SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/net/benet/be_main.c

   1 /*
   2  * Copyright (C) 2005 - 2009 ServerEngines
   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@serverengines.com
  12  *
  13  * ServerEngines
  14  * 209 N. Fair Oaks Ave
  15  * Sunnyvale, CA 94085
  16  */
  17
  18 #include "be.h"
  19 #include <asm/div64.h>
  20
  21 MODULE_VERSION(DRV_VER);
  22 MODULE_DEVICE_TABLE(pci, be_dev_ids);
  23 MODULE_DESCRIPTION(DRV_DESC " " DRV_VER);
  24 MODULE_AUTHOR("ServerEngines Corporation");
  25 MODULE_LICENSE("GPL");
  26
  27 static unsigned int rx_frag_size = 2048;
  28 module_param(rx_frag_size, uint, S_IRUGO);
  29 MODULE_PARM_DESC(rx_frag_size, "Size of a fragment that holds rcvd data.");
  30
  31 static DEFINE_PCI_DEVICE_TABLE(be_dev_ids) = {
  32         { PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
  33         { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
  34         { PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
  35         { 0 }
  36 };
  37 MODULE_DEVICE_TABLE(pci, be_dev_ids);
  38
  39 static void be_queue_free(struct be_adapter *adapter, struct be_queue_info *q)
  40 {
  41         struct be_dma_mem *mem = &q->dma_mem;
  42         if (mem->va)
  43                 pci_free_consistent(adapter->pdev, mem->size,
  44                         mem->va, mem->dma);
  45 }
  46
  47 static int be_queue_alloc(struct be_adapter *adapter, struct be_queue_info *q,
  48                 u16 len, u16 entry_size)
  49 {
  50         struct be_dma_mem *mem = &q->dma_mem;
  51
  52         memset(q, 0, sizeof(*q));
  53         q->len = len;
  54         q->entry_size = entry_size;
  55         mem->size = len * entry_size;
  56         mem->va = pci_alloc_consistent(adapter->pdev, mem->size, &mem->dma);
  57         if (!mem->va)
  58                 return -1;
  59         memset(mem->va, 0, mem->size);
  60         return 0;
  61 }
  62
  63 static void be_intr_set(struct be_ctrl_info *ctrl, bool enable)
  64 {
  65         u8 __iomem *addr = ctrl->pcicfg + PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET;
  66         u32 reg = ioread32(addr);
  67         u32 enabled = reg & MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
  68         if (!enabled && enable) {
  69                 reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
  70         } else if (enabled && !enable) {
  71                 reg &= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
  72         } else {
  73                 printk(KERN_WARNING DRV_NAME
  74                         ": bad value in membar_int_ctrl reg=0x%x\n", reg);
  75                 return;
  76         }
  77         iowrite32(reg, addr);
  78 }
  79
  80 static void be_rxq_notify(struct be_ctrl_info *ctrl, u16 qid, u16 posted)
  81 {
  82         u32 val = 0;
  83         val |= qid & DB_RQ_RING_ID_MASK;
  84         val |= posted << DB_RQ_NUM_POSTED_SHIFT;
  85         iowrite32(val, ctrl->db + DB_RQ_OFFSET);
  86 }
  87
  88 static void be_txq_notify(struct be_ctrl_info *ctrl, u16 qid, u16 posted)
  89 {
  90         u32 val = 0;
  91         val |= qid & DB_TXULP_RING_ID_MASK;
  92         val |= (posted & DB_TXULP_NUM_POSTED_MASK) << DB_TXULP_NUM_POSTED_SHIFT;
  93         iowrite32(val, ctrl->db + DB_TXULP1_OFFSET);
  94 }
  95
  96 static void be_eq_notify(struct be_ctrl_info *ctrl, u16 qid,
  97                 bool arm, bool clear_int, u16 num_popped)
  98 {
  99         u32 val = 0;
 100         val |= qid & DB_EQ_RING_ID_MASK;
 101         if (arm)
 102                 val |= 1 << DB_EQ_REARM_SHIFT;
 103         if (clear_int)
 104                 val |= 1 << DB_EQ_CLR_SHIFT;
 105         val |= 1 << DB_EQ_EVNT_SHIFT;
 106         val |= num_popped << DB_EQ_NUM_POPPED_SHIFT;
 107         iowrite32(val, ctrl->db + DB_EQ_OFFSET);
 108 }
 109
 110 void be_cq_notify(struct be_ctrl_info *ctrl, u16 qid,
 111                 bool arm, u16 num_popped)
 112 {
 113         u32 val = 0;
 114         val |= qid & DB_CQ_RING_ID_MASK;
 115         if (arm)
 116                 val |= 1 << DB_CQ_REARM_SHIFT;
 117         val |= num_popped << DB_CQ_NUM_POPPED_SHIFT;
 118         iowrite32(val, ctrl->db + DB_CQ_OFFSET);
 119 }
 120
 121
 122 static int be_mac_addr_set(struct net_device *netdev, void *p)
 123 {
 124         struct be_adapter *adapter = netdev_priv(netdev);
 125         struct sockaddr *addr = p;
 126         int status = 0;
 127
 128         if (netif_running(netdev)) {
 129                 status = be_cmd_pmac_del(&adapter->ctrl, adapter->if_handle,
 130                                 adapter->pmac_id);
 131                 if (status)
 132                         return status;
 133
 134                 status = be_cmd_pmac_add(&adapter->ctrl, (u8 *)addr->sa_data,
 135                                 adapter->if_handle, &adapter->pmac_id);
 136         }
 137
 138         if (!status)
 139                 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
 140
 141         return status;
 142 }
 143
 144 static void netdev_stats_update(struct be_adapter *adapter)
 145 {
 146         struct be_hw_stats *hw_stats = hw_stats_from_cmd(adapter->stats.cmd.va);
 147         struct be_rxf_stats *rxf_stats = &hw_stats->rxf;
 148         struct be_port_rxf_stats *port_stats =
 149                         &rxf_stats->port[adapter->port_num];
 150         struct net_device_stats *dev_stats = &adapter->stats.net_stats;
 151         struct be_erx_stats *erx_stats = &hw_stats->erx;
 152
 153         dev_stats->rx_packets = port_stats->rx_total_frames;
 154         dev_stats->tx_packets = port_stats->tx_unicastframes +
 155                 port_stats->tx_multicastframes + port_stats->tx_broadcastframes;
 156         dev_stats->rx_bytes = (u64) port_stats->rx_bytes_msd << 32 |
 157                                 (u64) port_stats->rx_bytes_lsd;
 158         dev_stats->tx_bytes = (u64) port_stats->tx_bytes_msd << 32 |
 159                                 (u64) port_stats->tx_bytes_lsd;
 160
 161         /* bad pkts received */
 162         dev_stats->rx_errors = port_stats->rx_crc_errors +
 163                 port_stats->rx_alignment_symbol_errors +
 164                 port_stats->rx_in_range_errors +
 165                 port_stats->rx_out_range_errors +
 166                 port_stats->rx_frame_too_long +
 167                 port_stats->rx_dropped_too_small +
 168                 port_stats->rx_dropped_too_short +
 169                 port_stats->rx_dropped_header_too_small +
 170                 port_stats->rx_dropped_tcp_length +
 171                 port_stats->rx_dropped_runt +
 172                 port_stats->rx_tcp_checksum_errs +
 173                 port_stats->rx_ip_checksum_errs +
 174                 port_stats->rx_udp_checksum_errs;
 175
 176         /*  no space in linux buffers: best possible approximation */
 177         dev_stats->rx_dropped = erx_stats->rx_drops_no_fragments[0];
 178
 179         /* detailed rx errors */
 180         dev_stats->rx_length_errors = port_stats->rx_in_range_errors +
 181                 port_stats->rx_out_range_errors +
 182                 port_stats->rx_frame_too_long;
 183
 184         /* receive ring buffer overflow */
 185         dev_stats->rx_over_errors = 0;
 186
 187         dev_stats->rx_crc_errors = port_stats->rx_crc_errors;
 188
 189         /* frame alignment errors */
 190         dev_stats->rx_frame_errors = port_stats->rx_alignment_symbol_errors;
 191
 192         /* receiver fifo overrun */
 193         /* drops_no_pbuf is no per i/f, it's per BE card */
 194         dev_stats->rx_fifo_errors = port_stats->rx_fifo_overflow +
 195                                         port_stats->rx_input_fifo_overflow +
 196                                         rxf_stats->rx_drops_no_pbuf;
 197         /* receiver missed packetd */
 198         dev_stats->rx_missed_errors = 0;
 199
 200         /*  packet transmit problems */
 201         dev_stats->tx_errors = 0;
 202
 203         /* no space available in linux */
 204         dev_stats->tx_dropped = 0;
 205
 206         dev_stats->multicast = port_stats->tx_multicastframes;
 207         dev_stats->collisions = 0;
 208
 209         /* detailed tx_errors */
 210         dev_stats->tx_aborted_errors = 0;
 211         dev_stats->tx_carrier_errors = 0;
 212         dev_stats->tx_fifo_errors = 0;
 213         dev_stats->tx_heartbeat_errors = 0;
 214         dev_stats->tx_window_errors = 0;
 215 }
 216
 217 void be_link_status_update(void *ctxt, bool link_up)
 218 {
 219         struct be_adapter *adapter = ctxt;
 220         struct net_device *netdev = adapter->netdev;
 221
 222         /* If link came up or went down */
 223         if (adapter->link_up != link_up) {
 224                 if (link_up) {
 225                         netif_start_queue(netdev);
 226                         netif_carrier_on(netdev);
 227                         printk(KERN_INFO "%s: Link up\n", netdev->name);
 228                 } else {
 229                         netif_stop_queue(netdev);
 230                         netif_carrier_off(netdev);
 231                         printk(KERN_INFO "%s: Link down\n", netdev->name);
 232                 }
 233                 adapter->link_up = link_up;
 234         }
 235 }
 236
 237 /* Update the EQ delay n BE based on the RX frags consumed / sec */
 238 static void be_rx_eqd_update(struct be_adapter *adapter)
 239 {
 240         struct be_ctrl_info *ctrl = &adapter->ctrl;
 241         struct be_eq_obj *rx_eq = &adapter->rx_eq;
 242         struct be_drvr_stats *stats = &adapter->stats.drvr_stats;
