drivers/net/cxgb3/cxgb3_main.c

   1 /*
   2  * Copyright (c) 2003-2008 Chelsio, Inc. All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32 #include <linux/module.h>
  33 #include <linux/moduleparam.h>
  34 #include <linux/init.h>
  35 #include <linux/pci.h>
  36 #include <linux/dma-mapping.h>
  37 #include <linux/netdevice.h>
  38 #include <linux/etherdevice.h>
  39 #include <linux/if_vlan.h>
  40 #include <linux/mdio.h>
  41 #include <linux/sockios.h>
  42 #include <linux/workqueue.h>
  43 #include <linux/proc_fs.h>
  44 #include <linux/rtnetlink.h>
  45 #include <linux/firmware.h>
  46 #include <linux/log2.h>
  47 #include <linux/stringify.h>
  48 #include <asm/uaccess.h>
  49
  50 #include "common.h"
  51 #include "cxgb3_ioctl.h"
  52 #include "regs.h"
  53 #include "cxgb3_offload.h"
  54 #include "version.h"
  55
  56 #include "cxgb3_ctl_defs.h"
  57 #include "t3_cpl.h"
  58 #include "firmware_exports.h"
  59
  60 enum {
  61         MAX_TXQ_ENTRIES = 16384,
  62         MAX_CTRL_TXQ_ENTRIES = 1024,
  63         MAX_RSPQ_ENTRIES = 16384,
  64         MAX_RX_BUFFERS = 16384,
  65         MAX_RX_JUMBO_BUFFERS = 16384,
  66         MIN_TXQ_ENTRIES = 4,
  67         MIN_CTRL_TXQ_ENTRIES = 4,
  68         MIN_RSPQ_ENTRIES = 32,
  69         MIN_FL_ENTRIES = 32
  70 };
  71
  72 #define PORT_MASK ((1 << MAX_NPORTS) - 1)
  73
  74 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
  75                          NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
  76                          NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
  77
  78 #define EEPROM_MAGIC 0x38E2F10C
  79
  80 #define CH_DEVICE(devid, idx) \
  81         { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, idx }
  82
  83 static const struct pci_device_id cxgb3_pci_tbl[] = {
  84         CH_DEVICE(0x20, 0),     /* PE9000 */
  85         CH_DEVICE(0x21, 1),     /* T302E */
  86         CH_DEVICE(0x22, 2),     /* T310E */
  87         CH_DEVICE(0x23, 3),     /* T320X */
  88         CH_DEVICE(0x24, 1),     /* T302X */
  89         CH_DEVICE(0x25, 3),     /* T320E */
  90         CH_DEVICE(0x26, 2),     /* T310X */
  91         CH_DEVICE(0x30, 2),     /* T3B10 */
  92         CH_DEVICE(0x31, 3),     /* T3B20 */
  93         CH_DEVICE(0x32, 1),     /* T3B02 */
  94         CH_DEVICE(0x35, 6),     /* T3C20-derived T3C10 */
  95         CH_DEVICE(0x36, 3),     /* S320E-CR */
  96         CH_DEVICE(0x37, 7),     /* N320E-G2 */
  97         {0,}
  98 };
  99
 100 MODULE_DESCRIPTION(DRV_DESC);
 101 MODULE_AUTHOR("Chelsio Communications");
 102 MODULE_LICENSE("Dual BSD/GPL");
 103 MODULE_VERSION(DRV_VERSION);
 104 MODULE_DEVICE_TABLE(pci, cxgb3_pci_tbl);
 105
 106 static int dflt_msg_enable = DFLT_MSG_ENABLE;
 107
 108 module_param(dflt_msg_enable, int, 0644);
 109 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T3 default message enable bitmap");
 110
 111 /*
 112  * The driver uses the best interrupt scheme available on a platform in the
 113  * order MSI-X, MSI, legacy pin interrupts.  This parameter determines which
 114  * of these schemes the driver may consider as follows:
 115  *
 116  * msi = 2: choose from among all three options
 117  * msi = 1: only consider MSI and pin interrupts
 118  * msi = 0: force pin interrupts
 119  */
 120 static int msi = 2;
 121
 122 module_param(msi, int, 0644);
 123 MODULE_PARM_DESC(msi, "whether to use MSI or MSI-X");
 124
 125 /*
 126  * The driver enables offload as a default.
 127  * To disable it, use ofld_disable = 1.
 128  */
 129
 130 static int ofld_disable = 0;
 131
 132 module_param(ofld_disable, int, 0644);
 133 MODULE_PARM_DESC(ofld_disable, "whether to enable offload at init time or not");
 134
 135 /*
 136  * We have work elements that we need to cancel when an interface is taken
 137  * down.  Normally the work elements would be executed by keventd but that
 138  * can deadlock because of linkwatch.  If our close method takes the rtnl
 139  * lock and linkwatch is ahead of our work elements in keventd, linkwatch
 140  * will block keventd as it needs the rtnl lock, and we'll deadlock waiting
 141  * for our work to complete.  Get our own work queue to solve this.
 142  */
 143 static struct workqueue_struct *cxgb3_wq;
 144
 145 /**
 146  *      link_report - show link status and link speed/duplex
 147  *      @p: the port whose settings are to be reported
 148  *
 149  *      Shows the link status, speed, and duplex of a port.
 150  */
 151 static void link_report(struct net_device *dev)
 152 {
 153         if (!netif_carrier_ok(dev))
 154                 printk(KERN_INFO "%s: link down\n", dev->name);
 155         else {
 156                 const char *s = "10Mbps";
 157                 const struct port_info *p = netdev_priv(dev);
 158
 159                 switch (p->link_config.speed) {
 160                 case SPEED_10000:
 161                         s = "10Gbps";
 162                         break;
 163                 case SPEED_1000:
 164                         s = "1000Mbps";
 165                         break;
 166                 case SPEED_100:
 167                         s = "100Mbps";
 168                         break;
 169                 }
 170
 171                 printk(KERN_INFO "%s: link up, %s, %s-duplex\n", dev->name, s,
 172                        p->link_config.duplex == DUPLEX_FULL ? "full" : "half");
 173         }
 174 }
 175
 176 static void enable_tx_fifo_drain(struct adapter *adapter,
 177                                  struct port_info *pi)
 178 {
 179         t3_set_reg_field(adapter, A_XGM_TXFIFO_CFG + pi->mac.offset, 0,
 180                          F_ENDROPPKT);
 181         t3_write_reg(adapter, A_XGM_RX_CTRL + pi->mac.offset, 0);
 182         t3_write_reg(adapter, A_XGM_TX_CTRL + pi->mac.offset, F_TXEN);
 183         t3_write_reg(adapter, A_XGM_RX_CTRL + pi->mac.offset, F_RXEN);
 184 }
 185
 186 static void disable_tx_fifo_drain(struct adapter *adapter,
 187                                   struct port_info *pi)
 188 {
 189         t3_set_reg_field(adapter, A_XGM_TXFIFO_CFG + pi->mac.offset,
 190                          F_ENDROPPKT, 0);
 191 }
 192
 193 void t3_os_link_fault(struct adapter *adap, int port_id, int state)
 194 {
 195         struct net_device *dev = adap->port[port_id];
 196         struct port_info *pi = netdev_priv(dev);
 197
 198         if (state == netif_carrier_ok(dev))
 199                 return;
 200
 201         if (state) {
 202                 struct cmac *mac = &pi->mac;
 203
 204                 netif_carrier_on(dev);
 205
 206                 disable_tx_fifo_drain(adap, pi);
 207
 208                 /* Clear local faults */
 209                 t3_xgm_intr_disable(adap, pi->port_id);
 210                 t3_read_reg(adap, A_XGM_INT_STATUS +
 211                                     pi->mac.offset);
 212                 t3_write_reg(adap,
 213                              A_XGM_INT_CAUSE + pi->mac.offset,
 214                              F_XGM_INT);
 215
 216                 t3_set_reg_field(adap,
 217                                  A_XGM_INT_ENABLE +
 218                                  pi->mac.offset,
 219                                  F_XGM_INT, F_XGM_INT);
 220                 t3_xgm_intr_enable(adap, pi->port_id);
 221
 222                 t3_mac_enable(mac, MAC_DIRECTION_TX);
 223         } else {
 224                 netif_carrier_off(dev);
 225
 226                 /* Flush TX FIFO */
 227                 enable_tx_fifo_drain(adap, pi);
 228         }
 229         link_report(dev);
 230 }
 231
 232 /**
 233  *      t3_os_link_changed - handle link status changes
 234  *      @adapter: the adapter associated with the link change
 235  *      @port_id: the port index whose limk status has changed
 236  *      @link_stat: the new status of the link
 237  *      @speed: the new speed setting
 238  *      @duplex: the new duplex setting
 239  *      @pause: the new flow-control setting
 240  *
 241  *      This is the OS-dependent handler for link status changes.  The OS
 242  *      neutral handler takes care of most of the processing for these events,
 243  *      then calls this handler for any OS-specific processing.
 244  */
 245 void t3_os_link_changed(struct adapter *adapter, int port_id, int link_stat,
 246                         int speed, int duplex, int pause)
 247 {
 248         struct net_device *dev = adapter->port[port_id];
 249         struct port_info *pi = netdev_priv(dev);
 250         struct cmac *mac = &pi->mac;
 251
 252         /* Skip changes from disabled ports. */
 253         if (!netif_running(dev))
 254                 return;
 255
 256         if (link_stat != netif_carrier_ok(dev)) {
 257                 if (link_stat) {
 258                         disable_tx_fifo_drain(adapter, pi);
 259
 260                         t3_mac_enable(mac, MAC_DIRECTION_RX);
 261
 262                         /* Clear local faults */
 263                         t3_xgm_intr_disable(adapter, pi->port_id);
 264                         t3_read_reg(adapter, A_XGM_INT_STATUS +
 265                                     pi->mac.offset);
 266                         t3_write_reg(adapter,
 267                                      A_XGM_INT_CAUSE + pi->mac.offset,
 268                                      F_XGM_INT);
 269
 270                         t3_set_reg_field(adapter,
 271                                          A_XGM_INT_ENABLE + pi->mac.offset,
 272                                          F_XGM_INT, F_XGM_INT);
 273                         t3_xgm_intr_enable(adapter, pi->port_id);
 274
 275                         netif_carrier_on(dev);
 276                 } else {
 277                         netif_carrier_off(dev);
 278
 279                         t3_xgm_intr_disable(adapter, pi->port_id);
 280                         t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
 281                         t3_set_reg_field(adapter,
 282                                          A_XGM_INT_ENABLE + pi->mac.offset,
 283                                          F_XGM_INT, 0);
 284
 285                         if (is_10G(adapter))
 286                                 pi->phy.ops->power_down(&pi->phy, 1);
 287
 288                         t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
 289                         t3_mac_disable(mac, MAC_DIRECTION_RX);
 290                         t3_link_start(&pi->phy, mac, &pi->link_config);
 291
 292                         /* Flush TX FIFO */
 293                         enable_tx_fifo_drain(adapter, pi);
 294                 }
 295
 296                 link_report(dev);
 297         }
 298 }
 299
 300 /**
 301  *      t3_os_phymod_changed - handle PHY module changes
 302  *      @phy: the PHY reporting the module change
 303  *      @mod_type: new module type
 304  *
 305  *      This is the OS-dependent handler for PHY module changes.  It is
 306  *      invoked when a PHY module is removed or inserted for any OS-specific
 307  *      processing.
 308  */
 309 void t3_os_phymod_changed(struct adapter *adap, int port_id)
 310 {
 311         static const char *mod_str[] = {
 312                 NULL, "SR", "LR", "LRM", "TWINAX", "TWINAX", "unknown"
 313         };
 314
 315         const struct net_device *dev = adap->port[port_id];
 316         const struct port_info *pi = netdev_priv(dev);
 317
 318         if (pi->phy.modtype == phy_modtype_none)
 319                 printk(KERN_INFO "%s: PHY module unplugged\n", dev->name);
 320         else
 321                 printk(KERN_INFO "%s: %s PHY module inserted\n", dev->name,
 322                        mod_str[pi->phy.modtype]);
 323 }
 324
 325 static void cxgb_set_rxmode(struct net_device *dev)
 326 {
 327         struct t3_rx_mode rm;
 328         struct port_info *pi = netdev_priv(dev);
 329
 330         init_rx_mode(&rm, dev, dev->mc_list);
 331         t3_mac_set_rx_mode(&pi->mac, &rm);
 332 }
 333
 334 /**
 335  *      link_start - enable a port
 336  *      @dev: the device to enable
 337  *
 338  *      Performs the MAC and PHY actions needed to enable a port.
 339  */
 340 static void link_start(struct net_device *dev)
 341 {
 342         struct t3_rx_mode rm;
 343         struct port_info *pi = netdev_priv(dev);
 344         struct cmac *mac = &pi->mac;
 345
 346         init_rx_mode(&rm, dev, dev->mc_list);
 347         t3_mac_reset(mac);
 348         t3_mac_set_num_ucast(mac, MAX_MAC_IDX);
 349         t3_mac_set_mtu(mac, dev->mtu);
 350         t3_mac_set_address(mac, LAN_MAC_IDX, dev->dev_addr);
 351         t3_mac_set_address(mac, SAN_MAC_IDX, pi->iscsic.mac_addr);
 352         t3_mac_set_rx_mode(mac, &rm);
 353         t3_link_start(&pi->phy, mac, &pi->link_config);
 354         t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
 355 }
 356
 357 static inline void cxgb_disable_msi(struct adapter *adapter)
 358 {
 359         if (adapter->flags & USING_MSIX) {
 360                 pci_disable_msix(adapter->pdev);
 361                 adapter->flags &= ~USING_MSIX;
 362         } else if (adapter->flags & USING_MSI) {
 363                 pci_disable_msi(adapter->pdev);
 364                 adapter->flags &= ~USING_MSI;
 365         }
 366 }
 367
 368 /*
 369  * Interrupt handler for asynchronous events used with MSI-X.
 370  */
 371 static irqreturn_t t3_async_intr_handler(int irq, void *cookie)
 372 {
 373         t3_slow_intr_handler(cookie);
 374         return IRQ_HANDLED;
 375 }
 376
 377 /*
 378  * Name the MSI-X interrupts.
 379  */
 380 static void name_msix_vecs(struct adapter *adap)
 381 {
 382         int i, j, msi_idx = 1, n = sizeof(adap->msix_info[0].desc) - 1;
 383
 384         snprintf(adap->msix_info[0].desc, n, "%s", adap->name);
 385         adap->msix_info[0].desc[n] = 0;
 386
 387         for_each_port(adap, j) {
 388                 struct net_device *d = adap->port[j];
 389                 const struct port_info *pi = netdev_priv(d);
 390
 391                 for (i = 0; i < pi->nqsets; i++, msi_idx++) {
 392                         snprintf(adap->msix_info[msi_idx].desc, n,
 393                                  "%s-%d", d->name, pi->first_qset + i);
 394                         adap->msix_info[msi_idx].desc[n] = 0;
 395                 }
 396         }
 397 }
 398
 399 static int request_msix_data_irqs(struct adapter *adap)
 400 {
 401         int i, j, err, qidx = 0;
 402
 403         for_each_port(adap, i) {
 404                 int nqsets = adap2pinfo(adap, i)->nqsets;
 405
 406                 for (j = 0; j < nqsets; ++j) {
 407                         err = request_irq(adap->msix_info[qidx + 1].vec,
 408                                           t3_intr_handler(adap,
 409                                                           adap->sge.qs[qidx].
