drivers/net/ethernet/emulex/benet/be_main.c

   1 /*
   2  * Copyright (C) 2005 - 2015 Emulex
   3  * All rights reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License version 2
   7  * as published by the Free Software Foundation.  The full GNU General
   8  * Public License is included in this distribution in the file called COPYING.
   9  *
  10  * Contact Information:
  11  * linux-drivers@emulex.com
  12  *
  13  * Emulex
  14  * 3333 Susan Street
  15  * Costa Mesa, CA 92626
  16  */
  17
  18 #include <linux/prefetch.h>
  19 #include <linux/module.h>
  20 #include "be.h"
  21 #include "be_cmds.h"
  22 #include <asm/div64.h>
  23 #include <linux/aer.h>
  24 #include <linux/if_bridge.h>
  25 #include <net/busy_poll.h>
  26 #include <net/vxlan.h>
  27
  28 MODULE_VERSION(DRV_VER);
  29 MODULE_DESCRIPTION(DRV_DESC " " DRV_VER);
  30 MODULE_AUTHOR("Emulex Corporation");
  31 MODULE_LICENSE("GPL");
  32
  33 /* num_vfs module param is obsolete.
  34  * Use sysfs method to enable/disable VFs.
  35  */
  36 static unsigned int num_vfs;
  37 module_param(num_vfs, uint, S_IRUGO);
  38 MODULE_PARM_DESC(num_vfs, "Number of PCI VFs to initialize");
  39
  40 static ushort rx_frag_size = 2048;
  41 module_param(rx_frag_size, ushort, S_IRUGO);
  42 MODULE_PARM_DESC(rx_frag_size, "Size of a fragment that holds rcvd data.");
  43
  44 static const struct pci_device_id be_dev_ids[] = {
  45         { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
  46         { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID2) },
  47         { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
  48         { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
  49         { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID3)},
  50         { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID4)},
  51         { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID5)},
  52         { PCI_DEVICE(EMULEX_VENDOR_ID, OC_DEVICE_ID6)},
  53         { 0 }
  54 };
  55 MODULE_DEVICE_TABLE(pci, be_dev_ids);
  56 /* UE Status Low CSR */
  57 static const char * const ue_status_low_desc[] = {
  58         "CEV",
  59         "CTX",
  60         "DBUF",
  61         "ERX",
  62         "Host",
  63         "MPU",
  64         "NDMA",
  65         "PTC ",
  66         "RDMA ",
  67         "RXF ",
  68         "RXIPS ",
  69         "RXULP0 ",
  70         "RXULP1 ",
  71         "RXULP2 ",
  72         "TIM ",
  73         "TPOST ",
  74         "TPRE ",
  75         "TXIPS ",
  76         "TXULP0 ",
  77         "TXULP1 ",
  78         "UC ",
  79         "WDMA ",
  80         "TXULP2 ",
  81         "HOST1 ",
  82         "P0_OB_LINK ",
  83         "P1_OB_LINK ",
  84         "HOST_GPIO ",
  85         "MBOX ",
  86         "ERX2 ",
  87         "SPARE ",
  88         "JTAG ",
  89         "MPU_INTPEND "
  90 };
  91
  92 /* UE Status High CSR */
  93 static const char * const ue_status_hi_desc[] = {
  94         "LPCMEMHOST",
  95         "MGMT_MAC",
  96         "PCS0ONLINE",
  97         "MPU_IRAM",
  98         "PCS1ONLINE",
  99         "PCTL0",
 100         "PCTL1",
 101         "PMEM",
 102         "RR",
 103         "TXPB",
 104         "RXPP",
 105         "XAUI",
 106         "TXP",
 107         "ARM",
 108         "IPC",
 109         "HOST2",
 110         "HOST3",
 111         "HOST4",
 112         "HOST5",
 113         "HOST6",
 114         "HOST7",
 115         "ECRC",
 116         "Poison TLP",
 117         "NETC",
 118         "PERIPH",
 119         "LLTXULP",
 120         "D2P",
 121         "RCON",
 122         "LDMA",
 123         "LLTXP",
 124         "LLTXPB",
 125         "Unknown"
 126 };
 127
 128 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q)
 129 {
 130         struct be_dma_mem *mem = &q->dma_mem;
 131
 132         if (mem->va) {
 133                 dma_free_coherent(&adapter->pdev->dev, mem->size, mem->va,
 134                                   mem->dma);
 135                 mem->va = NULL;
 136         }
 137 }
 138
 139 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q,
 140                           u16 len, u16 entry_size)
 141 {
 142         struct be_dma_mem *mem = &q->dma_mem;
 143
 144         memset(q, 0, sizeof(*q));
 145         q->len = len;
 146         q->entry_size = entry_size;
 147         mem->size = len * entry_size;
 148         mem->va = dma_zalloc_coherent(&adapter->pdev->dev, mem->size, &mem->dma,
 149                                       GFP_KERNEL);
 150         if (!mem->va)
 151                 return -ENOMEM;
 152         return 0;
 153 }
 154
 155 static void be_reg_intr_set(struct be_adapter *adapter, bool enable)
 156 {
 157         u32 reg, enabled;
 158
 159         pci_read_config_dword(adapter->pdev, PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET,
 160                               &reg);
 161         enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
 162
 163         if (!enabled && enable)
 164                 reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
 165         else if (enabled && !enable)
 166                 reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
 167         else
 168                 return;
 169
 170         pci_write_config_dword(adapter->pdev,
 171                                PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET, reg);
 172 }
 173
 174 static void be_intr_set(struct be_adapter *adapter, bool enable)
 175 {
 176         int status = 0;
 177
 178         /* On lancer interrupts can't be controlled via this register */
 179         if (lancer_chip(adapter))
 180                 return;
 181
 182         if (be_check_error(adapter, BE_ERROR_EEH))
 183                 return;
 184
 185         status = be_cmd_intr_set(adapter, enable);
 186         if (status)
 187                 be_reg_intr_set(adapter, enable);
 188 }
 189
 190 static void be_rxq_notify(struct be_adapter *adapter, u16 qid, u16 posted)
 191 {
 192         u32 val = 0;
 193
 194         if (be_check_error(adapter, BE_ERROR_HW))
 195                 return;
 196
 197         val |= qid & DB_RQ_RING_ID_MASK;
 198         val |= posted << DB_RQ_NUM_POSTED_SHIFT;
 199
 200         wmb();
 201         iowrite32(val, adapter->db + DB_RQ_OFFSET);
 202 }
 203
 204 static void be_txq_notify(struct be_adapter *adapter, struct be_tx_obj *txo,
 205                           u16 posted)
 206 {
 207         u32 val = 0;
 208
 209         if (be_check_error(adapter, BE_ERROR_HW))
 210                 return;
 211
 212         val |= txo->q.id & DB_TXULP_RING_ID_MASK;
 213         val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT;
 214
 215         wmb();
 216         iowrite32(val, adapter->db + txo->db_offset);
 217 }
 218
 219 static void be_eq_notify(struct be_adapter *adapter, u16 qid,
 220                          bool arm, bool clear_int, u16 num_popped,
 221                          u32 eq_delay_mult_enc)
 222 {
 223         u32 val = 0;
 224
 225         val |= qid & DB_EQ_RING_ID_MASK;
 226         val |= ((qid & DB_EQ_RING_ID_EXT_MASK) << DB_EQ_RING_ID_EXT_MASK_SHIFT);
 227
 228         if (be_check_error(adapter, BE_ERROR_HW))
 229                 return;
 230
 231         if (arm)
 232                 val |= 1 << DB_EQ_REARM_SHIFT;
 233         if (clear_int)
 234                 val |= 1 << DB_EQ_CLR_SHIFT;
 235         val |= 1 << DB_EQ_EVNT_SHIFT;
 236         val |= num_popped << DB_EQ_NUM_POPPED_SHIFT;
 237         val |= eq_delay_mult_enc << DB_EQ_R2I_DLY_SHIFT;
 238         iowrite32(val, adapter->db + DB_EQ_OFFSET);
 239 }
 240
 241 void be_cq_notify(struct be_adapter *adapter, u16 qid, bool arm, u16 num_popped)
 242 {
 243         u32 val = 0;
 244
 245         val |= qid & DB_CQ_RING_ID_MASK;
 246         val |= ((qid & DB_CQ_RING_ID_EXT_MASK) <<
 247                         DB_CQ_RING_ID_EXT_MASK_SHIFT);
 248
 249         if (be_check_error(adapter, BE_ERROR_HW))
 250                 return;
 251
 252         if (arm)
 253                 val |= 1 << DB_CQ_REARM_SHIFT;
 254         val |= num_popped << DB_CQ_NUM_POPPED_SHIFT;
 255         iowrite32(val, adapter->db + DB_CQ_OFFSET);
 256 }
 257
 258 static int be_mac_addr_set(struct net_device *netdev, void *p)
 259 {
 260         struct be_adapter *adapter = netdev_priv(netdev);
 261         struct device *dev = &adapter->pdev->dev;
 262         struct sockaddr *addr = p;
 263         int status;
 264         u8 mac[ETH_ALEN];
 265         u32 old_pmac_id = adapter->pmac_id[0], curr_pmac_id = 0;
 266
 267         if (!is_valid_ether_addr(addr->sa_data))
 268                 return -EADDRNOTAVAIL;
 269
 270         /* Proceed further only if, User provided MAC is different
 271          * from active MAC
 272          */
 273         if (ether_addr_equal(addr->sa_data, netdev->dev_addr))
 274                 return 0;
 275
 276         /* if device is not running, copy MAC to netdev->dev_addr */
 277         if (!netif_running(netdev))
 278                 goto done;
 279
 280         /* The PMAC_ADD cmd may fail if the VF doesn't have FILTMGMT
 281          * privilege or if PF did not provision the new MAC address.
 282          * On BE3, this cmd will always fail if the VF doesn't have the
 283          * FILTMGMT privilege. This failure is OK, only if the PF programmed
 284          * the MAC for the VF.
 285          */
 286         status = be_cmd_pmac_add(adapter, (u8 *)addr->sa_data,
 287                                  adapter->if_handle, &adapter->pmac_id[0], 0);
 288         if (!status) {
 289                 curr_pmac_id = adapter->pmac_id[0];
 290
 291                 /* Delete the old programmed MAC. This call may fail if the
 292                  * old MAC was already deleted by the PF driver.
 293                  */
 294                 if (adapter->pmac_id[0] != old_pmac_id)
 295                         be_cmd_pmac_del(adapter, adapter->if_handle,
 296                                         old_pmac_id, 0);
 297         }
 298
 299         /* Decide if the new MAC is successfully activated only after
 300          * querying the FW
 301          */
 302         status = be_cmd_get_active_mac(adapter, curr_pmac_id, mac,
 303                                        adapter->if_handle, true, 0);
 304         if (status)
 305                 goto err;
 306
 307         /* The MAC change did not happen, either due to lack of privilege
 308          * or PF didn't pre-provision.
 309          */
 310         if (!ether_addr_equal(addr->sa_data, mac)) {
 311                 status = -EPERM;
 312                 goto err;
 313         }
 314 done:
 315         ether_addr_copy(netdev->dev_addr, addr->sa_data);
 316         dev_info(dev, "MAC address changed to %pM\n", addr->sa_data);
 317         return 0;
 318 err:
 319         dev_warn(dev, "MAC address change to %pM failed\n", addr->sa_data);
 320         return status;
 321 }
 322
 323 /* BE2 supports only v0 cmd */
 324 static void *hw_stats_from_cmd(struct be_adapter *adapter)
 325 {
 326         if (BE2_chip(adapter)) {
 327                 struct be_cmd_resp_get_stats_v0 *cmd = adapter->stats_cmd.va;
 328
 329                 return &cmd->hw_stats;
 330         } else if (BE3_chip(adapter)) {
 331                 struct be_cmd_resp_get_stats_v1 *cmd = adapter->stats_cmd.va;
 332
 333                 return &cmd->hw_stats;
 334         } else {
 335                 struct be_cmd_resp_get_stats_v2 *cmd = adapter->stats_cmd.va;
 336
 337                 return &cmd->hw_stats;
 338         }
 339 }
 340
 341 /* BE2 supports only v0 cmd */
 342 static void *be_erx_stats_from_cmd(struct be_adapter *adapter)
 343 {
 344         if (BE2_chip(adapter)) {
 345                 struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
 346
 347                 return &hw_stats->erx;
 348         } else if (BE3_chip(adapter)) {
 349                 struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
 350
 351                 return &hw_stats->erx;
 352         } else {
 353                 struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
 354
 355                 return &hw_stats->erx;
 356         }
 357 }
 358
 359 static void populate_be_v0_stats(struct be_adapter *adapter)
 360 {
 361         struct be_hw_stats_v0 *hw_stats = hw_stats_from_cmd(adapter);
 362         struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
 363         struct be_rxf_stats_v0 *rxf_stats = &hw_stats->rxf;
 364         struct be_port_rxf_stats_v0 *port_stats =
 365                                         &rxf_stats->port[adapter->port_num];
 366         struct be_drv_stats *drvs = &adapter->drv_stats;
 367
 368         be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
 369         drvs->rx_pause_frames = port_stats->rx_pause_frames;
 370         drvs->rx_crc_errors = port_stats->rx_crc_errors;
 371         drvs->rx_control_frames = port_stats->rx_control_frames;
 372         drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
 373         drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
 374         drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
 375         drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
 376         drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
 377         drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
 378         drvs->rxpp_fifo_overflow_drop = port_stats->rx_fifo_overflow;
 379         drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
 380         drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
 381         drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
 382         drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
 383         drvs->rx_input_fifo_overflow_drop = port_stats->rx_input_fifo_overflow;
 384         drvs->rx_dropped_header_too_small =
 385                 port_stats->rx_dropped_header_too_small;
 386         drvs->rx_address_filtered =
 387                                         port_stats->rx_address_filtered +
 388                                         port_stats->rx_vlan_filtered;
 389         drvs->rx_alignment_symbol_errors =
 390                 port_stats->rx_alignment_symbol_errors;
 391
 392         drvs->tx_pauseframes = port_stats->tx_pauseframes;
 393         drvs->tx_controlframes = port_stats->tx_controlframes;
 394
 395         if (adapter->port_num)
 396                 drvs->jabber_events = rxf_stats->port1_jabber_events;
 397         else
 398                 drvs->jabber_events = rxf_stats->port0_jabber_events;
 399         drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
 400         drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
 401         drvs->forwarded_packets = rxf_stats->forwarded_packets;
 402         drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
 403         drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
 404         drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
 405         adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
 406 }
 407
 408 static void populate_be_v1_stats(struct be_adapter *adapter)
 409 {
 410         struct be_hw_stats_v1 *hw_stats = hw_stats_from_cmd(adapter);
 411         struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
 412         struct be_rxf_stats_v1 *rxf_stats = &hw_stats->rxf;
 413         struct be_port_rxf_stats_v1 *port_stats =
 414                                         &rxf_stats->port[adapter->port_num];
 415         struct be_drv_stats *drvs = &adapter->drv_stats;
 416
 417         be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
 418         drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
 419         drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
 420         drvs->rx_pause_frames = port_stats->rx_pause_frames;
 421         drvs->rx_crc_errors = port_stats->rx_crc_errors;
 422         drvs->rx_control_frames = port_stats->rx_control_frames;
 423         drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
 424         drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
 425         drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
 426         drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
 427         drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
 428         drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
 429         drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
 430         drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
 431         drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
 432         drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
 433         drvs->rx_dropped_header_too_small =
 434                 port_stats->rx_dropped_header_too_small;
 435         drvs->rx_input_fifo_overflow_drop =
 436                 port_stats->rx_input_fifo_overflow_drop;
 437         drvs->rx_address_filtered = port_stats->rx_address_filtered;
 438         drvs->rx_alignment_symbol_errors =
 439                 port_stats->rx_alignment_symbol_errors;
 440         drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
 441         drvs->tx_pauseframes = port_stats->tx_pauseframes;
 442         drvs->tx_controlframes = port_stats->tx_controlframes;
 443         drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
 444         drvs->jabber_events = port_stats->jabber_events;
 445         drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
 446         drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
 447         drvs->forwarded_packets = rxf_stats->forwarded_packets;
 448         drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
 449         drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
 450         drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
 451         adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
 452 }
 453
 454 static void populate_be_v2_stats(struct be_adapter *adapter)
 455 {
 456         struct be_hw_stats_v2 *hw_stats = hw_stats_from_cmd(adapter);
 457         struct be_pmem_stats *pmem_sts = &hw_stats->pmem;
 458         struct be_rxf_stats_v2 *rxf_stats = &hw_stats->rxf;
 459         struct be_port_rxf_stats_v2 *port_stats =
 460                                         &rxf_stats->port[adapter->port_num];
 461         struct be_drv_stats *drvs = &adapter->drv_stats;
 462
 463         be_dws_le_to_cpu(hw_stats, sizeof(*hw_stats));
 464         drvs->pmem_fifo_overflow_drop = port_stats->pmem_fifo_overflow_drop;
 465         drvs->rx_priority_pause_frames = port_stats->rx_priority_pause_frames;
 466         drvs->rx_pause_frames = port_stats->rx_pause_frames;
 467         drvs->rx_crc_errors = port_stats->rx_crc_errors;
 468         drvs->rx_control_frames = port_stats->rx_control_frames;
 469         drvs->rx_in_range_errors = port_stats->rx_in_range_errors;
 470         drvs->rx_frame_too_long = port_stats->rx_frame_too_long;
 471         drvs->rx_dropped_runt = port_stats->rx_dropped_runt;
 472         drvs->rx_ip_checksum_errs = port_stats->rx_ip_checksum_errs;
 473         drvs->rx_tcp_checksum_errs = port_stats->rx_tcp_checksum_errs;
 474         drvs->rx_udp_checksum_errs = port_stats->rx_udp_checksum_errs;
 475         drvs->rx_dropped_tcp_length = port_stats->rx_dropped_tcp_length;
 476         drvs->rx_dropped_too_small = port_stats->rx_dropped_too_small;
 477         drvs->rx_dropped_too_short = port_stats->rx_dropped_too_short;
 478         drvs->rx_out_range_errors = port_stats->rx_out_range_errors;
 479         drvs->rx_dropped_header_too_small =
 480                 port_stats->rx_dropped_header_too_small;
 481         drvs->rx_input_fifo_overflow_drop =
 482                 port_stats->rx_input_fifo_overflow_drop;
 483         drvs->rx_address_filtered = port_stats->rx_address_filtered;
 484         drvs->rx_alignment_symbol_errors =
 485                 port_stats->rx_alignment_symbol_errors;
 486         drvs->rxpp_fifo_overflow_drop = port_stats->rxpp_fifo_overflow_drop;
 487         drvs->tx_pauseframes = port_stats->tx_pauseframes;
 488         drvs->tx_controlframes = port_stats->tx_controlframes;
 489         drvs->tx_priority_pauseframes = port_stats->tx_priority_pauseframes;
 490         drvs->jabber_events = port_stats->jabber_events;
 491         drvs->rx_drops_no_pbuf = rxf_stats->rx_drops_no_pbuf;
 492         drvs->rx_drops_no_erx_descr = rxf_stats->rx_drops_no_erx_descr;
 493         drvs->forwarded_packets = rxf_stats->forwarded_packets;
 494         drvs->rx_drops_mtu = rxf_stats->rx_drops_mtu;
 495         drvs->rx_drops_no_tpre_descr = rxf_stats->rx_drops_no_tpre_descr;
 496         drvs->rx_drops_too_many_frags = rxf_stats->rx_drops_too_many_frags;
 497         adapter->drv_stats.eth_red_drops = pmem_sts->eth_red_drops;
 498         if (be_roce_supported(adapter)) {
 499                 drvs->rx_roce_bytes_lsd = port_stats->roce_bytes_received_lsd;
 500                 drvs->rx_roce_bytes_msd = port_stats->roce_bytes_received_msd;
 501                 drvs->rx_roce_frames = port_stats->roce_frames_received;
 502                 drvs->roce_drops_crc = port_stats->roce_drops_crc;
 503                 drvs->roce_drops_payload_len =
 504                         port_stats->roce_drops_payload_len;
 505         }
 506 }
 507
 508 static void populate_lancer_stats(struct be_adapter *adapter)
 509 {
 510         struct be_drv_stats *drvs = &adapter->drv_stats;
 511         struct lancer_pport_stats *pport_stats = pport_stats_from_cmd(adapter);
 512
 513         be_dws_le_to_cpu(pport_stats, sizeof(*pport_stats));
 514         drvs->rx_pause_frames = pport_stats->rx_pause_frames_lo;
 515         drvs->rx_crc_errors = pport_stats->rx_crc_errors_lo;
 516         drvs->rx_control_frames = pport_stats->rx_control_frames_lo;
 517         drvs->rx_in_range_errors = pport_stats->rx_in_range_errors;
 518         drvs->rx_frame_too_long = pport_stats->rx_frames_too_long_lo;
 519         drvs->rx_dropped_runt = pport_stats->rx_dropped_runt;
 520         drvs->rx_ip_checksum_errs = pport_stats->rx_ip_checksum_errors;
 521         drvs->rx_tcp_checksum_errs = pport_stats->rx_tcp_checksum_errors;
 522         drvs->rx_udp_checksum_errs = pport_stats->rx_udp_checksum_errors;
 523         drvs->rx_dropped_tcp_length =
 524                                 pport_stats->rx_dropped_invalid_tcp_length;
 525         drvs->rx_dropped_too_small = pport_stats->rx_dropped_too_small;
 526         drvs->rx_dropped_too_short = pport_stats->rx_dropped_too_short;
 527         drvs->rx_out_range_errors = pport_stats->rx_out_of_range_errors;
 528         drvs->rx_dropped_header_too_small =
 529                                 pport_stats->rx_dropped_header_too_small;
 530         drvs->rx_input_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
 531         drvs->rx_address_filtered =
 532                                         pport_stats->rx_address_filtered +
 533                                         pport_stats->rx_vlan_filtered;
 534         drvs->rx_alignment_symbol_errors = pport_stats->rx_symbol_errors_lo;
 535         drvs->rxpp_fifo_overflow_drop = pport_stats->rx_fifo_overflow;
 536         drvs->tx_pauseframes = pport_stats->tx_pause_frames_lo;
 537         drvs->tx_controlframes = pport_stats->tx_control_frames_lo;
 538         drvs->jabber_events = pport_stats->rx_jabbers;
 539         drvs->forwarded_packets = pport_stats->num_forwards_lo;
 540         drvs->rx_drops_mtu = pport_stats->rx_drops_mtu_lo;
 541         drvs->rx_drops_too_many_frags =
 542                                 pport_stats->rx_drops_too_many_frags_lo;
 543 }
 544
 545 static void accumulate_16bit_val(u32 *acc, u16 val)
 546 {
 547 #define lo(x)                   (x & 0xFFFF)
 548 #define hi(x)                   (x & 0xFFFF0000)
 549         bool wrapped = val < lo(*acc);
 550         u32 newacc = hi(*acc) + val;
 551
 552         if (wrapped)
 553                 newacc += 65536;
 554         ACCESS_ONCE(*acc) = newacc;
 555 }
 556
 557 static void populate_erx_stats(struct be_adapter *adapter,
 558                                struct be_rx_obj *rxo, u32 erx_stat)
 559 {
 560         if (!BEx_chip(adapter))
 561                 rx_stats(rxo)->rx_drops_no_frags = erx_stat;
 562         else
 563                 /* below erx HW counter can actually wrap around after
 564                  * 65535. Driver accumulates a 32-bit value
 565                  */
 566                 accumulate_16bit_val(&rx_stats(rxo)->rx_drops_no_frags,
 567                                      (u16)erx_stat);
 568 }
 569
 570 void be_parse_stats(struct be_adapter *adapter)
 571 {
 572         struct be_erx_stats_v2 *erx = be_erx_stats_from_cmd(adapter);
 573         struct be_rx_obj *rxo;
 574         int i;
 575         u32 erx_stat;
 576
 577         if (lancer_chip(adapter)) {
 578                 populate_lancer_stats(adapter);
 579         } else {
 580                 if (BE2_chip(adapter))
 581                         populate_be_v0_stats(adapter);
 582                 else if (BE3_chip(adapter))
 583                         /* for BE3 */
 584                         populate_be_v1_stats(adapter);
 585                 else
 586                         populate_be_v2_stats(adapter);
 587
 588                 /* erx_v2 is longer than v0, v1. use v2 for v0, v1 access */
 589                 for_all_rx_queues(adapter, rxo, i) {
 590                         erx_stat = erx->rx_drops_no_fragments[rxo->q.id];
 591                         populate_erx_stats(adapter, rxo, erx_stat);
 592                 }
 593         }
 594 }
 595
 596 static struct rtnl_link_stats64 *be_get_stats64(struct net_device *netdev,
 597                                                 struct rtnl_link_stats64 *stats)
 598 {
 599         struct be_adapter *adapter = netdev_priv(netdev);
 600         struct be_drv_stats *drvs = &adapter->drv_stats;
 601         struct be_rx_obj *rxo;
 602         struct be_tx_obj *txo;
 603         u64 pkts, bytes;
 604         unsigned int start;
 605         int i;
 606
 607         for_all_rx_queues(adapter, rxo, i) {
 608                 const struct be_rx_stats *rx_stats = rx_stats(rxo);
 609
 610                 do {
 611                         start = u64_stats_fetch_begin_irq(&rx_stats->sync);
 612                         pkts = rx_stats(rxo)->rx_pkts;
 613                         bytes = rx_stats(rxo)->rx_bytes;
 614                 } while (u64_stats_fetch_retry_irq(&rx_stats->sync, start));
 615                 stats->rx_packets += pkts;
 616                 stats->rx_bytes += bytes;
 617                 stats->multicast += rx_stats(rxo)->rx_mcast_pkts;
 618                 stats->rx_dropped += rx_stats(rxo)->rx_drops_no_skbs +
 619                                         rx_stats(rxo)->rx_drops_no_frags;
 620         }
 621
 622         for_all_tx_queues(adapter, txo, i) {
 623                 const struct be_tx_stats *tx_stats = tx_stats(txo);
 624
 625                 do {
 626                         start = u64_stats_fetch_begin_irq(&tx_stats->sync);
 627                         pkts = tx_stats(txo)->tx_pkts;
 628                         bytes = tx_stats(txo)->tx_bytes;
 629                 } while (u64_stats_fetch_retry_irq(&tx_stats->sync, start));
 630                 stats->tx_packets += pkts;
 631                 stats->tx_bytes += bytes;
 632         }
 633
 634         /* bad pkts received */
 635         stats->rx_errors = drvs->rx_crc_errors +
 636                 drvs->rx_alignment_symbol_errors +
 637                 drvs->rx_in_range_errors +
 638                 drvs->rx_out_range_errors +
 639                 drvs->rx_frame_too_long +
 640                 drvs->rx_dropped_too_small +
 641                 drvs->rx_dropped_too_short +
 642                 drvs->rx_dropped_header_too_small +
 643                 drvs->rx_dropped_tcp_length +
 644                 drvs->rx_dropped_runt;
 645
 646         /* detailed rx errors */
 647         stats->rx_length_errors = drvs->rx_in_range_errors +
 648                 drvs->rx_out_range_errors +
 649                 drvs->rx_frame_too_long;
 650
 651         stats->rx_crc_errors = drvs->rx_crc_errors;
 652
 653         /* frame alignment errors */
 654         stats->rx_frame_errors = drvs->rx_alignment_symbol_errors;
 655
 656         /* receiver fifo overrun */
 657         /* drops_no_pbuf is no per i/f, it's per BE card */
 658         stats->rx_fifo_errors = drvs->rxpp_fifo_overflow_drop +
 659                                 drvs->rx_input_fifo_overflow_drop +
 660                                 drvs->rx_drops_no_pbuf;
 661         return stats;
 662 }
 663
 664 void be_link_status_update(struct be_adapter *adapter, u8 link_status)
 665 {
 666         struct net_device *netdev = adapter->netdev;
 667
 668         if (!(adapter->flags & BE_FLAGS_LINK_STATUS_INIT)) {
 669                 netif_carrier_off(netdev);
 670                 adapter->flags |= BE_FLAGS_LINK_STATUS_INIT;
 671         }
 672
 673         if (link_status)
 674                 netif_carrier_on(netdev);
 675         else
 676                 netif_carrier_off(netdev);
 677
 678         netdev_info(netdev, "Link is %s\n", link_status ? "Up" : "Down");
 679 }
 680
 681 static void be_tx_stats_update(struct be_tx_obj *txo, struct sk_buff *skb)
 682 {
 683         struct be_tx_stats *stats = tx_stats(txo);
 684         u64 tx_pkts = skb_shinfo(skb)->gso_segs ? : 1;
 685
 686         u64_stats_update_begin(&stats->sync);
 687         stats->tx_reqs++;
 688         stats->tx_bytes += skb->len;
 689         stats->tx_pkts += tx_pkts;
 690         if (skb->encapsulation && skb->ip_summed == CHECKSUM_PARTIAL)
 691                 stats->tx_vxlan_offload_pkts += tx_pkts;
 692         u64_stats_update_end(&stats->sync);
 693 }
 694
 695 /* Returns number of WRBs needed for the skb */
 696 static u32 skb_wrb_cnt(struct sk_buff *skb)
 697 {
 698         /* +1 for the header wrb */
 699         return 1 + (skb_headlen(skb) ? 1 : 0) + skb_shinfo(skb)->nr_frags;
 700 }
 701
 702 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
 703 {
 704         wrb->frag_pa_hi = cpu_to_le32(upper_32_bits(addr));
 705         wrb->frag_pa_lo = cpu_to_le32(lower_32_bits(addr));
 706         wrb->frag_len = cpu_to_le32(len & ETH_WRB_FRAG_LEN_MASK);
 707         wrb->rsvd0 = 0;
 708 }
 709
 710 /* A dummy wrb is just all zeros. Using a separate routine for dummy-wrb
 711  * to avoid the swap and shift/mask operations in wrb_fill().
