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/module.h> 9 #include <linux/pci.h> 10 #include <linux/netdevice.h> 11 #include <linux/etherdevice.h> 12 #include <linux/delay.h> 13 #include <linux/notifier.h> 14 #include <linux/ip.h> 15 #include <linux/tcp.h> 16 #include <linux/in.h> 17 #include <linux/ethtool.h> 18 #include <linux/topology.h> 19 #include <linux/gfp.h> 20 #include <linux/aer.h> 21 #include <linux/interrupt.h> 22 #include "net_driver.h" 23 #include <net/gre.h> 24 #include <net/udp_tunnel.h> 25 #include "efx.h" 26 #include "efx_common.h" 27 #include "efx_channels.h" 28 #include "ef100.h" 29 #include "rx_common.h" 30 #include "tx_common.h" 31 #include "nic.h" 32 #include "io.h" 33 #include "selftest.h" 34 #include "sriov.h" 35 36 #include "mcdi_port_common.h" 37 #include "mcdi_pcol.h" 38 #include "workarounds.h" 39 40 /************************************************************************** 41 * 42 * Configurable values 43 * 44 *************************************************************************/ 45 46 module_param_named(interrupt_mode, efx_interrupt_mode, uint, 0444); 47 MODULE_PARM_DESC(interrupt_mode, 48 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)"); 49 50 module_param(rss_cpus, uint, 0444); 51 MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling"); 52 53 /* 54 * Use separate channels for TX and RX events 55 * 56 * Set this to 1 to use separate channels for TX and RX. It allows us 57 * to control interrupt affinity separately for TX and RX. 58 * 59 * This is only used in MSI-X interrupt mode 60 */ 61 bool efx_separate_tx_channels; 62 module_param(efx_separate_tx_channels, bool, 0444); 63 MODULE_PARM_DESC(efx_separate_tx_channels, 64 "Use separate channels for TX and RX"); 65 66 /* Initial interrupt moderation settings. They can be modified after 67 * module load with ethtool. 68 * 69 * The default for RX should strike a balance between increasing the 70 * round-trip latency and reducing overhead. 71 */ 72 static unsigned int rx_irq_mod_usec = 60; 73 74 /* Initial interrupt moderation settings. They can be modified after 75 * module load with ethtool. 76 * 77 * This default is chosen to ensure that a 10G link does not go idle 78 * while a TX queue is stopped after it has become full. A queue is 79 * restarted when it drops below half full. The time this takes (assuming 80 * worst case 3 descriptors per packet and 1024 descriptors) is 81 * 512 / 3 * 1.2 = 205 usec. 82 */ 83 static unsigned int tx_irq_mod_usec = 150; 84 85 static bool phy_flash_cfg; 86 module_param(phy_flash_cfg, bool, 0644); 87 MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially"); 88 89 static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE | 90 NETIF_MSG_LINK | NETIF_MSG_IFDOWN | 91 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR | 92 NETIF_MSG_TX_ERR | NETIF_MSG_HW); 93 module_param(debug, uint, 0); 94 MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value"); 95 96 /************************************************************************** 97 * 98 * Utility functions and prototypes 99 * 100 *************************************************************************/ 101 102 static void efx_remove_port(struct efx_nic *efx); 103 static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog); 104 static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp); 105 static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs, 106 u32 flags); 107 108 #define EFX_ASSERT_RESET_SERIALISED(efx) \ 109 do { \ 110 if ((efx->state == STATE_READY) || \ 111 (efx->state == STATE_RECOVERY) || \ 112 (efx->state == STATE_DISABLED)) \ 113 ASSERT_RTNL(); \ 114 } while (0) 115 116 /************************************************************************** 117 * 118 * Port handling 119 * 120 **************************************************************************/ 121 122 static void efx_fini_port(struct efx_nic *efx); 123 124 static int efx_probe_port(struct efx_nic *efx) 125 { 126 int rc; 127 128 netif_dbg(efx, probe, efx->net_dev, "create port\n"); 129 130 if (phy_flash_cfg) 131 efx->phy_mode = PHY_MODE_SPECIAL; 132 133 /* Connect up MAC/PHY operations table */ 134 rc = efx->type->probe_port(efx); 135 if (rc) 136 return rc; 137 138 /* Initialise MAC address to permanent address */ 139 eth_hw_addr_set(efx->net_dev, efx->net_dev->perm_addr); 140 141 return 0; 142 } 143 144 static int efx_init_port(struct efx_nic *efx) 145 { 146 int rc; 147 148 netif_dbg(efx, drv, efx->net_dev, "init port\n"); 149 150 mutex_lock(&efx->mac_lock); 151 152 efx->port_initialized = true; 153 154 /* Ensure the PHY advertises the correct flow control settings */ 155 rc = efx_mcdi_port_reconfigure(efx); 156 if (rc && rc != -EPERM) 157 goto fail; 158 