 410                                                           rspq.polling), 0,
 411                                           adap->msix_info[qidx + 1].desc,
 412                                           &adap->sge.qs[qidx]);
 413                         if (err) {
 414                                 while (--qidx >= 0)
 415                                         free_irq(adap->msix_info[qidx + 1].vec,
 416                                                  &adap->sge.qs[qidx]);
 417                                 return err;
 418                         }
 419                         qidx++;
 420                 }
 421         }
 422         return 0;
 423 }
 424
 425 static void free_irq_resources(struct adapter *adapter)
 426 {
 427         if (adapter->flags & USING_MSIX) {
 428                 int i, n = 0;
 429
 430                 free_irq(adapter->msix_info[0].vec, adapter);
 431                 for_each_port(adapter, i)
 432                         n += adap2pinfo(adapter, i)->nqsets;
 433
 434                 for (i = 0; i < n; ++i)
 435                         free_irq(adapter->msix_info[i + 1].vec,
 436                                  &adapter->sge.qs[i]);
 437         } else
 438                 free_irq(adapter->pdev->irq, adapter);
 439 }
 440
 441 static int await_mgmt_replies(struct adapter *adap, unsigned long init_cnt,
 442                               unsigned long n)
 443 {
 444         int attempts = 5;
 445
 446         while (adap->sge.qs[0].rspq.offload_pkts < init_cnt + n) {
 447                 if (!--attempts)
 448                         return -ETIMEDOUT;
 449                 msleep(10);
 450         }
 451         return 0;
 452 }
 453
 454 static int init_tp_parity(struct adapter *adap)
 455 {
 456         int i;
 457         struct sk_buff *skb;
 458         struct cpl_set_tcb_field *greq;
 459         unsigned long cnt = adap->sge.qs[0].rspq.offload_pkts;
 460
 461         t3_tp_set_offload_mode(adap, 1);
 462
 463         for (i = 0; i < 16; i++) {
 464                 struct cpl_smt_write_req *req;
 465
 466                 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
 467                 if (!skb)
 468                         skb = adap->nofail_skb;
 469                 if (!skb)
 470                         goto alloc_skb_fail;
 471
 472                 req = (struct cpl_smt_write_req *)__skb_put(skb, sizeof(*req));
 473                 memset(req, 0, sizeof(*req));
 474                 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 475                 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, i));
 476                 req->mtu_idx = NMTUS - 1;
 477                 req->iff = i;
 478                 t3_mgmt_tx(adap, skb);
 479                 if (skb == adap->nofail_skb) {
 480                         await_mgmt_replies(adap, cnt, i + 1);
 481                         adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
 482                         if (!adap->nofail_skb)
 483                                 goto alloc_skb_fail;
 484                 }
 485         }
 486
 487         for (i = 0; i < 2048; i++) {
 488                 struct cpl_l2t_write_req *req;
 489
 490                 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
 491                 if (!skb)
 492                         skb = adap->nofail_skb;
 493                 if (!skb)
 494                         goto alloc_skb_fail;
 495
 496                 req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
 497                 memset(req, 0, sizeof(*req));
 498                 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 499                 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, i));
 500                 req->params = htonl(V_L2T_W_IDX(i));
 501                 t3_mgmt_tx(adap, skb);
 502                 if (skb == adap->nofail_skb) {
 503                         await_mgmt_replies(adap, cnt, 16 + i + 1);
 504                         adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
 505                         if (!adap->nofail_skb)
 506                                 goto alloc_skb_fail;
 507                 }
 508         }
 509
 510         for (i = 0; i < 2048; i++) {
 511                 struct cpl_rte_write_req *req;
 512
 513                 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
 514                 if (!skb)
 515                         skb = adap->nofail_skb;
 516                 if (!skb)
 517                         goto alloc_skb_fail;
 518
 519                 req = (struct cpl_rte_write_req *)__skb_put(skb, sizeof(*req));
 520                 memset(req, 0, sizeof(*req));
 521                 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 522                 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RTE_WRITE_REQ, i));
 523                 req->l2t_idx = htonl(V_L2T_W_IDX(i));
 524                 t3_mgmt_tx(adap, skb);
 525                 if (skb == adap->nofail_skb) {
 526                         await_mgmt_replies(adap, cnt, 16 + 2048 + i + 1);
 527                         adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
 528                         if (!adap->nofail_skb)
 529                                 goto alloc_skb_fail;
 530                 }
 531         }
 532
 533         skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
 534         if (!skb)
 535                 skb = adap->nofail_skb;
 536         if (!skb)
 537                 goto alloc_skb_fail;
 538
 539         greq = (struct cpl_set_tcb_field *)__skb_put(skb, sizeof(*greq));
 540         memset(greq, 0, sizeof(*greq));
 541         greq->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 542         OPCODE_TID(greq) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, 0));
 543         greq->mask = cpu_to_be64(1);
 544         t3_mgmt_tx(adap, skb);
 545
 546         i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
 547         if (skb == adap->nofail_skb) {
 548                 i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
 549                 adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
 550         }
 551
 552         t3_tp_set_offload_mode(adap, 0);
 553         return i;
 554
 555 alloc_skb_fail:
 556         t3_tp_set_offload_mode(adap, 0);
 557         return -ENOMEM;
 558 }
 559
 560 /**
 561  *      setup_rss - configure RSS
 562  *      @adap: the adapter
 563  *
 564  *      Sets up RSS to distribute packets to multiple receive queues.  We
 565  *      configure the RSS CPU lookup table to distribute to the number of HW
 566  *      receive queues, and the response queue lookup table to narrow that
 567  *      down to the response queues actually configured for each port.
 568  *      We always configure the RSS mapping for two ports since the mapping
 569  *      table has plenty of entries.
 570  */
 571 static void setup_rss(struct adapter *adap)
 572 {
 573         int i;
 574         unsigned int nq0 = adap2pinfo(adap, 0)->nqsets;
 575         unsigned int nq1 = adap->port[1] ? adap2pinfo(adap, 1)->nqsets : 1;
 576         u8 cpus[SGE_QSETS + 1];
 577         u16 rspq_map[RSS_TABLE_SIZE];
 578
 579         for (i = 0; i < SGE_QSETS; ++i)
 580                 cpus[i] = i;
 581         cpus[SGE_QSETS] = 0xff; /* terminator */
 582
 583         for (i = 0; i < RSS_TABLE_SIZE / 2; ++i) {
 584                 rspq_map[i] = i % nq0;
 585                 rspq_map[i + RSS_TABLE_SIZE / 2] = (i % nq1) + nq0;
 586         }
 587
 588         t3_config_rss(adap, F_RQFEEDBACKENABLE | F_TNLLKPEN | F_TNLMAPEN |
 589                       F_TNLPRTEN | F_TNL2TUPEN | F_TNL4TUPEN |
 590                       V_RRCPLCPUSIZE(6) | F_HASHTOEPLITZ, cpus, rspq_map);
 591 }
 592
 593 static void init_napi(struct adapter *adap)
 594 {
 595         int i;
 596
 597         for (i = 0; i < SGE_QSETS; i++) {
 598                 struct sge_qset *qs = &adap->sge.qs[i];
 599
 600                 if (qs->adap)
 601                         netif_napi_add(qs->netdev, &qs->napi, qs->napi.poll,
 602                                        64);
 603         }
 604
 605         /*
 606          * netif_napi_add() can be called only once per napi_struct because it
 607          * adds each new napi_struct to a list.  Be careful not to call it a
 608          * second time, e.g., during EEH recovery, by making a note of it.
 609          */
 610         adap->flags |= NAPI_INIT;
 611 }
 612
 613 /*
 614  * Wait until all NAPI handlers are descheduled.  This includes the handlers of
 615  * both netdevices representing interfaces and the dummy ones for the extra
 616  * queues.
 617  */
 618 static void quiesce_rx(struct adapter *adap)
 619 {
 620         int i;
 621
 622         for (i = 0; i < SGE_QSETS; i++)
 623                 if (adap->sge.qs[i].adap)
 624                         napi_disable(&adap->sge.qs[i].napi);
 625 }
 626
 627 static void enable_all_napi(struct adapter *adap)
 628 {
 629         int i;
 630         for (i = 0; i < SGE_QSETS; i++)
 631                 if (adap->sge.qs[i].adap)
 632                         napi_enable(&adap->sge.qs[i].napi);
 633 }
 634
 635 /**
 636  *      set_qset_lro - Turn a queue set's LRO capability on and off
 637  *      @dev: the device the qset is attached to
 638  *      @qset_idx: the queue set index
 639  *      @val: the LRO switch
 640  *
 641  *      Sets LRO on or off for a particular queue set.
 642  *      the device's features flag is updated to reflect the LRO
 643  *      capability when all queues belonging to the device are
 644  *      in the same state.
 645  */
 646 static void set_qset_lro(struct net_device *dev, int qset_idx, int val)
 647 {
 648         struct port_info *pi = netdev_priv(dev);
 649         struct adapter *adapter = pi->adapter;
 650
 651         adapter->params.sge.qset[qset_idx].lro = !!val;
 652         adapter->sge.qs[qset_idx].lro_enabled = !!val;
 653 }
 654
 655 /**
 656  *      setup_sge_qsets - configure SGE Tx/Rx/response queues
 657  *      @adap: the adapter
 658  *
 659  *      Determines how many sets of SGE queues to use and initializes them.
 660  *      We support multiple queue sets per port if we have MSI-X, otherwise
 661  *      just one queue set per port.
 662  */
 663 static int setup_sge_qsets(struct adapter *adap)
 664 {
 665         int i, j, err, irq_idx = 0, qset_idx = 0;
 666         unsigned int ntxq = SGE_TXQ_PER_SET;
 667
 668         if (adap->params.rev > 0 && !(adap->flags & USING_MSI))
 669                 irq_idx = -1;
 670
 671         for_each_port(adap, i) {
 672                 struct net_device *dev = adap->port[i];
 673                 struct port_info *pi = netdev_priv(dev);
 674
 675                 pi->qs = &adap->sge.qs[pi->first_qset];
 676                 for (j = 0; j < pi->nqsets; ++j, ++qset_idx) {
 677                         set_qset_lro(dev, qset_idx, pi->rx_offload & T3_LRO);
 678                         err = t3_sge_alloc_qset(adap, qset_idx, 1,
 679                                 (adap->flags & USING_MSIX) ? qset_idx + 1 :
 680                                                              irq_idx,
 681                                 &adap->params.sge.qset[qset_idx], ntxq, dev,
 682                                 netdev_get_tx_queue(dev, j));
 683                         if (err) {
 684                                 t3_free_sge_resources(adap);
 685                                 return err;
 686                         }
 687                 }
 688         }
 689
 690         return 0;
 691 }
 692
 693 static ssize_t attr_show(struct device *d, char *buf,
 694                          ssize_t(*format) (struct net_device *, char *))
 695 {
 696         ssize_t len;
 697
 698         /* Synchronize with ioctls that may shut down the device */
 699         rtnl_lock();
 700         len = (*format) (to_net_dev(d), buf);
 701         rtnl_unlock();
 702         return len;
 703 }
 704
 705 static ssize_t attr_store(struct device *d,
 706                           const char *buf, size_t len,
 707                           ssize_t(*set) (struct net_device *, unsigned int),
 708                           unsigned int min_val, unsigned int max_val)
 709 {
 710         char *endp;
 711         ssize_t ret;
 712         unsigned int val;
 713
 714         if (!capable(CAP_NET_ADMIN))
 715                 return -EPERM;
 716
 717         val = simple_strtoul(buf, &endp, 0);
 718         if (endp == buf || val < min_val || val > max_val)
 719                 return -EINVAL;
 720
 721         rtnl_lock();
 722         ret = (*set) (to_net_dev(d), val);
 723         if (!ret)
 724                 ret = len;
 725         rtnl_unlock();
 726         return ret;
 727 }
 728
 729 #define CXGB3_SHOW(name, val_expr) \
 730 static ssize_t format_##name(struct net_device *dev, char *buf) \
 731 { \
 732         struct port_info *pi = netdev_priv(dev); \
 733         struct adapter *adap = pi->adapter; \
 734         return sprintf(buf, "%u\n", val_expr); \
 735 } \
 736 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
 737                            char *buf) \
 738 { \
 739         return attr_show(d, buf, format_##name); \
 740 }
 741
 742 static ssize_t set_nfilters(struct net_device *dev, unsigned int val)
 743 {
 744         struct port_info *pi = netdev_priv(dev);
 745         struct adapter *adap = pi->adapter;
 746         int min_tids = is_offload(adap) ? MC5_MIN_TIDS : 0;
 747
 748         if (adap->flags & FULL_INIT_DONE)
 749                 return -EBUSY;
 750         if (val && adap->params.rev == 0)
 751                 return -EINVAL;
 752         if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nservers -
 753             min_tids)
 754                 return -EINVAL;
 755         adap->params.mc5.nfilters = val;
 756         return 0;
 757 }
 758
 759 static ssize_t store_nfilters(struct device *d, struct device_attribute *attr,
 760                               const char *buf, size_t len)
 761 {
 762         return attr_store(d, buf, len, set_nfilters, 0, ~0);
 763 }
 764
 765 static ssize_t set_nservers(struct net_device *dev, unsigned int val)
 766 {
 767         struct port_info *pi = netdev_priv(dev);
 768         struct adapter *adap = pi->adapter;
 769
 770         if (adap->flags & FULL_INIT_DONE)
 771                 return -EBUSY;
 772         if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nfilters -
 773             MC5_MIN_TIDS)
 774                 return -EINVAL;
 775         adap->params.mc5.nservers = val;
 776         return 0;
 777 }
 778
 779 static ssize_t store_nservers(struct device *d, struct device_attribute *attr,
 780                               const char *buf, size_t len)
 781 {
 782         return attr_store(d, buf, len, set_nservers, 0, ~0);
 783 }
 784
 785 #define CXGB3_ATTR_R(name, val_expr) \
 786 CXGB3_SHOW(name, val_expr) \
 787 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 788
 789 #define CXGB3_ATTR_RW(name, val_expr, store_method) \
 790 CXGB3_SHOW(name, val_expr) \
 791 static DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_method)
 792
 793 CXGB3_ATTR_R(cam_size, t3_mc5_size(&adap->mc5));
 794 CXGB3_ATTR_RW(nfilters, adap->params.mc5.nfilters, store_nfilters);
 795 CXGB3_ATTR_RW(nservers, adap->params.mc5.nservers, store_nservers);
 796
 797 static struct attribute *cxgb3_attrs[] = {
 798         &dev_attr_cam_size.attr,
 799         &dev_attr_nfilters.attr,
 800         &dev_attr_nservers.attr,
 801         NULL
 802 };
 803
 804 static struct attribute_group cxgb3_attr_group = {.attrs = cxgb3_attrs };
 805
 806 static ssize_t tm_attr_show(struct device *d,
 807                             char *buf, int sched)
 808 {
 809         struct port_info *pi = netdev_priv(to_net_dev(d));
 810         struct adapter *adap = pi->adapter;
 811         unsigned int v, addr, bpt, cpt;
 812         ssize_t len;
 813
 814         addr = A_TP_TX_MOD_Q1_Q0_RATE_LIMIT - sched / 2;
 815         rtnl_lock();
 816         t3_write_reg(adap, A_TP_TM_PIO_ADDR, addr);
 817         v = t3_read_reg(adap, A_TP_TM_PIO_DATA);
 818         if (sched & 1)
 819                 v >>= 16;
 820         bpt = (v >> 8) & 0xff;
 821         cpt = v & 0xff;
 822         if (!cpt)
 823                 len = sprintf(buf, "disabled\n");
 824         else {
 825                 v = (adap->params.vpd.cclk * 1000) / cpt;
 826                 len = sprintf(buf, "%u Kbps\n", (v * bpt) / 125);
 827         }
 828         rtnl_unlock();
 829         return len;
 830 }
 831
 832 static ssize_t tm_attr_store(struct device *d,
 833                              const char *buf, size_t len, int sched)
 834 {
 835         struct port_info *pi = netdev_priv(to_net_dev(d));
 836         struct adapter *adap = pi->adapter;
 837         unsigned int val;
 838         char *endp;
 839         ssize_t ret;
 840
 841         if (!capable(CAP_NET_ADMIN))
 842                 return -EPERM;
 843
 844         val = simple_strtoul(buf, &endp, 0);
 845         if (endp == buf || val > 10000000)
 846                 return -EINVAL;
 847
 848         rtnl_lock();
 849         ret = t3_config_sched(adap, val, sched);
 850         if (!ret)
 851                 ret = len;
 852         rtnl_unlock();
 853         return ret;
 854 }
 855
 856 #define TM_ATTR(name, sched) \
 857 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
 858                            char *buf) \
 859 { \
 860         return tm_attr_show(d, buf, sched); \
 861 } \
 862 static ssize_t store_##name(struct device *d, struct device_attribute *attr, \
 863                             const char *buf, size_t len) \
 864 { \
 865         return tm_attr_store(d, buf, len, sched); \
 866 } \
 867 static DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_##name)
 868
 869 TM_ATTR(sched0, 0);
 870 TM_ATTR(sched1, 1);
 871 TM_ATTR(sched2, 2);
 872 TM_ATTR(sched3, 3);
 873 TM_ATTR(sched4, 4);
 874 TM_ATTR(sched5, 5);
 875 TM_ATTR(sched6, 6);
 876 TM_ATTR(sched7, 7);
 877
 878 static struct attribute *offload_attrs[] = {
 879         &dev_attr_sched0.attr,
 880         &dev_attr_sched1.attr,
 881         &dev_attr_sched2.attr,
 882         &dev_attr_sched3.attr,
 883         &dev_attr_sched4.attr,
 884         &dev_attr_sched5.attr,
 885         &dev_attr_sched6.attr,
 886         &dev_attr_sched7.attr,
 887         NULL
 888 };
 889
 890 static struct attribute_group offload_attr_group = {.attrs = offload_attrs };
 891
 892 /*
 893  * Sends an sk_buff to an offload queue driver
 894  * after dealing with any active network taps.
