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 <linux/sched.h>
  49 #include <linux/slab.h>
  50 #include <asm/uaccess.h>
  51
  52 #include "common.h"
  53 #include "cxgb3_ioctl.h"
  54 #include "regs.h"
  55 #include "cxgb3_offload.h"
  56 #include "version.h"
  57
  58 #include "cxgb3_ctl_defs.h"
  59 #include "t3_cpl.h"
  60 #include "firmware_exports.h"
  61
  62 enum {
  63         MAX_TXQ_ENTRIES = 16384,
  64         MAX_CTRL_TXQ_ENTRIES = 1024,
  65         MAX_RSPQ_ENTRIES = 16384,
  66         MAX_RX_BUFFERS = 16384,
  67         MAX_RX_JUMBO_BUFFERS = 16384,
  68         MIN_TXQ_ENTRIES = 4,
  69         MIN_CTRL_TXQ_ENTRIES = 4,
  70         MIN_RSPQ_ENTRIES = 32,
  71         MIN_FL_ENTRIES = 32
  72 };
  73
  74 #define PORT_MASK ((1 << MAX_NPORTS) - 1)
  75
  76 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
  77                          NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
  78                          NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
  79
  80 #define EEPROM_MAGIC 0x38E2F10C
  81
  82 #define CH_DEVICE(devid, idx) \
  83         { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, idx }
  84
  85 static DEFINE_PCI_DEVICE_TABLE(cxgb3_pci_tbl) = {
  86         CH_DEVICE(0x20, 0),     /* PE9000 */
  87         CH_DEVICE(0x21, 1),     /* T302E */
  88         CH_DEVICE(0x22, 2),     /* T310E */
  89         CH_DEVICE(0x23, 3),     /* T320X */
  90         CH_DEVICE(0x24, 1),     /* T302X */
  91         CH_DEVICE(0x25, 3),     /* T320E */
  92         CH_DEVICE(0x26, 2),     /* T310X */
  93         CH_DEVICE(0x30, 2),     /* T3B10 */
  94         CH_DEVICE(0x31, 3),     /* T3B20 */
  95         CH_DEVICE(0x32, 1),     /* T3B02 */
  96         CH_DEVICE(0x35, 6),     /* T3C20-derived T3C10 */
  97         CH_DEVICE(0x36, 3),     /* S320E-CR */
  98         CH_DEVICE(0x37, 7),     /* N320E-G2 */
  99         {0,}
 100 };
 101
 102 MODULE_DESCRIPTION(DRV_DESC);
 103 MODULE_AUTHOR("Chelsio Communications");
 104 MODULE_LICENSE("Dual BSD/GPL");
 105 MODULE_VERSION(DRV_VERSION);
 106 MODULE_DEVICE_TABLE(pci, cxgb3_pci_tbl);
 107
 108 static int dflt_msg_enable = DFLT_MSG_ENABLE;
 109
 110 module_param(dflt_msg_enable, int, 0644);
 111 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T3 default message enable bitmap");
 112
 113 /*
 114  * The driver uses the best interrupt scheme available on a platform in the
 115  * order MSI-X, MSI, legacy pin interrupts.  This parameter determines which
 116  * of these schemes the driver may consider as follows:
 117  *
 118  * msi = 2: choose from among all three options
 119  * msi = 1: only consider MSI and pin interrupts
 120  * msi = 0: force pin interrupts
 121  */
 122 static int msi = 2;
 123
 124 module_param(msi, int, 0644);
 125 MODULE_PARM_DESC(msi, "whether to use MSI or MSI-X");
 126
 127 /*
 128  * The driver enables offload as a default.
 129  * To disable it, use ofld_disable = 1.
 130  */
 131
 132 static int ofld_disable = 0;
 133
 134 module_param(ofld_disable, int, 0644);
 135 MODULE_PARM_DESC(ofld_disable, "whether to enable offload at init time or not");
 136
 137 /*
 138  * We have work elements that we need to cancel when an interface is taken
 139  * down.  Normally the work elements would be executed by keventd but that
 140  * can deadlock because of linkwatch.  If our close method takes the rtnl
 141  * lock and linkwatch is ahead of our work elements in keventd, linkwatch
 142  * will block keventd as it needs the rtnl lock, and we'll deadlock waiting
 143  * for our work to complete.  Get our own work queue to solve this.
 144  */
 145 struct workqueue_struct *cxgb3_wq;
 146
 147 /**
 148  *      link_report - show link status and link speed/duplex
 149  *      @p: the port whose settings are to be reported
 150  *
 151  *      Shows the link status, speed, and duplex of a port.
 152  */
 153 static void link_report(struct net_device *dev)
 154 {
 155         if (!netif_carrier_ok(dev))
 156                 printk(KERN_INFO "%s: link down\n", dev->name);
 157         else {
 158                 const char *s = "10Mbps";
 159                 const struct port_info *p = netdev_priv(dev);
 160
 161                 switch (p->link_config.speed) {
 162                 case SPEED_10000:
 163                         s = "10Gbps";
 164                         break;
 165                 case SPEED_1000:
 166                         s = "1000Mbps";
 167                         break;
 168                 case SPEED_100:
 169                         s = "100Mbps";
 170                         break;
 171                 }
 172
 173                 printk(KERN_INFO "%s: link up, %s, %s-duplex\n", dev->name, s,
 174                        p->link_config.duplex == DUPLEX_FULL ? "full" : "half");
 175         }
 176 }
 177
 178 static void enable_tx_fifo_drain(struct adapter *adapter,
 179                                  struct port_info *pi)
 180 {
 181         t3_set_reg_field(adapter, A_XGM_TXFIFO_CFG + pi->mac.offset, 0,
 182                          F_ENDROPPKT);
 183         t3_write_reg(adapter, A_XGM_RX_CTRL + pi->mac.offset, 0);
 184         t3_write_reg(adapter, A_XGM_TX_CTRL + pi->mac.offset, F_TXEN);
 185         t3_write_reg(adapter, A_XGM_RX_CTRL + pi->mac.offset, F_RXEN);
 186 }
 187
 188 static void disable_tx_fifo_drain(struct adapter *adapter,
 189                                   struct port_info *pi)
 190 {
 191         t3_set_reg_field(adapter, A_XGM_TXFIFO_CFG + pi->mac.offset,
 192                          F_ENDROPPKT, 0);
 193 }
 194
 195 void t3_os_link_fault(struct adapter *adap, int port_id, int state)
 196 {
 197         struct net_device *dev = adap->port[port_id];
 198         struct port_info *pi = netdev_priv(dev);
 199
 200         if (state == netif_carrier_ok(dev))
 201                 return;
 202
 203         if (state) {
 204                 struct cmac *mac = &pi->mac;
 205
 206                 netif_carrier_on(dev);
 207
 208                 disable_tx_fifo_drain(adap, pi);
 209
 210                 /* Clear local faults */
 211                 t3_xgm_intr_disable(adap, pi->port_id);
 212                 t3_read_reg(adap, A_XGM_INT_STATUS +
 213                                     pi->mac.offset);
 214                 t3_write_reg(adap,
 215                              A_XGM_INT_CAUSE + pi->mac.offset,
 216                              F_XGM_INT);
 217
 218                 t3_set_reg_field(adap,
 219                                  A_XGM_INT_ENABLE +
 220                                  pi->mac.offset,
 221                                  F_XGM_INT, F_XGM_INT);
 222                 t3_xgm_intr_enable(adap, pi->port_id);
 223
 224                 t3_mac_enable(mac, MAC_DIRECTION_TX);
 225         } else {
 226                 netif_carrier_off(dev);
 227
 228                 /* Flush TX FIFO */
 229                 enable_tx_fifo_drain(adap, pi);
 230         }
 231         link_report(dev);
 232 }
 233
 234 /**
 235  *      t3_os_link_changed - handle link status changes
 236  *      @adapter: the adapter associated with the link change
 237  *      @port_id: the port index whose limk status has changed
 238  *      @link_stat: the new status of the link
 239  *      @speed: the new speed setting
 240  *      @duplex: the new duplex setting
 241  *      @pause: the new flow-control setting
 242  *
 243  *      This is the OS-dependent handler for link status changes.  The OS
 244  *      neutral handler takes care of most of the processing for these events,
 245  *      then calls this handler for any OS-specific processing.
 246  */
 247 void t3_os_link_changed(struct adapter *adapter, int port_id, int link_stat,
 248                         int speed, int duplex, int pause)
 249 {
 250         struct net_device *dev = adapter->port[port_id];
 251         struct port_info *pi = netdev_priv(dev);
 252         struct cmac *mac = &pi->mac;
 253
 254         /* Skip changes from disabled ports. */
 255         if (!netif_running(dev))
 256                 return;
 257
 258         if (link_stat != netif_carrier_ok(dev)) {
 259                 if (link_stat) {
 260                         disable_tx_fifo_drain(adapter, pi);
 261
 262                         t3_mac_enable(mac, MAC_DIRECTION_RX);
 263
 264                         /* Clear local faults */
 265                         t3_xgm_intr_disable(adapter, pi->port_id);
 266                         t3_read_reg(adapter, A_XGM_INT_STATUS +
 267                                     pi->mac.offset);
 268                         t3_write_reg(adapter,
 269                                      A_XGM_INT_CAUSE + pi->mac.offset,
 270                                      F_XGM_INT);
 271
 272                         t3_set_reg_field(adapter,
 273                                          A_XGM_INT_ENABLE + pi->mac.offset,
 274                                          F_XGM_INT, F_XGM_INT);
 275                         t3_xgm_intr_enable(adapter, pi->port_id);
 276
 277                         netif_carrier_on(dev);
 278                 } else {
 279                         netif_carrier_off(dev);
 280
 281                         t3_xgm_intr_disable(adapter, pi->port_id);
 282                         t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
 283                         t3_set_reg_field(adapter,
 284                                          A_XGM_INT_ENABLE + pi->mac.offset,
 285                                          F_XGM_INT, 0);
 286
 287                         if (is_10G(adapter))
 288                                 pi->phy.ops->power_down(&pi->phy, 1);
 289
 290                         t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
 291                         t3_mac_disable(mac, MAC_DIRECTION_RX);
 292                         t3_link_start(&pi->phy, mac, &pi->link_config);
 293
 294                         /* Flush TX FIFO */
 295                         enable_tx_fifo_drain(adapter, pi);
 296                 }
 297
 298                 link_report(dev);
 299         }
 300 }
 301
 302 /**
 303  *      t3_os_phymod_changed - handle PHY module changes
 304  *      @phy: the PHY reporting the module change
 305  *      @mod_type: new module type
 306  *
 307  *      This is the OS-dependent handler for PHY module changes.  It is
 308  *      invoked when a PHY module is removed or inserted for any OS-specific
 309  *      processing.
 310  */
 311 void t3_os_phymod_changed(struct adapter *adap, int port_id)
 312 {
 313         static const char *mod_str[] = {
 314                 NULL, "SR", "LR", "LRM", "TWINAX", "TWINAX", "unknown"
 315         };
 316
 317         const struct net_device *dev = adap->port[port_id];
 318         const struct port_info *pi = netdev_priv(dev);
 319
 320         if (pi->phy.modtype == phy_modtype_none)
 321                 printk(KERN_INFO "%s: PHY module unplugged\n", dev->name);
 322         else
 323                 printk(KERN_INFO "%s: %s PHY module inserted\n", dev->name,
 324                        mod_str[pi->phy.modtype]);
 325 }
 326
 327 static void cxgb_set_rxmode(struct net_device *dev)
 328 {
 329         struct port_info *pi = netdev_priv(dev);
 330
 331         t3_mac_set_rx_mode(&pi->mac, dev);
 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 port_info *pi = netdev_priv(dev);
 343         struct cmac *mac = &pi->mac;
 344
 345         t3_mac_reset(mac);
 346         t3_mac_set_num_ucast(mac, MAX_MAC_IDX);
 347         t3_mac_set_mtu(mac, dev->mtu);
 348         t3_mac_set_address(mac, LAN_MAC_IDX, dev->dev_addr);
 349         t3_mac_set_address(mac, SAN_MAC_IDX, pi->iscsic.mac_addr);
 350         t3_mac_set_rx_mode(mac, dev);
 351         t3_link_start(&pi->phy, mac, &pi->link_config);
 352         t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
 353 }
 354
 355 static inline void cxgb_disable_msi(struct adapter *adapter)
 356 {
 357         if (adapter->flags & USING_MSIX) {
 358                 pci_disable_msix(adapter->pdev);
 359                 adapter->flags &= ~USING_MSIX;
 360         } else if (adapter->flags & USING_MSI) {
 361                 pci_disable_msi(adapter->pdev);
 362                 adapter->flags &= ~USING_MSI;
 363         }
 364 }
 365
 366 /*
 367  * Interrupt handler for asynchronous events used with MSI-X.
 368  */
 369 static irqreturn_t t3_async_intr_handler(int irq, void *cookie)
 370 {
 371         t3_slow_intr_handler(cookie);
 372         return IRQ_HANDLED;
 373 }
 374
 375 /*
 376  * Name the MSI-X interrupts.
 377  */
 378 static void name_msix_vecs(struct adapter *adap)
 379 {
 380         int i, j, msi_idx = 1, n = sizeof(adap->msix_info[0].desc) - 1;
 381
 382         snprintf(adap->msix_info[0].desc, n, "%s", adap->name);
 383         adap->msix_info[0].desc[n] = 0;
 384
 385         for_each_port(adap, j) {
 386                 struct net_device *d = adap->port[j];
 387                 const struct port_info *pi = netdev_priv(d);
 388
 389                 for (i = 0; i < pi->nqsets; i++, msi_idx++) {
 390                         snprintf(adap->msix_info[msi_idx].desc, n,
 391                                  "%s-%d", d->name, pi->first_qset + i);
 392                         adap->msix_info[msi_idx].desc[n] = 0;
 393                 }
 394         }
 395 }
 396
 397 static int request_msix_data_irqs(struct adapter *adap)
 398 {
 399         int i, j, err, qidx = 0;
 400
 401         for_each_port(adap, i) {
 402                 int nqsets = adap2pinfo(adap, i)->nqsets;
 403
 404                 for (j = 0; j < nqsets; ++j) {
 405                         err = request_irq(adap->msix_info[qidx + 1].vec,
 406                                           t3_intr_handler(adap,
 407                                                           adap->sge.qs[qidx].
