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 #define EFX_ASSERT_RESET_SERIALISED(efx) \ 110 do { \ 111 if ((efx->state == STATE_READY) || \ 112 (efx->state == STATE_RECOVERY) || \ 113 (efx->state == STATE_DISABLED)) \ 114 ASSERT_RTNL(); \ 115 } while (0) 116 117 /************************************************************************** 118 * 119 * Port handling 120 * 121 **************************************************************************/ 122 123 static void efx_fini_port(struct efx_nic *efx); 124 125 static int efx_probe_port(struct efx_nic *efx) 126 { 127 int rc; 128 129 netif_dbg(efx, probe, efx->net_dev, "create port\n"); 130 131 if (phy_flash_cfg) 132 efx->phy_mode = PHY_MODE_SPECIAL; 133 134 /* Connect up MAC/PHY operations table */ 135 rc = efx->type->probe_port(efx); 136 if (rc) 137 return rc; 138 139 /* Initialise MAC address to permanent address */ 140 eth_hw_addr_set(efx->net_dev, efx->net_dev->perm_addr); 141 142 return 0; 143 } 144 145 static int efx_init_port(struct efx_nic *efx) 146 { 147 int rc; 148 149 netif_dbg(efx, drv, efx->net_dev, "init port\n"); 150 151 mutex_lock(&efx->mac_lock); 152 153 efx->port_initialized = true; 154 155 /* Ensure the PHY advertises the correct flow control settings */ 156 rc = efx_mcdi_port_reconfigure(efx); 157 if (rc && rc != -EPERM) 158 goto fail; 159 160 mutex_unlock(&efx->mac_lock); 161 return 0; 162 163 fail: 164 mutex_unlock(&efx->mac_lock); 165 return rc; 166 } 167 168 static void efx_fini_port(struct efx_nic *efx) 169 { 170 netif_dbg(efx, drv, efx->net_dev, "shut down port\n"); 171 172 if (!efx->port_initialized) 173 return; 174 175 efx->port_initialized = false; 176 177 efx->link_state.up = false; 178 efx_link_status_changed(efx); 179 } 180 181 static void efx_remove_port(struct efx_nic *efx) 182 { 183 netif_dbg(efx, drv, efx->net_dev, "destroying port\n"); 184 185 efx->type->remove_port(efx); 186 } 187 188 /************************************************************************** 189 * 190 * NIC handling 191 * 192 **************************************************************************/ 193 194 static LIST_HEAD(efx_primary_list); 195 static LIST_HEAD(efx_unassociated_list); 196 197 static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right) 198 { 199 return left->type == right->type && 200 left->vpd_sn && right->vpd_sn && 201 !strcmp(left->vpd_sn, right->vpd_sn); 202 } 203 204 static void efx_associate(struct efx_nic *efx) 205 { 206 struct efx_nic *other, *next; 207 208 if (efx->primary == efx) { 209 /* Adding primary function; look for secondaries */ 210 211 netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n"); 212 list_add_tail(&efx->node, &efx_primary_list); 213 214 list_for_each_entry_safe(other, next, &efx_unassociated_list, 215 node) { 216 if (efx_same_controller(efx, other)) { 217 list_del(&other->node); 218 netif_dbg(other, probe, other->net_dev, 219 "moving to secondary list of %s %s\n", 220 pci_name(efx->pci_dev), 221 efx->net_dev->name); 222 list_add_tail(&other->node, 223 &efx->secondary_list); 224 other->primary = efx; 225 } 226 } 227 } else { 228 /* Adding secondary function; look for primary */ 229 230 list_for_each_entry(other, &efx_primary_list, node) { 231 if (efx_same_controller(efx, other)) { 232 netif_dbg(efx, probe, efx->net_dev, 233 "adding to secondary list of %s %s\n", 234 pci_name(other->pci_dev), 235 other->net_dev->name); 236 list_add_tail(&efx->node, 237 &other->secondary_list); 238 efx->primary = other; 239 return; 240 } 241 } 242 243 netif_dbg(efx, probe, efx->net_dev, 244 "adding to unassociated list\n"); 245 list_add_tail(&efx->node, &efx_unassociated_list); 246 } 247 } 248 249 static void efx_dissociate(struct efx_nic *efx) 250 { 251 struct efx_nic *other, *next; 252 253 list_del(&efx->node); 254 efx->primary = NULL; 255 256 list_for_each_entry_safe(other, next, &efx->secondary_list, node) { 257 list_del(&other->node); 258 netif_dbg(other, probe, other->net_dev, 259 "moving to unassociated list\n"); 260 list_add_tail(&other->node, &efx_unassociated_list); 261 other->primary = NULL; 262 } 263 } 264 265 static int efx_probe_nic(struct efx_nic *efx) 266 { 267 int rc; 268 269 netif_dbg(efx, probe, efx->net_dev, "creating NIC\n"); 270 271 /* Carry out hardware-type specific initialisation */ 272 rc = efx->type->probe(efx); 273 if (rc) 274 return rc; 275 276 do { 277 if (!efx->max_channels || !efx->max_tx_channels) { 278 netif_err(efx, drv, efx->net_dev, 279 "Insufficient resources to allocate" 280 " any channels\n"); 281 rc = -ENOSPC; 282 goto fail1; 283 } 284 285 /* Determine the number of channels and queues by trying 286 * to hook in MSI-X interrupts. 