drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c

   1 /*
   2  * This file is part of the Chelsio T4 Ethernet driver for Linux.
   3  *
   4  * Copyright (c) 2003-2016 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.h>
  45 #include <linux/if_vlan.h>
  46 #include <linux/init.h>
  47 #include <linux/log2.h>
  48 #include <linux/mdio.h>
  49 #include <linux/module.h>
  50 #include <linux/moduleparam.h>
  51 #include <linux/mutex.h>
  52 #include <linux/netdevice.h>
  53 #include <linux/pci.h>
  54 #include <linux/aer.h>
  55 #include <linux/rtnetlink.h>
  56 #include <linux/sched.h>
  57 #include <linux/seq_file.h>
  58 #include <linux/sockios.h>
  59 #include <linux/vmalloc.h>
  60 #include <linux/workqueue.h>
  61 #include <net/neighbour.h>
  62 #include <net/netevent.h>
  63 #include <net/addrconf.h>
  64 #include <net/bonding.h>
  65 #include <net/addrconf.h>
  66 #include <linux/uaccess.h>
  67 #include <linux/crash_dump.h>
  68
  69 #include "cxgb4.h"
  70 #include "cxgb4_filter.h"
  71 #include "t4_regs.h"
  72 #include "t4_values.h"
  73 #include "t4_msg.h"
  74 #include "t4fw_api.h"
  75 #include "t4fw_version.h"
  76 #include "cxgb4_dcb.h"
  77 #include "cxgb4_debugfs.h"
  78 #include "clip_tbl.h"
  79 #include "l2t.h"
  80 #include "sched.h"
  81 #include "cxgb4_tc_u32.h"
  82
  83 char cxgb4_driver_name[] = KBUILD_MODNAME;
  84
  85 #ifdef DRV_VERSION
  86 #undef DRV_VERSION
  87 #endif
  88 #define DRV_VERSION "2.0.0-ko"
  89 const char cxgb4_driver_version[] = DRV_VERSION;
  90 #define DRV_DESC "Chelsio T4/T5/T6 Network Driver"
  91
  92 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
  93                          NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
  94                          NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
  95
  96 /* Macros needed to support the PCI Device ID Table ...
  97  */
  98 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
  99         static const struct pci_device_id cxgb4_pci_tbl[] = {
 100 #define CH_PCI_DEVICE_ID_FUNCTION 0x4
 101
 102 /* Include PCI Device IDs for both PF4 and PF0-3 so our PCI probe() routine is
 103  * called for both.
 104  */
 105 #define CH_PCI_DEVICE_ID_FUNCTION2 0x0
 106
 107 #define CH_PCI_ID_TABLE_ENTRY(devid) \
 108                 {PCI_VDEVICE(CHELSIO, (devid)), 4}
 109
 110 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
 111                 { 0, } \
 112         }
 113
 114 #include "t4_pci_id_tbl.h"
 115
 116 #define FW4_FNAME "cxgb4/t4fw.bin"
 117 #define FW5_FNAME "cxgb4/t5fw.bin"
 118 #define FW6_FNAME "cxgb4/t6fw.bin"
 119 #define FW4_CFNAME "cxgb4/t4-config.txt"
 120 #define FW5_CFNAME "cxgb4/t5-config.txt"
 121 #define FW6_CFNAME "cxgb4/t6-config.txt"
 122 #define PHY_AQ1202_FIRMWARE "cxgb4/aq1202_fw.cld"
 123 #define PHY_BCM84834_FIRMWARE "cxgb4/bcm8483.bin"
 124 #define PHY_AQ1202_DEVICEID 0x4409
 125 #define PHY_BCM84834_DEVICEID 0x4486
 126
 127 MODULE_DESCRIPTION(DRV_DESC);
 128 MODULE_AUTHOR("Chelsio Communications");
 129 MODULE_LICENSE("Dual BSD/GPL");
 130 MODULE_VERSION(DRV_VERSION);
 131 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
 132 MODULE_FIRMWARE(FW4_FNAME);
 133 MODULE_FIRMWARE(FW5_FNAME);
 134 MODULE_FIRMWARE(FW6_FNAME);
 135
 136 /*
 137  * The driver uses the best interrupt scheme available on a platform in the
 138  * order MSI-X, MSI, legacy INTx interrupts.  This parameter determines which
 139  * of these schemes the driver may consider as follows:
 140  *
 141  * msi = 2: choose from among all three options
 142  * msi = 1: only consider MSI and INTx interrupts
 143  * msi = 0: force INTx interrupts
 144  */
 145 static int msi = 2;
 146
 147 module_param(msi, int, 0644);
 148 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
 149
 150 /*
 151  * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
 152  * offset by 2 bytes in order to have the IP headers line up on 4-byte
 153  * boundaries.  This is a requirement for many architectures which will throw
 154  * a machine check fault if an attempt is made to access one of the 4-byte IP
 155  * header fields on a non-4-byte boundary.  And it's a major performance issue
 156  * even on some architectures which allow it like some implementations of the
 157  * x86 ISA.  However, some architectures don't mind this and for some very
 158  * edge-case performance sensitive applications (like forwarding large volumes
 159  * of small packets), setting this DMA offset to 0 will decrease the number of
 160  * PCI-E Bus transfers enough to measurably affect performance.
 161  */
 162 static int rx_dma_offset = 2;
 163
 164 /* TX Queue select used to determine what algorithm to use for selecting TX
 165  * queue. Select between the kernel provided function (select_queue=0) or user
 166  * cxgb_select_queue function (select_queue=1)
 167  *
 168  * Default: select_queue=0
 169  */
 170 static int select_queue;
 171 module_param(select_queue, int, 0644);
 172 MODULE_PARM_DESC(select_queue,
 173                  "Select between kernel provided method of selecting or driver method of selecting TX queue. Default is kernel method.");
 174
 175 static struct dentry *cxgb4_debugfs_root;
 176
 177 LIST_HEAD(adapter_list);
 178 DEFINE_MUTEX(uld_mutex);
 179
 180 static void link_report(struct net_device *dev)
 181 {
 182         if (!netif_carrier_ok(dev))
 183                 netdev_info(dev, "link down\n");
 184         else {
 185                 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
 186
 187                 const char *s;
 188                 const struct port_info *p = netdev_priv(dev);
 189
 190                 switch (p->link_cfg.speed) {
 191                 case 100:
 192                         s = "100Mbps";
 193                         break;
 194                 case 1000:
 195                         s = "1Gbps";
 196                         break;
 197                 case 10000:
 198                         s = "10Gbps";
 199                         break;
 200                 case 25000:
 201                         s = "25Gbps";
 202                         break;
 203                 case 40000:
 204                         s = "40Gbps";
 205                         break;
 206                 case 100000:
 207                         s = "100Gbps";
 208                         break;
 209                 default:
 210                         pr_info("%s: unsupported speed: %d\n",
 211                                 dev->name, p->link_cfg.speed);
 212                         return;
 213                 }
 214
 215                 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
 216                             fc[p->link_cfg.fc]);
 217         }
 218 }
 219
 220 #ifdef CONFIG_CHELSIO_T4_DCB
 221 /* Set up/tear down Data Center Bridging Priority mapping for a net device. */
 222 static void dcb_tx_queue_prio_enable(struct net_device *dev, int enable)
 223 {
 224         struct port_info *pi = netdev_priv(dev);
 225         struct adapter *adap = pi->adapter;
 226         struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
 227         int i;
 228
 229         /* We use a simple mapping of Port TX Queue Index to DCB
 230          * Priority when we're enabling DCB.
 231          */
 232         for (i = 0; i < pi->nqsets; i++, txq++) {
 233                 u32 name, value;
 234                 int err;
 235
 236                 name = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
 237                         FW_PARAMS_PARAM_X_V(
 238                                 FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH) |
 239                         FW_PARAMS_PARAM_YZ_V(txq->q.cntxt_id));
 240                 value = enable ? i : 0xffffffff;
 241
 242                 /* Since we can be called while atomic (from "interrupt
 243                  * level") we need to issue the Set Parameters Commannd
 244                  * without sleeping (timeout < 0).
 245                  */
 246                 err = t4_set_params_timeout(adap, adap->mbox, adap->pf, 0, 1,
 247                                             &name, &value,
 248                                             -FW_CMD_MAX_TIMEOUT);
 249
 250                 if (err)
 251                         dev_err(adap->pdev_dev,
 252                                 "Can't %s DCB Priority on port %d, TX Queue %d: err=%d\n",
 253                                 enable ? "set" : "unset", pi->port_id, i, -err);
 254                 else
 255                         txq->dcb_prio = value;
 256         }
 257 }
 258
 259 static int cxgb4_dcb_enabled(const struct net_device *dev)
 260 {
 261         struct port_info *pi = netdev_priv(dev);
 262
 263         if (!pi->dcb.enabled)
 264                 return 0;
 265
 266         return ((pi->dcb.state == CXGB4_DCB_STATE_FW_ALLSYNCED) ||
 267                 (pi->dcb.state == CXGB4_DCB_STATE_HOST));
 268 }
 269 #endif /* CONFIG_CHELSIO_T4_DCB */
 270
 271 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
 272 {
 273         struct net_device *dev = adapter->port[port_id];
 274
 275         /* Skip changes from disabled ports. */
 276         if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
 277                 if (link_stat)
 278                         netif_carrier_on(dev);
 279                 else {
 280 #ifdef CONFIG_CHELSIO_T4_DCB
 281                         if (cxgb4_dcb_enabled(dev)) {
 282                                 cxgb4_dcb_state_init(dev);
 283                                 dcb_tx_queue_prio_enable(dev, false);
 284                         }
 285 #endif /* CONFIG_CHELSIO_T4_DCB */
 286                         netif_carrier_off(dev);
 287                 }
 288
 289                 link_report(dev);
 290         }
 291 }
 292
 293 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
 294 {
 295         static const char *mod_str[] = {
 296                 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
 297         };
 298
 299         const struct net_device *dev = adap->port[port_id];
 300         const struct port_info *pi = netdev_priv(dev);
 301
 302         if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
 303                 netdev_info(dev, "port module unplugged\n");
 304         else if (pi->mod_type < ARRAY_SIZE(mod_str))
 305                 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
 306         else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
 307                 netdev_info(dev, "%s: unsupported port module inserted\n",
 308                             dev->name);
 309         else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
 310                 netdev_info(dev, "%s: unknown port module inserted\n",
 311                             dev->name);
 312         else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
 313                 netdev_info(dev, "%s: transceiver module error\n", dev->name);
 314         else
 315                 netdev_info(dev, "%s: unknown module type %d inserted\n",
 316                             dev->name, pi->mod_type);
 317 }
 318
 319 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
 320 module_param(dbfifo_int_thresh, int, 0644);
 321 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
 322
 323 /*
 324  * usecs to sleep while draining the dbfifo
 325  */
 326 static int dbfifo_drain_delay = 1000;
 327 module_param(dbfifo_drain_delay, int, 0644);
 328 MODULE_PARM_DESC(dbfifo_drain_delay,
 329                  "usecs to sleep while draining the dbfifo");
 330
 331 static inline int cxgb4_set_addr_hash(struct port_info *pi)
 332 {
 333         struct adapter *adap = pi->adapter;
 334         u64 vec = 0;
 335         bool ucast = false;
 336         struct hash_mac_addr *entry;
 337
 338         /* Calculate the hash vector for the updated list and program it */
 339         list_for_each_entry(entry, &adap->mac_hlist, list) {
 340                 ucast |= is_unicast_ether_addr(entry->addr);
 341                 vec |= (1ULL << hash_mac_addr(entry->addr));
 342         }
 343         return t4_set_addr_hash(adap, adap->mbox, pi->viid, ucast,
 344                                 vec, false);
 345 }
 346
 347 static int cxgb4_mac_sync(struct net_device *netdev, const u8 *mac_addr)
 348 {
 349         struct port_info *pi = netdev_priv(netdev);
 350         struct adapter *adap = pi->adapter;
 351         int ret;
 352         u64 mhash = 0;
 353         u64 uhash = 0;
 354         bool free = false;
 355         bool ucast = is_unicast_ether_addr(mac_addr);
 356         const u8 *maclist[1] = {mac_addr};
 357         struct hash_mac_addr *new_entry;
 358
 359         ret = t4_alloc_mac_filt(adap, adap->mbox, pi->viid, free, 1, maclist,
 360                                 NULL, ucast ? &uhash : &mhash, false);
 361         if (ret < 0)
 362                 goto out;
 363         /* if hash != 0, then add the addr to hash addr list
 364          * so on the end we will calculate the hash for the
 365          * list and program it
 366          */
 367         if (uhash || mhash) {
 368                 new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
 369                 if (!new_entry)
 370                         return -ENOMEM;
 371                 ether_addr_copy(new_entry->addr, mac_addr);
 372                 list_add_tail(&new_entry->list, &adap->mac_hlist);
 373                 ret = cxgb4_set_addr_hash(pi);
 374         }
 375 out:
 376         return ret < 0 ? ret : 0;
 377 }
 378
 379 static int cxgb4_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
 380 {
 381         struct port_info *pi = netdev_priv(netdev);
 382         struct adapter *adap = pi->adapter;
 383         int ret;
 384         const u8 *maclist[1] = {mac_addr};
 385         struct hash_mac_addr *entry, *tmp;
 386
 387         /* If the MAC address to be removed is in the hash addr
 388          * list, delete it from the list and update hash vector
 389          */
 390         list_for_each_entry_safe(entry, tmp, &adap->mac_hlist, list) {
 391                 if (ether_addr_equal(entry->addr, mac_addr)) {
 392                         list_del(&entry->list);
 393                         kfree(entry);
 394                         return cxgb4_set_addr_hash(pi);
 395                 }
 396         }
 397
 398         ret = t4_free_mac_filt(adap, adap->mbox, pi->viid, 1, maclist, false);
 399         return ret < 0 ? -EINVAL : 0;
 400 }
 401
 402 /*
 403  * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
 404  * If @mtu is -1 it is left unchanged.
 405  */
 406 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
 407 {
 408         struct port_info *pi = netdev_priv(dev);
 409         struct adapter *adapter = pi->adapter;
 410
 411         __dev_uc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
 412         __dev_mc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
 413
 414         return t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu,
 415                              (dev->flags & IFF_PROMISC) ? 1 : 0,
 416                              (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
 417                              sleep_ok);
 418 }
 419
 420 /**
 421  *      link_start - enable a port
 422  *      @dev: the port to enable
 423  *
 424  *      Performs the MAC and PHY actions needed to enable a port.
 425  */
 426 static int link_start(struct net_device *dev)
 427 {
 428         int ret;
 429         struct port_info *pi = netdev_priv(dev);
 430         unsigned int mb = pi->adapter->pf;
 431
 432         /*
 433          * We do not set address filters and promiscuity here, the stack does
 434          * that step explicitly.
 435          */
 436         ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
 437                             !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
 438         if (ret == 0) {
 439                 ret = t4_change_mac(pi->adapter, mb, pi->viid,
 440                                     pi->xact_addr_filt, dev->dev_addr, true,
 441                                     true);
 442                 if (ret >= 0) {
 443                         pi->xact_addr_filt = ret;
 444                         ret = 0;
 445                 }
 446         }
 447         if (ret == 0)
 448                 ret = t4_link_l1cfg(pi->adapter, mb, pi->tx_chan,
 449                                     &pi->link_cfg);
 450         if (ret == 0) {
 451                 local_bh_disable();
 452                 ret = t4_enable_vi_params(pi->adapter, mb, pi->viid, true,
 453                                           true, CXGB4_DCB_ENABLED);
 454                 local_bh_enable();
 455         }
 456
 457         return ret;
 458 }
 459
 460 #ifdef CONFIG_CHELSIO_T4_DCB
 461 /* Handle a Data Center Bridging update message from the firmware. */
 462 static void dcb_rpl(struct adapter *adap, const struct fw_port_cmd *pcmd)
 463 {
 464         int port = FW_PORT_CMD_PORTID_G(ntohl(pcmd->op_to_portid));
 465         struct net_device *dev = adap->port[adap->chan_map[port]];
 466         int old_dcb_enabled = cxgb4_dcb_enabled(dev);
 467         int new_dcb_enabled;
 468
 469         cxgb4_dcb_handle_fw_update(adap, pcmd);
 470         new_dcb_enabled = cxgb4_dcb_enabled(dev);
 471
 472         /* If the DCB has become enabled or disabled on the port then we're
 473          * going to need to set up/tear down DCB Priority parameters for the
 474          * TX Queues associated with the port.
 475          */
 476         if (new_dcb_enabled != old_dcb_enabled)
 477                 dcb_tx_queue_prio_enable(dev, new_dcb_enabled);
 478 }
 479 #endif /* CONFIG_CHELSIO_T4_DCB */
 480
 481 /* Response queue handler for the FW event queue.
 482  */
 483 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
 484                           const struct pkt_gl *gl)
 485 {
 486         u8 opcode = ((const struct rss_header *)rsp)->opcode;
 487
 488         rsp++;                                          /* skip RSS header */
 489
 490         /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
 491          */
 492         if (unlikely(opcode == CPL_FW4_MSG &&
 493            ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
 494                 rsp++;
 495                 opcode = ((const struct rss_header *)rsp)->opcode;
 496                 rsp++;
 497                 if (opcode != CPL_SGE_EGR_UPDATE) {
 498                         dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
 499                                 , opcode);
 500                         goto out;
 501                 }
 502         }
 503
 504         if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
 505                 const struct cpl_sge_egr_update *p = (void *)rsp;
 506                 unsigned int qid = EGR_QID_G(ntohl(p->opcode_qid));
 507                 struct sge_txq *txq;
 508
 509                 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
 510                 txq->restarts++;
 511                 if (txq->q_type == CXGB4_TXQ_ETH) {
 512                         struct sge_eth_txq *eq;
 513
 514                         eq = container_of(txq, struct sge_eth_txq, q);
 515                         netif_tx_wake_queue(eq->txq);
 516                 } else {
 517                         struct sge_uld_txq *oq;
 518
 519                         oq = container_of(txq, struct sge_uld_txq, q);
 520                         tasklet_schedule(&oq->qresume_tsk);
 521                 }
 522         } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
 523                 const struct cpl_fw6_msg *p = (void *)rsp;
 524
 525 #ifdef CONFIG_CHELSIO_T4_DCB
 526                 const struct fw_port_cmd *pcmd = (const void *)p->data;
 527                 unsigned int cmd = FW_CMD_OP_G(ntohl(pcmd->op_to_portid));
 528                 unsigned int action =
 529                         FW_PORT_CMD_ACTION_G(ntohl(pcmd->action_to_len16));
 530
 531                 if (cmd == FW_PORT_CMD &&
 532                     action == FW_PORT_ACTION_GET_PORT_INFO) {
 533                         int port = FW_PORT_CMD_PORTID_G(
 534                                         be32_to_cpu(pcmd->op_to_portid));
 535                         struct net_device *dev =
 536                                 q->adap->port[q->adap->chan_map[port]];
 537                         int state_input = ((pcmd->u.info.dcbxdis_pkd &
 538                                             FW_PORT_CMD_DCBXDIS_F)
 539                                            ? CXGB4_DCB_INPUT_FW_DISABLED
 540                                            : CXGB4_DCB_INPUT_FW_ENABLED);
 541
 542                         cxgb4_dcb_state_fsm(dev, state_input);
 543                 }
 544
 545                 if (cmd == FW_PORT_CMD &&
 546                     action == FW_PORT_ACTION_L2_DCB_CFG)
 547                         dcb_rpl(q->adap, pcmd);
 548                 else
 549 #endif
 550                         if (p->type == 0)
 551                                 t4_handle_fw_rpl(q->adap, p->data);
 552         } else if (opcode == CPL_L2T_WRITE_RPL) {
 553                 const struct cpl_l2t_write_rpl *p = (void *)rsp;
 554
 555                 do_l2t_write_rpl(q->adap, p);
 556         } else if (opcode == CPL_SET_TCB_RPL) {
 557                 const struct cpl_set_tcb_rpl *p = (void *)rsp;
 558
 559                 filter_rpl(q->adap, p);
 560         } else
 561                 dev_err(q->adap->pdev_dev,
 562                         "unexpected CPL %#x on FW event queue\n", opcode);
 563 out:
 564         return 0;
 565 }
 566
 567 static void disable_msi(struct adapter *adapter)
 568 {
 569         if (adapter->flags & USING_MSIX) {
 570                 pci_disable_msix(adapter->pdev);
 571                 adapter->flags &= ~USING_MSIX;
 572         } else if (adapter->flags & USING_MSI) {
 573                 pci_disable_msi(adapter->pdev);
 574                 adapter->flags &= ~USING_MSI;
 575         }
 576 }
 577
 578 /*
 579  * Interrupt handler for non-data events used with MSI-X.
 580  */
 581 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
 582 {
 583         struct adapter *adap = cookie;
 584         u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A));
 585
 586         if (v & PFSW_F) {
 587                 adap->swintr = 1;
 588                 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A), v);
 589         }
 590         if (adap->flags & MASTER_PF)
 591                 t4_slow_intr_handler(adap);
 592         return IRQ_HANDLED;
 593 }
 594
 595 /*
 596  * Name the MSI-X interrupts.
 597  */
 598 static void name_msix_vecs(struct adapter *adap)
 599 {
 600         int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
 601
 602         /* non-data interrupts */
 603         snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
 604
 605         /* FW events */
 606         snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
 607                  adap->port[0]->name);
 608
 609         /* Ethernet queues */
 610         for_each_port(adap, j) {
 611                 struct net_device *d = adap->port[j];
 612                 const struct port_info *pi = netdev_priv(d);
 613
 614                 for (i = 0; i < pi->nqsets; i++, msi_idx++)
 615                         snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
 616                                  d->name, i);
 617         }
 618 }
 619
 620 static int request_msix_queue_irqs(struct adapter *adap)
 621 {
 622         struct sge *s = &adap->sge;
 623         int err, ethqidx;
 624         int msi_index = 2;
 625
 626         err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
 627                           adap->msix_info[1].desc, &s->fw_evtq);
 628         if (err)
 629                 return err;
 630
 631         for_each_ethrxq(s, ethqidx) {
 632                 err = request_irq(adap->msix_info[msi_index].vec,
 633                                   t4_sge_intr_msix, 0,
 634                                   adap->msix_info[msi_index].desc,
 635                                   &s->ethrxq[ethqidx].rspq);
 636                 if (err)
 637                         goto unwind;
 638                 msi_index++;
 639         }
 640         return 0;
 641
 642 unwind:
 643         while (--ethqidx >= 0)
 644                 free_irq(adap->msix_info[--msi_index].vec,
 645                          &s->ethrxq[ethqidx].rspq);
 646         free_irq(adap->msix_info[1].vec, &s->fw_evtq);
 647         return err;
 648 }
 649
 650 static void free_msix_queue_irqs(struct adapter *adap)
 651 {
 652         int i, msi_index = 2;
 653         struct sge *s = &adap->sge;
 654
 655         free_irq(adap->msix_info[1].vec, &s->fw_evtq);
 656         for_each_ethrxq(s, i)
 657                 free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
 658 }
 659
 660 /**
 661  *      cxgb4_write_rss - write the RSS table for a given port
 662  *      @pi: the port
 663  *      @queues: array of queue indices for RSS
 664  *
 665  *      Sets up the portion of the HW RSS table for the port's VI to distribute
 666  *      packets to the Rx queues in @queues.
 667  *      Should never be called before setting up sge eth rx queues
 668  */
 669 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
 670 {
 671         u16 *rss;
 672         int i, err;
 673         struct adapter *adapter = pi->adapter;
 674         const struct sge_eth_rxq *rxq;
 675
 676         rxq = &adapter->sge.ethrxq[pi->first_qset];
 677         rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
 678         if (!rss)
 679                 return -ENOMEM;
 680
 681         /* map the queue indices to queue ids */
 682         for (i = 0; i < pi->rss_size; i++, queues++)
 683                 rss[i] = rxq[*queues].rspq.abs_id;
 684
 685         err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
 686                                   pi->rss_size, rss, pi->rss_size);
 687         /* If Tunnel All Lookup isn't specified in the global RSS
 688          * Configuration, then we need to specify a default Ingress
 689          * Queue for any ingress packets which aren't hashed.  We'll
 690          * use our first ingress queue ...