 408                                                           rspq.polling), 0,
 409                                           adap->msix_info[qidx + 1].desc,
 410                                           &adap->sge.qs[qidx]);
 411                         if (err) {
 412                                 while (--qidx >= 0)
 413                                         free_irq(adap->msix_info[qidx + 1].vec,
 414                                                  &adap->sge.qs[qidx]);
 415                                 return err;
 416                         }
 417                         qidx++;
 418                 }
 419         }
 420         return 0;
 421 }
 422
 423 static void free_irq_resources(struct adapter *adapter)
 424 {
 425         if (adapter->flags & USING_MSIX) {
 426                 int i, n = 0;
 427
 428                 free_irq(adapter->msix_info[0].vec, adapter);
 429                 for_each_port(adapter, i)
 430                         n += adap2pinfo(adapter, i)->nqsets;
 431
 432                 for (i = 0; i < n; ++i)
 433                         free_irq(adapter->msix_info[i + 1].vec,
 434                                  &adapter->sge.qs[i]);
 435         } else
 436                 free_irq(adapter->pdev->irq, adapter);
 437 }
 438
 439 static int await_mgmt_replies(struct adapter *adap, unsigned long init_cnt,
 440                               unsigned long n)
 441 {
 442         int attempts = 10;
 443
 444         while (adap->sge.qs[0].rspq.offload_pkts < init_cnt + n) {
 445                 if (!--attempts)
 446                         return -ETIMEDOUT;
 447                 msleep(10);
 448         }
 449         return 0;
 450 }
 451
 452 static int init_tp_parity(struct adapter *adap)
 453 {
 454         int i;
 455         struct sk_buff *skb;
 456         struct cpl_set_tcb_field *greq;
 457         unsigned long cnt = adap->sge.qs[0].rspq.offload_pkts;
 458
 459         t3_tp_set_offload_mode(adap, 1);
 460
 461         for (i = 0; i < 16; i++) {
 462                 struct cpl_smt_write_req *req;
 463
 464                 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
 465                 if (!skb)
 466                         skb = adap->nofail_skb;
 467                 if (!skb)
 468                         goto alloc_skb_fail;
 469
 470                 req = (struct cpl_smt_write_req *)__skb_put(skb, sizeof(*req));
 471                 memset(req, 0, sizeof(*req));
 472                 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 473                 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, i));
 474                 req->mtu_idx = NMTUS - 1;
 475                 req->iff = i;
 476                 t3_mgmt_tx(adap, skb);
 477                 if (skb == adap->nofail_skb) {
 478                         await_mgmt_replies(adap, cnt, i + 1);
 479                         adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
 480                         if (!adap->nofail_skb)
 481                                 goto alloc_skb_fail;
 482                 }
 483         }
 484
 485         for (i = 0; i < 2048; i++) {
 486                 struct cpl_l2t_write_req *req;
 487
 488                 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
 489                 if (!skb)
 490                         skb = adap->nofail_skb;
 491                 if (!skb)
 492                         goto alloc_skb_fail;
 493
 494                 req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
 495                 memset(req, 0, sizeof(*req));
 496                 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 497                 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, i));
 498                 req->params = htonl(V_L2T_W_IDX(i));
 499                 t3_mgmt_tx(adap, skb);
 500                 if (skb == adap->nofail_skb) {
 501                         await_mgmt_replies(adap, cnt, 16 + i + 1);
 502                         adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
 503                         if (!adap->nofail_skb)
 504                                 goto alloc_skb_fail;
 505                 }
 506         }
 507
 508         for (i = 0; i < 2048; i++) {
 509                 struct cpl_rte_write_req *req;
 510
 511                 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
 512                 if (!skb)
 513                         skb = adap->nofail_skb;
 514                 if (!skb)
 515                         goto alloc_skb_fail;
 516
 517                 req = (struct cpl_rte_write_req *)__skb_put(skb, sizeof(*req));
 518                 memset(req, 0, sizeof(*req));
 519                 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 520                 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RTE_WRITE_REQ, i));
 521                 req->l2t_idx = htonl(V_L2T_W_IDX(i));
 522                 t3_mgmt_tx(adap, skb);
 523                 if (skb == adap->nofail_skb) {
 524                         await_mgmt_replies(adap, cnt, 16 + 2048 + i + 1);
 525                         adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
 526                         if (!adap->nofail_skb)
 527                                 goto alloc_skb_fail;
 528                 }
 529         }
 530
 531         skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
 532         if (!skb)
 533                 skb = adap->nofail_skb;
 534         if (!skb)
 535                 goto alloc_skb_fail;
 536
 537         greq = (struct cpl_set_tcb_field *)__skb_put(skb, sizeof(*greq));
 538         memset(greq, 0, sizeof(*greq));
 539         greq->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 540         OPCODE_TID(greq) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, 0));
 541         greq->mask = cpu_to_be64(1);
 542         t3_mgmt_tx(adap, skb);
 543
 544         i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
 545         if (skb == adap->nofail_skb) {
 546                 i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
 547                 adap->nofail_skb = alloc_skb(sizeof(*greq), GFP_KERNEL);
 548         }
 549
 550         t3_tp_set_offload_mode(adap, 0);
 551         return i;
 552
 553 alloc_skb_fail:
 554         t3_tp_set_offload_mode(adap, 0);
 555         return -ENOMEM;
 556 }
 557
 558 /**
 559  *      setup_rss - configure RSS
 560  *      @adap: the adapter
 561  *
 562  *      Sets up RSS to distribute packets to multiple receive queues.  We
 563  *      configure the RSS CPU lookup table to distribute to the number of HW
 564  *      receive queues, and the response queue lookup table to narrow that
 565  *      down to the response queues actually configured for each port.
 566  *      We always configure the RSS mapping for two ports since the mapping
 567  *      table has plenty of entries.
 568  */
 569 static void setup_rss(struct adapter *adap)
 570 {
 571         int i;
 572         unsigned int nq0 = adap2pinfo(adap, 0)->nqsets;
 573         unsigned int nq1 = adap->port[1] ? adap2pinfo(adap, 1)->nqsets : 1;
 574         u8 cpus[SGE_QSETS + 1];
 575         u16 rspq_map[RSS_TABLE_SIZE];
 576
 577         for (i = 0; i < SGE_QSETS; ++i)
 578                 cpus[i] = i;
 579         cpus[SGE_QSETS] = 0xff; /* terminator */
 580
 581         for (i = 0; i < RSS_TABLE_SIZE / 2; ++i) {
 582                 rspq_map[i] = i % nq0;
 583                 rspq_map[i + RSS_TABLE_SIZE / 2] = (i % nq1) + nq0;
 584         }
 585
 586         t3_config_rss(adap, F_RQFEEDBACKENABLE | F_TNLLKPEN | F_TNLMAPEN |
 587                       F_TNLPRTEN | F_TNL2TUPEN | F_TNL4TUPEN |
 588                       V_RRCPLCPUSIZE(6) | F_HASHTOEPLITZ, cpus, rspq_map);
 589 }
 590
 591 static void ring_dbs(struct adapter *adap)
 592 {
 593         int i, j;
 594
 595         for (i = 0; i < SGE_QSETS; i++) {
 596                 struct sge_qset *qs = &adap->sge.qs[i];
 597
 598                 if (qs->adap)
 599                         for (j = 0; j < SGE_TXQ_PER_SET; j++)
 600                                 t3_write_reg(adap, A_SG_KDOORBELL, F_SELEGRCNTX | V_EGRCNTX(qs->txq[j].cntxt_id));
 601         }
 602 }
 603
 604 static void init_napi(struct adapter *adap)
 605 {
 606         int i;
 607
 608         for (i = 0; i < SGE_QSETS; i++) {
 609                 struct sge_qset *qs = &adap->sge.qs[i];
 610
 611                 if (qs->adap)
 612                         netif_napi_add(qs->netdev, &qs->napi, qs->napi.poll,
 613                                        64);
 614         }
 615
 616         /*
 617          * netif_napi_add() can be called only once per napi_struct because it
 618          * adds each new napi_struct to a list.  Be careful not to call it a
 619          * second time, e.g., during EEH recovery, by making a note of it.
 620          */
 621         adap->flags |= NAPI_INIT;
 622 }
 623
 624 /*
 625  * Wait until all NAPI handlers are descheduled.  This includes the handlers of
 626  * both netdevices representing interfaces and the dummy ones for the extra
 627  * queues.
 628  */
 629 static void quiesce_rx(struct adapter *adap)
 630 {
 631         int i;
 632
 633         for (i = 0; i < SGE_QSETS; i++)
 634                 if (adap->sge.qs[i].adap)
 635                         napi_disable(&adap->sge.qs[i].napi);
 636 }
 637
 638 static void enable_all_napi(struct adapter *adap)
 639 {
 640         int i;
 641         for (i = 0; i < SGE_QSETS; i++)
 642                 if (adap->sge.qs[i].adap)
 643                         napi_enable(&adap->sge.qs[i].napi);
 644 }
 645
 646 /**
 647  *      set_qset_lro - Turn a queue set's LRO capability on and off
 648  *      @dev: the device the qset is attached to
 649  *      @qset_idx: the queue set index
 650  *      @val: the LRO switch
 651  *
 652  *      Sets LRO on or off for a particular queue set.
 653  *      the device's features flag is updated to reflect the LRO
 654  *      capability when all queues belonging to the device are
 655  *      in the same state.
 656  */
 657 static void set_qset_lro(struct net_device *dev, int qset_idx, int val)
 658 {
 659         struct port_info *pi = netdev_priv(dev);
 660         struct adapter *adapter = pi->adapter;
 661
 662         adapter->params.sge.qset[qset_idx].lro = !!val;
 663         adapter->sge.qs[qset_idx].lro_enabled = !!val;
 664 }
 665
 666 /**
 667  *      setup_sge_qsets - configure SGE Tx/Rx/response queues
 668  *      @adap: the adapter
 669  *
 670  *      Determines how many sets of SGE queues to use and initializes them.
 671  *      We support multiple queue sets per port if we have MSI-X, otherwise
 672  *      just one queue set per port.
 673  */
 674 static int setup_sge_qsets(struct adapter *adap)
 675 {
 676         int i, j, err, irq_idx = 0, qset_idx = 0;
 677         unsigned int ntxq = SGE_TXQ_PER_SET;
 678
 679         if (adap->params.rev > 0 && !(adap->flags & USING_MSI))
 680                 irq_idx = -1;
 681
 682         for_each_port(adap, i) {
 683                 struct net_device *dev = adap->port[i];
 684                 struct port_info *pi = netdev_priv(dev);
 685
 686                 pi->qs = &adap->sge.qs[pi->first_qset];
 687                 for (j = 0; j < pi->nqsets; ++j, ++qset_idx) {
 688                         set_qset_lro(dev, qset_idx, pi->rx_offload & T3_LRO);
 689                         err = t3_sge_alloc_qset(adap, qset_idx, 1,
 690                                 (adap->flags & USING_MSIX) ? qset_idx + 1 :
 691                                                              irq_idx,
 692                                 &adap->params.sge.qset[qset_idx], ntxq, dev,
 693                                 netdev_get_tx_queue(dev, j));
 694                         if (err) {
 695                                 t3_free_sge_resources(adap);
 696                                 return err;
 697                         }
 698                 }
 699         }
 700
 701         return 0;
 702 }
 703
 704 static ssize_t attr_show(struct device *d, char *buf,
 705                          ssize_t(*format) (struct net_device *, char *))
 706 {
 707         ssize_t len;
 708
 709         /* Synchronize with ioctls that may shut down the device */
 710         rtnl_lock();
 711         len = (*format) (to_net_dev(d), buf);
 712         rtnl_unlock();
 713         return len;
 714 }
 715
 716 static ssize_t attr_store(struct device *d,
 717                           const char *buf, size_t len,
 718                           ssize_t(*set) (struct net_device *, unsigned int),
 719                           unsigned int min_val, unsigned int max_val)
 720 {
 721         char *endp;
 722         ssize_t ret;
 723         unsigned int val;
 724
 725         if (!capable(CAP_NET_ADMIN))
 726                 return -EPERM;
 727
 728         val = simple_strtoul(buf, &endp, 0);
 729         if (endp == buf || val < min_val || val > max_val)
 730                 return -EINVAL;
 731
 732         rtnl_lock();
 733         ret = (*set) (to_net_dev(d), val);
 734         if (!ret)
 735                 ret = len;
 736         rtnl_unlock();
 737         return ret;
 738 }
 739
 740 #define CXGB3_SHOW(name, val_expr) \
 741 static ssize_t format_##name(struct net_device *dev, char *buf) \
 742 { \
 743         struct port_info *pi = netdev_priv(dev); \
 744         struct adapter *adap = pi->adapter; \
 745         return sprintf(buf, "%u\n", val_expr); \
 746 } \
 747 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
 748                            char *buf) \
 749 { \
 750         return attr_show(d, buf, format_##name); \
 751 }
 752
 753 static ssize_t set_nfilters(struct net_device *dev, unsigned int val)
 754 {
 755         struct port_info *pi = netdev_priv(dev);
 756         struct adapter *adap = pi->adapter;
 757         int min_tids = is_offload(adap) ? MC5_MIN_TIDS : 0;
 758
 759         if (adap->flags & FULL_INIT_DONE)
 760                 return -EBUSY;
 761         if (val && adap->params.rev == 0)
 762                 return -EINVAL;
 763         if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nservers -
 764             min_tids)
 765                 return -EINVAL;
 766         adap->params.mc5.nfilters = val;
 767         return 0;
 768 }
 769
 770 static ssize_t store_nfilters(struct device *d, struct device_attribute *attr,
 771                               const char *buf, size_t len)
 772 {
 773         return attr_store(d, buf, len, set_nfilters, 0, ~0);
 774 }
 775
 776 static ssize_t set_nservers(struct net_device *dev, unsigned int val)
 777 {
 778         struct port_info *pi = netdev_priv(dev);
 779         struct adapter *adap = pi->adapter;
 780
 781         if (adap->flags & FULL_INIT_DONE)
 782                 return -EBUSY;
 783         if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nfilters -
 784             MC5_MIN_TIDS)
 785                 return -EINVAL;
 786         adap->params.mc5.nservers = val;
 787         return 0;
 788 }
 789
 790 static ssize_t store_nservers(struct device *d, struct device_attribute *attr,
 791                               const char *buf, size_t len)
 792 {
 793         return attr_store(d, buf, len, set_nservers, 0, ~0);
 794 }
 795
 796 #define CXGB3_ATTR_R(name, val_expr) \
 797 CXGB3_SHOW(name, val_expr) \
 798 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 799
 800 #define CXGB3_ATTR_RW(name, val_expr, store_method) \
 801 CXGB3_SHOW(name, val_expr) \
 802 static DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_method)
 803
 804 CXGB3_ATTR_R(cam_size, t3_mc5_size(&adap->mc5));
 805 CXGB3_ATTR_RW(nfilters, adap->params.mc5.nfilters, store_nfilters);
 806 CXGB3_ATTR_RW(nservers, adap->params.mc5.nservers, store_nservers);
 807
 808 static struct attribute *cxgb3_attrs[] = {
 809         &dev_attr_cam_size.attr,
 810         &dev_attr_nfilters.attr,
 811         &dev_attr_nservers.attr,
 812         NULL
 813 };
 814
 815 static struct attribute_group cxgb3_attr_group = {.attrs = cxgb3_attrs };
 816
 817 static ssize_t tm_attr_show(struct device *d,
 818                             char *buf, int sched)
 819 {
 820         struct port_info *pi = netdev_priv(to_net_dev(d));
 821         struct adapter *adap = pi->adapter;
 822         unsigned int v, addr, bpt, cpt;
 823         ssize_t len;
 824
 825         addr = A_TP_TX_MOD_Q1_Q0_RATE_LIMIT - sched / 2;
 826         rtnl_lock();
 827         t3_write_reg(adap, A_TP_TM_PIO_ADDR, addr);
 828         v = t3_read_reg(adap, A_TP_TM_PIO_DATA);
 829         if (sched & 1)
 830                 v >>= 16;
 831         bpt = (v >> 8) & 0xff;
 832         cpt = v & 0xff;
 833         if (!cpt)
 834                 len = sprintf(buf, "disabled\n");
 835         else {
 836                 v = (adap->params.vpd.cclk * 1000) / cpt;
 837                 len = sprintf(buf, "%u Kbps\n", (v * bpt) / 125);
 838         }
 839         rtnl_unlock();
 840         return len;
 841 }
 842
 843 static ssize_t tm_attr_store(struct device *d,
 844                              const char *buf, size_t len, int sched)
 845 {
 846         struct port_info *pi = netdev_priv(to_net_dev(d));
 847         struct adapter *adap = pi->adapter;
 848         unsigned int val;
 849         char *endp;
 850         ssize_t ret;
 851
 852         if (!capable(CAP_NET_ADMIN))
 853                 return -EPERM;
 854
 855         val = simple_strtoul(buf, &endp, 0);
 856         if (endp == buf || val > 10000000)
 857                 return -EINVAL;
 858
 859         rtnl_lock();
 860         ret = t3_config_sched(adap, val, sched);
 861         if (!ret)
 862                 ret = len;
 863         rtnl_unlock();
 864         return ret;
 865 }
 866
 867 #define TM_ATTR(name, sched) \
 868 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
 869                            char *buf) \
 870 { \
 871         return tm_attr_show(d, buf, sched); \
 872 } \
 873 static ssize_t store_##name(struct device *d, struct device_attribute *attr, \
 874                             const char *buf, size_t len) \
 875 { \
 876         return tm_attr_store(d, buf, len, sched); \
 877 } \
 878 static DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_##name)
 879
 880 TM_ATTR(sched0, 0);
 881 TM_ATTR(sched1, 1);
 882 TM_ATTR(sched2, 2);
 883 TM_ATTR(sched3, 3);
 884 TM_ATTR(sched4, 4);
 885 TM_ATTR(sched5, 5);
 886 TM_ATTR(sched6, 6);
 887 TM_ATTR(sched7, 7);
 888
 889 static struct attribute *offload_attrs[] = {
 890         &dev_attr_sched0.attr,
 891         &dev_attr_sched1.attr,
 892         &dev_attr_sched2.attr,
 893         &dev_attr_sched3.attr,
 894         &dev_attr_sched4.attr,
 895         &dev_attr_sched5.attr,
 896         &dev_attr_sched6.attr,
 897         &dev_attr_sched7.attr,
 898         NULL
 899 };
 900
 901 static struct attribute_group offload_attr_group = {.attrs = offload_attrs };
 902
 903 /*
 904  * Sends an sk_buff to an offload queue driver
 905  * after dealing with any active network taps.
