRESHOLD_3(M_THRESHOLD_3), V_THRESHOLD_3(32U));
1110
1111         t4_set_reg_field(adapter, A_SGE_CONTROL2, V_IDMAARBROUNDROBIN(1U),
1112                          V_IDMAARBROUNDROBIN(1U));
1113
1114         /*
1115          * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
1116          * adds the pseudo header itself.
1117          */
1118         t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG,
1119                                F_CSUM_HAS_PSEUDO_HDR, 0);
1120
1121         return 0;
1122 }
1123
1124 /*
1125  * Attempt to initialize the adapter via a Firmware Configuration File.
1126  */
1127 static int adap_init0_config(struct adapter *adapter, int reset)
1128 {
1129         struct fw_caps_config_cmd caps_cmd;
1130         unsigned long mtype = 0, maddr = 0;
1131         u32 finiver, finicsum, cfcsum;
1132         int ret;
1133         int config_issued = 0;
1134         int cfg_addr;
1135         char config_name[20];
1136
1137         /*
1138          * Reset device if necessary.
1139          */
1140         if (reset) {
1141                 ret = t4_fw_reset(adapter, adapter->mbox,
1142                                   F_PIORSTMODE | F_PIORST);
1143                 if (ret < 0) {
1144                         dev_warn(adapter, "Firmware reset failed, error %d\n",
1145                                  -ret);
1146                         goto bye;
1147                 }
1148         }
1149
1150         cfg_addr = t4_flash_cfg_addr(adapter);
1151         if (cfg_addr < 0) {
1152                 ret = cfg_addr;
1153                 dev_warn(adapter, "Finding address for firmware config file in flash failed, error %d\n",
1154                          -ret);
1155                 goto bye;
1156         }
1157
1158         strcpy(config_name, "On Flash");
1159         mtype = FW_MEMTYPE_CF_FLASH;
1160         maddr = cfg_addr;
1161
1162         /*
1163          * Issue a Capability Configuration command to the firmware to get it
1164          * to parse the Configuration File.  We don't use t4_fw_config_file()
1165          * because we want the ability to modify various features after we've
1166          * processed the configuration file ...
1167          */
1168         memset(&caps_cmd, 0, sizeof(caps_cmd));
1169         caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1170                                            F_FW_CMD_REQUEST | F_FW_CMD_READ);
1171         caps_cmd.cfvalid_to_len16 =
1172                 cpu_to_be32(F_FW_CAPS_CONFIG_CMD_CFVALID |
1173                             V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
1174                             V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
1175                             FW_LEN16(caps_cmd));
1176         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
1177                          &caps_cmd);
1178         /*
1179          * If the CAPS_CONFIG failed with an ENOENT (for a Firmware
1180          * Configuration File in FLASH), our last gasp effort is to use the
1181          * Firmware Configuration File which is embedded in the firmware.  A
1182          * very few early versions of the firmware didn't have one embedded
1183          * but we can ignore those.
1184          */
1185         if (ret == -ENOENT) {
1186                 dev_info(adapter, "%s: Going for embedded config in firmware..\n",
1187                          __func__);
1188
1189                 memset(&caps_cmd, 0, sizeof(caps_cmd));
1190                 caps_cmd.op_to_write =
1191                         cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1192                                     F_FW_CMD_REQUEST | F_FW_CMD_READ);
1193                 caps_cmd.cfvalid_to_len16 = cpu_to_be32(FW_LEN16(caps_cmd));
1194                 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
1195                                  sizeof(caps_cmd), &caps_cmd);
1196                 strcpy(config_name, "Firmware Default");
1197         }
1198
1199         config_issued = 1;
1200         if (ret < 0)
1201                 goto bye;
1202
1203         finiver = be32_to_cpu(caps_cmd.finiver);
1204         finicsum = be32_to_cpu(caps_cmd.finicsum);
1205         cfcsum = be32_to_cpu(caps_cmd.cfcsum);
1206         if (finicsum != cfcsum)
1207                 dev_warn(adapter, "Configuration File checksum mismatch: [fini] csum=%#x, computed csum=%#x\n",
1208                          finicsum, cfcsum);
1209
1210         /*
1211          * If we're a pure NIC driver then disable all offloading facilities.
1212          * This will allow the firmware to optimize aspects of the hardware
1213          * configuration which will result in improved performance.
1214          */
1215         caps_cmd.niccaps &= cpu_to_be16(~FW_CAPS_CONFIG_NIC_ETHOFLD);
1216         caps_cmd.toecaps = 0;
1217         caps_cmd.iscsicaps = 0;
1218         caps_cmd.rdmacaps = 0;
1219         caps_cmd.fcoecaps = 0;
1220
1221         /*
1222          * And now tell the firmware to use the configuration we just loaded.
1223          */
1224         caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1225                                            F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
1226         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
1227         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
1228                          NULL);
1229         if (ret < 0) {
1230                 dev_warn(adapter, "Unable to finalize Firmware Capabilities %d\n",
1231                          -ret);
1232                 goto bye;
1233         }
1234
1235         /*
1236          * Tweak configuration based on system architecture, etc.
1237          */
1238         ret = adap_init0_tweaks(adapter);
1239         if (ret < 0) {
1240                 dev_warn(adapter, "Unable to do init0-tweaks %d\n", -ret);
1241                 goto bye;
1242         }
1243
1244         /*
1245          * And finally tell the firmware to initialize itself using the
1246          * parameters from the Configuration File.