 691          */
 692         if (!err)
 693                 err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
 694                                        FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN_F |
 695                                        FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN_F |
 696                                        FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN_F |
 697                                        FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN_F |
 698                                        FW_RSS_VI_CONFIG_CMD_UDPEN_F,
 699                                        rss[0]);
 700         kfree(rss);
 701         return err;
 702 }
 703
 704 /**
 705  *      setup_rss - configure RSS
 706  *      @adap: the adapter
 707  *
 708  *      Sets up RSS for each port.
 709  */
 710 static int setup_rss(struct adapter *adap)
 711 {
 712         int i, j, err;
 713
 714         for_each_port(adap, i) {
 715                 const struct port_info *pi = adap2pinfo(adap, i);
 716
 717                 /* Fill default values with equal distribution */
 718                 for (j = 0; j < pi->rss_size; j++)
 719                         pi->rss[j] = j % pi->nqsets;
 720
 721                 err = cxgb4_write_rss(pi, pi->rss);
 722                 if (err)
 723                         return err;
 724         }
 725         return 0;
 726 }
 727
 728 /*
 729  * Return the channel of the ingress queue with the given qid.
 730  */
 731 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
 732 {
 733         qid -= p->ingr_start;
 734         return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
 735 }
 736
 737 /*
 738  * Wait until all NAPI handlers are descheduled.
 739  */
 740 static void quiesce_rx(struct adapter *adap)
 741 {
 742         int i;
 743
 744         for (i = 0; i < adap->sge.ingr_sz; i++) {
 745                 struct sge_rspq *q = adap->sge.ingr_map[i];
 746
 747                 if (q && q->handler)
 748                         napi_disable(&q->napi);
 749         }
 750 }
 751
 752 /* Disable interrupt and napi handler */
 753 static void disable_interrupts(struct adapter *adap)
 754 {
 755         if (adap->flags & FULL_INIT_DONE) {
 756                 t4_intr_disable(adap);
 757                 if (adap->flags & USING_MSIX) {
 758                         free_msix_queue_irqs(adap);
 759                         free_irq(adap->msix_info[0].vec, adap);
 760                 } else {
 761                         free_irq(adap->pdev->irq, adap);
 762                 }
 763                 quiesce_rx(adap);
 764         }
 765 }
 766
 767 /*
 768  * Enable NAPI scheduling and interrupt generation for all Rx queues.
 769  */
 770 static void enable_rx(struct adapter *adap)
 771 {
 772         int i;
 773
 774         for (i = 0; i < adap->sge.ingr_sz; i++) {
 775                 struct sge_rspq *q = adap->sge.ingr_map[i];
 776
 777                 if (!q)
 778                         continue;
 779                 if (q->handler)
 780                         napi_enable(&q->napi);
 781
 782                 /* 0-increment GTS to start the timer and enable interrupts */
 783                 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS_A),
 784                              SEINTARM_V(q->intr_params) |
 785                              INGRESSQID_V(q->cntxt_id));
 786         }
 787 }
 788
 789
 790 static int setup_fw_sge_queues(struct adapter *adap)
 791 {
 792         struct sge *s = &adap->sge;
 793         int err = 0;
 794
 795         bitmap_zero(s->starving_fl, s->egr_sz);
 796         bitmap_zero(s->txq_maperr, s->egr_sz);
 797
 798         if (adap->flags & USING_MSIX)
 799                 adap->msi_idx = 1;         /* vector 0 is for non-queue interrupts */
 800         else {
 801                 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
 802                                        NULL, NULL, NULL, -1);
 803                 if (err)
 804                         return err;
 805                 adap->msi_idx = -((int)s->intrq.abs_id + 1);
 806         }
 807
 808         err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
 809                                adap->msi_idx, NULL, fwevtq_handler, NULL, -1);
 810         if (err)
 811                 t4_free_sge_resources(adap);
 812         return err;
 813 }
 814
 815 /**
 816  *      setup_sge_queues - configure SGE Tx/Rx/response queues
 817  *      @adap: the adapter
 818  *
 819  *      Determines how many sets of SGE queues to use and initializes them.
 820  *      We support multiple queue sets per port if we have MSI-X, otherwise
 821  *      just one queue set per port.
 822  */
 823 static int setup_sge_queues(struct adapter *adap)
 824 {
 825         int err, i, j;
 826         struct sge *s = &adap->sge;
 827         struct sge_uld_rxq_info *rxq_info = s->uld_rxq_info[CXGB4_ULD_RDMA];
 828         unsigned int cmplqid = 0;
 829
 830         for_each_port(adap, i) {
 831                 struct net_device *dev = adap->port[i];
 832                 struct port_info *pi = netdev_priv(dev);
 833                 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
 834                 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
 835
 836                 for (j = 0; j < pi->nqsets; j++, q++) {
 837                         if (adap->msi_idx > 0)
 838                                 adap->msi_idx++;
 839                         err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
 840                                                adap->msi_idx, &q->fl,
 841                                                t4_ethrx_handler,
 842                                                NULL,
 843                                                t4_get_mps_bg_map(adap,
 844                                                                  pi->tx_chan));
 845                         if (err)
 846                                 goto freeout;
 847                         q->rspq.idx = j;
 848                         memset(&q->stats, 0, sizeof(q->stats));
 849                 }
 850                 for (j = 0; j < pi->nqsets; j++, t++) {
 851                         err = t4_sge_alloc_eth_txq(adap, t, dev,
 852                                         netdev_get_tx_queue(dev, j),
 853                                         s->fw_evtq.cntxt_id);
 854                         if (err)
 855                                 goto freeout;
 856                 }
 857         }
 858
 859         for_each_port(adap, i) {
 860                 /* Note that cmplqid below is 0 if we don't
 861                  * have RDMA queues, and that's the right value.
 862                  */
 863                 if (rxq_info)
 864                         cmplqid = rxq_info->uldrxq[i].rspq.cntxt_id;
 865
 866                 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
 867                                             s->fw_evtq.cntxt_id, cmplqid);
 868                 if (err)
 869                         goto freeout;
 870         }
 871
 872         t4_write_reg(adap, is_t4(adap->params.chip) ?
 873                                 MPS_TRC_RSS_CONTROL_A :
 874                                 MPS_T5_TRC_RSS_CONTROL_A,
 875                      RSSCONTROL_V(netdev2pinfo(adap->port[0])->tx_chan) |
 876                      QUEUENUMBER_V(s->ethrxq[0].rspq.abs_id));
 877         return 0;
 878 freeout:
 879         t4_free_sge_resources(adap);
 880         return err;
 881 }
 882
 883 static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
 884                              void *accel_priv, select_queue_fallback_t fallback)
 885 {
 886         int txq;
 887
 888 #ifdef CONFIG_CHELSIO_T4_DCB
 889         /* If a Data Center Bridging has been successfully negotiated on this
 890          * link then we'll use the skb's priority to map it to a TX Queue.
 891          * The skb's priority is determined via the VLAN Tag Priority Code
 892          * Point field.
 893          */
 894         if (cxgb4_dcb_enabled(dev)) {
 895                 u16 vlan_tci;
 896                 int err;
 897
 898                 err = vlan_get_tag(skb, &vlan_tci);
 899                 if (unlikely(err)) {
 900                         if (net_ratelimit())
 901                                 netdev_warn(dev,
 902                                             "TX Packet without VLAN Tag on DCB Link\n");
 903                         txq = 0;
 904                 } else {
 905                         txq = (vlan_tci & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
 906 #ifdef CONFIG_CHELSIO_T4_FCOE
 907                         if (skb->protocol == htons(ETH_P_FCOE))
 908                                 txq = skb->priority & 0x7;
 909 #endif /* CONFIG_CHELSIO_T4_FCOE */
 910                 }
 911                 return txq;
 912         }
 913 #endif /* CONFIG_CHELSIO_T4_DCB */
 914
 915         if (select_queue) {
 916                 txq = (skb_rx_queue_recorded(skb)
 917                         ? skb_get_rx_queue(skb)
 918                         : smp_processor_id());
 919
 920                 while (unlikely(txq >= dev->real_num_tx_queues))
 921                         txq -= dev->real_num_tx_queues;
 922
 923                 return txq;
 924         }
 925
 926         return fallback(dev, skb) % dev->real_num_tx_queues;
 927 }
 928
 929 static int closest_timer(const struct sge *s, int time)
 930 {
 931         int i, delta, match = 0, min_delta = INT_MAX;
 932
 933         for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
 934                 delta = time - s->timer_val[i];
 935                 if (delta < 0)
 936                         delta = -delta;
 937                 if (delta < min_delta) {
 938                         min_delta = delta;
 939                         match = i;
 940                 }
 941         }
 942         return match;
 943 }
 944
 945 static int closest_thres(const struct sge *s, int thres)
 946 {
 947         int i, delta, match = 0, min_delta = INT_MAX;
 948
 949         for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
 950                 delta = thres - s->counter_val[i];
 951                 if (delta < 0)
 952                         delta = -delta;
 953                 if (delta < min_delta) {
 954                         min_delta = delta;
 955                         match = i;
 956                 }
 957         }
 958         return match;
 959 }
 960
 961 /**
 962  *      cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
 963  *      @q: the Rx queue
 964  *      @us: the hold-off time in us, or 0 to disable timer
 965  *      @cnt: the hold-off packet count, or 0 to disable counter
 966  *
 967  *      Sets an Rx queue's interrupt hold-off time and packet count.  At least
 968  *      one of the two needs to be enabled for the queue to generate interrupts.
 969  */
 970 int cxgb4_set_rspq_intr_params(struct sge_rspq *q,
 971                                unsigned int us, unsigned int cnt)
 972 {
 973         struct adapter *adap = q->adap;
 974
 975         if ((us | cnt) == 0)
 976                 cnt = 1;
 977
 978         if (cnt) {
 979                 int err;
 980                 u32 v, new_idx;
 981
 982                 new_idx = closest_thres(&adap->sge, cnt);
 983                 if (q->desc && q->pktcnt_idx != new_idx) {
 984                         /* the queue has already been created, update it */
 985                         v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
 986                             FW_PARAMS_PARAM_X_V(
 987                                         FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
 988                             FW_PARAMS_PARAM_YZ_V(q->cntxt_id);
 989                         err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
 990                                             &v, &new_idx);
 991                         if (err)
 992                                 return err;
 993                 }
 994                 q->pktcnt_idx = new_idx;
 995         }
 996
 997         us = us == 0 ? 6 : closest_timer(&adap->sge, us);
 998         q->intr_params = QINTR_TIMER_IDX_V(us) | QINTR_CNT_EN_V(cnt > 0);
 999         return 0;
1000 }
1001
1002 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
1003 {
1004         const struct port_info *pi = netdev_priv(dev);
1005         netdev_features_t changed = dev->features ^ features;
1006         int err;
1007
1008         if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
1009                 return 0;
1010
1011         err = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, -1,
1012                             -1, -1, -1,
1013                             !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
1014         if (unlikely(err))
1015                 dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
1016         return err;
1017 }
1018
1019 static int setup_debugfs(struct adapter *adap)
1020 {
1021         if (IS_ERR_OR_NULL(adap->debugfs_root))
1022                 return -1;
1023
1024 #ifdef CONFIG_DEBUG_FS
1025         t4_setup_debugfs(adap);
1026 #endif
1027         return 0;
1028 }
1029
1030 /*
1031  * upper-layer driver support
1032  */
1033
1034 /*
1035  * Allocate an active-open TID and set it to the supplied value.
1036  */
1037 int cxgb4_alloc_atid(struct tid_info *t, void *data)
1038 {
1039         int atid = -1;
1040
1041         spin_lock_bh(&t->atid_lock);
1042         if (t->afree) {
1043                 union aopen_entry *p = t->afree;
1044
1045                 atid = (p - t->atid_tab) + t->atid_base;
1046                 t->afree = p->next;
1047                 p->data = data;
1048                 t->atids_in_use++;
1049         }
1050         spin_unlock_bh(&t->atid_lock);
1051         return atid;
1052 }
1053 EXPORT_SYMBOL(cxgb4_alloc_atid);
1054
1055 /*
1056  * Release an active-open TID.
1057  */
1058 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
1059 {
1060         union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
1061
1062         spin_lock_bh(&t->atid_lock);
1063         p->next = t->afree;
1064         t->afree = p;
1065         t->atids_in_use--;
1066         spin_unlock_bh(&t->atid_lock);
1067 }
1068 EXPORT_SYMBOL(cxgb4_free_atid);
1069
1070 /*
1071  * Allocate a server TID and set it to the supplied value.
1072  */
1073 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
1074 {
1075         int stid;
1076
1077         spin_lock_bh(&t->stid_lock);
1078         if (family == PF_INET) {
1079                 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
1080                 if (stid < t->nstids)
1081                         __set_bit(stid, t->stid_bmap);
1082                 else
1083                         stid = -1;
1084         } else {
1085                 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 1);
1086                 if (stid < 0)
1087                         stid = -1;
1088         }
1089         if (stid >= 0) {
1090                 t->stid_tab[stid].data = data;
1091                 stid += t->stid_base;
1092                 /* IPv6 requires max of 520 bits or 16 cells in TCAM
1093                  * This is equivalent to 4 TIDs. With CLIP enabled it
1094                  * needs 2 TIDs.
1095                  */
1096                 if (family == PF_INET)
1097                         t->stids_in_use++;
1098                 else
1099                         t->stids_in_use += 2;
1100         }
1101         spin_unlock_bh(&t->stid_lock);
1102         return stid;
1103 }
1104 EXPORT_SYMBOL(cxgb4_alloc_stid);
1105
1106 /* Allocate a server filter TID and set it to the supplied value.
1107  */
1108 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
1109 {
1110         int stid;
1111
1112         spin_lock_bh(&t->stid_lock);
1113         if (family == PF_INET) {
1114                 stid = find_next_zero_bit(t->stid_bmap,
1115                                 t->nstids + t->nsftids, t->nstids);
1116                 if (stid < (t->nstids + t->nsftids))
1117                         __set_bit(stid, t->stid_bmap);
1118                 else
1119                         stid = -1;
1120         } else {
1121                 stid = -1;
1122         }
1123         if (stid >= 0) {
1124                 t->stid_tab[stid].data = data;
1125                 stid -= t->nstids;
1126                 stid += t->sftid_base;
1127                 t->sftids_in_use++;
1128         }
1129         spin_unlock_bh(&t->stid_lock);
1130         return stid;
1131 }
1132 EXPORT_SYMBOL(cxgb4_alloc_sftid);
1133
1134 /* Release a server TID.
1135  */
1136 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
1137 {
1138         /* Is it a server filter TID? */
1139         if (t->nsftids && (stid >= t->sftid_base)) {
1140                 stid -= t->sftid_base;
1141                 stid += t->nstids;
1142         } else {
1143                 stid -= t->stid_base;
1144         }
1145
1146         spin_lock_bh(&t->stid_lock);
1147         if (family == PF_INET)
1148                 __clear_bit(stid, t->stid_bmap);
1149         else
1150                 bitmap_release_region(t->stid_bmap, stid, 1);
1151         t->stid_tab[stid].data = NULL;
1152         if (stid < t->nstids) {
1153                 if (family == PF_INET)
1154                         t->stids_in_use--;
1155                 else
1156                         t->stids_in_use -= 2;
1157         } else {
1158                 t->sftids_in_use--;
1159         }
1160         spin_unlock_bh(&t->stid_lock);
1161 }
1162 EXPORT_SYMBOL(cxgb4_free_stid);
1163
1164 /*
1165  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
1166  */
1167 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
1168                            unsigned int tid)
1169 {
1170         struct cpl_tid_release *req;
1171
1172         set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1173         req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
1174         INIT_TP_WR(req, tid);
1175         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
1176 }
1177
1178 /*
1179  * Queue a TID release request and if necessary schedule a work queue to
1180  * process it.
1181  */
1182 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
1183                                     unsigned int tid)
1184 {
1185         void **p = &t->tid_tab[tid];
1186         struct adapter *adap = container_of(t, struct adapter, tids);
1187
1188         spin_lock_bh(&adap->tid_release_lock);
1189         *p = adap->tid_release_head;
1190         /* Low 2 bits encode the Tx channel number */
1191         adap->tid_release_head = (void **)((uintptr_t)p | chan);
1192         if (!adap->tid_release_task_busy) {
1193                 adap->tid_release_task_busy = true;
1194                 queue_work(adap->workq, &adap->tid_release_task);
1195         }
1196         spin_unlock_bh(&adap->tid_release_lock);
1197 }
1198
1199 /*
1200  * Process the list of pending TID release requests.
1201  */
1202 static void process_tid_release_list(struct work_struct *work)
1203 {
1204         struct sk_buff *skb;
1205         struct adapter *adap;
1206
1207         adap = container_of(work, struct adapter, tid_release_task);
1208
1209         spin_lock_bh(&adap->tid_release_lock);
1210         while (adap->tid_release_head) {
1211                 void **p = adap->tid_release_head;
1212                 unsigned int chan = (uintptr_t)p & 3;
1213                 p = (void *)p - chan;
1214
1215                 adap->tid_release_head = *p;
1216                 *p = NULL;
1217                 spin_unlock_bh(&adap->tid_release_lock);
1218
1219                 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
1220                                          GFP_KERNEL)))
1221                         schedule_timeout_uninterruptible(1);
1222
1223                 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
1224                 t4_ofld_send(adap, skb);
1225                 spin_lock_bh(&adap->tid_release_lock);
1226         }
1227         adap->tid_release_task_busy = false;
1228         spin_unlock_bh(&adap->tid_release_lock);
1229 }
1230
1231 /*
1232  * Release a TID and inform HW.  If we are unable to allocate the release
1233  * message we defer to a work queue.
1234  */
1235 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
1236 {
1237         struct sk_buff *skb;
1238         struct adapter *adap = container_of(t, struct adapter, tids);
1239
1240         WARN_ON(tid >= t->ntids);
1241
1242         if (t->tid_tab[tid]) {
1243                 t->tid_tab[tid] = NULL;
1244                 if (t->hash_base && (tid >= t->hash_base))
1245                         atomic_dec(&t->hash_tids_in_use);
1246                 else
1247                         atomic_dec(&t->tids_in_use);
1248         }
1249
1250         skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
1251         if (likely(skb)) {
1252                 mk_tid_release(skb, chan, tid);
1253                 t4_ofld_send(adap, skb);
1254         } else
1255                 cxgb4_queue_tid_release(t, chan, tid);
1256 }
1257 EXPORT_SYMBOL(cxgb4_remove_tid);
1258
1259 /*
1260  * Allocate and initialize the TID tables.  Returns 0 on success.
1261  */
1262 static int tid_init(struct tid_info *t)
1263 {
1264         struct adapter *adap = container_of(t, struct adapter, tids);
1265         unsigned int max_ftids = t->nftids + t->nsftids;
1266         unsigned int natids = t->natids;
1267         unsigned int stid_bmap_size;
1268         unsigned int ftid_bmap_size;
1269         size_t size;
1270
1271         stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
1272         ftid_bmap_size = BITS_TO_LONGS(t->nftids);
1273         size = t->ntids * sizeof(*t->tid_tab) +
1274                natids * sizeof(*t->atid_tab) +
1275                t->nstids * sizeof(*t->stid_tab) +
1276                t->nsftids * sizeof(*t->stid_tab) +
1277                stid_bmap_size * sizeof(long) +
1278                max_ftids * sizeof(*t->ftid_tab) +
1279                ftid_bmap_size * sizeof(long);
1280
1281         t->tid_tab = kvzalloc(size, GFP_KERNEL);
1282         if (!t->tid_tab)
1283                 return -ENOMEM;
1284
1285         t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
1286         t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
1287         t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
1288         t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
1289         t->ftid_bmap = (unsigned long *)&t->ftid_tab[max_ftids];
1290         spin_lock_init(&t->stid_lock);
1291         spin_lock_init(&t->atid_lock);
1292         spin_lock_init(&t->ftid_lock);
1293
1294         t->stids_in_use = 0;
1295         t->sftids_in_use = 0;
1296         t->afree = NULL;
1297         t->atids_in_use = 0;
1298         atomic_set(&t->tids_in_use, 0);
1299         atomic_set(&t->hash_tids_in_use, 0);
1300
1301         /* Setup the free list for atid_tab and clear the stid bitmap. */
1302         if (natids) {
1303                 while (--natids)
1304                         t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1305                 t->afree = t->atid_tab;
1306         }
1307
1308         if (is_offload(adap)) {
1309                 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
1310                 /* Reserve stid 0 for T4/T5 adapters */
1311                 if (!t->stid_base &&
1312                     CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1313                         __set_bit(0, t->stid_bmap);
1314         }
1315
1316         bitmap_zero(t->ftid_bmap, t->nftids);
1317         return 0;
1318 }
1319
1320 /**
1321  *      cxgb4_create_server - create an IP server
1322  *      @dev: the device
1323  *      @stid: the server TID
1324  *      @sip: local IP address to bind server to
1325  *      @sport: the server's TCP port
1326  *      @queue: queue to direct messages from this server to
1327  *
1328  *      Create an IP server for the given port and address.
1329  *      Returns <0 on error and one of the %NET_XMIT_* values on success.
1330  */
1331 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
1332                         __be32 sip, __be16 sport, __be16 vlan,
1333                         unsigned int queue)
1334 {
1335         unsigned int chan;
1336         struct sk_buff *skb;
1337         struct adapter *adap;
1338         struct cpl_pass_open_req *req;
1339         int ret;
1340
1341         skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1342         if (!skb)
1343                 return -ENOMEM;
1344
1345         adap = netdev2adap(dev);
1346         req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
1347         INIT_TP_WR(req, 0);
1348         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
1349         req->local_port = sport;
1350         req->peer_port = htons(0);
1351         req->local_ip = sip;
1352         req->peer_ip = htonl(0);
1353         chan = rxq_to_chan(&adap->sge, queue);
1354         req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1355         req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1356                                 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1357         ret = t4_mgmt_tx(adap, skb);
1358         return net_xmit_eval(ret);
1359 }
1360 EXPORT_SYMBOL(cxgb4_create_server);
1361
1362 /*      cxgb4_create_server6 - create an IPv6 server
1363  *      @dev: the device
1364  *      @stid: the server TID
1365  *      @sip: local IPv6 address to bind server to
1366  *      @sport: the server's TCP port
1367  *      @queue: queue to direct messages from this server to
1368  *
1369  *      Create an IPv6 server for the given port and address.
1370  *      Returns <0 on error and one of the %NET_XMIT_* values on success.