 895  */
 896 static inline int offload_tx(struct t3cdev *tdev, struct sk_buff *skb)
 897 {
 898         int ret;
 899
 900         local_bh_disable();
 901         ret = t3_offload_tx(tdev, skb);
 902         local_bh_enable();
 903         return ret;
 904 }
 905
 906 static int write_smt_entry(struct adapter *adapter, int idx)
 907 {
 908         struct cpl_smt_write_req *req;
 909         struct port_info *pi = netdev_priv(adapter->port[idx]);
 910         struct sk_buff *skb = alloc_skb(sizeof(*req), GFP_KERNEL);
 911
 912         if (!skb)
 913                 return -ENOMEM;
 914
 915         req = (struct cpl_smt_write_req *)__skb_put(skb, sizeof(*req));
 916         req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 917         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, idx));
 918         req->mtu_idx = NMTUS - 1;       /* should be 0 but there's a T3 bug */
 919         req->iff = idx;
 920         memcpy(req->src_mac0, adapter->port[idx]->dev_addr, ETH_ALEN);
 921         memcpy(req->src_mac1, pi->iscsic.mac_addr, ETH_ALEN);
 922         skb->priority = 1;
 923         offload_tx(&adapter->tdev, skb);
 924         return 0;
 925 }
 926
 927 static int init_smt(struct adapter *adapter)
 928 {
 929         int i;
 930
 931         for_each_port(adapter, i)
 932             write_smt_entry(adapter, i);
 933         return 0;
 934 }
 935
 936 static void init_port_mtus(struct adapter *adapter)
 937 {
 938         unsigned int mtus = adapter->port[0]->mtu;
 939
 940         if (adapter->port[1])
 941                 mtus |= adapter->port[1]->mtu << 16;
 942         t3_write_reg(adapter, A_TP_MTU_PORT_TABLE, mtus);
 943 }
 944
 945 static int send_pktsched_cmd(struct adapter *adap, int sched, int qidx, int lo,
 946                               int hi, int port)
 947 {
 948         struct sk_buff *skb;
 949         struct mngt_pktsched_wr *req;
 950         int ret;
 951
 952         skb = alloc_skb(sizeof(*req), GFP_KERNEL);
 953         if (!skb)
 954                 skb = adap->nofail_skb;
 955         if (!skb)
 956                 return -ENOMEM;
 957
 958         req = (struct mngt_pktsched_wr *)skb_put(skb, sizeof(*req));
 959         req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_MNGT));
 960         req->mngt_opcode = FW_MNGTOPCODE_PKTSCHED_SET;
 961         req->sched = sched;
 962         req->idx = qidx;
 963         req->min = lo;
 964         req->max = hi;
 965         req->binding = port;
 966         ret = t3_mgmt_tx(adap, skb);
 967         if (skb == adap->nofail_skb) {
 968                 adap->nofail_skb = alloc_skb(sizeof(struct cpl_set_tcb_field),
 969                                              GFP_KERNEL);
 970                 if (!adap->nofail_skb)
 971                         ret = -ENOMEM;
 972         }
 973
 974         return ret;
 975 }
 976
 977 static int bind_qsets(struct adapter *adap)
 978 {
 979         int i, j, err = 0;
 980
 981         for_each_port(adap, i) {
 982                 const struct port_info *pi = adap2pinfo(adap, i);
 983
 984                 for (j = 0; j < pi->nqsets; ++j) {
 985                         int ret = send_pktsched_cmd(adap, 1,
 986                                                     pi->first_qset + j, -1,
 987                                                     -1, i);
 988                         if (ret)
 989                                 err = ret;
 990                 }
 991         }
 992
 993         return err;
 994 }
 995
 996 #define FW_VERSION __stringify(FW_VERSION_MAJOR) "."                    \
 997         __stringify(FW_VERSION_MINOR) "." __stringify(FW_VERSION_MICRO)
 998 #define FW_FNAME "cxgb3/t3fw-" FW_VERSION ".bin"
 999 #define TPSRAM_VERSION __stringify(TP_VERSION_MAJOR) "."                \
1000         __stringify(TP_VERSION_MINOR) "." __stringify(TP_VERSION_MICRO)
1001 #define TPSRAM_NAME "cxgb3/t3%c_psram-" TPSRAM_VERSION ".bin"
1002 #define AEL2005_OPT_EDC_NAME "cxgb3/ael2005_opt_edc.bin"
1003 #define AEL2005_TWX_EDC_NAME "cxgb3/ael2005_twx_edc.bin"
1004 #define AEL2020_TWX_EDC_NAME "cxgb3/ael2020_twx_edc.bin"
1005 MODULE_FIRMWARE(FW_FNAME);
1006 MODULE_FIRMWARE("cxgb3/t3b_psram-" TPSRAM_VERSION ".bin");
1007 MODULE_FIRMWARE("cxgb3/t3c_psram-" TPSRAM_VERSION ".bin");
1008 MODULE_FIRMWARE(AEL2005_OPT_EDC_NAME);
1009 MODULE_FIRMWARE(AEL2005_TWX_EDC_NAME);
1010 MODULE_FIRMWARE(AEL2020_TWX_EDC_NAME);
1011
1012 static inline const char *get_edc_fw_name(int edc_idx)
1013 {
1014         const char *fw_name = NULL;
1015
1016         switch (edc_idx) {
1017         case EDC_OPT_AEL2005:
1018                 fw_name = AEL2005_OPT_EDC_NAME;
1019                 break;
1020         case EDC_TWX_AEL2005:
1021                 fw_name = AEL2005_TWX_EDC_NAME;
1022                 break;
1023         case EDC_TWX_AEL2020:
1024                 fw_name = AEL2020_TWX_EDC_NAME;
1025                 break;
1026         }
1027         return fw_name;
1028 }
1029
1030 int t3_get_edc_fw(struct cphy *phy, int edc_idx, int size)
1031 {
1032         struct adapter *adapter = phy->adapter;
1033         const struct firmware *fw;
1034         char buf[64];
1035         u32 csum;
1036         const __be32 *p;
1037         u16 *cache = phy->phy_cache;
1038         int i, ret;
1039
1040         snprintf(buf, sizeof(buf), get_edc_fw_name(edc_idx));
1041
1042         ret = request_firmware(&fw, buf, &adapter->pdev->dev);
1043         if (ret < 0) {
1044                 dev_err(&adapter->pdev->dev,
1045                         "could not upgrade firmware: unable to load %s\n",
1046                         buf);
1047                 return ret;
1048         }
1049
1050         /* check size, take checksum in account */
1051         if (fw->size > size + 4) {
1052                 CH_ERR(adapter, "firmware image too large %u, expected %d\n",
1053                        (unsigned int)fw->size, size + 4);
1054                 ret = -EINVAL;
1055         }
1056
1057         /* compute checksum */
1058         p = (const __be32 *)fw->data;
1059         for (csum = 0, i = 0; i < fw->size / sizeof(csum); i++)
1060                 csum += ntohl(p[i]);
1061
1062         if (csum != 0xffffffff) {
1063                 CH_ERR(adapter, "corrupted firmware image, checksum %u\n",
1064                        csum);
1065                 ret = -EINVAL;
1066         }
1067
1068         for (i = 0; i < size / 4 ; i++) {
1069                 *cache++ = (be32_to_cpu(p[i]) & 0xffff0000) >> 16;
1070                 *cache++ = be32_to_cpu(p[i]) & 0xffff;
1071         }
1072
1073         release_firmware(fw);
1074
1075         return ret;
1076 }
1077
1078 static int upgrade_fw(struct adapter *adap)
1079 {
1080         int ret;
1081         const struct firmware *fw;
1082         struct device *dev = &adap->pdev->dev;
1083
1084         ret = request_firmware(&fw, FW_FNAME, dev);
1085         if (ret < 0) {
1086                 dev_err(dev, "could not upgrade firmware: unable to load %s\n",
1087                         FW_FNAME);
1088                 return ret;
1089         }
1090         ret = t3_load_fw(adap, fw->data, fw->size);
1091         release_firmware(fw);
1092
1093         if (ret == 0)
1094                 dev_info(dev, "successful upgrade to firmware %d.%d.%d\n",
1095                          FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
1096         else
1097                 dev_err(dev, "failed to upgrade to firmware %d.%d.%d\n",
1098                         FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
1099
1100         return ret;
1101 }
1102
1103 static inline char t3rev2char(struct adapter *adapter)
1104 {
1105         char rev = 0;
1106
1107         switch(adapter->params.rev) {
1108         case T3_REV_B:
1109         case T3_REV_B2:
1110                 rev = 'b';
1111                 break;
1112         case T3_REV_C:
1113                 rev = 'c';
1114                 break;
1115         }
1116         return rev;
1117 }
1118
1119 static int update_tpsram(struct adapter *adap)
1120 {
1121         const struct firmware *tpsram;
1122         char buf[64];
1123         struct device *dev = &adap->pdev->dev;
1124         int ret;
1125         char rev;
1126
1127         rev = t3rev2char(adap);
1128         if (!rev)
1129                 return 0;
1130
1131         snprintf(buf, sizeof(buf), TPSRAM_NAME, rev);
1132
1133         ret = request_firmware(&tpsram, buf, dev);
1134         if (ret < 0) {
1135                 dev_err(dev, "could not load TP SRAM: unable to load %s\n",
1136                         buf);
1137                 return ret;
1138         }
1139
1140         ret = t3_check_tpsram(adap, tpsram->data, tpsram->size);
1141         if (ret)
1142                 goto release_tpsram;
1143
1144         ret = t3_set_proto_sram(adap, tpsram->data);
1145         if (ret == 0)
1146                 dev_info(dev,
1147                          "successful update of protocol engine "
1148                          "to %d.%d.%d\n",
1149                          TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
1150         else
1151                 dev_err(dev, "failed to update of protocol engine %d.%d.%d\n",
1152                         TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
1153         if (ret)
1154                 dev_err(dev, "loading protocol SRAM failed\n");
1155
1156 release_tpsram:
1157         release_firmware(tpsram);
1158
1159         return ret;
1160 }
1161
1162 /**
1163  *      cxgb_up - enable the adapter
1164  *      @adapter: adapter being enabled
1165  *
1166  *      Called when the first port is enabled, this function performs the
1167  *      actions necessary to make an adapter operational, such as completing
1168  *      the initialization of HW modules, and enabling interrupts.
1169  *
1170  *      Must be called with the rtnl lock held.
1171  */
1172 static int cxgb_up(struct adapter *adap)
1173 {
1174         int err;
1175
1176         if (!(adap->flags & FULL_INIT_DONE)) {
1177                 err = t3_check_fw_version(adap);
1178                 if (err == -EINVAL) {
1179                         err = upgrade_fw(adap);
1180                         CH_WARN(adap, "FW upgrade to %d.%d.%d %s\n",
1181                                 FW_VERSION_MAJOR, FW_VERSION_MINOR,
1182                                 FW_VERSION_MICRO, err ? "failed" : "succeeded");
1183                 }
1184
1185                 err = t3_check_tpsram_version(adap);
1186                 if (err == -EINVAL) {
1187                         err = update_tpsram(adap);
1188                         CH_WARN(adap, "TP upgrade to %d.%d.%d %s\n",
1189                                 TP_VERSION_MAJOR, TP_VERSION_MINOR,
1190                                 TP_VERSION_MICRO, err ? "failed" : "succeeded");
1191                 }
1192
1193                 /*
1194                  * Clear interrupts now to catch errors if t3_init_hw fails.
1195                  * We clear them again later as initialization may trigger
1196                  * conditions that can interrupt.
1197                  */
1198                 t3_intr_clear(adap);
1199
1200                 err = t3_init_hw(adap, 0);
1201                 if (err)
1202                         goto out;
1203
1204                 t3_set_reg_field(adap, A_TP_PARA_REG5, 0, F_RXDDPOFFINIT);
1205                 t3_write_reg(adap, A_ULPRX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
1206
1207                 err = setup_sge_qsets(adap);
1208                 if (err)
1209                         goto out;
1210
1211                 setup_rss(adap);
1212                 if (!(adap->flags & NAPI_INIT))
1213                         init_napi(adap);
1214
1215                 t3_start_sge_timers(adap);
1216                 adap->flags |= FULL_INIT_DONE;
1217         }
1218
1219         t3_intr_clear(adap);
1220
1221         if (adap->flags & USING_MSIX) {
1222                 name_msix_vecs(adap);
1223                 err = request_irq(adap->msix_info[0].vec,
1224                                   t3_async_intr_handler, 0,
1225                                   adap->msix_info[0].desc, adap);
1226                 if (err)
1227                         goto irq_err;
1228
1229                 err = request_msix_data_irqs(adap);
1230                 if (err) {
1231                         free_irq(adap->msix_info[0].vec, adap);
1232                         goto irq_err;
1233                 }
1234         } else if ((err = request_irq(adap->pdev->irq,
1235                                       t3_intr_handler(adap,
1236                                                       adap->sge.qs[0].rspq.
1237                                                       polling),
1238                                       (adap->flags & USING_MSI) ?
1239                                        0 : IRQF_SHARED,
1240                                       adap->name, adap)))
1241                 goto irq_err;
1242
1243         enable_all_napi(adap);
1244         t3_sge_start(adap);
1245         t3_intr_enable(adap);
1246
1247         if (adap->params.rev >= T3_REV_C && !(adap->flags & TP_PARITY_INIT) &&
1248             is_offload(adap) && init_tp_parity(adap) == 0)
1249                 adap->flags |= TP_PARITY_INIT;
1250
1251         if (adap->flags & TP_PARITY_INIT) {
1252                 t3_write_reg(adap, A_TP_INT_CAUSE,
1253                              F_CMCACHEPERR | F_ARPLUTPERR);
1254                 t3_write_reg(adap, A_TP_INT_ENABLE, 0x7fbfffff);
1255         }
1256
1257         if (!(adap->flags & QUEUES_BOUND)) {
1258                 err = bind_qsets(adap);
1259                 if (err) {
1260                         CH_ERR(adap, "failed to bind qsets, err %d\n", err);
1261                         t3_intr_disable(adap);
1262                         free_irq_resources(adap);
1263                         goto out;
1264                 }
1265                 adap->flags |= QUEUES_BOUND;
1266         }
1267
1268 out:
1269         return err;
1270 irq_err:
1271         CH_ERR(adap, "request_irq failed, err %d\n", err);
1272         goto out;
1273 }
1274
1275 /*
1276  * Release resources when all the ports and offloading have been stopped.
1277  */
1278 static void cxgb_down(struct adapter *adapter)
1279 {
1280         t3_sge_stop(adapter);
1281         spin_lock_irq(&adapter->work_lock);     /* sync with PHY intr task */
1282         t3_intr_disable(adapter);
1283         spin_unlock_irq(&adapter->work_lock);
1284
1285         free_irq_resources(adapter);
1286         quiesce_rx(adapter);
1287         flush_workqueue(cxgb3_wq);      /* wait for external IRQ handler */
1288 }
1289
1290 static void schedule_chk_task(struct adapter *adap)
1291 {
1292         unsigned int timeo;
1293
1294         timeo = adap->params.linkpoll_period ?
