drivers/net/ethernet/intel/ice/ice_dcb_lib.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright (c) 2019, Intel Corporation. */
   3
   4 #include "ice_dcb_lib.h"
   5 #include "ice_dcb_nl.h"
   6 #include "ice_devlink.h"
   7
   8 /**
   9  * ice_dcb_get_ena_tc - return bitmap of enabled TCs
  10  * @dcbcfg: DCB config to evaluate for enabled TCs
  11  */
  12 static u8 ice_dcb_get_ena_tc(struct ice_dcbx_cfg *dcbcfg)
  13 {
  14         u8 i, num_tc, ena_tc = 1;
  15
  16         num_tc = ice_dcb_get_num_tc(dcbcfg);
  17
  18         for (i = 0; i < num_tc; i++)
  19                 ena_tc |= BIT(i);
  20
  21         return ena_tc;
  22 }
  23
  24 /**
  25  * ice_is_pfc_causing_hung_q
  26  * @pf: pointer to PF structure
  27  * @txqueue: Tx queue which is supposedly hung queue
  28  *
  29  * find if PFC is causing the hung queue, if yes return true else false
  30  */
  31 bool ice_is_pfc_causing_hung_q(struct ice_pf *pf, unsigned int txqueue)
  32 {
  33         u8 num_tcs = 0, i, tc, up_mapped_tc, up_in_tc = 0;
  34         u64 ref_prio_xoff[ICE_MAX_UP];
  35         struct ice_vsi *vsi;
  36         u32 up2tc;
  37
  38         vsi = ice_get_main_vsi(pf);
  39         if (!vsi)
  40                 return false;
  41
  42         ice_for_each_traffic_class(i)
  43                 if (vsi->tc_cfg.ena_tc & BIT(i))
  44                         num_tcs++;
  45
  46         /* first find out the TC to which the hung queue belongs to */
  47         for (tc = 0; tc < num_tcs - 1; tc++)
  48                 if (ice_find_q_in_range(vsi->tc_cfg.tc_info[tc].qoffset,
  49                                         vsi->tc_cfg.tc_info[tc + 1].qoffset,
  50                                         txqueue))
  51                         break;
  52
  53         /* Build a bit map of all UPs associated to the suspect hung queue TC,
  54          * so that we check for its counter increment.
  55          */
  56         up2tc = rd32(&pf->hw, PRTDCB_TUP2TC);
  57         for (i = 0; i < ICE_MAX_UP; i++) {
  58                 up_mapped_tc = (up2tc >> (i * 3)) & 0x7;
  59                 if (up_mapped_tc == tc)
  60                         up_in_tc |= BIT(i);
  61         }
  62
  63         /* Now that we figured out that hung queue is PFC enabled, still the
  64          * Tx timeout can be legitimate. So to make sure Tx timeout is
  65          * absolutely caused by PFC storm, check if the counters are
  66          * incrementing.
  67          */
  68         for (i = 0; i < ICE_MAX_UP; i++)
  69                 if (up_in_tc & BIT(i))
  70                         ref_prio_xoff[i] = pf->stats.priority_xoff_rx[i];
  71
  72         ice_update_dcb_stats(pf);
  73
  74         for (i = 0; i < ICE_MAX_UP; i++)
  75                 if (up_in_tc & BIT(i))
  76                         if (pf->stats.priority_xoff_rx[i] > ref_prio_xoff[i])
  77                                 return true;
  78
  79         return false;
  80 }
  81
  82 /**
  83  * ice_dcb_get_mode - gets the DCB mode
  84  * @port_info: pointer to port info structure
  85  * @host: if set it's HOST if not it's MANAGED
  86  */
  87 static u8 ice_dcb_get_mode(struct ice_port_info *port_info, bool host)
  88 {
  89         u8 mode;
  90
  91         if (host)
  92                 mode = DCB_CAP_DCBX_HOST;
  93         else
  94                 mode = DCB_CAP_DCBX_LLD_MANAGED;
  95
  96         if (port_info->qos_cfg.local_dcbx_cfg.dcbx_mode & ICE_DCBX_MODE_CEE)
  97                 return mode | DCB_CAP_DCBX_VER_CEE;
  98         else
  99                 return mode | DCB_CAP_DCBX_VER_IEEE;
 100 }
 101
 102 /**
 103  * ice_dcb_get_num_tc - Get the number of TCs from DCBX config
 104  * @dcbcfg: config to retrieve number of TCs from
 105  */
 106 u8 ice_dcb_get_num_tc(struct ice_dcbx_cfg *dcbcfg)
 107 {
 108         bool tc_unused = false;
 109         u8 num_tc = 0;
 110         u8 ret = 0;
 111         int i;
 112
 113         /* Scan the ETS Config Priority Table to find traffic classes
 114          * enabled and create a bitmask of enabled TCs
 115          */
 116         for (i = 0; i < CEE_DCBX_MAX_PRIO; i++)
 117                 num_tc |= BIT(dcbcfg->etscfg.prio_table[i]);
 118
 119         /* Scan bitmask for contiguous TCs starting with TC0 */
 120         for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++) {
 121                 if (num_tc & BIT(i)) {
 122                         if (!tc_unused) {
 123                                 ret++;
 124                         } else {
 125                                 pr_err("Non-contiguous TCs - Disabling DCB\n");
 126                                 return 1;
 127                         }
 128                 } else {
 129                         tc_unused = true;
 130                 }
 131         }
 132
 133         /* There is always at least 1 TC */
 134         if (!ret)
 135                 ret = 1;
 136
 137         return ret;
 138 }
 139
 140 /**
 141  * ice_get_first_droptc - returns number of first droptc
 142  * @vsi: used to find the first droptc
 143  *
 144  * This function returns the value of first_droptc.
