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_VERSION(DRV_VERSION); 84 MODULE_DEVICE_TABLE(pci, enic_id_table); 85 86 #define ENIC_LARGE_PKT_THRESHOLD 1000 87 #define ENIC_MAX_COALESCE_TIMERS 10 88 /* Interrupt moderation table, which will be used to decide the 89 * coalescing timer values 90 * {rx_rate in Mbps, mapping percentage of the range} 91 */ 92 static struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = { 93 {4000, 0}, 94 {4400, 10}, 95 {5060, 20}, 96 {5230, 30}, 97 {5540, 40}, 98 {5820, 50}, 99 {6120, 60}, 100 {6435, 70}, 101 {6745, 80}, 102 {7000, 90}, 103 {0xFFFFFFFF, 100} 104 }; 105 106 /* This table helps the driver to pick different ranges for rx coalescing 107 * timer depending on the link speed. 108 */ 109 static struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = { 110 {0, 0}, /* 0 - 4 Gbps */ 111 {0, 3}, /* 4 - 10 Gbps */ 112 {3, 6}, /* 10 - 40 Gbps */ 113 }; 114 115 static void enic_init_affinity_hint(struct enic *enic) 116 { 117 int numa_node = dev_to_node(&enic->pdev->dev); 118 int i; 119 120 for (i = 0; i < enic->intr_count; i++) { 121 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i) || 122 (enic->msix[i].affinity_mask && 123 !cpumask_empty(enic->msix[i].affinity_mask))) 124 continue; 125 if (zalloc_cpumask_var(&enic->msix[i].affinity_mask, 126 GFP_KERNEL)) 127 cpumask_set_cpu(cpumask_local_spread(i, numa_node), 128 enic->msix[i].affinity_mask); 129 } 130 } 131 132 static void enic_free_affinity_hint(struct enic *enic) 133 { 134 int i; 135 136 for (i = 0; i < enic->intr_count; i++) { 137 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i)) 138 continue; 139 free_cpumask_var(enic->msix[i].affinity_mask); 140 } 141 } 142 143 static void enic_set_affinity_hint(struct enic *enic) 144 { 145 int i; 146 int err; 147 148 for (i = 0; i < enic->intr_count; i++) { 149 if (enic_is_err_intr(enic, i) || 150 enic_is_notify_intr(enic, i) || 151 !enic->msix[i].affinity_mask || 152 cpumask_empty(enic->msix[i].affinity_mask)) 153 continue; 154 err = irq_set_affinity_hint(enic->msix_entry[i].vector, 155 enic->msix[i].affinity_mask); 156 if (err) 157 netdev_warn(enic->netdev, "irq_set_affinity_hint failed, err %d\n", 158 err); 159 } 160 161 for (i = 0; i < enic->wq_count; i++) { 162 int wq_intr = enic_msix_wq_intr(enic, i); 163 164 if (enic->msix[wq_intr].affinity_mask && 165 !cpumask_empty(enic->msix[wq_intr].affinity_mask)) 166 netif_set_xps_queue(enic->netdev, 167 enic->msix[wq_intr].affinity_mask, 168 i); 169 } 170 } 171 172 static void enic_unset_affinity_hint(struct enic *enic) 173 { 174 int i; 175 176 for (i = 0; i < enic->intr_count; i++) 177 irq_set_affinity_hint(enic->msix_entry[i].vector, NULL); 178 } 179 180 static void enic_udp_tunnel_add(struct net_device *netdev, 181 struct udp_tunnel_info *ti) 182 { 183 struct enic *enic = netdev_priv(netdev); 184 __be16 port = ti->port; 185 int err; 186 187 spin_lock_bh(&enic->devcmd_lock); 188 189 if (ti->type != UDP_TUNNEL_TYPE_VXLAN) { 190 netdev_info(netdev, "udp_tnl: only vxlan tunnel offload supported"); 191 goto error; 192 } 193 194 switch (ti->sa_family) { 195 case AF_INET6: 196 if (!(enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6)) { 197 netdev_info(netdev, "vxlan: only IPv4 offload supported"); 198 goto error; 199 } 200 /* Fall through */ 201 case AF_INET: 202 break; 203 default: 204 goto error; 205 } 206 207 if (enic->vxlan.vxlan_udp_port_number) { 208 if (ntohs(port) == enic->vxlan.vxlan_udp_port_number) 209 netdev_warn(netdev, "vxlan: udp port already offloaded"); 210 else 211 netdev_info(netdev, "vxlan: offload supported for only one UDP port"); 212 213 goto error; 214 } 215 if ((vnic_dev_get_res_count(enic->vdev, RES_TYPE_WQ) != 1) && 216 !(enic->vxlan.flags & ENIC_VXLAN_MULTI_WQ)) { 217 netdev_info(netdev, "vxlan: vxlan offload with multi wq not supported on this adapter"); 218 goto error; 219 } 220 221 err = vnic_dev_overlay_offload_cfg(enic->vdev, 222 OVERLAY_CFG_VXLAN_PORT_UPDATE, 223 ntohs(port)); 224 if (err) 225 goto error; 226 227 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN, 228 enic->vxlan.patch_level); 229 if (err) 230 goto error; 231 232 enic->vxlan.vxlan_udp_port_number = ntohs(port); 233 234 netdev_info(netdev, "vxlan fw-vers-%d: offload enabled for udp port: %d, sa_family: %d ", 235 (int)enic->vxlan.patch_level, ntohs(port), ti->sa_family); 236 237 goto unlock; 238 239 error: 240 netdev_info(netdev, "failed to offload udp port: %d, sa_family: %d, type: %d", 241 ntohs(port), ti->sa_family, ti->type); 242 unlock: 243 spin_unlock_bh(&enic->devcmd_lock); 244 } 245 246 static void enic_udp_tunnel_del(struct net_device *netdev, 247 struct udp_tunnel_info *ti) 248 { 249 struct enic *enic = netdev_priv(netdev); 250 int err; 251 252 spin_lock_bh(&enic->devcmd_lock); 253 254 if ((ntohs(ti->port) != enic->vxlan.vxlan_udp_port_number) || 255 ti->type != UDP_TUNNEL_TYPE_VXLAN) { 256 netdev_info(netdev, "udp_tnl: port:%d, sa_family: %d, type: %d not offloaded", 257 ntohs(ti->port), ti->sa_family, ti->type); 258 goto unlock; 259 } 260 261 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN, 262 OVERLAY_OFFLOAD_DISABLE); 263 if (err) { 264 netdev_err(netdev, "vxlan: del offload udp port: %d failed", 265 ntohs(ti->port)); 266 goto unlock; 267 } 268 269 enic->vxlan.vxlan_udp_port_number = 0; 270 271 netdev_info(netdev, "vxlan: del offload udp port %d, family %d\n", 272 ntohs(ti->port), ti->sa_family); 273 274 unlock: 275 spin_unlock_bh(&enic->devcmd_lock); 276 } 277 278 static netdev_features_t enic_features_check(struct sk_buff *skb, 279 struct net_device *dev, 280 netdev_features_t features) 281 { 282 const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb); 283 struct enic *enic = netdev_priv(dev); 284 struct udphdr *udph; 285 u16 port = 0; 286 u8 proto; 287 288 if (!skb->encapsulation) 289 return features; 290 291 features = vxlan_features_check(skb, features); 292 293 switch (vlan_get_protocol(skb)) { 294 case htons(ETH_P_IPV6): 295 if (!(enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6)) 296 goto out; 297 proto = ipv6_hdr(skb)->nexthdr; 298 break; 299 case htons(ETH_P_IP): 300 proto = ip_hdr(skb)->protocol; 301 break; 302 default: 303 goto out; 304 } 305 306 switch (eth->h_proto) { 307 case ntohs(ETH_P_IPV6): 308 if (!(enic->vxlan.flags & ENIC_VXLAN_INNER_IPV6)) 309 goto out; 310 /* Fall through */ 311 case ntohs(ETH_P_IP): 312 break; 313 default: 314 goto out; 315 } 316 317 318 if (proto == IPPROTO_UDP) { 319 udph = udp_hdr(skb); 320 port = be16_to_cpu(udph->dest); 321 } 322 323 /* HW supports offload of only one UDP port. Remove CSUM and GSO MASK 324 * for other UDP port tunnels 325 */ 326 if (port != enic->vxlan.vxlan_udp_port_number) 327 goto out; 328 329 return features; 330 331 out: 332 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 333 } 334 335 int enic_is_dynamic(struct enic *enic) 336 { 337 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN; 338 } 339 340 int enic_sriov_enabled(struct enic *enic) 341 { 342 return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0; 343 } 344 345 static int enic_is_sriov_vf(struct enic *enic) 346 { 347 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF; 348 } 349 350 int enic_is_valid_vf(struct enic *enic, int vf) 351 { 352 #ifdef CONFIG_PCI_IOV 353 return vf >= 0 && vf < enic->num_vfs; 354 #else 355 return 0; 356 #endif 357 } 358 359 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf) 360 { 361 struct enic *enic = vnic_dev_priv(wq->vdev); 362 363 if (buf->sop) 364 pci_unmap_single(enic->pdev, buf->dma_addr, 365 buf->len, PCI_DMA_TODEVICE); 366 else 367 pci_unmap_page(enic->pdev, buf->dma_addr, 368 buf->len, PCI_DMA_TODEVICE); 369 370 if (buf->os_buf) 371 dev_kfree_skb_any(buf->os_buf); 372 } 373 374 static void enic_wq_free_buf(struct vnic_wq *wq, 375 struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque) 376 { 377 enic_free_wq_buf(wq, buf); 378 } 379 380 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc, 381 u8 type, u16 q_number, u16 completed_index, void *opaque) 382 { 383 struct enic *enic = vnic_dev_priv(vdev); 384 385 spin_lock(&enic->wq_lock[q_number]); 386 387 vnic_wq_service(&enic->wq[q_number], cq_desc, 388 completed_index, enic_wq_free_buf, 389 opaque); 390 391 if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) && 392 vnic_wq_desc_avail(&enic->wq[q_number]) >= 393 (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)) 394 netif_wake_subqueue(enic->netdev, q_number); 395 396 spin_unlock(&enic->wq_lock[q_number]); 397 398 return 0; 399 } 400 401 static bool enic_log_q_error(struct enic *enic) 402 { 403 unsigned int i; 404 u32 error_status; 405 bool err = false; 406 407 for (i = 0; i < enic->wq_count; i++) { 408 error_status = vnic_wq_error_status(&enic->wq[i]); 409 err |= error_status; 410 if (error_status) 411 netdev_err(enic->netdev, "WQ[%d] error_status %d\n", 412 i, error_status); 413 } 414 415 for (i = 0; i < enic->rq_count; i++) { 416 error_status = vnic_rq_error_status(&enic->rq[i]); 417 err |= error_status; 418 if (error_status) 419 netdev_err(enic->netdev, "RQ[%d] error_status %d\n", 420 i, error_status); 421 } 422 423 return err; 424 } 425 426 static void enic_msglvl_check(struct enic *enic) 427 { 428 u32 msg_enable = vnic_dev_msg_lvl(enic->vdev); 429 430 if (msg_enable != enic->msg_enable) { 431 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n", 432 enic->msg_enable, msg_enable); 433 enic->msg_enable = msg_enable; 434 } 435 } 436 437 static void enic_mtu_check(struct enic *enic) 438 { 439 u32 mtu = vnic_dev_mtu(enic->vdev); 440 struct net_device *netdev = enic->netdev; 441 442 if (mtu && mtu != enic->port_mtu) { 443 enic->port_mtu = mtu; 444 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) { 445 mtu = max_t(int, ENIC_MIN_MTU, 446 min_t(int, ENIC_MAX_MTU, mtu)); 447 if (mtu != netdev->mtu) 448 schedule_work(&enic->change_mtu_work); 449 } else { 450 if (mtu < netdev->mtu) 451 netdev_warn(netdev, 452 "interface MTU (%d) set higher " 453 "than switch port MTU (%d)\n", 454 netdev->mtu, mtu); 455 } 456 } 457 } 458 459 static void enic_link_check(struct enic *enic) 460 { 461 int link_status = vnic_dev_link_status(enic->vdev); 462 int carrier_ok = netif_carrier_ok(enic->netdev); 463 464 if (link_status && !carrier_ok) { 465 netdev_info(enic->netdev, "Link UP\n"); 466 netif_carrier_on(enic->netdev); 467 } else if (!link_status && carrier_ok) { 468 netdev_info(enic->netdev, "Link DOWN\n"); 469 netif_carrier_off(enic->netdev); 470 } 471 } 472 473 static void enic_notify_check(struct enic *enic) 474 { 475 enic_msglvl_check(enic); 476 enic_mtu_check(enic); 477 enic_link_check(enic); 478 } 479 480 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i)) 481 482 static irqreturn_t enic_isr_legacy(int irq, void *data) 483 { 484 struct net_device *netdev = data; 485 struct enic *enic = netdev_priv(netdev); 486 unsigned int io_intr = enic_legacy_io_intr(); 487 unsigned int err_intr = enic_legacy_err_intr(); 488 unsigned int notify_intr = enic_legacy_notify_intr(); 489 u32 pba; 490 491 vnic_intr_mask(&enic->intr[io_intr]); 492 493 pba = vnic_intr_legacy_pba(enic->legacy_pba); 494 if (!pba) { 495 vnic_intr_unmask(&enic->intr[io_intr]); 496 return IRQ_NONE; /* not our interrupt */ 497 } 498 499 if (ENIC_TEST_INTR(pba, notify_intr)) { 500 enic_notify_check(enic); 501 vnic_intr_return_all_credits(&enic->intr[notify_intr]); 502 } 503 504 if (ENIC_TEST_INTR(pba, err_intr)) { 505 vnic_intr_return_all_credits(&enic->intr[err_intr]); 506 enic_log_q_error(enic); 507 /* schedule recovery from WQ/RQ error */ 508 schedule_work(&enic->reset); 509 return IRQ_HANDLED; 510 } 511 512 if (ENIC_TEST_INTR(pba, io_intr)) 513 napi_schedule_irqoff(&enic->napi[0]); 514 else 515 vnic_intr_unmask(&enic->intr[io_intr]); 516 517 return IRQ_HANDLED; 518 } 519 520 static irqreturn_t enic_isr_msi(int irq, void *data) 521 { 522 struct enic *enic = data; 523 524 /* With MSI, there is no sharing of interrupts, so this is 525 * our interrupt and there is no need to ack it. The device 526 * is not providing per-vector masking, so the OS will not 527 * write to PCI config space to mask/unmask the interrupt. 