drivers/net/cxgb4/cxgb4_main.c

   1 /*
   2  * This file is part of the Chelsio T4 Ethernet driver for Linux.
   3  *
   4  * Copyright (c) 2003-2010 Chelsio Communications, Inc. All rights reserved.
   5  *
   6  * This software is available to you under a choice of one of two
   7  * licenses.  You may choose to be licensed under the terms of the GNU
   8  * General Public License (GPL) Version 2, available from the file
   9  * COPYING in the main directory of this source tree, or the
  10  * OpenIB.org BSD license below:
  11  *
  12  *     Redistribution and use in source and binary forms, with or
  13  *     without modification, are permitted provided that the following
  14  *     conditions are met:
  15  *
  16  *      - Redistributions of source code must retain the above
  17  *        copyright notice, this list of conditions and the following
  18  *        disclaimer.
  19  *
  20  *      - Redistributions in binary form must reproduce the above
  21  *        copyright notice, this list of conditions and the following
  22  *        disclaimer in the documentation and/or other materials
  23  *        provided with the distribution.
  24  *
  25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  32  * SOFTWARE.
  33  */
  34
  35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  36
  37 #include <linux/bitmap.h>
  38 #include <linux/crc32.h>
  39 #include <linux/ctype.h>
  40 #include <linux/debugfs.h>
  41 #include <linux/err.h>
  42 #include <linux/etherdevice.h>
  43 #include <linux/firmware.h>
  44 #include <linux/if_vlan.h>
  45 #include <linux/init.h>
  46 #include <linux/log2.h>
  47 #include <linux/mdio.h>
  48 #include <linux/module.h>
  49 #include <linux/moduleparam.h>
  50 #include <linux/mutex.h>
  51 #include <linux/netdevice.h>
  52 #include <linux/pci.h>
  53 #include <linux/aer.h>
  54 #include <linux/rtnetlink.h>
  55 #include <linux/sched.h>
  56 #include <linux/seq_file.h>
  57 #include <linux/sockios.h>
  58 #include <linux/vmalloc.h>
  59 #include <linux/workqueue.h>
  60 #include <net/neighbour.h>
  61 #include <net/netevent.h>
  62 #include <asm/uaccess.h>
  63
  64 #include "cxgb4.h"
  65 #include "t4_regs.h"
  66 #include "t4_msg.h"
  67 #include "t4fw_api.h"
  68 #include "l2t.h"
  69
  70 #define DRV_VERSION "1.0.0-ko"
  71 #define DRV_DESC "Chelsio T4 Network Driver"
  72
  73 /*
  74  * Max interrupt hold-off timer value in us.  Queues fall back to this value
  75  * under extreme memory pressure so it's largish to give the system time to
  76  * recover.
  77  */
  78 #define MAX_SGE_TIMERVAL 200U
  79
  80 enum {
  81         MEMWIN0_APERTURE = 65536,
  82         MEMWIN0_BASE     = 0x30000,
  83         MEMWIN1_APERTURE = 32768,
  84         MEMWIN1_BASE     = 0x28000,
  85         MEMWIN2_APERTURE = 2048,
  86         MEMWIN2_BASE     = 0x1b800,
  87 };
  88
  89 enum {
  90         MAX_TXQ_ENTRIES      = 16384,
  91         MAX_CTRL_TXQ_ENTRIES = 1024,
  92         MAX_RSPQ_ENTRIES     = 16384,
  93         MAX_RX_BUFFERS       = 16384,
  94         MIN_TXQ_ENTRIES      = 32,
  95         MIN_CTRL_TXQ_ENTRIES = 32,
  96         MIN_RSPQ_ENTRIES     = 128,
  97         MIN_FL_ENTRIES       = 16
  98 };
  99
 100 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
 101                          NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
 102                          NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
 103
 104 #define CH_DEVICE(devid) { PCI_VDEVICE(CHELSIO, devid), 0 }
 105
 106 static DEFINE_PCI_DEVICE_TABLE(cxgb4_pci_tbl) = {
 107         CH_DEVICE(0xa000),  /* PE10K */
 108         { 0, }
 109 };
 110
 111 #define FW_FNAME "cxgb4/t4fw.bin"
 112
 113 MODULE_DESCRIPTION(DRV_DESC);
 114 MODULE_AUTHOR("Chelsio Communications");
 115 MODULE_LICENSE("Dual BSD/GPL");
 116 MODULE_VERSION(DRV_VERSION);
 117 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
 118 MODULE_FIRMWARE(FW_FNAME);
 119
 120 static int dflt_msg_enable = DFLT_MSG_ENABLE;
 121
 122 module_param(dflt_msg_enable, int, 0644);
 123 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T4 default message enable bitmap");
 124
 125 /*
 126  * The driver uses the best interrupt scheme available on a platform in the
 127  * order MSI-X, MSI, legacy INTx interrupts.  This parameter determines which
 128  * of these schemes the driver may consider as follows:
 129  *
 130  * msi = 2: choose from among all three options
 131  * msi = 1: only consider MSI and INTx interrupts
 132  * msi = 0: force INTx interrupts
 133  */
 134 static int msi = 2;
 135
 136 module_param(msi, int, 0644);
 137 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
 138
 139 /*
 140  * Queue interrupt hold-off timer values.  Queues default to the first of these
 141  * upon creation.
 142  */
 143 static unsigned int intr_holdoff[SGE_NTIMERS - 1] = { 5, 10, 20, 50, 100 };
 144
 145 module_param_array(intr_holdoff, uint, NULL, 0644);
 146 MODULE_PARM_DESC(intr_holdoff, "values for queue interrupt hold-off timers "
 147                  "0..4 in microseconds");
 148
 149 static unsigned int intr_cnt[SGE_NCOUNTERS - 1] = { 4, 8, 16 };
 150
 151 module_param_array(intr_cnt, uint, NULL, 0644);
 152 MODULE_PARM_DESC(intr_cnt,
 153                  "thresholds 1..3 for queue interrupt packet counters");
 154
 155 static int vf_acls;
 156
 157 #ifdef CONFIG_PCI_IOV
 158 module_param(vf_acls, bool, 0644);
 159 MODULE_PARM_DESC(vf_acls, "if set enable virtualization L2 ACL enforcement");
 160
 161 static unsigned int num_vf[4];
 162
 163 module_param_array(num_vf, uint, NULL, 0644);
 164 MODULE_PARM_DESC(num_vf, "number of VFs for each of PFs 0-3");
 165 #endif
 166
 167 static struct dentry *cxgb4_debugfs_root;
 168
 169 static LIST_HEAD(adapter_list);
 170 static DEFINE_MUTEX(uld_mutex);
 171 static struct cxgb4_uld_info ulds[CXGB4_ULD_MAX];
 172 static const char *uld_str[] = { "RDMA", "iSCSI" };
 173
 174 static void link_report(struct net_device *dev)
 175 {
 176         if (!netif_carrier_ok(dev))
 177                 netdev_info(dev, "link down\n");
 178         else {
 179                 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
 180
 181                 const char *s = "10Mbps";
 182                 const struct port_info *p = netdev_priv(dev);
 183
 184                 switch (p->link_cfg.speed) {
 185                 case SPEED_10000:
 186                         s = "10Gbps";
 187                         break;
 188                 case SPEED_1000:
 189                         s = "1000Mbps";
 190                         break;
 191                 case SPEED_100:
 192                         s = "100Mbps";
 193                         break;
 194                 }
 195
 196                 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
 197                             fc[p->link_cfg.fc]);
 198         }
 199 }
 200
 201 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
 202 {
 203         struct net_device *dev = adapter->port[port_id];
 204
 205         /* Skip changes from disabled ports. */
 206         if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
 207                 if (link_stat)
 208                         netif_carrier_on(dev);
 209                 else
 210                         netif_carrier_off(dev);
 211
 212                 link_report(dev);
 213         }
 214 }
 215
 216 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
 217 {
 218         static const char *mod_str[] = {
 219                 NULL, "LR", "SR", "ER", "passive DA", "active DA"
 220         };
 221
 222         const struct net_device *dev = adap->port[port_id];
 223         const struct port_info *pi = netdev_priv(dev);
 224
 225         if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
 226                 netdev_info(dev, "port module unplugged\n");
 227         else
 228                 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
 229 }
 230
 231 /*
 232  * Configure the exact and hash address filters to handle a port's multicast
 233  * and secondary unicast MAC addresses.
 234  */
 235 static int set_addr_filters(const struct net_device *dev, bool sleep)
 236 {
 237         u64 mhash = 0;
 238         u64 uhash = 0;
 239         bool free = true;
 240         u16 filt_idx[7];
 241         const u8 *addr[7];
 242         int ret, naddr = 0;
 243         const struct netdev_hw_addr *ha;
 244         int uc_cnt = netdev_uc_count(dev);
 245         int mc_cnt = netdev_mc_count(dev);
 246         const struct port_info *pi = netdev_priv(dev);
 247
 248         /* first do the secondary unicast addresses */
 249         netdev_for_each_uc_addr(ha, dev) {
 250                 addr[naddr++] = ha->addr;
 251                 if (--uc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
 252                         ret = t4_alloc_mac_filt(pi->adapter, 0, pi->viid, free,
 253                                         naddr, addr, filt_idx, &uhash, sleep);
 254                         if (ret < 0)
 255                                 return ret;
 256
 257                         free = false;
 258                         naddr = 0;
 259                 }
 260         }
 261
 262         /* next set up the multicast addresses */
 263         netdev_for_each_mc_addr(ha, dev) {
 264                 addr[naddr++] = ha->addr;
 265                 if (--mc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
 266                         ret = t4_alloc_mac_filt(pi->adapter, 0, pi->viid, free,
 267                                         naddr, addr, filt_idx, &mhash, sleep);
 268                         if (ret < 0)
 269                                 return ret;
 270
 271                         free = false;
 272                         naddr = 0;
 273                 }
 274         }
 275
 276         return t4_set_addr_hash(pi->adapter, 0, pi->viid, uhash != 0,
 277                                 uhash | mhash, sleep);
 278 }
 279
 280 /*
 281  * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
 282  * If @mtu is -1 it is left unchanged.
 283  */
 284 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
 285 {
 286         int ret;
 287         struct port_info *pi = netdev_priv(dev);
 288
 289         ret = set_addr_filters(dev, sleep_ok);
 290         if (ret == 0)
 291                 ret = t4_set_rxmode(pi->adapter, 0, pi->viid, mtu,
 292                                     (dev->flags & IFF_PROMISC) ? 1 : 0,
 293                                     (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
 294                                     sleep_ok);
 295         return ret;
 296 }
 297
 298 /**
 299  *      link_start - enable a port
 300  *      @dev: the port to enable
 301  *
 302  *      Performs the MAC and PHY actions needed to enable a port.
 303  */
 304 static int link_start(struct net_device *dev)
 305 {
 306         int ret;
 307         struct port_info *pi = netdev_priv(dev);
 308
 309         /*
 310          * We do not set address filters and promiscuity here, the stack does
 311          * that step explicitly.
 312          */
 313         ret = t4_set_rxmode(pi->adapter, 0, pi->viid, dev->mtu, -1, -1, -1,
 314                             pi->vlan_grp != NULL, true);
 315         if (ret == 0) {
 316                 ret = t4_change_mac(pi->adapter, 0, pi->viid,
 317                                     pi->xact_addr_filt, dev->dev_addr, true,
 318                                     true);
 319                 if (ret >= 0) {
 320                         pi->xact_addr_filt = ret;
 321                         ret = 0;
 322                 }
 323         }
 324         if (ret == 0)
 325                 ret = t4_link_start(pi->adapter, 0, pi->tx_chan, &pi->link_cfg);
 326         if (ret == 0)
 327                 ret = t4_enable_vi(pi->adapter, 0, pi->viid, true, true);
 328         return ret;
 329 }
 330
 331 /*
 332  * Response queue handler for the FW event queue.
 333  */
 334 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
 335                           const struct pkt_gl *gl)
 336 {
 337         u8 opcode = ((const struct rss_header *)rsp)->opcode;
 338
 339         rsp++;                                          /* skip RSS header */
 340         if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
 341                 const struct cpl_sge_egr_update *p = (void *)rsp;
 342                 unsigned int qid = EGR_QID(ntohl(p->opcode_qid));
 343                 struct sge_txq *txq = q->adap->sge.egr_map[qid];
 344
 345                 txq->restarts++;
 346                 if ((u8 *)txq < (u8 *)q->adap->sge.ethrxq) {
 347                         struct sge_eth_txq *eq;
 348
 349                         eq = container_of(txq, struct sge_eth_txq, q);
 350                         netif_tx_wake_queue(eq->txq);
 351                 } else {
 352                         struct sge_ofld_txq *oq;
 353
 354                         oq = container_of(txq, struct sge_ofld_txq, q);
 355                         tasklet_schedule(&oq->qresume_tsk);
 356                 }
 357         } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
 358                 const struct cpl_fw6_msg *p = (void *)rsp;
 359
 360                 if (p->type == 0)
 361                         t4_handle_fw_rpl(q->adap, p->data);
 362         } else if (opcode == CPL_L2T_WRITE_RPL) {
 363                 const struct cpl_l2t_write_rpl *p = (void *)rsp;
 364
 365                 do_l2t_write_rpl(q->adap, p);
 366         } else
 367                 dev_err(q->adap->pdev_dev,
 368                         "unexpected CPL %#x on FW event queue\n", opcode);
 369         return 0;
 370 }
 371
 372 /**
 373  *      uldrx_handler - response queue handler for ULD queues
 374  *      @q: the response queue that received the packet
 375  *      @rsp: the response queue descriptor holding the offload message
 376  *      @gl: the gather list of packet fragments
 377  *
 378  *      Deliver an ingress offload packet to a ULD.  All processing is done by
 379  *      the ULD, we just maintain statistics.
 380  */
 381 static int uldrx_handler(struct sge_rspq *q, const __be64 *rsp,
 382                          const struct pkt_gl *gl)
 383 {
 384         struct sge_ofld_rxq *rxq = container_of(q, struct sge_ofld_rxq, rspq);
 385
 386         if (ulds[q->uld].rx_handler(q->adap->uld_handle[q->uld], rsp, gl)) {
 387                 rxq->stats.nomem++;
 388                 return -1;
 389         }
 390         if (gl == NULL)
 391                 rxq->stats.imm++;
 392         else if (gl == CXGB4_MSG_AN)
 393                 rxq->stats.an++;
 394         else
 395                 rxq->stats.pkts++;
 396         return 0;
 397 }
 398
 399 static void disable_msi(struct adapter *adapter)
 400 {
 401         if (adapter->flags & USING_MSIX) {
 402                 pci_disable_msix(adapter->pdev);
 403                 adapter->flags &= ~USING_MSIX;
 404         } else if (adapter->flags & USING_MSI) {
 405                 pci_disable_msi(adapter->pdev);
 406                 adapter->flags &= ~USING_MSI;
 407         }
 408 }
 409
 410 /*
 411  * Interrupt handler for non-data events used with MSI-X.
 412  */
 413 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
 414 {
 415         struct adapter *adap = cookie;
 416
 417         u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE));
 418         if (v & PFSW) {
 419                 adap->swintr = 1;
 420                 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE), v);
 421         }
 422         t4_slow_intr_handler(adap);
 423         return IRQ_HANDLED;
 424 }
 425
 426 /*
 427  * Name the MSI-X interrupts.