 906  */
 907 static inline int offload_tx(struct t3cdev *tdev, struct sk_buff *skb)
 908 {
 909         int ret;
 910
 911         local_bh_disable();
 912         ret = t3_offload_tx(tdev, skb);
 913         local_bh_enable();
 914         return ret;
 915 }
 916
 917 static int write_smt_entry(struct adapter *adapter, int idx)
 918 {
 919         struct cpl_smt_write_req *req;
 920         struct port_info *pi = netdev_priv(adapter->port[idx]);
 921         struct sk_buff *skb = alloc_skb(sizeof(*req), GFP_KERNEL);
 922
 923         if (!skb)
 924                 return -ENOMEM;
 925
 926         req = (struct cpl_smt_write_req *)__skb_put(skb, sizeof(*req));
 927         req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 928         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, idx));
 929         req->mtu_idx = NMTUS - 1;       /* should be 0 but there's a T3 bug */
 930         req->iff = idx;
 931         memcpy(req->src_mac0, adapter->port[idx]->dev_addr, ETH_ALEN);
 932         memcpy(req->src_mac1, pi->iscsic.mac_addr, ETH_ALEN);
 933         skb->priority = 1;
 934         offload_tx(&adapter->tdev, skb);
 935         return 0;
 936 }
 937
 938 static int init_smt(struct adapter *adapter)
 939 {
 940         int i;
 941
 942         for_each_port(adapter, i)
 943             write_smt_entry(adapter, i);
 944         return 0;
 945 }
 946
 947 static void init_port_mtus(struct adapter *adapter)
 948 {
 949         unsigned int mtus = adapter->port[0]->mtu;
 950
 951         if (adapter->port[1])
 952                 mtus |= adapter->port[1]->mtu << 16;
 953         t3_write_reg(adapter, A_TP_MTU_PORT_TABLE, mtus);
 954 }
 955
 956 static int send_pktsched_cmd(struct adapter *adap, int sched, int qidx, int lo,
 957                               int hi, int port)
 958 {
 959         struct sk_buff *skb;
 960         struct mngt_pktsched_wr *req;
 961         int ret;
 962
 963         skb = alloc_skb(sizeof(*req), GFP_KERNEL);
 964         if (!skb)
 965                 skb = adap->nofail_skb;
 966         if (!skb)
 967                 return -ENOMEM;
 968
 969         req = (struct mngt_pktsched_wr *)skb_put(skb, sizeof(*req));
 970         req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_MNGT));
 971         req->mngt_opcode = FW_MNGTOPCODE_PKTSCHED_SET;
 972         req->sched = sched;
 973         req->idx = qidx;
 974         req->min = lo;
 975         req->max = hi;
 976         req->binding = port;
 977         ret = t3_mgmt_tx(adap, skb);
 978         if (skb == adap->nofail_skb) {
 979                 adap->nofail_skb = alloc_skb(sizeof(struct cpl_set_tcb_field),
 980                                              GFP_KERNEL);
 981                 if (!adap->nofail_skb)
 982                         ret = -ENOMEM;
 983         }
 984
 985         return ret;
 986 }
 987
 988 static int bind_qsets(struct adapter *adap)
 989 {
 990         int i, j, err = 0;
 991
 992         for_each_port(adap, i) {
 993                 const struct port_info *pi = adap2pinfo(adap, i);
 994
 995                 for (j = 0; j < pi->nqsets; ++j) {
 996                         int ret = send_pktsched_cmd(adap, 1,
 997                                                     pi->first_qset + j, -1,
 998                                                     -1, i);
 999                         if (ret)
1000                                 err = ret;
1001                 }
1002         }
1003
1004         return err;
1005 }
1006
1007 #define FW_VERSION __stringify(FW_VERSION_MAJOR) "."                    \
1008         __stringify(FW_VERSION_MINOR) "." __stringify(FW_VERSION_MICRO)
1009 #define FW_FNAME "cxgb3/t3fw-" FW_VERSION ".bin"
1010 #define TPSRAM_VERSION __stringify(TP_VERSION_MAJOR) "."                \
1011         __stringify(TP_VERSION_MINOR) "." __stringify(TP_VERSION_MICRO)
1012 #define TPSRAM_NAME "cxgb3/t3%c_psram-" TPSRAM_VERSION ".bin"
1013 #define AEL2005_OPT_EDC_NAME "cxgb3/ael2005_opt_edc.bin"
1014 #define AEL2005_TWX_EDC_NAME "cxgb3/ael2005_twx_edc.bin"
1015 #define AEL2020_TWX_EDC_NAME "cxgb3/ael2020_twx_edc.bin"
1016 MODULE_FIRMWARE(FW_FNAME);
1017 MODULE_FIRMWARE("cxgb3/t3b_psram-" TPSRAM_VERSION ".bin");
1018 MODULE_FIRMWARE("cxgb3/t3c_psram-" TPSRAM_VERSION ".bin");
1019 MODULE_FIRMWARE(AEL2005_OPT_EDC_NAME);
1020 MODULE_FIRMWARE(AEL2005_TWX_EDC_NAME);
1021 MODULE_FIRMWARE(AEL2020_TWX_EDC_NAME);
1022
1023 static inline const char *get_edc_fw_name(int edc_idx)
1024 {
1025         const char *fw_name = NULL;
1026
1027         switch (edc_idx) {
1028         case EDC_OPT_AEL2005:
1029                 fw_name = AEL2005_OPT_EDC_NAME;
1030                 break;
1031         case EDC_TWX_AEL2005:
1032                 fw_name = AEL2005_TWX_EDC_NAME;
1033                 break;
1034         case EDC_TWX_AEL2020:
1035                 fw_name = AEL2020_TWX_EDC_NAME;
1036                 break;
1037         }
1038         return fw_name;
1039 }
1040
1041 int t3_get_edc_fw(struct cphy *phy, int edc_idx, int size)
1042 {
1043         struct adapter *adapter = phy->adapter;
1044         const struct firmware *fw;
1045         char buf[64];
1046         u32 csum;
1047         const __be32 *p;
1048         u16 *cache = phy->phy_cache;
1049         int i, ret;
1050
1051         snprintf(buf, sizeof(buf), get_edc_fw_name(edc_idx));
1052
1053         ret = request_firmware(&fw, buf, &adapter->pdev->dev);
1054         if (ret < 0) {
1055                 dev_err(&adapter->pdev->dev,
1056                         "could not upgrade firmware: unable to load %s\n",
1057                         buf);
1058                 return ret;
1059         }
1060
1061         /* check size, take checksum in account */
1062         if (fw->size > size + 4) {
1063                 CH_ERR(adapter, "firmware image too large %u, expected %d\n",
1064                        (unsigned int)fw->size, size + 4);
1065                 ret = -EINVAL;
1066         }
1067
1068         /* compute checksum */
1069         p = (const __be32 *)fw->data;
1070         for (csum = 0, i = 0; i < fw->size / sizeof(csum); i++)
1071                 csum += ntohl(p[i]);
1072
1073         if (csum != 0xffffffff) {
1074                 CH_ERR(adapter, "corrupted firmware image, checksum %u\n",
1075                        csum);
1076                 ret = -EINVAL;
1077         }
1078
1079         for (i = 0; i < size / 4 ; i++) {
1080                 *cache++ = (be32_to_cpu(p[i]) & 0xffff0000) >> 16;
1081                 *cache++ = be32_to_cpu(p[i]) & 0xffff;
1082         }
1083
1084         release_firmware(fw);
1085
1086         return ret;
1087 }
1088
1089 static int upgrade_fw(struct adapter *adap)
1090 {
1091         int ret;
1092         const struct firmware *fw;
1093         struct device *dev = &adap->pdev->dev;
1094
1095         ret = request_firmware(&fw, FW_FNAME, dev);
1096         if (ret < 0) {
1097                 dev_err(dev, "could not upgrade firmware: unable to load %s\n",
1098                         FW_FNAME);
1099                 return ret;
1100         }
1101         ret = t3_load_fw(adap, fw->data, fw->size);
1102         release_firmware(fw);
1103
1104         if (ret == 0)
1105                 dev_info(dev, "successful upgrade to firmware %d.%d.%d\n",
1106                          FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
1107         else
1108                 dev_err(dev, "failed to upgrade to firmware %d.%d.%d\n",
1109                         FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
1110
1111         return ret;
1112 }
1113
1114 static inline char t3rev2char(struct adapter *adapter)
1115 {
1116         char rev = 0;
1117
1118         switch(adapter->params.rev) {
1119         case T3_REV_B:
1120         case T3_REV_B2:
1121                 rev = 'b';
1122                 break;
1123         case T3_REV_C:
1124                 rev = 'c';
1125                 break;
1126         }
1127         return rev;
1128 }
1129
1130 static int update_tpsram(struct adapter *adap)
1131 {
1132         const struct firmware *tpsram;
1133         char buf[64];
1134         struct device *dev = &adap->pdev->dev;
1135         int ret;
1136         char rev;
1137
1138         rev = t3rev2char(adap);
1139         if (!rev)
1140                 return 0;
1141
1142         snprintf(buf, sizeof(buf), TPSRAM_NAME, rev);
1143
1144         ret = request_firmware(&tpsram, buf, dev);
1145         if (ret < 0) {
1146                 dev_err(dev, "could not load TP SRAM: unable to load %s\n",
1147                         buf);
1148                 return ret;
1149         }
1150
1151         ret = t3_check_tpsram(adap, tpsram->data, tpsram->size);
1152         if (ret)
1153                 goto release_tpsram;
1154
1155         ret = t3_set_proto_sram(adap, tpsram->data);
1156         if (ret == 0)
1157                 dev_info(dev,
1158                          "successful update of protocol engine "
1159                          "to %d.%d.%d\n",
1160                          TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
1161         else
1162                 dev_err(dev, "failed to update of protocol engine %d.%d.%d\n",
1163                         TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
1164         if (ret)
1165                 dev_err(dev, "loading protocol SRAM failed\n");
1166
1167 release_tpsram:
1168         release_firmware(tpsram);
1169
1170         return ret;
1171 }
1172
1173 /**
1174  *      cxgb_up - enable the adapter
1175  *      @adapter: adapter being enabled
1176  *
1177  *      Called when the first port is enabled, this function performs the
1178  *      actions necessary to make an adapter operational, such as completing
1179  *      the initialization of HW modules, and enabling interrupts.
1180  *
1181  *      Must be called with the rtnl lock held.
1182  */
1183 static int cxgb_up(struct adapter *adap)
1184 {
1185         int err;
1186
1187         if (!(adap->flags & FULL_INIT_DONE)) {
1188                 err = t3_check_fw_version(adap);
1189                 if (err == -EINVAL) {
1190                         err = upgrade_fw(adap);
1191                         CH_WARN(adap, "FW upgrade to %d.%d.%d %s\n",
1192                                 FW_VERSION_MAJOR, FW_VERSION_MINOR,
1193                                 FW_VERSION_MICRO, err ? "failed" : "succeeded");
1194                 }
1195
1196                 err = t3_check_tpsram_version(adap);
1197                 if (err == -EINVAL) {
1198                         err = update_tpsram(adap);
1199                         CH_WARN(adap, "TP upgrade to %d.%d.%d %s\n",
1200                                 TP_VERSION_MAJOR, TP_VERSION_MINOR,
1201                                 TP_VERSION_MICRO, err ? "failed" : "succeeded");
1202                 }
1203
1204                 /*
1205                  * Clear interrupts now to catch errors if t3_init_hw fails.
1206                  * We clear them again later as initialization may trigger
1207                  * conditions that can interrupt.
1208                  */
1209                 t3_intr_clear(adap);
1210
1211                 err = t3_init_hw(adap, 0);
1212                 if (err)
1213                         goto out;
1214
1215                 t3_set_reg_field(adap, A_TP_PARA_REG5, 0, F_RXDDPOFFINIT);
1216                 t3_write_reg(adap, A_ULPRX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
1217
1218                 err = setup_sge_qsets(adap);
1219                 if (err)
1220                         goto out;
1221
1222                 setup_rss(adap);
1223                 if (!(adap->flags & NAPI_INIT))
1224                         init_napi(adap);
1225
1226                 t3_start_sge_timers(adap);
1227                 adap->flags |= FULL_INIT_DONE;
1228         }
1229
1230         t3_intr_clear(adap);
1231
1232         if (adap->flags & USING_MSIX) {
1233                 name_msix_vecs(adap);
1234                 err = request_irq(adap->msix_info[0].vec,
1235                                   t3_async_intr_handler, 0,
1236                                   adap->msix_info[0].desc, adap);
1237                 if (err)
1238                         goto irq_err;
1239
1240                 err = request_msix_data_irqs(adap);
1241                 if (err) {
1242                         free_irq(adap->msix_info[0].vec, adap);
1243                         goto irq_err;
1244                 }
1245         } else if ((err = request_irq(adap->pdev->irq,
1246                                       t3_intr_handler(adap,
1247                                                       adap->sge.qs[0].rspq.
1248                                                       polling),
1249                                       (adap->flags & USING_MSI) ?
1250                                        0 : IRQF_SHARED,
1251                                       adap->name, adap)))
1252                 goto irq_err;
1253
1254         enable_all_napi(adap);
1255         t3_sge_start(adap);
1256         t3_intr_enable(adap);
1257
1258         if (adap->params.rev >= T3_REV_C && !(adap->flags & TP_PARITY_INIT) &&
1259             is_offload(adap) && init_tp_parity(adap) == 0)
1260                 adap->flags |= TP_PARITY_INIT;
1261
1262         if (adap->flags & TP_PARITY_INIT) {
1263                 t3_write_reg(adap, A_TP_INT_CAUSE,
1264                              F_CMCACHEPERR | F_ARPLUTPERR);
1265                 t3_write_reg(adap, A_TP_INT_ENABLE, 0x7fbfffff);
1266         }
1267
1268         if (!(adap->flags & QUEUES_BOUND)) {
1269                 err = bind_qsets(adap);
1270                 if (err) {
1271                         CH_ERR(adap, "failed to bind qsets, err %d\n", err);
1272                         t3_intr_disable(adap);
1273                         free_irq_resources(adap);
1274                         goto out;
1275                 }
1276                 adap->flags |= QUEUES_BOUND;
1277         }
1278
1279 out:
1280         return err;
1281 irq_err:
1282         CH_ERR(adap, "request_irq failed, err %d\n", err);
1283         goto out;
1284 }
1285
1286 /*
1287  * Release resources when all the ports and offloading have been stopped.
1288  */
1289 static void cxgb_down(struct adapter *adapter, int on_wq)
1290 {
1291         t3_sge_stop(adapter);
1292         spin_lock_irq(&adapter->work_lock);     /* sync with PHY intr task */
1293         t3_intr_disable(adapter);
1294         spin_unlock_irq(&adapter->work_lock);
1295
1296         free_irq_resources(adapter);
1297         quiesce_rx(adapter);
1298         t3_sge_stop(adapter);
1299         if (!on_wq)
1300                 flush_workqueue(cxgb3_wq);/* wait for external IRQ handler */
1301 }
1302
1303 static void schedule_chk_task(struct adapter *adap)
1304 {
1305         unsigned int timeo;
1306
1307         timeo = adap->params.linkpoll_period ?
1308             (HZ * adap->params.linkpoll_period) / 10 :
1309             adap->params.stats_update_period * HZ;
1310         if (timeo)
1311                 queue_delayed_work(cxgb3_wq, &adap->adap_check_task, timeo);
1312 }
1313
1314 static int offload_open(struct net_device *dev)
1315 {
1316         struct port_info *pi = netdev_priv(dev);
1317         struct adapter *adapter = pi->adapter;
1318         struct t3cdev *tdev = dev2t3cdev(dev);
1319         int adap_up = adapter->open_device_map & PORT_MASK;
1320         int err;
1321
1322         if (test_and_set_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
1323                 return 0;
1324
1325         if (!adap_up && (err = cxgb_up(adapter)) < 0)
1326                 goto out;
1327
1328         t3_tp_set_offload_mode(adapter, 1);
1329         tdev->lldev = adapter->port[0];
1330         err = cxgb3_offload_activate(adapter);
1331         if (err)
1332                 goto out;
1333
1334         init_port_mtus(adapter);
1335         t3_load_mtus(adapter, adapter->params.mtus, adapter->params.a_wnd,
1336                      adapter->params.b_wnd,
1337                      adapter->params.rev == 0 ?