1247          */
1248         ret = t4_fw_initialize(adapter, adapter->mbox);
1249         if (ret < 0) {
1250                 dev_warn(adapter, "Initializing Firmware failed, error %d\n",
1251                          -ret);
1252                 goto bye;
1253         }
1254
1255         /*
1256          * Return successfully and note that we're operating with parameters
1257          * not supplied by the driver, rather than from hard-wired
1258          * initialization constants buried in the driver.
1259          */
1260         dev_info(adapter,
1261                  "Successfully configured using Firmware Configuration File \"%s\", version %#x, computed checksum %#x\n",
1262                  config_name, finiver, cfcsum);
1263
1264         return 0;
1265
1266         /*
1267          * Something bad happened.  Return the error ...  (If the "error"
1268          * is that there's no Configuration File on the adapter we don't
1269          * want to issue a warning since this is fairly common.)
1270          */
1271 bye:
1272         if (config_issued && ret != -ENOENT)
1273                 dev_warn(adapter, "\"%s\" configuration file error %d\n",
1274                          config_name, -ret);
1275
1276         dev_debug(adapter, "%s: returning ret = %d ..\n", __func__, ret);
1277         return ret;
1278 }
1279
1280 static int adap_init0(struct adapter *adap)
1281 {
1282         struct fw_caps_config_cmd caps_cmd;
1283         int ret = 0;
1284         u32 v, port_vec;
1285         enum dev_state state;
1286         u32 params[7], val[7];
1287         int reset = 1;
1288         int mbox = adap->mbox;
1289
1290         /*
1291          * Contact FW, advertising Master capability.
1292          */
1293         ret = t4_fw_hello(adap, adap->mbox, adap->mbox, MASTER_MAY, &state);
1294         if (ret < 0) {
1295                 dev_err(adap, "%s: could not connect to FW, error %d\n",
1296                         __func__, -ret);
1297                 goto bye;
1298         }
1299
1300         CXGBE_DEBUG_MBOX(adap, "%s: adap->mbox = %d; ret = %d\n", __func__,
1301                          adap->mbox, ret);
1302
1303         if (ret == mbox)
1304                 adap->flags |= MASTER_PF;
1305
1306         if (state == DEV_STATE_INIT) {
1307                 /*
1308                  * Force halt and reset FW because a previous instance may have
1309                  * exited abnormally without properly shutting down
1310                  */
1311                 ret = t4_fw_halt(adap, adap->mbox, reset);
1312                 if (ret < 0) {
1313                         dev_err(adap, "Failed to halt. Exit.\n");
1314                         goto bye;
1315                 }
1316
1317                 ret = t4_fw_restart(adap, adap->mbox, reset);
1318                 if (ret < 0) {
1319                         dev_err(adap, "Failed to restart. Exit.\n");
1320                         goto bye;
1321                 }
1322                 state = (enum dev_state)((unsigned)state & ~DEV_STATE_INIT);
1323         }
1324
1325         t4_get_version_info(adap);
1326
1327         ret = t4_get_core_clock(adap, &adap->params.vpd);
1328         if (ret < 0) {
1329                 dev_err(adap, "%s: could not get core clock, error %d\n",
1330                         __func__, -ret);
1331                 goto bye;
1332         }
1333
1334         /*
1335          * If the firmware is initialized already (and we're not forcing a
1336          * master initialization), note that we're living with existing
1337          * adapter parameters.  Otherwise, it's time to try initializing the
1338          * adapter ...
1339          */
1340         if (state == DEV_STATE_INIT) {
1341                 dev_info(adap, "Coming up as %s: Adapter already initialized\n",
1342                          adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
1343         } else {
1344                 dev_info(adap, "Coming up as MASTER: Initializing adapter\n");
1345
1346                 ret = adap_init0_config(adap, reset);
1347                 if (ret == -ENOENT) {
1348                         dev_err(adap,
1349                                 "No Configuration File present on adapter. Using hard-wired configuration parameters.\n");
1350                         goto bye;
1351                 }
1352         }
1353         if (ret < 0) {
1354                 dev_err(adap, "could not initialize adapter, error %d\n", -ret);
1355                 goto bye;
1356         }
1357
1358         /* Now that we've successfully configured and initialized the adapter
1359          * (or found it already initialized), we can ask the Firmware what
1360          * resources it has provisioned for us.
1361          */
1362         ret = t4_get_pfres(adap);
1363         if (ret) {
1364                 dev_err(adap->pdev_dev,
1365                         "Unable to retrieve resource provisioning info\n");
1366                 goto bye;
1367         }
1368
1369         /* Find out what ports are available to us. */
1370         v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
1371             V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
1372         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
1373         if (ret < 0) {
1374                 dev_err(adap, "%s: failure in t4_query_params; error = %d\n",
1375                         __func__, ret);
1376                 goto bye;
1377         }
1378
1379         adap->params.nports = hweight32(port_vec);
1380         adap->params.portvec = port_vec;
1381
1382         dev_debug(adap, "%s: adap->params.nports = %u\n", __func__,
1383                   adap->params.nports);
1384
1385         /*
1386          * Give the SGE code a chance to pull in anything that it needs ...
1387          * Note that this must be called after we retrieve our VPD parameters
1388          * in order to know how to convert core ticks to seconds, etc.
1389          */
1390         ret = t4_sge_init(adap);
1391         if (ret < 0) {
1392                 dev_err(adap, "t4_sge_init failed with error %d\n",
1393                         -ret);
1394                 goto bye;
1395         }
1396
1397         /*
1398          * Grab some of our basic fundamental operating parameters.