 145  * When DCB is enabled, first droptc information is derived from enabled_tc
 146  * and PFC enabled bits. otherwise this function returns 0 as there is one
 147  * TC without DCB (tc0)
 148  */
 149 static u8 ice_get_first_droptc(struct ice_vsi *vsi)
 150 {
 151         struct ice_dcbx_cfg *cfg = &vsi->port_info->qos_cfg.local_dcbx_cfg;
 152         struct device *dev = ice_pf_to_dev(vsi->back);
 153         u8 num_tc, ena_tc_map, pfc_ena_map;
 154         u8 i;
 155
 156         num_tc = ice_dcb_get_num_tc(cfg);
 157
 158         /* get bitmap of enabled TCs */
 159         ena_tc_map = ice_dcb_get_ena_tc(cfg);
 160
 161         /* get bitmap of PFC enabled TCs */
 162         pfc_ena_map = cfg->pfc.pfcena;
 163
 164         /* get first TC that is not PFC enabled */
 165         for (i = 0; i < num_tc; i++) {
 166                 if ((ena_tc_map & BIT(i)) && (!(pfc_ena_map & BIT(i)))) {
 167                         dev_dbg(dev, "first drop tc = %d\n", i);
 168                         return i;
 169                 }
 170         }
 171
 172         dev_dbg(dev, "first drop tc = 0\n");
 173         return 0;
 174 }
 175
 176 /**
 177  * ice_vsi_set_dcb_tc_cfg - Set VSI's TC based on DCB configuration
 178  * @vsi: pointer to the VSI instance
 179  */
 180 void ice_vsi_set_dcb_tc_cfg(struct ice_vsi *vsi)
 181 {
 182         struct ice_dcbx_cfg *cfg = &vsi->port_info->qos_cfg.local_dcbx_cfg;
 183
 184         switch (vsi->type) {
 185         case ICE_VSI_PF:
 186                 vsi->tc_cfg.ena_tc = ice_dcb_get_ena_tc(cfg);
 187                 vsi->tc_cfg.numtc = ice_dcb_get_num_tc(cfg);
 188                 break;
 189         case ICE_VSI_CHNL:
 190                 vsi->tc_cfg.ena_tc = BIT(ice_get_first_droptc(vsi));
 191                 vsi->tc_cfg.numtc = 1;
 192                 break;
 193         case ICE_VSI_CTRL:
 194         case ICE_VSI_LB:
 195         default:
 196                 vsi->tc_cfg.ena_tc = ICE_DFLT_TRAFFIC_CLASS;
 197                 vsi->tc_cfg.numtc = 1;
 198         }
 199 }
 200
 201 /**
 202  * ice_dcb_get_tc - Get the TC associated with the queue
 203  * @vsi: ptr to the VSI
 204  * @queue_index: queue number associated with VSI
 205  */
 206 u8 ice_dcb_get_tc(struct ice_vsi *vsi, int queue_index)
 207 {
 208         return vsi->tx_rings[queue_index]->dcb_tc;
 209 }
 210
 211 /**
 212  * ice_vsi_cfg_dcb_rings - Update rings to reflect DCB TC
 213  * @vsi: VSI owner of rings being updated
 214  */
 215 void ice_vsi_cfg_dcb_rings(struct ice_vsi *vsi)
 216 {
 217         struct ice_tx_ring *tx_ring;
 218         struct ice_rx_ring *rx_ring;
 219         u16 qoffset, qcount;
 220         int i, n;
 221
 222         if (!test_bit(ICE_FLAG_DCB_ENA, vsi->back->flags)) {
 223                 /* Reset the TC information */
 224                 ice_for_each_txq(vsi, i) {
 225                         tx_ring = vsi->tx_rings[i];
 226                         tx_ring->dcb_tc = 0;
 227                 }
 228                 ice_for_each_rxq(vsi, i) {
 229                         rx_ring = vsi->rx_rings[i];
 230                         rx_ring->dcb_tc = 0;
 231                 }
 232                 return;
 233         }
 234
 235         ice_for_each_traffic_class(n) {
 236                 if (!(vsi->tc_cfg.ena_tc & BIT(n)))
 237                         break;
 238
 239                 qoffset = vsi->tc_cfg.tc_info[n].qoffset;
 240                 qcount = vsi->tc_cfg.tc_info[n].qcount_tx;
 241                 for (i = qoffset; i < (qoffset + qcount); i++)
 242                         vsi->tx_rings[i]->dcb_tc = n;
 243
 244                 qcount = vsi->tc_cfg.tc_info[n].qcount_rx;
 245                 for (i = qoffset; i < (qoffset + qcount); i++)
 246                         vsi->rx_rings[i]->dcb_tc = n;
 247         }
 248         /* applicable only if "all_enatc" is set, which will be set from
 249          * setup_tc method as part of configuring channels
 250          */
 251         if (vsi->all_enatc) {
 252                 u8 first_droptc = ice_get_first_droptc(vsi);
 253
 254                 /* When DCB is configured, TC for ADQ queues (which are really
 255                  * PF queues) should be the first drop TC of the main VSI
 256                  */
 257                 ice_for_each_chnl_tc(n) {
 258                         if (!(vsi->all_enatc & BIT(n)))
 259                                 break;
 260
 261                         qoffset = vsi->mqprio_qopt.qopt.offset[n];
 262                         qcount = vsi->mqprio_qopt.qopt.count[n];
 263                         for (i = qoffset; i < (qoffset + qcount); i++) {
 264                                 vsi->tx_rings[i]->dcb_tc = first_droptc;
 265                                 vsi->rx_rings[i]->dcb_tc = first_droptc;
 266                         }
 267                 }
 268         }
 269 }
 270
 271 /**
 272  * ice_dcb_ena_dis_vsi - disable certain VSIs for DCB config/reconfig
 273  * @pf: pointer to the PF instance
 274  * @ena: true to enable VSIs, false to disable
 275  * @locked: true if caller holds RTNL lock, false otherwise
 276  *
 277  * Before a new DCB configuration can be applied, VSIs of type PF, SWITCHDEV
 278  * and CHNL need to be brought down. Following completion of DCB configuration
 279  * the VSIs that were downed need to be brought up again. This helper function
 280  * does both.