1338                      adapter->port[0]->mtu : 0xffff);
1339         init_smt(adapter);
1340
1341         if (sysfs_create_group(&tdev->lldev->dev.kobj, &offload_attr_group))
1342                 dev_dbg(&dev->dev, "cannot create sysfs group\n");
1343
1344         /* Call back all registered clients */
1345         cxgb3_add_clients(tdev);
1346
1347 out:
1348         /* restore them in case the offload module has changed them */
1349         if (err) {
1350                 t3_tp_set_offload_mode(adapter, 0);
1351                 clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
1352                 cxgb3_set_dummy_ops(tdev);
1353         }
1354         return err;
1355 }
1356
1357 static int offload_close(struct t3cdev *tdev)
1358 {
1359         struct adapter *adapter = tdev2adap(tdev);
1360
1361         if (!test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
1362                 return 0;
1363
1364         /* Call back all registered clients */
1365         cxgb3_remove_clients(tdev);
1366
1367         sysfs_remove_group(&tdev->lldev->dev.kobj, &offload_attr_group);
1368
1369         /* Flush work scheduled while releasing TIDs */
1370         flush_scheduled_work();
1371
1372         tdev->lldev = NULL;
1373         cxgb3_set_dummy_ops(tdev);
1374         t3_tp_set_offload_mode(adapter, 0);
1375         clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
1376
1377         if (!adapter->open_device_map)
1378                 cxgb_down(adapter, 0);
1379
1380         cxgb3_offload_deactivate(adapter);
1381         return 0;
1382 }
1383
1384 static int cxgb_open(struct net_device *dev)
1385 {
1386         struct port_info *pi = netdev_priv(dev);
1387         struct adapter *adapter = pi->adapter;
1388         int other_ports = adapter->open_device_map & PORT_MASK;
1389         int err;
1390
1391         if (!adapter->open_device_map && (err = cxgb_up(adapter)) < 0)
1392                 return err;
1393
1394         set_bit(pi->port_id, &adapter->open_device_map);
1395         if (is_offload(adapter) && !ofld_disable) {
1396                 err = offload_open(dev);
1397                 if (err)
1398                         printk(KERN_WARNING
1399                                "Could not initialize offload capabilities\n");
1400         }
1401
1402         dev->real_num_tx_queues = pi->nqsets;
1403         link_start(dev);
1404         t3_port_intr_enable(adapter, pi->port_id);
1405         netif_tx_start_all_queues(dev);
1406         if (!other_ports)
1407                 schedule_chk_task(adapter);
1408
1409         cxgb3_event_notify(&adapter->tdev, OFFLOAD_PORT_UP, pi->port_id);
1410         return 0;
1411 }
1412
1413 static int __cxgb_close(struct net_device *dev, int on_wq)
1414 {
1415         struct port_info *pi = netdev_priv(dev);
1416         struct adapter *adapter = pi->adapter;
1417
1418
1419         if (!adapter->open_device_map)
1420                 return 0;
1421
1422         /* Stop link fault interrupts */
1423         t3_xgm_intr_disable(adapter, pi->port_id);
1424         t3_read_reg(adapter, A_XGM_INT_STATUS + pi->mac.offset);
1425
1426         t3_port_intr_disable(adapter, pi->port_id);
1427         netif_tx_stop_all_queues(dev);
1428         pi->phy.ops->power_down(&pi->phy, 1);
1429         netif_carrier_off(dev);
1430         t3_mac_disable(&pi->mac, MAC_DIRECTION_TX | MAC_DIRECTION_RX);
1431
1432         spin_lock_irq(&adapter->work_lock);     /* sync with update task */
1433         clear_bit(pi->port_id, &adapter->open_device_map);
1434         spin_unlock_irq(&adapter->work_lock);
1435
1436         if (!(adapter->open_device_map & PORT_MASK))
1437                 cancel_delayed_work_sync(&adapter->adap_check_task);
1438
1439         if (!adapter->open_device_map)
1440                 cxgb_down(adapter, on_wq);
1441
1442         cxgb3_event_notify(&adapter->tdev, OFFLOAD_PORT_DOWN, pi->port_id);
1443         return 0;
1444 }
1445
1446 static int cxgb_close(struct net_device *dev)
1447 {
1448         return __cxgb_close(dev, 0);
1449 }
1450
1451 static struct net_device_stats *cxgb_get_stats(struct net_device *dev)
1452 {
1453         struct port_info *pi = netdev_priv(dev);
1454         struct adapter *adapter = pi->adapter;
1455         struct net_device_stats *ns = &pi->netstats;
1456         const struct mac_stats *pstats;
1457
1458         spin_lock(&adapter->stats_lock);
1459         pstats = t3_mac_update_stats(&pi->mac);
1460         spin_unlock(&adapter->stats_lock);
1461
1462         ns->tx_bytes = pstats->tx_octets;
1463         ns->tx_packets = pstats->tx_frames;
1464         ns->rx_bytes = pstats->rx_octets;
1465         ns->rx_packets = pstats->rx_frames;
1466         ns->multicast = pstats->rx_mcast_frames;
1467
1468         ns->tx_errors = pstats->tx_underrun;
1469         ns->rx_errors = pstats->rx_symbol_errs + pstats->rx_fcs_errs +
1470             pstats->rx_too_long + pstats->rx_jabber + pstats->rx_short +
1471             pstats->rx_fifo_ovfl;
1472
1473         /* detailed rx_errors */
1474         ns->rx_length_errors = pstats->rx_jabber + pstats->rx_too_long;
1475         ns->rx_over_errors = 0;
1476         ns->rx_crc_errors = pstats->rx_fcs_errs;
1477         ns->rx_frame_errors = pstats->rx_symbol_errs;
1478         ns->rx_fifo_errors = pstats->rx_fifo_ovfl;
1479         ns->rx_missed_errors = pstats->rx_cong_drops;
1480
1481         /* detailed tx_errors */
1482         ns->tx_aborted_errors = 0;
1483         ns->tx_carrier_errors = 0;
1484         ns->tx_fifo_errors = pstats->tx_underrun;
1485         ns->tx_heartbeat_errors = 0;
1486         ns->tx_window_errors = 0;
1487         return ns;
1488 }
1489
1490 static u32 get_msglevel(struct net_device *dev)
1491 {
1492         struct port_info *pi = netdev_priv(dev);
1493         struct adapter *adapter = pi->adapter;
1494
1495         return adapter->msg_enable;
1496 }
1497
1498 static void set_msglevel(struct net_device *dev, u32 val)
1499 {
1500         struct port_info *pi = netdev_priv(dev);
1501         struct adapter *adapter = pi->adapter;
1502
1503         adapter->msg_enable = val;
1504 }
1505
1506 static char stats_strings[][ETH_GSTRING_LEN] = {
1507         "TxOctetsOK         ",
1508         "TxFramesOK         ",
1509         "TxMulticastFramesOK",
1510         "TxBroadcastFramesOK",
1511         "TxPauseFrames      ",
1512         "TxUnderrun         ",
1513         "TxExtUnderrun      ",
1514
1515         "TxFrames64         ",
1516         "TxFrames65To127    ",
1517         "TxFrames128To255   ",
1518         "TxFrames256To511   ",
1519         "TxFrames512To1023  ",
1520         "TxFrames1024To1518 ",
1521         "TxFrames1519ToMax  ",
1522
1523         "RxOctetsOK         ",
1524         "RxFramesOK         ",
1525         "RxMulticastFramesOK",
1526         "RxBroadcastFramesOK",
1527         "RxPauseFrames      ",
1528         "RxFCSErrors        ",
1529         "RxSymbolErrors     ",
1530         "RxShortErrors      ",
1531         "RxJabberErrors     ",
1532         "RxLengthErrors     ",
1533         "RxFIFOoverflow     ",
1534
1535         "RxFrames64         ",
1536         "RxFrames65To127    ",
1537         "RxFrames128To255   ",
1538         "RxFrames256To511   ",
1539         "RxFrames512To1023  ",
1540         "RxFrames1024To1518 ",
1541         "RxFrames1519ToMax  ",
1542
1543         "PhyFIFOErrors      ",
1544         "TSO                ",
1545         "VLANextractions    ",
1546         "VLANinsertions     ",
1547         "TxCsumOffload      ",
1548         "RxCsumGood         ",
1549         "LroAggregated      ",
1550         "LroFlushed         ",
1551         "LroNoDesc          ",
1552         "RxDrops            ",
1553
1554         "CheckTXEnToggled   ",
1555         "CheckResets        ",
1556
1557         "LinkFaults         ",
1558 };
1559
1560 static int get_sset_count(struct net_device *dev, int sset)
1561 {
1562         switch (sset) {
1563         case ETH_SS_STATS:
1564                 return ARRAY_SIZE(stats_strings);
1565         default:
1566                 return -EOPNOTSUPP;
1567         }
1568 }
1569
1570 #define T3_REGMAP_SIZE (3 * 1024)
1571
1572 static int get_regs_len(struct net_device *dev)
1573 {
1574         return T3_REGMAP_SIZE;
1575 }
1576
1577 static int get_eeprom_len(struct net_device *dev)
1578 {
1579         return EEPROMSIZE;
1580 }
1581
1582 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1583 {
1584         struct port_info *pi = netdev_priv(dev);
1585         struct adapter *adapter = pi->adapter;
1586         u32 fw_vers = 0;
1587         u32 tp_vers = 0;
1588
1589         spin_lock(&adapter->stats_lock);
1590         t3_get_fw_version(adapter, &fw_vers);
1591         t3_get_tp_version(adapter, &tp_vers);
1592         spin_unlock(&adapter->stats_lock);
1593
1594         strcpy(info->driver, DRV_NAME);
1595         strcpy(info->version, DRV_VERSION);
1596         strcpy(info->bus_info, pci_name(adapter->pdev));
1597         if (!fw_vers)
1598                 strcpy(info->fw_version, "N/A");
1599         else {
1600                 snprintf(info->fw_version, sizeof(info->fw_version),
1601                          "%s %u.%u.%u TP %u.%u.%u",
1602                          G_FW_VERSION_TYPE(fw_vers) ? "T" : "N",
1603                          G_FW_VERSION_MAJOR(fw_vers),
1604                          G_FW_VERSION_MINOR(fw_vers),
1605                          G_FW_VERSION_MICRO(fw_vers),
1606                          G_TP_VERSION_MAJOR(tp_vers),
1607                          G_TP_VERSION_MINOR(tp_vers),
1608                          G_TP_VERSION_MICRO(tp_vers));
1609         }
1610 }
1611
1612 static void get_strings(struct net_device *dev, u32 stringset, u8 * data)
1613 {
1614         if (stringset == ETH_SS_STATS)
1615                 memcpy(data, stats_strings, sizeof(stats_strings));
1616 }
1617
1618 static unsigned long collect_sge_port_stats(struct adapter *adapter,
1619                                             struct port_info *p, int idx)
1620 {
1621         int i;
1622         unsigned long tot = 0;
1623
1624         for (i = p->first_qset; i < p->first_qset + p->nqsets; ++i)
1625                 tot += adapter->sge.qs[i].port_stats[idx];
1626         return tot;
1627 }
1628
1629 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1630                       u64 *data)
1631 {
1632         struct port_info *pi = netdev_priv(dev);
1633         struct adapter *adapter = pi->adapter;
1634         const struct mac_stats *s;
1635
1636         spin_lock(&adapter->stats_lock);
1637         s = t3_mac_update_stats(&pi->mac);
1638         spin_unlock(&adapter->stats_lock);
1639
1640         *data++ = s->tx_octets;
1641         *data++ = s->tx_frames;
1642         *data++ = s->tx_mcast_frames;
1643         *data++ = s->tx_bcast_frames;
1644         *data++ = s->tx_pause;
1645         *data++ = s->tx_underrun;
1646         *data++ = s->tx_fifo_urun;
1647
1648         *data++ = s->tx_frames_64;
1649         *data++ = s->tx_frames_65_127;
1650         *data++ = s->tx_frames_128_255;
1651         *data++ = s->tx_frames_256_511;
1652         *data++ = s->tx_frames_512_1023;
1653         *data++ = s->tx_frames_1024_1518;
1654         *data++ = s->tx_frames_1519_max;
1655
1656         *data++ = s->rx_octets;
1657         *data++ = s->rx_frames;
1658         *data++ = s->rx_mcast_frames;
1659         *data++ = s->rx_bcast_frames;
1660         *data++ = s->rx_pause;
1661         *data++ = s->rx_fcs_errs;
1662         *data++ = s->rx_symbol_errs;
1663         *data++ = s->rx_short;
1664         *data++ = s->rx_jabber;
1665         *data++ = s->rx_too_long;
1666         *data++ = s->rx_fifo_ovfl;
1667
1668         *data++ = s->rx_frames_64;
1669         *data++ = s->rx_frames_65_127;
1670         *data++ = s->rx_frames_128_255;
1671         *data++ = s->rx_frames_256_511;
1672         *data++ = s->rx_frames_512_1023;
1673         *data++ = s->rx_frames_1024_1518;
1674         *data++ = s->rx_frames_1519_max;
1675
1676         *data++ = pi->phy.fifo_errors;
1677
1678         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TSO);
1679         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANEX);
1680         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANINS);
1681         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TX_CSUM);
1682         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_RX_CSUM_GOOD);
1683         *data++ = 0;
1684         *data++ = 0;
1685         *data++ = 0;
1686         *data++ = s->rx_cong_drops;
1687
1688         *data++ = s->num_toggled;
1689         *data++ = s->num_resets;
1690
1691         *data++ = s->link_faults;
1692 }
1693
1694 static inline void reg_block_dump(struct adapter *ap, void *buf,
1695                                   unsigned int start, unsigned int end)
1696 {
1697         u32 *p = buf + start;
1698
1699         for (; start <= end; start += sizeof(u32))
1700                 *p++ = t3_read_reg(ap, start);
1701 }
1702
1703 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1704                      void *buf)
1705 {
1706         struct port_info *pi = netdev_priv(dev);
1707         struct adapter *ap = pi->adapter;
1708
1709         /*
1710          * Version scheme:
1711          * bits 0..9: chip version
1712          * bits 10..15: chip revision
1713          * bit 31: set for PCIe cards
1714          */
1715         regs->version = 3 | (ap->params.rev << 10) | (is_pcie(ap) << 31);
1716
1717         /*
1718          * We skip the MAC statistics registers because they are clear-on-read.
1719          * Also reading multi-register stats would need to synchronize with the
1720          * periodic mac stats accumulation.  Hard to justify the complexity.