 428  */
 429 static void name_msix_vecs(struct adapter *adap)
 430 {
 431         int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc) - 1;
 432
 433         /* non-data interrupts */
 434         snprintf(adap->msix_info[0].desc, n, "%s", adap->name);
 435         adap->msix_info[0].desc[n] = 0;
 436
 437         /* FW events */
 438         snprintf(adap->msix_info[1].desc, n, "%s-FWeventq", adap->name);
 439         adap->msix_info[1].desc[n] = 0;
 440
 441         /* Ethernet queues */
 442         for_each_port(adap, j) {
 443                 struct net_device *d = adap->port[j];
 444                 const struct port_info *pi = netdev_priv(d);
 445
 446                 for (i = 0; i < pi->nqsets; i++, msi_idx++) {
 447                         snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
 448                                  d->name, i);
 449                         adap->msix_info[msi_idx].desc[n] = 0;
 450                 }
 451         }
 452
 453         /* offload queues */
 454         for_each_ofldrxq(&adap->sge, i) {
 455                 snprintf(adap->msix_info[msi_idx].desc, n, "%s-ofld%d",
 456                          adap->name, i);
 457                 adap->msix_info[msi_idx++].desc[n] = 0;
 458         }
 459         for_each_rdmarxq(&adap->sge, i) {
 460                 snprintf(adap->msix_info[msi_idx].desc, n, "%s-rdma%d",
 461                          adap->name, i);
 462                 adap->msix_info[msi_idx++].desc[n] = 0;
 463         }
 464 }
 465
 466 static int request_msix_queue_irqs(struct adapter *adap)
 467 {
 468         struct sge *s = &adap->sge;
 469         int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, msi = 2;
 470
 471         err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
 472                           adap->msix_info[1].desc, &s->fw_evtq);
 473         if (err)
 474                 return err;
 475
 476         for_each_ethrxq(s, ethqidx) {
 477                 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
 478                                   adap->msix_info[msi].desc,
 479                                   &s->ethrxq[ethqidx].rspq);
 480                 if (err)
 481                         goto unwind;
 482                 msi++;
 483         }
 484         for_each_ofldrxq(s, ofldqidx) {
 485                 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
 486                                   adap->msix_info[msi].desc,
 487                                   &s->ofldrxq[ofldqidx].rspq);
 488                 if (err)
 489                         goto unwind;
 490                 msi++;
 491         }
 492         for_each_rdmarxq(s, rdmaqidx) {
 493                 err = request_irq(adap->msix_info[msi].vec, t4_sge_intr_msix, 0,
 494                                   adap->msix_info[msi].desc,
 495                                   &s->rdmarxq[rdmaqidx].rspq);
 496                 if (err)
 497                         goto unwind;
 498                 msi++;
 499         }
 500         return 0;
 501
 502 unwind:
 503         while (--rdmaqidx >= 0)
 504                 free_irq(adap->msix_info[--msi].vec,
 505                          &s->rdmarxq[rdmaqidx].rspq);
 506         while (--ofldqidx >= 0)
 507                 free_irq(adap->msix_info[--msi].vec,
 508                          &s->ofldrxq[ofldqidx].rspq);
 509         while (--ethqidx >= 0)
 510                 free_irq(adap->msix_info[--msi].vec, &s->ethrxq[ethqidx].rspq);
 511         free_irq(adap->msix_info[1].vec, &s->fw_evtq);
 512         return err;
 513 }
 514
 515 static void free_msix_queue_irqs(struct adapter *adap)
 516 {
 517         int i, msi = 2;
 518         struct sge *s = &adap->sge;
 519
 520         free_irq(adap->msix_info[1].vec, &s->fw_evtq);
 521         for_each_ethrxq(s, i)
 522                 free_irq(adap->msix_info[msi++].vec, &s->ethrxq[i].rspq);
 523         for_each_ofldrxq(s, i)
 524                 free_irq(adap->msix_info[msi++].vec, &s->ofldrxq[i].rspq);
 525         for_each_rdmarxq(s, i)
 526                 free_irq(adap->msix_info[msi++].vec, &s->rdmarxq[i].rspq);
 527 }
 528
 529 /**
 530  *      setup_rss - configure RSS
 531  *      @adap: the adapter
 532  *
 533  *      Sets up RSS to distribute packets to multiple receive queues.  We
 534  *      configure the RSS CPU lookup table to distribute to the number of HW
 535  *      receive queues, and the response queue lookup table to narrow that
 536  *      down to the response queues actually configured for each port.
 537  *      We always configure the RSS mapping for all ports since the mapping
 538  *      table has plenty of entries.
 539  */
 540 static int setup_rss(struct adapter *adap)
 541 {
 542         int i, j, err;
 543         u16 rss[MAX_ETH_QSETS];
 544
 545         for_each_port(adap, i) {
 546                 const struct port_info *pi = adap2pinfo(adap, i);
 547                 const struct sge_eth_rxq *q = &adap->sge.ethrxq[pi->first_qset];
 548
 549                 for (j = 0; j < pi->nqsets; j++)
 550                         rss[j] = q[j].rspq.abs_id;
 551
 552                 err = t4_config_rss_range(adap, 0, pi->viid, 0, pi->rss_size,
 553                                           rss, pi->nqsets);
 554                 if (err)
 555                         return err;
 556         }
 557         return 0;
 558 }
 559
 560 /*
 561  * Wait until all NAPI handlers are descheduled.
 562  */
 563 static void quiesce_rx(struct adapter *adap)
 564 {
 565         int i;
 566
 567         for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
 568                 struct sge_rspq *q = adap->sge.ingr_map[i];
 569
 570                 if (q && q->handler)
 571                         napi_disable(&q->napi);
 572         }
 573 }
 574
 575 /*
 576  * Enable NAPI scheduling and interrupt generation for all Rx queues.
 577  */
 578 static void enable_rx(struct adapter *adap)
 579 {
 580         int i;
 581
 582         for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
 583                 struct sge_rspq *q = adap->sge.ingr_map[i];
 584
 585                 if (!q)
 586                         continue;
 587                 if (q->handler)
 588                         napi_enable(&q->napi);
 589                 /* 0-increment GTS to start the timer and enable interrupts */
 590                 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
 591                              SEINTARM(q->intr_params) |
 592                              INGRESSQID(q->cntxt_id));
 593         }
 594 }
 595
 596 /**
 597  *      setup_sge_queues - configure SGE Tx/Rx/response queues
 598  *      @adap: the adapter
 599  *
 600  *      Determines how many sets of SGE queues to use and initializes them.
 601  *      We support multiple queue sets per port if we have MSI-X, otherwise
 602  *      just one queue set per port.
 603  */
 604 static int setup_sge_queues(struct adapter *adap)
 605 {
 606         int err, msi_idx, i, j;
 607         struct sge *s = &adap->sge;
 608
 609         bitmap_zero(s->starving_fl, MAX_EGRQ);
 610         bitmap_zero(s->txq_maperr, MAX_EGRQ);
 611
 612         if (adap->flags & USING_MSIX)
 613                 msi_idx = 1;         /* vector 0 is for non-queue interrupts */
 614         else {
 615                 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
 616                                        NULL, NULL);
 617                 if (err)
 618                         return err;
 619                 msi_idx = -((int)s->intrq.abs_id + 1);
 620         }
 621
 622         err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
 623                                msi_idx, NULL, fwevtq_handler);
 624         if (err) {
 625 freeout:        t4_free_sge_resources(adap);
 626                 return err;
 627         }
 628
 629         for_each_port(adap, i) {
 630                 struct net_device *dev = adap->port[i];
 631                 struct port_info *pi = netdev_priv(dev);
 632                 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
 633                 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
 634
 635                 for (j = 0; j < pi->nqsets; j++, q++) {
 636                         if (msi_idx > 0)
 637                                 msi_idx++;
 638                         err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
 639                                                msi_idx, &q->fl,
 640                                                t4_ethrx_handler);
 641                         if (err)
 642                                 goto freeout;
 643                         q->rspq.idx = j;
 644                         memset(&q->stats, 0, sizeof(q->stats));
 645                 }
 646                 for (j = 0; j < pi->nqsets; j++, t++) {
 647                         err = t4_sge_alloc_eth_txq(adap, t, dev,
 648                                         netdev_get_tx_queue(dev, j),
 649                                         s->fw_evtq.cntxt_id);
 650                         if (err)
 651                                 goto freeout;
 652                 }
 653         }
 654
 655         j = s->ofldqsets / adap->params.nports; /* ofld queues per channel */
 656         for_each_ofldrxq(s, i) {
 657                 struct sge_ofld_rxq *q = &s->ofldrxq[i];
 658                 struct net_device *dev = adap->port[i / j];
 659
 660                 if (msi_idx > 0)
 661                         msi_idx++;
 662                 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
 663                                        &q->fl, uldrx_handler);
 664                 if (err)
 665                         goto freeout;
 666                 memset(&q->stats, 0, sizeof(q->stats));
 667                 s->ofld_rxq[i] = q->rspq.abs_id;
 668                 err = t4_sge_alloc_ofld_txq(adap, &s->ofldtxq[i], dev,
 669                                             s->fw_evtq.cntxt_id);
 670                 if (err)
 671                         goto freeout;
 672         }
 673
 674         for_each_rdmarxq(s, i) {
 675                 struct sge_ofld_rxq *q = &s->rdmarxq[i];
 676
 677                 if (msi_idx > 0)
 678                         msi_idx++;
 679                 err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
 680                                        msi_idx, &q->fl, uldrx_handler);
 681                 if (err)
 682                         goto freeout;
 683                 memset(&q->stats, 0, sizeof(q->stats));
 684                 s->rdma_rxq[i] = q->rspq.abs_id;
 685         }
 686
 687         for_each_port(adap, i) {
 688                 /*
 689                  * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
 690                  * have RDMA queues, and that's the right value.
 691                  */
 692                 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
 693                                             s->fw_evtq.cntxt_id,
 694                                             s->rdmarxq[i].rspq.cntxt_id);
 695                 if (err)
 696                         goto freeout;
 697         }
 698
 699         t4_write_reg(adap, MPS_TRC_RSS_CONTROL,
 700                      RSSCONTROL(netdev2pinfo(adap->port[0])->tx_chan) |
 701                      QUEUENUMBER(s->ethrxq[0].rspq.abs_id));
 702         return 0;
 703 }
 704
 705 /*
 706  * Returns 0 if new FW was successfully loaded, a positive errno if a load was
 707  * started but failed, and a negative errno if flash load couldn't start.
 708  */
 709 static int upgrade_fw(struct adapter *adap)
 710 {
 711         int ret;
 712         u32 vers;
 713         const struct fw_hdr *hdr;
 714         const struct firmware *fw;
 715         struct device *dev = adap->pdev_dev;
 716
 717         ret = request_firmware(&fw, FW_FNAME, dev);
 718         if (ret < 0) {
 719                 dev_err(dev, "unable to load firmware image " FW_FNAME
 720                         ", error %d\n", ret);
 721                 return ret;
 722         }
 723
 724         hdr = (const struct fw_hdr *)fw->data;
 725         vers = ntohl(hdr->fw_ver);
 726         if (FW_HDR_FW_VER_MAJOR_GET(vers) != FW_VERSION_MAJOR) {
 727                 ret = -EINVAL;              /* wrong major version, won't do */
 728                 goto out;
 729         }
 730
 731         /*
 732          * If the flash FW is unusable or we found something newer, load it.
 733          */
 734         if (FW_HDR_FW_VER_MAJOR_GET(adap->params.fw_vers) != FW_VERSION_MAJOR ||
 735             vers > adap->params.fw_vers) {
 736                 ret = -t4_load_fw(adap, fw->data, fw->size);
 737                 if (!ret)
 738                         dev_info(dev, "firmware upgraded to version %pI4 from "
 739                                  FW_FNAME "\n", &hdr->fw_ver);
 740         }
 741 out:    release_firmware(fw);
 742         return ret;
 743 }
 744
 745 /*
 746  * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
 747  * The allocated memory is cleared.
 748  */
 749 void *t4_alloc_mem(size_t size)
 750 {
 751         void *p = kmalloc(size, GFP_KERNEL);
 752
 753         if (!p)
 754                 p = vmalloc(size);
 755         if (p)
 756                 memset(p, 0, size);
 757         return p;
 758 }
 759
 760 /*
 761  * Free memory allocated through alloc_mem().
 762  */
 763 void t4_free_mem(void *addr)
 764 {
 765         if (is_vmalloc_addr(addr))
 766                 vfree(addr);
 767         else
 768                 kfree(addr);
 769 }
 770
 771 static inline int is_offload(const struct adapter *adap)
 772 {
 773         return adap->params.offload;
 774 }
 775
 776 /*
 777  * Implementation of ethtool operations.
 778  */
 779
 780 static u32 get_msglevel(struct net_device *dev)
 781 {
 782         return netdev2adap(dev)->msg_enable;
 783 }
 784
 785 static void set_msglevel(struct net_device *dev, u32 val)
 786 {
 787         netdev2adap(dev)->msg_enable = val;
 788 }
 789
 790 static char stats_strings[][ETH_GSTRING_LEN] = {
 791         "TxOctetsOK         ",
 792         "TxFramesOK         ",
 793         "TxBroadcastFrames  ",
 794         "TxMulticastFrames  ",
 795         "TxUnicastFrames    ",
 796         "TxErrorFrames      ",
 797
 798         "TxFrames64         ",
 799         "TxFrames65To127    ",
 800         "TxFrames128To255   ",
 801         "TxFrames256To511   ",
 802         "TxFrames512To1023  ",
 803         "TxFrames1024To1518 ",
 804         "TxFrames1519ToMax  ",
 805
 806         "TxFramesDropped    ",
 807         "TxPauseFrames      ",
 808         "TxPPP0Frames       ",
 809         "TxPPP1Frames       ",
 810         "TxPPP2Frames       ",
 811         "TxPPP3Frames       ",
 812         "TxPPP4Frames       ",
 813         "TxPPP5Frames       ",
 814         "TxPPP6Frames       ",
 815         "TxPPP7Frames       ",
 816
 817         "RxOctetsOK         ",
 818         "RxFramesOK         ",
 819         "RxBroadcastFrames  ",
 820         "RxMulticastFrames  ",
 821         "RxUnicastFrames    ",
 822
 823         "RxFramesTooLong    ",
 824         "RxJabberErrors     ",
 825         "RxFCSErrors        ",
 826         "RxLengthErrors     ",
 827         "RxSymbolErrors     ",
 828         "RxRuntFrames       ",
 829
 830         "RxFrames64         ",
 831         "RxFrames65To127    ",
 832         "RxFrames128To255   ",
 833         "RxFrames256To511   ",
 834         "RxFrames512To1023  ",
 835         "RxFrames1024To1518 ",
 836         "RxFrames1519ToMax  ",
 837
 838         "RxPauseFrames      ",
 839         "RxPPP0Frames       ",
 840         "RxPPP1Frames       ",
 841         "RxPPP2Frames       ",
 842         "RxPPP3Frames       ",
 843         "RxPPP4Frames       ",
 844         "RxPPP5Frames       ",
 845         "RxPPP6Frames       ",
 846         "RxPPP7Frames       ",
 847
 848         "RxBG0FramesDropped ",
 849         "RxBG1FramesDropped ",
 850         "RxBG2FramesDropped ",
 851         "RxBG3FramesDropped ",
 852         "RxBG0FramesTrunc   ",
 853         "RxBG1FramesTrunc   ",
 854         "RxBG2FramesTrunc   ",
 855         "RxBG3FramesTrunc   ",
 856
 857         "TSO                ",
 858         "TxCsumOffload      ",
 859         "RxCsumGood         ",
 860         "VLANextractions    ",
 861         "VLANinsertions     ",
 862         "GROpackets         ",
 863         "GROmerged          ",
 864 };
 865
 866 static int get_sset_count(struct net_device *dev, int sset)
 867 {
 868         switch (sset) {
 869         case ETH_SS_STATS:
 870                 return ARRAY_SIZE(stats_strings);
 871         default:
 872                 return -EOPNOTSUPP;
 873         }
 874 }
 875
 876 #define T4_REGMAP_SIZE (160 * 1024)
 877
 878 static int get_regs_len(struct net_device *dev)
 879 {
 880         return T4_REGMAP_SIZE;
 881 }
 882
 883 static int get_eeprom_len(struct net_device *dev)
 884 {
 885         return EEPROMSIZE;
 886 }
 887
 888 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
 889 {
 890         struct adapter *adapter = netdev2adap(dev);
 891
 892         strcpy(info->driver, KBUILD_MODNAME);
 893         strcpy(info->version, DRV_VERSION);
 894         strcpy(info->bus_info, pci_name(adapter->pdev));
 895
 896         if (!adapter->params.fw_vers)
 897                 strcpy(info->fw_version, "N/A");
 898         else
 899                 snprintf(info->fw_version, sizeof(info->fw_version),
 900                         "%u.%u.%u.%u, TP %u.%u.%u.%u",
 901                         FW_HDR_FW_VER_MAJOR_GET(adapter->params.fw_vers),
 902                         FW_HDR_FW_VER_MINOR_GET(adapter->params.fw_vers),
 903                         FW_HDR_FW_VER_MICRO_GET(adapter->params.fw_vers),
 904                         FW_HDR_FW_VER_BUILD_GET(adapter->params.fw_vers),
 905                         FW_HDR_FW_VER_MAJOR_GET(adapter->params.tp_vers),
 906                         FW_HDR_FW_VER_MINOR_GET(adapter->params.tp_vers),
 907                         FW_HDR_FW_VER_MICRO_GET(adapter->params.tp_vers),
 908                         FW_HDR_FW_VER_BUILD_GET(adapter->params.tp_vers));
 909 }
 910
 911 static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
 912 {
 913         if (stringset == ETH_SS_STATS)
 914                 memcpy(data, stats_strings, sizeof(stats_strings));
 915 }
 916
 917 /*
 918  * port stats maintained per queue of the port.  They should be in the same
 919  * order as in stats_strings above.
