1 // SPDX-License-Identifier: GPL-2.0-only 2 /**************************************************************************** 3 * Driver for Solarflare network controllers and boards 4 * Copyright 2005-2006 Fen Systems Ltd. 5 * Copyright 2005-2013 Solarflare Communications Inc. 6 */ 7 8 #include <linux/filter.h> 9 #include <linux/module.h> 10 #include <linux/pci.h> 11 #include <linux/netdevice.h> 12 #include <linux/etherdevice.h> 13 #include <linux/delay.h> 14 #include <linux/notifier.h> 15 #include <linux/ip.h> 16 #include <linux/tcp.h> 17 #include <linux/in.h> 18 #include <linux/ethtool.h> 19 #include <linux/topology.h> 20 #include <linux/gfp.h> 21 #include <linux/aer.h> 22 #include <linux/interrupt.h> 23 #include "net_driver.h" 24 #include <net/gre.h> 25 #include <net/udp_tunnel.h> 26 #include "efx.h" 27 #include "efx_common.h" 28 #include "efx_channels.h" 29 #include "ef100.h" 30 #include "rx_common.h" 31 #include "tx_common.h" 32 #include "nic.h" 33 #include "io.h" 34 #include "selftest.h" 35 #include "sriov.h" 36 37 #include "mcdi_port_common.h" 38 #include "mcdi_pcol.h" 39 #include "workarounds.h" 40 41 /************************************************************************** 42 * 43 * Configurable values 44 * 45 *************************************************************************/ 46 47 module_param_named(interrupt_mode, efx_interrupt_mode, uint, 0444); 48 MODULE_PARM_DESC(interrupt_mode, 49 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)"); 50 51 module_param(rss_cpus, uint, 0444); 52 MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling"); 53 54 /* 55 * Use separate channels for TX and RX events 56 * 57 * Set this to 1 to use separate channels for TX and RX. It allows us 58 * to control interrupt affinity separately for TX and RX. 59 * 60 * This is only used in MSI-X interrupt mode 61 */ 62 bool efx_separate_tx_channels; 63 module_param(efx_separate_tx_channels, bool, 0444); 64 MODULE_PARM_DESC(efx_separate_tx_channels, 65 "Use separate channels for TX and RX"); 66 67 /* Initial interrupt moderation settings. They can be modified after 68 * module load with ethtool. 69 * 70 * The default for RX should strike a balance between increasing the 71 * round-trip latency and reducing overhead. 72 */ 73 static unsigned int rx_irq_mod_usec = 60; 74 75 /* Initial interrupt moderation settings. They can be modified after 76 * module load with ethtool. 77 * 78 * This default is chosen to ensure that a 10G link does not go idle 79 * while a TX queue is stopped after it has become full. A queue is 80 * restarted when it drops below half full. The time this takes (assuming 81 * worst case 3 descriptors per packet and 1024 descriptors) is 82 * 512 / 3 * 1.2 = 205 usec. 83 */ 84 static unsigned int tx_irq_mod_usec = 150; 85 86 static bool phy_flash_cfg; 87 module_param(phy_flash_cfg, bool, 0644); 88 MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially"); 89 90 static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE | 91 NETIF_MSG_LINK | NETIF_MSG_IFDOWN | 92 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR | 93 NETIF_MSG_TX_ERR | NETIF_MSG_HW); 94 module_param(debug, uint, 0); 95 MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value"); 96 97 /************************************************************************** 98 * 99 * Utility functions and prototypes 100 * 101 *************************************************************************/ 102 103 static void efx_remove_port(struct efx_nic *efx); 104 static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog); 105 static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp); 106 static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs, 107 u32 flags); 108 109 /************************************************************************** 110 * 111 * Port handling 112 * 113 **************************************************************************/ 114 115 static void efx_fini_port(struct efx_nic *efx); 116 117 static int efx_probe_port(struct efx_nic *efx) 118 { 119 int rc; 120 121 netif_dbg(efx, probe, efx->net_dev, "create port\n"); 122 123 if (phy_flash_cfg) 124 efx->phy_mode = PHY_MODE_SPECIAL; 125 126 /* Connect up MAC/PHY operations table */ 127 rc = efx->type->probe_port(efx); 128 if (rc) 129 return rc; 130 131 /* Initialise MAC address to permanent address */ 132 eth_hw_addr_set(efx->net_dev, efx->net_dev->perm_addr); 133 134 return 0; 135 } 136 137 static int efx_init_port(struct efx_nic *efx) 138 { 139 int rc; 140 141 netif_dbg(efx, drv, efx->net_dev, "init port\n"); 142 143 mutex_lock(&efx->mac_lock); 144 145 efx->port_initialized = true; 146 147 /* Ensure the PHY advertises the correct flow control settings */ 148 rc = efx_mcdi_port_reconfigure(efx); 149 if (rc && rc != -EPERM) 