 712  */
 713 static inline void wrb_fill_dummy(struct be_eth_wrb *wrb)
 714 {
 715         wrb->frag_pa_hi = 0;
 716         wrb->frag_pa_lo = 0;
 717         wrb->frag_len = 0;
 718         wrb->rsvd0 = 0;
 719 }
 720
 721 static inline u16 be_get_tx_vlan_tag(struct be_adapter *adapter,
 722                                      struct sk_buff *skb)
 723 {
 724         u8 vlan_prio;
 725         u16 vlan_tag;
 726
 727         vlan_tag = skb_vlan_tag_get(skb);
 728         vlan_prio = (vlan_tag & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
 729         /* If vlan priority provided by OS is NOT in available bmap */
 730         if (!(adapter->vlan_prio_bmap & (1 << vlan_prio)))
 731                 vlan_tag = (vlan_tag & ~VLAN_PRIO_MASK) |
 732                                 adapter->recommended_prio_bits;
 733
 734         return vlan_tag;
 735 }
 736
 737 /* Used only for IP tunnel packets */
 738 static u16 skb_inner_ip_proto(struct sk_buff *skb)
 739 {
 740         return (inner_ip_hdr(skb)->version == 4) ?
 741                 inner_ip_hdr(skb)->protocol : inner_ipv6_hdr(skb)->nexthdr;
 742 }
 743
 744 static u16 skb_ip_proto(struct sk_buff *skb)
 745 {
 746         return (ip_hdr(skb)->version == 4) ?
 747                 ip_hdr(skb)->protocol : ipv6_hdr(skb)->nexthdr;
 748 }
 749
 750 static inline bool be_is_txq_full(struct be_tx_obj *txo)
 751 {
 752         return atomic_read(&txo->q.used) + BE_MAX_TX_FRAG_COUNT >= txo->q.len;
 753 }
 754
 755 static inline bool be_can_txq_wake(struct be_tx_obj *txo)
 756 {
 757         return atomic_read(&txo->q.used) < txo->q.len / 2;
 758 }
 759
 760 static inline bool be_is_tx_compl_pending(struct be_tx_obj *txo)
 761 {
 762         return atomic_read(&txo->q.used) > txo->pend_wrb_cnt;
 763 }
 764
 765 static void be_get_wrb_params_from_skb(struct be_adapter *adapter,
 766                                        struct sk_buff *skb,
 767                                        struct be_wrb_params *wrb_params)
 768 {
 769         u16 proto;
 770
 771         if (skb_is_gso(skb)) {
 772                 BE_WRB_F_SET(wrb_params->features, LSO, 1);
 773                 wrb_params->lso_mss = skb_shinfo(skb)->gso_size;
 774                 if (skb_is_gso_v6(skb) && !lancer_chip(adapter))
 775                         BE_WRB_F_SET(wrb_params->features, LSO6, 1);
 776         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 777                 if (skb->encapsulation) {
 778                         BE_WRB_F_SET(wrb_params->features, IPCS, 1);
 779                         proto = skb_inner_ip_proto(skb);
 780                 } else {
 781                         proto = skb_ip_proto(skb);
 782                 }
 783                 if (proto == IPPROTO_TCP)
 784                         BE_WRB_F_SET(wrb_params->features, TCPCS, 1);
 785                 else if (proto == IPPROTO_UDP)
 786                         BE_WRB_F_SET(wrb_params->features, UDPCS, 1);
 787         }
 788
 789         if (skb_vlan_tag_present(skb)) {
 790                 BE_WRB_F_SET(wrb_params->features, VLAN, 1);
 791                 wrb_params->vlan_tag = be_get_tx_vlan_tag(adapter, skb);
 792         }
 793
 794         BE_WRB_F_SET(wrb_params->features, CRC, 1);
 795 }
 796
 797 static void wrb_fill_hdr(struct be_adapter *adapter,
 798                          struct be_eth_hdr_wrb *hdr,
 799                          struct be_wrb_params *wrb_params,
 800                          struct sk_buff *skb)
 801 {
 802         memset(hdr, 0, sizeof(*hdr));
 803
 804         SET_TX_WRB_HDR_BITS(crc, hdr,
 805                             BE_WRB_F_GET(wrb_params->features, CRC));
 806         SET_TX_WRB_HDR_BITS(ipcs, hdr,
 807                             BE_WRB_F_GET(wrb_params->features, IPCS));
 808         SET_TX_WRB_HDR_BITS(tcpcs, hdr,
 809                             BE_WRB_F_GET(wrb_params->features, TCPCS));
 810         SET_TX_WRB_HDR_BITS(udpcs, hdr,
 811                             BE_WRB_F_GET(wrb_params->features, UDPCS));
 812
 813         SET_TX_WRB_HDR_BITS(lso, hdr,
 814                             BE_WRB_F_GET(wrb_params->features, LSO));
 815         SET_TX_WRB_HDR_BITS(lso6, hdr,
 816                             BE_WRB_F_GET(wrb_params->features, LSO6));
 817         SET_TX_WRB_HDR_BITS(lso_mss, hdr, wrb_params->lso_mss);
 818
 819         /* Hack to skip HW VLAN tagging needs evt = 1, compl = 0. When this
 820          * hack is not needed, the evt bit is set while ringing DB.
 821          */
 822         SET_TX_WRB_HDR_BITS(event, hdr,
 823                             BE_WRB_F_GET(wrb_params->features, VLAN_SKIP_HW));
 824         SET_TX_WRB_HDR_BITS(vlan, hdr,
 825                             BE_WRB_F_GET(wrb_params->features, VLAN));
 826         SET_TX_WRB_HDR_BITS(vlan_tag, hdr, wrb_params->vlan_tag);
 827
 828         SET_TX_WRB_HDR_BITS(num_wrb, hdr, skb_wrb_cnt(skb));
 829         SET_TX_WRB_HDR_BITS(len, hdr, skb->len);
 830         SET_TX_WRB_HDR_BITS(mgmt, hdr,
 831                             BE_WRB_F_GET(wrb_params->features, OS2BMC));
 832 }
 833
 834 static void unmap_tx_frag(struct device *dev, struct be_eth_wrb *wrb,
 835                           bool unmap_single)
 836 {
 837         dma_addr_t dma;
 838         u32 frag_len = le32_to_cpu(wrb->frag_len);
 839
 840
 841         dma = (u64)le32_to_cpu(wrb->frag_pa_hi) << 32 |
 842                 (u64)le32_to_cpu(wrb->frag_pa_lo);
 843         if (frag_len) {
 844                 if (unmap_single)
 845                         dma_unmap_single(dev, dma, frag_len, DMA_TO_DEVICE);
 846                 else
 847                         dma_unmap_page(dev, dma, frag_len, DMA_TO_DEVICE);
 848         }
 849 }
 850
 851 /* Grab a WRB header for xmit */
 852 static u16 be_tx_get_wrb_hdr(struct be_tx_obj *txo)
 853 {
 854         u16 head = txo->q.head;
 855
 856         queue_head_inc(&txo->q);
 857         return head;
 858 }
 859
 860 /* Set up the WRB header for xmit */
 861 static void be_tx_setup_wrb_hdr(struct be_adapter *adapter,
 862                                 struct be_tx_obj *txo,
 863                                 struct be_wrb_params *wrb_params,
 864                                 struct sk_buff *skb, u16 head)
 865 {
 866         u32 num_frags = skb_wrb_cnt(skb);
 867         struct be_queue_info *txq = &txo->q;
 868         struct be_eth_hdr_wrb *hdr = queue_index_node(txq, head);
 869
 870         wrb_fill_hdr(adapter, hdr, wrb_params, skb);
 871         be_dws_cpu_to_le(hdr, sizeof(*hdr));
 872
 873         BUG_ON(txo->sent_skb_list[head]);
 874         txo->sent_skb_list[head] = skb;
 875         txo->last_req_hdr = head;
 876         atomic_add(num_frags, &txq->used);
 877         txo->last_req_wrb_cnt = num_frags;
 878         txo->pend_wrb_cnt += num_frags;
 879 }
 880
 881 /* Setup a WRB fragment (buffer descriptor) for xmit */
 882 static void be_tx_setup_wrb_frag(struct be_tx_obj *txo, dma_addr_t busaddr,
 883                                  int len)
 884 {
 885         struct be_eth_wrb *wrb;
 886         struct be_queue_info *txq = &txo->q;
 887
 888         wrb = queue_head_node(txq);
 889         wrb_fill(wrb, busaddr, len);
 890         queue_head_inc(txq);
 891 }
 892
 893 /* Bring the queue back to the state it was in before be_xmit_enqueue() routine
 894  * was invoked. The producer index is restored to the previous packet and the
 895  * WRBs of the current packet are unmapped. Invoked to handle tx setup errors.
 896  */
 897 static void be_xmit_restore(struct be_adapter *adapter,
 898                             struct be_tx_obj *txo, u16 head, bool map_single,
 899                             u32 copied)
 900 {
 901         struct device *dev;
 902         struct be_eth_wrb *wrb;
 903         struct be_queue_info *txq = &txo->q;
 904
 905         dev = &adapter->pdev->dev;
 906         txq->head = head;
 907
 908         /* skip the first wrb (hdr); it's not mapped */
 909         queue_head_inc(txq);
 910         while (copied) {
 911                 wrb = queue_head_node(txq);
 912                 unmap_tx_frag(dev, wrb, map_single);
 913                 map_single = false;
 914                 copied -= le32_to_cpu(wrb->frag_len);
 915                 queue_head_inc(txq);
 916         }
 917
 918         txq->head = head;
 919 }
 920
 921 /* Enqueue the given packet for transmit. This routine allocates WRBs for the
 922  * packet, dma maps the packet buffers and sets up the WRBs. Returns the number
 923  * of WRBs used up by the packet.
 924  */
 925 static u32 be_xmit_enqueue(struct be_adapter *adapter, struct be_tx_obj *txo,
 926                            struct sk_buff *skb,
 927                            struct be_wrb_params *wrb_params)
 928 {
 929         u32 i, copied = 0, wrb_cnt = skb_wrb_cnt(skb);
 930         struct device *dev = &adapter->pdev->dev;
 931         struct be_queue_info *txq = &txo->q;
 932         bool map_single = false;
 933         u16 head = txq->head;
 934         dma_addr_t busaddr;
 935         int len;
 936
 937         head = be_tx_get_wrb_hdr(txo);
 938
 939         if (skb->len > skb->data_len) {
 940                 len = skb_headlen(skb);
 941
 942                 busaddr = dma_map_single(dev, skb->data, len, DMA_TO_DEVICE);
 943                 if (dma_mapping_error(dev, busaddr))
 944                         goto dma_err;
 945                 map_single = true;
 946                 be_tx_setup_wrb_frag(txo, busaddr, len);
 947                 copied += len;
 948         }
 949
 950         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 951                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 952                 len = skb_frag_size(frag);
 953
 954                 busaddr = skb_frag_dma_map(dev, frag, 0, len, DMA_TO_DEVICE);
 955                 if (dma_mapping_error(dev, busaddr))
 956                         goto dma_err;
 957                 be_tx_setup_wrb_frag(txo, busaddr, len);
 958                 copied += len;
 959         }
 960
 961         be_tx_setup_wrb_hdr(adapter, txo, wrb_params, skb, head);
 962
 963         be_tx_stats_update(txo, skb);
 964         return wrb_cnt;
 965
 966 dma_err:
 967         adapter->drv_stats.dma_map_errors++;
 968         be_xmit_restore(adapter, txo, head, map_single, copied);
 969         return 0;
 970 }
 971
 972 static inline int qnq_async_evt_rcvd(struct be_adapter *adapter)
 973 {
 974         return adapter->flags & BE_FLAGS_QNQ_ASYNC_EVT_RCVD;
 975 }
 976
 977 static struct sk_buff *be_insert_vlan_in_pkt(struct be_adapter *adapter,
 978                                              struct sk_buff *skb,
 979                                              struct be_wrb_params
 980                                              *wrb_params)
 981 {
 982         u16 vlan_tag = 0;
 983
 984         skb = skb_share_check(skb, GFP_ATOMIC);
 985         if (unlikely(!skb))
 986                 return skb;
 987
 988         if (skb_vlan_tag_present(skb))
 989                 vlan_tag = be_get_tx_vlan_tag(adapter, skb);
 990
 991         if (qnq_async_evt_rcvd(adapter) && adapter->pvid) {
 992                 if (!vlan_tag)
 993                         vlan_tag = adapter->pvid;
 994                 /* f/w workaround to set skip_hw_vlan = 1, informs the F/W to
 995                  * skip VLAN insertion
 996                  */
 997                 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
 998         }
 999
1000         if (vlan_tag) {
1001                 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1002                                                 vlan_tag);
1003                 if (unlikely(!skb))
1004                         return skb;
1005                 skb->vlan_tci = 0;
1006         }
1007
1008         /* Insert the outer VLAN, if any */
1009         if (adapter->qnq_vid) {
1010                 vlan_tag = adapter->qnq_vid;
1011                 skb = vlan_insert_tag_set_proto(skb, htons(ETH_P_8021Q),
1012                                                 vlan_tag);
1013                 if (unlikely(!skb))
1014                         return skb;
1015                 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1016         }
1017
1018         return skb;
1019 }
1020
1021 static bool be_ipv6_exthdr_check(struct sk_buff *skb)
1022 {
1023         struct ethhdr *eh = (struct ethhdr *)skb->data;
1024         u16 offset = ETH_HLEN;
1025
1026         if (eh->h_proto == htons(ETH_P_IPV6)) {
1027                 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + offset);
1028
1029                 offset += sizeof(struct ipv6hdr);
1030                 if (ip6h->nexthdr != NEXTHDR_TCP &&
1031                     ip6h->nexthdr != NEXTHDR_UDP) {
1032                         struct ipv6_opt_hdr *ehdr =
1033                                 (struct ipv6_opt_hdr *)(skb->data + offset);
1034
1035                         /* offending pkt: 2nd byte following IPv6 hdr is 0xff */
1036                         if (ehdr->hdrlen == 0xff)
1037                                 return true;
1038                 }
1039         }
1040         return false;
1041 }
1042
1043 static int be_vlan_tag_tx_chk(struct be_adapter *adapter, struct sk_buff *skb)
1044 {
1045         return skb_vlan_tag_present(skb) || adapter->pvid || adapter->qnq_vid;
1046 }
1047
1048 static int be_ipv6_tx_stall_chk(struct be_adapter *adapter, struct sk_buff *skb)
1049 {
1050         return BE3_chip(adapter) && be_ipv6_exthdr_check(skb);
1051 }
1052
1053 static struct sk_buff *be_lancer_xmit_workarounds(struct be_adapter *adapter,
1054                                                   struct sk_buff *skb,
1055                                                   struct be_wrb_params
1056                                                   *wrb_params)
1057 {
1058         struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1059         unsigned int eth_hdr_len;
1060         struct iphdr *ip;
1061
1062         /* For padded packets, BE HW modifies tot_len field in IP header
1063          * incorrecly when VLAN tag is inserted by HW.
1064          * For padded packets, Lancer computes incorrect checksum.
1065          */
1066         eth_hdr_len = ntohs(skb->protocol) == ETH_P_8021Q ?
1067                                                 VLAN_ETH_HLEN : ETH_HLEN;
1068         if (skb->len <= 60 &&
1069             (lancer_chip(adapter) || skb_vlan_tag_present(skb)) &&
1070             is_ipv4_pkt(skb)) {
1071                 ip = (struct iphdr *)ip_hdr(skb);
1072                 pskb_trim(skb, eth_hdr_len + ntohs(ip->tot_len));
1073         }
1074
1075         /* If vlan tag is already inlined in the packet, skip HW VLAN
1076          * tagging in pvid-tagging mode
1077          */
1078         if (be_pvid_tagging_enabled(adapter) &&
1079             veh->h_vlan_proto == htons(ETH_P_8021Q))
1080                 BE_WRB_F_SET(wrb_params->features, VLAN_SKIP_HW, 1);
1081
1082         /* HW has a bug wherein it will calculate CSUM for VLAN
1083          * pkts even though it is disabled.
1084          * Manually insert VLAN in pkt.
1085          */
1086         if (skb->ip_summed != CHECKSUM_PARTIAL &&
1087             skb_vlan_tag_present(skb)) {
1088                 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1089                 if (unlikely(!skb))
1090                         goto err;
1091         }
1092
1093         /* HW may lockup when VLAN HW tagging is requested on
1094          * certain ipv6 packets. Drop such pkts if the HW workaround to
1095          * skip HW tagging is not enabled by FW.
1096          */
1097         if (unlikely(be_ipv6_tx_stall_chk(adapter, skb) &&
1098                      (adapter->pvid || adapter->qnq_vid) &&
1099                      !qnq_async_evt_rcvd(adapter)))
1100                 goto tx_drop;
1101
1102         /* Manual VLAN tag insertion to prevent:
1103          * ASIC lockup when the ASIC inserts VLAN tag into
1104          * certain ipv6 packets. Insert VLAN tags in driver,
1105          * and set event, completion, vlan bits accordingly
1106          * in the Tx WRB.
1107          */
1108         if (be_ipv6_tx_stall_chk(adapter, skb) &&
1109             be_vlan_tag_tx_chk(adapter, skb)) {
1110                 skb = be_insert_vlan_in_pkt(adapter, skb, wrb_params);
1111                 if (unlikely(!skb))
1112                         goto err;
1113         }
1114
1115         return skb;
1116 tx_drop:
1117         dev_kfree_skb_any(skb);
1118 err:
1119         return NULL;
1120 }
1121
1122 static struct sk_buff *be_xmit_workarounds(struct be_adapter *adapter,
1123                                            struct sk_buff *skb,
1124                                            struct be_wrb_params *wrb_params)
1125 {
1126         /* Lancer, SH and BE3 in SRIOV mode have a bug wherein
1127          * packets that are 32b or less may cause a transmit stall
1128          * on that port. The workaround is to pad such packets
1129          * (len <= 32 bytes) to a minimum length of 36b.
1130          */
1131         if (skb->len <= 32) {
1132                 if (skb_put_padto(skb, 36))
1133                         return NULL;
1134         }
1135
1136         if (BEx_chip(adapter) || lancer_chip(adapter)) {
1137                 skb = be_lancer_xmit_workarounds(adapter, skb, wrb_params);
1138                 if (!skb)
1139                         return NULL;
1140         }
1141
1142         return skb;
1143 }
1144
1145 static void be_xmit_flush(struct be_adapter *adapter, struct be_tx_obj *txo)
1146 {
1147         struct be_queue_info *txq = &txo->q;
1148         struct be_eth_hdr_wrb *hdr = queue_index_node(txq, txo->last_req_hdr);
1149
1150         /* Mark the last request eventable if it hasn't been marked already */
1151         if (!(hdr->dw[2] & cpu_to_le32(TX_HDR_WRB_EVT)))
1152                 hdr->dw[2] |= cpu_to_le32(TX_HDR_WRB_EVT | TX_HDR_WRB_COMPL);
1153
1154         /* compose a dummy wrb if there are odd set of wrbs to notify */
1155         if (!lancer_chip(adapter) && (txo->pend_wrb_cnt & 1)) {
1156                 wrb_fill_dummy(queue_head_node(txq));
1157                 queue_head_inc(txq);
1158                 atomic_inc(&txq->used);
1159                 txo->pend_wrb_cnt++;
1160                 hdr->dw[2] &= ~cpu_to_le32(TX_HDR_WRB_NUM_MASK <<
1161                                            TX_HDR_WRB_NUM_SHIFT);
1162                 hdr->dw[2] |= cpu_to_le32((txo->last_req_wrb_cnt + 1) <<
1163                                           TX_HDR_WRB_NUM_SHIFT);
1164         }
1165         be_txq_notify(adapter, txo, txo->pend_wrb_cnt);
1166         txo->pend_wrb_cnt = 0;
1167 }
1168
1169 /* OS2BMC related */
1170
1171 #define DHCP_CLIENT_PORT        68
1172 #define DHCP_SERVER_PORT        67
1173 #define NET_BIOS_PORT1          137
1174 #define NET_BIOS_PORT2          138
1175 #define DHCPV6_RAS_PORT         547
1176
1177 #define is_mc_allowed_on_bmc(adapter, eh)       \
1178         (!is_multicast_filt_enabled(adapter) && \
1179          is_multicast_ether_addr(eh->h_dest) && \
1180          !is_broadcast_ether_addr(eh->h_dest))
1181
1182 #define is_bc_allowed_on_bmc(adapter, eh)       \
1183         (!is_broadcast_filt_enabled(adapter) && \
1184          is_broadcast_ether_addr(eh->h_dest))
1185
1186 #define is_arp_allowed_on_bmc(adapter, skb)     \
1187         (is_arp(skb) && is_arp_filt_enabled(adapter))
1188
1189 #define is_broadcast_packet(eh, adapter)        \
1190                 (is_multicast_ether_addr(eh->h_dest) && \
1191                 !compare_ether_addr(eh->h_dest, adapter->netdev->broadcast))
1192
1193 #define is_arp(skb)     (skb->protocol == htons(ETH_P_ARP))
1194
1195 #define is_arp_filt_enabled(adapter)    \
1196                 (adapter->bmc_filt_mask & (BMC_FILT_BROADCAST_ARP))
1197
1198 #define is_dhcp_client_filt_enabled(adapter)    \
1199                 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_CLIENT)
1200
1201 #define is_dhcp_srvr_filt_enabled(adapter)      \
1202                 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_DHCP_SERVER)
1203
1204 #define is_nbios_filt_enabled(adapter)  \
1205                 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST_NET_BIOS)
1206
1207 #define is_ipv6_na_filt_enabled(adapter)        \
1208                 (adapter->bmc_filt_mask &       \
1209                         BMC_FILT_MULTICAST_IPV6_NEIGH_ADVER)
1210
1211 #define is_ipv6_ra_filt_enabled(adapter)        \
1212                 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RA)
1213
1214 #define is_ipv6_ras_filt_enabled(adapter)       \
1215                 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST_IPV6_RAS)
1216
1217 #define is_broadcast_filt_enabled(adapter)      \
1218                 (adapter->bmc_filt_mask & BMC_FILT_BROADCAST)
1219
1220 #define is_multicast_filt_enabled(adapter)      \
1221                 (adapter->bmc_filt_mask & BMC_FILT_MULTICAST)
1222
1223 static bool be_send_pkt_to_bmc(struct be_adapter *adapter,
1224                                struct sk_buff **skb)
1225 {
1226         struct ethhdr *eh = (struct ethhdr *)(*skb)->data;
1227         bool os2bmc = false;
1228
1229         if (!be_is_os2bmc_enabled(adapter))
1230                 goto done;
1231
1232         if (!is_multicast_ether_addr(eh->h_dest))
1233                 goto done;
1234
1235         if (is_mc_allowed_on_bmc(adapter, eh) ||
1236             is_bc_allowed_on_bmc(adapter, eh) ||
1237             is_arp_allowed_on_bmc(adapter, (*skb))) {
1238                 os2bmc = true;
1239                 goto done;
1240         }
1241
1242         if ((*skb)->protocol == htons(ETH_P_IPV6)) {
1243                 struct ipv6hdr *hdr = ipv6_hdr((*skb));
1244                 u8 nexthdr = hdr->nexthdr;
1245
1246                 if (nexthdr == IPPROTO_ICMPV6) {
1247                         struct icmp6hdr *icmp6 = icmp6_hdr((*skb));
1248
1249                         switch (icmp6->icmp6_type) {
1250                         case NDISC_ROUTER_ADVERTISEMENT:
1251                                 os2bmc = is_ipv6_ra_filt_enabled(adapter);
1252                                 goto done;
1253                         case NDISC_NEIGHBOUR_ADVERTISEMENT:
1254                                 os2bmc = is_ipv6_na_filt_enabled(adapter);
1255                                 goto done;
1256                         default:
1257                                 break;
1258                         }
1259                 }
1260         }
1261
1262         if (is_udp_pkt((*skb))) {
1263                 struct udphdr *udp = udp_hdr((*skb));
1264
1265                 switch (ntohs(udp->dest)) {
1266                 case DHCP_CLIENT_PORT:
1267                         os2bmc = is_dhcp_client_filt_enabled(adapter);
1268                         goto done;
1269                 case DHCP_SERVER_PORT:
1270                         os2bmc = is_dhcp_srvr_filt_enabled(adapter);
1271                         goto done;
1272                 case NET_BIOS_PORT1:
1273                 case NET_BIOS_PORT2:
1274                         os2bmc = is_nbios_filt_enabled(adapter);
1275                         goto done;
1276                 case DHCPV6_RAS_PORT:
1277                         os2bmc = is_ipv6_ras_filt_enabled(adapter);
1278                         goto done;
1279                 default:
1280                         break;
1281                 }
1282         }
1283 done:
1284         /* For packets over a vlan, which are destined
1285          * to BMC, asic expects the vlan to be inline in the packet.