1399          */
1400         params[0] = CXGBE_FW_PARAM_PFVF(L2T_START);
1401         params[1] = CXGBE_FW_PARAM_PFVF(L2T_END);
1402         params[2] = CXGBE_FW_PARAM_PFVF(FILTER_START);
1403         params[3] = CXGBE_FW_PARAM_PFVF(FILTER_END);
1404         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 4, params, val);
1405         if (ret < 0)
1406                 goto bye;
1407         adap->l2t_start = val[0];
1408         adap->l2t_end = val[1];
1409         adap->tids.ftid_base = val[2];
1410         adap->tids.nftids = val[3] - val[2] + 1;
1411
1412         params[0] = CXGBE_FW_PARAM_PFVF(CLIP_START);
1413         params[1] = CXGBE_FW_PARAM_PFVF(CLIP_END);
1414         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
1415         if (ret < 0)
1416                 goto bye;
1417         adap->clipt_start = val[0];
1418         adap->clipt_end = val[1];
1419
1420         /*
1421          * Get device capabilities so we can determine what resources we need
1422          * to manage.
1423          */
1424         memset(&caps_cmd, 0, sizeof(caps_cmd));
1425         caps_cmd.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1426                                      F_FW_CMD_REQUEST | F_FW_CMD_READ);
1427         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
1428         ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
1429                          &caps_cmd);
1430         if (ret < 0)
1431                 goto bye;
1432
1433         if ((caps_cmd.niccaps & cpu_to_be16(FW_CAPS_CONFIG_NIC_HASHFILTER)) &&
1434             is_t6(adap->params.chip)) {
1435                 if (cxgbe_init_hash_filter(adap) < 0)
1436                         goto bye;
1437         }
1438
1439         /* See if FW supports FW_FILTER2 work request */
1440         if (is_t4(adap->params.chip)) {
1441                 adap->params.filter2_wr_support = 0;
1442         } else {
1443                 params[0] = CXGBE_FW_PARAM_DEV(FILTER2_WR);
1444                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1445                                       1, params, val);
1446                 adap->params.filter2_wr_support = (ret == 0 && val[0] != 0);
1447         }
1448
1449         /* Check if FW supports returning vin.
1450          * If this is not supported, driver will interpret
1451          * these values from viid.
1452          */
1453         params[0] = CXGBE_FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
1454         ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1455                               1, params, val);
1456         adap->params.viid_smt_extn_support = (ret == 0 && val[0] != 0);
1457
1458         /* query tid-related parameters */
1459         params[0] = CXGBE_FW_PARAM_DEV(NTID);
1460         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
1461                               params, val);
1462         if (ret < 0)
1463                 goto bye;
1464         adap->tids.ntids = val[0];
1465         adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
1466
1467         /* If we're running on newer firmware, let it know that we're
1468          * prepared to deal with encapsulated CPL messages.  Older
1469          * firmware won't understand this and we'll just get
1470          * unencapsulated messages ...
1471          */
1472         params[0] = CXGBE_FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
1473         val[0] = 1;
1474         (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1475
1476         /*
1477          * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
1478          * capability.  Earlier versions of the firmware didn't have the
1479          * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
1480          * permission to use ULPTX MEMWRITE DSGL.
1481          */
1482         if (is_t4(adap->params.chip)) {
1483                 adap->params.ulptx_memwrite_dsgl = false;
1484         } else {
1485                 params[0] = CXGBE_FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
1486                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1487                                       1, params, val);
1488                 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
1489         }
1490
1491         /* Query for max number of packets that can be coalesced for Tx */
1492         params[0] = CXGBE_FW_PARAM_PFVF(MAX_PKTS_PER_ETH_TX_PKTS_WR);
1493         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1494         if (!ret && val[0] > 0)
1495                 adap->params.max_tx_coalesce_num = val[0];
1496         else
1497                 adap->params.max_tx_coalesce_num = ETH_COALESCE_PKT_NUM;
1498
1499         /*
1500          * The MTU/MSS Table is initialized by now, so load their values.  If
1501          * we're initializing the adapter, then we'll make any modifications
1502          * we want to the MTU/MSS Table and also initialize the congestion
1503          * parameters.
1504          */
1505         t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
1506         if (state != DEV_STATE_INIT) {
1507                 int i;
1508
1509                 /*
1510                  * The default MTU Table contains values 1492 and 1500.
1511                  * However, for TCP, it's better to have two values which are
1512                  * a multiple of 8 +/- 4 bytes apart near this popular MTU.
1513                  * This allows us to have a TCP Data Payload which is a
1514                  * multiple of 8 regardless of what combination of TCP Options
1515                  * are in use (always a multiple of 4 bytes) which is
1516                  * important for performance reasons.  For instance, if no
1517                  * options are in use, then we have a 20-byte IP header and a
1518                  * 20-byte TCP header.  In this case, a 1500-byte MSS would
1519                  * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
1520                  * which is not a multiple of 8.  So using an MSS of 1488 in
1521                  * this case results in a TCP Data Payload of 1448 bytes which
1522                  * is a multiple of 8.  On the other hand, if 12-byte TCP Time
1523                  * Stamps have been negotiated, then an MTU of 1500 bytes
1524                  * results in a TCP Data Payload of 1448 bytes which, as
1525                  * above, is a multiple of 8 bytes ...