 243         ulong now = jiffies;
 244         u32 eqd;
 245
 246         if (!rx_eq->enable_aic)
 247                 return;
 248
 249         /* Wrapped around */
 250         if (time_before(now, stats->rx_fps_jiffies)) {
 251                 stats->rx_fps_jiffies = now;
 252                 return;
 253         }
 254
 255         /* Update once a second */
 256         if ((now - stats->rx_fps_jiffies) < HZ)
 257                 return;
 258
 259         stats->be_rx_fps = (stats->be_rx_frags - stats->be_prev_rx_frags) /
 260                         ((now - stats->rx_fps_jiffies) / HZ);
 261
 262         stats->rx_fps_jiffies = now;
 263         stats->be_prev_rx_frags = stats->be_rx_frags;
 264         eqd = stats->be_rx_fps / 110000;
 265         eqd = eqd << 3;
 266         if (eqd > rx_eq->max_eqd)
 267                 eqd = rx_eq->max_eqd;
 268         if (eqd < rx_eq->min_eqd)
 269                 eqd = rx_eq->min_eqd;
 270         if (eqd < 10)
 271                 eqd = 0;
 272         if (eqd != rx_eq->cur_eqd)
 273                 be_cmd_modify_eqd(ctrl, rx_eq->q.id, eqd);
 274
 275         rx_eq->cur_eqd = eqd;
 276 }
 277
 278 static struct net_device_stats *be_get_stats(struct net_device *dev)
 279 {
 280         struct be_adapter *adapter = netdev_priv(dev);
 281
 282         return &adapter->stats.net_stats;
 283 }
 284
 285 static u32 be_calc_rate(u64 bytes, unsigned long ticks)
 286 {
 287         u64 rate = bytes;
 288
 289         do_div(rate, ticks / HZ);
 290         rate <<= 3;                     /* bytes/sec -> bits/sec */
 291         do_div(rate, 1000000ul);        /* MB/Sec */
 292
 293         return rate;
 294 }
 295
 296 static void be_tx_rate_update(struct be_adapter *adapter)
 297 {
 298         struct be_drvr_stats *stats = drvr_stats(adapter);
 299         ulong now = jiffies;
 300
 301         /* Wrapped around? */
 302         if (time_before(now, stats->be_tx_jiffies)) {
 303                 stats->be_tx_jiffies = now;
 304                 return;
 305         }
 306
 307         /* Update tx rate once in two seconds */
 308         if ((now - stats->be_tx_jiffies) > 2 * HZ) {
 309                 stats->be_tx_rate = be_calc_rate(stats->be_tx_bytes
 310                                                   - stats->be_tx_bytes_prev,
 311                                                  now - stats->be_tx_jiffies);
 312                 stats->be_tx_jiffies = now;
 313                 stats->be_tx_bytes_prev = stats->be_tx_bytes;
 314         }
 315 }
 316
 317 static void be_tx_stats_update(struct be_adapter *adapter,
 318                         u32 wrb_cnt, u32 copied, bool stopped)
 319 {
 320         struct be_drvr_stats *stats = drvr_stats(adapter);
 321         stats->be_tx_reqs++;
 322         stats->be_tx_wrbs += wrb_cnt;
 323         stats->be_tx_bytes += copied;
 324         if (stopped)
 325                 stats->be_tx_stops++;
 326 }
 327
 328 /* Determine number of WRB entries needed to xmit data in an skb */
 329 static u32 wrb_cnt_for_skb(struct sk_buff *skb, bool *dummy)
 330 {
 331         int cnt = (skb->len > skb->data_len);
 332
 333         cnt += skb_shinfo(skb)->nr_frags;
 334
 335         /* to account for hdr wrb */
 336         cnt++;
 337         if (cnt & 1) {
 338                 /* add a dummy to make it an even num */
 339                 cnt++;
 340                 *dummy = true;
 341         } else
 342                 *dummy = false;
 343         BUG_ON(cnt > BE_MAX_TX_FRAG_COUNT);
 344         return cnt;
 345 }
 346
 347 static inline void wrb_fill(struct be_eth_wrb *wrb, u64 addr, int len)
 348 {
 349         wrb->frag_pa_hi = upper_32_bits(addr);
 350         wrb->frag_pa_lo = addr & 0xFFFFFFFF;
 351         wrb->frag_len = len & ETH_WRB_FRAG_LEN_MASK;
 352 }
 353
 354 static void wrb_fill_hdr(struct be_eth_hdr_wrb *hdr, struct sk_buff *skb,
 355                 bool vlan, u32 wrb_cnt, u32 len)
 356 {
 357         memset(hdr, 0, sizeof(*hdr));
 358
 359         AMAP_SET_BITS(struct amap_eth_hdr_wrb, crc, hdr, 1);
 360
 361         if (skb_shinfo(skb)->gso_segs > 1 && skb_shinfo(skb)->gso_size) {
 362                 AMAP_SET_BITS(struct amap_eth_hdr_wrb, lso, hdr, 1);
 363                 AMAP_SET_BITS(struct amap_eth_hdr_wrb, lso_mss,
 364                         hdr, skb_shinfo(skb)->gso_size);
 365         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 366                 if (is_tcp_pkt(skb))
 367                         AMAP_SET_BITS(struct amap_eth_hdr_wrb, tcpcs, hdr, 1);
 368                 else if (is_udp_pkt(skb))
 369                         AMAP_SET_BITS(struct amap_eth_hdr_wrb, udpcs, hdr, 1);
 370         }
 371
 372         if (vlan && vlan_tx_tag_present(skb)) {
 373                 AMAP_SET_BITS(struct amap_eth_hdr_wrb, vlan, hdr, 1);
 374                 AMAP_SET_BITS(struct amap_eth_hdr_wrb, vlan_tag,
 375                         hdr, vlan_tx_tag_get(skb));
 376         }
 377
 378         AMAP_SET_BITS(struct amap_eth_hdr_wrb, event, hdr, 1);
 379         AMAP_SET_BITS(struct amap_eth_hdr_wrb, complete, hdr, 1);
 380         AMAP_SET_BITS(struct amap_eth_hdr_wrb, num_wrb, hdr, wrb_cnt);
 381         AMAP_SET_BITS(struct amap_eth_hdr_wrb, len, hdr, len);
 382 }
 383
 384
 385 static int make_tx_wrbs(struct be_adapter *adapter,
 386                 struct sk_buff *skb, u32 wrb_cnt, bool dummy_wrb)
 387 {
 388         u64 busaddr;
 389         u32 i, copied = 0;
 390         struct pci_dev *pdev = adapter->pdev;
 391         struct sk_buff *first_skb = skb;
 392         struct be_queue_info *txq = &adapter->tx_obj.q;
 393         struct be_eth_wrb *wrb;
 394         struct be_eth_hdr_wrb *hdr;
 395
 396         atomic_add(wrb_cnt, &txq->used);
 397         hdr = queue_head_node(txq);
 398         queue_head_inc(txq);
 399
 400         if (skb->len > skb->data_len) {
 401                 int len = skb->len - skb->data_len;
 402                 busaddr = pci_map_single(pdev, skb->data, len,
 403                                          PCI_DMA_TODEVICE);
 404                 wrb = queue_head_node(txq);
 405                 wrb_fill(wrb, busaddr, len);
 406                 be_dws_cpu_to_le(wrb, sizeof(*wrb));
 407                 queue_head_inc(txq);
 408                 copied += len;
 409         }
 410
 411         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 412                 struct skb_frag_struct *frag =
 413                         &skb_shinfo(skb)->frags[i];
 414                 busaddr = pci_map_page(pdev, frag->page,
 415                                        frag->page_offset,
 416                                        frag->size, PCI_DMA_TODEVICE);
 417                 wrb = queue_head_node(txq);
 418                 wrb_fill(wrb, busaddr, frag->size);
 419                 be_dws_cpu_to_le(wrb, sizeof(*wrb));
 420                 queue_head_inc(txq);
 421                 copied += frag->size;
 422         }
 423
 424         if (dummy_wrb) {
 425                 wrb = queue_head_node(txq);
 426                 wrb_fill(wrb, 0, 0);
 427                 be_dws_cpu_to_le(wrb, sizeof(*wrb));
 428                 queue_head_inc(txq);
 429         }
 430
 431         wrb_fill_hdr(hdr, first_skb, adapter->vlan_grp ? true : false,
 432                 wrb_cnt, copied);
 433         be_dws_cpu_to_le(hdr, sizeof(*hdr));
 434
 435         return copied;
 436 }
 437
 438 static int be_xmit(struct sk_buff *skb, struct net_device *netdev)
 439 {
 440         struct be_adapter *adapter = netdev_priv(netdev);
 441         struct be_tx_obj *tx_obj = &adapter->tx_obj;
 442         struct be_queue_info *txq = &tx_obj->q;
 443         u32 wrb_cnt = 0, copied = 0;
 444         u32 start = txq->head;
 445         bool dummy_wrb, stopped = false;
 446
 447         wrb_cnt = wrb_cnt_for_skb(skb, &dummy_wrb);
 448
 449         copied = make_tx_wrbs(adapter, skb, wrb_cnt, dummy_wrb);
 450
 451         /* record the sent skb in the sent_skb table */
 452         BUG_ON(tx_obj->sent_skb_list[start]);
 453         tx_obj->sent_skb_list[start] = skb;
 454
 455         /* Ensure that txq has space for the next skb; Else stop the queue
 456          * *BEFORE* ringing the tx doorbell, so that we serialze the
 457          * tx compls of the current transmit which'll wake up the queue
 458          */
 459         if ((BE_MAX_TX_FRAG_COUNT + atomic_read(&txq->used)) >= txq->len) {
 460                 netif_stop_queue(netdev);
 461                 stopped = true;
 462         }
 463
 464         be_txq_notify(&adapter->ctrl, txq->id, wrb_cnt);
 465
 466         be_tx_stats_update(adapter, wrb_cnt, copied, stopped);
 467         return NETDEV_TX_OK;
 468 }
 469
 470 static int be_change_mtu(struct net_device *netdev, int new_mtu)
 471 {
 472         struct be_adapter *adapter = netdev_priv(netdev);
 473         if (new_mtu < BE_MIN_MTU ||
 474                         new_mtu > BE_MAX_JUMBO_FRAME_SIZE) {
 475                 dev_info(&adapter->pdev->dev,
 476                         "MTU must be between %d and %d bytes\n",
 477                         BE_MIN_MTU, BE_MAX_JUMBO_FRAME_SIZE);
 478                 return -EINVAL;
 479         }
 480         dev_info(&adapter->pdev->dev, "MTU changed from %d to %d bytes\n",
 481                         netdev->mtu, new_mtu);
 482         netdev->mtu = new_mtu;
 483         return 0;
 484 }
 485
 486 /*
 487  * if there are BE_NUM_VLANS_SUPPORTED or lesser number of VLANS configured,
 488  * program them in BE.  If more than BE_NUM_VLANS_SUPPORTED are configured,
 489  * set the BE in promiscuous VLAN mode.