1295             (HZ * adap->params.linkpoll_period) / 10 :
1296             adap->params.stats_update_period * HZ;
1297         if (timeo)
1298                 queue_delayed_work(cxgb3_wq, &adap->adap_check_task, timeo);
1299 }
1300
1301 static int offload_open(struct net_device *dev)
1302 {
1303         struct port_info *pi = netdev_priv(dev);
1304         struct adapter *adapter = pi->adapter;
1305         struct t3cdev *tdev = dev2t3cdev(dev);
1306         int adap_up = adapter->open_device_map & PORT_MASK;
1307         int err;
1308
1309         if (test_and_set_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
1310                 return 0;
1311
1312         if (!adap_up && (err = cxgb_up(adapter)) < 0)
1313                 goto out;
1314
1315         t3_tp_set_offload_mode(adapter, 1);
1316         tdev->lldev = adapter->port[0];
1317         err = cxgb3_offload_activate(adapter);
1318         if (err)
1319                 goto out;
1320
1321         init_port_mtus(adapter);
1322         t3_load_mtus(adapter, adapter->params.mtus, adapter->params.a_wnd,
1323                      adapter->params.b_wnd,
1324                      adapter->params.rev == 0 ?
1325                      adapter->port[0]->mtu : 0xffff);
1326         init_smt(adapter);
1327
1328         if (sysfs_create_group(&tdev->lldev->dev.kobj, &offload_attr_group))
1329                 dev_dbg(&dev->dev, "cannot create sysfs group\n");
1330
1331         /* Call back all registered clients */
1332         cxgb3_add_clients(tdev);
1333
1334 out:
1335         /* restore them in case the offload module has changed them */
1336         if (err) {
1337                 t3_tp_set_offload_mode(adapter, 0);
1338                 clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
1339                 cxgb3_set_dummy_ops(tdev);
1340         }
1341         return err;
1342 }
1343
1344 static int offload_close(struct t3cdev *tdev)
1345 {
1346         struct adapter *adapter = tdev2adap(tdev);
1347
1348         if (!test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
1349                 return 0;
1350
1351         /* Call back all registered clients */
1352         cxgb3_remove_clients(tdev);
1353
1354         sysfs_remove_group(&tdev->lldev->dev.kobj, &offload_attr_group);
1355
1356         /* Flush work scheduled while releasing TIDs */
1357         flush_scheduled_work();
1358
1359         tdev->lldev = NULL;
1360         cxgb3_set_dummy_ops(tdev);
1361         t3_tp_set_offload_mode(adapter, 0);
1362         clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
1363
1364         if (!adapter->open_device_map)
1365                 cxgb_down(adapter);
1366
1367         cxgb3_offload_deactivate(adapter);
1368         return 0;
1369 }
1370
1371 static int cxgb_open(struct net_device *dev)
1372 {
1373         struct port_info *pi = netdev_priv(dev);
1374         struct adapter *adapter = pi->adapter;
1375         int other_ports = adapter->open_device_map & PORT_MASK;
1376         int err;
1377
1378         if (!adapter->open_device_map && (err = cxgb_up(adapter)) < 0)
1379                 return err;
1380
1381         set_bit(pi->port_id, &adapter->open_device_map);
1382         if (is_offload(adapter) && !ofld_disable) {
1383                 err = offload_open(dev);
1384                 if (err)
1385                         printk(KERN_WARNING
1386                                "Could not initialize offload capabilities\n");
1387         }
1388
1389         dev->real_num_tx_queues = pi->nqsets;
1390         link_start(dev);
1391         t3_port_intr_enable(adapter, pi->port_id);
1392         netif_tx_start_all_queues(dev);
1393         if (!other_ports)
1394                 schedule_chk_task(adapter);
1395
1396         cxgb3_event_notify(&adapter->tdev, OFFLOAD_PORT_UP, pi->port_id);
1397         return 0;
1398 }
1399
1400 static int cxgb_close(struct net_device *dev)
1401 {
1402         struct port_info *pi = netdev_priv(dev);
1403         struct adapter *adapter = pi->adapter;
1404
1405
1406         if (!adapter->open_device_map)
1407                 return 0;
1408
1409         /* Stop link fault interrupts */
1410         t3_xgm_intr_disable(adapter, pi->port_id);
1411         t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
1412
1413         t3_port_intr_disable(adapter, pi->port_id);
1414         netif_tx_stop_all_queues(dev);
1415         pi->phy.ops->power_down(&pi->phy, 1);
1416         netif_carrier_off(dev);
1417         t3_mac_disable(&pi->mac, MAC_DIRECTION_TX | MAC_DIRECTION_RX);
1418
1419         spin_lock_irq(&adapter->work_lock);     /* sync with update task */
1420         clear_bit(pi->port_id, &adapter->open_device_map);
1421         spin_unlock_irq(&adapter->work_lock);
1422
1423         if (!(adapter->open_device_map & PORT_MASK))
1424                 cancel_delayed_work_sync(&adapter->adap_check_task);
1425
1426         if (!adapter->open_device_map)
1427                 cxgb_down(adapter);
1428
1429         cxgb3_event_notify(&adapter->tdev, OFFLOAD_PORT_DOWN, pi->port_id);
1430         return 0;
1431 }
1432
1433 static struct net_device_stats *cxgb_get_stats(struct net_device *dev)
1434 {
1435         struct port_info *pi = netdev_priv(dev);
1436         struct adapter *adapter = pi->adapter;
1437         struct net_device_stats *ns = &pi->netstats;
1438         const struct mac_stats *pstats;
1439
1440         spin_lock(&adapter->stats_lock);
1441         pstats = t3_mac_update_stats(&pi->mac);
1442         spin_unlock(&adapter->stats_lock);
1443
1444         ns->tx_bytes = pstats->tx_octets;
1445         ns->tx_packets = pstats->tx_frames;
1446         ns->rx_bytes = pstats->rx_octets;
1447         ns->rx_packets = pstats->rx_frames;
1448         ns->multicast = pstats->rx_mcast_frames;
1449
1450         ns->tx_errors = pstats->tx_underrun;
1451         ns->rx_errors = pstats->rx_symbol_errs + pstats->rx_fcs_errs +
1452             pstats->rx_too_long + pstats->rx_jabber + pstats->rx_short +
1453             pstats->rx_fifo_ovfl;
1454
1455         /* detailed rx_errors */
1456         ns->rx_length_errors = pstats->rx_jabber + pstats->rx_too_long;
1457         ns->rx_over_errors = 0;
1458         ns->rx_crc_errors = pstats->rx_fcs_errs;
1459         ns->rx_frame_errors = pstats->rx_symbol_errs;
1460         ns->rx_fifo_errors = pstats->rx_fifo_ovfl;
1461         ns->rx_missed_errors = pstats->rx_cong_drops;
1462
1463         /* detailed tx_errors */
1464         ns->tx_aborted_errors = 0;
1465         ns->tx_carrier_errors = 0;
1466         ns->tx_fifo_errors = pstats->tx_underrun;
1467         ns->tx_heartbeat_errors = 0;
1468         ns->tx_window_errors = 0;
1469         return ns;
1470 }
1471
1472 static u32 get_msglevel(struct net_device *dev)
1473 {
1474         struct port_info *pi = netdev_priv(dev);
1475         struct adapter *adapter = pi->adapter;
1476
1477         return adapter->msg_enable;
1478 }
1479
1480 static void set_msglevel(struct net_device *dev, u32 val)
1481 {
1482         struct port_info *pi = netdev_priv(dev);
1483         struct adapter *adapter = pi->adapter;
1484
1485         adapter->msg_enable = val;
1486 }
1487
1488 static char stats_strings[][ETH_GSTRING_LEN] = {
1489         "TxOctetsOK         ",
1490         "TxFramesOK         ",
1491         "TxMulticastFramesOK",
1492         "TxBroadcastFramesOK",
1493         "TxPauseFrames      ",
1494         "TxUnderrun         ",
1495         "TxExtUnderrun      ",
1496
1497         "TxFrames64         ",
1498         "TxFrames65To127    ",
1499         "TxFrames128To255   ",
1500         "TxFrames256To511   ",
1501         "TxFrames512To1023  ",
1502         "TxFrames1024To1518 ",
1503         "TxFrames1519ToMax  ",
1504
1505         "RxOctetsOK         ",
1506         "RxFramesOK         ",
1507         "RxMulticastFramesOK",
1508         "RxBroadcastFramesOK",
1509         "RxPauseFrames      ",
1510         "RxFCSErrors        ",
1511         "RxSymbolErrors     ",
1512         "RxShortErrors      ",
1513         "RxJabberErrors     ",
1514         "RxLengthErrors     ",
1515         "RxFIFOoverflow     ",
1516
1517         "RxFrames64         ",
1518         "RxFrames65To127    ",
1519         "RxFrames128To255   ",
1520         "RxFrames256To511   ",
1521         "RxFrames512To1023  ",
1522         "RxFrames1024To1518 ",
1523         "RxFrames1519ToMax  ",
1524
1525         "PhyFIFOErrors      ",
1526         "TSO                ",
1527         "VLANextractions    ",
1528         "VLANinsertions     ",
1529         "TxCsumOffload      ",
1530         "RxCsumGood         ",
1531         "LroAggregated      ",
1532         "LroFlushed         ",
1533         "LroNoDesc          ",
1534         "RxDrops            ",
1535
1536         "CheckTXEnToggled   ",
1537         "CheckResets        ",
1538
1539         "LinkFaults         ",
1540 };
1541
1542 static int get_sset_count(struct net_device *dev, int sset)
1543 {
1544         switch (sset) {
1545         case ETH_SS_STATS:
1546                 return ARRAY_SIZE(stats_strings);
1547         default:
1548                 return -EOPNOTSUPP;
1549         }
1550 }
1551
1552 #define T3_REGMAP_SIZE (3 * 1024)
1553
1554 static int get_regs_len(struct net_device *dev)
1555 {
1556         return T3_REGMAP_SIZE;
1557 }
1558
1559 static int get_eeprom_len(struct net_device *dev)
1560 {
1561         return EEPROMSIZE;
1562 }
1563
1564 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1565 {
1566         struct port_info *pi = netdev_priv(dev);
1567         struct adapter *adapter = pi->adapter;
1568         u32 fw_vers = 0;
1569         u32 tp_vers = 0;
1570
1571         spin_lock(&adapter->stats_lock);
1572         t3_get_fw_version(adapter, &fw_vers);
1573         t3_get_tp_version(adapter, &tp_vers);
1574         spin_unlock(&adapter->stats_lock);
1575
1576         strcpy(info->driver, DRV_NAME);
1577         strcpy(info->version, DRV_VERSION);
1578         strcpy(info->bus_info, pci_name(adapter->pdev));
1579         if (!fw_vers)
1580                 strcpy(info->fw_version, "N/A");
1581         else {
1582                 snprintf(info->fw_version, sizeof(info->fw_version),
1583                          "%s %u.%u.%u TP %u.%u.%u",
1584                          G_FW_VERSION_TYPE(fw_vers) ? "T" : "N",
1585                          G_FW_VERSION_MAJOR(fw_vers),
1586                          G_FW_VERSION_MINOR(fw_vers),
1587                          G_FW_VERSION_MICRO(fw_vers),
1588                          G_TP_VERSION_MAJOR(tp_vers),
1589                          G_TP_VERSION_MINOR(tp_vers),
1590                          G_TP_VERSION_MICRO(tp_vers));
1591         }
1592 }
1593
1594 static void get_strings(struct net_device *dev, u32 stringset, u8 * data)
1595 {
1596         if (stringset == ETH_SS_STATS)
1597                 memcpy(data, stats_strings, sizeof(stats_strings));
1598 }
1599
1600 static unsigned long collect_sge_port_stats(struct adapter *adapter,
1601                                             struct port_info *p, int idx)
1602 {
1603         int i;
1604         unsigned long tot = 0;
1605
1606         for (i = p->first_qset; i < p->first_qset + p->nqsets; ++i)
1607                 tot += adapter->sge.qs[i].port_stats[idx];
1608         return tot;
1609 }
1610
1611 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1612                       u64 *data)
1613 {
1614         struct port_info *pi = netdev_priv(dev);
1615         struct adapter *adapter = pi->adapter;
1616         const struct mac_stats *s;
1617
1618         spin_lock(&adapter->stats_lock);
1619         s = t3_mac_update_stats(&pi->mac);
1620         spin_unlock(&adapter->stats_lock);
1621
1622         *data++ = s->tx_octets;
1623         *data++ = s->tx_frames;
1624         *data++ = s->tx_mcast_frames;
1625         *data++ = s->tx_bcast_frames;
1626         *data++ = s->tx_pause;
1627         *data++ = s->tx_underrun;
1628         *data++ = s->tx_fifo_urun;
1629
1630         *data++ = s->tx_frames_64;
1631         *data++ = s->tx_frames_65_127;
1632         *data++ = s->tx_frames_128_255;
1633         *data++ = s->tx_frames_256_511;
1634         *data++ = s->tx_frames_512_1023;
1635         *data++ = s->tx_frames_1024_1518;
1636         *data++ = s->tx_frames_1519_max;
1637
1638         *data++ = s->rx_octets;
1639         *data++ = s->rx_frames;
1640         *data++ = s->rx_mcast_frames;
1641         *data++ = s->rx_bcast_frames;
1642         *data++ = s->rx_pause;
1643         *data++ = s->rx_fcs_errs;
1644         *data++ = s->rx_symbol_errs;
1645         *data++ = s->rx_short;
1646         *data++ = s->rx_jabber;
1647         *data++ = s->rx_too_long;
1648         *data++ = s->rx_fifo_ovfl;
1649
1650         *data++ = s->rx_frames_64;
1651         *data++ = s->rx_frames_65_127;
1652         *data++ = s->rx_frames_128_255;
1653         *data++ = s->rx_frames_256_511;
1654         *data++ = s->rx_frames_512_1023;
1655         *data++ = s->rx_frames_1024_1518;
1656         *data++ = s->rx_frames_1519_max;
1657
1658         *data++ = pi->phy.fifo_errors;
1659
1660         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TSO);
1661         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANEX);
1662         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANINS);
1663         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TX_CSUM);
1664         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_RX_CSUM_GOOD);
1665         *data++ = 0;
1666         *data++ = 0;
1667         *data++ = 0;
1668         *data++ = s->rx_cong_drops;
1669
1670         *data++ = s->num_toggled;
1671         *data++ = s->num_resets;
1672
1673         *data++ = s->link_faults;
1674 }
1675
1676 static inline void reg_block_dump(struct adapter *ap, void *buf,
1677                                   unsigned int start, unsigned int end)
1678 {
1679         u32 *p = buf + start;
1680
1681         for (; start <= end; start += sizeof(u32))
1682                 *p++ = t3_read_reg(ap, start);
1683 }
1684
1685 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1686                      void *buf)
1687 {
1688         struct port_info *pi = netdev_priv(dev);
1689         struct adapter *ap = pi->adapter;
1690
1691         /*
1692          * Version scheme:
1693          * bits 0..9: chip version
1694          * bits 10..15: chip revision
1695          * bit 31: set for PCIe cards
1696          */
1697         regs->version = 3 | (ap->params.rev << 10) | (is_pcie(ap) << 31);
1698
1699         /*
1700          * We skip the MAC statistics registers because they are clear-on-read.
1701          * Also reading multi-register stats would need to synchronize with the
1702          * periodic mac stats accumulation.  Hard to justify the complexity.