528 * We're using mask_on_assertion for MSI, so the device 529 * automatically masks the interrupt when the interrupt is 530 * generated. Later, when exiting polling, the interrupt 531 * will be unmasked (see enic_poll). 532 * 533 * Also, the device uses the same PCIe Traffic Class (TC) 534 * for Memory Write data and MSI, so there are no ordering 535 * issues; the MSI will always arrive at the Root Complex 536 * _after_ corresponding Memory Writes (i.e. descriptor 537 * writes). 538 */ 539 540 napi_schedule_irqoff(&enic->napi[0]); 541 542 return IRQ_HANDLED; 543 } 544 545 static irqreturn_t enic_isr_msix(int irq, void *data) 546 { 547 struct napi_struct *napi = data; 548 549 napi_schedule_irqoff(napi); 550 551 return IRQ_HANDLED; 552 } 553 554 static irqreturn_t enic_isr_msix_err(int irq, void *data) 555 { 556 struct enic *enic = data; 557 unsigned int intr = enic_msix_err_intr(enic); 558 559 vnic_intr_return_all_credits(&enic->intr[intr]); 560 561 if (enic_log_q_error(enic)) 562 /* schedule recovery from WQ/RQ error */ 563 schedule_work(&enic->reset); 564 565 return IRQ_HANDLED; 566 } 567 568 static irqreturn_t enic_isr_msix_notify(int irq, void *data) 569 { 570 struct enic *enic = data; 571 unsigned int intr = enic_msix_notify_intr(enic); 572 573 enic_notify_check(enic); 574 vnic_intr_return_all_credits(&enic->intr[intr]); 575 576 return IRQ_HANDLED; 577 } 578 579 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq, 580 struct sk_buff *skb, unsigned int len_left, 581 int loopback) 582 { 583 const skb_frag_t *frag; 584 dma_addr_t dma_addr; 585 586 /* Queue additional data fragments */ 587 for (frag = skb_shinfo(skb)->frags; len_left; frag++) { 588 len_left -= skb_frag_size(frag); 589 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 0, 590 skb_frag_size(frag), 591 DMA_TO_DEVICE); 592 if (unlikely(enic_dma_map_check(enic, dma_addr))) 593 return -ENOMEM; 594 enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag), 595 (len_left == 0), /* EOP? */ 596 loopback); 597 } 598 599 return 0; 600 } 601 602 static int enic_queue_wq_skb_vlan(struct enic *enic, struct vnic_wq *wq, 603 struct sk_buff *skb, int vlan_tag_insert, 604 unsigned int vlan_tag, int loopback) 605 { 606 unsigned int head_len = skb_headlen(skb); 607 unsigned int len_left = skb->len - head_len; 608 int eop = (len_left == 0); 609 dma_addr_t dma_addr; 610 int err = 0; 611 612 dma_addr = pci_map_single(enic->pdev, skb->data, head_len, 613 PCI_DMA_TODEVICE); 614 if (unlikely(enic_dma_map_check(enic, dma_addr))) 615 return -ENOMEM; 616 617 /* Queue the main skb fragment. The fragments are no larger 618 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less 619 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor 620 * per fragment is queued. 621 */ 622 enic_queue_wq_desc(wq, skb, dma_addr, head_len, vlan_tag_insert, 623 vlan_tag, eop, loopback); 624 625 if (!eop) 626 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback); 627 628 return err; 629 } 630 631 static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq, 632 struct sk_buff *skb, int vlan_tag_insert, 633 unsigned int vlan_tag, int loopback) 634 { 635 unsigned int head_len = skb_headlen(skb); 636 unsigned int len_left = skb->len - head_len; 637 unsigned int hdr_len = skb_checksum_start_offset(skb); 638 unsigned int csum_offset = hdr_len + skb->csum_offset; 639 int eop = (len_left == 0); 640 dma_addr_t dma_addr; 641 int err = 0; 642 643 dma_addr = pci_map_single(enic->pdev, skb->data, head_len, 644 PCI_DMA_TODEVICE); 645 if (unlikely(enic_dma_map_check(enic, dma_addr))) 646 return -ENOMEM; 647 648 /* Queue the main skb fragment. The fragments are no larger 649 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less 650 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor 651 * per fragment is queued. 652 */ 653 enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len, csum_offset, 654 hdr_len, vlan_tag_insert, vlan_tag, eop, 655 loopback); 656 657 if (!eop) 658 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback); 659 660 return err; 661 } 662 663 static void enic_preload_tcp_csum_encap(struct sk_buff *skb) 664 { 665 const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb); 666 667 switch (eth->h_proto) { 668 case ntohs(ETH_P_IP): 669 inner_ip_hdr(skb)->check = 0; 670 inner_tcp_hdr(skb)->check = 671 ~csum_tcpudp_magic(inner_ip_hdr(skb)->saddr, 672 inner_ip_hdr(skb)->daddr, 0, 673 IPPROTO_TCP, 0); 674 break; 675 case ntohs(ETH_P_IPV6): 676 inner_tcp_hdr(skb)->check = 677 ~csum_ipv6_magic(&inner_ipv6_hdr(skb)->saddr, 678 &inner_ipv6_hdr(skb)->daddr, 0, 679 IPPROTO_TCP, 0); 680 break; 681 default: 682 WARN_ONCE(1, "Non ipv4/ipv6 inner pkt for encap offload"); 683 break; 684 } 685 } 686 687 static void enic_preload_tcp_csum(struct sk_buff *skb) 688 { 689 /* Preload TCP csum field with IP pseudo hdr calculated 690 * with IP length set to zero. HW will later add in length 691 * to each TCP segment resulting from the TSO. 692 */ 693 694 if (skb->protocol == cpu_to_be16(ETH_P_IP)) { 695 ip_hdr(skb)->check = 0; 696 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 697 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); 698 } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) { 699 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 700 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); 701 } 702 } 703 704 static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq, 705 struct sk_buff *skb, unsigned int mss, 706 int vlan_tag_insert, unsigned int vlan_tag, 707 int loopback) 708 { 709 unsigned int frag_len_left = skb_headlen(skb); 710 unsigned int len_left = skb->len - frag_len_left; 711 int eop = (len_left == 0); 712 unsigned int offset = 0; 713 unsigned int hdr_len; 714 dma_addr_t dma_addr; 715 unsigned int len; 716 skb_frag_t *frag; 717 718 if (skb->encapsulation) { 719 hdr_len = skb_inner_transport_header(skb) - skb->data; 720 hdr_len += inner_tcp_hdrlen(skb); 721 enic_preload_tcp_csum_encap(skb); 722 } else { 723 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb); 724 enic_preload_tcp_csum(skb); 725 } 726 727 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors 728 * for the main skb fragment 729 */ 730 while (frag_len_left) { 731 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN); 732 dma_addr = pci_map_single(enic->pdev, skb->data + offset, len, 733 PCI_DMA_TODEVICE); 734 if (unlikely(enic_dma_map_check(enic, dma_addr))) 735 return -ENOMEM; 736 enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len, 737 vlan_tag_insert, vlan_tag, 738 eop && (len == frag_len_left), loopback); 739 frag_len_left -= len; 740 offset += len; 741 } 742 743 if (eop) 744 return 0; 745 746 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors 747 * for additional data fragments 748 */ 749 for (frag = skb_shinfo(skb)->frags; len_left; frag++) { 750 len_left -= skb_frag_size(frag); 751 frag_len_left = skb_frag_size(frag); 752 offset = 0; 753 754 while (frag_len_left) { 755 len = min(frag_len_left, 756 (unsigned int)WQ_ENET_MAX_DESC_LEN); 757 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 758 offset, len, 759 DMA_TO_DEVICE); 760 if (unlikely(enic_dma_map_check(enic, dma_addr))) 761 return -ENOMEM; 762 enic_queue_wq_desc_cont(wq, skb, dma_addr, len, 763 (len_left == 0) && 764 (len == frag_len_left),/*EOP*/ 765 loopback); 766 frag_len_left -= len; 767 offset += len; 768 } 769 } 770 771 return 0; 772 } 773 774 static inline int enic_queue_wq_skb_encap(struct enic *enic, struct vnic_wq *wq, 775 struct sk_buff *skb, 776 int vlan_tag_insert, 777 unsigned int vlan_tag, int loopback) 778 { 779 unsigned int head_len = skb_headlen(skb); 780 unsigned int len_left = skb->len - head_len; 781 /* Hardware will overwrite the checksum fields, calculating from 782 * scratch and ignoring the value placed by software. 783 * Offload mode = 00 784 * mss[2], mss[1], mss[0] bits are set 785 */ 786 unsigned int mss_or_csum = 7; 787 int eop = (len_left == 0); 788 dma_addr_t dma_addr; 789 int err = 0; 790 791 dma_addr = pci_map_single(enic->pdev, skb->data, head_len, 792 PCI_DMA_TODEVICE); 793 if (unlikely(enic_dma_map_check(enic, dma_addr))) 794 return -ENOMEM; 795 796 enic_queue_wq_desc_ex(wq, skb, dma_addr, head_len, mss_or_csum, 0, 797 vlan_tag_insert, vlan_tag, 798 WQ_ENET_OFFLOAD_MODE_CSUM, eop, 1 /* SOP */, eop, 799 loopback); 800 if (!eop) 801 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback); 802 803 return err; 804 } 805 806 static inline void enic_queue_wq_skb(struct enic *enic, 807 struct vnic_wq *wq, struct sk_buff *skb) 808 { 809 unsigned int mss = skb_shinfo(skb)->gso_size; 810 unsigned int vlan_tag = 0; 811 int vlan_tag_insert = 0; 812 int loopback = 0; 813 int err; 814 815 if (skb_vlan_tag_present(skb)) { 816 /* VLAN tag from trunking driver */ 817 vlan_tag_insert = 1; 818 vlan_tag = skb_vlan_tag_get(skb); 819 } else if (enic->loop_enable) { 820 vlan_tag = enic->loop_tag; 821 loopback = 1; 822 } 823 824 if (mss) 825 err = enic_queue_wq_skb_tso(enic, wq, skb, mss, 826 vlan_tag_insert, vlan_tag, 827 loopback); 828 else if (skb->encapsulation) 829 err = enic_queue_wq_skb_encap(enic, wq, skb, vlan_tag_insert, 830 vlan_tag, loopback); 831 else if (skb->ip_summed == CHECKSUM_PARTIAL) 832 err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert, 833 vlan_tag, loopback); 834 else 835 err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert, 836 vlan_tag, loopback); 837 if (unlikely(err)) { 838 struct vnic_wq_buf *buf; 839 840 buf = wq->to_use->prev; 841 /* while not EOP of previous pkt && queue not empty. 