1 /* 2 * Copyright 2008-2010 Cisco Systems, Inc. All rights reserved. 3 * Copyright 2007 Nuova Systems, Inc. All rights reserved. 4 * 5 * This program is free software; you may redistribute it and/or modify 6 * it under the terms of the GNU General Public License as published by 7 * the Free Software Foundation; version 2 of the License. 8 * 9 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 10 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 11 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 12 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 13 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 14 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 15 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 16 * SOFTWARE. 17 * 18 */ 19 20 #include <linux/module.h> 21 #include <linux/kernel.h> 22 #include <linux/string.h> 23 #include <linux/errno.h> 24 #include <linux/types.h> 25 #include <linux/init.h> 26 #include <linux/interrupt.h> 27 #include <linux/workqueue.h> 28 #include <linux/pci.h> 29 #include <linux/netdevice.h> 30 #include <linux/etherdevice.h> 31 #include <linux/if.h> 32 #include <linux/if_ether.h> 33 #include <linux/if_vlan.h> 34 #include <linux/in.h> 35 #include <linux/ip.h> 36 #include <linux/ipv6.h> 37 #include <linux/tcp.h> 38 #include <linux/rtnetlink.h> 39 #include <linux/prefetch.h> 40 #include <net/ip6_checksum.h> 41 #include <linux/ktime.h> 42 #include <linux/numa.h> 43 #ifdef CONFIG_RFS_ACCEL 44 #include <linux/cpu_rmap.h> 45 #endif 46 #include <linux/crash_dump.h> 47 #include <net/busy_poll.h> 48 #include <net/vxlan.h> 49 50 #include "cq_enet_desc.h" 51 #include "vnic_dev.h" 52 #include "vnic_intr.h" 53 #include "vnic_stats.h" 54 #include "vnic_vic.h" 55 #include "enic_res.h" 56 #include "enic.h" 57 #include "enic_dev.h" 58 #include "enic_pp.h" 59 #include "enic_clsf.h" 60 61 #define ENIC_NOTIFY_TIMER_PERIOD (2 * HZ) 62 #define WQ_ENET_MAX_DESC_LEN (1 << WQ_ENET_LEN_BITS) 63 #define MAX_TSO (1 << 16) 64 #define ENIC_DESC_MAX_SPLITS (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1) 65 66 #define PCI_DEVICE_ID_CISCO_VIC_ENET 0x0043 /* ethernet vnic */ 67 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN 0x0044 /* enet dynamic vnic */ 68 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF 0x0071 /* enet SRIOV VF */ 69 70 #define RX_COPYBREAK_DEFAULT 256 71 72 /* Supported devices */ 73 static const struct pci_device_id enic_id_table[] = { 74 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) }, 75 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) }, 76 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) }, 77 { 0, } /* end of table */ 78 }; 79 80 MODULE_DESCRIPTION(DRV_DESCRIPTION); 81 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>"); 82 MODULE_LICENSE("GPL"); 83 MODULE_DEVICE_TABLE(pci, enic_id_table); 84 85 #define ENIC_LARGE_PKT_THRESHOLD 1000 86 #define ENIC_MAX_COALESCE_TIMERS 10 87 /* Interrupt moderation table, which will be used to decide the 88 * coalescing timer values 89 * {rx_rate in Mbps, mapping percentage of the range} 90 */ 91 static struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = { 92 {4000, 0}, 93 {4400, 10}, 94 {5060, 20}, 95 {5230, 30}, 96 {5540, 40}, 97 {5820, 50}, 98 {6120, 60}, 99 {6435, 70}, 100 {6745, 80}, 101 {7000, 90}, 102 {0xFFFFFFFF, 100} 103 }; 104 105 /* This table helps the driver to pick different ranges for rx coalescing 106 * timer depending on the link speed. 107 */ 108 static struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = { 109 {0, 0}, /* 0 - 4 Gbps */ 110 {0, 3}, /* 4 - 10 Gbps */ 111 {3, 6}, /* 10 - 40 Gbps */ 112 }; 113 114 static void enic_init_affinity_hint(struct enic *enic) 115 { 116 int numa_node = dev_to_node(&enic->pdev->dev); 117 int i; 118 119 for (i = 0; i < enic->intr_count; i++) { 120 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i) || 121 (cpumask_available(enic->msix[i].affinity_mask) && 122 !cpumask_empty(enic->msix[i].affinity_mask))) 123 continue; 124 if (zalloc_cpumask_var(&enic->msix[i].affinity_mask, 125 GFP_KERNEL)) 126 cpumask_set_cpu(cpumask_local_spread(i, numa_node), 127 enic->msix[i].affinity_mask); 128 } 129 } 130 131 static void enic_free_affinity_hint(struct enic *enic) 132 { 133 int i; 134 135 for (i = 0; i < enic->intr_count; i++) { 136 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i)) 137 continue; 138 free_cpumask_var(enic->msix[i].affinity_mask); 139 } 140 } 141 142 static void enic_set_affinity_hint(struct enic *enic) 143 { 144 int i; 145 int err; 146 147 for (i = 0; i < enic->intr_count; i++) { 148 if (enic_is_err_intr(enic, i) || 149 enic_is_notify_intr(enic, i) || 150 !cpumask_available(enic->msix[i].affinity_mask) || 151 cpumask_empty(enic->msix[i].affinity_mask)) 152 continue; 153 err = irq_update_affinity_hint(enic->msix_entry[i].vector, 154 enic->msix[i].affinity_mask); 155 if (err) 156 netdev_warn(enic->netdev, "irq_update_affinity_hint failed, err %d\n", 157 err); 158 } 159 160 for (i = 0; i < enic->wq_count; i++) { 161 int wq_intr = enic_msix_wq_intr(enic, i); 162 163 if (cpumask_available(enic->msix[wq_intr].affinity_mask) && 164 !cpumask_empty(enic->msix[wq_intr].affinity_mask)) 165 netif_set_xps_queue(enic->netdev, 166 enic->msix[wq_intr].affinity_mask, 167 i); 168 } 169 } 170 171 static void enic_unset_affinity_hint(struct enic *enic) 172 { 173 int i; 174 175 for (i = 0; i < enic->intr_count; i++) 176 irq_update_affinity_hint(enic->msix_entry[i].vector, NULL); 177 } 178 179 static int enic_udp_tunnel_set_port(struct net_device *netdev, 180 unsigned int table, unsigned int entry, 181 struct udp_tunnel_info *ti) 182 { 183 struct enic *enic = netdev_priv(netdev); 184 int err; 185 186 spin_lock_bh(&enic->devcmd_lock); 187 188 err = vnic_dev_overlay_offload_cfg(enic->vdev, 189 OVERLAY_CFG_VXLAN_PORT_UPDATE, 190 ntohs(ti->port)); 191 if (err) 192 goto error; 193 194 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN, 195 enic->vxlan.patch_level); 196 if (err) 197 goto error; 198 199 enic->vxlan.vxlan_udp_port_number = ntohs(ti->port); 200 error: 201 spin_unlock_bh(&enic->devcmd_lock); 202 203 return err; 204 } 205 206 static int enic_udp_tunnel_unset_port(struct net_device *netdev, 207 unsigned int table, unsigned int entry, 208 struct udp_tunnel_info *ti) 209 { 210 struct enic *enic = netdev_priv(netdev); 211 int err; 212 213 spin_lock_bh(&enic->devcmd_lock); 214 215 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN, 216 OVERLAY_OFFLOAD_DISABLE); 217 if (err) 218 goto unlock; 219 220 enic->vxlan.vxlan_udp_port_number = 0; 221 222 unlock: 223 spin_unlock_bh(&enic->devcmd_lock); 224 225 return err; 226 } 227 228 static const struct udp_tunnel_nic_info enic_udp_tunnels = { 229 .set_port = enic_udp_tunnel_set_port, 230 .unset_port = enic_udp_tunnel_unset_port, 231 .tables = { 232 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, }, 233 }, 234 }, enic_udp_tunnels_v4 = { 235 .set_port = enic_udp_tunnel_set_port, 236 .unset_port = enic_udp_tunnel_unset_port, 237 .flags = UDP_TUNNEL_NIC_INFO_IPV4_ONLY, 238 .tables = { 239 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, }, 240 }, 241 }; 242 243 static netdev_features_t enic_features_check(struct sk_buff *skb, 244 struct net_device *dev, 245 netdev_features_t features) 246 { 247 const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb); 248 struct enic *enic = netdev_priv(dev); 249 struct udphdr *udph; 250 u16 port = 0; 251 u8 proto; 252 253 if (!skb->encapsulation) 254 return features; 255 256 features = vxlan_features_check(skb, features); 257 258 switch (vlan_get_protocol(skb)) { 259 case htons(ETH_P_IPV6): 260 if (!(enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6)) 261 goto out; 262 proto = ipv6_hdr(skb)->nexthdr; 263 break; 264 case htons(ETH_P_IP): 265 proto = ip_hdr(skb)->protocol; 266 break; 267 default: 268 goto out; 269 } 270 271 switch (eth->h_proto) { 272 case ntohs(ETH_P_IPV6): 273 if (!(enic->vxlan.flags & ENIC_VXLAN_INNER_IPV6)) 274 goto out; 275 fallthrough; 276 case ntohs(ETH_P_IP): 277 break; 278 default: 279 goto out; 280 } 281 282 283 if (proto == IPPROTO_UDP) { 284 udph = udp_hdr(skb); 285 port = be16_to_cpu(udph->dest); 286 } 287 288 /* HW supports offload of only one UDP port. Remove CSUM and GSO MASK 289 * for other UDP port tunnels 290 */ 291 if (port != enic->vxlan.vxlan_udp_port_number) 292 goto out; 293 294 return features; 295 296 out: 297 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 298 } 299 300 int enic_is_dynamic(struct enic *enic) 301 { 302 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN; 303 } 304 305 int enic_sriov_enabled(struct enic *enic) 306 { 307 return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0; 308 } 309 310 static int enic_is_sriov_vf(struct enic *enic) 311 { 312 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF; 313 } 314 315 int enic_is_valid_vf(struct enic *enic, int vf) 316 { 317 #ifdef CONFIG_PCI_IOV 318 return vf >= 0 && vf < enic->num_vfs; 319 #else 320 return 0; 321 #endif 322 } 323 324 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf) 325 { 326 struct enic *enic = vnic_dev_priv(wq->vdev); 327 328 if (buf->sop) 329 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len, 330 DMA_TO_DEVICE); 331 else 332 dma_unmap_page(&enic->pdev->dev, buf->dma_addr, buf->len, 333 DMA_TO_DEVICE); 334 335 if (buf->os_buf) 336 dev_kfree_skb_any(buf->os_buf); 337 } 338 339 static void enic_wq_free_buf(struct vnic_wq *wq, 340 struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque) 341 { 342 enic_free_wq_buf(wq, buf); 343 } 344 345 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc, 346 u8 type, u16 q_number, u16 completed_index, void *opaque) 347 { 348 struct enic *enic = vnic_dev_priv(vdev); 349 350 spin_lock(&enic->wq_lock[q_number]); 351 352 vnic_wq_service(&enic->wq[q_number], cq_desc, 353 completed_index, enic_wq_free_buf, 354 opaque); 355 356 if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) && 357 vnic_wq_desc_avail(&enic->wq[q_number]) >= 358 (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)) 359 netif_wake_subqueue(enic->netdev, q_number); 360 361 spin_unlock(&enic->wq_lock[q_number]); 362 363 return 0; 364 } 365 366 static bool enic_log_q_error(struct enic *enic) 367 { 368 unsigned int i; 369 u32 error_status; 370 bool err = false; 371 372 for (i = 0; i < enic->wq_count; i++) { 373 error_status = vnic_wq_error_status(&enic->wq[i]); 374 err |= error_status; 375 if (error_status) 376 netdev_err(enic->netdev, "WQ[%d] error_status %d\n", 377 i, error_status); 378 } 379 380 for (i = 0; i < enic->rq_count; i++) { 381 error_status = vnic_rq_error_status(&enic->rq[i]); 382 err |= error_status; 383 if (error_status) 384 netdev_err(enic->netdev, "RQ[%d] error_status %d\n", 385 i, error_status); 386 } 387 388 return err; 389 } 390 391 static void enic_msglvl_check(struct enic *enic) 392 { 393 u32 msg_enable = vnic_dev_msg_lvl(enic->vdev); 394 395 if (msg_enable != enic->msg_enable) { 396 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n", 397 enic->msg_enable, msg_enable); 398 enic->msg_enable = msg_enable; 399 } 400 } 401 402 static void enic_mtu_check(struct enic *enic) 403 { 404 u32 mtu = vnic_dev_mtu(enic->vdev); 405 struct net_device *netdev = enic->netdev; 406 407 if (mtu && mtu != enic->port_mtu) { 408 enic->port_mtu = mtu; 409 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) { 410 mtu = max_t(int, ENIC_MIN_MTU, 411 min_t(int, ENIC_MAX_MTU, mtu)); 412 if (mtu != netdev->mtu) 413 schedule_work(&enic->change_mtu_work); 414 } else { 415 if (mtu < netdev->mtu) 416 netdev_warn(netdev, 417 "interface MTU (%d) set higher " 418 "than switch port MTU (%d)\n", 419 netdev->mtu, mtu); 420 } 421 } 422 } 423 424 static void enic_link_check(struct enic *enic) 425 { 426 int link_status = vnic_dev_link_status(enic->vdev); 427 int carrier_ok = netif_carrier_ok(enic->netdev); 428 429 if (link_status && !carrier_ok) { 430 netdev_info(enic->netdev, "Link UP\n"); 431 netif_carrier_on(enic->netdev); 432 } else if (!link_status && carrier_ok) { 433 netdev_info(enic->netdev, "Link DOWN\n"); 434 netif_carrier_off(enic->netdev); 435 } 436 } 437 438 static void enic_notify_check(struct enic *enic) 439 { 440 enic_msglvl_check(enic); 441 enic_mtu_check(enic); 442 enic_link_check(enic); 443 } 444 445 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i)) 446 447 static irqreturn_t enic_isr_legacy(int irq, void *data) 448 { 449 struct net_device *netdev = data; 450 struct enic *enic = netdev_priv(netdev); 451 unsigned int io_intr = ENIC_LEGACY_IO_INTR; 452 unsigned int err_intr = ENIC_LEGACY_ERR_INTR; 453 unsigned int notify_intr = ENIC_LEGACY_NOTIFY_INTR; 454 u32 pba; 455 456 vnic_intr_mask(&enic->intr[io_intr]); 457 458 pba = vnic_intr_legacy_pba(enic->legacy_pba); 459 if (!pba) { 460 vnic_intr_unmask(&enic->intr[io_intr]); 461 return IRQ_NONE; /* not our interrupt */ 462 } 463 464 if (ENIC_TEST_INTR(pba, notify_intr)) { 465 enic_notify_check(enic); 466 vnic_intr_return_all_credits(&enic->intr[notify_intr]); 467 } 468 469 if (ENIC_TEST_INTR(pba, err_intr)) { 470 vnic_intr_return_all_credits(&enic->intr[err_intr]); 471 enic_log_q_error(enic); 472 /* schedule recovery from WQ/RQ error */ 473 schedule_work(&enic->reset); 474 return IRQ_HANDLED; 475 } 476 477 if (ENIC_TEST_INTR(pba, io_intr)) 478 napi_schedule_irqoff(&enic->napi[0]); 479 else 480 vnic_intr_unmask(&enic->intr[io_intr]); 481 482 return IRQ_HANDLED; 483 } 484 485 static irqreturn_t enic_isr_msi(int irq, void *data) 486 { 487 struct enic *enic = data; 488 489 /* With MSI, there is no sharing of interrupts, so this is 490 * our interrupt and there is no need to ack it. The device 491 * is not providing per-vector masking, so the OS will not 492 * write to PCI config space to mask/unmask the interrupt. 