1526                  */
1527                 for (i = 0; i < NMTUS; i++)
1528                         if (adap->params.mtus[i] == 1492) {
1529                                 adap->params.mtus[i] = 1488;
1530                                 break;
1531                         }
1532
1533                 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
1534                              adap->params.b_wnd);
1535         }
1536         t4_init_sge_params(adap);
1537         ret = configure_filter_mode_mask(adap);
1538         if (ret < 0)
1539                 goto bye;
1540         t4_init_tp_params(adap);
1541         configure_pcie_ext_tag(adap);
1542         configure_vlan_types(adap);
1543         cxgbe_configure_max_ethqsets(adap);
1544
1545         adap->params.drv_memwin = MEMWIN_NIC;
1546         adap->flags |= FW_OK;
1547         dev_debug(adap, "%s: returning zero..\n", __func__);
1548         return 0;
1549
1550         /*
1551          * Something bad happened.  If a command timed out or failed with EIO
1552          * FW does not operate within its spec or something catastrophic
1553          * happened to HW/FW, stop issuing commands.
1554          */
1555 bye:
1556         if (ret != -ETIMEDOUT && ret != -EIO)
1557                 t4_fw_bye(adap, adap->mbox);
1558         return ret;
1559 }
1560
1561 /**
1562  * t4_os_portmod_changed - handle port module changes
1563  * @adap: the adapter associated with the module change
1564  * @port_id: the port index whose module status has changed
1565  *
1566  * This is the OS-dependent handler for port module changes.  It is
1567  * invoked when a port module is removed or inserted for any OS-specific
1568  * processing.
1569  */
1570 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
1571 {
1572         static const char * const mod_str[] = {
1573                 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
1574         };
1575
1576         const struct port_info *pi = adap2pinfo(adap, port_id);
1577
1578         if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
1579                 dev_info(adap, "Port%d: port module unplugged\n", pi->port_id);
1580         else if (pi->mod_type < ARRAY_SIZE(mod_str))
1581                 dev_info(adap, "Port%d: %s port module inserted\n", pi->port_id,
1582                          mod_str[pi->mod_type]);
1583         else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
1584                 dev_info(adap, "Port%d: unsupported port module inserted\n",
1585                          pi->port_id);
1586         else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
1587                 dev_info(adap, "Port%d: unknown port module inserted\n",
1588                          pi->port_id);
1589         else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
1590                 dev_info(adap, "Port%d: transceiver module error\n",
1591                          pi->port_id);
1592         else
1593                 dev_info(adap, "Port%d: unknown module type %d inserted\n",
1594                          pi->port_id, pi->mod_type);
1595 }
1596
1597 bool cxgbe_force_linkup(struct adapter *adap)
1598 {
1599         if (is_pf4(adap))
1600                 return false;   /* force_linkup not required for pf driver */
1601
1602         return adap->devargs.force_link_up;
1603 }
1604
1605 /**
1606  * link_start - enable a port
1607  * @dev: the port to enable
1608  *
1609  * Performs the MAC and PHY actions needed to enable a port.
1610  */
1611 int cxgbe_link_start(struct port_info *pi)
1612 {
1613         struct adapter *adapter = pi->adapter;
1614         u64 conf_offloads;
1615         unsigned int mtu;
1616         int ret;
1617
1618         mtu = pi->eth_dev->data->dev_conf.rxmode.max_rx_pkt_len -
1619               (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN);
1620
1621         conf_offloads = pi->eth_dev->data->dev_conf.rxmode.offloads;
1622
1623         /*
1624          * We do not set address filters and promiscuity here, the stack does
1625          * that step explicitly.
1626          */
1627         ret = t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu, -1, -1, -1,
1628                             !!(conf_offloads & DEV_RX_OFFLOAD_VLAN_STRIP),
1629                             true);
1630         if (ret == 0) {
1631                 ret = cxgbe_mpstcam_modify(pi, (int)pi->xact_addr_filt,
1632                                 (u8 *)&pi->eth_dev->data->mac_addrs[0]);
1633                 if (ret >= 0) {
1634                         pi->xact_addr_filt = ret;
1635                         ret = 0;
1636                 }
1637         }
1638         if (ret == 0 && is_pf4(adapter))
1639                 ret = t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
1640                                     &pi->link_cfg);
1641         if (ret == 0) {
1642                 /*
1643                  * Enabling a Virtual Interface can result in an interrupt
1644                  * during the processing of the VI Enable command and, in some
1645                  * paths, result in an attempt to issue another command in the
1646                  * interrupt context.  Thus, we disable interrupts during the
1647                  * course of the VI Enable command ...
1648                  */
1649                 ret = t4_enable_vi_params(adapter, adapter->mbox, pi->viid,
1650                                           true, true, false);
1651         }
1652
1653         if (ret == 0 && cxgbe_force_linkup(adapter))
1654                 pi->eth_dev->data->dev_link.link_status = ETH_LINK_UP;
1655         return ret;
1656 }
1657
1658 /**
1659  * cxgbe_write_rss_conf - flash the RSS configuration for a given port
1660  * @pi: the port
1661  * @rss_hf: Hash configuration to apply
1662  */
1663 int cxgbe_write_rss_conf(const struct port_info *pi, uint64_t rss_hf)
1664 {
1665         struct adapter *adapter = pi->adapter;
1666         const struct sge_eth_rxq *rxq;
1667         u64 flags = 0;
1668         u16 rss;
1669         int err;
1670
1671         /*  Should never be called before setting up sge eth rx queues */
1672         if (!(adapter->flags & FULL_INIT_DONE)) {
1673                 dev_err(adap, "%s No RXQs available on port %d\n",
1674                         __func__, pi->port_id);
1675                 return -EINVAL;
1676         }
1677
1678         /* Don't allow unsupported hash functions */
1679         if (rss_hf & ~CXGBE_RSS_HF_ALL)
1680                 return -EINVAL;
1681
1682         if (rss_hf & CXGBE_RSS_HF_IPV4_MASK)
1683                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
1684
1685         if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
1686                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
1687
1688         if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
1689                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
1690                          F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1691
1692         if (rss_hf & CXGBE_RSS_HF_IPV6_MASK)
1693                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
1694
1695         if (rss_hf & CXGBE_RSS_HF_TCP_IPV6_MASK)
1696                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1697                          F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
1698
1699         if (rss_hf & CXGBE_RSS_HF_UDP_IPV6_MASK)
1700                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1701                          F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
1702                          F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1703
1704         rxq = &adapter->sge.ethrxq[pi->first_qset];
1705         rss = rxq[0].rspq.abs_id;
1706
1707         /* If Tunnel All Lookup isn't specified in the global RSS
1708          * Configuration, then we need to specify a default Ingress
1709          * Queue for any ingress packets which aren't hashed.  We'll
1710          * use our first ingress queue ...