842 * For all non EOP bufs, os_buf is NULL. 843 */ 844 while (!buf->os_buf && (buf->next != wq->to_clean)) { 845 enic_free_wq_buf(wq, buf); 846 wq->ring.desc_avail++; 847 buf = buf->prev; 848 } 849 wq->to_use = buf->next; 850 dev_kfree_skb(skb); 851 } 852 } 853 854 /* netif_tx_lock held, process context with BHs disabled, or BH */ 855 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb, 856 struct net_device *netdev) 857 { 858 struct enic *enic = netdev_priv(netdev); 859 struct vnic_wq *wq; 860 unsigned int txq_map; 861 struct netdev_queue *txq; 862 863 if (skb->len <= 0) { 864 dev_kfree_skb_any(skb); 865 return NETDEV_TX_OK; 866 } 867 868 txq_map = skb_get_queue_mapping(skb) % enic->wq_count; 869 wq = &enic->wq[txq_map]; 870 txq = netdev_get_tx_queue(netdev, txq_map); 871 872 /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs, 873 * which is very likely. In the off chance it's going to take 874 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb. 875 */ 876 877 if (skb_shinfo(skb)->gso_size == 0 && 878 skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC && 879 skb_linearize(skb)) { 880 dev_kfree_skb_any(skb); 881 return NETDEV_TX_OK; 882 } 883 884 spin_lock(&enic->wq_lock[txq_map]); 885 886 if (vnic_wq_desc_avail(wq) < 887 skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) { 888 netif_tx_stop_queue(txq); 889 /* This is a hard error, log it */ 890 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n"); 891 spin_unlock(&enic->wq_lock[txq_map]); 892 return NETDEV_TX_BUSY; 893 } 894 895 enic_queue_wq_skb(enic, wq, skb); 896 897 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS) 898 netif_tx_stop_queue(txq); 899 skb_tx_timestamp(skb); 900 if (!skb->xmit_more || netif_xmit_stopped(txq)) 901 vnic_wq_doorbell(wq); 902 903 spin_unlock(&enic->wq_lock[txq_map]); 904 905 return NETDEV_TX_OK; 906 } 907 908 /* dev_base_lock rwlock held, nominally process context */ 909 static void enic_get_stats(struct net_device *netdev, 910 struct rtnl_link_stats64 *net_stats) 911 { 912 struct enic *enic = netdev_priv(netdev); 913 struct vnic_stats *stats; 914 int err; 915 916 err = enic_dev_stats_dump(enic, &stats); 917 /* return only when pci_zalloc_consistent fails in vnic_dev_stats_dump 918 * For other failures, like devcmd failure, we return previously 919 * recorded stats. 920 */ 921 if (err == -ENOMEM) 922 return; 923 924 net_stats->tx_packets = stats->tx.tx_frames_ok; 925 net_stats->tx_bytes = stats->tx.tx_bytes_ok; 926 net_stats->tx_errors = stats->tx.tx_errors; 927 net_stats->tx_dropped = stats->tx.tx_drops; 928 929 net_stats->rx_packets = stats->rx.rx_frames_ok; 930 net_stats->rx_bytes = stats->rx.rx_bytes_ok; 931 net_stats->rx_errors = stats->rx.rx_errors; 932 net_stats->multicast = stats->rx.rx_multicast_frames_ok; 933 net_stats->rx_over_errors = enic->rq_truncated_pkts; 934 net_stats->rx_crc_errors = enic->rq_bad_fcs; 935 net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop; 936 } 937 938 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr) 939 { 940 struct enic *enic = netdev_priv(netdev); 941 942 if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) { 943 unsigned int mc_count = netdev_mc_count(netdev); 944 945 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n", 946 ENIC_MULTICAST_PERFECT_FILTERS, mc_count); 947 948 return -ENOSPC; 949 } 950 951 enic_dev_add_addr(enic, mc_addr); 952 enic->mc_count++; 953 954 return 0; 955 } 956 957 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr) 958 { 959 struct enic *enic = netdev_priv(netdev); 960 961 enic_dev_del_addr(enic, mc_addr); 962 enic->mc_count--; 963 964 return 0; 965 } 966 967 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr) 968 { 969 struct enic *enic = netdev_priv(netdev); 970 971 if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) { 972 unsigned int uc_count = netdev_uc_count(netdev); 973 974 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n", 975 ENIC_UNICAST_PERFECT_FILTERS, uc_count); 976 977 return -ENOSPC; 978 } 979 980 enic_dev_add_addr(enic, uc_addr); 981 enic->uc_count++; 982 983 return 0; 984 } 985 986 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr) 987 { 988 struct enic *enic = netdev_priv(netdev); 989 990 enic_dev_del_addr(enic, uc_addr); 991 enic->uc_count--; 992 993 return 0; 994 } 995 996 void enic_reset_addr_lists(struct enic *enic) 997 { 998 struct net_device *netdev = enic->netdev; 999 1000 __dev_uc_unsync(netdev, NULL); 1001 __dev_mc_unsync(netdev, NULL); 1002 1003 enic->mc_count = 0; 1004 enic->uc_count = 0; 1005 enic->flags = 0; 1006 } 1007 1008 static int enic_set_mac_addr(struct net_device *netdev, char *addr) 1009 { 1010 struct enic *enic = netdev_priv(netdev); 1011 1012 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) { 1013 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr)) 1014 return -EADDRNOTAVAIL; 1015 } else { 1016 if (!is_valid_ether_addr(addr)) 1017 return -EADDRNOTAVAIL; 1018 } 1019 1020 memcpy(netdev->dev_addr, addr, netdev->addr_len); 1021 1022 return 0; 1023 } 1024 1025 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p) 1026 { 1027 struct enic *enic = netdev_priv(netdev); 1028 struct sockaddr *saddr = p; 1029 char *addr = saddr->sa_data; 1030 int err; 1031 1032 if (netif_running(enic->netdev)) { 1033 err = enic_dev_del_station_addr(enic); 1034 if (err) 1035 return err; 1036 } 1037 1038 err = enic_set_mac_addr(netdev, addr); 1039 if (err) 1040 return err; 1041 1042 if (netif_running(enic->netdev)) { 1043 err = enic_dev_add_station_addr(enic); 1044 if (err) 1045 return err; 1046 } 1047 1048 return err; 1049 } 1050 1051 static int enic_set_mac_address(struct net_device *netdev, void *p) 1052 { 1053 struct sockaddr *saddr = p; 1054 char *addr = saddr->sa_data; 1055 struct enic *enic = netdev_priv(netdev); 1056 int err; 1057 1058 err = enic_dev_del_station_addr(enic); 1059 if (err) 1060 return err; 1061 1062 err = enic_set_mac_addr(netdev, addr); 1063 if (err) 1064 return err; 1065 1066 return enic_dev_add_station_addr(enic); 1067 } 1068 1069 /* netif_tx_lock held, BHs disabled */ 1070 static void enic_set_rx_mode(struct net_device *netdev) 1071 { 1072 struct enic *enic = netdev_priv(netdev); 1073 int directed = 1; 1074 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0; 1075 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0; 1076 int promisc = (netdev->flags & IFF_PROMISC) || 1077 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS; 1078 int allmulti = (netdev->flags & IFF_ALLMULTI) || 1079 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS; 1080 unsigned int flags = netdev->flags | 1081 (allmulti ? IFF_ALLMULTI : 0) | 1082 (promisc ? IFF_PROMISC : 0); 1083 1084 if (enic->flags != flags) { 1085 enic->flags = flags; 1086 enic_dev_packet_filter(enic, directed, 1087 multicast, broadcast, promisc, allmulti); 1088 } 1089 1090 if (!promisc) { 1091 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync); 1092 if (!allmulti) 1093 __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync); 1094 } 1095 } 1096 1097 /* netif_tx_lock held, BHs disabled */ 1098 static void enic_tx_timeout(struct net_device *netdev) 1099 { 1100 struct enic *enic = netdev_priv(netdev); 1101 schedule_work(&enic->tx_hang_reset); 1102 } 1103 1104 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac) 1105 { 1106 struct enic *enic = netdev_priv(netdev); 1107 struct enic_port_profile *pp; 1108 int err; 1109 1110 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1111 if (err) 1112 return err; 1113 1114 if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) { 1115 if (vf == PORT_SELF_VF) { 1116 memcpy(pp->vf_mac, mac, ETH_ALEN); 1117 return 0; 1118 } else { 1119 /* 1120 * For sriov vf's set the mac in hw 1121 */ 1122 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic, 1123 vnic_dev_set_mac_addr, mac); 1124 return enic_dev_status_to_errno(err); 1125 } 1126 } else 1127 return -EINVAL; 1128 } 1129 1130 static int enic_set_vf_port(struct net_device *netdev, int vf, 1131 struct nlattr *port[]) 1132 { 1133 struct enic *enic = netdev_priv(netdev); 1134 struct enic_port_profile prev_pp; 1135 struct enic_port_profile *pp; 1136 int err = 0, restore_pp = 1; 1137 1138 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1139 if (err) 1140 return err; 1141 1142 if (!port[IFLA_PORT_REQUEST]) 1143 return -EOPNOTSUPP; 1144 1145 memcpy(&prev_pp, pp, sizeof(*enic->pp)); 1146 memset(pp, 0, sizeof(*enic->pp)); 1147 1148 pp->set |= ENIC_SET_REQUEST; 1149 pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]); 1150 1151 if (port[IFLA_PORT_PROFILE]) { 1152 pp->set |= ENIC_SET_NAME; 1153 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]), 1154 PORT_PROFILE_MAX); 1155 } 1156 1157 if (port[IFLA_PORT_INSTANCE_UUID]) { 1158 pp->set |= ENIC_SET_INSTANCE; 1159 memcpy(pp->instance_uuid, 1160 nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX); 1161 } 1162 1163 if (port[IFLA_PORT_HOST_UUID]) { 1164 pp->set |= ENIC_SET_HOST; 1165 memcpy(pp->host_uuid, 1166 nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX); 1167 } 1168 1169 if (vf == PORT_SELF_VF) { 1170 /* Special case handling: mac came from IFLA_VF_MAC */ 1171 if (!is_zero_ether_addr(prev_pp.vf_mac)) 1172 memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN); 1173 1174 if (is_zero_ether_addr(netdev->dev_addr)) 1175 eth_hw_addr_random(netdev); 1176 } else { 1177 /* SR-IOV VF: get mac from adapter */ 1178 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic, 1179 vnic_dev_get_mac_addr, pp->mac_addr); 1180 if (err) { 1181 netdev_err(netdev, "Error getting mac for vf %d\n", vf); 1182 memcpy(pp, &prev_pp, sizeof(*pp)); 1183 return enic_dev_status_to_errno(err); 1184 } 1185 } 1186 1187 err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp); 1188 if (err) { 1189 if (restore_pp) { 1190 /* Things are still the way they were: Implicit 1191 * DISASSOCIATE failed 1192 */ 1193 memcpy(pp, &prev_pp, sizeof(*pp)); 1194 } else { 1195 memset(pp, 0, sizeof(*pp)); 1196 if (vf == PORT_SELF_VF) 1197 eth_zero_addr(netdev->dev_addr); 1198 } 1199 } else { 1200 /* Set flag to indicate that the port assoc/disassoc 1201 * request has been sent out to fw 1202 */ 1203 pp->set |= ENIC_PORT_REQUEST_APPLIED; 1204 1205 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */ 1206 if (pp->request == PORT_REQUEST_DISASSOCIATE) { 1207 eth_zero_addr(pp->mac_addr); 1208 if (vf == PORT_SELF_VF) 1209 eth_zero_addr(netdev->dev_addr); 1210 } 1211 } 1212 1213 if (vf == PORT_SELF_VF) 1214 eth_zero_addr(pp->vf_mac); 1215 1216 return err; 1217 } 1218 1219 static int enic_get_vf_port(struct net_device *netdev, int vf, 1220 struct sk_buff *skb) 1221 { 1222 struct enic *enic = netdev_priv(netdev); 1223 u16 response = PORT_PROFILE_RESPONSE_SUCCESS; 1224 struct enic_port_profile *pp; 1225 int err; 1226 1227 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1228 if (err) 1229 return err; 1230 1231 if (!