287 */ 288 rc = efx_probe_interrupts(efx); 289 if (rc) 290 goto fail1; 291 292 rc = efx_set_channels(efx); 293 if (rc) 294 goto fail1; 295 296 /* dimension_resources can fail with EAGAIN */ 297 rc = efx->type->dimension_resources(efx); 298 if (rc != 0 && rc != -EAGAIN) 299 goto fail2; 300 301 if (rc == -EAGAIN) 302 /* try again with new max_channels */ 303 efx_remove_interrupts(efx); 304 305 } while (rc == -EAGAIN); 306 307 if (efx->n_channels > 1) 308 netdev_rss_key_fill(efx->rss_context.rx_hash_key, 309 sizeof(efx->rss_context.rx_hash_key)); 310 efx_set_default_rx_indir_table(efx, &efx->rss_context); 311 312 /* Initialise the interrupt moderation settings */ 313 efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000); 314 efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true, 315 true); 316 317 return 0; 318 319 fail2: 320 efx_remove_interrupts(efx); 321 fail1: 322 efx->type->remove(efx); 323 return rc; 324 } 325 326 static void efx_remove_nic(struct efx_nic *efx) 327 { 328 netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n"); 329 330 efx_remove_interrupts(efx); 331 efx->type->remove(efx); 332 } 333 334 /************************************************************************** 335 * 336 * NIC startup/shutdown 337 * 338 *************************************************************************/ 339 340 static int efx_probe_all(struct efx_nic *efx) 341 { 342 int rc; 343 344 rc = efx_probe_nic(efx); 345 if (rc) { 346 netif_err(efx, probe, efx->net_dev, "failed to create NIC\n"); 347 goto fail1; 348 } 349 350 rc = efx_probe_port(efx); 351 if (rc) { 352 netif_err(efx, probe, efx->net_dev, "failed to create port\n"); 353 goto fail2; 354 } 355 356 BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT); 357 if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) { 358 rc = -EINVAL; 359 goto fail3; 360 } 361 362 #ifdef CONFIG_SFC_SRIOV 363 rc = efx->type->vswitching_probe(efx); 364 if (rc) /* not fatal; the PF will still work fine */ 365 netif_warn(efx, probe, efx->net_dev, 366 "failed to setup vswitching rc=%d;" 367 " VFs may not function\n", rc); 368 #endif 369 370 rc = efx_probe_filters(efx); 371 if (rc) { 372 netif_err(efx, probe, efx->net_dev, 373 "failed to create filter tables\n"); 374 goto fail4; 375 } 376 377 rc = efx_probe_channels(efx); 378 if (rc) 379 goto fail5; 380 381 return 0; 382 383 fail5: 384 efx_remove_filters(efx); 385 fail4: 386 #ifdef CONFIG_SFC_SRIOV 387 efx->type->vswitching_remove(efx); 388 #endif 389 fail3: 390 efx_remove_port(efx); 391 fail2: 392 efx_remove_nic(efx); 393 fail1: 394 return rc; 395 } 396 397 static void efx_remove_all(struct efx_nic *efx) 398 { 399 rtnl_lock(); 400 efx_xdp_setup_prog(efx, NULL); 401 rtnl_unlock(); 402 403 efx_remove_channels(efx); 404 efx_remove_filters(efx); 405 #ifdef CONFIG_SFC_SRIOV 406 efx->type->vswitching_remove(efx); 407 #endif 408 efx_remove_port(efx); 409 efx_remove_nic(efx); 410 } 411 412 /************************************************************************** 413 * 414 * Interrupt moderation 415 * 416 **************************************************************************/ 417 unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs) 418 { 419 if (usecs == 0) 420 return 0; 421 if (usecs * 1000 < efx->timer_quantum_ns) 422 return 1; /* never round down to 0 */ 423 return usecs * 1000 / efx->timer_quantum_ns; 424 } 425 426 unsigned int efx_ticks_to_usecs(struct efx_nic *efx, unsigned int ticks) 427 { 428 /* We must round up when converting ticks to microseconds 429 * because we round down when converting the other way. 430 */ 431 return DIV_ROUND_UP(ticks * efx->timer_quantum_ns, 1000); 432 } 433 434 /* Set interrupt moderation parameters */ 435 int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs, 436 unsigned int rx_usecs, bool rx_adaptive, 437 bool rx_may_override_tx) 438 { 439 struct efx_channel *channel; 440 unsigned int timer_max_us; 441 442 EFX_ASSERT_RESET_SERIALISED(efx); 