150 goto fail; 151 152 mutex_unlock(&efx->mac_lock); 153 return 0; 154 155 fail: 156 mutex_unlock(&efx->mac_lock); 157 return rc; 158 } 159 160 static void efx_fini_port(struct efx_nic *efx) 161 { 162 netif_dbg(efx, drv, efx->net_dev, "shut down port\n"); 163 164 if (!efx->port_initialized) 165 return; 166 167 efx->port_initialized = false; 168 169 efx->link_state.up = false; 170 efx_link_status_changed(efx); 171 } 172 173 static void efx_remove_port(struct efx_nic *efx) 174 { 175 netif_dbg(efx, drv, efx->net_dev, "destroying port\n"); 176 177 efx->type->remove_port(efx); 178 } 179 180 /************************************************************************** 181 * 182 * NIC handling 183 * 184 **************************************************************************/ 185 186 static LIST_HEAD(efx_primary_list); 187 static LIST_HEAD(efx_unassociated_list); 188 189 static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right) 190 { 191 return left->type == right->type && 192 left->vpd_sn && right->vpd_sn && 193 !strcmp(left->vpd_sn, right->vpd_sn); 194 } 195 196 static void efx_associate(struct efx_nic *efx) 197 { 198 struct efx_nic *other, *next; 199 200 if (efx->primary == efx) { 201 /* Adding primary function; look for secondaries */ 202 203 netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n"); 204 list_add_tail(&efx->node, &efx_primary_list); 205 206 list_for_each_entry_safe(other, next, &efx_unassociated_list, 207 node) { 208 if (efx_same_controller(efx, other)) { 209 list_del(&other->node); 210 netif_dbg(other, probe, other->net_dev, 211 "moving to secondary list of %s %s\n", 212 pci_name(efx->pci_dev), 213 efx->net_dev->name); 214 list_add_tail(&other->node, 215 &efx->secondary_list); 216 other->primary = efx; 217 } 218 } 219 } else { 220 /* Adding secondary function; look for primary */ 221 222 list_for_each_entry(other, &efx_primary_list, node) { 223 if (efx_same_controller(efx, other)) { 224 netif_dbg(efx, probe, efx->net_dev, 225 "adding to secondary list of %s %s\n", 226 pci_name(other->pci_dev), 227 other->net_dev->name); 228 list_add_tail(&efx->node, 229 &other->secondary_list); 230 efx->primary = other; 231 return; 232 } 233 } 234 235 netif_dbg(efx, probe, efx->net_dev, 236 "adding to unassociated list\n"); 237 list_add_tail(&efx->node, &efx_unassociated_list); 238 } 239 } 240 241 static void efx_dissociate(struct efx_nic *efx) 242 { 243 struct efx_nic *other, *next; 244 245 list_del(&efx->node); 246 efx->primary = NULL; 247 248 list_for_each_entry_safe(other, next, &efx->secondary_list, node) { 249 list_del(&other->node); 250 netif_dbg(other, probe, other->net_dev, 251 "moving to unassociated list\n"); 252 list_add_tail(&other->node, &efx_unassociated_list); 253 other->primary = NULL; 254 } 255 } 256 257 static int efx_probe_nic(struct efx_nic *efx) 258 { 259 int rc; 260 261 netif_dbg(efx, probe, efx->net_dev, "creating NIC\n"); 262 263 /* Carry out hardware-type specific initialisation */ 264 rc = efx->type->probe(efx); 265 if (rc) 266 return rc; 267 268 do { 269 if (!efx->max_channels || !efx->max_tx_channels) { 270 netif_err(efx, drv, efx->net_dev, 271 "Insufficient resources to allocate" 272 " any channels\n"); 273 rc = -ENOSPC; 274 goto fail1; 275 } 276 277 /* Determine the number of channels and queues by trying 278 * to hook in MSI-X interrupts. 279 */ 280 rc = efx_probe_interrupts(efx); 281 if (rc) 282 goto fail1; 283 284 rc = efx_set_channels(efx); 285 if (rc) 286 goto fail1; 287 288 /* dimension_resources can fail with EAGAIN */ 289 rc = efx->type->dimension_resources(efx); 290 if (rc != 0 && rc != -EAGAIN) 291 goto fail2; 292 293 if (rc == -EAGAIN) 294 /* try again with new max_channels */ 295 efx_remove_interrupts(efx); 296 297 } while (rc == -EAGAIN); 298 299 if (efx->n_channels > 1) 300 netdev_rss_key_fill(efx->rss_context.rx_hash_key, 301 sizeof(efx->rss_context.rx_hash_key)); 302 efx_set_default_rx_indir_table(efx, &efx->rss_context); 303 304 /* Initialise the interrupt moderation settings */ 305 efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000); 306 efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true, 307 true); 308 309 return 0; 310 311 fail2: 312 efx_remove_interrupts(efx); 313 fail1: 314 efx->type->remove(efx); 315 return rc; 316 } 317 318 static void efx_remove_nic(struct efx_nic *efx) 319 { 320 netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n"); 321 322 efx_remove_interrupts(efx); 323 efx->type->remove(efx); 324 } 325 326 /************************************************************************** 327 * 328 * NIC startup/shutdown 329 * 330 *************************************************************************/ 331 332 static int efx_probe_all(struct efx_nic *efx) 333 { 334 int rc; 335 336 rc = efx_probe_nic(efx); 