159 mutex_unlock(&efx->mac_lock); 160 return 0; 161 162 fail: 163 mutex_unlock(&efx->mac_lock); 164 return rc; 165 } 166 167 static void efx_fini_port(struct efx_nic *efx) 168 { 169 netif_dbg(efx, drv, efx->net_dev, "shut down port\n"); 170 171 if (!efx->port_initialized) 172 return; 173 174 efx->port_initialized = false; 175 176 efx->link_state.up = false; 177 efx_link_status_changed(efx); 178 } 179 180 static void efx_remove_port(struct efx_nic *efx) 181 { 182 netif_dbg(efx, drv, efx->net_dev, "destroying port\n"); 183 184 efx->type->remove_port(efx); 185 } 186 187 /************************************************************************** 188 * 189 * NIC handling 190 * 191 **************************************************************************/ 192 193 static LIST_HEAD(efx_primary_list); 194 static LIST_HEAD(efx_unassociated_list); 195 196 static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right) 197 { 198 return left->type == right->type && 199 left->vpd_sn && right->vpd_sn && 200 !strcmp(left->vpd_sn, right->vpd_sn); 201 } 202 203 static void efx_associate(struct efx_nic *efx) 204 { 205 struct efx_nic *other, *next; 206 207 if (efx->primary == efx) { 208 /* Adding primary function; look for secondaries */ 209 210 netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n"); 211 list_add_tail(&efx->node, &efx_primary_list); 212 213 list_for_each_entry_safe(other, next, &efx_unassociated_list, 214 node) { 215 if (efx_same_controller(efx, other)) { 216 list_del(&other->node); 217 netif_dbg(other, probe, other->net_dev, 218 "moving to secondary list of %s %s\n", 219 pci_name(efx->pci_dev), 220 efx->net_dev->name); 221 list_add_tail(&other->node, 222 &efx->secondary_list); 223 other->primary = efx; 224 } 225 } 226 } else { 227 /* Adding secondary function; look for primary */ 228 229 list_for_each_entry(other, &efx_primary_list, node) { 230 if (efx_same_controller(efx, other)) { 231 netif_dbg(efx, probe, efx->net_dev, 232 "adding to secondary list of %s %s\n", 233 pci_name(other->pci_dev), 234 other->net_dev->name); 235 list_add_tail(&efx->node, 236 &other->secondary_list); 237 efx->primary = other; 238 return; 239 } 240 } 241 242 netif_dbg(efx, probe, efx->net_dev, 243 "adding to unassociated list\n"); 244 list_add_tail(&efx->node, &efx_unassociated_list); 245 } 246 } 247 248 static void efx_dissociate(struct efx_nic *efx) 249 { 250 struct efx_nic *other, *next; 251 252 list_del(&efx->node); 253 efx->primary = NULL; 254 255 list_for_each_entry_safe(other, next, &efx->secondary_list, node) { 256 list_del(&other->node); 257 netif_dbg(other, probe, other->net_dev, 258 "moving to unassociated list\n"); 259 list_add_tail(&other->node, &efx_unassociated_list); 260 other->primary = NULL; 261 } 262 } 263 264 static int efx_probe_nic(struct efx_nic *efx) 265 { 266 int rc; 267 268 netif_dbg(efx, probe, efx->net_dev, "creating NIC\n"); 269 270 /* Carry out hardware-type specific initialisation */ 271 rc = efx->type->probe(efx); 272 if (rc) 273 return rc; 274 275 do { 276 if (!efx->max_channels || !efx->max_tx_channels) { 277 netif_err(efx, drv, efx->net_dev, 278 "Insufficient resources to allocate" 279 " any channels\n"); 280 rc = -ENOSPC; 281 goto fail1; 282 } 283 284 /* Determine the number of channels and queues by trying 285 * to hook in MSI-X interrupts. 286 */ 287 rc = efx_probe_interrupts(efx); 288 if (rc) 289 goto fail1; 290 291 rc = efx_set_channels(efx); 292 if (rc) 293 goto fail1; 294 295 /* dimension_resources can fail with EAGAIN */ 296 rc = efx->type->dimension_resources(efx); 297 if (rc != 0 && rc != -EAGAIN) 298 goto fail2; 299 300 if (rc == -EAGAIN) 301 /* try again with new max_channels */ 302 efx_remove_interrupts(efx); 303 304 } while (rc == -EAGAIN); 305 306 if (efx->n_channels > 1) 307 netdev_rss_key_fill(efx->rss_context.rx_hash_key, 308 sizeof(efx->rss_context.rx_hash_key)); 309 efx_set_default_rx_indir_table(efx, &efx->rss_context); 310 311 /* Initialise the interrupt moderation settings */ 312 efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000); 313 efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true, 314 true); 315 316 return 0; 317 318 fail2: 319 efx_remove_interrupts(efx); 320 fail1: 321 efx->type->remove(efx); 322 return rc; 323 } 324 325 static void efx_remove_nic(struct efx_nic *efx) 326 { 327 netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n"); 328 329 efx_remove_interrupts(efx); 330 efx->type->remove(efx); 331 } 332 333 /************************************************************************** 334 * 335 * NIC startup/shutdown 336 * 337 *************************************************************************/ 338 339 static int efx_probe_all(struct efx_nic *efx) 340 { 341 int