1286          */
1287         if (os2bmc)
1288                 *skb = be_insert_vlan_in_pkt(adapter, *skb, NULL);
1289
1290         return os2bmc;
1291 }
1292
1293 static netdev_tx_t be_xmit(struct sk_buff *skb, struct net_device *netdev)
1294 {
1295         struct be_adapter *adapter = netdev_priv(netdev);
1296         u16 q_idx = skb_get_queue_mapping(skb);
1297         struct be_tx_obj *txo = &adapter->tx_obj[q_idx];
1298         struct be_wrb_params wrb_params = { 0 };
1299         bool flush = !skb->xmit_more;
1300         u16 wrb_cnt;
1301
1302         skb = be_xmit_workarounds(adapter, skb, &wrb_params);
1303         if (unlikely(!skb))
1304                 goto drop;
1305
1306         be_get_wrb_params_from_skb(adapter, skb, &wrb_params);
1307
1308         wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1309         if (unlikely(!wrb_cnt)) {
1310                 dev_kfree_skb_any(skb);
1311                 goto drop;
1312         }
1313
1314         /* if os2bmc is enabled and if the pkt is destined to bmc,
1315          * enqueue the pkt a 2nd time with mgmt bit set.
1316          */
1317         if (be_send_pkt_to_bmc(adapter, &skb)) {
1318                 BE_WRB_F_SET(wrb_params.features, OS2BMC, 1);
1319                 wrb_cnt = be_xmit_enqueue(adapter, txo, skb, &wrb_params);
1320                 if (unlikely(!wrb_cnt))
1321                         goto drop;
1322                 else
1323                         skb_get(skb);
1324         }
1325
1326         if (be_is_txq_full(txo)) {
1327                 netif_stop_subqueue(netdev, q_idx);
1328                 tx_stats(txo)->tx_stops++;
1329         }
1330
1331         if (flush || __netif_subqueue_stopped(netdev, q_idx))
1332                 be_xmit_flush(adapter, txo);
1333
1334         return NETDEV_TX_OK;
1335 drop:
1336         tx_stats(txo)->tx_drv_drops++;
1337         /* Flush the already enqueued tx requests */
1338         if (flush && txo->pend_wrb_cnt)
1339                 be_xmit_flush(adapter, txo);
1340
1341         return NETDEV_TX_OK;
1342 }
1343
1344 static int be_change_mtu(struct net_device *netdev, int new_mtu)
1345 {
1346         struct be_adapter *adapter = netdev_priv(netdev);
1347         struct device *dev = &adapter->pdev->dev;
1348
1349         if (new_mtu < BE_MIN_MTU || new_mtu > BE_MAX_MTU) {
1350                 dev_info(dev, "MTU must be between %d and %d bytes\n",
1351                          BE_MIN_MTU, BE_MAX_MTU);
1352                 return -EINVAL;
1353         }
1354
1355         dev_info(dev, "MTU changed from %d to %d bytes\n",
1356                  netdev->mtu, new_mtu);
1357         netdev->mtu = new_mtu;
1358         return 0;
1359 }
1360
1361 static inline bool be_in_all_promisc(struct be_adapter *adapter)
1362 {
1363         return (adapter->if_flags & BE_IF_FLAGS_ALL_PROMISCUOUS) ==
1364                         BE_IF_FLAGS_ALL_PROMISCUOUS;
1365 }
1366
1367 static int be_set_vlan_promisc(struct be_adapter *adapter)
1368 {
1369         struct device *dev = &adapter->pdev->dev;
1370         int status;
1371
1372         if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS)
1373                 return 0;
1374
1375         status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, ON);
1376         if (!status) {
1377                 dev_info(dev, "Enabled VLAN promiscuous mode\n");
1378                 adapter->if_flags |= BE_IF_FLAGS_VLAN_PROMISCUOUS;
1379         } else {
1380                 dev_err(dev, "Failed to enable VLAN promiscuous mode\n");
1381         }
1382         return status;
1383 }
1384
1385 static int be_clear_vlan_promisc(struct be_adapter *adapter)
1386 {
1387         struct device *dev = &adapter->pdev->dev;
1388         int status;
1389
1390         status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_VLAN_PROMISCUOUS, OFF);
1391         if (!status) {
1392                 dev_info(dev, "Disabling VLAN promiscuous mode\n");
1393                 adapter->if_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
1394         }
1395         return status;
1396 }
1397
1398 /*
1399  * A max of 64 (BE_NUM_VLANS_SUPPORTED) vlans can be configured in BE.
1400  * If the user configures more, place BE in vlan promiscuous mode.
1401  */
1402 static int be_vid_config(struct be_adapter *adapter)
1403 {
1404         struct device *dev = &adapter->pdev->dev;
1405         u16 vids[BE_NUM_VLANS_SUPPORTED];
1406         u16 num = 0, i = 0;
1407         int status = 0;
1408
1409         /* No need to further configure vids if in promiscuous mode */
1410         if (be_in_all_promisc(adapter))
1411                 return 0;
1412
1413         if (adapter->vlans_added > be_max_vlans(adapter))
1414                 return be_set_vlan_promisc(adapter);
1415
1416         /* Construct VLAN Table to give to HW */
1417         for_each_set_bit(i, adapter->vids, VLAN_N_VID)
1418                 vids[num++] = cpu_to_le16(i);
1419
1420         status = be_cmd_vlan_config(adapter, adapter->if_handle, vids, num, 0);
1421         if (status) {
1422                 dev_err(dev, "Setting HW VLAN filtering failed\n");
1423                 /* Set to VLAN promisc mode as setting VLAN filter failed */
1424                 if (addl_status(status) == MCC_ADDL_STATUS_INSUFFICIENT_VLANS ||
1425                     addl_status(status) ==
1426                                 MCC_ADDL_STATUS_INSUFFICIENT_RESOURCES)
1427                         return be_set_vlan_promisc(adapter);
1428         } else if (adapter->if_flags & BE_IF_FLAGS_VLAN_PROMISCUOUS) {
1429                 status = be_clear_vlan_promisc(adapter);
1430         }
1431         return status;
1432 }
1433
1434 static int be_vlan_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1435 {
1436         struct be_adapter *adapter = netdev_priv(netdev);
1437         int status = 0;
1438
1439         /* Packets with VID 0 are always received by Lancer by default */
1440         if (lancer_chip(adapter) && vid == 0)
1441                 return status;
1442
1443         if (test_bit(vid, adapter->vids))
1444                 return status;
1445
1446         set_bit(vid, adapter->vids);
1447         adapter->vlans_added++;
1448
1449         status = be_vid_config(adapter);
1450         if (status) {
1451                 adapter->vlans_added--;
1452                 clear_bit(vid, adapter->vids);
1453         }
1454
1455         return status;
1456 }
1457
1458 static int be_vlan_rem_vid(struct net_device *netdev, __be16 proto, u16 vid)
1459 {
1460         struct be_adapter *adapter = netdev_priv(netdev);
1461
1462         /* Packets with VID 0 are always received by Lancer by default */
1463         if (lancer_chip(adapter) && vid == 0)
1464                 return 0;
1465
1466         if (!test_bit(vid, adapter->vids))
1467                 return 0;
1468
1469         clear_bit(vid, adapter->vids);
1470         adapter->vlans_added--;
1471
1472         return be_vid_config(adapter);
1473 }
1474
1475 static void be_clear_all_promisc(struct be_adapter *adapter)
1476 {
1477         be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, OFF);
1478         adapter->if_flags &= ~BE_IF_FLAGS_ALL_PROMISCUOUS;
1479 }
1480
1481 static void be_set_all_promisc(struct be_adapter *adapter)
1482 {
1483         be_cmd_rx_filter(adapter, BE_IF_FLAGS_ALL_PROMISCUOUS, ON);
1484         adapter->if_flags |= BE_IF_FLAGS_ALL_PROMISCUOUS;
1485 }
1486
1487 static void be_set_mc_promisc(struct be_adapter *adapter)
1488 {
1489         int status;
1490
1491         if (adapter->if_flags & BE_IF_FLAGS_MCAST_PROMISCUOUS)
1492                 return;
1493
1494         status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MCAST_PROMISCUOUS, ON);
1495         if (!status)
1496                 adapter->if_flags |= BE_IF_FLAGS_MCAST_PROMISCUOUS;
1497 }
1498
1499 static void be_set_mc_list(struct be_adapter *adapter)
1500 {
1501         int status;
1502
1503         status = be_cmd_rx_filter(adapter, BE_IF_FLAGS_MULTICAST, ON);
1504         if (!status)
1505                 adapter->if_flags &= ~BE_IF_FLAGS_MCAST_PROMISCUOUS;
1506         else
1507                 be_set_mc_promisc(adapter);
1508 }
1509
1510 static void be_set_uc_list(struct be_adapter *adapter)
1511 {
1512         struct netdev_hw_addr *ha;
1513         int i = 1; /* First slot is claimed by the Primary MAC */
1514
1515         for (; adapter->uc_macs > 0; adapter->uc_macs--, i++)
1516                 be_cmd_pmac_del(adapter, adapter->if_handle,
1517                                 adapter->pmac_id[i], 0);
1518
1519         if (netdev_uc_count(adapter->netdev) > be_max_uc(adapter)) {
1520                 be_set_all_promisc(adapter);
1521                 return;
1522         }
1523
1524         netdev_for_each_uc_addr(ha, adapter->netdev) {
1525                 adapter->uc_macs++; /* First slot is for Primary MAC */
1526                 be_cmd_pmac_add(adapter, (u8 *)ha->addr, adapter->if_handle,
1527                                 &adapter->pmac_id[adapter->uc_macs], 0);
1528         }
1529 }
1530
1531 static void be_clear_uc_list(struct be_adapter *adapter)
1532 {
1533         int i;
1534
1535         for (i = 1; i < (adapter->uc_macs + 1); i++)
1536                 be_cmd_pmac_del(adapter, adapter->if_handle,
1537                                 adapter->pmac_id[i], 0);
1538         adapter->uc_macs = 0;
1539 }
1540
1541 static void be_set_rx_mode(struct net_device *netdev)
1542 {
1543         struct be_adapter *adapter = netdev_priv(netdev);
1544
1545         if (netdev->flags & IFF_PROMISC) {
1546                 be_set_all_promisc(adapter);
1547                 return;
1548         }
1549
1550         /* Interface was previously in promiscuous mode; disable it */
1551         if (be_in_all_promisc(adapter)) {
1552                 be_clear_all_promisc(adapter);
1553                 if (adapter->vlans_added)
1554                         be_vid_config(adapter);
1555         }
1556
1557         /* Enable multicast promisc if num configured exceeds what we support */
1558         if (netdev->flags & IFF_ALLMULTI ||
1559             netdev_mc_count(netdev) > be_max_mc(adapter)) {
1560                 be_set_mc_promisc(adapter);
1561                 return;
1562         }
1563
1564         if (netdev_uc_count(netdev) != adapter->uc_macs)
1565                 be_set_uc_list(adapter);
1566
1567         be_set_mc_list(adapter);
1568 }
1569
1570 static int be_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1571 {
1572         struct be_adapter *adapter = netdev_priv(netdev);
1573         struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1574         int status;
1575
1576         if (!sriov_enabled(adapter))
1577                 return -EPERM;
1578
1579         if (!is_valid_ether_addr(mac) || vf >= adapter->num_vfs)
1580                 return -EINVAL;
1581
1582         /* Proceed further only if user provided MAC is different
1583          * from active MAC
1584          */
1585         if (ether_addr_equal(mac, vf_cfg->mac_addr))
1586                 return 0;
1587
1588         if (BEx_chip(adapter)) {
1589                 be_cmd_pmac_del(adapter, vf_cfg->if_handle, vf_cfg->pmac_id,
1590                                 vf + 1);
1591
1592                 status = be_cmd_pmac_add(adapter, mac, vf_cfg->if_handle,
1593                                          &vf_cfg->pmac_id, vf + 1);
1594         } else {
1595                 status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
1596                                         vf + 1);
1597         }
1598
1599         if (status) {
1600                 dev_err(&adapter->pdev->dev, "MAC %pM set on VF %d Failed: %#x",
1601                         mac, vf, status);
1602                 return be_cmd_status(status);
1603         }
1604
1605         ether_addr_copy(vf_cfg->mac_addr, mac);
1606
1607         return 0;
1608 }
1609
1610 static int be_get_vf_config(struct net_device *netdev, int vf,
1611                             struct ifla_vf_info *vi)
1612 {
1613         struct be_adapter *adapter = netdev_priv(netdev);
1614         struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1615
1616         if (!sriov_enabled(adapter))
1617                 return -EPERM;
1618
1619         if (vf >= adapter->num_vfs)
1620                 return -EINVAL;
1621
1622         vi->vf = vf;
1623         vi->max_tx_rate = vf_cfg->tx_rate;
1624         vi->min_tx_rate = 0;
1625         vi->vlan = vf_cfg->vlan_tag & VLAN_VID_MASK;
1626         vi->qos = vf_cfg->vlan_tag >> VLAN_PRIO_SHIFT;
1627         memcpy(&vi->mac, vf_cfg->mac_addr, ETH_ALEN);
1628         vi->linkstate = adapter->vf_cfg[vf].plink_tracking;
1629         vi->spoofchk = adapter->vf_cfg[vf].spoofchk;
1630
1631         return 0;
1632 }
1633
1634 static int be_set_vf_tvt(struct be_adapter *adapter, int vf, u16 vlan)
1635 {
1636         struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1637         u16 vids[BE_NUM_VLANS_SUPPORTED];
1638         int vf_if_id = vf_cfg->if_handle;
1639         int status;
1640
1641         /* Enable Transparent VLAN Tagging */
1642         status = be_cmd_set_hsw_config(adapter, vlan, vf + 1, vf_if_id, 0, 0);
1643         if (status)
1644                 return status;
1645
1646         /* Clear pre-programmed VLAN filters on VF if any, if TVT is enabled */
1647         vids[0] = 0;
1648         status = be_cmd_vlan_config(adapter, vf_if_id, vids, 1, vf + 1);
1649         if (!status)
1650                 dev_info(&adapter->pdev->dev,
1651                          "Cleared guest VLANs on VF%d", vf);
1652
1653         /* After TVT is enabled, disallow VFs to program VLAN filters */
1654         if (vf_cfg->privileges & BE_PRIV_FILTMGMT) {
1655                 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges &
1656                                                   ~BE_PRIV_FILTMGMT, vf + 1);
1657                 if (!status)
1658                         vf_cfg->privileges &= ~BE_PRIV_FILTMGMT;
1659         }
1660         return 0;
1661 }
1662
1663 static int be_clear_vf_tvt(struct be_adapter *adapter, int vf)
1664 {
1665         struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1666         struct device *dev = &adapter->pdev->dev;
1667         int status;
1668
1669         /* Reset Transparent VLAN Tagging. */
1670         status = be_cmd_set_hsw_config(adapter, BE_RESET_VLAN_TAG_ID, vf + 1,
1671                                        vf_cfg->if_handle, 0, 0);
1672         if (status)
1673                 return status;
1674
1675         /* Allow VFs to program VLAN filtering */
1676         if (!(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
1677                 status = be_cmd_set_fn_privileges(adapter, vf_cfg->privileges |
1678                                                   BE_PRIV_FILTMGMT, vf + 1);
1679                 if (!status) {
1680                         vf_cfg->privileges |= BE_PRIV_FILTMGMT;
1681                         dev_info(dev, "VF%d: FILTMGMT priv enabled", vf);
1682                 }
1683         }
1684
1685         dev_info(dev,
1686                  "Disable/re-enable i/f in VM to clear Transparent VLAN tag");
1687         return 0;
1688 }
1689
1690 static int be_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan, u8 qos)
1691 {
1692         struct be_adapter *adapter = netdev_priv(netdev);
1693         struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1694         int status;
1695
1696         if (!sriov_enabled(adapter))
1697                 return -EPERM;
1698
1699         if (vf >= adapter->num_vfs || vlan > 4095 || qos > 7)
1700                 return -EINVAL;
1701
1702         if (vlan || qos) {
1703                 vlan |= qos << VLAN_PRIO_SHIFT;
1704                 status = be_set_vf_tvt(adapter, vf, vlan);
1705         } else {
1706                 status = be_clear_vf_tvt(adapter, vf);
1707         }
1708
1709         if (status) {
1710                 dev_err(&adapter->pdev->dev,
1711                         "VLAN %d config on VF %d failed : %#x\n", vlan, vf,
1712                         status);
1713                 return be_cmd_status(status);
1714         }
1715
1716         vf_cfg->vlan_tag = vlan;
1717         return 0;
1718 }
1719
1720 static int be_set_vf_tx_rate(struct net_device *netdev, int vf,
1721                              int min_tx_rate, int max_tx_rate)
1722 {
1723         struct be_adapter *adapter = netdev_priv(netdev);
1724         struct device *dev = &adapter->pdev->dev;
1725         int percent_rate, status = 0;
1726         u16 link_speed = 0;
1727         u8 link_status;
1728
1729         if (!sriov_enabled(adapter))
1730                 return -EPERM;
1731
1732         if (vf >= adapter->num_vfs)
1733                 return -EINVAL;
1734
1735         if (min_tx_rate)
1736                 return -EINVAL;
1737
1738         if (!max_tx_rate)
1739                 goto config_qos;
1740
1741         status = be_cmd_link_status_query(adapter, &link_speed,
1742                                           &link_status, 0);
1743         if (status)
1744                 goto err;
1745
1746         if (!link_status) {
1747                 dev_err(dev, "TX-rate setting not allowed when link is down\n");
1748                 status = -ENETDOWN;
1749                 goto err;
1750         }
1751
1752         if (max_tx_rate < 100 || max_tx_rate > link_speed) {
1753                 dev_err(dev, "TX-rate must be between 100 and %d Mbps\n",
1754                         link_speed);
1755                 status = -EINVAL;
1756                 goto err;
1757         }
1758
1759         /* On Skyhawk the QOS setting must be done only as a % value */
1760         percent_rate = link_speed / 100;
1761         if (skyhawk_chip(adapter) && (max_tx_rate % percent_rate)) {
1762                 dev_err(dev, "TX-rate must be a multiple of %d Mbps\n",
1763                         percent_rate);
1764                 status = -EINVAL;
1765                 goto err;
1766         }
1767
1768 config_qos:
1769         status = be_cmd_config_qos(adapter, max_tx_rate, link_speed, vf + 1);
1770         if (status)
1771                 goto err;
1772
1773         adapter->vf_cfg[vf].tx_rate = max_tx_rate;
1774         return 0;
1775
1776 err:
1777         dev_err(dev, "TX-rate setting of %dMbps on VF%d failed\n",
1778                 max_tx_rate, vf);
1779         return be_cmd_status(status);
1780 }
1781
1782 static int be_set_vf_link_state(struct net_device *netdev, int vf,
1783                                 int link_state)
1784 {
1785         struct be_adapter *adapter = netdev_priv(netdev);
1786         int status;
1787
1788         if (!sriov_enabled(adapter))
1789                 return -EPERM;
1790
1791         if (vf >= adapter->num_vfs)
1792                 return -EINVAL;
1793
1794         status = be_cmd_set_logical_link_config(adapter, link_state, vf+1);
1795         if (status) {
1796                 dev_err(&adapter->pdev->dev,
1797                         "Link state change on VF %d failed: %#x\n", vf, status);
1798                 return be_cmd_status(status);
1799         }
1800
1801         adapter->vf_cfg[vf].plink_tracking = link_state;
1802
1803         return 0;
1804 }
1805
1806 static int be_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1807 {
1808         struct be_adapter *adapter = netdev_priv(netdev);
1809         struct be_vf_cfg *vf_cfg = &adapter->vf_cfg[vf];
1810         u8 spoofchk;
1811         int status;
1812
1813         if (!sriov_enabled(adapter))
1814                 return -EPERM;
1815
1816         if (vf >= adapter->num_vfs)
1817                 return -EINVAL;
1818
1819         if (BEx_chip(adapter))
1820                 return -EOPNOTSUPP;
1821
1822         if (enable == vf_cfg->spoofchk)
1823                 return 0;
1824
1825         spoofchk = enable ? ENABLE_MAC_SPOOFCHK : DISABLE_MAC_SPOOFCHK;
1826
1827         status = be_cmd_set_hsw_config(adapter, 0, vf + 1, vf_cfg->if_handle,
1828                                        0, spoofchk);
1829         if (status) {
1830                 dev_err(&adapter->pdev->dev,
1831                         "Spoofchk change on VF %d failed: %#x\n", vf, status);
1832                 return be_cmd_status(status);
1833         }
1834
1835         vf_cfg->spoofchk = enable;
1836         return 0;
1837 }
1838
1839 static void be_aic_update(struct be_aic_obj *aic, u64 rx_pkts, u64 tx_pkts,
1840                           ulong now)
1841 {
1842         aic->rx_pkts_prev = rx_pkts;
1843         aic->tx_reqs_prev = tx_pkts;
1844         aic->jiffies = now;
1845 }
1846
1847 static int be_get_new_eqd(struct be_eq_obj *eqo)
1848 {
1849         struct be_adapter *adapter = eqo->adapter;
1850         int eqd, start;
1851         struct be_aic_obj *aic;
1852         struct be_rx_obj *rxo;
1853         struct be_tx_obj *txo;
1854         u64 rx_pkts = 0, tx_pkts = 0;
1855         ulong now;
1856         u32 pps, delta;
1857         int i;
1858
1859         aic = &adapter->aic_obj[eqo->idx];
1860         if (!aic->enable) {
1861                 if (aic->jiffies)
1862                         aic->jiffies = 0;
1863                 eqd = aic->et_eqd;
1864                 return eqd;
1865         }
1866
1867         for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
1868                 do {
1869                         start = u64_stats_fetch_begin_irq(&rxo->stats.sync);
1870                         rx_pkts += rxo->stats.rx_pkts;
1871                 } while (u64_stats_fetch_retry_irq(&rxo->stats.sync, start));
1872         }
1873
1874         for_all_tx_queues_on_eq(adapter, eqo, txo, i) {
1875                 do {
1876                         start = u64_stats_fetch_begin_irq(&txo->stats.sync);
1877                         tx_pkts += txo->stats.tx_reqs;
1878                 } while (u64_stats_fetch_retry_irq(&txo->stats.sync, start));
1879         }
1880
1881         /* Skip, if wrapped around or first calculation */
1882         now = jiffies;
1883         if (!aic->jiffies || time_before(now, aic->jiffies) ||
1884             rx_pkts < aic->rx_pkts_prev ||
1885             tx_pkts < aic->tx_reqs_prev) {
1886                 be_aic_update(aic, rx_pkts, tx_pkts, now);
1887                 return aic->prev_eqd;
1888         }
1889
1890         delta = jiffies_to_msecs(now - aic->jiffies);
1891         if (delta == 0)
1892                 return aic->prev_eqd;
1893
1894         pps = (((u32)(rx_pkts - aic->rx_pkts_prev) * 1000) / delta) +
1895                 (((u32)(tx_pkts - aic->tx_reqs_prev) * 1000) / delta);
1896         eqd = (pps / 15000) << 2;
1897
1898         if (eqd < 8)
1899                 eqd = 0;
1900         eqd = min_t(u32, eqd, aic->max_eqd);
1901         eqd = max_t(u32, eqd, aic->min_eqd);
1902
1903         be_aic_update(aic, rx_pkts, tx_pkts, now);
1904
1905         return eqd;
1906 }
1907
1908 /* For Skyhawk-R only */
1909 static u32 be_get_eq_delay_mult_enc(struct be_eq_obj *eqo)
1910 {
1911         struct be_adapter *adapter = eqo->adapter;
1912         struct be_aic_obj *aic = &adapter->aic_obj[eqo->idx];
1913         ulong now = jiffies;
1914         int eqd;
1915         u32 mult_enc;
1916
1917         if (!aic->enable)
1918                 return 0;
1919
1920         if (jiffies_to_msecs(now - aic->jiffies) < 1)
1921                 eqd = aic->prev_eqd;
1922         else
1923                 eqd = be_get_new_eqd(eqo);
1924
1925         if (eqd > 100)
1926                 mult_enc = R2I_DLY_ENC_1;
1927         else if (eqd > 60)
1928                 mult_enc = R2I_DLY_ENC_2;
1929         else if (eqd > 20)
1930                 mult_enc = R2I_DLY_ENC_3;
1931         else
1932                 mult_enc = R2I_DLY_ENC_0;
1933
1934         aic->prev_eqd = eqd;
1935
1936         return mult_enc;
1937 }
1938
1939 void be_eqd_update(struct be_adapter *adapter, bool force_update)
1940 {
1941         struct be_set_eqd set_eqd[MAX_EVT_QS];
1942         struct be_aic_obj *aic;
1943         struct be_eq_obj *eqo;
1944         int i, num = 0, eqd;
1945
1946         for_all_evt_queues(adapter, eqo, i) {
1947                 aic = &adapter->aic_obj[eqo->idx];
1948                 eqd = be_get_new_eqd(eqo);
1949                 if (force_update || eqd != aic->prev_eqd) {
1950                         set_eqd[num].delay_multiplier = (eqd * 65)/100;
1951                         set_eqd[num].eq_id = eqo->q.id;
1952                         aic->prev_eqd = eqd;
1953                         num++;
1954                 }
1955         }
1956
1957         if (num)
1958                 be_cmd_modify_eqd(adapter, set_eqd, num);
1959 }
1960
1961 static void be_rx_stats_update(struct be_rx_obj *rxo,
1962                                struct be_rx_compl_info *rxcp)
1963 {
1964         struct be_rx_stats *stats = rx_stats(rxo);
1965
1966         u64_stats_update_begin(&stats->sync);
1967         stats->rx_compl++;
1968         stats->rx_bytes += rxcp->pkt_size;
1969         stats->rx_pkts++;
1970         if (rxcp->tunneled)
1971                 stats->rx_vxlan_offload_pkts++;
1972         if (rxcp->pkt_type == BE_MULTICAST_PACKET)
1973                 stats->rx_mcast_pkts++;
1974         if (rxcp->err)
1975                 stats->rx_compl_err++;
1976         u64_stats_update_end(&stats->sync);
1977 }
1978
1979 static inline bool csum_passed(struct be_rx_compl_info *rxcp)
1980 {
1981         /* L4 checksum is not reliable for non TCP/UDP packets.