 281  */
 282 static void ice_dcb_ena_dis_vsi(struct ice_pf *pf, bool ena, bool locked)
 283 {
 284         int i;
 285
 286         ice_for_each_vsi(pf, i) {
 287                 struct ice_vsi *vsi = pf->vsi[i];
 288
 289                 if (!vsi)
 290                         continue;
 291
 292                 switch (vsi->type) {
 293                 case ICE_VSI_CHNL:
 294                 case ICE_VSI_SWITCHDEV_CTRL:
 295                 case ICE_VSI_PF:
 296                         if (ena)
 297                                 ice_ena_vsi(vsi, locked);
 298                         else
 299                                 ice_dis_vsi(vsi, locked);
 300                         break;
 301                 default:
 302                         continue;
 303                 }
 304         }
 305 }
 306
 307 /**
 308  * ice_dcb_bwchk - check if ETS bandwidth input parameters are correct
 309  * @pf: pointer to the PF struct
 310  * @dcbcfg: pointer to DCB config structure
 311  */
 312 int ice_dcb_bwchk(struct ice_pf *pf, struct ice_dcbx_cfg *dcbcfg)
 313 {
 314         struct ice_dcb_ets_cfg *etscfg = &dcbcfg->etscfg;
 315         u8 num_tc, total_bw = 0;
 316         int i;
 317
 318         /* returns number of contigous TCs and 1 TC for non-contigous TCs,
 319          * since at least 1 TC has to be configured
 320          */
 321         num_tc = ice_dcb_get_num_tc(dcbcfg);
 322
 323         /* no bandwidth checks required if there's only one TC, so assign
 324          * all bandwidth to TC0 and return
 325          */
 326         if (num_tc == 1) {
 327                 etscfg->tcbwtable[0] = ICE_TC_MAX_BW;
 328                 return 0;
 329         }
 330
 331         for (i = 0; i < num_tc; i++)
 332                 total_bw += etscfg->tcbwtable[i];
 333
 334         if (!total_bw) {
 335                 etscfg->tcbwtable[0] = ICE_TC_MAX_BW;
 336         } else if (total_bw != ICE_TC_MAX_BW) {
 337                 dev_err(ice_pf_to_dev(pf), "Invalid config, total bandwidth must equal 100\n");
 338                 return -EINVAL;
 339         }
 340
 341         return 0;
 342 }
 343
 344 /**
 345  * ice_pf_dcb_cfg - Apply new DCB configuration
 346  * @pf: pointer to the PF struct
 347  * @new_cfg: DCBX config to apply
 348  * @locked: is the RTNL held
 349  */
 350 int ice_pf_dcb_cfg(struct ice_pf *pf, struct ice_dcbx_cfg *new_cfg, bool locked)
 351 {
 352         struct ice_aqc_port_ets_elem buf = { 0 };
 353         struct ice_dcbx_cfg *old_cfg, *curr_cfg;
 354         struct device *dev = ice_pf_to_dev(pf);
 355         int ret = ICE_DCB_NO_HW_CHG;
 356         struct iidc_event *event;
 357         struct ice_vsi *pf_vsi;
 358
 359         curr_cfg = &pf->hw.port_info->qos_cfg.local_dcbx_cfg;
 360
 361         /* FW does not care if change happened */
 362         if (!pf->hw.port_info->qos_cfg.is_sw_lldp)
 363                 ret = ICE_DCB_HW_CHG_RST;
 364
 365         /* Enable DCB tagging only when more than one TC */
 366         if (ice_dcb_get_num_tc(new_cfg) > 1) {
 367                 dev_dbg(dev, "DCB tagging enabled (num TC > 1)\n");
 368                 if (pf->hw.port_info->is_custom_tx_enabled) {
 369                         dev_err(dev, "Custom Tx scheduler feature enabled, can't configure DCB\n");
 370                         return -EBUSY;
 371                 }
 372                 ice_tear_down_devlink_rate_tree(pf);
 373
 374                 set_bit(ICE_FLAG_DCB_ENA, pf->flags);
 375         } else {
 376                 dev_dbg(dev, "DCB tagging disabled (num TC = 1)\n");
 377                 clear_bit(ICE_FLAG_DCB_ENA, pf->flags);
 378         }
 379
 380         if (!memcmp(new_cfg, curr_cfg, sizeof(*new_cfg))) {
 381                 dev_dbg(dev, "No change in DCB config required\n");
 382                 return ret;
 383         }
 384
 385         if (ice_dcb_bwchk(pf, new_cfg))
 386                 return -EINVAL;
 387
 388         /* Store old config in case FW config fails */
 389         old_cfg = kmemdup(curr_cfg, sizeof(*old_cfg), GFP_KERNEL);
 390         if (!old_cfg)
 391                 return -ENOMEM;
 392
 393         dev_info(dev, "Commit DCB Configuration to the hardware\n");
 394         pf_vsi = ice_get_main_vsi(pf);
 395         if (!pf_vsi) {
 396                 dev_dbg(dev, "PF VSI doesn't exist\n");
 397                 ret = -EINVAL;
 398                 goto free_cfg;
 399         }
 400
 401         /* Notify AUX drivers about impending change to TCs */
 402         event = kzalloc(sizeof(*event), GFP_KERNEL);
 403         if (!event) {
 404                 ret = -ENOMEM;
 405                 goto free_cfg;
 406         }
 407
 408         set_bit(IIDC_EVENT_BEFORE_TC_CHANGE, event->type);
 409         ice_send_event_to_aux(pf, event);
 410         kfree(event);
 411
 412         /* avoid race conditions by holding the lock while disabling and
 413          * re-enabling the VSI
 414          */
 415         if (!locked)
 416                 rtnl_lock();
 417
 418         /* disable VSIs affected by DCB changes */
 419         ice_dcb_ena_dis_vsi(pf, false, true);
 420
 421         memcpy(curr_cfg, new_cfg, sizeof(*curr_cfg));
 422         memcpy(&curr_cfg->etsrec, &curr_cfg->etscfg, sizeof(curr_cfg->etsrec));
 423         memcpy(&new_cfg->etsrec, &curr_cfg->etscfg, sizeof(curr_cfg->etsrec));
 424
 425         /* Only send new config to HW if we are in SW LLDP mode. Otherwise,
 426          * the new config came from the HW in the first place.