 490  */
 491 static void be_vid_config(struct net_device *netdev)
 492 {
 493         struct be_adapter *adapter = netdev_priv(netdev);
 494         u16 vtag[BE_NUM_VLANS_SUPPORTED];
 495         u16 ntags = 0, i;
 496
 497         if (adapter->num_vlans <= BE_NUM_VLANS_SUPPORTED)  {
 498                 /* Construct VLAN Table to give to HW */
 499                 for (i = 0; i < VLAN_GROUP_ARRAY_LEN; i++) {
 500                         if (adapter->vlan_tag[i]) {
 501                                 vtag[ntags] = cpu_to_le16(i);
 502                                 ntags++;
 503                         }
 504                 }
 505                 be_cmd_vlan_config(&adapter->ctrl, adapter->if_handle,
 506                         vtag, ntags, 1, 0);
 507         } else {
 508                 be_cmd_vlan_config(&adapter->ctrl, adapter->if_handle,
 509                         NULL, 0, 1, 1);
 510         }
 511 }
 512
 513 static void be_vlan_register(struct net_device *netdev, struct vlan_group *grp)
 514 {
 515         struct be_adapter *adapter = netdev_priv(netdev);
 516         struct be_eq_obj *rx_eq = &adapter->rx_eq;
 517         struct be_eq_obj *tx_eq = &adapter->tx_eq;
 518         struct be_ctrl_info *ctrl = &adapter->ctrl;
 519
 520         be_eq_notify(ctrl, rx_eq->q.id, false, false, 0);
 521         be_eq_notify(ctrl, tx_eq->q.id, false, false, 0);
 522         adapter->vlan_grp = grp;
 523         be_eq_notify(ctrl, rx_eq->q.id, true, false, 0);
 524         be_eq_notify(ctrl, tx_eq->q.id, true, false, 0);
 525 }
 526
 527 static void be_vlan_add_vid(struct net_device *netdev, u16 vid)
 528 {
 529         struct be_adapter *adapter = netdev_priv(netdev);
 530
 531         adapter->num_vlans++;
 532         adapter->vlan_tag[vid] = 1;
 533
 534         be_vid_config(netdev);
 535 }
 536
 537 static void be_vlan_rem_vid(struct net_device *netdev, u16 vid)
 538 {
 539         struct be_adapter *adapter = netdev_priv(netdev);
 540
 541         adapter->num_vlans--;
 542         adapter->vlan_tag[vid] = 0;
 543
 544         vlan_group_set_device(adapter->vlan_grp, vid, NULL);
 545         be_vid_config(netdev);
 546 }
 547
 548 static void be_set_multicast_list(struct net_device *netdev)
 549 {
 550         struct be_adapter *adapter = netdev_priv(netdev);
 551         struct be_ctrl_info *ctrl = &adapter->ctrl;
 552
 553         if (netdev->flags & IFF_PROMISC) {
 554                 be_cmd_promiscuous_config(ctrl, adapter->port_num, 1);
 555                 adapter->promiscuous = true;
 556                 goto done;
 557         }
 558
 559         /* BE was previously in promiscous mode; disable it */
 560         if (adapter->promiscuous) {
 561                 adapter->promiscuous = false;
 562                 be_cmd_promiscuous_config(ctrl, adapter->port_num, 0);
 563         }
 564
 565         if (netdev->flags & IFF_ALLMULTI) {
 566                 be_cmd_multicast_set(ctrl, adapter->if_handle, NULL, 0);
 567                 goto done;
 568         }
 569
 570         be_cmd_multicast_set(ctrl, adapter->if_handle, netdev->mc_list,
 571                 netdev->mc_count);
 572 done:
 573         return;
 574 }
 575
 576 static void be_rx_rate_update(struct be_adapter *adapter)
 577 {
 578         struct be_drvr_stats *stats = drvr_stats(adapter);
 579         ulong now = jiffies;
 580
 581         /* Wrapped around */
 582         if (time_before(now, stats->be_rx_jiffies)) {
 583                 stats->be_rx_jiffies = now;
 584                 return;
 585         }
 586
 587         /* Update the rate once in two seconds */
 588         if ((now - stats->be_rx_jiffies) < 2 * HZ)
 589                 return;
 590
 591         stats->be_rx_rate = be_calc_rate(stats->be_rx_bytes
 592                                           - stats->be_rx_bytes_prev,
 593                                          now - stats->be_rx_jiffies);
 594         stats->be_rx_jiffies = now;
 595         stats->be_rx_bytes_prev = stats->be_rx_bytes;
 596 }
 597
 598 static void be_rx_stats_update(struct be_adapter *adapter,
 599                 u32 pktsize, u16 numfrags)
 600 {
 601         struct be_drvr_stats *stats = drvr_stats(adapter);
 602
 603         stats->be_rx_compl++;
 604         stats->be_rx_frags += numfrags;
 605         stats->be_rx_bytes += pktsize;
 606 }
 607
 608 static inline bool do_pkt_csum(struct be_eth_rx_compl *rxcp, bool cso)
 609 {
 610         u8 l4_cksm, ip_version, ipcksm, tcpf = 0, udpf = 0, ipv6_chk;
 611
 612         l4_cksm = AMAP_GET_BITS(struct amap_eth_rx_compl, l4_cksm, rxcp);
 613         ipcksm = AMAP_GET_BITS(struct amap_eth_rx_compl, ipcksm, rxcp);
 614         ip_version = AMAP_GET_BITS(struct amap_eth_rx_compl, ip_version, rxcp);
 615         if (ip_version) {
 616                 tcpf = AMAP_GET_BITS(struct amap_eth_rx_compl, tcpf, rxcp);
 617                 udpf = AMAP_GET_BITS(struct amap_eth_rx_compl, udpf, rxcp);
 618         }
 619         ipv6_chk = (ip_version && (tcpf || udpf));
 620
 621         return ((l4_cksm && ipv6_chk && ipcksm) && cso) ? false : true;
 622 }
 623
 624 static struct be_rx_page_info *
 625 get_rx_page_info(struct be_adapter *adapter, u16 frag_idx)
 626 {
 627         struct be_rx_page_info *rx_page_info;
 628         struct be_queue_info *rxq = &adapter->rx_obj.q;
 629
 630         rx_page_info = &adapter->rx_obj.page_info_tbl[frag_idx];
 631         BUG_ON(!rx_page_info->page);
 632
 633         if (rx_page_info->last_page_user)
 634                 pci_unmap_page(adapter->pdev, pci_unmap_addr(rx_page_info, bus),
 635                         adapter->big_page_size, PCI_DMA_FROMDEVICE);
 636
 637         atomic_dec(&rxq->used);
 638         return rx_page_info;
 639 }
 640
 641 /* Throwaway the data in the Rx completion */
 642 static void be_rx_compl_discard(struct be_adapter *adapter,
 643                         struct be_eth_rx_compl *rxcp)
 644 {
 645         struct be_queue_info *rxq = &adapter->rx_obj.q;
 646         struct be_rx_page_info *page_info;
 647         u16 rxq_idx, i, num_rcvd;
 648
 649         rxq_idx = AMAP_GET_BITS(struct amap_eth_rx_compl, fragndx, rxcp);
 650         num_rcvd = AMAP_GET_BITS(struct amap_eth_rx_compl, numfrags, rxcp);
 651
 652         for (i = 0; i < num_rcvd; i++) {
 653                 page_info = get_rx_page_info(adapter, rxq_idx);
 654                 put_page(page_info->page);
 655                 memset(page_info, 0, sizeof(*page_info));
 656                 index_inc(&rxq_idx, rxq->len);
 657         }
 658 }
 659
 660 /*
 661  * skb_fill_rx_data forms a complete skb for an ether frame
 662  * indicated by rxcp.
