1 /* 2 * Copyright 2008-2010 Cisco Systems, Inc. All rights reserved. 3 * Copyright 2007 Nuova Systems, Inc. All rights reserved. 4 * 5 * This program is free software; you may redistribute it and/or modify 6 * it under the terms of the GNU General Public License as published by 7 * the Free Software Foundation; version 2 of the License. 8 * 9 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 10 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 11 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 12 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 13 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 14 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 15 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 16 * SOFTWARE. 17 * 18 */ 19 20 #include <linux/module.h> 21 #include <linux/kernel.h> 22 #include <linux/string.h> 23 #include <linux/errno.h> 24 #include <linux/types.h> 25 #include <linux/init.h> 26 #include <linux/interrupt.h> 27 #include <linux/workqueue.h> 28 #include <linux/pci.h> 29 #include <linux/netdevice.h> 30 #include <linux/etherdevice.h> 31 #include <linux/if.h> 32 #include <linux/if_ether.h> 33 #include <linux/if_vlan.h> 34 #include <linux/in.h> 35 #include <linux/ip.h> 36 #include <linux/ipv6.h> 37 #include <linux/tcp.h> 38 #include <linux/rtnetlink.h> 39 #include <linux/prefetch.h> 40 #include <net/ip6_checksum.h> 41 #include <linux/ktime.h> 42 #include <linux/numa.h> 43 #ifdef CONFIG_RFS_ACCEL 44 #include <linux/cpu_rmap.h> 45 #endif 46 #include <linux/crash_dump.h> 47 #include <net/busy_poll.h> 48 #include <net/vxlan.h> 49 50 #include "cq_enet_desc.h" 51 #include "vnic_dev.h" 52 #include "vnic_intr.h" 53 #include "vnic_stats.h" 54 #include "vnic_vic.h" 55 #include "enic_res.h" 56 #include "enic.h" 57 #include "enic_dev.h" 58 #include "enic_pp.h" 59 #include "enic_clsf.h" 60 61 #define ENIC_NOTIFY_TIMER_PERIOD (2 * HZ) 62 #define WQ_ENET_MAX_DESC_LEN (1 << WQ_ENET_LEN_BITS) 63 #define MAX_TSO (1 << 16) 64 #define ENIC_DESC_MAX_SPLITS (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1) 65 66 #define PCI_DEVICE_ID_CISCO_VIC_ENET 0x0043 /* ethernet vnic */ 67 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN 0x0044 /* enet dynamic vnic */ 68 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF 0x0071 /* enet SRIOV VF */ 69 70 #define RX_COPYBREAK_DEFAULT 256 71 72 /* Supported devices */ 73 static const struct pci_device_id enic_id_table[] = { 74 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) }, 75 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) }, 76 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) }, 77 { 0, } /* end of table */ 78 }; 79 80 MODULE_DESCRIPTION(DRV_DESCRIPTION); 81 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>"); 82 MODULE_LICENSE("GPL"); 83 MODULE_DEVICE_TABLE(pci, enic_id_table); 84 85 #define ENIC_LARGE_PKT_THRESHOLD 1000 86 #define ENIC_MAX_COALESCE_TIMERS 10 87 /* Interrupt moderation table, which will be used to decide the 88 * coalescing timer values 89 * {rx_rate in Mbps, mapping percentage of the range} 90 */ 91 static struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = { 92 {4000, 0}, 93 {4400, 10}, 94 {5060, 20}, 95 {5230, 30}, 96 {5540, 40}, 97 {5820, 50}, 98 {6120, 60}, 99 {6435, 70}, 100 {6745, 80}, 101 {7000, 90}, 102 {0xFFFFFFFF, 100} 103 }; 104 105 /* This table helps the driver to pick different ranges for rx coalescing 106 * timer depending on the link speed. 107 */ 108 static struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = { 109 {0, 0}, /* 0 - 4 Gbps */ 110 {0, 3}, /* 4 - 10 Gbps */ 111 {3, 6}, /* 10 - 40 Gbps */ 112 }; 113 114 static void enic_init_affinity_hint(struct enic *enic) 115 { 116 int numa_node = dev_to_node(&enic->pdev->dev); 117 int i; 118 119 for (i = 0; i < enic->intr_count; i++) { 120 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i) || 121 (cpumask_available(enic->msix[i].affinity_mask) && 122 !cpumask_empty(enic->msix[i].affinity_mask))) 123 continue; 124 if (zalloc_cpumask_var(&enic->msix[i].affinity_mask, 125 GFP_KERNEL)) 126 cpumask_set_cpu(cpumask_local_spread(i, numa_node), 127 enic->msix[i].affinity_mask); 128 } 129 } 130 131 static void enic_free_affinity_hint(struct enic *enic) 132 { 133 int i; 134 135 for (i = 0; i < enic->intr_count; i++) { 136 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i)) 137 continue; 138 free_cpumask_var(enic->msix[i].affinity_mask); 139 } 140 } 141 142 static void enic_set_affinity_hint(struct enic *enic) 143 { 144 int i; 145 int err; 146 147 for (i = 0; i < enic->intr_count; i++) { 148 if (enic_is_err_intr(enic, i) || 149 enic_is_notify_intr(enic, i) || 150 !cpumask_available(enic->msix[i].affinity_mask) || 151 cpumask_empty(enic->msix[i].affinity_mask)) 152 continue; 153 err = irq_update_affinity_hint(enic->msix_entry[i].vector, 154 enic->msix[i].affinity_mask); 155 if (err) 156 netdev_warn(enic->netdev, "irq_update_affinity_hint failed, err %d\n", 157 err); 158 } 159 160 for (i = 0; i < enic->wq_count; i++) { 161 int wq_intr = enic_msix_wq_intr(enic, i); 162 163 if (cpumask_available(enic->msix[wq_intr].affinity_mask) && 164 !cpumask_empty(enic->msix[wq_intr].affinity_mask)) 165 netif_set_xps_queue(enic->netdev, 166 enic->msix[wq_intr].affinity_mask, 167 i); 168 } 169 } 170 171 static void enic_unset_affinity_hint(struct enic *enic) 172 { 173 int i; 174 175 for (i = 0; i < enic->intr_count; i++) 176 irq_update_affinity_hint(enic->msix_entry[i].vector, NULL); 177 } 178 179 static int enic_udp_tunnel_set_port(struct net_device *netdev, 180 unsigned int table, unsigned int entry, 181 struct udp_tunnel_info *ti) 182 { 183 struct enic *enic = netdev_priv(netdev); 184 int err; 185 186 spin_lock_bh(&enic->devcmd_lock); 187 188 err = vnic_dev_overlay_offload_cfg(enic->vdev, 189 OVERLAY_CFG_VXLAN_PORT_UPDATE, 190 ntohs(ti->port)); 191 if (err) 192 goto error; 193 194 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN, 195 enic->vxlan.patch_level); 196 if (err) 197 goto error; 198 199 enic->vxlan.vxlan_udp_port_number = ntohs(ti->port); 200 error: 201 spin_unlock_bh(&enic->devcmd_lock); 202 203 return err; 204 } 205 206 static int enic_udp_tunnel_unset_port(struct net_device *netdev, 207 unsigned int table, unsigned int entry, 208 struct udp_tunnel_info *ti) 209 { 210 struct enic *enic = netdev_priv(netdev); 211 int err; 212 213 spin_lock_bh(&enic->devcmd_lock); 214 215 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN, 216 OVERLAY_OFFLOAD_DISABLE); 217 if (err) 218 goto unlock; 219 220 enic->vxlan.vxlan_udp_port_number = 0; 221 222 unlock: 223 spin_unlock_bh(&enic->devcmd_lock); 224 225 return err; 226 } 227 228 static const struct udp_tunnel_nic_info enic_udp_tunnels = { 229 .set_port = enic_udp_tunnel_set_port, 230 .unset_port = enic_udp_tunnel_unset_port, 231 .tables = { 232 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, }, 233 }, 234 }, enic_udp_tunnels_v4 = { 235 .set_port = enic_udp_tunnel_set_port, 236 .unset_port = enic_udp_tunnel_unset_port, 237 .flags = UDP_TUNNEL_NIC_INFO_IPV4_ONLY, 238 .tables = { 239 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, }, 240 }, 241 }; 242 243 static netdev_features_t enic_features_check(struct sk_buff *skb, 244 struct net_device *dev, 245 netdev_features_t features) 246 { 247 const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb); 248 struct enic *enic = netdev_priv(dev); 249 struct udphdr *udph; 250 u16 port = 0; 251 u8 proto; 252 253 if (!skb->encapsulation) 254 return features; 255 256 features = vxlan_features_check(skb, features); 257 258 switch (vlan_get_protocol(skb)) { 259 case htons(ETH_P_IPV6): 260 if (!(enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6)) 261 goto out; 262 proto = ipv6_hdr(skb)->nexthdr; 263 break; 264 case htons(ETH_P_IP): 265 proto = ip_hdr(skb)->protocol; 266 break; 267 default: 268 goto out; 269 } 270 271 switch (eth->h_proto) { 272 case ntohs(ETH_P_IPV6): 273 if (!(enic->vxlan.flags & ENIC_VXLAN_INNER_IPV6)) 274 goto out; 275 fallthrough; 276 case ntohs(ETH_P_IP): 277 break; 278 default: 279 goto out; 280 } 281 282 283 if (proto == IPPROTO_UDP) { 284 udph = udp_hdr(skb); 285 port = be16_to_cpu(udph->dest); 286 } 287 288 /* HW supports offload of only one UDP port. Remove CSUM and GSO MASK 289 * for other UDP port tunnels 290 */ 291 if (port != enic->vxlan.vxlan_udp_port_number) 292 goto out; 293 294 return features; 295 296 out: 297 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 298 } 299 300 int enic_is_dynamic(struct enic *enic) 301 { 302 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN; 303 } 304 305 int enic_sriov_enabled(struct enic *enic) 306 { 307 return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0; 308 } 309 310 static int enic_is_sriov_vf(struct enic *enic) 311 { 312 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF; 313 } 314 315 int enic_is_valid_vf(struct enic *enic, int vf) 316 { 317 #ifdef CONFIG_PCI_IOV 318 return vf >= 0 && vf < enic->num_vfs; 319 #else 320 return 0; 321 #endif 322 } 323 324 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf) 325 { 326 struct enic *enic = vnic_dev_priv(wq->vdev); 327 328 if (buf->sop) 329 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len, 330 DMA_TO_DEVICE); 331 else 332 dma_unmap_page(&enic->pdev->dev, buf->dma_addr, buf->len, 333 DMA_TO_DEVICE); 334 335 if (buf->os_buf) 336 dev_kfree_skb_any(buf->os_buf); 337 } 338 339 static void enic_wq_free_buf(struct vnic_wq *wq, 340 struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque) 341 { 342 enic_free_wq_buf(wq, buf); 343 } 344 345 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc, 346 u8 type, u16 q_number, u16 completed_index, void *opaque) 347 { 348 struct enic *enic = vnic_dev_priv(vdev); 349 350 spin_lock(&enic->wq_lock[q_number]); 351 352 vnic_wq_service(&enic->wq[q_number], cq_desc, 353 completed_index, enic_wq_free_buf, 354 opaque); 355 356 if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) && 357 vnic_wq_desc_avail(&enic->wq[q_number]) >= 358 (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)) 359 netif_wake_subqueue(enic->netdev, q_number); 360 361 spin_unlock(&enic->wq_lock[q_number]); 362 363 return 0; 364 } 365 366 static bool enic_log_q_error(struct enic *enic) 367 { 368 unsigned int i; 369 u32 error_status; 370 bool err = false; 371 372 for (i = 0; i < enic->wq_count; i++) { 373 error_status = vnic_wq_error_status(&enic->wq[i]); 374 err |= error_status; 375 if (error_status) 376 netdev_err(enic->netdev, "WQ[%d] error_status %d\n", 377 i, error_status); 378 } 379 380 for (i = 0; i < enic->rq_count; i++) { 381 error_status = vnic_rq_error_status(&enic->rq[i]); 382 err |= error_status; 383 if (error_status) 384 netdev_err(enic->netdev, "RQ[%d] error_status %d\n", 385 i, error_status); 386 } 387 388 return