 427          */
 428         if (pf->hw.port_info->qos_cfg.is_sw_lldp) {
 429                 ret = ice_set_dcb_cfg(pf->hw.port_info);
 430                 if (ret) {
 431                         dev_err(dev, "Set DCB Config failed\n");
 432                         /* Restore previous settings to local config */
 433                         memcpy(curr_cfg, old_cfg, sizeof(*curr_cfg));
 434                         goto out;
 435                 }
 436         }
 437
 438         ret = ice_query_port_ets(pf->hw.port_info, &buf, sizeof(buf), NULL);
 439         if (ret) {
 440                 dev_err(dev, "Query Port ETS failed\n");
 441                 goto out;
 442         }
 443
 444         ice_pf_dcb_recfg(pf);
 445
 446 out:
 447         /* enable previously downed VSIs */
 448         ice_dcb_ena_dis_vsi(pf, true, true);
 449         if (!locked)
 450                 rtnl_unlock();
 451 free_cfg:
 452         kfree(old_cfg);
 453         return ret;
 454 }
 455
 456 /**
 457  * ice_cfg_etsrec_defaults - Set default ETS recommended DCB config
 458  * @pi: port information structure
 459  */
 460 static void ice_cfg_etsrec_defaults(struct ice_port_info *pi)
 461 {
 462         struct ice_dcbx_cfg *dcbcfg = &pi->qos_cfg.local_dcbx_cfg;
 463         u8 i;
 464
 465         /* Ensure ETS recommended DCB configuration is not already set */
 466         if (dcbcfg->etsrec.maxtcs)
 467                 return;
 468
 469         /* In CEE mode, set the default to 1 TC */
 470         dcbcfg->etsrec.maxtcs = 1;
 471         for (i = 0; i < ICE_MAX_TRAFFIC_CLASS; i++) {
 472                 dcbcfg->etsrec.tcbwtable[i] = i ? 0 : 100;
 473                 dcbcfg->etsrec.tsatable[i] = i ? ICE_IEEE_TSA_STRICT :
 474                                                  ICE_IEEE_TSA_ETS;
 475         }
 476 }
 477
 478 /**
 479  * ice_dcb_need_recfg - Check if DCB needs reconfig
 480  * @pf: board private structure
 481  * @old_cfg: current DCB config
 482  * @new_cfg: new DCB config
 483  */
 484 static bool
 485 ice_dcb_need_recfg(struct ice_pf *pf, struct ice_dcbx_cfg *old_cfg,
 486                    struct ice_dcbx_cfg *new_cfg)
 487 {
 488         struct device *dev = ice_pf_to_dev(pf);
 489         bool need_reconfig = false;
 490
 491         /* Check if ETS configuration has changed */
 492         if (memcmp(&new_cfg->etscfg, &old_cfg->etscfg,
 493                    sizeof(new_cfg->etscfg))) {
 494                 /* If Priority Table has changed reconfig is needed */
 495                 if (memcmp(&new_cfg->etscfg.prio_table,
 496                            &old_cfg->etscfg.prio_table,
 497                            sizeof(new_cfg->etscfg.prio_table))) {
 498                         need_reconfig = true;
 499                         dev_dbg(dev, "ETS UP2TC changed.\n");
 500                 }
 501
 502                 if (memcmp(&new_cfg->etscfg.tcbwtable,
 503                            &old_cfg->etscfg.tcbwtable,
 504                            sizeof(new_cfg->etscfg.tcbwtable)))
 505                         dev_dbg(dev, "ETS TC BW Table changed.\n");
 506
 507                 if (memcmp(&new_cfg->etscfg.tsatable,
 508                            &old_cfg->etscfg.tsatable,
 509                            sizeof(new_cfg->etscfg.tsatable)))
 510                         dev_dbg(dev, "ETS TSA Table changed.\n");
 511         }
 512
 513         /* Check if PFC configuration has changed */
 514         if (memcmp(&new_cfg->pfc, &old_cfg->pfc, sizeof(new_cfg->pfc))) {
 515                 need_reconfig = true;
 516                 dev_dbg(dev, "PFC config change detected.\n");
 517         }
 518
 519         /* Check if APP Table has changed */
 520         if (memcmp(&new_cfg->app, &old_cfg->app, sizeof(new_cfg->app))) {
 521                 need_reconfig = true;
 522                 dev_dbg(dev, "APP Table change detected.\n");
 523         }
 524
 525         dev_dbg(dev, "dcb need_reconfig=%d\n", need_reconfig);
 526         return need_reconfig;
 527 }
 528
 529 /**
 530  * ice_dcb_rebuild - rebuild DCB post reset
 531  * @pf: physical function instance
 532  */
 533 void ice_dcb_rebuild(struct ice_pf *pf)
 534 {
 535         struct ice_aqc_port_ets_elem buf = { 0 };
 536         struct device *dev = ice_pf_to_dev(pf);
 537         struct ice_dcbx_cfg *err_cfg;
 538         int ret;
 539
 540         ret = ice_query_port_ets(pf->hw.port_info, &buf, sizeof(buf), NULL);
 541         if (ret) {
 542                 dev_err(dev, "Query Port ETS failed\n");
 543                 goto dcb_error;
 544         }
 545
 546         mutex_lock(&pf->tc_mutex);
 547
 548         if (!pf->hw.port_info->qos_cfg.is_sw_lldp)
 549                 ice_cfg_etsrec_defaults(pf->hw.port_info);
 550
 551         ret = ice_set_dcb_cfg(pf->hw.port_info);
 552         if (ret) {
 553                 dev_err(dev, "Failed to set DCB config in rebuild\n");
 554                 goto dcb_error;
 555         }
 556
 557         if (!pf->hw.port_info->qos_cfg.is_sw_lldp) {
 558                 ret = ice_cfg_lldp_mib_change(&pf->hw, true);
 559                 if (ret && !pf->hw.port_info->qos_cfg.is_sw_lldp) {
