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_set_affinity_hint(enic->msix_entry[i].vector, 154 enic->msix[i].affinity_mask); 155 if (err) 156 netdev_warn(enic->netdev, "irq_set_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_set_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 void 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 } 818 819 /* netif_tx_lock held, process context with BHs disabled, or BH */ 820 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb, 821 struct net_device *netdev) 822 { 823 struct enic *enic = netdev_priv(netdev); 824 struct vnic_wq *wq; 825 unsigned int txq_map; 826 struct netdev_queue *txq; 827 828 if (skb->len <= 0) { 829 dev_kfree_skb_any(skb); 830 return NETDEV_TX_OK; 831 } 832 833 txq_map = skb_get_queue_mapping(skb) % enic->wq_count; 834 wq = &enic->wq[txq_map]; 835 txq = netdev_get_tx_queue(netdev, txq_map); 836 837 /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs, 838 * which is very likely. In the off chance it's going to take 839 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb. 840 */ 841 842 if (skb_shinfo(skb)->gso_size == 0 && 843 skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC && 844 skb_linearize(skb)) { 845 dev_kfree_skb_any(skb); 846 return NETDEV_TX_OK; 847 } 848 849 spin_lock(&enic->wq_lock[txq_map]); 850 851 if (vnic_wq_desc_avail(wq) < 852 skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) { 853 netif_tx_stop_queue(txq); 854 /* This is a hard error, log it */ 855 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n"); 856 spin_unlock(&enic->wq_lock[txq_map]); 857 return NETDEV_TX_BUSY; 858 } 859 860 enic_queue_wq_skb(enic, wq, skb); 861 862 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS) 863 netif_tx_stop_queue(txq); 864 skb_tx_timestamp(skb); 865 if (!netdev_xmit_more() || netif_xmit_stopped(txq)) 866 vnic_wq_doorbell(wq); 867 868 spin_unlock(&enic->wq_lock[txq_map]); 869 870 return NETDEV_TX_OK; 871 } 872 873 /* dev_base_lock rwlock held, nominally process context */ 874 static void enic_get_stats(struct net_device *netdev, 875 struct rtnl_link_stats64 *net_stats) 876 { 877 struct enic *enic = netdev_priv(netdev); 878 struct vnic_stats *stats; 879 int err; 880 881 err = enic_dev_stats_dump(enic, &stats); 882 /* return only when pci_zalloc_consistent fails in vnic_dev_stats_dump 883 * For other failures, like devcmd failure, we return previously 884 * recorded stats. 885 */ 886 if (err == -ENOMEM) 887 return; 888 889 net_stats->tx_packets = stats->tx.tx_frames_ok; 890 net_stats->tx_bytes = stats->tx.tx_bytes_ok; 891 net_stats->tx_errors = stats->tx.tx_errors; 892 net_stats->tx_dropped = stats->tx.tx_drops; 893 894 net_stats->rx_packets = stats->rx.rx_frames_ok; 895 net_stats->rx_bytes = stats->rx.rx_bytes_ok; 896 net_stats->rx_errors = stats->rx.rx_errors; 897 net_stats->multicast = stats->rx.rx_multicast_frames_ok; 898 net_stats->rx_over_errors = enic->rq_truncated_pkts; 899 net_stats->rx_crc_errors = enic->rq_bad_fcs; 900 net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop; 901 } 902 903 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr) 904 { 905 struct enic *enic = netdev_priv(netdev); 906 907 if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) { 908 unsigned int mc_count = netdev_mc_count(netdev); 909 910 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n", 911 ENIC_MULTICAST_PERFECT_FILTERS, mc_count); 912 913 return -ENOSPC; 914 } 915 916 enic_dev_add_addr(enic, mc_addr); 917 enic->mc_count++; 918 919 return 0; 920 } 921 922 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr) 923 { 924 struct enic *enic = netdev_priv(netdev); 925 926 enic_dev_del_addr(enic, mc_addr); 927 enic->mc_count--; 928 929 return 0; 930 } 931 932 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr) 933 { 934 struct enic *enic = netdev_priv(netdev); 935 936 if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) { 937 unsigned int uc_count = netdev_uc_count(netdev); 938 939 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n", 940 ENIC_UNICAST_PERFECT_FILTERS, uc_count); 941 942 return -ENOSPC; 943 } 944 945 enic_dev_add_addr(enic, uc_addr); 946 enic->uc_count++; 947 948 return 0; 949 } 950 951 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr) 952 { 953 struct enic *enic = netdev_priv(netdev); 954 955 enic_dev_del_addr(enic, uc_addr); 956 enic->uc_count--; 957 958 return 0; 959 } 960 961 void enic_reset_addr_lists(struct enic *enic) 962 { 963 struct net_device *netdev = enic->netdev; 964 965 __dev_uc_unsync(netdev, NULL); 966 __dev_mc_unsync(netdev, NULL); 967 968 enic->mc_count = 0; 969 enic->uc_count = 0; 970 enic->flags = 0; 971 } 972 973 static int enic_set_mac_addr(struct net_device *netdev, char *addr) 974 { 975 struct enic *enic = netdev_priv(netdev); 976 977 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) { 978 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr)) 979 return -EADDRNOTAVAIL; 980 } else { 981 if (!is_valid_ether_addr(addr)) 982 return -EADDRNOTAVAIL; 983 } 984 985 memcpy(netdev->dev_addr, addr, netdev->addr_len); 986 987 return 0; 988 } 989 990 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p) 991 { 992 struct enic *enic = netdev_priv(netdev); 993 struct sockaddr *saddr = p; 994 char *addr = saddr->sa_data; 995 int err; 996 997 if (netif_running(enic->netdev)) { 998 err = enic_dev_del_station_addr(enic); 999 if (err) 1000 return err; 1001 } 1002 1003 err = enic_set_mac_addr(netdev, addr); 1004 if (err) 1005 return err; 1006 1007 if (netif_running(enic->netdev)) { 1008 err = enic_dev_add_station_addr(enic); 1009 if (err) 1010 return err; 1011 } 1012 1013 return err; 1014 } 1015 1016 static int enic_set_mac_address(struct net_device *netdev, void *p) 1017 { 1018 struct sockaddr *saddr = p; 1019 char *addr = saddr->sa_data; 1020 struct enic *enic = netdev_priv(netdev); 1021 int err; 1022 1023 err = enic_dev_del_station_addr(enic); 1024 if (err) 1025 return err; 1026 1027 err = enic_set_mac_addr(netdev, addr); 1028 if (err) 1029 return err; 1030 1031 return enic_dev_add_station_addr(enic); 1032 } 1033 1034 /* netif_tx_lock held, BHs disabled */ 1035 static void enic_set_rx_mode(struct net_device *netdev) 1036 { 1037 struct enic *enic = netdev_priv(netdev); 1038 int directed = 1; 1039 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0; 1040 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0; 1041 int promisc = (netdev->flags & IFF_PROMISC) || 1042 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS; 1043 int allmulti = (netdev->flags & IFF_ALLMULTI) || 1044 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS; 1045 unsigned int flags = netdev->flags | 1046 (allmulti ? IFF_ALLMULTI : 0) | 1047 (promisc ? IFF_PROMISC : 0); 1048 1049 if (enic->flags != flags) { 1050 enic->flags = flags; 1051 enic_dev_packet_filter(enic, directed, 1052 multicast, broadcast, promisc, allmulti); 1053 } 1054 1055 if (!promisc) { 1056 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync); 1057 if (!allmulti) 1058 __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync); 1059 } 1060 } 1061 1062 /* netif_tx_lock held, BHs disabled */ 1063 static void enic_tx_timeout(struct net_device *netdev, unsigned int txqueue) 1064 { 1065 struct enic *enic = netdev_priv(netdev); 1066 schedule_work(&enic->tx_hang_reset); 1067 } 1068 1069 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac) 1070 { 1071 