 920  */
 921 struct queue_port_stats {
 922         u64 tso;
 923         u64 tx_csum;
 924         u64 rx_csum;
 925         u64 vlan_ex;
 926         u64 vlan_ins;
 927         u64 gro_pkts;
 928         u64 gro_merged;
 929 };
 930
 931 static void collect_sge_port_stats(const struct adapter *adap,
 932                 const struct port_info *p, struct queue_port_stats *s)
 933 {
 934         int i;
 935         const struct sge_eth_txq *tx = &adap->sge.ethtxq[p->first_qset];
 936         const struct sge_eth_rxq *rx = &adap->sge.ethrxq[p->first_qset];
 937
 938         memset(s, 0, sizeof(*s));
 939         for (i = 0; i < p->nqsets; i++, rx++, tx++) {
 940                 s->tso += tx->tso;
 941                 s->tx_csum += tx->tx_cso;
 942                 s->rx_csum += rx->stats.rx_cso;
 943                 s->vlan_ex += rx->stats.vlan_ex;
 944                 s->vlan_ins += tx->vlan_ins;
 945                 s->gro_pkts += rx->stats.lro_pkts;
 946                 s->gro_merged += rx->stats.lro_merged;
 947         }
 948 }
 949
 950 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
 951                       u64 *data)
 952 {
 953         struct port_info *pi = netdev_priv(dev);
 954         struct adapter *adapter = pi->adapter;
 955
 956         t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
 957
 958         data += sizeof(struct port_stats) / sizeof(u64);
 959         collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
 960 }
 961
 962 /*
 963  * Return a version number to identify the type of adapter.  The scheme is:
 964  * - bits 0..9: chip version
 965  * - bits 10..15: chip revision
 966  */
 967 static inline unsigned int mk_adap_vers(const struct adapter *ap)
 968 {
 969         return 4 | (ap->params.rev << 10);
 970 }
 971
 972 static void reg_block_dump(struct adapter *ap, void *buf, unsigned int start,
 973                            unsigned int end)
 974 {
 975         u32 *p = buf + start;
 976
 977         for ( ; start <= end; start += sizeof(u32))
 978                 *p++ = t4_read_reg(ap, start);
 979 }
 980
 981 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
 982                      void *buf)
 983 {
 984         static const unsigned int reg_ranges[] = {
 985                 0x1008, 0x1108,
 986                 0x1180, 0x11b4,
 987                 0x11fc, 0x123c,
 988                 0x1300, 0x173c,
 989                 0x1800, 0x18fc,
 990                 0x3000, 0x30d8,
 991                 0x30e0, 0x5924,
 992                 0x5960, 0x59d4,
 993                 0x5a00, 0x5af8,
 994                 0x6000, 0x6098,
 995                 0x6100, 0x6150,
 996                 0x6200, 0x6208,
 997                 0x6240, 0x6248,
 998                 0x6280, 0x6338,
 999                 0x6370, 0x638c,
1000                 0x6400, 0x643c,
1001                 0x6500, 0x6524,
1002                 0x6a00, 0x6a38,
1003                 0x6a60, 0x6a78,
1004                 0x6b00, 0x6b84,
1005                 0x6bf0, 0x6c84,
1006                 0x6cf0, 0x6d84,
1007                 0x6df0, 0x6e84,
1008                 0x6ef0, 0x6f84,
1009                 0x6ff0, 0x7084,
1010                 0x70f0, 0x7184,
1011                 0x71f0, 0x7284,
1012                 0x72f0, 0x7384,
1013                 0x73f0, 0x7450,
1014                 0x7500, 0x7530,
1015                 0x7600, 0x761c,
1016                 0x7680, 0x76cc,
1017                 0x7700, 0x7798,
1018                 0x77c0, 0x77fc,
1019                 0x7900, 0x79fc,
1020                 0x7b00, 0x7c38,
1021                 0x7d00, 0x7efc,
1022                 0x8dc0, 0x8e1c,
1023                 0x8e30, 0x8e78,
1024                 0x8ea0, 0x8f6c,
1025                 0x8fc0, 0x9074,
1026                 0x90fc, 0x90fc,
1027                 0x9400, 0x9458,
1028                 0x9600, 0x96bc,
1029                 0x9800, 0x9808,
1030                 0x9820, 0x983c,
1031                 0x9850, 0x9864,
1032                 0x9c00, 0x9c6c,
1033                 0x9c80, 0x9cec,
1034                 0x9d00, 0x9d6c,
1035                 0x9d80, 0x9dec,
1036                 0x9e00, 0x9e6c,
1037                 0x9e80, 0x9eec,
1038                 0x9f00, 0x9f6c,
1039                 0x9f80, 0x9fec,
1040                 0xd004, 0xd03c,
1041                 0xdfc0, 0xdfe0,
1042                 0xe000, 0xea7c,
1043                 0xf000, 0x11190,
1044                 0x19040, 0x19124,
1045                 0x19150, 0x191b0,
1046                 0x191d0, 0x191e8,
1047                 0x19238, 0x1924c,
1048                 0x193f8, 0x19474,
1049                 0x19490, 0x194f8,
1050                 0x19800, 0x19f30,
1051                 0x1a000, 0x1a06c,
1052                 0x1a0b0, 0x1a120,
1053                 0x1a128, 0x1a138,
1054                 0x1a190, 0x1a1c4,
1055                 0x1a1fc, 0x1a1fc,
1056                 0x1e040, 0x1e04c,
1057                 0x1e240, 0x1e28c,
1058                 0x1e2c0, 0x1e2c0,
1059                 0x1e2e0, 0x1e2e0,
1060                 0x1e300, 0x1e384,
1061                 0x1e3c0, 0x1e3c8,
1062                 0x1e440, 0x1e44c,
1063                 0x1e640, 0x1e68c,
1064                 0x1e6c0, 0x1e6c0,
1065                 0x1e6e0, 0x1e6e0,
1066                 0x1e700, 0x1e784,
1067                 0x1e7c0, 0x1e7c8,
1068                 0x1e840, 0x1e84c,
1069                 0x1ea40, 0x1ea8c,
1070                 0x1eac0, 0x1eac0,
1071                 0x1eae0, 0x1eae0,
1072                 0x1eb00, 0x1eb84,
1073                 0x1ebc0, 0x1ebc8,
1074                 0x1ec40, 0x1ec4c,
1075                 0x1ee40, 0x1ee8c,
1076                 0x1eec0, 0x1eec0,
1077                 0x1eee0, 0x1eee0,
1078                 0x1ef00, 0x1ef84,
1079                 0x1efc0, 0x1efc8,
1080                 0x1f040, 0x1f04c,
1081                 0x1f240, 0x1f28c,
1082                 0x1f2c0, 0x1f2c0,
1083                 0x1f2e0, 0x1f2e0,
1084                 0x1f300, 0x1f384,
1085                 0x1f3c0, 0x1f3c8,
1086                 0x1f440, 0x1f44c,
1087                 0x1f640, 0x1f68c,
1088                 0x1f6c0, 0x1f6c0,
1089                 0x1f6e0, 0x1f6e0,
1090                 0x1f700, 0x1f784,
1091                 0x1f7c0, 0x1f7c8,
1092                 0x1f840, 0x1f84c,
1093                 0x1fa40, 0x1fa8c,
1094                 0x1fac0, 0x1fac0,
1095                 0x1fae0, 0x1fae0,
1096                 0x1fb00, 0x1fb84,
1097                 0x1fbc0, 0x1fbc8,
1098                 0x1fc40, 0x1fc4c,
1099                 0x1fe40, 0x1fe8c,
1100                 0x1fec0, 0x1fec0,
1101                 0x1fee0, 0x1fee0,
1102                 0x1ff00, 0x1ff84,
1103                 0x1ffc0, 0x1ffc8,
1104                 0x20000, 0x2002c,
1105                 0x20100, 0x2013c,
1106                 0x20190, 0x201c8,
1107                 0x20200, 0x20318,
1108                 0x20400, 0x20528,
1109                 0x20540, 0x20614,
1110                 0x21000, 0x21040,
1111                 0x2104c, 0x21060,
1112                 0x210c0, 0x210ec,
1113                 0x21200, 0x21268,
1114                 0x21270, 0x21284,
1115                 0x212fc, 0x21388,
1116                 0x21400, 0x21404,
1117                 0x21500, 0x21518,
1118                 0x2152c, 0x2153c,
1119                 0x21550, 0x21554,
1120                 0x21600, 0x21600,
1121                 0x21608, 0x21628,
1122                 0x21630, 0x2163c,
1123                 0x21700, 0x2171c,
1124                 0x21780, 0x2178c,
1125                 0x21800, 0x21c38,
1126                 0x21c80, 0x21d7c,
1127                 0x21e00, 0x21e04,
1128                 0x22000, 0x2202c,
1129                 0x22100, 0x2213c,
1130                 0x22190, 0x221c8,
1131                 0x22200, 0x22318,
1132                 0x22400, 0x22528,
1133                 0x22540, 0x22614,
1134                 0x23000, 0x23040,
1135                 0x2304c, 0x23060,
1136                 0x230c0, 0x230ec,
1137                 0x23200, 0x23268,
1138                 0x23270, 0x23284,
1139                 0x232fc, 0x23388,
1140                 0x23400, 0x23404,
1141                 0x23500, 0x23518,
1142                 0x2352c, 0x2353c,
1143                 0x23550, 0x23554,
1144                 0x23600, 0x23600,
1145                 0x23608, 0x23628,
1146                 0x23630, 0x2363c,
1147                 0x23700, 0x2371c,
1148                 0x23780, 0x2378c,
1149                 0x23800, 0x23c38,
1150                 0x23c80, 0x23d7c,
1151                 0x23e00, 0x23e04,
1152                 0x24000, 0x2402c,
1153                 0x24100, 0x2413c,
1154                 0x24190, 0x241c8,
1155                 0x24200, 0x24318,
1156                 0x24400, 0x24528,
1157                 0x24540, 0x24614,
1158                 0x25000, 0x25040,
1159                 0x2504c, 0x25060,
1160                 0x250c0, 0x250ec,
1161                 0x25200, 0x25268,
1162                 0x25270, 0x25284,
1163                 0x252fc, 0x25388,
1164                 0x25400, 0x25404,
1165                 0x25500, 0x25518,
1166                 0x2552c, 0x2553c,
1167                 0x25550, 0x25554,
1168                 0x25600, 0x25600,
1169                 0x25608, 0x25628,
1170                 0x25630, 0x2563c,
1171                 0x25700, 0x2571c,
1172                 0x25780, 0x2578c,
1173                 0x25800, 0x25c38,
1174                 0x25c80, 0x25d7c,
1175                 0x25e00, 0x25e04,
1176                 0x26000, 0x2602c,
1177                 0x26100, 0x2613c,
1178                 0x26190, 0x261c8,
1179                 0x26200, 0x26318,
1180                 0x26400, 0x26528,
1181                 0x26540, 0x26614,
1182                 0x27000, 0x27040,
1183                 0x2704c, 0x27060,
1184                 0x270c0, 0x270ec,
1185                 0x27200, 0x27268,
1186                 0x27270, 0x27284,
1187                 0x272fc, 0x27388,
1188                 0x27400, 0x27404,
1189                 0x27500, 0x27518,
1190                 0x2752c, 0x2753c,
1191                 0x27550, 0x27554,
1192                 0x27600, 0x27600,
1193                 0x27608, 0x27628,
1194                 0x27630, 0x2763c,
1195                 0x27700, 0x2771c,
1196                 0x27780, 0x2778c,
1197                 0x27800, 0x27c38,
1198                 0x27c80, 0x27d7c,
1199                 0x27e00, 0x27e04
1200         };
1201
1202         int i;
1203         struct adapter *ap = netdev2adap(dev);
1204
1205         regs->version = mk_adap_vers(ap);
1206
1207         memset(buf, 0, T4_REGMAP_SIZE);
1208         for (i = 0; i < ARRAY_SIZE(reg_ranges); i += 2)
1209                 reg_block_dump(ap, buf, reg_ranges[i], reg_ranges[i + 1]);
1210 }
1211
1212 static int restart_autoneg(struct net_device *dev)
1213 {
1214         struct port_info *p = netdev_priv(dev);
1215
1216         if (!netif_running(dev))
1217                 return -EAGAIN;
1218         if (p->link_cfg.autoneg != AUTONEG_ENABLE)
1219                 return -EINVAL;
1220         t4_restart_aneg(p->adapter, 0, p->tx_chan);
1221         return 0;
1222 }
1223
1224 static int identify_port(struct net_device *dev, u32 data)
1225 {
1226         if (data == 0)
1227                 data = 2;     /* default to 2 seconds */
1228
1229         return t4_identify_port(netdev2adap(dev), 0, netdev2pinfo(dev)->viid,
1230                                 data * 5);
1231 }
1232
1233 static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps)
1234 {
1235         unsigned int v = 0;
1236
1237         if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XAUI) {
1238                 v |= SUPPORTED_TP;
1239                 if (caps & FW_PORT_CAP_SPEED_100M)
1240                         v |= SUPPORTED_100baseT_Full;
1241                 if (caps & FW_PORT_CAP_SPEED_1G)
1242                         v |= SUPPORTED_1000baseT_Full;
1243                 if (caps & FW_PORT_CAP_SPEED_10G)
1244                         v |= SUPPORTED_10000baseT_Full;
1245         } else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) {
1246                 v |= SUPPORTED_Backplane;
1247                 if (caps & FW_PORT_CAP_SPEED_1G)
1248                         v |= SUPPORTED_1000baseKX_Full;
1249                 if (caps & FW_PORT_CAP_SPEED_10G)
1250                         v |= SUPPORTED_10000baseKX4_Full;
1251         } else if (type == FW_PORT_TYPE_KR)
1252                 v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full;
1253         else if (type == FW_PORT_TYPE_FIBER)
1254                 v |= SUPPORTED_FIBRE;
1255
1256         if (caps & FW_PORT_CAP_ANEG)
1257                 v |= SUPPORTED_Autoneg;
1258         return v;
1259 }
1260
1261 static unsigned int to_fw_linkcaps(unsigned int caps)
1262 {
1263         unsigned int v = 0;
1264
1265         if (caps & ADVERTISED_100baseT_Full)
1266                 v |= FW_PORT_CAP_SPEED_100M;
1267         if (caps & ADVERTISED_1000baseT_Full)
1268                 v |= FW_PORT_CAP_SPEED_1G;
1269         if (caps & ADVERTISED_10000baseT_Full)
1270                 v |= FW_PORT_CAP_SPEED_10G;
1271         return v;
1272 }
1273
1274 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1275 {
1276         const struct port_info *p = netdev_priv(dev);
1277
1278         if (p->port_type == FW_PORT_TYPE_BT_SGMII ||
1279             p->port_type == FW_PORT_TYPE_BT_XAUI)
1280                 cmd->port = PORT_TP;
1281         else if (p->port_type == FW_PORT_TYPE_FIBER)
1282                 cmd->port = PORT_FIBRE;
1283         else if (p->port_type == FW_PORT_TYPE_TWINAX)
1284                 cmd->port = PORT_DA;
1285         else
1286                 cmd->port = PORT_OTHER;
1287
1288         if (p->mdio_addr >= 0) {
1289                 cmd->phy_address = p->mdio_addr;
1290                 cmd->transceiver = XCVR_EXTERNAL;
1291                 cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ?