443 444 timer_max_us = efx->timer_max_ns / 1000; 445 446 if (tx_usecs > timer_max_us || rx_usecs > timer_max_us) 447 return -EINVAL; 448 449 if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 && 450 !rx_may_override_tx) { 451 netif_err(efx, drv, efx->net_dev, "Channels are shared. " 452 "RX and TX IRQ moderation must be equal\n"); 453 return -EINVAL; 454 } 455 456 efx->irq_rx_adaptive = rx_adaptive; 457 efx->irq_rx_moderation_us = rx_usecs; 458 efx_for_each_channel(channel, efx) { 459 if (efx_channel_has_rx_queue(channel)) 460 channel->irq_moderation_us = rx_usecs; 461 else if (efx_channel_has_tx_queues(channel)) 462 channel->irq_moderation_us = tx_usecs; 463 else if (efx_channel_is_xdp_tx(channel)) 464 channel->irq_moderation_us = tx_usecs; 465 } 466 467 return 0; 468 } 469 470 void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs, 471 unsigned int *rx_usecs, bool *rx_adaptive) 472 { 473 *rx_adaptive = efx->irq_rx_adaptive; 474 *rx_usecs = efx->irq_rx_moderation_us; 475 476 /* If channels are shared between RX and TX, so is IRQ 477 * moderation. Otherwise, IRQ moderation is the same for all 478 * TX channels and is not adaptive. 479 */ 480 if (efx->tx_channel_offset == 0) { 481 *tx_usecs = *rx_usecs; 482 } else { 483 struct efx_channel *tx_channel; 484 485 tx_channel = efx->channel[efx->tx_channel_offset]; 486 *tx_usecs = tx_channel->irq_moderation_us; 487 } 488 } 489 490 /************************************************************************** 491 * 492 * ioctls 493 * 494 *************************************************************************/ 495 496 /* Net device ioctl 497 * Context: process, rtnl_lock() held. 498 */ 499 static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd) 500 { 501 struct efx_nic *efx = netdev_priv(net_dev); 502 struct mii_ioctl_data *data = if_mii(ifr); 503 504 if (cmd == SIOCSHWTSTAMP) 505 return efx_ptp_set_ts_config(efx, ifr); 506 if (cmd == SIOCGHWTSTAMP) 507 return efx_ptp_get_ts_config(efx, ifr); 508 509 /* Convert phy_id from older PRTAD/DEVAD format */ 510 if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) && 511 (data->phy_id & 0xfc00) == 0x0400) 512 data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400; 513 514 return mdio_mii_ioctl(&efx->mdio, data, cmd); 515 } 516 517 /************************************************************************** 518 * 519 * Kernel net device interface 520 * 521 *************************************************************************/ 522 523 /* Context: process, rtnl_lock() held. */ 524 int efx_net_open(struct net_device *net_dev) 525 { 526 struct efx_nic *efx = netdev_priv(net_dev); 527 int rc; 528 529 netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n", 530 raw_smp_processor_id()); 531 532 rc = efx_check_disabled(efx); 533 if (rc) 534 return rc; 535 if (efx->phy_mode & PHY_MODE_SPECIAL) 536 return -EBUSY; 537 if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL)) 538 return -EIO; 539 540 /* Notify the kernel of the link state polled during driver load, 541 * before the monitor starts running */ 542 efx_link_status_changed(efx); 543 544 efx_start_all(efx); 545 if (efx->state == STATE_DISABLED || efx->reset_pending) 546 netif_device_detach(efx->net_dev); 547 efx_selftest_async_start(efx); 548 return 0; 549 } 550 551 /* Context: process, rtnl_lock() held. 552 * Note that the kernel will ignore our return code; this method 553 * should really be a void. 554 */ 555 int efx_net_stop(struct net_device *net_dev) 556 { 557 struct efx_nic *efx = netdev_priv(net_dev); 558 559 netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n", 560 raw_smp_processor_id()); 561 562 /* Stop the device and flush all the channels */ 563 efx_stop_all(efx); 564 565 return 0; 566 } 567 568 static int efx_vlan_rx_add_vid(struct net_device *net_dev, __be16 proto, u16 vid) 569 { 570 struct efx_nic *efx = netdev_priv(net_dev); 571 572 if (efx->type->vlan_rx_add_vid) 573 return efx->type->vlan_rx_add_vid(efx, proto, vid); 574 else 575 return -EOPNOTSUPP; 576 } 577 578 static int efx_vlan_rx_kill_vid(struct net_device *net_dev, __be16 proto, u16 vid) 579 { 580 struct efx_nic *efx = netdev_priv(net_dev); 581 582 if (efx->type->vlan_rx_kill_vid) 583 return efx->type->vlan_rx_kill_vid(efx, proto, vid); 584 else 585 return -EOPNOTSUPP; 586 } 587 