337 if (rc) { 338 netif_err(efx, probe, efx->net_dev, "failed to create NIC\n"); 339 goto fail1; 340 } 341 342 rc = efx_probe_port(efx); 343 if (rc) { 344 netif_err(efx, probe, efx->net_dev, "failed to create port\n"); 345 goto fail2; 346 } 347 348 BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT); 349 if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) { 350 rc = -EINVAL; 351 goto fail3; 352 } 353 354 #ifdef CONFIG_SFC_SRIOV 355 rc = efx->type->vswitching_probe(efx); 356 if (rc) /* not fatal; the PF will still work fine */ 357 netif_warn(efx, probe, efx->net_dev, 358 "failed to setup vswitching rc=%d;" 359 " VFs may not function\n", rc); 360 #endif 361 362 rc = efx_probe_filters(efx); 363 if (rc) { 364 netif_err(efx, probe, efx->net_dev, 365 "failed to create filter tables\n"); 366 goto fail4; 367 } 368 369 rc = efx_probe_channels(efx); 370 if (rc) 371 goto fail5; 372 373 efx->state = STATE_NET_DOWN; 374 375 return 0; 376 377 fail5: 378 efx_remove_filters(efx); 379 fail4: 380 #ifdef CONFIG_SFC_SRIOV 381 efx->type->vswitching_remove(efx); 382 #endif 383 fail3: 384 efx_remove_port(efx); 385 fail2: 386 efx_remove_nic(efx); 387 fail1: 388 return rc; 389 } 390 391 static void efx_remove_all(struct efx_nic *efx) 392 { 393 rtnl_lock(); 394 efx_xdp_setup_prog(efx, NULL); 395 rtnl_unlock(); 396 397 efx_remove_channels(efx); 398 efx_remove_filters(efx); 399 #ifdef CONFIG_SFC_SRIOV 400 efx->type->vswitching_remove(efx); 401 #endif 402 efx_remove_port(efx); 403 efx_remove_nic(efx); 404 } 405 406 /************************************************************************** 407 * 408 * Interrupt moderation 409 * 410 **************************************************************************/ 411 unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs) 412 { 413 if (usecs == 0) 414 return 0; 415 if (usecs * 1000 < efx->timer_quantum_ns) 416 return 1; /* never round down to 0 */ 417 return usecs * 1000 / efx->timer_quantum_ns; 418 } 419 420 unsigned int efx_ticks_to_usecs(struct efx_nic *efx, unsigned int ticks) 421 { 422 /* We must round up when converting ticks to microseconds 423 * because we round down when converting the other way. 424 */ 425 return DIV_ROUND_UP(ticks * efx->timer_quantum_ns, 1000); 426 } 427 428 /* Set interrupt moderation parameters */ 429 int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs, 430 unsigned int rx_usecs, bool rx_adaptive, 431 bool rx_may_override_tx) 432 { 433 struct efx_channel *channel; 434 unsigned int timer_max_us; 435 436 EFX_ASSERT_RESET_SERIALISED(efx); 437 438 timer_max_us = efx->timer_max_ns / 1000; 439 440 if (tx_usecs > timer_max_us || rx_usecs > timer_max_us) 441 return -EINVAL; 442 443 if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 && 444 !rx_may_override_tx) { 445 netif_err(efx, drv, efx->net_dev, "Channels are shared. " 446 "RX and TX IRQ moderation must be equal\n"); 447 return -EINVAL; 448 } 449 450 efx->irq_rx_adaptive = rx_adaptive; 451 efx->irq_rx_moderation_us = rx_usecs; 452 efx_for_each_channel(channel, efx) { 453 if (efx_channel_has_rx_queue(channel)) 454 channel->irq_moderation_us = rx_usecs; 455 else if (efx_channel_has_tx_queues(channel)) 456 channel->irq_moderation_us = tx_usecs; 457 else if (efx_channel_is_xdp_tx(channel)) 458 channel->irq_moderation_us = tx_usecs; 459 } 460 461 return 0; 462 } 463 464 void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs, 465 unsigned int *rx_usecs, bool *rx_adaptive) 466 { 467 *rx_adaptive = efx->irq_rx_adaptive; 468 *rx_usecs = efx->irq_rx_moderation_us; 469 470 /* If channels are shared between RX and TX, so is IRQ 471 * moderation. Otherwise, IRQ moderation is the same for all 472 * TX channels and is not adaptive. 473 */ 474 if (efx->tx_channel_offset == 0) { 475 *tx_usecs = *rx_usecs; 476 } else { 477 struct efx_channel *tx_channel; 478 479 tx_channel = efx->channel[efx->tx_channel_offset]; 480 *tx_usecs = tx_channel->irq_moderation_us; 481 } 482 } 483 484 /************************************************************************** 485 * 486 * ioctls 487 * 488 *************************************************************************/ 489 490 /* Net device ioctl 491 * Context: process, rtnl_lock() held. 492 */ 493 static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd) 494 { 495 struct efx_nic *efx = efx_netdev_priv(net_dev); 496 struct mii_ioctl_data *data = if_mii(ifr); 497 498 if (cmd == SIOCSHWTSTAMP) 499 return efx_ptp_set_ts_config(efx, ifr); 500 if (cmd == SIOCGHWTSTAMP) 501 return efx_ptp_get_ts_config(efx, ifr); 502 503 /* Convert phy_id from older PRTAD/DEVAD format */ 504 if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) && 505 (data->phy_id & 0xfc00) == 0x0400) 506 data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400; 