1703          */
1704         memset(buf, 0, T3_REGMAP_SIZE);
1705         reg_block_dump(ap, buf, 0, A_SG_RSPQ_CREDIT_RETURN);
1706         reg_block_dump(ap, buf, A_SG_HI_DRB_HI_THRSH, A_ULPRX_PBL_ULIMIT);
1707         reg_block_dump(ap, buf, A_ULPTX_CONFIG, A_MPS_INT_CAUSE);
1708         reg_block_dump(ap, buf, A_CPL_SWITCH_CNTRL, A_CPL_MAP_TBL_DATA);
1709         reg_block_dump(ap, buf, A_SMB_GLOBAL_TIME_CFG, A_XGM_SERDES_STAT3);
1710         reg_block_dump(ap, buf, A_XGM_SERDES_STATUS0,
1711                        XGM_REG(A_XGM_SERDES_STAT3, 1));
1712         reg_block_dump(ap, buf, XGM_REG(A_XGM_SERDES_STATUS0, 1),
1713                        XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1));
1714 }
1715
1716 static int restart_autoneg(struct net_device *dev)
1717 {
1718         struct port_info *p = netdev_priv(dev);
1719
1720         if (!netif_running(dev))
1721                 return -EAGAIN;
1722         if (p->link_config.autoneg != AUTONEG_ENABLE)
1723                 return -EINVAL;
1724         p->phy.ops->autoneg_restart(&p->phy);
1725         return 0;
1726 }
1727
1728 static int cxgb3_phys_id(struct net_device *dev, u32 data)
1729 {
1730         struct port_info *pi = netdev_priv(dev);
1731         struct adapter *adapter = pi->adapter;
1732         int i;
1733
1734         if (data == 0)
1735                 data = 2;
1736
1737         for (i = 0; i < data * 2; i++) {
1738                 t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
1739                                  (i & 1) ? F_GPIO0_OUT_VAL : 0);
1740                 if (msleep_interruptible(500))
1741                         break;
1742         }
1743         t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
1744                          F_GPIO0_OUT_VAL);
1745         return 0;
1746 }
1747
1748 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1749 {
1750         struct port_info *p = netdev_priv(dev);
1751
1752         cmd->supported = p->link_config.supported;
1753         cmd->advertising = p->link_config.advertising;
1754
1755         if (netif_carrier_ok(dev)) {
1756                 cmd->speed = p->link_config.speed;
1757                 cmd->duplex = p->link_config.duplex;
1758         } else {
1759                 cmd->speed = -1;
1760                 cmd->duplex = -1;
1761         }
1762
1763         cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1764         cmd->phy_address = p->phy.mdio.prtad;
1765         cmd->transceiver = XCVR_EXTERNAL;
1766         cmd->autoneg = p->link_config.autoneg;
1767         cmd->maxtxpkt = 0;
1768         cmd->maxrxpkt = 0;
1769         return 0;
1770 }
1771
1772 static int speed_duplex_to_caps(int speed, int duplex)
1773 {
1774         int cap = 0;
1775
1776         switch (speed) {
1777         case SPEED_10:
1778                 if (duplex == DUPLEX_FULL)
1779                         cap = SUPPORTED_10baseT_Full;
1780                 else
1781                         cap = SUPPORTED_10baseT_Half;
1782                 break;
1783         case SPEED_100:
1784                 if (duplex == DUPLEX_FULL)
1785                         cap = SUPPORTED_100baseT_Full;
1786                 else
1787                         cap = SUPPORTED_100baseT_Half;
1788                 break;
1789         case SPEED_1000:
1790                 if (duplex == DUPLEX_FULL)
1791                         cap = SUPPORTED_1000baseT_Full;
1792                 else
1793                         cap = SUPPORTED_1000baseT_Half;
1794                 break;
1795         case SPEED_10000:
1796                 if (duplex == DUPLEX_FULL)
1797                         cap = SUPPORTED_10000baseT_Full;
1798         }
1799         return cap;
1800 }
1801
1802 #define ADVERTISED_MASK (ADVERTISED_10baseT_Half | ADVERTISED_10baseT_Full | \
1803                       ADVERTISED_100baseT_Half | ADVERTISED_100baseT_Full | \
1804                       ADVERTISED_1000baseT_Half | ADVERTISED_1000baseT_Full | \
1805                       ADVERTISED_10000baseT_Full)
1806
1807 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1808 {
1809         struct port_info *p = netdev_priv(dev);
1810         struct link_config *lc = &p->link_config;
1811
1812         if (!(lc->supported & SUPPORTED_Autoneg)) {
1813                 /*
1814                  * PHY offers a single speed/duplex.  See if that's what's
1815                  * being requested.
1816                  */
1817                 if (cmd->autoneg == AUTONEG_DISABLE) {
1818                         int cap = speed_duplex_to_caps(cmd->speed, cmd->duplex);
1819                         if (lc->supported & cap)
1820                                 return 0;
1821                 }
1822                 return -EINVAL;
1823         }
1824
1825         if (cmd->autoneg == AUTONEG_DISABLE) {
1826                 int cap = speed_duplex_to_caps(cmd->speed, cmd->duplex);
1827
1828                 if (!(lc->supported & cap) || cmd->speed == SPEED_1000)
1829                         return -EINVAL;
1830                 lc->requested_speed = cmd->speed;
1831                 lc->requested_duplex = cmd->duplex;
1832                 lc->advertising = 0;
1833         } else {
1834                 cmd->advertising &= ADVERTISED_MASK;
1835                 cmd->advertising &= lc->supported;
1836                 if (!cmd->advertising)
1837                         return -EINVAL;
1838                 lc->requested_speed = SPEED_INVALID;
1839                 lc->requested_duplex = DUPLEX_INVALID;
1840                 lc->advertising = cmd->advertising | ADVERTISED_Autoneg;
1841         }
1842         lc->autoneg = cmd->autoneg;
1843         if (netif_running(dev))
1844                 t3_link_start(&p->phy, &p->mac, lc);
1845         return 0;
1846 }
1847
1848 static void get_pauseparam(struct net_device *dev,
1849                            struct ethtool_pauseparam *epause)
1850 {
1851         struct port_info *p = netdev_priv(dev);
1852
1853         epause->autoneg = (p->link_config.requested_fc & PAUSE_AUTONEG) != 0;
1854         epause->rx_pause = (p->link_config.fc & PAUSE_RX) != 0;
1855         epause->tx_pause = (p->link_config.fc & PAUSE_TX) != 0;
1856 }
1857
1858 static int set_pauseparam(struct net_device *dev,
1859                           struct ethtool_pauseparam *epause)
1860 {
1861         struct port_info *p = netdev_priv(dev);
1862         struct link_config *lc = &p->link_config;
1863
1864         if (epause->autoneg == AUTONEG_DISABLE)
1865                 lc->requested_fc = 0;
1866         else if (lc->supported & SUPPORTED_Autoneg)
1867                 lc->requested_fc = PAUSE_AUTONEG;
1868         else
1869                 return -EINVAL;
1870
1871         if (epause->rx_pause)
1872                 lc->requested_fc |= PAUSE_RX;
1873         if (epause->tx_pause)
1874                 lc->requested_fc |= PAUSE_TX;
1875         if (lc->autoneg == AUTONEG_ENABLE) {
1876                 if (netif_running(dev))
1877                         t3_link_start(&p->phy, &p->mac, lc);
1878         } else {
1879                 lc->fc = lc->requested_fc & (PAUSE_RX | PAUSE_TX);
1880                 if (netif_running(dev))
1881                         t3_mac_set_speed_duplex_fc(&p->mac, -1, -1, lc->fc);
1882         }
1883         return 0;
1884 }
1885
1886 static u32 get_rx_csum(struct net_device *dev)
1887 {
1888         struct port_info *p = netdev_priv(dev);
1889
1890         return p->rx_offload & T3_RX_CSUM;
1891 }
1892
1893 static int set_rx_csum(struct net_device *dev, u32 data)
1894 {
1895         struct port_info *p = netdev_priv(dev);
1896
1897         if (data) {
1898                 p->rx_offload |= T3_RX_CSUM;
1899         } else {
1900                 int i;
1901
1902                 p->rx_offload &= ~(T3_RX_CSUM | T3_LRO);
1903                 for (i = p->first_qset; i < p->first_qset + p->nqsets; i++)
1904                         set_qset_lro(dev, i, 0);
1905         }
1906         return 0;
1907 }
1908
1909 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1910 {
1911         struct port_info *pi = netdev_priv(dev);
1912         struct adapter *adapter = pi->adapter;
1913         const struct qset_params *q = &adapter->params.sge.qset[pi->first_qset];
1914
1915         e->rx_max_pending = MAX_RX_BUFFERS;
1916         e->rx_mini_max_pending = 0;
1917         e->rx_jumbo_max_pending = MAX_RX_JUMBO_BUFFERS;
1918         e->tx_max_pending = MAX_TXQ_ENTRIES;
1919
1920         e->rx_pending = q->fl_size;
1921         e->rx_mini_pending = q->rspq_size;
1922         e->rx_jumbo_pending = q->jumbo_size;
1923         e->tx_pending = q->txq_size[0];
1924 }
1925
1926 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1927 {
1928         struct port_info *pi = netdev_priv(dev);
1929         struct adapter *adapter = pi->adapter;
1930         struct qset_params *q;
1931         int i;
1932
1933         if (e->rx_pending > MAX_RX_BUFFERS ||
1934             e->rx_jumbo_pending > MAX_RX_JUMBO_BUFFERS ||
1935             e->tx_pending > MAX_TXQ_ENTRIES ||
1936             e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1937             e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1938             e->rx_pending < MIN_FL_ENTRIES ||
1939             e->rx_jumbo_pending < MIN_FL_ENTRIES ||
1940             e->tx_pending < adapter->params.nports * MIN_TXQ_ENTRIES)
1941                 return -EINVAL;
1942
1943         if (adapter->flags & FULL_INIT_DONE)
1944                 return -EBUSY;
1945
1946         q = &adapter->params.sge.qset[pi->first_qset];
1947         for (i = 0; i < pi->nqsets; ++i, ++q) {
1948                 q->rspq_size = e->rx_mini_pending;
1949                 q->fl_size = e->rx_pending;
1950                 q->jumbo_size = e->rx_jumbo_pending;
1951                 q->txq_size[0] = e->tx_pending;
1952                 q->txq_size[1] = e->tx_pending;
1953                 q->txq_size[2] = e->tx_pending;
1954         }
1955         return 0;
1956 }
1957
1958 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1959 {
1960         struct port_info *pi = netdev_priv(dev);
1961         struct adapter *adapter = pi->adapter;
1962         struct qset_params *qsp = &adapter->params.sge.qset[0];
1963         struct sge_qset *qs = &adapter->sge.qs[0];
1964
1965         if (c->rx_coalesce_usecs * 10 > M_NEWTIMER)
1966                 return -EINVAL;
1967
1968         qsp->coalesce_usecs = c->rx_coalesce_usecs;
1969         t3_update_qset_coalesce(qs, qsp);
1970         return 0;
1971 }
1972
1973 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1974 {
1975         struct port_info *pi = netdev_priv(dev);
1976         struct adapter *adapter = pi->adapter;
1977         struct qset_params *q = adapter->params.sge.qset;
1978
1979         c->rx_coalesce_usecs = q->coalesce_usecs;
1980         return 0;
1981 }
1982
1983 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
1984                       u8 * data)
1985 {
1986         struct port_info *pi = netdev_priv(dev);
1987         struct adapter *adapter = pi->adapter;
1988         int i, err = 0;
1989
1990         u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
1991         if (!buf)
1992                 return -ENOMEM;
1993
1994         e->magic = EEPROM_MAGIC;
1995         for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
1996                 err = t3_seeprom_read(adapter, i, (__le32 *) & buf[i]);
1997
1998         if (!err)
1999                 memcpy(data, buf + e->offset, e->len);
2000         kfree(buf);
2001         return err;
2002 }
2003
2004 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2005                       u8 * data)
2006 {
2007         struct port_info *pi = netdev_priv(dev);
2008         struct adapter *adapter = pi->adapter;
2009         u32 aligned_offset, aligned_len;
2010         __le32 *p;
2011         u8 *buf;
2012         int err;
2013
2014         if (eeprom->magic != EEPROM_MAGIC)
2015                 return -EINVAL;
2016
2017         aligned_offset = eeprom->offset & ~3;
2018         aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2019
2020         if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2021                 buf = kmalloc(aligned_len, GFP_KERNEL);
2022                 if (!buf)
2023                         return -ENOMEM;
2024                 err = t3_seeprom_read(adapter, aligned_offset, (__le32 *) buf);
2025                 if (!err && aligned_len > 4)
2026                         err = t3_seeprom_read(adapter,
2027                                               aligned_offset + aligned_len - 4,
2028                                               (__le32 *) & buf[aligned_len - 4]);
2029                 if (err)
2030                         goto out;
2031                 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
2032         } else
2033                 buf = data;
2034
2035         err = t3_seeprom_wp(adapter, 0);
2036         if (err)
2037                 goto out;
2038
2039         for (p = (__le32 *) buf; !err && aligned_len; aligned_len -= 4, p++) {
2040                 err = t3_seeprom_write(adapter, aligned_offset, *p);
2041                 aligned_offset += 4;
2042         }
2043
2044         if (!err)
2045                 err = t3_seeprom_wp(adapter, 1);
2046 out:
2047         if (buf != data)
2048                 kfree(buf);
2049         return err;
2050 }
2051
2052 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2053 {
2054         wol->supported = 0;
2055         wol->wolopts = 0;
2056         memset(&wol->sopass, 0, sizeof(wol->sopass));
2057 }
2058
2059 static const struct ethtool_ops cxgb_ethtool_ops = {
2060         .get_settings = get_settings,
2061         .set_settings = set_settings,
2062         .get_drvinfo = get_drvinfo,
2063         .get_msglevel = get_msglevel,
2064         .set_msglevel = set_msglevel,
2065         .get_ringparam = get_sge_param,
2066         .set_ringparam = set_sge_param,
2067         .get_coalesce = get_coalesce,
2068         .set_coalesce = set_coalesce,
2069         .get_eeprom_len = get_eeprom_len,
2070         .get_eeprom = get_eeprom,
2071         .set_eeprom = set_eeprom,
2072         .get_pauseparam = get_pauseparam,
2073         .set_pauseparam = set_pauseparam,
2074         .get_rx_csum = get_rx_csum,
2075         .set_rx_csum = set_rx_csum,
2076         .set_tx_csum = ethtool_op_set_tx_csum,
2077         .set_sg = ethtool_op_set_sg,
2078         .get_link = ethtool_op_get_link,
2079         .get_strings = get_strings,
2080         .phys_id = cxgb3_phys_id,
2081         .nway_reset = restart_autoneg,
2082         .get_sset_count = get_sset_count,
2083         .get_ethtool_stats = get_stats,
2084         .get_regs_len = get_regs_len,
2085         .get_regs = get_regs,
2086         .get_wol = get_wol,
2087         .set_tso = ethtool_op_set_tso,
2088 };
2089
2090 static int in_range(int val, int lo, int hi)
2091 {
2092         return val < 0 || (val <= hi && val >= lo);
2093 }
2094
2095 static int cxgb_extension_ioctl(struct net_device *dev, void __user *useraddr)
2096 {
2097         struct port_info *pi = netdev_priv(dev);
2098         struct adapter *adapter = pi->adapter;
2099         u32 cmd;
2100         int ret;
2101
2102         if (copy_from_user(&cmd, useraddr, sizeof(cmd)))
2103                 return -EFAULT;
2104
2105         switch (cmd) {
2106         case CHELSIO_SET_QSET_PARAMS:{
2107                 int i;
2108                 struct qset_params *q;
2109                 struct ch_qset_params t;
2110                 int q1 = pi->first_qset;
2111                 int nqsets = pi->nqsets;
2112
2113                 if (!capable(CAP_NET_ADMIN))
2114                         return -EPERM;
2115                 if (copy_from_user(&t, useraddr, sizeof(t)))
2116                         return -EFAULT;
2117                 if (t.qset_idx >= SGE_QSETS)
2118                         return -EINVAL;
2119                 if (!in_range(t.intr_lat, 0, M_NEWTIMER) ||
2120                     !in_range(t.cong_thres, 0, 255) ||
2121                     !in_range(t.txq_size[0], MIN_TXQ_ENTRIES,
2122                               MAX_TXQ_ENTRIES) ||
2123                     !in_range(t.txq_size[1], MIN_TXQ_ENTRIES,
2124                               MAX_TXQ_ENTRIES) ||
2125                     !in_range(t.txq_size[2], MIN_CTRL_TXQ_ENTRIES,
2126                               MAX_CTRL_TXQ_ENTRIES) ||
2127                     !in_range(t.fl_size[0], MIN_FL_ENTRIES,
2128                               MAX_RX_BUFFERS) ||
2129                     !in_range(t.fl_size[1], MIN_FL_ENTRIES,
2130                               MAX_RX_JUMBO_BUFFERS) ||
2131                     !in_range(t.rspq_size, MIN_RSPQ_ENTRIES,
2132                               MAX_RSPQ_ENTRIES))
2133                         return -EINVAL;
2134
2135                 if ((adapter->flags & FULL_INIT_DONE) && t.lro > 0)
2136                         for_each_port(adapter, i) {
2137                                 pi = adap2pinfo(adapter, i);
2138                                 if (t.qset_idx >= pi->first_qset &&
2139                                     t.qset_idx < pi->first_qset + pi->nqsets &&
2140                                     !(pi->rx_offload & T3_RX_CSUM))
2141                                         return -EINVAL;
2142                         }
2143
2144                 if ((adapter->flags & FULL_INIT_DONE) &&
2145                         (t.rspq_size >= 0 || t.fl_size[0] >= 0 ||
2146                         t.fl_size[1] >= 0 || t.txq_size[0] >= 0 ||
2147                         t.txq_size[1] >= 0 || t.txq_size[2] >= 0 ||
2148                         t.polling >= 0 || t.cong_thres >= 0))
2149                         return -EBUSY;
2150
2151                 /* Allow setting of any available qset when offload enabled */
2152                 if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2153                         q1 = 0;
2154                         for_each_port(adapter, i) {
2155                                 pi = adap2pinfo(adapter, i);
2156                                 nqsets += pi->first_qset + pi->nqsets;
2157                         }
2158                 }
2159
2160                 if (t.qset_idx < q1)
2161                         return -EINVAL;
2162                 if (t.qset_idx > q1 + nqsets - 1)
2163                         return -EINVAL;
2164
2165                 q = &adapter->params.sge.qset[t.qset_idx];
2166
2167                 if (t.rspq_size >= 0)
2168                         q->rspq_size = t.rspq_size;
2169                 if (t.fl_size[0] >= 0)
2170                         q->fl_size = t.fl_size[0];
2171                 if (t.fl_size[1] >= 0)
2172                         q->jumbo_size = t.fl_size[1];
2173                 if (t.txq_size[0] >= 0)
2174                         q->txq_size[0] = t.txq_size[0];
2175                 if (t.txq_size[1] >= 0)
2176                         q->txq_size[1] = t.txq_size[1];
2177                 if (t.txq_size[2] >= 0)
2178                         q->txq_size[2] = t.txq_size[2];
2179                 if (t.cong_thres >= 0)
2180                         q->cong_thres = t.cong_thres;
2181                 if (t.intr_lat >= 0) {
2182                         struct sge_qset *qs =
2183                                 &adapter->sge.qs[t.qset_idx];
2184
2185                         q->coalesce_usecs = t.intr_lat;
2186                         t3_update_qset_coalesce(qs, q);
2187                 }
2188                 if (t.polling >= 0) {
2189                         if (adapter->flags & USING_MSIX)
2190                                 q->polling = t.polling;
2191                         else {
2192                                 /* No polling with INTx for T3A */
2193                                 if (adapter->params.rev == 0 &&
2194                                         !(adapter->flags & USING_MSI))
2195                                         t.polling = 0;
2196
2197                                 for (i = 0; i < SGE_QSETS; i++) {
2198                                         q = &adapter->params.sge.