1292                         MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45;
1293         } else {
1294                 cmd->phy_address = 0;  /* not really, but no better option */
1295                 cmd->transceiver = XCVR_INTERNAL;
1296                 cmd->mdio_support = 0;
1297         }
1298
1299         cmd->supported = from_fw_linkcaps(p->port_type, p->link_cfg.supported);
1300         cmd->advertising = from_fw_linkcaps(p->port_type,
1301                                             p->link_cfg.advertising);
1302         cmd->speed = netif_carrier_ok(dev) ? p->link_cfg.speed : 0;
1303         cmd->duplex = DUPLEX_FULL;
1304         cmd->autoneg = p->link_cfg.autoneg;
1305         cmd->maxtxpkt = 0;
1306         cmd->maxrxpkt = 0;
1307         return 0;
1308 }
1309
1310 static unsigned int speed_to_caps(int speed)
1311 {
1312         if (speed == SPEED_100)
1313                 return FW_PORT_CAP_SPEED_100M;
1314         if (speed == SPEED_1000)
1315                 return FW_PORT_CAP_SPEED_1G;
1316         if (speed == SPEED_10000)
1317                 return FW_PORT_CAP_SPEED_10G;
1318         return 0;
1319 }
1320
1321 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1322 {
1323         unsigned int cap;
1324         struct port_info *p = netdev_priv(dev);
1325         struct link_config *lc = &p->link_cfg;
1326
1327         if (cmd->duplex != DUPLEX_FULL)     /* only full-duplex supported */
1328                 return -EINVAL;
1329
1330         if (!(lc->supported & FW_PORT_CAP_ANEG)) {
1331                 /*
1332                  * PHY offers a single speed.  See if that's what's
1333                  * being requested.
1334                  */
1335                 if (cmd->autoneg == AUTONEG_DISABLE &&
1336                     (lc->supported & speed_to_caps(cmd->speed)))
1337                                 return 0;
1338                 return -EINVAL;
1339         }
1340
1341         if (cmd->autoneg == AUTONEG_DISABLE) {
1342                 cap = speed_to_caps(cmd->speed);
1343
1344                 if (!(lc->supported & cap) || cmd->speed == SPEED_1000 ||
1345                     cmd->speed == SPEED_10000)
1346                         return -EINVAL;
1347                 lc->requested_speed = cap;
1348                 lc->advertising = 0;
1349         } else {
1350                 cap = to_fw_linkcaps(cmd->advertising);
1351                 if (!(lc->supported & cap))
1352                         return -EINVAL;
1353                 lc->requested_speed = 0;
1354                 lc->advertising = cap | FW_PORT_CAP_ANEG;
1355         }
1356         lc->autoneg = cmd->autoneg;
1357
1358         if (netif_running(dev))
1359                 return t4_link_start(p->adapter, 0, p->tx_chan, lc);
1360         return 0;
1361 }
1362
1363 static void get_pauseparam(struct net_device *dev,
1364                            struct ethtool_pauseparam *epause)
1365 {
1366         struct port_info *p = netdev_priv(dev);
1367
1368         epause->autoneg = (p->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
1369         epause->rx_pause = (p->link_cfg.fc & PAUSE_RX) != 0;
1370         epause->tx_pause = (p->link_cfg.fc & PAUSE_TX) != 0;
1371 }
1372
1373 static int set_pauseparam(struct net_device *dev,
1374                           struct ethtool_pauseparam *epause)
1375 {
1376         struct port_info *p = netdev_priv(dev);
1377         struct link_config *lc = &p->link_cfg;
1378
1379         if (epause->autoneg == AUTONEG_DISABLE)
1380                 lc->requested_fc = 0;
1381         else if (lc->supported & FW_PORT_CAP_ANEG)
1382                 lc->requested_fc = PAUSE_AUTONEG;
1383         else
1384                 return -EINVAL;
1385
1386         if (epause->rx_pause)
1387                 lc->requested_fc |= PAUSE_RX;
1388         if (epause->tx_pause)
1389                 lc->requested_fc |= PAUSE_TX;
1390         if (netif_running(dev))
1391                 return t4_link_start(p->adapter, 0, p->tx_chan, lc);
1392         return 0;
1393 }
1394
1395 static u32 get_rx_csum(struct net_device *dev)
1396 {
1397         struct port_info *p = netdev_priv(dev);
1398
1399         return p->rx_offload & RX_CSO;
1400 }
1401
1402 static int set_rx_csum(struct net_device *dev, u32 data)
1403 {
1404         struct port_info *p = netdev_priv(dev);
1405
1406         if (data)
1407                 p->rx_offload |= RX_CSO;
1408         else
1409                 p->rx_offload &= ~RX_CSO;
1410         return 0;
1411 }
1412
1413 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1414 {
1415         const struct port_info *pi = netdev_priv(dev);
1416         const struct sge *s = &pi->adapter->sge;
1417
1418         e->rx_max_pending = MAX_RX_BUFFERS;
1419         e->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
1420         e->rx_jumbo_max_pending = 0;
1421         e->tx_max_pending = MAX_TXQ_ENTRIES;
1422
1423         e->rx_pending = s->ethrxq[pi->first_qset].fl.size - 8;
1424         e->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
1425         e->rx_jumbo_pending = 0;
1426         e->tx_pending = s->ethtxq[pi->first_qset].q.size;
1427 }
1428
1429 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1430 {
1431         int i;
1432         const struct port_info *pi = netdev_priv(dev);
1433         struct adapter *adapter = pi->adapter;
1434         struct sge *s = &adapter->sge;
1435
1436         if (e->rx_pending > MAX_RX_BUFFERS || e->rx_jumbo_pending ||
1437             e->tx_pending > MAX_TXQ_ENTRIES ||
1438             e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1439             e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1440             e->rx_pending < MIN_FL_ENTRIES || e->tx_pending < MIN_TXQ_ENTRIES)
1441                 return -EINVAL;
1442
1443         if (adapter->flags & FULL_INIT_DONE)
1444                 return -EBUSY;
1445
1446         for (i = 0; i < pi->nqsets; ++i) {
1447                 s->ethtxq[pi->first_qset + i].q.size = e->tx_pending;
1448                 s->ethrxq[pi->first_qset + i].fl.size = e->rx_pending + 8;
1449                 s->ethrxq[pi->first_qset + i].rspq.size = e->rx_mini_pending;
1450         }
1451         return 0;
1452 }
1453
1454 static int closest_timer(const struct sge *s, int time)
1455 {
1456         int i, delta, match = 0, min_delta = INT_MAX;
1457
1458         for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
1459                 delta = time - s->timer_val[i];
1460                 if (delta < 0)
1461                         delta = -delta;
1462                 if (delta < min_delta) {
1463                         min_delta = delta;
1464                         match = i;
1465                 }
1466         }
1467         return match;
1468 }
1469
1470 static int closest_thres(const struct sge *s, int thres)
1471 {
1472         int i, delta, match = 0, min_delta = INT_MAX;
1473
1474         for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
1475                 delta = thres - s->counter_val[i];
1476                 if (delta < 0)
1477                         delta = -delta;
1478                 if (delta < min_delta) {
1479                         min_delta = delta;
1480                         match = i;
1481                 }
1482         }
1483         return match;
1484 }
1485
1486 /*
1487  * Return a queue's interrupt hold-off time in us.  0 means no timer.
1488  */
1489 static unsigned int qtimer_val(const struct adapter *adap,
1490                                const struct sge_rspq *q)
1491 {
1492         unsigned int idx = q->intr_params >> 1;
1493
1494         return idx < SGE_NTIMERS ? adap->sge.timer_val[idx] : 0;
1495 }
1496
1497 /**
1498  *      set_rxq_intr_params - set a queue's interrupt holdoff parameters
1499  *      @adap: the adapter
1500  *      @q: the Rx queue
1501  *      @us: the hold-off time in us, or 0 to disable timer
1502  *      @cnt: the hold-off packet count, or 0 to disable counter
1503  *
1504  *      Sets an Rx queue's interrupt hold-off time and packet count.  At least
1505  *      one of the two needs to be enabled for the queue to generate interrupts.
1506  */
1507 static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
1508                                unsigned int us, unsigned int cnt)
1509 {
1510         if ((us | cnt) == 0)
1511                 cnt = 1;
1512
1513         if (cnt) {
1514                 int err;
1515                 u32 v, new_idx;
1516
1517                 new_idx = closest_thres(&adap->sge, cnt);
1518                 if (q->desc && q->pktcnt_idx != new_idx) {
1519                         /* the queue has already been created, update it */
1520                         v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
1521                             FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1522                             FW_PARAMS_PARAM_YZ(q->cntxt_id);
1523                         err = t4_set_params(adap, 0, 0, 0, 1, &v, &new_idx);
1524                         if (err)
1525                                 return err;
1526                 }
1527                 q->pktcnt_idx = new_idx;
1528         }
1529
1530         us = us == 0 ? 6 : closest_timer(&adap->sge, us);
1531         q->intr_params = QINTR_TIMER_IDX(us) | (cnt > 0 ? QINTR_CNT_EN : 0);
1532         return 0;
1533 }
1534
1535 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1536 {
1537         const struct port_info *pi = netdev_priv(dev);
1538         struct adapter *adap = pi->adapter;
1539
1540         return set_rxq_intr_params(adap, &adap->sge.ethrxq[pi->first_qset].rspq,
1541                         c->rx_coalesce_usecs, c->rx_max_coalesced_frames);
1542 }
1543
1544 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1545 {
1546         const struct port_info *pi = netdev_priv(dev);
1547         const struct adapter *adap = pi->adapter;
1548         const struct sge_rspq *rq = &adap->sge.ethrxq[pi->first_qset].rspq;
1549
1550         c->rx_coalesce_usecs = qtimer_val(adap, rq);
1551         c->rx_max_coalesced_frames = (rq->intr_params & QINTR_CNT_EN) ?
1552                 adap->sge.counter_val[rq->pktcnt_idx] : 0;
1553         return 0;
1554 }
1555
1556 /*
1557  * Translate a physical EEPROM address to virtual.  The first 1K is accessed
1558  * through virtual addresses starting at 31K, the rest is accessed through
1559  * virtual addresses starting at 0.  This mapping is correct only for PF0.
1560  */
1561 static int eeprom_ptov(unsigned int phys_addr)
1562 {
1563         if (phys_addr < 1024)
1564                 return phys_addr + (31 << 10);
1565         if (phys_addr < EEPROMSIZE)
1566                 return phys_addr - 1024;
1567         return -EINVAL;
1568 }
1569
1570 /*
1571  * The next two routines implement eeprom read/write from physical addresses.
1572  * The physical->virtual translation is correct only for PF0.
1573  */
1574 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
1575 {
1576         int vaddr = eeprom_ptov(phys_addr);
1577
1578         if (vaddr >= 0)
1579                 vaddr = pci_read_vpd(adap->pdev, vaddr, sizeof(u32), v);
1580         return vaddr < 0 ? vaddr : 0;
1581 }
1582
1583 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
1584 {
1585         int vaddr = eeprom_ptov(phys_addr);
1586
1587         if (vaddr >= 0)
1588                 vaddr = pci_write_vpd(adap->pdev, vaddr, sizeof(u32), &v);
1589         return vaddr < 0 ? vaddr : 0;
1590 }
1591
1592 #define EEPROM_MAGIC 0x38E2F10C
1593
1594 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
1595                       u8 *data)
1596 {
1597         int i, err = 0;
1598         struct adapter *adapter = netdev2adap(dev);
1599
1600         u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
1601         if (!buf)
1602                 return -ENOMEM;
1603
1604         e->magic = EEPROM_MAGIC;
1605         for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
1606                 err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
1607
1608         if (!err)
1609                 memcpy(data, buf + e->offset, e->len);
1610         kfree(buf);
1611         return err;
1612 }
1613
1614 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
1615                       u8 *data)
1616 {
1617         u8 *buf;
1618         int err = 0;
1619         u32 aligned_offset, aligned_len, *p;
1620         struct adapter *adapter = netdev2adap(dev);
1621
1622         if (eeprom->magic != EEPROM_MAGIC)
1623                 return -EINVAL;
1624
1625         aligned_offset = eeprom->offset & ~3;
1626         aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
1627
1628         if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
1629                 /*
1630                  * RMW possibly needed for first or last words.