err; 389 } 390 391 static void enic_msglvl_check(struct enic *enic) 392 { 393 u32 msg_enable = vnic_dev_msg_lvl(enic->vdev); 394 395 if (msg_enable != enic->msg_enable) { 396 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n", 397 enic->msg_enable, msg_enable); 398 enic->msg_enable = msg_enable; 399 } 400 } 401 402 static void enic_mtu_check(struct enic *enic) 403 { 404 u32 mtu = vnic_dev_mtu(enic->vdev); 405 struct net_device *netdev = enic->netdev; 406 407 if (mtu && mtu != enic->port_mtu) { 408 enic->port_mtu = mtu; 409 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) { 410 mtu = max_t(int, ENIC_MIN_MTU, 411 min_t(int, ENIC_MAX_MTU, mtu)); 412 if (mtu != netdev->mtu) 413 schedule_work(&enic->change_mtu_work); 414 } else { 415 if (mtu < netdev->mtu) 416 netdev_warn(netdev, 417 "interface MTU (%d) set higher " 418 "than switch port MTU (%d)\n", 419 netdev->mtu, mtu); 420 } 421 } 422 } 423 424 static void enic_link_check(struct enic *enic) 425 { 426 int link_status = vnic_dev_link_status(enic->vdev); 427 int carrier_ok = netif_carrier_ok(enic->netdev); 428 429 if (link_status && !carrier_ok) { 430 netdev_info(enic->netdev, "Link UP\n"); 431 netif_carrier_on(enic->netdev); 432 } else if (!link_status && carrier_ok) { 433 netdev_info(enic->netdev, "Link DOWN\n"); 434 netif_carrier_off(enic->netdev); 435 } 436 } 437 438 static void enic_notify_check(struct enic *enic) 439 { 440 enic_msglvl_check(enic); 441 enic_mtu_check(enic); 442 enic_link_check(enic); 443 } 444 445 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i)) 446 447 static irqreturn_t enic_isr_legacy(int irq, void *data) 448 { 449 struct net_device *netdev = data; 450 struct enic *enic = netdev_priv(netdev); 451 unsigned int io_intr = enic_legacy_io_intr(); 452 unsigned int err_intr = enic_legacy_err_intr(); 453 unsigned int notify_intr = enic_legacy_notify_intr(); 454 u32 pba; 455 456 vnic_intr_mask(&enic->intr[io_intr]); 457 458 pba = vnic_intr_legacy_pba(enic->legacy_pba); 459 if (!pba) { 460 vnic_intr_unmask(&enic->intr[io_intr]); 461 return IRQ_NONE; /* not our interrupt */ 462 } 463 464 if (ENIC_TEST_INTR(pba, notify_intr)) { 465 enic_notify_check(enic); 466 vnic_intr_return_all_credits(&enic->intr[notify_intr]); 467 } 468 469 if (ENIC_TEST_INTR(pba, err_intr)) { 470 vnic_intr_return_all_credits(&enic->intr[err_intr]); 471 enic_log_q_error(enic); 472 /* schedule recovery from WQ/RQ error */ 473 schedule_work(&enic->reset); 474 return IRQ_HANDLED; 475 } 476 477 if (ENIC_TEST_INTR(pba, io_intr)) 478 napi_schedule_irqoff(&enic->napi[0]); 479 else 480 vnic_intr_unmask(&enic->intr[io_intr]); 481 482 return IRQ_HANDLED; 483 } 484 485 static irqreturn_t enic_isr_msi(int irq, void *data) 486 { 487 struct enic *enic = data; 488 489 /* With MSI, there is no sharing of interrupts, so this is 490 * our interrupt and there is no need to ack it. The device 491 * is not providing per-vector masking, so the OS will not 492 * write to PCI config space to mask/unmask the interrupt. 493 * We're using mask_on_assertion for MSI, so the device 494 * automatically masks the interrupt when the interrupt is 495 * generated. Later, when exiting polling, the interrupt 496 * will be unmasked (see enic_poll). 497 * 498 * Also, the device uses the same PCIe Traffic Class (TC) 499 * for Memory Write data and MSI, so there are no ordering 500 * issues; the MSI will always arrive at the Root Complex 501 * _after_ corresponding Memory Writes (i.e. descriptor 502 * writes). 503 */ 504 505 napi_schedule_irqoff(&enic->napi[0]); 506 507 return IRQ_HANDLED; 508 } 509 510 static irqreturn_t enic_isr_msix(int irq, void *data) 511 { 512 struct napi_struct *napi = data; 513 514 napi_schedule_irqoff(napi); 515 516 return IRQ_HANDLED; 517 } 518 519 static irqreturn_t enic_isr_msix_err(int irq, void *data) 520 { 521 struct enic *enic = data; 522 unsigned int intr = enic_msix_err_intr(enic); 523 524 vnic_intr_return_all_credits(&enic->intr[intr]); 525 526 if (enic_log_q_error(enic)) 527 /* schedule recovery from WQ/RQ error */ 528 schedule_work(&enic->reset); 529 530 return IRQ_HANDLED; 531 } 532 533 static irqreturn_t enic_isr_msix_notify(int irq, void *data) 534 { 535 struct enic *enic = data; 536 unsigned int intr = enic_msix_notify_intr(enic); 537 538 enic_notify_check(enic); 539 vnic_intr_return_all_credits(&enic->intr[intr]); 540 541 return IRQ_HANDLED; 542 } 543 544 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq, 545 struct sk_buff *skb, unsigned int len_left, 546 int loopback) 547 { 548 const skb_frag_t *frag; 549 dma_addr_t dma_addr; 550 551 /* Queue additional data fragments */ 552 for (frag = skb_shinfo(skb)->frags; len_left; frag++) { 553 len_left -= skb_frag_size(frag); 554 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 0, 555 skb_frag_size(frag), 556 DMA_TO_DEVICE); 557 if (unlikely(enic_dma_map_check(enic, dma_addr))) 558 return -ENOMEM; 559 enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag), 560 (len_left == 0), /* EOP? */ 561 loopback); 562 } 563 564 return 0; 565 } 566 567 static int enic_queue_wq_skb_vlan(struct enic *enic, struct vnic_wq *wq, 568 struct sk_buff *skb, int vlan_tag_insert, 569 unsigned int vlan_tag, int loopback) 570 { 571 unsigned int head_len = skb_headlen(skb); 572 unsigned int len_left = skb->len - head_len; 573 int eop = (len_left == 0); 574 dma_addr_t dma_addr; 575 int err = 0; 576 577 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len, 578 DMA_TO_DEVICE); 579 if (unlikely(enic_dma_map_check(enic, dma_addr))) 580 return -ENOMEM; 581 582 /* Queue the main skb fragment. The fragments are no larger 583 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less 584 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor 585 * per fragment is queued. 586 */ 587 enic_queue_wq_desc(wq, skb, dma_addr, head_len, vlan_tag_insert, 588 vlan_tag, eop, loopback); 589 590 if (!eop) 591 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback); 592 593 return err; 594 } 595 596 static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq, 597 struct sk_buff *skb, int vlan_tag_insert, 598 unsigned int vlan_tag, int loopback) 599 { 600 unsigned int head_len = skb_headlen(skb); 601 unsigned int len_left = skb->len - head_len; 602 unsigned int hdr_len = skb_checksum_start_offset(skb); 603 unsigned int csum_offset = hdr_len + skb->csum_offset; 604 int eop = (len_left == 0); 605 dma_addr_t dma_addr; 606 int err = 0; 607 608 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len, 609 DMA_TO_DEVICE); 610 if (unlikely(enic_dma_map_check(enic, dma_addr))) 611 return -ENOMEM; 612 613 /* Queue the main skb fragment. The fragments are no larger 614 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less 615 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor 616 * per fragment is queued. 617 */ 618 enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len, csum_offset, 619 hdr_len, vlan_tag_insert, vlan_tag, eop, 620 loopback); 621 622 if (!eop) 623 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback); 624 625 return err; 626 } 627 628 static void enic_preload_tcp_csum_encap(struct sk_buff *skb) 629 { 630 const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb); 631 632 switch (eth->h_proto) { 633 case ntohs(ETH_P_IP): 634 inner_ip_hdr(skb)->check = 0; 635 inner_tcp_hdr(skb)->check = 636 ~csum_tcpudp_magic(inner_ip_hdr(skb)->saddr, 637 inner_ip_hdr(skb)->daddr, 0, 638 IPPROTO_TCP, 0); 639 break; 640 case ntohs(ETH_P_IPV6): 641 inner_tcp_hdr(skb)->check = 642 ~csum_ipv6_magic(&inner_ipv6_hdr(skb)->saddr, 643 &inner_ipv6_hdr(skb)->daddr, 0, 644 IPPROTO_TCP, 0); 645 break; 646 default: 647 WARN_ONCE(1, "Non ipv4/ipv6 inner pkt for encap offload"); 648 break; 649 } 650 } 651 652 static void enic_preload_tcp_csum(struct sk_buff *skb) 653 { 654 /* Preload TCP csum field with IP pseudo hdr calculated 655 * with IP length set to zero. HW will later add in length 656 * to each TCP segment resulting from the TSO. 657 */ 658 659 if (skb->protocol == cpu_to_be16(ETH_P_IP)) { 660 ip_hdr(skb)->check = 0; 661 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 662 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); 663 } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) { 664 tcp_v6_gso_csum_prep(skb); 665 } 666 } 667 668 static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq, 669 struct sk_buff *skb, unsigned int mss, 670 int vlan_tag_insert, unsigned int vlan_tag, 671 int loopback) 672 { 673 unsigned int frag_len_left = skb_headlen(skb); 674 unsigned int len_left = skb->len - frag_len_left; 675 int eop = (len_left == 0); 676 unsigned int offset = 0; 677 unsigned int hdr_len; 678 dma_addr_t dma_addr; 679 unsigned int len; 680 skb_frag_t *frag; 681 682 if (skb->encapsulation) { 683 hdr_len = skb_inner_transport_header(skb) - skb->data; 684 hdr_len += inner_tcp_hdrlen(skb); 685 enic_preload_tcp_csum_encap(skb); 686 } else { 687 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb); 688 enic_preload_tcp_csum(skb); 689 } 690 691 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors 692 * for the main skb fragment 693 */ 694 while (frag_len_left) { 695 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN); 696 dma_addr = dma_map_single(&enic->pdev->dev, 697 skb->data + offset, len, 698 DMA_TO_DEVICE); 699 if (unlikely(enic_dma_map_check(enic, dma_addr))) 700 return -ENOMEM; 701 enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len, 702 vlan_tag_insert, vlan_tag, 703 eop && (len == frag_len_left), loopback); 704 frag_len_left -= len; 705 offset += len; 706 } 707 708 if (eop) 709 return 0; 710 711 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors 712 * for additional data fragments 713 */ 714 for (frag = skb_shinfo(skb)->frags; len_left; frag++) { 715 len_left -= skb_frag_size(frag); 716 frag_len_left = skb_frag_size(frag); 717 offset = 0; 718 719 while (frag_len_left) { 720 len = min(frag_len_left, 721 (unsigned int)WQ_ENET_MAX_DESC_LEN); 722 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 723 offset, len, 724 DMA_TO_DEVICE); 725 if (unlikely(enic_dma_map_check(enic, dma_addr))) 726 return -ENOMEM; 727 enic_queue_wq_desc_cont(wq, skb, dma_addr, len, 728 (len_left == 0) && 729 (len == frag_len_left),/*EOP*/ 730 loopback); 731 frag_len_left -= len; 732 offset += len; 733 } 734 } 735 736 return 0; 737 } 738 739 static inline int enic_queue_wq_skb_encap(struct enic *enic, struct vnic_wq *wq, 740 struct sk_buff *skb, 741 int vlan_tag_insert, 742 unsigned int vlan_tag, int loopback) 743 { 744 unsigned int head_len = skb_headlen(skb); 745 unsigned int len_left = skb->len - head_len; 746 /* Hardware will overwrite the checksum fields, calculating from 747 * scratch and ignoring the value placed by software. 