493 * We're using mask_on_assertion for MSI, so the device 494 * automatically masks the interrupt when the interrupt is 495 * generated. Later, when exiting polling, the interrupt 496 * will be unmasked (see enic_poll). 497 * 498 * Also, the device uses the same PCIe Traffic Class (TC) 499 * for Memory Write data and MSI, so there are no ordering 500 * issues; the MSI will always arrive at the Root Complex 501 * _after_ corresponding Memory Writes (i.e. descriptor 502 * writes). 503 */ 504 505 napi_schedule_irqoff(&enic->napi[0]); 506 507 return IRQ_HANDLED; 508 } 509 510 static irqreturn_t enic_isr_msix(int irq, void *data) 511 { 512 struct napi_struct *napi = data; 513 514 napi_schedule_irqoff(napi); 515 516 return IRQ_HANDLED; 517 } 518 519 static irqreturn_t enic_isr_msix_err(int irq, void *data) 520 { 521 struct enic *enic = data; 522 unsigned int intr = enic_msix_err_intr(enic); 523 524 vnic_intr_return_all_credits(&enic->intr[intr]); 525 526 if (enic_log_q_error(enic)) 527 /* schedule recovery from WQ/RQ error */ 528 schedule_work(&enic->reset); 529 530 return IRQ_HANDLED; 531 } 532 533 static irqreturn_t enic_isr_msix_notify(int irq, void *data) 534 { 535 struct enic *enic = data; 536 unsigned int intr = enic_msix_notify_intr(enic); 537 538 enic_notify_check(enic); 539 vnic_intr_return_all_credits(&enic->intr[intr]); 540 541 return IRQ_HANDLED; 542 } 543 544 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq, 545 struct sk_buff *skb, unsigned int len_left, 546 int loopback) 547 { 548 const skb_frag_t *frag; 549 dma_addr_t dma_addr; 550 551 /* Queue additional data fragments */ 552 for (frag = skb_shinfo(skb)->frags; len_left; frag++) { 553 len_left -= skb_frag_size(frag); 554 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 0, 555 skb_frag_size(frag), 556 DMA_TO_DEVICE); 557 if (unlikely(enic_dma_map_check(enic, dma_addr))) 558 return -ENOMEM; 559 enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag), 560 (len_left == 0), /* EOP? */ 561 loopback); 562 } 563 564 return 0; 565 } 566 567 static int enic_queue_wq_skb_vlan(struct enic *enic, struct vnic_wq *wq, 568 struct sk_buff *skb, int vlan_tag_insert, 569 unsigned int vlan_tag, int loopback) 570 { 571 unsigned int head_len = skb_headlen(skb); 572 unsigned int len_left = skb->len - head_len; 573 int eop = (len_left == 0); 574 dma_addr_t dma_addr; 575 int err = 0; 576 577 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len, 578 DMA_TO_DEVICE); 579 if (unlikely(enic_dma_map_check(enic, dma_addr))) 580 return -ENOMEM; 581 582 /* Queue the main skb fragment. The fragments are no larger 583 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less 584 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor 585 * per fragment is queued. 586 */ 587 enic_queue_wq_desc(wq, skb, dma_addr, head_len, vlan_tag_insert, 588 vlan_tag, eop, loopback); 589 590 if (!eop) 591 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback); 592 593 return err; 594 } 595 596 static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq, 597 struct sk_buff *skb, int vlan_tag_insert, 598 unsigned int vlan_tag, int loopback) 599 { 600 unsigned int head_len = skb_headlen(skb); 601 unsigned int len_left = skb->len - head_len; 602 unsigned int hdr_len = skb_checksum_start_offset(skb); 603 unsigned int csum_offset = hdr_len + skb->csum_offset; 604 int eop = (len_left == 0); 605 dma_addr_t dma_addr; 606 int err = 0; 607 608 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len, 609 DMA_TO_DEVICE); 610 if (unlikely(enic_dma_map_check(enic, dma_addr))) 611 return -ENOMEM; 612 613 /* Queue the main skb fragment. The fragments are no larger 614 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less 615 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor 616 * per fragment is queued. 617 */ 618 enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len, csum_offset, 619 hdr_len, vlan_tag_insert, vlan_tag, eop, 620 loopback); 621 622 if (!eop) 623 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback); 624 625 return err; 626 } 627 628 static void enic_preload_tcp_csum_encap(struct sk_buff *skb) 629 { 630 const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb); 631 632 switch (eth->h_proto) { 633 case ntohs(ETH_P_IP): 634 inner_ip_hdr(skb)->check = 0; 635 inner_tcp_hdr(skb)->check = 636 ~csum_tcpudp_magic(inner_ip_hdr(skb)->saddr, 637 inner_ip_hdr(skb)->daddr, 0, 638 IPPROTO_TCP, 0); 639 break; 640 case ntohs(ETH_P_IPV6): 641 inner_tcp_hdr(skb)->check = 642 ~csum_ipv6_magic(&inner_ipv6_hdr(skb)->saddr, 643 &inner_ipv6_hdr(skb)->daddr, 0, 644 IPPROTO_TCP, 0); 645 break; 646 default: 647 WARN_ONCE(1, "Non ipv4/ipv6 inner pkt for encap offload"); 648 break; 649 } 650 } 651 652 static void enic_preload_tcp_csum(struct sk_buff *skb) 653 { 654 /* Preload TCP csum field with IP pseudo hdr calculated 655 * with IP length set to zero. HW will later add in length 656 * to each TCP segment resulting from the TSO. 657 */ 658 659 if (skb->protocol == cpu_to_be16(ETH_P_IP)) { 660 ip_hdr(skb)->check = 0; 661 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 662 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); 663 } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) { 664 tcp_v6_gso_csum_prep(skb); 665 } 666 } 667 668 static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq, 669 struct sk_buff *skb, unsigned int mss, 670 int vlan_tag_insert, unsigned int vlan_tag, 671 int loopback) 672 { 673 unsigned int frag_len_left = skb_headlen(skb); 674 unsigned int len_left = skb->len - frag_len_left; 675 int eop = (len_left == 0); 676 unsigned int offset = 0; 677 unsigned int hdr_len; 678 dma_addr_t dma_addr; 679 unsigned int len; 680 skb_frag_t *frag; 681 682 if (skb->encapsulation) { 683 hdr_len = skb_inner_tcp_all_headers(skb); 684 enic_preload_tcp_csum_encap(skb); 685 } else { 686 hdr_len = skb_tcp_all_headers(skb); 687 enic_preload_tcp_csum(skb); 688 } 689 690 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors 691 * for the main skb fragment 692 */ 693 while (frag_len_left) { 694 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN); 695 dma_addr = dma_map_single(&enic->pdev->dev, 696 skb->data + offset, len, 697 DMA_TO_DEVICE); 698 if (unlikely(enic_dma_map_check(enic, dma_addr))) 699 return -ENOMEM; 700 enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len, 701 vlan_tag_insert, vlan_tag, 702 eop && (len == frag_len_left), loopback); 703 frag_len_left -= len; 704 offset += len; 705 } 706 707 if (eop) 708 return 0; 709 710 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors 711 * for additional data fragments 712 */ 713 for (frag = skb_shinfo(skb)->frags; len_left; frag++) { 714 len_left -= skb_frag_size(frag); 715 frag_len_left = skb_frag_size(frag); 716 offset = 0; 717 718 while (frag_len_left) { 719 len = min(frag_len_left, 720 (unsigned int)WQ_ENET_MAX_DESC_LEN); 721 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 722 offset, len, 723 DMA_TO_DEVICE); 724 if (unlikely(enic_dma_map_check(enic, dma_addr))) 725 return -ENOMEM; 726 enic_queue_wq_desc_cont(wq, skb, dma_addr, len, 727 (len_left == 0) && 728 (len == frag_len_left),/*EOP*/ 729 loopback); 730 frag_len_left -= len; 731 offset += len; 732 } 733 } 734 735 return 0; 736 } 737 738 static inline int enic_queue_wq_skb_encap(struct enic *enic, struct vnic_wq *wq, 739 struct sk_buff *skb, 740 int vlan_tag_insert, 741 unsigned int vlan_tag, int loopback) 742 { 743 unsigned int head_len = skb_headlen(skb); 744 unsigned int len_left = skb->len - head_len; 745 /* Hardware will overwrite the checksum fields, calculating from 746 * scratch and ignoring the value placed by software. 