1631                  */
1632                 buf = kmalloc(aligned_len, GFP_KERNEL);
1633                 if (!buf)
1634                         return -ENOMEM;
1635                 err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
1636                 if (!err && aligned_len > 4)
1637                         err = eeprom_rd_phys(adapter,
1638                                              aligned_offset + aligned_len - 4,
1639                                              (u32 *)&buf[aligned_len - 4]);
1640                 if (err)
1641                         goto out;
1642                 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
1643         } else
1644                 buf = data;
1645
1646         err = t4_seeprom_wp(adapter, false);
1647         if (err)
1648                 goto out;
1649
1650         for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
1651                 err = eeprom_wr_phys(adapter, aligned_offset, *p);
1652                 aligned_offset += 4;
1653         }
1654
1655         if (!err)
1656                 err = t4_seeprom_wp(adapter, true);
1657 out:
1658         if (buf != data)
1659                 kfree(buf);
1660         return err;
1661 }
1662
1663 static int set_flash(struct net_device *netdev, struct ethtool_flash *ef)
1664 {
1665         int ret;
1666         const struct firmware *fw;
1667         struct adapter *adap = netdev2adap(netdev);
1668
1669         ef->data[sizeof(ef->data) - 1] = '\0';
1670         ret = request_firmware(&fw, ef->data, adap->pdev_dev);
1671         if (ret < 0)
1672                 return ret;
1673
1674         ret = t4_load_fw(adap, fw->data, fw->size);
1675         release_firmware(fw);
1676         if (!ret)
1677                 dev_info(adap->pdev_dev, "loaded firmware %s\n", ef->data);
1678         return ret;
1679 }
1680
1681 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
1682 #define BCAST_CRC 0xa0ccc1a6
1683
1684 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1685 {
1686         wol->supported = WAKE_BCAST | WAKE_MAGIC;
1687         wol->wolopts = netdev2adap(dev)->wol;
1688         memset(&wol->sopass, 0, sizeof(wol->sopass));
1689 }
1690
1691 static int set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1692 {
1693         int err = 0;
1694         struct port_info *pi = netdev_priv(dev);
1695
1696         if (wol->wolopts & ~WOL_SUPPORTED)
1697                 return -EINVAL;
1698         t4_wol_magic_enable(pi->adapter, pi->tx_chan,
1699                             (wol->wolopts & WAKE_MAGIC) ? dev->dev_addr : NULL);
1700         if (wol->wolopts & WAKE_BCAST) {
1701                 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0xfe, ~0ULL,
1702                                         ~0ULL, 0, false);
1703                 if (!err)
1704                         err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 1,
1705                                                 ~6ULL, ~0ULL, BCAST_CRC, true);
1706         } else
1707                 t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0, 0, 0, 0, false);
1708         return err;
1709 }
1710
1711 static int set_tso(struct net_device *dev, u32 value)
1712 {
1713         if (value)
1714                 dev->features |= NETIF_F_TSO | NETIF_F_TSO6;
1715         else
1716                 dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
1717         return 0;
1718 }
1719
1720 static int set_flags(struct net_device *dev, u32 flags)
1721 {
1722         if (flags & ~ETH_FLAG_RXHASH)
1723                 return -EOPNOTSUPP;
1724
1725         if (flags & ETH_FLAG_RXHASH)
1726                 dev->features |= NETIF_F_RXHASH;
1727         else
1728                 dev->features &= ~NETIF_F_RXHASH;
1729         return 0;
1730 }
1731
1732 static struct ethtool_ops cxgb_ethtool_ops = {
1733         .get_settings      = get_settings,
1734         .set_settings      = set_settings,
1735         .get_drvinfo       = get_drvinfo,
1736         .get_msglevel      = get_msglevel,
1737         .set_msglevel      = set_msglevel,
1738         .get_ringparam     = get_sge_param,
1739         .set_ringparam     = set_sge_param,
1740         .get_coalesce      = get_coalesce,
1741         .set_coalesce      = set_coalesce,
1742         .get_eeprom_len    = get_eeprom_len,
1743         .get_eeprom        = get_eeprom,
1744         .set_eeprom        = set_eeprom,
1745         .get_pauseparam    = get_pauseparam,
1746         .set_pauseparam    = set_pauseparam,
1747         .get_rx_csum       = get_rx_csum,
1748         .set_rx_csum       = set_rx_csum,
1749         .set_tx_csum       = ethtool_op_set_tx_ipv6_csum,
1750         .set_sg            = ethtool_op_set_sg,
1751         .get_link          = ethtool_op_get_link,
1752         .get_strings       = get_strings,
1753         .phys_id           = identify_port,
1754         .nway_reset        = restart_autoneg,
1755         .get_sset_count    = get_sset_count,
1756         .get_ethtool_stats = get_stats,
1757         .get_regs_len      = get_regs_len,
1758         .get_regs          = get_regs,
1759         .get_wol           = get_wol,
1760         .set_wol           = set_wol,
1761         .set_tso           = set_tso,
1762         .set_flags         = set_flags,
1763         .flash_device      = set_flash,
1764 };
1765
1766 /*
1767  * debugfs support
1768  */
1769
1770 static int mem_open(struct inode *inode, struct file *file)
1771 {
1772         file->private_data = inode->i_private;
1773         return 0;
1774 }
1775
1776 static ssize_t mem_read(struct file *file, char __user *buf, size_t count,
1777                         loff_t *ppos)
1778 {
1779         loff_t pos = *ppos;
1780         loff_t avail = file->f_path.dentry->d_inode->i_size;
1781         unsigned int mem = (uintptr_t)file->private_data & 3;
1782         struct adapter *adap = file->private_data - mem;
1783
1784         if (pos < 0)
1785                 return -EINVAL;
1786         if (pos >= avail)
1787                 return 0;
1788         if (count > avail - pos)
1789                 count = avail - pos;
1790
1791         while (count) {
1792                 size_t len;
1793                 int ret, ofst;
1794                 __be32 data[16];
1795
1796                 if (mem == MEM_MC)
1797                         ret = t4_mc_read(adap, pos, data, NULL);
1798                 else
1799                         ret = t4_edc_read(adap, mem, pos, data, NULL);
1800                 if (ret)
1801                         return ret;
1802
1803                 ofst = pos % sizeof(data);
1804                 len = min(count, sizeof(data) - ofst);
1805                 if (copy_to_user(buf, (u8 *)data + ofst, len))
1806                         return -EFAULT;
1807
1808                 buf += len;
1809                 pos += len;
1810                 count -= len;
1811         }
1812         count = pos - *ppos;
1813         *ppos = pos;
1814         return count;
1815 }
1816
1817 static const struct file_operations mem_debugfs_fops = {
1818         .owner   = THIS_MODULE,
1819         .open    = mem_open,
1820         .read    = mem_read,
1821 };
1822
1823 static void __devinit add_debugfs_mem(struct adapter *adap, const char *name,
1824                                       unsigned int idx, unsigned int size_mb)
1825 {
1826         struct dentry *de;
1827
1828         de = debugfs_create_file(name, S_IRUSR, adap->debugfs_root,
1829                                  (void *)adap + idx, &mem_debugfs_fops);
1830         if (de && de->d_inode)
1831                 de->d_inode->i_size = size_mb << 20;
1832 }
1833
1834 static int __devinit setup_debugfs(struct adapter *adap)
1835 {
1836         int i;
1837
1838         if (IS_ERR_OR_NULL(adap->debugfs_root))
1839                 return -1;
1840
1841         i = t4_read_reg(adap, MA_TARGET_MEM_ENABLE);
1842         if (i & EDRAM0_ENABLE)
1843                 add_debugfs_mem(adap, "edc0", MEM_EDC0, 5);
1844         if (i & EDRAM1_ENABLE)
1845                 add_debugfs_mem(adap, "edc1", MEM_EDC1, 5);
1846         if (i & EXT_MEM_ENABLE)
1847                 add_debugfs_mem(adap, "mc", MEM_MC,
1848                         EXT_MEM_SIZE_GET(t4_read_reg(adap, MA_EXT_MEMORY_BAR)));
1849         if (adap->l2t)
1850                 debugfs_create_file("l2t", S_IRUSR, adap->debugfs_root, adap,
1851                                     &t4_l2t_fops);
1852         return 0;
1853 }
1854
1855 /*
1856  * upper-layer driver support
1857  */
1858
1859 /*
1860  * Allocate an active-open TID and set it to the supplied value.
1861  */
1862 int cxgb4_alloc_atid(struct tid_info *t, void *data)
1863 {
1864         int atid = -1;
1865
1866         spin_lock_bh(&t->atid_lock);
1867         if (t->afree) {
1868                 union aopen_entry *p = t->afree;
1869
1870                 atid = p - t->atid_tab;
1871                 t->afree = p->next;
1872                 p->data = data;
1873                 t->atids_in_use++;
1874         }
1875         spin_unlock_bh(&t->atid_lock);
1876         return atid;
1877 }
1878 EXPORT_SYMBOL(cxgb4_alloc_atid);
1879
1880 /*
1881  * Release an active-open TID.
1882  */
1883 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
1884 {
1885         union aopen_entry *p = &t->atid_tab[atid];
1886
1887         spin_lock_bh(&t->atid_lock);
1888         p->next = t->afree;
1889         t->afree = p;
1890         t->atids_in_use--;
1891         spin_unlock_bh(&t->atid_lock);
1892 }
1893 EXPORT_SYMBOL(cxgb4_free_atid);
1894
1895 /*
1896  * Allocate a server TID and set it to the supplied value.
1897  */
1898 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
1899 {
1900         int stid;
1901
1902         spin_lock_bh(&t->stid_lock);
1903         if (family == PF_INET) {
1904                 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
1905                 if (stid < t->nstids)
1906                         __set_bit(stid, t->stid_bmap);
1907                 else
1908                         stid = -1;
1909         } else {
1910                 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
1911                 if (stid < 0)
1912                         stid = -1;
1913         }
1914         if (stid >= 0) {
1915                 t->stid_tab[stid].data = data;
1916                 stid += t->stid_base;
1917                 t->stids_in_use++;
1918         }
1919         spin_unlock_bh(&t->stid_lock);
1920         return stid;
1921 }
1922 EXPORT_SYMBOL(cxgb4_alloc_stid);
1923
1924 /*
1925  * Release a server TID.
1926  */
1927 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
1928 {
1929         stid -= t->stid_base;
1930         spin_lock_bh(&t->stid_lock);
1931         if (family == PF_INET)
1932                 __clear_bit(stid, t->stid_bmap);
1933         else
1934                 bitmap_release_region(t->stid_bmap, stid, 2);
1935         t->stid_tab[stid].data = NULL;
1936         t->stids_in_use--;
1937         spin_unlock_bh(&t->stid_lock);
1938 }
1939 EXPORT_SYMBOL(cxgb4_free_stid);
1940
1941 /*
1942  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
1943  */
1944 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
1945                            unsigned int tid)
1946 {
1947         struct cpl_tid_release *req;
1948
1949         set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1950         req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
1951         INIT_TP_WR(req, tid);
1952         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
1953 }
1954
1955 /*
1956  * Queue a TID release request and if necessary schedule a work queue to
1957  * process it.
1958  */
1959 void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
1960                              unsigned int tid)
1961 {
1962         void **p = &t->tid_tab[tid];
1963         struct adapter *adap = container_of(t, struct adapter, tids);
1964
1965         spin_lock_bh(&adap->tid_release_lock);
1966         *p = adap->tid_release_head;
1967         /* Low 2 bits encode the Tx channel number */
1968         adap->tid_release_head = (void **)((uintptr_t)p | chan);
1969         if (!adap->tid_release_task_busy) {
1970                 adap->tid_release_task_busy = true;
1971                 schedule_work(&adap->tid_release_task);
1972         }
1973         spin_unlock_bh(&adap->tid_release_lock);
1974 }
1975 EXPORT_SYMBOL(cxgb4_queue_tid_release);
1976
1977 /*
1978  * Process the list of pending TID release requests.
1979  */
1980 static void process_tid_release_list(struct work_struct *work)
1981 {
1982         struct sk_buff *skb;
1983         struct adapter *adap;
1984
1985         adap = container_of(work, struct adapter, tid_release_task);
1986
1987         spin_lock_bh(&adap->tid_release_lock);
1988         while (adap->tid_release_head) {
1989                 void **p = adap->tid_release_head;
1990                 unsigned int chan = (uintptr_t)p & 3;
1991                 p = (void *)p - chan;
1992
1993                 adap->tid_release_head = *p;
1994                 *p = NULL;
1995                 spin_unlock_bh(&adap->tid_release_lock);
1996
1997                 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
1998                                          GFP_KERNEL)))
1999                         schedule_timeout_uninterruptible(1);
2000
2001                 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
2002                 t4_ofld_send(adap, skb);
2003                 spin_lock_bh(&adap->tid_release_lock);
2004         }
2005         adap->tid_release_task_busy = false;
2006         spin_unlock_bh(&adap->tid_release_lock);
2007 }
2008
2009 /*
2010  * Release a TID and inform HW.  If we are unable to allocate the release
2011  * message we defer to a work queue.
2012  */
2013 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
2014 {
2015         void *old;
2016         struct sk_buff *skb;
2017         struct adapter *adap = container_of(t, struct adapter, tids);
2018
2019         old = t->tid_tab[tid];
2020         skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
2021         if (likely(skb)) {
2022                 t->tid_tab[tid] = NULL;
2023                 mk_tid_release(skb, chan, tid);
2024                 t4_ofld_send(adap, skb);
2025         } else
2026                 cxgb4_queue_tid_release(t, chan, tid);
2027         if (old)
2028                 atomic_dec(&t->tids_in_use);
2029 }
2030 EXPORT_SYMBOL(cxgb4_remove_tid);
2031
2032 /*
2033  * Allocate and initialize the TID tables.  Returns 0 on success.
2034  */
2035 static int tid_init(struct tid_info *t)
2036 {
2037         size_t size;
2038         unsigned int natids = t->natids;
2039
2040         size = t->ntids * sizeof(*t->tid_tab) + natids * sizeof(*t->atid_tab) +
2041                t->nstids * sizeof(*t->stid_tab) +
2042                BITS_TO_LONGS(t->nstids) * sizeof(long);
2043         t->tid_tab = t4_alloc_mem(size);
2044         if (!t->tid_tab)
2045                 return -ENOMEM;
2046
2047         t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
2048         t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
2049         t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids];
2050         spin_lock_init(&t->stid_lock);
2051         spin_lock_init(&t->atid_lock);
2052
2053         t->stids_in_use = 0;
2054         t->afree = NULL;
2055         t->atids_in_use = 0;
2056         atomic_set(&t->tids_in_use, 0);
2057
2058         /* Setup the free list for atid_tab and clear the stid bitmap. */
2059         if (natids) {
2060                 while (--natids)
2061                         t->atid_tab[natids - 1].next = &t->atid_tab[natids];
2062                 t->afree = t->atid_tab;
2063         }
2064         bitmap_zero(t->stid_bmap, t->nstids);
2065         return 0;
2066 }
2067
2068 /**
2069  *      cxgb4_create_server - create an IP server
2070  *      @dev: the device
2071  *      @stid: the server TID
2072  *      @sip: local IP address to bind server to
2073  *      @sport: the server's TCP port
2074  *      @queue: queue to direct messages from this server to
2075  *
2076  *      Create an IP server for the given port and address.
2077  *      Returns <0 on error and one of the %NET_XMIT_* values on success.
2078  */
2079 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
2080                         __be32 sip, __be16 sport, unsigned int queue)
2081 {
2082         unsigned int chan;
2083         struct sk_buff *skb;
2084         struct adapter *adap;
2085         struct cpl_pass_open_req *req;
2086
2087         skb = alloc_skb(sizeof(*req), GFP_KERNEL);
2088         if (!skb)
2089                 return -ENOMEM;
2090
2091         adap = netdev2adap(dev);
2092         req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
2093         INIT_TP_WR(req, 0);
2094         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
2095         req->local_port = sport;
2096         req->peer_port = htons(0);
2097         req->local_ip = sip;
2098         req->peer_ip = htonl(0);
2099         chan = netdev2pinfo(adap->sge.ingr_map[queue]->netdev)->tx_chan;
2100         req->opt0 = cpu_to_be64(TX_CHAN(chan));
2101         req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
2102                                 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
2103         return t4_mgmt_tx(adap, skb);
2104 }
2105 EXPORT_SYMBOL(cxgb4_create_server);
2106
2107 /**
2108  *      cxgb4_create_server6 - create an IPv6 server
2109  *      @dev: the device
2110  *      @stid: the server TID
2111  *      @sip: local IPv6 address to bind server to
2112  *      @sport: the server's TCP port
2113  *      @queue: queue to direct messages from this server to
2114  *
2115  *      Create an IPv6 server for the given port and address.
2116  *      Returns <0 on error and one of the %NET_XMIT_* values on success.