rc; 342 343 rc = efx_probe_nic(efx); 344 if (rc) { 345 netif_err(efx, probe, efx->net_dev, "failed to create NIC\n"); 346 goto fail1; 347 } 348 349 rc = efx_probe_port(efx); 350 if (rc) { 351 netif_err(efx, probe, efx->net_dev, "failed to create port\n"); 352 goto fail2; 353 } 354 355 BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT); 356 if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) { 357 rc = -EINVAL; 358 goto fail3; 359 } 360 361 #ifdef CONFIG_SFC_SRIOV 362 rc = efx->type->vswitching_probe(efx); 363 if (rc) /* not fatal; the PF will still work fine */ 364 netif_warn(efx, probe, efx->net_dev, 365 "failed to setup vswitching rc=%d;" 366 " VFs may not function\n", rc); 367 #endif 368 369 rc = efx_probe_filters(efx); 370 if (rc) { 371 netif_err(efx, probe, efx->net_dev, 372 "failed to create filter tables\n"); 373 goto fail4; 374 } 375 376 rc = efx_probe_channels(efx); 377 if (rc) 378 goto fail5; 379 380 return 0; 381 382 fail5: 383 efx_remove_filters(efx); 384 fail4: 385 #ifdef CONFIG_SFC_SRIOV 386 efx->type->vswitching_remove(efx); 387 #endif 388 fail3: 389 efx_remove_port(efx); 390 fail2: 391 efx_remove_nic(efx); 392 fail1: 393 return rc; 394 } 395 396 static void efx_remove_all(struct efx_nic *efx) 397 { 398 rtnl_lock(); 399 efx_xdp_setup_prog(efx, NULL); 400 rtnl_unlock(); 401 402 efx_remove_channels(efx); 403 efx_remove_filters(efx); 404 #ifdef CONFIG_SFC_SRIOV 405 efx->type->vswitching_remove(efx); 406 #endif 407 efx_remove_port(efx); 408 efx_remove_nic(efx); 409 } 410 411 /************************************************************************** 412 * 413 * Interrupt moderation 414 * 415 **************************************************************************/ 416 unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs) 417 { 418 if (usecs == 0) 419 return 0; 420 if (usecs * 1000 < efx->timer_quantum_ns) 421 return 1; /* never round down to 0 */ 422 return usecs * 1000 / efx->timer_quantum_ns; 423 } 424 425 unsigned int efx_ticks_to_usecs(struct efx_nic *efx, unsigned int ticks) 426 { 427 /* We must round up when converting ticks to microseconds 428 * because we round down when converting the other way. 429 */ 430 return DIV_ROUND_UP(ticks * efx->timer_quantum_ns, 1000); 431 } 432 433 /* Set interrupt moderation parameters */ 434 int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs, 435 unsigned int rx_usecs, bool rx_adaptive, 436 bool rx_may_override_tx) 437 { 438 struct efx_channel *channel; 439 unsigned int timer_max_us; 440 441 EFX_ASSERT_RESET_SERIALISED(efx); 442 443 timer_max_us = efx->timer_max_ns / 1000; 444 445 if (tx_usecs > timer_max_us || rx_usecs > timer_max_us) 446 return -EINVAL; 447 448 if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 && 449 !rx_may_override_tx) { 450 netif_err(efx, drv, efx->net_dev, "Channels are shared. " 451 "RX and TX IRQ moderation must be equal\n"); 452 return -EINVAL; 453 } 454 455 efx->irq_rx_adaptive = rx_adaptive; 456 efx->irq_rx_moderation_us = rx_usecs; 457 efx_for_each_channel(channel, efx) { 458 if (efx_channel_has_rx_queue(channel)) 459 channel->irq_moderation_us = rx_usecs; 460 else if (efx_channel_has_tx_queues(channel)) 461 channel->irq_moderation_us = tx_usecs; 462 else if (efx_channel_is_xdp_tx(channel)) 463 channel->irq_moderation_us = tx_usecs; 464 } 465 466 return 0; 467 } 468 469 void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs, 470 unsigned int *rx_usecs, bool *rx_adaptive) 471 { 472 *rx_adaptive = efx->irq_rx_adaptive; 473 *rx_usecs = efx->irq_rx_moderation_us; 474 475 /* If channels are shared between RX and TX, so is IRQ 476 * moderation. Otherwise, IRQ moderation is the same for all 477 * TX channels and is not adaptive. 478 */ 479 if (efx->tx_channel_offset == 0) { 480 *tx_usecs = *rx_usecs; 481 } else { 482 struct efx_channel *tx_channel; 483 484 tx_channel = efx->channel[efx->tx_channel_offset]; 485 *tx_usecs = tx_channel->irq_moderation_us; 486 } 487 } 488 489 /************************************************************************** 490 * 491 * ioctls 492 * 493 *************************************************************************/ 494 495 /* Net device ioctl 496 * Context: process, rtnl_lock() held. 497 */ 498 static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd) 499 { 500 struct efx_nic *efx = netdev_priv(net_dev); 501 struct mii_ioctl_data *data = if_mii(ifr); 502 503 if (cmd == SIOCSHWTSTAMP) 504 return efx_ptp_set_ts_config(efx, ifr); 505 if (cmd == SIOCGHWTSTAMP) 506 return efx_ptp_get_ts_config(efx, ifr); 507 508 /* Convert phy_id from older PRTAD/DEVAD format */ 509 if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) && 510 (data->phy_id & 0xfc00) == 0x0400) 511 data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400; 