588 static const struct net_device_ops efx_netdev_ops = { 589 .ndo_open = efx_net_open, 590 .ndo_stop = efx_net_stop, 591 .ndo_get_stats64 = efx_net_stats, 592 .ndo_tx_timeout = efx_watchdog, 593 .ndo_start_xmit = efx_hard_start_xmit, 594 .ndo_validate_addr = eth_validate_addr, 595 .ndo_eth_ioctl = efx_ioctl, 596 .ndo_change_mtu = efx_change_mtu, 597 .ndo_set_mac_address = efx_set_mac_address, 598 .ndo_set_rx_mode = efx_set_rx_mode, 599 .ndo_set_features = efx_set_features, 600 .ndo_features_check = efx_features_check, 601 .ndo_vlan_rx_add_vid = efx_vlan_rx_add_vid, 602 .ndo_vlan_rx_kill_vid = efx_vlan_rx_kill_vid, 603 #ifdef CONFIG_SFC_SRIOV 604 .ndo_set_vf_mac = efx_sriov_set_vf_mac, 605 .ndo_set_vf_vlan = efx_sriov_set_vf_vlan, 606 .ndo_set_vf_spoofchk = efx_sriov_set_vf_spoofchk, 607 .ndo_get_vf_config = efx_sriov_get_vf_config, 608 .ndo_set_vf_link_state = efx_sriov_set_vf_link_state, 609 #endif 610 .ndo_get_phys_port_id = efx_get_phys_port_id, 611 .ndo_get_phys_port_name = efx_get_phys_port_name, 612 .ndo_setup_tc = efx_setup_tc, 613 #ifdef CONFIG_RFS_ACCEL 614 .ndo_rx_flow_steer = efx_filter_rfs, 615 #endif 616 .ndo_xdp_xmit = efx_xdp_xmit, 617 .ndo_bpf = efx_xdp 618 }; 619 620 static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog) 621 { 622 struct bpf_prog *old_prog; 623 624 if (efx->xdp_rxq_info_failed) { 625 netif_err(efx, drv, efx->net_dev, 626 "Unable to bind XDP program due to previous failure of rxq_info\n"); 627 return -EINVAL; 628 } 629 630 if (prog && efx->net_dev->mtu > efx_xdp_max_mtu(efx)) { 631 netif_err(efx, drv, efx->net_dev, 632 "Unable to configure XDP with MTU of %d (max: %d)\n", 633 efx->net_dev->mtu, efx_xdp_max_mtu(efx)); 634 return -EINVAL; 635 } 636 637 old_prog = rtnl_dereference(efx->xdp_prog); 638 rcu_assign_pointer(efx->xdp_prog, prog); 639 /* Release the reference that was originally passed by the caller. */ 640 if (old_prog) 641 bpf_prog_put(old_prog); 642 643 return 0; 644 } 645 646 /* Context: process, rtnl_lock() held. */ 647 static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp) 648 { 649 struct efx_nic *efx = netdev_priv(dev); 650 651 switch (xdp->command) { 652 case XDP_SETUP_PROG: 653 return efx_xdp_setup_prog(efx, xdp->prog); 654 default: 655 return -EINVAL; 656 } 657 } 658 659 static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs, 660 u32 flags) 661 { 662 struct efx_nic *efx = netdev_priv(dev); 663 664 if (!netif_running(dev)) 665 return -EINVAL; 666 667 return efx_xdp_tx_buffers(efx, n, xdpfs, flags & XDP_XMIT_FLUSH); 668 } 669 670 static void efx_update_name(struct efx_nic *efx) 671 { 672 strcpy(efx->name, efx->net_dev->name); 673 efx_mtd_rename(efx); 674 efx_set_channel_names(efx); 675 } 676 677 static int efx_netdev_event(struct notifier_block *this, 678 unsigned long event, void *ptr) 679 { 680 struct net_device *net_dev = netdev_notifier_info_to_dev(ptr); 681 682 if ((net_dev->netdev_ops == &efx_netdev_ops) && 683 event == NETDEV_CHANGENAME) 684 efx_update_name(netdev_priv(net_dev)); 685 686 return NOTIFY_DONE; 687 } 688 689 static struct notifier_block efx_netdev_notifier = { 690 .notifier_call = efx_netdev_event, 691 }; 692 693 static ssize_t phy_type_show(struct device *dev, 694 struct device_attribute *attr, char *buf) 695 { 696 struct efx_nic *efx = dev_get_drvdata(dev); 697 return sprintf(buf, "%d\n", efx->phy_type); 698 } 699 static DEVICE_ATTR_RO(phy_type); 700 701 static int efx_register_netdev(struct efx_nic *efx) 702 { 703 struct net_device *net_dev = efx->net_dev; 704 struct efx_channel *channel; 705 int rc; 706 707 net_dev->watchdog_timeo = 5 * HZ; 708 net_dev->irq = efx->pci_dev->irq; 709 net_dev->netdev_ops = &efx_netdev_ops; 710 if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0) 711 net_dev->priv_flags |= IFF_UNICAST_FLT; 712 net_dev->ethtool_ops = &efx_ethtool_ops; 713 netif_set_gso_max_segs(net_dev, EFX_TSO_MAX_SEGS); 714 net_dev->min_mtu = EFX_MIN_MTU; 715 net_dev->max_mtu = EFX_MAX_MTU; 716 717 rtnl_lock(); 718 719 /* Enable resets to be scheduled and check whether any were 720 * already requested. If so, the NIC is probably hosed so we 721 * abort. 