struct enic *enic = netdev_priv(netdev); 1072 struct enic_port_profile *pp; 1073 int err; 1074 1075 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1076 if (err) 1077 return err; 1078 1079 if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) { 1080 if (vf == PORT_SELF_VF) { 1081 memcpy(pp->vf_mac, mac, ETH_ALEN); 1082 return 0; 1083 } else { 1084 /* 1085 * For sriov vf's set the mac in hw 1086 */ 1087 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic, 1088 vnic_dev_set_mac_addr, mac); 1089 return enic_dev_status_to_errno(err); 1090 } 1091 } else 1092 return -EINVAL; 1093 } 1094 1095 static int enic_set_vf_port(struct net_device *netdev, int vf, 1096 struct nlattr *port[]) 1097 { 1098 struct enic *enic = netdev_priv(netdev); 1099 struct enic_port_profile prev_pp; 1100 struct enic_port_profile *pp; 1101 int err = 0, restore_pp = 1; 1102 1103 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1104 if (err) 1105 return err; 1106 1107 if (!port[IFLA_PORT_REQUEST]) 1108 return -EOPNOTSUPP; 1109 1110 memcpy(&prev_pp, pp, sizeof(*enic->pp)); 1111 memset(pp, 0, sizeof(*enic->pp)); 1112 1113 pp->set |= ENIC_SET_REQUEST; 1114 pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]); 1115 1116 if (port[IFLA_PORT_PROFILE]) { 1117 pp->set |= ENIC_SET_NAME; 1118 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]), 1119 PORT_PROFILE_MAX); 1120 } 1121 1122 if (port[IFLA_PORT_INSTANCE_UUID]) { 1123 pp->set |= ENIC_SET_INSTANCE; 1124 memcpy(pp->instance_uuid, 1125 nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX); 1126 } 1127 1128 if (port[IFLA_PORT_HOST_UUID]) { 1129 pp->set |= ENIC_SET_HOST; 1130 memcpy(pp->host_uuid, 1131 nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX); 1132 } 1133 1134 if (vf == PORT_SELF_VF) { 1135 /* Special case handling: mac came from IFLA_VF_MAC */ 1136 if (!is_zero_ether_addr(prev_pp.vf_mac)) 1137 memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN); 1138 1139 if (is_zero_ether_addr(netdev->dev_addr)) 1140 eth_hw_addr_random(netdev); 1141 } else { 1142 /* SR-IOV VF: get mac from adapter */ 1143 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic, 1144 vnic_dev_get_mac_addr, pp->mac_addr); 1145 if (err) { 1146 netdev_err(netdev, "Error getting mac for vf %d\n", vf); 1147 memcpy(pp, &prev_pp, sizeof(*pp)); 1148 return enic_dev_status_to_errno(err); 1149 } 1150 } 1151 1152 err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp); 1153 if (err) { 1154 if (restore_pp) { 1155 /* Things are still the way they were: Implicit 1156 * DISASSOCIATE failed 1157 */ 1158 memcpy(pp, &prev_pp, sizeof(*pp)); 1159 } else { 1160 memset(pp, 0, sizeof(*pp)); 1161 if (vf == PORT_SELF_VF) 1162 eth_zero_addr(netdev->dev_addr); 1163 } 1164 } else { 1165 /* Set flag to indicate that the port assoc/disassoc 1166 * request has been sent out to fw 1167 */ 1168 pp->set |= ENIC_PORT_REQUEST_APPLIED; 1169 1170 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */ 1171 if (pp->request == PORT_REQUEST_DISASSOCIATE) { 1172 eth_zero_addr(pp->mac_addr); 1173 if (vf == PORT_SELF_VF) 1174 eth_zero_addr(netdev->dev_addr); 1175 } 1176 } 1177 1178 if (vf == PORT_SELF_VF) 1179 eth_zero_addr(pp->vf_mac); 1180 1181 return err; 1182 } 1183 1184 static int enic_get_vf_port(struct net_device *netdev, int vf, 1185 struct sk_buff *skb) 1186 { 1187 struct enic *enic = netdev_priv(netdev); 1188 u16 response = PORT_PROFILE_RESPONSE_SUCCESS; 1189 struct enic_port_profile *pp; 1190 int err; 1191 1192 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1193 if (err) 1194 return err; 1195 1196 if (!(pp->set & ENIC_PORT_REQUEST_APPLIED)) 1197 return -ENODATA; 1198 1199 err = enic_process_get_pp_request(enic, vf, pp->request, &response); 1200 if (err) 1201 return err; 1202 1203 if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) || 1204 nla_put_u16(skb, IFLA_PORT_RESPONSE, response) || 1205 ((pp->set & ENIC_SET_NAME) && 1206 nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) || 1207 ((pp->set & ENIC_SET_INSTANCE) && 1208 nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX, 1209 pp->instance_uuid)) || 1210 ((pp->set & ENIC_SET_HOST) && 1211 nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid))) 1212 goto nla_put_failure; 1213 return 0; 1214 1215 nla_put_failure: 1216 return -EMSGSIZE; 1217 } 1218 1219 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf) 1220 { 1221 struct enic *enic = vnic_dev_priv(rq->vdev); 1222 1223 if (!buf->os_buf) 1224 return; 1225 1226 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len, 1227 DMA_FROM_DEVICE); 1228 dev_kfree_skb_any(buf->os_buf); 1229 buf->os_buf = NULL; 1230 } 1231 1232 static int enic_rq_alloc_buf(struct vnic_rq *rq) 1233 { 1234 struct enic *enic = vnic_dev_priv(rq->vdev); 1235 struct net_device *netdev = enic->netdev; 1236 struct sk_buff *skb; 1237 unsigned int len = netdev->mtu + VLAN_ETH_HLEN; 1238 unsigned int os_buf_index = 0; 1239 dma_addr_t dma_addr; 1240 struct vnic_rq_buf *buf = rq->to_use; 1241 1242 if (buf->os_buf) { 1243 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr, 1244 buf->len); 1245 1246 return 0; 1247 } 1248 skb = netdev_alloc_skb_ip_align(netdev, len); 1249 if (!skb) 1250 return -ENOMEM; 1251 1252 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, len, 1253 DMA_FROM_DEVICE); 1254 if (unlikely(enic_dma_map_check(enic, dma_addr))) { 1255 dev_kfree_skb(skb); 1256 return -ENOMEM; 1257 } 1258 1259 enic_queue_rq_desc(rq, skb, os_buf_index, 1260 dma_addr, len); 1261 1262 return 0; 1263 } 1264 1265 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size, 1266 u32 pkt_len) 1267 { 1268 if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len) 1269 pkt_size->large_pkt_bytes_cnt += pkt_len; 1270 else 1271 pkt_size->small_pkt_bytes_cnt += pkt_len; 1272 } 1273 1274 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb, 1275 struct vnic_rq_buf *buf, u16 len) 1276 { 1277 struct enic *enic = netdev_priv(netdev); 1278 struct sk_buff *new_skb; 1279 1280 if (len > enic->rx_copybreak) 1281 return false; 1282 new_skb = netdev_alloc_skb_ip_align(netdev, len); 1283 if (!new_skb) 1284 return false; 1285 dma_sync_single_for_cpu(&enic->pdev->dev, buf->dma_addr, len, 1286 DMA_FROM_DEVICE); 1287 memcpy(new_skb->data, (*skb)->data, len); 1288 *skb = new_skb; 1289 1290 return true; 1291 } 1292 1293 static void enic_rq_indicate_buf(struct vnic_rq *rq, 1294 struct cq_desc *cq_desc, struct vnic_rq_buf *buf, 1295 int skipped, void *opaque) 1296 { 1297 struct enic *enic = vnic_dev_priv(rq->vdev); 1298 struct net_device *netdev = enic->netdev; 1299 struct sk_buff *skb; 1300 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1301 1302 u8 type, color, eop, sop, ingress_port, vlan_stripped; 1303 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof; 1304 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok; 1305 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc; 1306 u8 packet_error; 1307 u16 q_number, completed_index, bytes_written, vlan_tci, checksum; 1308 u32 rss_hash; 1309 bool outer_csum_ok = true, encap = false; 1310 1311 if (skipped) 1312 return; 1313 1314 skb = buf->os_buf; 1315 1316 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc, 1317 &type, &color, &q_number, &completed_index, 1318 &ingress_port, &fcoe, &eop, &sop, &rss_type, 1319 &csum_not_calc, &rss_hash, &bytes_written, 1320 &packet_error, &vlan_stripped, &vlan_tci, &checksum, 1321 