(pp->set & ENIC_PORT_REQUEST_APPLIED)) 1232 return -ENODATA; 1233 1234 err = enic_process_get_pp_request(enic, vf, pp->request, &response); 1235 if (err) 1236 return err; 1237 1238 if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) || 1239 nla_put_u16(skb, IFLA_PORT_RESPONSE, response) || 1240 ((pp->set & ENIC_SET_NAME) && 1241 nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) || 1242 ((pp->set & ENIC_SET_INSTANCE) && 1243 nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX, 1244 pp->instance_uuid)) || 1245 ((pp->set & ENIC_SET_HOST) && 1246 nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid))) 1247 goto nla_put_failure; 1248 return 0; 1249 1250 nla_put_failure: 1251 return -EMSGSIZE; 1252 } 1253 1254 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf) 1255 { 1256 struct enic *enic = vnic_dev_priv(rq->vdev); 1257 1258 if (!buf->os_buf) 1259 return; 1260 1261 pci_unmap_single(enic->pdev, buf->dma_addr, 1262 buf->len, PCI_DMA_FROMDEVICE); 1263 dev_kfree_skb_any(buf->os_buf); 1264 buf->os_buf = NULL; 1265 } 1266 1267 static int enic_rq_alloc_buf(struct vnic_rq *rq) 1268 { 1269 struct enic *enic = vnic_dev_priv(rq->vdev); 1270 struct net_device *netdev = enic->netdev; 1271 struct sk_buff *skb; 1272 unsigned int len = netdev->mtu + VLAN_ETH_HLEN; 1273 unsigned int os_buf_index = 0; 1274 dma_addr_t dma_addr; 1275 struct vnic_rq_buf *buf = rq->to_use; 1276 1277 if (buf->os_buf) { 1278 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr, 1279 buf->len); 1280 1281 return 0; 1282 } 1283 skb = netdev_alloc_skb_ip_align(netdev, len); 1284 if (!skb) 1285 return -ENOMEM; 1286 1287 dma_addr = pci_map_single(enic->pdev, skb->data, len, 1288 PCI_DMA_FROMDEVICE); 1289 if (unlikely(enic_dma_map_check(enic, dma_addr))) { 1290 dev_kfree_skb(skb); 1291 return -ENOMEM; 1292 } 1293 1294 enic_queue_rq_desc(rq, skb, os_buf_index, 1295 dma_addr, len); 1296 1297 return 0; 1298 } 1299 1300 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size, 1301 u32 pkt_len) 1302 { 1303 if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len) 1304 pkt_size->large_pkt_bytes_cnt += pkt_len; 1305 else 1306 pkt_size->small_pkt_bytes_cnt += pkt_len; 1307 } 1308 1309 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb, 1310 struct vnic_rq_buf *buf, u16 len) 1311 { 1312 struct enic *enic = netdev_priv(netdev); 1313 struct sk_buff *new_skb; 1314 1315 if (len > enic->rx_copybreak) 1316 return false; 1317 new_skb = netdev_alloc_skb_ip_align(netdev, len); 1318 if (!new_skb) 1319 return false; 1320 pci_dma_sync_single_for_cpu(enic->pdev, buf->dma_addr, len, 1321 DMA_FROM_DEVICE); 1322 memcpy(new_skb->data, (*skb)->data, len); 1323 *skb = new_skb; 1324 1325 return true; 1326 } 1327 1328 static void enic_rq_indicate_buf(struct vnic_rq *rq, 1329 struct cq_desc *cq_desc, struct vnic_rq_buf *buf, 1330 int skipped, void *opaque) 1331 { 1332 struct enic *enic = vnic_dev_priv(rq->vdev); 1333 struct net_device *netdev = enic->netdev; 1334 struct sk_buff *skb; 1335 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1336 1337 u8 type, color, eop, sop, ingress_port, vlan_stripped; 1338 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof; 1339 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok; 1340 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc; 1341 u8 packet_error; 1342 u16 q_number, completed_index, bytes_written, vlan_tci, checksum; 1343 u32 rss_hash; 1344 bool outer_csum_ok = true, encap = false; 1345 1346 if (skipped) 1347 return; 1348 1349 skb = buf->os_buf; 1350 1351 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc, 1352 &type, &color, &q_number, &completed_index, 1353 &ingress_port, &fcoe, &eop, &sop, &rss_type, 1354 &csum_not_calc, &rss_hash, &bytes_written, 1355 &packet_error, &vlan_stripped, &vlan_tci, &checksum, 1356 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error, 1357 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp, 1358 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment, 1359 &fcs_ok); 1360 1361 if (packet_error) { 1362 1363 if (!fcs_ok) { 1364 if (bytes_written > 0) 1365 enic->rq_bad_fcs++; 1366 else if (bytes_written == 0) 1367 enic->rq_truncated_pkts++; 1368 } 1369 1370 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len, 1371 PCI_DMA_FROMDEVICE); 1372 dev_kfree_skb_any(skb); 1373 buf->os_buf = NULL; 1374 1375 return; 1376 } 1377 1378 if (eop && bytes_written > 0) { 1379 1380 /* Good receive 1381 */ 1382 1383 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) { 1384 buf->os_buf = NULL; 1385 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len, 1386 PCI_DMA_FROMDEVICE); 1387 } 1388 prefetch(skb->data - NET_IP_ALIGN); 1389 1390 skb_put(skb, bytes_written); 1391 skb->protocol = eth_type_trans(skb, netdev); 1392 skb_record_rx_queue(skb, q_number); 1393 if ((netdev->features & NETIF_F_RXHASH) && rss_hash && 1394 (type == 3)) { 1395 switch (rss_type) { 1396 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4: 1397 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6: 1398 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX: 1399 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4); 1400 break; 1401 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4: 1402 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6: 1403 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX: 1404 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3); 1405 break; 1406 } 1407 } 1408 if (enic->vxlan.vxlan_udp_port_number) { 1409 switch (enic->vxlan.patch_level) { 1410 case 0: 1411 if (fcoe) { 1412 encap = true; 1413 outer_csum_ok = fcoe_fc_crc_ok; 1414 } 1415 break; 1416 case 2: 1417 if ((type == 7) && 1418 (rss_hash & BIT(0))) { 1419 encap = true; 1420 outer_csum_ok = (rss_hash & BIT(1)) && 1421 (rss_hash & BIT(2)); 1422 } 1423 break; 1424 } 1425 } 1426 1427 /* Hardware does not provide whole packet checksum. It only 1428 * provides pseudo checksum. Since hw validates the packet 1429 * checksum but not provide us the checksum value. use 1430 * CHECSUM_UNNECESSARY. 1431 * 1432 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is 1433 * inner csum_ok. outer_csum_ok is set by hw when outer udp 1434 * csum is correct or is zero. 1435 */ 1436 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc && 1437 tcp_udp_csum_ok && ipv4_csum_ok && outer_csum_ok) { 1438 skb->ip_summed = CHECKSUM_UNNECESSARY; 1439 skb->csum_level = encap; 1440 } 1441 1442 if (vlan_stripped) 1443 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci); 1444 1445 skb_mark_napi_id(skb, &enic->napi[rq->index]); 1446 if (!(netdev->features & NETIF_F_GRO)) 1447 netif_receive_skb(skb); 1448 else 1449 napi_gro_receive(&enic->napi[q_number], skb); 1450 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1451 enic_intr_update_pkt_size(&cq->pkt_size_counter, 1452 bytes_written); 1453 } else { 1454 1455 /* Buffer overflow 1456 */ 1457 1458 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len, 1459 PCI_DMA_FROMDEVICE); 1460 dev_kfree_skb_any(skb); 1461 buf->os_buf = NULL; 1462 } 1463 } 1464 1465 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc, 1466 u8 type, u16 q_number, u16 completed_index, void *opaque) 1467 { 1468 struct enic *enic = vnic_dev_priv(vdev); 1469 1470 vnic_rq_service(&enic->rq[q_number], cq_desc, 1471 completed_index, VNIC_RQ_RETURN_DESC, 1472 enic_rq_indicate_buf, opaque); 1473 1474 return 0; 1475 } 1476 1477 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq) 1478 { 1479 unsigned int intr = enic_msix_rq_intr(enic, rq->index); 1480 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1481 u32 timer = cq->tobe_rx_coal_timeval; 1482 1483 if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) { 1484 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer); 1485 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval; 1486 } 1487 } 1488 1489 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq) 1490 { 1491 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting; 1492 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1493 struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter; 1494 int index; 1495 u32 timer; 1496 u32 range_start; 1497 u32 traffic; 1498 u64 delta; 1499 ktime_t now = ktime_get(); 1500 1501 delta = ktime_us_delta(now, cq->prev_ts); 1502 if (delta < ENIC_AIC_TS_BREAK) 1503 return; 1504 cq->prev_ts = now; 1505 1506 traffic = pkt_size_counter->large_pkt_bytes_cnt + 1507 pkt_size_counter->small_pkt_bytes_cnt; 1508 /* The table takes Mbps 1509 * traffic *= 8 => bits 1510 * traffic *= (10^6 / delta) => bps 1511 * traffic /= 10^6 => Mbps 1512 * 1513 * Combining, traffic *= (8 / delta) 1514 */ 1515 1516 traffic <<= 3; 1517 traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta; 1518 1519 for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++) 1520 if (traffic < mod_table[index].rx_rate) 1521 break; 1522 range_start = (pkt_size_counter->small_pkt_bytes_cnt > 1523 pkt_size_counter->large_pkt_bytes_cnt << 1) ? 1524 rx_coal->small_pkt_range_start : 1525 rx_coal->large_pkt_range_start; 1526 timer = range_start + ((rx_coal->range_end - range_start) * 1527 mod_table[index].range_percent / 100); 1528 /* Damping */ 1529 cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1; 1530 1531 pkt_size_counter->large_pkt_bytes_cnt = 0; 1532 pkt_size_counter->small_pkt_bytes_cnt = 0; 1533 } 1534 1535 static int enic_poll(struct napi_struct *napi, int budget) 1536 { 1537 struct net_device *netdev = napi->dev; 1538 struct enic *enic = netdev_priv(netdev); 1539 unsigned int cq_rq = enic_cq_rq(enic, 0); 1540 unsigned int cq_wq = enic_cq_wq(enic, 0); 1541 unsigned int intr = enic_legacy_io_intr(); 1542 unsigned int rq_work_to_do = budget; 1543 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET; 1544 unsigned int work_done, rq_work_done = 0, wq_work_done; 1545 int err; 1546 1547 wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do, 1548 enic_wq_service, NULL); 1549 1550 if (budget > 0) 1551 rq_work_done = vnic_cq_service(&enic->cq[cq_rq], 1552 rq_work_to_do, enic_rq_service, NULL); 1553 1554 /* Accumulate intr event credits for this polling 1555 * cycle. An intr event is the completion of a 1556 * a WQ or RQ packet. 