1721          */
1722         memset(buf, 0, T3_REGMAP_SIZE);
1723         reg_block_dump(ap, buf, 0, A_SG_RSPQ_CREDIT_RETURN);
1724         reg_block_dump(ap, buf, A_SG_HI_DRB_HI_THRSH, A_ULPRX_PBL_ULIMIT);
1725         reg_block_dump(ap, buf, A_ULPTX_CONFIG, A_MPS_INT_CAUSE);
1726         reg_block_dump(ap, buf, A_CPL_SWITCH_CNTRL, A_CPL_MAP_TBL_DATA);
1727         reg_block_dump(ap, buf, A_SMB_GLOBAL_TIME_CFG, A_XGM_SERDES_STAT3);
1728         reg_block_dump(ap, buf, A_XGM_SERDES_STATUS0,
1729                        XGM_REG(A_XGM_SERDES_STAT3, 1));
1730         reg_block_dump(ap, buf, XGM_REG(A_XGM_SERDES_STATUS0, 1),
1731                        XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1));
1732 }
1733
1734 static int restart_autoneg(struct net_device *dev)
1735 {
1736         struct port_info *p = netdev_priv(dev);
1737
1738         if (!netif_running(dev))
1739                 return -EAGAIN;
1740         if (p->link_config.autoneg != AUTONEG_ENABLE)
1741                 return -EINVAL;
1742         p->phy.ops->autoneg_restart(&p->phy);
1743         return 0;
1744 }
1745
1746 static int cxgb3_phys_id(struct net_device *dev, u32 data)
1747 {
1748         struct port_info *pi = netdev_priv(dev);
1749         struct adapter *adapter = pi->adapter;
1750         int i;
1751
1752         if (data == 0)
1753                 data = 2;
1754
1755         for (i = 0; i < data * 2; i++) {
1756                 t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
1757                                  (i & 1) ? F_GPIO0_OUT_VAL : 0);
1758                 if (msleep_interruptible(500))
1759                         break;
1760         }
1761         t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
1762                          F_GPIO0_OUT_VAL);
1763         return 0;
1764 }
1765
1766 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1767 {
1768         struct port_info *p = netdev_priv(dev);
1769
1770         cmd->supported = p->link_config.supported;
1771         cmd->advertising = p->link_config.advertising;
1772
1773         if (netif_carrier_ok(dev)) {
1774                 cmd->speed = p->link_config.speed;
1775                 cmd->duplex = p->link_config.duplex;
1776         } else {
1777                 cmd->speed = -1;
1778                 cmd->duplex = -1;
1779         }
1780
1781         cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1782         cmd->phy_address = p->phy.mdio.prtad;
1783         cmd->transceiver = XCVR_EXTERNAL;
1784         cmd->autoneg = p->link_config.autoneg;
1785         cmd->maxtxpkt = 0;
1786         cmd->maxrxpkt = 0;
1787         return 0;
1788 }
1789
1790 static int speed_duplex_to_caps(int speed, int duplex)
1791 {
1792         int cap = 0;
1793
1794         switch (speed) {
1795         case SPEED_10:
1796                 if (duplex == DUPLEX_FULL)
1797                         cap = SUPPORTED_10baseT_Full;
1798                 else
1799                         cap = SUPPORTED_10baseT_Half;
1800                 break;
1801         case SPEED_100:
1802                 if (duplex == DUPLEX_FULL)
1803                         cap = SUPPORTED_100baseT_Full;
1804                 else
1805                         cap = SUPPORTED_100baseT_Half;
1806                 break;
1807         case SPEED_1000:
1808                 if (duplex == DUPLEX_FULL)
1809                         cap = SUPPORTED_1000baseT_Full;
1810                 else
1811                         cap = SUPPORTED_1000baseT_Half;
1812                 break;
1813         case SPEED_10000:
1814                 if (duplex == DUPLEX_FULL)
1815                         cap = SUPPORTED_10000baseT_Full;
1816         }
1817         return cap;
1818 }
1819
1820 #define ADVERTISED_MASK (ADVERTISED_10baseT_Half | ADVERTISED_10baseT_Full | \
1821                       ADVERTISED_100baseT_Half | ADVERTISED_100baseT_Full | \
1822                       ADVERTISED_1000baseT_Half | ADVERTISED_1000baseT_Full | \
1823                       ADVERTISED_10000baseT_Full)
1824
1825 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1826 {
1827         struct port_info *p = netdev_priv(dev);
1828         struct link_config *lc = &p->link_config;
1829
1830         if (!(lc->supported & SUPPORTED_Autoneg)) {
1831                 /*
1832                  * PHY offers a single speed/duplex.  See if that's what's
1833                  * being requested.
1834                  */
1835                 if (cmd->autoneg == AUTONEG_DISABLE) {
1836                         int cap = speed_duplex_to_caps(cmd->speed, cmd->duplex);
1837                         if (lc->supported & cap)
1838                                 return 0;
1839                 }
1840                 return -EINVAL;
1841         }
1842
1843         if (cmd->autoneg == AUTONEG_DISABLE) {
1844                 int cap = speed_duplex_to_caps(cmd->speed, cmd->duplex);
1845
1846                 if (!(lc->supported & cap) || cmd->speed == SPEED_1000)
1847                         return -EINVAL;
1848                 lc->requested_speed = cmd->speed;
1849                 lc->requested_duplex = cmd->duplex;
1850                 lc->advertising = 0;
1851         } else {
1852                 cmd->advertising &= ADVERTISED_MASK;
1853                 cmd->advertising &= lc->supported;
1854                 if (!cmd->advertising)
1855                         return -EINVAL;
1856                 lc->requested_speed = SPEED_INVALID;
1857                 lc->requested_duplex = DUPLEX_INVALID;
1858                 lc->advertising = cmd->advertising | ADVERTISED_Autoneg;
1859         }
1860         lc->autoneg = cmd->autoneg;
1861         if (netif_running(dev))
1862                 t3_link_start(&p->phy, &p->mac, lc);
1863         return 0;
1864 }
1865
1866 static void get_pauseparam(struct net_device *dev,
1867                            struct ethtool_pauseparam *epause)
1868 {
1869         struct port_info *p = netdev_priv(dev);
1870
1871         epause->autoneg = (p->link_config.requested_fc & PAUSE_AUTONEG) != 0;
1872         epause->rx_pause = (p->link_config.fc & PAUSE_RX) != 0;
1873         epause->tx_pause = (p->link_config.fc & PAUSE_TX) != 0;
1874 }
1875
1876 static int set_pauseparam(struct net_device *dev,
1877                           struct ethtool_pauseparam *epause)
1878 {
1879         struct port_info *p = netdev_priv(dev);
1880         struct link_config *lc = &p->link_config;
1881
1882         if (epause->autoneg == AUTONEG_DISABLE)
1883                 lc->requested_fc = 0;
1884         else if (lc->supported & SUPPORTED_Autoneg)
1885                 lc->requested_fc = PAUSE_AUTONEG;
1886         else
1887                 return -EINVAL;
1888
1889         if (epause->rx_pause)
1890                 lc->requested_fc |= PAUSE_RX;
1891         if (epause->tx_pause)
1892                 lc->requested_fc |= PAUSE_TX;
1893         if (lc->autoneg == AUTONEG_ENABLE) {
1894                 if (netif_running(dev))
1895                         t3_link_start(&p->phy, &p->mac, lc);
1896         } else {
1897                 lc->fc = lc->requested_fc & (PAUSE_RX | PAUSE_TX);
1898                 if (netif_running(dev))
1899                         t3_mac_set_speed_duplex_fc(&p->mac, -1, -1, lc->fc);
1900         }
1901         return 0;
1902 }
1903
1904 static u32 get_rx_csum(struct net_device *dev)
1905 {
1906         struct port_info *p = netdev_priv(dev);
1907
1908         return p->rx_offload & T3_RX_CSUM;
1909 }
1910
1911 static int set_rx_csum(struct net_device *dev, u32 data)
1912 {
1913         struct port_info *p = netdev_priv(dev);
1914
1915         if (data) {
1916                 p->rx_offload |= T3_RX_CSUM;
1917         } else {
1918                 int i;
1919
1920                 p->rx_offload &= ~(T3_RX_CSUM | T3_LRO);
1921                 for (i = p->first_qset; i < p->first_qset + p->nqsets; i++)
1922                         set_qset_lro(dev, i, 0);
1923         }
1924         return 0;
1925 }
1926
1927 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1928 {
1929         struct port_info *pi = netdev_priv(dev);
1930         struct adapter *adapter = pi->adapter;
1931         const struct qset_params *q = &adapter->params.sge.qset[pi->first_qset];
1932
1933         e->rx_max_pending = MAX_RX_BUFFERS;
1934         e->rx_mini_max_pending = 0;
1935         e->rx_jumbo_max_pending = MAX_RX_JUMBO_BUFFERS;
1936         e->tx_max_pending = MAX_TXQ_ENTRIES;
1937
1938         e->rx_pending = q->fl_size;
1939         e->rx_mini_pending = q->rspq_size;
1940         e->rx_jumbo_pending = q->jumbo_size;
1941         e->tx_pending = q->txq_size[0];
1942 }
1943
1944 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1945 {
1946         struct port_info *pi = netdev_priv(dev);
1947         struct adapter *adapter = pi->adapter;
1948         struct qset_params *q;
1949         int i;
1950
1951         if (e->rx_pending > MAX_RX_BUFFERS ||
1952             e->rx_jumbo_pending > MAX_RX_JUMBO_BUFFERS ||
1953             e->tx_pending > MAX_TXQ_ENTRIES ||
1954             e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1955             e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1956             e->rx_pending < MIN_FL_ENTRIES ||
1957             e->rx_jumbo_pending < MIN_FL_ENTRIES ||
1958             e->tx_pending < adapter->params.nports * MIN_TXQ_ENTRIES)
1959                 return -EINVAL;
1960
1961         if (adapter->flags & FULL_INIT_DONE)
1962                 return -EBUSY;
1963
1964         q = &adapter->params.sge.qset[pi->first_qset];
1965         for (i = 0; i < pi->nqsets; ++i, ++q) {
1966                 q->rspq_size = e->rx_mini_pending;
1967                 q->fl_size = e->rx_pending;
1968                 q->jumbo_size = e->rx_jumbo_pending;
1969                 q->txq_size[0] = e->tx_pending;
1970                 q->txq_size[1] = e->tx_pending;
1971                 q->txq_size[2] = e->tx_pending;
1972         }
1973         return 0;
1974 }
1975
1976 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1977 {
1978         struct port_info *pi = netdev_priv(dev);
1979         struct adapter *adapter = pi->adapter;
1980         struct qset_params *qsp = &adapter->params.sge.qset[0];
1981         struct sge_qset *qs = &adapter->sge.qs[0];
1982
1983         if (c->rx_coalesce_usecs * 10 > M_NEWTIMER)
1984                 return -EINVAL;
1985
1986         qsp->coalesce_usecs = c->rx_coalesce_usecs;
1987         t3_update_qset_coalesce(qs, qsp);
1988         return 0;
1989 }
1990
1991 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1992 {
1993         struct port_info *pi = netdev_priv(dev);
1994         struct adapter *adapter = pi->adapter;
1995         struct qset_params *q = adapter->params.sge.qset;
1996
1997         c->rx_coalesce_usecs = q->coalesce_usecs;
1998         return 0;
1999 }
2000
2001 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
2002                       u8 * data)
2003 {
2004         struct port_info *pi = netdev_priv(dev);
2005         struct adapter *adapter = pi->adapter;
2006         int i, err = 0;
2007
2008         u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
2009         if (!buf)
2010                 return -ENOMEM;
2011
2012         e->magic = EEPROM_MAGIC;
2013         for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
2014                 err = t3_seeprom_read(adapter, i, (__le32 *) & buf[i]);
2015
2016         if (!err)
2017                 memcpy(data, buf + e->offset, e->len);
2018         kfree(buf);
2019         return err;
2020 }
2021
2022 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2023                       u8 * data)
2024 {
2025         struct port_info *pi = netdev_priv(dev);
2026         struct adapter *adapter = pi->adapter;
2027         u32 aligned_offset, aligned_len;
2028         __le32 *p;
2029         u8 *buf;
2030         int err;
2031
2032         if (eeprom->magic != EEPROM_MAGIC)
2033                 return -EINVAL;
2034
2035         aligned_offset = eeprom->offset & ~3;
2036         aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2037
2038         if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2039                 buf = kmalloc(aligned_len, GFP_KERNEL);
2040                 if (!buf)
2041                         return -ENOMEM;
2042                 err = t3_seeprom_read(adapter, aligned_offset, (__le32 *) buf);
2043                 if (!err && aligned_len > 4)
2044                         err = t3_seeprom_read(adapter,
2045                                               aligned_offset + aligned_len - 4,
2046                                               (__le32 *) & buf[aligned_len - 4]);
2047                 if (err)
2048                         goto out;
2049                 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
2050         } else
2051                 buf = data;
2052
2053         err = t3_seeprom_wp(adapter, 0);
2054         if (err)
2055                 goto out;
2056
2057         for (p = (__le32 *) buf; !err && aligned_len; aligned_len -= 4, p++) {
2058                 err = t3_seeprom_write(adapter, aligned_offset, *p);
2059                 aligned_offset += 4;
2060         }
2061
2062         if (!err)
2063                 err = t3_seeprom_wp(adapter, 1);
2064 out:
2065         if (buf != data)
2066                 kfree(buf);
2067         return err;
2068 }
2069
2070 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2071 {
2072         wol->supported = 0;
2073         wol->wolopts = 0;
2074         memset(&wol->sopass, 0, sizeof(wol->sopass));
2075 }
2076
2077 static const struct ethtool_ops cxgb_ethtool_ops = {
2078         .get_settings = get_settings,
2079         .set_settings = set_settings,
2080         .get_drvinfo = get_drvinfo,
2081         .get_msglevel = get_msglevel,
2082         .set_msglevel = set_msglevel,
2083         .get_ringparam = get_sge_param,
2084         .set_ringparam = set_sge_param,
2085         .get_coalesce = get_coalesce,
2086         .set_coalesce = set_coalesce,
2087         .get_eeprom_len = get_eeprom_len,
2088         .get_eeprom = get_eeprom,
2089         .set_eeprom = set_eeprom,
2090         .get_pauseparam = get_pauseparam,
2091         .set_pauseparam = set_pauseparam,
2092         .get_rx_csum = get_rx_csum,
2093         .set_rx_csum = set_rx_csum,
2094         .set_tx_csum = ethtool_op_set_tx_csum,
2095         .set_sg = ethtool_op_set_sg,
2096         .get_link = ethtool_op_get_link,
2097         .get_strings = get_strings,
2098         .phys_id = cxgb3_phys_id,
2099         .nway_reset = restart_autoneg,
2100         .get_sset_count = get_sset_count,
2101         .get_ethtool_stats = get_stats,
2102         .get_regs_len = get_regs_len,
2103         .get_regs = get_regs,
2104         .get_wol = get_wol,
2105         .set_tso = ethtool_op_set_tso,
2106 };
2107
2108 static int in_range(int val, int lo, int hi)
2109 {
2110         return val < 0 || (val <= hi && val >= lo);
2111 }
2112
2113 static int cxgb_extension_ioctl(struct net_device *dev, void __user *useraddr)
2114 {
2115         struct port_info *pi = netdev_priv(dev);
2116         struct adapter *adapter = pi->adapter;
2117         u32 cmd;
2118         int ret;
2119
2120         if (copy_from_user(&cmd, useraddr, sizeof(cmd)))
2121                 return -EFAULT;
2122
2123         switch (cmd) {
2124         case CHELSIO_SET_QSET_PARAMS:{
2125                 int i;
2126                 struct qset_params *q;
2127                 struct ch_qset_params t;
2128                 int q1 = pi->first_qset;
2129                 int nqsets = pi->nqsets;
2130
2131                 if (!capable(CAP_NET_ADMIN))
2132                         return -EPERM;
2133                 if (copy_from_user(&t, useraddr, sizeof(t)))
2134                         return -EFAULT;
2135                 if (t.qset_idx >= SGE_QSETS)
2136                         return -EINVAL;
2137                 if (!in_range(t.intr_lat, 0, M_NEWTIMER) ||
2138                     !in_range(t.cong_thres, 0, 255) ||
2139                     !in_range(t.txq_size[0], MIN_TXQ_ENTRIES,
2140                               MAX_TXQ_ENTRIES) ||
2141                     !in_range(t.txq_size[1], MIN_TXQ_ENTRIES,
2142                               MAX_TXQ_ENTRIES) ||
2143                     !in_range(t.txq_size[2], MIN_CTRL_TXQ_ENTRIES,
2144                               MAX_CTRL_TXQ_ENTRIES) ||
2145                     !in_range(t.fl_size[0], MIN_FL_ENTRIES,
2146                               MAX_RX_BUFFERS) ||
2147                     !in_range(t.fl_size[1], MIN_FL_ENTRIES,
2148                               MAX_RX_JUMBO_BUFFERS) ||
2149                     !in_range(t.rspq_size, MIN_RSPQ_ENTRIES,
2150                               MAX_RSPQ_ENTRIES))
2151                         return -EINVAL;
2152
2153                 if ((adapter->flags & FULL_INIT_DONE) && t.lro > 0)
2154                         for_each_port(adapter, i) {
2155                                 pi = adap2pinfo(adapter, i);
2156                                 if (t.qset_idx >= pi->first_qset &&
2157                                     t.qset_idx < pi->first_qset + pi->nqsets &&
2158                                     !(pi->rx_offload & T3_RX_CSUM))
2159                                         return -EINVAL;
2160                         }
2161
2162                 if ((adapter->flags & FULL_INIT_DONE) &&
2163                         (t.rspq_size >= 0 || t.fl_size[0] >= 0 ||
2164                         t.fl_size[1] >= 0 || t.txq_size[0] >= 0 ||
2165                         t.txq_size[1] >= 0 || t.txq_size[2] >= 0 ||
2166                         t.polling >= 0 || t.cong_thres >= 0))
2167                         return -EBUSY;
2168
2169                 /* Allow setting of any available qset when offload enabled */
2170                 if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2171                         q1 = 0;
2172                         for_each_port(adapter, i) {
2173                                 pi = adap2pinfo(adapter, i);
2174                                 nqsets += pi->first_qset + pi->nqsets;
2175                         }
2176                 }
2177
2178                 if (t.qset_idx < q1)
2179                         return -EINVAL;
2180                 if (t.qset_idx > q1 + nqsets - 1)
2181                         return -EINVAL;
2182
2183                 q = &adapter->params.sge.qset[t.qset_idx];
2184
2185                 if (t.rspq_size >= 0)
2186                         q->rspq_size = t.rspq_size;
2187                 if (t.fl_size[0] >= 0)
2188                         q->fl_size = t.fl_size[0];
2189                 if (t.fl_size[1] >= 0)
2190                         q->jumbo_size = t.fl_size[1];
2191                 if (t.txq_size[0] >= 0)
2192                         q->txq_size[0] = t.txq_size[0];
2193                 if (t.txq_size[1] >= 0)
2194                         q->txq_size[1] = t.txq_size[1];
2195                 if (t.txq_size[2] >= 0)
2196                         q->txq_size[2] = t.txq_size[2];
2197                 if (t.cong_thres >= 0)
2198                         q->cong_thres = t.cong_thres;
2199                 if (t.intr_lat >= 0) {
2200                         struct sge_qset *qs =
2201                                 &adapter->sge.qs[t.qset_idx];
2202
2203                         q->coalesce_usecs = t.intr_lat;
2204                         t3_update_qset_coalesce(qs, q);
2205                 }
2206                 if (t.polling >= 0) {
2207                         if (adapter->flags & USING_MSIX)
2208                                 q->polling = t.polling;
2209                         else {
2210                                 /* No polling with INTx for T3A */
2211                                 if (adapter->params.rev == 0 &&
2212                                         !(adapter->flags & USING_MSI))
2213                                         t.polling = 0;
2214
2215                                 for (i = 0; i < SGE_QSETS; i++) {
2216                                         q = &adapter->params.sge.