722 */ 723 efx->state = STATE_READY; 724 smp_mb(); /* ensure we change state before checking reset_pending */ 725 if (efx->reset_pending) { 726 pci_err(efx->pci_dev, "aborting probe due to scheduled reset\n"); 727 rc = -EIO; 728 goto fail_locked; 729 } 730 731 rc = dev_alloc_name(net_dev, net_dev->name); 732 if (rc < 0) 733 goto fail_locked; 734 efx_update_name(efx); 735 736 /* Always start with carrier off; PHY events will detect the link */ 737 netif_carrier_off(net_dev); 738 739 rc = register_netdevice(net_dev); 740 if (rc) 741 goto fail_locked; 742 743 efx_for_each_channel(channel, efx) { 744 struct efx_tx_queue *tx_queue; 745 efx_for_each_channel_tx_queue(tx_queue, channel) 746 efx_init_tx_queue_core_txq(tx_queue); 747 } 748 749 efx_associate(efx); 750 751 rtnl_unlock(); 752 753 rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type); 754 if (rc) { 755 netif_err(efx, drv, efx->net_dev, 756 "failed to init net dev attributes\n"); 757 goto fail_registered; 758 } 759 760 efx_init_mcdi_logging(efx); 761 762 return 0; 763 764 fail_registered: 765 rtnl_lock(); 766 efx_dissociate(efx); 767 unregister_netdevice(net_dev); 768 fail_locked: 769 efx->state = STATE_UNINIT; 770 rtnl_unlock(); 771 netif_err(efx, drv, efx->net_dev, "could not register net dev\n"); 772 return rc; 773 } 774 775 static void efx_unregister_netdev(struct efx_nic *efx) 776 { 777 if (!efx->net_dev) 778 return; 779 780 BUG_ON(netdev_priv(efx->net_dev) != efx); 781 782 if (efx_dev_registered(efx)) { 783 strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name)); 784 efx_fini_mcdi_logging(efx); 785 device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type); 786 unregister_netdev(efx->net_dev); 787 } 788 } 789 790 /************************************************************************** 791 * 792 * List of NICs we support 793 * 794 **************************************************************************/ 795 796 /* PCI device ID table */ 797 static const struct pci_device_id efx_pci_table[] = { 798 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803), /* SFC9020 */ 799 .driver_data = (unsigned long) &siena_a0_nic_type}, 800 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813), /* SFL9021 */ 801 .driver_data = (unsigned long) &siena_a0_nic_type}, 802 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903), /* SFC9120 PF */ 803 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 804 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903), /* SFC9120 VF */ 805 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 806 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923), /* SFC9140 PF */ 807 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 808 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923), /* SFC9140 VF */ 809 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 810 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03), /* SFC9220 PF */ 811 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 812 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03), /* SFC9220 VF */ 813 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 814 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0b03), /* SFC9250 PF */ 815 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 816 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1b03), /* SFC9250 VF */ 817 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 818 {0} /* end of list */ 819 }; 820 821 /************************************************************************** 822 * 823 * Data housekeeping 824 * 825 **************************************************************************/ 826 827 void efx_update_sw_stats(struct efx_nic *efx, u64 *stats) 828 { 829 u64 n_rx_nodesc_trunc = 0; 830 struct efx_channel *channel; 831 832 efx_for_each_channel(channel, efx) 833 n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc; 834 stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc; 835 stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops); 836 } 837 838 /************************************************************************** 839 * 840 * PCI interface 841 * 842 **************************************************************************/ 843 844 /* Main body of final NIC shutdown code 845 * This is called only at module unload (or hotplug removal). 846 */ 847 static void efx_pci_remove_main(struct efx_nic *efx) 848 { 849 /* Flush reset_work. It can no longer be scheduled since we 850 * are not READY. 