1557 */ 1558 1559 work_done = rq_work_done + wq_work_done; 1560 1561 if (work_done > 0) 1562 vnic_intr_return_credits(&enic->intr[intr], 1563 work_done, 1564 0 /* don't unmask intr */, 1565 0 /* don't reset intr timer */); 1566 1567 err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf); 1568 1569 /* Buffer allocation failed. Stay in polling 1570 * mode so we can try to fill the ring again. 1571 */ 1572 1573 if (err) 1574 rq_work_done = rq_work_to_do; 1575 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1576 /* Call the function which refreshes the intr coalescing timer 1577 * value based on the traffic. 1578 */ 1579 enic_calc_int_moderation(enic, &enic->rq[0]); 1580 1581 if ((rq_work_done < budget) && napi_complete_done(napi, rq_work_done)) { 1582 1583 /* Some work done, but not enough to stay in polling, 1584 * exit polling 1585 */ 1586 1587 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1588 enic_set_int_moderation(enic, &enic->rq[0]); 1589 vnic_intr_unmask(&enic->intr[intr]); 1590 } 1591 1592 return rq_work_done; 1593 } 1594 1595 #ifdef CONFIG_RFS_ACCEL 1596 static void enic_free_rx_cpu_rmap(struct enic *enic) 1597 { 1598 free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap); 1599 enic->netdev->rx_cpu_rmap = NULL; 1600 } 1601 1602 static void enic_set_rx_cpu_rmap(struct enic *enic) 1603 { 1604 int i, res; 1605 1606 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) { 1607 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count); 1608 if (unlikely(!enic->netdev->rx_cpu_rmap)) 1609 return; 1610 for (i = 0; i < enic->rq_count; i++) { 1611 res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap, 1612 enic->msix_entry[i].vector); 1613 if (unlikely(res)) { 1614 enic_free_rx_cpu_rmap(enic); 1615 return; 1616 } 1617 } 1618 } 1619 } 1620 1621 #else 1622 1623 static void enic_free_rx_cpu_rmap(struct enic *enic) 1624 { 1625 } 1626 1627 static void enic_set_rx_cpu_rmap(struct enic *enic) 1628 { 1629 } 1630 1631 #endif /* CONFIG_RFS_ACCEL */ 1632 1633 static int enic_poll_msix_wq(struct napi_struct *napi, int budget) 1634 { 1635 struct net_device *netdev = napi->dev; 1636 struct enic *enic = netdev_priv(netdev); 1637 unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count; 1638 struct vnic_wq *wq = &enic->wq[wq_index]; 1639 unsigned int cq; 1640 unsigned int intr; 1641 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET; 1642 unsigned int wq_work_done; 1643 unsigned int wq_irq; 1644 1645 wq_irq = wq->index; 1646 cq = enic_cq_wq(enic, wq_irq); 1647 intr = enic_msix_wq_intr(enic, wq_irq); 1648 wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do, 1649 enic_wq_service, NULL); 1650 1651 vnic_intr_return_credits(&enic->intr[intr], wq_work_done, 1652 0 /* don't unmask intr */, 1653 1 /* reset intr timer */); 1654 if (!wq_work_done) { 1655 napi_complete(napi); 1656 vnic_intr_unmask(&enic->intr[intr]); 1657 return 0; 1658 } 1659 1660 return budget; 1661 } 1662 1663 static int enic_poll_msix_rq(struct napi_struct *napi, int budget) 1664 { 1665 struct net_device *netdev = napi->dev; 1666 struct enic *enic = netdev_priv(netdev); 1667 unsigned int rq = (napi - &enic->napi[0]); 1668 unsigned int cq = enic_cq_rq(enic, rq); 1669 unsigned int intr = enic_msix_rq_intr(enic, rq); 1670 unsigned int work_to_do = budget; 1671 unsigned int work_done = 0; 1672 int err; 1673 1674 /* Service RQ 1675 */ 1676 1677 if (budget > 0) 1678 work_done = vnic_cq_service(&enic->cq[cq], 1679 work_to_do, enic_rq_service, NULL); 1680 1681 /* Return intr event credits for this polling 1682 * cycle. An intr event is the completion of a 1683 * RQ packet. 1684 */ 1685 1686 if (work_done > 0) 1687 vnic_intr_return_credits(&enic->intr[intr], 1688 work_done, 1689 0 /* don't unmask intr */, 1690 0 /* don't reset intr timer */); 1691 1692 err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf); 1693 1694 /* Buffer allocation failed. Stay in polling mode 1695 * so we can try to fill the ring again. 1696 */ 1697 1698 if (err) 1699 work_done = work_to_do; 1700 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1701 /* Call the function which refreshes the intr coalescing timer 1702 * value based on the traffic. 1703 */ 1704 enic_calc_int_moderation(enic, &enic->rq[rq]); 1705 1706 if ((work_done < budget) && napi_complete_done(napi, work_done)) { 1707 1708 /* Some work done, but not enough to stay in polling, 1709 * exit polling 1710 */ 1711 1712 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1713 enic_set_int_moderation(enic, &enic->rq[rq]); 1714 vnic_intr_unmask(&enic->intr[intr]); 1715 } 1716 1717 return work_done; 1718 } 1719 1720 static void enic_notify_timer(struct timer_list *t) 1721 { 1722 struct enic *enic = from_timer(enic, t, notify_timer); 1723 1724 enic_notify_check(enic); 1725 1726 mod_timer(&enic->notify_timer, 1727 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD)); 1728 } 1729 1730 static void enic_free_intr(struct enic *enic) 1731 { 1732 struct net_device *netdev = enic->netdev; 1733 unsigned int i; 1734 1735 enic_free_rx_cpu_rmap(enic); 1736 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1737 case VNIC_DEV_INTR_MODE_INTX: 1738 free_irq(enic->pdev->irq, netdev); 1739 break; 1740 case VNIC_DEV_INTR_MODE_MSI: 1741 free_irq(enic->pdev->irq, enic); 1742 break; 1743 case VNIC_DEV_INTR_MODE_MSIX: 1744 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) 1745 if (enic->msix[i].requested) 1746 free_irq(enic->msix_entry[i].vector, 1747 enic->msix[i].devid); 1748 break; 1749 default: 1750 break; 1751 } 1752 } 1753 1754 static int enic_request_intr(struct enic *enic) 1755 { 1756 struct net_device *netdev = enic->netdev; 1757 unsigned int i, intr; 1758 int err = 0; 1759 1760 enic_set_rx_cpu_rmap(enic); 1761 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1762 1763 case VNIC_DEV_INTR_MODE_INTX: 1764 1765 err = request_irq(enic->pdev->irq, enic_isr_legacy, 1766 IRQF_SHARED, netdev->name, netdev); 1767 break; 1768 1769 case VNIC_DEV_INTR_MODE_MSI: 1770 1771 err = request_irq(enic->pdev->irq, enic_isr_msi, 1772 0, netdev->name, enic); 1773 break; 1774 1775 case VNIC_DEV_INTR_MODE_MSIX: 1776 1777 for (i = 0; i < enic->rq_count; i++) { 1778 intr = enic_msix_rq_intr(enic, i); 1779 snprintf(enic->msix[intr].devname, 1780 sizeof(enic->msix[intr].devname), 1781 "%s-rx-%u", netdev->name, i); 1782 enic->msix[intr].isr = enic_isr_msix; 1783 enic->msix[intr].devid = &enic->napi[i]; 1784 } 1785 1786 for (i = 0; i < enic->wq_count; i++) { 1787 int wq = enic_cq_wq(enic, i); 1788 1789 intr = enic_msix_wq_intr(enic, i); 1790 snprintf(enic->msix[intr].devname, 1791 sizeof(enic->msix[intr].devname), 1792 "%s-tx-%u", netdev->name, i); 1793 enic->msix[intr].isr = enic_isr_msix; 1794 enic->msix[intr].devid = &enic->napi[wq]; 1795 } 1796 1797 intr = enic_msix_err_intr(enic); 1798 snprintf(enic->msix[intr].devname, 1799 sizeof(enic->msix[intr].devname), 1800 "%s-err", netdev->name); 1801 enic->msix[intr].isr = enic_isr_msix_err; 1802 enic->msix[intr].devid = enic; 1803 1804 intr = enic_msix_notify_intr(enic); 1805 snprintf(enic->msix[intr].devname, 1806 sizeof(enic->msix[intr].devname), 1807 "%s-notify", netdev->name); 1808 enic->msix[intr].isr = enic_isr_msix_notify; 1809 enic->msix[intr].devid = enic; 1810 1811 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) 1812 enic->msix[i].requested = 0; 1813 1814 for (i = 0; i < enic->intr_count; i++) { 1815 err = request_irq(enic->msix_entry[i].vector, 1816 enic->msix[i].isr, 0, 1817 enic->msix[i].devname, 1818 enic->msix[i].devid); 1819 if (err) { 1820 enic_free_intr(enic); 1821 break; 1822 } 1823 enic->msix[i].requested = 1; 1824 } 1825 1826 break; 1827 1828 default: 1829 break; 1830 } 1831 1832 return err; 1833 } 1834 1835 static void enic_synchronize_irqs(struct enic *enic) 1836 { 1837 unsigned int i; 1838 1839 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1840 case VNIC_DEV_INTR_MODE_INTX: 1841 case VNIC_DEV_INTR_MODE_MSI: 1842 synchronize_irq(enic->pdev->irq); 1843 break; 1844 case VNIC_DEV_INTR_MODE_MSIX: 1845 for (i = 0; i < enic->intr_count; i++) 1846 synchronize_irq(enic->msix_entry[i].vector); 1847 break; 1848 default: 1849 break; 1850 } 1851 } 1852 1853 static void enic_set_rx_coal_setting(struct enic *enic) 1854 { 1855 unsigned int speed; 1856 int index = -1; 1857 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting; 1858 1859 /* 1. Read the link speed from fw 1860 * 2. Pick the default range for the speed 1861 * 3. Update it in enic->rx_coalesce_setting 1862 */ 1863 speed = vnic_dev_port_speed(enic->vdev); 1864 if (ENIC_LINK_SPEED_10G < speed) 1865 index = ENIC_LINK_40G_INDEX; 1866 else if (ENIC_LINK_SPEED_4G < speed) 1867 index = ENIC_LINK_10G_INDEX; 1868 else 1869 index = ENIC_LINK_4G_INDEX; 1870 1871 rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start; 1872 rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start; 1873 rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END; 1874 1875 /* Start with the value provided by UCSM */ 1876 for (index = 0; index < enic->rq_count; index++) 1877 enic->cq[index].cur_rx_coal_timeval = 1878 enic->config.intr_timer_usec; 1879 1880 rx_coal->use_adaptive_rx_coalesce = 1; 1881 } 1882 1883 static int enic_dev_notify_set(struct enic *enic) 1884 { 1885 int err; 1886 1887 spin_lock_bh(&enic->devcmd_lock); 1888 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1889 case VNIC_DEV_INTR_MODE_INTX: 1890 err = vnic_dev_notify_set(enic->vdev, 1891 enic_legacy_notify_intr()); 1892 break; 1893 case VNIC_DEV_INTR_MODE_MSIX: 1894 err = vnic_dev_notify_set(enic->vdev, 1895 enic_msix_notify_intr(enic)); 1896 break; 1897 default: 1898 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */); 1899 break; 1900 } 1901 spin_unlock_bh(&enic->devcmd_lock); 1902 1903 return err; 1904 } 1905 1906 static void enic_notify_timer_start(struct enic *enic) 1907 { 1908 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1909 case VNIC_DEV_INTR_MODE_MSI: 1910 mod_timer(&enic->notify_timer, jiffies); 1911 break; 1912 default: 1913 /* Using intr for notification for INTx/MSI-X */ 1914 break; 1915 } 1916 } 1917 1918 /* rtnl lock is held, process context */ 1919 static int enic_open(struct net_device *netdev) 1920 { 1921 struct enic *enic = netdev_priv(netdev); 1922 unsigned int i; 1923 int err; 1924 1925 err = enic_request_intr(enic); 1926 if (err) { 1927 netdev_err(netdev, "Unable to request irq.\n"); 1928 return err; 1929 } 1930 enic_init_affinity_hint(enic); 1931 enic_set_affinity_hint(enic); 1932 1933 err = enic_dev_notify_set(enic); 1934 if (err) { 1935 netdev_err(netdev, 1936 "Failed to alloc notify buffer, aborting.