512 513 return mdio_mii_ioctl(&efx->mdio, data, cmd); 514 } 515 516 /************************************************************************** 517 * 518 * Kernel net device interface 519 * 520 *************************************************************************/ 521 522 /* Context: process, rtnl_lock() held. */ 523 int efx_net_open(struct net_device *net_dev) 524 { 525 struct efx_nic *efx = netdev_priv(net_dev); 526 int rc; 527 528 netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n", 529 raw_smp_processor_id()); 530 531 rc = efx_check_disabled(efx); 532 if (rc) 533 return rc; 534 if (efx->phy_mode & PHY_MODE_SPECIAL) 535 return -EBUSY; 536 if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL)) 537 return -EIO; 538 539 /* Notify the kernel of the link state polled during driver load, 540 * before the monitor starts running */ 541 efx_link_status_changed(efx); 542 543 efx_start_all(efx); 544 if (efx->state == STATE_DISABLED || efx->reset_pending) 545 netif_device_detach(efx->net_dev); 546 efx_selftest_async_start(efx); 547 return 0; 548 } 549 550 /* Context: process, rtnl_lock() held. 551 * Note that the kernel will ignore our return code; this method 552 * should really be a void. 553 */ 554 int efx_net_stop(struct net_device *net_dev) 555 { 556 struct efx_nic *efx = netdev_priv(net_dev); 557 558 netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n", 559 raw_smp_processor_id()); 560 561 /* Stop the device and flush all the channels */ 562 efx_stop_all(efx); 563 564 return 0; 565 } 566 567 static int efx_vlan_rx_add_vid(struct net_device *net_dev, __be16 proto, u16 vid) 568 { 569 struct efx_nic *efx = netdev_priv(net_dev); 570 571 if (efx->type->vlan_rx_add_vid) 572 return efx->type->vlan_rx_add_vid(efx, proto, vid); 573 else 574 return -EOPNOTSUPP; 575 } 576 577 static int efx_vlan_rx_kill_vid(struct net_device *net_dev, __be16 proto, u16 vid) 578 { 579 struct efx_nic *efx = netdev_priv(net_dev); 580 581 if (efx->type->vlan_rx_kill_vid) 582 return efx->type->vlan_rx_kill_vid(efx, proto, vid); 583 else 584 return -EOPNOTSUPP; 585 } 586 587 static const struct net_device_ops efx_netdev_ops = { 588 .ndo_open = efx_net_open, 589 .ndo_stop = efx_net_stop, 590 .ndo_get_stats64 = efx_net_stats, 591 .ndo_tx_timeout = efx_watchdog, 592 .ndo_start_xmit = efx_hard_start_xmit, 593 .ndo_validate_addr = eth_validate_addr, 594 .ndo_eth_ioctl = efx_ioctl, 595 .ndo_change_mtu = efx_change_mtu, 596 .ndo_set_mac_address = efx_set_mac_address, 597 .ndo_set_rx_mode = efx_set_rx_mode, 598 .ndo_set_features = efx_set_features, 599 .ndo_features_check = efx_features_check, 600 .ndo_vlan_rx_add_vid = efx_vlan_rx_add_vid, 601 .ndo_vlan_rx_kill_vid = efx_vlan_rx_kill_vid, 602 #ifdef CONFIG_SFC_SRIOV 603 .ndo_set_vf_mac = efx_sriov_set_vf_mac, 604 .ndo_set_vf_vlan = efx_sriov_set_vf_vlan, 605 .ndo_set_vf_spoofchk = efx_sriov_set_vf_spoofchk, 606 .ndo_get_vf_config = efx_sriov_get_vf_config, 607 .ndo_set_vf_link_state = efx_sriov_set_vf_link_state, 608 #endif 609 .ndo_get_phys_port_id = efx_get_phys_port_id, 610 .ndo_get_phys_port_name = efx_get_phys_port_name, 611 .ndo_setup_tc = efx_setup_tc, 612 #ifdef CONFIG_RFS_ACCEL 613 .ndo_rx_flow_steer = efx_filter_rfs, 614 #endif 615 .ndo_xdp_xmit = efx_xdp_xmit, 616 .ndo_bpf = efx_xdp 617 }; 618 619 static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog) 620 { 621 struct bpf_prog *old_prog; 622 623 if (efx->xdp_rxq_info_failed) { 624 netif_err(efx, drv, efx->net_dev, 625 "Unable to bind XDP program due to previous failure of rxq_info\n"); 626 return -EINVAL; 627 } 628 629 if (prog && efx->net_dev->mtu > efx_xdp_max_mtu(efx)) { 630 netif_err(efx, drv, efx->net_dev, 631 "Unable to configure XDP with MTU of %d (max: %d)\n", 632 efx->net_dev->mtu, efx_xdp_max_mtu(efx)); 633 return -EINVAL; 634 } 635 636 old_prog = rtnl_dereference(efx->xdp_prog); 637 rcu_assign_pointer(efx->xdp_prog, prog); 638 /* Release the reference that was originally passed by the caller. */ 639 if (old_prog) 640 bpf_prog_put(old_prog); 641 642 return 0; 643 } 644 645 /* Context: process, rtnl_lock() held. */ 646 static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp) 647 { 648 struct efx_nic *efx = netdev_priv(dev); 649 650 switch (xdp->command) { 651 case XDP_SETUP_PROG: 652 return efx_xdp_setup_prog(efx, xdp->prog); 653 default: 654 return -EINVAL; 655 } 656 } 657 658 static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs, 659 u32 flags) 660 { 661 struct efx_nic *efx = netdev_priv(dev); 662 663 if (!netif_running(dev)) 664 return -EINVAL; 665 666 return efx_xdp_tx_buffers(efx, n, xdpfs, flags & XDP_XMIT_FLUSH); 667 } 668 669 static void efx_update_name(struct efx_nic *efx) 670 { 671 