 663  */
 664 static void skb_fill_rx_data(struct be_adapter *adapter,
 665                         struct sk_buff *skb, struct be_eth_rx_compl *rxcp)
 666 {
 667         struct be_queue_info *rxq = &adapter->rx_obj.q;
 668         struct be_rx_page_info *page_info;
 669         u16 rxq_idx, i, num_rcvd, j;
 670         u32 pktsize, hdr_len, curr_frag_len;
 671         u8 *start;
 672
 673         rxq_idx = AMAP_GET_BITS(struct amap_eth_rx_compl, fragndx, rxcp);
 674         pktsize = AMAP_GET_BITS(struct amap_eth_rx_compl, pktsize, rxcp);
 675         num_rcvd = AMAP_GET_BITS(struct amap_eth_rx_compl, numfrags, rxcp);
 676
 677         page_info = get_rx_page_info(adapter, rxq_idx);
 678
 679         start = page_address(page_info->page) + page_info->page_offset;
 680         prefetch(start);
 681
 682         /* Copy data in the first descriptor of this completion */
 683         curr_frag_len = min(pktsize, rx_frag_size);
 684
 685         /* Copy the header portion into skb_data */
 686         hdr_len = min((u32)BE_HDR_LEN, curr_frag_len);
 687         memcpy(skb->data, start, hdr_len);
 688         skb->len = curr_frag_len;
 689         if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
 690                 /* Complete packet has now been moved to data */
 691                 put_page(page_info->page);
 692                 skb->data_len = 0;
 693                 skb->tail += curr_frag_len;
 694         } else {
 695                 skb_shinfo(skb)->nr_frags = 1;
 696                 skb_shinfo(skb)->frags[0].page = page_info->page;
 697                 skb_shinfo(skb)->frags[0].page_offset =
 698                                         page_info->page_offset + hdr_len;
 699                 skb_shinfo(skb)->frags[0].size = curr_frag_len - hdr_len;
 700                 skb->data_len = curr_frag_len - hdr_len;
 701                 skb->tail += hdr_len;
 702         }
 703         memset(page_info, 0, sizeof(*page_info));
 704
 705         if (pktsize <= rx_frag_size) {
 706                 BUG_ON(num_rcvd != 1);
 707                 goto done;
 708         }
 709
 710         /* More frags present for this completion */
 711         pktsize -= curr_frag_len; /* account for above copied frag */
 712         for (i = 1, j = 0; i < num_rcvd; i++) {
 713                 index_inc(&rxq_idx, rxq->len);
 714                 page_info = get_rx_page_info(adapter, rxq_idx);
 715
 716                 curr_frag_len = min(pktsize, rx_frag_size);
 717
 718                 /* Coalesce all frags from the same physical page in one slot */
 719                 if (page_info->page_offset == 0) {
 720                         /* Fresh page */
 721                         j++;
 722                         skb_shinfo(skb)->frags[j].page = page_info->page;
 723                         skb_shinfo(skb)->frags[j].page_offset =
 724                                                         page_info->page_offset;
 725                         skb_shinfo(skb)->frags[j].size = 0;
 726                         skb_shinfo(skb)->nr_frags++;
 727                 } else {
 728                         put_page(page_info->page);
 729                 }
 730
 731                 skb_shinfo(skb)->frags[j].size += curr_frag_len;
 732                 skb->len += curr_frag_len;
 733                 skb->data_len += curr_frag_len;
 734                 pktsize -= curr_frag_len;
 735
 736                 memset(page_info, 0, sizeof(*page_info));
 737         }
 738         BUG_ON(j > MAX_SKB_FRAGS);
 739
 740 done:
 741         be_rx_stats_update(adapter, pktsize, num_rcvd);
 742         return;
 743 }
 744
 745 /* Process the RX completion indicated by rxcp when LRO is disabled */
 746 static void be_rx_compl_process(struct be_adapter *adapter,
 747                         struct be_eth_rx_compl *rxcp)
 748 {
 749         struct sk_buff *skb;
 750         u32 vtp, vid;
 751
 752         vtp = AMAP_GET_BITS(struct amap_eth_rx_compl, vtp, rxcp);
 753
 754         skb = netdev_alloc_skb(adapter->netdev, BE_HDR_LEN + NET_IP_ALIGN);
 755         if (!skb) {
 756                 if (net_ratelimit())
 757                         dev_warn(&adapter->pdev->dev, "skb alloc failed\n");
 758                 be_rx_compl_discard(adapter, rxcp);
 759                 return;
 760         }
 761
 762         skb_reserve(skb, NET_IP_ALIGN);
 763
 764         skb_fill_rx_data(adapter, skb, rxcp);
 765
 766         if (do_pkt_csum(rxcp, adapter->rx_csum))
 767                 skb->ip_summed = CHECKSUM_NONE;
 768         else
 769                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 770
 771         skb->truesize = skb->len + sizeof(struct sk_buff);
 772         skb->protocol = eth_type_trans(skb, adapter->netdev);
 773         skb->dev = adapter->netdev;
 774
 775         if (vtp) {
 776                 if (!adapter->vlan_grp || adapter->num_vlans == 0) {
 777                         kfree_skb(skb);
 778                         return;
 779                 }
 780                 vid = AMAP_GET_BITS(struct amap_eth_rx_compl, vlan_tag, rxcp);
 781                 vid = be16_to_cpu(vid);
 782                 vlan_hwaccel_receive_skb(skb, adapter->vlan_grp, vid);
 783         } else {
 784                 netif_receive_skb(skb);
 785         }
 786
 787         adapter->netdev->last_rx = jiffies;
 788
 789         return;
 790 }
 791
 792 /* Process the RX completion indicated by rxcp when LRO is enabled */
 793 static void be_rx_compl_process_lro(struct be_adapter *adapter,
 794                         struct be_eth_rx_compl *rxcp)
 795 {
 796         struct be_rx_page_info *page_info;
 797         struct skb_frag_struct rx_frags[BE_MAX_FRAGS_PER_FRAME];
 798         struct be_queue_info *rxq = &adapter->rx_obj.q;
 799         u32 num_rcvd, pkt_size, remaining, vlanf, curr_frag_len;
 800         u16 i, rxq_idx = 0, vid, j;
 801
 802         num_rcvd = AMAP_GET_BITS(struct amap_eth_rx_compl, numfrags, rxcp);
 803         pkt_size = AMAP_GET_BITS(struct amap_eth_rx_compl, pktsize, rxcp);
 804         vlanf = AMAP_GET_BITS(struct amap_eth_rx_compl, vtp, rxcp);
 805         rxq_idx = AMAP_GET_BITS(struct amap_eth_rx_compl, fragndx, rxcp);
 806
 807         remaining = pkt_size;
 808         for (i = 0, j = -1; i < num_rcvd; i++) {
 809                 page_info = get_rx_page_info(adapter, rxq_idx);
 810
 811                 curr_frag_len = min(remaining, rx_frag_size);
 812
 813                 /* Coalesce all frags from the same physical page in one slot */
 814                 if (i == 0 || page_info->page_offset == 0) {
 815                         /* First frag or Fresh page */
 816                         j++;
 817                         rx_frags[j].page = page_info->page;
 818                         rx_frags[j].page_offset = page_info->page_offset;
 819                         rx_frags[j].size = 0;
 820                 } else {
 821                         put_page(page_info->page);
 822                 }
 823                 rx_frags[j].size += curr_frag_len;
 824
 825                 remaining -= curr_frag_len;
 826                 index_inc(&rxq_idx, rxq->len);
 827                 memset(page_info, 0, sizeof(*page_info));
 828         }
 829         BUG_ON(j > MAX_SKB_FRAGS);
 830
 831         if (likely(!vlanf)) {
 832                 lro_receive_frags(&adapter->rx_obj.lro_mgr, rx_frags, pkt_size,
 833                                 pkt_size, NULL, 0);
 834         } else {
 835                 vid = AMAP_GET_BITS(struct amap_eth_rx_compl, vlan_tag, rxcp);
 836                 vid = be16_to_cpu(vid);
 837
 838                 if (!adapter->vlan_grp || adapter->num_vlans == 0)
 839                         return;
 840
 841                 lro_vlan_hwaccel_receive_frags(&adapter->rx_obj.lro_mgr,
 842                         rx_frags, pkt_size, pkt_size, adapter->vlan_grp,
 843                         vid, NULL, 0);
 844         }
 845
 846         be_rx_stats_update(adapter, pkt_size, num_rcvd);
 847         return;
 848 }
 849
 850 static struct be_eth_rx_compl *be_rx_compl_get(struct be_adapter *adapter)
 851 {
 852         struct be_eth_rx_compl *rxcp = queue_tail_node(&adapter->rx_obj.cq);
 853
 854         if (rxcp->dw[offsetof(struct amap_eth_rx_compl, valid) / 32] == 0)
 855                 return NULL;
 856
 857         be_dws_le_to_cpu(rxcp, sizeof(*rxcp));
 858
 859         queue_tail_inc(&adapter->rx_obj.cq);
 860         return rxcp;
 861 }
 862
 863 /* To reset the valid bit, we need to reset the whole word as
 864  * when walking the queue the valid entries are little-endian
 865  * and invalid entries are host endian
 866  */
 867 static inline void be_rx_compl_reset(struct be_eth_rx_compl *rxcp)
 868 {
 869         rxcp->dw[offsetof(struct amap_eth_rx_compl, valid) / 32] = 0;
 870 }
 871
 872 static inline struct page *be_alloc_pages(u32 size)
 873 {
 874         gfp_t alloc_flags = GFP_ATOMIC;
 875         u32 order = get_order(size);
 876         if (order > 0)
 877                 alloc_flags |= __GFP_COMP;
 878         return  alloc_pages(alloc_flags, order);
 879 }
 880
 881 /*
 882  * Allocate a page, split it to fragments of size rx_frag_size and post as
 883  * receive buffers to BE
 884  */
 885 static void be_post_rx_frags(struct be_adapter *adapter)
 886 {
 887         struct be_rx_page_info *page_info_tbl = adapter->rx_obj.page_info_tbl;
 888         struct be_rx_page_info *page_info = NULL;
 889         struct be_queue_info *rxq = &adapter->rx_obj.q;
 890         struct page *pagep = NULL;
 891         struct be_eth_rx_d *rxd;
 892         u64 page_dmaaddr = 0, frag_dmaaddr;
 893         u32 posted, page_offset = 0;
 894
 895         page_info = &page_info_tbl[rxq->head];
 896         for (posted = 0; posted < MAX_RX_POST && !page_info->page; posted++) {
 897                 if (!pagep) {
 898                         pagep = be_alloc_pages(adapter->big_page_size);
 899                         if (unlikely(!pagep)) {
 900                                 drvr_stats(adapter)->be_ethrx_post_fail++;
 901                                 break;
 902                         }
 903                         page_dmaaddr = pci_map_page(adapter->pdev, pagep, 0,
 904                                                 adapter->big_page_size,