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error, 1322 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp, 1323 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment, 1324 &fcs_ok); 1325 1326 if (packet_error) { 1327 1328 if (!fcs_ok) { 1329 if (bytes_written > 0) 1330 enic->rq_bad_fcs++; 1331 else if (bytes_written == 0) 1332 enic->rq_truncated_pkts++; 1333 } 1334 1335 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len, 1336 DMA_FROM_DEVICE); 1337 dev_kfree_skb_any(skb); 1338 buf->os_buf = NULL; 1339 1340 return; 1341 } 1342 1343 if (eop && bytes_written > 0) { 1344 1345 /* Good receive 1346 */ 1347 1348 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) { 1349 buf->os_buf = NULL; 1350 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, 1351 buf->len, DMA_FROM_DEVICE); 1352 } 1353 prefetch(skb->data - NET_IP_ALIGN); 1354 1355 skb_put(skb, bytes_written); 1356 skb->protocol = eth_type_trans(skb, netdev); 1357 skb_record_rx_queue(skb, q_number); 1358 if ((netdev->features & NETIF_F_RXHASH) && rss_hash && 1359 (type == 3)) { 1360 switch (rss_type) { 1361 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4: 1362 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6: 1363 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX: 1364 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4); 1365 break; 1366 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4: 1367 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6: 1368 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX: 1369 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3); 1370 break; 1371 } 1372 } 1373 if (enic->vxlan.vxlan_udp_port_number) { 1374 switch (enic->vxlan.patch_level) { 1375 case 0: 1376 if (fcoe) { 1377 encap = true; 1378 outer_csum_ok = fcoe_fc_crc_ok; 1379 } 1380 break; 1381 case 2: 1382 if ((type == 7) && 1383 (rss_hash & BIT(0))) { 1384 encap = true; 1385 outer_csum_ok = (rss_hash & BIT(1)) && 1386 (rss_hash & BIT(2)); 1387 } 1388 break; 1389 } 1390 } 1391 1392 /* Hardware does not provide whole packet checksum. It only 1393 * provides pseudo checksum. Since hw validates the packet 1394 * checksum but not provide us the checksum value. use 1395 * CHECSUM_UNNECESSARY. 1396 * 1397 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is 1398 * inner csum_ok. outer_csum_ok is set by hw when outer udp 1399 * csum is correct or is zero. 1400 */ 1401 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc && 1402 tcp_udp_csum_ok && outer_csum_ok && 1403 (ipv4_csum_ok || ipv6)) { 1404 skb->ip_summed = CHECKSUM_UNNECESSARY; 1405 skb->csum_level = encap; 1406 } 1407 1408 if (vlan_stripped) 1409 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci); 1410 1411 skb_mark_napi_id(skb, &enic->napi[rq->index]); 1412 if (!(netdev->features & NETIF_F_GRO)) 1413 netif_receive_skb(skb); 1414 else 1415 napi_gro_receive(&enic->napi[q_number], skb); 1416 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1417 enic_intr_update_pkt_size(&cq->pkt_size_counter, 1418 bytes_written); 1419 } else { 1420 1421 /* Buffer overflow 1422 */ 1423 1424 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len, 1425 DMA_FROM_DEVICE); 1426 dev_kfree_skb_any(skb); 1427 buf->os_buf = NULL; 1428 } 1429 } 1430 1431 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc, 1432 u8 type, u16 q_number, u16 completed_index, void *opaque) 1433 { 1434 struct enic *enic = vnic_dev_priv(vdev); 1435 1436 vnic_rq_service(&enic->rq[q_number], cq_desc, 1437 completed_index, VNIC_RQ_RETURN_DESC, 1438 enic_rq_indicate_buf, opaque); 1439 1440 return 0; 1441 } 1442 1443 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq) 1444 { 1445 unsigned int intr = enic_msix_rq_intr(enic, rq->index); 1446 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1447 u32 timer = cq->tobe_rx_coal_timeval; 1448 1449 if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) { 1450 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer); 1451 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval; 1452 } 1453 } 1454 1455 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq) 1456 { 1457 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting; 1458 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1459 struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter; 1460 int index; 1461 u32 timer; 1462 u32 range_start; 1463 u32 traffic; 1464 u64 delta; 1465 ktime_t now = ktime_get(); 1466 1467 delta = ktime_us_delta(now, cq->prev_ts); 1468 if (delta < ENIC_AIC_TS_BREAK) 1469 return; 1470 cq->prev_ts = now; 1471 1472 traffic = pkt_size_counter->large_pkt_bytes_cnt + 1473 pkt_size_counter->small_pkt_bytes_cnt; 1474 /* The table takes Mbps 1475 * traffic *= 8 => bits 1476 * traffic *= (10^6 / delta) => bps 1477 * traffic /= 10^6 => Mbps 1478 * 1479 * Combining, traffic *= (8 / delta) 1480 */ 1481 1482 traffic <<= 3; 1483 traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta; 1484 1485 for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++) 1486 if (traffic < mod_table[index].rx_rate) 1487 break; 1488 range_start = (pkt_size_counter->small_pkt_bytes_cnt > 1489 pkt_size_counter->large_pkt_bytes_cnt << 1) ? 1490 rx_coal->small_pkt_range_start : 1491 rx_coal->large_pkt_range_start; 1492 timer = range_start + ((rx_coal->range_end - range_start) * 1493 mod_table[index].range_percent / 100); 1494 /* Damping */ 1495 cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1; 1496 1497 pkt_size_counter->large_pkt_bytes_cnt = 0; 1498 pkt_size_counter->small_pkt_bytes_cnt = 0; 1499 } 1500 1501 static int enic_poll(struct napi_struct *napi, int budget) 1502 { 1503 struct net_device *netdev = napi->dev; 1504 struct enic *enic = netdev_priv(netdev); 1505 unsigned int cq_rq = enic_cq_rq(enic, 0); 1506 unsigned int cq_wq = enic_cq_wq(enic, 0); 1507 unsigned int intr = enic_legacy_io_intr(); 1508 unsigned int rq_work_to_do = budget; 1509 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET; 1510 unsigned int work_done, rq_work_done = 0, wq_work_done; 1511 int err; 1512 1513 wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do, 1514 enic_wq_service, NULL); 1515 1516 if (budget > 0) 1517 rq_work_done = vnic_cq_service(&enic->cq[cq_rq], 1518 rq_work_to_do, enic_rq_service, NULL); 1519 1520 /* Accumulate intr event credits for this polling 1521 * cycle. An intr event is the completion of a 1522 * a WQ or RQ packet. 1523 */ 1524 1525 work_done = rq_work_done + wq_work_done; 1526 1527 if (work_done > 0) 1528 vnic_intr_return_credits(&enic->intr[intr], 1529 work_done, 1530 0 /* don't unmask intr */, 1531 0 /* don't reset intr timer */); 1532 1533 err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf); 1534 1535 /* Buffer allocation failed. Stay in polling 1536 * mode so we can try to fill the ring again. 1537 */ 1538 1539 if (err) 1540 rq_work_done = rq_work_to_do; 1541 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1542 /* Call the function which refreshes the intr coalescing timer 1543 * value based on the traffic. 