2199                                                 qset[i];
2200                                         q->polling = t.polling;
2201                                 }
2202                         }
2203                 }
2204                 if (t.lro >= 0)
2205                         set_qset_lro(dev, t.qset_idx, t.lro);
2206
2207                 break;
2208         }
2209         case CHELSIO_GET_QSET_PARAMS:{
2210                 struct qset_params *q;
2211                 struct ch_qset_params t;
2212                 int q1 = pi->first_qset;
2213                 int nqsets = pi->nqsets;
2214                 int i;
2215
2216                 if (copy_from_user(&t, useraddr, sizeof(t)))
2217                         return -EFAULT;
2218
2219                 /* Display qsets for all ports when offload enabled */
2220                 if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2221                         q1 = 0;
2222                         for_each_port(adapter, i) {
2223                                 pi = adap2pinfo(adapter, i);
2224                                 nqsets = pi->first_qset + pi->nqsets;
2225                         }
2226                 }
2227
2228                 if (t.qset_idx >= nqsets)
2229                         return -EINVAL;
2230
2231                 q = &adapter->params.sge.qset[q1 + t.qset_idx];
2232                 t.rspq_size = q->rspq_size;
2233                 t.txq_size[0] = q->txq_size[0];
2234                 t.txq_size[1] = q->txq_size[1];
2235                 t.txq_size[2] = q->txq_size[2];
2236                 t.fl_size[0] = q->fl_size;
2237                 t.fl_size[1] = q->jumbo_size;
2238                 t.polling = q->polling;
2239                 t.lro = q->lro;
2240                 t.intr_lat = q->coalesce_usecs;
2241                 t.cong_thres = q->cong_thres;
2242                 t.qnum = q1;
2243
2244                 if (adapter->flags & USING_MSIX)
2245                         t.vector = adapter->msix_info[q1 + t.qset_idx + 1].vec;
2246                 else
2247                         t.vector = adapter->pdev->irq;
2248
2249                 if (copy_to_user(useraddr, &t, sizeof(t)))
2250                         return -EFAULT;
2251                 break;
2252         }
2253         case CHELSIO_SET_QSET_NUM:{
2254                 struct ch_reg edata;
2255                 unsigned int i, first_qset = 0, other_qsets = 0;
2256
2257                 if (!capable(CAP_NET_ADMIN))
2258                         return -EPERM;
2259                 if (adapter->flags & FULL_INIT_DONE)
2260                         return -EBUSY;
2261                 if (copy_from_user(&edata, useraddr, sizeof(edata)))
2262                         return -EFAULT;
2263                 if (edata.val < 1 ||
2264                         (edata.val > 1 && !(adapter->flags & USING_MSIX)))
2265                         return -EINVAL;
2266
2267                 for_each_port(adapter, i)
2268                         if (adapter->port[i] && adapter->port[i] != dev)
2269                                 other_qsets += adap2pinfo(adapter, i)->nqsets;
2270
2271                 if (edata.val + other_qsets > SGE_QSETS)
2272                         return -EINVAL;
2273
2274                 pi->nqsets = edata.val;
2275
2276                 for_each_port(adapter, i)
2277                         if (adapter->port[i]) {
2278                                 pi = adap2pinfo(adapter, i);
2279                                 pi->first_qset = first_qset;
2280                                 first_qset += pi->nqsets;
2281                         }
2282                 break;
2283         }
2284         case CHELSIO_GET_QSET_NUM:{
2285                 struct ch_reg edata;
2286
2287                 edata.cmd = CHELSIO_GET_QSET_NUM;
2288                 edata.val = pi->nqsets;
2289                 if (copy_to_user(useraddr, &edata, sizeof(edata)))
2290                         return -EFAULT;
2291                 break;
2292         }
2293         case CHELSIO_LOAD_FW:{
2294                 u8 *fw_data;
2295                 struct ch_mem_range t;
2296
2297                 if (!capable(CAP_SYS_RAWIO))
2298                         return -EPERM;
2299                 if (copy_from_user(&t, useraddr, sizeof(t)))
2300                         return -EFAULT;
2301                 /* Check t.len sanity ? */
2302                 fw_data = kmalloc(t.len, GFP_KERNEL);
2303                 if (!fw_data)
2304                         return -ENOMEM;
2305
2306                 if (copy_from_user
2307                         (fw_data, useraddr + sizeof(t), t.len)) {
2308                         kfree(fw_data);
2309                         return -EFAULT;
2310                 }
2311
2312                 ret = t3_load_fw(adapter, fw_data, t.len);
2313                 kfree(fw_data);
2314                 if (ret)
2315                         return ret;
2316                 break;
2317         }
2318         case CHELSIO_SETMTUTAB:{
2319                 struct ch_mtus m;
2320                 int i;
2321
2322                 if (!is_offload(adapter))
2323                         return -EOPNOTSUPP;
2324                 if (!capable(CAP_NET_ADMIN))
2325                         return -EPERM;
2326                 if (offload_running(adapter))
2327                         return -EBUSY;
2328                 if (copy_from_user(&m, useraddr, sizeof(m)))
2329                         return -EFAULT;
2330                 if (m.nmtus != NMTUS)
2331                         return -EINVAL;
2332                 if (m.mtus[0] < 81)     /* accommodate SACK */
2333                         return -EINVAL;
2334
2335                 /* MTUs must be in ascending order */
2336                 for (i = 1; i < NMTUS; ++i)
2337                         if (m.mtus[i] < m.mtus[i - 1])
2338                                 return -EINVAL;
2339
2340                 memcpy(adapter->params.mtus, m.mtus,
2341                         sizeof(adapter->params.mtus));
2342                 break;
2343         }
2344         case CHELSIO_GET_PM:{
2345                 struct tp_params *p = &adapter->params.tp;
2346                 struct ch_pm m = {.cmd = CHELSIO_GET_PM };
2347
2348                 if (!is_offload(adapter))
2349                         return -EOPNOTSUPP;
2350                 m.tx_pg_sz = p->tx_pg_size;
2351                 m.tx_num_pg = p->tx_num_pgs;
2352                 m.rx_pg_sz = p->rx_pg_size;
2353                 m.rx_num_pg = p->rx_num_pgs;
2354                 m.pm_total = p->pmtx_size + p->chan_rx_size * p->nchan;
2355                 if (copy_to_user(useraddr, &m, sizeof(m)))
2356                         return -EFAULT;
2357                 break;
2358         }
2359         case CHELSIO_SET_PM:{
2360                 struct ch_pm m;
2361                 struct tp_params *p = &adapter->params.tp;
2362
2363                 if (!is_offload(adapter))
2364                         return -EOPNOTSUPP;
2365                 if (!capable(CAP_NET_ADMIN))
2366                         return -EPERM;
2367                 if (adapter->flags & FULL_INIT_DONE)
2368                         return -EBUSY;
2369                 if (copy_from_user(&m, useraddr, sizeof(m)))
2370                         return -EFAULT;
2371                 if (!is_power_of_2(m.rx_pg_sz) ||
2372                         !is_power_of_2(m.tx_pg_sz))
2373                         return -EINVAL; /* not power of 2 */
2374                 if (!(m.rx_pg_sz & 0x14000))
2375                         return -EINVAL; /* not 16KB or 64KB */
2376                 if (!(m.tx_pg_sz & 0x1554000))
2377                         return -EINVAL;
2378                 if (m.tx_num_pg == -1)
2379                         m.tx_num_pg = p->tx_num_pgs;
2380                 if (m.rx_num_pg == -1)
2381                         m.rx_num_pg = p->rx_num_pgs;
2382                 if (m.tx_num_pg % 24 || m.rx_num_pg % 24)
2383                         return -EINVAL;
2384                 if (m.rx_num_pg * m.rx_pg_sz > p->chan_rx_size ||
2385                         m.tx_num_pg * m.tx_pg_sz > p->chan_tx_size)
2386                         return -EINVAL;
2387                 p->rx_pg_size = m.rx_pg_sz;
2388                 p->tx_pg_size = m.tx_pg_sz;
2389                 p->rx_num_pgs = m.rx_num_pg;
2390                 p->tx_num_pgs = m.tx_num_pg;
2391                 break;
2392         }
2393         case CHELSIO_GET_MEM:{
2394                 struct ch_mem_range t;
2395                 struct mc7 *mem;
2396                 u64 buf[32];
2397
2398                 if (!is_offload(adapter))
2399                         return -EOPNOTSUPP;
2400                 if (!(adapter->flags & FULL_INIT_DONE))
2401                         return -EIO;    /* need the memory controllers */
2402                 if (copy_from_user(&t, useraddr, sizeof(t)))
2403                         return -EFAULT;
2404                 if ((t.addr & 7) || (t.len & 7))
2405                         return -EINVAL;
2406                 if (t.mem_id == MEM_CM)
2407                         mem = &adapter->cm;
2408                 else if (t.mem_id == MEM_PMRX)
2409                         mem = &adapter->pmrx;
2410                 else if (t.mem_id == MEM_PMTX)
2411                         mem = &adapter->pmtx;
2412                 else
2413                         return -EINVAL;
2414
2415                 /*
2416                  * Version scheme:
2417                  * bits 0..9: chip version
2418                  * bits 10..15: chip revision
2419                  */
2420                 t.version = 3 | (adapter->params.rev << 10);
2421                 if (copy_to_user(useraddr, &t, sizeof(t)))
2422                         return -EFAULT;
2423
2424                 /*
2425                  * Read 256 bytes at a time as len can be large and we don't
2426                  * want to use huge intermediate buffers.
2427                  */
2428                 useraddr += sizeof(t);  /* advance to start of buffer */
2429                 while (t.len) {
2430                         unsigned int chunk =
2431                                 min_t(unsigned int, t.len, sizeof(buf));
2432
2433                         ret =
2434                                 t3_mc7_bd_read(mem, t.addr / 8, chunk / 8,
2435                                                 buf);
2436                         if (ret)
2437                                 return ret;
2438                         if (copy_to_user(useraddr, buf, chunk))
2439                                 return -EFAULT;
2440                         useraddr += chunk;
2441                         t.addr += chunk;
2442                         t.len -= chunk;
2443                 }
2444                 break;
2445         }
2446         case CHELSIO_SET_TRACE_FILTER:{
2447                 struct ch_trace t;
2448                 const struct trace_params *tp;
2449
2450                 if (!capable(CAP_NET_ADMIN))
2451                         return -EPERM;
2452                 if (!offload_running(adapter))
2453                         return -EAGAIN;
2454                 if (copy_from_user(&t, useraddr, sizeof(t)))
2455                         return -EFAULT;
2456
2457                 tp = (const struct trace_params *)&t.sip;
2458                 if (t.config_tx)
2459                         t3_config_trace_filter(adapter, tp, 0,
2460                                                 t.invert_match,
2461                                                 t.trace_tx);
2462                 if (t.config_rx)
2463                         t3_config_trace_filter(adapter, tp, 1,
2464                                                 t.invert_match,
2465                                                 t.trace_rx);
2466                 break;
2467         }
2468         default:
2469                 return -EOPNOTSUPP;
2470         }
2471         return 0;
2472 }
2473
2474 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2475 {
2476         struct mii_ioctl_data *data = if_mii(req);
2477         struct port_info *pi = netdev_priv(dev);
2478         struct adapter *adapter = pi->adapter;
2479
2480         switch (cmd) {
2481         case SIOCGMIIREG:
2482         case SIOCSMIIREG:
2483                 /* Convert phy_id from older PRTAD/DEVAD format */
2484                 if (is_10G(adapter) &&
2485                     !mdio_phy_id_is_c45(data->phy_id) &&
2486                     (data->phy_id & 0x1f00) &&
2487                     !(data->phy_id & 0xe0e0))
2488                         data->phy_id = mdio_phy_id_c45(data->phy_id >> 8,
2489                                                        data->phy_id & 0x1f);
2490                 /* FALLTHRU */
2491         case SIOCGMIIPHY:
2492                 return mdio_mii_ioctl(&pi->phy.mdio, data, cmd);
2493         case SIOCCHIOCTL:
2494                 return cxgb_extension_ioctl(dev, req->ifr_data);
2495         default:
2496                 return -EOPNOTSUPP;
2497         }
2498 }
2499
2500 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2501 {
2502         struct port_info *pi = netdev_priv(dev);
2503         struct adapter *adapter = pi->adapter;
2504         int ret;
2505
2506         if (new_mtu < 81)       /* accommodate SACK */
2507                 return -EINVAL;
2508         if ((ret = t3_mac_set_mtu(&pi->mac, new_mtu)))
2509                 return ret;
2510         dev->mtu = new_mtu;
2511         init_port_mtus(adapter);
2512         if (adapter->params.rev == 0 && offload_running(adapter))
2513                 t3_load_mtus(adapter, adapter->params.mtus,
2514                              adapter->params.a_wnd, adapter->params.b_wnd,
2515                              adapter->port[0]->mtu);
2516         return 0;
2517 }
2518
2519 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2520 {
2521         struct port_info *pi = netdev_priv(dev);
2522         struct adapter *adapter = pi->adapter;
2523         struct sockaddr *addr = p;
2524
2525         if (!is_valid_ether_addr(addr->sa_data))
2526                 return -EINVAL;
2527
2528         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2529         t3_mac_set_address(&pi->mac, LAN_MAC_IDX, dev->dev_addr);
2530         if (offload_running(adapter))
2531                 write_smt_entry(adapter, pi->port_id);
2532         return 0;
2533 }
2534
2535 /**
2536  * t3_synchronize_rx - wait for current Rx processing on a port to complete
2537  * @adap: the adapter
2538  * @p: the port
2539  *
2540  * Ensures that current Rx processing on any of the queues associated with
2541  * the given port completes before returning.  We do this by acquiring and
2542  * releasing the locks of the response queues associated with the port.