strcpy(efx->name, efx->net_dev->name); 672 efx_mtd_rename(efx); 673 efx_set_channel_names(efx); 674 } 675 676 static int efx_netdev_event(struct notifier_block *this, 677 unsigned long event, void *ptr) 678 { 679 struct net_device *net_dev = netdev_notifier_info_to_dev(ptr); 680 681 if ((net_dev->netdev_ops == &efx_netdev_ops) && 682 event == NETDEV_CHANGENAME) 683 efx_update_name(netdev_priv(net_dev)); 684 685 return NOTIFY_DONE; 686 } 687 688 static struct notifier_block efx_netdev_notifier = { 689 .notifier_call = efx_netdev_event, 690 }; 691 692 static ssize_t phy_type_show(struct device *dev, 693 struct device_attribute *attr, char *buf) 694 { 695 struct efx_nic *efx = dev_get_drvdata(dev); 696 return sprintf(buf, "%d\n", efx->phy_type); 697 } 698 static DEVICE_ATTR_RO(phy_type); 699 700 static int efx_register_netdev(struct efx_nic *efx) 701 { 702 struct net_device *net_dev = efx->net_dev; 703 struct efx_channel *channel; 704 int rc; 705 706 net_dev->watchdog_timeo = 5 * HZ; 707 net_dev->irq = efx->pci_dev->irq; 708 net_dev->netdev_ops = &efx_netdev_ops; 709 if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0) 710 net_dev->priv_flags |= IFF_UNICAST_FLT; 711 net_dev->ethtool_ops = &efx_ethtool_ops; 712 net_dev->gso_max_segs = EFX_TSO_MAX_SEGS; 713 net_dev->min_mtu = EFX_MIN_MTU; 714 net_dev->max_mtu = EFX_MAX_MTU; 715 716 rtnl_lock(); 717 718 /* Enable resets to be scheduled and check whether any were 719 * already requested. If so, the NIC is probably hosed so we 720 * abort. 721 */ 722 efx->state = STATE_READY; 723 smp_mb(); /* ensure we change state before checking reset_pending */ 724 if (efx->reset_pending) { 725 pci_err(efx->pci_dev, "aborting probe due to scheduled reset\n"); 726 rc = -EIO; 727 goto fail_locked; 728 } 729 730 rc = dev_alloc_name(net_dev, net_dev->name); 731 if (rc < 0) 732 goto fail_locked; 733 efx_update_name(efx); 734 735 /* Always start with carrier off; PHY events will detect the link */ 736 netif_carrier_off(net_dev); 737 738 rc = register_netdevice(net_dev); 739 if (rc) 740 goto fail_locked; 741 742 efx_for_each_channel(channel, efx) { 743 struct efx_tx_queue *tx_queue; 744 efx_for_each_channel_tx_queue(tx_queue, channel) 745 efx_init_tx_queue_core_txq(tx_queue); 746 } 747 748 efx_associate(efx); 749 750 rtnl_unlock(); 751 752 rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type); 753 if (rc) { 754 netif_err(efx, drv, efx->net_dev, 755 "failed to init net dev attributes\n"); 756 goto fail_registered; 757 } 758 759 efx_init_mcdi_logging(efx); 760 761 return 0; 762 763 fail_registered: 764 rtnl_lock(); 765 efx_dissociate(efx); 766 unregister_netdevice(net_dev); 767 fail_locked: 768 efx->state = STATE_UNINIT; 769 rtnl_unlock(); 770 netif_err(efx, drv, efx->net_dev, "could not register net dev\n"); 771 return rc; 772 } 773 774 static void efx_unregister_netdev(struct efx_nic *efx) 775 { 776 if (!efx->net_dev) 777 return; 778 779 BUG_ON(netdev_priv(efx->net_dev) != efx); 780 781 if (efx_dev_registered(efx)) { 782 strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name)); 783 efx_fini_mcdi_logging(efx); 784 device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type); 785 unregister_netdev(efx->net_dev); 786 } 787 } 788 789 /************************************************************************** 790 * 791 * List of NICs we support 792 * 793 **************************************************************************/ 794 795 /* PCI device ID table */ 796 static const struct pci_device_id efx_pci_table[] = { 797 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803), /* SFC9020 */ 798 .driver_data = (unsigned long) &siena_a0_nic_type}, 799 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813), /* SFL9021 */ 800 .driver_data = (unsigned long) &siena_a0_nic_type}, 801 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903), /* SFC9120 PF */ 802 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 803 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903), /* SFC9120 VF */ 804 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 805 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923), /* SFC9140 PF */ 806 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 807 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923), /* SFC9140 VF */ 808 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 809 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03), /* SFC9220 PF */ 810 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 811 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03), /* SFC9220 VF */ 812 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 813 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0b03), /* SFC9250 PF */ 814 .driver_data = (unsigned