748 * Offload mode = 00 749 * mss[2], mss[1], mss[0] bits are set 750 */ 751 unsigned int mss_or_csum = 7; 752 int eop = (len_left == 0); 753 dma_addr_t dma_addr; 754 int err = 0; 755 756 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len, 757 DMA_TO_DEVICE); 758 if (unlikely(enic_dma_map_check(enic, dma_addr))) 759 return -ENOMEM; 760 761 enic_queue_wq_desc_ex(wq, skb, dma_addr, head_len, mss_or_csum, 0, 762 vlan_tag_insert, vlan_tag, 763 WQ_ENET_OFFLOAD_MODE_CSUM, eop, 1 /* SOP */, eop, 764 loopback); 765 if (!eop) 766 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback); 767 768 return err; 769 } 770 771 static inline int enic_queue_wq_skb(struct enic *enic, 772 struct vnic_wq *wq, struct sk_buff *skb) 773 { 774 unsigned int mss = skb_shinfo(skb)->gso_size; 775 unsigned int vlan_tag = 0; 776 int vlan_tag_insert = 0; 777 int loopback = 0; 778 int err; 779 780 if (skb_vlan_tag_present(skb)) { 781 /* VLAN tag from trunking driver */ 782 vlan_tag_insert = 1; 783 vlan_tag = skb_vlan_tag_get(skb); 784 } else if (enic->loop_enable) { 785 vlan_tag = enic->loop_tag; 786 loopback = 1; 787 } 788 789 if (mss) 790 err = enic_queue_wq_skb_tso(enic, wq, skb, mss, 791 vlan_tag_insert, vlan_tag, 792 loopback); 793 else if (skb->encapsulation) 794 err = enic_queue_wq_skb_encap(enic, wq, skb, vlan_tag_insert, 795 vlan_tag, loopback); 796 else if (skb->ip_summed == CHECKSUM_PARTIAL) 797 err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert, 798 vlan_tag, loopback); 799 else 800 err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert, 801 vlan_tag, loopback); 802 if (unlikely(err)) { 803 struct vnic_wq_buf *buf; 804 805 buf = wq->to_use->prev; 806 /* while not EOP of previous pkt && queue not empty. 807 * For all non EOP bufs, os_buf is NULL. 808 */ 809 while (!buf->os_buf && (buf->next != wq->to_clean)) { 810 enic_free_wq_buf(wq, buf); 811 wq->ring.desc_avail++; 812 buf = buf->prev; 813 } 814 wq->to_use = buf->next; 815 dev_kfree_skb(skb); 816 } 817 return err; 818 } 819 820 /* netif_tx_lock held, process context with BHs disabled, or BH */ 821 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb, 822 struct net_device *netdev) 823 { 824 struct enic *enic = netdev_priv(netdev); 825 struct vnic_wq *wq; 826 unsigned int txq_map; 827 struct netdev_queue *txq; 828 829 if (skb->len <= 0) { 830 dev_kfree_skb_any(skb); 831 return NETDEV_TX_OK; 832 } 833 834 txq_map = skb_get_queue_mapping(skb) % enic->wq_count; 835 wq = &enic->wq[txq_map]; 836 txq = netdev_get_tx_queue(netdev, txq_map); 837 838 /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs, 839 * which is very likely. In the off chance it's going to take 840 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb. 841 */ 842 843 if (skb_shinfo(skb)->gso_size == 0 && 844 skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC && 845 skb_linearize(skb)) { 846 dev_kfree_skb_any(skb); 847 return NETDEV_TX_OK; 848 } 849 850 spin_lock(&enic->wq_lock[txq_map]); 851 852 if (vnic_wq_desc_avail(wq) < 853 skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) { 854 netif_tx_stop_queue(txq); 855 /* This is a hard error, log it */ 856 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n"); 857 spin_unlock(&enic->wq_lock[txq_map]); 858 return NETDEV_TX_BUSY; 859 } 860 861 if (enic_queue_wq_skb(enic, wq, skb)) 862 goto error; 863 864 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS) 865 netif_tx_stop_queue(txq); 866 skb_tx_timestamp(skb); 867 if (!netdev_xmit_more() || netif_xmit_stopped(txq)) 868 vnic_wq_doorbell(wq); 869 870 error: 871 spin_unlock(&enic->wq_lock[txq_map]); 872 873 return NETDEV_TX_OK; 874 } 875 876 /* dev_base_lock rwlock held, nominally process context */ 877 static void enic_get_stats(struct net_device *netdev, 878 struct rtnl_link_stats64 *net_stats) 879 { 880 struct enic *enic = netdev_priv(netdev); 881 struct vnic_stats *stats; 882 int err; 883 884 err = enic_dev_stats_dump(enic, &stats); 885 /* return only when dma_alloc_coherent fails in vnic_dev_stats_dump 886 * For other failures, like devcmd failure, we return previously 887 * recorded stats. 888 */ 889 if (err == -ENOMEM) 890 return; 891 892 net_stats->tx_packets = stats->tx.tx_frames_ok; 893 net_stats->tx_bytes = stats->tx.tx_bytes_ok; 894 net_stats->tx_errors = stats->tx.tx_errors; 895 net_stats->tx_dropped = stats->tx.tx_drops; 896 897 net_stats->rx_packets = stats->rx.rx_frames_ok; 898 net_stats->rx_bytes = stats->rx.rx_bytes_ok; 899 net_stats->rx_errors = stats->rx.rx_errors; 900 net_stats->multicast = stats->rx.rx_multicast_frames_ok; 901 net_stats->rx_over_errors = enic->rq_truncated_pkts; 902 net_stats->rx_crc_errors = enic->rq_bad_fcs; 903 net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop; 904 } 905 906 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr) 907 { 908 struct enic *enic = netdev_priv(netdev); 909 910 if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) { 911 unsigned int mc_count = netdev_mc_count(netdev); 912 913 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n", 914 ENIC_MULTICAST_PERFECT_FILTERS, mc_count); 915 916 return -ENOSPC; 917 } 918 919 enic_dev_add_addr(enic, mc_addr); 920 enic->mc_count++; 921 922 return 0; 923 } 924 925 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr) 926 { 927 struct enic *enic = netdev_priv(netdev); 928 929 enic_dev_del_addr(enic, mc_addr); 930 enic->mc_count--; 931 932 return 0; 933 } 934 935 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr) 936 { 937 struct enic *enic = netdev_priv(netdev); 938 939 if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) { 940 unsigned int uc_count = netdev_uc_count(netdev); 941 942 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n", 943 ENIC_UNICAST_PERFECT_FILTERS, uc_count); 944 945 return -ENOSPC; 946 } 947 948 enic_dev_add_addr(enic, uc_addr); 949 enic->uc_count++; 950 951 return 0; 952 } 953 954 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr) 955 { 956 struct enic *enic = netdev_priv(netdev); 957 958 enic_dev_del_addr(enic, uc_addr); 959 enic->uc_count--; 960 961 return 0; 962 } 963 964 void enic_reset_addr_lists(struct enic *enic) 965 { 966 struct net_device *netdev = enic->netdev; 967 968 __dev_uc_unsync(netdev, NULL); 969 __dev_mc_unsync(netdev, NULL); 970 971 enic->mc_count = 0; 972 enic->uc_count = 0; 973 enic->flags = 0; 974 } 975 976 static int enic_set_mac_addr(struct net_device *netdev, char *addr) 977 { 978 struct enic *enic = netdev_priv(netdev); 979 980 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) { 981 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr)) 982 return -EADDRNOTAVAIL; 983 } else { 984 if (!is_valid_ether_addr(addr)) 985 return -EADDRNOTAVAIL; 986 } 987 988 eth_hw_addr_set(netdev, addr); 989 990 return 0; 991 } 992 993 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p) 994 { 995 struct enic *enic = netdev_priv(netdev); 996 struct sockaddr *saddr = p; 997 char *addr = saddr->sa_data; 998 int err; 999 1000 if (netif_running(enic->netdev)) { 1001 err = enic_dev_del_station_addr(enic); 1002 if (err) 1003 return err; 1004 } 1005 1006 err = enic_set_mac_addr(netdev, addr); 1007 if (err) 1008 return err; 1009 1010 if (netif_running(enic->netdev)) { 1011 err = enic_dev_add_station_addr(enic); 1012 if (err) 1013 return err; 1014 } 1015 1016 return err; 1017 } 1018 1019 static int enic_set_mac_address(struct net_device *netdev, void *p) 1020 { 1021 struct sockaddr *saddr = p; 1022 char *addr = saddr->sa_data; 1023 struct enic *enic = netdev_priv(netdev); 1024 int err; 1025 1026 err = enic_dev_del_station_addr(enic); 1027 if (err) 1028 return err; 1029 1030 err = enic_set_mac_addr(netdev, addr); 1031 if (err) 1032 return err; 1033 1034 return enic_dev_add_station_addr(enic); 1035 } 1036 1037 /* netif_tx_lock held, BHs disabled */ 1038 static void enic_set_rx_mode(struct net_device *netdev) 1039 { 1040 struct enic *enic = netdev_priv(netdev); 1041 int directed = 1; 1042 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0; 1043 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0; 1044 int promisc = (netdev->flags & IFF_PROMISC) || 1045 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS; 1046 int allmulti = (netdev->flags & IFF_ALLMULTI) || 1047 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS; 1048 unsigned int flags = netdev->flags | 1049 (allmulti ? IFF_ALLMULTI : 0) | 1050 (promisc ? IFF_PROMISC : 0); 1051 1052 if (enic->flags != flags) { 1053 enic->flags = flags; 1054 enic_dev_packet_filter(enic, directed, 1055 multicast, broadcast, promisc, allmulti); 1056 } 1057 1058 if (!promisc) { 1059 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync); 1060 if (!allmulti) 1061 __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync); 1062 } 1063 } 1064 1065 /* netif_tx_lock held, BHs disabled */ 1066 static void enic_tx_timeout(struct net_device *netdev, unsigned int txqueue) 1067 { 1068 struct enic *enic = netdev_priv(netdev); 1069 schedule_work(&enic->tx_hang_reset); 1070 } 1071 1072 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac) 1073 { 1074 struct enic *enic = netdev_priv(netdev); 1075 struct enic_port_profile *pp; 1076 int err; 1077 1078 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1079 if (err) 1080 return err; 1081 1082 if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) { 1083 if (vf == PORT_SELF_VF) { 1084 memcpy(pp->vf_mac, mac, ETH_ALEN); 1085 return 0; 1086 } else { 1087 /* 1088 * For sriov vf's set the mac in hw 1089 */ 1090 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic, 1091 vnic_dev_set_mac_addr, mac); 1092 return enic_dev_status_to_errno(err); 1093 } 1094 } else 1095 return -EINVAL; 1096 } 1097 1098 static int enic_set_vf_port(struct net_device *netdev, int vf, 1099 struct nlattr *port[]) 1100 { 1101 static const u8 zero_addr[ETH_ALEN] = {}; 1102 struct enic *enic = netdev_priv(netdev); 1103 struct enic_port_profile prev_pp; 1104 struct enic_port_profile *pp; 1105 int err = 0, restore_pp = 1; 1106 1107 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1108 if (err) 1109 return err; 1110 1111 if (!port[IFLA_PORT_REQUEST]) 1112 return -EOPNOTSUPP; 1113 1114 memcpy(&prev_pp, pp, sizeof(*enic->pp)); 1115 memset(pp, 0, sizeof(*enic->pp)); 1116 1117 pp->set |= ENIC_SET_REQUEST; 1118 pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]); 1119 1120 if (port[IFLA_PORT_PROFILE]) { 1121 pp->set |= ENIC_SET_NAME; 1122 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]), 1123 PORT_PROFILE_MAX); 1124 } 1125 1126 if (port[IFLA_PORT_INSTANCE_UUID]) { 1127 pp->set |= ENIC_SET_INSTANCE; 1128 memcpy(pp->instance_uuid, 1129 nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX); 1130 } 1131 1132 if (port[IFLA_PORT_HOST_UUID]) { 1133 pp->set |= ENIC_SET_HOST; 1134 memcpy(pp->host_uuid, 1135 nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX); 1136 } 1137 1138 if (vf == PORT_SELF_VF) { 