1544 */ 1545 enic_calc_int_moderation(enic, &enic->rq[0]); 1546 1547 if ((rq_work_done < budget) && napi_complete_done(napi, rq_work_done)) { 1548 1549 /* Some work done, but not enough to stay in polling, 1550 * exit polling 1551 */ 1552 1553 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1554 enic_set_int_moderation(enic, &enic->rq[0]); 1555 vnic_intr_unmask(&enic->intr[intr]); 1556 } 1557 1558 return rq_work_done; 1559 } 1560 1561 #ifdef CONFIG_RFS_ACCEL 1562 static void enic_free_rx_cpu_rmap(struct enic *enic) 1563 { 1564 free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap); 1565 enic->netdev->rx_cpu_rmap = NULL; 1566 } 1567 1568 static void enic_set_rx_cpu_rmap(struct enic *enic) 1569 { 1570 int i, res; 1571 1572 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) { 1573 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count); 1574 if (unlikely(!enic->netdev->rx_cpu_rmap)) 1575 return; 1576 for (i = 0; i < enic->rq_count; i++) { 1577 res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap, 1578 enic->msix_entry[i].vector); 1579 if (unlikely(res)) { 1580 enic_free_rx_cpu_rmap(enic); 1581 return; 1582 } 1583 } 1584 } 1585 } 1586 1587 #else 1588 1589 static void enic_free_rx_cpu_rmap(struct enic *enic) 1590 { 1591 } 1592 1593 static void enic_set_rx_cpu_rmap(struct enic *enic) 1594 { 1595 } 1596 1597 #endif /* CONFIG_RFS_ACCEL */ 1598 1599 static int enic_poll_msix_wq(struct napi_struct *napi, int budget) 1600 { 1601 struct net_device *netdev = napi->dev; 1602 struct enic *enic = netdev_priv(netdev); 1603 unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count; 1604 struct vnic_wq *wq = &enic->wq[wq_index]; 1605 unsigned int cq; 1606 unsigned int intr; 1607 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET; 1608 unsigned int wq_work_done; 1609 unsigned int wq_irq; 1610 1611 wq_irq = wq->index; 1612 cq = enic_cq_wq(enic, wq_irq); 1613 intr = enic_msix_wq_intr(enic, wq_irq); 1614 wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do, 1615 enic_wq_service, NULL); 1616 1617 vnic_intr_return_credits(&enic->intr[intr], wq_work_done, 1618 0 /* don't unmask intr */, 1619 1 /* reset intr timer */); 1620 if (!wq_work_done) { 1621 napi_complete(napi); 1622 vnic_intr_unmask(&enic->intr[intr]); 1623 return 0; 1624 } 1625 1626 return budget; 1627 } 1628 1629 static int enic_poll_msix_rq(struct napi_struct *napi, int budget) 1630 { 1631 struct net_device *netdev = napi->dev; 1632 struct enic *enic = netdev_priv(netdev); 1633 unsigned int rq = (napi - &enic->napi[0]); 1634 unsigned int cq = enic_cq_rq(enic, rq); 1635 unsigned int intr = enic_msix_rq_intr(enic, rq); 1636 unsigned int work_to_do = budget; 1637 unsigned int work_done = 0; 1638 int err; 1639 1640 /* Service RQ 1641 */ 1642 1643 if (budget > 0) 1644 work_done = vnic_cq_service(&enic->cq[cq], 1645 work_to_do, enic_rq_service, NULL); 1646 1647 /* Return intr event credits for this polling 1648 * cycle. An intr event is the completion of a 1649 * RQ packet. 1650 */ 1651 1652 if (work_done > 0) 1653 vnic_intr_return_credits(&enic->intr[intr], 1654 work_done, 1655 0 /* don't unmask intr */, 1656 0 /* don't reset intr timer */); 1657 1658 err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf); 1659 1660 /* Buffer allocation failed. Stay in polling mode 1661 * so we can try to fill the ring again. 1662 */ 1663 1664 if (err) 1665 work_done = work_to_do; 1666 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1667 /* Call the function which refreshes the intr coalescing timer 1668 * value based on the traffic. 1669 */ 1670 enic_calc_int_moderation(enic, &enic->rq[rq]); 1671 1672 if ((work_done < budget) && napi_complete_done(napi, work_done)) { 1673 1674 /* Some work done, but not enough to stay in polling, 1675 * exit polling 1676 */ 1677 1678 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1679 enic_set_int_moderation(enic, &enic->rq[rq]); 1680 vnic_intr_unmask(&enic->intr[intr]); 1681 } 1682 1683 return work_done; 1684 } 1685 1686 static void enic_notify_timer(struct timer_list *t) 1687 { 1688 struct enic *enic = from_timer(enic, t, notify_timer); 1689 1690 enic_notify_check(enic); 1691 1692 mod_timer(&enic->notify_timer, 1693 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD)); 1694 } 1695 1696 static void enic_free_intr(struct enic *enic) 1697 { 1698 struct net_device *netdev = enic->netdev; 1699 unsigned int i; 1700 1701 enic_free_rx_cpu_rmap(enic); 1702 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1703 case VNIC_DEV_INTR_MODE_INTX: 1704 free_irq(enic->pdev->irq, netdev); 1705 break; 1706 case VNIC_DEV_INTR_MODE_MSI: 1707 free_irq(enic->pdev->irq, enic); 1708 break; 1709 case VNIC_DEV_INTR_MODE_MSIX: 1710 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) 1711 if (enic->msix[i].requested) 1712 free_irq(enic->msix_entry[i].vector, 1713 enic->msix[i].devid); 1714 break; 1715 default: 1716 break; 1717 } 1718 } 1719 1720 static int enic_request_intr(struct enic *enic) 1721 { 1722 struct net_device *netdev = enic->netdev; 1723 unsigned int i, intr; 1724 int err = 0; 1725 1726 enic_set_rx_cpu_rmap(enic); 1727 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1728 1729 case VNIC_DEV_INTR_MODE_INTX: 1730 1731 err = request_irq(enic->pdev->irq, enic_isr_legacy, 1732 IRQF_SHARED, netdev->name, netdev); 1733 break; 1734 1735 case VNIC_DEV_INTR_MODE_MSI: 1736 1737 err = request_irq(enic->pdev->irq, enic_isr_msi, 1738 0, netdev->name, enic); 1739 break; 1740 1741 case VNIC_DEV_INTR_MODE_MSIX: 1742 1743 for (i = 0; i < enic->rq_count; i++) { 1744 intr = enic_msix_rq_intr(enic, i); 1745 snprintf(enic->msix[intr].devname, 1746 sizeof(enic->msix[intr].devname), 1747 "%s-rx-%u", netdev->name, i); 1748 enic->msix[intr].isr = enic_isr_msix; 1749 enic->msix[intr].devid = &enic->napi[i]; 1750 } 1751 1752 for (i = 0; i < enic->wq_count; i++) { 1753 int wq = enic_cq_wq(enic, i); 1754 1755 intr = enic_msix_wq_intr(enic, i); 1756 snprintf(enic->msix[intr].devname, 1757 sizeof(enic->msix[intr].devname), 1758 "%s-tx-%u", netdev->name, i); 1759 enic->msix[intr].isr = enic_isr_msix; 1760 enic->msix[intr].devid = &enic->napi[wq]; 1761 } 1762 1763 intr = enic_msix_err_intr(enic); 1764 snprintf(enic->msix[intr].devname, 1765 sizeof(enic->msix[intr].devname), 1766 "%s-err", netdev->name); 1767 enic->msix[intr].isr = enic_isr_msix_err; 1768 enic->msix[intr].devid = enic; 1769 1770 intr = enic_msix_notify_intr(enic); 1771 snprintf(enic->msix[intr].devname, 1772 sizeof(enic->msix[intr].devname), 1773 "%s-notify", netdev->name); 1774 enic->msix[intr].isr = enic_isr_msix_notify; 1775 enic->msix[intr].devid = enic; 1776 1777 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) 1778 enic->msix[i].requested = 0; 1779 1780 for (i = 0; i < enic->intr_count; i++) { 1781 err = request_irq(enic->msix_entry[i].vector, 1782 enic->msix[i].isr, 0, 1783 enic->msix[i].devname, 1784 enic->msix[i].devid); 1785 if (err) { 1786 enic_free_intr(enic); 1787 break; 1788 } 1789 enic->msix[i].requested = 1; 1790 } 1791 1792 break; 1793 1794 default: 1795 break; 1796 } 1797 1798 return err; 1799 } 1800 1801 static void enic_synchronize_irqs(struct enic *enic) 1802 { 1803 unsigned int i; 1804 1805 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1806 case VNIC_DEV_INTR_MODE_INTX: 1807 case VNIC_DEV_INTR_MODE_MSI: 1808 synchronize_irq(enic->pdev->irq); 1809 break; 1810 case VNIC_DEV_INTR_MODE_MSIX: 1811 for (i = 0; i < enic->intr_count; i++) 1812 synchronize_irq(enic->msix_entry[i].vector); 1813 break; 1814 default: 1815 break; 1816 } 1817 } 1818 1819 static void enic_set_rx_coal_setting(struct enic *enic) 1820 { 1821 unsigned int speed; 1822 int index = -1; 1823 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting; 1824 1825 /* 1. Read the link speed from fw 1826 * 2. Pick the default range for the