\n"); 1937 goto err_out_free_intr; 1938 } 1939 1940 for (i = 0; i < enic->rq_count; i++) { 1941 /* enable rq before updating rq desc */ 1942 vnic_rq_enable(&enic->rq[i]); 1943 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf); 1944 /* Need at least one buffer on ring to get going */ 1945 if (vnic_rq_desc_used(&enic->rq[i]) == 0) { 1946 netdev_err(netdev, "Unable to alloc receive buffers\n"); 1947 err = -ENOMEM; 1948 goto err_out_free_rq; 1949 } 1950 } 1951 1952 for (i = 0; i < enic->wq_count; i++) 1953 vnic_wq_enable(&enic->wq[i]); 1954 1955 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic)) 1956 enic_dev_add_station_addr(enic); 1957 1958 enic_set_rx_mode(netdev); 1959 1960 netif_tx_wake_all_queues(netdev); 1961 1962 for (i = 0; i < enic->rq_count; i++) 1963 napi_enable(&enic->napi[i]); 1964 1965 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) 1966 for (i = 0; i < enic->wq_count; i++) 1967 napi_enable(&enic->napi[enic_cq_wq(enic, i)]); 1968 enic_dev_enable(enic); 1969 1970 for (i = 0; i < enic->intr_count; i++) 1971 vnic_intr_unmask(&enic->intr[i]); 1972 1973 enic_notify_timer_start(enic); 1974 enic_rfs_flw_tbl_init(enic); 1975 1976 return 0; 1977 1978 err_out_free_rq: 1979 for (i = 0; i < enic->rq_count; i++) { 1980 err = vnic_rq_disable(&enic->rq[i]); 1981 if (err) 1982 return err; 1983 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf); 1984 } 1985 enic_dev_notify_unset(enic); 1986 err_out_free_intr: 1987 enic_unset_affinity_hint(enic); 1988 enic_free_intr(enic); 1989 1990 return err; 1991 } 1992 1993 /* rtnl lock is held, process context */ 1994 static int enic_stop(struct net_device *netdev) 1995 { 1996 struct enic *enic = netdev_priv(netdev); 1997 unsigned int i; 1998 int err; 1999 2000 for (i = 0; i < enic->intr_count; i++) { 2001 vnic_intr_mask(&enic->intr[i]); 2002 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */ 2003 } 2004 2005 enic_synchronize_irqs(enic); 2006 2007 del_timer_sync(&enic->notify_timer); 2008 enic_rfs_flw_tbl_free(enic); 2009 2010 enic_dev_disable(enic); 2011 2012 for (i = 0; i < enic->rq_count; i++) 2013 napi_disable(&enic->napi[i]); 2014 2015 netif_carrier_off(netdev); 2016 netif_tx_disable(netdev); 2017 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) 2018 for (i = 0; i < enic->wq_count; i++) 2019 napi_disable(&enic->napi[enic_cq_wq(enic, i)]); 2020 2021 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic)) 2022 enic_dev_del_station_addr(enic); 2023 2024 for (i = 0; i < enic->wq_count; i++) { 2025 err = vnic_wq_disable(&enic->wq[i]); 2026 if (err) 2027 return err; 2028 } 2029 for (i = 0; i < enic->rq_count; i++) { 2030 err = vnic_rq_disable(&enic->rq[i]); 2031 if (err) 2032 return err; 2033 } 2034 2035 enic_dev_notify_unset(enic); 2036 enic_unset_affinity_hint(enic); 2037 enic_free_intr(enic); 2038 2039 for (i = 0; i < enic->wq_count; i++) 2040 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf); 2041 for (i = 0; i < enic->rq_count; i++) 2042 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf); 2043 for (i = 0; i < enic->cq_count; i++) 2044 vnic_cq_clean(&enic->cq[i]); 2045 for (i = 0; i < enic->intr_count; i++) 2046 vnic_intr_clean(&enic->intr[i]); 2047 2048 return 0; 2049 } 2050 2051 static int enic_change_mtu(struct net_device *netdev, int new_mtu) 2052 { 2053 struct enic *enic = netdev_priv(netdev); 2054 int running = netif_running(netdev); 2055 2056 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) 2057 return -EOPNOTSUPP; 2058 2059 if (running) 2060 enic_stop(netdev); 2061 2062 netdev->mtu = new_mtu; 2063 2064 if (netdev->mtu > enic->port_mtu) 2065 netdev_warn(netdev, 2066 "interface MTU (%d) set higher than port MTU (%d)\n", 2067 netdev->mtu, enic->port_mtu); 2068 2069 if (running) 2070 enic_open(netdev); 2071 2072 return 0; 2073 } 2074 2075 static void enic_change_mtu_work(struct work_struct *work) 2076 { 2077 struct enic *enic = container_of(work, struct enic, change_mtu_work); 2078 struct net_device *netdev = enic->netdev; 2079 int new_mtu = vnic_dev_mtu(enic->vdev); 2080 int err; 2081 unsigned int i; 2082 2083 new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu)); 2084 2085 rtnl_lock(); 2086 2087 /* Stop RQ */ 2088 del_timer_sync(&enic->notify_timer); 2089 2090 for (i = 0; i < enic->rq_count; i++) 2091 napi_disable(&enic->napi[i]); 2092 2093 vnic_intr_mask(&enic->intr[0]); 2094 enic_synchronize_irqs(enic); 2095 err = vnic_rq_disable(&enic->rq[0]); 2096 if (err) { 2097 rtnl_unlock(); 2098 netdev_err(netdev, "Unable to disable RQ.\n"); 2099 return; 2100 } 2101 vnic_rq_clean(&enic->rq[0], enic_free_rq_buf); 2102 vnic_cq_clean(&enic->cq[0]); 2103 vnic_intr_clean(&enic->intr[0]); 2104 2105 /* Fill RQ with new_mtu-sized buffers */ 2106 netdev->mtu = new_mtu; 2107 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf); 2108 /* Need at least one buffer on ring to get going */ 2109 if (vnic_rq_desc_used(&enic->rq[0]) == 0) { 2110 rtnl_unlock(); 2111 netdev_err(netdev, "Unable to alloc receive buffers.\n"); 2112 return; 2113 } 2114 2115 /* Start RQ */ 2116 vnic_rq_enable(&enic->rq[0]); 2117 napi_enable(&enic->napi[0]); 2118 vnic_intr_unmask(&enic->intr[0]); 2119 enic_notify_timer_start(enic); 2120 2121 rtnl_unlock(); 2122 2123 netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu); 2124 } 2125 2126 #ifdef CONFIG_NET_POLL_CONTROLLER 2127 static void enic_poll_controller(struct net_device *netdev) 2128 { 2129 struct enic *enic = netdev_priv(netdev); 2130 struct vnic_dev *vdev = enic->vdev; 2131 unsigned int i, intr; 2132 2133 switch (vnic_dev_get_intr_mode(vdev)) { 2134 case VNIC_DEV_INTR_MODE_MSIX: 2135 for (i = 0; i < enic->rq_count; i++) { 2136 intr = enic_msix_rq_intr(enic, i); 2137 enic_isr_msix(enic->msix_entry[intr].vector, 2138 &enic->napi[i]); 2139 } 2140 2141 for (i = 0; i < enic->wq_count; i++) { 2142 intr = enic_msix_wq_intr(enic, i); 2143 enic_isr_msix(enic->msix_entry[intr].vector, 2144 &enic->napi[enic_cq_wq(enic, i)]); 2145 } 2146 2147 break; 2148 case VNIC_DEV_INTR_MODE_MSI: 2149 enic_isr_msi(enic->pdev->irq, enic); 2150 break; 2151 case VNIC_DEV_INTR_MODE_INTX: 2152 enic_isr_legacy(enic->pdev->irq, netdev); 2153 break; 2154 default: 2155 break; 2156 } 2157 } 2158 #endif 2159 2160 static int enic_dev_wait(struct vnic_dev *vdev, 2161 int (*start)(struct vnic_dev *, int), 2162 int (*finished)(struct vnic_dev *, int *), 2163 int arg) 2164 { 2165 unsigned long time; 2166 int done; 2167 int err; 2168 2169 BUG_ON(in_interrupt()); 2170 2171 err = start(vdev, arg); 2172 if (err) 2173 return err; 2174 2175 /* Wait for func to complete...2 seconds max 2176 */ 2177 2178 time = jiffies + (HZ * 2); 2179 do { 2180 2181 err = finished(vdev, &done); 2182 if (err) 2183 return err; 2184 2185 if (done) 2186 return 0; 2187 2188 schedule_timeout_uninterruptible(HZ / 10); 2189 2190 } while (time_after(time, jiffies)); 2191 2192 return -ETIMEDOUT; 2193 } 2194 2195 static int enic_dev_open(struct enic *enic) 2196 { 2197 int err; 2198 u32 flags = CMD_OPENF_IG_DESCCACHE; 2199 2200 err = enic_dev_wait(enic->vdev, vnic_dev_open, 2201 vnic_dev_open_done, flags); 2202 if (err) 2203 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n", 2204 err); 2205 2206 return err; 2207 } 2208 2209 static int enic_dev_soft_reset(struct enic *enic) 2210 { 2211 int err; 2212 2213 err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset, 2214 vnic_dev_soft_reset_done, 0); 2215 if (err) 2216 netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n", 2217 err); 2218 2219 return err; 2220 } 2221 2222 static int enic_dev_hang_reset(struct enic *enic) 2223 { 2224 int err; 2225 2226 err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset, 2227 vnic_dev_hang_reset_done, 0); 2228 if (err) 2229 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n", 2230 err); 2231 2232 return err; 2233 } 2234 2235 int __enic_set_rsskey(struct enic *enic) 2236 { 2237 union vnic_rss_key *rss_key_buf_va; 2238 dma_addr_t rss_key_buf_pa; 2239 int i, kidx, bidx, err; 2240 2241 rss_key_buf_va = pci_zalloc_consistent(enic->pdev, 2242 sizeof(union vnic_rss_key), 2243 &rss_key_buf_pa); 2244 if (!rss_key_buf_va) 2245 return -ENOMEM; 2246 2247 for (i = 0; i < ENIC_RSS_LEN; i++) { 2248 kidx = i / ENIC_RSS_BYTES_PER_KEY; 2249 bidx = i % ENIC_RSS_BYTES_PER_KEY; 2250 rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i]; 2251 } 2252 spin_lock_bh(&enic->devcmd_lock); 2253 err = enic_set_rss_key(enic, 2254 rss_key_buf_pa, 2255 sizeof(union vnic_rss_key)); 2256 spin_unlock_bh(&enic->devcmd_lock); 2257 2258 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key), 2259 rss_key_buf_va, rss_key_buf_pa); 2260 2261 return err; 2262 } 2263 2264 static int enic_set_rsskey(struct enic *enic) 2265 { 2266 netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN); 2267 2268 return __enic_set_rsskey(enic); 2269 } 2270 2271 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits) 2272 { 2273 dma_addr_t rss_cpu_buf_pa; 2274 union vnic_rss_cpu *rss_cpu_buf_va = NULL; 2275 unsigned int i; 2276 int err; 2277 2278 rss_cpu_buf_va = pci_alloc_consistent(enic->pdev, 2279 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa); 2280 if (!rss_cpu_buf_va) 2281 return -ENOMEM; 2282 2283 for (i = 0; i < (1 << rss_hash_bits); i++) 2284 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count; 2285 2286 spin_lock_bh(&enic->devcmd_lock); 2287 err = enic_set_rss_cpu(enic, 2288 rss_cpu_buf_pa, 2289 sizeof(union vnic_rss_cpu)); 2290 spin_unlock_bh(&enic->devcmd_lock); 2291 2292 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu), 2293 rss_cpu_buf_va, rss_cpu_buf_pa); 2294 2295 return err; 2296 } 2297 2298 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu, 2299 u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable) 2300 { 2301 const u8 tso_ipid_split_en = 0; 2302 const u8 ig_vlan_strip_en = 1; 2303 int err; 2304 2305 /* Enable VLAN tag stripping. 2306 */ 2307 2308 spin_lock_bh(&enic->devcmd_lock); 2309 err = enic_set_nic_cfg(enic, 2310 rss_default_cpu, rss_hash_type, 2311 rss_hash_bits, rss_base_cpu, 2312 rss_enable, tso_ipid_split_en, 2313 ig_vlan_strip_en); 2314 spin_unlock_bh(&enic->devcmd_lock); 2315 2316 return err; 2317 } 2318 2319 static int enic_set_rss_nic_cfg(struct enic *enic) 2320 { 2321 struct device *dev = enic_get_dev(enic); 2322 const u8 rss_default_cpu = 0; 2323 u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 | 2324 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 | 2325 NIC_CFG_RSS_HASH_TYPE_IPV6 | 2326 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6; 2327 const u8 rss_hash_bits = 7; 2328 const u8 rss_base_cpu = 0; 2329 u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1); 2330 2331 if (vnic_dev_capable_udp_rss(enic->vdev)) 2332 rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_UDP; 2333 if (rss_enable) { 2334 if (!enic_set_rsskey(enic)) { 2335 if (enic_set_rsscpu(enic, rss_hash_bits)) { 2336 rss_enable = 0; 2337 dev_warn(dev, "RSS disabled, " 2338 "Failed to set RSS cpu indirection table."); 2339 } 2340 } else { 2341 rss_enable = 0; 2342 dev_warn(dev, "RSS disabled, Failed to set RSS key.