1982          * Also ignore ipcksm for ipv6 pkts
1983          */
1984         return (rxcp->tcpf || rxcp->udpf) && rxcp->l4_csum &&
1985                 (rxcp->ip_csum || rxcp->ipv6) && !rxcp->err;
1986 }
1987
1988 static struct be_rx_page_info *get_rx_page_info(struct be_rx_obj *rxo)
1989 {
1990         struct be_adapter *adapter = rxo->adapter;
1991         struct be_rx_page_info *rx_page_info;
1992         struct be_queue_info *rxq = &rxo->q;
1993         u16 frag_idx = rxq->tail;
1994
1995         rx_page_info = &rxo->page_info_tbl[frag_idx];
1996         BUG_ON(!rx_page_info->page);
1997
1998         if (rx_page_info->last_frag) {
1999                 dma_unmap_page(&adapter->pdev->dev,
2000                                dma_unmap_addr(rx_page_info, bus),
2001                                adapter->big_page_size, DMA_FROM_DEVICE);
2002                 rx_page_info->last_frag = false;
2003         } else {
2004                 dma_sync_single_for_cpu(&adapter->pdev->dev,
2005                                         dma_unmap_addr(rx_page_info, bus),
2006                                         rx_frag_size, DMA_FROM_DEVICE);
2007         }
2008
2009         queue_tail_inc(rxq);
2010         atomic_dec(&rxq->used);
2011         return rx_page_info;
2012 }
2013
2014 /* Throwaway the data in the Rx completion */
2015 static void be_rx_compl_discard(struct be_rx_obj *rxo,
2016                                 struct be_rx_compl_info *rxcp)
2017 {
2018         struct be_rx_page_info *page_info;
2019         u16 i, num_rcvd = rxcp->num_rcvd;
2020
2021         for (i = 0; i < num_rcvd; i++) {
2022                 page_info = get_rx_page_info(rxo);
2023                 put_page(page_info->page);
2024                 memset(page_info, 0, sizeof(*page_info));
2025         }
2026 }
2027
2028 /*
2029  * skb_fill_rx_data forms a complete skb for an ether frame
2030  * indicated by rxcp.
2031  */
2032 static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
2033                              struct be_rx_compl_info *rxcp)
2034 {
2035         struct be_rx_page_info *page_info;
2036         u16 i, j;
2037         u16 hdr_len, curr_frag_len, remaining;
2038         u8 *start;
2039
2040         page_info = get_rx_page_info(rxo);
2041         start = page_address(page_info->page) + page_info->page_offset;
2042         prefetch(start);
2043
2044         /* Copy data in the first descriptor of this completion */
2045         curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
2046
2047         skb->len = curr_frag_len;
2048         if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
2049                 memcpy(skb->data, start, curr_frag_len);
2050                 /* Complete packet has now been moved to data */
2051                 put_page(page_info->page);
2052                 skb->data_len = 0;
2053                 skb->tail += curr_frag_len;
2054         } else {
2055                 hdr_len = ETH_HLEN;
2056                 memcpy(skb->data, start, hdr_len);
2057                 skb_shinfo(skb)->nr_frags = 1;
2058                 skb_frag_set_page(skb, 0, page_info->page);
2059                 skb_shinfo(skb)->frags[0].page_offset =
2060                                         page_info->page_offset + hdr_len;
2061                 skb_frag_size_set(&skb_shinfo(skb)->frags[0],
2062                                   curr_frag_len - hdr_len);
2063                 skb->data_len = curr_frag_len - hdr_len;
2064                 skb->truesize += rx_frag_size;
2065                 skb->tail += hdr_len;
2066         }
2067         page_info->page = NULL;
2068
2069         if (rxcp->pkt_size <= rx_frag_size) {
2070                 BUG_ON(rxcp->num_rcvd != 1);
2071                 return;
2072         }
2073
2074         /* More frags present for this completion */
2075         remaining = rxcp->pkt_size - curr_frag_len;
2076         for (i = 1, j = 0; i < rxcp->num_rcvd; i++) {
2077                 page_info = get_rx_page_info(rxo);
2078                 curr_frag_len = min(remaining, rx_frag_size);
2079
2080                 /* Coalesce all frags from the same physical page in one slot */
2081                 if (page_info->page_offset == 0) {
2082                         /* Fresh page */
2083                         j++;
2084                         skb_frag_set_page(skb, j, page_info->page);
2085                         skb_shinfo(skb)->frags[j].page_offset =
2086                                                         page_info->page_offset;
2087                         skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2088                         skb_shinfo(skb)->nr_frags++;
2089                 } else {
2090                         put_page(page_info->page);
2091                 }
2092
2093                 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2094                 skb->len += curr_frag_len;
2095                 skb->data_len += curr_frag_len;
2096                 skb->truesize += rx_frag_size;
2097                 remaining -= curr_frag_len;
2098                 page_info->page = NULL;
2099         }
2100         BUG_ON(j > MAX_SKB_FRAGS);
2101 }
2102
2103 /* Process the RX completion indicated by rxcp when GRO is disabled */
2104 static void be_rx_compl_process(struct be_rx_obj *rxo, struct napi_struct *napi,
2105                                 struct be_rx_compl_info *rxcp)
2106 {
2107         struct be_adapter *adapter = rxo->adapter;
2108         struct net_device *netdev = adapter->netdev;
2109         struct sk_buff *skb;
2110
2111         skb = netdev_alloc_skb_ip_align(netdev, BE_RX_SKB_ALLOC_SIZE);
2112         if (unlikely(!skb)) {
2113                 rx_stats(rxo)->rx_drops_no_skbs++;
2114                 be_rx_compl_discard(rxo, rxcp);
2115                 return;
2116         }
2117
2118         skb_fill_rx_data(rxo, skb, rxcp);
2119
2120         if (likely((netdev->features & NETIF_F_RXCSUM) && csum_passed(rxcp)))
2121                 skb->ip_summed = CHECKSUM_UNNECESSARY;
2122         else
2123                 skb_checksum_none_assert(skb);
2124
2125         skb->protocol = eth_type_trans(skb, netdev);
2126         skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2127         if (netdev->features & NETIF_F_RXHASH)
2128                 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2129
2130         skb->csum_level = rxcp->tunneled;
2131         skb_mark_napi_id(skb, napi);
2132
2133         if (rxcp->vlanf)
2134                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2135
2136         netif_receive_skb(skb);
2137 }
2138
2139 /* Process the RX completion indicated by rxcp when GRO is enabled */
2140 static void be_rx_compl_process_gro(struct be_rx_obj *rxo,
2141                                     struct napi_struct *napi,
2142                                     struct be_rx_compl_info *rxcp)
2143 {
2144         struct be_adapter *adapter = rxo->adapter;
2145         struct be_rx_page_info *page_info;
2146         struct sk_buff *skb = NULL;
2147         u16 remaining, curr_frag_len;
2148         u16 i, j;
2149
2150         skb = napi_get_frags(napi);
2151         if (!skb) {
2152                 be_rx_compl_discard(rxo, rxcp);
2153                 return;
2154         }
2155
2156         remaining = rxcp->pkt_size;
2157         for (i = 0, j = -1; i < rxcp->num_rcvd; i++) {
2158                 page_info = get_rx_page_info(rxo);
2159
2160                 curr_frag_len = min(remaining, rx_frag_size);
2161
2162                 /* Coalesce all frags from the same physical page in one slot */
2163                 if (i == 0 || page_info->page_offset == 0) {
2164                         /* First frag or Fresh page */
2165                         j++;
2166                         skb_frag_set_page(skb, j, page_info->page);
2167                         skb_shinfo(skb)->frags[j].page_offset =
2168                                                         page_info->page_offset;
2169                         skb_frag_size_set(&skb_shinfo(skb)->frags[j], 0);
2170                 } else {
2171                         put_page(page_info->page);
2172                 }
2173                 skb_frag_size_add(&skb_shinfo(skb)->frags[j], curr_frag_len);
2174                 skb->truesize += rx_frag_size;
2175                 remaining -= curr_frag_len;
2176                 memset(page_info, 0, sizeof(*page_info));
2177         }
2178         BUG_ON(j > MAX_SKB_FRAGS);
2179
2180         skb_shinfo(skb)->nr_frags = j + 1;
2181         skb->len = rxcp->pkt_size;
2182         skb->data_len = rxcp->pkt_size;
2183         skb->ip_summed = CHECKSUM_UNNECESSARY;
2184         skb_record_rx_queue(skb, rxo - &adapter->rx_obj[0]);
2185         if (adapter->netdev->features & NETIF_F_RXHASH)
2186                 skb_set_hash(skb, rxcp->rss_hash, PKT_HASH_TYPE_L3);
2187
2188         skb->csum_level = rxcp->tunneled;
2189
2190         if (rxcp->vlanf)
2191                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rxcp->vlan_tag);
2192
2193         napi_gro_frags(napi);
2194 }
2195
2196 static void be_parse_rx_compl_v1(struct be_eth_rx_compl *compl,
2197                                  struct be_rx_compl_info *rxcp)
2198 {
2199         rxcp->pkt_size = GET_RX_COMPL_V1_BITS(pktsize, compl);
2200         rxcp->vlanf = GET_RX_COMPL_V1_BITS(vtp, compl);
2201         rxcp->err = GET_RX_COMPL_V1_BITS(err, compl);
2202         rxcp->tcpf = GET_RX_COMPL_V1_BITS(tcpf, compl);
2203         rxcp->udpf = GET_RX_COMPL_V1_BITS(udpf, compl);
2204         rxcp->ip_csum = GET_RX_COMPL_V1_BITS(ipcksm, compl);
2205         rxcp->l4_csum = GET_RX_COMPL_V1_BITS(l4_cksm, compl);
2206         rxcp->ipv6 = GET_RX_COMPL_V1_BITS(ip_version, compl);
2207         rxcp->num_rcvd = GET_RX_COMPL_V1_BITS(numfrags, compl);
2208         rxcp->pkt_type = GET_RX_COMPL_V1_BITS(cast_enc, compl);
2209         rxcp->rss_hash = GET_RX_COMPL_V1_BITS(rsshash, compl);
2210         if (rxcp->vlanf) {
2211                 rxcp->qnq = GET_RX_COMPL_V1_BITS(qnq, compl);
2212                 rxcp->vlan_tag = GET_RX_COMPL_V1_BITS(vlan_tag, compl);
2213         }
2214         rxcp->port = GET_RX_COMPL_V1_BITS(port, compl);
2215         rxcp->tunneled =
2216                 GET_RX_COMPL_V1_BITS(tunneled, compl);
2217 }
2218
2219 static void be_parse_rx_compl_v0(struct be_eth_rx_compl *compl,
2220                                  struct be_rx_compl_info *rxcp)
2221 {
2222         rxcp->pkt_size = GET_RX_COMPL_V0_BITS(pktsize, compl);
2223         rxcp->vlanf = GET_RX_COMPL_V0_BITS(vtp, compl);
2224         rxcp->err = GET_RX_COMPL_V0_BITS(err, compl);
2225         rxcp->tcpf = GET_RX_COMPL_V0_BITS(tcpf, compl);
2226         rxcp->udpf = GET_RX_COMPL_V0_BITS(udpf, compl);
2227         rxcp->ip_csum = GET_RX_COMPL_V0_BITS(ipcksm, compl);
2228         rxcp->l4_csum = GET_RX_COMPL_V0_BITS(l4_cksm, compl);
2229         rxcp->ipv6 = GET_RX_COMPL_V0_BITS(ip_version, compl);
2230         rxcp->num_rcvd = GET_RX_COMPL_V0_BITS(numfrags, compl);
2231         rxcp->pkt_type = GET_RX_COMPL_V0_BITS(cast_enc, compl);
2232         rxcp->rss_hash = GET_RX_COMPL_V0_BITS(rsshash, compl);
2233         if (rxcp->vlanf) {
2234                 rxcp->qnq = GET_RX_COMPL_V0_BITS(qnq, compl);
2235                 rxcp->vlan_tag = GET_RX_COMPL_V0_BITS(vlan_tag, compl);
2236         }
2237         rxcp->port = GET_RX_COMPL_V0_BITS(port, compl);
2238         rxcp->ip_frag = GET_RX_COMPL_V0_BITS(ip_frag, compl);
2239 }
2240
2241 static struct be_rx_compl_info *be_rx_compl_get(struct be_rx_obj *rxo)
2242 {
2243         struct be_eth_rx_compl *compl = queue_tail_node(&rxo->cq);
2244         struct be_rx_compl_info *rxcp = &rxo->rxcp;
2245         struct be_adapter *adapter = rxo->adapter;
2246
2247         /* For checking the valid bit it is Ok to use either definition as the
2248          * valid bit is at the same position in both v0 and v1 Rx compl */
2249         if (compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] == 0)
2250                 return NULL;
2251
2252         rmb();
2253         be_dws_le_to_cpu(compl, sizeof(*compl));
2254
2255         if (adapter->be3_native)
2256                 be_parse_rx_compl_v1(compl, rxcp);
2257         else
2258                 be_parse_rx_compl_v0(compl, rxcp);
2259
2260         if (rxcp->ip_frag)
2261                 rxcp->l4_csum = 0;
2262
2263         if (rxcp->vlanf) {
2264                 /* In QNQ modes, if qnq bit is not set, then the packet was
2265                  * tagged only with the transparent outer vlan-tag and must
2266                  * not be treated as a vlan packet by host
2267                  */
2268                 if (be_is_qnq_mode(adapter) && !rxcp->qnq)
2269                         rxcp->vlanf = 0;
2270
2271                 if (!lancer_chip(adapter))
2272                         rxcp->vlan_tag = swab16(rxcp->vlan_tag);
2273
2274                 if (adapter->pvid == (rxcp->vlan_tag & VLAN_VID_MASK) &&
2275                     !test_bit(rxcp->vlan_tag, adapter->vids))
2276                         rxcp->vlanf = 0;
2277         }
2278
2279         /* As the compl has been parsed, reset it; we wont touch it again */
2280         compl->dw[offsetof(struct amap_eth_rx_compl_v1, valid) / 32] = 0;
2281
2282         queue_tail_inc(&rxo->cq);
2283         return rxcp;
2284 }
2285
2286 static inline struct page *be_alloc_pages(u32 size, gfp_t gfp)
2287 {
2288         u32 order = get_order(size);
2289
2290         if (order > 0)
2291                 gfp |= __GFP_COMP;
2292         return  alloc_pages(gfp, order);
2293 }
2294
2295 /*
2296  * Allocate a page, split it to fragments of size rx_frag_size and post as
2297  * receive buffers to BE
2298  */
2299 static void be_post_rx_frags(struct be_rx_obj *rxo, gfp_t gfp, u32 frags_needed)
2300 {
2301         struct be_adapter *adapter = rxo->adapter;
2302         struct be_rx_page_info *page_info = NULL, *prev_page_info = NULL;
2303         struct be_queue_info *rxq = &rxo->q;
2304         struct page *pagep = NULL;
2305         struct device *dev = &adapter->pdev->dev;
2306         struct be_eth_rx_d *rxd;
2307         u64 page_dmaaddr = 0, frag_dmaaddr;
2308         u32 posted, page_offset = 0, notify = 0;
2309
2310         page_info = &rxo->page_info_tbl[rxq->head];
2311         for (posted = 0; posted < frags_needed && !page_info->page; posted++) {
2312                 if (!pagep) {
2313                         pagep = be_alloc_pages(adapter->big_page_size, gfp);
2314                         if (unlikely(!pagep)) {
2315                                 rx_stats(rxo)->rx_post_fail++;
2316                                 break;
2317                         }
2318                         page_dmaaddr = dma_map_page(dev, pagep, 0,
2319                                                     adapter->big_page_size,
2320                                                     DMA_FROM_DEVICE);
2321                         if (dma_mapping_error(dev, page_dmaaddr)) {
2322                                 put_page(pagep);
2323                                 pagep = NULL;
2324                                 adapter->drv_stats.dma_map_errors++;
2325                                 break;
2326                         }
2327                         page_offset = 0;
2328                 } else {
2329                         get_page(pagep);
2330                         page_offset += rx_frag_size;
2331                 }
2332                 page_info->page_offset = page_offset;
2333                 page_info->page = pagep;
2334
2335                 rxd = queue_head_node(rxq);
2336                 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
2337                 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
2338                 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
2339
2340                 /* Any space left in the current big page for another frag? */
2341                 if ((page_offset + rx_frag_size + rx_frag_size) >
2342                                         adapter->big_page_size) {
2343                         pagep = NULL;
2344                         page_info->last_frag = true;
2345                         dma_unmap_addr_set(page_info, bus, page_dmaaddr);
2346                 } else {
2347                         dma_unmap_addr_set(page_info, bus, frag_dmaaddr);
2348                 }
2349
2350                 prev_page_info = page_info;
2351                 queue_head_inc(rxq);
2352                 page_info = &rxo->page_info_tbl[rxq->head];
2353         }
2354
2355         /* Mark the last frag of a page when we break out of the above loop
2356          * with no more slots available in the RXQ
2357          */
2358         if (pagep) {
2359                 prev_page_info->last_frag = true;
2360                 dma_unmap_addr_set(prev_page_info, bus, page_dmaaddr);
2361         }
2362
2363         if (posted) {
2364                 atomic_add(posted, &rxq->used);
2365                 if (rxo->rx_post_starved)
2366                         rxo->rx_post_starved = false;
2367                 do {
2368                         notify = min(MAX_NUM_POST_ERX_DB, posted);
2369                         be_rxq_notify(adapter, rxq->id, notify);
2370                         posted -= notify;
2371                 } while (posted);
2372         } else if (atomic_read(&rxq->used) == 0) {
2373                 /* Let be_worker replenish when memory is available */
2374                 rxo->rx_post_starved = true;
2375         }
2376 }
2377
2378 static struct be_tx_compl_info *be_tx_compl_get(struct be_tx_obj *txo)
2379 {
2380         struct be_queue_info *tx_cq = &txo->cq;
2381         struct be_tx_compl_info *txcp = &txo->txcp;
2382         struct be_eth_tx_compl *compl = queue_tail_node(tx_cq);
2383
2384         if (compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
2385                 return NULL;
2386
2387         /* Ensure load ordering of valid bit dword and other dwords below */
2388         rmb();
2389         be_dws_le_to_cpu(compl, sizeof(*compl));
2390
2391         txcp->status = GET_TX_COMPL_BITS(status, compl);
2392         txcp->end_index = GET_TX_COMPL_BITS(wrb_index, compl);
2393
2394         compl->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
2395         queue_tail_inc(tx_cq);
2396         return txcp;
2397 }
2398
2399 static u16 be_tx_compl_process(struct be_adapter *adapter,
2400                                struct be_tx_obj *txo, u16 last_index)
2401 {
2402         struct sk_buff **sent_skbs = txo->sent_skb_list;
2403         struct be_queue_info *txq = &txo->q;
2404         u16 frag_index, num_wrbs = 0;
2405         struct sk_buff *skb = NULL;
2406         bool unmap_skb_hdr = false;
2407         struct be_eth_wrb *wrb;
2408
2409         do {
2410                 if (sent_skbs[txq->tail]) {
2411                         /* Free skb from prev req */
2412                         if (skb)
2413                                 dev_consume_skb_any(skb);
2414                         skb = sent_skbs[txq->tail];
2415                         sent_skbs[txq->tail] = NULL;
2416                         queue_tail_inc(txq);  /* skip hdr wrb */
2417                         num_wrbs++;
2418                         unmap_skb_hdr = true;
2419                 }
2420                 wrb = queue_tail_node(txq);
2421                 frag_index = txq->tail;
2422                 unmap_tx_frag(&adapter->pdev->dev, wrb,
2423                               (unmap_skb_hdr && skb_headlen(skb)));
2424                 unmap_skb_hdr = false;
2425                 queue_tail_inc(txq);
2426                 num_wrbs++;
2427         } while (frag_index != last_index);
2428         dev_consume_skb_any(skb);
2429
2430         return num_wrbs;
2431 }
2432
2433 /* Return the number of events in the event queue */
2434 static inline int events_get(struct be_eq_obj *eqo)
2435 {
2436         struct be_eq_entry *eqe;
2437         int num = 0;
2438
2439         do {
2440                 eqe = queue_tail_node(&eqo->q);
2441                 if (eqe->evt == 0)
2442                         break;
2443
2444                 rmb();
2445                 eqe->evt = 0;
2446                 num++;
2447                 queue_tail_inc(&eqo->q);
2448         } while (true);
2449
2450         return num;
2451 }
2452
2453 /* Leaves the EQ is disarmed state */
2454 static void be_eq_clean(struct be_eq_obj *eqo)
2455 {
2456         int num = events_get(eqo);
2457
2458         be_eq_notify(eqo->adapter, eqo->q.id, false, true, num, 0);
2459 }
2460
2461 /* Free posted rx buffers that were not used */
2462 static void be_rxq_clean(struct be_rx_obj *rxo)
2463 {
2464         struct be_queue_info *rxq = &rxo->q;
2465         struct be_rx_page_info *page_info;
2466
2467         while (atomic_read(&rxq->used) > 0) {
2468                 page_info = get_rx_page_info(rxo);
2469                 put_page(page_info->page);
2470                 memset(page_info, 0, sizeof(*page_info));
2471         }
2472         BUG_ON(atomic_read(&rxq->used));
2473         rxq->tail = 0;
2474         rxq->head = 0;
2475 }
2476
2477 static void be_rx_cq_clean(struct be_rx_obj *rxo)
2478 {
2479         struct be_queue_info *rx_cq = &rxo->cq;
2480         struct be_rx_compl_info *rxcp;
2481         struct be_adapter *adapter = rxo->adapter;
2482         int flush_wait = 0;
2483
2484         /* Consume pending rx completions.
2485          * Wait for the flush completion (identified by zero num_rcvd)
2486          * to arrive. Notify CQ even when there are no more CQ entries
2487          * for HW to flush partially coalesced CQ entries.
2488          * In Lancer, there is no need to wait for flush compl.
2489          */
2490         for (;;) {
2491                 rxcp = be_rx_compl_get(rxo);
2492                 if (!rxcp) {
2493                         if (lancer_chip(adapter))
2494                                 break;
2495
2496                         if (flush_wait++ > 50 ||
2497                             be_check_error(adapter,
2498                                            BE_ERROR_HW)) {
2499                                 dev_warn(&adapter->pdev->dev,
2500                                          "did not receive flush compl\n");
2501                                 break;
2502                         }
2503                         be_cq_notify(adapter, rx_cq->id, true, 0);
2504                         mdelay(1);
2505                 } else {
2506                         be_rx_compl_discard(rxo, rxcp);
2507                         be_cq_notify(adapter, rx_cq->id, false, 1);
2508                         if (rxcp->num_rcvd == 0)
2509                                 break;
2510                 }
2511         }
2512
2513         /* After cleanup, leave the CQ in unarmed state */
2514         be_cq_notify(adapter, rx_cq->id, false, 0);
2515 }
2516
2517 static void be_tx_compl_clean(struct be_adapter *adapter)
2518 {
2519         u16 end_idx, notified_idx, cmpl = 0, timeo = 0, num_wrbs = 0;
2520         struct device *dev = &adapter->pdev->dev;
2521         struct be_tx_compl_info *txcp;
2522         struct be_queue_info *txq;
2523         struct be_tx_obj *txo;
2524         int i, pending_txqs;
2525
2526         /* Stop polling for compls when HW has been silent for 10ms */
2527         do {
2528                 pending_txqs = adapter->num_tx_qs;
2529
2530                 for_all_tx_queues(adapter, txo, i) {
2531                         cmpl = 0;
2532                         num_wrbs = 0;
2533                         txq = &txo->q;
2534                         while ((txcp = be_tx_compl_get(txo))) {
2535                                 num_wrbs +=
2536                                         be_tx_compl_process(adapter, txo,
2537                                                             txcp->end_index);
2538                                 cmpl++;
2539                         }
2540                         if (cmpl) {
2541                                 be_cq_notify(adapter, txo->cq.id, false, cmpl);
2542                                 atomic_sub(num_wrbs, &txq->used);
2543                                 timeo = 0;
2544                         }
2545                         if (!be_is_tx_compl_pending(txo))
2546                                 pending_txqs--;
2547                 }
2548
2549                 if (pending_txqs == 0 || ++timeo > 10 ||
2550                     be_check_error(adapter, BE_ERROR_HW))
2551                         break;
2552
2553                 mdelay(1);
2554         } while (true);
2555
2556         /* Free enqueued TX that was never notified to HW */
2557         for_all_tx_queues(adapter, txo, i) {
2558                 txq = &txo->q;
2559
2560                 if (atomic_read(&txq->used)) {
2561                         dev_info(dev, "txq%d: cleaning %d pending tx-wrbs\n",
2562                                  i, atomic_read(&txq->used));
2563                         notified_idx = txq->tail;
2564                         end_idx = txq->tail;
2565                         index_adv(&end_idx, atomic_read(&txq->used) - 1,
2566                                   txq->len);
2567                         /* Use the tx-compl process logic to handle requests
2568                          * that were not sent to the HW.