speed 1827 * 3. Update it in enic->rx_coalesce_setting 1828 */ 1829 speed = vnic_dev_port_speed(enic->vdev); 1830 if (ENIC_LINK_SPEED_10G < speed) 1831 index = ENIC_LINK_40G_INDEX; 1832 else if (ENIC_LINK_SPEED_4G < speed) 1833 index = ENIC_LINK_10G_INDEX; 1834 else 1835 index = ENIC_LINK_4G_INDEX; 1836 1837 rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start; 1838 rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start; 1839 rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END; 1840 1841 /* Start with the value provided by UCSM */ 1842 for (index = 0; index < enic->rq_count; index++) 1843 enic->cq[index].cur_rx_coal_timeval = 1844 enic->config.intr_timer_usec; 1845 1846 rx_coal->use_adaptive_rx_coalesce = 1; 1847 } 1848 1849 static int enic_dev_notify_set(struct enic *enic) 1850 { 1851 int err; 1852 1853 spin_lock_bh(&enic->devcmd_lock); 1854 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1855 case VNIC_DEV_INTR_MODE_INTX: 1856 err = vnic_dev_notify_set(enic->vdev, 1857 enic_legacy_notify_intr()); 1858 break; 1859 case VNIC_DEV_INTR_MODE_MSIX: 1860 err = vnic_dev_notify_set(enic->vdev, 1861 enic_msix_notify_intr(enic)); 1862 break; 1863 default: 1864 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */); 1865 break; 1866 } 1867 spin_unlock_bh(&enic->devcmd_lock); 1868 1869 return err; 1870 } 1871 1872 static void enic_notify_timer_start(struct enic *enic) 1873 { 1874 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1875 case VNIC_DEV_INTR_MODE_MSI: 1876 mod_timer(&enic->notify_timer, jiffies); 1877 break; 1878 default: 1879 /* Using intr for notification for INTx/MSI-X */ 1880 break; 1881 } 1882 } 1883 1884 /* rtnl lock is held, process context */ 1885 static int enic_open(struct net_device *netdev) 1886 { 1887 struct enic *enic = netdev_priv(netdev); 1888 unsigned int i; 1889 int err, ret; 1890 1891 err = enic_request_intr(enic); 1892 if (err) { 1893 netdev_err(netdev, "Unable to request irq.\n"); 1894 return err; 1895 } 1896 enic_init_affinity_hint(enic); 1897 enic_set_affinity_hint(enic); 1898 1899 err = enic_dev_notify_set(enic); 1900 if (err) { 1901 netdev_err(netdev, 1902 "Failed to alloc notify buffer, aborting.\n"); 1903 goto err_out_free_intr; 1904 } 1905 1906 for (i = 0; i < enic->rq_count; i++) { 1907 /* enable rq before updating rq desc */ 1908 vnic_rq_enable(&enic->rq[i]); 1909 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf); 1910 /* Need at least one buffer on ring to get going */ 1911 if (vnic_rq_desc_used(&enic->rq[i]) == 0) { 1912 netdev_err(netdev, "Unable to alloc receive buffers\n"); 1913 err = -ENOMEM; 1914 goto err_out_free_rq; 1915 } 1916 } 1917 1918 for (i = 0; i < enic->wq_count; i++) 1919 vnic_wq_enable(&enic->wq[i]); 1920 1921 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic)) 1922 enic_dev_add_station_addr(enic); 1923 1924 enic_set_rx_mode(netdev); 1925 1926 netif_tx_wake_all_queues(netdev); 1927 1928 for (i = 0; i < enic->rq_count; i++) 1929 napi_enable(&enic->napi[i]); 1930 1931 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) 1932 for (i = 0; i < enic->wq_count; i++) 1933 napi_enable(&enic->napi[enic_cq_wq(enic, i)]); 1934 enic_dev_enable(enic); 1935 1936 for (i = 0; i < enic->intr_count; i++) 1937 vnic_intr_unmask(&enic->intr[i]); 1938 1939 enic_notify_timer_start(enic); 1940 enic_rfs_timer_start(enic); 1941 1942 return 0; 1943 1944 err_out_free_rq: 1945 for (i = 0; i < enic->rq_count; i++) { 1946 ret = vnic_rq_disable(&enic->rq[i]); 1947 if (!ret) 1948 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf); 1949 } 1950 enic_dev_notify_unset(enic); 1951 err_out_free_intr: 1952 enic_unset_affinity_hint(enic); 1953 enic_free_intr(enic); 1954 1955 return err; 1956 } 1957 1958 /* rtnl lock is held, process context */ 1959 static int enic_stop(struct net_device *netdev) 1960 { 1961 struct enic *enic = netdev_priv(netdev); 1962 unsigned int i; 1963 int err; 1964 1965 for (i = 0; i < enic->intr_count; i++) { 1966 vnic_intr_mask(&enic->intr[i]); 1967 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */ 1968 } 1969 1970 enic_synchronize_irqs(enic); 1971 1972 del_timer_sync(&enic->notify_timer); 1973 enic_rfs_flw_tbl_free(enic); 1974 1975 enic_dev_disable(enic); 1976 1977 for (i = 0; i < enic->rq_count; i++) 1978 napi_disable(&enic->napi[i]); 1979 1980 netif_carrier_off(netdev); 1981 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) 1982 for (i = 0; i < enic->wq_count; i++) 1983 napi_disable(&enic->napi[enic_cq_wq(enic, i)]); 1984 netif_tx_disable(netdev); 1985 1986 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic)) 1987 enic_dev_del_station_addr(enic); 1988 1989 for (i = 0; i < enic->wq_count; i++) { 1990 err = vnic_wq_disable(&enic->wq[i]); 1991 if (err) 1992 return err; 1993 } 1994 for (i = 0; i < enic->rq_count; i++) { 1995 err = vnic_rq_disable(&enic->rq[i]); 1996 if (err) 1997 return err; 1998 } 1999 2000 enic_dev_notify_unset(enic); 2001 enic_unset_affinity_hint(enic); 2002 enic_free_intr(enic); 2003 2004 for (i = 0; i < enic->wq_count; i++) 2005 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf); 2006 for (i = 0; i < enic->rq_count; i++) 2007 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf); 2008 for (i = 0; i < enic->cq_count; i++) 2009 vnic_cq_clean(&enic->cq[i]); 2010 for (i = 0; i < enic->intr_count; i++) 2011 vnic_intr_clean(&enic->intr[i]); 2012 2013 return 0; 2014 } 2015 2016 static int _enic_change_mtu(struct net_device *netdev, int new_mtu) 2017 { 2018 bool running = netif_running(netdev); 2019 int err = 0; 2020 2021 ASSERT_RTNL(); 2022 if (running) { 2023 err = enic_stop(netdev); 2024 if (err) 2025 return err; 2026 } 2027 2028 netdev->mtu = new_mtu; 2029 2030 if (running) { 2031 err = enic_open(netdev); 2032 if (err) 2033 return err; 2034 } 2035 2036 return 0; 2037 } 2038 2039 static int enic_change_mtu(struct net_device *netdev, int new_mtu) 2040 { 2041 struct enic *enic = netdev_priv(netdev); 2042 2043 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) 2044 return -EOPNOTSUPP; 2045 2046 if (netdev->mtu > enic->port_mtu) 2047 netdev_warn(netdev, 2048 "interface MTU (%d) set higher than port MTU (%d)\n", 2049 netdev->mtu, enic->port_mtu); 2050 2051 return _enic_change_mtu(netdev, new_mtu); 2052 } 2053 2054 static void enic_change_mtu_work(struct work_struct *work) 2055 { 2056 struct enic *enic = container_of(work, struct enic, change_mtu_work); 2057 struct net_device *netdev = enic->netdev; 2058 int new_mtu = vnic_dev_mtu(enic->vdev); 2059 2060 rtnl_lock(); 2061 (void)_enic_change_mtu(netdev, new_mtu); 2062 rtnl_unlock(); 2063 2064 netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu); 2065 } 2066 2067 #ifdef CONFIG_NET_POLL_CONTROLLER 2068 static void enic_poll_controller(struct net_device *netdev) 2069 { 2070 struct enic *enic = netdev_priv(netdev); 2071 struct vnic_dev *vdev = enic->vdev; 2072 unsigned int i, intr; 2073 2074 switch (vnic_dev_get_intr_mode(vdev)) { 2075 case VNIC_DEV_INTR_MODE_MSIX: 2076 for (i = 0; i < enic->rq_count; i++) { 2077 intr = enic_msix_rq_intr(enic, i); 2078 enic_isr_msix(enic->msix_entry[intr].vector, 2079 &enic->napi[i]); 2080 } 2081 2082 for (i = 0; i < enic->wq_count; i++) { 2083 intr = enic_msix_wq_intr(enic, i); 2084 enic_isr_msix(enic->msix_entry[intr].vector, 2085 &enic->napi[enic_cq_wq(enic, i)]); 2086 } 2087 2088 break; 2089 case VNIC_DEV_INTR_MODE_MSI: 2090 enic_isr_msi(enic->pdev->irq, enic); 