2117  */
2118 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
2119                          const struct in6_addr *sip, __be16 sport,
2120                          unsigned int queue)
2121 {
2122         unsigned int chan;
2123         struct sk_buff *skb;
2124         struct adapter *adap;
2125         struct cpl_pass_open_req6 *req;
2126
2127         skb = alloc_skb(sizeof(*req), GFP_KERNEL);
2128         if (!skb)
2129                 return -ENOMEM;
2130
2131         adap = netdev2adap(dev);
2132         req = (struct cpl_pass_open_req6 *)__skb_put(skb, sizeof(*req));
2133         INIT_TP_WR(req, 0);
2134         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
2135         req->local_port = sport;
2136         req->peer_port = htons(0);
2137         req->local_ip_hi = *(__be64 *)(sip->s6_addr);
2138         req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
2139         req->peer_ip_hi = cpu_to_be64(0);
2140         req->peer_ip_lo = cpu_to_be64(0);
2141         chan = netdev2pinfo(adap->sge.ingr_map[queue]->netdev)->tx_chan;
2142         req->opt0 = cpu_to_be64(TX_CHAN(chan));
2143         req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
2144                                 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
2145         return t4_mgmt_tx(adap, skb);
2146 }
2147 EXPORT_SYMBOL(cxgb4_create_server6);
2148
2149 /**
2150  *      cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
2151  *      @mtus: the HW MTU table
2152  *      @mtu: the target MTU
2153  *      @idx: index of selected entry in the MTU table
2154  *
2155  *      Returns the index and the value in the HW MTU table that is closest to
2156  *      but does not exceed @mtu, unless @mtu is smaller than any value in the
2157  *      table, in which case that smallest available value is selected.
2158  */
2159 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
2160                             unsigned int *idx)
2161 {
2162         unsigned int i = 0;
2163
2164         while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
2165                 ++i;
2166         if (idx)
2167                 *idx = i;
2168         return mtus[i];
2169 }
2170 EXPORT_SYMBOL(cxgb4_best_mtu);
2171
2172 /**
2173  *      cxgb4_port_chan - get the HW channel of a port
2174  *      @dev: the net device for the port
2175  *
2176  *      Return the HW Tx channel of the given port.
2177  */
2178 unsigned int cxgb4_port_chan(const struct net_device *dev)
2179 {
2180         return netdev2pinfo(dev)->tx_chan;
2181 }
2182 EXPORT_SYMBOL(cxgb4_port_chan);
2183
2184 /**
2185  *      cxgb4_port_viid - get the VI id of a port
2186  *      @dev: the net device for the port
2187  *
2188  *      Return the VI id of the given port.
2189  */
2190 unsigned int cxgb4_port_viid(const struct net_device *dev)
2191 {
2192         return netdev2pinfo(dev)->viid;
2193 }
2194 EXPORT_SYMBOL(cxgb4_port_viid);
2195
2196 /**
2197  *      cxgb4_port_idx - get the index of a port
2198  *      @dev: the net device for the port
2199  *
2200  *      Return the index of the given port.
2201  */
2202 unsigned int cxgb4_port_idx(const struct net_device *dev)
2203 {
2204         return netdev2pinfo(dev)->port_id;
2205 }
2206 EXPORT_SYMBOL(cxgb4_port_idx);
2207
2208 /**
2209  *      cxgb4_netdev_by_hwid - return the net device of a HW port
2210  *      @pdev: identifies the adapter
2211  *      @id: the HW port id
2212  *
2213  *      Return the net device associated with the interface with the given HW
2214  *      id.
2215  */
2216 struct net_device *cxgb4_netdev_by_hwid(struct pci_dev *pdev, unsigned int id)
2217 {
2218         const struct adapter *adap = pci_get_drvdata(pdev);
2219
2220         if (!adap || id >= NCHAN)
2221                 return NULL;
2222         id = adap->chan_map[id];
2223         return id < MAX_NPORTS ? adap->port[id] : NULL;
2224 }
2225 EXPORT_SYMBOL(cxgb4_netdev_by_hwid);
2226
2227 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
2228                          struct tp_tcp_stats *v6)
2229 {
2230         struct adapter *adap = pci_get_drvdata(pdev);
2231
2232         spin_lock(&adap->stats_lock);
2233         t4_tp_get_tcp_stats(adap, v4, v6);
2234         spin_unlock(&adap->stats_lock);
2235 }
2236 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
2237
2238 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
2239                       const unsigned int *pgsz_order)
2240 {
2241         struct adapter *adap = netdev2adap(dev);
2242
2243         t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK, tag_mask);
2244         t4_write_reg(adap, ULP_RX_ISCSI_PSZ, HPZ0(pgsz_order[0]) |
2245                      HPZ1(pgsz_order[1]) | HPZ2(pgsz_order[2]) |
2246                      HPZ3(pgsz_order[3]));
2247 }
2248 EXPORT_SYMBOL(cxgb4_iscsi_init);
2249
2250 static struct pci_driver cxgb4_driver;
2251
2252 static void check_neigh_update(struct neighbour *neigh)
2253 {
2254         const struct device *parent;
2255         const struct net_device *netdev = neigh->dev;
2256
2257         if (netdev->priv_flags & IFF_802_1Q_VLAN)
2258                 netdev = vlan_dev_real_dev(netdev);
2259         parent = netdev->dev.parent;
2260         if (parent && parent->driver == &cxgb4_driver.driver)
2261                 t4_l2t_update(dev_get_drvdata(parent), neigh);
2262 }
2263
2264 static int netevent_cb(struct notifier_block *nb, unsigned long event,
2265                        void *data)
2266 {
2267         switch (event) {
2268         case NETEVENT_NEIGH_UPDATE:
2269                 check_neigh_update(data);
2270                 break;
2271         case NETEVENT_PMTU_UPDATE:
2272         case NETEVENT_REDIRECT:
2273         default:
2274                 break;
2275         }
2276         return 0;
2277 }
2278
2279 static bool netevent_registered;
2280 static struct notifier_block cxgb4_netevent_nb = {
2281         .notifier_call = netevent_cb
2282 };
2283
2284 static void uld_attach(struct adapter *adap, unsigned int uld)
2285 {
2286         void *handle;
2287         struct cxgb4_lld_info lli;
2288
2289         lli.pdev = adap->pdev;
2290         lli.l2t = adap->l2t;
2291         lli.tids = &adap->tids;
2292         lli.ports = adap->port;
2293         lli.vr = &adap->vres;
2294         lli.mtus = adap->params.mtus;
2295         if (uld == CXGB4_ULD_RDMA) {
2296                 lli.rxq_ids = adap->sge.rdma_rxq;
2297                 lli.nrxq = adap->sge.rdmaqs;
2298         } else if (uld == CXGB4_ULD_ISCSI) {
2299                 lli.rxq_ids = adap->sge.ofld_rxq;
2300                 lli.nrxq = adap->sge.ofldqsets;
2301         }
2302         lli.ntxq = adap->sge.ofldqsets;
2303         lli.nchan = adap->params.nports;
2304         lli.nports = adap->params.nports;
2305         lli.wr_cred = adap->params.ofldq_wr_cred;
2306         lli.adapter_type = adap->params.rev;
2307         lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
2308         lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
2309                         t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF));
2310         lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
2311                         t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF));
2312         lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS);
2313         lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL);
2314         lli.fw_vers = adap->params.fw_vers;
2315
2316         handle = ulds[uld].add(&lli);
2317         if (IS_ERR(handle)) {
2318                 dev_warn(adap->pdev_dev,
2319                          "could not attach to the %s driver, error %ld\n",
2320                          uld_str[uld], PTR_ERR(handle));
2321                 return;
2322         }
2323
2324         adap->uld_handle[uld] = handle;
2325
2326         if (!netevent_registered) {
2327                 register_netevent_notifier(&cxgb4_netevent_nb);
2328                 netevent_registered = true;
2329         }
2330 }
2331
2332 static void attach_ulds(struct adapter *adap)
2333 {
2334         unsigned int i;
2335
2336         mutex_lock(&uld_mutex);
2337         list_add_tail(&adap->list_node, &adapter_list);
2338         for (i = 0; i < CXGB4_ULD_MAX; i++)
2339                 if (ulds[i].add)
2340                         uld_attach(adap, i);
2341         mutex_unlock(&uld_mutex);
2342 }
2343
2344 static void detach_ulds(struct adapter *adap)
2345 {
2346         unsigned int i;
2347
2348         mutex_lock(&uld_mutex);
2349         list_del(&adap->list_node);
2350         for (i = 0; i < CXGB4_ULD_MAX; i++)
2351                 if (adap->uld_handle[i]) {
2352                         ulds[i].state_change(adap->uld_handle[i],
2353                                              CXGB4_STATE_DETACH);
2354                         adap->uld_handle[i] = NULL;
2355                 }
2356         if (netevent_registered && list_empty(&adapter_list)) {
2357                 unregister_netevent_notifier(&cxgb4_netevent_nb);
2358                 netevent_registered = false;
2359         }
2360         mutex_unlock(&uld_mutex);
2361 }
2362
2363 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2364 {
2365         unsigned int i;
2366
2367         mutex_lock(&uld_mutex);
2368         for (i = 0; i < CXGB4_ULD_MAX; i++)
2369                 if (adap->uld_handle[i])
2370                         ulds[i].state_change(adap->uld_handle[i], new_state);
2371         mutex_unlock(&uld_mutex);
2372 }
2373
2374 /**
2375  *      cxgb4_register_uld - register an upper-layer driver
2376  *      @type: the ULD type
2377  *      @p: the ULD methods
2378  *
2379  *      Registers an upper-layer driver with this driver and notifies the ULD
2380  *      about any presently available devices that support its type.  Returns
2381  *      %-EBUSY if a ULD of the same type is already registered.
2382  */
2383 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
2384 {
2385         int ret = 0;
2386         struct adapter *adap;
2387
2388         if (type >= CXGB4_ULD_MAX)
2389                 return -EINVAL;
2390         mutex_lock(&uld_mutex);
2391         if (ulds[type].add) {
2392                 ret = -EBUSY;
2393                 goto out;
2394         }
2395         ulds[type] = *p;
2396         list_for_each_entry(adap, &adapter_list, list_node)
2397                 uld_attach(adap, type);
2398 out:    mutex_unlock(&uld_mutex);
2399         return ret;
2400 }
2401 EXPORT_SYMBOL(cxgb4_register_uld);
2402
2403 /**
2404  *      cxgb4_unregister_uld - unregister an upper-layer driver
2405  *      @type: the ULD type
2406  *
2407  *      Unregisters an existing upper-layer driver.
2408  */
2409 int cxgb4_unregister_uld(enum cxgb4_uld type)
2410 {
2411         struct adapter *adap;
2412
2413         if (type >= CXGB4_ULD_MAX)
2414                 return -EINVAL;
2415         mutex_lock(&uld_mutex);
2416         list_for_each_entry(adap, &adapter_list, list_node)
2417                 adap->uld_handle[type] = NULL;
2418         ulds[type].add = NULL;
2419         mutex_unlock(&uld_mutex);
2420         return 0;
2421 }
2422 EXPORT_SYMBOL(cxgb4_unregister_uld);
2423
2424 /**
2425  *      cxgb_up - enable the adapter
2426  *      @adap: adapter being enabled
2427  *
2428  *      Called when the first port is enabled, this function performs the
2429  *      actions necessary to make an adapter operational, such as completing
2430  *      the initialization of HW modules, and enabling interrupts.
2431  *
2432  *      Must be called with the rtnl lock held.
2433  */
2434 static int cxgb_up(struct adapter *adap)
2435 {
2436         int err;
2437
2438         err = setup_sge_queues(adap);
2439         if (err)
2440                 goto out;
2441         err = setup_rss(adap);
2442         if (err)
2443                 goto freeq;
2444
2445         if (adap->flags & USING_MSIX) {
2446                 name_msix_vecs(adap);
2447                 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2448                                   adap->msix_info[0].desc, adap);
2449                 if (err)
2450                         goto irq_err;
2451
2452                 err = request_msix_queue_irqs(adap);
2453                 if (err) {
2454                         free_irq(adap->msix_info[0].vec, adap);
2455                         goto irq_err;
2456                 }
2457         } else {
2458                 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2459                                   (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
2460                                   adap->name, adap);
2461                 if (err)
2462                         goto irq_err;
2463         }
2464         enable_rx(adap);
2465         t4_sge_start(adap);
2466         t4_intr_enable(adap);
2467         adap->flags |= FULL_INIT_DONE;
2468         notify_ulds(adap, CXGB4_STATE_UP);
2469  out:
2470         return err;
2471  irq_err:
2472         dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2473  freeq:
2474         t4_free_sge_resources(adap);
2475         goto out;
2476 }
2477
2478 static void cxgb_down(struct adapter *adapter)
2479 {
2480         t4_intr_disable(adapter);
2481         cancel_work_sync(&adapter->tid_release_task);
2482         adapter->tid_release_task_busy = false;
2483
2484         if (adapter->flags & USING_MSIX) {
2485                 free_msix_queue_irqs(adapter);
2486                 free_irq(adapter->msix_info[0].vec, adapter);
2487         } else
2488                 free_irq(adapter->pdev->irq, adapter);
2489         quiesce_rx(adapter);
2490         t4_sge_stop(adapter);
2491         t4_free_sge_resources(adapter);
2492         adapter->flags &= ~FULL_INIT_DONE;
2493 }
2494
2495 /*
2496  * net_device operations
2497  */
2498 static int cxgb_open(struct net_device *dev)
2499 {
2500         int err;
2501         struct port_info *pi = netdev_priv(dev);
2502         struct adapter *adapter = pi->adapter;
2503
2504         if (!(adapter->flags & FULL_INIT_DONE)) {
2505                 err = cxgb_up(adapter);
2506                 if (err < 0)
2507                         return err;
2508         }
2509
2510         dev->real_num_tx_queues = pi->nqsets;
2511         link_start(dev);
2512         netif_tx_start_all_queues(dev);
2513         return 0;
2514 }
2515
2516 static int cxgb_close(struct net_device *dev)
2517 {
2518         struct port_info *pi = netdev_priv(dev);
2519         struct adapter *adapter = pi->adapter;
2520
2521         netif_tx_stop_all_queues(dev);
2522         netif_carrier_off(dev);
2523         return t4_enable_vi(adapter, 0, pi->viid, false, false);
2524 }
2525
2526 static struct net_device_stats *cxgb_get_stats(struct net_device *dev)
2527 {
2528         struct port_stats stats;
2529         struct port_info *p = netdev_priv(dev);
2530         struct adapter *adapter = p->adapter;
2531         struct net_device_stats *ns = &dev->stats;
2532
2533         spin_lock(&adapter->stats_lock);
2534         t4_get_port_stats(adapter, p->tx_chan, &stats);
2535         spin_unlock(&adapter->stats_lock);
2536
2537         ns->tx_bytes   = stats.tx_octets;
2538         ns->tx_packets = stats.tx_frames;
2539         ns->rx_bytes   = stats.rx_octets;
2540         ns->rx_packets = stats.rx_frames;
2541         ns->multicast  = stats.rx_mcast_frames;
2542
2543         /* detailed rx_errors */
2544         ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2545                                stats.rx_runt;
2546         ns->rx_over_errors   = 0;
2547         ns->rx_crc_errors    = stats.rx_fcs_err;
2548         ns->rx_frame_errors  = stats.rx_symbol_err;
2549         ns->rx_fifo_errors   = stats.rx_ovflow0 + stats.rx_ovflow1 +
2550                                stats.rx_ovflow2 + stats.rx_ovflow3 +
2551                                stats.rx_trunc0 + stats.rx_trunc1 +
2552                                stats.rx_trunc2 + stats.rx_trunc3;
2553         ns->rx_missed_errors = 0;
2554
2555         /* detailed tx_errors */
2556         ns->tx_aborted_errors   = 0;
2557         ns->tx_carrier_errors   = 0;
2558         ns->tx_fifo_errors      = 0;