507 508 return mdio_mii_ioctl(&efx->mdio, data, cmd); 509 } 510 511 /************************************************************************** 512 * 513 * Kernel net device interface 514 * 515 *************************************************************************/ 516 517 /* Context: process, rtnl_lock() held. */ 518 int efx_net_open(struct net_device *net_dev) 519 { 520 struct efx_nic *efx = efx_netdev_priv(net_dev); 521 int rc; 522 523 netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n", 524 raw_smp_processor_id()); 525 526 rc = efx_check_disabled(efx); 527 if (rc) 528 return rc; 529 if (efx->phy_mode & PHY_MODE_SPECIAL) 530 return -EBUSY; 531 if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL)) 532 return -EIO; 533 534 /* Notify the kernel of the link state polled during driver load, 535 * before the monitor starts running */ 536 efx_link_status_changed(efx); 537 538 efx_start_all(efx); 539 if (efx->state == STATE_DISABLED || efx->reset_pending) 540 netif_device_detach(efx->net_dev); 541 else 542 efx->state = STATE_NET_UP; 543 544 efx_selftest_async_start(efx); 545 return 0; 546 } 547 548 /* Context: process, rtnl_lock() held. 549 * Note that the kernel will ignore our return code; this method 550 * should really be a void. 551 */ 552 int efx_net_stop(struct net_device *net_dev) 553 { 554 struct efx_nic *efx = efx_netdev_priv(net_dev); 555 556 netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n", 557 raw_smp_processor_id()); 558 559 /* Stop the device and flush all the channels */ 560 efx_stop_all(efx); 561 562 return 0; 563 } 564 565 static int efx_vlan_rx_add_vid(struct net_device *net_dev, __be16 proto, u16 vid) 566 { 567 struct efx_nic *efx = efx_netdev_priv(net_dev); 568 569 if (efx->type->vlan_rx_add_vid) 570 return efx->type->vlan_rx_add_vid(efx, proto, vid); 571 else 572 return -EOPNOTSUPP; 573 } 574 575 static int efx_vlan_rx_kill_vid(struct net_device *net_dev, __be16 proto, u16 vid) 576 { 577 struct efx_nic *efx = efx_netdev_priv(net_dev); 578 579 if (efx->type->vlan_rx_kill_vid) 580 return efx->type->vlan_rx_kill_vid(efx, proto, vid); 581 else 582 return -EOPNOTSUPP; 583 } 584 585 static const struct net_device_ops efx_netdev_ops = { 586 .ndo_open = efx_net_open, 587 .ndo_stop = efx_net_stop, 588 .ndo_get_stats64 = efx_net_stats, 589 .ndo_tx_timeout = efx_watchdog, 590 .ndo_start_xmit = efx_hard_start_xmit, 591 .ndo_validate_addr = eth_validate_addr, 592 .ndo_eth_ioctl = efx_ioctl, 593 .ndo_change_mtu = efx_change_mtu, 594 .ndo_set_mac_address = efx_set_mac_address, 595 .ndo_set_rx_mode = efx_set_rx_mode, 596 .ndo_set_features = efx_set_features, 597 .ndo_features_check = efx_features_check, 598 .ndo_vlan_rx_add_vid = efx_vlan_rx_add_vid, 599 .ndo_vlan_rx_kill_vid = efx_vlan_rx_kill_vid, 600 #ifdef CONFIG_SFC_SRIOV 601 .ndo_set_vf_mac = efx_sriov_set_vf_mac, 602 .ndo_set_vf_vlan = efx_sriov_set_vf_vlan, 603 .ndo_set_vf_spoofchk = efx_sriov_set_vf_spoofchk, 604 .ndo_get_vf_config = efx_sriov_get_vf_config, 605 .ndo_set_vf_link_state = efx_sriov_set_vf_link_state, 606 #endif 607 .ndo_get_phys_port_id = efx_get_phys_port_id, 608 .ndo_get_phys_port_name = efx_get_phys_port_name, 609 .ndo_setup_tc = efx_setup_tc, 610 #ifdef CONFIG_RFS_ACCEL 611 .ndo_rx_flow_steer = efx_filter_rfs, 612 #endif 613 .ndo_xdp_xmit = efx_xdp_xmit, 614 .ndo_bpf = efx_xdp 615 }; 616 617 static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog) 618 { 619 struct bpf_prog *old_prog; 620 621 if (efx->xdp_rxq_info_failed) { 622 netif_err(efx, drv, efx->net_dev, 623 "Unable to bind XDP program due to previous failure of rxq_info\n"); 624 return -EINVAL; 625 } 626 627 if (prog && efx->net_dev->mtu > efx_xdp_max_mtu(efx)) { 628 netif_err(efx, drv, efx->net_dev, 629 "Unable to configure XDP with MTU of %d (max: %d)\n", 630 efx->net_dev->mtu, efx_xdp_max_mtu(efx)); 631 return -EINVAL; 632 } 633 634 old_prog = rtnl_dereference(efx->xdp_prog); 635 rcu_assign_pointer(efx->xdp_prog, prog); 636 /* Release the reference that was originally passed by the caller. */ 637 if (old_prog) 638 bpf_prog_put(old_prog); 639 640 return 0; 641 } 642 643 /* Context: process, rtnl_lock() held. */ 644 static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp) 645 { 646 struct efx_nic *efx = efx_netdev_priv(dev); 647 648 switch (xdp->command) { 649 case XDP_SETUP_PROG: 650 return efx_xdp_setup_prog(efx, xdp->prog); 651 default: 652 return -EINVAL; 653 } 654 } 655 656 static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs, 657 u32 flags) 658 { 659 struct efx_nic *efx = efx_netdev_priv(dev); 660 661 if (!netif_running(dev)) 662 return -EINVAL; 663 664 return efx_xdp_tx_buffers(efx, n, xdpfs, flags & XDP_XMIT_FLUSH); 665 } 666 667 static void efx_update_name(struct efx_nic *efx) 668 { 669 