2217                                                 qset[i];
2218                                         q->polling = t.polling;
2219                                 }
2220                         }
2221                 }
2222                 if (t.lro >= 0)
2223                         set_qset_lro(dev, t.qset_idx, t.lro);
2224
2225                 break;
2226         }
2227         case CHELSIO_GET_QSET_PARAMS:{
2228                 struct qset_params *q;
2229                 struct ch_qset_params t;
2230                 int q1 = pi->first_qset;
2231                 int nqsets = pi->nqsets;
2232                 int i;
2233
2234                 if (copy_from_user(&t, useraddr, sizeof(t)))
2235                         return -EFAULT;
2236
2237                 /* Display qsets for all ports when offload enabled */
2238                 if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2239                         q1 = 0;
2240                         for_each_port(adapter, i) {
2241                                 pi = adap2pinfo(adapter, i);
2242                                 nqsets = pi->first_qset + pi->nqsets;
2243                         }
2244                 }
2245
2246                 if (t.qset_idx >= nqsets)
2247                         return -EINVAL;
2248
2249                 q = &adapter->params.sge.qset[q1 + t.qset_idx];
2250                 t.rspq_size = q->rspq_size;
2251                 t.txq_size[0] = q->txq_size[0];
2252                 t.txq_size[1] = q->txq_size[1];
2253                 t.txq_size[2] = q->txq_size[2];
2254                 t.fl_size[0] = q->fl_size;
2255                 t.fl_size[1] = q->jumbo_size;
2256                 t.polling = q->polling;
2257                 t.lro = q->lro;
2258                 t.intr_lat = q->coalesce_usecs;
2259                 t.cong_thres = q->cong_thres;
2260                 t.qnum = q1;
2261
2262                 if (adapter->flags & USING_MSIX)
2263                         t.vector = adapter->msix_info[q1 + t.qset_idx + 1].vec;
2264                 else
2265                         t.vector = adapter->pdev->irq;
2266
2267                 if (copy_to_user(useraddr, &t, sizeof(t)))
2268                         return -EFAULT;
2269                 break;
2270         }
2271         case CHELSIO_SET_QSET_NUM:{
2272                 struct ch_reg edata;
2273                 unsigned int i, first_qset = 0, other_qsets = 0;
2274
2275                 if (!capable(CAP_NET_ADMIN))
2276                         return -EPERM;
2277                 if (adapter->flags & FULL_INIT_DONE)
2278                         return -EBUSY;
2279                 if (copy_from_user(&edata, useraddr, sizeof(edata)))
2280                         return -EFAULT;
2281                 if (edata.val < 1 ||
2282                         (edata.val > 1 && !(adapter->flags & USING_MSIX)))
2283                         return -EINVAL;
2284
2285                 for_each_port(adapter, i)
2286                         if (adapter->port[i] && adapter->port[i] != dev)
2287                                 other_qsets += adap2pinfo(adapter, i)->nqsets;
2288
2289                 if (edata.val + other_qsets > SGE_QSETS)
2290                         return -EINVAL;
2291
2292                 pi->nqsets = edata.val;
2293
2294                 for_each_port(adapter, i)
2295                         if (adapter->port[i]) {
2296                                 pi = adap2pinfo(adapter, i);
2297                                 pi->first_qset = first_qset;
2298                                 first_qset += pi->nqsets;
2299                         }
2300                 break;
2301         }
2302         case CHELSIO_GET_QSET_NUM:{
2303                 struct ch_reg edata;
2304
2305                 edata.cmd = CHELSIO_GET_QSET_NUM;
2306                 edata.val = pi->nqsets;
2307                 if (copy_to_user(useraddr, &edata, sizeof(edata)))
2308                         return -EFAULT;
2309                 break;
2310         }
2311         case CHELSIO_LOAD_FW:{
2312                 u8 *fw_data;
2313                 struct ch_mem_range t;
2314
2315                 if (!capable(CAP_SYS_RAWIO))
2316                         return -EPERM;
2317                 if (copy_from_user(&t, useraddr, sizeof(t)))
2318                         return -EFAULT;
2319                 /* Check t.len sanity ? */
2320                 fw_data = memdup_user(useraddr + sizeof(t), t.len);
2321                 if (IS_ERR(fw_data))
2322                         return PTR_ERR(fw_data);
2323
2324                 ret = t3_load_fw(adapter, fw_data, t.len);
2325                 kfree(fw_data);
2326                 if (ret)
2327                         return ret;
2328                 break;
2329         }
2330         case CHELSIO_SETMTUTAB:{
2331                 struct ch_mtus m;
2332                 int i;
2333
2334                 if (!is_offload(adapter))
2335                         return -EOPNOTSUPP;
2336                 if (!capable(CAP_NET_ADMIN))
2337                         return -EPERM;
2338                 if (offload_running(adapter))
2339                         return -EBUSY;
2340                 if (copy_from_user(&m, useraddr, sizeof(m)))
2341                         return -EFAULT;
2342                 if (m.nmtus != NMTUS)
2343                         return -EINVAL;
2344                 if (m.mtus[0] < 81)     /* accommodate SACK */
2345                         return -EINVAL;
2346
2347                 /* MTUs must be in ascending order */
2348                 for (i = 1; i < NMTUS; ++i)
2349                         if (m.mtus[i] < m.mtus[i - 1])
2350                                 return -EINVAL;
2351
2352                 memcpy(adapter->params.mtus, m.mtus,
2353                         sizeof(adapter->params.mtus));
2354                 break;
2355         }
2356         case CHELSIO_GET_PM:{
2357                 struct tp_params *p = &adapter->params.tp;
2358                 struct ch_pm m = {.cmd = CHELSIO_GET_PM };
2359
2360                 if (!is_offload(adapter))
2361                         return -EOPNOTSUPP;
2362                 m.tx_pg_sz = p->tx_pg_size;
2363                 m.tx_num_pg = p->tx_num_pgs;
2364                 m.rx_pg_sz = p->rx_pg_size;
2365                 m.rx_num_pg = p->rx_num_pgs;
2366                 m.pm_total = p->pmtx_size + p->chan_rx_size * p->nchan;
2367                 if (copy_to_user(useraddr, &m, sizeof(m)))
2368                         return -EFAULT;
2369                 break;
2370         }
2371         case CHELSIO_SET_PM:{
2372                 struct ch_pm m;
2373                 struct tp_params *p = &adapter->params.tp;
2374
2375                 if (!is_offload(adapter))
2376                         return -EOPNOTSUPP;
2377                 if (!capable(CAP_NET_ADMIN))
2378                         return -EPERM;
2379                 if (adapter->flags & FULL_INIT_DONE)
2380                         return -EBUSY;
2381                 if (copy_from_user(&m, useraddr, sizeof(m)))
2382                         return -EFAULT;
2383                 if (!is_power_of_2(m.rx_pg_sz) ||
2384                         !is_power_of_2(m.tx_pg_sz))
2385                         return -EINVAL; /* not power of 2 */
2386                 if (!(m.rx_pg_sz & 0x14000))
2387                         return -EINVAL; /* not 16KB or 64KB */
2388                 if (!(m.tx_pg_sz & 0x1554000))
2389                         return -EINVAL;
2390                 if (m.tx_num_pg == -1)
2391                         m.tx_num_pg = p->tx_num_pgs;
2392                 if (m.rx_num_pg == -1)
2393                         m.rx_num_pg = p->rx_num_pgs;
2394                 if (m.tx_num_pg % 24 || m.rx_num_pg % 24)
2395                         return -EINVAL;
2396                 if (m.rx_num_pg * m.rx_pg_sz > p->chan_rx_size ||
2397                         m.tx_num_pg * m.tx_pg_sz > p->chan_tx_size)
2398                         return -EINVAL;
2399                 p->rx_pg_size = m.rx_pg_sz;
2400                 p->tx_pg_size = m.tx_pg_sz;
2401                 p->rx_num_pgs = m.rx_num_pg;
2402                 p->tx_num_pgs = m.tx_num_pg;
2403                 break;
2404         }
2405         case CHELSIO_GET_MEM:{
2406                 struct ch_mem_range t;
2407                 struct mc7 *mem;
2408                 u64 buf[32];
2409
2410                 if (!is_offload(adapter))
2411                         return -EOPNOTSUPP;
2412                 if (!(adapter->flags & FULL_INIT_DONE))
2413                         return -EIO;    /* need the memory controllers */
2414                 if (copy_from_user(&t, useraddr, sizeof(t)))
2415                         return -EFAULT;
2416                 if ((t.addr & 7) || (t.len & 7))
2417                         return -EINVAL;
2418                 if (t.mem_id == MEM_CM)
2419                         mem = &adapter->cm;
2420                 else if (t.mem_id == MEM_PMRX)
2421                         mem = &adapter->pmrx;
2422                 else if (t.mem_id == MEM_PMTX)
2423                         mem = &adapter->pmtx;
2424                 else
2425                         return -EINVAL;
2426
2427                 /*
2428                  * Version scheme:
2429                  * bits 0..9: chip version
2430                  * bits 10..15: chip revision
2431                  */
2432                 t.version = 3 | (adapter->params.rev << 10);
2433                 if (copy_to_user(useraddr, &t, sizeof(t)))
2434                         return -EFAULT;
2435
2436                 /*
2437                  * Read 256 bytes at a time as len can be large and we don't
2438                  * want to use huge intermediate buffers.
2439                  */
2440                 useraddr += sizeof(t);  /* advance to start of buffer */
2441                 while (t.len) {
2442                         unsigned int chunk =
2443                                 min_t(unsigned int, t.len, sizeof(buf));
2444
2445                         ret =
2446                                 t3_mc7_bd_read(mem, t.addr / 8, chunk / 8,
2447                                                 buf);
2448                         if (ret)
2449                                 return ret;
2450                         if (copy_to_user(useraddr, buf, chunk))
2451                                 return -EFAULT;
2452                         useraddr += chunk;
2453                         t.addr += chunk;
2454                         t.len -= chunk;
2455                 }
2456                 break;
2457         }
2458         case CHELSIO_SET_TRACE_FILTER:{
2459                 struct ch_trace t;
2460                 const struct trace_params *tp;
2461
2462                 if (!capable(CAP_NET_ADMIN))
2463                         return -EPERM;
2464                 if (!offload_running(adapter))
2465                         return -EAGAIN;
2466                 if (copy_from_user(&t, useraddr, sizeof(t)))
2467                         return -EFAULT;
2468
2469                 tp = (const struct trace_params *)&t.sip;
2470                 if (t.config_tx)
2471                         t3_config_trace_filter(adapter, tp, 0,
2472                                                 t.invert_match,
2473                                                 t.trace_tx);
2474                 if (t.config_rx)
2475                         t3_config_trace_filter(adapter, tp, 1,
2476                                                 t.invert_match,
2477                                                 t.trace_rx);
2478                 break;
2479         }
2480         default:
2481                 return -EOPNOTSUPP;
2482         }
2483         return 0;
2484 }
2485
2486 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2487 {
2488         struct mii_ioctl_data *data = if_mii(req);
2489         struct port_info *pi = netdev_priv(dev);
2490         struct adapter *adapter = pi->adapter;
2491
2492         switch (cmd) {
2493         case SIOCGMIIREG:
2494         case SIOCSMIIREG:
2495                 /* Convert phy_id from older PRTAD/DEVAD format */
2496                 if (is_10G(adapter) &&
2497                     !mdio_phy_id_is_c45(data->phy_id) &&
2498                     (data->phy_id & 0x1f00) &&
2499                     !(data->phy_id & 0xe0e0))
2500                         data->phy_id = mdio_phy_id_c45(data->phy_id >> 8,
2501                                                        data->phy_id & 0x1f);
2502                 /* FALLTHRU */
2503         case SIOCGMIIPHY:
2504                 return mdio_mii_ioctl(&pi->phy.mdio, data, cmd);
2505         case SIOCCHIOCTL:
2506                 return cxgb_extension_ioctl(dev, req->ifr_data);
2507         default:
2508                 return -EOPNOTSUPP;
2509         }
2510 }
2511
2512 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2513 {
2514         struct port_info *pi = netdev_priv(dev);
2515         struct adapter *adapter = pi->adapter;
2516         int ret;
2517
2518         if (new_mtu < 81)       /* accommodate SACK */
2519                 return -EINVAL;
2520         if ((ret = t3_mac_set_mtu(&pi->mac, new_mtu)))
2521                 return ret;
2522         dev->mtu = new_mtu;
2523         init_port_mtus(adapter);
2524         if (adapter->params.rev == 0 && offload_running(adapter))
2525                 t3_load_mtus(adapter, adapter->params.mtus,
2526                              adapter->params.a_wnd, adapter->params.b_wnd,
2527                              adapter->port[0]->mtu);
2528         return 0;
2529 }
2530
2531 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2532 {
2533         struct port_info *pi = netdev_priv(dev);
2534         struct adapter *adapter = pi->adapter;
2535         struct sockaddr *addr = p;
2536
2537         if (!is_valid_ether_addr(addr->sa_data))
2538                 return -EINVAL;
2539
2540         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2541         t3_mac_set_address(&pi->mac, LAN_MAC_IDX, dev->dev_addr);
2542         if (offload_running(adapter))
2543                 write_smt_entry(adapter, pi->port_id);
2544         return 0;
2545 }
2546
2547 /**
2548  * t3_synchronize_rx - wait for current Rx processing on a port to complete
2549  * @adap: the adapter
2550  * @p: the port
2551  *
2552  * Ensures that current Rx processing on any of the queues associated with
2553  * the given port completes before returning.  We do this by acquiring and
2554  * releasing the locks of the response queues associated with the port.