2559         ns->tx_heartbeat_errors = 0;
2560         ns->tx_window_errors    = 0;
2561
2562         ns->tx_errors = stats.tx_error_frames;
2563         ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2564                 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2565         return ns;
2566 }
2567
2568 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2569 {
2570         int ret = 0, prtad, devad;
2571         struct port_info *pi = netdev_priv(dev);
2572         struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2573
2574         switch (cmd) {
2575         case SIOCGMIIPHY:
2576                 if (pi->mdio_addr < 0)
2577                         return -EOPNOTSUPP;
2578                 data->phy_id = pi->mdio_addr;
2579                 break;
2580         case SIOCGMIIREG:
2581         case SIOCSMIIREG:
2582                 if (mdio_phy_id_is_c45(data->phy_id)) {
2583                         prtad = mdio_phy_id_prtad(data->phy_id);
2584                         devad = mdio_phy_id_devad(data->phy_id);
2585                 } else if (data->phy_id < 32) {
2586                         prtad = data->phy_id;
2587                         devad = 0;
2588                         data->reg_num &= 0x1f;
2589                 } else
2590                         return -EINVAL;
2591
2592                 if (cmd == SIOCGMIIREG)
2593                         ret = t4_mdio_rd(pi->adapter, 0, prtad, devad,
2594                                          data->reg_num, &data->val_out);
2595                 else
2596                         ret = t4_mdio_wr(pi->adapter, 0, prtad, devad,
2597                                          data->reg_num, data->val_in);
2598                 break;
2599         default:
2600                 return -EOPNOTSUPP;
2601         }
2602         return ret;
2603 }
2604
2605 static void cxgb_set_rxmode(struct net_device *dev)
2606 {
2607         /* unfortunately we can't return errors to the stack */
2608         set_rxmode(dev, -1, false);
2609 }
2610
2611 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2612 {
2613         int ret;
2614         struct port_info *pi = netdev_priv(dev);
2615
2616         if (new_mtu < 81 || new_mtu > MAX_MTU)         /* accommodate SACK */
2617                 return -EINVAL;
2618         ret = t4_set_rxmode(pi->adapter, 0, pi->viid, new_mtu, -1, -1, -1, -1,
2619                             true);
2620         if (!ret)
2621                 dev->mtu = new_mtu;
2622         return ret;
2623 }
2624
2625 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2626 {
2627         int ret;
2628         struct sockaddr *addr = p;
2629         struct port_info *pi = netdev_priv(dev);
2630
2631         if (!is_valid_ether_addr(addr->sa_data))
2632                 return -EINVAL;
2633
2634         ret = t4_change_mac(pi->adapter, 0, pi->viid, pi->xact_addr_filt,
2635                             addr->sa_data, true, true);
2636         if (ret < 0)
2637                 return ret;
2638
2639         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2640         pi->xact_addr_filt = ret;
2641         return 0;
2642 }
2643
2644 static void vlan_rx_register(struct net_device *dev, struct vlan_group *grp)
2645 {
2646         struct port_info *pi = netdev_priv(dev);
2647
2648         pi->vlan_grp = grp;
2649         t4_set_rxmode(pi->adapter, 0, pi->viid, -1, -1, -1, -1, grp != NULL,
2650                       true);
2651 }
2652
2653 #ifdef CONFIG_NET_POLL_CONTROLLER
2654 static void cxgb_netpoll(struct net_device *dev)
2655 {
2656         struct port_info *pi = netdev_priv(dev);
2657         struct adapter *adap = pi->adapter;
2658
2659         if (adap->flags & USING_MSIX) {
2660                 int i;
2661                 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
2662
2663                 for (i = pi->nqsets; i; i--, rx++)
2664                         t4_sge_intr_msix(0, &rx->rspq);
2665         } else
2666                 t4_intr_handler(adap)(0, adap);
2667 }
2668 #endif
2669
2670 static const struct net_device_ops cxgb4_netdev_ops = {
2671         .ndo_open             = cxgb_open,
2672         .ndo_stop             = cxgb_close,
2673         .ndo_start_xmit       = t4_eth_xmit,
2674         .ndo_get_stats        = cxgb_get_stats,
2675         .ndo_set_rx_mode      = cxgb_set_rxmode,
2676         .ndo_set_mac_address  = cxgb_set_mac_addr,
2677         .ndo_validate_addr    = eth_validate_addr,
2678         .ndo_do_ioctl         = cxgb_ioctl,
2679         .ndo_change_mtu       = cxgb_change_mtu,
2680         .ndo_vlan_rx_register = vlan_rx_register,
2681 #ifdef CONFIG_NET_POLL_CONTROLLER
2682         .ndo_poll_controller  = cxgb_netpoll,
2683 #endif
2684 };
2685
2686 void t4_fatal_err(struct adapter *adap)
2687 {
2688         t4_set_reg_field(adap, SGE_CONTROL, GLOBALENABLE, 0);
2689         t4_intr_disable(adap);
2690         dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
2691 }
2692
2693 static void setup_memwin(struct adapter *adap)
2694 {
2695         u32 bar0;
2696
2697         bar0 = pci_resource_start(adap->pdev, 0);  /* truncation intentional */
2698         t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 0),
2699                      (bar0 + MEMWIN0_BASE) | BIR(0) |
2700                      WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
2701         t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 1),
2702                      (bar0 + MEMWIN1_BASE) | BIR(0) |
2703                      WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
2704         t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2),
2705                      (bar0 + MEMWIN2_BASE) | BIR(0) |
2706                      WINDOW(ilog2(MEMWIN2_APERTURE) - 10));
2707 }
2708
2709 /*
2710  * Max # of ATIDs.  The absolute HW max is 16K but we keep it lower.
2711  */
2712 #define MAX_ATIDS 8192U
2713
2714 /*
2715  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
2716  */
2717 static int adap_init0(struct adapter *adap)
2718 {
2719         int ret;
2720         u32 v, port_vec;
2721         enum dev_state state;
2722         u32 params[7], val[7];
2723         struct fw_caps_config_cmd c;
2724
2725         ret = t4_check_fw_version(adap);
2726         if (ret == -EINVAL || ret > 0) {
2727                 if (upgrade_fw(adap) >= 0)             /* recache FW version */
2728                         ret = t4_check_fw_version(adap);
2729         }
2730         if (ret < 0)
2731                 return ret;
2732
2733         /* contact FW, request master */
2734         ret = t4_fw_hello(adap, 0, 0, MASTER_MUST, &state);
2735         if (ret < 0) {
2736                 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
2737                         ret);
2738                 return ret;
2739         }
2740
2741         /* reset device */
2742         ret = t4_fw_reset(adap, 0, PIORSTMODE | PIORST);
2743         if (ret < 0)
2744                 goto bye;
2745
2746         /* get device capabilities */
2747         memset(&c, 0, sizeof(c));
2748         c.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2749                               FW_CMD_REQUEST | FW_CMD_READ);
2750         c.retval_len16 = htonl(FW_LEN16(c));
2751         ret = t4_wr_mbox(adap, 0, &c, sizeof(c), &c);
2752         if (ret < 0)
2753                 goto bye;
2754
2755         /* select capabilities we'll be using */
2756         if (c.niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
2757                 if (!vf_acls)
2758                         c.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
2759                 else
2760                         c.niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
2761         } else if (vf_acls) {
2762                 dev_err(adap->pdev_dev, "virtualization ACLs not supported");
2763                 goto bye;
2764         }
2765         c.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2766                               FW_CMD_REQUEST | FW_CMD_WRITE);
2767         ret = t4_wr_mbox(adap, 0, &c, sizeof(c), NULL);
2768         if (ret < 0)
2769                 goto bye;
2770
2771         ret = t4_config_glbl_rss(adap, 0,
2772                                  FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
2773                                  FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
2774                                  FW_RSS_GLB_CONFIG_CMD_TNLALLLKP);
2775         if (ret < 0)
2776                 goto bye;
2777
2778         ret = t4_cfg_pfvf(adap, 0, 0, 0, 64, 64, 64, 0, 0, 4, 0xf, 0xf, 16,
2779                           FW_CMD_CAP_PF, FW_CMD_CAP_PF);
2780         if (ret < 0)
2781                 goto bye;
2782
2783         for (v = 0; v < SGE_NTIMERS - 1; v++)
2784                 adap->sge.timer_val[v] = min(intr_holdoff[v], MAX_SGE_TIMERVAL);
2785         adap->sge.timer_val[SGE_NTIMERS - 1] = MAX_SGE_TIMERVAL;
2786         adap->sge.counter_val[0] = 1;
2787         for (v = 1; v < SGE_NCOUNTERS; v++)
2788                 adap->sge.counter_val[v] = min(intr_cnt[v - 1],
2789                                                THRESHOLD_3_MASK);
2790         t4_sge_init(adap);
2791
2792         /* get basic stuff going */
2793         ret = t4_early_init(adap, 0);
2794         if (ret < 0)
2795                 goto bye;
2796
2797 #define FW_PARAM_DEV(param) \
2798         (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
2799          FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
2800
2801 #define FW_PARAM_PFVF(param) \
2802         (FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
2803          FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
2804
2805         params[0] = FW_PARAM_DEV(PORTVEC);
2806         params[1] = FW_PARAM_PFVF(L2T_START);
2807         params[2] = FW_PARAM_PFVF(L2T_END);
2808         params[3] = FW_PARAM_PFVF(FILTER_START);
2809         params[4] = FW_PARAM_PFVF(FILTER_END);
2810         ret = t4_query_params(adap, 0, 0, 0, 5, params, val);
2811         if (ret < 0)
2812                 goto bye;
2813         port_vec = val[0];
2814         adap->tids.ftid_base = val[3];
2815         adap->tids.nftids = val[4] - val[3] + 1;
2816
2817         if (c.ofldcaps) {
2818                 /* query offload-related parameters */
2819                 params[0] = FW_PARAM_DEV(NTID);
2820                 params[1] = FW_PARAM_PFVF(SERVER_START);
2821                 params[2] = FW_PARAM_PFVF(SERVER_END);
2822                 params[3] = FW_PARAM_PFVF(TDDP_START);
2823                 params[4] = FW_PARAM_PFVF(TDDP_END);
2824                 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
2825                 ret = t4_query_params(adap, 0, 0, 0, 6, params, val);
2826                 if (ret < 0)
2827                         goto bye;
2828                 adap->tids.ntids = val[0];
2829                 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
2830                 adap->tids.stid_base = val[1];
2831                 adap->tids.nstids = val[2] - val[1] + 1;
2832                 adap->vres.ddp.start = val[3];
2833                 adap->vres.ddp.size = val[4] - val[3] + 1;
2834                 adap->params.ofldq_wr_cred = val[5];
2835                 adap->params.offload = 1;
2836         }
2837         if (c.rdmacaps) {
2838                 params[0] = FW_PARAM_PFVF(STAG_START);
2839                 params[1] = FW_PARAM_PFVF(STAG_END);
2840                 params[2] = FW_PARAM_PFVF(RQ_START);
2841                 params[3] = FW_PARAM_PFVF(RQ_END);
2842                 params[4] = FW_PARAM_PFVF(PBL_START);
2843                 params[5] = FW_PARAM_PFVF(PBL_END);
2844                 ret = t4_query_params(adap, 0, 0, 0, 6, params, val);
2845                 if (ret < 0)
2846                         goto bye;
2847                 adap->vres.stag.start = val[0];
2848                 adap->vres.stag.size = val[1] - val[0] + 1;
2849                 adap->vres.rq.start = val[2];
2850                 adap->vres.rq.size = val[3] - val[2] + 1;
2851                 adap->vres.pbl.start = val[4];
2852                 adap->vres.pbl.size = val[5] - val[4] + 1;
2853         }
2854         if (c.iscsicaps) {
2855                 params[0] = FW_PARAM_PFVF(ISCSI_START);
2856                 params[1] = FW_PARAM_PFVF(ISCSI_END);
2857                 ret = t4_query_params(adap, 0, 0, 0, 2, params, val);
2858                 if (ret < 0)
2859                         goto bye;
2860                 adap->vres.iscsi.start = val[0];
2861                 adap->vres.iscsi.size = val[1] - val[0] + 1;
2862         }
2863 #undef FW_PARAM_PFVF
2864 #undef FW_PARAM_DEV
2865
2866         adap->params.nports = hweight32(port_vec);
2867         adap->params.portvec = port_vec;
2868         adap->flags |= FW_OK;
2869
2870         /* These are finalized by FW initialization, load their values now */
2871         v = t4_read_reg(adap, TP_TIMER_RESOLUTION);
2872         adap->params.tp.tre = TIMERRESOLUTION_GET(v);
2873         t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
2874         t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
2875                      adap->params.b_wnd);
2876
2877         /* tweak some settings */
2878         t4_write_reg(adap, TP_SHIFT_CNT, 0x64f8849);
2879         t4_write_reg(adap, ULP_RX_TDDP_PSZ, HPZ0(PAGE_SHIFT - 12));
2880         t4_write_reg(adap, TP_PIO_ADDR, TP_INGRESS_CONFIG);
2881         v = t4_read_reg(adap, TP_PIO_DATA);
2882         t4_write_reg(adap, TP_PIO_DATA, v & ~CSUM_HAS_PSEUDO_HDR);
2883         setup_memwin(adap);
2884         return 0;
2885
2886         /*
2887          * If a command timed out or failed with EIO FW does not operate within
2888          * its spec or something catastrophic happened to HW/FW, stop issuing
2889          * commands.
2890          */
2891 bye:    if (ret != -ETIMEDOUT && ret != -EIO)
2892                 t4_fw_bye(adap, 0);
2893         return ret;
2894 }
2895
2896 static inline bool is_10g_port(const struct link_config *lc)
2897 {
2898         return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0;
2899 }
2900
2901 static inline void init_rspq(struct sge_rspq *q, u8 timer_idx, u8 pkt_cnt_idx,
2902                              unsigned int size, unsigned int iqe_size)
2903 {
2904         q->intr_params = QINTR_TIMER_IDX(timer_idx) |
2905                          (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0);
2906         q->pktcnt_idx = pkt_cnt_idx < SGE_NCOUNTERS ? pkt_cnt_idx : 0;
2907         q->iqe_len = iqe_size;
2908         q->size = size;
2909 }
2910
2911 /*
2912  * Perform default configuration of DMA queues depending on the number and type
2913  * of ports we found and the number of available CPUs.  Most settings can be
2914  * modified by the admin prior to actual use.
2915  */
2916 static void __devinit cfg_queues(struct adapter *adap)
2917 {
2918         struct sge *s = &adap->sge;
2919         int i, q10g = 0, n10g = 0, qidx = 0;
2920
2921         for_each_port(adap, i)
2922                 n10g += is_10g_port(&adap2pinfo(adap, i)->link_cfg);
2923
2924         /*
2925          * We default to 1 queue per non-10G port and up to # of cores queues
2926          * per 10G port.
2927          */
2928         if (n10g)
2929                 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
2930         if (q10g > num_online_cpus())
2931                 q10g = num_online_cpus();
2932
2933         for_each_port(adap, i) {
2934                 struct port_info *pi = adap2pinfo(adap, i);
2935
2936                 pi->first_qset = qidx;
2937                 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
2938                 qidx += pi->nqsets;
2939         }
2940
2941         s->ethqsets = qidx;
2942         s->max_ethqsets = qidx;   /* MSI-X may lower it later */
2943
2944         if (is_offload(adap)) {
2945                 /*
2946                  * For offload we use 1 queue/channel if all ports are up to 1G,
2947                  * otherwise we divide all available queues amongst the channels
2948                  * capped by the number of available cores.