1139 /* Special case handling: mac came from IFLA_VF_MAC */ 1140 if (!is_zero_ether_addr(prev_pp.vf_mac)) 1141 memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN); 1142 1143 if (is_zero_ether_addr(netdev->dev_addr)) 1144 eth_hw_addr_random(netdev); 1145 } else { 1146 /* SR-IOV VF: get mac from adapter */ 1147 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic, 1148 vnic_dev_get_mac_addr, pp->mac_addr); 1149 if (err) { 1150 netdev_err(netdev, "Error getting mac for vf %d\n", vf); 1151 memcpy(pp, &prev_pp, sizeof(*pp)); 1152 return enic_dev_status_to_errno(err); 1153 } 1154 } 1155 1156 err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp); 1157 if (err) { 1158 if (restore_pp) { 1159 /* Things are still the way they were: Implicit 1160 * DISASSOCIATE failed 1161 */ 1162 memcpy(pp, &prev_pp, sizeof(*pp)); 1163 } else { 1164 memset(pp, 0, sizeof(*pp)); 1165 if (vf == PORT_SELF_VF) 1166 eth_hw_addr_set(netdev, zero_addr); 1167 } 1168 } else { 1169 /* Set flag to indicate that the port assoc/disassoc 1170 * request has been sent out to fw 1171 */ 1172 pp->set |= ENIC_PORT_REQUEST_APPLIED; 1173 1174 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */ 1175 if (pp->request == PORT_REQUEST_DISASSOCIATE) { 1176 eth_zero_addr(pp->mac_addr); 1177 if (vf == PORT_SELF_VF) 1178 eth_hw_addr_set(netdev, zero_addr); 1179 } 1180 } 1181 1182 if (vf == PORT_SELF_VF) 1183 eth_zero_addr(pp->vf_mac); 1184 1185 return err; 1186 } 1187 1188 static int enic_get_vf_port(struct net_device *netdev, int vf, 1189 struct sk_buff *skb) 1190 { 1191 struct enic *enic = netdev_priv(netdev); 1192 u16 response = PORT_PROFILE_RESPONSE_SUCCESS; 1193 struct enic_port_profile *pp; 1194 int err; 1195 1196 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1197 if (err) 1198 return err; 1199 1200 if (!(pp->set & ENIC_PORT_REQUEST_APPLIED)) 1201 return -ENODATA; 1202 1203 err = enic_process_get_pp_request(enic, vf, pp->request, &response); 1204 if (err) 1205 return err; 1206 1207 if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) || 1208 nla_put_u16(skb, IFLA_PORT_RESPONSE, response) || 1209 ((pp->set & ENIC_SET_NAME) && 1210 nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) || 1211 ((pp->set & ENIC_SET_INSTANCE) && 1212 nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX, 1213 pp->instance_uuid)) || 1214 ((pp->set & ENIC_SET_HOST) && 1215 nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid))) 1216 goto nla_put_failure; 1217 return 0; 1218 1219 nla_put_failure: 1220 return -EMSGSIZE; 1221 } 1222 1223 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf) 1224 { 1225 struct enic *enic = vnic_dev_priv(rq->vdev); 1226 1227 if (!buf->os_buf) 1228 return; 1229 1230 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len, 1231 DMA_FROM_DEVICE); 1232 dev_kfree_skb_any(buf->os_buf); 1233 buf->os_buf = NULL; 1234 } 1235 1236 static int enic_rq_alloc_buf(struct vnic_rq *rq) 1237 { 1238 struct enic *enic = vnic_dev_priv(rq->vdev); 1239 struct net_device *netdev = enic->netdev; 1240 struct sk_buff *skb; 1241 unsigned int len = netdev->mtu + VLAN_ETH_HLEN; 1242 unsigned int os_buf_index = 0; 1243 dma_addr_t dma_addr; 1244 struct vnic_rq_buf *buf = rq->to_use; 1245 1246 if (buf->os_buf) { 1247 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr, 1248 buf->len); 1249 1250 return 0; 1251 } 1252 skb = netdev_alloc_skb_ip_align(netdev, len); 1253 if (!skb) 1254 return -ENOMEM; 1255 1256 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, len, 1257 DMA_FROM_DEVICE); 1258 if (unlikely(enic_dma_map_check(enic, dma_addr))) { 1259 dev_kfree_skb(skb); 1260 return -ENOMEM; 1261 } 1262 1263 enic_queue_rq_desc(rq, skb, os_buf_index, 1264 dma_addr, len); 1265 1266 return 0; 1267 } 1268 1269 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size, 1270 u32 pkt_len) 1271 { 1272 if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len) 1273 pkt_size->large_pkt_bytes_cnt += pkt_len; 1274 else 1275 pkt_size->small_pkt_bytes_cnt += pkt_len; 1276 } 1277 1278 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb, 1279 struct vnic_rq_buf *buf, u16 len) 1280 { 1281 struct enic *enic = netdev_priv(netdev); 1282 struct sk_buff *new_skb; 1283 1284 if (len > enic->rx_copybreak) 1285 return false; 1286 new_skb = netdev_alloc_skb_ip_align(netdev, len); 1287 if (!new_skb) 1288 return false; 1289 dma_sync_single_for_cpu(&enic->pdev->dev, buf->dma_addr, len, 1290 DMA_FROM_DEVICE); 1291 memcpy(new_skb->data, (*skb)->data, len); 1292 *skb = new_skb; 1293 1294 return true; 1295 } 1296 1297 static void enic_rq_indicate_buf(struct vnic_rq *rq, 1298 struct cq_desc *cq_desc, struct vnic_rq_buf *buf, 1299 int skipped, void *opaque) 1300 { 1301 struct enic *enic = vnic_dev_priv(rq->vdev); 1302 struct net_device *netdev = enic->netdev; 1303 struct sk_buff *skb; 1304 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1305 1306 u8 type, color, eop, sop, ingress_port, vlan_stripped; 1307 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof; 1308 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok; 1309 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc; 1310 u8 packet_error; 1311 u16 q_number, completed_index, bytes_written, vlan_tci, checksum; 1312 u32 rss_hash; 1313 bool outer_csum_ok = true, encap = false; 1314 1315 if (skipped) 1316 return; 1317 1318 skb = buf->os_buf; 1319 1320 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc, 1321 &type, &color, &q_number, &completed_index, 1322 &ingress_port, &fcoe, &eop, &sop, &rss_type, 1323 &csum_not_calc, &rss_hash, &bytes_written, 1324 &packet_error, &vlan_stripped, &vlan_tci, &checksum, 1325 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error, 1326 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp, 1327 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment, 1328 &fcs_ok); 1329 1330 if (packet_error) { 1331 1332 if (!fcs_ok) { 1333 if (bytes_written > 0) 1334 enic->rq_bad_fcs++; 1335 else if (bytes_written == 0) 1336 enic->rq_truncated_pkts++; 1337 } 1338 1339 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len, 1340 DMA_FROM_DEVICE); 1341 dev_kfree_skb_any(skb); 1342 buf->os_buf = NULL; 1343 1344 return; 1345 } 1346 1347 if (eop && bytes_written > 0) { 1348 1349 /* Good receive 1350 */ 1351 1352 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) { 1353 buf->os_buf = NULL; 1354 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, 1355 buf->len, DMA_FROM_DEVICE); 1356 } 1357 prefetch(skb->data - NET_IP_ALIGN); 1358 1359 skb_put(skb, bytes_written); 1360 skb->protocol = eth_type_trans(skb, netdev); 1361 skb_record_rx_queue(skb, q_number); 1362 if ((netdev->features & NETIF_F_RXHASH) && rss_hash && 1363 (type == 3)) { 1364 switch (rss_type) { 1365 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4: 1366 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6: 1367 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX: 1368 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4); 1369 break; 1370 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4: 1371 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6: 1372 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX: 1373 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3); 1374 break; 1375 } 1376 } 1377 if (enic->vxlan.vxlan_udp_port_number) { 1378 switch (enic->vxlan.patch_level) { 1379 case 0: 1380 if (fcoe) { 1381 encap = true; 1382 outer_csum_ok = fcoe_fc_crc_ok; 1383 } 1384 break; 1385 case 2: 1386 if ((type == 7) && 1387 (rss_hash & BIT(0))) { 1388 encap = true; 1389 outer_csum_ok = (rss_hash & BIT(1)) && 1390 (rss_hash & BIT(2)); 1391 } 1392 break; 1393 } 1394 } 1395 1396 /* Hardware does not provide whole packet checksum. It only 1397 * provides pseudo checksum. Since hw validates the packet 1398 * checksum but not provide us the checksum value. use 1399 * CHECSUM_UNNECESSARY. 1400 * 1401 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is 1402 * inner csum_ok. outer_csum_ok is set by hw when outer udp 1403 * csum is correct or is zero. 1404 */ 1405 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc && 1406 tcp_udp_csum_ok && outer_csum_ok && 1407 (ipv4_csum_ok || ipv6)) { 1408 skb->ip_summed = CHECKSUM_UNNECESSARY; 1409 skb->csum_level = encap; 1410 } 1411 1412 if (vlan_stripped) 1413 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci); 1414 1415 skb_mark_napi_id(skb, &enic->napi[rq->index]); 1416 if (!(netdev->features & NETIF_F_GRO)) 1417 netif_receive_skb(skb); 1418 else 1419 napi_gro_receive(&enic->napi[q_number], skb); 1420 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1421 enic_intr_update_pkt_size(&cq->pkt_size_counter, 1422 bytes_written); 1423 } else { 1424 1425 /* Buffer overflow 1426 */ 1427 1428 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len, 1429 DMA_FROM_DEVICE); 1430 dev_kfree_skb_any(skb); 1431 buf->os_buf = NULL; 1432 } 1433 } 1434 1435 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc, 1436 u8 type, u16 q_number, u16 completed_index, void *opaque) 1437 { 1438 struct enic *enic = vnic_dev_priv(vdev); 1439 1440 vnic_rq_service(&enic->rq[q_number], cq_desc, 1441 completed_index, VNIC_RQ_RETURN_DESC, 1442 enic_rq_indicate_buf, opaque); 1443 1444 return 0; 1445 } 1446 1447 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq) 1448 { 1449 unsigned int intr = enic_msix_rq_intr(enic, rq->index); 1450 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1451 u32 timer = cq->tobe_rx_coal_timeval; 1452 1453 if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) { 1454 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer); 1455 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval; 1456 } 1457 } 1458 1459 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq) 1460 { 1461 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting; 1462 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1463 struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter; 1464 int index; 1465 u32 timer; 1466 u32 range_start; 1467 u32 traffic; 1468 u64 delta; 1469 ktime_t now = ktime_get(); 1470 1471 delta = ktime_us_delta(now, cq->prev_ts); 1472 if (delta < ENIC_AIC_TS_BREAK) 1473 return; 1474 cq->prev_ts = now; 1475 1476 traffic = pkt_size_counter->large_pkt_bytes_cnt + 1477 pkt_size_counter->small_pkt_bytes_cnt; 1478 /* The table takes Mbps 1479 * traffic *= 8 => bits 1480 * traffic *= (10^6 / delta) => bps 1481 * traffic /= 10^6 => Mbps 1482 * 1483 * Combining, traffic *= (8 / delta) 1484 */ 1485 1486 traffic <<= 3; 1487 traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta; 1488 1489 for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++) 1490 if (traffic < mod_table[index].rx_rate) 1491 break; 1492 range_start = (pkt_size_counter->small_pkt_bytes_cnt > 1493 pkt_size_counter->large_pkt_bytes_cnt << 1) ? 