2543  */
2544 static void t3_synchronize_rx(struct adapter *adap, const struct port_info *p)
2545 {
2546         int i;
2547
2548         for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
2549                 struct sge_rspq *q = &adap->sge.qs[i].rspq;
2550
2551                 spin_lock_irq(&q->lock);
2552                 spin_unlock_irq(&q->lock);
2553         }
2554 }
2555
2556 static void vlan_rx_register(struct net_device *dev, struct vlan_group *grp)
2557 {
2558         struct port_info *pi = netdev_priv(dev);
2559         struct adapter *adapter = pi->adapter;
2560
2561         pi->vlan_grp = grp;
2562         if (adapter->params.rev > 0)
2563                 t3_set_vlan_accel(adapter, 1 << pi->port_id, grp != NULL);
2564         else {
2565                 /* single control for all ports */
2566                 unsigned int i, have_vlans = 0;
2567                 for_each_port(adapter, i)
2568                     have_vlans |= adap2pinfo(adapter, i)->vlan_grp != NULL;
2569
2570                 t3_set_vlan_accel(adapter, 1, have_vlans);
2571         }
2572         t3_synchronize_rx(adapter, pi);
2573 }
2574
2575 #ifdef CONFIG_NET_POLL_CONTROLLER
2576 static void cxgb_netpoll(struct net_device *dev)
2577 {
2578         struct port_info *pi = netdev_priv(dev);
2579         struct adapter *adapter = pi->adapter;
2580         int qidx;
2581
2582         for (qidx = pi->first_qset; qidx < pi->first_qset + pi->nqsets; qidx++) {
2583                 struct sge_qset *qs = &adapter->sge.qs[qidx];
2584                 void *source;
2585
2586                 if (adapter->flags & USING_MSIX)
2587                         source = qs;
2588                 else
2589                         source = adapter;
2590
2591                 t3_intr_handler(adapter, qs->rspq.polling) (0, source);
2592         }
2593 }
2594 #endif
2595
2596 /*
2597  * Periodic accumulation of MAC statistics.
2598  */
2599 static void mac_stats_update(struct adapter *adapter)
2600 {
2601         int i;
2602
2603         for_each_port(adapter, i) {
2604                 struct net_device *dev = adapter->port[i];
2605                 struct port_info *p = netdev_priv(dev);
2606
2607                 if (netif_running(dev)) {
2608                         spin_lock(&adapter->stats_lock);
2609                         t3_mac_update_stats(&p->mac);
2610                         spin_unlock(&adapter->stats_lock);
2611                 }
2612         }
2613 }
2614
2615 static void check_link_status(struct adapter *adapter)
2616 {
2617         int i;
2618
2619         for_each_port(adapter, i) {
2620                 struct net_device *dev = adapter->port[i];
2621                 struct port_info *p = netdev_priv(dev);
2622                 int link_fault;
2623
2624                 spin_lock_irq(&adapter->work_lock);
2625                 link_fault = p->link_fault;
2626                 spin_unlock_irq(&adapter->work_lock);
2627
2628                 if (link_fault) {
2629                         t3_link_fault(adapter, i);
2630                         continue;
2631                 }
2632
2633                 if (!(p->phy.caps & SUPPORTED_IRQ) && netif_running(dev)) {
2634                         t3_xgm_intr_disable(adapter, i);
2635                         t3_read_reg(adapter, A_XGM_INT_STATUS + p->mac.offset);
2636
2637                         t3_link_changed(adapter, i);
2638                         t3_xgm_intr_enable(adapter, i);
2639                 }
2640         }
2641 }
2642
2643 static void check_t3b2_mac(struct adapter *adapter)
2644 {
2645         int i;
2646
2647         if (!rtnl_trylock())    /* synchronize with ifdown */
2648                 return;
2649
2650         for_each_port(adapter, i) {
2651                 struct net_device *dev = adapter->port[i];
2652                 struct port_info *p = netdev_priv(dev);
2653                 int status;
2654
2655                 if (!netif_running(dev))
2656                         continue;
2657
2658                 status = 0;
2659                 if (netif_running(dev) && netif_carrier_ok(dev))
2660                         status = t3b2_mac_watchdog_task(&p->mac);
2661                 if (status == 1)
2662                         p->mac.stats.num_toggled++;
2663                 else if (status == 2) {
2664                         struct cmac *mac = &p->mac;
2665
2666                         t3_mac_set_mtu(mac, dev->mtu);
2667                         t3_mac_set_address(mac, LAN_MAC_IDX, dev->dev_addr);
2668                         cxgb_set_rxmode(dev);
2669                         t3_link_start(&p->phy, mac, &p->link_config);
2670                         t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
2671                         t3_port_intr_enable(adapter, p->port_id);
2672                         p->mac.stats.num_resets++;
2673                 }
2674         }
2675         rtnl_unlock();
2676 }
2677
2678
2679 static void t3_adap_check_task(struct work_struct *work)
2680 {
2681         struct adapter *adapter = container_of(work, struct adapter,
2682                                                adap_check_task.work);
2683         const struct adapter_params *p = &adapter->params;
2684         int port;
2685         unsigned int v, status, reset;
2686
2687         adapter->check_task_cnt++;
2688
2689         check_link_status(adapter);
2690
2691         /* Accumulate MAC stats if needed */
2692         if (!p->linkpoll_period ||
2693             (adapter->check_task_cnt * p->linkpoll_period) / 10 >=
2694             p->stats_update_period) {
2695                 mac_stats_update(adapter);
2696                 adapter->check_task_cnt = 0;
2697         }
2698
2699         if (p->rev == T3_REV_B2)
2700                 check_t3b2_mac(adapter);
2701
2702         /*
2703          * Scan the XGMAC's to check for various conditions which we want to
2704          * monitor in a periodic polling manner rather than via an interrupt
2705          * condition.  This is used for conditions which would otherwise flood
2706          * the system with interrupts and we only really need to know that the
2707          * conditions are "happening" ...  For each condition we count the
2708          * detection of the condition and reset it for the next polling loop.
2709          */
2710         for_each_port(adapter, port) {
2711                 struct cmac *mac =  &adap2pinfo(adapter, port)->mac;
2712                 u32 cause;
2713
2714                 cause = t3_read_reg(adapter, A_XGM_INT_CAUSE + mac->offset);
2715                 reset = 0;
2716                 if (cause & F_RXFIFO_OVERFLOW) {
2717                         mac->stats.rx_fifo_ovfl++;
2718                         reset |= F_RXFIFO_OVERFLOW;
2719                 }
2720
2721                 t3_write_reg(adapter, A_XGM_INT_CAUSE + mac->offset, reset);
2722         }
2723
2724         /*
2725          * We do the same as above for FL_EMPTY interrupts.
2726          */
2727         status = t3_read_reg(adapter, A_SG_INT_CAUSE);
2728         reset = 0;
2729
2730         if (status & F_FLEMPTY) {
2731                 struct sge_qset *qs = &adapter->sge.qs[0];
2732                 int i = 0;
2733
2734                 reset |= F_FLEMPTY;
2735
2736                 v = (t3_read_reg(adapter, A_SG_RSPQ_FL_STATUS) >> S_FL0EMPTY) &
2737                     0xffff;
2738
2739                 while (v) {
2740                         qs->fl[i].empty += (v & 1);
2741                         if (i)
2742                                 qs++;
2743                         i ^= 1;
2744                         v >>= 1;
2745                 }
2746         }
2747
2748         t3_write_reg(adapter, A_SG_INT_CAUSE, reset);
2749
2750         /* Schedule the next check update if any port is active. */
2751         spin_lock_irq(&adapter->work_lock);
2752         if (adapter->open_device_map & PORT_MASK)
2753                 schedule_chk_task(adapter);
2754         spin_unlock_irq(&adapter->work_lock);
2755 }
2756
2757 /*
2758  * Processes external (PHY) interrupts in process context.
2759  */
2760 static void ext_intr_task(struct work_struct *work)
2761 {
2762         struct adapter *adapter = container_of(work, struct adapter,
2763                                                ext_intr_handler_task);
2764         int i;
2765
2766         /* Disable link fault interrupts */
2767         for_each_port(adapter, i) {
2768                 struct net_device *dev = adapter->port[i];
2769                 struct port_info *p = netdev_priv(dev);
2770
2771                 t3_xgm_intr_disable(adapter, i);
2772                 t3_read_reg(adapter, A_XGM_INT_STATUS + p->mac.offset);
2773         }
2774
2775         /* Re-enable link fault interrupts */
2776         t3_phy_intr_handler(adapter);
2777
2778         for_each_port(adapter, i)
2779                 t3_xgm_intr_enable(adapter, i);
2780
2781         /* Now reenable external interrupts */
2782         spin_lock_irq(&adapter->work_lock);
2783         if (adapter->slow_intr_mask) {
2784                 adapter->slow_intr_mask |= F_T3DBG;
2785                 t3_write_reg(adapter, A_PL_INT_CAUSE0, F_T3DBG);
2786                 t3_write_reg(adapter, A_PL_INT_ENABLE0,
2787                              adapter->slow_intr_mask);
2788         }
2789         spin_unlock_irq(&adapter->work_lock);
2790 }
2791
2792 /*
2793  * Interrupt-context handler for external (PHY) interrupts.
2794  */
2795 void t3_os_ext_intr_handler(struct adapter *adapter)
2796 {
2797         /*
2798          * Schedule a task to handle external interrupts as they may be slow
2799          * and we use a mutex to protect MDIO registers.  We disable PHY
2800          * interrupts in the meantime and let the task reenable them when
2801          * it's done.
2802          */
2803         spin_lock(&adapter->work_lock);
2804         if (adapter->slow_intr_mask) {
2805                 adapter->slow_intr_mask &= ~F_T3DBG;
2806                 t3_write_reg(adapter, A_PL_INT_ENABLE0,
2807                              adapter->slow_intr_mask);
2808                 queue_work(cxgb3_wq, &adapter->ext_intr_handler_task);
2809         }
2810         spin_unlock(&adapter->work_lock);
2811 }
2812
2813 void t3_os_link_fault_handler(struct adapter *adapter, int port_id)
2814 {
2815         struct net_device *netdev = adapter->port[port_id];
2816         struct port_info *pi = netdev_priv(netdev);
2817
2818         spin_lock(&adapter->work_lock);
2819         pi->link_fault = 1;
2820         spin_unlock(&adapter->work_lock);
2821 }
2822
2823 static int t3_adapter_error(struct adapter *adapter, int reset)
2824 {
2825         int i, ret = 0;
2826
2827         if (is_offload(adapter) &&
2828             test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2829                 cxgb3_event_notify(&adapter->tdev, OFFLOAD_STATUS_DOWN, 0);
2830                 offload_close(&adapter->tdev);
2831         }
2832
2833         /* Stop all ports */
2834         for_each_port(adapter, i) {
2835                 struct net_device *netdev = adapter->port[i];
2836
2837                 if (netif_running(netdev))
2838                         cxgb_close(netdev);
2839         }
2840
2841         /* Stop SGE timers */
2842         t3_stop_sge_timers(adapter);
2843
2844         adapter->flags &= ~FULL_INIT_DONE;
2845
2846         if (reset)
2847                 ret = t3_reset_adapter(adapter);
2848
2849         pci_disable_device(adapter->pdev);
2850
2851         return ret;
2852 }
2853
2854 static int t3_reenable_adapter(struct adapter *adapter)
2855 {
2856         if (pci_enable_device(adapter->pdev)) {
2857                 dev_err(&adapter->pdev->dev,
2858                         "Cannot re-enable PCI device after reset.\n");
2859                 goto err;
2860         }
2861         pci_set_master(adapter->pdev);
2862         pci_restore_state(adapter->pdev);
2863         pci_save_state(adapter->pdev);
2864
2865         /* Free sge resources */
2866         t3_free_sge_resources(adapter);
2867
2868         if (t3_replay_prep_adapter(adapter))
2869                 goto err;
2870
2871         return 0;
2872 err:
2873         return -1;
2874 }
2875
2876 static void t3_resume_ports(struct adapter *adapter)
2877 {
2878         int i;
2879
2880         /* Restart the ports */
2881         for_each_port(adapter, i) {
2882                 struct net_device *netdev = adapter->port[i];
2883
2884                 if (netif_running(netdev)) {
2885                         if (cxgb_open(netdev)) {
2886                                 dev_err(&adapter->pdev->dev,
2887                                         "can't bring device back up"
2888                                         " after reset\n");
2889                                 continue;
2890                         }
2891                 }
2892         }
2893
2894         if (is_offload(adapter) && !ofld_disable)
2895                 cxgb3_event_notify(&adapter->tdev, OFFLOAD_STATUS_UP, 0);
2896 }
2897
2898 /*
2899  * processes a fatal error.
2900  * Bring the ports down, reset the chip, bring the ports back up.
2901  */
2902 static void fatal_error_task(struct work_struct *work)
2903 {
2904         struct adapter *adapter = container_of(work, struct adapter,
2905                                                fatal_error_handler_task);
2906         int err = 0;
2907
2908         rtnl_lock();
2909         err = t3_adapter_error(adapter, 1);
2910         if (!err)
2911                 err = t3_reenable_adapter(adapter);
2912         if (!err)
2913                 t3_resume_ports(adapter);
2914
2915         CH_ALERT(adapter, "adapter reset %s\n", err ? "failed" : "succeeded");
2916         rtnl_unlock();
2917 }
2918
2919 void t3_fatal_err(struct adapter *adapter)
2920 {
2921         unsigned int fw_status[4];
2922
2923         if (adapter->flags & FULL_INIT_DONE) {
2924                 t3_sge_stop(adapter);
2925                 t3_write_reg(adapter, A_XGM_TX_CTRL, 0);
2926                 t3_write_reg(adapter, A_XGM_RX_CTRL, 0);
2927                 t3_write_reg(adapter, XGM_REG(A_XGM_TX_CTRL, 1), 0);
2928                 t3_write_reg(adapter, XGM_REG(A_XGM_RX_CTRL, 1), 0);
2929
2930                 spin_lock(&adapter->work_lock);
2931                 t3_intr_disable(adapter);
2932                 queue_work(cxgb3_wq, &adapter->fatal_error_handler_task);
2933                 spin_unlock(&adapter->work_lock);
2934         }
2935         CH_ALERT(adapter, "encountered fatal error, operation suspended\n");
2936         if (!t3_cim_ctl_blk_read(adapter, 0xa0, 4, fw_status))
2937                 CH_ALERT(adapter, "FW status: 0x%x, 0x%x, 0x%x, 0x%x\n",
2938                          fw_status[0], fw_status[1],
2939                          fw_status[2], fw_status[3]);
2940 }
2941
2942 /**
2943  * t3_io_error_detected - called when PCI error is detected
2944  * @pdev: Pointer to PCI device
2945  * @state: The current pci connection state
2946  *
2947  * This function is called after a PCI bus error affecting
2948  * this device has been detected.
2949  */
2950 static pci_ers_result_t t3_io_error_detected(struct pci_dev *pdev,
2951                                              pci_channel_state_t state)
2952 {
2953         struct adapter *adapter = pci_get_drvdata(pdev);
2954         int ret;
2955
2956         if (state == pci_channel_io_perm_failure)
2957                 return PCI_ERS_RESULT_DISCONNECT;
2958
2959         ret = t3_adapter_error(adapter, 0);
2960
2961         /* Request a slot reset. */
2962         return PCI_ERS_RESULT_NEED_RESET;
2963 }
2964
2965 /**
2966  * t3_io_slot_reset - called after the pci bus has been reset.
2967  * @pdev: Pointer to PCI device
2968  *
2969  * Restart the card from scratch, as if from a cold-boot.