2949                  */
2950                 if (n10g) {
2951                         i = min_t(int, ARRAY_SIZE(s->ofldrxq),
2952                                   num_online_cpus());
2953                         s->ofldqsets = roundup(i, adap->params.nports);
2954                 } else
2955                         s->ofldqsets = adap->params.nports;
2956                 /* For RDMA one Rx queue per channel suffices */
2957                 s->rdmaqs = adap->params.nports;
2958         }
2959
2960         for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
2961                 struct sge_eth_rxq *r = &s->ethrxq[i];
2962
2963                 init_rspq(&r->rspq, 0, 0, 1024, 64);
2964                 r->fl.size = 72;
2965         }
2966
2967         for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
2968                 s->ethtxq[i].q.size = 1024;
2969
2970         for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
2971                 s->ctrlq[i].q.size = 512;
2972
2973         for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
2974                 s->ofldtxq[i].q.size = 1024;
2975
2976         for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
2977                 struct sge_ofld_rxq *r = &s->ofldrxq[i];
2978
2979                 init_rspq(&r->rspq, 0, 0, 1024, 64);
2980                 r->rspq.uld = CXGB4_ULD_ISCSI;
2981                 r->fl.size = 72;
2982         }
2983
2984         for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
2985                 struct sge_ofld_rxq *r = &s->rdmarxq[i];
2986
2987                 init_rspq(&r->rspq, 0, 0, 511, 64);
2988                 r->rspq.uld = CXGB4_ULD_RDMA;
2989                 r->fl.size = 72;
2990         }
2991
2992         init_rspq(&s->fw_evtq, 6, 0, 512, 64);
2993         init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
2994 }
2995
2996 /*
2997  * Reduce the number of Ethernet queues across all ports to at most n.
2998  * n provides at least one queue per port.
2999  */
3000 static void __devinit reduce_ethqs(struct adapter *adap, int n)
3001 {
3002         int i;
3003         struct port_info *pi;
3004
3005         while (n < adap->sge.ethqsets)
3006                 for_each_port(adap, i) {
3007                         pi = adap2pinfo(adap, i);
3008                         if (pi->nqsets > 1) {
3009                                 pi->nqsets--;
3010                                 adap->sge.ethqsets--;
3011                                 if (adap->sge.ethqsets <= n)
3012                                         break;
3013                         }
3014                 }
3015
3016         n = 0;
3017         for_each_port(adap, i) {
3018                 pi = adap2pinfo(adap, i);
3019                 pi->first_qset = n;
3020                 n += pi->nqsets;
3021         }
3022 }
3023
3024 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
3025 #define EXTRA_VECS 2
3026
3027 static int __devinit enable_msix(struct adapter *adap)
3028 {
3029         int ofld_need = 0;
3030         int i, err, want, need;
3031         struct sge *s = &adap->sge;
3032         unsigned int nchan = adap->params.nports;
3033         struct msix_entry entries[MAX_INGQ + 1];
3034
3035         for (i = 0; i < ARRAY_SIZE(entries); ++i)
3036                 entries[i].entry = i;
3037
3038         want = s->max_ethqsets + EXTRA_VECS;
3039         if (is_offload(adap)) {
3040                 want += s->rdmaqs + s->ofldqsets;
3041                 /* need nchan for each possible ULD */
3042                 ofld_need = 2 * nchan;
3043         }
3044         need = adap->params.nports + EXTRA_VECS + ofld_need;
3045
3046         while ((err = pci_enable_msix(adap->pdev, entries, want)) >= need)
3047                 want = err;
3048
3049         if (!err) {
3050                 /*
3051                  * Distribute available vectors to the various queue groups.
3052                  * Every group gets its minimum requirement and NIC gets top
3053                  * priority for leftovers.
3054                  */
3055                 i = want - EXTRA_VECS - ofld_need;
3056                 if (i < s->max_ethqsets) {
3057                         s->max_ethqsets = i;
3058                         if (i < s->ethqsets)
3059                                 reduce_ethqs(adap, i);
3060                 }
3061                 if (is_offload(adap)) {
3062                         i = want - EXTRA_VECS - s->max_ethqsets;
3063                         i -= ofld_need - nchan;
3064                         s->ofldqsets = (i / nchan) * nchan;  /* round down */
3065                 }
3066                 for (i = 0; i < want; ++i)
3067                         adap->msix_info[i].vec = entries[i].vector;
3068         } else if (err > 0)
3069                 dev_info(adap->pdev_dev,
3070                          "only %d MSI-X vectors left, not using MSI-X\n", err);
3071         return err;
3072 }
3073
3074 #undef EXTRA_VECS
3075
3076 static void __devinit print_port_info(struct adapter *adap)
3077 {
3078         static const char *base[] = {
3079                 "R", "KX4", "T", "KX", "T", "KR", "CX4"
3080         };
3081
3082         int i;
3083         char buf[80];
3084         const char *spd = "";
3085
3086         if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
3087                 spd = " 2.5 GT/s";
3088         else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
3089                 spd = " 5 GT/s";
3090
3091         for_each_port(adap, i) {
3092                 struct net_device *dev = adap->port[i];
3093                 const struct port_info *pi = netdev_priv(dev);
3094                 char *bufp = buf;
3095
3096                 if (!test_bit(i, &adap->registered_device_map))
3097                         continue;
3098
3099                 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
3100                         bufp += sprintf(bufp, "100/");
3101                 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
3102                         bufp += sprintf(bufp, "1000/");
3103                 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
3104                         bufp += sprintf(bufp, "10G/");
3105                 if (bufp != buf)
3106                         --bufp;
3107                 sprintf(bufp, "BASE-%s", base[pi->port_type]);
3108
3109                 netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
3110                             adap->params.vpd.id, adap->params.rev,
3111                             buf, is_offload(adap) ? "R" : "",
3112                             adap->params.pci.width, spd,
3113                             (adap->flags & USING_MSIX) ? " MSI-X" :
3114                             (adap->flags & USING_MSI) ? " MSI" : "");
3115                 if (adap->name == dev->name)
3116                         netdev_info(dev, "S/N: %s, E/C: %s\n",
3117                                     adap->params.vpd.sn, adap->params.vpd.ec);
3118         }
3119 }
3120
3121 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO | NETIF_F_TSO6 |\
3122                    NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
3123
3124 static int __devinit init_one(struct pci_dev *pdev,
3125                               const struct pci_device_id *ent)
3126 {
3127         int func, i, err;
3128         struct port_info *pi;
3129         unsigned int highdma = 0;
3130         struct adapter *adapter = NULL;
3131
3132         printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
3133
3134         err = pci_request_regions(pdev, KBUILD_MODNAME);
3135         if (err) {
3136                 /* Just info, some other driver may have claimed the device. */
3137                 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
3138                 return err;
3139         }
3140
3141         /* We control everything through PF 0 */
3142         func = PCI_FUNC(pdev->devfn);
3143         if (func > 0)
3144                 goto sriov;
3145
3146         err = pci_enable_device(pdev);
3147         if (err) {
3148                 dev_err(&pdev->dev, "cannot enable PCI device\n");
3149                 goto out_release_regions;
3150         }
3151
3152         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
3153                 highdma = NETIF_F_HIGHDMA;
3154                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3155                 if (err) {
3156                         dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
3157                                 "coherent allocations\n");
3158                         goto out_disable_device;
3159                 }
3160         } else {
3161                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3162                 if (err) {
3163                         dev_err(&pdev->dev, "no usable DMA configuration\n");
3164                         goto out_disable_device;
3165                 }
3166         }
3167
3168         pci_enable_pcie_error_reporting(pdev);
3169         pci_set_master(pdev);
3170         pci_save_state(pdev);
3171
3172         adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
3173         if (!adapter) {
3174                 err = -ENOMEM;
3175                 goto out_disable_device;
3176         }
3177
3178         adapter->regs = pci_ioremap_bar(pdev, 0);
3179         if (!adapter->regs) {
3180                 dev_err(&pdev->dev, "cannot map device registers\n");
3181                 err = -ENOMEM;
3182                 goto out_free_adapter;
3183         }
3184
3185         adapter->pdev = pdev;
3186         adapter->pdev_dev = &pdev->dev;
3187         adapter->name = pci_name(pdev);
3188         adapter->msg_enable = dflt_msg_enable;
3189         memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
3190
3191         spin_lock_init(&adapter->stats_lock);
3192         spin_lock_init(&adapter->tid_release_lock);
3193
3194         INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
3195
3196         err = t4_prep_adapter(adapter);
3197         if (err)
3198                 goto out_unmap_bar;
3199         err = adap_init0(adapter);
3200         if (err)
3201                 goto out_unmap_bar;
3202
3203         for_each_port(adapter, i) {
3204                 struct net_device *netdev;
3205
3206                 netdev = alloc_etherdev_mq(sizeof(struct port_info),
3207                                            MAX_ETH_QSETS);
3208                 if (!netdev) {
3209                         err = -ENOMEM;
3210                         goto out_free_dev;
3211                 }
3212
3213                 SET_NETDEV_DEV(netdev, &pdev->dev);
3214
3215                 adapter->port[i] = netdev;
3216                 pi = netdev_priv(netdev);
3217                 pi->adapter = adapter;
3218                 pi->xact_addr_filt = -1;
3219                 pi->rx_offload = RX_CSO;
3220                 pi->port_id = i;
3221                 netif_carrier_off(netdev);
3222                 netif_tx_stop_all_queues(netdev);
3223                 netdev->irq = pdev->irq;
3224
3225                 netdev->features |= NETIF_F_SG | NETIF_F_TSO | NETIF_F_TSO6;
3226                 netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3227                 netdev->features |= NETIF_F_GRO | NETIF_F_RXHASH | highdma;
3228                 netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
3229                 netdev->vlan_features = netdev->features & VLAN_FEAT;
3230
3231                 netdev->netdev_ops = &cxgb4_netdev_ops;
3232                 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
3233         }
3234
3235         pci_set_drvdata(pdev, adapter);
3236
3237         if (adapter->flags & FW_OK) {
3238                 err = t4_port_init(adapter, 0, 0, 0);
3239                 if (err)
3240                         goto out_free_dev;
3241         }
3242
3243         /*
3244          * Configure queues and allocate tables now, they can be needed as
3245          * soon as the first register_netdev completes.
3246          */
3247         cfg_queues(adapter);
3248
3249         adapter->l2t = t4_init_l2t();
3250         if (!adapter->l2t) {
3251                 /* We tolerate a lack of L2T, giving up some functionality */
3252                 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
3253                 adapter->params.offload = 0;
3254         }
3255
3256         if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
3257                 dev_warn(&pdev->dev, "could not allocate TID table, "
3258                          "continuing\n");
3259                 adapter->params.offload = 0;
3260         }
3261
3262         /*
3263          * The card is now ready to go.  If any errors occur during device
3264          * registration we do not fail the whole card but rather proceed only
3265          * with the ports we manage to register successfully.  However we must
3266          * register at least one net device.
3267          */
3268         for_each_port(adapter, i) {
3269                 err = register_netdev(adapter->port[i]);
3270                 if (err)
3271                         dev_warn(&pdev->dev,
3272                                  "cannot register net device %s, skipping\n",
3273                                  adapter->port[i]->name);
3274                 else {
3275                         /*
3276                          * Change the name we use for messages to the name of
3277                          * the first successfully registered interface.
3278                          */
3279                         if (!adapter->registered_device_map)
3280                                 adapter->name = adapter->port[i]->name;
3281
3282                         __set_bit(i, &adapter->registered_device_map);
3283                         adapter->chan_map[adap2pinfo(adapter, i)->tx_chan] = i;
3284                 }
3285         }
3286         if (!adapter->registered_device_map) {
3287                 dev_err(&pdev->dev, "could not register any net devices\n");
3288                 goto out_free_dev;
3289         }
3290
3291         if (cxgb4_debugfs_root) {
3292                 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
3293                                                            cxgb4_debugfs_root);
3294                 setup_debugfs(adapter);
3295         }
3296
3297         /* See what interrupts we'll be using */
3298         if (msi > 1 && enable_msix(adapter) == 0)
3299                 adapter->flags |= USING_MSIX;
3300         else if (msi > 0 && pci_enable_msi(pdev) == 0)
3301                 adapter->flags |= USING_MSI;
3302
3303         if (is_offload(adapter))
3304                 attach_ulds(adapter);
3305
3306         print_port_info(adapter);
3307
3308 sriov:
3309 #ifdef CONFIG_PCI_IOV
3310         if (func < ARRAY_SIZE(num_vf) && num_vf[func] > 0)
3311                 if (pci_enable_sriov(pdev, num_vf[func]) == 0)
3312                         dev_info(&pdev->dev,
3313                                  "instantiated %u virtual functions\n",
3314                                  num_vf[func]);
3315 #endif
3316         return 0;
3317
3318  out_free_dev:
3319         t4_free_mem(adapter->tids.tid_tab);
3320         t4_free_mem(adapter->l2t);
3321         for_each_port(adapter, i)
3322                 if (adapter->port[i])
3323                         free_netdev(adapter->port[i]);
3324         if (adapter->flags & FW_OK)
3325                 t4_fw_bye(adapter, 0);
3326  out_unmap_bar:
3327         iounmap(adapter->regs);
3328  out_free_adapter:
3329         kfree(adapter);
3330  out_disable_device:
3331         pci_disable_pcie_error_reporting(pdev);
3332         pci_disable_device(pdev);
3333  out_release_regions:
3334         pci_release_regions(pdev);
3335         pci_set_drvdata(pdev, NULL);
3336         return err;
3337 }
3338
3339 static void __devexit remove_one(struct pci_dev *pdev)
3340 {
3341         struct adapter *adapter = pci_get_drvdata(pdev);
3342
3343         pci_disable_sriov(pdev);
3344
3345         if (adapter) {
3346                 int i;
3347
3348                 if (is_offload(adapter))
3349                         detach_ulds(adapter);
3350
3351                 for_each_port(adapter, i)
3352                         if (test_bit(i, &adapter->registered_device_map))
3353                                 unregister_netdev(adapter->port[i]);
3354
3355                 if (adapter->debugfs_root)
3356                         debugfs_remove_recursive(adapter->debugfs_root);
3357
3358                 if (adapter->flags & FULL_INIT_DONE)
3359                         cxgb_down(adapter);
3360                 t4_free_mem(adapter->l2t);
3361                 t4_free_mem(adapter->tids.tid_tab);
3362                 disable_msi(adapter);
3363
3364                 for_each_port(adapter, i)
3365                         if (adapter->port[i])
3366                                 free_netdev(adapter->port[i]);
3367
3368                 if (adapter->flags & FW_OK)
3369                         t4_fw_bye(adapter, 0);
3370                 iounmap(adapter->regs);
3371                 kfree(adapter);
3372                 pci_disable_pcie_error_reporting(pdev);
3373                 pci_disable_device(pdev);
3374                 pci_release_regions(pdev);
3375                 pci_set_drvdata(pdev, NULL);
3376         } else if (PCI_FUNC(pdev->devfn) > 0)
3377                 pci_release_regions(pdev);
3378 }
3379
3380 static struct pci_driver cxgb4_driver = {
3381         .name     = KBUILD_MODNAME,
3382         .id_table = cxgb4_pci_tbl,
3383         .probe    = init_one,
3384         .remove   = __devexit_p(remove_one),
3385 };
3386
3387 static int __init cxgb4_init_module(void)
3388 {
3389         int ret;
3390
3391         /* Debugfs support is optional, just warn if this fails */
3392         cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
3393         if (!cxgb4_debugfs_root)
3394                 pr_warning("could not create debugfs entry, continuing\n");
3395
3396         ret = pci_register_driver(&cxgb4_driver);
3397         if (ret < 0)
3398                 debugfs_remove(cxgb4_debugfs_root);
3399         return ret;
3400 }
3401
3402 static void __exit cxgb4_cleanup_module(void)
3403 {
3404         pci_unregister_driver(&cxgb4_driver);
3405         debugfs_remove(cxgb4_debugfs_root);  /* NULL ok */
3406 }
3407
3408 module_init(cxgb4_init_module);
3409 module_exit(cxgb4_cleanup_module);