1494 rx_coal->small_pkt_range_start : 1495 rx_coal->large_pkt_range_start; 1496 timer = range_start + ((rx_coal->range_end - range_start) * 1497 mod_table[index].range_percent / 100); 1498 /* Damping */ 1499 cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1; 1500 1501 pkt_size_counter->large_pkt_bytes_cnt = 0; 1502 pkt_size_counter->small_pkt_bytes_cnt = 0; 1503 } 1504 1505 static int enic_poll(struct napi_struct *napi, int budget) 1506 { 1507 struct net_device *netdev = napi->dev; 1508 struct enic *enic = netdev_priv(netdev); 1509 unsigned int cq_rq = enic_cq_rq(enic, 0); 1510 unsigned int cq_wq = enic_cq_wq(enic, 0); 1511 unsigned int intr = enic_legacy_io_intr(); 1512 unsigned int rq_work_to_do = budget; 1513 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET; 1514 unsigned int work_done, rq_work_done = 0, wq_work_done; 1515 int err; 1516 1517 wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do, 1518 enic_wq_service, NULL); 1519 1520 if (budget > 0) 1521 rq_work_done = vnic_cq_service(&enic->cq[cq_rq], 1522 rq_work_to_do, enic_rq_service, NULL); 1523 1524 /* Accumulate intr event credits for this polling 1525 * cycle. An intr event is the completion of a 1526 * a WQ or RQ packet. 1527 */ 1528 1529 work_done = rq_work_done + wq_work_done; 1530 1531 if (work_done > 0) 1532 vnic_intr_return_credits(&enic->intr[intr], 1533 work_done, 1534 0 /* don't unmask intr */, 1535 0 /* don't reset intr timer */); 1536 1537 err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf); 1538 1539 /* Buffer allocation failed. Stay in polling 1540 * mode so we can try to fill the ring again. 1541 */ 1542 1543 if (err) 1544 rq_work_done = rq_work_to_do; 1545 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1546 /* Call the function which refreshes the intr coalescing timer 1547 * value based on the traffic. 1548 */ 1549 enic_calc_int_moderation(enic, &enic->rq[0]); 1550 1551 if ((rq_work_done < budget) && napi_complete_done(napi, rq_work_done)) { 1552 1553 /* Some work done, but not enough to stay in polling, 1554 * exit polling 1555 */ 1556 1557 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1558 enic_set_int_moderation(enic, &enic->rq[0]); 1559 vnic_intr_unmask(&enic->intr[intr]); 1560 } 1561 1562 return rq_work_done; 1563 } 1564 1565 #ifdef CONFIG_RFS_ACCEL 1566 static void enic_free_rx_cpu_rmap(struct enic *enic) 1567 { 1568 free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap); 1569 enic->netdev->rx_cpu_rmap = NULL; 1570 } 1571 1572 static void enic_set_rx_cpu_rmap(struct enic *enic) 1573 { 1574 int i, res; 1575 1576 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) { 1577 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count); 1578 if (unlikely(!enic->netdev->rx_cpu_rmap)) 1579 return; 1580 for (i = 0; i < enic->rq_count; i++) { 1581 res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap, 1582 enic->msix_entry[i].vector); 1583 if (unlikely(res)) { 1584 enic_free_rx_cpu_rmap(enic); 1585 return; 1586 } 1587 } 1588 } 1589 } 1590 1591 #else 1592 1593 static void enic_free_rx_cpu_rmap(struct enic *enic) 1594 { 1595 } 1596 1597 static void enic_set_rx_cpu_rmap(struct enic *enic) 1598 { 1599 } 1600 1601 #endif /* CONFIG_RFS_ACCEL */ 1602 1603 static int enic_poll_msix_wq(struct napi_struct *napi, int budget) 1604 { 1605 struct net_device *netdev = napi->dev; 1606 struct enic *enic = netdev_priv(netdev); 1607 unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count; 1608 struct vnic_wq *wq = &enic->wq[wq_index]; 1609 unsigned int cq; 1610 unsigned int intr; 1611 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET; 1612 unsigned int wq_work_done; 1613 unsigned int wq_irq; 1614 1615 wq_irq = wq->index; 1616 cq = enic_cq_wq(enic, wq_irq); 1617 intr = enic_msix_wq_intr(enic, wq_irq); 1618 wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do, 1619 enic_wq_service, NULL); 1620 1621 vnic_intr_return_credits(&enic->intr[intr], wq_work_done, 1622 0 /* don't unmask intr */, 1623 1 /* reset intr timer */); 1624 if (!wq_work_done) { 1625 napi_complete(napi); 1626 vnic_intr_unmask(&enic->intr[intr]); 1627 return 0; 1628 } 1629 1630 return budget; 1631 } 1632 1633 static int enic_poll_msix_rq(struct napi_struct *napi, int budget) 1634 { 1635 struct net_device *netdev = napi->dev; 1636 struct enic *enic = netdev_priv(netdev); 1637 unsigned int rq = (napi - &enic->napi[0]); 1638 unsigned int cq = enic_cq_rq(enic, rq); 1639 unsigned int intr = enic_msix_rq_intr(enic, rq); 1640 unsigned int work_to_do = budget; 1641 unsigned int work_done = 0; 1642 int err; 1643 1644 /* Service RQ 1645 */ 1646 1647 if (budget > 0) 1648 work_done = vnic_cq_service(&enic->cq[cq], 1649 work_to_do, enic_rq_service, NULL); 1650 1651 /* Return intr event credits for this polling 1652 * cycle. An intr event is the completion of a 1653 * RQ packet. 1654 */ 1655 1656 if (work_done > 0) 1657 vnic_intr_return_credits(&enic->intr[intr], 1658 work_done, 1659 0 /* don't unmask intr */, 1660 0 /* don't reset intr timer */); 1661 1662 err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf); 1663 1664 /* Buffer allocation failed. Stay in polling mode 1665 * so we can try to fill the ring again. 1666 */ 1667 1668 if (err) 1669 work_done = work_to_do; 1670 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1671 /* Call the function which refreshes the intr coalescing timer 1672 * value based on the traffic. 1673 */ 1674 enic_calc_int_moderation(enic, &enic->rq[rq]); 1675 1676 if ((work_done < budget) && napi_complete_done(napi, work_done)) { 1677 1678 /* Some work done, but not enough to stay in polling, 1679 * exit polling 1680 */ 1681 1682 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1683 enic_set_int_moderation(enic, &enic->rq[rq]); 1684 vnic_intr_unmask(&enic->intr[intr]); 1685 } 1686 1687 return work_done; 1688 } 1689 1690 static void enic_notify_timer(struct timer_list *t) 1691 { 1692 struct enic *enic = from_timer(enic, t, notify_timer); 1693 1694 enic_notify_check(enic); 1695 1696 mod_timer(&enic->notify_timer, 1697 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD)); 1698 } 1699 1700 static void enic_free_intr(struct enic *enic) 1701 { 1702 struct net_device *netdev = enic->netdev; 1703 unsigned int i; 1704 1705 enic_free_rx_cpu_rmap(enic); 1706 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1707 case VNIC_DEV_INTR_MODE_INTX: 1708 free_irq(enic->pdev->irq, netdev); 1709 break; 1710 case VNIC_DEV_INTR_MODE_MSI: 1711 free_irq(enic->pdev->irq, enic); 1712 break; 1713 case VNIC_DEV_INTR_MODE_MSIX: 1714 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) 1715 if (enic->msix[i].requested) 1716 free_irq(enic->msix_entry[i].vector, 1717 enic->msix[i].devid); 1718 break; 1719 default: 1720 break; 1721 } 1722 } 1723 1724 static int enic_request_intr(struct enic *enic) 1725 { 1726 struct net_device *netdev = enic->netdev; 1727 unsigned int i, intr; 1728 int err = 0; 1729 1730 enic_set_rx_cpu_rmap(enic); 1731 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1732 1733 case VNIC_DEV_INTR_MODE_INTX: 1734 1735 err = request_irq(enic->pdev->irq, enic_isr_legacy, 1736 IRQF_SHARED, netdev->name, netdev); 1737 break; 1738 1739 case VNIC_DEV_INTR_MODE_MSI: 1740 1741 err = request_irq(enic->pdev->irq, enic_isr_msi, 1742 0, netdev->name, enic); 1743 break; 1744 1745 case VNIC_DEV_INTR_MODE_MSIX: 1746 1747 for (i = 0; i < enic->rq_count; i++) { 1748 intr = enic_msix_rq_intr(enic, i); 1749 snprintf(enic->msix[intr].devname, 1750 sizeof(enic->msix[intr].devname), 1751 "%s-rx-%u", netdev->name, i); 1752 enic->msix[intr].isr = enic_isr_msix; 1753 enic->msix[intr].devid = &enic->napi[i]; 1754 } 1755 1756 for (i = 0; i < enic->wq_count; i++) { 1757 int wq = enic_cq_wq(enic, i); 1758 1759 intr = enic_msix_wq_intr(enic, i); 1760 snprintf(enic->msix[intr].devname, 1761 sizeof(enic->msix[intr].devname), 1762 "%s-tx-%u", netdev->name, i); 1763 enic->msix[intr].isr = enic_isr_msix; 1764 enic->msix[intr].devid = &enic->napi[wq]; 1765 } 1766 1767 intr = enic_msix_err_intr(enic); 1768 snprintf(enic->msix[intr].devname, 1769 sizeof(enic->msix[intr].devname), 1770 "%s-err", netdev->name); 1771 enic->msix[intr].isr = enic_isr_msix_err; 1772 enic->msix[intr].devid = enic; 1773 1774 intr = enic_msix_notify_intr(enic); 1775 snprintf(enic->msix[intr].devname, 1776 sizeof(enic->msix[intr].devname), 1777 "%s-notify", netdev->name); 1778 enic->msix[intr].isr = enic_isr_msix_notify; 1779 enic->msix[intr].devid = enic; 1780 1781 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) 1782 enic->msix[i].requested = 0; 1783 1784 for (i = 0; i < enic->intr_count; i++) { 1785 err = request_irq(enic->msix_entry[i].vector, 1786 enic->msix[i].isr, 0, 1787 enic->msix[i].devname, 1788 enic->msix[i].devid); 1789 if (err) { 1790 enic_free_intr(enic); 1791 break; 1792 } 1793 enic->msix[i].requested = 1; 1794 } 1795 1796 break; 1797 1798 default: 1799 break; 1800 } 1801 1802 return err; 1803 } 1804 1805 static void enic_synchronize_irqs(struct enic *enic) 1806 { 1807 unsigned int i; 1808 1809 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1810 case VNIC_DEV_INTR_MODE_INTX: 1811 case VNIC_DEV_INTR_MODE_MSI: 1812 synchronize_irq(enic->pdev->irq); 1813 break; 1814 case VNIC_DEV_INTR_MODE_MSIX: 1815 for (i = 0; i < enic->intr_count; i++) 1816 synchronize_irq(enic->msix_entry[i].vector); 1817 break; 1818 default: 1819 break; 1820 } 1821 } 1822 1823 static void enic_set_rx_coal_setting(struct enic *enic) 1824 { 1825 unsigned int speed; 1826 int index = -1; 1827 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting; 1828 1829 /* 1. Read the link speed from fw 1830 * 2. Pick the default range for the speed 1831 * 3. Update it in enic->rx_coalesce_setting 1832 */ 1833 speed = vnic_dev_port_speed(enic->vdev); 1834 if (ENIC_LINK_SPEED_10G < speed) 1835 index = ENIC_LINK_40G_INDEX; 1836 else if (ENIC_LINK_SPEED_4G < speed) 1837 index = ENIC_LINK_10G_INDEX; 1838 else 1839 index = ENIC_LINK_4G_INDEX; 1840 1841 rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start; 1842 rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start; 1843 rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END; 1844 1845 /* Start with the value provided by UCSM */ 1846 for (index = 0; index < enic->rq_count; index++) 1847 enic->cq[index].cur_rx_coal_timeval = 1848 enic->config.intr_timer_usec; 1849 1850 rx_coal->use_adaptive_rx_coalesce = 1; 1851 } 1852 1853 static int enic_dev_notify_set(struct enic *enic) 1854 { 1855 int err; 1856 1857 spin_lock_bh(&enic->devcmd_lock); 1858 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1859 case VNIC_DEV_INTR_MODE_INTX: 1860 err = vnic_dev_notify_set(enic->vdev, 1861 enic_legacy_notify_intr()); 1862 break; 1863 case VNIC_DEV_INTR_MODE_MSIX: 1864 err = vnic_dev_notify_set(enic->vdev, 1865 enic_msix_notify_intr(enic)); 1866 break; 1867 default: 1868 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */); 1869 break; 1870 } 1871 spin_unlock_bh(&enic->devcmd_lock); 1872 1873 return err; 1874 } 1875 1876 static void enic_notify_timer_start(struct enic *enic) 1877 { 1878 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1879 case VNIC_DEV_INTR_MODE_MSI: 1880 mod_timer(&enic->notify_timer, jiffies); 1881 break; 1882 default: 1883 /* Using intr for notification for INTx/MSI-X */ 1884 break; 1885 } 1886 } 1887 1888 /* rtnl lock is held, process context */ 1889 static int enic_open(struct net_device *netdev) 1890 { 1891 struct enic *enic = netdev_priv(netdev); 1892 unsigned int i; 1893 int err, ret; 1894 1895 err = enic_request_intr(enic); 1896 if (err) { 1897 netdev_err(netdev, "Unable to request irq.