 905                                                 PCI_DMA_FROMDEVICE);
 906                         page_info->page_offset = 0;
 907                 } else {
 908                         get_page(pagep);
 909                         page_info->page_offset = page_offset + rx_frag_size;
 910                 }
 911                 page_offset = page_info->page_offset;
 912                 page_info->page = pagep;
 913                 pci_unmap_addr_set(page_info, bus, page_dmaaddr);
 914                 frag_dmaaddr = page_dmaaddr + page_info->page_offset;
 915
 916                 rxd = queue_head_node(rxq);
 917                 rxd->fragpa_lo = cpu_to_le32(frag_dmaaddr & 0xFFFFFFFF);
 918                 rxd->fragpa_hi = cpu_to_le32(upper_32_bits(frag_dmaaddr));
 919                 queue_head_inc(rxq);
 920
 921                 /* Any space left in the current big page for another frag? */
 922                 if ((page_offset + rx_frag_size + rx_frag_size) >
 923                                         adapter->big_page_size) {
 924                         pagep = NULL;
 925                         page_info->last_page_user = true;
 926                 }
 927                 page_info = &page_info_tbl[rxq->head];
 928         }
 929         if (pagep)
 930                 page_info->last_page_user = true;
 931
 932         if (posted) {
 933                 atomic_add(posted, &rxq->used);
 934                 be_rxq_notify(&adapter->ctrl, rxq->id, posted);
 935         } else if (atomic_read(&rxq->used) == 0) {
 936                 /* Let be_worker replenish when memory is available */
 937                 adapter->rx_post_starved = true;
 938         }
 939
 940         return;
 941 }
 942
 943 static struct be_eth_tx_compl *be_tx_compl_get(struct be_queue_info *tx_cq)
 944 {
 945         struct be_eth_tx_compl *txcp = queue_tail_node(tx_cq);
 946
 947         if (txcp->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] == 0)
 948                 return NULL;
 949
 950         be_dws_le_to_cpu(txcp, sizeof(*txcp));
 951
 952         txcp->dw[offsetof(struct amap_eth_tx_compl, valid) / 32] = 0;
 953
 954         queue_tail_inc(tx_cq);
 955         return txcp;
 956 }
 957
 958 static void be_tx_compl_process(struct be_adapter *adapter, u16 last_index)
 959 {
 960         struct be_queue_info *txq = &adapter->tx_obj.q;
 961         struct be_eth_wrb *wrb;
 962         struct sk_buff **sent_skbs = adapter->tx_obj.sent_skb_list;
 963         struct sk_buff *sent_skb;
 964         u64 busaddr;
 965         u16 cur_index, num_wrbs = 0;
 966
 967         cur_index = txq->tail;
 968         sent_skb = sent_skbs[cur_index];
 969         BUG_ON(!sent_skb);
 970         sent_skbs[cur_index] = NULL;
 971
 972         do {
 973                 cur_index = txq->tail;
 974                 wrb = queue_tail_node(txq);
 975                 be_dws_le_to_cpu(wrb, sizeof(*wrb));
 976                 busaddr = ((u64)wrb->frag_pa_hi << 32) | (u64)wrb->frag_pa_lo;
 977                 if (busaddr != 0) {
 978                         pci_unmap_single(adapter->pdev, busaddr,
 979                                 wrb->frag_len, PCI_DMA_TODEVICE);
 980                 }
 981                 num_wrbs++;
 982                 queue_tail_inc(txq);
 983         } while (cur_index != last_index);
 984
 985         atomic_sub(num_wrbs, &txq->used);
 986
 987         kfree_skb(sent_skb);
 988 }
 989
 990 static void be_rx_q_clean(struct be_adapter *adapter)
 991 {
 992         struct be_rx_page_info *page_info;
 993         struct be_queue_info *rxq = &adapter->rx_obj.q;
 994         struct be_queue_info *rx_cq = &adapter->rx_obj.cq;
 995         struct be_eth_rx_compl *rxcp;
 996         u16 tail;
 997
 998         /* First cleanup pending rx completions */
 999         while ((rxcp = be_rx_compl_get(adapter)) != NULL) {
1000                 be_rx_compl_discard(adapter, rxcp);
1001                 be_rx_compl_reset(rxcp);
1002                 be_cq_notify(&adapter->ctrl, rx_cq->id, true, 1);
1003         }
1004
1005         /* Then free posted rx buffer that were not used */
1006         tail = (rxq->head + rxq->len - atomic_read(&rxq->used)) % rxq->len;
1007         for (; tail != rxq->head; index_inc(&tail, rxq->len)) {
1008                 page_info = get_rx_page_info(adapter, tail);
1009                 put_page(page_info->page);
1010                 memset(page_info, 0, sizeof(*page_info));
1011         }
1012         BUG_ON(atomic_read(&rxq->used));
1013 }
1014
1015 static void be_tx_q_clean(struct be_adapter *adapter)
1016 {
1017         struct sk_buff **sent_skbs = adapter->tx_obj.sent_skb_list;
1018         struct sk_buff *sent_skb;
1019         struct be_queue_info *txq = &adapter->tx_obj.q;
1020         u16 last_index;
1021         bool dummy_wrb;
1022
1023         while (atomic_read(&txq->used)) {
1024                 sent_skb = sent_skbs[txq->tail];
1025                 last_index = txq->tail;
1026                 index_adv(&last_index,
1027                         wrb_cnt_for_skb(sent_skb, &dummy_wrb) - 1, txq->len);
1028                 be_tx_compl_process(adapter, last_index);
1029         }
1030 }
1031
1032 static void be_mcc_queues_destroy(struct be_adapter *adapter)
1033 {
1034         struct be_queue_info *q;
1035         struct be_ctrl_info *ctrl = &adapter->ctrl;
1036
1037         q = &ctrl->mcc_obj.q;
1038         if (q->created)
1039                 be_cmd_q_destroy(ctrl, q, QTYPE_MCCQ);
1040         be_queue_free(adapter, q);
1041
1042         q = &ctrl->mcc_obj.cq;
1043         if (q->created)
1044                 be_cmd_q_destroy(ctrl, q, QTYPE_CQ);
1045         be_queue_free(adapter, q);
1046 }
1047
1048 /* Must be called only after TX qs are created as MCC shares TX EQ */
1049 static int be_mcc_queues_create(struct be_adapter *adapter)
1050 {
1051         struct be_queue_info *q, *cq;
1052         struct be_ctrl_info *ctrl = &adapter->ctrl;
1053
1054         /* Alloc MCC compl queue */
1055         cq = &ctrl->mcc_obj.cq;
1056         if (be_queue_alloc(adapter, cq, MCC_CQ_LEN,
1057                         sizeof(struct be_mcc_cq_entry)))
1058                 goto err;
1059
1060         /* Ask BE to create MCC compl queue; share TX's eq */
1061         if (be_cmd_cq_create(ctrl, cq, &adapter->tx_eq.q, false, true, 0))
1062                 goto mcc_cq_free;
1063
1064         /* Alloc MCC queue */
1065         q = &ctrl->mcc_obj.q;
1066         if (be_queue_alloc(adapter, q, MCC_Q_LEN, sizeof(struct be_mcc_wrb)))
1067                 goto mcc_cq_destroy;
1068
1069         /* Ask BE to create MCC queue */
1070         if (be_cmd_mccq_create(ctrl, q, cq))
1071                 goto mcc_q_free;
1072
1073         return 0;
1074
1075 mcc_q_free:
1076         be_queue_free(adapter, q);
1077 mcc_cq_destroy:
1078         be_cmd_q_destroy(ctrl, cq, QTYPE_CQ);
1079 mcc_cq_free:
1080         be_queue_free(adapter, cq);
1081 err:
1082         return -1;
1083 }
1084
1085 static void be_tx_queues_destroy(struct be_adapter *adapter)
1086 {
1087         struct be_queue_info *q;
1088
1089         q = &adapter->tx_obj.q;
1090         if (q->created) {
1091                 be_cmd_q_destroy(&adapter->ctrl, q, QTYPE_TXQ);
1092
1093                 /* No more tx completions can be rcvd now; clean up if there
1094                  * are any pending completions or pending tx requests */
1095                 be_tx_q_clean(adapter);
1096         }
1097         be_queue_free(adapter, q);
1098
1099         q = &adapter->tx_obj.cq;
1100         if (q->created)
1101                 be_cmd_q_destroy(&adapter->ctrl, q, QTYPE_CQ);
1102         be_queue_free(adapter, q);
1103
1104         q = &adapter->tx_eq.q;
1105         if (q->created)
1106                 be_cmd_q_destroy(&adapter->ctrl, q, QTYPE_EQ);
1107         be_queue_free(adapter, q);
1108 }
1109
1110 static int be_tx_queues_create(struct be_adapter *adapter)
1111 {
1112         struct be_queue_info *eq, *q, *cq;
1113
1114         adapter->tx_eq.max_eqd = 0;
1115         adapter->tx_eq.min_eqd = 0;
1116         adapter->tx_eq.cur_eqd = 96;
1117         adapter->tx_eq.enable_aic = false;
1118         /* Alloc Tx Event queue */
1119         eq = &adapter->tx_eq.q;
1120         if (be_queue_alloc(adapter, eq, EVNT_Q_LEN, sizeof(struct be_eq_entry)))
1121                 return -1;
1122
1123         /* Ask BE to create Tx Event queue */
1124         if (be_cmd_eq_create(&adapter->ctrl, eq, adapter->tx_eq.cur_eqd))
1125                 goto tx_eq_free;
1126         /* Alloc TX eth compl queue */
1127         cq = &adapter->tx_obj.cq;
1128         if (be_queue_alloc(adapter, cq, TX_CQ_LEN,
1129                         sizeof(struct be_eth_tx_compl)))
1130                 goto tx_eq_destroy;
1131
1132         /* Ask BE to create Tx eth compl queue */
1133         if (be_cmd_cq_create(&adapter->ctrl, cq, eq, false, false, 3))
1134                 goto tx_cq_free;
1135
1136         /* Alloc TX eth queue */
1137         q = &adapter->tx_obj.q;
1138         if (be_queue_alloc(adapter, q, TX_Q_LEN, sizeof(struct be_eth_wrb)))
1139                 goto tx_cq_destroy;
1140
1141         /* Ask BE to create Tx eth queue */
1142         if (be_cmd_txq_create(&adapter->ctrl, q, cq))
1143                 goto tx_q_free;
1144         return 0;
1145
1146 tx_q_free:
1147         be_queue_free(adapter, q);
1148 tx_cq_destroy:
1149         be_cmd_q_destroy(&adapter->ctrl, cq, QTYPE_CQ);
1150 tx_cq_free:
1151         be_queue_free(adapter, cq);
1152 tx_eq_destroy:
1153         be_cmd_q_destroy(&adapter->ctrl, eq, QTYPE_EQ);
1154 tx_eq_free:
1155         be_queue_free(adapter, eq);
1156         return -1;
1157 }
1158
1159 static void be_rx_queues_destroy(struct be_adapter *adapter)
1160 {
1161         struct be_queue_info *q;
1162
1163         q = &adapter->rx_obj.q;
1164         if (q->created) {
1165                 be_cmd_q_destroy(&adapter->ctrl, q, QTYPE_RXQ);
1166                 be_rx_q_clean(adapter);
1167         }
1168         be_queue_free(adapter, q);
1169
1170         q = &adapter->rx_obj.cq;
1171         if (q->created)
1172                 be_cmd_q_destroy(&adapter->ctrl, q, QTYPE_CQ);
1173         be_queue_free(adapter, q);
1174
1175         q = &adapter->rx_eq.q;
1176         if (q->created)
1177                 be_cmd_q_destroy(&adapter->ctrl, q, QTYPE_EQ);