2569                          */
2570                         num_wrbs = be_tx_compl_process(adapter, txo, end_idx);
2571                         atomic_sub(num_wrbs, &txq->used);
2572                         BUG_ON(atomic_read(&txq->used));
2573                         txo->pend_wrb_cnt = 0;
2574                         /* Since hw was never notified of these requests,
2575                          * reset TXQ indices
2576                          */
2577                         txq->head = notified_idx;
2578                         txq->tail = notified_idx;
2579                 }
2580         }
2581 }
2582
2583 static void be_evt_queues_destroy(struct be_adapter *adapter)
2584 {
2585         struct be_eq_obj *eqo;
2586         int i;
2587
2588         for_all_evt_queues(adapter, eqo, i) {
2589                 if (eqo->q.created) {
2590                         be_eq_clean(eqo);
2591                         be_cmd_q_destroy(adapter, &eqo->q, QTYPE_EQ);
2592                         napi_hash_del(&eqo->napi);
2593                         netif_napi_del(&eqo->napi);
2594                         free_cpumask_var(eqo->affinity_mask);
2595                 }
2596                 be_queue_free(adapter, &eqo->q);
2597         }
2598 }
2599
2600 static int be_evt_queues_create(struct be_adapter *adapter)
2601 {
2602         struct be_queue_info *eq;
2603         struct be_eq_obj *eqo;
2604         struct be_aic_obj *aic;
2605         int i, rc;
2606
2607         adapter->num_evt_qs = min_t(u16, num_irqs(adapter),
2608                                     adapter->cfg_num_qs);
2609
2610         for_all_evt_queues(adapter, eqo, i) {
2611                 int numa_node = dev_to_node(&adapter->pdev->dev);
2612
2613                 aic = &adapter->aic_obj[i];
2614                 eqo->adapter = adapter;
2615                 eqo->idx = i;
2616                 aic->max_eqd = BE_MAX_EQD;
2617                 aic->enable = true;
2618
2619                 eq = &eqo->q;
2620                 rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
2621                                     sizeof(struct be_eq_entry));
2622                 if (rc)
2623                         return rc;
2624
2625                 rc = be_cmd_eq_create(adapter, eqo);
2626                 if (rc)
2627                         return rc;
2628
2629                 if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
2630                         return -ENOMEM;
2631                 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
2632                                 eqo->affinity_mask);
2633                 netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
2634                                BE_NAPI_WEIGHT);
2635         }
2636         return 0;
2637 }
2638
2639 static void be_mcc_queues_destroy(struct be_adapter *adapter)
2640 {
2641         struct be_queue_info *q;
2642
2643         q = &adapter->mcc_obj.q;
2644         if (q->created)
2645                 be_cmd_q_destroy(adapter, q, QTYPE_MCCQ);
2646         be_queue_free(adapter, q);
2647
2648         q = &adapter->mcc_obj.cq;
2649         if (q->created)
2650                 be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2651         be_queue_free(adapter, q);
2652 }
2653
2654 /* Must be called only after TX qs are created as MCC shares TX EQ */
2655 static int be_mcc_queues_create(struct be_adapter *adapter)
2656 {
2657         struct be_queue_info *q, *cq;
2658
2659         cq = &adapter->mcc_obj.cq;
2660         if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
2661                            sizeof(struct be_mcc_compl)))
2662                 goto err;
2663
2664         /* Use the default EQ for MCC completions */
2665         if (be_cmd_cq_create(adapter, cq, &mcc_eqo(adapter)->q, true, 0))
2666                 goto mcc_cq_free;
2667
2668         q = &adapter->mcc_obj.q;
2669         if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
2670                 goto mcc_cq_destroy;
2671
2672         if (be_cmd_mccq_create(adapter, q, cq))
2673                 goto mcc_q_free;
2674
2675         return 0;
2676
2677 mcc_q_free:
2678         be_queue_free(adapter, q);
2679 mcc_cq_destroy:
2680         be_cmd_q_destroy(adapter, cq, QTYPE_CQ);
2681 mcc_cq_free:
2682         be_queue_free(adapter, cq);
2683 err:
2684         return -1;
2685 }
2686
2687 static void be_tx_queues_destroy(struct be_adapter *adapter)
2688 {
2689         struct be_queue_info *q;
2690         struct be_tx_obj *txo;
2691         u8 i;
2692
2693         for_all_tx_queues(adapter, txo, i) {
2694                 q = &txo->q;
2695                 if (q->created)
2696                         be_cmd_q_destroy(adapter, q, QTYPE_TXQ);
2697                 be_queue_free(adapter, q);
2698
2699                 q = &txo->cq;
2700                 if (q->created)
2701                         be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2702                 be_queue_free(adapter, q);
2703         }
2704 }
2705
2706 static int be_tx_qs_create(struct be_adapter *adapter)
2707 {
2708         struct be_queue_info *cq;
2709         struct be_tx_obj *txo;
2710         struct be_eq_obj *eqo;
2711         int status, i;
2712
2713         adapter->num_tx_qs = min(adapter->num_evt_qs, be_max_txqs(adapter));
2714
2715         for_all_tx_queues(adapter, txo, i) {
2716                 cq = &txo->cq;
2717                 status = be_queue_alloc(adapter, cq, TX_CQ_LEN,
2718                                         sizeof(struct be_eth_tx_compl));
2719                 if (status)
2720                         return status;
2721
2722                 u64_stats_init(&txo->stats.sync);
2723                 u64_stats_init(&txo->stats.sync_compl);
2724
2725                 /* If num_evt_qs is less than num_tx_qs, then more than
2726                  * one txq share an eq
2727                  */
2728                 eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
2729                 status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
2730                 if (status)
2731                         return status;
2732
2733                 status = be_queue_alloc(adapter, &txo->q, TX_Q_LEN,
2734                                         sizeof(struct be_eth_wrb));
2735                 if (status)
2736                         return status;
2737
2738                 status = be_cmd_txq_create(adapter, txo);
2739                 if (status)
2740                         return status;
2741
2742                 netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
2743                                     eqo->idx);
2744         }
2745
2746         dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
2747                  adapter->num_tx_qs);
2748         return 0;
2749 }
2750
2751 static void be_rx_cqs_destroy(struct be_adapter *adapter)
2752 {
2753         struct be_queue_info *q;
2754         struct be_rx_obj *rxo;
2755         int i;
2756
2757         for_all_rx_queues(adapter, rxo, i) {
2758                 q = &rxo->cq;
2759                 if (q->created)
2760                         be_cmd_q_destroy(adapter, q, QTYPE_CQ);
2761                 be_queue_free(adapter, q);
2762         }
2763 }
2764
2765 static int be_rx_cqs_create(struct be_adapter *adapter)
2766 {
2767         struct be_queue_info *eq, *cq;
2768         struct be_rx_obj *rxo;
2769         int rc, i;
2770
2771         /* We can create as many RSS rings as there are EQs. */
2772         adapter->num_rss_qs = adapter->num_evt_qs;
2773
2774         /* We'll use RSS only if atleast 2 RSS rings are supported. */
2775         if (adapter->num_rss_qs <= 1)
2776                 adapter->num_rss_qs = 0;
2777
2778         adapter->num_rx_qs = adapter->num_rss_qs + adapter->need_def_rxq;
2779
2780         /* When the interface is not capable of RSS rings (and there is no
2781          * need to create a default RXQ) we'll still need one RXQ
2782          */
2783         if (adapter->num_rx_qs == 0)
2784                 adapter->num_rx_qs = 1;
2785
2786         adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
2787         for_all_rx_queues(adapter, rxo, i) {
2788                 rxo->adapter = adapter;
2789                 cq = &rxo->cq;
2790                 rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
2791                                     sizeof(struct be_eth_rx_compl));
2792                 if (rc)
2793                         return rc;
2794
2795                 u64_stats_init(&rxo->stats.sync);
2796                 eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
2797                 rc = be_cmd_cq_create(adapter, cq, eq, false, 3);
2798                 if (rc)
2799                         return rc;
2800         }
2801
2802         dev_info(&adapter->pdev->dev,
2803                  "created %d RX queue(s)\n", adapter->num_rx_qs);
2804         return 0;
2805 }
2806
2807 static irqreturn_t be_intx(int irq, void *dev)
2808 {
2809         struct be_eq_obj *eqo = dev;
2810         struct be_adapter *adapter = eqo->adapter;
2811         int num_evts = 0;
2812
2813         /* IRQ is not expected when NAPI is scheduled as the EQ
2814          * will not be armed.
2815          * But, this can happen on Lancer INTx where it takes
2816          * a while to de-assert INTx or in BE2 where occasionaly
2817          * an interrupt may be raised even when EQ is unarmed.
2818          * If NAPI is already scheduled, then counting & notifying
2819          * events will orphan them.
2820          */
2821         if (napi_schedule_prep(&eqo->napi)) {
2822                 num_evts = events_get(eqo);
2823                 __napi_schedule(&eqo->napi);
2824                 if (num_evts)
2825                         eqo->spurious_intr = 0;
2826         }
2827         be_eq_notify(adapter, eqo->q.id, false, true, num_evts, 0);
2828
2829         /* Return IRQ_HANDLED only for the the first spurious intr
2830          * after a valid intr to stop the kernel from branding
2831          * this irq as a bad one!
2832          */
2833         if (num_evts || eqo->spurious_intr++ == 0)
2834                 return IRQ_HANDLED;
2835         else
2836                 return IRQ_NONE;
2837 }
2838
2839 static irqreturn_t be_msix(int irq, void *dev)
2840 {
2841         struct be_eq_obj *eqo = dev;
2842
2843         be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
2844         napi_schedule(&eqo->napi);
2845         return IRQ_HANDLED;
2846 }
2847
2848 static inline bool do_gro(struct be_rx_compl_info *rxcp)
2849 {
2850         return (rxcp->tcpf && !rxcp->err && rxcp->l4_csum) ? true : false;
2851 }
2852
2853 static int be_process_rx(struct be_rx_obj *rxo, struct napi_struct *napi,
2854                          int budget, int polling)
2855 {
2856         struct be_adapter *adapter = rxo->adapter;
2857         struct be_queue_info *rx_cq = &rxo->cq;
2858         struct be_rx_compl_info *rxcp;
2859         u32 work_done;
2860         u32 frags_consumed = 0;
2861
2862         for (work_done = 0; work_done < budget; work_done++) {
2863                 rxcp = be_rx_compl_get(rxo);
2864                 if (!rxcp)
2865                         break;
2866
2867                 /* Is it a flush compl that has no data */
2868                 if (unlikely(rxcp->num_rcvd == 0))
2869                         goto loop_continue;
2870
2871                 /* Discard compl with partial DMA Lancer B0 */
2872                 if (unlikely(!rxcp->pkt_size)) {
2873                         be_rx_compl_discard(rxo, rxcp);
2874                         goto loop_continue;
2875                 }
2876
2877                 /* On BE drop pkts that arrive due to imperfect filtering in
2878                  * promiscuous mode on some skews
2879                  */
2880                 if (unlikely(rxcp->port != adapter->port_num &&
2881                              !lancer_chip(adapter))) {
2882                         be_rx_compl_discard(rxo, rxcp);
2883                         goto loop_continue;
2884                 }
2885
2886                 /* Don't do gro when we're busy_polling */
2887                 if (do_gro(rxcp) && polling != BUSY_POLLING)
2888                         be_rx_compl_process_gro(rxo, napi, rxcp);
2889                 else
2890                         be_rx_compl_process(rxo, napi, rxcp);
2891
2892 loop_continue:
2893                 frags_consumed += rxcp->num_rcvd;
2894                 be_rx_stats_update(rxo, rxcp);
2895         }
2896
2897         if (work_done) {
2898                 be_cq_notify(adapter, rx_cq->id, true, work_done);
2899
2900                 /* When an rx-obj gets into post_starved state, just
2901                  * let be_worker do the posting.
2902                  */
2903                 if (atomic_read(&rxo->q.used) < RX_FRAGS_REFILL_WM &&
2904                     !rxo->rx_post_starved)
2905                         be_post_rx_frags(rxo, GFP_ATOMIC,
2906                                          max_t(u32, MAX_RX_POST,
2907                                                frags_consumed));
2908         }
2909
2910         return work_done;
2911 }
2912
2913 static inline void be_update_tx_err(struct be_tx_obj *txo, u8 status)
2914 {
2915         switch (status) {
2916         case BE_TX_COMP_HDR_PARSE_ERR:
2917                 tx_stats(txo)->tx_hdr_parse_err++;
2918                 break;
2919         case BE_TX_COMP_NDMA_ERR:
2920                 tx_stats(txo)->tx_dma_err++;
2921                 break;
2922         case BE_TX_COMP_ACL_ERR:
2923                 tx_stats(txo)->tx_spoof_check_err++;
2924                 break;
2925         }
2926 }
2927
2928 static inline void lancer_update_tx_err(struct be_tx_obj *txo, u8 status)
2929 {
2930         switch (status) {
2931         case LANCER_TX_COMP_LSO_ERR:
2932                 tx_stats(txo)->tx_tso_err++;
2933                 break;
2934         case LANCER_TX_COMP_HSW_DROP_MAC_ERR:
2935         case LANCER_TX_COMP_HSW_DROP_VLAN_ERR:
2936                 tx_stats(txo)->tx_spoof_check_err++;
2937                 break;
2938         case LANCER_TX_COMP_QINQ_ERR:
2939                 tx_stats(txo)->tx_qinq_err++;
2940                 break;
2941         case LANCER_TX_COMP_PARITY_ERR:
2942                 tx_stats(txo)->tx_internal_parity_err++;
2943                 break;
2944         case LANCER_TX_COMP_DMA_ERR:
2945                 tx_stats(txo)->tx_dma_err++;
2946                 break;
2947         }
2948 }
2949
2950 static void be_process_tx(struct be_adapter *adapter, struct be_tx_obj *txo,
2951                           int idx)
2952 {
2953         int num_wrbs = 0, work_done = 0;
2954         struct be_tx_compl_info *txcp;
2955
2956         while ((txcp = be_tx_compl_get(txo))) {
2957                 num_wrbs += be_tx_compl_process(adapter, txo, txcp->end_index);
2958                 work_done++;
2959
2960                 if (txcp->status) {
2961                         if (lancer_chip(adapter))
2962                                 lancer_update_tx_err(txo, txcp->status);
2963                         else
2964                                 be_update_tx_err(txo, txcp->status);
2965                 }
2966         }
2967
2968         if (work_done) {
2969                 be_cq_notify(adapter, txo->cq.id, true, work_done);
2970                 atomic_sub(num_wrbs, &txo->q.used);
2971
2972                 /* As Tx wrbs have been freed up, wake up netdev queue
2973                  * if it was stopped due to lack of tx wrbs.  */
2974                 if (__netif_subqueue_stopped(adapter->netdev, idx) &&
2975                     be_can_txq_wake(txo)) {
2976                         netif_wake_subqueue(adapter->netdev, idx);
2977                 }
2978
2979                 u64_stats_update_begin(&tx_stats(txo)->sync_compl);
2980                 tx_stats(txo)->tx_compl += work_done;
2981                 u64_stats_update_end(&tx_stats(txo)->sync_compl);
2982         }
2983 }
2984
2985 #ifdef CONFIG_NET_RX_BUSY_POLL
2986 static inline bool be_lock_napi(struct be_eq_obj *eqo)
2987 {
2988         bool status = true;
2989
2990         spin_lock(&eqo->lock); /* BH is already disabled */
2991         if (eqo->state & BE_EQ_LOCKED) {
2992                 WARN_ON(eqo->state & BE_EQ_NAPI);
2993                 eqo->state |= BE_EQ_NAPI_YIELD;
2994                 status = false;
2995         } else {
2996                 eqo->state = BE_EQ_NAPI;
2997         }
2998         spin_unlock(&eqo->lock);
2999         return status;
3000 }
3001
3002 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3003 {
3004         spin_lock(&eqo->lock); /* BH is already disabled */
3005
3006         WARN_ON(eqo->state & (BE_EQ_POLL | BE_EQ_NAPI_YIELD));
3007         eqo->state = BE_EQ_IDLE;
3008
3009         spin_unlock(&eqo->lock);
3010 }
3011
3012 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3013 {
3014         bool status = true;
3015
3016         spin_lock_bh(&eqo->lock);
3017         if (eqo->state & BE_EQ_LOCKED) {
3018                 eqo->state |= BE_EQ_POLL_YIELD;
3019                 status = false;
3020         } else {
3021                 eqo->state |= BE_EQ_POLL;
3022         }
3023         spin_unlock_bh(&eqo->lock);
3024         return status;
3025 }
3026
3027 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3028 {
3029         spin_lock_bh(&eqo->lock);
3030
3031         WARN_ON(eqo->state & (BE_EQ_NAPI));
3032         eqo->state = BE_EQ_IDLE;
3033
3034         spin_unlock_bh(&eqo->lock);
3035 }
3036
3037 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3038 {
3039         spin_lock_init(&eqo->lock);
3040         eqo->state = BE_EQ_IDLE;
3041 }
3042
3043 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3044 {
3045         local_bh_disable();
3046
3047         /* It's enough to just acquire napi lock on the eqo to stop
3048          * be_busy_poll() from processing any queueus.
3049          */
3050         while (!be_lock_napi(eqo))
3051                 mdelay(1);
3052
3053         local_bh_enable();
3054 }
3055
3056 #else /* CONFIG_NET_RX_BUSY_POLL */
3057
3058 static inline bool be_lock_napi(struct be_eq_obj *eqo)
3059 {
3060         return true;
3061 }
3062
3063 static inline void be_unlock_napi(struct be_eq_obj *eqo)
3064 {
3065 }
3066
3067 static inline bool be_lock_busy_poll(struct be_eq_obj *eqo)
3068 {
3069         return false;
3070 }
3071
3072 static inline void be_unlock_busy_poll(struct be_eq_obj *eqo)
3073 {
3074 }
3075
3076 static inline void be_enable_busy_poll(struct be_eq_obj *eqo)
3077 {
3078 }
3079
3080 static inline void be_disable_busy_poll(struct be_eq_obj *eqo)
3081 {
3082 }
3083 #endif /* CONFIG_NET_RX_BUSY_POLL */
3084
3085 int be_poll(struct napi_struct *napi, int budget)
3086 {
3087         struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3088         struct be_adapter *adapter = eqo->adapter;
3089         int max_work = 0, work, i, num_evts;
3090         struct be_rx_obj *rxo;
3091         struct be_tx_obj *txo;
3092         u32 mult_enc = 0;
3093
3094         num_evts = events_get(eqo);
3095
3096         for_all_tx_queues_on_eq(adapter, eqo, txo, i)
3097                 be_process_tx(adapter, txo, i);
3098
3099         if (be_lock_napi(eqo)) {
3100                 /* This loop will iterate twice for EQ0 in which
3101                  * completions of the last RXQ (default one) are also processed
3102                  * For other EQs the loop iterates only once
3103                  */
3104                 for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3105                         work = be_process_rx(rxo, napi, budget, NAPI_POLLING);
3106                         max_work = max(work, max_work);
3107                 }
3108                 be_unlock_napi(eqo);
3109         } else {
3110                 max_work = budget;
3111         }
3112
3113         if (is_mcc_eqo(eqo))
3114                 be_process_mcc(adapter);
3115
3116         if (max_work < budget) {
3117                 napi_complete(napi);
3118
3119                 /* Skyhawk EQ_DB has a provision to set the rearm to interrupt
3120                  * delay via a delay multiplier encoding value
3121                  */
3122                 if (skyhawk_chip(adapter))
3123                         mult_enc = be_get_eq_delay_mult_enc(eqo);
3124
3125                 be_eq_notify(adapter, eqo->q.id, true, false, num_evts,
3126                              mult_enc);
3127         } else {
3128                 /* As we'll continue in polling mode, count and clear events */
3129                 be_eq_notify(adapter, eqo->q.id, false, false, num_evts, 0);
3130         }
3131         return max_work;
3132 }
3133
3134 #ifdef CONFIG_NET_RX_BUSY_POLL
3135 static int be_busy_poll(struct napi_struct *napi)
3136 {
3137         struct be_eq_obj *eqo = container_of(napi, struct be_eq_obj, napi);
3138         struct be_adapter *adapter = eqo->adapter;
3139         struct be_rx_obj *rxo;
3140         int i, work = 0;
3141
3142         if (!be_lock_busy_poll(eqo))
3143                 return LL_FLUSH_BUSY;
3144
3145         for_all_rx_queues_on_eq(adapter, eqo, rxo, i) {
3146                 work = be_process_rx(rxo, napi, 4, BUSY_POLLING);
3147                 if (work)
3148                         break;
3149         }
3150
3151         be_unlock_busy_poll(eqo);
3152         return work;
3153 }
3154 #endif
3155
3156 void be_detect_error(struct be_adapter *adapter)
3157 {
3158         u32 ue_lo = 0, ue_hi = 0, ue_lo_mask = 0, ue_hi_mask = 0;
3159         u32 sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
3160         u32 i;
3161         struct device *dev = &adapter->pdev->dev;
3162
3163         if (be_check_error(adapter, BE_ERROR_HW))
3164                 return;
3165
3166         if (lancer_chip(adapter)) {
3167                 sliport_status = ioread32(adapter->db + SLIPORT_STATUS_OFFSET);
3168                 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
3169                         be_set_error(adapter, BE_ERROR_UE);
3170                         sliport_err1 = ioread32(adapter->db +
3171                                                 SLIPORT_ERROR1_OFFSET);
3172                         sliport_err2 = ioread32(adapter->db +
3173                                                 SLIPORT_ERROR2_OFFSET);
3174                         /* Do not log error messages if its a FW reset */
3175                         if (sliport_err1 == SLIPORT_ERROR_FW_RESET1 &&
3176                             sliport_err2 == SLIPORT_ERROR_FW_RESET2) {
3177                                 dev_info(dev, "Firmware update in progress\n");
3178                         } else {
3179                                 dev_err(dev, "Error detected in the card\n");
3180                                 dev_err(dev, "ERR: sliport status 0x%x\n",
3181                                         sliport_status);
3182                                 dev_err(dev, "ERR: sliport error1 0x%x\n",
3183                                         sliport_err1);
3184                                 dev_err(dev, "ERR: sliport error2 0x%x\n",
3185                                         sliport_err2);
3186                         }
3187                 }
3188         } else {
3189                 ue_lo = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_LOW);
3190                 ue_hi = ioread32(adapter->pcicfg + PCICFG_UE_STATUS_HIGH);
3191                 ue_lo_mask = ioread32(adapter->pcicfg +
3192                                       PCICFG_UE_STATUS_LOW_MASK);
3193                 ue_hi_mask = ioread32(adapter->pcicfg +
3194                                       PCICFG_UE_STATUS_HI_MASK);
3195
3196                 ue_lo = (ue_lo & ~ue_lo_mask);
3197                 ue_hi = (ue_hi & ~ue_hi_mask);
3198
3199                 /* On certain platforms BE hardware can indicate spurious UEs.
3200                  * Allow HW to stop working completely in case of a real UE.
3201                  * Hence not setting the hw_error for UE detection.
3202                  */
3203
3204                 if (ue_lo || ue_hi) {
3205                         dev_err(dev,
3206                                 "Unrecoverable Error detected in the adapter");
3207                         dev_err(dev, "Please reboot server to recover");
3208                         if (skyhawk_chip(adapter))
3209                                 be_set_error(adapter, BE_ERROR_UE);
3210
3211                         for (i = 0; ue_lo; ue_lo >>= 1, i++) {
3212                                 if (ue_lo & 1)
3213                                         dev_err(dev, "UE: %s bit set\n",
3214                                                 ue_status_low_desc[i]);
3215                         }
3216                         for (i = 0; ue_hi; ue_hi >>= 1, i++) {
3217                                 if (ue_hi & 1)
3218                                         dev_err(dev, "UE: %s bit set\n",
3219                                                 ue_status_hi_desc[i]);
3220                         }
3221                 }
3222         }
3223 }
3224
3225 static void be_msix_disable(struct be_adapter *adapter)
3226 {
3227         if (msix_enabled(adapter)) {
3228                 pci_disable_msix(adapter->pdev);
3229                 adapter->num_msix_vec = 0;
3230                 adapter->num_msix_roce_vec = 0;
3231         }
3232 }
3233
3234 static int be_msix_enable(struct be_adapter *adapter)
3235 {
3236         int i, num_vec;
3237         struct device *dev = &adapter->pdev->dev;
3238
3239         /* If RoCE is supported, program the max number of NIC vectors that
3240          * may be configured via set-channels, along with vectors needed for
3241          * RoCe. Else, just program the number we'll use initially.
3242          */
3243         if (be_roce_supported(adapter))
3244                 num_vec = min_t(int, 2 * be_max_eqs(adapter),
3245                                 2 * num_online_cpus());
3246         else
3247                 num_vec = adapter->cfg_num_qs;
3248
3249         for (i = 0; i < num_vec; i++)
3250                 adapter->msix_entries[i].entry = i;
3251
3252         num_vec = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
3253                                         MIN_MSIX_VECTORS, num_vec);
3254         if (num_vec < 0)
3255                 goto fail;
3256
3257         if (be_roce_supported(adapter) && num_vec > MIN_MSIX_VECTORS) {
3258                 adapter->num_msix_roce_vec = num_vec / 2;
3259                 dev_info(dev, "enabled %d MSI-x vector(s) for RoCE\n",
3260                          adapter->num_msix_roce_vec);
3261         }
3262
3263         adapter->num_msix_vec = num_vec - adapter->num_msix_roce_vec;
3264
3265         dev_info(dev, "enabled %d MSI-x vector(s) for NIC\n",
3266                  adapter->num_msix_vec);
3267         return 0;
3268
3269 fail:
3270         dev_warn(dev, "MSIx enable failed\n");
3271
3272         /* INTx is not supported in VFs, so fail probe if enable_msix fails */
3273         if (be_virtfn(adapter))
3274                 return num_vec;
3275         return 0;
3276 }
3277
3278 static inline int be_msix_vec_get(struct be_adapter *adapter,
3279                                   struct be_eq_obj *eqo)
3280 {
3281         return adapter->msix_entries[eqo->msix_idx].vector;
3282 }
3283
3284 static int be_msix_register(struct be_adapter *adapter)
3285 {
3286         struct net_device *netdev = adapter->netdev;
3287         struct be_eq_obj *eqo;
3288         int status, i, vec;
3289
3290         for_all_evt_queues(adapter, eqo, i) {
3291                 sprintf(eqo->desc, "%s-q%d", netdev->name, i);
3292                 vec = be_msix_vec_get(adapter, eqo);
3293                 status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
3294                 if (status)
3295                         goto err_msix;
3296
3297                 irq_set_affinity_hint(vec, eqo->affinity_mask);
3298         }
3299
3300         return 0;
3301 err_msix:
3302         for (i--; i >= 0; i--) {
3303                 eqo = &adapter->eq_obj[i];
3304                 free_irq(be_msix_vec_get(adapter, eqo), eqo);
3305         }
3306         dev_warn(&adapter->pdev->dev, "MSIX Request IRQ failed - err %d\n",
3307                  status);
3308         be_msix_disable(adapter);
3309         return status;
3310 }
3311
3312 static int be_irq_register(struct be_adapter *adapter)
3313 {
3314         struct net_device *netdev = adapter->netdev;
3315         int status;
3316
3317         if (msix_enabled(adapter)) {
3318                 status = be_msix_register(adapter);
3319                 if (status == 0)
3320                         goto done;
3321                 /* INTx is not supported for VF */
3322                 if (be_virtfn(adapter))
3323                         return status;
3324         }
3325
3326         /* INTx: only the first EQ is used */
3327         netdev->irq = adapter->pdev->irq;
3328         status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
3329                              &adapter->eq_obj[0]);
3330         if (status) {
3331                 dev_err(&adapter->pdev->dev,
3332                         "INTx request IRQ failed - err %d\n", status);
3333                 return status;
3334         }
3335 done:
3336         adapter->isr_registered = true;
3337         return 0;
3338 }
3339
3340 static void be_irq_unregister(struct be_adapter *adapter)
3341 {
3342         struct net_device *netdev = adapter->netdev;
3343         struct be_eq_obj *eqo;
3344         int i, vec;
3345
3346         if (!adapter->isr_registered)
3347                 return;
3348
3349         /* INTx */
3350         if (!msix_enabled(adapter)) {
3351                 free_irq(netdev->irq, &adapter->eq_obj[0]);
3352                 goto done;
3353         }
3354
3355         /* MSIx */
3356         for_all_evt_queues(adapter, eqo, i) {
3357                 vec = be_msix_vec_get(adapter, eqo);
3358                 irq_set_affinity_hint(vec, NULL);
3359                 free_irq(vec, eqo);
3360         }
3361
3362 done:
3363         adapter->isr_registered = false;
3364 }
3365
3366 static void be_rx_qs_destroy(struct be_adapter *adapter)
3367 {
3368         struct be_queue_info *q;
3369         struct be_rx_obj *rxo;
3370         int i;
3371
3372         for_all_rx_queues(adapter, rxo, i) {
3373                 q = &rxo->q;
3374                 if (q->created) {
3375                         /* If RXQs are destroyed while in an "out of buffer"
3376                          * state, there is a possibility of an HW stall on
3377                          * Lancer. So, post 64 buffers to each queue to relieve
3378                          * the "out of buffer" condition.
3379                          * Make sure there's space in the RXQ before posting.