\n"); 2343 } 2344 } 2345 2346 return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type, 2347 rss_hash_bits, rss_base_cpu, rss_enable); 2348 } 2349 2350 static void enic_reset(struct work_struct *work) 2351 { 2352 struct enic *enic = container_of(work, struct enic, reset); 2353 2354 if (!netif_running(enic->netdev)) 2355 return; 2356 2357 rtnl_lock(); 2358 2359 spin_lock(&enic->enic_api_lock); 2360 enic_stop(enic->netdev); 2361 enic_dev_soft_reset(enic); 2362 enic_reset_addr_lists(enic); 2363 enic_init_vnic_resources(enic); 2364 enic_set_rss_nic_cfg(enic); 2365 enic_dev_set_ig_vlan_rewrite_mode(enic); 2366 enic_open(enic->netdev); 2367 spin_unlock(&enic->enic_api_lock); 2368 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev); 2369 2370 rtnl_unlock(); 2371 } 2372 2373 static void enic_tx_hang_reset(struct work_struct *work) 2374 { 2375 struct enic *enic = container_of(work, struct enic, tx_hang_reset); 2376 2377 rtnl_lock(); 2378 2379 spin_lock(&enic->enic_api_lock); 2380 enic_dev_hang_notify(enic); 2381 enic_stop(enic->netdev); 2382 enic_dev_hang_reset(enic); 2383 enic_reset_addr_lists(enic); 2384 enic_init_vnic_resources(enic); 2385 enic_set_rss_nic_cfg(enic); 2386 enic_dev_set_ig_vlan_rewrite_mode(enic); 2387 enic_open(enic->netdev); 2388 spin_unlock(&enic->enic_api_lock); 2389 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev); 2390 2391 rtnl_unlock(); 2392 } 2393 2394 static int enic_set_intr_mode(struct enic *enic) 2395 { 2396 unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX); 2397 unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX); 2398 unsigned int i; 2399 2400 /* Set interrupt mode (INTx, MSI, MSI-X) depending 2401 * on system capabilities. 2402 * 2403 * Try MSI-X first 2404 * 2405 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs 2406 * (the second to last INTR is used for WQ/RQ errors) 2407 * (the last INTR is used for notifications) 2408 */ 2409 2410 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2); 2411 for (i = 0; i < n + m + 2; i++) 2412 enic->msix_entry[i].entry = i; 2413 2414 /* Use multiple RQs if RSS is enabled 2415 */ 2416 2417 if (ENIC_SETTING(enic, RSS) && 2418 enic->config.intr_mode < 1 && 2419 enic->rq_count >= n && 2420 enic->wq_count >= m && 2421 enic->cq_count >= n + m && 2422 enic->intr_count >= n + m + 2) { 2423 2424 if (pci_enable_msix_range(enic->pdev, enic->msix_entry, 2425 n + m + 2, n + m + 2) > 0) { 2426 2427 enic->rq_count = n; 2428 enic->wq_count = m; 2429 enic->cq_count = n + m; 2430 enic->intr_count = n + m + 2; 2431 2432 vnic_dev_set_intr_mode(enic->vdev, 2433 VNIC_DEV_INTR_MODE_MSIX); 2434 2435 return 0; 2436 } 2437 } 2438 2439 if (enic->config.intr_mode < 1 && 2440 enic->rq_count >= 1 && 2441 enic->wq_count >= m && 2442 enic->cq_count >= 1 + m && 2443 enic->intr_count >= 1 + m + 2) { 2444 if (pci_enable_msix_range(enic->pdev, enic->msix_entry, 2445 1 + m + 2, 1 + m + 2) > 0) { 2446 2447 enic->rq_count = 1; 2448 enic->wq_count = m; 2449 enic->cq_count = 1 + m; 2450 enic->intr_count = 1 + m + 2; 2451 2452 vnic_dev_set_intr_mode(enic->vdev, 2453 VNIC_DEV_INTR_MODE_MSIX); 2454 2455 return 0; 2456 } 2457 } 2458 2459 /* Next try MSI 2460 * 2461 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR 2462 */ 2463 2464 if (enic->config.intr_mode < 2 && 2465 enic->rq_count >= 1 && 2466 enic->wq_count >= 1 && 2467 enic->cq_count >= 2 && 2468 enic->intr_count >= 1 && 2469 !pci_enable_msi(enic->pdev)) { 2470 2471 enic->rq_count = 1; 2472 enic->wq_count = 1; 2473 enic->cq_count = 2; 2474 enic->intr_count = 1; 2475 2476 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI); 2477 2478 return 0; 2479 } 2480 2481 /* Next try INTx 2482 * 2483 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs 2484 * (the first INTR is used for WQ/RQ) 2485 * (the second INTR is used for WQ/RQ errors) 2486 * (the last INTR is used for notifications) 2487 */ 2488 2489 if (enic->config.intr_mode < 3 && 2490 enic->rq_count >= 1 && 2491 enic->wq_count >= 1 && 2492 enic->cq_count >= 2 && 2493 enic->intr_count >= 3) { 2494 2495 enic->rq_count = 1; 2496 enic->wq_count = 1; 2497 enic->cq_count = 2; 2498 enic->intr_count = 3; 2499 2500 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX); 2501 2502 return 0; 2503 } 2504 2505 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN); 2506 2507 return -EINVAL; 2508 } 2509 2510 static void enic_clear_intr_mode(struct enic *enic) 2511 { 2512 switch (vnic_dev_get_intr_mode(enic->vdev)) { 2513 case VNIC_DEV_INTR_MODE_MSIX: 2514 pci_disable_msix(enic->pdev); 2515 break; 2516 case VNIC_DEV_INTR_MODE_MSI: 2517 pci_disable_msi(enic->pdev); 2518 break; 2519 default: 2520 break; 2521 } 2522 2523 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN); 2524 } 2525 2526 static const struct net_device_ops enic_netdev_dynamic_ops = { 2527 .ndo_open = enic_open, 2528 .ndo_stop = enic_stop, 2529 .ndo_start_xmit = enic_hard_start_xmit, 2530 .ndo_get_stats64 = enic_get_stats, 2531 .ndo_validate_addr = eth_validate_addr, 2532 .ndo_set_rx_mode = enic_set_rx_mode, 2533 .ndo_set_mac_address = enic_set_mac_address_dynamic, 2534 .ndo_change_mtu = enic_change_mtu, 2535 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid, 2536 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid, 2537 .ndo_tx_timeout = enic_tx_timeout, 2538 .ndo_set_vf_port = enic_set_vf_port, 2539 .ndo_get_vf_port = enic_get_vf_port, 2540 .ndo_set_vf_mac = enic_set_vf_mac, 2541 #ifdef CONFIG_NET_POLL_CONTROLLER 2542 .ndo_poll_controller = enic_poll_controller, 2543 #endif 2544 #ifdef CONFIG_RFS_ACCEL 2545 .ndo_rx_flow_steer = enic_rx_flow_steer, 2546 #endif 2547 .ndo_udp_tunnel_add = enic_udp_tunnel_add, 2548 .ndo_udp_tunnel_del = enic_udp_tunnel_del, 2549 .ndo_features_check = enic_features_check, 2550 }; 2551 2552 static const struct net_device_ops enic_netdev_ops = { 2553 .ndo_open = enic_open, 2554 .ndo_stop = enic_stop, 2555 .ndo_start_xmit = enic_hard_start_xmit, 2556 .ndo_get_stats64 = enic_get_stats, 2557 .ndo_validate_addr = eth_validate_addr, 2558 .ndo_set_mac_address = enic_set_mac_address, 2559 .ndo_set_rx_mode = enic_set_rx_mode, 2560 .ndo_change_mtu = enic_change_mtu, 2561 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid, 2562 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid, 2563 .ndo_tx_timeout = enic_tx_timeout, 2564 .ndo_set_vf_port = enic_set_vf_port, 2565 .ndo_get_vf_port = enic_get_vf_port, 2566 .ndo_set_vf_mac = enic_set_vf_mac, 2567 #ifdef CONFIG_NET_POLL_CONTROLLER 2568 .ndo_poll_controller = enic_poll_controller, 2569 #endif 2570 #ifdef CONFIG_RFS_ACCEL 2571 .ndo_rx_flow_steer = enic_rx_flow_steer, 2572 #endif 2573 .ndo_udp_tunnel_add = enic_udp_tunnel_add, 2574 .ndo_udp_tunnel_del = enic_udp_tunnel_del, 2575 .ndo_features_check = enic_features_check, 2576 }; 2577 2578 static void enic_dev_deinit(struct enic *enic) 2579 { 2580 unsigned int i; 2581 2582 for (i = 0; i < enic->rq_count; i++) { 2583 napi_hash_del(&enic->napi[i]); 2584 netif_napi_del(&enic->napi[i]); 2585 } 2586 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) 2587 for (i = 0; i < enic->wq_count; i++) 2588 netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]); 2589 2590 enic_free_vnic_resources(enic); 2591 enic_clear_intr_mode(enic); 2592 enic_free_affinity_hint(enic); 2593 } 2594 2595 static void enic_kdump_kernel_config(struct enic *enic) 2596 { 2597 if (is_kdump_kernel()) { 2598 dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n"); 2599 enic->rq_count = 1; 2600 enic->wq_count = 1; 2601 enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS; 2602 enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS; 2603 enic->config.mtu = min_t(u16, 1500, enic->config.mtu); 2604 } 2605 } 2606 2607 static int enic_dev_init(struct enic *enic) 2608 { 2609 struct device *dev = enic_get_dev(enic); 2610 struct net_device *netdev = enic->netdev; 2611 unsigned int i; 2612 int err; 2613 2614 /* Get interrupt coalesce timer info */ 2615 err = enic_dev_intr_coal_timer_info(enic); 2616 if (err) { 2617 dev_warn(dev, "Using default conversion factor for " 2618 "interrupt coalesce timer\n"); 2619 vnic_dev_intr_coal_timer_info_default(enic->vdev); 2620 } 2621 2622 /* Get vNIC configuration 2623 */ 2624 2625 err = enic_get_vnic_config(enic); 2626 if (err) { 2627 dev_err(dev, "Get vNIC configuration failed, aborting\n"); 2628 return err; 2629 } 2630 2631 /* Get available resource counts 2632 */ 2633 2634 enic_get_res_counts(enic); 2635 2636 /* modify resource count if we are in kdump_kernel 2637 */ 2638 enic_kdump_kernel_config(enic); 2639 2640 /* Set interrupt mode based on resource counts and system 2641 * capabilities 2642 */ 2643 2644 err = enic_set_intr_mode(enic); 2645 if (err) { 2646 dev_err(dev, "Failed to set intr mode based on resource " 2647 "counts and system capabilities, aborting\n"); 2648 return err; 2649 } 2650 2651 /* Allocate and configure vNIC resources 2652 */ 2653 2654 err = enic_alloc_vnic_resources(enic); 2655 if (err) { 2656 dev_err(dev, "Failed to alloc vNIC resources, aborting\n"); 2657 goto err_out_free_vnic_resources; 2658 } 2659 2660 enic_init_vnic_resources(enic); 2661 2662 err = enic_set_rss_nic_cfg(enic); 2663 if (err) { 2664 dev_err(dev, "Failed to config nic, aborting\n"); 2665 goto err_out_free_vnic_resources; 2666 } 2667 2668 switch (vnic_dev_get_intr_mode(enic->vdev)) { 2669 default: 2670 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64); 2671 break; 2672 case VNIC_DEV_INTR_MODE_MSIX: 2673 for (i = 0; i < enic->rq_count; i++) { 2674 netif_napi_add(netdev, &enic->napi[i], 2675 enic_poll_msix_rq, NAPI_POLL_WEIGHT); 2676 } 2677 for (i = 0; i < enic->wq_count; i++) 2678 netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)], 2679 enic_poll_msix_wq, NAPI_POLL_WEIGHT); 2680 break; 2681 } 2682 2683 return 0; 2684 2685 err_out_free_vnic_resources: 2686 enic_free_affinity_hint(enic); 2687 enic_clear_intr_mode(enic); 2688 enic_free_vnic_resources(enic); 2689 2690 return err; 2691 } 2692 2693 static void enic_iounmap(struct enic *enic) 2694 { 2695 unsigned int i; 2696 2697 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) 2698 if (enic->bar[i].vaddr) 2699 iounmap(enic->bar[i].vaddr); 2700 } 2701 2702 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 2703 { 2704 struct device *dev = &pdev->dev; 2705 struct net_device *netdev; 2706 struct enic *enic; 2707 int using_dac = 0; 2708 unsigned int i; 2709 int err; 2710 #ifdef CONFIG_PCI_IOV 2711 int pos = 0; 