1711          */
1712         err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
1713                                flags, rss);
1714         return err;
1715 }
1716
1717 /**
1718  * cxgbe_write_rss - write the RSS table for a given port
1719  * @pi: the port
1720  * @queues: array of queue indices for RSS
1721  *
1722  * Sets up the portion of the HW RSS table for the port's VI to distribute
1723  * packets to the Rx queues in @queues.
1724  */
1725 int cxgbe_write_rss(const struct port_info *pi, const u16 *queues)
1726 {
1727         u16 *rss;
1728         int i, err;
1729         struct adapter *adapter = pi->adapter;
1730         const struct sge_eth_rxq *rxq;
1731
1732         /*  Should never be called before setting up sge eth rx queues */
1733         BUG_ON(!(adapter->flags & FULL_INIT_DONE));
1734
1735         rxq = &adapter->sge.ethrxq[pi->first_qset];
1736         rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
1737         if (!rss)
1738                 return -ENOMEM;
1739
1740         /* map the queue indices to queue ids */
1741         for (i = 0; i < pi->rss_size; i++, queues++)
1742                 rss[i] = rxq[*queues].rspq.abs_id;
1743
1744         err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
1745                                   pi->rss_size, rss, pi->rss_size);
1746         rte_free(rss);
1747         return err;
1748 }
1749
1750 /**
1751  * setup_rss - configure RSS
1752  * @adapter: the adapter
1753  *
1754  * Sets up RSS to distribute packets to multiple receive queues.  We
1755  * configure the RSS CPU lookup table to distribute to the number of HW
1756  * receive queues, and the response queue lookup table to narrow that
1757  * down to the response queues actually configured for each port.
1758  * We always configure the RSS mapping for all ports since the mapping
1759  * table has plenty of entries.
1760  */
1761 int cxgbe_setup_rss(struct port_info *pi)
1762 {
1763         int j, err;
1764         struct adapter *adapter = pi->adapter;
1765
1766         dev_debug(adapter, "%s:  pi->rss_size = %u; pi->n_rx_qsets = %u\n",
1767                   __func__, pi->rss_size, pi->n_rx_qsets);
1768
1769         if (!(pi->flags & PORT_RSS_DONE)) {
1770                 if (adapter->flags & FULL_INIT_DONE) {
1771                         /* Fill default values with equal distribution */
1772                         for (j = 0; j < pi->rss_size; j++)
1773                                 pi->rss[j] = j % pi->n_rx_qsets;
1774
1775                         err = cxgbe_write_rss(pi, pi->rss);
1776                         if (err)
1777                                 return err;
1778
1779                         err = cxgbe_write_rss_conf(pi, pi->rss_hf);
1780                         if (err)
1781                                 return err;
1782                         pi->flags |= PORT_RSS_DONE;
1783                 }
1784         }
1785         return 0;
1786 }
1787
1788 /*
1789  * Enable NAPI scheduling and interrupt generation for all Rx queues.
1790  */
1791 static void enable_rx(struct adapter *adap, struct sge_rspq *q)
1792 {
1793         /* 0-increment GTS to start the timer and enable interrupts */
1794         t4_write_reg(adap, is_pf4(adap) ? MYPF_REG(A_SGE_PF_GTS) :
1795                                           T4VF_SGE_BASE_ADDR + A_SGE_VF_GTS,
1796                      V_SEINTARM(q->intr_params) |
1797                      V_INGRESSQID(q->cntxt_id));
1798 }
1799
1800 void cxgbe_enable_rx_queues(struct port_info *pi)
1801 {
1802         struct adapter *adap = pi->adapter;
1803         struct sge *s = &adap->sge;
1804         unsigned int i;
1805
1806         for (i = 0; i < pi->n_rx_qsets; i++)
1807                 enable_rx(adap, &s->ethrxq[pi->first_qset + i].rspq);
1808 }
1809
1810 /**
1811  * fw_caps_to_speed_caps - translate Firmware Port Caps to Speed Caps.
1812  * @port_type: Firmware Port Type
1813  * @fw_caps: Firmware Port Capabilities
1814  * @speed_caps: Device Info Speed Capabilities
1815  *
1816  * Translate a Firmware Port Capabilities specification to Device Info
1817  * Speed Capabilities.