1371  */
1372 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
1373                          const struct in6_addr *sip, __be16 sport,
1374                          unsigned int queue)
1375 {
1376         unsigned int chan;
1377         struct sk_buff *skb;
1378         struct adapter *adap;
1379         struct cpl_pass_open_req6 *req;
1380         int ret;
1381
1382         skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1383         if (!skb)
1384                 return -ENOMEM;
1385
1386         adap = netdev2adap(dev);
1387         req = (struct cpl_pass_open_req6 *)__skb_put(skb, sizeof(*req));
1388         INIT_TP_WR(req, 0);
1389         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
1390         req->local_port = sport;
1391         req->peer_port = htons(0);
1392         req->local_ip_hi = *(__be64 *)(sip->s6_addr);
1393         req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
1394         req->peer_ip_hi = cpu_to_be64(0);
1395         req->peer_ip_lo = cpu_to_be64(0);
1396         chan = rxq_to_chan(&adap->sge, queue);
1397         req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1398         req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1399                                 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1400         ret = t4_mgmt_tx(adap, skb);
1401         return net_xmit_eval(ret);
1402 }
1403 EXPORT_SYMBOL(cxgb4_create_server6);
1404
1405 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
1406                         unsigned int queue, bool ipv6)
1407 {
1408         struct sk_buff *skb;
1409         struct adapter *adap;
1410         struct cpl_close_listsvr_req *req;
1411         int ret;
1412
1413         adap = netdev2adap(dev);
1414
1415         skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1416         if (!skb)
1417                 return -ENOMEM;
1418
1419         req = (struct cpl_close_listsvr_req *)__skb_put(skb, sizeof(*req));
1420         INIT_TP_WR(req, 0);
1421         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
1422         req->reply_ctrl = htons(NO_REPLY_V(0) | (ipv6 ? LISTSVR_IPV6_V(1) :
1423                                 LISTSVR_IPV6_V(0)) | QUEUENO_V(queue));
1424         ret = t4_mgmt_tx(adap, skb);
1425         return net_xmit_eval(ret);
1426 }
1427 EXPORT_SYMBOL(cxgb4_remove_server);
1428
1429 /**
1430  *      cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
1431  *      @mtus: the HW MTU table
1432  *      @mtu: the target MTU
1433  *      @idx: index of selected entry in the MTU table
1434  *
1435  *      Returns the index and the value in the HW MTU table that is closest to
1436  *      but does not exceed @mtu, unless @mtu is smaller than any value in the
1437  *      table, in which case that smallest available value is selected.
1438  */
1439 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
1440                             unsigned int *idx)
1441 {
1442         unsigned int i = 0;
1443
1444         while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
1445                 ++i;
1446         if (idx)
1447                 *idx = i;
1448         return mtus[i];
1449 }
1450 EXPORT_SYMBOL(cxgb4_best_mtu);
1451
1452 /**
1453  *     cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
1454  *     @mtus: the HW MTU table
1455  *     @header_size: Header Size
1456  *     @data_size_max: maximum Data Segment Size
1457  *     @data_size_align: desired Data Segment Size Alignment (2^N)
1458  *     @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
1459  *
1460  *     Similar to cxgb4_best_mtu() but instead of searching the Hardware
1461  *     MTU Table based solely on a Maximum MTU parameter, we break that
1462  *     parameter up into a Header Size and Maximum Data Segment Size, and
1463  *     provide a desired Data Segment Size Alignment.  If we find an MTU in
1464  *     the Hardware MTU Table which will result in a Data Segment Size with
1465  *     the requested alignment _and_ that MTU isn't "too far" from the
1466  *     closest MTU, then we'll return that rather than the closest MTU.
1467  */
1468 unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
1469                                     unsigned short header_size,
1470                                     unsigned short data_size_max,
1471                                     unsigned short data_size_align,
1472                                     unsigned int *mtu_idxp)
1473 {
1474         unsigned short max_mtu = header_size + data_size_max;
1475         unsigned short data_size_align_mask = data_size_align - 1;
1476         int mtu_idx, aligned_mtu_idx;
1477
1478         /* Scan the MTU Table till we find an MTU which is larger than our
1479          * Maximum MTU or we reach the end of the table.  Along the way,
1480          * record the last MTU found, if any, which will result in a Data
1481          * Segment Length matching the requested alignment.
1482          */
1483         for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
1484                 unsigned short data_size = mtus[mtu_idx] - header_size;
1485
1486                 /* If this MTU minus the Header Size would result in a
1487                  * Data Segment Size of the desired alignment, remember it.
1488                  */
1489                 if ((data_size & data_size_align_mask) == 0)
1490                         aligned_mtu_idx = mtu_idx;
1491
1492                 /* If we're not at the end of the Hardware MTU Table and the
1493                  * next element is larger than our Maximum MTU, drop out of
1494                  * the loop.
1495                  */
1496                 if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
1497                         break;
1498         }
1499
1500         /* If we fell out of the loop because we ran to the end of the table,
1501          * then we just have to use the last [largest] entry.
1502          */
1503         if (mtu_idx == NMTUS)
1504                 mtu_idx--;
1505
1506         /* If we found an MTU which resulted in the requested Data Segment
1507          * Length alignment and that's "not far" from the largest MTU which is
1508          * less than or equal to the maximum MTU, then use that.
1509          */
1510         if (aligned_mtu_idx >= 0 &&
1511             mtu_idx - aligned_mtu_idx <= 1)
1512                 mtu_idx = aligned_mtu_idx;
1513
1514         /* If the caller has passed in an MTU Index pointer, pass the
1515          * MTU Index back.  Return the MTU value.
1516          */
1517         if (mtu_idxp)
1518                 *mtu_idxp = mtu_idx;
1519         return mtus[mtu_idx];
1520 }
1521 EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
1522
1523 /**
1524  *      cxgb4_tp_smt_idx - Get the Source Mac Table index for this VI
1525  *      @chip: chip type
1526  *      @viid: VI id of the given port
1527  *
1528  *      Return the SMT index for this VI.
1529  */
1530 unsigned int cxgb4_tp_smt_idx(enum chip_type chip, unsigned int viid)
1531 {
1532         /* In T4/T5, SMT contains 256 SMAC entries organized in
1533          * 128 rows of 2 entries each.
1534          * In T6, SMT contains 256 SMAC entries in 256 rows.
1535          * TODO: The below code needs to be updated when we add support
1536          * for 256 VFs.
1537          */
1538         if (CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5)
1539                 return ((viid & 0x7f) << 1);
1540         else
1541                 return (viid & 0x7f);
1542 }
1543 EXPORT_SYMBOL(cxgb4_tp_smt_idx);
1544
1545 /**
1546  *      cxgb4_port_chan - get the HW channel of a port
1547  *      @dev: the net device for the port
1548  *
1549  *      Return the HW Tx channel of the given port.
1550  */
1551 unsigned int cxgb4_port_chan(const struct net_device *dev)
1552 {
1553         return netdev2pinfo(dev)->tx_chan;
1554 }
1555 EXPORT_SYMBOL(cxgb4_port_chan);
1556
1557 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
1558 {
1559         struct adapter *adap = netdev2adap(dev);
1560         u32 v1, v2, lp_count, hp_count;
1561
1562         v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1563         v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1564         if (is_t4(adap->params.chip)) {
1565                 lp_count = LP_COUNT_G(v1);
1566                 hp_count = HP_COUNT_G(v1);
1567         } else {
1568                 lp_count = LP_COUNT_T5_G(v1);
1569                 hp_count = HP_COUNT_T5_G(v2);
1570         }
1571         return lpfifo ? lp_count : hp_count;
1572 }
1573 EXPORT_SYMBOL(cxgb4_dbfifo_count);
1574
1575 /**
1576  *      cxgb4_port_viid - get the VI id of a port
1577  *      @dev: the net device for the port
1578  *
1579  *      Return the VI id of the given port.
1580  */
1581 unsigned int cxgb4_port_viid(const struct net_device *dev)
1582 {
1583         return netdev2pinfo(dev)->viid;
1584 }
1585 EXPORT_SYMBOL(cxgb4_port_viid);
1586
1587 /**
1588  *      cxgb4_port_idx - get the index of a port
1589  *      @dev: the net device for the port
1590  *
1591  *      Return the index of the given port.
1592  */
1593 unsigned int cxgb4_port_idx(const struct net_device *dev)
1594 {
1595         return netdev2pinfo(dev)->port_id;
1596 }
1597 EXPORT_SYMBOL(cxgb4_port_idx);
1598
1599 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
1600                          struct tp_tcp_stats *v6)
1601 {
1602         struct adapter *adap = pci_get_drvdata(pdev);
1603
1604         spin_lock(&adap->stats_lock);
1605         t4_tp_get_tcp_stats(adap, v4, v6);
1606         spin_unlock(&adap->stats_lock);
1607 }
1608 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
1609
1610 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
1611                       const unsigned int *pgsz_order)
1612 {
1613         struct adapter *adap = netdev2adap(dev);
1614
1615         t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK_A, tag_mask);
1616         t4_write_reg(adap, ULP_RX_ISCSI_PSZ_A, HPZ0_V(pgsz_order[0]) |
1617                      HPZ1_V(pgsz_order[1]) | HPZ2_V(pgsz_order[2]) |
1618                      HPZ3_V(pgsz_order[3]));
1619 }
1620 EXPORT_SYMBOL(cxgb4_iscsi_init);
1621
1622 int cxgb4_flush_eq_cache(struct net_device *dev)
1623 {
1624         struct adapter *adap = netdev2adap(dev);
1625
1626         return t4_sge_ctxt_flush(adap, adap->mbox);
1627 }
1628 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
1629
1630 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
1631 {
1632         u32 addr = t4_read_reg(adap, SGE_DBQ_CTXT_BADDR_A) + 24 * qid + 8;
1633         __be64 indices;
1634         int ret;
1635
1636         spin_lock(&adap->win0_lock);
1637         ret = t4_memory_rw(adap, 0, MEM_EDC0, addr,
1638                            sizeof(indices), (__be32 *)&indices,
1639                            T4_MEMORY_READ);
1640         spin_unlock(&adap->win0_lock);
1641         if (!ret) {
1642                 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
1643                 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
1644         }
1645         return ret;
1646 }
1647
1648 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
1649                         u16 size)
1650 {
1651         struct adapter *adap = netdev2adap(dev);
1652         u16 hw_pidx, hw_cidx;
1653         int ret;
1654
1655         ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
1656         if (ret)
1657                 goto out;
1658
1659         if (pidx != hw_pidx) {
1660                 u16 delta;
1661                 u32 val;
1662
1663                 if (pidx >= hw_pidx)
1664                         delta = pidx - hw_pidx;
1665                 else
1666                         delta = size - hw_pidx + pidx;
1667
1668                 if (is_t4(adap->params.chip))
1669                         val = PIDX_V(delta);
1670                 else
1671                         val = PIDX_T5_V(delta);
1672                 wmb();
1673                 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1674                              QID_V(qid) | val);
1675         }
1676 out:
1677         return ret;
1678 }
1679 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
1680
1681 int cxgb4_read_tpte(struct net_device *dev, u32 stag, __be32 *tpte)
1682 {
1683         struct adapter *adap;
1684         u32 offset, memtype, memaddr;
1685         u32 edc0_size, edc1_size, mc0_size, mc1_size, size;
1686         u32 edc0_end, edc1_end, mc0_end, mc1_end;
1687         int ret;
1688
1689         adap = netdev2adap(dev);
1690
1691         offset = ((stag >> 8) * 32) + adap->vres.stag.start;
1692
1693         /* Figure out where the offset lands in the Memory Type/Address scheme.
1694          * This code assumes that the memory is laid out starting at offset 0
1695          * with no breaks as: EDC0, EDC1, MC0, MC1. All cards have both EDC0
1696          * and EDC1.  Some cards will have neither MC0 nor MC1, most cards have
1697          * MC0, and some have both MC0 and MC1.
1698          */
1699         size = t4_read_reg(adap, MA_EDRAM0_BAR_A);
1700         edc0_size = EDRAM0_SIZE_G(size) << 20;
1701         size = t4_read_reg(adap, MA_EDRAM1_BAR_A);
1702         edc1_size = EDRAM1_SIZE_G(size) << 20;
1703         size = t4_read_reg(adap, MA_EXT_MEMORY0_BAR_A);
1704         mc0_size = EXT_MEM0_SIZE_G(size) << 20;
1705
1706         edc0_end = edc0_size;
1707         edc1_end = edc0_end + edc1_size;
1708         mc0_end = edc1_end + mc0_size;
1709
1710         if (offset < edc0_end) {
1711                 memtype = MEM_EDC0;
1712                 memaddr = offset;
1713         } else if (offset < edc1_end) {
1714                 memtype = MEM_EDC1;
1715                 memaddr = offset - edc0_end;
1716         } else {
1717                 if (offset < mc0_end) {
1718                         memtype = MEM_MC0;
1719                         memaddr = offset - edc1_end;
1720                 } else if (is_t5(adap->params.chip)) {
1721                         size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
1722                         mc1_size = EXT_MEM1_SIZE_G(size) << 20;
1723                         mc1_end = mc0_end + mc1_size;
1724                         if (offset < mc1_end) {
1725                                 memtype = MEM_MC1;
1726                                 memaddr = offset - mc0_end;
1727                         } else {
1728                                 /* offset beyond the end of any memory */
1729                                 goto err;
1730                         }
1731                 } else {
1732                         /* T4/T6 only has a single memory channel */
1733                         goto err;
1734                 }
1735         }
1736
1737         spin_lock(&adap->win0_lock);
1738         ret = t4_memory_rw(adap, 0, memtype, memaddr, 32, tpte, T4_MEMORY_READ);
1739         spin_unlock(&adap->win0_lock);
1740         return ret;
1741
1742 err:
1743         dev_err(adap->pdev_dev, "stag %#x, offset %#x out of range\n",
1744                 stag, offset);
1745         return -EINVAL;
1746 }
1747 EXPORT_SYMBOL(cxgb4_read_tpte);
1748
1749 u64 cxgb4_read_sge_timestamp(struct net_device *dev)
1750 {
1751         u32 hi, lo;
1752         struct adapter *adap;
1753
1754         adap = netdev2adap(dev);
1755         lo = t4_read_reg(adap, SGE_TIMESTAMP_LO_A);
1756         hi = TSVAL_G(t4_read_reg(adap, SGE_TIMESTAMP_HI_A));
1757
1758         return ((u64)hi << 32) | (u64)lo;
1759 }
1760 EXPORT_SYMBOL(cxgb4_read_sge_timestamp);
1761
1762 int cxgb4_bar2_sge_qregs(struct net_device *dev,
1763                          unsigned int qid,
1764                          enum cxgb4_bar2_qtype qtype,
1765                          int user,
1766                          u64 *pbar2_qoffset,
1767                          unsigned int *pbar2_qid)
1768 {
1769         return t4_bar2_sge_qregs(netdev2adap(dev),
1770                                  qid,
1771                                  (qtype == CXGB4_BAR2_QTYPE_EGRESS
1772                                   ? T4_BAR2_QTYPE_EGRESS
1773                                   : T4_BAR2_QTYPE_INGRESS),
1774                                  user,
1775                                  pbar2_qoffset,
1776                                  pbar2_qid);
1777 }
1778 EXPORT_SYMBOL(cxgb4_bar2_sge_qregs);
1779
1780 static struct pci_driver cxgb4_driver;
1781
1782 static void check_neigh_update(struct neighbour *neigh)
1783 {
1784         const struct device *parent;
1785         const struct net_device *netdev = neigh->dev;
1786
1787         if (is_vlan_dev(netdev))
1788                 netdev = vlan_dev_real_dev(netdev);
1789         parent = netdev->dev.parent;
1790         if (parent && parent->driver == &cxgb4_driver.driver)
1791                 t4_l2t_update(dev_get_drvdata(parent), neigh);
1792 }
1793
1794 static int netevent_cb(struct notifier_block *nb, unsigned long event,
1795                        void *data)
1796 {
1797         switch (event) {
1798         case NETEVENT_NEIGH_UPDATE:
1799                 check_neigh_update(data);
1800                 break;
1801         case NETEVENT_REDIRECT:
1802         default:
1803                 break;
1804         }
1805         return 0;
1806 }
1807
1808 static bool netevent_registered;
1809 static struct notifier_block cxgb4_netevent_nb = {
1810         .notifier_call = netevent_cb
1811 };
1812
1813 static void drain_db_fifo(struct adapter *adap, int usecs)
1814 {
1815         u32 v1, v2, lp_count, hp_count;
1816
1817         do {
1818                 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1819                 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1820                 if (is_t4(adap->params.chip)) {
1821                         lp_count = LP_COUNT_G(v1);
1822                         hp_count = HP_COUNT_G(v1);
1823                 } else {
1824                         lp_count = LP_COUNT_T5_G(v1);
1825                         hp_count = HP_COUNT_T5_G(v2);
1826                 }
1827
1828                 if (lp_count == 0 && hp_count == 0)
1829                         break;
1830                 set_current_state(TASK_UNINTERRUPTIBLE);
1831                 schedule_timeout(usecs_to_jiffies(usecs));
1832         } while (1);
1833 }
1834
1835 static void disable_txq_db(struct sge_txq *q)
1836 {
1837         unsigned long flags;
1838
1839         spin_lock_irqsave(&q->db_lock, flags);
1840         q->db_disabled = 1;
1841         spin_unlock_irqrestore(&q->db_lock, flags);
1842 }
1843
1844 static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
1845 {
1846         spin_lock_irq(&q->db_lock);
1847         if (q->db_pidx_inc) {
1848                 /* Make sure that all writes to the TX descriptors
1849                  * are committed before we tell HW about them.
1850                  */
1851                 wmb();
1852                 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1853                              QID_V(q->cntxt_id) | PIDX_V(q->db_pidx_inc));
1854                 q->db_pidx_inc = 0;
1855         }
1856         q->db_disabled = 0;
1857         spin_unlock_irq(&q->db_lock);
1858 }
1859
1860 static void disable_dbs(struct adapter *adap)
1861 {
1862         int i;
1863
1864         for_each_ethrxq(&adap->sge, i)
1865                 disable_txq_db(&adap->sge.ethtxq[i].q);
1866         if (is_offload(adap)) {
1867                 struct sge_uld_txq_info *txq_info =
1868                         adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1869
1870                 if (txq_info) {
1871                         for_each_ofldtxq(&adap->sge, i) {
1872                                 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1873
1874                                 disable_txq_db(&txq->q);
1875                         }
1876                 }
1877         }
1878         for_each_port(adap, i)
1879                 disable_txq_db(&adap->sge.ctrlq[i].q);
1880 }
1881
1882 static void enable_dbs(struct adapter *adap)
1883 {
1884         int i;
1885
1886         for_each_ethrxq(&adap->sge, i)
1887                 enable_txq_db(adap, &adap->sge.ethtxq[i].q);
1888         if (is_offload(adap)) {
1889                 struct sge_uld_txq_info *txq_info =
1890                         adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1891
1892                 if (txq_info) {
1893                         for_each_ofldtxq(&adap->sge, i) {
1894                                 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1895
1896                                 enable_txq_db(adap, &txq->q);
1897                         }
1898                 }
1899         }
1900         for_each_port(adap, i)
1901                 enable_txq_db(adap, &adap->sge.ctrlq[i].q);
1902 }
1903
1904 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
1905 {
1906         enum cxgb4_uld type = CXGB4_ULD_RDMA;
1907
1908         if (adap->uld && adap->uld[type].handle)
1909                 adap->uld[type].control(adap->uld[type].handle, cmd);
1910 }
1911
1912 static void process_db_full(struct work_struct *work)
1913 {
1914         struct adapter *adap;
1915
1916         adap = container_of(work, struct adapter, db_full_task);
1917
1918         drain_db_fifo(adap, dbfifo_drain_delay);
1919         enable_dbs(adap);
1920         notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
1921         if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1922                 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1923                                  DBFIFO_HP_INT_F | DBFIFO_LP_INT_F,
1924                                  DBFIFO_HP_INT_F | DBFIFO_LP_INT_F);
1925         else
1926                 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1927                                  DBFIFO_LP_INT_F, DBFIFO_LP_INT_F);
1928 }
1929
1930 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
1931 {
1932         u16 hw_pidx, hw_cidx;
1933         int ret;
1934
1935         spin_lock_irq(&q->db_lock);
1936         ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
1937         if (ret)
1938                 goto out;
1939         if (q->db_pidx != hw_pidx) {
1940                 u16 delta;
1941                 u32 val;
1942
1943                 if (q->db_pidx >= hw_pidx)
1944                         delta = q->db_pidx - hw_pidx;
1945                 else
1946                         delta = q->size - hw_pidx + q->db_pidx;
1947
1948                 if (is_t4(adap->params.chip))
1949                         val = PIDX_V(delta);
1950                 else
1951                         val = PIDX_T5_V(delta);
1952                 wmb();
1953                 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1954                              QID_V(q->cntxt_id) | val);
1955         }
1956 out:
1957         q->db_disabled = 0;
1958         q->db_pidx_inc = 0;
1959         spin_unlock_irq(&q->db_lock);
1960         if (ret)
1961                 CH_WARN(adap, "DB drop recovery failed.\n");
1962 }
1963
1964 static void recover_all_queues(struct adapter *adap)
1965 {
1966         int i;
1967
1968         for_each_ethrxq(&adap->sge, i)
1969                 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
1970         if (is_offload(adap)) {
1971                 struct sge_uld_txq_info *txq_info =
1972                         adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1973                 if (txq_info) {
1974                         for_each_ofldtxq(&adap->sge, i) {
1975                                 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1976
1977                                 sync_txq_pidx(adap, &txq->q);
1978                         }
1979                 }
1980         }
1981         for_each_port(adap, i)
1982                 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
1983 }
1984
1985 static void process_db_drop(struct work_struct *work)
1986 {
1987         struct adapter *adap;
1988
1989         adap = container_of(work, struct adapter, db_drop_task);
1990
1991         if (is_t4(adap->params.chip)) {
1992                 drain_db_fifo(adap, dbfifo_drain_delay);
1993                 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
1994                 drain_db_fifo(adap, dbfifo_drain_delay);
1995                 recover_all_queues(adap);
1996                 drain_db_fifo(adap, dbfifo_drain_delay);
1997                 enable_dbs(adap);
1998                 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
1999         } else if (is_t5(adap->params.chip)) {
2000                 u32 dropped_db = t4_read_reg(adap, 0x010ac);
2001                 u16 qid = (dropped_db >> 15) & 0x1ffff;
2002                 u16 pidx_inc = dropped_db & 0x1fff;
2003                 u64 bar2_qoffset;
2004                 unsigned int bar2_qid;
2005                 int ret;
2006
2007                 ret = t4_bar2_sge_qregs(adap, qid, T4_BAR2_QTYPE_EGRESS,
2008                                         0, &bar2_qoffset, &bar2_qid);
2009                 if (ret)
2010                         dev_err(adap->pdev_dev, "doorbell drop recovery: "
2011                                 "qid=%d, pidx_inc=%d\n", qid, pidx_inc);
2012                 else
2013                         writel(PIDX_T5_V(pidx_inc) | QID_V(bar2_qid),
2014                                adap->bar2 + bar2_qoffset + SGE_UDB_KDOORBELL);
2015
2016                 /* Re-enable BAR2 WC */
2017                 t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
2018         }
2019
2020         if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2021                 t4_set_reg_field(adap, SGE_DOORBELL_CONTROL_A, DROPPED_DB_F, 0);
2022 }
2023
2024 void t4_db_full(struct adapter *adap)
2025 {
2026         if (is_t4(adap->params.chip)) {
2027                 disable_dbs(adap);
2028                 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2029                 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2030                                  DBFIFO_HP_INT_F | DBFIFO_LP_INT_F, 0);
2031                 queue_work(adap->workq, &adap->db_full_task);
2032         }
2033 }
2034
2035 void t4_db_dropped(struct adapter *adap)
2036 {
2037         if (is_t4(adap->params.chip)) {
2038                 disable_dbs(adap);
2039                 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2040         }
2041         queue_work(adap->workq, &adap->db_drop_task);
2042 }
2043
2044 void t4_register_netevent_notifier(void)
2045 {
2046         if (!netevent_registered) {
2047                 register_netevent_notifier(&cxgb4_netevent_nb);
2048                 netevent_registered = true;
2049         }
2050 }
2051
2052 static void detach_ulds(struct adapter *adap)
2053 {
2054         unsigned int i;
2055
2056         mutex_lock(&uld_mutex);
2057         list_del(&adap->list_node);
2058         for (i = 0; i < CXGB4_ULD_MAX; i++)
2059                 if (adap->uld && adap->uld[i].handle) {
2060                         adap->uld[i].state_change(adap->uld[i].handle,
2061                                              CXGB4_STATE_DETACH);
2062                         adap->uld[i].handle = NULL;
2063                 }
2064         if (netevent_registered && list_empty(&adapter_list)) {
2065                 unregister_netevent_notifier(&cxgb4_netevent_nb);
2066                 netevent_registered = false;
2067         }
2068         mutex_unlock(&uld_mutex);
2069 }
2070
2071 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2072 {
2073         unsigned int i;
2074
2075         mutex_lock(&uld_mutex);
2076         for (i = 0; i < CXGB4_ULD_MAX; i++)
2077                 if (adap->uld && adap->uld[i].handle)
2078                         adap->uld[i].state_change(adap->uld[i].handle,
2079                                                   new_state);
2080         mutex_unlock(&uld_mutex);
2081 }
2082
2083 #if IS_ENABLED(CONFIG_IPV6)
2084 static int cxgb4_inet6addr_handler(struct notifier_block *this,
2085                                    unsigned long event, void *data)
2086 {
2087         struct inet6_ifaddr *ifa = data;
2088         struct net_device *event_dev = ifa->idev->dev;
2089         const struct device *parent = NULL;
2090 #if IS_ENABLED(CONFIG_BONDING)
2091         struct adapter *adap;
2092 #endif
2093         if (is_vlan_dev(event_dev))
2094                 event_dev = vlan_dev_real_dev(event_dev);
2095 #if IS_ENABLED(CONFIG_BONDING)
2096         if (event_dev->flags & IFF_MASTER) {
2097                 list_for_each_entry(adap, &adapter_list, list_node) {
2098                         switch (event) {
2099                         case NETDEV_UP:
2100                                 cxgb4_clip_get(adap->port[0],
2101                                                (const u32 *)ifa, 1);
2102                                 break;
2103                         case NETDEV_DOWN:
2104                                 cxgb4_clip_release(adap->port[0],
2105                                                    (const u32 *)ifa, 1);
2106                                 break;
2107                         default:
2108                                 break;
2109                         }
2110                 }
2111                 return NOTIFY_OK;
2112         }
2113 #endif
2114
2115         if (event_dev)
2116                 parent = event_dev->dev.parent;
2117
2118         if (parent && parent->driver == &cxgb4_driver.driver) {
2119                 switch (event) {
2120                 case NETDEV_UP:
2121                         cxgb4_clip_get(event_dev, (const u32 *)ifa, 1);
2122                         break;
2123                 case NETDEV_DOWN:
2124                         cxgb4_clip_release(event_dev, (const u32 *)ifa, 1);
2125                         break;
2126                 default:
2127                         break;
2128                 }
2129         }
2130         return NOTIFY_OK;
2131 }
2132
2133 static bool inet6addr_registered;
2134 static struct notifier_block cxgb4_inet6addr_notifier = {
2135         .notifier_call = cxgb4_inet6addr_handler
2136 };
2137
2138 static void update_clip(const struct adapter *adap)
2139 {
2140         int i;
2141         struct net_device *dev;
2142         int ret;
2143
2144         rcu_read_lock();
2145
2146         for (i = 0; i < MAX_NPORTS; i++) {
2147                 dev = adap->port[i];
2148                 ret = 0;
2149
2150                 if (dev)
2151                         ret = cxgb4_update_root_dev_clip(dev);
2152
2153                 if (ret < 0)
2154                         break;
2155         }
2156         rcu_read_unlock();
2157 }
2158 #endif /* IS_ENABLED(CONFIG_IPV6) */
2159
2160 /**
2161  *      cxgb_up - enable the adapter
2162  *      @adap: adapter being enabled
2163  *
2164  *      Called when the first port is enabled, this function performs the
2165  *      actions necessary to make an adapter operational, such as completing
2166  *      the initialization of HW modules, and enabling interrupts.