 560                         dev_err(dev, "Failed to register for MIB changes\n");
 561                         goto dcb_error;
 562                 }
 563         }
 564
 565         dev_info(dev, "DCB info restored\n");
 566         ret = ice_query_port_ets(pf->hw.port_info, &buf, sizeof(buf), NULL);
 567         if (ret) {
 568                 dev_err(dev, "Query Port ETS failed\n");
 569                 goto dcb_error;
 570         }
 571
 572         mutex_unlock(&pf->tc_mutex);
 573
 574         return;
 575
 576 dcb_error:
 577         dev_err(dev, "Disabling DCB until new settings occur\n");
 578         err_cfg = kzalloc(sizeof(*err_cfg), GFP_KERNEL);
 579         if (!err_cfg) {
 580                 mutex_unlock(&pf->tc_mutex);
 581                 return;
 582         }
 583
 584         err_cfg->etscfg.willing = true;
 585         err_cfg->etscfg.tcbwtable[0] = ICE_TC_MAX_BW;
 586         err_cfg->etscfg.tsatable[0] = ICE_IEEE_TSA_ETS;
 587         memcpy(&err_cfg->etsrec, &err_cfg->etscfg, sizeof(err_cfg->etsrec));
 588         /* Coverity warns the return code of ice_pf_dcb_cfg() is not checked
 589          * here as is done for other calls to that function. That check is
 590          * not necessary since this is in this function's error cleanup path.
 591          * Suppress the Coverity warning with the following comment...
 592          */
 593         /* coverity[check_return] */
 594         ice_pf_dcb_cfg(pf, err_cfg, false);
 595         kfree(err_cfg);
 596
 597         mutex_unlock(&pf->tc_mutex);
 598 }
 599
 600 /**
 601  * ice_dcb_init_cfg - set the initial DCB config in SW
 602  * @pf: PF to apply config to
 603  * @locked: Is the RTNL held
 604  */
 605 static int ice_dcb_init_cfg(struct ice_pf *pf, bool locked)
 606 {
 607         struct ice_dcbx_cfg *newcfg;
 608         struct ice_port_info *pi;
 609         int ret = 0;
 610
 611         pi = pf->hw.port_info;
 612         newcfg = kmemdup(&pi->qos_cfg.local_dcbx_cfg, sizeof(*newcfg),
 613                          GFP_KERNEL);
 614         if (!newcfg)
 615                 return -ENOMEM;
 616
 617         memset(&pi->qos_cfg.local_dcbx_cfg, 0, sizeof(*newcfg));
 618
 619         dev_info(ice_pf_to_dev(pf), "Configuring initial DCB values\n");
 620         if (ice_pf_dcb_cfg(pf, newcfg, locked))
 621                 ret = -EINVAL;
 622
 623         kfree(newcfg);
 624
 625         return ret;
 626 }
 627
 628 /**
 629  * ice_dcb_sw_dflt_cfg - Apply a default DCB config
 630  * @pf: PF to apply config to
 631  * @ets_willing: configure ETS willing
 632  * @locked: was this function called with RTNL held
 633  */
 634 int ice_dcb_sw_dflt_cfg(struct ice_pf *pf, bool ets_willing, bool locked)
 635 {
 636         struct ice_aqc_port_ets_elem buf = { 0 };
 637         struct ice_dcbx_cfg *dcbcfg;
 638         struct ice_port_info *pi;
 639         struct ice_hw *hw;
 640         int ret;
 641
 642         hw = &pf->hw;
 643         pi = hw->port_info;
 644         dcbcfg = kzalloc(sizeof(*dcbcfg), GFP_KERNEL);
 645         if (!dcbcfg)
 646                 return -ENOMEM;
 647
 648         memset(&pi->qos_cfg.local_dcbx_cfg, 0, sizeof(*dcbcfg));
 649
 650         dcbcfg->etscfg.willing = ets_willing ? 1 : 0;
 651         dcbcfg->etscfg.maxtcs = hw->func_caps.common_cap.maxtc;
 652         dcbcfg->etscfg.tcbwtable[0] = 100;
 653         dcbcfg->etscfg.tsatable[0] = ICE_IEEE_TSA_ETS;
 654
 655         memcpy(&dcbcfg->etsrec, &dcbcfg->etscfg,
 656                sizeof(dcbcfg->etsrec));
 657         dcbcfg->etsrec.willing = 0;
 658
 659         dcbcfg->pfc.willing = 1;
 660         dcbcfg->pfc.pfccap = hw->func_caps.common_cap.maxtc;
 661
 662         dcbcfg->numapps = 1;
 663         dcbcfg->app[0].selector = ICE_APP_SEL_ETHTYPE;
 664         dcbcfg->app[0].priority = 3;
 665         dcbcfg->app[0].prot_id = ETH_P_FCOE;
 666
 667         ret = ice_pf_dcb_cfg(pf, dcbcfg, locked);
 668         kfree(dcbcfg);
 669         if (ret)
 670                 return ret;
 671
 672         return ice_query_port_ets(pi, &buf, sizeof(buf), NULL);
 673 }
 674
 675 /**
 676  * ice_dcb_tc_contig - Check that TCs are contiguous
 677  * @prio_table: pointer to priority table
 678  *
 679  * Check if TCs begin with TC0 and are contiguous
 680  */
 681 static bool ice_dcb_tc_contig(u8 *prio_table)
 682 {
 683         bool found_empty = false;
 684         u8 used_tc = 0;
 685         int i;
 686
 687         /* Create a bitmap of used TCs */
 688         for (i = 0; i < CEE_DCBX_MAX_PRIO; i++)
 689                 used_tc |= BIT(prio_table[i]);
 690
 691         for (i = 0; i < CEE_DCBX_MAX_PRIO; i++) {
 692                 if (used_tc & BIT(i)) {
 693                         if (found_empty)
 694                                 return false;
 695                 } else {
 696                         found_empty = true;
 697                 }
 698         }
 699
 700         return true;
 701 }
 702
 703 /**
 704  * ice_dcb_noncontig_cfg - Configure DCB for non-contiguous TCs
 705  * @pf: pointer to the PF struct
 706  *
 707  * If non-contiguous TCs, then configure SW DCB with TC0 and ETS non-willing
 708  */
 709 static int ice_dcb_noncontig_cfg(struct ice_pf *pf)
 710 {