3380                          */
3381                         if (lancer_chip(adapter)) {
3382                                 be_rx_cq_clean(rxo);
3383                                 if (atomic_read(&q->used) == 0)
3384                                         be_post_rx_frags(rxo, GFP_KERNEL,
3385                                                          MAX_RX_POST);
3386                         }
3387
3388                         be_cmd_rxq_destroy(adapter, q);
3389                         be_rx_cq_clean(rxo);
3390                         be_rxq_clean(rxo);
3391                 }
3392                 be_queue_free(adapter, q);
3393         }
3394 }
3395
3396 static void be_disable_if_filters(struct be_adapter *adapter)
3397 {
3398         be_cmd_pmac_del(adapter, adapter->if_handle,
3399                         adapter->pmac_id[0], 0);
3400
3401         be_clear_uc_list(adapter);
3402
3403         /* The IFACE flags are enabled in the open path and cleared
3404          * in the close path. When a VF gets detached from the host and
3405          * assigned to a VM the following happens:
3406          *      - VF's IFACE flags get cleared in the detach path
3407          *      - IFACE create is issued by the VF in the attach path
3408          * Due to a bug in the BE3/Skyhawk-R FW
3409          * (Lancer FW doesn't have the bug), the IFACE capability flags
3410          * specified along with the IFACE create cmd issued by a VF are not
3411          * honoured by FW.  As a consequence, if a *new* driver
3412          * (that enables/disables IFACE flags in open/close)
3413          * is loaded in the host and an *old* driver is * used by a VM/VF,
3414          * the IFACE gets created *without* the needed flags.
3415          * To avoid this, disable RX-filter flags only for Lancer.
3416          */
3417         if (lancer_chip(adapter)) {
3418                 be_cmd_rx_filter(adapter, BE_IF_ALL_FILT_FLAGS, OFF);
3419                 adapter->if_flags &= ~BE_IF_ALL_FILT_FLAGS;
3420         }
3421 }
3422
3423 static int be_close(struct net_device *netdev)
3424 {
3425         struct be_adapter *adapter = netdev_priv(netdev);
3426         struct be_eq_obj *eqo;
3427         int i;
3428
3429         /* This protection is needed as be_close() may be called even when the
3430          * adapter is in cleared state (after eeh perm failure)
3431          */
3432         if (!(adapter->flags & BE_FLAGS_SETUP_DONE))
3433                 return 0;
3434
3435         be_disable_if_filters(adapter);
3436
3437         if (adapter->flags & BE_FLAGS_NAPI_ENABLED) {
3438                 for_all_evt_queues(adapter, eqo, i) {
3439                         napi_disable(&eqo->napi);
3440                         be_disable_busy_poll(eqo);
3441                 }
3442                 adapter->flags &= ~BE_FLAGS_NAPI_ENABLED;
3443         }
3444
3445         be_async_mcc_disable(adapter);
3446
3447         /* Wait for all pending tx completions to arrive so that
3448          * all tx skbs are freed.
3449          */
3450         netif_tx_disable(netdev);
3451         be_tx_compl_clean(adapter);
3452
3453         be_rx_qs_destroy(adapter);
3454
3455         for_all_evt_queues(adapter, eqo, i) {
3456                 if (msix_enabled(adapter))
3457                         synchronize_irq(be_msix_vec_get(adapter, eqo));
3458                 else
3459                         synchronize_irq(netdev->irq);
3460                 be_eq_clean(eqo);
3461         }
3462
3463         be_irq_unregister(adapter);
3464
3465         return 0;
3466 }
3467
3468 static int be_rx_qs_create(struct be_adapter *adapter)
3469 {
3470         struct rss_info *rss = &adapter->rss_info;
3471         u8 rss_key[RSS_HASH_KEY_LEN];
3472         struct be_rx_obj *rxo;
3473         int rc, i, j;
3474
3475         for_all_rx_queues(adapter, rxo, i) {
3476                 rc = be_queue_alloc(adapter, &rxo->q, RX_Q_LEN,
3477                                     sizeof(struct be_eth_rx_d));
3478                 if (rc)
3479                         return rc;
3480         }
3481
3482         if (adapter->need_def_rxq || !adapter->num_rss_qs) {
3483                 rxo = default_rxo(adapter);
3484                 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3485                                        rx_frag_size, adapter->if_handle,
3486                                        false, &rxo->rss_id);
3487                 if (rc)
3488                         return rc;
3489         }
3490
3491         for_all_rss_queues(adapter, rxo, i) {
3492                 rc = be_cmd_rxq_create(adapter, &rxo->q, rxo->cq.id,
3493                                        rx_frag_size, adapter->if_handle,
3494                                        true, &rxo->rss_id);
3495                 if (rc)
3496                         return rc;
3497         }
3498
3499         if (be_multi_rxq(adapter)) {
3500                 for (j = 0; j < RSS_INDIR_TABLE_LEN; j += adapter->num_rss_qs) {
3501                         for_all_rss_queues(adapter, rxo, i) {
3502                                 if ((j + i) >= RSS_INDIR_TABLE_LEN)
3503                                         break;
3504                                 rss->rsstable[j + i] = rxo->rss_id;
3505                                 rss->rss_queue[j + i] = i;
3506                         }
3507                 }
3508                 rss->rss_flags = RSS_ENABLE_TCP_IPV4 | RSS_ENABLE_IPV4 |
3509                         RSS_ENABLE_TCP_IPV6 | RSS_ENABLE_IPV6;
3510
3511                 if (!BEx_chip(adapter))
3512                         rss->rss_flags |= RSS_ENABLE_UDP_IPV4 |
3513                                 RSS_ENABLE_UDP_IPV6;
3514         } else {
3515                 /* Disable RSS, if only default RX Q is created */
3516                 rss->rss_flags = RSS_ENABLE_NONE;
3517         }
3518
3519         netdev_rss_key_fill(rss_key, RSS_HASH_KEY_LEN);
3520         rc = be_cmd_rss_config(adapter, rss->rsstable, rss->rss_flags,
3521                                RSS_INDIR_TABLE_LEN, rss_key);
3522         if (rc) {
3523                 rss->rss_flags = RSS_ENABLE_NONE;
3524                 return rc;
3525         }
3526
3527         memcpy(rss->rss_hkey, rss_key, RSS_HASH_KEY_LEN);
3528
3529         /* Post 1 less than RXQ-len to avoid head being equal to tail,
3530          * which is a queue empty condition
3531          */
3532         for_all_rx_queues(adapter, rxo, i)
3533                 be_post_rx_frags(rxo, GFP_KERNEL, RX_Q_LEN - 1);
3534
3535         return 0;
3536 }
3537
3538 static int be_enable_if_filters(struct be_adapter *adapter)
3539 {
3540         int status;
3541
3542         status = be_cmd_rx_filter(adapter, BE_IF_EN_FLAGS, ON);
3543         if (status)
3544                 return status;
3545
3546         /* For BE3 VFs, the PF programs the initial MAC address */
3547         if (!(BEx_chip(adapter) && be_virtfn(adapter))) {
3548                 status = be_cmd_pmac_add(adapter, adapter->netdev->dev_addr,
3549                                          adapter->if_handle,
3550                                          &adapter->pmac_id[0], 0);
3551                 if (status)
3552                         return status;
3553         }
3554
3555         if (adapter->vlans_added)
3556                 be_vid_config(adapter);
3557
3558         be_set_rx_mode(adapter->netdev);
3559
3560         return 0;
3561 }
3562
3563 static int be_open(struct net_device *netdev)
3564 {
3565         struct be_adapter *adapter = netdev_priv(netdev);
3566         struct be_eq_obj *eqo;
3567         struct be_rx_obj *rxo;
3568         struct be_tx_obj *txo;
3569         u8 link_status;
3570         int status, i;
3571
3572         status = be_rx_qs_create(adapter);
3573         if (status)
3574                 goto err;
3575
3576         status = be_enable_if_filters(adapter);
3577         if (status)
3578                 goto err;
3579
3580         status = be_irq_register(adapter);
3581         if (status)
3582                 goto err;
3583
3584         for_all_rx_queues(adapter, rxo, i)
3585                 be_cq_notify(adapter, rxo->cq.id, true, 0);
3586
3587         for_all_tx_queues(adapter, txo, i)
3588                 be_cq_notify(adapter, txo->cq.id, true, 0);
3589
3590         be_async_mcc_enable(adapter);
3591
3592         for_all_evt_queues(adapter, eqo, i) {
3593                 napi_enable(&eqo->napi);
3594                 be_enable_busy_poll(eqo);
3595                 be_eq_notify(adapter, eqo->q.id, true, true, 0, 0);
3596         }
3597         adapter->flags |= BE_FLAGS_NAPI_ENABLED;
3598
3599         status = be_cmd_link_status_query(adapter, NULL, &link_status, 0);
3600         if (!status)
3601                 be_link_status_update(adapter, link_status);
3602
3603         netif_tx_start_all_queues(netdev);
3604 #ifdef CONFIG_BE2NET_VXLAN
3605         if (skyhawk_chip(adapter))
3606                 vxlan_get_rx_port(netdev);
3607 #endif
3608
3609         return 0;
3610 err:
3611         be_close(adapter->netdev);
3612         return -EIO;
3613 }
3614
3615 static int be_setup_wol(struct be_adapter *adapter, bool enable)
3616 {
3617         struct device *dev = &adapter->pdev->dev;
3618         struct be_dma_mem cmd;
3619         u8 mac[ETH_ALEN];
3620         int status;
3621
3622         eth_zero_addr(mac);
3623
3624         cmd.size = sizeof(struct be_cmd_req_acpi_wol_magic_config);
3625         cmd.va = dma_zalloc_coherent(dev, cmd.size, &cmd.dma, GFP_KERNEL);
3626         if (!cmd.va)
3627                 return -ENOMEM;
3628
3629         if (enable) {
3630                 status = pci_write_config_dword(adapter->pdev,
3631                                                 PCICFG_PM_CONTROL_OFFSET,
3632                                                 PCICFG_PM_CONTROL_MASK);
3633                 if (status) {
3634                         dev_err(dev, "Could not enable Wake-on-lan\n");
3635                         goto err;
3636                 }
3637         } else {
3638                 ether_addr_copy(mac, adapter->netdev->dev_addr);
3639         }
3640
3641         status = be_cmd_enable_magic_wol(adapter, mac, &cmd);
3642         pci_enable_wake(adapter->pdev, PCI_D3hot, enable);
3643         pci_enable_wake(adapter->pdev, PCI_D3cold, enable);
3644 err:
3645         dma_free_coherent(dev, cmd.size, cmd.va, cmd.dma);
3646         return status;
3647 }
3648
3649 static void be_vf_eth_addr_generate(struct be_adapter *adapter, u8 *mac)
3650 {
3651         u32 addr;
3652
3653         addr = jhash(adapter->netdev->dev_addr, ETH_ALEN, 0);
3654
3655         mac[5] = (u8)(addr & 0xFF);
3656         mac[4] = (u8)((addr >> 8) & 0xFF);
3657         mac[3] = (u8)((addr >> 16) & 0xFF);
3658         /* Use the OUI from the current MAC address */
3659         memcpy(mac, adapter->netdev->dev_addr, 3);
3660 }
3661
3662 /*
3663  * Generate a seed MAC address from the PF MAC Address using jhash.
3664  * MAC Address for VFs are assigned incrementally starting from the seed.
3665  * These addresses are programmed in the ASIC by the PF and the VF driver
3666  * queries for the MAC address during its probe.
3667  */
3668 static int be_vf_eth_addr_config(struct be_adapter *adapter)
3669 {
3670         u32 vf;
3671         int status = 0;
3672         u8 mac[ETH_ALEN];
3673         struct be_vf_cfg *vf_cfg;
3674
3675         be_vf_eth_addr_generate(adapter, mac);
3676
3677         for_all_vfs(adapter, vf_cfg, vf) {
3678                 if (BEx_chip(adapter))
3679                         status = be_cmd_pmac_add(adapter, mac,
3680                                                  vf_cfg->if_handle,
3681                                                  &vf_cfg->pmac_id, vf + 1);
3682                 else
3683                         status = be_cmd_set_mac(adapter, mac, vf_cfg->if_handle,
3684                                                 vf + 1);
3685
3686                 if (status)
3687                         dev_err(&adapter->pdev->dev,
3688                                 "Mac address assignment failed for VF %d\n",
3689                                 vf);
3690                 else
3691                         memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3692
3693                 mac[5] += 1;
3694         }
3695         return status;
3696 }
3697
3698 static int be_vfs_mac_query(struct be_adapter *adapter)
3699 {
3700         int status, vf;
3701         u8 mac[ETH_ALEN];
3702         struct be_vf_cfg *vf_cfg;
3703
3704         for_all_vfs(adapter, vf_cfg, vf) {
3705                 status = be_cmd_get_active_mac(adapter, vf_cfg->pmac_id,
3706                                                mac, vf_cfg->if_handle,
3707                                                false, vf+1);
3708                 if (status)
3709                         return status;
3710                 memcpy(vf_cfg->mac_addr, mac, ETH_ALEN);
3711         }
3712         return 0;
3713 }
3714
3715 static void be_vf_clear(struct be_adapter *adapter)
3716 {
3717         struct be_vf_cfg *vf_cfg;
3718         u32 vf;
3719
3720         if (pci_vfs_assigned(adapter->pdev)) {
3721                 dev_warn(&adapter->pdev->dev,
3722                          "VFs are assigned to VMs: not disabling VFs\n");
3723                 goto done;
3724         }
3725
3726         pci_disable_sriov(adapter->pdev);
3727
3728         for_all_vfs(adapter, vf_cfg, vf) {
3729                 if (BEx_chip(adapter))
3730                         be_cmd_pmac_del(adapter, vf_cfg->if_handle,
3731                                         vf_cfg->pmac_id, vf + 1);
3732                 else
3733                         be_cmd_set_mac(adapter, NULL, vf_cfg->if_handle,
3734                                        vf + 1);
3735
3736                 be_cmd_if_destroy(adapter, vf_cfg->if_handle, vf + 1);
3737         }
3738 done:
3739         kfree(adapter->vf_cfg);
3740         adapter->num_vfs = 0;
3741         adapter->flags &= ~BE_FLAGS_SRIOV_ENABLED;
3742 }
3743
3744 static void be_clear_queues(struct be_adapter *adapter)
3745 {
3746         be_mcc_queues_destroy(adapter);
3747         be_rx_cqs_destroy(adapter);
3748         be_tx_queues_destroy(adapter);
3749         be_evt_queues_destroy(adapter);
3750 }
3751
3752 static void be_cancel_worker(struct be_adapter *adapter)
3753 {
3754         if (adapter->flags & BE_FLAGS_WORKER_SCHEDULED) {
3755                 cancel_delayed_work_sync(&adapter->work);
3756                 adapter->flags &= ~BE_FLAGS_WORKER_SCHEDULED;
3757         }
3758 }
3759
3760 static void be_cancel_err_detection(struct be_adapter *adapter)
3761 {
3762         if (adapter->flags & BE_FLAGS_ERR_DETECTION_SCHEDULED) {
3763                 cancel_delayed_work_sync(&adapter->be_err_detection_work);
3764                 adapter->flags &= ~BE_FLAGS_ERR_DETECTION_SCHEDULED;
3765         }
3766 }
3767
3768 #ifdef CONFIG_BE2NET_VXLAN
3769 static void be_disable_vxlan_offloads(struct be_adapter *adapter)
3770 {
3771         struct net_device *netdev = adapter->netdev;
3772
3773         if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS)
3774                 be_cmd_manage_iface(adapter, adapter->if_handle,
3775                                     OP_CONVERT_TUNNEL_TO_NORMAL);
3776
3777         if (adapter->vxlan_port)
3778                 be_cmd_set_vxlan_port(adapter, 0);
3779
3780         adapter->flags &= ~BE_FLAGS_VXLAN_OFFLOADS;
3781         adapter->vxlan_port = 0;
3782
3783         netdev->hw_enc_features = 0;
3784         netdev->hw_features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3785         netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL);
3786 }
3787 #endif
3788
3789 static u16 be_calculate_vf_qs(struct be_adapter *adapter, u16 num_vfs)
3790 {
3791         struct be_resources res = adapter->pool_res;
3792         u16 num_vf_qs = 1;
3793
3794         /* Distribute the queue resources among the PF and it's VFs
3795          * Do not distribute queue resources in multi-channel configuration.
3796          */
3797         if (num_vfs && !be_is_mc(adapter)) {
3798                  /* Divide the qpairs evenly among the VFs and the PF, capped
3799                   * at VF-EQ-count. Any remainder qpairs belong to the PF.
3800                   */
3801                 num_vf_qs = min(SH_VF_MAX_NIC_EQS,
3802                                 res.max_rss_qs / (num_vfs + 1));
3803
3804                 /* Skyhawk-R chip supports only MAX_RSS_IFACES RSS capable
3805                  * interfaces per port. Provide RSS on VFs, only if number
3806                  * of VFs requested is less than MAX_RSS_IFACES limit.
3807                  */
3808                 if (num_vfs >= MAX_RSS_IFACES)
3809                         num_vf_qs = 1;
3810         }
3811         return num_vf_qs;
3812 }
3813
3814 static int be_clear(struct be_adapter *adapter)
3815 {
3816         struct pci_dev *pdev = adapter->pdev;
3817         u16 num_vf_qs;
3818
3819         be_cancel_worker(adapter);
3820
3821         if (sriov_enabled(adapter))
3822                 be_vf_clear(adapter);
3823
3824         /* Re-configure FW to distribute resources evenly across max-supported
3825          * number of VFs, only when VFs are not already enabled.
3826          */
3827         if (skyhawk_chip(adapter) && be_physfn(adapter) &&
3828             !pci_vfs_assigned(pdev)) {
3829                 num_vf_qs = be_calculate_vf_qs(adapter,
3830                                                pci_sriov_get_totalvfs(pdev));
3831                 be_cmd_set_sriov_config(adapter, adapter->pool_res,
3832                                         pci_sriov_get_totalvfs(pdev),
3833                                         num_vf_qs);
3834         }
3835
3836 #ifdef CONFIG_BE2NET_VXLAN
3837         be_disable_vxlan_offloads(adapter);
3838 #endif
3839         kfree(adapter->pmac_id);
3840         adapter->pmac_id = NULL;
3841
3842         be_cmd_if_destroy(adapter, adapter->if_handle,  0);
3843
3844         be_clear_queues(adapter);
3845
3846         be_msix_disable(adapter);
3847         adapter->flags &= ~BE_FLAGS_SETUP_DONE;
3848         return 0;
3849 }
3850
3851 static int be_vfs_if_create(struct be_adapter *adapter)
3852 {
3853         struct be_resources res = {0};
3854         u32 cap_flags, en_flags, vf;
3855         struct be_vf_cfg *vf_cfg;
3856         int status;
3857
3858         /* If a FW profile exists, then cap_flags are updated */
3859         cap_flags = BE_IF_FLAGS_UNTAGGED | BE_IF_FLAGS_BROADCAST |
3860                     BE_IF_FLAGS_MULTICAST | BE_IF_FLAGS_PASS_L3L4_ERRORS;
3861
3862         for_all_vfs(adapter, vf_cfg, vf) {
3863                 if (!BE3_chip(adapter)) {
3864                         status = be_cmd_get_profile_config(adapter, &res,
3865                                                            RESOURCE_LIMITS,
3866                                                            vf + 1);
3867                         if (!status) {
3868                                 cap_flags = res.if_cap_flags;
3869                                 /* Prevent VFs from enabling VLAN promiscuous
3870                                  * mode
3871                                  */
3872                                 cap_flags &= ~BE_IF_FLAGS_VLAN_PROMISCUOUS;
3873                         }
3874                 }
3875
3876                 en_flags = cap_flags & (BE_IF_FLAGS_UNTAGGED |
3877                                         BE_IF_FLAGS_BROADCAST |
3878                                         BE_IF_FLAGS_MULTICAST |
3879                                         BE_IF_FLAGS_PASS_L3L4_ERRORS);
3880                 status = be_cmd_if_create(adapter, cap_flags, en_flags,
3881                                           &vf_cfg->if_handle, vf + 1);
3882                 if (status)
3883                         return status;
3884         }
3885
3886         return 0;
3887 }
3888
3889 static int be_vf_setup_init(struct be_adapter *adapter)
3890 {
3891         struct be_vf_cfg *vf_cfg;
3892         int vf;
3893
3894         adapter->vf_cfg = kcalloc(adapter->num_vfs, sizeof(*vf_cfg),
3895                                   GFP_KERNEL);
3896         if (!adapter->vf_cfg)
3897                 return -ENOMEM;
3898
3899         for_all_vfs(adapter, vf_cfg, vf) {
3900                 vf_cfg->if_handle = -1;
3901                 vf_cfg->pmac_id = -1;
3902         }
3903         return 0;
3904 }
3905
3906 static int be_vf_setup(struct be_adapter *adapter)
3907 {
3908         struct device *dev = &adapter->pdev->dev;
3909         struct be_vf_cfg *vf_cfg;
3910         int status, old_vfs, vf;
3911         bool spoofchk;
3912
3913         old_vfs = pci_num_vf(adapter->pdev);
3914
3915         status = be_vf_setup_init(adapter);
3916         if (status)
3917                 goto err;
3918
3919         if (old_vfs) {
3920                 for_all_vfs(adapter, vf_cfg, vf) {
3921                         status = be_cmd_get_if_id(adapter, vf_cfg, vf);
3922                         if (status)
3923                                 goto err;
3924                 }
3925
3926                 status = be_vfs_mac_query(adapter);
3927                 if (status)
3928                         goto err;
3929         } else {
3930                 status = be_vfs_if_create(adapter);
3931                 if (status)
3932                         goto err;
3933
3934                 status = be_vf_eth_addr_config(adapter);
3935                 if (status)
3936                         goto err;
3937         }
3938
3939         for_all_vfs(adapter, vf_cfg, vf) {
3940                 /* Allow VFs to programs MAC/VLAN filters */
3941                 status = be_cmd_get_fn_privileges(adapter, &vf_cfg->privileges,
3942                                                   vf + 1);
3943                 if (!status && !(vf_cfg->privileges & BE_PRIV_FILTMGMT)) {
3944                         status = be_cmd_set_fn_privileges(adapter,
3945                                                           vf_cfg->privileges |
3946                                                           BE_PRIV_FILTMGMT,
3947                                                           vf + 1);
3948                         if (!status) {
3949                                 vf_cfg->privileges |= BE_PRIV_FILTMGMT;
3950                                 dev_info(dev, "VF%d has FILTMGMT privilege\n",
3951                                          vf);
3952                         }
3953                 }
3954
3955                 /* Allow full available bandwidth */
3956                 if (!old_vfs)
3957                         be_cmd_config_qos(adapter, 0, 0, vf + 1);
3958
3959                 status = be_cmd_get_hsw_config(adapter, NULL, vf + 1,
3960                                                vf_cfg->if_handle, NULL,
3961                                                &spoofchk);
3962                 if (!status)
3963                         vf_cfg->spoofchk = spoofchk;
3964
3965                 if (!old_vfs) {
3966                         be_cmd_enable_vf(adapter, vf + 1);
3967                         be_cmd_set_logical_link_config(adapter,
3968                                                        IFLA_VF_LINK_STATE_AUTO,
3969                                                        vf+1);
3970                 }
3971         }
3972
3973         if (!old_vfs) {
3974                 status = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
3975                 if (status) {
3976                         dev_err(dev, "SRIOV enable failed\n");
3977                         adapter->num_vfs = 0;
3978                         goto err;
3979                 }
3980         }
3981
3982         adapter->flags |= BE_FLAGS_SRIOV_ENABLED;
3983         return 0;
3984 err:
3985         dev_err(dev, "VF setup failed\n");
3986         be_vf_clear(adapter);
3987         return status;
3988 }
3989
3990 /* Converting function_mode bits on BE3 to SH mc_type enums */
3991
3992 static u8 be_convert_mc_type(u32 function_mode)
3993 {
3994         if (function_mode & VNIC_MODE && function_mode & QNQ_MODE)
3995                 return vNIC1;
3996         else if (function_mode & QNQ_MODE)
3997                 return FLEX10;
3998         else if (function_mode & VNIC_MODE)
3999                 return vNIC2;
4000         else if (function_mode & UMC_ENABLED)
4001                 return UMC;
4002         else
4003                 return MC_NONE;
4004 }
4005
4006 /* On BE2/BE3 FW does not suggest the supported limits */
4007 static void BEx_get_resources(struct be_adapter *adapter,
4008                               struct be_resources *res)
4009 {
4010         bool use_sriov = adapter->num_vfs ? 1 : 0;
4011
4012         if (be_physfn(adapter))
4013                 res->max_uc_mac = BE_UC_PMAC_COUNT;
4014         else
4015                 res->max_uc_mac = BE_VF_UC_PMAC_COUNT;
4016
4017         adapter->mc_type = be_convert_mc_type(adapter->function_mode);
4018
4019         if (be_is_mc(adapter)) {
4020                 /* Assuming that there are 4 channels per port,
4021                  * when multi-channel is enabled
4022                  */
4023                 if (be_is_qnq_mode(adapter))
4024                         res->max_vlans = BE_NUM_VLANS_SUPPORTED/8;
4025                 else
4026                         /* In a non-qnq multichannel mode, the pvid
4027                          * takes up one vlan entry
4028                          */
4029                         res->max_vlans = (BE_NUM_VLANS_SUPPORTED / 4) - 1;
4030         } else {
4031                 res->max_vlans = BE_NUM_VLANS_SUPPORTED;
4032         }
4033
4034         res->max_mcast_mac = BE_MAX_MC;
4035
4036         /* 1) For BE3 1Gb ports, FW does not support multiple TXQs
4037          * 2) Create multiple TX rings on a BE3-R multi-channel interface
4038          *    *only* if it is RSS-capable.
4039          */
4040         if (BE2_chip(adapter) || use_sriov ||  (adapter->port_num > 1) ||
4041             be_virtfn(adapter) ||
4042             (be_is_mc(adapter) &&
4043              !(adapter->function_caps & BE_FUNCTION_CAPS_RSS))) {
4044                 res->max_tx_qs = 1;
4045         } else if (adapter->function_caps & BE_FUNCTION_CAPS_SUPER_NIC) {
4046                 struct be_resources super_nic_res = {0};
4047
4048                 /* On a SuperNIC profile, the driver needs to use the
4049                  * GET_PROFILE_CONFIG cmd to query the per-function TXQ limits
4050                  */
4051                 be_cmd_get_profile_config(adapter, &super_nic_res,
4052                                           RESOURCE_LIMITS, 0);
4053                 /* Some old versions of BE3 FW don't report max_tx_qs value */
4054                 res->max_tx_qs = super_nic_res.max_tx_qs ? : BE3_MAX_TX_QS;
4055         } else {
4056                 res->max_tx_qs = BE3_MAX_TX_QS;
4057         }
4058
4059         if ((adapter->function_caps & BE_FUNCTION_CAPS_RSS) &&
4060             !use_sriov && be_physfn(adapter))
4061                 res->max_rss_qs = (adapter->be3_native) ?
4062                                            BE3_MAX_RSS_QS : BE2_MAX_RSS_QS;
4063         res->max_rx_qs = res->max_rss_qs + 1;
4064
4065         if (be_physfn(adapter))
4066                 res->max_evt_qs = (be_max_vfs(adapter) > 0) ?