747 * Offload mode = 00 748 * mss[2], mss[1], mss[0] bits are set 749 */ 750 unsigned int mss_or_csum = 7; 751 int eop = (len_left == 0); 752 dma_addr_t dma_addr; 753 int err = 0; 754 755 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len, 756 DMA_TO_DEVICE); 757 if (unlikely(enic_dma_map_check(enic, dma_addr))) 758 return -ENOMEM; 759 760 enic_queue_wq_desc_ex(wq, skb, dma_addr, head_len, mss_or_csum, 0, 761 vlan_tag_insert, vlan_tag, 762 WQ_ENET_OFFLOAD_MODE_CSUM, eop, 1 /* SOP */, eop, 763 loopback); 764 if (!eop) 765 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback); 766 767 return err; 768 } 769 770 static inline int enic_queue_wq_skb(struct enic *enic, 771 struct vnic_wq *wq, struct sk_buff *skb) 772 { 773 unsigned int mss = skb_shinfo(skb)->gso_size; 774 unsigned int vlan_tag = 0; 775 int vlan_tag_insert = 0; 776 int loopback = 0; 777 int err; 778 779 if (skb_vlan_tag_present(skb)) { 780 /* VLAN tag from trunking driver */ 781 vlan_tag_insert = 1; 782 vlan_tag = skb_vlan_tag_get(skb); 783 } else if (enic->loop_enable) { 784 vlan_tag = enic->loop_tag; 785 loopback = 1; 786 } 787 788 if (mss) 789 err = enic_queue_wq_skb_tso(enic, wq, skb, mss, 790 vlan_tag_insert, vlan_tag, 791 loopback); 792 else if (skb->encapsulation) 793 err = enic_queue_wq_skb_encap(enic, wq, skb, vlan_tag_insert, 794 vlan_tag, loopback); 795 else if (skb->ip_summed == CHECKSUM_PARTIAL) 796 err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert, 797 vlan_tag, loopback); 798 else 799 err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert, 800 vlan_tag, loopback); 801 if (unlikely(err)) { 802 struct vnic_wq_buf *buf; 803 804 buf = wq->to_use->prev; 805 /* while not EOP of previous pkt && queue not empty. 806 * For all non EOP bufs, os_buf is NULL. 807 */ 808 while (!buf->os_buf && (buf->next != wq->to_clean)) { 809 enic_free_wq_buf(wq, buf); 810 wq->ring.desc_avail++; 811 buf = buf->prev; 812 } 813 wq->to_use = buf->next; 814 dev_kfree_skb(skb); 815 } 816 return err; 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 if (enic_queue_wq_skb(enic, wq, skb)) 861 goto error; 862 863 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS) 864 netif_tx_stop_queue(txq); 865 skb_tx_timestamp(skb); 866 if (!netdev_xmit_more() || netif_xmit_stopped(txq)) 867 vnic_wq_doorbell(wq); 868 869 error: 870 spin_unlock(&enic->wq_lock[txq_map]); 871 872 return NETDEV_TX_OK; 873 } 874 875 /* dev_base_lock rwlock held, nominally process context */ 876 static void enic_get_stats(struct net_device *netdev, 877 struct rtnl_link_stats64 *net_stats) 878 { 879 struct enic *enic = netdev_priv(netdev); 880 struct vnic_stats *stats; 881 int err; 882 883 err = enic_dev_stats_dump(enic, &stats); 884 /* return only when dma_alloc_coherent fails in vnic_dev_stats_dump 885 * For other failures, like devcmd failure, we return previously 886 * recorded stats. 887 */ 888 if (err == -ENOMEM) 889 return; 890 891 net_stats->tx_packets = stats->tx.tx_frames_ok; 892 net_stats->tx_bytes = stats->tx.tx_bytes_ok; 893 net_stats->tx_errors = stats->tx.tx_errors; 894 net_stats->tx_dropped = stats->tx.tx_drops; 895 896 net_stats->rx_packets = stats->rx.rx_frames_ok; 897 net_stats->rx_bytes = stats->rx.rx_bytes_ok; 898 net_stats->rx_errors = stats->rx.rx_errors; 899 net_stats->multicast = stats->rx.rx_multicast_frames_ok; 900 net_stats->rx_over_errors = enic->rq_truncated_pkts; 901 net_stats->rx_crc_errors = enic->rq_bad_fcs; 902 net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop; 903 } 904 905 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr) 906 { 907 struct enic *enic = netdev_priv(netdev); 908 909 if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) { 910 unsigned int mc_count = netdev_mc_count(netdev); 911 912 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n", 913 ENIC_MULTICAST_PERFECT_FILTERS, mc_count); 914 915 return -ENOSPC; 916 } 917 918 enic_dev_add_addr(enic, mc_addr); 919 enic->mc_count++; 920 921 return 0; 922 } 923 924 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr) 925 { 926 struct enic *enic = netdev_priv(netdev); 927 928 enic_dev_del_addr(enic, mc_addr); 929 enic->mc_count--; 930 931 return 0; 932 } 933 934 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr) 935 { 936 struct enic *enic = netdev_priv(netdev); 937 938 if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) { 939 unsigned int uc_count = netdev_uc_count(netdev); 940 941 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n", 942 ENIC_UNICAST_PERFECT_FILTERS, uc_count); 943 944 return -ENOSPC; 945 } 946 947 enic_dev_add_addr(enic, uc_addr); 948 enic->uc_count++; 949 950 return 0; 951 } 952 953 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr) 954 { 955 struct enic *enic = netdev_priv(netdev); 956 957 enic_dev_del_addr(enic, uc_addr); 958 enic->uc_count--; 959 960 return 0; 961 } 962 963 void enic_reset_addr_lists(struct enic *enic) 964 { 965 struct net_device *netdev = enic->netdev; 966 967 __dev_uc_unsync(netdev, NULL); 968 __dev_mc_unsync(netdev, NULL); 969 970 enic->mc_count = 0; 971 enic->uc_count = 0; 972 enic->flags = 0; 973 } 974 975 static int enic_set_mac_addr(struct net_device *netdev, char *addr) 976 { 977 struct enic *enic = netdev_priv(netdev); 978 979 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) { 980 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr)) 981 return -EADDRNOTAVAIL; 982 } else { 983 if (!is_valid_ether_addr(addr)) 984 return -EADDRNOTAVAIL; 985 } 986 987 eth_hw_addr_set(netdev, addr); 988 989 return 0; 990 } 991 992 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p) 993 { 994 struct enic *enic = netdev_priv(netdev); 995 struct sockaddr *saddr = p; 996 char *addr = saddr->sa_data; 997 int err; 998 999 if (netif_running(enic->netdev)) { 1000 err = enic_dev_del_station_addr(enic); 1001 if (err) 1002 return err; 1003 } 1004 1005 err = enic_set_mac_addr(netdev, addr); 1006 if (err) 1007 return err; 1008 1009 if (netif_running(enic->netdev)) { 1010 err = enic_dev_add_station_addr(enic); 1011 if (err) 1012 return err; 1013 } 1014 1015 return err; 1016 } 1017 1018 static int enic_set_mac_address(struct net_device *netdev, void *p) 1019 { 1020 struct sockaddr *saddr = p; 1021 char *addr = saddr->sa_data; 1022 struct enic *enic = netdev_priv(netdev); 1023 int err; 1024 1025 err = enic_dev_del_station_addr(enic); 1026 if (err) 1027 return err; 1028 1029 err = enic_set_mac_addr(netdev, addr); 1030 if (err) 1031 return err; 1032 1033 return enic_dev_add_station_addr(enic); 1034 } 1035 1036 /* netif_tx_lock held, BHs disabled */ 1037 static void enic_set_rx_mode(struct net_device *netdev) 1038 { 1039 struct enic *enic = netdev_priv(netdev); 1040 int directed = 1; 1041 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0; 1042 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0; 1043 int promisc = (netdev->flags & IFF_PROMISC) || 1044 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS; 1045 int allmulti = (netdev->flags & IFF_ALLMULTI) || 1046 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS; 1047 unsigned int flags = netdev->flags | 1048 (allmulti ? IFF_ALLMULTI : 0) | 1049 (promisc ? IFF_PROMISC : 0); 1050 1051 if (enic->flags != flags) { 1052 enic->flags = flags; 1053 enic_dev_packet_filter(enic, directed, 1054 multicast, broadcast, promisc, allmulti); 1055 } 1056 1057 if (!promisc) { 1058 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync); 1059 if (!allmulti) 1060 __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync); 1061 } 1062 } 1063 1064 /* netif_tx_lock held, BHs disabled */ 1065 static void enic_tx_timeout(struct net_device *netdev, unsigned int txqueue) 1066 { 1067 struct enic *enic = netdev_priv(netdev); 1068 schedule_work(&enic->tx_hang_reset); 1069 } 1070 1071 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac) 1072 { 1073 struct enic *enic = netdev_priv(netdev); 1074 struct enic_port_profile *pp; 1075 int err; 1076 1077 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1078 if (err) 1079 return err; 1080 1081 if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) { 1082 if (vf == PORT_SELF_VF) { 1083 memcpy(pp->vf_mac, mac, ETH_ALEN); 1084 return 0; 1085 } else { 1086 /* 1087 * For sriov vf's set the mac in hw 1088 */ 1089 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic, 1090 vnic_dev_set_mac_addr, mac); 1091 return enic_dev_status_to_errno(err); 1092 } 1093 } else 1094 return -EINVAL; 1095 } 1096 1097 static int enic_set_vf_port(struct net_device *netdev, int vf, 1098 struct nlattr *port[]) 1099 { 1100 static const u8 zero_addr[ETH_ALEN] = {}; 1101 struct enic *enic = netdev_priv(netdev); 1102 struct enic_port_profile prev_pp; 1103 struct enic_port_profile *pp; 1104 int err = 0, restore_pp = 1; 1105 1106 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1107 if (err) 1108 return err; 1109 1110 if (!port[IFLA_PORT_REQUEST]) 1111 return -EOPNOTSUPP; 1112 1113 memcpy(&prev_pp, pp, sizeof(*enic->pp)); 1114 memset(pp, 0, sizeof(*enic->pp)); 1115 1116 pp->set |= ENIC_SET_REQUEST; 1117 pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]); 1118 1119 if (port[IFLA_PORT_PROFILE]) { 1120 if (nla_len(port[IFLA_PORT_PROFILE]) != PORT_PROFILE_MAX) { 1121 memcpy(pp, &prev_pp, sizeof(*pp)); 1122 return -EINVAL; 1123 } 1124 pp->set |= ENIC_SET_NAME; 1125 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]), 1126 PORT_PROFILE_MAX); 1127 } 1128 1129 if (port[IFLA_PORT_INSTANCE_UUID]) { 1130 if (nla_len(port[IFLA_PORT_INSTANCE_UUID]) != PORT_UUID_MAX) { 1131 memcpy(pp, &prev_pp, sizeof(*pp)); 1132 return -EINVAL; 1133 } 1134 pp->set |= ENIC_SET_INSTANCE; 1135 memcpy(pp->instance_uuid, 1136 nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX); 1137 } 1138 1139 if (port[IFLA_PORT_HOST_UUID]) { 1140 if (nla_len(port[IFLA_PORT_HOST_UUID]) != PORT_UUID_MAX) { 1141 memcpy(pp, &prev_pp, sizeof(*pp)); 1142 return -EINVAL; 1143 } 1144 pp->set |= ENIC_SET_HOST; 1145 memcpy(pp->host_uuid, 1146 nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX); 1147 } 1148 1149 if (vf == PORT_SELF_VF) { 1150 /* Special case handling: mac came from IFLA_VF_MAC */ 1151 if (!is_zero_ether_addr(prev_pp.vf_mac)) 1152 memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN); 1153 1154 if (is_zero_ether_addr(netdev->dev_addr)) 1155 eth_hw_addr_random(netdev); 1156 } else { 1157 /* SR-IOV VF: get mac from adapter */ 1158 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic, 1159 vnic_dev_get_mac_addr, pp->mac_addr); 1160 if (err) { 1161 netdev_err(netdev, "Error getting mac for vf %d\n", vf); 1162 memcpy(pp, &prev_pp, sizeof(*pp)); 1163 return enic_dev_status_to_errno(err); 1164 } 1165 } 1166 1167 err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp); 1168 if (err) { 1169 if (restore_pp) { 1170 /* Things are still the way they were: Implicit 1171 * DISASSOCIATE failed 1172 */ 1173 memcpy(pp, &prev_pp, sizeof(*pp)); 1174 } else { 1175 memset(pp, 0, sizeof(*pp)); 1176 if (vf == PORT_SELF_VF) 1177 eth_hw_addr_set(netdev, zero_addr); 1178 } 1179 } else { 1180 /* Set flag to indicate that the port assoc/disassoc 1181 * request has been sent out to fw 1182 */ 1183 pp->set |= ENIC_PORT_REQUEST_APPLIED; 1184 1185 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */ 1186 if (pp->request == PORT_REQUEST_DISASSOCIATE) { 1187 eth_zero_addr(pp->mac_addr); 1188 if (vf == PORT_SELF_VF) 1189 eth_hw_addr_set(netdev, zero_addr); 1190 } 1191 } 1192 1193 if (vf == PORT_SELF_VF) 1194 eth_zero_addr(pp->vf_mac); 1195 1196 return err; 1197 } 1198 1199 static int enic_get_vf_port(struct net_device *netdev, int vf, 1200 struct sk_buff *skb) 1201 { 1202 struct enic *enic = netdev_priv(netdev); 1203 u16 response = PORT_PROFILE_RESPONSE_SUCCESS; 1204 struct enic_port_profile *pp; 1205 int err; 1206 1207 ENIC_PP_BY_INDEX(enic, vf, pp, &err); 1208 if (err) 1209 return err; 1210 1211 if (!(pp->set & ENIC_PORT_REQUEST_APPLIED)) 1212 return -ENODATA; 1213 1214 err = enic_process_get_pp_request(enic, vf, pp->request, &response); 1215 if (err) 1216 return err; 1217 1218 if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) || 1219 nla_put_u16(skb, IFLA_PORT_RESPONSE, response) || 1220 ((pp->set & ENIC_SET_NAME) && 1221 nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) || 1222 ((pp->set & ENIC_SET_INSTANCE) && 1223 nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX, 1224 pp->instance_uuid)) || 1225 ((pp->set & ENIC_SET_HOST) && 1226 nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid))) 1227 goto nla_put_failure; 1228 return 0; 1229 1230 nla_put_failure: 1231 return -EMSGSIZE; 1232 } 1233 1234 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf) 1235 { 1236 struct enic *enic = vnic_dev_priv(rq->vdev); 1237 1238 if (!buf->os_buf) 1239 return; 1240 1241 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len, 1242 DMA_FROM_DEVICE); 1243 dev_kfree_skb_any(buf->os_buf); 1244 buf->os_buf = NULL; 1245 } 1246 1247 static int enic_rq_alloc_buf(struct vnic_rq *rq) 1248 { 1249 struct enic *enic = vnic_dev_priv(rq->vdev); 1250 struct net_device *netdev = enic->netdev; 1251 struct sk_buff *skb; 1252 unsigned int len = netdev->mtu + VLAN_ETH_HLEN; 1253 unsigned int os_buf_index = 0; 1254 dma_addr_t dma_addr; 1255 struct vnic_rq_buf *buf = rq->to_use; 1256 1257 if (buf->os_buf) { 1258 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr, 