1818  */
1819 static void fw_caps_to_speed_caps(enum fw_port_type port_type,
1820                                   unsigned int fw_caps,
1821                                   u32 *speed_caps)
1822 {
1823 #define SET_SPEED(__speed_name) \
1824         do { \
1825                 *speed_caps |= ETH_LINK_ ## __speed_name; \
1826         } while (0)
1827
1828 #define FW_CAPS_TO_SPEED(__fw_name) \
1829         do { \
1830                 if (fw_caps & FW_PORT_CAP32_ ## __fw_name) \
1831                         SET_SPEED(__fw_name); \
1832         } while (0)
1833
1834         switch (port_type) {
1835         case FW_PORT_TYPE_BT_SGMII:
1836         case FW_PORT_TYPE_BT_XFI:
1837         case FW_PORT_TYPE_BT_XAUI:
1838                 FW_CAPS_TO_SPEED(SPEED_100M);
1839                 FW_CAPS_TO_SPEED(SPEED_1G);
1840                 FW_CAPS_TO_SPEED(SPEED_10G);
1841                 break;
1842
1843         case FW_PORT_TYPE_KX4:
1844         case FW_PORT_TYPE_KX:
1845         case FW_PORT_TYPE_FIBER_XFI:
1846         case FW_PORT_TYPE_FIBER_XAUI:
1847         case FW_PORT_TYPE_SFP:
1848         case FW_PORT_TYPE_QSFP_10G:
1849         case FW_PORT_TYPE_QSA:
1850                 FW_CAPS_TO_SPEED(SPEED_1G);
1851                 FW_CAPS_TO_SPEED(SPEED_10G);
1852                 break;
1853
1854         case FW_PORT_TYPE_KR:
1855                 SET_SPEED(SPEED_10G);
1856                 break;
1857
1858         case FW_PORT_TYPE_BP_AP:
1859         case FW_PORT_TYPE_BP4_AP:
1860                 SET_SPEED(SPEED_1G);
1861                 SET_SPEED(SPEED_10G);
1862                 break;
1863
1864         case FW_PORT_TYPE_BP40_BA:
1865         case FW_PORT_TYPE_QSFP:
1866                 SET_SPEED(SPEED_40G);
1867                 break;
1868
1869         case FW_PORT_TYPE_CR_QSFP:
1870         case FW_PORT_TYPE_SFP28:
1871         case FW_PORT_TYPE_KR_SFP28:
1872                 FW_CAPS_TO_SPEED(SPEED_1G);
1873                 FW_CAPS_TO_SPEED(SPEED_10G);
1874                 FW_CAPS_TO_SPEED(SPEED_25G);
1875                 break;
1876
1877         case FW_PORT_TYPE_CR2_QSFP:
1878                 SET_SPEED(SPEED_50G);
1879                 break;
1880
1881         case FW_PORT_TYPE_KR4_100G:
1882         case FW_PORT_TYPE_CR4_QSFP:
1883                 FW_CAPS_TO_SPEED(SPEED_25G);
1884                 FW_CAPS_TO_SPEED(SPEED_40G);
1885                 FW_CAPS_TO_SPEED(SPEED_50G);
1886                 FW_CAPS_TO_SPEED(SPEED_100G);
1887                 break;
1888
1889         default:
1890                 break;
1891         }
1892
1893 #undef FW_CAPS_TO_SPEED
1894 #undef SET_SPEED
1895 }
1896
1897 /**
1898  * cxgbe_get_speed_caps - Fetch supported speed capabilities
1899  * @pi: Underlying port's info
1900  * @speed_caps: Device Info speed capabilities
1901  *
1902  * Fetch supported speed capabilities of the underlying port.
1903  */
1904 void cxgbe_get_speed_caps(struct port_info *pi, u32 *speed_caps)
1905 {
1906         *speed_caps = 0;
1907
1908         fw_caps_to_speed_caps(pi->port_type, pi->link_cfg.pcaps,
1909                               speed_caps);
1910
1911         if (!(pi->link_cfg.pcaps & FW_PORT_CAP32_ANEG))
1912                 *speed_caps |= ETH_LINK_SPEED_FIXED;
1913 }
1914
1915 /**
1916  * cxgbe_set_link_status - Set device link up or down.
1917  * @pi: Underlying port's info
1918  * @status: 0 - down, 1 - up
1919  *
1920  * Set the device link up or down.
1921  */
1922 int cxgbe_set_link_status(struct port_info *pi, bool status)
1923 {
1924         struct adapter *adapter = pi->adapter;
1925         int err = 0;
1926
1927         err = t4_enable_vi(adapter, adapter->mbox, pi->viid, status, status);
1928         if (err) {
1929                 dev_err(adapter, "%s: disable_vi failed: %d\n", __func__, err);
1930                 return err;
1931         }
1932
1933         if (!status)
1934                 t4_reset_link_config(adapter, pi->pidx);
1935
1936         return 0;
1937 }
1938
1939 /**
1940  * cxgb_up - enable the adapter
1941  * @adap: adapter being enabled
1942  *
1943  * Called when the first port is enabled, this function performs the
1944  * actions necessary to make an adapter operational, such as completing
1945  * the initialization of HW modules, and enabling interrupts.
1946  */
1947 int cxgbe_up(struct adapter *adap)
1948 {
1949         enable_rx(adap, &adap->sge.fw_evtq);
1950         t4_sge_tx_monitor_start(adap);
1951         if (is_pf4(adap))
1952                 t4_intr_enable(adap);
1953         adap->flags |= FULL_INIT_DONE;
1954
1955         /* TODO: deadman watchdog ?? */
1956         return 0;
1957 }
1958
1959 /*
1960  * Close the port
1961  */
1962 int cxgbe_down(struct port_info *pi)
1963 {
1964         return cxgbe_set_link_status(pi, false);
1965 }
1966
1967 /*
1968  * Release resources when all the ports have been stopped.