4067                                         BE3_SRIOV_MAX_EVT_QS : BE3_MAX_EVT_QS;
4068         else
4069                 res->max_evt_qs = 1;
4070
4071         res->if_cap_flags = BE_IF_CAP_FLAGS_WANT;
4072         res->if_cap_flags &= ~BE_IF_FLAGS_DEFQ_RSS;
4073         if (!(adapter->function_caps & BE_FUNCTION_CAPS_RSS))
4074                 res->if_cap_flags &= ~BE_IF_FLAGS_RSS;
4075 }
4076
4077 static void be_setup_init(struct be_adapter *adapter)
4078 {
4079         adapter->vlan_prio_bmap = 0xff;
4080         adapter->phy.link_speed = -1;
4081         adapter->if_handle = -1;
4082         adapter->be3_native = false;
4083         adapter->if_flags = 0;
4084         if (be_physfn(adapter))
4085                 adapter->cmd_privileges = MAX_PRIVILEGES;
4086         else
4087                 adapter->cmd_privileges = MIN_PRIVILEGES;
4088 }
4089
4090 static int be_get_sriov_config(struct be_adapter *adapter)
4091 {
4092         struct be_resources res = {0};
4093         int max_vfs, old_vfs;
4094
4095         be_cmd_get_profile_config(adapter, &res, RESOURCE_LIMITS, 0);
4096
4097         /* Some old versions of BE3 FW don't report max_vfs value */
4098         if (BE3_chip(adapter) && !res.max_vfs) {
4099                 max_vfs = pci_sriov_get_totalvfs(adapter->pdev);
4100                 res.max_vfs = max_vfs > 0 ? min(MAX_VFS, max_vfs) : 0;
4101         }
4102
4103         adapter->pool_res = res;
4104
4105         /* If during previous unload of the driver, the VFs were not disabled,
4106          * then we cannot rely on the PF POOL limits for the TotalVFs value.
4107          * Instead use the TotalVFs value stored in the pci-dev struct.
4108          */
4109         old_vfs = pci_num_vf(adapter->pdev);
4110         if (old_vfs) {
4111                 dev_info(&adapter->pdev->dev, "%d VFs are already enabled\n",
4112                          old_vfs);
4113
4114                 adapter->pool_res.max_vfs =
4115                         pci_sriov_get_totalvfs(adapter->pdev);
4116                 adapter->num_vfs = old_vfs;
4117         }
4118
4119         return 0;
4120 }
4121
4122 static void be_alloc_sriov_res(struct be_adapter *adapter)
4123 {
4124         int old_vfs = pci_num_vf(adapter->pdev);
4125         u16 num_vf_qs;
4126         int status;
4127
4128         be_get_sriov_config(adapter);
4129
4130         if (!old_vfs)
4131                 pci_sriov_set_totalvfs(adapter->pdev, be_max_vfs(adapter));
4132
4133         /* When the HW is in SRIOV capable configuration, the PF-pool
4134          * resources are given to PF during driver load, if there are no
4135          * old VFs. This facility is not available in BE3 FW.
4136          * Also, this is done by FW in Lancer chip.
4137          */
4138         if (skyhawk_chip(adapter) && be_max_vfs(adapter) && !old_vfs) {
4139                 num_vf_qs = be_calculate_vf_qs(adapter, 0);
4140                 status = be_cmd_set_sriov_config(adapter, adapter->pool_res, 0,
4141                                                  num_vf_qs);
4142                 if (status)
4143                         dev_err(&adapter->pdev->dev,
4144                                 "Failed to optimize SRIOV resources\n");
4145         }
4146 }
4147
4148 static int be_get_resources(struct be_adapter *adapter)
4149 {
4150         struct device *dev = &adapter->pdev->dev;
4151         struct be_resources res = {0};
4152         int status;
4153
4154         if (BEx_chip(adapter)) {
4155                 BEx_get_resources(adapter, &res);
4156                 adapter->res = res;
4157         }
4158
4159         /* For Lancer, SH etc read per-function resource limits from FW.
4160          * GET_FUNC_CONFIG returns per function guaranteed limits.
4161          * GET_PROFILE_CONFIG returns PCI-E related limits PF-pool limits
4162          */
4163         if (!BEx_chip(adapter)) {
4164                 status = be_cmd_get_func_config(adapter, &res);
4165                 if (status)
4166                         return status;
4167
4168                 /* If a deafault RXQ must be created, we'll use up one RSSQ*/
4169                 if (res.max_rss_qs && res.max_rss_qs == res.max_rx_qs &&
4170                     !(res.if_cap_flags & BE_IF_FLAGS_DEFQ_RSS))
4171                         res.max_rss_qs -= 1;
4172
4173                 /* If RoCE may be enabled stash away half the EQs for RoCE */
4174                 if (be_roce_supported(adapter))
4175                         res.max_evt_qs /= 2;
4176                 adapter->res = res;
4177         }
4178
4179         /* If FW supports RSS default queue, then skip creating non-RSS
4180          * queue for non-IP traffic.
4181          */
4182         adapter->need_def_rxq = (be_if_cap_flags(adapter) &
4183                                  BE_IF_FLAGS_DEFQ_RSS) ? 0 : 1;
4184
4185         dev_info(dev, "Max: txqs %d, rxqs %d, rss %d, eqs %d, vfs %d\n",
4186                  be_max_txqs(adapter), be_max_rxqs(adapter),
4187                  be_max_rss(adapter), be_max_eqs(adapter),
4188                  be_max_vfs(adapter));
4189         dev_info(dev, "Max: uc-macs %d, mc-macs %d, vlans %d\n",
4190                  be_max_uc(adapter), be_max_mc(adapter),
4191                  be_max_vlans(adapter));
4192
4193         /* Sanitize cfg_num_qs based on HW and platform limits */
4194         adapter->cfg_num_qs = min_t(u16, netif_get_num_default_rss_queues(),
4195                                     be_max_qs(adapter));
4196         return 0;
4197 }
4198
4199 static int be_get_config(struct be_adapter *adapter)
4200 {
4201         int status, level;
4202         u16 profile_id;
4203
4204         status = be_cmd_get_cntl_attributes(adapter);
4205         if (status)
4206                 return status;
4207
4208         status = be_cmd_query_fw_cfg(adapter);
4209         if (status)
4210                 return status;
4211
4212         if (!lancer_chip(adapter) && be_physfn(adapter))
4213                 be_cmd_get_fat_dump_len(adapter, &adapter->fat_dump_len);
4214
4215         if (BEx_chip(adapter)) {
4216                 level = be_cmd_get_fw_log_level(adapter);
4217                 adapter->msg_enable =
4218                         level <= FW_LOG_LEVEL_DEFAULT ? NETIF_MSG_HW : 0;
4219         }
4220
4221         be_cmd_get_acpi_wol_cap(adapter);
4222
4223         be_cmd_query_port_name(adapter);
4224
4225         if (be_physfn(adapter)) {
4226                 status = be_cmd_get_active_profile(adapter, &profile_id);
4227                 if (!status)
4228                         dev_info(&adapter->pdev->dev,
4229                                  "Using profile 0x%x\n", profile_id);
4230         }
4231
4232         status = be_get_resources(adapter);
4233         if (status)
4234                 return status;
4235
4236         adapter->pmac_id = kcalloc(be_max_uc(adapter),
4237                                    sizeof(*adapter->pmac_id), GFP_KERNEL);
4238         if (!adapter->pmac_id)
4239                 return -ENOMEM;
4240
4241         return 0;
4242 }
4243
4244 static int be_mac_setup(struct be_adapter *adapter)
4245 {
4246         u8 mac[ETH_ALEN];
4247         int status;
4248
4249         if (is_zero_ether_addr(adapter->netdev->dev_addr)) {
4250                 status = be_cmd_get_perm_mac(adapter, mac);
4251                 if (status)
4252                         return status;
4253
4254                 memcpy(adapter->netdev->dev_addr, mac, ETH_ALEN);
4255                 memcpy(adapter->netdev->perm_addr, mac, ETH_ALEN);
4256         }
4257
4258         return 0;
4259 }
4260
4261 static void be_schedule_worker(struct be_adapter *adapter)
4262 {
4263         schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
4264         adapter->flags |= BE_FLAGS_WORKER_SCHEDULED;
4265 }
4266
4267 static void be_schedule_err_detection(struct be_adapter *adapter, u32 delay)
4268 {
4269         schedule_delayed_work(&adapter->be_err_detection_work,
4270                               msecs_to_jiffies(delay));
4271         adapter->flags |= BE_FLAGS_ERR_DETECTION_SCHEDULED;
4272 }
4273
4274 static int be_setup_queues(struct be_adapter *adapter)
4275 {
4276         struct net_device *netdev = adapter->netdev;
4277         int status;
4278
4279         status = be_evt_queues_create(adapter);
4280         if (status)
4281                 goto err;
4282
4283         status = be_tx_qs_create(adapter);
4284         if (status)
4285                 goto err;
4286
4287         status = be_rx_cqs_create(adapter);
4288         if (status)
4289                 goto err;
4290
4291         status = be_mcc_queues_create(adapter);
4292         if (status)
4293                 goto err;
4294
4295         status = netif_set_real_num_rx_queues(netdev, adapter->num_rx_qs);
4296         if (status)
4297                 goto err;
4298
4299         status = netif_set_real_num_tx_queues(netdev, adapter->num_tx_qs);
4300         if (status)
4301                 goto err;
4302
4303         return 0;
4304 err:
4305         dev_err(&adapter->pdev->dev, "queue_setup failed\n");
4306         return status;
4307 }
4308
4309 int be_update_queues(struct be_adapter *adapter)
4310 {
4311         struct net_device *netdev = adapter->netdev;
4312         int status;
4313
4314         if (netif_running(netdev))
4315                 be_close(netdev);
4316
4317         be_cancel_worker(adapter);
4318
4319         /* If any vectors have been shared with RoCE we cannot re-program
4320          * the MSIx table.
4321          */
4322         if (!adapter->num_msix_roce_vec)
4323                 be_msix_disable(adapter);
4324
4325         be_clear_queues(adapter);
4326
4327         if (!msix_enabled(adapter)) {
4328                 status = be_msix_enable(adapter);
4329                 if (status)
4330                         return status;
4331         }
4332
4333         status = be_setup_queues(adapter);
4334         if (status)
4335                 return status;
4336
4337         be_schedule_worker(adapter);
4338
4339         if (netif_running(netdev))
4340                 status = be_open(netdev);
4341
4342         return status;
4343 }
4344
4345 static inline int fw_major_num(const char *fw_ver)
4346 {
4347         int fw_major = 0, i;
4348
4349         i = sscanf(fw_ver, "%d.", &fw_major);
4350         if (i != 1)
4351                 return 0;
4352
4353         return fw_major;
4354 }
4355
4356 /* If any VFs are already enabled don't FLR the PF */
4357 static bool be_reset_required(struct be_adapter *adapter)
4358 {
4359         return pci_num_vf(adapter->pdev) ? false : true;
4360 }
4361
4362 /* Wait for the FW to be ready and perform the required initialization */
4363 static int be_func_init(struct be_adapter *adapter)
4364 {
4365         int status;
4366
4367         status = be_fw_wait_ready(adapter);
4368         if (status)
4369                 return status;
4370
4371         if (be_reset_required(adapter)) {
4372                 status = be_cmd_reset_function(adapter);
4373                 if (status)
4374                         return status;
4375
4376                 /* Wait for interrupts to quiesce after an FLR */
4377                 msleep(100);
4378
4379                 /* We can clear all errors when function reset succeeds */
4380                 be_clear_error(adapter, BE_CLEAR_ALL);
4381         }
4382
4383         /* Tell FW we're ready to fire cmds */
4384         status = be_cmd_fw_init(adapter);
4385         if (status)
4386                 return status;
4387
4388         /* Allow interrupts for other ULPs running on NIC function */
4389         be_intr_set(adapter, true);
4390
4391         return 0;
4392 }
4393
4394 static int be_setup(struct be_adapter *adapter)
4395 {
4396         struct device *dev = &adapter->pdev->dev;
4397         u32 en_flags;
4398         int status;
4399
4400         status = be_func_init(adapter);
4401         if (status)
4402                 return status;
4403
4404         be_setup_init(adapter);
4405
4406         if (!lancer_chip(adapter))
4407                 be_cmd_req_native_mode(adapter);
4408
4409         /* invoke this cmd first to get pf_num and vf_num which are needed
4410          * for issuing profile related cmds
4411          */
4412         if (!BEx_chip(adapter)) {
4413                 status = be_cmd_get_func_config(adapter, NULL);
4414                 if (status)
4415                         return status;
4416         }
4417
4418         if (!BE2_chip(adapter) && be_physfn(adapter))
4419                 be_alloc_sriov_res(adapter);
4420
4421         status = be_get_config(adapter);
4422         if (status)
4423                 goto err;
4424
4425         status = be_msix_enable(adapter);
4426         if (status)
4427                 goto err;
4428
4429         /* will enable all the needed filter flags in be_open() */
4430         en_flags = BE_IF_FLAGS_RSS | BE_IF_FLAGS_DEFQ_RSS;
4431         en_flags = en_flags & be_if_cap_flags(adapter);
4432         status = be_cmd_if_create(adapter, be_if_cap_flags(adapter), en_flags,
4433                                   &adapter->if_handle, 0);
4434         if (status)
4435                 goto err;
4436
4437         /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
4438         rtnl_lock();
4439         status = be_setup_queues(adapter);
4440         rtnl_unlock();
4441         if (status)
4442                 goto err;
4443
4444         be_cmd_get_fn_privileges(adapter, &adapter->cmd_privileges, 0);
4445
4446         status = be_mac_setup(adapter);
4447         if (status)
4448                 goto err;
4449
4450         be_cmd_get_fw_ver(adapter);
4451         dev_info(dev, "FW version is %s\n", adapter->fw_ver);
4452
4453         if (BE2_chip(adapter) && fw_major_num(adapter->fw_ver) < 4) {
4454                 dev_err(dev, "Firmware on card is old(%s), IRQs may not work",
4455                         adapter->fw_ver);
4456                 dev_err(dev, "Please upgrade firmware to version >= 4.0\n");
4457         }
4458
4459         status = be_cmd_set_flow_control(adapter, adapter->tx_fc,
4460                                          adapter->rx_fc);
4461         if (status)
4462                 be_cmd_get_flow_control(adapter, &adapter->tx_fc,
4463                                         &adapter->rx_fc);
4464
4465         dev_info(&adapter->pdev->dev, "HW Flow control - TX:%d RX:%d\n",
4466                  adapter->tx_fc, adapter->rx_fc);
4467
4468         if (be_physfn(adapter))
4469                 be_cmd_set_logical_link_config(adapter,
4470                                                IFLA_VF_LINK_STATE_AUTO, 0);
4471
4472         if (adapter->num_vfs)
4473                 be_vf_setup(adapter);
4474
4475         status = be_cmd_get_phy_info(adapter);
4476         if (!status && be_pause_supported(adapter))
4477                 adapter->phy.fc_autoneg = 1;
4478
4479         be_schedule_worker(adapter);
4480         adapter->flags |= BE_FLAGS_SETUP_DONE;
4481         return 0;
4482 err:
4483         be_clear(adapter);
4484         return status;
4485 }
4486
4487 #ifdef CONFIG_NET_POLL_CONTROLLER
4488 static void be_netpoll(struct net_device *netdev)
4489 {
4490         struct be_adapter *adapter = netdev_priv(netdev);
4491         struct be_eq_obj *eqo;
4492         int i;
4493
4494         for_all_evt_queues(adapter, eqo, i) {
4495                 be_eq_notify(eqo->adapter, eqo->q.id, false, true, 0, 0);
4496                 napi_schedule(&eqo->napi);
4497         }
4498 }
4499 #endif
4500
4501 int be_load_fw(struct be_adapter *adapter, u8 *fw_file)
4502 {
4503         const struct firmware *fw;
4504         int status;
4505
4506         if (!netif_running(adapter->netdev)) {
4507                 dev_err(&adapter->pdev->dev,
4508                         "Firmware load not allowed (interface is down)\n");
4509                 return -ENETDOWN;
4510         }
4511
4512         status = request_firmware(&fw, fw_file, &adapter->pdev->dev);
4513         if (status)
4514                 goto fw_exit;
4515
4516         dev_info(&adapter->pdev->dev, "Flashing firmware file %s\n", fw_file);
4517
4518         if (lancer_chip(adapter))
4519                 status = lancer_fw_download(adapter, fw);
4520         else
4521                 status = be_fw_download(adapter, fw);
4522
4523         if (!status)
4524                 be_cmd_get_fw_ver(adapter);
4525
4526 fw_exit:
4527         release_firmware(fw);
4528         return status;
4529 }
4530
4531 static int be_ndo_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
4532                                  u16 flags)
4533 {
4534         struct be_adapter *adapter = netdev_priv(dev);
4535         struct nlattr *attr, *br_spec;
4536         int rem;
4537         int status = 0;
4538         u16 mode = 0;
4539
4540         if (!sriov_enabled(adapter))
4541                 return -EOPNOTSUPP;
4542
4543         br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
4544         if (!br_spec)
4545                 return -EINVAL;
4546
4547         nla_for_each_nested(attr, br_spec, rem) {
4548                 if (nla_type(attr) != IFLA_BRIDGE_MODE)
4549                         continue;
4550
4551                 if (nla_len(attr) < sizeof(mode))
4552                         return -EINVAL;
4553
4554                 mode = nla_get_u16(attr);
4555                 if (BE3_chip(adapter) && mode == BRIDGE_MODE_VEPA)
4556                         return -EOPNOTSUPP;
4557
4558                 if (mode != BRIDGE_MODE_VEPA && mode != BRIDGE_MODE_VEB)
4559                         return -EINVAL;
4560
4561                 status = be_cmd_set_hsw_config(adapter, 0, 0,
4562                                                adapter->if_handle,
4563                                                mode == BRIDGE_MODE_VEPA ?
4564                                                PORT_FWD_TYPE_VEPA :
4565                                                PORT_FWD_TYPE_VEB, 0);
4566                 if (status)
4567                         goto err;
4568
4569                 dev_info(&adapter->pdev->dev, "enabled switch mode: %s\n",
4570                          mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4571
4572                 return status;
4573         }
4574 err:
4575         dev_err(&adapter->pdev->dev, "Failed to set switch mode %s\n",
4576                 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
4577
4578         return status;
4579 }
4580
4581 static int be_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
4582                                  struct net_device *dev, u32 filter_mask,
4583                                  int nlflags)
4584 {
4585         struct be_adapter *adapter = netdev_priv(dev);
4586         int status = 0;
4587         u8 hsw_mode;
4588
4589         /* BE and Lancer chips support VEB mode only */
4590         if (BEx_chip(adapter) || lancer_chip(adapter)) {
4591                 hsw_mode = PORT_FWD_TYPE_VEB;
4592         } else {
4593                 status = be_cmd_get_hsw_config(adapter, NULL, 0,
4594                                                adapter->if_handle, &hsw_mode,
4595                                                NULL);
4596                 if (status)
4597                         return 0;
4598
4599                 if (hsw_mode == PORT_FWD_TYPE_PASSTHRU)
4600                         return 0;
4601         }
4602
4603         return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
4604                                        hsw_mode == PORT_FWD_TYPE_VEPA ?
4605                                        BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB,
4606                                        0, 0, nlflags, filter_mask, NULL);
4607 }
4608
4609 #ifdef CONFIG_BE2NET_VXLAN
4610 /* VxLAN offload Notes:
4611  *
4612  * The stack defines tunnel offload flags (hw_enc_features) for IP and doesn't
4613  * distinguish various types of transports (VxLAN, GRE, NVGRE ..). So, offload
4614  * is expected to work across all types of IP tunnels once exported. Skyhawk
4615  * supports offloads for either VxLAN or NVGRE, exclusively. So we export VxLAN
4616  * offloads in hw_enc_features only when a VxLAN port is added. If other (non
4617  * VxLAN) tunnels are configured while VxLAN offloads are enabled, offloads for
4618  * those other tunnels are unexported on the fly through ndo_features_check().
4619  *
4620  * Skyhawk supports VxLAN offloads only for one UDP dport. So, if the stack
4621  * adds more than one port, disable offloads and don't re-enable them again
4622  * until after all the tunnels are removed.
4623  */
4624 static void be_add_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
4625                               __be16 port)
4626 {
4627         struct be_adapter *adapter = netdev_priv(netdev);
4628         struct device *dev = &adapter->pdev->dev;
4629         int status;
4630
4631         if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
4632                 return;
4633
4634         if (adapter->vxlan_port == port && adapter->vxlan_port_count) {
4635                 adapter->vxlan_port_aliases++;
4636                 return;
4637         }
4638
4639         if (adapter->flags & BE_FLAGS_VXLAN_OFFLOADS) {
4640                 dev_info(dev,
4641                          "Only one UDP port supported for VxLAN offloads\n");
4642                 dev_info(dev, "Disabling VxLAN offloads\n");
4643                 adapter->vxlan_port_count++;
4644                 goto err;
4645         }
4646
4647         if (adapter->vxlan_port_count++ >= 1)
4648                 return;
4649
4650         status = be_cmd_manage_iface(adapter, adapter->if_handle,
4651                                      OP_CONVERT_NORMAL_TO_TUNNEL);
4652         if (status) {
4653                 dev_warn(dev, "Failed to convert normal interface to tunnel\n");
4654                 goto err;
4655         }
4656
4657         status = be_cmd_set_vxlan_port(adapter, port);
4658         if (status) {
4659                 dev_warn(dev, "Failed to add VxLAN port\n");
4660                 goto err;
4661         }
4662         adapter->flags |= BE_FLAGS_VXLAN_OFFLOADS;
4663         adapter->vxlan_port = port;
4664
4665         netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4666                                    NETIF_F_TSO | NETIF_F_TSO6 |
4667                                    NETIF_F_GSO_UDP_TUNNEL;
4668         netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
4669         netdev->features |= NETIF_F_GSO_UDP_TUNNEL;
4670
4671         dev_info(dev, "Enabled VxLAN offloads for UDP port %d\n",
4672                  be16_to_cpu(port));
4673         return;
4674 err:
4675         be_disable_vxlan_offloads(adapter);
4676 }
4677
4678 static void be_del_vxlan_port(struct net_device *netdev, sa_family_t sa_family,
4679                               __be16 port)
4680 {
4681         struct be_adapter *adapter = netdev_priv(netdev);
4682
4683         if (lancer_chip(adapter) || BEx_chip(adapter) || be_is_mc(adapter))
4684                 return;
4685
4686         if (adapter->vxlan_port != port)
4687                 goto done;
4688
4689         if (adapter->vxlan_port_aliases) {
4690                 adapter->vxlan_port_aliases--;
4691                 return;
4692         }
4693
4694         be_disable_vxlan_offloads(adapter);
4695
4696         dev_info(&adapter->pdev->dev,
4697                  "Disabled VxLAN offloads for UDP port %d\n",
4698                  be16_to_cpu(port));
4699 done:
4700         adapter->vxlan_port_count--;
4701 }
4702
4703 static netdev_features_t be_features_check(struct sk_buff *skb,
4704                                            struct net_device *dev,
4705                                            netdev_features_t features)
4706 {
4707         struct be_adapter *adapter = netdev_priv(dev);
4708         u8 l4_hdr = 0;
4709
4710         /* The code below restricts offload features for some tunneled packets.
4711          * Offload features for normal (non tunnel) packets are unchanged.
4712          */
4713         if (!skb->encapsulation ||
4714             !(adapter->flags & BE_FLAGS_VXLAN_OFFLOADS))
4715                 return features;
4716
4717         /* It's an encapsulated packet and VxLAN offloads are enabled. We
4718          * should disable tunnel offload features if it's not a VxLAN packet,
4719          * as tunnel offloads have been enabled only for VxLAN. This is done to
4720          * allow other tunneled traffic like GRE work fine while VxLAN
4721          * offloads are configured in Skyhawk-R.
4722          */
4723         switch (vlan_get_protocol(skb)) {
4724         case htons(ETH_P_IP):
4725                 l4_hdr = ip_hdr(skb)->protocol;
4726                 break;
4727         case htons(ETH_P_IPV6):
4728                 l4_hdr = ipv6_hdr(skb)->nexthdr;
4729                 break;
4730         default:
4731                 return features;
4732         }
4733
4734         if (l4_hdr != IPPROTO_UDP ||
4735             skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
4736             skb->inner_protocol != htons(ETH_P_TEB) ||
4737             skb_inner_mac_header(skb) - skb_transport_header(skb) !=
4738             sizeof(struct udphdr) + sizeof(struct vxlanhdr))
4739                 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
4740
4741         return features;
4742 }
4743 #endif
4744
4745 static int be_get_phys_port_id(struct net_device *dev,
4746                                struct netdev_phys_item_id *ppid)
4747 {
4748         int i, id_len = CNTL_SERIAL_NUM_WORDS * CNTL_SERIAL_NUM_WORD_SZ + 1;
4749         struct be_adapter *adapter = netdev_priv(dev);
4750         u8 *id;
4751
4752         if (MAX_PHYS_ITEM_ID_LEN < id_len)
4753                 return -ENOSPC;
4754
4755         ppid->id[0] = adapter->hba_port_num + 1;
4756         id = &ppid->id[1];
4757         for (i = CNTL_SERIAL_NUM_WORDS - 1; i >= 0;
4758              i--, id += CNTL_SERIAL_NUM_WORD_SZ)
4759                 memcpy(id, &adapter->serial_num[i], CNTL_SERIAL_NUM_WORD_SZ);
4760
4761         ppid->id_len = id_len;
4762
4763         return 0;
4764 }
4765
4766 static const struct net_device_ops be_netdev_ops = {
4767         .ndo_open               = be_open,
4768         .ndo_stop               = be_close,
4769         .ndo_start_xmit         = be_xmit,
4770         .ndo_set_rx_mode        = be_set_rx_mode,
4771         .ndo_set_mac_address    = be_mac_addr_set,
4772         .ndo_change_mtu         = be_change_mtu,
4773         .ndo_get_stats64        = be_get_stats64,
4774         .ndo_validate_addr      = eth_validate_addr,
4775         .ndo_vlan_rx_add_vid    = be_vlan_add_vid,
4776         .ndo_vlan_rx_kill_vid   = be_vlan_rem_vid,
4777         .ndo_set_vf_mac         = be_set_vf_mac,
4778         .ndo_set_vf_vlan        = be_set_vf_vlan,
4779         .ndo_set_vf_rate        = be_set_vf_tx_rate,
4780         .ndo_get_vf_config      = be_get_vf_config,
4781         .ndo_set_vf_link_state  = be_set_vf_link_state,
4782         .ndo_set_vf_spoofchk    = be_set_vf_spoofchk,
4783 #ifdef CONFIG_NET_POLL_CONTROLLER
4784         .ndo_poll_controller    = be_netpoll,
4785 #endif
4786         .ndo_bridge_setlink     = be_ndo_bridge_setlink,
4787         .ndo_bridge_getlink     = be_ndo_bridge_getlink,
4788 #ifdef CONFIG_NET_RX_BUSY_POLL
4789         .ndo_busy_poll          = be_busy_poll,
4790 #endif
4791 #ifdef CONFIG_BE2NET_VXLAN
4792         .ndo_add_vxlan_port     = be_add_vxlan_port,
4793         .ndo_del_vxlan_port     = be_del_vxlan_port,
4794         .ndo_features_check     = be_features_check,
4795 #endif
4796         .ndo_get_phys_port_id   = be_get_phys_port_id,
4797 };
4798
4799 static void be_netdev_init(struct net_device *netdev)
4800 {
4801         struct be_adapter *adapter = netdev_priv(netdev);
4802
4803         netdev->hw_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4804                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM |
4805                 NETIF_F_HW_VLAN_CTAG_TX;
4806         if (be_multi_rxq(adapter))
4807                 netdev->hw_features |= NETIF_F_RXHASH;
4808
4809         netdev->features |= netdev->hw_features |
4810                 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_HW_VLAN_CTAG_FILTER;
4811
4812         netdev->vlan_features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6 |
4813                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4814
4815         netdev->priv_flags |= IFF_UNICAST_FLT;
4816
4817         netdev->flags |= IFF_MULTICAST;
4818
4819         netif_set_gso_max_size(netdev, 65535 - ETH_HLEN);
4820
4821         netdev->netdev_ops = &be_netdev_ops;
4822
4823         netdev->ethtool_ops = &be_ethtool_ops;
4824 }
4825
4826 static void be_cleanup(struct be_adapter *adapter)
4827 {
4828         struct net_device *netdev = adapter->netdev;
4829
4830         rtnl_lock();
4831         netif_device_detach(netdev);
4832         if (netif_running(netdev))
4833                 be_close(netdev);
4834         rtnl_unlock();
4835
4836         be_clear(adapter);
4837 }
4838
4839 static int be_resume(struct be_adapter *adapter)
4840 {
4841         struct net_device *netdev = adapter->netdev;
4842         int status;
4843
4844         status = be_setup(adapter);
4845         if (status)
4846                 return status;
4847
4848         if (netif_running(netdev)) {
4849                 status = be_open(netdev);
4850                 if (status)
4851                         return status;
4852         }
4853
4854         netif_device_attach(netdev);
4855
4856         return 0;
4857 }
4858
4859 static int be_err_recover(struct be_adapter *adapter)
4860 {
4861         int status;
4862
4863         /* Error recovery is supported only Lancer as of now */
4864         if (!lancer_chip(adapter))
4865                 return -EIO;
4866
4867         /* Wait for adapter to reach quiescent state before
4868          * destroying queues
4869          */
4870         status = be_fw_wait_ready(adapter);
4871         if (status)
4872                 goto err;
4873
4874         be_cleanup(adapter);
4875
4876         status = be_resume(adapter);
4877         if (status)
4878                 goto err;
4879
4880         return 0;
4881 err:
4882         return status;
4883 }
4884
4885 static void be_err_detection_task(struct work_struct *work)
4886 {
4887         struct be_adapter *adapter =
4888                                 container_of(work, struct be_adapter,
4889                                              be_err_detection_work.work);
4890         struct device *dev = &adapter->pdev->dev;
4891         int recovery_status;
4892         int delay = ERR_DETECTION_DELAY;
4893
4894         be_detect_error(adapter);
4895
4896         if (be_check_error(adapter, BE_ERROR_HW))
4897                 recovery_status = be_err_recover(adapter);
4898         else
4899                 goto reschedule_task;
4900
4901         if (!recovery_status) {
4902                 adapter->recovery_retries = 0;
4903                 dev_info(dev, "Adapter recovery successful\n");
4904                 goto reschedule_task;
4905         } else if (be_virtfn(adapter)) {
4906                 /* For VFs, check if PF have allocated resources
4907                  * every second.