strcpy(efx->name, efx->net_dev->name); 670 efx_mtd_rename(efx); 671 efx_set_channel_names(efx); 672 } 673 674 static int efx_netdev_event(struct notifier_block *this, 675 unsigned long event, void *ptr) 676 { 677 struct net_device *net_dev = netdev_notifier_info_to_dev(ptr); 678 679 if ((net_dev->netdev_ops == &efx_netdev_ops) && 680 event == NETDEV_CHANGENAME) 681 efx_update_name(efx_netdev_priv(net_dev)); 682 683 return NOTIFY_DONE; 684 } 685 686 static struct notifier_block efx_netdev_notifier = { 687 .notifier_call = efx_netdev_event, 688 }; 689 690 static ssize_t phy_type_show(struct device *dev, 691 struct device_attribute *attr, char *buf) 692 { 693 struct efx_nic *efx = dev_get_drvdata(dev); 694 return sprintf(buf, "%d\n", efx->phy_type); 695 } 696 static DEVICE_ATTR_RO(phy_type); 697 698 static int efx_register_netdev(struct efx_nic *efx) 699 { 700 struct net_device *net_dev = efx->net_dev; 701 struct efx_channel *channel; 702 int rc; 703 704 net_dev->watchdog_timeo = 5 * HZ; 705 net_dev->irq = efx->pci_dev->irq; 706 net_dev->netdev_ops = &efx_netdev_ops; 707 if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0) 708 net_dev->priv_flags |= IFF_UNICAST_FLT; 709 net_dev->ethtool_ops = &efx_ethtool_ops; 710 netif_set_tso_max_segs(net_dev, EFX_TSO_MAX_SEGS); 711 net_dev->min_mtu = EFX_MIN_MTU; 712 net_dev->max_mtu = EFX_MAX_MTU; 713 714 rtnl_lock(); 715 716 /* Enable resets to be scheduled and check whether any were 717 * already requested. If so, the NIC is probably hosed so we 718 * abort. 719 */ 720 if (efx->reset_pending) { 721 pci_err(efx->pci_dev, "aborting probe due to scheduled reset\n"); 722 rc = -EIO; 723 goto fail_locked; 724 } 725 726 rc = dev_alloc_name(net_dev, net_dev->name); 727 if (rc < 0) 728 goto fail_locked; 729 efx_update_name(efx); 730 731 /* Always start with carrier off; PHY events will detect the link */ 732 netif_carrier_off(net_dev); 733 734 rc = register_netdevice(net_dev); 735 if (rc) 736 goto fail_locked; 737 738 efx_for_each_channel(channel, efx) { 739 struct efx_tx_queue *tx_queue; 740 efx_for_each_channel_tx_queue(tx_queue, channel) 741 efx_init_tx_queue_core_txq(tx_queue); 742 } 743 744 efx_associate(efx); 745 746 efx->state = STATE_NET_DOWN; 747 748 rtnl_unlock(); 749 750 rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type); 751 if (rc) { 752 netif_err(efx, drv, efx->net_dev, 753 "failed to init net dev attributes\n"); 754 goto fail_registered; 755 } 756 757 efx_init_mcdi_logging(efx); 758 759 return 0; 760 761 fail_registered: 762 rtnl_lock(); 763 efx_dissociate(efx); 764 unregister_netdevice(net_dev); 765 fail_locked: 766 efx->state = STATE_UNINIT; 767 rtnl_unlock(); 768 netif_err(efx, drv, efx->net_dev, "could not register net dev\n"); 769 return rc; 770 } 771 772 static void efx_unregister_netdev(struct efx_nic *efx) 773 { 774 if (!efx->net_dev) 775 return; 776 777 if (WARN_ON(efx_netdev_priv(efx->net_dev) != efx)) 778 return; 779 780 if (efx_dev_registered(efx)) { 781 strscpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name)); 782 efx_fini_mcdi_logging(efx); 783 device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type); 784 unregister_netdev(efx->net_dev); 785 } 786 } 787 788 /************************************************************************** 789 * 790 * List of NICs we support 791 * 792 **************************************************************************/ 793 794 /* PCI device ID table */ 795 static const struct pci_device_id efx_pci_table[] = { 796 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903), /* SFC9120 PF */ 797 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 798 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903), /* SFC9120 VF */ 799 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 800 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923), /* SFC9140 PF */ 801 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 802 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923), /* SFC9140 VF */ 803 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 804 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03), /* SFC9220 PF */ 805 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 806 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03), /* SFC9220 VF */ 807 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 808 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0b03), /* SFC9250 PF */ 809 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 810 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1b03), /* SFC9250 VF */ 811 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 812 {0} /* end of list */ 813 }; 814 815 /************************************************************************** 816 * 817 * Data housekeeping 818 * 819 **************************************************************************/ 820 821 void efx_update_sw_stats(struct efx_nic *efx, u64 *stats) 822 { 823 u64 n_rx_nodesc_trunc = 0; 824 struct efx_channel *channel; 825 826 efx_for_each_channel(channel, efx) 827 n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc; 828 stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc; 829 stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops); 830 } 831 832 /************************************************************************** 833 * 834 * PCI interface 835 * 836 **************************************************************************/ 837 838 /* Main body of final NIC shutdown code 839 * This is called only at module unload (or hotplug removal). 840 */ 841 static void efx_pci_remove_main(struct efx_nic *efx) 842 { 843 /* Flush reset_work. It can no longer be scheduled since we 844 * are not READY. 845 */ 846 WARN_ON(efx_net_active(efx->state)); 847 efx_flush_reset_workqueue(efx); 848 849 efx_disable_interrupts(efx); 850 efx_clear_interrupt_affinity(efx); 851 efx_nic_fini_interrupt(efx); 852 efx_fini_port(efx); 853 efx->type->fini(efx); 854 efx_fini_napi(efx); 855 efx_remove_all(efx); 856 } 857 858 /* Final NIC shutdown 859 * This is called only at module unload (or hotplug removal). A PF can call 860 * this on its VFs to ensure they are unbound first. 861 */ 862 static void efx_pci_remove(struct pci_dev *pci_dev) 863 { 864 struct efx_probe_data *probe_data; 865 struct efx_nic *efx; 866 867 efx = pci_get_drvdata(pci_dev); 868 if (!efx) 869 return; 870 871 /* Mark the NIC as fini, then stop the interface */ 872 rtnl_lock(); 873 efx_dissociate(efx); 874 dev_close(efx->net_dev); 875 efx_disable_interrupts(efx); 876 efx->state = STATE_UNINIT; 877 rtnl_unlock(); 878 879 if (efx->type->sriov_fini) 880 efx->type->sriov_fini(efx); 881 882 efx_unregister_netdev(efx); 883 884 efx_mtd_remove(efx); 885 886 efx_pci_remove_main(efx); 887 888 efx_fini_io(efx); 889 pci_dbg(efx->pci_dev, "shutdown successful\n"); 890 891 efx_fini_struct(efx); 892 free_netdev(efx->net_dev); 893 probe_data = container_of(efx, struct efx_probe_data, efx); 894 kfree(probe_data); 895 896 pci_disable_pcie_error_reporting(pci_dev); 897 }; 898 899 /* NIC VPD information 900 * Called during probe to display the part number of the 901 * installed NIC. 902 */ 903 static void efx_probe_vpd_strings(struct efx_nic *efx) 904 { 905 struct pci_dev *dev = efx->pci_dev; 906 unsigned int vpd_size, kw_len; 907 u8 *vpd_data; 908 int start; 909 910 vpd_data = pci_vpd_alloc(dev, &vpd_size); 911 if (IS_ERR(vpd_data)) { 912 pci_warn(dev, "Unable to read VPD\n"); 913 return; 914 } 915 916 start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size, 917 PCI_VPD_RO_KEYWORD_PARTNO, &kw_len); 918 if (start < 0) 919 pci_err(dev, "Part number not found or incomplete\n"); 920 else 921 pci_info(dev, "Part Number : %.*s\n", kw_len, vpd_data + start); 922 923 start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size, 924 PCI_VPD_RO_KEYWORD_SERIALNO, &kw_len); 925 if (start < 0) 926 pci_err(dev, "Serial number not found or incomplete\n"); 927 else 928 efx->vpd_sn = kmemdup_nul(vpd_data + start, kw_len, GFP_KERNEL); 929 930 kfree(vpd_data); 931 } 932 933 934 /* Main body of NIC initialisation 935 * This is called at module load (or hotplug insertion, theoretically). 936 */ 937 static int efx_pci_probe_main(struct efx_nic *efx) 938 { 939 int rc; 940 941 /* Do start-of-day initialisation */ 942 rc = efx_probe_all(efx); 943 if (rc) 944 goto fail1; 945 946 efx_init_napi(efx); 947 948 down_write(&efx->filter_sem); 949 rc = efx->type->init(efx); 950 up_write(&efx->filter_sem); 951 if (rc) { 952 pci_err(efx->pci_dev, "failed to initialise NIC\n"); 953 goto fail3; 954 } 955 956 rc = efx_init_port(efx); 957 if (rc) { 958 netif_err(efx, probe, efx->net_dev, 959 "failed to initialise port\n"); 960 goto fail4; 961 } 962 963 rc = efx_nic_init_interrupt(efx); 964 if (rc) 965 goto fail5; 966 967 efx_set_interrupt_affinity(efx); 968 rc = efx_enable_interrupts(efx); 969 if (rc) 970 goto fail6; 971 972 return 0; 973 974 fail6: 975 efx_clear_interrupt_affinity(efx); 976 efx_nic_fini_interrupt(efx); 977 fail5: 978 efx_fini_port(efx); 979 fail4: 980 efx->type->fini(efx); 981 fail3: 982 efx_fini_napi(efx); 983 efx_remove_all(efx); 984 fail1: 985 return rc; 986 } 987 988 static int efx_pci_probe_post_io(struct efx_nic *efx) 989 { 990 struct net_device *net_dev = efx->net_dev; 991 int rc = efx_pci_probe_main(efx); 992 993 if (rc) 994 return rc; 995 996 if (efx->type->sriov_init) { 997 rc = efx->type->sriov_init(efx); 998 if (rc) 999 pci_err(efx->pci_dev, "SR-IOV can't be enabled rc %d\n", 1000 rc); 1001 } 1002 1003 /* Determine netdevice features */ 1004 net_dev->features |= (efx->type->offload_features | NETIF_F_SG | 1005 NETIF_F_TSO | NETIF_F_RXCSUM | NETIF_F_RXALL); 1006 if (efx->type->offload_features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM)) 1007 net_dev->features |= NETIF_F_TSO6; 1008 /* Check whether device supports TSO */ 1009 if (!efx->type->tso_versions || !efx->type->tso_versions(efx)) 1010 net_dev->features &= ~NETIF_F_ALL_TSO; 1011 /* Mask for features that also apply to VLAN devices */ 1012 net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG | 1013 NETIF_F_HIGHDMA | NETIF_F_ALL_TSO | 1014 NETIF_F_RXCSUM); 1015 1016 net_dev->hw_features |= net_dev->features & ~efx->fixed_features; 1017 1018 /* Disable receiving frames with bad FCS, by default. */ 1019 net_dev->features &= ~NETIF_F_RXALL; 1020 1021 /* Disable VLAN filtering by default. It may be enforced if 1022 * the feature is fixed (i.e. VLAN filters are required to 1023 * receive VLAN tagged packets due to vPort restrictions). 1024 */ 1025 net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER; 1026 net_dev->features |= efx->fixed_features; 1027 1028 rc = efx_register_netdev(efx); 1029 if (!rc) 1030 return 0; 1031 1032 efx_pci_remove_main(efx); 1033 return rc; 1034 } 1035 1036 /* NIC initialisation 1037 * 1038 * This is called at module load (or hotplug insertion, 1039 * theoretically). It sets up PCI mappings, resets the NIC, 1040 * sets up and registers the network devices with the kernel and hooks 1041 * the interrupt service routine. It does not prepare the device for 1042 * transmission; this is left to the first time one of the network 1043 * interfaces is brought up (i.e. efx_net_open). 1044 */ 1045 static int efx_pci_probe(struct pci_dev *pci_dev, 1046 const struct pci_device_id *entry) 1047 { 1048 struct efx_probe_data *probe_data, **probe_ptr; 1049 struct net_device *net_dev; 1050 struct efx_nic *efx; 1051 int rc; 1052 1053 /* Allocate probe data and struct efx_nic */ 1054 probe_data = kzalloc(sizeof(*probe_data), GFP_KERNEL); 1055 if (!probe_data) 1056 return -ENOMEM; 1057 probe_data->pci_dev = pci_dev; 1058 efx = &probe_data->efx; 1059 1060 /* Allocate and initialise a struct net_device */ 1061 net_dev = alloc_etherdev_mq(sizeof(probe_data), EFX_MAX_CORE_TX_QUEUES); 1062 if (!net_dev) 1063 return -ENOMEM; 1064 probe_ptr = netdev_priv(net_dev); 1065 *probe_ptr = probe_data; 1066 efx->net_dev = net_dev; 1067 efx->type = (const struct efx_nic_type *) entry->driver_data; 1068 efx->fixed_features |= NETIF_F_HIGHDMA; 1069 1070 pci_set_drvdata(pci_dev, efx); 1071 SET_NETDEV_DEV(net_dev, &pci_dev->dev); 1072 rc = efx_init_struct(efx, pci_dev); 1073 if (rc) 1074 goto fail1; 1075 efx->mdio.dev = net_dev; 1076 1077 pci_info(pci_dev, "Solarflare NIC detected\n"); 1078 1079 if (!efx->type->is_vf) 1080 efx_probe_vpd_strings(efx); 1081 1082 /* Set up basic I/O (BAR mappings etc) */ 1083 rc = efx_init_io(efx, efx->type->mem_bar(efx), efx->type->max_dma_mask, 1084 efx->type->mem_map_size(efx)); 1085 if (rc) 1086 goto fail2; 1087 1088 rc = efx_pci_probe_post_io(efx); 1089 if (rc) { 1090 /* On failure, retry once immediately. 1091 * If we aborted probe due to a scheduled reset, dismiss it. 1092 */ 1093 efx->reset_pending = 0; 1094 rc = efx_pci_probe_post_io(efx); 1095 if (rc) { 1096 /* On another failure, retry once more 1097 * after a 50-305ms delay. 1098 */ 1099 unsigned char r; 1100 1101 get_random_bytes(&r, 1); 1102 msleep((unsigned int)r + 50); 1103 efx->reset_pending = 0; 1104 rc = efx_pci_probe_post_io(efx); 1105 } 1106 } 1107 if (rc) 1108 goto fail3; 1109 1110 netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n"); 1111 1112 /* Try to create MTDs, but allow this to fail */ 1113 rtnl_lock(); 1114 rc = efx_mtd_probe(efx); 1115 rtnl_unlock(); 1116 if (rc && rc != -EPERM) 1117 netif_warn(efx, probe, efx->net_dev, 1118 "failed to create MTDs (%d)\n", rc); 1119 1120 (void)pci_enable_pcie_error_reporting(pci_dev); 1121 1122 if (efx->type->udp_tnl_push_ports) 1123 efx->type->udp_tnl_push_ports(efx); 1124 1125 return 0; 1126 1127 fail3: 1128 efx_fini_io(efx); 1129 fail2: 1130 efx_fini_struct(efx); 1131 fail1: 1132 WARN_ON(rc > 0); 1133 netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc); 1134 free_netdev(net_dev); 1135 return rc; 1136 } 1137 1138 /* efx_pci_sriov_configure returns the actual number of Virtual Functions 1139 * enabled on success 1140 */ 1141 #ifdef CONFIG_SFC_SRIOV 1142 static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs) 1143 { 1144 int rc; 1145 struct efx_nic *efx = pci_get_drvdata(dev); 1146 1147 if (efx->type->sriov_configure) { 1148 rc = efx->type->sriov_configure(efx, num_vfs); 1149 if (rc) 1150 return rc; 1151 else 1152 return num_vfs; 1153 } else 1154 return -EOPNOTSUPP; 1155 } 1156 #endif 1157 1158 static int efx_pm_freeze(struct