1969  */
1970 void cxgbe_close(struct adapter *adapter)
1971 {
1972         struct port_info *pi;
1973         int i;
1974
1975         if (adapter->flags & FULL_INIT_DONE) {
1976                 tid_free(&adapter->tids);
1977                 t4_cleanup_mpstcam(adapter);
1978                 t4_cleanup_clip_tbl(adapter);
1979                 t4_cleanup_l2t(adapter);
1980                 t4_cleanup_smt(adapter);
1981                 if (is_pf4(adapter))
1982                         t4_intr_disable(adapter);
1983                 t4_sge_tx_monitor_stop(adapter);
1984                 t4_free_sge_resources(adapter);
1985                 for_each_port(adapter, i) {
1986                         pi = adap2pinfo(adapter, i);
1987                         if (pi->viid != 0)
1988                                 t4_free_vi(adapter, adapter->mbox,
1989                                            adapter->pf, 0, pi->viid);
1990                         rte_eth_dev_release_port(pi->eth_dev);
1991                 }
1992                 adapter->flags &= ~FULL_INIT_DONE;
1993         }
1994
1995         if (is_pf4(adapter) && (adapter->flags & FW_OK))
1996                 t4_fw_bye(adapter, adapter->mbox);
1997 }
1998
1999 static void adap_smt_index(struct adapter *adapter, u32 *smt_start_idx,
2000                            u32 *smt_size)
2001 {
2002         u32 params[2], smt_val[2];
2003         int ret;
2004
2005         params[0] = CXGBE_FW_PARAM_PFVF(GET_SMT_START);
2006         params[1] = CXGBE_FW_PARAM_PFVF(GET_SMT_SIZE);
2007
2008         ret = t4_query_params(adapter, adapter->mbox, adapter->pf, 0,
2009                               2, params, smt_val);
2010
2011         /* if FW doesn't recognize this command then set it to default setting
2012          * which is start index as 0 and size as 256.
2013          */
2014         if (ret < 0) {
2015                 *smt_start_idx = 0;
2016                 *smt_size = SMT_SIZE;
2017         } else {
2018                 *smt_start_idx = smt_val[0];
2019                 /* smt size can be zero, if nsmt is not yet configured in
2020                  * the config file or set as zero, then configure all the
2021                  * remaining entries to this PF itself.
2022                  */
2023                 if (!smt_val[1])
2024                         *smt_size = SMT_SIZE - *smt_start_idx;
2025                 else
2026                         *smt_size = smt_val[1];
2027         }
2028 }
2029
2030 int cxgbe_probe(struct adapter *adapter)
2031 {
2032         u32 smt_start_idx, smt_size;
2033         struct port_info *pi;
2034         int func, i;
2035         int err = 0;
2036         u32 whoami;
2037         int chip;
2038
2039         whoami = t4_read_reg(adapter, A_PL_WHOAMI);
2040         chip = t4_get_chip_type(adapter,
2041                         CHELSIO_PCI_ID_VER(adapter->pdev->id.device_id));
2042         if (chip < 0)
2043                 return chip;
2044
2045         func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
2046                G_SOURCEPF(whoami) : G_T6_SOURCEPF(whoami);
2047
2048         adapter->mbox = func;
2049         adapter->pf = func;
2050
2051         t4_os_lock_init(&adapter->mbox_lock);
2052         TAILQ_INIT(&adapter->mbox_list);
2053         t4_os_lock_init(&adapter->win0_lock);
2054
2055         err = t4_prep_adapter(adapter);
2056         if (err)
2057                 return err;
2058
2059         setup_memwin(adapter);
2060         err = adap_init0(adapter);
2061         if (err) {
2062                 dev_err(adapter, "%s: Adapter initialization failed, error %d\n",
2063                         __func__, err);
2064                 goto out_free;
2065         }
2066
2067         if (!is_t4(adapter->params.chip)) {
2068                 /*
2069                  * The userspace doorbell BAR is split evenly into doorbell
2070                  * regions, each associated with an egress queue.  If this
2071                  * per-queue region is large enough (at least UDBS_SEG_SIZE)
2072                  * then it can be used to submit a tx work request with an
2073                  * implied doorbell.  Enable write combining on the BAR if
2074                  * there is room for such work requests.