2167  *
2168  *      Must be called with the rtnl lock held.
2169  */
2170 static int cxgb_up(struct adapter *adap)
2171 {
2172         int err;
2173
2174         err = setup_sge_queues(adap);
2175         if (err)
2176                 goto out;
2177         err = setup_rss(adap);
2178         if (err)
2179                 goto freeq;
2180
2181         if (adap->flags & USING_MSIX) {
2182                 name_msix_vecs(adap);
2183                 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2184                                   adap->msix_info[0].desc, adap);
2185                 if (err)
2186                         goto irq_err;
2187                 err = request_msix_queue_irqs(adap);
2188                 if (err) {
2189                         free_irq(adap->msix_info[0].vec, adap);
2190                         goto irq_err;
2191                 }
2192         } else {
2193                 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2194                                   (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
2195                                   adap->port[0]->name, adap);
2196                 if (err)
2197                         goto irq_err;
2198         }
2199         enable_rx(adap);
2200         t4_sge_start(adap);
2201         t4_intr_enable(adap);
2202         adap->flags |= FULL_INIT_DONE;
2203         notify_ulds(adap, CXGB4_STATE_UP);
2204 #if IS_ENABLED(CONFIG_IPV6)
2205         update_clip(adap);
2206 #endif
2207         /* Initialize hash mac addr list*/
2208         INIT_LIST_HEAD(&adap->mac_hlist);
2209  out:
2210         return err;
2211  irq_err:
2212         dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2213  freeq:
2214         t4_free_sge_resources(adap);
2215         goto out;
2216 }
2217
2218 static void cxgb_down(struct adapter *adapter)
2219 {
2220         cancel_work_sync(&adapter->tid_release_task);
2221         cancel_work_sync(&adapter->db_full_task);
2222         cancel_work_sync(&adapter->db_drop_task);
2223         adapter->tid_release_task_busy = false;
2224         adapter->tid_release_head = NULL;
2225
2226         t4_sge_stop(adapter);
2227         t4_free_sge_resources(adapter);
2228         adapter->flags &= ~FULL_INIT_DONE;
2229 }
2230
2231 /*
2232  * net_device operations
2233  */
2234 static int cxgb_open(struct net_device *dev)
2235 {
2236         int err;
2237         struct port_info *pi = netdev_priv(dev);
2238         struct adapter *adapter = pi->adapter;
2239
2240         netif_carrier_off(dev);
2241
2242         if (!(adapter->flags & FULL_INIT_DONE)) {
2243                 err = cxgb_up(adapter);
2244                 if (err < 0)
2245                         return err;
2246         }
2247
2248         err = link_start(dev);
2249         if (!err)
2250                 netif_tx_start_all_queues(dev);
2251         return err;
2252 }
2253
2254 static int cxgb_close(struct net_device *dev)
2255 {
2256         struct port_info *pi = netdev_priv(dev);
2257         struct adapter *adapter = pi->adapter;
2258
2259         netif_tx_stop_all_queues(dev);
2260         netif_carrier_off(dev);
2261         return t4_enable_vi(adapter, adapter->pf, pi->viid, false, false);
2262 }
2263
2264 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
2265                 __be32 sip, __be16 sport, __be16 vlan,
2266                 unsigned int queue, unsigned char port, unsigned char mask)
2267 {
2268         int ret;
2269         struct filter_entry *f;
2270         struct adapter *adap;
2271         int i;
2272         u8 *val;
2273
2274         adap = netdev2adap(dev);
2275
2276         /* Adjust stid to correct filter index */
2277         stid -= adap->tids.sftid_base;
2278         stid += adap->tids.nftids;
2279
2280         /* Check to make sure the filter requested is writable ...
2281          */
2282         f = &adap->tids.ftid_tab[stid];
2283         ret = writable_filter(f);
2284         if (ret)
2285                 return ret;
2286
2287         /* Clear out any old resources being used by the filter before
2288          * we start constructing the new filter.
2289          */
2290         if (f->valid)
2291                 clear_filter(adap, f);
2292
2293         /* Clear out filter specifications */
2294         memset(&f->fs, 0, sizeof(struct ch_filter_specification));
2295         f->fs.val.lport = cpu_to_be16(sport);
2296         f->fs.mask.lport  = ~0;
2297         val = (u8 *)&sip;
2298         if ((val[0] | val[1] | val[2] | val[3]) != 0) {
2299                 for (i = 0; i < 4; i++) {
2300                         f->fs.val.lip[i] = val[i];
2301                         f->fs.mask.lip[i] = ~0;
2302                 }
2303                 if (adap->params.tp.vlan_pri_map & PORT_F) {
2304                         f->fs.val.iport = port;
2305                         f->fs.mask.iport = mask;
2306                 }
2307         }
2308
2309         if (adap->params.tp.vlan_pri_map & PROTOCOL_F) {
2310                 f->fs.val.proto = IPPROTO_TCP;
2311                 f->fs.mask.proto = ~0;
2312         }
2313
2314         f->fs.dirsteer = 1;
2315         f->fs.iq = queue;
2316         /* Mark filter as locked */
2317         f->locked = 1;
2318         f->fs.rpttid = 1;
2319
2320         /* Save the actual tid. We need this to get the corresponding
2321          * filter entry structure in filter_rpl.
2322          */
2323         f->tid = stid + adap->tids.ftid_base;
2324         ret = set_filter_wr(adap, stid);
2325         if (ret) {
2326                 clear_filter(adap, f);
2327                 return ret;
2328         }
2329
2330         return 0;
2331 }
2332 EXPORT_SYMBOL(cxgb4_create_server_filter);
2333
2334 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
2335                 unsigned int queue, bool ipv6)
2336 {
2337         struct filter_entry *f;
2338         struct adapter *adap;
2339
2340         adap = netdev2adap(dev);
2341
2342         /* Adjust stid to correct filter index */
2343         stid -= adap->tids.sftid_base;
2344         stid += adap->tids.nftids;
2345
2346         f = &adap->tids.ftid_tab[stid];
2347         /* Unlock the filter */
2348         f->locked = 0;
2349
2350         return delete_filter(adap, stid);
2351 }
2352 EXPORT_SYMBOL(cxgb4_remove_server_filter);
2353
2354 static void cxgb_get_stats(struct net_device *dev,
2355                            struct rtnl_link_stats64 *ns)
2356 {
2357         struct port_stats stats;
2358         struct port_info *p = netdev_priv(dev);
2359         struct adapter *adapter = p->adapter;
2360
2361         /* Block retrieving statistics during EEH error
2362          * recovery. Otherwise, the recovery might fail
2363          * and the PCI device will be removed permanently
2364          */
2365         spin_lock(&adapter->stats_lock);
2366         if (!netif_device_present(dev)) {
2367                 spin_unlock(&adapter->stats_lock);
2368                 return;
2369         }
2370         t4_get_port_stats_offset(adapter, p->tx_chan, &stats,
2371                                  &p->stats_base);
2372         spin_unlock(&adapter->stats_lock);
2373
2374         ns->tx_bytes   = stats.tx_octets;
2375         ns->tx_packets = stats.tx_frames;
2376         ns->rx_bytes   = stats.rx_octets;
2377         ns->rx_packets = stats.rx_frames;
2378         ns->multicast  = stats.rx_mcast_frames;
2379
2380         /* detailed rx_errors */
2381         ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2382                                stats.rx_runt;
2383         ns->rx_over_errors   = 0;
2384         ns->rx_crc_errors    = stats.rx_fcs_err;
2385         ns->rx_frame_errors  = stats.rx_symbol_err;
2386         ns->rx_dropped       = stats.rx_ovflow0 + stats.rx_ovflow1 +
2387                                stats.rx_ovflow2 + stats.rx_ovflow3 +
2388                                stats.rx_trunc0 + stats.rx_trunc1 +
2389                                stats.rx_trunc2 + stats.rx_trunc3;
2390         ns->rx_missed_errors = 0;
2391
2392         /* detailed tx_errors */
2393         ns->tx_aborted_errors   = 0;
2394         ns->tx_carrier_errors   = 0;
2395         ns->tx_fifo_errors      = 0;
2396         ns->tx_heartbeat_errors = 0;
2397         ns->tx_window_errors    = 0;
2398
2399         ns->tx_errors = stats.tx_error_frames;
2400         ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2401                 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2402 }
2403
2404 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2405 {
2406         unsigned int mbox;
2407         int ret = 0, prtad, devad;
2408         struct port_info *pi = netdev_priv(dev);
2409         struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2410
2411         switch (cmd) {
2412         case SIOCGMIIPHY:
2413                 if (pi->mdio_addr < 0)
2414                         return -EOPNOTSUPP;
2415                 data->phy_id = pi->mdio_addr;
2416                 break;
2417         case SIOCGMIIREG:
2418         case SIOCSMIIREG:
2419                 if (mdio_phy_id_is_c45(data->phy_id)) {
2420                         prtad = mdio_phy_id_prtad(data->phy_id);
2421                         devad = mdio_phy_id_devad(data->phy_id);
2422                 } else if (data->phy_id < 32) {
2423                         prtad = data->phy_id;
2424                         devad = 0;
2425                         data->reg_num &= 0x1f;
2426                 } else
2427                         return -EINVAL;
2428
2429                 mbox = pi->adapter->pf;
2430                 if (cmd == SIOCGMIIREG)
2431                         ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
2432                                          data->reg_num, &data->val_out);
2433                 else
2434                         ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
2435                                          data->reg_num, data->val_in);
2436                 break;
2437         case SIOCGHWTSTAMP:
2438                 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2439                                     sizeof(pi->tstamp_config)) ?
2440                         -EFAULT : 0;
2441         case SIOCSHWTSTAMP:
2442                 if (copy_from_user(&pi->tstamp_config, req->ifr_data,
2443                                    sizeof(pi->tstamp_config)))
2444                         return -EFAULT;
2445
2446                 switch (pi->tstamp_config.rx_filter) {
2447                 case HWTSTAMP_FILTER_NONE:
2448                         pi->rxtstamp = false;
2449                         break;
2450                 case HWTSTAMP_FILTER_ALL:
2451                         pi->rxtstamp = true;
2452                         break;
2453                 default:
2454                         pi->tstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
2455                         return -ERANGE;
2456                 }
2457
2458                 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2459                                     sizeof(pi->tstamp_config)) ?
2460                         -EFAULT : 0;
2461         default:
2462                 return -EOPNOTSUPP;
2463         }
2464         return ret;
2465 }
2466
2467 static void cxgb_set_rxmode(struct net_device *dev)
2468 {
2469         /* unfortunately we can't return errors to the stack */
2470         set_rxmode(dev, -1, false);
2471 }
2472
2473 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2474 {
2475         int ret;
2476         struct port_info *pi = netdev_priv(dev);
2477
2478         ret = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, new_mtu, -1,
2479                             -1, -1, -1, true);
2480         if (!ret)
2481                 dev->mtu = new_mtu;
2482         return ret;
2483 }
2484
2485 #ifdef CONFIG_PCI_IOV
2486 static int dummy_open(struct net_device *dev)
2487 {
2488         /* Turn carrier off since we don't have to transmit anything on this
2489          * interface.
2490          */
2491         netif_carrier_off(dev);
2492         return 0;
2493 }
2494
2495 /* Fill MAC address that will be assigned by the FW */
2496 static void fill_vf_station_mac_addr(struct adapter *adap)
2497 {
2498         unsigned int i;
2499         u8 hw_addr[ETH_ALEN], macaddr[ETH_ALEN];
2500         int err;
2501         u8 *na;
2502         u16 a, b;
2503
2504         err = t4_get_raw_vpd_params(adap, &adap->params.vpd);
2505         if (!err) {
2506                 na = adap->params.vpd.na;
2507                 for (i = 0; i < ETH_ALEN; i++)
2508                         hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
2509                                       hex2val(na[2 * i + 1]));
2510                 a = (hw_addr[0] << 8) | hw_addr[1];
2511                 b = (hw_addr[1] << 8) | hw_addr[2];
2512                 a ^= b;
2513                 a |= 0x0200;    /* locally assigned Ethernet MAC address */
2514                 a &= ~0x0100;   /* not a multicast Ethernet MAC address */
2515                 macaddr[0] = a >> 8;
2516                 macaddr[1] = a & 0xff;
2517
2518                 for (i = 2; i < 5; i++)
2519                         macaddr[i] = hw_addr[i + 1];
2520
2521                 for (i = 0; i < adap->num_vfs; i++) {
2522                         macaddr[5] = adap->pf * 16 + i;
2523                         ether_addr_copy(adap->vfinfo[i].vf_mac_addr, macaddr);
2524                 }
2525         }
2526 }
2527
2528 static int cxgb_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
2529 {
2530         struct port_info *pi = netdev_priv(dev);
2531         struct adapter *adap = pi->adapter;
2532         int ret;
2533
2534         /* verify MAC addr is valid */
2535         if (!is_valid_ether_addr(mac)) {
2536                 dev_err(pi->adapter->pdev_dev,
2537                         "Invalid Ethernet address %pM for VF %d\n",
2538                         mac, vf);
2539                 return -EINVAL;
2540         }
2541
2542         dev_info(pi->adapter->pdev_dev,
2543                  "Setting MAC %pM on VF %d\n", mac, vf);
2544         ret = t4_set_vf_mac_acl(adap, vf + 1, 1, mac);
2545         if (!ret)
2546                 ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, mac);
2547         return ret;
2548 }
2549
2550 static int cxgb_get_vf_config(struct net_device *dev,
2551                               int vf, struct ifla_vf_info *ivi)
2552 {
2553         struct port_info *pi = netdev_priv(dev);
2554         struct adapter *adap = pi->adapter;
2555
2556         if (vf >= adap->num_vfs)
2557                 return -EINVAL;
2558         ivi->vf = vf;
2559         ether_addr_copy(ivi->mac, adap->vfinfo[vf].vf_mac_addr);
2560         return 0;
2561 }
2562
2563 static int cxgb_get_phys_port_id(struct net_device *dev,
2564                                  struct netdev_phys_item_id *ppid)
2565 {
2566         struct port_info *pi = netdev_priv(dev);
2567         unsigned int phy_port_id;
2568
2569         phy_port_id = pi->adapter->adap_idx * 10 + pi->port_id;
2570         ppid->id_len = sizeof(phy_port_id);
2571         memcpy(ppid->id, &phy_port_id, ppid->id_len);
2572         return 0;
2573 }
2574
2575 #endif
2576
2577 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2578 {
2579         int ret;
2580         struct sockaddr *addr = p;
2581         struct port_info *pi = netdev_priv(dev);
2582
2583         if (!is_valid_ether_addr(addr->sa_data))
2584                 return -EADDRNOTAVAIL;
2585
2586         ret = t4_change_mac(pi->adapter, pi->adapter->pf, pi->viid,
2587                             pi->xact_addr_filt, addr->sa_data, true, true);
2588         if (ret < 0)
2589                 return ret;
2590
2591         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2592         pi->xact_addr_filt = ret;
2593         return 0;
2594 }
2595
2596 #ifdef CONFIG_NET_POLL_CONTROLLER
2597 static void cxgb_netpoll(struct net_device *dev)
2598 {
2599         struct port_info *pi = netdev_priv(dev);
2600         struct adapter *adap = pi->adapter;
2601
2602         if (adap->flags & USING_MSIX) {
2603                 int i;
2604                 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
2605
2606                 for (i = pi->nqsets; i; i--, rx++)
2607                         t4_sge_intr_msix(0, &rx->rspq);
2608         } else
2609                 t4_intr_handler(adap)(0, adap);
2610 }
2611 #endif
2612
2613 static int cxgb_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
2614 {
2615         struct port_info *pi = netdev_priv(dev);
2616         struct adapter *adap = pi->adapter;
2617         struct sched_class *e;
2618         struct ch_sched_params p;
2619         struct ch_sched_queue qe;
2620         u32 req_rate;
2621         int err = 0;
2622
2623         if (!can_sched(dev))
2624                 return -ENOTSUPP;
2625
2626         if (index < 0 || index > pi->nqsets - 1)
2627                 return -EINVAL;
2628
2629         if (!(adap->flags & FULL_INIT_DONE)) {
2630                 dev_err(adap->pdev_dev,
2631                         "Failed to rate limit on queue %d. Link Down?\n",
2632                         index);
2633                 return -EINVAL;
2634         }
2635
2636         /* Convert from Mbps to Kbps */
2637         req_rate = rate << 10;
2638
2639         /* Max rate is 10 Gbps */
2640         if (req_rate >= SCHED_MAX_RATE_KBPS) {
2641                 dev_err(adap->pdev_dev,
2642                         "Invalid rate %u Mbps, Max rate is %u Gbps\n",
2643                         rate, SCHED_MAX_RATE_KBPS);
2644                 return -ERANGE;
2645         }
2646
2647         /* First unbind the queue from any existing class */
2648         memset(&qe, 0, sizeof(qe));
2649         qe.queue = index;
2650         qe.class = SCHED_CLS_NONE;
2651
2652         err = cxgb4_sched_class_unbind(dev, (void *)(&qe), SCHED_QUEUE);
2653         if (err) {
2654                 dev_err(adap->pdev_dev,
2655                         "Unbinding Queue %d on port %d fail. Err: %d\n",
2656                         index, pi->port_id, err);
2657                 return err;
2658         }
2659
2660         /* Queue already unbound */
2661         if (!req_rate)
2662                 return 0;
2663
2664         /* Fetch any available unused or matching scheduling class */
2665         memset(&p, 0, sizeof(p));
2666         p.type = SCHED_CLASS_TYPE_PACKET;
2667         p.u.params.level    = SCHED_CLASS_LEVEL_CL_RL;
2668         p.u.params.mode     = SCHED_CLASS_MODE_CLASS;
2669         p.u.params.rateunit = SCHED_CLASS_RATEUNIT_BITS;
2670         p.u.params.ratemode = SCHED_CLASS_RATEMODE_ABS;
2671         p.u.params.channel  = pi->tx_chan;
2672         p.u.params.class    = SCHED_CLS_NONE;
2673         p.u.params.minrate  = 0;
2674         p.u.params.maxrate  = req_rate;
2675         p.u.params.weight   = 0;
2676         p.u.params.pktsize  = dev->mtu;
2677
2678         e = cxgb4_sched_class_alloc(dev, &p);
2679         if (!e)
2680                 return -ENOMEM;
2681
2682         /* Bind the queue to a scheduling class */
2683         memset(&qe, 0, sizeof(qe));
2684         qe.queue = index;
2685         qe.class = e->idx;
2686
2687         err = cxgb4_sched_class_bind(dev, (void *)(&qe), SCHED_QUEUE);
2688         if (err)
2689                 dev_err(adap->pdev_dev,
2690                         "Queue rate limiting failed. Err: %d\n", err);
2691         return err;
2692 }
2693
2694 static int cxgb_setup_tc(struct net_device *dev, u32 handle, __be16 proto,
2695                          struct tc_to_netdev *tc)
2696 {
2697         struct port_info *pi = netdev2pinfo(dev);
2698         struct adapter *adap = netdev2adap(dev);
2699
2700         if (!(adap->flags & FULL_INIT_DONE)) {
2701                 dev_err(adap->pdev_dev,
2702                         "Failed to setup tc on port %d. Link Down?\n",
2703                         pi->port_id);
2704                 return -EINVAL;
2705         }
2706
2707         if (TC_H_MAJ(handle) == TC_H_MAJ(TC_H_INGRESS) &&
2708             tc->type == TC_SETUP_CLSU32) {
2709                 switch (tc->cls_u32->command) {
2710                 case TC_CLSU32_NEW_KNODE:
2711                 case TC_CLSU32_REPLACE_KNODE:
2712                         return cxgb4_config_knode(dev, proto, tc->cls_u32);
2713                 case TC_CLSU32_DELETE_KNODE:
2714                         return cxgb4_delete_knode(dev, proto, tc->cls_u32);
2715                 default:
2716                         return -EOPNOTSUPP;
2717                 }
2718         }
2719
2720         return -EOPNOTSUPP;
2721 }
2722
2723 static const struct net_device_ops cxgb4_netdev_ops = {
2724         .ndo_open             = cxgb_open,
2725         .ndo_stop             = cxgb_close,
2726         .ndo_start_xmit       = t4_eth_xmit,
2727         .ndo_select_queue     = cxgb_select_queue,
2728         .ndo_get_stats64      = cxgb_get_stats,
2729         .ndo_set_rx_mode      = cxgb_set_rxmode,
2730         .ndo_set_mac_address  = cxgb_set_mac_addr,
2731         .ndo_set_features     = cxgb_set_features,
2732         .ndo_validate_addr    = eth_validate_addr,
2733         .ndo_do_ioctl         = cxgb_ioctl,
2734         .ndo_change_mtu       = cxgb_change_mtu,
2735 #ifdef CONFIG_NET_POLL_CONTROLLER
2736         .ndo_poll_controller  = cxgb_netpoll,
2737 #endif
2738 #ifdef CONFIG_CHELSIO_T4_FCOE
2739         .ndo_fcoe_enable      = cxgb_fcoe_enable,
2740         .ndo_fcoe_disable     = cxgb_fcoe_disable,
2741 #endif /* CONFIG_CHELSIO_T4_FCOE */
2742         .ndo_set_tx_maxrate   = cxgb_set_tx_maxrate,
2743         .ndo_setup_tc         = cxgb_setup_tc,
2744 };
2745
2746 #ifdef CONFIG_PCI_IOV
2747 static const struct net_device_ops cxgb4_mgmt_netdev_ops = {
2748         .ndo_open             = dummy_open,
2749         .ndo_set_vf_mac       = cxgb_set_vf_mac,
2750         .ndo_get_vf_config    = cxgb_get_vf_config,
2751         .ndo_get_phys_port_id = cxgb_get_phys_port_id,
2752 };
2753 #endif
2754
2755 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2756 {
2757         struct adapter *adapter = netdev2adap(dev);
2758
2759         strlcpy(info->driver, cxgb4_driver_name, sizeof(info->driver));
2760         strlcpy(info->version, cxgb4_driver_version,
2761                 sizeof(info->version));
2762         strlcpy(info->bus_info, pci_name(adapter->pdev),
2763                 sizeof(info->bus_info));
2764 }
2765
2766 static const struct ethtool_ops cxgb4_mgmt_ethtool_ops = {
2767         .get_drvinfo       = get_drvinfo,
2768 };
2769
2770 void t4_fatal_err(struct adapter *adap)
2771 {
2772         int port;
2773
2774         if (pci_channel_offline(adap->pdev))
2775                 return;
2776
2777         /* Disable the SGE since ULDs are going to free resources that
2778          * could be exposed to the adapter.  RDMA MWs for example...