851 */ 852 BUG_ON(efx->state == STATE_READY); 853 efx_flush_reset_workqueue(efx); 854 855 efx_disable_interrupts(efx); 856 efx_clear_interrupt_affinity(efx); 857 efx_nic_fini_interrupt(efx); 858 efx_fini_port(efx); 859 efx->type->fini(efx); 860 efx_fini_napi(efx); 861 efx_remove_all(efx); 862 } 863 864 /* Final NIC shutdown 865 * This is called only at module unload (or hotplug removal). A PF can call 866 * this on its VFs to ensure they are unbound first. 867 */ 868 static void efx_pci_remove(struct pci_dev *pci_dev) 869 { 870 struct efx_nic *efx; 871 872 efx = pci_get_drvdata(pci_dev); 873 if (!efx) 874 return; 875 876 /* Mark the NIC as fini, then stop the interface */ 877 rtnl_lock(); 878 efx_dissociate(efx); 879 dev_close(efx->net_dev); 880 efx_disable_interrupts(efx); 881 efx->state = STATE_UNINIT; 882 rtnl_unlock(); 883 884 if (efx->type->sriov_fini) 885 efx->type->sriov_fini(efx); 886 887 efx_unregister_netdev(efx); 888 889 efx_mtd_remove(efx); 890 891 efx_pci_remove_main(efx); 892 893 efx_fini_io(efx); 894 netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n"); 895 896 efx_fini_struct(efx); 897 free_netdev(efx->net_dev); 898 899 pci_disable_pcie_error_reporting(pci_dev); 900 }; 901 902 /* NIC VPD information 903 * Called during probe to display the part number of the 904 * installed NIC. 905 */ 906 static void efx_probe_vpd_strings(struct efx_nic *efx) 907 { 908 struct pci_dev *dev = efx->pci_dev; 909 unsigned int vpd_size, kw_len; 910 u8 *vpd_data; 911 int start; 912 913 vpd_data = pci_vpd_alloc(dev, &vpd_size); 914 if (IS_ERR(vpd_data)) { 915 pci_warn(dev, "Unable to read VPD\n"); 916 return; 917 } 918 919 start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size, 920 PCI_VPD_RO_KEYWORD_PARTNO, &kw_len); 921 if (start < 0) 922 pci_err(dev, "Part number not found or incomplete\n"); 923 else 924 pci_info(dev, "Part Number : %.*s\n", kw_len, vpd_data + start); 925 926 start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size, 927 PCI_VPD_RO_KEYWORD_SERIALNO, &kw_len); 928 if (start < 0) 929 pci_err(dev, "Serial number not found or incomplete\n"); 930 else 931 efx->vpd_sn = kmemdup_nul(vpd_data + start, kw_len, GFP_KERNEL); 932 933 kfree(vpd_data); 934 } 935 936 937 /* Main body of NIC initialisation 938 * This is called at module load (or hotplug insertion, theoretically). 939 */ 940 static int efx_pci_probe_main(struct efx_nic *efx) 941 { 942 int rc; 943 944 /* Do start-of-day initialisation */ 945 rc = efx_probe_all(efx); 946 if (rc) 947 goto fail1; 948 949 efx_init_napi(efx); 950 951 down_write(&efx->filter_sem); 952 rc = efx->type->init(efx); 953 up_write(&efx->filter_sem); 954 if (rc) { 955 pci_err(efx->pci_dev, "failed to initialise NIC\n"); 956 goto fail3; 957 } 958 959 rc = efx_init_port(efx); 960 if (rc) { 961 netif_err(efx, probe, efx->net_dev, 962 "failed to initialise port\n"); 963 goto fail4; 964 } 965 966 rc = efx_nic_init_interrupt(efx); 967 if (rc) 968 goto fail5; 969 970 efx_set_interrupt_affinity(efx); 971 rc = efx_enable_interrupts(efx); 972 if (rc) 973 goto fail6; 974 975 return 0; 976 977 fail6: 978 efx_clear_interrupt_affinity(efx); 979 efx_nic_fini_interrupt(efx); 980 fail5: 981 efx_fini_port(efx); 982 fail4: 983 efx->type->fini(efx); 984 fail3: 985 efx_fini_napi(efx); 986 efx_remove_all(efx); 987 fail1: 988 return rc; 989 } 990 991 static int efx_pci_probe_post_io(struct efx_nic *efx) 992 { 993 struct net_device *net_dev = efx->net_dev; 994 int rc = efx_pci_probe_main(efx); 995 996 if (rc) 997 return rc; 998 999 if (efx->type->sriov_init) { 1000 rc = efx->type->sriov_init(efx); 1001 if (rc) 1002 pci_err(efx->pci_dev, "SR-IOV can't be enabled rc %d\n", 1003 rc); 1004 } 1005 1006 /* Determine netdevice features */ 1007 net_dev->features |= (efx->type->offload_features | NETIF_F_SG | 1008 NETIF_F_TSO | NETIF_F_RXCSUM | NETIF_F_RXALL); 1009 if (efx->type->offload_features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM)) 1010 net_dev->features |= NETIF_F_TSO6; 1011 /* Check whether device supports TSO */ 1012 if (!efx->type->tso_versions || !efx->type->tso_versions(efx)) 1013 net_dev->features &= ~NETIF_F_ALL_TSO; 1014 /* Mask for features that also apply to VLAN devices */ 1015 net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG | 1016 NETIF_F_HIGHDMA | NETIF_F_ALL_TSO | 1017 NETIF_F_RXCSUM); 1018 1019 net_dev->hw_features |= net_dev->features & ~efx->fixed_features; 1020 1021 /* Disable receiving frames with bad FCS, by default. */ 1022 net_dev->features &= ~NETIF_F_RXALL; 1023 1024 /* Disable VLAN filtering by default. It may be enforced if 1025 * the feature is fixed (i.e. VLAN filters are required to 1026 * receive VLAN tagged packets due to vPort restrictions). 