\n"); 1898 return err; 1899 } 1900 enic_init_affinity_hint(enic); 1901 enic_set_affinity_hint(enic); 1902 1903 err = enic_dev_notify_set(enic); 1904 if (err) { 1905 netdev_err(netdev, 1906 "Failed to alloc notify buffer, aborting.\n"); 1907 goto err_out_free_intr; 1908 } 1909 1910 for (i = 0; i < enic->rq_count; i++) { 1911 /* enable rq before updating rq desc */ 1912 vnic_rq_enable(&enic->rq[i]); 1913 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf); 1914 /* Need at least one buffer on ring to get going */ 1915 if (vnic_rq_desc_used(&enic->rq[i]) == 0) { 1916 netdev_err(netdev, "Unable to alloc receive buffers\n"); 1917 err = -ENOMEM; 1918 goto err_out_free_rq; 1919 } 1920 } 1921 1922 for (i = 0; i < enic->wq_count; i++) 1923 vnic_wq_enable(&enic->wq[i]); 1924 1925 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic)) 1926 enic_dev_add_station_addr(enic); 1927 1928 enic_set_rx_mode(netdev); 1929 1930 netif_tx_wake_all_queues(netdev); 1931 1932 for (i = 0; i < enic->rq_count; i++) 1933 napi_enable(&enic->napi[i]); 1934 1935 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) 1936 for (i = 0; i < enic->wq_count; i++) 1937 napi_enable(&enic->napi[enic_cq_wq(enic, i)]); 1938 enic_dev_enable(enic); 1939 1940 for (i = 0; i < enic->intr_count; i++) 1941 vnic_intr_unmask(&enic->intr[i]); 1942 1943 enic_notify_timer_start(enic); 1944 enic_rfs_timer_start(enic); 1945 1946 return 0; 1947 1948 err_out_free_rq: 1949 for (i = 0; i < enic->rq_count; i++) { 1950 ret = vnic_rq_disable(&enic->rq[i]); 1951 if (!ret) 1952 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf); 1953 } 1954 enic_dev_notify_unset(enic); 1955 err_out_free_intr: 1956 enic_unset_affinity_hint(enic); 1957 enic_free_intr(enic); 1958 1959 return err; 1960 } 1961 1962 /* rtnl lock is held, process context */ 1963 static int enic_stop(struct net_device *netdev) 1964 { 1965 struct enic *enic = netdev_priv(netdev); 1966 unsigned int i; 1967 int err; 1968 1969 for (i = 0; i < enic->intr_count; i++) { 1970 vnic_intr_mask(&enic->intr[i]); 1971 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */ 1972 } 1973 1974 enic_synchronize_irqs(enic); 1975 1976 del_timer_sync(&enic->notify_timer); 1977 enic_rfs_flw_tbl_free(enic); 1978 1979 enic_dev_disable(enic); 1980 1981 for (i = 0; i < enic->rq_count; i++) 1982 napi_disable(&enic->napi[i]); 1983 1984 netif_carrier_off(netdev); 1985 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) 1986 for (i = 0; i < enic->wq_count; i++) 1987 napi_disable(&enic->napi[enic_cq_wq(enic, i)]); 1988 netif_tx_disable(netdev); 1989 1990 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic)) 1991 enic_dev_del_station_addr(enic); 1992 1993 for (i = 0; i < enic->wq_count; i++) { 1994 err = vnic_wq_disable(&enic->wq[i]); 1995 if (err) 1996 return err; 1997 } 1998 for (i = 0; i < enic->rq_count; i++) { 1999 err = vnic_rq_disable(&enic->rq[i]); 2000 if (err) 2001 return err; 2002 } 2003 2004 enic_dev_notify_unset(enic); 2005 enic_unset_affinity_hint(enic); 2006 enic_free_intr(enic); 2007 2008 for (i = 0; i < enic->wq_count; i++) 2009 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf); 2010 for (i = 0; i < enic->rq_count; i++) 2011 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf); 2012 for (i = 0; i < enic->cq_count; i++) 2013 vnic_cq_clean(&enic->cq[i]); 2014 for (i = 0; i < enic->intr_count; i++) 2015 vnic_intr_clean(&enic->intr[i]); 2016 2017 return 0; 2018 } 2019 2020 static int _enic_change_mtu(struct net_device *netdev, int new_mtu) 2021 { 2022 bool running = netif_running(netdev); 2023 int err = 0; 2024 2025 ASSERT_RTNL(); 2026 if (running) { 2027 err = enic_stop(netdev); 2028 if (err) 2029 return err; 2030 } 2031 2032 netdev->mtu = new_mtu; 2033 2034 if (running) { 2035 err = enic_open(netdev); 2036 if (err) 2037 return err; 2038 } 2039 2040 return 0; 2041 } 2042 2043 static int enic_change_mtu(struct net_device *netdev, int new_mtu) 2044 { 2045 struct enic *enic = netdev_priv(netdev); 2046 2047 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) 2048 return -EOPNOTSUPP; 2049 2050 if (netdev->mtu > enic->port_mtu) 2051 netdev_warn(netdev, 2052 "interface MTU (%d) set higher than port MTU (%d)\n", 2053 netdev->mtu, enic->port_mtu); 2054 2055 return _enic_change_mtu(netdev, new_mtu); 2056 } 2057 2058 static void enic_change_mtu_work(struct work_struct *work) 2059 { 2060 struct enic *enic = container_of(work, struct enic, change_mtu_work); 2061 struct net_device *netdev = enic->netdev; 2062 int new_mtu = vnic_dev_mtu(enic->vdev); 2063 2064 rtnl_lock(); 2065 (void)_enic_change_mtu(netdev, new_mtu); 2066 rtnl_unlock(); 2067 2068 netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu); 2069 } 2070 2071 #ifdef CONFIG_NET_POLL_CONTROLLER 2072 static void enic_poll_controller(struct net_device *netdev) 2073 { 2074 struct enic *enic = netdev_priv(netdev); 2075 struct vnic_dev *vdev = enic->vdev; 2076 unsigned int i, intr; 2077 2078 switch (vnic_dev_get_intr_mode(vdev)) { 2079 case VNIC_DEV_INTR_MODE_MSIX: 2080 for (i = 0; i < enic->rq_count; i++) { 2081 intr = enic_msix_rq_intr(enic, i); 2082 enic_isr_msix(enic->msix_entry[intr].vector, 2083 &enic->napi[i]); 2084 } 2085 2086 for (i = 0; i < enic->wq_count; i++) { 2087 intr = enic_msix_wq_intr(enic, i); 2088 enic_isr_msix(enic->msix_entry[intr].vector, 2089 &enic->napi[enic_cq_wq(enic, i)]); 2090 } 2091 2092 break; 2093 case VNIC_DEV_INTR_MODE_MSI: 2094 enic_isr_msi(enic->pdev->irq, enic); 2095 break; 2096 case VNIC_DEV_INTR_MODE_INTX: 2097 enic_isr_legacy(enic->pdev->irq, netdev); 2098 break; 2099 default: 2100 break; 2101 } 2102 } 2103 #endif 2104 2105 static int enic_dev_wait(struct vnic_dev *vdev, 2106 int (*start)(struct vnic_dev *, int), 2107 int (*finished)(struct vnic_dev *, int *), 2108 int arg) 2109 { 2110 unsigned long time; 2111 int done; 2112 int err; 2113 2114 err = start(vdev, arg); 2115 if (err) 2116 return err; 2117 2118 /* Wait for func to complete...2 seconds max 2119 */ 2120 2121 time = jiffies + (HZ * 2); 2122 do { 2123 2124 err = finished(vdev, &done); 2125 if (err) 2126 return err; 2127 2128 if (done) 2129 return 0; 2130 2131 schedule_timeout_uninterruptible(HZ / 10); 2132 2133 } while (time_after(time, jiffies)); 2134 2135 return -ETIMEDOUT; 2136 } 2137 2138 static int enic_dev_open(struct enic *enic) 2139 { 2140 int err; 2141 u32 flags = CMD_OPENF_IG_DESCCACHE; 2142 2143 err = enic_dev_wait(enic->vdev, vnic_dev_open, 2144 vnic_dev_open_done, flags); 2145 if (err) 2146 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n", 2147 err); 2148 2149 return err; 2150 } 2151 2152 static int enic_dev_soft_reset(struct enic *enic) 2153 { 2154 int err; 2155 2156 err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset, 2157 vnic_dev_soft_reset_done, 0); 2158 if (err) 2159 netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n", 2160 err); 2161 2162 return err; 2163 } 2164 2165 static int enic_dev_hang_reset(struct enic *enic) 2166 { 2167 int err; 2168 2169 err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset, 2170 vnic_dev_hang_reset_done, 0); 2171 if (err) 2172 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n", 2173 err); 2174 2175 return err; 2176 } 2177 2178 int __enic_set_rsskey(struct enic *enic) 2179 { 2180 union vnic_rss_key *rss_key_buf_va; 2181 dma_addr_t rss_key_buf_pa; 2182 int i, kidx, bidx, err; 2183 2184 rss_key_buf_va = dma_alloc_coherent(&enic->pdev->dev, 2185 sizeof(union vnic_rss_key), 2186 &rss_key_buf_pa, GFP_ATOMIC); 2187 if (!rss_key_buf_va) 2188 return -ENOMEM; 2189 2190 for (i = 0; i < ENIC_RSS_LEN; i++) { 2191 kidx = i / ENIC_RSS_BYTES_PER_KEY; 2192 bidx = i % ENIC_RSS_BYTES_PER_KEY; 2193 rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i]; 2194 } 2195 spin_lock_bh(&enic->devcmd_lock); 2196 err = enic_set_rss_key(enic, 2197 rss_key_buf_pa, 2198 sizeof(union vnic_rss_key)); 2199 spin_unlock_bh(&enic->devcmd_lock); 2200 2201 dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_key), 2202 rss_key_buf_va, rss_key_buf_pa); 2203 2204 return err; 2205 } 2206 2207 static int enic_set_rsskey(struct enic *enic) 2208 { 2209 netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN); 2210 2211 return __enic_set_rsskey(enic); 2212 } 2213 2214 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits) 2215 { 2216 dma_addr_t rss_cpu_buf_pa; 2217 union vnic_rss_cpu *rss_cpu_buf_va = NULL; 2218 unsigned int i; 2219 int err; 2220 2221 rss_cpu_buf_va = dma_alloc_coherent(&enic->pdev->dev, 2222 sizeof(union vnic_rss_cpu), 2223 &rss_cpu_buf_pa, GFP_ATOMIC); 2224 if (!rss_cpu_buf_va) 2225 return -ENOMEM; 2226 2227 for (i = 0; i < (1 << rss_hash_bits); i++) 2228 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count; 2229 2230 spin_lock_bh(&enic->devcmd_lock); 2231 err = enic_set_rss_cpu(enic, 2232 rss_cpu_buf_pa, 2233 sizeof(union vnic_rss_cpu)); 2234 spin_unlock_bh(&enic->devcmd_lock); 2235 2236 dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_cpu), 2237 rss_cpu_buf_va, rss_cpu_buf_pa); 2238 2239 return err; 2240 } 2241 2242 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu, 2243 u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable) 2244 { 2245 const u8 tso_ipid_split_en = 0; 2246 const u8 ig_vlan_strip_en = 1; 2247 int err; 2248 2249 /* Enable VLAN tag stripping. 2250 */ 2251 2252 spin_lock_bh(&enic->devcmd_lock); 2253 err = enic_set_nic_cfg(enic, 2254 rss_default_cpu, rss_hash_type, 2255 rss_hash_bits, rss_base_cpu, 2256 rss_enable, tso_ipid_split_en, 2257 ig_vlan_strip_en); 2258 spin_unlock_bh(&enic->devcmd_lock); 2259 2260 return err; 2261 } 2262 2263 static int enic_set_rss_nic_cfg(struct enic *enic) 2264 { 2265 struct device *dev = enic_get_dev(enic); 2266 const u8 rss_default_cpu = 0; 2267 const u8 rss_hash_bits = 7; 2268 const u8 rss_base_cpu = 0; 2269 u8 rss_hash_type; 2270 int res; 2271 u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1); 2272 2273 spin_lock_bh(&enic->devcmd_lock); 2274 res = vnic_dev_capable_rss_hash_type(enic->vdev, &rss_hash_type); 2275 spin_unlock_bh(&enic->devcmd_lock); 2276 if (res) { 2277 /* defaults for old adapters 2278 */ 2279 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 | 2280 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 | 2281 NIC_CFG_RSS_HASH_TYPE_IPV6 | 2282 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6; 2283 } 2284 2285 if (rss_enable) { 2286 if (!enic_set_rsskey(enic)) { 2287 if (enic_set_rsscpu(enic, rss_hash_bits)) { 2288 rss_enable = 0; 2289 dev_warn(dev, "RSS disabled, " 2290 "Failed to set RSS cpu indirection table."); 2291 } 2292 } else { 2293 rss_enable = 0; 2294 dev_warn(dev, "RSS disabled, Failed to set RSS key.