2712 #endif 2713 int num_pps = 1; 2714 2715 /* Allocate net device structure and initialize. Private 2716 * instance data is initialized to zero. 2717 */ 2718 2719 netdev = alloc_etherdev_mqs(sizeof(struct enic), 2720 ENIC_RQ_MAX, ENIC_WQ_MAX); 2721 if (!netdev) 2722 return -ENOMEM; 2723 2724 pci_set_drvdata(pdev, netdev); 2725 2726 SET_NETDEV_DEV(netdev, &pdev->dev); 2727 2728 enic = netdev_priv(netdev); 2729 enic->netdev = netdev; 2730 enic->pdev = pdev; 2731 2732 /* Setup PCI resources 2733 */ 2734 2735 err = pci_enable_device_mem(pdev); 2736 if (err) { 2737 dev_err(dev, "Cannot enable PCI device, aborting\n"); 2738 goto err_out_free_netdev; 2739 } 2740 2741 err = pci_request_regions(pdev, DRV_NAME); 2742 if (err) { 2743 dev_err(dev, "Cannot request PCI regions, aborting\n"); 2744 goto err_out_disable_device; 2745 } 2746 2747 pci_set_master(pdev); 2748 2749 /* Query PCI controller on system for DMA addressing 2750 * limitation for the device. Try 64-bit first, and 2751 * fail to 32-bit. 2752 */ 2753 2754 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64)); 2755 if (err) { 2756 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32)); 2757 if (err) { 2758 dev_err(dev, "No usable DMA configuration, aborting\n"); 2759 goto err_out_release_regions; 2760 } 2761 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)); 2762 if (err) { 2763 dev_err(dev, "Unable to obtain %u-bit DMA " 2764 "for consistent allocations, aborting\n", 32); 2765 goto err_out_release_regions; 2766 } 2767 } else { 2768 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)); 2769 if (err) { 2770 dev_err(dev, "Unable to obtain %u-bit DMA " 2771 "for consistent allocations, aborting\n", 64); 2772 goto err_out_release_regions; 2773 } 2774 using_dac = 1; 2775 } 2776 2777 /* Map vNIC resources from BAR0-5 2778 */ 2779 2780 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) { 2781 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM)) 2782 continue; 2783 enic->bar[i].len = pci_resource_len(pdev, i); 2784 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len); 2785 if (!enic->bar[i].vaddr) { 2786 dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i); 2787 err = -ENODEV; 2788 goto err_out_iounmap; 2789 } 2790 enic->bar[i].bus_addr = pci_resource_start(pdev, i); 2791 } 2792 2793 /* Register vNIC device 2794 */ 2795 2796 enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar, 2797 ARRAY_SIZE(enic->bar)); 2798 if (!enic->vdev) { 2799 dev_err(dev, "vNIC registration failed, aborting\n"); 2800 err = -ENODEV; 2801 goto err_out_iounmap; 2802 } 2803 2804 err = vnic_devcmd_init(enic->vdev); 2805 2806 if (err) 2807 goto err_out_vnic_unregister; 2808 2809 #ifdef CONFIG_PCI_IOV 2810 /* Get number of subvnics */ 2811 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV); 2812 if (pos) { 2813 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF, 2814 &enic->num_vfs); 2815 if (enic->num_vfs) { 2816 err = pci_enable_sriov(pdev, enic->num_vfs); 2817 if (err) { 2818 dev_err(dev, "SRIOV enable failed, aborting." 2819 " pci_enable_sriov() returned %d\n", 2820 err); 2821 goto err_out_vnic_unregister; 2822 } 2823 enic->priv_flags |= ENIC_SRIOV_ENABLED; 2824 num_pps = enic->num_vfs; 2825 } 2826 } 2827 #endif 2828 2829 /* Allocate structure for port profiles */ 2830 enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL); 2831 if (!enic->pp) { 2832 err = -ENOMEM; 2833 goto err_out_disable_sriov_pp; 2834 } 2835 2836 /* Issue device open to get device in known state 2837 */ 2838 2839 err = enic_dev_open(enic); 2840 if (err) { 2841 dev_err(dev, "vNIC dev open failed, aborting\n"); 2842 goto err_out_disable_sriov; 2843 } 2844 2845 /* Setup devcmd lock 2846 */ 2847 2848 spin_lock_init(&enic->devcmd_lock); 2849 spin_lock_init(&enic->enic_api_lock); 2850 2851 /* 2852 * Set ingress vlan rewrite mode before vnic initialization 2853 */ 2854 2855 err = enic_dev_set_ig_vlan_rewrite_mode(enic); 2856 if (err) { 2857 dev_err(dev, 2858 "Failed to set ingress vlan rewrite mode, aborting.\n"); 2859 goto err_out_dev_close; 2860 } 2861 2862 /* Issue device init to initialize the vnic-to-switch link. 2863 * We'll start with carrier off and wait for link UP 2864 * notification later to turn on carrier. We don't need 2865 * to wait here for the vnic-to-switch link initialization 2866 * to complete; link UP notification is the indication that 2867 * the process is complete. 2868 */ 2869 2870 netif_carrier_off(netdev); 2871 2872 /* Do not call dev_init for a dynamic vnic. 2873 * For a dynamic vnic, init_prov_info will be 2874 * called later by an upper layer. 2875 */ 2876 2877 if (!enic_is_dynamic(enic)) { 2878 err = vnic_dev_init(enic->vdev, 0); 2879 if (err) { 2880 dev_err(dev, "vNIC dev init failed, aborting\n"); 2881 goto err_out_dev_close; 2882 } 2883 } 2884 2885 err = enic_dev_init(enic); 2886 if (err) { 2887 dev_err(dev, "Device initialization failed, aborting\n"); 2888 goto err_out_dev_close; 2889 } 2890 2891 netif_set_real_num_tx_queues(netdev, enic->wq_count); 2892 netif_set_real_num_rx_queues(netdev, enic->rq_count); 2893 2894 /* Setup notification timer, HW reset task, and wq locks 2895 */ 2896 2897 timer_setup(&enic->notify_timer, enic_notify_timer, 0); 2898 2899 enic_set_rx_coal_setting(enic); 2900 INIT_WORK(&enic->reset, enic_reset); 2901 INIT_WORK(&enic->tx_hang_reset, enic_tx_hang_reset); 2902 INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work); 2903 2904 for (i = 0; i < enic->wq_count; i++) 2905 spin_lock_init(&enic->wq_lock[i]); 2906 2907 /* Register net device 2908 */ 2909 2910 enic->port_mtu = enic->config.mtu; 2911 (void)enic_change_mtu(netdev, enic->port_mtu); 2912 2913 err = enic_set_mac_addr(netdev, enic->mac_addr); 2914 if (err) { 2915 dev_err(dev, "Invalid MAC address, aborting\n"); 2916 goto err_out_dev_deinit; 2917 } 2918 2919 enic->tx_coalesce_usecs = enic->config.intr_timer_usec; 2920 /* rx coalesce time already got initialized. This gets used 2921 * if adaptive coal is turned off 2922 */ 2923 enic->rx_coalesce_usecs = enic->tx_coalesce_usecs; 2924 2925 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) 2926 netdev->netdev_ops = &enic_netdev_dynamic_ops; 2927 else 2928 netdev->netdev_ops = &enic_netdev_ops; 2929 2930 netdev->watchdog_timeo = 2 * HZ; 2931 enic_set_ethtool_ops(netdev); 2932 2933 netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX; 2934 if (ENIC_SETTING(enic, LOOP)) { 2935 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX; 2936 enic->loop_enable = 1; 2937 enic->loop_tag = enic->config.loop_tag; 2938 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag); 2939 } 2940 if (ENIC_SETTING(enic, TXCSUM)) 2941 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM; 2942 if (ENIC_SETTING(enic, TSO)) 2943 netdev->hw_features |= NETIF_F_TSO | 2944 NETIF_F_TSO6 | NETIF_F_TSO_ECN; 2945 if (ENIC_SETTING(enic, RSS)) 2946 netdev->hw_features |= NETIF_F_RXHASH; 2947 if (ENIC_SETTING(enic, RXCSUM)) 2948 netdev->hw_features |= NETIF_F_RXCSUM; 2949 if (ENIC_SETTING(enic, VXLAN)) { 2950 u64 patch_level; 2951 u64 a1 = 0; 2952 2953 netdev->hw_enc_features |= NETIF_F_RXCSUM | 2954 NETIF_F_TSO | 2955 NETIF_F_TSO6 | 2956 NETIF_F_TSO_ECN | 2957 NETIF_F_GSO_UDP_TUNNEL | 2958 NETIF_F_HW_CSUM | 2959 NETIF_F_GSO_UDP_TUNNEL_CSUM; 2960 netdev->hw_features |= netdev->hw_enc_features; 2961 /* get bit mask from hw about supported offload bit level 2962 * BIT(0) = fw supports patch_level 0 2963 * fcoe bit = encap 2964 * fcoe_fc_crc_ok = outer csum ok 2965 * BIT(1) = always set by fw 2966 * BIT(2) = fw supports patch_level 2 2967 * BIT(0) in rss_hash = encap 2968 * BIT(1,2) in rss_hash = outer_ip_csum_ok/ 2969 * outer_tcp_csum_ok 2970 * used in enic_rq_indicate_buf 2971 */ 2972 err = vnic_dev_get_supported_feature_ver(enic->vdev, 2973 VIC_FEATURE_VXLAN, 2974 &patch_level, &a1); 2975 if (err) 2976 patch_level = 0; 2977 enic->vxlan.flags = (u8)a1; 2978 /* mask bits that are supported by driver 2979 */ 2980 patch_level &= BIT_ULL(0) | BIT_ULL(2); 2981 patch_level = fls(patch_level); 2982 patch_level = patch_level ? patch_level - 1 : 0; 2983 enic->vxlan.patch_level = patch_level; 2984 } 2985 2986 netdev->features |= netdev->hw_features; 2987 netdev->vlan_features |= netdev->features; 2988 2989 #ifdef CONFIG_RFS_ACCEL 2990 netdev->hw_features |= NETIF_F_NTUPLE; 2991 #endif 2992 2993 if (using_dac) 2994 netdev->features |= NETIF_F_HIGHDMA; 2995 2996 netdev->priv_flags |= IFF_UNICAST_FLT; 2997 2998 /* MTU range: 68 - 9000 */ 2999 netdev->min_mtu = ENIC_MIN_MTU; 3000 netdev->max_mtu = ENIC_MAX_MTU; 3001 3002 err = register_netdev(netdev); 3003 if (err) { 3004 dev_err(dev, "Cannot register net device, aborting\n"); 3005 goto err_out_dev_deinit; 3006 } 3007 enic->rx_copybreak = RX_COPYBREAK_DEFAULT; 3008 3009 return 0; 3010 3011 err_out_dev_deinit: 3012 enic_dev_deinit(enic); 3013 err_out_dev_close: 3014 vnic_dev_close(enic->vdev); 3015 err_out_disable_sriov: 3016 kfree(enic->pp); 3017 err_out_disable_sriov_pp: 3018 #ifdef CONFIG_PCI_IOV 3019 if (enic_sriov_enabled(enic)) { 3020 pci_disable_sriov(pdev); 3021 enic->priv_flags &= ~ENIC_SRIOV_ENABLED; 3022 } 3023 #endif 3024 err_out_vnic_unregister: 3025 vnic_dev_unregister(enic->vdev); 3026 err_out_iounmap: 3027 enic_iounmap(enic); 3028 err_out_release_regions: 3029 pci_release_regions(pdev); 3030 err_out_disable_device: 3031 pci_disable_device(pdev); 3032 err_out_free_netdev: 3033 free_netdev(netdev); 3034 3035 return err; 3036 } 3037 3038 static void enic_remove(struct pci_dev *pdev) 3039 { 3040 struct net_device *netdev = pci_get_drvdata(pdev); 3041 3042 if (netdev) { 3043 struct enic *enic = netdev_priv(netdev); 3044 3045 cancel_work_sync(&enic->reset); 3046 cancel_work_sync(&enic->change_mtu_work); 3047 unregister_netdev(netdev); 3048 enic_dev_deinit(enic); 3049 vnic_dev_close(enic->vdev); 3050 #ifdef CONFIG_PCI_IOV 3051 if (enic_sriov_enabled(enic)) { 3052 pci_disable_sriov(pdev); 3053 enic->priv_flags &= ~ENIC_SRIOV_ENABLED; 3054 } 3055 #endif 3056 kfree(enic->pp); 3057 vnic_dev_unregister(enic->vdev); 3058 enic_iounmap(enic); 3059 pci_release_regions(pdev); 3060 pci_disable_device(pdev); 3061 free_netdev(netdev); 3062 } 3063 } 3064 3065 static struct pci_driver enic_driver = { 3066 .name = DRV_NAME, 3067 .id_table = enic_id_table, 3068 .probe = enic_probe, 3069 .remove = enic_remove, 3070 }; 3071 3072 static int __init enic_init_module(void) 3073 { 3074 pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION); 3075 3076 return pci_register_driver(&enic_driver); 3077 } 3078 3079 static void __exit enic_cleanup_module(void) 3080 { 3081 pci_unregister_driver(&enic_driver); 3082 } 3083 3084 module_init(enic_init_module); 3085 module_exit(enic_cleanup_module); 3086