1178         be_queue_free(adapter, q);
1179 }
1180
1181 static int be_rx_queues_create(struct be_adapter *adapter)
1182 {
1183         struct be_queue_info *eq, *q, *cq;
1184         int rc;
1185
1186         adapter->max_rx_coal = BE_MAX_FRAGS_PER_FRAME;
1187         adapter->big_page_size = (1 << get_order(rx_frag_size)) * PAGE_SIZE;
1188         adapter->rx_eq.max_eqd = BE_MAX_EQD;
1189         adapter->rx_eq.min_eqd = 0;
1190         adapter->rx_eq.cur_eqd = 0;
1191         adapter->rx_eq.enable_aic = true;
1192
1193         /* Alloc Rx Event queue */
1194         eq = &adapter->rx_eq.q;
1195         rc = be_queue_alloc(adapter, eq, EVNT_Q_LEN,
1196                                 sizeof(struct be_eq_entry));
1197         if (rc)
1198                 return rc;
1199
1200         /* Ask BE to create Rx Event queue */
1201         rc = be_cmd_eq_create(&adapter->ctrl, eq, adapter->rx_eq.cur_eqd);
1202         if (rc)
1203                 goto rx_eq_free;
1204
1205         /* Alloc RX eth compl queue */
1206         cq = &adapter->rx_obj.cq;
1207         rc = be_queue_alloc(adapter, cq, RX_CQ_LEN,
1208                         sizeof(struct be_eth_rx_compl));
1209         if (rc)
1210                 goto rx_eq_destroy;
1211
1212         /* Ask BE to create Rx eth compl queue */
1213         rc = be_cmd_cq_create(&adapter->ctrl, cq, eq, false, false, 3);
1214         if (rc)
1215                 goto rx_cq_free;
1216
1217         /* Alloc RX eth queue */
1218         q = &adapter->rx_obj.q;
1219         rc = be_queue_alloc(adapter, q, RX_Q_LEN, sizeof(struct be_eth_rx_d));
1220         if (rc)
1221                 goto rx_cq_destroy;
1222
1223         /* Ask BE to create Rx eth queue */
1224         rc = be_cmd_rxq_create(&adapter->ctrl, q, cq->id, rx_frag_size,
1225                 BE_MAX_JUMBO_FRAME_SIZE, adapter->if_handle, false);
1226         if (rc)
1227                 goto rx_q_free;
1228
1229         return 0;
1230 rx_q_free:
1231         be_queue_free(adapter, q);
1232 rx_cq_destroy:
1233         be_cmd_q_destroy(&adapter->ctrl, cq, QTYPE_CQ);
1234 rx_cq_free:
1235         be_queue_free(adapter, cq);
1236 rx_eq_destroy:
1237         be_cmd_q_destroy(&adapter->ctrl, eq, QTYPE_EQ);
1238 rx_eq_free:
1239         be_queue_free(adapter, eq);
1240         return rc;
1241 }
1242 static bool event_get(struct be_eq_obj *eq_obj, u16 *rid)
1243 {
1244         struct be_eq_entry *entry = queue_tail_node(&eq_obj->q);
1245         u32 evt = entry->evt;
1246
1247         if (!evt)
1248                 return false;
1249
1250         evt = le32_to_cpu(evt);
1251         *rid = (evt >> EQ_ENTRY_RES_ID_SHIFT) & EQ_ENTRY_RES_ID_MASK;
1252         entry->evt = 0;
1253         queue_tail_inc(&eq_obj->q);
1254         return true;
1255 }
1256
1257 static int event_handle(struct be_ctrl_info *ctrl,
1258                         struct be_eq_obj *eq_obj)
1259 {
1260         u16 rid = 0, num = 0;
1261
1262         while (event_get(eq_obj, &rid))
1263                 num++;
1264
1265         /* We can see an interrupt and no event */
1266         be_eq_notify(ctrl, eq_obj->q.id, true, true, num);
1267         if (num)
1268                 napi_schedule(&eq_obj->napi);
1269
1270         return num;
1271 }
1272
1273 static irqreturn_t be_intx(int irq, void *dev)
1274 {
1275         struct be_adapter *adapter = dev;
1276         struct be_ctrl_info *ctrl = &adapter->ctrl;
1277         int rx, tx;
1278
1279         tx = event_handle(ctrl, &adapter->tx_eq);
1280         rx = event_handle(ctrl, &adapter->rx_eq);
1281
1282         if (rx || tx)
1283                 return IRQ_HANDLED;
1284         else
1285                 return IRQ_NONE;
1286 }
1287
1288 static irqreturn_t be_msix_rx(int irq, void *dev)
1289 {
1290         struct be_adapter *adapter = dev;
1291
1292         event_handle(&adapter->ctrl, &adapter->rx_eq);
1293
1294         return IRQ_HANDLED;
1295 }
1296
1297 static irqreturn_t be_msix_tx_mcc(int irq, void *dev)
1298 {
1299         struct be_adapter *adapter = dev;
1300
1301         event_handle(&adapter->ctrl, &adapter->tx_eq);
1302
1303         return IRQ_HANDLED;
1304 }
1305
1306 static inline bool do_lro(struct be_adapter *adapter,
1307                         struct be_eth_rx_compl *rxcp)
1308 {
1309         int err = AMAP_GET_BITS(struct amap_eth_rx_compl, err, rxcp);
1310         int tcp_frame = AMAP_GET_BITS(struct amap_eth_rx_compl, tcpf, rxcp);
1311
1312         if (err)
1313                 drvr_stats(adapter)->be_rxcp_err++;
1314
1315         return (!tcp_frame || err || (adapter->max_rx_coal <= 1)) ?
1316                 false : true;
1317 }
1318
1319 int be_poll_rx(struct napi_struct *napi, int budget)
1320 {
1321         struct be_eq_obj *rx_eq = container_of(napi, struct be_eq_obj, napi);
1322         struct be_adapter *adapter =
1323                 container_of(rx_eq, struct be_adapter, rx_eq);
1324         struct be_queue_info *rx_cq = &adapter->rx_obj.cq;
1325         struct be_eth_rx_compl *rxcp;
1326         u32 work_done;
1327
1328         for (work_done = 0; work_done < budget; work_done++) {
1329                 rxcp = be_rx_compl_get(adapter);
1330                 if (!rxcp)
1331                         break;
1332
1333                 if (do_lro(adapter, rxcp))
1334                         be_rx_compl_process_lro(adapter, rxcp);
1335                 else
1336                         be_rx_compl_process(adapter, rxcp);
1337
1338                 be_rx_compl_reset(rxcp);
1339         }
1340
1341         lro_flush_all(&adapter->rx_obj.lro_mgr);
1342
1343         /* Refill the queue */
1344         if (atomic_read(&adapter->rx_obj.q.used) < RX_FRAGS_REFILL_WM)
1345                 be_post_rx_frags(adapter);
1346
1347         /* All consumed */
1348         if (work_done < budget) {
1349                 napi_complete(napi);
1350                 be_cq_notify(&adapter->ctrl, rx_cq->id, true, work_done);
1351         } else {
1352                 /* More to be consumed; continue with interrupts disabled */
1353                 be_cq_notify(&adapter->ctrl, rx_cq->id, false, work_done);
1354         }
1355         return work_done;
1356 }
1357
1358 void be_process_tx(struct be_adapter *adapter)
1359 {
1360         struct be_queue_info *txq = &adapter->tx_obj.q;
1361         struct be_queue_info *tx_cq = &adapter->tx_obj.cq;
1362         struct be_eth_tx_compl *txcp;
1363         u32 num_cmpl = 0;
1364         u16 end_idx;
1365
1366         while ((txcp = be_tx_compl_get(tx_cq))) {
1367                 end_idx = AMAP_GET_BITS(struct amap_eth_tx_compl,
1368                                         wrb_index, txcp);
1369                 be_tx_compl_process(adapter, end_idx);
1370                 num_cmpl++;
1371         }
1372
1373         if (num_cmpl) {
1374                 be_cq_notify(&adapter->ctrl, tx_cq->id, true, num_cmpl);
1375
1376                 /* As Tx wrbs have been freed up, wake up netdev queue if
1377                  * it was stopped due to lack of tx wrbs.
1378                  */
1379                 if (netif_queue_stopped(adapter->netdev) &&
1380                         atomic_read(&txq->used) < txq->len / 2) {
1381                         netif_wake_queue(adapter->netdev);
1382                 }
1383
1384                 drvr_stats(adapter)->be_tx_events++;
1385                 drvr_stats(adapter)->be_tx_compl += num_cmpl;
1386         }
1387 }
1388
1389 /* As TX and MCC share the same EQ check for both TX and MCC completions.
1390  * For TX/MCC we don't honour budget; consume everything
1391  */
1392 static int be_poll_tx_mcc(struct napi_struct *napi, int budget)
1393 {
1394         struct be_eq_obj *tx_eq = container_of(napi, struct be_eq_obj, napi);
1395         struct be_adapter *adapter =
1396                 container_of(tx_eq, struct be_adapter, tx_eq);
1397
1398         napi_complete(napi);
1399
1400         be_process_tx(adapter);
1401
1402         be_process_mcc(&adapter->ctrl);
1403
1404         return 1;
1405 }
1406
1407 static void be_worker(struct work_struct *work)
1408 {
1409         struct be_adapter *adapter =
1410                 container_of(work, struct be_adapter, work.work);
1411         int status;
1412
1413         /* Get Stats */
1414         status = be_cmd_get_stats(&adapter->ctrl, &adapter->stats.cmd);
1415         if (!status)
1416                 netdev_stats_update(adapter);
1417
1418         /* Set EQ delay */
1419         be_rx_eqd_update(adapter);
1420
1421         be_tx_rate_update(adapter);
1422         be_rx_rate_update(adapter);
1423
1424         if (adapter->rx_post_starved) {
1425                 adapter->rx_post_starved = false;
1426                 be_post_rx_frags(adapter);
1427         }
1428
1429         schedule_delayed_work(&adapter->work, msecs_to_jiffies(1000));
1430 }
1431
1432 static void be_msix_enable(struct be_adapter *adapter)
1433 {
1434         int i, status;
1435
1436         for (i = 0; i < BE_NUM_MSIX_VECTORS; i++)
1437                 adapter->msix_entries[i].entry = i;
1438
1439         status = pci_enable_msix(adapter->pdev, adapter->msix_entries,
1440                 BE_NUM_MSIX_VECTORS);
1441         if (status == 0)
1442                 adapter->msix_enabled = true;
1443         return;
1444 }
1445
1446 static inline int be_msix_vec_get(struct be_adapter *adapter, u32 eq_id)
1447 {
1448         return adapter->msix_entries[eq_id -
1449                         8 * adapter->ctrl.pci_func].vector;
1450 }
1451
1452 static int be_msix_register(struct be_adapter *adapter)
1453 {
1454         struct net_device *netdev = adapter->netdev;
1455         struct be_eq_obj *tx_eq = &adapter->tx_eq;
1456         struct be_eq_obj *rx_eq = &adapter->rx_eq;
1457         int status, vec;
1458
1459         sprintf(tx_eq->desc, "%s-tx", netdev->name);
1460         vec = be_msix_vec_get(adapter, tx_eq->q.id);
1461         status = request_irq(vec, be_msix_tx_mcc, 0, tx_eq->desc, adapter);
1462         if (status)
1463                 goto err;
1464
1465         sprintf(rx_eq->desc, "%s-rx", netdev->name);
1466         vec = be_msix_vec_get(adapter, rx_eq->q.id);
1467         status = request_irq(vec, be_msix_rx, 0, rx_eq->desc, adapter);
1468         if (status) { /* Free TX IRQ */
1469                 vec = be_msix_vec_get(adapter, tx_eq->q.id);