2091 break; 2092 case VNIC_DEV_INTR_MODE_INTX: 2093 enic_isr_legacy(enic->pdev->irq, netdev); 2094 break; 2095 default: 2096 break; 2097 } 2098 } 2099 #endif 2100 2101 static int enic_dev_wait(struct vnic_dev *vdev, 2102 int (*start)(struct vnic_dev *, int), 2103 int (*finished)(struct vnic_dev *, int *), 2104 int arg) 2105 { 2106 unsigned long time; 2107 int done; 2108 int err; 2109 2110 err = start(vdev, arg); 2111 if (err) 2112 return err; 2113 2114 /* Wait for func to complete...2 seconds max 2115 */ 2116 2117 time = jiffies + (HZ * 2); 2118 do { 2119 2120 err = finished(vdev, &done); 2121 if (err) 2122 return err; 2123 2124 if (done) 2125 return 0; 2126 2127 schedule_timeout_uninterruptible(HZ / 10); 2128 2129 } while (time_after(time, jiffies)); 2130 2131 return -ETIMEDOUT; 2132 } 2133 2134 static int enic_dev_open(struct enic *enic) 2135 { 2136 int err; 2137 u32 flags = CMD_OPENF_IG_DESCCACHE; 2138 2139 err = enic_dev_wait(enic->vdev, vnic_dev_open, 2140 vnic_dev_open_done, flags); 2141 if (err) 2142 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n", 2143 err); 2144 2145 return err; 2146 } 2147 2148 static int enic_dev_soft_reset(struct enic *enic) 2149 { 2150 int err; 2151 2152 err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset, 2153 vnic_dev_soft_reset_done, 0); 2154 if (err) 2155 netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n", 2156 err); 2157 2158 return err; 2159 } 2160 2161 static int enic_dev_hang_reset(struct enic *enic) 2162 { 2163 int err; 2164 2165 err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset, 2166 vnic_dev_hang_reset_done, 0); 2167 if (err) 2168 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n", 2169 err); 2170 2171 return err; 2172 } 2173 2174 int __enic_set_rsskey(struct enic *enic) 2175 { 2176 union vnic_rss_key *rss_key_buf_va; 2177 dma_addr_t rss_key_buf_pa; 2178 int i, kidx, bidx, err; 2179 2180 rss_key_buf_va = dma_alloc_coherent(&enic->pdev->dev, 2181 sizeof(union vnic_rss_key), 2182 &rss_key_buf_pa, GFP_ATOMIC); 2183 if (!rss_key_buf_va) 2184 return -ENOMEM; 2185 2186 for (i = 0; i < ENIC_RSS_LEN; i++) { 2187 kidx = i / ENIC_RSS_BYTES_PER_KEY; 2188 bidx = i % ENIC_RSS_BYTES_PER_KEY; 2189 rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i]; 2190 } 2191 spin_lock_bh(&enic->devcmd_lock); 2192 err = enic_set_rss_key(enic, 2193 rss_key_buf_pa, 2194 sizeof(union vnic_rss_key)); 2195 spin_unlock_bh(&enic->devcmd_lock); 2196 2197 dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_key), 2198 rss_key_buf_va, rss_key_buf_pa); 2199 2200 return err; 2201 } 2202 2203 static int enic_set_rsskey(struct enic *enic) 2204 { 2205 netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN); 2206 2207 return __enic_set_rsskey(enic); 2208 } 2209 2210 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits) 2211 { 2212 dma_addr_t rss_cpu_buf_pa; 2213 union vnic_rss_cpu *rss_cpu_buf_va = NULL; 2214 unsigned int i; 2215 int err; 2216 2217 rss_cpu_buf_va = dma_alloc_coherent(&enic->pdev->dev, 2218 sizeof(union vnic_rss_cpu), 2219 &rss_cpu_buf_pa, GFP_ATOMIC); 2220 if (!rss_cpu_buf_va) 2221 return -ENOMEM; 2222 2223 for (i = 0; i < (1 << rss_hash_bits); i++) 2224 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count; 2225 2226 spin_lock_bh(&enic->devcmd_lock); 2227 err = enic_set_rss_cpu(enic, 2228 rss_cpu_buf_pa, 2229 sizeof(union vnic_rss_cpu)); 2230 spin_unlock_bh(&enic->devcmd_lock); 2231 2232 dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_cpu), 2233 rss_cpu_buf_va, rss_cpu_buf_pa); 2234 2235 return err; 2236 } 2237 2238 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu, 2239 u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable) 2240 { 2241 const u8 tso_ipid_split_en = 0; 2242 const u8 ig_vlan_strip_en = 1; 2243 int err; 2244 2245 /* Enable VLAN tag stripping. 2246 */ 2247 2248 spin_lock_bh(&enic->devcmd_lock); 2249 err = enic_set_nic_cfg(enic, 2250 rss_default_cpu, rss_hash_type, 2251 rss_hash_bits, rss_base_cpu, 2252 rss_enable, tso_ipid_split_en, 2253 ig_vlan_strip_en); 2254 spin_unlock_bh(&enic->devcmd_lock); 2255 2256 return err; 2257 } 2258 2259 static int enic_set_rss_nic_cfg(struct enic *enic) 2260 { 2261 struct device *dev = enic_get_dev(enic); 2262 const u8 rss_default_cpu = 0; 2263 const u8 rss_hash_bits = 7; 2264 const u8 rss_base_cpu = 0; 2265 u8 rss_hash_type; 2266 int res; 2267 u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1); 2268 2269 spin_lock_bh(&enic->devcmd_lock); 2270 res = vnic_dev_capable_rss_hash_type(enic->vdev, &rss_hash_type); 2271 spin_unlock_bh(&enic->devcmd_lock); 2272 if (res) { 2273 /* defaults for old adapters 2274 */ 2275 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 | 2276 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 | 2277 NIC_CFG_RSS_HASH_TYPE_IPV6 | 2278 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6; 2279 } 2280 2281 if (rss_enable) { 2282 if (!enic_set_rsskey(enic)) { 2283 if (enic_set_rsscpu(enic, rss_hash_bits)) { 2284 rss_enable = 0; 2285 dev_warn(dev, "RSS disabled, " 2286 "Failed to set RSS cpu indirection table."); 2287 } 2288 } else { 2289 rss_enable = 0; 2290 dev_warn(dev, "RSS disabled, Failed to set RSS key.\n"); 2291 } 2292 } 2293 2294 return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type, 2295 rss_hash_bits, rss_base_cpu, rss_enable); 2296 } 2297 2298 static void enic_set_api_busy(struct enic *enic, bool busy) 2299 { 2300 spin_lock(&enic->enic_api_lock); 2301 enic->enic_api_busy = busy; 2302 spin_unlock(&enic->enic_api_lock); 2303 } 2304 2305 static void enic_reset(struct work_struct *work) 2306 { 2307 struct enic *enic = container_of(work, struct enic, reset); 2308 2309 if (!netif_running(enic->netdev)) 2310 return; 2311 2312 rtnl_lock(); 2313 2314 /* Stop any activity from infiniband */ 2315 enic_set_api_busy(enic, true); 2316 2317 enic_stop(enic->netdev); 2318 enic_dev_soft_reset(enic); 2319 enic_reset_addr_lists(enic); 2320 enic_init_vnic_resources(enic); 2321 enic_set_rss_nic_cfg(enic); 2322 enic_dev_set_ig_vlan_rewrite_mode(enic); 2323 enic_open(enic->netdev); 2324 2325 /* Allow infiniband to fiddle with the device again */ 2326 enic_set_api_busy(enic, false); 2327 2328 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev); 2329 2330 rtnl_unlock(); 2331 } 2332 2333 static void enic_tx_hang_reset(struct work_struct *work) 2334 { 2335 struct enic *enic = container_of(work, struct enic, tx_hang_reset); 2336 2337 rtnl_lock(); 2338 2339 /* Stop any activity from infiniband */ 2340 enic_set_api_busy(enic, true); 2341 2342 enic_dev_hang_notify(enic); 2343 enic_stop(enic->netdev); 2344 enic_dev_hang_reset(enic); 2345 enic_reset_addr_lists(enic); 2346 enic_init_vnic_resources(enic); 2347 enic_set_rss_nic_cfg(enic); 2348 enic_dev_set_ig_vlan_rewrite_mode(enic); 2349 enic_open(enic->netdev); 2350 2351 /* Allow infiniband to fiddle with the device again */ 2352 enic_set_api_busy(enic, false); 2353 2354 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev); 2355 2356 rtnl_unlock(); 2357 } 2358 2359 static int enic_set_intr_mode(struct enic *enic) 2360 { 2361 unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX); 2362 unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX); 2363 unsigned int i; 2364 2365 /* Set interrupt mode (INTx, MSI, MSI-X) depending 2366 * on system capabilities. 