2555  */
2556 static void t3_synchronize_rx(struct adapter *adap, const struct port_info *p)
2557 {
2558         int i;
2559
2560         for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
2561                 struct sge_rspq *q = &adap->sge.qs[i].rspq;
2562
2563                 spin_lock_irq(&q->lock);
2564                 spin_unlock_irq(&q->lock);
2565         }
2566 }
2567
2568 static void vlan_rx_register(struct net_device *dev, struct vlan_group *grp)
2569 {
2570         struct port_info *pi = netdev_priv(dev);
2571         struct adapter *adapter = pi->adapter;
2572
2573         pi->vlan_grp = grp;
2574         if (adapter->params.rev > 0)
2575                 t3_set_vlan_accel(adapter, 1 << pi->port_id, grp != NULL);
2576         else {
2577                 /* single control for all ports */
2578                 unsigned int i, have_vlans = 0;
2579                 for_each_port(adapter, i)
2580                     have_vlans |= adap2pinfo(adapter, i)->vlan_grp != NULL;
2581
2582                 t3_set_vlan_accel(adapter, 1, have_vlans);
2583         }
2584         t3_synchronize_rx(adapter, pi);
2585 }
2586
2587 #ifdef CONFIG_NET_POLL_CONTROLLER
2588 static void cxgb_netpoll(struct net_device *dev)
2589 {
2590         struct port_info *pi = netdev_priv(dev);
2591         struct adapter *adapter = pi->adapter;
2592         int qidx;
2593
2594         for (qidx = pi->first_qset; qidx < pi->first_qset + pi->nqsets; qidx++) {
2595                 struct sge_qset *qs = &adapter->sge.qs[qidx];
2596                 void *source;
2597
2598                 if (adapter->flags & USING_MSIX)
2599                         source = qs;
2600                 else
2601                         source = adapter;
2602
2603                 t3_intr_handler(adapter, qs->rspq.polling) (0, source);
2604         }
2605 }
2606 #endif
2607
2608 /*
2609  * Periodic accumulation of MAC statistics.
2610  */
2611 static void mac_stats_update(struct adapter *adapter)
2612 {
2613         int i;
2614
2615         for_each_port(adapter, i) {
2616                 struct net_device *dev = adapter->port[i];
2617                 struct port_info *p = netdev_priv(dev);
2618
2619                 if (netif_running(dev)) {
2620                         spin_lock(&adapter->stats_lock);
2621                         t3_mac_update_stats(&p->mac);
2622                         spin_unlock(&adapter->stats_lock);
2623                 }
2624         }
2625 }
2626
2627 static void check_link_status(struct adapter *adapter)
2628 {
2629         int i;
2630
2631         for_each_port(adapter, i) {
2632                 struct net_device *dev = adapter->port[i];
2633                 struct port_info *p = netdev_priv(dev);
2634                 int link_fault;
2635
2636                 spin_lock_irq(&adapter->work_lock);
2637                 link_fault = p->link_fault;
2638                 spin_unlock_irq(&adapter->work_lock);
2639
2640                 if (link_fault) {
2641                         t3_link_fault(adapter, i);
2642                         continue;
2643                 }
2644
2645                 if (!(p->phy.caps & SUPPORTED_IRQ) && netif_running(dev)) {
2646                         t3_xgm_intr_disable(adapter, i);
2647                         t3_read_reg(adapter, A_XGM_INT_STATUS + p->mac.offset);
2648
2649                         t3_link_changed(adapter, i);
2650                         t3_xgm_intr_enable(adapter, i);
2651                 }
2652         }
2653 }
2654
2655 static void check_t3b2_mac(struct adapter *adapter)
2656 {
2657         int i;
2658
2659         if (!rtnl_trylock())    /* synchronize with ifdown */
2660                 return;
2661
2662         for_each_port(adapter, i) {
2663                 struct net_device *dev = adapter->port[i];
2664                 struct port_info *p = netdev_priv(dev);
2665                 int status;
2666
2667                 if (!netif_running(dev))
2668                         continue;
2669
2670                 status = 0;
2671                 if (netif_running(dev) && netif_carrier_ok(dev))
2672                         status = t3b2_mac_watchdog_task(&p->mac);
2673                 if (status == 1)
2674                         p->mac.stats.num_toggled++;
2675                 else if (status == 2) {
2676                         struct cmac *mac = &p->mac;
2677
2678                         t3_mac_set_mtu(mac, dev->mtu);
2679                         t3_mac_set_address(mac, LAN_MAC_IDX, dev->dev_addr);
2680                         cxgb_set_rxmode(dev);
2681                         t3_link_start(&p->phy, mac, &p->link_config);
2682                         t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
2683                         t3_port_intr_enable(adapter, p->port_id);
2684                         p->mac.stats.num_resets++;
2685                 }
2686         }
2687         rtnl_unlock();
2688 }
2689
2690
2691 static void t3_adap_check_task(struct work_struct *work)
2692 {
2693         struct adapter *adapter = container_of(work, struct adapter,
2694                                                adap_check_task.work);
2695         const struct adapter_params *p = &adapter->params;
2696         int port;
2697         unsigned int v, status, reset;
2698
2699         adapter->check_task_cnt++;
2700
2701         check_link_status(adapter);
2702
2703         /* Accumulate MAC stats if needed */
2704         if (!p->linkpoll_period ||
2705             (adapter->check_task_cnt * p->linkpoll_period) / 10 >=
2706             p->stats_update_period) {
2707                 mac_stats_update(adapter);
2708                 adapter->check_task_cnt = 0;
2709         }
2710
2711         if (p->rev == T3_REV_B2)
2712                 check_t3b2_mac(adapter);
2713
2714         /*
2715          * Scan the XGMAC's to check for various conditions which we want to
2716          * monitor in a periodic polling manner rather than via an interrupt
2717          * condition.  This is used for conditions which would otherwise flood
2718          * the system with interrupts and we only really need to know that the
2719          * conditions are "happening" ...  For each condition we count the
2720          * detection of the condition and reset it for the next polling loop.
2721          */
2722         for_each_port(adapter, port) {
2723                 struct cmac *mac =  &adap2pinfo(adapter, port)->mac;
2724                 u32 cause;
2725
2726                 cause = t3_read_reg(adapter, A_XGM_INT_CAUSE + mac->offset);
2727                 reset = 0;
2728                 if (cause & F_RXFIFO_OVERFLOW) {
2729                         mac->stats.rx_fifo_ovfl++;
2730                         reset |= F_RXFIFO_OVERFLOW;
2731                 }
2732
2733                 t3_write_reg(adapter, A_XGM_INT_CAUSE + mac->offset, reset);
2734         }
2735
2736         /*
2737          * We do the same as above for FL_EMPTY interrupts.
2738          */
2739         status = t3_read_reg(adapter, A_SG_INT_CAUSE);
2740         reset = 0;
2741
2742         if (status & F_FLEMPTY) {
2743                 struct sge_qset *qs = &adapter->sge.qs[0];
2744                 int i = 0;
2745
2746                 reset |= F_FLEMPTY;
2747
2748                 v = (t3_read_reg(adapter, A_SG_RSPQ_FL_STATUS) >> S_FL0EMPTY) &
2749                     0xffff;
2750
2751                 while (v) {
2752                         qs->fl[i].empty += (v & 1);
2753                         if (i)
2754                                 qs++;
2755                         i ^= 1;
2756                         v >>= 1;
2757                 }
2758         }
2759
2760         t3_write_reg(adapter, A_SG_INT_CAUSE, reset);
2761
2762         /* Schedule the next check update if any port is active. */
2763         spin_lock_irq(&adapter->work_lock);
2764         if (adapter->open_device_map & PORT_MASK)
2765                 schedule_chk_task(adapter);
2766         spin_unlock_irq(&adapter->work_lock);
2767 }
2768
2769 static void db_full_task(struct work_struct *work)
2770 {
2771         struct adapter *adapter = container_of(work, struct adapter,
2772                                                db_full_task);
2773
2774         cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_FULL, 0);
2775 }
2776
2777 static void db_empty_task(struct work_struct *work)
2778 {
2779         struct adapter *adapter = container_of(work, struct adapter,
2780                                                db_empty_task);
2781
2782         cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_EMPTY, 0);
2783 }
2784
2785 static void db_drop_task(struct work_struct *work)
2786 {
2787         struct adapter *adapter = container_of(work, struct adapter,
2788                                                db_drop_task);
2789         unsigned long delay = 1000;
2790         unsigned short r;
2791
2792         cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_DROP, 0);
2793
2794         /*
2795          * Sleep a while before ringing the driver qset dbs.
2796          * The delay is between 1000-2023 usecs.
2797          */
2798         get_random_bytes(&r, 2);
2799         delay += r & 1023;
2800         set_current_state(TASK_UNINTERRUPTIBLE);
2801         schedule_timeout(usecs_to_jiffies(delay));
2802         ring_dbs(adapter);
2803 }
2804
2805 /*
2806  * Processes external (PHY) interrupts in process context.
2807  */
2808 static void ext_intr_task(struct work_struct *work)
2809 {
2810         struct adapter *adapter = container_of(work, struct adapter,
2811                                                ext_intr_handler_task);
2812         int i;
2813
2814         /* Disable link fault interrupts */
2815         for_each_port(adapter, i) {
2816                 struct net_device *dev = adapter->port[i];
2817                 struct port_info *p = netdev_priv(dev);
2818
2819                 t3_xgm_intr_disable(adapter, i);
2820                 t3_read_reg(adapter, A_XGM_INT_STATUS + p->mac.offset);
2821         }
2822
2823         /* Re-enable link fault interrupts */
2824         t3_phy_intr_handler(adapter);
2825
2826         for_each_port(adapter, i)
2827                 t3_xgm_intr_enable(adapter, i);
2828
2829         /* Now reenable external interrupts */
2830         spin_lock_irq(&adapter->work_lock);
2831         if (adapter->slow_intr_mask) {
2832                 adapter->slow_intr_mask |= F_T3DBG;
2833                 t3_write_reg(adapter, A_PL_INT_CAUSE0, F_T3DBG);
2834                 t3_write_reg(adapter, A_PL_INT_ENABLE0,
2835                              adapter->slow_intr_mask);
2836         }
2837         spin_unlock_irq(&adapter->work_lock);
2838 }
2839
2840 /*
2841  * Interrupt-context handler for external (PHY) interrupts.
2842  */
2843 void t3_os_ext_intr_handler(struct adapter *adapter)
2844 {
2845         /*
2846          * Schedule a task to handle external interrupts as they may be slow
2847          * and we use a mutex to protect MDIO registers.  We disable PHY
2848          * interrupts in the meantime and let the task reenable them when
2849          * it's done.
2850          */
2851         spin_lock(&adapter->work_lock);
2852         if (adapter->slow_intr_mask) {
2853                 adapter->slow_intr_mask &= ~F_T3DBG;
2854                 t3_write_reg(adapter, A_PL_INT_ENABLE0,
2855                              adapter->slow_intr_mask);
2856                 queue_work(cxgb3_wq, &adapter->ext_intr_handler_task);
2857         }
2858         spin_unlock(&adapter->work_lock);
2859 }
2860
2861 void t3_os_link_fault_handler(struct adapter *adapter, int port_id)
2862 {
2863         struct net_device *netdev = adapter->port[port_id];
2864         struct port_info *pi = netdev_priv(netdev);
2865
2866         spin_lock(&adapter->work_lock);
2867         pi->link_fault = 1;
2868         spin_unlock(&adapter->work_lock);
2869 }
2870
2871 static int t3_adapter_error(struct adapter *adapter, int reset, int on_wq)
2872 {
2873         int i, ret = 0;
2874
2875         if (is_offload(adapter) &&
2876             test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2877                 cxgb3_event_notify(&adapter->tdev, OFFLOAD_STATUS_DOWN, 0);
2878                 offload_close(&adapter->tdev);
2879         }
2880
2881         /* Stop all ports */
2882         for_each_port(adapter, i) {
2883                 struct net_device *netdev = adapter->port[i];
2884
2885                 if (netif_running(netdev))
2886                         __cxgb_close(netdev, on_wq);
2887         }
2888
2889         /* Stop SGE timers */
2890         t3_stop_sge_timers(adapter);
2891
2892         adapter->flags &= ~FULL_INIT_DONE;
2893
2894         if (reset)
2895                 ret = t3_reset_adapter(adapter);
2896
2897         pci_disable_device(adapter->pdev);
2898
2899         return ret;
2900 }
2901
2902 static int t3_reenable_adapter(struct adapter *adapter)
2903 {
2904         if (pci_enable_device(adapter->pdev)) {
2905                 dev_err(&adapter->pdev->dev,
2906                         "Cannot re-enable PCI device after reset.\n");
2907                 goto err;
2908         }
2909         pci_set_master(adapter->pdev);
2910         pci_restore_state(adapter->pdev);
2911         pci_save_state(adapter->pdev);
2912
2913         /* Free sge resources */
2914         t3_free_sge_resources(adapter);
2915
2916         if (t3_replay_prep_adapter(adapter))
2917                 goto err;
2918
2919         return 0;
2920 err:
2921         return -1;
2922 }
2923
2924 static void t3_resume_ports(struct adapter *adapter)
2925 {
2926         int i;
2927
2928         /* Restart the ports */
2929         for_each_port(adapter, i) {
2930                 struct net_device *netdev = adapter->port[i];
2931
2932                 if (netif_running(netdev)) {
2933                         if (cxgb_open(netdev)) {
2934                                 dev_err(&adapter->pdev->dev,
2935                                         "can't bring device back up"
2936                                         " after reset\n");
2937                                 continue;
2938                         }
2939                 }
2940         }
2941
2942         if (is_offload(adapter) && !ofld_disable)
2943                 cxgb3_event_notify(&adapter->tdev, OFFLOAD_STATUS_UP, 0);
2944 }
2945
2946 /*
2947  * processes a fatal error.
2948  * Bring the ports down, reset the chip, bring the ports back up.
2949  */
2950 static void fatal_error_task(struct work_struct *work)
2951 {
2952         struct adapter *adapter = container_of(work, struct adapter,
2953                                                fatal_error_handler_task);
2954         int err = 0;
2955
2956         rtnl_lock();
2957         err = t3_adapter_error(adapter, 1, 1);
2958         if (!err)
2959                 err = t3_reenable_adapter(adapter);
2960         if (!err)
2961                 t3_resume_ports(adapter);
2962
2963         CH_ALERT(adapter, "adapter reset %s\n", err ? "failed" : "succeeded");
2964         rtnl_unlock();
2965 }
2966
2967 void t3_fatal_err(struct adapter *adapter)
2968 {
2969         unsigned int fw_status[4];
2970
2971         if (adapter->flags & FULL_INIT_DONE) {
2972                 t3_sge_stop(adapter);
2973                 t3_write_reg(adapter, A_XGM_TX_CTRL, 0);
2974                 t3_write_reg(adapter, A_XGM_RX_CTRL, 0);
2975                 t3_write_reg(adapter, XGM_REG(A_XGM_TX_CTRL, 1), 0);
2976                 t3_write_reg(adapter, XGM_REG(A_XGM_RX_CTRL, 1), 0);
2977
2978                 spin_lock(&adapter->work_lock);
2979                 t3_intr_disable(adapter);
2980                 queue_work(cxgb3_wq, &adapter->fatal_error_handler_task);
2981                 spin_unlock(&adapter->work_lock);
2982         }
2983         CH_ALERT(adapter, "encountered fatal error, operation suspended\n");
2984         if (!t3_cim_ctl_blk_read(adapter, 0xa0, 4, fw_status))
2985                 CH_ALERT(adapter, "FW status: 0x%x, 0x%x, 0x%x, 0x%x\n",
2986                          fw_status[0], fw_status[1],
2987                          fw_status[2], fw_status[3]);
2988 }
2989
2990 /**
2991  * t3_io_error_detected - called when PCI error is detected
2992  * @pdev: Pointer to PCI device
2993  * @state: The current pci connection state
2994  *
2995  * This function is called after a PCI bus error affecting
2996  * this device has been detected.
2997  */
2998 static pci_ers_result_t t3_io_error_detected(struct pci_dev *pdev,
2999                                              pci_channel_state_t state)
3000 {
3001         struct adapter *adapter = pci_get_drvdata(pdev);
3002         int ret;
3003
3004         if (state == pci_channel_io_perm_failure)
3005                 return PCI_ERS_RESULT_DISCONNECT;
3006
3007         ret = t3_adapter_error(adapter, 0, 0);
3008
3009         /* Request a slot reset. */
3010         return PCI_ERS_RESULT_NEED_RESET;
3011 }
3012
3013 /**
3014  * t3_io_slot_reset - called after the pci bus has been reset.
3015  * @pdev: Pointer to PCI device
3016  *
3017  * Restart the card from scratch, as if from a cold-boot.