1470                 free_irq(vec, adapter);
1471                 goto err;
1472         }
1473         return 0;
1474 err:
1475         dev_warn(&adapter->pdev->dev,
1476                 "MSIX Request IRQ failed - err %d\n", status);
1477         pci_disable_msix(adapter->pdev);
1478         adapter->msix_enabled = false;
1479         return status;
1480 }
1481
1482 static int be_irq_register(struct be_adapter *adapter)
1483 {
1484         struct net_device *netdev = adapter->netdev;
1485         int status;
1486
1487         if (adapter->msix_enabled) {
1488                 status = be_msix_register(adapter);
1489                 if (status == 0)
1490                         goto done;
1491         }
1492
1493         /* INTx */
1494         netdev->irq = adapter->pdev->irq;
1495         status = request_irq(netdev->irq, be_intx, IRQF_SHARED, netdev->name,
1496                         adapter);
1497         if (status) {
1498                 dev_err(&adapter->pdev->dev,
1499                         "INTx request IRQ failed - err %d\n", status);
1500                 return status;
1501         }
1502 done:
1503         adapter->isr_registered = true;
1504         return 0;
1505 }
1506
1507 static void be_irq_unregister(struct be_adapter *adapter)
1508 {
1509         struct net_device *netdev = adapter->netdev;
1510         int vec;
1511
1512         if (!adapter->isr_registered)
1513                 return;
1514
1515         /* INTx */
1516         if (!adapter->msix_enabled) {
1517                 free_irq(netdev->irq, adapter);
1518                 goto done;
1519         }
1520
1521         /* MSIx */
1522         vec = be_msix_vec_get(adapter, adapter->tx_eq.q.id);
1523         free_irq(vec, adapter);
1524         vec = be_msix_vec_get(adapter, adapter->rx_eq.q.id);
1525         free_irq(vec, adapter);
1526 done:
1527         adapter->isr_registered = false;
1528         return;
1529 }
1530
1531 static int be_open(struct net_device *netdev)
1532 {
1533         struct be_adapter *adapter = netdev_priv(netdev);
1534         struct be_ctrl_info *ctrl = &adapter->ctrl;
1535         struct be_eq_obj *rx_eq = &adapter->rx_eq;
1536         struct be_eq_obj *tx_eq = &adapter->tx_eq;
1537         bool link_up;
1538         int status;
1539
1540         /* First time posting */
1541         be_post_rx_frags(adapter);
1542
1543         napi_enable(&rx_eq->napi);
1544         napi_enable(&tx_eq->napi);
1545
1546         be_irq_register(adapter);
1547
1548         be_intr_set(ctrl, true);
1549
1550         /* The evt queues are created in unarmed state; arm them */
1551         be_eq_notify(ctrl, rx_eq->q.id, true, false, 0);
1552         be_eq_notify(ctrl, tx_eq->q.id, true, false, 0);
1553
1554         /* Rx compl queue may be in unarmed state; rearm it */
1555         be_cq_notify(ctrl, adapter->rx_obj.cq.id, true, 0);
1556
1557         status = be_cmd_link_status_query(ctrl, &link_up);
1558         if (status)
1559                 return status;
1560         be_link_status_update(adapter, link_up);
1561
1562         schedule_delayed_work(&adapter->work, msecs_to_jiffies(100));
1563         return 0;
1564 }
1565
1566 static int be_setup(struct be_adapter *adapter)
1567 {
1568         struct be_ctrl_info *ctrl = &adapter->ctrl;
1569         struct net_device *netdev = adapter->netdev;
1570         u32 if_flags;
1571         int status;
1572
1573         if_flags = BE_IF_FLAGS_BROADCAST | BE_IF_FLAGS_PROMISCUOUS |
1574                 BE_IF_FLAGS_MCAST_PROMISCUOUS | BE_IF_FLAGS_UNTAGGED |
1575                 BE_IF_FLAGS_PASS_L3L4_ERRORS;
1576         status = be_cmd_if_create(ctrl, if_flags, netdev->dev_addr,
1577                         false/* pmac_invalid */, &adapter->if_handle,
1578                         &adapter->pmac_id);
1579         if (status != 0)
1580                 goto do_none;
1581
1582         be_vid_config(netdev);
1583
1584         status = be_cmd_set_flow_control(ctrl, true, true);
1585         if (status != 0)
1586                 goto if_destroy;
1587
1588         status = be_tx_queues_create(adapter);
1589         if (status != 0)
1590                 goto if_destroy;
1591
1592         status = be_rx_queues_create(adapter);
1593         if (status != 0)
1594                 goto tx_qs_destroy;
1595
1596         status = be_mcc_queues_create(adapter);
1597         if (status != 0)
1598                 goto rx_qs_destroy;
1599
1600         return 0;
1601
1602 rx_qs_destroy:
1603         be_rx_queues_destroy(adapter);
1604 tx_qs_destroy:
1605         be_tx_queues_destroy(adapter);
1606 if_destroy:
1607         be_cmd_if_destroy(ctrl, adapter->if_handle);
1608 do_none:
1609         return status;
1610 }
1611
1612 static int be_clear(struct be_adapter *adapter)
1613 {
1614         struct be_ctrl_info *ctrl = &adapter->ctrl;
1615
1616         be_rx_queues_destroy(adapter);
1617         be_tx_queues_destroy(adapter);
1618
1619         be_cmd_if_destroy(ctrl, adapter->if_handle);
1620
1621         be_mcc_queues_destroy(adapter);
1622         return 0;
1623 }
1624
1625 static int be_close(struct net_device *netdev)
1626 {
1627         struct be_adapter *adapter = netdev_priv(netdev);
1628         struct be_ctrl_info *ctrl = &adapter->ctrl;
1629         struct be_eq_obj *rx_eq = &adapter->rx_eq;
1630         struct be_eq_obj *tx_eq = &adapter->tx_eq;
1631         int vec;
1632
1633         cancel_delayed_work_sync(&adapter->work);
1634
1635         netif_stop_queue(netdev);
1636         netif_carrier_off(netdev);
1637         adapter->link_up = false;
1638
1639         be_intr_set(ctrl, false);
1640
1641         if (adapter->msix_enabled) {
1642                 vec = be_msix_vec_get(adapter, tx_eq->q.id);
1643                 synchronize_irq(vec);
1644                 vec = be_msix_vec_get(adapter, rx_eq->q.id);
1645                 synchronize_irq(vec);
1646         } else {
1647                 synchronize_irq(netdev->irq);
1648         }
1649         be_irq_unregister(adapter);
1650
1651         napi_disable(&rx_eq->napi);
1652         napi_disable(&tx_eq->napi);
1653
1654         return 0;
1655 }
1656
1657 static int be_get_frag_header(struct skb_frag_struct *frag, void **mac_hdr,
1658                                 void **ip_hdr, void **tcpudp_hdr,
1659                                 u64 *hdr_flags, void *priv)
1660 {
1661         struct ethhdr *eh;
1662         struct vlan_ethhdr *veh;
1663         struct iphdr *iph;
1664         u8 *va = page_address(frag->page) + frag->page_offset;
1665         unsigned long ll_hlen;
1666
1667         prefetch(va);
1668         eh = (struct ethhdr *)va;
1669         *mac_hdr = eh;
1670         ll_hlen = ETH_HLEN;
1671         if (eh->h_proto != htons(ETH_P_IP)) {
1672                 if (eh->h_proto == htons(ETH_P_8021Q)) {
1673                         veh = (struct vlan_ethhdr *)va;
1674                         if (veh->h_vlan_encapsulated_proto != htons(ETH_P_IP))
1675                                 return -1;
1676
1677                         ll_hlen += VLAN_HLEN;
1678                 } else {
1679                         return -1;
1680                 }
1681         }
1682         *hdr_flags = LRO_IPV4;
1683         iph = (struct iphdr *)(va + ll_hlen);
1684         *ip_hdr = iph;
1685         if (iph->protocol != IPPROTO_TCP)
1686                 return -1;
1687         *hdr_flags |= LRO_TCP;
1688         *tcpudp_hdr = (u8 *) (*ip_hdr) + (iph->ihl << 2);
1689
1690         return 0;
1691 }
1692
1693 static void be_lro_init(struct be_adapter *adapter, struct net_device *netdev)
1694 {
1695         struct net_lro_mgr *lro_mgr;
1696
1697         lro_mgr = &adapter->rx_obj.lro_mgr;
1698         lro_mgr->dev = netdev;
1699         lro_mgr->features = LRO_F_NAPI;
1700         lro_mgr->ip_summed = CHECKSUM_UNNECESSARY;
1701         lro_mgr->ip_summed_aggr = CHECKSUM_UNNECESSARY;
1702         lro_mgr->max_desc = BE_MAX_LRO_DESCRIPTORS;
1703         lro_mgr->lro_arr = adapter->rx_obj.lro_desc;
1704         lro_mgr->get_frag_header = be_get_frag_header;
1705         lro_mgr->max_aggr = BE_MAX_FRAGS_PER_FRAME;
1706 }
1707
1708 static struct net_device_ops be_netdev_ops = {
1709         .ndo_open               = be_open,
1710         .ndo_stop               = be_close,
1711         .ndo_start_xmit         = be_xmit,
1712         .ndo_get_stats          = be_get_stats,
1713         .ndo_set_rx_mode        = be_set_multicast_list,
1714         .ndo_set_mac_address    = be_mac_addr_set,
1715         .ndo_change_mtu         = be_change_mtu,
1716         .ndo_validate_addr      = eth_validate_addr,
1717         .ndo_vlan_rx_register   = be_vlan_register,
1718         .ndo_vlan_rx_add_vid    = be_vlan_add_vid,
1719         .ndo_vlan_rx_kill_vid   = be_vlan_rem_vid,
1720 };
1721
1722 static void be_netdev_init(struct net_device *netdev)
1723 {
1724         struct be_adapter *adapter = netdev_priv(netdev);
1725
1726         netdev->features |= NETIF_F_SG | NETIF_F_HW_VLAN_RX | NETIF_F_TSO |
1727                 NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_FILTER | NETIF_F_IP_CSUM |
1728                 NETIF_F_IPV6_CSUM;
1729
1730         netdev->flags |= IFF_MULTICAST;
1731
1732         adapter->rx_csum = true;
1733
1734         BE_SET_NETDEV_OPS(netdev, &be_netdev_ops);
1735
1736         SET_ETHTOOL_OPS(netdev, &be_ethtool_ops);
1737
1738         be_lro_init(adapter, netdev);
1739
1740         netif_napi_add(netdev, &adapter->rx_eq.napi, be_poll_rx,
1741                 BE_NAPI_WEIGHT);
1742         netif_napi_add(netdev, &adapter->tx_eq.napi, be_poll_tx_mcc,
1743                 BE_NAPI_WEIGHT);
1744
1745         netif_carrier_off(netdev);
1746         netif_stop_queue(netdev);
1747 }
1748
1749 static void be_unmap_pci_bars(struct be_adapter *adapter)
1750 {
1751         struct be_ctrl_info *ctrl = &adapter->ctrl;
1752         if (ctrl->csr)
1753                 iounmap(ctrl->csr);
1754         if (ctrl->db)
1755                 iounmap(ctrl->db);
1756         if (ctrl->pcicfg)
1757                 iounmap(ctrl->pcicfg);
1758 }
1759
1760 static int be_map_pci_bars(struct be_adapter *adapter)
1761 {
1762         u8 __iomem *addr;
1763
1764         addr = ioremap_nocache(pci_resource_start(adapter->pdev, 2),
1765                         pci_resource_len(adapter->pdev, 2));
1766         if (addr == NULL)