long) &efx_hunt_a0_nic_type}, 815 {PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1b03), /* SFC9250 VF */ 816 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type}, 817 {0} /* end of list */ 818 }; 819 820 /************************************************************************** 821 * 822 * Data housekeeping 823 * 824 **************************************************************************/ 825 826 void efx_update_sw_stats(struct efx_nic *efx, u64 *stats) 827 { 828 u64 n_rx_nodesc_trunc = 0; 829 struct efx_channel *channel; 830 831 efx_for_each_channel(channel, efx) 832 n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc; 833 stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc; 834 stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops); 835 } 836 837 /************************************************************************** 838 * 839 * PCI interface 840 * 841 **************************************************************************/ 842 843 /* Main body of final NIC shutdown code 844 * This is called only at module unload (or hotplug removal). 845 */ 846 static void efx_pci_remove_main(struct efx_nic *efx) 847 { 848 /* Flush reset_work. It can no longer be scheduled since we 849 * are not READY. 850 */ 851 BUG_ON(efx->state == STATE_READY); 852 efx_flush_reset_workqueue(efx); 853 854 efx_disable_interrupts(efx); 855 efx_clear_interrupt_affinity(efx); 856 efx_nic_fini_interrupt(efx); 857 efx_fini_port(efx); 858 efx->type->fini(efx); 859 efx_fini_napi(efx); 860 efx_remove_all(efx); 861 } 862 863 /* Final NIC shutdown 864 * This is called only at module unload (or hotplug removal). A PF can call 865 * this on its VFs to ensure they are unbound first. 866 */ 867 static void efx_pci_remove(struct pci_dev *pci_dev) 868 { 869 struct efx_nic *efx; 870 871 efx = pci_get_drvdata(pci_dev); 872 if (!efx) 873 return; 874 875 /* Mark the NIC as fini, then stop the interface */ 876 rtnl_lock(); 877 efx_dissociate(efx); 878 dev_close(efx->net_dev); 879 efx_disable_interrupts(efx); 880 efx->state = STATE_UNINIT; 881 rtnl_unlock(); 882 883 if (efx->type->sriov_fini) 884 efx->type->sriov_fini(efx); 885 886 efx_unregister_netdev(efx); 887 888 efx_mtd_remove(efx); 889 890 efx_pci_remove_main(efx); 891 892 efx_fini_io(efx); 893 netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n"); 894 895 efx_fini_struct(efx); 896 free_netdev(efx->net_dev); 897 898 pci_disable_pcie_error_reporting(pci_dev); 899 }; 900 901 /* NIC VPD information 902 * Called during probe to display the part number of the 903 * installed NIC. 904 */ 905 static void efx_probe_vpd_strings(struct efx_nic *efx) 906 { 907 struct pci_dev *dev = efx->pci_dev; 908 unsigned int vpd_size, kw_len; 909 u8 *vpd_data; 910 int start; 911 912 vpd_data = pci_vpd_alloc(dev, &vpd_size); 913 if (IS_ERR(vpd_data)) { 914 pci_warn(dev, "Unable to read VPD\n"); 915 return; 916 } 917 918 start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size, 919 PCI_VPD_RO_KEYWORD_PARTNO, &kw_len); 920 if (start < 0) 921 pci_err(dev, "Part number not found or incomplete\n"); 922 else 923 pci_info(dev, "Part Number : %.*s\n", kw_len, vpd_data + start); 924 925 start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size, 926 PCI_VPD_RO_KEYWORD_SERIALNO, &kw_len); 927 if (start < 0) 928 pci_err(dev, "Serial number not found or incomplete\n"); 929 else 930 efx->vpd_sn = kmemdup_nul(vpd_data + start, kw_len, GFP_KERNEL); 931 932 kfree(vpd_data); 933 } 934 935 936 /* Main body of NIC initialisation 937 * This is called at module load (or hotplug insertion, theoretically). 938 */ 939 static int efx_pci_probe_main(struct efx_nic *efx) 940 { 941 int rc; 942 943 /* Do start-of-day initialisation */ 944 rc = efx_probe_all(efx); 945 if (rc) 946 goto fail1; 947 948 efx_init_napi(efx); 949 950 down_write(&efx->filter_sem); 951 rc = efx->type->init(efx); 952 up_write(&efx->filter_sem); 953 if (rc) { 954 pci_err(efx->pci_dev, "failed to initialise NIC\n"); 955 goto fail3; 956 } 957 958 rc = efx_init_port(efx); 959 if (rc) { 960 netif_err(efx, probe, efx->net_dev, 961 "failed to initialise port\n"); 962 goto fail4; 963 } 964 965 rc = efx_nic_init_interrupt(efx); 966 if (rc) 967 goto fail5; 968 969 efx_set_interrupt_affinity(efx); 970 rc = efx_enable_interrupts(efx); 971 if (rc) 972 goto fail6; 973 974 return 0; 975 976 fail6: 977 efx_clear_interrupt_affinity(efx); 978 efx_nic_fini_interrupt(efx); 979 fail5: 980 efx_fini_port(efx); 981 fail4: 982 efx->type->fini(efx); 983 fail3: 984 efx_fini_napi(efx); 985 efx_remove_all(efx); 986 fail1: 987 return rc; 988 } 989 990 static int efx_pci_probe_post_io(struct efx_nic *efx) 991 { 992 struct net_device *net_dev = efx->net_dev; 993 int rc = efx_pci_probe_main(efx); 