2779          */
2780         t4_shutdown_adapter(adap);
2781         for_each_port(adap, port) {
2782                 struct net_device *dev = adap->port[port];
2783
2784                 /* If we get here in very early initialization the network
2785                  * devices may not have been set up yet.
2786                  */
2787                 if (!dev)
2788                         continue;
2789
2790                 netif_tx_stop_all_queues(dev);
2791                 netif_carrier_off(dev);
2792         }
2793         dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
2794 }
2795
2796 static void setup_memwin(struct adapter *adap)
2797 {
2798         u32 nic_win_base = t4_get_util_window(adap);
2799
2800         t4_setup_memwin(adap, nic_win_base, MEMWIN_NIC);
2801 }
2802
2803 static void setup_memwin_rdma(struct adapter *adap)
2804 {
2805         if (adap->vres.ocq.size) {
2806                 u32 start;
2807                 unsigned int sz_kb;
2808
2809                 start = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_2);
2810                 start &= PCI_BASE_ADDRESS_MEM_MASK;
2811                 start += OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
2812                 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
2813                 t4_write_reg(adap,
2814                              PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN_A, 3),
2815                              start | BIR_V(1) | WINDOW_V(ilog2(sz_kb)));
2816                 t4_write_reg(adap,
2817                              PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3),
2818                              adap->vres.ocq.start);
2819                 t4_read_reg(adap,
2820                             PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3));
2821         }
2822 }
2823
2824 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
2825 {
2826         u32 v;
2827         int ret;
2828
2829         /* get device capabilities */
2830         memset(c, 0, sizeof(*c));
2831         c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
2832                                FW_CMD_REQUEST_F | FW_CMD_READ_F);
2833         c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
2834         ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), c);
2835         if (ret < 0)
2836                 return ret;
2837
2838         c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
2839                                FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
2840         ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), NULL);
2841         if (ret < 0)
2842                 return ret;
2843
2844         ret = t4_config_glbl_rss(adap, adap->pf,
2845                                  FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
2846                                  FW_RSS_GLB_CONFIG_CMD_TNLMAPEN_F |
2847                                  FW_RSS_GLB_CONFIG_CMD_TNLALLLKP_F);
2848         if (ret < 0)
2849                 return ret;
2850
2851         ret = t4_cfg_pfvf(adap, adap->mbox, adap->pf, 0, adap->sge.egr_sz, 64,
2852                           MAX_INGQ, 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF,
2853                           FW_CMD_CAP_PF);
2854         if (ret < 0)
2855                 return ret;
2856
2857         t4_sge_init(adap);
2858
2859         /* tweak some settings */
2860         t4_write_reg(adap, TP_SHIFT_CNT_A, 0x64f8849);
2861         t4_write_reg(adap, ULP_RX_TDDP_PSZ_A, HPZ0_V(PAGE_SHIFT - 12));
2862         t4_write_reg(adap, TP_PIO_ADDR_A, TP_INGRESS_CONFIG_A);
2863         v = t4_read_reg(adap, TP_PIO_DATA_A);
2864         t4_write_reg(adap, TP_PIO_DATA_A, v & ~CSUM_HAS_PSEUDO_HDR_F);
2865
2866         /* first 4 Tx modulation queues point to consecutive Tx channels */
2867         adap->params.tp.tx_modq_map = 0xE4;
2868         t4_write_reg(adap, TP_TX_MOD_QUEUE_REQ_MAP_A,
2869                      TX_MOD_QUEUE_REQ_MAP_V(adap->params.tp.tx_modq_map));
2870
2871         /* associate each Tx modulation queue with consecutive Tx channels */
2872         v = 0x84218421;
2873         t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
2874                           &v, 1, TP_TX_SCHED_HDR_A);
2875         t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
2876                           &v, 1, TP_TX_SCHED_FIFO_A);
2877         t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
2878                           &v, 1, TP_TX_SCHED_PCMD_A);
2879
2880 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
2881         if (is_offload(adap)) {
2882                 t4_write_reg(adap, TP_TX_MOD_QUEUE_WEIGHT0_A,
2883                              TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2884                              TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2885                              TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2886                              TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
2887                 t4_write_reg(adap, TP_TX_MOD_CHANNEL_WEIGHT_A,
2888                              TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2889                              TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2890                              TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2891                              TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
2892         }
2893
2894         /* get basic stuff going */
2895         return t4_early_init(adap, adap->pf);
2896 }
2897
2898 /*
2899  * Max # of ATIDs.  The absolute HW max is 16K but we keep it lower.
2900  */
2901 #define MAX_ATIDS 8192U
2902
2903 /*
2904  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
2905  *
2906  * If the firmware we're dealing with has Configuration File support, then
2907  * we use that to perform all configuration
2908  */
2909
2910 /*
2911  * Tweak configuration based on module parameters, etc.  Most of these have
2912  * defaults assigned to them by Firmware Configuration Files (if we're using
2913  * them) but need to be explicitly set if we're using hard-coded
2914  * initialization.  But even in the case of using Firmware Configuration
2915  * Files, we'd like to expose the ability to change these via module
2916  * parameters so these are essentially common tweaks/settings for
2917  * Configuration Files and hard-coded initialization ...
2918  */
2919 static int adap_init0_tweaks(struct adapter *adapter)
2920 {
2921         /*
2922          * Fix up various Host-Dependent Parameters like Page Size, Cache
2923          * Line Size, etc.  The firmware default is for a 4KB Page Size and
2924          * 64B Cache Line Size ...
2925          */
2926         t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
2927
2928         /*
2929          * Process module parameters which affect early initialization.
2930          */
2931         if (rx_dma_offset != 2 && rx_dma_offset != 0) {
2932                 dev_err(&adapter->pdev->dev,
2933                         "Ignoring illegal rx_dma_offset=%d, using 2\n",
2934                         rx_dma_offset);
2935                 rx_dma_offset = 2;
2936         }
2937         t4_set_reg_field(adapter, SGE_CONTROL_A,
2938                          PKTSHIFT_V(PKTSHIFT_M),
2939                          PKTSHIFT_V(rx_dma_offset));
2940
2941         /*
2942          * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
2943          * adds the pseudo header itself.
2944          */
2945         t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG_A,
2946                                CSUM_HAS_PSEUDO_HDR_F, 0);
2947
2948         return 0;
2949 }
2950
2951 /* 10Gb/s-BT PHY Support. chip-external 10Gb/s-BT PHYs are complex chips
2952  * unto themselves and they contain their own firmware to perform their
2953  * tasks ...
2954  */
2955 static int phy_aq1202_version(const u8 *phy_fw_data,
2956                               size_t phy_fw_size)
2957 {
2958         int offset;
2959
2960         /* At offset 0x8 you're looking for the primary image's
2961          * starting offset which is 3 Bytes wide
2962          *
2963          * At offset 0xa of the primary image, you look for the offset
2964          * of the DRAM segment which is 3 Bytes wide.
2965          *
2966          * The FW version is at offset 0x27e of the DRAM and is 2 Bytes
2967          * wide
2968          */
2969         #define be16(__p) (((__p)[0] << 8) | (__p)[1])
2970         #define le16(__p) ((__p)[0] | ((__p)[1] << 8))
2971         #define le24(__p) (le16(__p) | ((__p)[2] << 16))
2972
2973         offset = le24(phy_fw_data + 0x8) << 12;
2974         offset = le24(phy_fw_data + offset + 0xa);
2975         return be16(phy_fw_data + offset + 0x27e);
2976
2977         #undef be16
2978         #undef le16
2979         #undef le24
2980 }
2981
2982 static struct info_10gbt_phy_fw {
2983         unsigned int phy_fw_id;         /* PCI Device ID */
2984         char *phy_fw_file;              /* /lib/firmware/ PHY Firmware file */
2985         int (*phy_fw_version)(const u8 *phy_fw_data, size_t phy_fw_size);
2986         int phy_flash;                  /* Has FLASH for PHY Firmware */
2987 } phy_info_array[] = {
2988         {
2989                 PHY_AQ1202_DEVICEID,
2990                 PHY_AQ1202_FIRMWARE,
2991                 phy_aq1202_version,
2992                 1,
2993         },
2994         {
2995                 PHY_BCM84834_DEVICEID,
2996                 PHY_BCM84834_FIRMWARE,
2997                 NULL,
2998                 0,
2999         },
3000         { 0, NULL, NULL },
3001 };
3002
3003 static struct info_10gbt_phy_fw *find_phy_info(int devid)
3004 {
3005         int i;
3006
3007         for (i = 0; i < ARRAY_SIZE(phy_info_array); i++) {
3008                 if (phy_info_array[i].phy_fw_id == devid)
3009                         return &phy_info_array[i];
3010         }
3011         return NULL;
3012 }
3013
3014 /* Handle updating of chip-external 10Gb/s-BT PHY firmware.  This needs to
3015  * happen after the FW_RESET_CMD but before the FW_INITIALIZE_CMD.  On error
3016  * we return a negative error number.  If we transfer new firmware we return 1
3017  * (from t4_load_phy_fw()).  If we don't do anything we return 0.
3018  */
3019 static int adap_init0_phy(struct adapter *adap)
3020 {
3021         const struct firmware *phyf;
3022         int ret;
3023         struct info_10gbt_phy_fw *phy_info;
3024
3025         /* Use the device ID to determine which PHY file to flash.
3026          */
3027         phy_info = find_phy_info(adap->pdev->device);
3028         if (!phy_info) {
3029                 dev_warn(adap->pdev_dev,
3030                          "No PHY Firmware file found for this PHY\n");
3031                 return -EOPNOTSUPP;
3032         }
3033
3034         /* If we have a T4 PHY firmware file under /lib/firmware/cxgb4/, then
3035          * use that. The adapter firmware provides us with a memory buffer
3036          * where we can load a PHY firmware file from the host if we want to
3037          * override the PHY firmware File in flash.
3038          */
3039         ret = request_firmware_direct(&phyf, phy_info->phy_fw_file,
3040                                       adap->pdev_dev);
3041         if (ret < 0) {
3042                 /* For adapters without FLASH attached to PHY for their
3043                  * firmware, it's obviously a fatal error if we can't get the
3044                  * firmware to the adapter.  For adapters with PHY firmware
3045                  * FLASH storage, it's worth a warning if we can't find the
3046                  * PHY Firmware but we'll neuter the error ...
3047                  */
3048                 dev_err(adap->pdev_dev, "unable to find PHY Firmware image "
3049                         "/lib/firmware/%s, error %d\n",
3050                         phy_info->phy_fw_file, -ret);
3051                 if (phy_info->phy_flash) {
3052                         int cur_phy_fw_ver = 0;
3053
3054                         t4_phy_fw_ver(adap, &cur_phy_fw_ver);
3055                         dev_warn(adap->pdev_dev, "continuing with, on-adapter "
3056                                  "FLASH copy, version %#x\n", cur_phy_fw_ver);
3057                         ret = 0;
3058                 }
3059
3060                 return ret;
3061         }
3062
3063         /* Load PHY Firmware onto adapter.
3064          */
3065         ret = t4_load_phy_fw(adap, MEMWIN_NIC, &adap->win0_lock,
3066                              phy_info->phy_fw_version,
3067                              (u8 *)phyf->data, phyf->size);
3068         if (ret < 0)
3069                 dev_err(adap->pdev_dev, "PHY Firmware transfer error %d\n",
3070                         -ret);
3071         else if (ret > 0) {
3072                 int new_phy_fw_ver = 0;
3073
3074                 if (phy_info->phy_fw_version)
3075                         new_phy_fw_ver = phy_info->phy_fw_version(phyf->data,
3076                                                                   phyf->size);
3077                 dev_info(adap->pdev_dev, "Successfully transferred PHY "
3078                          "Firmware /lib/firmware/%s, version %#x\n",
3079                          phy_info->phy_fw_file, new_phy_fw_ver);
3080         }
3081
3082         release_firmware(phyf);
3083
3084         return ret;
3085 }
3086
3087 /*
3088  * Attempt to initialize the adapter via a Firmware Configuration File.
3089  */
3090 static int adap_init0_config(struct adapter *adapter, int reset)
3091 {
3092         struct fw_caps_config_cmd caps_cmd;
3093         const struct firmware *cf;
3094         unsigned long mtype = 0, maddr = 0;
3095         u32 finiver, finicsum, cfcsum;
3096         int ret;
3097         int config_issued = 0;
3098         char *fw_config_file, fw_config_file_path[256];
3099         char *config_name = NULL;
3100
3101         /*
3102          * Reset device if necessary.
3103          */
3104         if (reset) {
3105                 ret = t4_fw_reset(adapter, adapter->mbox,
3106                                   PIORSTMODE_F | PIORST_F);
3107                 if (ret < 0)
3108                         goto bye;
3109         }
3110
3111         /* If this is a 10Gb/s-BT adapter make sure the chip-external
3112          * 10Gb/s-BT PHYs have up-to-date firmware.  Note that this step needs
3113          * to be performed after any global adapter RESET above since some
3114          * PHYs only have local RAM copies of the PHY firmware.
3115          */
3116         if (is_10gbt_device(adapter->pdev->device)) {
3117                 ret = adap_init0_phy(adapter);
3118                 if (ret < 0)
3119                         goto bye;
3120         }
3121         /*
3122          * If we have a T4 configuration file under /lib/firmware/cxgb4/,
3123          * then use that.  Otherwise, use the configuration file stored
3124          * in the adapter flash ...
3125          */
3126         switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
3127         case CHELSIO_T4:
3128                 fw_config_file = FW4_CFNAME;
3129                 break;
3130         case CHELSIO_T5:
3131                 fw_config_file = FW5_CFNAME;
3132                 break;
3133         case CHELSIO_T6:
3134                 fw_config_file = FW6_CFNAME;
3135                 break;
3136         default:
3137                 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
3138                        adapter->pdev->device);
3139                 ret = -EINVAL;
3140                 goto bye;
3141         }
3142
3143         ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
3144         if (ret < 0) {
3145                 config_name = "On FLASH";
3146                 mtype = FW_MEMTYPE_CF_FLASH;
3147                 maddr = t4_flash_cfg_addr(adapter);
3148         } else {
3149                 u32 params[7], val[7];
3150
3151                 sprintf(fw_config_file_path,
3152                         "/lib/firmware/%s", fw_config_file);
3153                 config_name = fw_config_file_path;
3154
3155                 if (cf->size >= FLASH_CFG_MAX_SIZE)
3156                         ret = -ENOMEM;
3157                 else {
3158                         params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3159                              FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3160                         ret = t4_query_params(adapter, adapter->mbox,
3161                                               adapter->pf, 0, 1, params, val);
3162                         if (ret == 0) {
3163                                 /*
3164                                  * For t4_memory_rw() below addresses and
3165                                  * sizes have to be in terms of multiples of 4
3166                                  * bytes.  So, if the Configuration File isn't
3167                                  * a multiple of 4 bytes in length we'll have
3168                                  * to write that out separately since we can't
3169                                  * guarantee that the bytes following the
3170                                  * residual byte in the buffer returned by
3171                                  * request_firmware() are zeroed out ...
3172                                  */
3173                                 size_t resid = cf->size & 0x3;
3174                                 size_t size = cf->size & ~0x3;
3175                                 __be32 *data = (__be32 *)cf->data;
3176
3177                                 mtype = FW_PARAMS_PARAM_Y_G(val[0]);
3178                                 maddr = FW_PARAMS_PARAM_Z_G(val[0]) << 16;
3179
3180                                 spin_lock(&adapter->win0_lock);
3181                                 ret = t4_memory_rw(adapter, 0, mtype, maddr,
3182                                                    size, data, T4_MEMORY_WRITE);
3183                                 if (ret == 0 && resid != 0) {
3184                                         union {
3185                                                 __be32 word;
3186                                                 char buf[4];
3187                                         } last;
3188                                         int i;
3189
3190                                         last.word = data[size >> 2];
3191                                         for (i = resid; i < 4; i++)
3192                                                 last.buf[i] = 0;
3193                                         ret = t4_memory_rw(adapter, 0, mtype,
3194                                                            maddr + size,
3195                                                            4, &last.word,
3196                                                            T4_MEMORY_WRITE);
3197                                 }
3198                                 spin_unlock(&adapter->win0_lock);
3199                         }
3200                 }
3201
3202                 release_firmware(cf);
3203                 if (ret)
3204                         goto bye;
3205         }
3206
3207         /*
3208          * Issue a Capability Configuration command to the firmware to get it
3209          * to parse the Configuration File.  We don't use t4_fw_config_file()
3210          * because we want the ability to modify various features after we've
3211          * processed the configuration file ...
3212          */
3213         memset(&caps_cmd, 0, sizeof(caps_cmd));
3214         caps_cmd.op_to_write =
3215                 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3216                       FW_CMD_REQUEST_F |
3217                       FW_CMD_READ_F);
3218         caps_cmd.cfvalid_to_len16 =
3219                 htonl(FW_CAPS_CONFIG_CMD_CFVALID_F |
3220                       FW_CAPS_CONFIG_CMD_MEMTYPE_CF_V(mtype) |
3221                       FW_CAPS_CONFIG_CMD_MEMADDR64K_CF_V(maddr >> 16) |
3222                       FW_LEN16(caps_cmd));
3223         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3224                          &caps_cmd);
3225
3226         /* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
3227          * Configuration File in FLASH), our last gasp effort is to use the
3228          * Firmware Configuration File which is embedded in the firmware.  A
3229          * very few early versions of the firmware didn't have one embedded
3230          * but we can ignore those.
3231          */
3232         if (ret == -ENOENT) {
3233                 memset(&caps_cmd, 0, sizeof(caps_cmd));
3234                 caps_cmd.op_to_write =
3235                         htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3236                                         FW_CMD_REQUEST_F |
3237                                         FW_CMD_READ_F);
3238                 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3239                 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
3240                                 sizeof(caps_cmd), &caps_cmd);
3241                 config_name = "Firmware Default";
3242         }
3243
3244         config_issued = 1;
3245         if (ret < 0)
3246                 goto bye;
3247
3248         finiver = ntohl(caps_cmd.finiver);
3249         finicsum = ntohl(caps_cmd.finicsum);
3250         cfcsum = ntohl(caps_cmd.cfcsum);
3251         if (finicsum != cfcsum)
3252                 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
3253                          "mismatch: [fini] csum=%#x, computed csum=%#x\n",
3254                          finicsum, cfcsum);
3255
3256         /*
3257          * And now tell the firmware to use the configuration we just loaded.
3258          */
3259         caps_cmd.op_to_write =
3260                 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3261                       FW_CMD_REQUEST_F |
3262                       FW_CMD_WRITE_F);
3263         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3264         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3265                          NULL);
3266         if (ret < 0)
3267                 goto bye;
3268
3269         /*
3270          * Tweak configuration based on system architecture, module
3271          * parameters, etc.
3272          */
3273         ret = adap_init0_tweaks(adapter);
3274         if (ret < 0)
3275                 goto bye;
3276
3277         /*
3278          * And finally tell the firmware to initialize itself using the
3279          * parameters from the Configuration File.
3280          */
3281         ret = t4_fw_initialize(adapter, adapter->mbox);
3282         if (ret < 0)
3283                 goto bye;
3284
3285         /* Emit Firmware Configuration File information and return
3286          * successfully.
3287          */
3288         dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
3289                  "Configuration File \"%s\", version %#x, computed checksum %#x\n",
3290                  config_name, finiver, cfcsum);
3291         return 0;
3292
3293         /*
3294          * Something bad happened.  Return the error ...  (If the "error"
3295          * is that there's no Configuration File on the adapter we don't
3296          * want to issue a warning since this is fairly common.)
3297          */
3298 bye:
3299         if (config_issued && ret != -ENOENT)
3300                 dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
3301                          config_name, -ret);
3302         return ret;
3303 }
3304
3305 static struct fw_info fw_info_array[] = {
3306         {
3307                 .chip = CHELSIO_T4,
3308                 .fs_name = FW4_CFNAME,
3309                 .fw_mod_name = FW4_FNAME,
3310                 .fw_hdr = {
3311                         .chip = FW_HDR_CHIP_T4,
3312                         .fw_ver = __cpu_to_be32(FW_VERSION(T4)),
3313                         .intfver_nic = FW_INTFVER(T4, NIC),
3314                         .intfver_vnic = FW_INTFVER(T4, VNIC),
3315                         .intfver_ri = FW_INTFVER(T4, RI),
3316                         .intfver_iscsi = FW_INTFVER(T4, ISCSI),
3317                         .intfver_fcoe = FW_INTFVER(T4, FCOE),
3318                 },
3319         }, {
3320                 .chip = CHELSIO_T5,
3321                 .fs_name = FW5_CFNAME,
3322                 .fw_mod_name = FW5_FNAME,
3323                 .fw_hdr = {
3324                         .chip = FW_HDR_CHIP_T5,
3325                         .fw_ver = __cpu_to_be32(FW_VERSION(T5)),
3326                         .intfver_nic = FW_INTFVER(T5, NIC),
3327                         .intfver_vnic = FW_INTFVER(T5, VNIC),
3328                         .intfver_ri = FW_INTFVER(T5, RI),
3329                         .intfver_iscsi = FW_INTFVER(T5, ISCSI),
3330                         .intfver_fcoe = FW_INTFVER(T5, FCOE),
3331                 },
3332         }, {
3333                 .chip = CHELSIO_T6,
3334                 .fs_name = FW6_CFNAME,
3335                 .fw_mod_name = FW6_FNAME,
3336                 .fw_hdr = {
3337                         .chip = FW_HDR_CHIP_T6,
3338                         .fw_ver = __cpu_to_be32(FW_VERSION(T6)),
3339                         .intfver_nic = FW_INTFVER(T6, NIC),
3340                         .intfver_vnic = FW_INTFVER(T6, VNIC),
3341                         .intfver_ofld = FW_INTFVER(T6, OFLD),
3342                         .intfver_ri = FW_INTFVER(T6, RI),
3343                         .intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3344                         .intfver_iscsi = FW_INTFVER(T6, ISCSI),
3345                         .intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3346                         .intfver_fcoe = FW_INTFVER(T6, FCOE),
3347                 },
3348         }
3349
3350 };
3351
3352 static struct fw_info *find_fw_info(int chip)
3353 {
3354         int i;
3355
3356         for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
3357                 if (fw_info_array[i].chip == chip)
3358                         return &fw_info_array[i];
3359         }
3360         return NULL;
3361 }
3362
3363 /*
3364  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3365  */
3366 static int adap_init0(struct adapter *adap)
3367 {
3368         int ret;
3369         u32 v, port_vec;
3370         enum dev_state state;
3371         u32 params[7], val[7];
3372         struct fw_caps_config_cmd caps_cmd;
3373         int reset = 1;
3374
3375         /* Grab Firmware Device Log parameters as early as possible so we have
3376          * access to it for debugging, etc.