1767                 return -ENOMEM;
1768         adapter->ctrl.csr = addr;
1769
1770         addr = ioremap_nocache(pci_resource_start(adapter->pdev, 4),
1771                         128 * 1024);
1772         if (addr == NULL)
1773                 goto pci_map_err;
1774         adapter->ctrl.db = addr;
1775
1776         addr = ioremap_nocache(pci_resource_start(adapter->pdev, 1),
1777                         pci_resource_len(adapter->pdev, 1));
1778         if (addr == NULL)
1779                 goto pci_map_err;
1780         adapter->ctrl.pcicfg = addr;
1781
1782         return 0;
1783 pci_map_err:
1784         be_unmap_pci_bars(adapter);
1785         return -ENOMEM;
1786 }
1787
1788
1789 static void be_ctrl_cleanup(struct be_adapter *adapter)
1790 {
1791         struct be_dma_mem *mem = &adapter->ctrl.mbox_mem_alloced;
1792
1793         be_unmap_pci_bars(adapter);
1794
1795         if (mem->va)
1796                 pci_free_consistent(adapter->pdev, mem->size,
1797                         mem->va, mem->dma);
1798 }
1799
1800 /* Initialize the mbox required to send cmds to BE */
1801 static int be_ctrl_init(struct be_adapter *adapter)
1802 {
1803         struct be_ctrl_info *ctrl = &adapter->ctrl;
1804         struct be_dma_mem *mbox_mem_alloc = &ctrl->mbox_mem_alloced;
1805         struct be_dma_mem *mbox_mem_align = &ctrl->mbox_mem;
1806         int status;
1807         u32 val;
1808
1809         status = be_map_pci_bars(adapter);
1810         if (status)
1811                 return status;
1812
1813         mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
1814         mbox_mem_alloc->va = pci_alloc_consistent(adapter->pdev,
1815                                 mbox_mem_alloc->size, &mbox_mem_alloc->dma);
1816         if (!mbox_mem_alloc->va) {
1817                 be_unmap_pci_bars(adapter);
1818                 return -1;
1819         }
1820         mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
1821         mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
1822         mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
1823         memset(mbox_mem_align->va, 0, sizeof(struct be_mcc_mailbox));
1824         spin_lock_init(&ctrl->mbox_lock);
1825         spin_lock_init(&ctrl->mcc_lock);
1826         spin_lock_init(&ctrl->mcc_cq_lock);
1827
1828         ctrl->async_cb = be_link_status_update;
1829         ctrl->adapter_ctxt = adapter;
1830
1831         val = ioread32(ctrl->pcicfg + PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET);
1832         ctrl->pci_func = (val >> MEMBAR_CTRL_INT_CTRL_PFUNC_SHIFT) &
1833                                         MEMBAR_CTRL_INT_CTRL_PFUNC_MASK;
1834         return 0;
1835 }
1836
1837 static void be_stats_cleanup(struct be_adapter *adapter)
1838 {
1839         struct be_stats_obj *stats = &adapter->stats;
1840         struct be_dma_mem *cmd = &stats->cmd;
1841
1842         if (cmd->va)
1843                 pci_free_consistent(adapter->pdev, cmd->size,
1844                         cmd->va, cmd->dma);
1845 }
1846
1847 static int be_stats_init(struct be_adapter *adapter)
1848 {
1849         struct be_stats_obj *stats = &adapter->stats;
1850         struct be_dma_mem *cmd = &stats->cmd;
1851
1852         cmd->size = sizeof(struct be_cmd_req_get_stats);
1853         cmd->va = pci_alloc_consistent(adapter->pdev, cmd->size, &cmd->dma);
1854         if (cmd->va == NULL)
1855                 return -1;
1856         return 0;
1857 }
1858
1859 static void __devexit be_remove(struct pci_dev *pdev)
1860 {
1861         struct be_adapter *adapter = pci_get_drvdata(pdev);
1862         if (!adapter)
1863                 return;
1864
1865         unregister_netdev(adapter->netdev);
1866
1867         be_clear(adapter);
1868
1869         be_stats_cleanup(adapter);
1870
1871         be_ctrl_cleanup(adapter);
1872
1873         if (adapter->msix_enabled) {
1874                 pci_disable_msix(adapter->pdev);
1875                 adapter->msix_enabled = false;
1876         }
1877
1878         pci_set_drvdata(pdev, NULL);
1879         pci_release_regions(pdev);
1880         pci_disable_device(pdev);
1881
1882         free_netdev(adapter->netdev);
1883 }
1884
1885 static int be_hw_up(struct be_adapter *adapter)
1886 {
1887         struct be_ctrl_info *ctrl = &adapter->ctrl;
1888         int status;
1889
1890         status = be_cmd_POST(ctrl);
1891         if (status)
1892                 return status;
1893
1894         status = be_cmd_get_fw_ver(ctrl, adapter->fw_ver);
1895         if (status)
1896                 return status;
1897
1898         status = be_cmd_query_fw_cfg(ctrl, &adapter->port_num);
1899         return status;
1900 }
1901
1902 static int __devinit be_probe(struct pci_dev *pdev,
1903                         const struct pci_device_id *pdev_id)
1904 {
1905         int status = 0;
1906         struct be_adapter *adapter;
1907         struct net_device *netdev;
1908         struct be_ctrl_info *ctrl;
1909         u8 mac[ETH_ALEN];
1910
1911         status = pci_enable_device(pdev);
1912         if (status)
1913                 goto do_none;
1914
1915         status = pci_request_regions(pdev, DRV_NAME);
1916         if (status)
1917                 goto disable_dev;
1918         pci_set_master(pdev);
1919
1920         netdev = alloc_etherdev(sizeof(struct be_adapter));
1921         if (netdev == NULL) {
1922                 status = -ENOMEM;
1923                 goto rel_reg;
1924         }
1925         adapter = netdev_priv(netdev);
1926         adapter->pdev = pdev;
1927         pci_set_drvdata(pdev, adapter);
1928         adapter->netdev = netdev;
1929
1930         be_msix_enable(adapter);
1931
1932         status = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1933         if (!status) {
1934                 netdev->features |= NETIF_F_HIGHDMA;
1935         } else {
1936                 status = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1937                 if (status) {
1938                         dev_err(&pdev->dev, "Could not set PCI DMA Mask\n");
1939                         goto free_netdev;
1940                 }
1941         }
1942
1943         ctrl = &adapter->ctrl;
1944         status = be_ctrl_init(adapter);
1945         if (status)
1946                 goto free_netdev;
1947
1948         status = be_stats_init(adapter);
1949         if (status)
1950                 goto ctrl_clean;
1951
1952         status = be_hw_up(adapter);
1953         if (status)
1954                 goto stats_clean;
1955
1956         status = be_cmd_mac_addr_query(ctrl, mac, MAC_ADDRESS_TYPE_NETWORK,
1957                         true /* permanent */, 0);
1958         if (status)
1959                 goto stats_clean;
1960         memcpy(netdev->dev_addr, mac, ETH_ALEN);
1961
1962         INIT_DELAYED_WORK(&adapter->work, be_worker);
1963         be_netdev_init(netdev);
1964         SET_NETDEV_DEV(netdev, &adapter->pdev->dev);
1965
1966         status = be_setup(adapter);
1967         if (status)
1968                 goto stats_clean;
1969         status = register_netdev(netdev);
1970         if (status != 0)
1971                 goto unsetup;
1972
1973         dev_info(&pdev->dev, "%s port %d\n", nic_name(pdev), adapter->port_num);
1974         return 0;
1975
1976 unsetup:
1977         be_clear(adapter);
1978 stats_clean:
1979         be_stats_cleanup(adapter);
1980 ctrl_clean:
1981         be_ctrl_cleanup(adapter);
1982 free_netdev:
1983         free_netdev(adapter->netdev);
1984 rel_reg:
1985         pci_release_regions(pdev);
1986 disable_dev:
1987         pci_disable_device(pdev);
1988 do_none:
1989         dev_err(&pdev->dev, "%s initialization failed\n", nic_name(pdev));
1990         return status;
1991 }
1992
1993 static int be_suspend(struct pci_dev *pdev, pm_message_t state)
1994 {
1995         struct be_adapter *adapter = pci_get_drvdata(pdev);
1996         struct net_device *netdev =  adapter->netdev;
1997
1998         netif_device_detach(netdev);
1999         if (netif_running(netdev)) {
2000                 rtnl_lock();
2001                 be_close(netdev);
2002                 be_clear(adapter);
2003                 rtnl_unlock();
2004         }
2005
2006         pci_save_state(pdev);
2007         pci_disable_device(pdev);
2008         pci_set_power_state(pdev, pci_choose_state(pdev, state));
2009         return 0;
2010 }
2011
2012 static int be_resume(struct pci_dev *pdev)
2013 {
2014         int status = 0;
2015         struct be_adapter *adapter = pci_get_drvdata(pdev);
2016         struct net_device *netdev =  adapter->netdev;
2017
2018         netif_device_detach(netdev);
2019
2020         status = pci_enable_device(pdev);
2021         if (status)
2022                 return status;
2023
2024         pci_set_power_state(pdev, 0);
2025         pci_restore_state(pdev);
2026
2027         if (netif_running(netdev)) {
2028                 rtnl_lock();
2029                 be_setup(adapter);
2030                 be_open(netdev);
2031                 rtnl_unlock();
2032         }
2033         netif_device_attach(netdev);
2034         return 0;
2035 }
2036
2037 static struct pci_driver be_driver = {
2038         .name = DRV_NAME,
2039         .id_table = be_dev_ids,
2040         .probe = be_probe,
2041         .remove = be_remove,
2042         .suspend = be_suspend,
2043         .resume = be_resume
2044 };
2045
2046 static int __init be_init_module(void)
2047 {
2048         if (rx_frag_size != 8192 && rx_frag_size != 4096
2049                 && rx_frag_size != 2048) {
2050                 printk(KERN_WARNING DRV_NAME
2051                         " : Module param rx_frag_size must be 2048/4096/8192."
2052                         " Using 2048\n");
2053                 rx_frag_size = 2048;
2054         }
2055         /* Ensure rx_frag_size is aligned to chache line */
2056         if (SKB_DATA_ALIGN(rx_frag_size) != rx_frag_size) {
2057                 printk(KERN_WARNING DRV_NAME
2058                         " : Bad module param rx_frag_size. Using 2048\n");
2059                 rx_frag_size = 2048;
2060         }
2061
2062         return pci_register_driver(&be_driver);
2063 }
2064 module_init(be_init_module);
2065
2066 static void __exit be_exit_module(void)
2067 {
2068         pci_unregister_driver(&be_driver);
2069 }
2070 module_exit(be_exit_module);