994 995 if (rc) 996 return rc; 997 998 if (efx->type->sriov_init) { 999 rc = efx->type->sriov_init(efx); 1000 if (rc) 1001 pci_err(efx->pci_dev, "SR-IOV can't be enabled rc %d\n", 1002 rc); 1003 } 1004 1005 /* Determine netdevice features */ 1006 net_dev->features |= (efx->type->offload_features | NETIF_F_SG | 1007 NETIF_F_TSO | NETIF_F_RXCSUM | NETIF_F_RXALL); 1008 if (efx->type->offload_features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM)) 1009 net_dev->features |= NETIF_F_TSO6; 1010 /* Check whether device supports TSO */ 1011 if (!efx->type->tso_versions || !efx->type->tso_versions(efx)) 1012 net_dev->features &= ~NETIF_F_ALL_TSO; 1013 /* Mask for features that also apply to VLAN devices */ 1014 net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG | 1015 NETIF_F_HIGHDMA | NETIF_F_ALL_TSO | 1016 NETIF_F_RXCSUM); 1017 1018 net_dev->hw_features |= net_dev->features & ~efx->fixed_features; 1019 1020 /* Disable receiving frames with bad FCS, by default. */ 1021 net_dev->features &= ~NETIF_F_RXALL; 1022 1023 /* Disable VLAN filtering by default. It may be enforced if 1024 * the feature is fixed (i.e. VLAN filters are required to 1025 * receive VLAN tagged packets due to vPort restrictions). 1026 */ 1027 net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER; 1028 net_dev->features |= efx->fixed_features; 1029 1030 rc = efx_register_netdev(efx); 1031 if (!rc) 1032 return 0; 1033 1034 efx_pci_remove_main(efx); 1035 return rc; 1036 } 1037 1038 /* NIC initialisation 1039 * 1040 * This is called at module load (or hotplug insertion, 1041 * theoretically). It sets up PCI mappings, resets the NIC, 1042 * sets up and registers the network devices with the kernel and hooks 1043 * the interrupt service routine. It does not prepare the device for 1044 * transmission; this is left to the first time one of the network 1045 * interfaces is brought up (i.e. efx_net_open). 1046 */ 1047 static int efx_pci_probe(struct pci_dev *pci_dev, 1048 const struct pci_device_id *entry) 1049 { 1050 struct net_device *net_dev; 1051 struct efx_nic *efx; 1052 int rc; 1053 1054 /* Allocate and initialise a struct net_device and struct efx_nic */ 1055 net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES, 1056 EFX_MAX_RX_QUEUES); 1057 if (!net_dev) 1058 return -ENOMEM; 1059 efx = netdev_priv(net_dev); 1060 efx->type = (const struct efx_nic_type *) entry->driver_data; 1061 efx->fixed_features |= NETIF_F_HIGHDMA; 1062 1063 pci_set_drvdata(pci_dev, efx); 1064 SET_NETDEV_DEV(net_dev, &pci_dev->dev); 1065 rc = efx_init_struct(efx, pci_dev, net_dev); 1066 if (rc) 1067 goto fail1; 1068 1069 pci_info(pci_dev, "Solarflare NIC detected\n"); 1070 1071 if (!efx->type->is_vf) 1072 efx_probe_vpd_strings(efx); 1073 1074 /* Set up basic I/O (BAR mappings etc) */ 1075 rc = efx_init_io(efx, efx->type->mem_bar(efx), efx->type->max_dma_mask, 1076 efx->type->mem_map_size(efx)); 1077 if (rc) 1078 goto fail2; 1079 1080 rc = efx_pci_probe_post_io(efx); 1081 if (rc) { 1082 /* On failure, retry once immediately. 1083 * If we aborted probe due to a scheduled reset, dismiss it. 1084 */ 1085 efx->reset_pending = 0; 1086 rc = efx_pci_probe_post_io(efx); 1087 if (rc) { 1088 /* On another failure, retry once more 1089 * after a 50-305ms delay. 1090 */ 1091 unsigned char r; 1092 1093 get_random_bytes(&r, 1); 1094 msleep((unsigned int)r + 50); 1095 efx->reset_pending = 0; 1096 rc = efx_pci_probe_post_io(efx); 1097 } 1098 } 1099 if (rc) 1100 goto fail3; 1101 1102 netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n"); 1103 1104 /* Try to create MTDs, but allow this to fail */ 1105 rtnl_lock(); 1106 rc = efx_mtd_probe(efx); 1107 rtnl_unlock(); 1108 if (rc && rc != -EPERM) 1109 netif_warn(efx, probe, efx->net_dev, 1110 "failed to create MTDs (%d)\n", rc); 1111 1112 (void)pci_enable_pcie_error_reporting(pci_dev); 1113 1114 if (efx->type->udp_tnl_push_ports) 1115 efx->type->udp_tnl_push_ports(efx); 1116 1117 return 0; 1118 1119 fail3: 1120 efx_fini_io(efx); 1121 fail2: 1122 efx_fini_struct(efx); 1123 fail1: 1124 WARN_ON(rc > 0); 1125 netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc); 1126 free_netdev(net_dev); 1127 return rc; 1128 } 1129 1130 /* efx_pci_sriov_configure returns the actual number of Virtual Functions 1131 * enabled on success 1132 */ 1133 #ifdef CONFIG_SFC_SRIOV 1134 static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs) 1135 { 1136 int rc; 1137 struct efx_nic *efx = pci_get_drvdata(dev); 1138 1139 if (efx->type->sriov_configure) { 1140 rc = efx->type->sriov_configure(efx, num_vfs); 1141 if (rc) 1142 return rc; 1143 else 1144 return num_vfs; 1145 } else 1146 return -EOPNOTSUPP; 1147 } 1148 #endif 1149 1150 static int efx_pm_freeze(struct device *dev) 