2367 * 2368 * Try MSI-X first 2369 * 2370 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs 2371 * (the second to last INTR is used for WQ/RQ errors) 2372 * (the last INTR is used for notifications) 2373 */ 2374 2375 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2); 2376 for (i = 0; i < n + m + 2; i++) 2377 enic->msix_entry[i].entry = i; 2378 2379 /* Use multiple RQs if RSS is enabled 2380 */ 2381 2382 if (ENIC_SETTING(enic, RSS) && 2383 enic->config.intr_mode < 1 && 2384 enic->rq_count >= n && 2385 enic->wq_count >= m && 2386 enic->cq_count >= n + m && 2387 enic->intr_count >= n + m + 2) { 2388 2389 if (pci_enable_msix_range(enic->pdev, enic->msix_entry, 2390 n + m + 2, n + m + 2) > 0) { 2391 2392 enic->rq_count = n; 2393 enic->wq_count = m; 2394 enic->cq_count = n + m; 2395 enic->intr_count = n + m + 2; 2396 2397 vnic_dev_set_intr_mode(enic->vdev, 2398 VNIC_DEV_INTR_MODE_MSIX); 2399 2400 return 0; 2401 } 2402 } 2403 2404 if (enic->config.intr_mode < 1 && 2405 enic->rq_count >= 1 && 2406 enic->wq_count >= m && 2407 enic->cq_count >= 1 + m && 2408 enic->intr_count >= 1 + m + 2) { 2409 if (pci_enable_msix_range(enic->pdev, enic->msix_entry, 2410 1 + m + 2, 1 + m + 2) > 0) { 2411 2412 enic->rq_count = 1; 2413 enic->wq_count = m; 2414 enic->cq_count = 1 + m; 2415 enic->intr_count = 1 + m + 2; 2416 2417 vnic_dev_set_intr_mode(enic->vdev, 2418 VNIC_DEV_INTR_MODE_MSIX); 2419 2420 return 0; 2421 } 2422 } 2423 2424 /* Next try MSI 2425 * 2426 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR 2427 */ 2428 2429 if (enic->config.intr_mode < 2 && 2430 enic->rq_count >= 1 && 2431 enic->wq_count >= 1 && 2432 enic->cq_count >= 2 && 2433 enic->intr_count >= 1 && 2434 !pci_enable_msi(enic->pdev)) { 2435 2436 enic->rq_count = 1; 2437 enic->wq_count = 1; 2438 enic->cq_count = 2; 2439 enic->intr_count = 1; 2440 2441 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI); 2442 2443 return 0; 2444 } 2445 2446 /* Next try INTx 2447 * 2448 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs 2449 * (the first INTR is used for WQ/RQ) 2450 * (the second INTR is used for WQ/RQ errors) 2451 * (the last INTR is used for notifications) 2452 */ 2453 2454 if (enic->config.intr_mode < 3 && 2455 enic->rq_count >= 1 && 2456 enic->wq_count >= 1 && 2457 enic->cq_count >= 2 && 2458 enic->intr_count >= 3) { 2459 2460 enic->rq_count = 1; 2461 enic->wq_count = 1; 2462 enic->cq_count = 2; 2463 enic->intr_count = 3; 2464 2465 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX); 2466 2467 return 0; 2468 } 2469 2470 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN); 2471 2472 return -EINVAL; 2473 } 2474 2475 static void enic_clear_intr_mode(struct enic *enic) 2476 { 2477 switch (vnic_dev_get_intr_mode(enic->vdev)) { 2478 case VNIC_DEV_INTR_MODE_MSIX: 2479 pci_disable_msix(enic->pdev); 2480 break; 2481 case VNIC_DEV_INTR_MODE_MSI: 2482 pci_disable_msi(enic->pdev); 2483 break; 2484 default: 2485 break; 2486 } 2487 2488 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN); 2489 } 2490 2491 static const struct net_device_ops enic_netdev_dynamic_ops = { 2492 .ndo_open = enic_open, 2493 .ndo_stop = enic_stop, 2494 .ndo_start_xmit = enic_hard_start_xmit, 2495 .ndo_get_stats64 = enic_get_stats, 2496 .ndo_validate_addr = eth_validate_addr, 2497 .ndo_set_rx_mode = enic_set_rx_mode, 2498 .ndo_set_mac_address = enic_set_mac_address_dynamic, 2499 .ndo_change_mtu = enic_change_mtu, 2500 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid, 2501 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid, 2502 .ndo_tx_timeout = enic_tx_timeout, 2503 .ndo_set_vf_port = enic_set_vf_port, 2504 .ndo_get_vf_port = enic_get_vf_port, 2505 .ndo_set_vf_mac = enic_set_vf_mac, 2506 #ifdef CONFIG_NET_POLL_CONTROLLER 2507 .ndo_poll_controller = enic_poll_controller, 2508 #endif 2509 #ifdef CONFIG_RFS_ACCEL 2510 .ndo_rx_flow_steer = enic_rx_flow_steer, 2511 #endif 2512 .ndo_udp_tunnel_add = udp_tunnel_nic_add_port, 2513 .ndo_udp_tunnel_del = udp_tunnel_nic_del_port, 2514 .ndo_features_check = enic_features_check, 2515 }; 2516 2517 static const struct net_device_ops enic_netdev_ops = { 2518 .ndo_open = enic_open, 2519 .ndo_stop = enic_stop, 2520 .ndo_start_xmit = enic_hard_start_xmit, 2521 .ndo_get_stats64 = enic_get_stats, 2522 .ndo_validate_addr = eth_validate_addr, 2523 .ndo_set_mac_address = enic_set_mac_address, 2524 .ndo_set_rx_mode = enic_set_rx_mode, 2525 .ndo_change_mtu = enic_change_mtu, 2526 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid, 2527 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid, 2528 .ndo_tx_timeout = enic_tx_timeout, 2529 .ndo_set_vf_port = enic_set_vf_port, 2530 .ndo_get_vf_port = enic_get_vf_port, 2531 .ndo_set_vf_mac = enic_set_vf_mac, 2532 #ifdef CONFIG_NET_POLL_CONTROLLER 2533 .ndo_poll_controller = enic_poll_controller, 2534 #endif 2535 #ifdef CONFIG_RFS_ACCEL 2536 .ndo_rx_flow_steer = enic_rx_flow_steer, 2537 #endif 2538 .ndo_udp_tunnel_add = udp_tunnel_nic_add_port, 2539 .ndo_udp_tunnel_del = udp_tunnel_nic_del_port, 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(&pdev->dev, DMA_BIT_MASK(47)); 2722 if (err) { 2723 err = dma_set_mask(&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 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32)); 2729 if (err) { 2730 dev_err(dev, "Unable to obtain %u-bit DMA " 2731 "for consistent allocations, aborting\n", 32); 2732 goto err_out_release_regions; 2733 } 2734 } else { 2735 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(47)); 2736 if (err) { 2737 dev_err(dev, "Unable to obtain %u-bit DMA " 2738 "for consistent allocations, aborting\n", 47); 2739 goto err_out_release_regions; 2740 } 2741 using_dac = 1; 2742 } 2743 2744 /* Map vNIC resources from BAR0-5 2745 */ 2746 2747 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) { 2748 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM)) 2749 continue; 2750 enic->bar[i].len = pci_resource_len(pdev, i); 2751 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len); 2752 if (!enic->bar[i].vaddr) { 2753 dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i); 2754 err = -ENODEV; 2755 goto err_out_iounmap; 2756 } 2757 enic->bar[i].bus_addr = pci_resource_start(pdev, i); 2758 } 2759 2760 /* Register vNIC device 2761 */ 2762 2763 enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar, 2764 ARRAY_SIZE(enic->bar)); 2765 if (!enic->vdev) { 2766 dev_err(dev, "vNIC registration failed, aborting\n"); 2767 err = -ENODEV; 2768 goto err_out_iounmap; 2769 } 2770 2771 err = vnic_devcmd_init(enic->vdev); 2772 2773 if (err) 2774 goto err_out_vnic_unregister; 2775 2776 #ifdef CONFIG_PCI_IOV 2777 /* Get number of subvnics */ 2778 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV); 2779 if (pos) { 2780 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF, 2781 &enic->num_vfs); 2782 if (enic->num_vfs) { 2783 err = pci_enable_sriov(pdev, enic->num_vfs); 2784 if (err) { 2785 dev_err(dev, "SRIOV enable failed, aborting." 2786 " pci_enable_sriov() returned %d\n", 2787 err); 2788 goto err_out_vnic_unregister; 2789 } 2790 enic->priv_flags |= ENIC_SRIOV_ENABLED; 2791 num_pps = enic->num_vfs; 2792 } 2793 } 2794 #endif 2795 2796 /* Allocate structure for port profiles */ 2797 enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL); 2798 if (!enic->pp) { 2799 err = -ENOMEM; 2800 goto err_out_disable_sriov_pp; 2801 } 2802 2803 /* Issue device open to get device in known state 2804 */ 2805 2806 err = enic_dev_open(enic); 2807 if (err) { 2808 dev_err(dev, "vNIC dev open failed, aborting\n"); 2809 goto err_out_disable_sriov; 2810 } 2811 2812 /* Setup devcmd lock 2813 */ 2814 2815 spin_lock_init(&enic->devcmd_lock); 2816 spin_lock_init(&enic->enic_api_lock); 2817 2818 /* 2819 * Set ingress vlan rewrite mode before vnic initialization 2820 */ 2821 2822 err = enic_dev_set_ig_vlan_rewrite_mode(enic); 2823 if (err) { 2824 dev_err(dev, 2825 "Failed to set ingress vlan rewrite mode, aborting.