2075                  */
2076                 int s_qpp, qpp, num_seg;
2077
2078                 s_qpp = (S_QUEUESPERPAGEPF0 +
2079                         (S_QUEUESPERPAGEPF1 - S_QUEUESPERPAGEPF0) *
2080                         adapter->pf);
2081                 qpp = 1 << ((t4_read_reg(adapter,
2082                                 A_SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp)
2083                                 & M_QUEUESPERPAGEPF0);
2084                 num_seg = CXGBE_PAGE_SIZE / UDBS_SEG_SIZE;
2085                 if (qpp > num_seg)
2086                         dev_warn(adapter, "Incorrect SGE EGRESS QUEUES_PER_PAGE configuration, continuing in debug mode\n");
2087
2088                 adapter->bar2 = (void *)adapter->pdev->mem_resource[2].addr;
2089                 if (!adapter->bar2) {
2090                         dev_err(adapter, "cannot map device bar2 region\n");
2091                         err = -ENOMEM;
2092                         goto out_free;
2093                 }
2094                 t4_write_reg(adapter, A_SGE_STAT_CFG, V_STATSOURCE_T5(7) |
2095                              V_STATMODE(0));
2096         }
2097
2098         for_each_port(adapter, i) {
2099                 const unsigned int numa_node = rte_socket_id();
2100                 char name[RTE_ETH_NAME_MAX_LEN];
2101                 struct rte_eth_dev *eth_dev;
2102
2103                 snprintf(name, sizeof(name), "%s_%d",
2104                          adapter->pdev->device.name, i);
2105
2106                 if (i == 0) {
2107                         /* First port is already allocated by DPDK */
2108                         eth_dev = adapter->eth_dev;
2109                         goto allocate_mac;
2110                 }
2111
2112                 /*
2113                  * now do all data allocation - for eth_dev structure,
2114                  * and internal (private) data for the remaining ports
2115                  */
2116
2117                 /* reserve an ethdev entry */
2118                 eth_dev = rte_eth_dev_allocate(name);
2119                 if (!eth_dev)
2120                         goto out_free;
2121
2122                 eth_dev->data->dev_private =
2123                         rte_zmalloc_socket(name, sizeof(struct port_info),
2124                                            RTE_CACHE_LINE_SIZE, numa_node);
2125                 if (!eth_dev->data->dev_private)
2126                         goto out_free;
2127
2128 allocate_mac:
2129                 pi = eth_dev->data->dev_private;
2130                 adapter->port[i] = pi;
2131                 pi->eth_dev = eth_dev;
2132                 pi->adapter = adapter;
2133                 pi->xact_addr_filt = -1;
2134                 pi->port_id = i;
2135                 pi->pidx = i;
2136
2137                 pi->eth_dev->device = &adapter->pdev->device;
2138                 pi->eth_dev->dev_ops = adapter->eth_dev->dev_ops;
2139                 pi->eth_dev->tx_pkt_burst = adapter->eth_dev->tx_pkt_burst;
2140                 pi->eth_dev->rx_pkt_burst = adapter->eth_dev->rx_pkt_burst;
2141
2142                 rte_eth_copy_pci_info(pi->eth_dev, adapter->pdev);
2143
2144                 pi->eth_dev->data->mac_addrs = rte_zmalloc(name,
2145                                                         RTE_ETHER_ADDR_LEN, 0);
2146                 if (!pi->eth_dev->data->mac_addrs) {
2147                         dev_err(adapter, "%s: Mem allocation failed for storing mac addr, aborting\n",
2148                                 __func__);
2149                         err = -1;
2150                         goto out_free;
2151                 }
2152
2153                 if (i > 0) {
2154                         /* First port will be notified by upper layer */
2155                         rte_eth_dev_probing_finish(eth_dev);
2156                 }
2157         }
2158
2159         if (adapter->flags & FW_OK) {
2160                 err = t4_port_init(adapter, adapter->mbox, adapter->pf, 0);
2161                 if (err) {
2162                         dev_err(adapter, "%s: t4_port_init failed with err %d\n",
2163                                 __func__, err);
2164                         goto out_free;
2165                 }
2166         }
2167
2168         cxgbe_cfg_queues(adapter->eth_dev);
2169
2170         cxgbe_print_adapter_info(adapter);
2171         cxgbe_print_port_info(adapter);
2172
2173         adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
2174                                           adapter->clipt_end);
2175         if (!adapter->clipt) {
2176                 /* We tolerate a lack of clip_table, giving up some
2177                  * functionality
2178                  */
2179                 dev_warn(adapter, "could not allocate CLIP. Continuing\n");
2180         }
2181
2182         adap_smt_index(adapter, &smt_start_idx, &smt_size);
2183         adapter->smt = t4_init_smt(smt_start_idx, smt_size);
2184         if (!adapter->smt)
2185                 dev_warn(adapter, "could not allocate SMT, continuing\n");
2186
2187         adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
2188         if (!adapter->l2t) {
2189                 /* We tolerate a lack of L2T, giving up some functionality */
2190                 dev_warn(adapter, "could not allocate L2T. Continuing\n");
2191         }
2192
2193         if (tid_init(&adapter->tids) < 0) {
2194                 /* Disable filtering support */
2195                 dev_warn(adapter, "could not allocate TID table, "
2196                          "filter support disabled. Continuing\n");
2197         }
2198
2199         t4_os_lock_init(&adapter->flow_lock);
2200
2201         adapter->mpstcam = t4_init_mpstcam(adapter);
2202         if (!adapter->mpstcam)
2203                 dev_warn(adapter, "could not allocate mps tcam table."
2204                          " Continuing\n");
2205
2206         if (is_hashfilter(adapter)) {
2207                 if (t4_read_reg(adapter, A_LE_DB_CONFIG) & F_HASHEN) {
2208                         u32 hash_base, hash_reg;
2209
2210                         hash_reg = A_LE_DB_TID_HASHBASE;
2211                         hash_base = t4_read_reg(adapter, hash_reg);
2212                         adapter->tids.hash_base = hash_base / 4;
2213                 }
2214         } else {
2215                 /* Disable hash filtering support */
2216                 dev_warn(adapter,
2217                          "Maskless filter support disabled. Continuing\n");
2218         }
2219
2220         err = cxgbe_init_rss(adapter);
2221         if (err)
2222                 goto out_free;
2223
2224         return 0;
2225
2226 out_free:
2227         for_each_port(adapter, i) {
2228                 pi = adap2pinfo(adapter, i);
2229                 if (pi->viid != 0)
2230                         t4_free_vi(adapter, adapter->mbox, adapter->pf,
2231                                    0, pi->viid);
2232                 rte_eth_dev_release_port(pi->eth_dev);
2233         }
2234
2235         if (adapter->flags & FW_OK)
2236                 t4_fw_bye(adapter, adapter->mbox);
2237         return -err;
2238 }