\n"); 2295 } 2296 } 2297 2298 return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type, 2299 rss_hash_bits, rss_base_cpu, rss_enable); 2300 } 2301 2302 static void enic_set_api_busy(struct enic *enic, bool busy) 2303 { 2304 spin_lock(&enic->enic_api_lock); 2305 enic->enic_api_busy = busy; 2306 spin_unlock(&enic->enic_api_lock); 2307 } 2308 2309 static void enic_reset(struct work_struct *work) 2310 { 2311 struct enic *enic = container_of(work, struct enic, reset); 2312 2313 if (!netif_running(enic->netdev)) 2314 return; 2315 2316 rtnl_lock(); 2317 2318 /* Stop any activity from infiniband */ 2319 enic_set_api_busy(enic, true); 2320 2321 enic_stop(enic->netdev); 2322 enic_dev_soft_reset(enic); 2323 enic_reset_addr_lists(enic); 2324 enic_init_vnic_resources(enic); 2325 enic_set_rss_nic_cfg(enic); 2326 enic_dev_set_ig_vlan_rewrite_mode(enic); 2327 enic_open(enic->netdev); 2328 2329 /* Allow infiniband to fiddle with the device again */ 2330 enic_set_api_busy(enic, false); 2331 2332 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev); 2333 2334 rtnl_unlock(); 2335 } 2336 2337 static void enic_tx_hang_reset(struct work_struct *work) 2338 { 2339 struct enic *enic = container_of(work, struct enic, tx_hang_reset); 2340 2341 rtnl_lock(); 2342 2343 /* Stop any activity from infiniband */ 2344 enic_set_api_busy(enic, true); 2345 2346 enic_dev_hang_notify(enic); 2347 enic_stop(enic->netdev); 2348 enic_dev_hang_reset(enic); 2349 enic_reset_addr_lists(enic); 2350 enic_init_vnic_resources(enic); 2351 enic_set_rss_nic_cfg(enic); 2352 enic_dev_set_ig_vlan_rewrite_mode(enic); 2353 enic_open(enic->netdev); 2354 2355 /* Allow infiniband to fiddle with the device again */ 2356 enic_set_api_busy(enic, false); 2357 2358 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev); 2359 2360 rtnl_unlock(); 2361 } 2362 2363 static int enic_set_intr_mode(struct enic *enic) 2364 { 2365 unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX); 2366 unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX); 2367 unsigned int i; 2368 2369 /* Set interrupt mode (INTx, MSI, MSI-X) depending 2370 * on system capabilities. 2371 * 2372 * Try MSI-X first 2373 * 2374 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs 2375 * (the second to last INTR is used for WQ/RQ errors) 2376 * (the last INTR is used for notifications) 2377 */ 2378 2379 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2); 2380 for (i = 0; i < n + m + 2; i++) 2381 enic->msix_entry[i].entry = i; 2382 2383 /* Use multiple RQs if RSS is enabled 2384 */ 2385 2386 if (ENIC_SETTING(enic, RSS) && 2387 enic->config.intr_mode < 1 && 2388 enic->rq_count >= n && 2389 enic->wq_count >= m && 2390 enic->cq_count >= n + m && 2391 enic->intr_count >= n + m + 2) { 2392 2393 if (pci_enable_msix_range(enic->pdev, enic->msix_entry, 2394 n + m + 2, n + m + 2) > 0) { 2395 2396 enic->rq_count = n; 2397 enic->wq_count = m; 2398 enic->cq_count = n + m; 2399 enic->intr_count = n + m + 2; 2400 2401 vnic_dev_set_intr_mode(enic->vdev, 2402 VNIC_DEV_INTR_MODE_MSIX); 2403 2404 return 0; 2405 } 2406 } 2407 2408 if (enic->config.intr_mode < 1 && 2409 enic->rq_count >= 1 && 2410 enic->wq_count >= m && 2411 enic->cq_count >= 1 + m && 2412 enic->intr_count >= 1 + m + 2) { 2413 if (pci_enable_msix_range(enic->pdev, enic->msix_entry, 2414 1 + m + 2, 1 + m + 2) > 0) { 2415 2416 enic->rq_count = 1; 2417 enic->wq_count = m; 2418 enic->cq_count = 1 + m; 2419 enic->intr_count = 1 + m + 2; 2420 2421 vnic_dev_set_intr_mode(enic->vdev, 2422 VNIC_DEV_INTR_MODE_MSIX); 2423 2424 return 0; 2425 } 2426 } 2427 2428 /* Next try MSI 2429 * 2430 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR 2431 */ 2432 2433 if (enic->config.intr_mode < 2 && 2434 enic->rq_count >= 1 && 2435 enic->wq_count >= 1 && 2436 enic->cq_count >= 2 && 2437 enic->intr_count >= 1 && 2438 !pci_enable_msi(enic->pdev)) { 2439 2440 enic->rq_count = 1; 2441 enic->wq_count = 1; 2442 enic->cq_count = 2; 2443 enic->intr_count = 1; 2444 2445 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI); 2446 2447 return 0; 2448 } 2449 2450 /* Next try INTx 2451 * 2452 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs 2453 * (the first INTR is used for WQ/RQ) 2454 * (the second INTR is used for WQ/RQ errors) 2455 * (the last INTR is used for notifications) 2456 */ 2457 2458 if (enic->config.intr_mode < 3 && 2459 enic->rq_count >= 1 && 2460 enic->wq_count >= 1 && 2461 enic->cq_count >= 2 && 2462 enic->intr_count >= 3) { 2463 2464 enic->rq_count = 1; 2465 enic->wq_count = 1; 2466 enic->cq_count = 2; 2467 enic->intr_count = 3; 2468 2469 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX); 2470 2471 return 0; 2472 } 2473 2474 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN); 2475 2476 return -EINVAL; 2477 } 2478 2479 static void enic_clear_intr_mode(struct enic *enic) 2480 { 2481 switch (vnic_dev_get_intr_mode(enic->vdev)) { 2482 case VNIC_DEV_INTR_MODE_MSIX: 2483 pci_disable_msix(enic->pdev); 2484 break; 2485 case VNIC_DEV_INTR_MODE_MSI: 2486 pci_disable_msi(enic->pdev); 2487 break; 2488 default: 2489 break; 2490 } 2491 2492 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN); 2493 } 2494 2495 static const struct net_device_ops enic_netdev_dynamic_ops = { 2496 .ndo_open = enic_open, 2497 .ndo_stop = enic_stop, 2498 .ndo_start_xmit = enic_hard_start_xmit, 2499 .ndo_get_stats64 = enic_get_stats, 2500 .ndo_validate_addr = eth_validate_addr, 2501 .ndo_set_rx_mode = enic_set_rx_mode, 2502 .ndo_set_mac_address = enic_set_mac_address_dynamic, 2503 .ndo_change_mtu = enic_change_mtu, 2504 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid, 2505 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid, 2506 .ndo_tx_timeout = enic_tx_timeout, 2507 .ndo_set_vf_port = enic_set_vf_port, 2508 .ndo_get_vf_port = enic_get_vf_port, 2509 .ndo_set_vf_mac = enic_set_vf_mac, 2510 #ifdef CONFIG_NET_POLL_CONTROLLER 2511 .ndo_poll_controller = enic_poll_controller, 2512 #endif 2513 #ifdef CONFIG_RFS_ACCEL 2514 .ndo_rx_flow_steer = enic_rx_flow_steer, 2515 #endif 2516 .ndo_features_check = enic_features_check, 2517 }; 2518 2519 static const struct net_device_ops enic_netdev_ops = { 2520 .ndo_open = enic_open, 2521 .ndo_stop = enic_stop, 2522 .ndo_start_xmit = enic_hard_start_xmit, 2523 .ndo_get_stats64 = enic_get_stats, 2524 .ndo_validate_addr = eth_validate_addr, 2525 .ndo_set_mac_address = enic_set_mac_address, 2526 .ndo_set_rx_mode = enic_set_rx_mode, 2527 .ndo_change_mtu = enic_change_mtu, 2528 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid, 2529 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid, 2530 .ndo_tx_timeout = enic_tx_timeout, 2531 .ndo_set_vf_port = enic_set_vf_port, 2532 .ndo_get_vf_port = enic_get_vf_port, 2533 .ndo_set_vf_mac = enic_set_vf_mac, 2534 #ifdef CONFIG_NET_POLL_CONTROLLER 2535 .ndo_poll_controller = enic_poll_controller, 2536 #endif 2537 #ifdef CONFIG_RFS_ACCEL 2538 .ndo_rx_flow_steer = enic_rx_flow_steer, 2539 #endif 2540 .ndo_features_check = enic_features_check, 2541 }; 2542 2543 static void enic_dev_deinit(struct enic *enic) 2544 { 2545 unsigned int i; 2546 2547 for (i = 0; i < enic->rq_count; i++) 2548 __netif_napi_del(&enic->napi[i]); 2549 2550 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) 2551 for (i = 0; i < enic->wq_count; i++) 2552 __netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]); 2553 2554 /* observe RCU grace period after __netif_napi_del() calls */ 2555 synchronize_net(); 2556 2557 enic_free_vnic_resources(enic); 2558 enic_clear_intr_mode(enic); 2559 enic_free_affinity_hint(enic); 2560 } 2561 2562 static void enic_kdump_kernel_config(struct enic *enic) 2563 { 2564 if (is_kdump_kernel()) { 2565 dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n"); 2566 enic->rq_count = 1; 2567 enic->wq_count = 1; 2568 enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS; 2569 enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS; 2570 enic->config.mtu = min_t(u16, 1500, enic->config.mtu); 2571 } 2572 } 2573 2574 static int enic_dev_init(struct enic *enic) 2575 { 2576 struct device *dev = enic_get_dev(enic); 2577 struct net_device *netdev = enic->netdev; 2578 unsigned int i; 2579 int err; 2580 2581 /* Get interrupt coalesce timer info */ 2582 err = enic_dev_intr_coal_timer_info(enic); 2583 if (err) { 2584 dev_warn(dev, "Using default conversion factor for " 2585 "interrupt coalesce timer\n"); 2586 vnic_dev_intr_coal_timer_info_default(enic->vdev); 2587 } 2588 2589 /* Get vNIC configuration 2590 */ 2591 2592 err = enic_get_vnic_config(enic); 2593 if (err) { 2594 dev_err(dev, "Get vNIC configuration failed, aborting\n"); 2595 return err; 2596 } 2597 2598 /* Get available resource counts 2599 */ 2600 2601 enic_get_res_counts(enic); 2602 2603 /* modify resource count if we are in kdump_kernel 2604 */ 2605 enic_kdump_kernel_config(enic); 2606 2607 /* Set interrupt mode based on resource counts and system 2608 * capabilities 2609 */ 2610 2611 err = enic_set_intr_mode(enic); 2612 if (err) { 2613 dev_err(dev, "Failed to set intr mode based on resource " 2614 "counts and system capabilities, aborting\n"); 2615 return err; 2616 } 2617 2618 /* Allocate and configure vNIC resources 2619 */ 2620 2621 err = enic_alloc_vnic_resources(enic); 2622 if (err) { 2623 dev_err(dev, "Failed to alloc vNIC resources, aborting\n"); 2624 goto err_out_free_vnic_resources; 2625 } 2626 2627 enic_init_vnic_resources(enic); 2628 2629 err = enic_set_rss_nic_cfg(enic); 2630 if (err) { 2631 dev_err(dev, "Failed to config nic, aborting\n"); 2632 goto err_out_free_vnic_resources; 2633 } 2634 2635 switch (vnic_dev_get_intr_mode(enic->vdev)) { 2636 default: 2637 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64); 2638 break; 2639 case VNIC_DEV_INTR_MODE_MSIX: 2640 for (i = 0; i < enic->rq_count; i++) { 2641 netif_napi_add(netdev, &enic->napi[i], 2642 enic_poll_msix_rq, NAPI_POLL_WEIGHT); 2643 } 2644 for (i = 0; i < enic->wq_count; i++) 2645 netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)], 2646 enic_poll_msix_wq, NAPI_POLL_WEIGHT); 2647 break; 2648 } 2649 2650 return 0; 2651 2652 err_out_free_vnic_resources: 2653 enic_free_affinity_hint(enic); 2654 enic_clear_intr_mode(enic); 