 711         struct ice_dcbx_cfg *dcbcfg = &pf->hw.port_info->qos_cfg.local_dcbx_cfg;
 712         struct device *dev = ice_pf_to_dev(pf);
 713         int ret;
 714
 715         /* Configure SW DCB default with ETS non-willing */
 716         ret = ice_dcb_sw_dflt_cfg(pf, false, true);
 717         if (ret) {
 718                 dev_err(dev, "Failed to set local DCB config %d\n", ret);
 719                 return ret;
 720         }
 721
 722         /* Reconfigure with ETS willing so that FW will send LLDP MIB event */
 723         dcbcfg->etscfg.willing = 1;
 724         ret = ice_set_dcb_cfg(pf->hw.port_info);
 725         if (ret)
 726                 dev_err(dev, "Failed to set DCB to unwilling\n");
 727
 728         return ret;
 729 }
 730
 731 /**
 732  * ice_pf_dcb_recfg - Reconfigure all VEBs and VSIs
 733  * @pf: pointer to the PF struct
 734  *
 735  * Assumed caller has already disabled all VSIs before
 736  * calling this function. Reconfiguring DCB based on
 737  * local_dcbx_cfg.
 738  */
 739 void ice_pf_dcb_recfg(struct ice_pf *pf)
 740 {
 741         struct ice_dcbx_cfg *dcbcfg = &pf->hw.port_info->qos_cfg.local_dcbx_cfg;
 742         struct iidc_event *event;
 743         u8 tc_map = 0;
 744         int v, ret;
 745
 746         /* Update each VSI */
 747         ice_for_each_vsi(pf, v) {
 748                 struct ice_vsi *vsi = pf->vsi[v];
 749
 750                 if (!vsi)
 751                         continue;
 752
 753                 if (vsi->type == ICE_VSI_PF) {
 754                         tc_map = ice_dcb_get_ena_tc(dcbcfg);
 755
 756                         /* If DCBX request non-contiguous TC, then configure
 757                          * default TC
 758                          */
 759                         if (!ice_dcb_tc_contig(dcbcfg->etscfg.prio_table)) {
 760                                 tc_map = ICE_DFLT_TRAFFIC_CLASS;
 761                                 ice_dcb_noncontig_cfg(pf);
 762                         }
 763                 } else if (vsi->type == ICE_VSI_CHNL) {
 764                         tc_map = BIT(ice_get_first_droptc(vsi));
 765                 } else {
 766                         tc_map = ICE_DFLT_TRAFFIC_CLASS;
 767                 }
 768
 769                 ret = ice_vsi_cfg_tc(vsi, tc_map);
 770                 if (ret) {
 771                         dev_err(ice_pf_to_dev(pf), "Failed to config TC for VSI index: %d\n",
 772                                 vsi->idx);
 773                         continue;
 774                 }
 775                 /* no need to proceed with remaining cfg if it is CHNL
 776                  * or switchdev VSI
 777                  */
 778                 if (vsi->type == ICE_VSI_CHNL ||
 779                     vsi->type == ICE_VSI_SWITCHDEV_CTRL)
 780                         continue;
 781
 782                 ice_vsi_map_rings_to_vectors(vsi);
 783                 if (vsi->type == ICE_VSI_PF)
 784                         ice_dcbnl_set_all(vsi);
 785         }
 786         /* Notify the AUX drivers that TC change is finished */
 787         event = kzalloc(sizeof(*event), GFP_KERNEL);
 788         if (!event)
 789                 return;
 790
 791         set_bit(IIDC_EVENT_AFTER_TC_CHANGE, event->type);
 792         ice_send_event_to_aux(pf, event);
 793         kfree(event);
 794 }
 795
 796 /**
 797  * ice_init_pf_dcb - initialize DCB for a PF
 798  * @pf: PF to initialize DCB for
 799  * @locked: Was function called with RTNL held
 800  */
 801 int ice_init_pf_dcb(struct ice_pf *pf, bool locked)
 802 {
 803         struct device *dev = ice_pf_to_dev(pf);
 804         struct ice_port_info *port_info;
 805         struct ice_hw *hw = &pf->hw;
 806         int err;
 807
 808         port_info = hw->port_info;
 809
 810         err = ice_init_dcb(hw, false);
 811         if (err && !port_info->qos_cfg.is_sw_lldp) {
 812                 dev_err(dev, "Error initializing DCB %d\n", err);
 813                 goto dcb_init_err;
 814         }
 815
 816         dev_info(dev, "DCB is enabled in the hardware, max number of TCs supported on this port are %d\n",
 817                  pf->hw.func_caps.common_cap.maxtc);
 818         if (err) {
 819                 struct ice_vsi *pf_vsi;
 820
 821                 /* FW LLDP is disabled, activate SW DCBX/LLDP mode */
 822                 dev_info(dev, "FW LLDP is disabled, DCBx/LLDP in SW mode.\n");
 823                 clear_bit(ICE_FLAG_FW_LLDP_AGENT, pf->flags);
 824                 err = ice_aq_set_pfc_mode(&pf->hw, ICE_AQC_PFC_VLAN_BASED_PFC,
 825                                           NULL);
 826                 if (err)
 827                         dev_info(dev, "Failed to set VLAN PFC mode\n");
 828
 829                 err = ice_dcb_sw_dflt_cfg(pf, true, locked);
 830                 if (err) {
 831                         dev_err(dev, "Failed to set local DCB config %d\n",
 832                                 err);
 833                         err = -EIO;
 834                         goto dcb_init_err;
 835                 }
 836
 837                 /* If the FW DCBX engine is not running then Rx LLDP packets
 838                  * need to be redirected up the stack.