3018  */
3019 static pci_ers_result_t t3_io_slot_reset(struct pci_dev *pdev)
3020 {
3021         struct adapter *adapter = pci_get_drvdata(pdev);
3022
3023         if (!t3_reenable_adapter(adapter))
3024                 return PCI_ERS_RESULT_RECOVERED;
3025
3026         return PCI_ERS_RESULT_DISCONNECT;
3027 }
3028
3029 /**
3030  * t3_io_resume - called when traffic can start flowing again.
3031  * @pdev: Pointer to PCI device
3032  *
3033  * This callback is called when the error recovery driver tells us that
3034  * its OK to resume normal operation.
3035  */
3036 static void t3_io_resume(struct pci_dev *pdev)
3037 {
3038         struct adapter *adapter = pci_get_drvdata(pdev);
3039
3040         CH_ALERT(adapter, "adapter recovering, PEX ERR 0x%x\n",
3041                  t3_read_reg(adapter, A_PCIE_PEX_ERR));
3042
3043         t3_resume_ports(adapter);
3044 }
3045
3046 static struct pci_error_handlers t3_err_handler = {
3047         .error_detected = t3_io_error_detected,
3048         .slot_reset = t3_io_slot_reset,
3049         .resume = t3_io_resume,
3050 };
3051
3052 /*
3053  * Set the number of qsets based on the number of CPUs and the number of ports,
3054  * not to exceed the number of available qsets, assuming there are enough qsets
3055  * per port in HW.
3056  */
3057 static void set_nqsets(struct adapter *adap)
3058 {
3059         int i, j = 0;
3060         int num_cpus = num_online_cpus();
3061         int hwports = adap->params.nports;
3062         int nqsets = adap->msix_nvectors - 1;
3063
3064         if (adap->params.rev > 0 && adap->flags & USING_MSIX) {
3065                 if (hwports == 2 &&
3066                     (hwports * nqsets > SGE_QSETS ||
3067                      num_cpus >= nqsets / hwports))
3068                         nqsets /= hwports;
3069                 if (nqsets > num_cpus)
3070                         nqsets = num_cpus;
3071                 if (nqsets < 1 || hwports == 4)
3072                         nqsets = 1;
3073         } else
3074                 nqsets = 1;
3075
3076         for_each_port(adap, i) {
3077                 struct port_info *pi = adap2pinfo(adap, i);
3078
3079                 pi->first_qset = j;
3080                 pi->nqsets = nqsets;
3081                 j = pi->first_qset + nqsets;
3082
3083                 dev_info(&adap->pdev->dev,
3084                          "Port %d using %d queue sets.\n", i, nqsets);
3085         }
3086 }
3087
3088 static int __devinit cxgb_enable_msix(struct adapter *adap)
3089 {
3090         struct msix_entry entries[SGE_QSETS + 1];
3091         int vectors;
3092         int i, err;
3093
3094         vectors = ARRAY_SIZE(entries);
3095         for (i = 0; i < vectors; ++i)
3096                 entries[i].entry = i;
3097
3098         while ((err = pci_enable_msix(adap->pdev, entries, vectors)) > 0)
3099                 vectors = err;
3100
3101         if (err < 0)
3102                 pci_disable_msix(adap->pdev);
3103
3104         if (!err && vectors < (adap->params.nports + 1)) {
3105                 pci_disable_msix(adap->pdev);
3106                 err = -1;
3107         }
3108
3109         if (!err) {
3110                 for (i = 0; i < vectors; ++i)
3111                         adap->msix_info[i].vec = entries[i].vector;
3112                 adap->msix_nvectors = vectors;
3113         }
3114
3115         return err;
3116 }
3117
3118 static void __devinit print_port_info(struct adapter *adap,
3119                                       const struct adapter_info *ai)
3120 {
3121         static const char *pci_variant[] = {
3122                 "PCI", "PCI-X", "PCI-X ECC", "PCI-X 266", "PCI Express"
3123         };
3124
3125         int i;
3126         char buf[80];
3127
3128         if (is_pcie(adap))
3129                 snprintf(buf, sizeof(buf), "%s x%d",
3130                          pci_variant[adap->params.pci.variant],
3131                          adap->params.pci.width);
3132         else
3133                 snprintf(buf, sizeof(buf), "%s %dMHz/%d-bit",
3134                          pci_variant[adap->params.pci.variant],
3135                          adap->params.pci.speed, adap->params.pci.width);
3136
3137         for_each_port(adap, i) {
3138                 struct net_device *dev = adap->port[i];
3139                 const struct port_info *pi = netdev_priv(dev);
3140
3141                 if (!test_bit(i, &adap->registered_device_map))
3142                         continue;
3143                 printk(KERN_INFO "%s: %s %s %sNIC (rev %d) %s%s\n",
3144                        dev->name, ai->desc, pi->phy.desc,
3145                        is_offload(adap) ? "R" : "", adap->params.rev, buf,
3146                        (adap->flags & USING_MSIX) ? " MSI-X" :
3147                        (adap->flags & USING_MSI) ? " MSI" : "");
3148                 if (adap->name == dev->name && adap->params.vpd.mclk)
3149                         printk(KERN_INFO
3150                                "%s: %uMB CM, %uMB PMTX, %uMB PMRX, S/N: %s\n",
3151                                adap->name, t3_mc7_size(&adap->cm) >> 20,
3152                                t3_mc7_size(&adap->pmtx) >> 20,
3153                                t3_mc7_size(&adap->pmrx) >> 20,
3154                                adap->params.vpd.sn);
3155         }
3156 }
3157
3158 static const struct net_device_ops cxgb_netdev_ops = {
3159         .ndo_open               = cxgb_open,
3160         .ndo_stop               = cxgb_close,
3161         .ndo_start_xmit         = t3_eth_xmit,
3162         .ndo_get_stats          = cxgb_get_stats,
3163         .ndo_validate_addr      = eth_validate_addr,
3164         .ndo_set_multicast_list = cxgb_set_rxmode,
3165         .ndo_do_ioctl           = cxgb_ioctl,
3166         .ndo_change_mtu         = cxgb_change_mtu,
3167         .ndo_set_mac_address    = cxgb_set_mac_addr,
3168         .ndo_vlan_rx_register   = vlan_rx_register,
3169 #ifdef CONFIG_NET_POLL_CONTROLLER
3170         .ndo_poll_controller    = cxgb_netpoll,
3171 #endif
3172 };
3173
3174 static void __devinit cxgb3_init_iscsi_mac(struct net_device *dev)
3175 {
3176         struct port_info *pi = netdev_priv(dev);
3177
3178         memcpy(pi->iscsic.mac_addr, dev->dev_addr, ETH_ALEN);
3179         pi->iscsic.mac_addr[3] |= 0x80;
3180 }
3181
3182 static int __devinit init_one(struct pci_dev *pdev,
3183                               const struct pci_device_id *ent)
3184 {
3185         static int version_printed;
3186
3187         int i, err, pci_using_dac = 0;
3188         resource_size_t mmio_start, mmio_len;
3189         const struct adapter_info *ai;
3190         struct adapter *adapter = NULL;
3191         struct port_info *pi;
3192
3193         if (!version_printed) {
3194                 printk(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
3195                 ++version_printed;
3196         }
3197
3198         if (!cxgb3_wq) {
3199                 cxgb3_wq = create_singlethread_workqueue(DRV_NAME);
3200                 if (!cxgb3_wq) {
3201                         printk(KERN_ERR DRV_NAME
3202                                ": cannot initialize work queue\n");
3203                         return -ENOMEM;
3204                 }
3205         }
3206
3207         err = pci_enable_device(pdev);
3208         if (err) {
3209                 dev_err(&pdev->dev, "cannot enable PCI device\n");
3210                 goto out;
3211         }
3212
3213         err = pci_request_regions(pdev, DRV_NAME);
3214         if (err) {
3215                 /* Just info, some other driver may have claimed the device. */
3216                 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
3217                 goto out_disable_device;
3218         }
3219
3220         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
3221                 pci_using_dac = 1;
3222                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3223                 if (err) {
3224                         dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
3225                                "coherent allocations\n");
3226                         goto out_release_regions;
3227                 }
3228         } else if ((err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32))) != 0) {
3229                 dev_err(&pdev->dev, "no usable DMA configuration\n");
3230                 goto out_release_regions;
3231         }
3232
3233         pci_set_master(pdev);
3234         pci_save_state(pdev);
3235
3236         mmio_start = pci_resource_start(pdev, 0);
3237         mmio_len = pci_resource_len(pdev, 0);
3238         ai = t3_get_adapter_info(ent->driver_data);
3239
3240         adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
3241         if (!adapter) {
3242                 err = -ENOMEM;
3243                 goto out_release_regions;
3244         }
3245
3246         adapter->nofail_skb =
3247                 alloc_skb(sizeof(struct cpl_set_tcb_field), GFP_KERNEL);
3248         if (!adapter->nofail_skb) {
3249                 dev_err(&pdev->dev, "cannot allocate nofail buffer\n");
3250                 err = -ENOMEM;
3251                 goto out_free_adapter;
3252         }
3253
3254         adapter->regs = ioremap_nocache(mmio_start, mmio_len);
3255         if (!adapter->regs) {
3256                 dev_err(&pdev->dev, "cannot map device registers\n");
3257                 err = -ENOMEM;
3258                 goto out_free_adapter;
3259         }
3260
3261         adapter->pdev = pdev;
3262         adapter->name = pci_name(pdev);
3263         adapter->msg_enable = dflt_msg_enable;
3264         adapter->mmio_len = mmio_len;
3265
3266         mutex_init(&adapter->mdio_lock);
3267         spin_lock_init(&adapter->work_lock);
3268         spin_lock_init(&adapter->stats_lock);
3269
3270         INIT_LIST_HEAD(&adapter->adapter_list);
3271         INIT_WORK(&adapter->ext_intr_handler_task, ext_intr_task);
3272         INIT_WORK(&adapter->fatal_error_handler_task, fatal_error_task);
3273
3274         INIT_WORK(&adapter->db_full_task, db_full_task);
3275         INIT_WORK(&adapter->db_empty_task, db_empty_task);
3276         INIT_WORK(&adapter->db_drop_task, db_drop_task);
3277
3278         INIT_DELAYED_WORK(&adapter->adap_check_task, t3_adap_check_task);
3279
3280         for (i = 0; i < ai->nports0 + ai->nports1; ++i) {
3281                 struct net_device *netdev;
3282
3283                 netdev = alloc_etherdev_mq(sizeof(struct port_info), SGE_QSETS);
3284                 if (!netdev) {
3285                         err = -ENOMEM;
3286                         goto out_free_dev;
3287                 }
3288
3289                 SET_NETDEV_DEV(netdev, &pdev->dev);
3290
3291                 adapter->port[i] = netdev;
3292                 pi = netdev_priv(netdev);
3293                 pi->adapter = adapter;
3294                 pi->rx_offload = T3_RX_CSUM | T3_LRO;
3295                 pi->port_id = i;
3296                 netif_carrier_off(netdev);
3297                 netif_tx_stop_all_queues(netdev);
3298                 netdev->irq = pdev->irq;
3299                 netdev->mem_start = mmio_start;
3300                 netdev->mem_end = mmio_start + mmio_len - 1;
3301                 netdev->features |= NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
3302                 netdev->features |= NETIF_F_GRO;
3303                 if (pci_using_dac)
3304                         netdev->features |= NETIF_F_HIGHDMA;
3305
3306                 netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
3307                 netdev->netdev_ops = &cxgb_netdev_ops;
3308                 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
3309         }
3310
3311         pci_set_drvdata(pdev, adapter);
3312         if (t3_prep_adapter(adapter, ai, 1) < 0) {
3313                 err = -ENODEV;
3314                 goto out_free_dev;
3315         }
3316
3317         /*
3318          * The card is now ready to go.  If any errors occur during device
3319          * registration we do not fail the whole card but rather proceed only
3320          * with the ports we manage to register successfully.  However we must
3321          * register at least one net device.
3322          */
3323         for_each_port(adapter, i) {
3324                 err = register_netdev(adapter->port[i]);
3325                 if (err)
3326                         dev_warn(&pdev->dev,
3327                                  "cannot register net device %s, skipping\n",
3328                                  adapter->port[i]->name);
3329                 else {
3330                         /*
3331                          * Change the name we use for messages to the name of
3332                          * the first successfully registered interface.
3333                          */
3334                         if (!adapter->registered_device_map)
3335                                 adapter->name = adapter->port[i]->name;
3336
3337                         __set_bit(i, &adapter->registered_device_map);
3338                 }
3339         }
3340         if (!adapter->registered_device_map) {
3341                 dev_err(&pdev->dev, "could not register any net devices\n");
3342                 goto out_free_dev;
3343         }
3344
3345         for_each_port(adapter, i)
3346                 cxgb3_init_iscsi_mac(adapter->port[i]);
3347
3348         /* Driver's ready. Reflect it on LEDs */
3349         t3_led_ready(adapter);
3350
3351         if (is_offload(adapter)) {
3352                 __set_bit(OFFLOAD_DEVMAP_BIT, &adapter->registered_device_map);
3353                 cxgb3_adapter_ofld(adapter);
3354         }
3355
3356         /* See what interrupts we'll be using */
3357         if (msi > 1 && cxgb_enable_msix(adapter) == 0)
3358                 adapter->flags |= USING_MSIX;
3359         else if (msi > 0 && pci_enable_msi(pdev) == 0)
3360                 adapter->flags |= USING_MSI;
3361
3362         set_nqsets(adapter);
3363
3364         err = sysfs_create_group(&adapter->port[0]->dev.kobj,
3365                                  &cxgb3_attr_group);
3366
3367         print_port_info(adapter, ai);
3368         return 0;
3369
3370 out_free_dev:
3371         iounmap(adapter->regs);
3372         for (i = ai->nports0 + ai->nports1 - 1; i >= 0; --i)
3373                 if (adapter->port[i])
3374                         free_netdev(adapter->port[i]);
3375
3376 out_free_adapter:
3377         kfree(adapter);
3378
3379 out_release_regions:
3380         pci_release_regions(pdev);
3381 out_disable_device:
3382         pci_disable_device(pdev);
3383         pci_set_drvdata(pdev, NULL);
3384 out:
3385         return err;
3386 }
3387
3388 static void __devexit remove_one(struct pci_dev *pdev)
3389 {
3390         struct adapter *adapter = pci_get_drvdata(pdev);
3391
3392         if (adapter) {
3393                 int i;
3394
3395                 t3_sge_stop(adapter);
3396                 sysfs_remove_group(&adapter->port[0]->dev.kobj,
3397                                    &cxgb3_attr_group);
3398
3399                 if (is_offload(adapter)) {
3400                         cxgb3_adapter_unofld(adapter);
3401                         if (test_bit(OFFLOAD_DEVMAP_BIT,
3402                                      &adapter->open_device_map))
3403                                 offload_close(&adapter->tdev);
3404                 }
3405
3406                 for_each_port(adapter, i)
3407                     if (test_bit(i, &adapter->registered_device_map))
3408                         unregister_netdev(adapter->port[i]);
3409
3410                 t3_stop_sge_timers(adapter);
3411                 t3_free_sge_resources(adapter);
3412                 cxgb_disable_msi(adapter);
3413
3414                 for_each_port(adapter, i)
3415                         if (adapter->port[i])
3416                                 free_netdev(adapter->port[i]);
3417
3418                 iounmap(adapter->regs);
3419                 if (adapter->nofail_skb)
3420                         kfree_skb(adapter->nofail_skb);
3421                 kfree(adapter);
3422                 pci_release_regions(pdev);
3423                 pci_disable_device(pdev);
3424                 pci_set_drvdata(pdev, NULL);
3425         }
3426 }
3427
3428 static struct pci_driver driver = {
3429         .name = DRV_NAME,
3430         .id_table = cxgb3_pci_tbl,
3431         .probe = init_one,
3432         .remove = __devexit_p(remove_one),
3433         .err_handler = &t3_err_handler,
3434 };
3435
3436 static int __init cxgb3_init_module(void)
3437 {
3438         int ret;
3439
3440         cxgb3_offload_init();
3441
3442         ret = pci_register_driver(&driver);
3443         return ret;
3444 }
3445
3446 static void __exit cxgb3_cleanup_module(void)
3447 {
3448         pci_unregister_driver(&driver);
3449         if (cxgb3_wq)
3450                 destroy_workqueue(cxgb3_wq);
3451 }
3452
3453 module_init(cxgb3_init_module);
3454 module_exit(cxgb3_cleanup_module);