3377          */
3378         ret = t4_init_devlog_params(adap);
3379         if (ret < 0)
3380                 return ret;
3381
3382         /* Contact FW, advertising Master capability */
3383         ret = t4_fw_hello(adap, adap->mbox, adap->mbox,
3384                           is_kdump_kernel() ? MASTER_MUST : MASTER_MAY, &state);
3385         if (ret < 0) {
3386                 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
3387                         ret);
3388                 return ret;
3389         }
3390         if (ret == adap->mbox)
3391                 adap->flags |= MASTER_PF;
3392
3393         /*
3394          * If we're the Master PF Driver and the device is uninitialized,
3395          * then let's consider upgrading the firmware ...  (We always want
3396          * to check the firmware version number in order to A. get it for
3397          * later reporting and B. to warn if the currently loaded firmware
3398          * is excessively mismatched relative to the driver.)
3399          */
3400         t4_get_fw_version(adap, &adap->params.fw_vers);
3401         t4_get_bs_version(adap, &adap->params.bs_vers);
3402         t4_get_tp_version(adap, &adap->params.tp_vers);
3403         t4_get_exprom_version(adap, &adap->params.er_vers);
3404
3405         ret = t4_check_fw_version(adap);
3406         /* If firmware is too old (not supported by driver) force an update. */
3407         if (ret)
3408                 state = DEV_STATE_UNINIT;
3409         if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
3410                 struct fw_info *fw_info;
3411                 struct fw_hdr *card_fw;
3412                 const struct firmware *fw;
3413                 const u8 *fw_data = NULL;
3414                 unsigned int fw_size = 0;
3415
3416                 /* This is the firmware whose headers the driver was compiled
3417                  * against
3418                  */
3419                 fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
3420                 if (fw_info == NULL) {
3421                         dev_err(adap->pdev_dev,
3422                                 "unable to get firmware info for chip %d.\n",
3423                                 CHELSIO_CHIP_VERSION(adap->params.chip));
3424                         return -EINVAL;
3425                 }
3426
3427                 /* allocate memory to read the header of the firmware on the
3428                  * card
3429                  */
3430                 card_fw = kvzalloc(sizeof(*card_fw), GFP_KERNEL);
3431
3432                 /* Get FW from from /lib/firmware/ */
3433                 ret = request_firmware(&fw, fw_info->fw_mod_name,
3434                                        adap->pdev_dev);
3435                 if (ret < 0) {
3436                         dev_err(adap->pdev_dev,
3437                                 "unable to load firmware image %s, error %d\n",
3438                                 fw_info->fw_mod_name, ret);
3439                 } else {
3440                         fw_data = fw->data;
3441                         fw_size = fw->size;
3442                 }
3443
3444                 /* upgrade FW logic */
3445                 ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
3446                                  state, &reset);
3447
3448                 /* Cleaning up */
3449                 release_firmware(fw);
3450                 kvfree(card_fw);
3451
3452                 if (ret < 0)
3453                         goto bye;
3454         }
3455
3456         /*
3457          * Grab VPD parameters.  This should be done after we establish a
3458          * connection to the firmware since some of the VPD parameters
3459          * (notably the Core Clock frequency) are retrieved via requests to
3460          * the firmware.  On the other hand, we need these fairly early on
3461          * so we do this right after getting ahold of the firmware.
3462          */
3463         ret = t4_get_vpd_params(adap, &adap->params.vpd);
3464         if (ret < 0)
3465                 goto bye;
3466
3467         /*
3468          * Find out what ports are available to us.  Note that we need to do
3469          * this before calling adap_init0_no_config() since it needs nports
3470          * and portvec ...
3471          */
3472         v =
3473             FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3474             FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
3475         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
3476         if (ret < 0)
3477                 goto bye;
3478
3479         adap->params.nports = hweight32(port_vec);
3480         adap->params.portvec = port_vec;
3481
3482         /* If the firmware is initialized already, emit a simply note to that
3483          * effect. Otherwise, it's time to try initializing the adapter.
3484          */
3485         if (state == DEV_STATE_INIT) {
3486                 dev_info(adap->pdev_dev, "Coming up as %s: "\
3487                          "Adapter already initialized\n",
3488                          adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
3489         } else {
3490                 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
3491                          "Initializing adapter\n");
3492
3493                 /* Find out whether we're dealing with a version of the
3494                  * firmware which has configuration file support.
3495                  */
3496                 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3497                              FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3498                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
3499                                       params, val);
3500
3501                 /* If the firmware doesn't support Configuration Files,
3502                  * return an error.
3503                  */
3504                 if (ret < 0) {
3505                         dev_err(adap->pdev_dev, "firmware doesn't support "
3506                                 "Firmware Configuration Files\n");
3507                         goto bye;
3508                 }
3509
3510                 /* The firmware provides us with a memory buffer where we can
3511                  * load a Configuration File from the host if we want to
3512                  * override the Configuration File in flash.
3513                  */
3514                 ret = adap_init0_config(adap, reset);
3515                 if (ret == -ENOENT) {
3516                         dev_err(adap->pdev_dev, "no Configuration File "
3517                                 "present on adapter.\n");
3518                         goto bye;
3519                 }
3520                 if (ret < 0) {
3521                         dev_err(adap->pdev_dev, "could not initialize "
3522                                 "adapter, error %d\n", -ret);
3523                         goto bye;
3524                 }
3525         }
3526
3527         /* Give the SGE code a chance to pull in anything that it needs ...
3528          * Note that this must be called after we retrieve our VPD parameters
3529          * in order to know how to convert core ticks to seconds, etc.
3530          */
3531         ret = t4_sge_init(adap);
3532         if (ret < 0)
3533                 goto bye;
3534
3535         if (is_bypass_device(adap->pdev->device))
3536                 adap->params.bypass = 1;
3537
3538         /*
3539          * Grab some of our basic fundamental operating parameters.
3540          */
3541 #define FW_PARAM_DEV(param) \
3542         (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) | \
3543         FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_##param))
3544
3545 #define FW_PARAM_PFVF(param) \
3546         FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | \
3547         FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_##param)|  \
3548         FW_PARAMS_PARAM_Y_V(0) | \
3549         FW_PARAMS_PARAM_Z_V(0)
3550
3551         params[0] = FW_PARAM_PFVF(EQ_START);
3552         params[1] = FW_PARAM_PFVF(L2T_START);
3553         params[2] = FW_PARAM_PFVF(L2T_END);
3554         params[3] = FW_PARAM_PFVF(FILTER_START);
3555         params[4] = FW_PARAM_PFVF(FILTER_END);
3556         params[5] = FW_PARAM_PFVF(IQFLINT_START);
3557         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params, val);
3558         if (ret < 0)
3559                 goto bye;
3560         adap->sge.egr_start = val[0];
3561         adap->l2t_start = val[1];
3562         adap->l2t_end = val[2];
3563         adap->tids.ftid_base = val[3];
3564         adap->tids.nftids = val[4] - val[3] + 1;
3565         adap->sge.ingr_start = val[5];
3566
3567         /* qids (ingress/egress) returned from firmware can be anywhere
3568          * in the range from EQ(IQFLINT)_START to EQ(IQFLINT)_END.
3569          * Hence driver needs to allocate memory for this range to
3570          * store the queue info. Get the highest IQFLINT/EQ index returned
3571          * in FW_EQ_*_CMD.alloc command.
3572          */
3573         params[0] = FW_PARAM_PFVF(EQ_END);
3574         params[1] = FW_PARAM_PFVF(IQFLINT_END);
3575         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3576         if (ret < 0)
3577                 goto bye;
3578         adap->sge.egr_sz = val[0] - adap->sge.egr_start + 1;
3579         adap->sge.ingr_sz = val[1] - adap->sge.ingr_start + 1;
3580
3581         adap->sge.egr_map = kcalloc(adap->sge.egr_sz,
3582                                     sizeof(*adap->sge.egr_map), GFP_KERNEL);
3583         if (!adap->sge.egr_map) {
3584                 ret = -ENOMEM;
3585                 goto bye;
3586         }
3587
3588         adap->sge.ingr_map = kcalloc(adap->sge.ingr_sz,
3589                                      sizeof(*adap->sge.ingr_map), GFP_KERNEL);
3590         if (!adap->sge.ingr_map) {
3591                 ret = -ENOMEM;
3592                 goto bye;
3593         }
3594
3595         /* Allocate the memory for the vaious egress queue bitmaps
3596          * ie starving_fl, txq_maperr and blocked_fl.
3597          */
3598         adap->sge.starving_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3599                                         sizeof(long), GFP_KERNEL);
3600         if (!adap->sge.starving_fl) {
3601                 ret = -ENOMEM;
3602                 goto bye;
3603         }
3604
3605         adap->sge.txq_maperr = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3606                                        sizeof(long), GFP_KERNEL);
3607         if (!adap->sge.txq_maperr) {
3608                 ret = -ENOMEM;
3609                 goto bye;
3610         }
3611
3612 #ifdef CONFIG_DEBUG_FS
3613         adap->sge.blocked_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3614                                        sizeof(long), GFP_KERNEL);
3615         if (!adap->sge.blocked_fl) {
3616                 ret = -ENOMEM;
3617                 goto bye;
3618         }
3619 #endif
3620
3621         params[0] = FW_PARAM_PFVF(CLIP_START);
3622         params[1] = FW_PARAM_PFVF(CLIP_END);
3623         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3624         if (ret < 0)
3625                 goto bye;
3626         adap->clipt_start = val[0];
3627         adap->clipt_end = val[1];
3628
3629         /* We don't yet have a PARAMs calls to retrieve the number of Traffic
3630          * Classes supported by the hardware/firmware so we hard code it here
3631          * for now.
3632          */
3633         adap->params.nsched_cls = is_t4(adap->params.chip) ? 15 : 16;
3634
3635         /* query params related to active filter region */
3636         params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
3637         params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
3638         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3639         /* If Active filter size is set we enable establishing
3640          * offload connection through firmware work request
3641          */
3642         if ((val[0] != val[1]) && (ret >= 0)) {
3643                 adap->flags |= FW_OFLD_CONN;
3644                 adap->tids.aftid_base = val[0];
3645                 adap->tids.aftid_end = val[1];
3646         }
3647
3648         /* If we're running on newer firmware, let it know that we're
3649          * prepared to deal with encapsulated CPL messages.  Older
3650          * firmware won't understand this and we'll just get
3651          * unencapsulated messages ...
3652          */
3653         params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3654         val[0] = 1;
3655         (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
3656
3657         /*
3658          * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
3659          * capability.  Earlier versions of the firmware didn't have the
3660          * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
3661          * permission to use ULPTX MEMWRITE DSGL.
3662          */
3663         if (is_t4(adap->params.chip)) {
3664                 adap->params.ulptx_memwrite_dsgl = false;
3665         } else {
3666                 params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
3667                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
3668                                       1, params, val);
3669                 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
3670         }
3671
3672         /* See if FW supports FW_RI_FR_NSMR_TPTE_WR work request */
3673         params[0] = FW_PARAM_DEV(RI_FR_NSMR_TPTE_WR);
3674         ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
3675                               1, params, val);
3676         adap->params.fr_nsmr_tpte_wr_support = (ret == 0 && val[0] != 0);
3677
3678         /*
3679          * Get device capabilities so we can determine what resources we need
3680          * to manage.
3681          */
3682         memset(&caps_cmd, 0, sizeof(caps_cmd));
3683         caps_cmd.op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3684                                      FW_CMD_REQUEST_F | FW_CMD_READ_F);
3685         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3686         ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
3687                          &caps_cmd);
3688         if (ret < 0)
3689                 goto bye;
3690
3691         if (caps_cmd.ofldcaps) {
3692                 /* query offload-related parameters */
3693                 params[0] = FW_PARAM_DEV(NTID);
3694                 params[1] = FW_PARAM_PFVF(SERVER_START);
3695                 params[2] = FW_PARAM_PFVF(SERVER_END);
3696                 params[3] = FW_PARAM_PFVF(TDDP_START);
3697                 params[4] = FW_PARAM_PFVF(TDDP_END);
3698                 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3699                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
3700                                       params, val);
3701                 if (ret < 0)
3702                         goto bye;
3703                 adap->tids.ntids = val[0];
3704                 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
3705                 adap->tids.stid_base = val[1];
3706                 adap->tids.nstids = val[2] - val[1] + 1;
3707                 /*
3708                  * Setup server filter region. Divide the available filter
3709                  * region into two parts. Regular filters get 1/3rd and server
3710                  * filters get 2/3rd part. This is only enabled if workarond
3711                  * path is enabled.
3712                  * 1. For regular filters.
3713                  * 2. Server filter: This are special filters which are used
3714                  * to redirect SYN packets to offload queue.
3715                  */
3716                 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
3717                         adap->tids.sftid_base = adap->tids.ftid_base +
3718                                         DIV_ROUND_UP(adap->tids.nftids, 3);
3719                         adap->tids.nsftids = adap->tids.nftids -
3720                                          DIV_ROUND_UP(adap->tids.nftids, 3);
3721                         adap->tids.nftids = adap->tids.sftid_base -
3722                                                 adap->tids.ftid_base;
3723                 }
3724                 adap->vres.ddp.start = val[3];
3725                 adap->vres.ddp.size = val[4] - val[3] + 1;
3726                 adap->params.ofldq_wr_cred = val[5];
3727
3728                 adap->params.offload = 1;
3729                 adap->num_ofld_uld += 1;
3730         }
3731         if (caps_cmd.rdmacaps) {
3732                 params[0] = FW_PARAM_PFVF(STAG_START);
3733                 params[1] = FW_PARAM_PFVF(STAG_END);
3734                 params[2] = FW_PARAM_PFVF(RQ_START);
3735                 params[3] = FW_PARAM_PFVF(RQ_END);
3736                 params[4] = FW_PARAM_PFVF(PBL_START);
3737                 params[5] = FW_PARAM_PFVF(PBL_END);
3738                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
3739                                       params, val);
3740                 if (ret < 0)
3741                         goto bye;
3742                 adap->vres.stag.start = val[0];
3743                 adap->vres.stag.size = val[1] - val[0] + 1;
3744                 adap->vres.rq.start = val[2];
3745                 adap->vres.rq.size = val[3] - val[2] + 1;
3746                 adap->vres.pbl.start = val[4];
3747                 adap->vres.pbl.size = val[5] - val[4] + 1;
3748
3749                 params[0] = FW_PARAM_PFVF(SQRQ_START);
3750                 params[1] = FW_PARAM_PFVF(SQRQ_END);
3751                 params[2] = FW_PARAM_PFVF(CQ_START);
3752                 params[3] = FW_PARAM_PFVF(CQ_END);
3753                 params[4] = FW_PARAM_PFVF(OCQ_START);
3754                 params[5] = FW_PARAM_PFVF(OCQ_END);
3755                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params,
3756                                       val);
3757                 if (ret < 0)
3758                         goto bye;
3759                 adap->vres.qp.start = val[0];
3760                 adap->vres.qp.size = val[1] - val[0] + 1;
3761                 adap->vres.cq.start = val[2];
3762                 adap->vres.cq.size = val[3] - val[2] + 1;
3763                 adap->vres.ocq.start = val[4];
3764                 adap->vres.ocq.size = val[5] - val[4] + 1;
3765
3766                 params[0] = FW_PARAM_DEV(MAXORDIRD_QP);
3767                 params[1] = FW_PARAM_DEV(MAXIRD_ADAPTER);
3768                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params,
3769                                       val);
3770                 if (ret < 0) {
3771                         adap->params.max_ordird_qp = 8;
3772                         adap->params.max_ird_adapter = 32 * adap->tids.ntids;
3773                         ret = 0;
3774                 } else {
3775                         adap->params.max_ordird_qp = val[0];
3776                         adap->params.max_ird_adapter = val[1];
3777                 }
3778                 dev_info(adap->pdev_dev,
3779                          "max_ordird_qp %d max_ird_adapter %d\n",
3780                          adap->params.max_ordird_qp,
3781                          adap->params.max_ird_adapter);
3782                 adap->num_ofld_uld += 2;
3783         }
3784         if (caps_cmd.iscsicaps) {
3785                 params[0] = FW_PARAM_PFVF(ISCSI_START);
3786                 params[1] = FW_PARAM_PFVF(ISCSI_END);
3787                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
3788                                       params, val);
3789                 if (ret < 0)
3790                         goto bye;
3791                 adap->vres.iscsi.start = val[0];
3792                 adap->vres.iscsi.size = val[1] - val[0] + 1;
3793                 /* LIO target and cxgb4i initiaitor */
3794                 adap->num_ofld_uld += 2;
3795         }
3796         if (caps_cmd.cryptocaps) {
3797                 /* Should query params here...TODO */
3798                 params[0] = FW_PARAM_PFVF(NCRYPTO_LOOKASIDE);
3799                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
3800                                       params, val);
3801                 if (ret < 0) {
3802                         if (ret != -EINVAL)
3803                                 goto bye;
3804                 } else {
3805                         adap->vres.ncrypto_fc = val[0];
3806                 }
3807                 adap->params.crypto |= ULP_CRYPTO_LOOKASIDE;
3808                 adap->num_uld += 1;
3809         }
3810 #undef FW_PARAM_PFVF
3811 #undef FW_PARAM_DEV
3812
3813         /* The MTU/MSS Table is initialized by now, so load their values.  If
3814          * we're initializing the adapter, then we'll make any modifications
3815          * we want to the MTU/MSS Table and also initialize the congestion
3816          * parameters.
3817          */
3818         t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
3819         if (state != DEV_STATE_INIT) {
3820                 int i;
3821
3822                 /* The default MTU Table contains values 1492 and 1500.
3823                  * However, for TCP, it's better to have two values which are
3824                  * a multiple of 8 +/- 4 bytes apart near this popular MTU.
3825                  * This allows us to have a TCP Data Payload which is a
3826                  * multiple of 8 regardless of what combination of TCP Options
3827                  * are in use (always a multiple of 4 bytes) which is
3828                  * important for performance reasons.  For instance, if no
3829                  * options are in use, then we have a 20-byte IP header and a
3830                  * 20-byte TCP header.  In this case, a 1500-byte MSS would
3831                  * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
3832                  * which is not a multiple of 8.  So using an MSS of 1488 in
3833                  * this case results in a TCP Data Payload of 1448 bytes which
3834                  * is a multiple of 8.  On the other hand, if 12-byte TCP Time
3835                  * Stamps have been negotiated, then an MTU of 1500 bytes
3836                  * results in a TCP Data Payload of 1448 bytes which, as
3837                  * above, is a multiple of 8 bytes ...
3838                  */
3839                 for (i = 0; i < NMTUS; i++)
3840                         if (adap->params.mtus[i] == 1492) {
3841                                 adap->params.mtus[i] = 1488;
3842                                 break;
3843                         }
3844
3845                 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
3846                              adap->params.b_wnd);
3847         }
3848         t4_init_sge_params(adap);
3849         adap->flags |= FW_OK;
3850         t4_init_tp_params(adap);
3851         return 0;
3852
3853         /*
3854          * Something bad happened.  If a command timed out or failed with EIO
3855          * FW does not operate within its spec or something catastrophic
3856          * happened to HW/FW, stop issuing commands.
3857          */
3858 bye:
3859         kfree(adap->sge.egr_map);
3860         kfree(adap->sge.ingr_map);
3861         kfree(adap->sge.starving_fl);
3862         kfree(adap->sge.txq_maperr);
3863 #ifdef CONFIG_DEBUG_FS
3864         kfree(adap->sge.blocked_fl);
3865 #endif
3866         if (ret != -ETIMEDOUT && ret != -EIO)
3867                 t4_fw_bye(adap, adap->mbox);
3868         return ret;
3869 }
3870
3871 /* EEH callbacks */
3872
3873 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
3874                                          pci_channel_state_t state)
3875 {
3876         int i;
3877         struct adapter *adap = pci_get_drvdata(pdev);
3878
3879         if (!adap)
3880                 goto out;
3881
3882         rtnl_lock();
3883         adap->flags &= ~FW_OK;
3884         notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
3885         spin_lock(&adap->stats_lock);
3886         for_each_port(adap, i) {
3887                 struct net_device *dev = adap->port[i];
3888                 if (dev) {
3889                         netif_device_detach(dev);
3890                         netif_carrier_off(dev);
3891                 }
3892         }
3893         spin_unlock(&adap->stats_lock);
3894         disable_interrupts(adap);
3895         if (adap->flags & FULL_INIT_DONE)
3896                 cxgb_down(adap);
3897         rtnl_unlock();
3898         if ((adap->flags & DEV_ENABLED)) {
3899                 pci_disable_device(pdev);
3900                 adap->flags &= ~DEV_ENABLED;
3901         }
3902 out:    return state == pci_channel_io_perm_failure ?
3903                 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
3904 }
3905
3906 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
3907 {
3908         int i, ret;
3909         struct fw_caps_config_cmd c;
3910         struct adapter *adap = pci_get_drvdata(pdev);
3911
3912         if (!adap) {
3913                 pci_restore_state(pdev);
3914                 pci_save_state(pdev);
3915                 return PCI_ERS_RESULT_RECOVERED;
3916         }
3917
3918         if (!(adap->flags & DEV_ENABLED)) {
3919                 if (pci_enable_device(pdev)) {
3920                         dev_err(&pdev->dev, "Cannot reenable PCI "
3921                                             "device after reset\n");
3922                         return PCI_ERS_RESULT_DISCONNECT;
3923                 }
3924                 adap->flags |= DEV_ENABLED;
3925         }
3926
3927         pci_set_master(pdev);
3928         pci_restore_state(pdev);
3929         pci_save_state(pdev);
3930         pci_cleanup_aer_uncorrect_error_status(pdev);
3931
3932         if (t4_wait_dev_ready(adap->regs) < 0)
3933                 return PCI_ERS_RESULT_DISCONNECT;
3934         if (t4_fw_hello(adap, adap->mbox, adap->pf, MASTER_MUST, NULL) < 0)
3935                 return PCI_ERS_RESULT_DISCONNECT;
3936         adap->flags |= FW_OK;
3937         if (adap_init1(adap, &c))
3938                 return PCI_ERS_RESULT_DISCONNECT;
3939
3940         for_each_port(adap, i) {
3941                 struct port_info *p = adap2pinfo(adap, i);
3942
3943                 ret = t4_alloc_vi(adap, adap->mbox, p->tx_chan, adap->pf, 0, 1,
3944                                   NULL, NULL);
3945                 if (ret < 0)
3946                         return PCI_ERS_RESULT_DISCONNECT;
3947                 p->viid = ret;
3948                 p->xact_addr_filt = -1;
3949         }
3950
3951         t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
3952                      adap->params.b_wnd);
3953         setup_memwin(adap);
3954         if (cxgb_up(adap))
3955                 return PCI_ERS_RESULT_DISCONNECT;
3956         return PCI_ERS_RESULT_RECOVERED;
3957 }
3958
3959 static void eeh_resume(struct pci_dev *pdev)
3960 {
3961         int i;
3962         struct adapter *adap = pci_get_drvdata(pdev);
3963
3964         if (!adap)
3965                 return;
3966
3967         rtnl_lock();
3968         for_each_port(adap, i) {
3969                 struct net_device *dev = adap->port[i];
3970                 if (dev) {
3971                         if (netif_running(dev)) {
3972                                 link_start(dev);
3973                                 cxgb_set_rxmode(dev);
3974                         }
3975                         netif_device_attach(dev);
3976                 }
3977         }
3978         rtnl_unlock();
3979 }
3980
3981 static const struct pci_error_handlers cxgb4_eeh = {
3982         .error_detected = eeh_err_detected,
3983         .slot_reset     = eeh_slot_reset,
3984         .resume         = eeh_resume,
3985 };
3986
3987 /* Return true if the Link Configuration supports "High Speeds" (those greater
3988  * than 1Gb/s).