1027 */ 1028 net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER; 1029 net_dev->features |= efx->fixed_features; 1030 1031 rc = efx_register_netdev(efx); 1032 if (!rc) 1033 return 0; 1034 1035 efx_pci_remove_main(efx); 1036 return rc; 1037 } 1038 1039 /* NIC initialisation 1040 * 1041 * This is called at module load (or hotplug insertion, 1042 * theoretically). It sets up PCI mappings, resets the NIC, 1043 * sets up and registers the network devices with the kernel and hooks 1044 * the interrupt service routine. It does not prepare the device for 1045 * transmission; this is left to the first time one of the network 1046 * interfaces is brought up (i.e. efx_net_open). 1047 */ 1048 static int efx_pci_probe(struct pci_dev *pci_dev, 1049 const struct pci_device_id *entry) 1050 { 1051 struct net_device *net_dev; 1052 struct efx_nic *efx; 1053 int rc; 1054 1055 /* Allocate and initialise a struct net_device and struct efx_nic */ 1056 net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES, 1057 EFX_MAX_RX_QUEUES); 1058 if (!net_dev) 1059 return -ENOMEM; 1060 efx = netdev_priv(net_dev); 1061 efx->type = (const struct efx_nic_type *) entry->driver_data; 1062 efx->fixed_features |= NETIF_F_HIGHDMA; 1063 1064 pci_set_drvdata(pci_dev, efx); 1065 SET_NETDEV_DEV(net_dev, &pci_dev->dev); 1066 rc = efx_init_struct(efx, pci_dev, net_dev); 1067 if (rc) 1068 goto fail1; 1069 1070 pci_info(pci_dev, "Solarflare NIC detected\n"); 1071 1072 if (!efx->type->is_vf) 1073 efx_probe_vpd_strings(efx); 1074 1075 /* Set up basic I/O (BAR mappings etc) */ 1076 rc = efx_init_io(efx, efx->type->mem_bar(efx), efx->type->max_dma_mask, 1077 efx->type->mem_map_size(efx)); 1078 if (rc) 1079 goto fail2; 1080 1081 rc = efx_pci_probe_post_io(efx); 1082 if (rc) { 1083 /* On failure, retry once immediately. 1084 * If we aborted probe due to a scheduled reset, dismiss it. 1085 */ 1086 efx->reset_pending = 0; 1087 rc = efx_pci_probe_post_io(efx); 1088 if (rc) { 1089 /* On another failure, retry once more 1090 * after a 50-305ms delay. 1091 */ 1092 unsigned char r; 1093 1094 get_random_bytes(&r, 1); 1095 msleep((unsigned int)r + 50); 1096 efx->reset_pending = 0; 1097 rc = efx_pci_probe_post_io(efx); 1098 } 1099 } 1100 if (rc) 1101 goto fail3; 1102 1103 netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n"); 1104 1105 /* Try to create MTDs, but allow this to fail */ 1106 rtnl_lock(); 1107 rc = efx_mtd_probe(efx); 1108 rtnl_unlock(); 1109 if (rc && rc != -EPERM) 1110 netif_warn(efx, probe, efx->net_dev, 1111 "failed to create MTDs (%d)\n", rc); 1112 1113 (void)pci_enable_pcie_error_reporting(pci_dev); 1114 1115 if (efx->type->udp_tnl_push_ports) 1116 efx->type->udp_tnl_push_ports(efx); 1117 1118 return 0; 1119 1120 fail3: 1121 efx_fini_io(efx); 1122 fail2: 1123 efx_fini_struct(efx); 1124 fail1: 1125 WARN_ON(rc > 0); 1126 netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc); 1127 free_netdev(net_dev); 1128 return rc; 1129 } 1130 1131 /* efx_pci_sriov_configure returns the actual number of Virtual Functions 1132 * enabled on success 1133 */ 1134 #ifdef CONFIG_SFC_SRIOV 1135 static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs) 1136 { 1137 int rc; 1138 struct efx_nic *efx = pci_get_drvdata(dev); 1139 1140 if (efx->type->sriov_configure) { 1141 rc = efx->type->sriov_configure(efx, num_vfs); 1142 if (rc) 1143 return rc; 1144 else 1145 return num_vfs; 1146 } else 1147 return -EOPNOTSUPP; 1148 } 1149 #endif 1150 1151 static int efx_pm_freeze(struct device *dev) 1152 { 1153 struct efx_nic *efx = dev_get_drvdata(dev); 1154 1155 rtnl_lock(); 1156 1157 if (efx->state != STATE_DISABLED) { 1158 efx->state = STATE_UNINIT; 1159 1160 efx_device_detach_sync(efx); 1161 1162 efx_stop_all(efx); 1163 efx_disable_interrupts(efx); 1164 } 1165 1166 rtnl_unlock(); 1167 1168 return 0; 1169 } 1170 1171 static int efx_pm_thaw(struct device *dev) 1172 { 1173 int rc; 1174 struct efx_nic *efx = dev_get_drvdata(dev); 1175 1176 rtnl_lock(); 1177 1178 if (efx->state != STATE_DISABLED) { 1179 rc = efx_enable_interrupts(efx); 1180 if (rc) 1181 goto fail; 1182 1183 mutex_lock(&efx->mac_lock); 1184 efx_mcdi_port_reconfigure(efx); 1185 mutex_unlock(&efx->mac_lock); 1186 1187 efx_start_all(efx); 1188 1189 efx_device_attach_if_not_resetting(efx); 1190 1191 efx->state = STATE_READY; 1192 1193 efx->type->resume_wol(efx); 1194 } 1195 1196 rtnl_unlock(); 1197 1198 /* Reschedule any quenched resets scheduled during efx_pm_freeze() */ 1199 efx_queue_reset_work(efx); 1200 1201 return 0; 1202 1203 fail: 1204 rtnl_unlock(); 1205 1206 return rc; 1207 } 1208 1209 static int efx_pm_poweroff(struct device *dev) 1210 { 1211 struct pci_dev *pci_dev = to_pci_dev(dev); 1212 struct efx_nic *efx = pci_get_drvdata(pci_dev); 1213 1214 efx->type->fini(efx); 1215 1216 efx->reset_pending = 0; 1217 1218 pci_save_state(pci_dev); 1219 return pci_set_power_state(pci_dev, PCI_D3hot); 1220 } 1221 1222 /* Used for both resume and restore */ 1223 static int efx_pm_resume(struct device *dev) 1224 { 1225 struct pci_dev *pci_dev = to_pci_dev(dev); 1226 struct efx_nic *efx = pci_get_drvdata(pci_dev); 1227 int rc; 1228 1229 rc = pci_set_power_state(pci_dev, PCI_D0); 1230 if (rc) 1231 return rc; 1232 pci_restore_state(pci_dev); 1233 rc = pci_enable_device(pci_dev); 1234 if (rc) 1235 return rc; 1236 pci_set_master(efx->pci_dev); 1237 rc = efx->type->reset(efx, RESET_TYPE_ALL); 1238 if (rc) 1239 return rc; 1240 down_write(&efx->filter_sem); 1241 rc = efx->type->init(efx); 1242 up_write(&efx->filter_sem); 1243 if (rc) 1244 return rc; 1245 rc = efx_pm_thaw(dev); 1246 return rc; 1247 } 1248 1249 static int efx_pm_suspend(struct device *dev) 1250 { 1251 int rc; 1252 1253 efx_pm_freeze(dev); 1254 rc = efx_pm_poweroff(dev); 1255 if (rc) 1256 efx_pm_resume(dev); 1257 return rc; 1258 } 1259 1260 static const struct dev_pm_ops efx_pm_ops = { 1261 .suspend = efx_pm_suspend, 1262 .resume = efx_pm_resume, 1263 .freeze = efx_pm_freeze, 1264 .thaw = efx_pm_thaw, 1265 .poweroff = efx_pm_poweroff, 1266 .restore = efx_pm_resume, 1267 }; 1268 1269 static struct pci_driver efx_pci_driver = { 1270 .name = KBUILD_MODNAME, 1271 .id_table = efx_pci_table, 1272 .probe = efx_pci_probe, 1273 .remove = efx_pci_remove, 1274 .driver.pm = &efx_pm_ops, 1275 .err_handler = &efx_err_handlers, 1276 #ifdef CONFIG_SFC_SRIOV 1277 .sriov_configure = efx_pci_sriov_configure, 1278 #endif 1279 }; 1280 1281 /************************************************************************** 1282 * 1283 * Kernel module interface 1284 * 1285 *************************************************************************/ 1286 1287 static int __init efx_init_module(void) 1288 { 1289 int rc; 1290 1291 printk(KERN_INFO "Solarflare NET driver\n"); 1292 1293 rc = register_netdevice_notifier(&efx_netdev_notifier); 1294 if (rc) 1295 goto err_notifier; 1296 1297 #ifdef CONFIG_SFC_SRIOV 1298 rc = efx_init_sriov(); 1299 if (rc) 1300 goto err_sriov; 1301 #endif 1302 1303 rc = efx_create_reset_workqueue(); 1304 if (rc) 1305 goto err_reset; 1306 1307 rc = pci_register_driver(&efx_pci_driver); 1308 if (rc < 0) 1309 goto err_pci; 1310 1311 rc = pci_register_driver(&ef100_pci_driver); 1312 if (rc < 0) 1313 goto err_pci_ef100; 1314 1315 return 0; 1316 1317 err_pci_ef100: 1318 pci_unregister_driver(&efx_pci_driver); 1319 err_pci: 1320 efx_destroy_reset_workqueue(); 1321 err_reset: 1322 #ifdef CONFIG_SFC_SRIOV 1323 efx_fini_sriov(); 1324 err_sriov: 1325 #endif 1326 unregister_netdevice_notifier(&efx_netdev_notifier); 1327 err_notifier: 1328 return rc; 1329 } 1330 1331 static void __exit efx_exit_module(void) 1332 { 1333 printk(KERN_INFO "Solarflare NET driver unloading\n"); 1334 1335 pci_unregister_driver(&ef100_pci_driver); 1336 pci_unregister_driver(&efx_pci_driver); 1337 efx_destroy_reset_workqueue(); 1338 #ifdef CONFIG_SFC_SRIOV 1339 efx_fini_sriov(); 1340 #endif 1341 unregister_netdevice_notifier(&efx_netdev_notifier); 1342 1343 } 1344 1345 module_init(efx_init_module); 1346 module_exit(efx_exit_module); 1347 1348 MODULE_AUTHOR("Solarflare Communications and " 1349 "Michael Brown <mbrown@fensystems.co.uk>"); 1350 MODULE_DESCRIPTION("Solarflare network driver"); 1351 MODULE_LICENSE("GPL"); 1352 MODULE_DEVICE_TABLE(pci, efx_pci_table); 1353