1259 buf->len); 1260 1261 return 0; 1262 } 1263 skb = netdev_alloc_skb_ip_align(netdev, len); 1264 if (!skb) 1265 return -ENOMEM; 1266 1267 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, len, 1268 DMA_FROM_DEVICE); 1269 if (unlikely(enic_dma_map_check(enic, dma_addr))) { 1270 dev_kfree_skb(skb); 1271 return -ENOMEM; 1272 } 1273 1274 enic_queue_rq_desc(rq, skb, os_buf_index, 1275 dma_addr, len); 1276 1277 return 0; 1278 } 1279 1280 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size, 1281 u32 pkt_len) 1282 { 1283 if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len) 1284 pkt_size->large_pkt_bytes_cnt += pkt_len; 1285 else 1286 pkt_size->small_pkt_bytes_cnt += pkt_len; 1287 } 1288 1289 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb, 1290 struct vnic_rq_buf *buf, u16 len) 1291 { 1292 struct enic *enic = netdev_priv(netdev); 1293 struct sk_buff *new_skb; 1294 1295 if (len > enic->rx_copybreak) 1296 return false; 1297 new_skb = netdev_alloc_skb_ip_align(netdev, len); 1298 if (!new_skb) 1299 return false; 1300 dma_sync_single_for_cpu(&enic->pdev->dev, buf->dma_addr, len, 1301 DMA_FROM_DEVICE); 1302 memcpy(new_skb->data, (*skb)->data, len); 1303 *skb = new_skb; 1304 1305 return true; 1306 } 1307 1308 static void enic_rq_indicate_buf(struct vnic_rq *rq, 1309 struct cq_desc *cq_desc, struct vnic_rq_buf *buf, 1310 int skipped, void *opaque) 1311 { 1312 struct enic *enic = vnic_dev_priv(rq->vdev); 1313 struct net_device *netdev = enic->netdev; 1314 struct sk_buff *skb; 1315 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1316 1317 u8 type, color, eop, sop, ingress_port, vlan_stripped; 1318 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof; 1319 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok; 1320 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc; 1321 u8 packet_error; 1322 u16 q_number, completed_index, bytes_written, vlan_tci, checksum; 1323 u32 rss_hash; 1324 bool outer_csum_ok = true, encap = false; 1325 1326 if (skipped) 1327 return; 1328 1329 skb = buf->os_buf; 1330 1331 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc, 1332 &type, &color, &q_number, &completed_index, 1333 &ingress_port, &fcoe, &eop, &sop, &rss_type, 1334 &csum_not_calc, &rss_hash, &bytes_written, 1335 &packet_error, &vlan_stripped, &vlan_tci, &checksum, 1336 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error, 1337 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp, 1338 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment, 1339 &fcs_ok); 1340 1341 if (packet_error) { 1342 1343 if (!fcs_ok) { 1344 if (bytes_written > 0) 1345 enic->rq_bad_fcs++; 1346 else if (bytes_written == 0) 1347 enic->rq_truncated_pkts++; 1348 } 1349 1350 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len, 1351 DMA_FROM_DEVICE); 1352 dev_kfree_skb_any(skb); 1353 buf->os_buf = NULL; 1354 1355 return; 1356 } 1357 1358 if (eop && bytes_written > 0) { 1359 1360 /* Good receive 1361 */ 1362 1363 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) { 1364 buf->os_buf = NULL; 1365 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, 1366 buf->len, DMA_FROM_DEVICE); 1367 } 1368 prefetch(skb->data - NET_IP_ALIGN); 1369 1370 skb_put(skb, bytes_written); 1371 skb->protocol = eth_type_trans(skb, netdev); 1372 skb_record_rx_queue(skb, q_number); 1373 if ((netdev->features & NETIF_F_RXHASH) && rss_hash && 1374 (type == 3)) { 1375 switch (rss_type) { 1376 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4: 1377 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6: 1378 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX: 1379 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4); 1380 break; 1381 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4: 1382 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6: 1383 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX: 1384 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3); 1385 break; 1386 } 1387 } 1388 if (enic->vxlan.vxlan_udp_port_number) { 1389 switch (enic->vxlan.patch_level) { 1390 case 0: 1391 if (fcoe) { 1392 encap = true; 1393 outer_csum_ok = fcoe_fc_crc_ok; 1394 } 1395 break; 1396 case 2: 1397 if ((type == 7) && 1398 (rss_hash & BIT(0))) { 1399 encap = true; 1400 outer_csum_ok = (rss_hash & BIT(1)) && 1401 (rss_hash & BIT(2)); 1402 } 1403 break; 1404 } 1405 } 1406 1407 /* Hardware does not provide whole packet checksum. It only 1408 * provides pseudo checksum. Since hw validates the packet 1409 * checksum but not provide us the checksum value. use 1410 * CHECSUM_UNNECESSARY. 1411 * 1412 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is 1413 * inner csum_ok. outer_csum_ok is set by hw when outer udp 1414 * csum is correct or is zero. 1415 */ 1416 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc && 1417 tcp_udp_csum_ok && outer_csum_ok && 1418 (ipv4_csum_ok || ipv6)) { 1419 skb->ip_summed = CHECKSUM_UNNECESSARY; 1420 skb->csum_level = encap; 1421 } 1422 1423 if (vlan_stripped) 1424 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci); 1425 1426 skb_mark_napi_id(skb, &enic->napi[rq->index]); 1427 if (!(netdev->features & NETIF_F_GRO)) 1428 netif_receive_skb(skb); 1429 else 1430 napi_gro_receive(&enic->napi[q_number], skb); 1431 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1432 enic_intr_update_pkt_size(&cq->pkt_size_counter, 1433 bytes_written); 1434 } else { 1435 1436 /* Buffer overflow 1437 */ 1438 1439 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len, 1440 DMA_FROM_DEVICE); 1441 dev_kfree_skb_any(skb); 1442 buf->os_buf = NULL; 1443 } 1444 } 1445 1446 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc, 1447 u8 type, u16 q_number, u16 completed_index, void *opaque) 1448 { 1449 struct enic *enic = vnic_dev_priv(vdev); 1450 1451 vnic_rq_service(&enic->rq[q_number], cq_desc, 1452 completed_index, VNIC_RQ_RETURN_DESC, 1453 enic_rq_indicate_buf, opaque); 1454 1455 return 0; 1456 } 1457 1458 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq) 1459 { 1460 unsigned int intr = enic_msix_rq_intr(enic, rq->index); 1461 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1462 u32 timer = cq->tobe_rx_coal_timeval; 1463 1464 if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) { 1465 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer); 1466 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval; 1467 } 1468 } 1469 1470 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq) 1471 { 1472 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting; 1473 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)]; 1474 struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter; 1475 int index; 1476 u32 timer; 1477 u32 range_start; 1478 u32 traffic; 1479 u64 delta; 1480 ktime_t now = ktime_get(); 1481 1482 delta = ktime_us_delta(now, cq->prev_ts); 1483 if (delta < ENIC_AIC_TS_BREAK) 1484 return; 1485 cq->prev_ts = now; 1486 1487 traffic = pkt_size_counter->large_pkt_bytes_cnt + 1488 pkt_size_counter->small_pkt_bytes_cnt; 1489 /* The table takes Mbps 1490 * traffic *= 8 => bits 1491 * traffic *= (10^6 / delta) => bps 1492 * traffic /= 10^6 => Mbps 1493 * 1494 * Combining, traffic *= (8 / delta) 1495 */ 1496 1497 traffic <<= 3; 1498 traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta; 1499 1500 for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++) 1501 if (traffic < mod_table[index].rx_rate) 1502 break; 1503 range_start = (pkt_size_counter->small_pkt_bytes_cnt > 1504 pkt_size_counter->large_pkt_bytes_cnt << 1) ? 