4908                  */
4909                 dev_err(dev, "Re-trying adapter recovery\n");
4910                 goto reschedule_task;
4911         } else if (adapter->recovery_retries++ <
4912                    MAX_ERR_RECOVERY_RETRY_COUNT) {
4913                 /* In case of another error during recovery, it takes 30 sec
4914                  * for adapter to come out of error. Retry error recovery after
4915                  * this time interval.
4916                  */
4917                 dev_err(&adapter->pdev->dev, "Re-trying adapter recovery\n");
4918                 delay = ERR_RECOVERY_RETRY_DELAY;
4919                 goto reschedule_task;
4920         } else {
4921                 dev_err(dev, "Adapter recovery failed\n");
4922         }
4923
4924         return;
4925 reschedule_task:
4926         be_schedule_err_detection(adapter, delay);
4927 }
4928
4929 static void be_log_sfp_info(struct be_adapter *adapter)
4930 {
4931         int status;
4932
4933         status = be_cmd_query_sfp_info(adapter);
4934         if (!status) {
4935                 dev_err(&adapter->pdev->dev,
4936                         "Unqualified SFP+ detected on %c from %s part no: %s",
4937                         adapter->port_name, adapter->phy.vendor_name,
4938                         adapter->phy.vendor_pn);
4939         }
4940         adapter->flags &= ~BE_FLAGS_EVT_INCOMPATIBLE_SFP;
4941 }
4942
4943 static void be_worker(struct work_struct *work)
4944 {
4945         struct be_adapter *adapter =
4946                 container_of(work, struct be_adapter, work.work);
4947         struct be_rx_obj *rxo;
4948         int i;
4949
4950         /* when interrupts are not yet enabled, just reap any pending
4951          * mcc completions
4952          */
4953         if (!netif_running(adapter->netdev)) {
4954                 local_bh_disable();
4955                 be_process_mcc(adapter);
4956                 local_bh_enable();
4957                 goto reschedule;
4958         }
4959
4960         if (!adapter->stats_cmd_sent) {
4961                 if (lancer_chip(adapter))
4962                         lancer_cmd_get_pport_stats(adapter,
4963                                                    &adapter->stats_cmd);
4964                 else
4965                         be_cmd_get_stats(adapter, &adapter->stats_cmd);
4966         }
4967
4968         if (be_physfn(adapter) &&
4969             MODULO(adapter->work_counter, adapter->be_get_temp_freq) == 0)
4970                 be_cmd_get_die_temperature(adapter);
4971
4972         for_all_rx_queues(adapter, rxo, i) {
4973                 /* Replenish RX-queues starved due to memory
4974                  * allocation failures.
4975                  */
4976                 if (rxo->rx_post_starved)
4977                         be_post_rx_frags(rxo, GFP_KERNEL, MAX_RX_POST);
4978         }
4979
4980         /* EQ-delay update for Skyhawk is done while notifying EQ */
4981         if (!skyhawk_chip(adapter))
4982                 be_eqd_update(adapter, false);
4983
4984         if (adapter->flags & BE_FLAGS_EVT_INCOMPATIBLE_SFP)
4985                 be_log_sfp_info(adapter);
4986
4987 reschedule:
4988         adapter->work_counter++;
4989         schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
4990 }
4991
4992 static void be_unmap_pci_bars(struct be_adapter *adapter)
4993 {
4994         if (adapter->csr)
4995                 pci_iounmap(adapter->pdev, adapter->csr);
4996         if (adapter->db)
4997                 pci_iounmap(adapter->pdev, adapter->db);
4998 }
4999
5000 static int db_bar(struct be_adapter *adapter)
5001 {
5002         if (lancer_chip(adapter) || be_virtfn(adapter))
5003                 return 0;
5004         else
5005                 return 4;
5006 }
5007
5008 static int be_roce_map_pci_bars(struct be_adapter *adapter)
5009 {
5010         if (skyhawk_chip(adapter)) {
5011                 adapter->roce_db.size = 4096;
5012                 adapter->roce_db.io_addr = pci_resource_start(adapter->pdev,
5013                                                               db_bar(adapter));
5014                 adapter->roce_db.total_size = pci_resource_len(adapter->pdev,
5015                                                                db_bar(adapter));
5016         }
5017         return 0;
5018 }
5019
5020 static int be_map_pci_bars(struct be_adapter *adapter)
5021 {
5022         struct pci_dev *pdev = adapter->pdev;
5023         u8 __iomem *addr;
5024         u32 sli_intf;
5025
5026         pci_read_config_dword(adapter->pdev, SLI_INTF_REG_OFFSET, &sli_intf);
5027         adapter->sli_family = (sli_intf & SLI_INTF_FAMILY_MASK) >>
5028                                 SLI_INTF_FAMILY_SHIFT;
5029         adapter->virtfn = (sli_intf & SLI_INTF_FT_MASK) ? 1 : 0;
5030
5031         if (BEx_chip(adapter) && be_physfn(adapter)) {
5032                 adapter->csr = pci_iomap(pdev, 2, 0);
5033                 if (!adapter->csr)
5034                         return -ENOMEM;
5035         }
5036
5037         addr = pci_iomap(pdev, db_bar(adapter), 0);
5038         if (!addr)
5039                 goto pci_map_err;
5040         adapter->db = addr;
5041
5042         if (skyhawk_chip(adapter) || BEx_chip(adapter)) {
5043                 if (be_physfn(adapter)) {
5044                         /* PCICFG is the 2nd BAR in BE2 */
5045                         addr = pci_iomap(pdev, BE2_chip(adapter) ? 1 : 0, 0);
5046                         if (!addr)
5047                                 goto pci_map_err;
5048                         adapter->pcicfg = addr;
5049                 } else {
5050                         adapter->pcicfg = adapter->db + SRIOV_VF_PCICFG_OFFSET;
5051                 }
5052         }
5053
5054         be_roce_map_pci_bars(adapter);
5055         return 0;
5056
5057 pci_map_err:
5058         dev_err(&pdev->dev, "Error in mapping PCI BARs\n");
5059         be_unmap_pci_bars(adapter);
5060         return -ENOMEM;
5061 }
5062
5063 static void be_drv_cleanup(struct be_adapter *adapter)
5064 {
5065         struct be_dma_mem *mem = &adapter->mbox_mem_alloced;
5066         struct device *dev = &adapter->pdev->dev;
5067
5068         if (mem->va)
5069                 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5070
5071         mem = &adapter->rx_filter;
5072         if (mem->va)
5073                 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5074
5075         mem = &adapter->stats_cmd;
5076         if (mem->va)
5077                 dma_free_coherent(dev, mem->size, mem->va, mem->dma);
5078 }
5079
5080 /* Allocate and initialize various fields in be_adapter struct */
5081 static int be_drv_init(struct be_adapter *adapter)
5082 {
5083         struct be_dma_mem *mbox_mem_alloc = &adapter->mbox_mem_alloced;
5084         struct be_dma_mem *mbox_mem_align = &adapter->mbox_mem;
5085         struct be_dma_mem *rx_filter = &adapter->rx_filter;
5086         struct be_dma_mem *stats_cmd = &adapter->stats_cmd;
5087         struct device *dev = &adapter->pdev->dev;
5088         int status = 0;
5089
5090         mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
5091         mbox_mem_alloc->va = dma_zalloc_coherent(dev, mbox_mem_alloc->size,
5092                                                  &mbox_mem_alloc->dma,
5093                                                  GFP_KERNEL);
5094         if (!mbox_mem_alloc->va)
5095                 return -ENOMEM;
5096
5097         mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
5098         mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
5099         mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
5100
5101         rx_filter->size = sizeof(struct be_cmd_req_rx_filter);
5102         rx_filter->va = dma_zalloc_coherent(dev, rx_filter->size,
5103                                             &rx_filter->dma, GFP_KERNEL);
5104         if (!rx_filter->va) {
5105                 status = -ENOMEM;
5106                 goto free_mbox;
5107         }
5108
5109         if (lancer_chip(adapter))
5110                 stats_cmd->size = sizeof(struct lancer_cmd_req_pport_stats);
5111         else if (BE2_chip(adapter))
5112                 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v0);
5113         else if (BE3_chip(adapter))
5114                 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v1);
5115         else
5116                 stats_cmd->size = sizeof(struct be_cmd_req_get_stats_v2);
5117         stats_cmd->va = dma_zalloc_coherent(dev, stats_cmd->size,
5118                                             &stats_cmd->dma, GFP_KERNEL);
5119         if (!stats_cmd->va) {
5120                 status = -ENOMEM;
5121                 goto free_rx_filter;
5122         }
5123
5124         mutex_init(&adapter->mbox_lock);
5125         spin_lock_init(&adapter->mcc_lock);
5126         spin_lock_init(&adapter->mcc_cq_lock);
5127         init_completion(&adapter->et_cmd_compl);
5128
5129         pci_save_state(adapter->pdev);
5130
5131         INIT_DELAYED_WORK(&adapter->work, be_worker);
5132         INIT_DELAYED_WORK(&adapter->be_err_detection_work,
5133                           be_err_detection_task);
5134
5135         adapter->rx_fc = true;
5136         adapter->tx_fc = true;
5137
5138         /* Must be a power of 2 or else MODULO will BUG_ON */
5139         adapter->be_get_temp_freq = 64;
5140
5141         return 0;
5142
5143 free_rx_filter:
5144         dma_free_coherent(dev, rx_filter->size, rx_filter->va, rx_filter->dma);
5145 free_mbox:
5146         dma_free_coherent(dev, mbox_mem_alloc->size, mbox_mem_alloc->va,
5147                           mbox_mem_alloc->dma);
5148         return status;
5149 }
5150
5151 static void be_remove(struct pci_dev *pdev)
5152 {
5153         struct be_adapter *adapter = pci_get_drvdata(pdev);
5154
5155         if (!adapter)
5156                 return;
5157
5158         be_roce_dev_remove(adapter);
5159         be_intr_set(adapter, false);
5160
5161         be_cancel_err_detection(adapter);
5162
5163         unregister_netdev(adapter->netdev);
5164
5165         be_clear(adapter);
5166
5167         /* tell fw we're done with firing cmds */
5168         be_cmd_fw_clean(adapter);
5169
5170         be_unmap_pci_bars(adapter);
5171         be_drv_cleanup(adapter);
5172
5173         pci_disable_pcie_error_reporting(pdev);
5174
5175         pci_release_regions(pdev);
5176         pci_disable_device(pdev);
5177
5178         free_netdev(adapter->netdev);
5179 }
5180
5181 static ssize_t be_hwmon_show_temp(struct device *dev,
5182                                   struct device_attribute *dev_attr,
5183                                   char *buf)
5184 {
5185         struct be_adapter *adapter = dev_get_drvdata(dev);
5186
5187         /* Unit: millidegree Celsius */
5188         if (adapter->hwmon_info.be_on_die_temp == BE_INVALID_DIE_TEMP)
5189                 return -EIO;
5190         else
5191                 return sprintf(buf, "%u\n",
5192                                adapter->hwmon_info.be_on_die_temp * 1000);
5193 }
5194
5195 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO,
5196                           be_hwmon_show_temp, NULL, 1);
5197
5198 static struct attribute *be_hwmon_attrs[] = {
5199         &sensor_dev_attr_temp1_input.dev_attr.attr,
5200         NULL
5201 };
5202
5203 ATTRIBUTE_GROUPS(be_hwmon);
5204
5205 static char *mc_name(struct be_adapter *adapter)
5206 {
5207         char *str = ""; /* default */
5208
5209         switch (adapter->mc_type) {
5210         case UMC:
5211                 str = "UMC";
5212                 break;
5213         case FLEX10:
5214                 str = "FLEX10";
5215                 break;
5216         case vNIC1:
5217                 str = "vNIC-1";
5218                 break;
5219         case nPAR:
5220                 str = "nPAR";
5221                 break;
5222         case UFP:
5223                 str = "UFP";
5224                 break;
5225         case vNIC2:
5226                 str = "vNIC-2";
5227                 break;
5228         default:
5229                 str = "";
5230         }
5231
5232         return str;
5233 }
5234
5235 static inline char *func_name(struct be_adapter *adapter)
5236 {
5237         return be_physfn(adapter) ? "PF" : "VF";
5238 }
5239
5240 static inline char *nic_name(struct pci_dev *pdev)
5241 {
5242         switch (pdev->device) {
5243         case OC_DEVICE_ID1:
5244                 return OC_NAME;
5245         case OC_DEVICE_ID2:
5246                 return OC_NAME_BE;
5247         case OC_DEVICE_ID3:
5248         case OC_DEVICE_ID4:
5249                 return OC_NAME_LANCER;
5250         case BE_DEVICE_ID2:
5251                 return BE3_NAME;
5252         case OC_DEVICE_ID5:
5253         case OC_DEVICE_ID6:
5254                 return OC_NAME_SH;
5255         default:
5256                 return BE_NAME;
5257         }
5258 }
5259
5260 static int be_probe(struct pci_dev *pdev, const struct pci_device_id *pdev_id)
5261 {
5262         struct be_adapter *adapter;
5263         struct net_device *netdev;
5264         int status = 0;
5265
5266         dev_info(&pdev->dev, "%s version is %s\n", DRV_NAME, DRV_VER);
5267
5268         status = pci_enable_device(pdev);
5269         if (status)
5270                 goto do_none;
5271
5272         status = pci_request_regions(pdev, DRV_NAME);
5273         if (status)
5274                 goto disable_dev;
5275         pci_set_master(pdev);
5276
5277         netdev = alloc_etherdev_mqs(sizeof(*adapter), MAX_TX_QS, MAX_RX_QS);
5278         if (!netdev) {
5279                 status = -ENOMEM;
5280                 goto rel_reg;
5281         }
5282         adapter = netdev_priv(netdev);
5283         adapter->pdev = pdev;
5284         pci_set_drvdata(pdev, adapter);
5285         adapter->netdev = netdev;
5286         SET_NETDEV_DEV(netdev, &pdev->dev);
5287
5288         status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
5289         if (!status) {
5290                 netdev->features |= NETIF_F_HIGHDMA;
5291         } else {
5292                 status = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
5293                 if (status) {
5294                         dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
5295                         goto free_netdev;
5296                 }
5297         }
5298
5299         status = pci_enable_pcie_error_reporting(pdev);
5300         if (!status)
5301                 dev_info(&pdev->dev, "PCIe error reporting enabled\n");
5302
5303         status = be_map_pci_bars(adapter);
5304         if (status)
5305                 goto free_netdev;
5306
5307         status = be_drv_init(adapter);
5308         if (status)
5309                 goto unmap_bars;
5310
5311         status = be_setup(adapter);
5312         if (status)
5313                 goto drv_cleanup;
5314
5315         be_netdev_init(netdev);
5316         status = register_netdev(netdev);
5317         if (status != 0)
5318                 goto unsetup;
5319
5320         be_roce_dev_add(adapter);
5321
5322         be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5323
5324         /* On Die temperature not supported for VF. */
5325         if (be_physfn(adapter) && IS_ENABLED(CONFIG_BE2NET_HWMON)) {
5326                 adapter->hwmon_info.hwmon_dev =
5327                         devm_hwmon_device_register_with_groups(&pdev->dev,
5328                                                                DRV_NAME,
5329                                                                adapter,
5330                                                                be_hwmon_groups);
5331                 adapter->hwmon_info.be_on_die_temp = BE_INVALID_DIE_TEMP;
5332         }
5333
5334         dev_info(&pdev->dev, "%s: %s %s port %c\n", nic_name(pdev),
5335                  func_name(adapter), mc_name(adapter), adapter->port_name);
5336
5337         return 0;
5338
5339 unsetup:
5340         be_clear(adapter);
5341 drv_cleanup:
5342         be_drv_cleanup(adapter);
5343 unmap_bars:
5344         be_unmap_pci_bars(adapter);
5345 free_netdev:
5346         free_netdev(netdev);
5347 rel_reg:
5348         pci_release_regions(pdev);
5349 disable_dev:
5350         pci_disable_device(pdev);
5351 do_none:
5352         dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
5353         return status;
5354 }
5355
5356 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
5357 {
5358         struct be_adapter *adapter = pci_get_drvdata(pdev);
5359
5360         if (adapter->wol_en)
5361                 be_setup_wol(adapter, true);
5362
5363         be_intr_set(adapter, false);
5364         be_cancel_err_detection(adapter);
5365
5366         be_cleanup(adapter);
5367
5368         pci_save_state(pdev);
5369         pci_disable_device(pdev);
5370         pci_set_power_state(pdev, pci_choose_state(pdev, state));
5371         return 0;
5372 }
5373
5374 static int be_pci_resume(struct pci_dev *pdev)
5375 {
5376         struct be_adapter *adapter = pci_get_drvdata(pdev);
5377         int status = 0;
5378
5379         status = pci_enable_device(pdev);
5380         if (status)
5381                 return status;
5382
5383         pci_restore_state(pdev);
5384
5385         status = be_resume(adapter);
5386         if (status)
5387                 return status;
5388
5389         be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5390
5391         if (adapter->wol_en)
5392                 be_setup_wol(adapter, false);
5393
5394         return 0;
5395 }
5396
5397 /*
5398  * An FLR will stop BE from DMAing any data.
5399  */
5400 static void be_shutdown(struct pci_dev *pdev)
5401 {
5402         struct be_adapter *adapter = pci_get_drvdata(pdev);
5403
5404         if (!adapter)
5405                 return;
5406
5407         be_roce_dev_shutdown(adapter);
5408         cancel_delayed_work_sync(&adapter->work);
5409         be_cancel_err_detection(adapter);
5410
5411         netif_device_detach(adapter->netdev);
5412
5413         be_cmd_reset_function(adapter);
5414
5415         pci_disable_device(pdev);
5416 }
5417
5418 static pci_ers_result_t be_eeh_err_detected(struct pci_dev *pdev,
5419                                             pci_channel_state_t state)
5420 {
5421         struct be_adapter *adapter = pci_get_drvdata(pdev);
5422
5423         dev_err(&adapter->pdev->dev, "EEH error detected\n");
5424
5425         if (!be_check_error(adapter, BE_ERROR_EEH)) {
5426                 be_set_error(adapter, BE_ERROR_EEH);
5427
5428                 be_cancel_err_detection(adapter);
5429
5430                 be_cleanup(adapter);
5431         }
5432
5433         if (state == pci_channel_io_perm_failure)
5434                 return PCI_ERS_RESULT_DISCONNECT;
5435
5436         pci_disable_device(pdev);
5437
5438         /* The error could cause the FW to trigger a flash debug dump.
5439          * Resetting the card while flash dump is in progress
5440          * can cause it not to recover; wait for it to finish.
5441          * Wait only for first function as it is needed only once per
5442          * adapter.
5443          */
5444         if (pdev->devfn == 0)
5445                 ssleep(30);
5446
5447         return PCI_ERS_RESULT_NEED_RESET;
5448 }
5449
5450 static pci_ers_result_t be_eeh_reset(struct pci_dev *pdev)
5451 {
5452         struct be_adapter *adapter = pci_get_drvdata(pdev);
5453         int status;
5454
5455         dev_info(&adapter->pdev->dev, "EEH reset\n");
5456
5457         status = pci_enable_device(pdev);
5458         if (status)
5459                 return PCI_ERS_RESULT_DISCONNECT;
5460
5461         pci_set_master(pdev);
5462         pci_restore_state(pdev);
5463
5464         /* Check if card is ok and fw is ready */
5465         dev_info(&adapter->pdev->dev,
5466                  "Waiting for FW to be ready after EEH reset\n");
5467         status = be_fw_wait_ready(adapter);
5468         if (status)
5469                 return PCI_ERS_RESULT_DISCONNECT;
5470
5471         pci_cleanup_aer_uncorrect_error_status(pdev);
5472         be_clear_error(adapter, BE_CLEAR_ALL);
5473         return PCI_ERS_RESULT_RECOVERED;
5474 }
5475
5476 static void be_eeh_resume(struct pci_dev *pdev)
5477 {
5478         int status = 0;
5479         struct be_adapter *adapter = pci_get_drvdata(pdev);
5480
5481         dev_info(&adapter->pdev->dev, "EEH resume\n");
5482
5483         pci_save_state(pdev);
5484
5485         status = be_resume(adapter);
5486         if (status)
5487                 goto err;
5488
5489         be_schedule_err_detection(adapter, ERR_DETECTION_DELAY);
5490         return;
5491 err:
5492         dev_err(&adapter->pdev->dev, "EEH resume failed\n");
5493 }
5494
5495 static int be_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
5496 {
5497         struct be_adapter *adapter = pci_get_drvdata(pdev);
5498         u16 num_vf_qs;
5499         int status;
5500
5501         if (!num_vfs)
5502                 be_vf_clear(adapter);
5503
5504         adapter->num_vfs = num_vfs;
5505
5506         if (adapter->num_vfs == 0 && pci_vfs_assigned(pdev)) {
5507                 dev_warn(&pdev->dev,
5508                          "Cannot disable VFs while they are assigned\n");
5509                 return -EBUSY;
5510         }
5511
5512         /* When the HW is in SRIOV capable configuration, the PF-pool resources
5513          * are equally distributed across the max-number of VFs. The user may
5514          * request only a subset of the max-vfs to be enabled.
5515          * Based on num_vfs, redistribute the resources across num_vfs so that
5516          * each VF will have access to more number of resources.
5517          * This facility is not available in BE3 FW.
5518          * Also, this is done by FW in Lancer chip.
5519          */
5520         if (skyhawk_chip(adapter) && !pci_num_vf(pdev)) {
5521                 num_vf_qs = be_calculate_vf_qs(adapter, adapter->num_vfs);
5522                 status = be_cmd_set_sriov_config(adapter, adapter->pool_res,
5523                                                  adapter->num_vfs, num_vf_qs);
5524                 if (status)
5525                         dev_err(&pdev->dev,
5526                                 "Failed to optimize SR-IOV resources\n");
5527         }
5528
5529         status = be_get_resources(adapter);
5530         if (status)
5531                 return be_cmd_status(status);
5532
5533         /* Updating real_num_tx/rx_queues() requires rtnl_lock() */
5534         rtnl_lock();
5535         status = be_update_queues(adapter);
5536         rtnl_unlock();
5537         if (status)
5538                 return be_cmd_status(status);
5539
5540         if (adapter->num_vfs)
5541                 status = be_vf_setup(adapter);
5542
5543         if (!status)
5544                 return adapter->num_vfs;
5545
5546         return 0;
5547 }
5548
5549 static const struct pci_error_handlers be_eeh_handlers = {
5550         .error_detected = be_eeh_err_detected,
5551         .slot_reset = be_eeh_reset,
5552         .resume = be_eeh_resume,
5553 };
5554
5555 static struct pci_driver be_driver = {
5556         .name = DRV_NAME,
5557         .id_table = be_dev_ids,
5558         .probe = be_probe,
5559         .remove = be_remove,
5560         .suspend = be_suspend,
5561         .resume = be_pci_resume,
5562         .shutdown = be_shutdown,
5563         .sriov_configure = be_pci_sriov_configure,
5564         .err_handler = &be_eeh_handlers
5565 };
5566
5567 static int __init be_init_module(void)
5568 {
5569         if (rx_frag_size != 8192 && rx_frag_size != 4096 &&
5570             rx_frag_size != 2048) {
5571                 printk(KERN_WARNING DRV_NAME
5572                         " : Module param rx_frag_size must be 2048/4096/8192."
5573                         " Using 2048\n");
5574                 rx_frag_size = 2048;
5575         }
5576
5577         if (num_vfs > 0) {
5578                 pr_info(DRV_NAME " : Module param num_vfs is obsolete.");
5579                 pr_info(DRV_NAME " : Use sysfs method to enable VFs\n");
5580         }
5581
5582         return pci_register_driver(&be_driver);
5583 }
5584 module_init(be_init_module);
5585
5586 static void __exit be_exit_module(void)
5587 {
5588         pci_unregister_driver(&be_driver);
5589 }
5590 module_exit(be_exit_module);