3989  */
3990 static inline bool is_x_10g_port(const struct link_config *lc)
3991 {
3992         unsigned int speeds, high_speeds;
3993
3994         speeds = FW_PORT_CAP_SPEED_V(FW_PORT_CAP_SPEED_G(lc->supported));
3995         high_speeds = speeds & ~(FW_PORT_CAP_SPEED_100M | FW_PORT_CAP_SPEED_1G);
3996
3997         return high_speeds != 0;
3998 }
3999
4000 /*
4001  * Perform default configuration of DMA queues depending on the number and type
4002  * of ports we found and the number of available CPUs.  Most settings can be
4003  * modified by the admin prior to actual use.
4004  */
4005 static void cfg_queues(struct adapter *adap)
4006 {
4007         struct sge *s = &adap->sge;
4008         int i = 0, n10g = 0, qidx = 0;
4009 #ifndef CONFIG_CHELSIO_T4_DCB
4010         int q10g = 0;
4011 #endif
4012
4013         /* Reduce memory usage in kdump environment, disable all offload.
4014          */
4015         if (is_kdump_kernel()) {
4016                 adap->params.offload = 0;
4017                 adap->params.crypto = 0;
4018         } else if (is_uld(adap) && t4_uld_mem_alloc(adap)) {
4019                 adap->params.offload = 0;
4020                 adap->params.crypto = 0;
4021         }
4022
4023         n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
4024 #ifdef CONFIG_CHELSIO_T4_DCB
4025         /* For Data Center Bridging support we need to be able to support up
4026          * to 8 Traffic Priorities; each of which will be assigned to its
4027          * own TX Queue in order to prevent Head-Of-Line Blocking.
4028          */
4029         if (adap->params.nports * 8 > MAX_ETH_QSETS) {
4030                 dev_err(adap->pdev_dev, "MAX_ETH_QSETS=%d < %d!\n",
4031                         MAX_ETH_QSETS, adap->params.nports * 8);
4032                 BUG_ON(1);
4033         }
4034
4035         for_each_port(adap, i) {
4036                 struct port_info *pi = adap2pinfo(adap, i);
4037
4038                 pi->first_qset = qidx;
4039                 pi->nqsets = 8;
4040                 qidx += pi->nqsets;
4041         }
4042 #else /* !CONFIG_CHELSIO_T4_DCB */
4043         /*
4044          * We default to 1 queue per non-10G port and up to # of cores queues
4045          * per 10G port.
4046          */
4047         if (n10g)
4048                 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
4049         if (q10g > netif_get_num_default_rss_queues())
4050                 q10g = netif_get_num_default_rss_queues();
4051
4052         for_each_port(adap, i) {
4053                 struct port_info *pi = adap2pinfo(adap, i);
4054
4055                 pi->first_qset = qidx;
4056                 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
4057                 qidx += pi->nqsets;
4058         }
4059 #endif /* !CONFIG_CHELSIO_T4_DCB */
4060
4061         s->ethqsets = qidx;
4062         s->max_ethqsets = qidx;   /* MSI-X may lower it later */
4063
4064         if (is_uld(adap)) {
4065                 /*
4066                  * For offload we use 1 queue/channel if all ports are up to 1G,
4067                  * otherwise we divide all available queues amongst the channels
4068                  * capped by the number of available cores.
4069                  */
4070                 if (n10g) {
4071                         i = min_t(int, MAX_OFLD_QSETS, num_online_cpus());
4072                         s->ofldqsets = roundup(i, adap->params.nports);
4073                 } else {
4074                         s->ofldqsets = adap->params.nports;
4075                 }
4076         }
4077
4078         for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
4079                 struct sge_eth_rxq *r = &s->ethrxq[i];
4080
4081                 init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
4082                 r->fl.size = 72;
4083         }
4084
4085         for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
4086                 s->ethtxq[i].q.size = 1024;
4087
4088         for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
4089                 s->ctrlq[i].q.size = 512;
4090
4091         init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
4092         init_rspq(adap, &s->intrq, 0, 1, 512, 64);
4093 }
4094
4095 /*
4096  * Reduce the number of Ethernet queues across all ports to at most n.
4097  * n provides at least one queue per port.
4098  */
4099 static void reduce_ethqs(struct adapter *adap, int n)
4100 {
4101         int i;
4102         struct port_info *pi;
4103
4104         while (n < adap->sge.ethqsets)
4105                 for_each_port(adap, i) {
4106                         pi = adap2pinfo(adap, i);
4107                         if (pi->nqsets > 1) {
4108                                 pi->nqsets--;
4109                                 adap->sge.ethqsets--;
4110                                 if (adap->sge.ethqsets <= n)
4111                                         break;
4112                         }
4113                 }
4114
4115         n = 0;
4116         for_each_port(adap, i) {
4117                 pi = adap2pinfo(adap, i);
4118                 pi->first_qset = n;
4119                 n += pi->nqsets;
4120         }
4121 }
4122
4123 static int get_msix_info(struct adapter *adap)
4124 {
4125         struct uld_msix_info *msix_info;
4126         unsigned int max_ingq = 0;
4127
4128         if (is_offload(adap))
4129                 max_ingq += MAX_OFLD_QSETS * adap->num_ofld_uld;
4130         if (is_pci_uld(adap))
4131                 max_ingq += MAX_OFLD_QSETS * adap->num_uld;
4132
4133         if (!max_ingq)
4134                 goto out;
4135
4136         msix_info = kcalloc(max_ingq, sizeof(*msix_info), GFP_KERNEL);
4137         if (!msix_info)
4138                 return -ENOMEM;
4139
4140         adap->msix_bmap_ulds.msix_bmap = kcalloc(BITS_TO_LONGS(max_ingq),
4141                                                  sizeof(long), GFP_KERNEL);
4142         if (!adap->msix_bmap_ulds.msix_bmap) {
4143                 kfree(msix_info);
4144                 return -ENOMEM;
4145         }
4146         spin_lock_init(&adap->msix_bmap_ulds.lock);
4147         adap->msix_info_ulds = msix_info;
4148 out:
4149         return 0;
4150 }
4151
4152 static void free_msix_info(struct adapter *adap)
4153 {
4154         if (!(adap->num_uld && adap->num_ofld_uld))
4155                 return;
4156
4157         kfree(adap->msix_info_ulds);
4158         kfree(adap->msix_bmap_ulds.msix_bmap);
4159 }
4160
4161 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
4162 #define EXTRA_VECS 2
4163
4164 static int enable_msix(struct adapter *adap)
4165 {
4166         int ofld_need = 0, uld_need = 0;
4167         int i, j, want, need, allocated;
4168         struct sge *s = &adap->sge;
4169         unsigned int nchan = adap->params.nports;
4170         struct msix_entry *entries;
4171         int max_ingq = MAX_INGQ;
4172
4173         if (is_pci_uld(adap))
4174                 max_ingq += (MAX_OFLD_QSETS * adap->num_uld);
4175         if (is_offload(adap))
4176                 max_ingq += (MAX_OFLD_QSETS * adap->num_ofld_uld);
4177         entries = kmalloc(sizeof(*entries) * (max_ingq + 1),
4178                           GFP_KERNEL);
4179         if (!entries)
4180                 return -ENOMEM;
4181
4182         /* map for msix */
4183         if (get_msix_info(adap)) {
4184                 adap->params.offload = 0;
4185                 adap->params.crypto = 0;
4186         }
4187
4188         for (i = 0; i < max_ingq + 1; ++i)
4189                 entries[i].entry = i;
4190
4191         want = s->max_ethqsets + EXTRA_VECS;
4192         if (is_offload(adap)) {
4193                 want += adap->num_ofld_uld * s->ofldqsets;
4194                 ofld_need = adap->num_ofld_uld * nchan;
4195         }
4196         if (is_pci_uld(adap)) {
4197                 want += adap->num_uld * s->ofldqsets;
4198                 uld_need = adap->num_uld * nchan;
4199         }
4200 #ifdef CONFIG_CHELSIO_T4_DCB
4201         /* For Data Center Bridging we need 8 Ethernet TX Priority Queues for
4202          * each port.
4203          */
4204         need = 8 * adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
4205 #else
4206         need = adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
4207 #endif
4208         allocated = pci_enable_msix_range(adap->pdev, entries, need, want);
4209         if (allocated < 0) {
4210                 dev_info(adap->pdev_dev, "not enough MSI-X vectors left,"
4211                          " not using MSI-X\n");
4212                 kfree(entries);
4213                 return allocated;
4214         }
4215
4216         /* Distribute available vectors to the various queue groups.
4217          * Every group gets its minimum requirement and NIC gets top
4218          * priority for leftovers.
4219          */
4220         i = allocated - EXTRA_VECS - ofld_need - uld_need;
4221         if (i < s->max_ethqsets) {
4222                 s->max_ethqsets = i;
4223                 if (i < s->ethqsets)
4224                         reduce_ethqs(adap, i);
4225         }
4226         if (is_uld(adap)) {
4227                 if (allocated < want)
4228                         s->nqs_per_uld = nchan;
4229                 else
4230                         s->nqs_per_uld = s->ofldqsets;
4231         }
4232
4233         for (i = 0; i < (s->max_ethqsets + EXTRA_VECS); ++i)
4234                 adap->msix_info[i].vec = entries[i].vector;
4235         if (is_uld(adap)) {
4236                 for (j = 0 ; i < allocated; ++i, j++) {
4237                         adap->msix_info_ulds[j].vec = entries[i].vector;
4238                         adap->msix_info_ulds[j].idx = i;
4239                 }
4240                 adap->msix_bmap_ulds.mapsize = j;
4241         }
4242         dev_info(adap->pdev_dev, "%d MSI-X vectors allocated, "
4243                  "nic %d per uld %d\n",
4244                  allocated, s->max_ethqsets, s->nqs_per_uld);
4245
4246         kfree(entries);
4247         return 0;
4248 }
4249
4250 #undef EXTRA_VECS
4251
4252 static int init_rss(struct adapter *adap)
4253 {
4254         unsigned int i;
4255         int err;
4256
4257         err = t4_init_rss_mode(adap, adap->mbox);
4258         if (err)
4259                 return err;
4260
4261         for_each_port(adap, i) {
4262                 struct port_info *pi = adap2pinfo(adap, i);
4263
4264                 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
4265                 if (!pi->rss)
4266                         return -ENOMEM;
4267         }
4268         return 0;
4269 }
4270
4271 static int cxgb4_get_pcie_dev_link_caps(struct adapter *adap,
4272                                         enum pci_bus_speed *speed,
4273                                         enum pcie_link_width *width)
4274 {
4275         u32 lnkcap1, lnkcap2;
4276         int err1, err2;
4277
4278 #define  PCIE_MLW_CAP_SHIFT 4   /* start of MLW mask in link capabilities */
4279
4280         *speed = PCI_SPEED_UNKNOWN;
4281         *width = PCIE_LNK_WIDTH_UNKNOWN;
4282
4283         err1 = pcie_capability_read_dword(adap->pdev, PCI_EXP_LNKCAP,
4284                                           &lnkcap1);
4285         err2 = pcie_capability_read_dword(adap->pdev, PCI_EXP_LNKCAP2,
4286                                           &lnkcap2);
4287         if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
4288                 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
4289                         *speed = PCIE_SPEED_8_0GT;
4290                 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
4291                         *speed = PCIE_SPEED_5_0GT;
4292                 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
4293                         *speed = PCIE_SPEED_2_5GT;
4294         }
4295         if (!err1) {
4296                 *width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
4297                 if (!lnkcap2) { /* pre-r3.0 */
4298                         if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
4299                                 *speed = PCIE_SPEED_5_0GT;
4300                         else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
4301                                 *speed = PCIE_SPEED_2_5GT;
4302                 }
4303         }
4304
4305         if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
4306                 return err1 ? err1 : err2 ? err2 : -EINVAL;
4307         return 0;
4308 }
4309
4310 static void cxgb4_check_pcie_caps(struct adapter *adap)
4311 {
4312         enum pcie_link_width width, width_cap;
4313         enum pci_bus_speed speed, speed_cap;
4314
4315 #define PCIE_SPEED_STR(speed) \
4316         (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
4317          speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
4318          speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
4319          "Unknown")
4320
4321         if (cxgb4_get_pcie_dev_link_caps(adap, &speed_cap, &width_cap)) {
4322                 dev_warn(adap->pdev_dev,
4323                          "Unable to determine PCIe device BW capabilities\n");
4324                 return;
4325         }
4326
4327         if (pcie_get_minimum_link(adap->pdev, &speed, &width) ||
4328             speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
4329                 dev_warn(adap->pdev_dev,
4330                          "Unable to determine PCI Express bandwidth.\n");
4331                 return;
4332         }
4333
4334         dev_info(adap->pdev_dev, "PCIe link speed is %s, device supports %s\n",
4335                  PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
4336         dev_info(adap->pdev_dev, "PCIe link width is x%d, device supports x%d\n",
4337                  width, width_cap);
4338         if (speed < speed_cap || width < width_cap)
4339                 dev_info(adap->pdev_dev,
4340                          "A slot with more lanes and/or higher speed is "
4341                          "suggested for optimal performance.\n");
4342 }
4343
4344 /* Dump basic information about the adapter */
4345 static void print_adapter_info(struct adapter *adapter)
4346 {
4347         /* Device information */
4348         dev_info(adapter->pdev_dev, "Chelsio %s rev %d\n",
4349                  adapter->params.vpd.id,
4350                  CHELSIO_CHIP_RELEASE(adapter->params.chip));
4351         dev_info(adapter->pdev_dev, "S/N: %s, P/N: %s\n",
4352                  adapter->params.vpd.sn, adapter->params.vpd.pn);
4353
4354         /* Firmware Version */
4355         if (!adapter->params.fw_vers)
4356                 dev_warn(adapter->pdev_dev, "No firmware loaded\n");
4357         else
4358                 dev_info(adapter->pdev_dev, "Firmware version: %u.%u.%u.%u\n",
4359                          FW_HDR_FW_VER_MAJOR_G(adapter->params.fw_vers),
4360                          FW_HDR_FW_VER_MINOR_G(adapter->params.fw_vers),
4361                          FW_HDR_FW_VER_MICRO_G(adapter->params.fw_vers),
4362                          FW_HDR_FW_VER_BUILD_G(adapter->params.fw_vers));
4363
4364         /* Bootstrap Firmware Version. (Some adapters don't have Bootstrap
4365          * Firmware, so dev_info() is more appropriate here.)
4366          */
4367         if (!adapter->params.bs_vers)
4368                 dev_info(adapter->pdev_dev, "No bootstrap loaded\n");
4369         else
4370                 dev_info(adapter->pdev_dev, "Bootstrap version: %u.%u.%u.%u\n",
4371                          FW_HDR_FW_VER_MAJOR_G(adapter->params.bs_vers),
4372                          FW_HDR_FW_VER_MINOR_G(adapter->params.bs_vers),
4373                          FW_HDR_FW_VER_MICRO_G(adapter->params.bs_vers),
4374                          FW_HDR_FW_VER_BUILD_G(adapter->params.bs_vers));
4375
4376         /* TP Microcode Version */
4377         if (!adapter->params.tp_vers)
4378                 dev_warn(adapter->pdev_dev, "No TP Microcode loaded\n");
4379         else
4380                 dev_info(adapter->pdev_dev,
4381                          "TP Microcode version: %u.%u.%u.%u\n",
4382                          FW_HDR_FW_VER_MAJOR_G(adapter->params.tp_vers),
4383                          FW_HDR_FW_VER_MINOR_G(adapter->params.tp_vers),
4384                          FW_HDR_FW_VER_MICRO_G(adapter->params.tp_vers),
4385                          FW_HDR_FW_VER_BUILD_G(adapter->params.tp_vers));
4386
4387         /* Expansion ROM version */
4388         if (!adapter->params.er_vers)
4389                 dev_info(adapter->pdev_dev, "No Expansion ROM loaded\n");
4390         else
4391                 dev_info(adapter->pdev_dev,
4392                          "Expansion ROM version: %u.%u.%u.%u\n",
4393                          FW_HDR_FW_VER_MAJOR_G(adapter->params.er_vers),
4394                          FW_HDR_FW_VER_MINOR_G(adapter->params.er_vers),
4395                          FW_HDR_FW_VER_MICRO_G(adapter->params.er_vers),
4396                          FW_HDR_FW_VER_BUILD_G(adapter->params.er_vers));
4397
4398         /* Software/Hardware configuration */
4399         dev_info(adapter->pdev_dev, "Configuration: %sNIC %s, %s capable\n",
4400                  is_offload(adapter) ? "R" : "",
4401                  ((adapter->flags & USING_MSIX) ? "MSI-X" :
4402                   (adapter->flags & USING_MSI) ? "MSI" : ""),
4403                  is_offload(adapter) ? "Offload" : "non-Offload");
4404 }
4405
4406 static void print_port_info(const struct net_device *dev)
4407 {
4408         char buf[80];
4409         char *bufp = buf;
4410         const char *spd = "";
4411         const struct port_info *pi = netdev_priv(dev);
4412         const struct adapter *adap = pi->adapter;
4413
4414         if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
4415                 spd = " 2.5 GT/s";
4416         else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
4417                 spd = " 5 GT/s";
4418         else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_8_0GB)
4419                 spd = " 8 GT/s";
4420
4421         if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
4422                 bufp += sprintf(bufp, "100M/");
4423         if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
4424                 bufp += sprintf(bufp, "1G/");
4425         if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
4426                 bufp += sprintf(bufp, "10G/");
4427         if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_25G)
4428                 bufp += sprintf(bufp, "25G/");
4429         if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G)
4430                 bufp += sprintf(bufp, "40G/");
4431         if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100G)
4432                 bufp += sprintf(bufp, "100G/");
4433         if (bufp != buf)
4434                 --bufp;
4435         sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
4436
4437         netdev_info(dev, "%s: Chelsio %s (%s) %s\n",
4438                     dev->name, adap->params.vpd.id, adap->name, buf);
4439 }
4440
4441 static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
4442 {
4443         pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
4444 }
4445
4446 /*
4447  * Free the following resources:
4448  * - memory used for tables
4449  * - MSI/MSI-X
4450  * - net devices
4451  * - resources FW is holding for us
4452  */
4453 static void free_some_resources(struct adapter *adapter)
4454 {
4455         unsigned int i;
4456
4457         kvfree(adapter->l2t);
4458         t4_cleanup_sched(adapter);
4459         kvfree(adapter->tids.tid_tab);
4460         cxgb4_cleanup_tc_u32(adapter);
4461         kfree(adapter->sge.egr_map);
4462         kfree(adapter->sge.ingr_map);
4463         kfree(adapter->sge.starving_fl);
4464         kfree(adapter->sge.txq_maperr);
4465 #ifdef CONFIG_DEBUG_FS
4466         kfree(adapter->sge.blocked_fl);
4467 #endif
4468         disable_msi(adapter);
4469
4470         for_each_port(adapter, i)
4471                 if (adapter->port[i]) {
4472                         struct port_info *pi = adap2pinfo(adapter, i);
4473
4474                         if (pi->viid != 0)
4475                                 t4_free_vi(adapter, adapter->mbox, adapter->pf,
4476                                            0, pi->viid);
4477                         kfree(adap2pinfo(adapter, i)->rss);
4478                         free_netdev(adapter->port[i]);
4479                 }
4480         if (adapter->flags & FW_OK)
4481                 t4_fw_bye(adapter, adapter->pf);
4482 }
4483
4484 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
4485 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
4486                    NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
4487 #define SEGMENT_SIZE 128
4488
4489 static int get_chip_type(struct pci_dev *pdev, u32 pl_rev)
4490 {
4491         u16 device_id;
4492
4493         /* Retrieve adapter's device ID */
4494         pci_read_config_word(pdev, PCI_DEVICE_ID, &device_id);
4495
4496         switch (device_id >> 12) {
4497         case CHELSIO_T4:
4498                 return CHELSIO_CHIP_CODE(CHELSIO_T4, pl_rev);
4499         case CHELSIO_T5:
4500                 return CHELSIO_CHIP_CODE(CHELSIO_T5, pl_rev);
4501         case CHELSIO_T6:
4502                 return CHELSIO_CHIP_CODE(CHELSIO_T6, pl_rev);
4503         default:
4504                 dev_err(&pdev->dev, "Device %d is not supported\n",
4505                         device_id);
4506         }
4507         return -EINVAL;
4508 }
4509
4510 #ifdef CONFIG_PCI_IOV
4511 static void dummy_setup(struct net_device *dev)
4512 {
4513         dev->type = ARPHRD_NONE;
4514         dev->mtu = 0;
4515         dev->hard_header_len = 0;
4516         dev->addr_len = 0;
4517         dev->tx_queue_len = 0;
4518         dev->flags |= IFF_NOARP;
4519         dev->priv_flags |= IFF_NO_QUEUE;
4520
4521         /* Initialize the device structure. */
4522         dev->netdev_ops = &cxgb4_mgmt_netdev_ops;
4523         dev->ethtool_ops = &cxgb4_mgmt_ethtool_ops;
4524         dev->destructor = free_netdev;
4525 }
4526
4527 static int config_mgmt_dev(struct pci_dev *pdev)
4528 {
4529         struct adapter *adap = pci_get_drvdata(pdev);
4530         struct net_device *netdev;
4531         struct port_info *pi;
4532         char name[IFNAMSIZ];
4533         int err;
4534
4535         snprintf(name, IFNAMSIZ, "mgmtpf%d%d", adap->adap_idx, adap->pf);
4536         netdev = alloc_netdev(sizeof(struct port_info), name, NET_NAME_UNKNOWN,
4537                               dummy_setup);
4538         if (!netdev)
4539                 return -ENOMEM;
4540
4541         pi = netdev_priv(netdev);
4542         pi->adapter = adap;
4543         pi->port_id = adap->pf % adap->params.nports;
4544         SET_NETDEV_DEV(netdev, &pdev->dev);
4545
4546         adap->port[0] = netdev;
4547
4548         err = register_netdev(adap->port[0]);
4549         if (err) {
4550                 pr_info("Unable to register VF mgmt netdev %s\n", name);
4551                 free_netdev(adap->port[0]);
4552                 adap->port[0] = NULL;
4553                 return err;
4554         }
4555         return 0;
4556 }
4557
4558 static int cxgb4_iov_configure(struct pci_dev *pdev, int num_vfs)
4559 {
4560         struct adapter *adap = pci_get_drvdata(pdev);
4561         int err = 0;
4562         int current_vfs = pci_num_vf(pdev);
4563         u32 pcie_fw;
4564
4565         pcie_fw = readl(adap->regs + PCIE_FW_A);
4566         /* Check if cxgb4 is the MASTER and fw is initialized */
4567         if (!(pcie_fw & PCIE_FW_INIT_F) ||
4568             !(pcie_fw & PCIE_FW_MASTER_VLD_F) ||
4569             PCIE_FW_MASTER_G(pcie_fw) != 4) {
4570                 dev_warn(&pdev->dev,
4571                          "cxgb4 driver needs to be MASTER to support SRIOV\n");
4572                 return -EOPNOTSUPP;
4573         }
4574
4575         /* If any of the VF's is already assigned to Guest OS, then
4576          * SRIOV for the same cannot be modified
4577          */
4578         if (current_vfs && pci_vfs_assigned(pdev)) {
4579                 dev_err(&pdev->dev,
4580                         "Cannot modify SR-IOV while VFs are assigned\n");
4581                 num_vfs = current_vfs;
4582                 return num_vfs;
4583         }
4584
4585         /* Disable SRIOV when zero is passed.