1505 rx_coal->small_pkt_range_start : 1506 rx_coal->large_pkt_range_start; 1507 timer = range_start + ((rx_coal->range_end - range_start) * 1508 mod_table[index].range_percent / 100); 1509 /* Damping */ 1510 cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1; 1511 1512 pkt_size_counter->large_pkt_bytes_cnt = 0; 1513 pkt_size_counter->small_pkt_bytes_cnt = 0; 1514 } 1515 1516 static int enic_poll(struct napi_struct *napi, int budget) 1517 { 1518 struct net_device *netdev = napi->dev; 1519 struct enic *enic = netdev_priv(netdev); 1520 unsigned int cq_rq = enic_cq_rq(enic, 0); 1521 unsigned int cq_wq = enic_cq_wq(enic, 0); 1522 unsigned int intr = ENIC_LEGACY_IO_INTR; 1523 unsigned int rq_work_to_do = budget; 1524 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET; 1525 unsigned int work_done, rq_work_done = 0, wq_work_done; 1526 int err; 1527 1528 wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do, 1529 enic_wq_service, NULL); 1530 1531 if (budget > 0) 1532 rq_work_done = vnic_cq_service(&enic->cq[cq_rq], 1533 rq_work_to_do, enic_rq_service, NULL); 1534 1535 /* Accumulate intr event credits for this polling 1536 * cycle. An intr event is the completion of a 1537 * a WQ or RQ packet. 1538 */ 1539 1540 work_done = rq_work_done + wq_work_done; 1541 1542 if (work_done > 0) 1543 vnic_intr_return_credits(&enic->intr[intr], 1544 work_done, 1545 0 /* don't unmask intr */, 1546 0 /* don't reset intr timer */); 1547 1548 err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf); 1549 1550 /* Buffer allocation failed. Stay in polling 1551 * mode so we can try to fill the ring again. 1552 */ 1553 1554 if (err) 1555 rq_work_done = rq_work_to_do; 1556 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1557 /* Call the function which refreshes the intr coalescing timer 1558 * value based on the traffic. 1559 */ 1560 enic_calc_int_moderation(enic, &enic->rq[0]); 1561 1562 if ((rq_work_done < budget) && napi_complete_done(napi, rq_work_done)) { 1563 1564 /* Some work done, but not enough to stay in polling, 1565 * exit polling 1566 */ 1567 1568 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1569 enic_set_int_moderation(enic, &enic->rq[0]); 1570 vnic_intr_unmask(&enic->intr[intr]); 1571 } 1572 1573 return rq_work_done; 1574 } 1575 1576 #ifdef CONFIG_RFS_ACCEL 1577 static void enic_free_rx_cpu_rmap(struct enic *enic) 1578 { 1579 free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap); 1580 enic->netdev->rx_cpu_rmap = NULL; 1581 } 1582 1583 static void enic_set_rx_cpu_rmap(struct enic *enic) 1584 { 1585 int i, res; 1586 1587 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) { 1588 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count); 1589 if (unlikely(!enic->netdev->rx_cpu_rmap)) 1590 return; 1591 for (i = 0; i < enic->rq_count; i++) { 1592 res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap, 1593 enic->msix_entry[i].vector); 1594 if (unlikely(res)) { 1595 enic_free_rx_cpu_rmap(enic); 1596 return; 1597 } 1598 } 1599 } 1600 } 1601 1602 #else 1603 1604 static void enic_free_rx_cpu_rmap(struct enic *enic) 1605 { 1606 } 1607 1608 static void enic_set_rx_cpu_rmap(struct enic *enic) 1609 { 1610 } 1611 1612 #endif /* CONFIG_RFS_ACCEL */ 1613 1614 static int enic_poll_msix_wq(struct napi_struct *napi, int budget) 1615 { 1616 struct net_device *netdev = napi->dev; 1617 struct enic *enic = netdev_priv(netdev); 1618 unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count; 1619 struct vnic_wq *wq = &enic->wq[wq_index]; 1620 unsigned int cq; 1621 unsigned int intr; 1622 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET; 1623 unsigned int wq_work_done; 1624 unsigned int wq_irq; 1625 1626 wq_irq = wq->index; 1627 cq = enic_cq_wq(enic, wq_irq); 1628 intr = enic_msix_wq_intr(enic, wq_irq); 1629 wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do, 1630 enic_wq_service, NULL); 1631 1632 vnic_intr_return_credits(&enic->intr[intr], wq_work_done, 1633 0 /* don't unmask intr */, 1634 1 /* reset intr timer */); 1635 if (!wq_work_done) { 1636 napi_complete(napi); 1637 vnic_intr_unmask(&enic->intr[intr]); 1638 return 0; 1639 } 1640 1641 return budget; 1642 } 1643 1644 static int enic_poll_msix_rq(struct napi_struct *napi, int budget) 1645 { 1646 struct net_device *netdev = napi->dev; 1647 struct enic *enic = netdev_priv(netdev); 1648 unsigned int rq = (napi - &enic->napi[0]); 1649 unsigned int cq = enic_cq_rq(enic, rq); 1650 unsigned int intr = enic_msix_rq_intr(enic, rq); 1651 unsigned int work_to_do = budget; 1652 unsigned int work_done = 0; 1653 int err; 1654 1655 /* Service RQ 1656 */ 1657 1658 if (budget > 0) 1659 work_done = vnic_cq_service(&enic->cq[cq], 1660 work_to_do, enic_rq_service, NULL); 1661 1662 /* Return intr event credits for this polling 1663 * cycle. An intr event is the completion of a 1664 * RQ packet. 1665 */ 1666 1667 if (work_done > 0) 1668 vnic_intr_return_credits(&enic->intr[intr], 1669 work_done, 1670 0 /* don't unmask intr */, 1671 0 /* don't reset intr timer */); 1672 1673 err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf); 1674 1675 /* Buffer allocation failed. Stay in polling mode 1676 * so we can try to fill the ring again. 1677 */ 1678 1679 if (err) 1680 work_done = work_to_do; 1681 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1682 /* Call the function which refreshes the intr coalescing timer 1683 * value based on the traffic. 