2970  */
2971 static pci_ers_result_t t3_io_slot_reset(struct pci_dev *pdev)
2972 {
2973         struct adapter *adapter = pci_get_drvdata(pdev);
2974
2975         if (!t3_reenable_adapter(adapter))
2976                 return PCI_ERS_RESULT_RECOVERED;
2977
2978         return PCI_ERS_RESULT_DISCONNECT;
2979 }
2980
2981 /**
2982  * t3_io_resume - called when traffic can start flowing again.
2983  * @pdev: Pointer to PCI device
2984  *
2985  * This callback is called when the error recovery driver tells us that
2986  * its OK to resume normal operation.
2987  */
2988 static void t3_io_resume(struct pci_dev *pdev)
2989 {
2990         struct adapter *adapter = pci_get_drvdata(pdev);
2991
2992         CH_ALERT(adapter, "adapter recovering, PEX ERR 0x%x\n",
2993                  t3_read_reg(adapter, A_PCIE_PEX_ERR));
2994
2995         t3_resume_ports(adapter);
2996 }
2997
2998 static struct pci_error_handlers t3_err_handler = {
2999         .error_detected = t3_io_error_detected,
3000         .slot_reset = t3_io_slot_reset,
3001         .resume = t3_io_resume,
3002 };
3003
3004 /*
3005  * Set the number of qsets based on the number of CPUs and the number of ports,
3006  * not to exceed the number of available qsets, assuming there are enough qsets
3007  * per port in HW.
3008  */
3009 static void set_nqsets(struct adapter *adap)
3010 {
3011         int i, j = 0;
3012         int num_cpus = num_online_cpus();
3013         int hwports = adap->params.nports;
3014         int nqsets = adap->msix_nvectors - 1;
3015
3016         if (adap->params.rev > 0 && adap->flags & USING_MSIX) {
3017                 if (hwports == 2 &&
3018                     (hwports * nqsets > SGE_QSETS ||
3019                      num_cpus >= nqsets / hwports))
3020                         nqsets /= hwports;
3021                 if (nqsets > num_cpus)
3022                         nqsets = num_cpus;
3023                 if (nqsets < 1 || hwports == 4)
3024                         nqsets = 1;
3025         } else
3026                 nqsets = 1;
3027
3028         for_each_port(adap, i) {
3029                 struct port_info *pi = adap2pinfo(adap, i);
3030
3031                 pi->first_qset = j;
3032                 pi->nqsets = nqsets;
3033                 j = pi->first_qset + nqsets;
3034
3035                 dev_info(&adap->pdev->dev,
3036                          "Port %d using %d queue sets.\n", i, nqsets);
3037         }
3038 }
3039
3040 static int __devinit cxgb_enable_msix(struct adapter *adap)
3041 {
3042         struct msix_entry entries[SGE_QSETS + 1];
3043         int vectors;
3044         int i, err;
3045
3046         vectors = ARRAY_SIZE(entries);
3047         for (i = 0; i < vectors; ++i)
3048                 entries[i].entry = i;
3049
3050         while ((err = pci_enable_msix(adap->pdev, entries, vectors)) > 0)
3051                 vectors = err;
3052
3053         if (err < 0)
3054                 pci_disable_msix(adap->pdev);
3055
3056         if (!err && vectors < (adap->params.nports + 1)) {
3057                 pci_disable_msix(adap->pdev);
3058                 err = -1;
3059         }
3060
3061         if (!err) {
3062                 for (i = 0; i < vectors; ++i)
3063                         adap->msix_info[i].vec = entries[i].vector;
3064                 adap->msix_nvectors = vectors;
3065         }
3066
3067         return err;
3068 }
3069
3070 static void __devinit print_port_info(struct adapter *adap,
3071                                       const struct adapter_info *ai)
3072 {
3073         static const char *pci_variant[] = {
3074                 "PCI", "PCI-X", "PCI-X ECC", "PCI-X 266", "PCI Express"
3075         };
3076
3077         int i;
3078         char buf[80];
3079
3080         if (is_pcie(adap))
3081                 snprintf(buf, sizeof(buf), "%s x%d",
3082                          pci_variant[adap->params.pci.variant],
3083                          adap->params.pci.width);
3084         else
3085                 snprintf(buf, sizeof(buf), "%s %dMHz/%d-bit",
3086                          pci_variant[adap->params.pci.variant],
3087                          adap->params.pci.speed, adap->params.pci.width);
3088
3089         for_each_port(adap, i) {
3090                 struct net_device *dev = adap->port[i];
3091                 const struct port_info *pi = netdev_priv(dev);
3092
3093                 if (!test_bit(i, &adap->registered_device_map))
3094                         continue;
3095                 printk(KERN_INFO "%s: %s %s %sNIC (rev %d) %s%s\n",
3096                        dev->name, ai->desc, pi->phy.desc,
3097                        is_offload(adap) ? "R" : "", adap->params.rev, buf,
3098                        (adap->flags & USING_MSIX) ? " MSI-X" :
3099                        (adap->flags & USING_MSI) ? " MSI" : "");
3100                 if (adap->name == dev->name && adap->params.vpd.mclk)
3101                         printk(KERN_INFO
3102                                "%s: %uMB CM, %uMB PMTX, %uMB PMRX, S/N: %s\n",
3103                                adap->name, t3_mc7_size(&adap->cm) >> 20,
3104                                t3_mc7_size(&adap->pmtx) >> 20,
3105                                t3_mc7_size(&adap->pmrx) >> 20,
3106                                adap->params.vpd.sn);
3107         }
3108 }
3109
3110 static const struct net_device_ops cxgb_netdev_ops = {
3111         .ndo_open               = cxgb_open,
3112         .ndo_stop               = cxgb_close,
3113         .ndo_start_xmit         = t3_eth_xmit,
3114         .ndo_get_stats          = cxgb_get_stats,
3115         .ndo_validate_addr      = eth_validate_addr,
3116         .ndo_set_multicast_list = cxgb_set_rxmode,
3117         .ndo_do_ioctl           = cxgb_ioctl,
3118         .ndo_change_mtu         = cxgb_change_mtu,
3119         .ndo_set_mac_address    = cxgb_set_mac_addr,
3120         .ndo_vlan_rx_register   = vlan_rx_register,
3121 #ifdef CONFIG_NET_POLL_CONTROLLER
3122         .ndo_poll_controller    = cxgb_netpoll,
3123 #endif
3124 };
3125
3126 static void __devinit cxgb3_init_iscsi_mac(struct net_device *dev)
3127 {
3128         struct port_info *pi = netdev_priv(dev);
3129
3130         memcpy(pi->iscsic.mac_addr, dev->dev_addr, ETH_ALEN);
3131         pi->iscsic.mac_addr[3] |= 0x80;
3132 }
3133
3134 static int __devinit init_one(struct pci_dev *pdev,
3135                               const struct pci_device_id *ent)
3136 {
3137         static int version_printed;
3138
3139         int i, err, pci_using_dac = 0;
3140         resource_size_t mmio_start, mmio_len;
3141         const struct adapter_info *ai;
3142         struct adapter *adapter = NULL;
3143         struct port_info *pi;
3144
3145         if (!version_printed) {
3146                 printk(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
3147                 ++version_printed;
3148         }
3149
3150         if (!cxgb3_wq) {
3151                 cxgb3_wq = create_singlethread_workqueue(DRV_NAME);
3152                 if (!cxgb3_wq) {
3153                         printk(KERN_ERR DRV_NAME
3154                                ": cannot initialize work queue\n");
3155                         return -ENOMEM;
3156                 }
3157         }
3158
3159         err = pci_request_regions(pdev, DRV_NAME);
3160         if (err) {
3161                 /* Just info, some other driver may have claimed the device. */
3162                 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
3163                 return err;
3164         }
3165
3166         err = pci_enable_device(pdev);
3167         if (err) {
3168                 dev_err(&pdev->dev, "cannot enable PCI device\n");
3169                 goto out_release_regions;
3170         }
3171
3172         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
3173                 pci_using_dac = 1;
3174                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3175                 if (err) {
3176                         dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
3177                                "coherent allocations\n");
3178                         goto out_disable_device;
3179                 }
3180         } else if ((err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32))) != 0) {
3181                 dev_err(&pdev->dev, "no usable DMA configuration\n");
3182                 goto out_disable_device;
3183         }
3184
3185         pci_set_master(pdev);
3186         pci_save_state(pdev);
3187
3188         mmio_start = pci_resource_start(pdev, 0);
3189         mmio_len = pci_resource_len(pdev, 0);
3190         ai = t3_get_adapter_info(ent->driver_data);
3191
3192         adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
3193         if (!adapter) {
3194                 err = -ENOMEM;
3195                 goto out_disable_device;
3196         }
3197
3198         adapter->nofail_skb =
3199                 alloc_skb(sizeof(struct cpl_set_tcb_field), GFP_KERNEL);
3200         if (!adapter->nofail_skb) {
3201                 dev_err(&pdev->dev, "cannot allocate nofail buffer\n");
3202                 err = -ENOMEM;
3203                 goto out_free_adapter;
3204         }
3205
3206         adapter->regs = ioremap_nocache(mmio_start, mmio_len);
3207         if (!adapter->regs) {
3208                 dev_err(&pdev->dev, "cannot map device registers\n");
3209                 err = -ENOMEM;
3210                 goto out_free_adapter;
3211         }
3212
3213         adapter->pdev = pdev;
3214         adapter->name = pci_name(pdev);
3215         adapter->msg_enable = dflt_msg_enable;
3216         adapter->mmio_len = mmio_len;
3217
3218         mutex_init(&adapter->mdio_lock);
3219         spin_lock_init(&adapter->work_lock);
3220         spin_lock_init(&adapter->stats_lock);
3221
3222         INIT_LIST_HEAD(&adapter->adapter_list);
3223         INIT_WORK(&adapter->ext_intr_handler_task, ext_intr_task);
3224         INIT_WORK(&adapter->fatal_error_handler_task, fatal_error_task);
3225         INIT_DELAYED_WORK(&adapter->adap_check_task, t3_adap_check_task);
3226
3227         for (i = 0; i < ai->nports0 + ai->nports1; ++i) {
3228                 struct net_device *netdev;
3229
3230                 netdev = alloc_etherdev_mq(sizeof(struct port_info), SGE_QSETS);
3231                 if (!netdev) {
3232                         err = -ENOMEM;
3233                         goto out_free_dev;
3234                 }
3235
3236                 SET_NETDEV_DEV(netdev, &pdev->dev);
3237
3238                 adapter->port[i] = netdev;
3239                 pi = netdev_priv(netdev);
3240                 pi->adapter = adapter;
3241                 pi->rx_offload = T3_RX_CSUM | T3_LRO;
3242                 pi->port_id = i;
3243                 netif_carrier_off(netdev);
3244                 netif_tx_stop_all_queues(netdev);
3245                 netdev->irq = pdev->irq;
3246                 netdev->mem_start = mmio_start;
3247                 netdev->mem_end = mmio_start + mmio_len - 1;
3248                 netdev->features |= NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
3249                 netdev->features |= NETIF_F_GRO;
3250                 if (pci_using_dac)
3251                         netdev->features |= NETIF_F_HIGHDMA;
3252
3253                 netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
3254                 netdev->netdev_ops = &cxgb_netdev_ops;
3255                 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
3256         }
3257
3258         pci_set_drvdata(pdev, adapter);
3259         if (t3_prep_adapter(adapter, ai, 1) < 0) {
3260                 err = -ENODEV;
3261                 goto out_free_dev;
3262         }
3263
3264         /*
3265          * The card is now ready to go.  If any errors occur during device
3266          * registration we do not fail the whole card but rather proceed only
3267          * with the ports we manage to register successfully.  However we must
3268          * register at least one net device.
3269          */
3270         for_each_port(adapter, i) {
3271                 err = register_netdev(adapter->port[i]);
3272                 if (err)
3273                         dev_warn(&pdev->dev,
3274                                  "cannot register net device %s, skipping\n",
3275                                  adapter->port[i]->name);
3276                 else {
3277                         /*
3278                          * Change the name we use for messages to the name of
3279                          * the first successfully registered interface.
3280                          */
3281                         if (!adapter->registered_device_map)
3282                                 adapter->name = adapter->port[i]->name;
3283
3284                         __set_bit(i, &adapter->registered_device_map);
3285                 }
3286         }
3287         if (!adapter->registered_device_map) {
3288                 dev_err(&pdev->dev, "could not register any net devices\n");
3289                 goto out_free_dev;
3290         }
3291
3292         for_each_port(adapter, i)
3293                 cxgb3_init_iscsi_mac(adapter->port[i]);
3294
3295         /* Driver's ready. Reflect it on LEDs */
3296         t3_led_ready(adapter);
3297
3298         if (is_offload(adapter)) {
3299                 __set_bit(OFFLOAD_DEVMAP_BIT, &adapter->registered_device_map);
3300                 cxgb3_adapter_ofld(adapter);
3301         }
3302
3303         /* See what interrupts we'll be using */
3304         if (msi > 1 && cxgb_enable_msix(adapter) == 0)
3305                 adapter->flags |= USING_MSIX;
3306         else if (msi > 0 && pci_enable_msi(pdev) == 0)
3307                 adapter->flags |= USING_MSI;
3308
3309         set_nqsets(adapter);
3310
3311         err = sysfs_create_group(&adapter->port[0]->dev.kobj,
3312                                  &cxgb3_attr_group);
3313
3314         print_port_info(adapter, ai);
3315         return 0;
3316
3317 out_free_dev:
3318         iounmap(adapter->regs);
3319         for (i = ai->nports0 + ai->nports1 - 1; i >= 0; --i)
3320                 if (adapter->port[i])
3321                         free_netdev(adapter->port[i]);
3322
3323 out_free_adapter:
3324         kfree(adapter);
3325
3326 out_disable_device:
3327         pci_disable_device(pdev);
3328 out_release_regions:
3329         pci_release_regions(pdev);
3330         pci_set_drvdata(pdev, NULL);
3331         return err;
3332 }
3333
3334 static void __devexit remove_one(struct pci_dev *pdev)
3335 {
3336         struct adapter *adapter = pci_get_drvdata(pdev);
3337
3338         if (adapter) {
3339                 int i;
3340
3341                 t3_sge_stop(adapter);
3342                 sysfs_remove_group(&adapter->port[0]->dev.kobj,
3343                                    &cxgb3_attr_group);
3344
3345                 if (is_offload(adapter)) {
3346                         cxgb3_adapter_unofld(adapter);
3347                         if (test_bit(OFFLOAD_DEVMAP_BIT,
3348                                      &adapter->open_device_map))
3349                                 offload_close(&adapter->tdev);
3350                 }
3351
3352                 for_each_port(adapter, i)
3353                     if (test_bit(i, &adapter->registered_device_map))
3354                         unregister_netdev(adapter->port[i]);
3355
3356                 t3_stop_sge_timers(adapter);
3357                 t3_free_sge_resources(adapter);
3358                 cxgb_disable_msi(adapter);
3359
3360                 for_each_port(adapter, i)
3361                         if (adapter->port[i])
3362                                 free_netdev(adapter->port[i]);
3363
3364                 iounmap(adapter->regs);
3365                 if (adapter->nofail_skb)
3366                         kfree_skb(adapter->nofail_skb);
3367                 kfree(adapter);
3368                 pci_release_regions(pdev);
3369                 pci_disable_device(pdev);
3370                 pci_set_drvdata(pdev, NULL);
3371         }
3372 }
3373
3374 static struct pci_driver driver = {
3375         .name = DRV_NAME,
3376         .id_table = cxgb3_pci_tbl,
3377         .probe = init_one,
3378         .remove = __devexit_p(remove_one),
3379         .err_handler = &t3_err_handler,
3380 };
3381
3382 static int __init cxgb3_init_module(void)
3383 {
3384         int ret;
3385
3386         cxgb3_offload_init();
3387
3388         ret = pci_register_driver(&driver);
3389         return ret;
3390 }
3391
3392 static void __exit cxgb3_cleanup_module(void)
3393 {
3394         pci_unregister_driver(&driver);
3395         if (cxgb3_wq)
3396                 destroy_workqueue(cxgb3_wq);
3397 }
3398
3399 module_init(cxgb3_init_module);
3400 module_exit(cxgb3_cleanup_module);