1151 { 1152 struct efx_nic *efx = dev_get_drvdata(dev); 1153 1154 rtnl_lock(); 1155 1156 if (efx->state != STATE_DISABLED) { 1157 efx->state = STATE_UNINIT; 1158 1159 efx_device_detach_sync(efx); 1160 1161 efx_stop_all(efx); 1162 efx_disable_interrupts(efx); 1163 } 1164 1165 rtnl_unlock(); 1166 1167 return 0; 1168 } 1169 1170 static int efx_pm_thaw(struct device *dev) 1171 { 1172 int rc; 1173 struct efx_nic *efx = dev_get_drvdata(dev); 1174 1175 rtnl_lock(); 1176 1177 if (efx->state != STATE_DISABLED) { 1178 rc = efx_enable_interrupts(efx); 1179 if (rc) 1180 goto fail; 1181 1182 mutex_lock(&efx->mac_lock); 1183 efx_mcdi_port_reconfigure(efx); 1184 mutex_unlock(&efx->mac_lock); 1185 1186 efx_start_all(efx); 1187 1188 efx_device_attach_if_not_resetting(efx); 1189 1190 efx->state = STATE_READY; 1191 1192 efx->type->resume_wol(efx); 1193 } 1194 1195 rtnl_unlock(); 1196 1197 /* Reschedule any quenched resets scheduled during efx_pm_freeze() */ 1198 efx_queue_reset_work(efx); 1199 1200 return 0; 1201 1202 fail: 1203 rtnl_unlock(); 1204 1205 return rc; 1206 } 1207 1208 static int efx_pm_poweroff(struct device *dev) 1209 { 1210 struct pci_dev *pci_dev = to_pci_dev(dev); 1211 struct efx_nic *efx = pci_get_drvdata(pci_dev); 1212 1213 efx->type->fini(efx); 1214 1215 efx->reset_pending = 0; 1216 1217 pci_save_state(pci_dev); 1218 return pci_set_power_state(pci_dev, PCI_D3hot); 1219 } 1220 1221 /* Used for both resume and restore */ 1222 static int efx_pm_resume(struct device *dev) 1223 { 1224 struct pci_dev *pci_dev = to_pci_dev(dev); 1225 struct efx_nic *efx = pci_get_drvdata(pci_dev); 1226 int rc; 1227 1228 rc = pci_set_power_state(pci_dev, PCI_D0); 1229 if (rc) 1230 return rc; 1231 pci_restore_state(pci_dev); 1232 rc = pci_enable_device(pci_dev); 1233 if (rc) 1234 return rc; 1235 pci_set_master(efx->pci_dev); 1236 rc = efx->type->reset(efx, RESET_TYPE_ALL); 1237 if (rc) 1238 return rc; 1239 down_write(&efx->filter_sem); 1240 rc = efx->type->init(efx); 1241 up_write(&efx->filter_sem); 1242 if (rc) 1243 return rc; 1244 rc = efx_pm_thaw(dev); 1245 return rc; 1246 } 1247 1248 static int efx_pm_suspend(struct device *dev) 1249 { 1250 int rc; 1251 1252 efx_pm_freeze(dev); 1253 rc = efx_pm_poweroff(dev); 1254 if (rc) 1255 efx_pm_resume(dev); 1256 return rc; 1257 } 1258 1259 static const struct dev_pm_ops efx_pm_ops = { 1260 .suspend = efx_pm_suspend, 1261 .resume = efx_pm_resume, 1262 .freeze = efx_pm_freeze, 1263 .thaw = efx_pm_thaw, 1264 .poweroff = efx_pm_poweroff, 1265 .restore = efx_pm_resume, 1266 }; 1267 1268 static struct pci_driver efx_pci_driver = { 1269 .name = KBUILD_MODNAME, 1270 .id_table = efx_pci_table, 1271 .probe = efx_pci_probe, 1272 .remove = efx_pci_remove, 1273 .driver.pm = &efx_pm_ops, 1274 .err_handler = &efx_err_handlers, 1275 #ifdef CONFIG_SFC_SRIOV 1276 .sriov_configure = efx_pci_sriov_configure, 1277 #endif 1278 }; 1279 1280 /************************************************************************** 1281 * 1282 * Kernel module interface 1283 * 1284 *************************************************************************/ 1285 1286 static int __init efx_init_module(void) 1287 { 1288 int rc; 1289 1290 printk(KERN_INFO "Solarflare NET driver\n"); 1291 1292 rc = register_netdevice_notifier(&efx_netdev_notifier); 1293 if (rc) 1294 goto err_notifier; 1295 1296 #ifdef CONFIG_SFC_SRIOV 1297 rc = efx_init_sriov(); 1298 if (rc) 1299 goto err_sriov; 1300 #endif 1301 1302 rc = efx_create_reset_workqueue(); 1303 if (rc) 1304 goto err_reset; 1305 1306 rc = pci_register_driver(&efx_pci_driver); 1307 if (rc < 0) 1308 goto err_pci; 1309 1310 rc = pci_register_driver(&ef100_pci_driver); 1311 if (rc < 0) 1312 goto err_pci_ef100; 1313 1314 return 0; 1315 1316 err_pci_ef100: 1317 pci_unregister_driver(&efx_pci_driver); 1318 err_pci: 1319 efx_destroy_reset_workqueue(); 1320 err_reset: 1321 #ifdef CONFIG_SFC_SRIOV 1322 efx_fini_sriov(); 1323 err_sriov: 1324 #endif 1325 unregister_netdevice_notifier(&efx_netdev_notifier); 1326 err_notifier: 1327 return rc; 1328 } 1329 1330 static void __exit efx_exit_module(void) 1331 { 1332 printk(KERN_INFO "Solarflare NET driver unloading\n"); 1333 1334 pci_unregister_driver(&ef100_pci_driver); 1335 pci_unregister_driver(&efx_pci_driver); 1336 efx_destroy_reset_workqueue(); 1337 #ifdef CONFIG_SFC_SRIOV 1338 efx_fini_sriov(); 1339 #endif 1340 unregister_netdevice_notifier(&efx_netdev_notifier); 1341 1342 } 1343 1344 module_init(efx_init_module); 1345 module_exit(efx_exit_module); 1346 1347 MODULE_AUTHOR("Solarflare Communications and " 1348 "Michael Brown <mbrown@fensystems.co.uk>"); 1349 MODULE_DESCRIPTION("Solarflare network driver"); 1350 MODULE_LICENSE("GPL"); 1351 MODULE_DEVICE_TABLE(pci, efx_pci_table); 1352