4586          * One needs to disable SRIOV before modifying it, else
4587          * stack throws the below warning:
4588          * " 'n' VFs already enabled. Disable before enabling 'm' VFs."
4589          */
4590         if (!num_vfs) {
4591                 pci_disable_sriov(pdev);
4592                 if (adap->port[0]) {
4593                         unregister_netdev(adap->port[0]);
4594                         adap->port[0] = NULL;
4595                 }
4596                 /* free VF resources */
4597                 kfree(adap->vfinfo);
4598                 adap->vfinfo = NULL;
4599                 adap->num_vfs = 0;
4600                 return num_vfs;
4601         }
4602
4603         if (num_vfs != current_vfs) {
4604                 err = pci_enable_sriov(pdev, num_vfs);
4605                 if (err)
4606                         return err;
4607
4608                 adap->num_vfs = num_vfs;
4609                 err = config_mgmt_dev(pdev);
4610                 if (err)
4611                         return err;
4612         }
4613
4614         adap->vfinfo = kcalloc(adap->num_vfs,
4615                                sizeof(struct vf_info), GFP_KERNEL);
4616         if (adap->vfinfo)
4617                 fill_vf_station_mac_addr(adap);
4618         return num_vfs;
4619 }
4620 #endif
4621
4622 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4623 {
4624         int func, i, err, s_qpp, qpp, num_seg;
4625         struct port_info *pi;
4626         bool highdma = false;
4627         struct adapter *adapter = NULL;
4628         struct net_device *netdev;
4629         void __iomem *regs;
4630         u32 whoami, pl_rev;
4631         enum chip_type chip;
4632         static int adap_idx = 1;
4633 #ifdef CONFIG_PCI_IOV
4634         u32 v, port_vec;
4635 #endif
4636
4637         printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
4638
4639         err = pci_request_regions(pdev, KBUILD_MODNAME);
4640         if (err) {
4641                 /* Just info, some other driver may have claimed the device. */
4642                 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
4643                 return err;
4644         }
4645
4646         err = pci_enable_device(pdev);
4647         if (err) {
4648                 dev_err(&pdev->dev, "cannot enable PCI device\n");
4649                 goto out_release_regions;
4650         }
4651
4652         regs = pci_ioremap_bar(pdev, 0);
4653         if (!regs) {
4654                 dev_err(&pdev->dev, "cannot map device registers\n");
4655                 err = -ENOMEM;
4656                 goto out_disable_device;
4657         }
4658
4659         err = t4_wait_dev_ready(regs);
4660         if (err < 0)
4661                 goto out_unmap_bar0;
4662
4663         /* We control everything through one PF */
4664         whoami = readl(regs + PL_WHOAMI_A);
4665         pl_rev = REV_G(readl(regs + PL_REV_A));
4666         chip = get_chip_type(pdev, pl_rev);
4667         func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
4668                 SOURCEPF_G(whoami) : T6_SOURCEPF_G(whoami);
4669         if (func != ent->driver_data) {
4670 #ifndef CONFIG_PCI_IOV
4671                 iounmap(regs);
4672 #endif
4673                 pci_disable_device(pdev);
4674                 pci_save_state(pdev);        /* to restore SR-IOV later */
4675                 goto sriov;
4676         }
4677
4678         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
4679                 highdma = true;
4680                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4681                 if (err) {
4682                         dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
4683                                 "coherent allocations\n");
4684                         goto out_unmap_bar0;
4685                 }
4686         } else {
4687                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4688                 if (err) {
4689                         dev_err(&pdev->dev, "no usable DMA configuration\n");
4690                         goto out_unmap_bar0;
4691                 }
4692         }
4693
4694         pci_enable_pcie_error_reporting(pdev);
4695         enable_pcie_relaxed_ordering(pdev);
4696         pci_set_master(pdev);
4697         pci_save_state(pdev);
4698
4699         adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
4700         if (!adapter) {
4701                 err = -ENOMEM;
4702                 goto out_unmap_bar0;
4703         }
4704         adap_idx++;
4705
4706         adapter->workq = create_singlethread_workqueue("cxgb4");
4707         if (!adapter->workq) {
4708                 err = -ENOMEM;
4709                 goto out_free_adapter;
4710         }
4711
4712         adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
4713                                     (sizeof(struct mbox_cmd) *
4714                                      T4_OS_LOG_MBOX_CMDS),
4715                                     GFP_KERNEL);
4716         if (!adapter->mbox_log) {
4717                 err = -ENOMEM;
4718                 goto out_free_adapter;
4719         }
4720         adapter->mbox_log->size = T4_OS_LOG_MBOX_CMDS;
4721
4722         /* PCI device has been enabled */
4723         adapter->flags |= DEV_ENABLED;
4724
4725         adapter->regs = regs;
4726         adapter->pdev = pdev;
4727         adapter->pdev_dev = &pdev->dev;
4728         adapter->name = pci_name(pdev);
4729         adapter->mbox = func;
4730         adapter->pf = func;
4731         adapter->msg_enable = DFLT_MSG_ENABLE;
4732         memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
4733
4734         spin_lock_init(&adapter->stats_lock);
4735         spin_lock_init(&adapter->tid_release_lock);
4736         spin_lock_init(&adapter->win0_lock);
4737         spin_lock_init(&adapter->mbox_lock);
4738
4739         INIT_LIST_HEAD(&adapter->mlist.list);
4740
4741         INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
4742         INIT_WORK(&adapter->db_full_task, process_db_full);
4743         INIT_WORK(&adapter->db_drop_task, process_db_drop);
4744
4745         err = t4_prep_adapter(adapter);
4746         if (err)
4747                 goto out_free_adapter;
4748
4749
4750         if (!is_t4(adapter->params.chip)) {
4751                 s_qpp = (QUEUESPERPAGEPF0_S +
4752                         (QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) *
4753                         adapter->pf);
4754                 qpp = 1 << QUEUESPERPAGEPF0_G(t4_read_reg(adapter,
4755                       SGE_EGRESS_QUEUES_PER_PAGE_PF_A) >> s_qpp);
4756                 num_seg = PAGE_SIZE / SEGMENT_SIZE;
4757
4758                 /* Each segment size is 128B. Write coalescing is enabled only
4759                  * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
4760                  * queue is less no of segments that can be accommodated in
4761                  * a page size.
4762                  */
4763                 if (qpp > num_seg) {
4764                         dev_err(&pdev->dev,
4765                                 "Incorrect number of egress queues per page\n");
4766                         err = -EINVAL;
4767                         goto out_free_adapter;
4768                 }
4769                 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
4770                 pci_resource_len(pdev, 2));
4771                 if (!adapter->bar2) {
4772                         dev_err(&pdev->dev, "cannot map device bar2 region\n");
4773                         err = -ENOMEM;
4774                         goto out_free_adapter;
4775                 }
4776         }
4777
4778         setup_memwin(adapter);
4779         err = adap_init0(adapter);
4780 #ifdef CONFIG_DEBUG_FS
4781         bitmap_zero(adapter->sge.blocked_fl, adapter->sge.egr_sz);
4782 #endif
4783         setup_memwin_rdma(adapter);
4784         if (err)
4785                 goto out_unmap_bar;
4786
4787         /* configure SGE_STAT_CFG_A to read WC stats */
4788         if (!is_t4(adapter->params.chip))
4789                 t4_write_reg(adapter, SGE_STAT_CFG_A, STATSOURCE_T5_V(7) |
4790                              (is_t5(adapter->params.chip) ? STATMODE_V(0) :
4791                               T6_STATMODE_V(0)));
4792
4793         for_each_port(adapter, i) {
4794                 netdev = alloc_etherdev_mq(sizeof(struct port_info),
4795                                            MAX_ETH_QSETS);
4796                 if (!netdev) {
4797                         err = -ENOMEM;
4798                         goto out_free_dev;
4799                 }
4800
4801                 SET_NETDEV_DEV(netdev, &pdev->dev);
4802
4803                 adapter->port[i] = netdev;
4804                 pi = netdev_priv(netdev);
4805                 pi->adapter = adapter;
4806                 pi->xact_addr_filt = -1;
4807                 pi->port_id = i;
4808                 netdev->irq = pdev->irq;
4809
4810                 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
4811                         NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4812                         NETIF_F_RXCSUM | NETIF_F_RXHASH |
4813                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
4814                         NETIF_F_HW_TC;
4815                 if (highdma)
4816                         netdev->hw_features |= NETIF_F_HIGHDMA;
4817                 netdev->features |= netdev->hw_features;
4818                 netdev->vlan_features = netdev->features & VLAN_FEAT;
4819
4820                 netdev->priv_flags |= IFF_UNICAST_FLT;
4821
4822                 /* MTU range: 81 - 9600 */
4823                 netdev->min_mtu = 81;
4824                 netdev->max_mtu = MAX_MTU;
4825
4826                 netdev->netdev_ops = &cxgb4_netdev_ops;
4827 #ifdef CONFIG_CHELSIO_T4_DCB
4828                 netdev->dcbnl_ops = &cxgb4_dcb_ops;
4829                 cxgb4_dcb_state_init(netdev);
4830 #endif
4831                 cxgb4_set_ethtool_ops(netdev);
4832         }
4833
4834         pci_set_drvdata(pdev, adapter);
4835
4836         if (adapter->flags & FW_OK) {
4837                 err = t4_port_init(adapter, func, func, 0);
4838                 if (err)
4839                         goto out_free_dev;
4840         } else if (adapter->params.nports == 1) {
4841                 /* If we don't have a connection to the firmware -- possibly
4842                  * because of an error -- grab the raw VPD parameters so we
4843                  * can set the proper MAC Address on the debug network
4844                  * interface that we've created.
4845                  */
4846                 u8 hw_addr[ETH_ALEN];
4847                 u8 *na = adapter->params.vpd.na;
4848
4849                 err = t4_get_raw_vpd_params(adapter, &adapter->params.vpd);
4850                 if (!err) {
4851                         for (i = 0; i < ETH_ALEN; i++)
4852                                 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
4853                                               hex2val(na[2 * i + 1]));
4854                         t4_set_hw_addr(adapter, 0, hw_addr);
4855                 }
4856         }
4857
4858         /* Configure queues and allocate tables now, they can be needed as
4859          * soon as the first register_netdev completes.
4860          */
4861         cfg_queues(adapter);
4862
4863         adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
4864         if (!adapter->l2t) {
4865                 /* We tolerate a lack of L2T, giving up some functionality */
4866                 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
4867                 adapter->params.offload = 0;
4868         }
4869
4870 #if IS_ENABLED(CONFIG_IPV6)
4871         if ((CHELSIO_CHIP_VERSION(adapter->params.chip) <= CHELSIO_T5) &&
4872             (!(t4_read_reg(adapter, LE_DB_CONFIG_A) & ASLIPCOMPEN_F))) {
4873                 /* CLIP functionality is not present in hardware,
4874                  * hence disable all offload features
4875                  */
4876                 dev_warn(&pdev->dev,
4877                          "CLIP not enabled in hardware, continuing\n");
4878                 adapter->params.offload = 0;
4879         } else {
4880                 adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
4881                                                   adapter->clipt_end);
4882                 if (!adapter->clipt) {
4883                         /* We tolerate a lack of clip_table, giving up
4884                          * some functionality
4885                          */
4886                         dev_warn(&pdev->dev,
4887                                  "could not allocate Clip table, continuing\n");
4888                         adapter->params.offload = 0;
4889                 }
4890         }
4891 #endif
4892
4893         for_each_port(adapter, i) {
4894                 pi = adap2pinfo(adapter, i);
4895                 pi->sched_tbl = t4_init_sched(adapter->params.nsched_cls);
4896                 if (!pi->sched_tbl)
4897                         dev_warn(&pdev->dev,
4898                                  "could not activate scheduling on port %d\n",
4899                                  i);
4900         }
4901
4902         if (tid_init(&adapter->tids) < 0) {
4903                 dev_warn(&pdev->dev, "could not allocate TID table, "
4904                          "continuing\n");
4905                 adapter->params.offload = 0;
4906         } else {
4907                 adapter->tc_u32 = cxgb4_init_tc_u32(adapter);
4908                 if (!adapter->tc_u32)
4909                         dev_warn(&pdev->dev,
4910                                  "could not offload tc u32, continuing\n");
4911         }
4912
4913         if (is_offload(adapter)) {
4914                 if (t4_read_reg(adapter, LE_DB_CONFIG_A) & HASHEN_F) {
4915                         u32 hash_base, hash_reg;
4916
4917                         if (chip <= CHELSIO_T5) {
4918                                 hash_reg = LE_DB_TID_HASHBASE_A;
4919                                 hash_base = t4_read_reg(adapter, hash_reg);
4920                                 adapter->tids.hash_base = hash_base / 4;
4921                         } else {
4922                                 hash_reg = T6_LE_DB_HASH_TID_BASE_A;
4923                                 hash_base = t4_read_reg(adapter, hash_reg);
4924                                 adapter->tids.hash_base = hash_base;
4925                         }
4926                 }
4927         }
4928
4929         /* See what interrupts we'll be using */
4930         if (msi > 1 && enable_msix(adapter) == 0)
4931                 adapter->flags |= USING_MSIX;
4932         else if (msi > 0 && pci_enable_msi(pdev) == 0) {
4933                 adapter->flags |= USING_MSI;
4934                 if (msi > 1)
4935                         free_msix_info(adapter);
4936         }
4937
4938         /* check for PCI Express bandwidth capabiltites */
4939         cxgb4_check_pcie_caps(adapter);
4940
4941         err = init_rss(adapter);
4942         if (err)
4943                 goto out_free_dev;
4944
4945         /*
4946          * The card is now ready to go.  If any errors occur during device
4947          * registration we do not fail the whole card but rather proceed only
4948          * with the ports we manage to register successfully.  However we must
4949          * register at least one net device.
4950          */
4951         for_each_port(adapter, i) {
4952                 pi = adap2pinfo(adapter, i);
4953                 adapter->port[i]->dev_port = pi->lport;
4954                 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
4955                 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
4956
4957                 err = register_netdev(adapter->port[i]);
4958                 if (err)
4959                         break;
4960                 adapter->chan_map[pi->tx_chan] = i;
4961                 print_port_info(adapter->port[i]);
4962         }
4963         if (i == 0) {
4964                 dev_err(&pdev->dev, "could not register any net devices\n");
4965                 goto out_free_dev;
4966         }
4967         if (err) {
4968                 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
4969                 err = 0;
4970         }
4971
4972         if (cxgb4_debugfs_root) {
4973                 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
4974                                                            cxgb4_debugfs_root);
4975                 setup_debugfs(adapter);
4976         }
4977
4978         /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
4979         pdev->needs_freset = 1;
4980
4981         if (is_uld(adapter)) {
4982                 mutex_lock(&uld_mutex);
4983                 list_add_tail(&adapter->list_node, &adapter_list);
4984                 mutex_unlock(&uld_mutex);
4985         }
4986
4987         print_adapter_info(adapter);
4988         setup_fw_sge_queues(adapter);
4989         return 0;
4990
4991 sriov:
4992 #ifdef CONFIG_PCI_IOV
4993         adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
4994         if (!adapter) {
4995                 err = -ENOMEM;
4996                 goto free_pci_region;
4997         }
4998
4999         adapter->pdev = pdev;
5000         adapter->pdev_dev = &pdev->dev;
5001         adapter->name = pci_name(pdev);
5002         adapter->mbox = func;
5003         adapter->pf = func;
5004         adapter->regs = regs;
5005         adapter->adap_idx = adap_idx;
5006         adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
5007                                     (sizeof(struct mbox_cmd) *
5008                                      T4_OS_LOG_MBOX_CMDS),
5009                                     GFP_KERNEL);
5010         if (!adapter->mbox_log) {
5011                 err = -ENOMEM;
5012                 goto free_adapter;
5013         }
5014         spin_lock_init(&adapter->mbox_lock);
5015         INIT_LIST_HEAD(&adapter->mlist.list);
5016
5017         v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
5018             FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
5019         err = t4_query_params(adapter, adapter->mbox, adapter->pf, 0, 1,
5020                               &v, &port_vec);
5021         if (err < 0) {
5022                 dev_err(adapter->pdev_dev, "Could not fetch port params\n");
5023                 goto free_adapter;
5024         }
5025
5026         adapter->params.nports = hweight32(port_vec);
5027         pci_set_drvdata(pdev, adapter);
5028         return 0;
5029
5030  free_adapter:
5031         kfree(adapter);
5032  free_pci_region:
5033         iounmap(regs);
5034         pci_disable_sriov(pdev);
5035         pci_release_regions(pdev);
5036         return err;
5037 #else
5038         return 0;
5039 #endif
5040
5041  out_free_dev:
5042         free_some_resources(adapter);
5043         if (adapter->flags & USING_MSIX)
5044                 free_msix_info(adapter);
5045         if (adapter->num_uld || adapter->num_ofld_uld)
5046                 t4_uld_mem_free(adapter);
5047  out_unmap_bar:
5048         if (!is_t4(adapter->params.chip))
5049                 iounmap(adapter->bar2);
5050  out_free_adapter:
5051         if (adapter->workq)
5052                 destroy_workqueue(adapter->workq);
5053
5054         kfree(adapter->mbox_log);
5055         kfree(adapter);
5056  out_unmap_bar0:
5057         iounmap(regs);
5058  out_disable_device:
5059         pci_disable_pcie_error_reporting(pdev);
5060         pci_disable_device(pdev);
5061  out_release_regions:
5062         pci_release_regions(pdev);
5063         return err;
5064 }
5065
5066 static void remove_one(struct pci_dev *pdev)
5067 {
5068         struct adapter *adapter = pci_get_drvdata(pdev);
5069
5070         if (!adapter) {
5071                 pci_release_regions(pdev);
5072                 return;
5073         }
5074
5075         if (adapter->pf == 4) {
5076                 int i;
5077
5078                 /* Tear down per-adapter Work Queue first since it can contain
5079                  * references to our adapter data structure.
5080                  */
5081                 destroy_workqueue(adapter->workq);
5082
5083                 if (is_uld(adapter))
5084                         detach_ulds(adapter);
5085
5086                 disable_interrupts(adapter);
5087
5088                 for_each_port(adapter, i)
5089                         if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5090                                 unregister_netdev(adapter->port[i]);
5091
5092                 debugfs_remove_recursive(adapter->debugfs_root);
5093
5094                 /* If we allocated filters, free up state associated with any
5095                  * valid filters ...
5096                  */
5097                 clear_all_filters(adapter);
5098
5099                 if (adapter->flags & FULL_INIT_DONE)
5100                         cxgb_down(adapter);
5101
5102                 if (adapter->flags & USING_MSIX)
5103                         free_msix_info(adapter);
5104                 if (adapter->num_uld || adapter->num_ofld_uld)
5105                         t4_uld_mem_free(adapter);
5106                 free_some_resources(adapter);
5107 #if IS_ENABLED(CONFIG_IPV6)
5108                 t4_cleanup_clip_tbl(adapter);
5109 #endif
5110                 iounmap(adapter->regs);
5111                 if (!is_t4(adapter->params.chip))
5112                         iounmap(adapter->bar2);
5113                 pci_disable_pcie_error_reporting(pdev);
5114                 if ((adapter->flags & DEV_ENABLED)) {
5115                         pci_disable_device(pdev);
5116                         adapter->flags &= ~DEV_ENABLED;
5117                 }
5118                 pci_release_regions(pdev);
5119                 kfree(adapter->mbox_log);
5120                 synchronize_rcu();
5121                 kfree(adapter);
5122         }
5123 #ifdef CONFIG_PCI_IOV
5124         else {
5125                 if (adapter->port[0])
5126                         unregister_netdev(adapter->port[0]);
5127                 iounmap(adapter->regs);
5128                 kfree(adapter->vfinfo);
5129                 kfree(adapter);
5130                 pci_disable_sriov(pdev);
5131                 pci_release_regions(pdev);
5132         }
5133 #endif
5134 }
5135
5136 /* "Shutdown" quiesces the device, stopping Ingress Packet and Interrupt
5137  * delivery.  This is essentially a stripped down version of the PCI remove()
5138  * function where we do the minimal amount of work necessary to shutdown any
5139  * further activity.
5140  */
5141 static void shutdown_one(struct pci_dev *pdev)
5142 {
5143         struct adapter *adapter = pci_get_drvdata(pdev);
5144
5145         /* As with remove_one() above (see extended comment), we only want do
5146          * do cleanup on PCI Devices which went all the way through init_one()
5147          * ...
5148          */
5149         if (!adapter) {
5150                 pci_release_regions(pdev);
5151                 return;
5152         }
5153
5154         if (adapter->pf == 4) {
5155                 int i;
5156
5157                 for_each_port(adapter, i)
5158                         if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5159                                 cxgb_close(adapter->port[i]);
5160
5161                 t4_uld_clean_up(adapter);
5162                 disable_interrupts(adapter);
5163                 disable_msi(adapter);
5164
5165                 t4_sge_stop(adapter);
5166                 if (adapter->flags & FW_OK)
5167                         t4_fw_bye(adapter, adapter->mbox);
5168         }
5169 #ifdef CONFIG_PCI_IOV
5170         else {
5171                 if (adapter->port[0])
5172                         unregister_netdev(adapter->port[0]);
5173                 iounmap(adapter->regs);
5174                 kfree(adapter->vfinfo);
5175                 kfree(adapter);
5176                 pci_disable_sriov(pdev);
5177                 pci_release_regions(pdev);
5178         }
5179 #endif
5180 }
5181
5182 static struct pci_driver cxgb4_driver = {
5183         .name     = KBUILD_MODNAME,
5184         .id_table = cxgb4_pci_tbl,
5185         .probe    = init_one,
5186         .remove   = remove_one,
5187         .shutdown = shutdown_one,
5188 #ifdef CONFIG_PCI_IOV
5189         .sriov_configure = cxgb4_iov_configure,
5190 #endif
5191         .err_handler = &cxgb4_eeh,
5192 };
5193
5194 static int __init cxgb4_init_module(void)
5195 {
5196         int ret;
5197
5198         /* Debugfs support is optional, just warn if this fails */
5199         cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
5200         if (!cxgb4_debugfs_root)
5201                 pr_warn("could not create debugfs entry, continuing\n");
5202
5203         ret = pci_register_driver(&cxgb4_driver);
5204         if (ret < 0)
5205                 debugfs_remove(cxgb4_debugfs_root);
5206
5207 #if IS_ENABLED(CONFIG_IPV6)
5208         if (!inet6addr_registered) {
5209                 register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
5210                 inet6addr_registered = true;
5211         }
5212 #endif
5213
5214         return ret;
5215 }
5216
5217 static void __exit cxgb4_cleanup_module(void)
5218 {
5219 #if IS_ENABLED(CONFIG_IPV6)
5220         if (inet6addr_registered) {
5221                 unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
5222                 inet6addr_registered = false;
5223         }
5224 #endif
5225         pci_unregister_driver(&cxgb4_driver);
5226         debugfs_remove(cxgb4_debugfs_root);  /* NULL ok */
5227 }
5228
5229 module_init(cxgb4_init_module);
5230 module_exit(cxgb4_cleanup_module);