\n"); 2826 goto err_out_dev_close; 2827 } 2828 2829 /* Issue device init to initialize the vnic-to-switch link. 2830 * We'll start with carrier off and wait for link UP 2831 * notification later to turn on carrier. We don't need 2832 * to wait here for the vnic-to-switch link initialization 2833 * to complete; link UP notification is the indication that 2834 * the process is complete. 2835 */ 2836 2837 netif_carrier_off(netdev); 2838 2839 /* Do not call dev_init for a dynamic vnic. 2840 * For a dynamic vnic, init_prov_info will be 2841 * called later by an upper layer. 2842 */ 2843 2844 if (!enic_is_dynamic(enic)) { 2845 err = vnic_dev_init(enic->vdev, 0); 2846 if (err) { 2847 dev_err(dev, "vNIC dev init failed, aborting\n"); 2848 goto err_out_dev_close; 2849 } 2850 } 2851 2852 err = enic_dev_init(enic); 2853 if (err) { 2854 dev_err(dev, "Device initialization failed, aborting\n"); 2855 goto err_out_dev_close; 2856 } 2857 2858 netif_set_real_num_tx_queues(netdev, enic->wq_count); 2859 netif_set_real_num_rx_queues(netdev, enic->rq_count); 2860 2861 /* Setup notification timer, HW reset task, and wq locks 2862 */ 2863 2864 timer_setup(&enic->notify_timer, enic_notify_timer, 0); 2865 2866 enic_rfs_flw_tbl_init(enic); 2867 enic_set_rx_coal_setting(enic); 2868 INIT_WORK(&enic->reset, enic_reset); 2869 INIT_WORK(&enic->tx_hang_reset, enic_tx_hang_reset); 2870 INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work); 2871 2872 for (i = 0; i < enic->wq_count; i++) 2873 spin_lock_init(&enic->wq_lock[i]); 2874 2875 /* Register net device 2876 */ 2877 2878 enic->port_mtu = enic->config.mtu; 2879 2880 err = enic_set_mac_addr(netdev, enic->mac_addr); 2881 if (err) { 2882 dev_err(dev, "Invalid MAC address, aborting\n"); 2883 goto err_out_dev_deinit; 2884 } 2885 2886 enic->tx_coalesce_usecs = enic->config.intr_timer_usec; 2887 /* rx coalesce time already got initialized. This gets used 2888 * if adaptive coal is turned off 2889 */ 2890 enic->rx_coalesce_usecs = enic->tx_coalesce_usecs; 2891 2892 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) 2893 netdev->netdev_ops = &enic_netdev_dynamic_ops; 2894 else 2895 netdev->netdev_ops = &enic_netdev_ops; 2896 2897 netdev->watchdog_timeo = 2 * HZ; 2898 enic_set_ethtool_ops(netdev); 2899 2900 netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX; 2901 if (ENIC_SETTING(enic, LOOP)) { 2902 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX; 2903 enic->loop_enable = 1; 2904 enic->loop_tag = enic->config.loop_tag; 2905 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag); 2906 } 2907 if (ENIC_SETTING(enic, TXCSUM)) 2908 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM; 2909 if (ENIC_SETTING(enic, TSO)) 2910 netdev->hw_features |= NETIF_F_TSO | 2911 NETIF_F_TSO6 | NETIF_F_TSO_ECN; 2912 if (ENIC_SETTING(enic, RSS)) 2913 netdev->hw_features |= NETIF_F_RXHASH; 2914 if (ENIC_SETTING(enic, RXCSUM)) 2915 netdev->hw_features |= NETIF_F_RXCSUM; 2916 if (ENIC_SETTING(enic, VXLAN)) { 2917 u64 patch_level; 2918 u64 a1 = 0; 2919 2920 netdev->hw_enc_features |= NETIF_F_RXCSUM | 2921 NETIF_F_TSO | 2922 NETIF_F_TSO6 | 2923 NETIF_F_TSO_ECN | 2924 NETIF_F_GSO_UDP_TUNNEL | 2925 NETIF_F_HW_CSUM | 2926 NETIF_F_GSO_UDP_TUNNEL_CSUM; 2927 netdev->hw_features |= netdev->hw_enc_features; 2928 /* get bit mask from hw about supported offload bit level 2929 * BIT(0) = fw supports patch_level 0 2930 * fcoe bit = encap 2931 * fcoe_fc_crc_ok = outer csum ok 2932 * BIT(1) = always set by fw 2933 * BIT(2) = fw supports patch_level 2 2934 * BIT(0) in rss_hash = encap 2935 * BIT(1,2) in rss_hash = outer_ip_csum_ok/ 2936 * outer_tcp_csum_ok 2937 * used in enic_rq_indicate_buf 2938 */ 2939 err = vnic_dev_get_supported_feature_ver(enic->vdev, 2940 VIC_FEATURE_VXLAN, 2941 &patch_level, &a1); 2942 if (err) 2943 patch_level = 0; 2944 enic->vxlan.flags = (u8)a1; 2945 /* mask bits that are supported by driver 2946 */ 2947 patch_level &= BIT_ULL(0) | BIT_ULL(2); 2948 patch_level = fls(patch_level); 2949 patch_level = patch_level ? patch_level - 1 : 0; 2950 enic->vxlan.patch_level = patch_level; 2951 2952 if (vnic_dev_get_res_count(enic->vdev, RES_TYPE_WQ) == 1 || 2953 enic->vxlan.flags & ENIC_VXLAN_MULTI_WQ) { 2954 netdev->udp_tunnel_nic_info = &enic_udp_tunnels_v4; 2955 if (enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6) 2956 netdev->udp_tunnel_nic_info = &enic_udp_tunnels; 2957 } 2958 } 2959 2960 netdev->features |= netdev->hw_features; 2961 netdev->vlan_features |= netdev->features; 2962 2963 #ifdef CONFIG_RFS_ACCEL 2964 netdev->hw_features |= NETIF_F_NTUPLE; 2965 #endif 2966 2967 if (using_dac) 2968 netdev->features |= NETIF_F_HIGHDMA; 2969 2970 netdev->priv_flags |= IFF_UNICAST_FLT; 2971 2972 /* MTU range: 68 - 9000 */ 2973 netdev->min_mtu = ENIC_MIN_MTU; 2974 netdev->max_mtu = ENIC_MAX_MTU; 2975 netdev->mtu = enic->port_mtu; 2976 2977 err = register_netdev(netdev); 2978 if (err) { 2979 dev_err(dev, "Cannot register net device, aborting\n"); 2980 goto err_out_dev_deinit; 2981 } 2982 enic->rx_copybreak = RX_COPYBREAK_DEFAULT; 2983 2984 return 0; 2985 2986 err_out_dev_deinit: 2987 enic_dev_deinit(enic); 2988 err_out_dev_close: 2989 vnic_dev_close(enic->vdev); 2990 err_out_disable_sriov: 2991 kfree(enic->pp); 2992 err_out_disable_sriov_pp: 2993 #ifdef CONFIG_PCI_IOV 2994 if (enic_sriov_enabled(enic)) { 2995 pci_disable_sriov(pdev); 2996 enic->priv_flags &= ~ENIC_SRIOV_ENABLED; 2997 } 2998 #endif 2999 err_out_vnic_unregister: 3000 vnic_dev_unregister(enic->vdev); 3001 err_out_iounmap: 3002 enic_iounmap(enic); 3003 err_out_release_regions: 3004 pci_release_regions(pdev); 3005 err_out_disable_device: 3006 pci_disable_device(pdev); 3007 err_out_free_netdev: 3008 free_netdev(netdev); 3009 3010 return err; 3011 } 3012 3013 static void enic_remove(struct pci_dev *pdev) 3014 { 3015 struct net_device *netdev = pci_get_drvdata(pdev); 3016 3017 if (netdev) { 3018 struct enic *enic = netdev_priv(netdev); 3019 3020 cancel_work_sync(&enic->reset); 3021 cancel_work_sync(&enic->change_mtu_work); 3022 unregister_netdev(netdev); 3023 enic_dev_deinit(enic); 3024 vnic_dev_close(enic->vdev); 3025 #ifdef CONFIG_PCI_IOV 3026 if (enic_sriov_enabled(enic)) { 3027 pci_disable_sriov(pdev); 3028 enic->priv_flags &= ~ENIC_SRIOV_ENABLED; 3029 } 3030 #endif 3031 kfree(enic->pp); 3032 vnic_dev_unregister(enic->vdev); 3033 enic_iounmap(enic); 3034 pci_release_regions(pdev); 3035 pci_disable_device(pdev); 3036 free_netdev(netdev); 3037 } 3038 } 3039 3040 static struct pci_driver enic_driver = { 3041 .name = DRV_NAME, 3042 .id_table = enic_id_table, 3043 .probe = enic_probe, 3044 .remove = enic_remove, 3045 }; 3046 3047 static int __init enic_init_module(void) 3048 { 3049 return pci_register_driver(&enic_driver); 3050 } 3051 3052 static void __exit enic_cleanup_module(void) 3053 { 3054 pci_unregister_driver(&enic_driver); 3055 } 3056 3057 module_init(enic_init_module); 3058 module_exit(enic_cleanup_module); 3059