2655 enic_free_vnic_resources(enic); 2656 2657 return err; 2658 } 2659 2660 static void enic_iounmap(struct enic *enic) 2661 { 2662 unsigned int i; 2663 2664 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) 2665 if (enic->bar[i].vaddr) 2666 iounmap(enic->bar[i].vaddr); 2667 } 2668 2669 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 2670 { 2671 struct device *dev = &pdev->dev; 2672 struct net_device *netdev; 2673 struct enic *enic; 2674 int using_dac = 0; 2675 unsigned int i; 2676 int err; 2677 #ifdef CONFIG_PCI_IOV 2678 int pos = 0; 2679 #endif 2680 int num_pps = 1; 2681 2682 /* Allocate net device structure and initialize. Private 2683 * instance data is initialized to zero. 2684 */ 2685 2686 netdev = alloc_etherdev_mqs(sizeof(struct enic), 2687 ENIC_RQ_MAX, ENIC_WQ_MAX); 2688 if (!netdev) 2689 return -ENOMEM; 2690 2691 pci_set_drvdata(pdev, netdev); 2692 2693 SET_NETDEV_DEV(netdev, &pdev->dev); 2694 2695 enic = netdev_priv(netdev); 2696 enic->netdev = netdev; 2697 enic->pdev = pdev; 2698 2699 /* Setup PCI resources 2700 */ 2701 2702 err = pci_enable_device_mem(pdev); 2703 if (err) { 2704 dev_err(dev, "Cannot enable PCI device, aborting\n"); 2705 goto err_out_free_netdev; 2706 } 2707 2708 err = pci_request_regions(pdev, DRV_NAME); 2709 if (err) { 2710 dev_err(dev, "Cannot request PCI regions, aborting\n"); 2711 goto err_out_disable_device; 2712 } 2713 2714 pci_set_master(pdev); 2715 2716 /* Query PCI controller on system for DMA addressing 2717 * limitation for the device. Try 47-bit first, and 2718 * fail to 32-bit. 2719 */ 2720 2721 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(47)); 2722 if (err) { 2723 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); 2724 if (err) { 2725 dev_err(dev, "No usable DMA configuration, aborting\n"); 2726 goto err_out_release_regions; 2727 } 2728 } else { 2729 using_dac = 1; 2730 } 2731 2732 /* Map vNIC resources from BAR0-5 2733 */ 2734 2735 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) { 2736 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM)) 2737 continue; 2738 enic->bar[i].len = pci_resource_len(pdev, i); 2739 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len); 2740 if (!enic->bar[i].vaddr) { 2741 dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i); 2742 err = -ENODEV; 2743 goto err_out_iounmap; 2744 } 2745 enic->bar[i].bus_addr = pci_resource_start(pdev, i); 2746 } 2747 2748 /* Register vNIC device 2749 */ 2750 2751 enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar, 2752 ARRAY_SIZE(enic->bar)); 2753 if (!enic->vdev) { 2754 dev_err(dev, "vNIC registration failed, aborting\n"); 2755 err = -ENODEV; 2756 goto err_out_iounmap; 2757 } 2758 2759 err = vnic_devcmd_init(enic->vdev); 2760 2761 if (err) 2762 goto err_out_vnic_unregister; 2763 2764 #ifdef CONFIG_PCI_IOV 2765 /* Get number of subvnics */ 2766 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV); 2767 if (pos) { 2768 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF, 2769 &enic->num_vfs); 2770 if (enic->num_vfs) { 2771 err = pci_enable_sriov(pdev, enic->num_vfs); 2772 if (err) { 2773 dev_err(dev, "SRIOV enable failed, aborting." 2774 " pci_enable_sriov() returned %d\n", 2775 err); 2776 goto err_out_vnic_unregister; 2777 } 2778 enic->priv_flags |= ENIC_SRIOV_ENABLED; 2779 num_pps = enic->num_vfs; 2780 } 2781 } 2782 #endif 2783 2784 /* Allocate structure for port profiles */ 2785 enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL); 2786 if (!enic->pp) { 2787 err = -ENOMEM; 2788 goto err_out_disable_sriov_pp; 2789 } 2790 2791 /* Issue device open to get device in known state 2792 */ 2793 2794 err = enic_dev_open(enic); 2795 if (err) { 2796 dev_err(dev, "vNIC dev open failed, aborting\n"); 2797 goto err_out_disable_sriov; 2798 } 2799 2800 /* Setup devcmd lock 2801 */ 2802 2803 spin_lock_init(&enic->devcmd_lock); 2804 spin_lock_init(&enic->enic_api_lock); 2805 2806 /* 2807 * Set ingress vlan rewrite mode before vnic initialization 2808 */ 2809 2810 err = enic_dev_set_ig_vlan_rewrite_mode(enic); 2811 if (err) { 2812 dev_err(dev, 2813 "Failed to set ingress vlan rewrite mode, aborting.\n"); 2814 goto err_out_dev_close; 2815 } 2816 2817 /* Issue device init to initialize the vnic-to-switch link. 2818 * We'll start with carrier off and wait for link UP 2819 * notification later to turn on carrier. We don't need 2820 * to wait here for the vnic-to-switch link initialization 2821 * to complete; link UP notification is the indication that 2822 * the process is complete. 2823 */ 2824 2825 netif_carrier_off(netdev); 2826 2827 /* Do not call dev_init for a dynamic vnic. 2828 * For a dynamic vnic, init_prov_info will be 2829 * called later by an upper layer. 2830 */ 2831 2832 if (!enic_is_dynamic(enic)) { 2833 err = vnic_dev_init(enic->vdev, 0); 2834 if (err) { 2835 dev_err(dev, "vNIC dev init failed, aborting\n"); 2836 goto err_out_dev_close; 2837 } 2838 } 2839 2840 err = enic_dev_init(enic); 2841 if (err) { 2842 dev_err(dev, "Device initialization failed, aborting\n"); 2843 goto err_out_dev_close; 2844 } 2845 2846 netif_set_real_num_tx_queues(netdev, enic->wq_count); 2847 netif_set_real_num_rx_queues(netdev, enic->rq_count); 2848 2849 /* Setup notification timer, HW reset task, and wq locks 2850 */ 2851 2852 timer_setup(&enic->notify_timer, enic_notify_timer, 0); 2853 2854 enic_rfs_flw_tbl_init(enic); 2855 enic_set_rx_coal_setting(enic); 2856 INIT_WORK(&enic->reset, enic_reset); 2857 INIT_WORK(&enic->tx_hang_reset, enic_tx_hang_reset); 2858 INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work); 2859 2860 for (i = 0; i < enic->wq_count; i++) 2861 spin_lock_init(&enic->wq_lock[i]); 2862 2863 /* Register net device 2864 */ 2865 2866 enic->port_mtu = enic->config.mtu; 2867 2868 err = enic_set_mac_addr(netdev, enic->mac_addr); 2869 if (err) { 2870 dev_err(dev, "Invalid MAC address, aborting\n"); 2871 goto err_out_dev_deinit; 2872 } 2873 2874 enic->tx_coalesce_usecs = enic->config.intr_timer_usec; 2875 /* rx coalesce time already got initialized. This gets used 2876 * if adaptive coal is turned off 2877 */ 2878 enic->rx_coalesce_usecs = enic->tx_coalesce_usecs; 2879 2880 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) 2881 netdev->netdev_ops = &enic_netdev_dynamic_ops; 2882 else 2883 netdev->netdev_ops = &enic_netdev_ops; 2884 2885 netdev->watchdog_timeo = 2 * HZ; 2886 enic_set_ethtool_ops(netdev); 2887 2888 netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX; 2889 if (ENIC_SETTING(enic, LOOP)) { 2890 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX; 2891 enic->loop_enable = 1; 2892 enic->loop_tag = enic->config.loop_tag; 2893 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag); 2894 } 2895 if (ENIC_SETTING(enic, TXCSUM)) 2896 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM; 2897 if (ENIC_SETTING(enic, TSO)) 2898 netdev->hw_features |= NETIF_F_TSO | 2899 NETIF_F_TSO6 | NETIF_F_TSO_ECN; 2900 if (ENIC_SETTING(enic, RSS)) 2901 netdev->hw_features |= NETIF_F_RXHASH; 2902 if (ENIC_SETTING(enic, RXCSUM)) 2903 netdev->hw_features |= NETIF_F_RXCSUM; 2904 if (ENIC_SETTING(enic, VXLAN)) { 2905 u64 patch_level; 2906 u64 a1 = 0; 2907 2908 netdev->hw_enc_features |= NETIF_F_RXCSUM | 2909 NETIF_F_TSO | 2910 NETIF_F_TSO6 | 2911 NETIF_F_TSO_ECN | 2912 NETIF_F_GSO_UDP_TUNNEL | 2913 NETIF_F_HW_CSUM | 2914 NETIF_F_GSO_UDP_TUNNEL_CSUM; 2915 netdev->hw_features |= netdev->hw_enc_features; 2916 /* get bit mask from hw about supported offload bit level 2917 * BIT(0) = fw supports patch_level 0 2918 * fcoe bit = encap 2919 * fcoe_fc_crc_ok = outer csum ok 2920 * BIT(1) = always set by fw 2921 * BIT(2) = fw supports patch_level 2 2922 * BIT(0) in rss_hash = encap 2923 * BIT(1,2) in rss_hash = outer_ip_csum_ok/ 2924 * outer_tcp_csum_ok 2925 * used in enic_rq_indicate_buf 2926 */ 2927 err = vnic_dev_get_supported_feature_ver(enic->vdev, 2928 VIC_FEATURE_VXLAN, 2929 &patch_level, &a1); 2930 if (err) 2931 patch_level = 0; 2932 enic->vxlan.flags = (u8)a1; 2933 /* mask bits that are supported by driver 2934 */ 2935 patch_level &= BIT_ULL(0) | BIT_ULL(2); 2936 patch_level = fls(patch_level); 2937 patch_level = patch_level ? patch_level - 1 : 0; 2938 enic->vxlan.patch_level = patch_level; 2939 2940 if (vnic_dev_get_res_count(enic->vdev, RES_TYPE_WQ) == 1 || 2941 enic->vxlan.flags & ENIC_VXLAN_MULTI_WQ) { 2942 netdev->udp_tunnel_nic_info = &enic_udp_tunnels_v4; 2943 if (enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6) 2944 netdev->udp_tunnel_nic_info = &enic_udp_tunnels; 2945 } 2946 } 2947 2948 netdev->features |= netdev->hw_features; 2949 netdev->vlan_features |= netdev->features; 2950 2951 #ifdef CONFIG_RFS_ACCEL 2952 netdev->hw_features |= NETIF_F_NTUPLE; 2953 #endif 2954 2955 if (using_dac) 2956 netdev->features |= NETIF_F_HIGHDMA; 2957 2958 netdev->priv_flags |= IFF_UNICAST_FLT; 2959 2960 /* MTU range: 68 - 9000 */ 2961 netdev->min_mtu = ENIC_MIN_MTU; 2962 netdev->max_mtu = ENIC_MAX_MTU; 2963 netdev->mtu = enic->port_mtu; 2964 2965 err = register_netdev(netdev); 2966 if (err) { 2967 dev_err(dev, "Cannot register net device, aborting\n"); 2968 goto err_out_dev_deinit; 2969 } 2970 enic->rx_copybreak = RX_COPYBREAK_DEFAULT; 2971 2972 return 0; 2973 2974 err_out_dev_deinit: 2975 enic_dev_deinit(enic); 2976 err_out_dev_close: 2977 vnic_dev_close(enic->vdev); 2978 err_out_disable_sriov: 2979 kfree(enic->pp); 2980 err_out_disable_sriov_pp: 2981 #ifdef CONFIG_PCI_IOV 2982 if (enic_sriov_enabled(enic)) { 2983 pci_disable_sriov(pdev); 2984 enic->priv_flags &= ~ENIC_SRIOV_ENABLED; 2985 } 2986 #endif 2987 err_out_vnic_unregister: 2988 vnic_dev_unregister(enic->vdev); 2989 err_out_iounmap: 2990 enic_iounmap(enic); 2991 err_out_release_regions: 2992 pci_release_regions(pdev); 2993 err_out_disable_device: 2994 pci_disable_device(pdev); 2995 err_out_free_netdev: 2996 free_netdev(netdev); 2997 2998 return err; 2999 } 3000 3001 static void enic_remove(struct pci_dev *pdev) 3002 { 3003 struct net_device *netdev = pci_get_drvdata(pdev); 3004 3005 if (netdev) { 3006 struct enic *enic = netdev_priv(netdev); 3007 3008 cancel_work_sync(&enic->reset); 3009 cancel_work_sync(&enic->change_mtu_work); 3010 unregister_netdev(netdev); 3011 enic_dev_deinit(enic); 3012 vnic_dev_close(enic->vdev); 3013 #ifdef CONFIG_PCI_IOV 3014 if (enic_sriov_enabled(enic)) { 3015 pci_disable_sriov(pdev); 3016 enic->priv_flags &= ~ENIC_SRIOV_ENABLED; 3017 } 3018 #endif 3019 kfree(enic->pp); 3020 vnic_dev_unregister(enic->vdev); 3021 enic_iounmap(enic); 3022 pci_release_regions(pdev); 3023 pci_disable_device(pdev); 3024 free_netdev(netdev); 3025 } 3026 } 3027 3028 static struct pci_driver enic_driver = { 3029 .name = DRV_NAME, 3030 .id_table = enic_id_table, 3031 .probe = enic_probe, 3032 .remove = enic_remove, 3033 }; 3034 3035 module_pci_driver(enic_driver); 3036