 839                  */
 840                 pf_vsi = ice_get_main_vsi(pf);
 841                 if (!pf_vsi) {
 842                         dev_err(dev, "Failed to set local DCB config\n");
 843                         err = -EIO;
 844                         goto dcb_init_err;
 845                 }
 846
 847                 ice_cfg_sw_lldp(pf_vsi, false, true);
 848
 849                 pf->dcbx_cap = ice_dcb_get_mode(port_info, true);
 850                 return 0;
 851         }
 852
 853         set_bit(ICE_FLAG_FW_LLDP_AGENT, pf->flags);
 854
 855         /* DCBX/LLDP enabled in FW, set DCBNL mode advertisement */
 856         pf->dcbx_cap = ice_dcb_get_mode(port_info, false);
 857
 858         err = ice_dcb_init_cfg(pf, locked);
 859         if (err)
 860                 goto dcb_init_err;
 861
 862         return err;
 863
 864 dcb_init_err:
 865         dev_err(dev, "DCB init failed\n");
 866         return err;
 867 }
 868
 869 /**
 870  * ice_update_dcb_stats - Update DCB stats counters
 871  * @pf: PF whose stats needs to be updated
 872  */
 873 void ice_update_dcb_stats(struct ice_pf *pf)
 874 {
 875         struct ice_hw_port_stats *prev_ps, *cur_ps;
 876         struct ice_hw *hw = &pf->hw;
 877         u8 port;
 878         int i;
 879
 880         port = hw->port_info->lport;
 881         prev_ps = &pf->stats_prev;
 882         cur_ps = &pf->stats;
 883
 884         if (ice_is_reset_in_progress(pf->state))
 885                 pf->stat_prev_loaded = false;
 886
 887         for (i = 0; i < 8; i++) {
 888                 ice_stat_update32(hw, GLPRT_PXOFFRXC(port, i),
 889                                   pf->stat_prev_loaded,
 890                                   &prev_ps->priority_xoff_rx[i],
 891                                   &cur_ps->priority_xoff_rx[i]);
 892                 ice_stat_update32(hw, GLPRT_PXONRXC(port, i),
 893                                   pf->stat_prev_loaded,
 894                                   &prev_ps->priority_xon_rx[i],
 895                                   &cur_ps->priority_xon_rx[i]);
 896                 ice_stat_update32(hw, GLPRT_PXONTXC(port, i),
 897                                   pf->stat_prev_loaded,
 898                                   &prev_ps->priority_xon_tx[i],
 899                                   &cur_ps->priority_xon_tx[i]);
 900                 ice_stat_update32(hw, GLPRT_PXOFFTXC(port, i),
 901                                   pf->stat_prev_loaded,
 902                                   &prev_ps->priority_xoff_tx[i],
 903                                   &cur_ps->priority_xoff_tx[i]);
 904                 ice_stat_update32(hw, GLPRT_RXON2OFFCNT(port, i),
 905                                   pf->stat_prev_loaded,
 906                                   &prev_ps->priority_xon_2_xoff[i],
 907                                   &cur_ps->priority_xon_2_xoff[i]);
 908         }
 909 }
 910
 911 /**
 912  * ice_tx_prepare_vlan_flags_dcb - prepare VLAN tagging for DCB
 913  * @tx_ring: ring to send buffer on
 914  * @first: pointer to struct ice_tx_buf
 915  *
 916  * This should not be called if the outer VLAN is software offloaded as the VLAN
 917  * tag will already be configured with the correct ID and priority bits
 918  */
 919 void
 920 ice_tx_prepare_vlan_flags_dcb(struct ice_tx_ring *tx_ring,
 921                               struct ice_tx_buf *first)
 922 {
 923         struct sk_buff *skb = first->skb;
 924
 925         if (!test_bit(ICE_FLAG_DCB_ENA, tx_ring->vsi->back->flags))
 926                 return;
 927
 928         /* Insert 802.1p priority into VLAN header */
 929         if ((first->tx_flags & ICE_TX_FLAGS_HW_VLAN ||
 930              first->tx_flags & ICE_TX_FLAGS_HW_OUTER_SINGLE_VLAN) ||
 931             skb->priority != TC_PRIO_CONTROL) {
 932                 first->tx_flags &= ~ICE_TX_FLAGS_VLAN_PR_M;
 933                 /* Mask the lower 3 bits to set the 802.1p priority */
 934                 first->tx_flags |= (skb->priority & 0x7) <<
 935                                    ICE_TX_FLAGS_VLAN_PR_S;
 936                 /* if this is not already set it means a VLAN 0 + priority needs
 937                  * to be offloaded
 938                  */
 939                 if (tx_ring->flags & ICE_TX_FLAGS_RING_VLAN_L2TAG2)
 940                         first->tx_flags |= ICE_TX_FLAGS_HW_OUTER_SINGLE_VLAN;
 941                 else
 942                         first->tx_flags |= ICE_TX_FLAGS_HW_VLAN;
 943         }
 944 }
 945
 946 /**