device *dev) 1159 { 1160 struct efx_nic *efx = dev_get_drvdata(dev); 1161 1162 rtnl_lock(); 1163 1164 if (efx_net_active(efx->state)) { 1165 efx_device_detach_sync(efx); 1166 1167 efx_stop_all(efx); 1168 efx_disable_interrupts(efx); 1169 1170 efx->state = efx_freeze(efx->state); 1171 } 1172 1173 rtnl_unlock(); 1174 1175 return 0; 1176 } 1177 1178 static void efx_pci_shutdown(struct pci_dev *pci_dev) 1179 { 1180 struct efx_nic *efx = pci_get_drvdata(pci_dev); 1181 1182 if (!efx) 1183 return; 1184 1185 efx_pm_freeze(&pci_dev->dev); 1186 pci_disable_device(pci_dev); 1187 } 1188 1189 static int efx_pm_thaw(struct device *dev) 1190 { 1191 int rc; 1192 struct efx_nic *efx = dev_get_drvdata(dev); 1193 1194 rtnl_lock(); 1195 1196 if (efx_frozen(efx->state)) { 1197 rc = efx_enable_interrupts(efx); 1198 if (rc) 1199 goto fail; 1200 1201 mutex_lock(&efx->mac_lock); 1202 efx_mcdi_port_reconfigure(efx); 1203 mutex_unlock(&efx->mac_lock); 1204 1205 efx_start_all(efx); 1206 1207 efx_device_attach_if_not_resetting(efx); 1208 1209 efx->state = efx_thaw(efx->state); 1210 1211 efx->type->resume_wol(efx); 1212 } 1213 1214 rtnl_unlock(); 1215 1216 /* Reschedule any quenched resets scheduled during efx_pm_freeze() */ 1217 efx_queue_reset_work(efx); 1218 1219 return 0; 1220 1221 fail: 1222 rtnl_unlock(); 1223 1224 return rc; 1225 } 1226 1227 static int efx_pm_poweroff(struct device *dev) 1228 { 1229 struct pci_dev *pci_dev = to_pci_dev(dev); 1230 struct efx_nic *efx = pci_get_drvdata(pci_dev); 1231 1232 efx->type->fini(efx); 1233 1234 efx->reset_pending = 0; 1235 1236 pci_save_state(pci_dev); 1237 return pci_set_power_state(pci_dev, PCI_D3hot); 1238 } 1239 1240 /* Used for both resume and restore */ 1241 static int efx_pm_resume(struct device *dev) 1242 { 1243 struct pci_dev *pci_dev = to_pci_dev(dev); 1244 struct efx_nic *efx = pci_get_drvdata(pci_dev); 1245 int rc; 1246 1247 rc = pci_set_power_state(pci_dev, PCI_D0); 1248 if (rc) 1249 return rc; 1250 pci_restore_state(pci_dev); 1251 rc = pci_enable_device(pci_dev); 1252 if (rc) 1253 return rc; 1254 pci_set_master(efx->pci_dev); 1255 rc = efx->type->reset(efx, RESET_TYPE_ALL); 1256 if (rc) 1257 return rc; 1258 down_write(&efx->filter_sem); 1259 rc = efx->type->init(efx); 1260 up_write(&efx->filter_sem); 1261 if (rc) 1262 return rc; 1263 rc = efx_pm_thaw(dev); 1264 return rc; 1265 } 1266 1267 static int efx_pm_suspend(struct device *dev) 1268 { 1269 int rc; 1270 1271 efx_pm_freeze(dev); 1272 rc = efx_pm_poweroff(dev); 1273 if (rc) 1274 efx_pm_resume(dev); 1275 return rc; 1276 } 1277 1278 static const struct dev_pm_ops efx_pm_ops = { 1279 .suspend = efx_pm_suspend, 1280 .resume = efx_pm_resume, 1281 .freeze = efx_pm_freeze, 1282 .thaw = efx_pm_thaw, 1283 .poweroff = efx_pm_poweroff, 1284 .restore = efx_pm_resume, 1285 }; 1286 1287 static struct pci_driver efx_pci_driver = { 1288 .name = KBUILD_MODNAME, 1289 .id_table = efx_pci_table, 1290 .probe = efx_pci_probe, 1291 .remove = efx_pci_remove, 1292 .driver.pm = &efx_pm_ops, 1293 .shutdown = efx_pci_shutdown, 1294 .err_handler = &efx_err_handlers, 1295 #ifdef CONFIG_SFC_SRIOV 1296 .sriov_configure = efx_pci_sriov_configure, 1297 #endif 1298 }; 1299 1300 /************************************************************************** 1301 * 1302 * Kernel module interface 1303 * 1304 *************************************************************************/ 1305 1306 static int __init efx_init_module(void) 1307 { 1308 int rc; 1309 1310 printk(KERN_INFO "Solarflare NET driver\n"); 1311 1312 rc = register_netdevice_notifier(&efx_netdev_notifier); 1313 if (rc) 1314 goto err_notifier; 1315 1316 rc = efx_create_reset_workqueue(); 1317 if (rc) 1318 goto err_reset; 1319 1320 rc = pci_register_driver(&efx_pci_driver); 1321 if (rc < 0) 1322 goto err_pci; 1323 1324 rc = pci_register_driver(&ef100_pci_driver); 1325 if (rc < 0) 1326 goto err_pci_ef100; 1327 1328 return 0; 1329 1330 err_pci_ef100: 1331 pci_unregister_driver(&efx_pci_driver); 1332 err_pci: 1333 efx_destroy_reset_workqueue(); 1334 err_reset: 1335 unregister_netdevice_notifier(&efx_netdev_notifier); 1336 err_notifier: 1337 return rc; 1338 } 1339 1340 static void __exit efx_exit_module(void) 1341 { 1342 printk(KERN_INFO "Solarflare NET driver unloading\n"); 1343 1344 pci_unregister_driver(&ef100_pci_driver); 1345 pci_unregister_driver(&efx_pci_driver); 1346 efx_destroy_reset_workqueue(); 1347 unregister_netdevice_notifier(&efx_netdev_notifier); 1348 1349 } 1350 1351 module_init(efx_init_module); 1352 module_exit(efx_exit_module); 1353 1354 MODULE_AUTHOR("Solarflare Communications and " 1355 "Michael Brown <mbrown@fensystems.co.uk>"); 1356 MODULE_DESCRIPTION("Solarflare network driver"); 1357 MODULE_LICENSE("GPL"); 1358 MODULE_DEVICE_TABLE(pci, efx_pci_table); 1359