 947  * ice_dcb_process_lldp_set_mib_change - Process MIB change
 948  * @pf: ptr to ice_pf
 949  * @event: pointer to the admin queue receive event
 950  */
 951 void
 952 ice_dcb_process_lldp_set_mib_change(struct ice_pf *pf,
 953                                     struct ice_rq_event_info *event)
 954 {
 955         struct ice_aqc_port_ets_elem buf = { 0 };
 956         struct device *dev = ice_pf_to_dev(pf);
 957         struct ice_aqc_lldp_get_mib *mib;
 958         struct ice_dcbx_cfg tmp_dcbx_cfg;
 959         bool need_reconfig = false;
 960         struct ice_port_info *pi;
 961         u8 mib_type;
 962         int ret;
 963
 964         /* Not DCB capable or capability disabled */
 965         if (!(test_bit(ICE_FLAG_DCB_CAPABLE, pf->flags)))
 966                 return;
 967
 968         if (pf->dcbx_cap & DCB_CAP_DCBX_HOST) {
 969                 dev_dbg(dev, "MIB Change Event in HOST mode\n");
 970                 return;
 971         }
 972
 973         pi = pf->hw.port_info;
 974         mib = (struct ice_aqc_lldp_get_mib *)&event->desc.params.raw;
 975         /* Ignore if event is not for Nearest Bridge */
 976         mib_type = ((mib->type >> ICE_AQ_LLDP_BRID_TYPE_S) &
 977                     ICE_AQ_LLDP_BRID_TYPE_M);
 978         dev_dbg(dev, "LLDP event MIB bridge type 0x%x\n", mib_type);
 979         if (mib_type != ICE_AQ_LLDP_BRID_TYPE_NEAREST_BRID)
 980                 return;
 981
 982         /* Check MIB Type and return if event for Remote MIB update */
 983         mib_type = mib->type & ICE_AQ_LLDP_MIB_TYPE_M;
 984         dev_dbg(dev, "LLDP event mib type %s\n", mib_type ? "remote" : "local");
 985         if (mib_type == ICE_AQ_LLDP_MIB_REMOTE) {
 986                 /* Update the remote cached instance and return */
 987                 ret = ice_aq_get_dcb_cfg(pi->hw, ICE_AQ_LLDP_MIB_REMOTE,
 988                                          ICE_AQ_LLDP_BRID_TYPE_NEAREST_BRID,
 989                                          &pi->qos_cfg.remote_dcbx_cfg);
 990                 if (ret) {
 991                         dev_err(dev, "Failed to get remote DCB config\n");
 992                         return;
 993                 }
 994         }
 995
 996         mutex_lock(&pf->tc_mutex);
 997
 998         /* store the old configuration */
 999         tmp_dcbx_cfg = pf->hw.port_info->qos_cfg.local_dcbx_cfg;
1000
1001         /* Reset the old DCBX configuration data */
1002         memset(&pi->qos_cfg.local_dcbx_cfg, 0,
1003                sizeof(pi->qos_cfg.local_dcbx_cfg));
1004
1005         /* Get updated DCBX data from firmware */
1006         ret = ice_get_dcb_cfg(pf->hw.port_info);
1007         if (ret) {
1008                 dev_err(dev, "Failed to get DCB config\n");
1009                 goto out;
1010         }
1011
1012         /* No change detected in DCBX configs */
1013         if (!memcmp(&tmp_dcbx_cfg, &pi->qos_cfg.local_dcbx_cfg,
1014                     sizeof(tmp_dcbx_cfg))) {
1015                 dev_dbg(dev, "No change detected in DCBX configuration.\n");
1016                 goto out;
1017         }
1018
1019         pf->dcbx_cap = ice_dcb_get_mode(pi, false);
1020
1021         need_reconfig = ice_dcb_need_recfg(pf, &tmp_dcbx_cfg,
1022                                            &pi->qos_cfg.local_dcbx_cfg);
1023         ice_dcbnl_flush_apps(pf, &tmp_dcbx_cfg, &pi->qos_cfg.local_dcbx_cfg);
1024         if (!need_reconfig)
1025                 goto out;
1026
1027         /* Enable DCB tagging only when more than one TC */
1028         if (ice_dcb_get_num_tc(&pi->qos_cfg.local_dcbx_cfg) > 1) {
1029                 dev_dbg(dev, "DCB tagging enabled (num TC > 1)\n");
1030                 set_bit(ICE_FLAG_DCB_ENA, pf->flags);
1031         } else {
1032                 dev_dbg(dev, "DCB tagging disabled (num TC = 1)\n");
1033                 clear_bit(ICE_FLAG_DCB_ENA, pf->flags);
1034         }
1035
1036         rtnl_lock();
1037         /* disable VSIs affected by DCB changes */
1038         ice_dcb_ena_dis_vsi(pf, false, true);
1039
1040         ret = ice_query_port_ets(pf->hw.port_info, &buf, sizeof(buf), NULL);
1041         if (ret) {
1042                 dev_err(dev, "Query Port ETS failed\n");
1043                 goto unlock_rtnl;
1044         }
1045
1046         /* changes in configuration update VSI */
1047         ice_pf_dcb_recfg(pf);
1048
1049         /* enable previously downed VSIs */
1050         ice_dcb_ena_dis_vsi(pf, true, true);
1051 unlock_rtnl:
1052         rtnl_unlock();
1053 out:
1054         mutex_unlock(&pf->tc_mutex);
1055 }