1684 */ 1685 enic_calc_int_moderation(enic, &enic->rq[rq]); 1686 1687 if ((work_done < budget) && napi_complete_done(napi, work_done)) { 1688 1689 /* Some work done, but not enough to stay in polling, 1690 * exit polling 1691 */ 1692 1693 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce) 1694 enic_set_int_moderation(enic, &enic->rq[rq]); 1695 vnic_intr_unmask(&enic->intr[intr]); 1696 } 1697 1698 return work_done; 1699 } 1700 1701 static void enic_notify_timer(struct timer_list *t) 1702 { 1703 struct enic *enic = from_timer(enic, t, notify_timer); 1704 1705 enic_notify_check(enic); 1706 1707 mod_timer(&enic->notify_timer, 1708 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD)); 1709 } 1710 1711 static void enic_free_intr(struct enic *enic) 1712 { 1713 struct net_device *netdev = enic->netdev; 1714 unsigned int i; 1715 1716 enic_free_rx_cpu_rmap(enic); 1717 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1718 case VNIC_DEV_INTR_MODE_INTX: 1719 free_irq(enic->pdev->irq, netdev); 1720 break; 1721 case VNIC_DEV_INTR_MODE_MSI: 1722 free_irq(enic->pdev->irq, enic); 1723 break; 1724 case VNIC_DEV_INTR_MODE_MSIX: 1725 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) 1726 if (enic->msix[i].requested) 1727 free_irq(enic->msix_entry[i].vector, 1728 enic->msix[i].devid); 1729 break; 1730 default: 1731 break; 1732 } 1733 } 1734 1735 static int enic_request_intr(struct enic *enic) 1736 { 1737 struct net_device *netdev = enic->netdev; 1738 unsigned int i, intr; 1739 int err = 0; 1740 1741 enic_set_rx_cpu_rmap(enic); 1742 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1743 1744 case VNIC_DEV_INTR_MODE_INTX: 1745 1746 err = request_irq(enic->pdev->irq, enic_isr_legacy, 1747 IRQF_SHARED, netdev->name, netdev); 1748 break; 1749 1750 case VNIC_DEV_INTR_MODE_MSI: 1751 1752 err = request_irq(enic->pdev->irq, enic_isr_msi, 1753 0, netdev->name, enic); 1754 break; 1755 1756 case VNIC_DEV_INTR_MODE_MSIX: 1757 1758 for (i = 0; i < enic->rq_count; i++) { 1759 intr = enic_msix_rq_intr(enic, i); 1760 snprintf(enic->msix[intr].devname, 1761 sizeof(enic->msix[intr].devname), 1762 "%s-rx-%u", netdev->name, i); 1763 enic->msix[intr].isr = enic_isr_msix; 1764 enic->msix[intr].devid = &enic->napi[i]; 1765 } 1766 1767 for (i = 0; i < enic->wq_count; i++) { 1768 int wq = enic_cq_wq(enic, i); 1769 1770 intr = enic_msix_wq_intr(enic, i); 1771 snprintf(enic->msix[intr].devname, 1772 sizeof(enic->msix[intr].devname), 1773 "%s-tx-%u", netdev->name, i); 1774 enic->msix[intr].isr = enic_isr_msix; 1775 enic->msix[intr].devid = &enic->napi[wq]; 1776 } 1777 1778 intr = enic_msix_err_intr(enic); 1779 snprintf(enic->msix[intr].devname, 1780 sizeof(enic->msix[intr].devname), 1781 "%s-err", netdev->name); 1782 enic->msix[intr].isr = enic_isr_msix_err; 1783 enic->msix[intr].devid = enic; 1784 1785 intr = enic_msix_notify_intr(enic); 1786 snprintf(enic->msix[intr].devname, 1787 sizeof(enic->msix[intr].devname), 1788 "%s-notify", netdev->name); 1789 enic->msix[intr].isr = enic_isr_msix_notify; 1790 enic->msix[intr].devid = enic; 1791 1792 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) 1793 enic->msix[i].requested = 0; 1794 1795 for (i = 0; i < enic->intr_count; i++) { 1796 err = request_irq(enic->msix_entry[i].vector, 1797 enic->msix[i].isr, 0, 1798 enic->msix[i].devname, 1799 enic->msix[i].devid); 1800 if (err) { 1801 enic_free_intr(enic); 1802 break; 1803 } 1804 enic->msix[i].requested = 1; 1805 } 1806 1807 break; 1808 1809 default: 1810 break; 1811 } 1812 1813 return err; 1814 } 1815 1816 static void enic_synchronize_irqs(struct enic *enic) 1817 { 1818 unsigned int i; 1819 1820 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1821 case VNIC_DEV_INTR_MODE_INTX: 1822 case VNIC_DEV_INTR_MODE_MSI: 1823 synchronize_irq(enic->pdev->irq); 1824 break; 1825 case VNIC_DEV_INTR_MODE_MSIX: 1826 for (i = 0; i < enic->intr_count; i++) 1827 synchronize_irq(enic->msix_entry[i].vector); 1828 break; 1829 default: 1830 break; 1831 } 1832 } 1833 1834 static void enic_set_rx_coal_setting(struct enic *enic) 1835 { 1836 unsigned int speed; 1837 int index = -1; 1838 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting; 1839 1840 /* 1. Read the link speed from fw 1841 * 2. Pick the default range for the speed 1842 * 3. Update it in enic->rx_coalesce_setting 1843 */ 1844 speed = vnic_dev_port_speed(enic->vdev); 1845 if (ENIC_LINK_SPEED_10G < speed) 1846 index = ENIC_LINK_40G_INDEX; 1847 else if (ENIC_LINK_SPEED_4G < speed) 1848 index = ENIC_LINK_10G_INDEX; 1849 else 1850 index = ENIC_LINK_4G_INDEX; 1851 1852 rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start; 1853 rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start; 1854 rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END; 1855 1856 /* Start with the value provided by UCSM */ 1857 for (index = 0; index < enic->rq_count; index++) 1858 enic->cq[index].cur_rx_coal_timeval = 1859 enic->config.intr_timer_usec; 1860 1861 rx_coal->use_adaptive_rx_coalesce = 1; 1862 } 1863 1864 static int enic_dev_notify_set(struct enic *enic) 1865 { 1866 int err; 1867 1868 spin_lock_bh(&enic->devcmd_lock); 1869 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1870 case VNIC_DEV_INTR_MODE_INTX: 1871 err = vnic_dev_notify_set(enic->vdev, ENIC_LEGACY_NOTIFY_INTR); 1872 break; 1873 case VNIC_DEV_INTR_MODE_MSIX: 1874 err = vnic_dev_notify_set(enic->vdev, 1875 enic_msix_notify_intr(enic)); 1876 break; 1877 default: 1878 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */); 1879 break; 1880 } 1881 spin_unlock_bh(&enic->devcmd_lock); 1882 1883 return err; 1884 } 1885 1886 static void enic_notify_timer_start(struct enic *enic) 1887 { 1888 switch (vnic_dev_get_intr_mode(enic->vdev)) { 1889 case VNIC_DEV_INTR_MODE_MSI: 1890 mod_timer(&enic->notify_timer, jiffies); 1891 break; 1892 default: 1893 /* Using intr for notification for INTx/MSI-X */ 1894 break; 1895 } 1896 } 1897 1898 /* rtnl lock is held, process context */ 1899 static int enic_open(struct net_device *netdev) 1900 { 1901 struct enic *enic = netdev_priv(netdev); 1902 unsigned int i; 1903 int err, ret; 1904 1905 err = enic_request_intr(enic); 1906 if (err) { 1907 netdev_err(netdev, "Unable to request irq.\n"); 1908 return err; 1909 } 1910 enic_init_affinity_hint(enic); 1911 enic_set_affinity_hint(enic); 1912 1913 err = enic_dev_notify_set(enic); 1914 if (err) { 1915 netdev_err(netdev, 1916 "Failed to alloc notify buffer, aborting.\n"); 1917 goto err_out_free_intr; 1918 } 1919 1920 for (i = 0; i < enic->rq_count; i++) { 1921 /* enable rq before updating rq desc */ 1922 vnic_rq_enable(&enic->rq[i]); 1923 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf); 1924 /* Need at least one buffer on ring to get going */ 1925 if (vnic_rq_desc_used(&enic->rq[i]) == 0) { 1926 netdev_err(netdev, "Unable to alloc receive buffers\n"); 1927 err = -ENOMEM; 1928 goto err_out_free_rq; 1929 } 1930 } 1931 1932 for (i = 0; i < enic->wq_count; i++) 1933 vnic_wq_enable(&enic->wq[i]); 1934 1935 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic)) 1936 enic_dev_add_station_addr(enic); 1937 1938 enic_set_rx_mode(netdev); 1939 1940 netif_tx_wake_all_queues(netdev); 1941 1942 for (i = 0; i < enic->rq_count; i++) 1943 napi_enable(&enic->napi[i]); 1944 1945 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) 1946 for (i = 0; i < enic->wq_count; i++) 1947 napi_enable(&enic->napi[enic_cq_wq(enic, i)]); 1948 enic_dev_enable(enic); 1949 1950 for (i = 0; i < enic->intr_count; i++) 1951 vnic_intr_unmask(&enic->intr[i]); 1952 1953 enic_notify_timer_start(enic); 1954 enic_rfs_timer_start(enic); 1955 1956 return 0; 1957 1958 err_out_free_rq: 1959 for (i = 0; i < enic->rq_count; i++) { 1960 ret = vnic_rq_disable(&enic->rq[i]); 1961 if (!ret) 1962 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf); 1963 } 1964 enic_dev_notify_unset(enic); 1965 err_out_free_intr: 1966 enic_unset_affinity_hint(enic); 1967 enic_free_intr(enic); 1968 1969 return err; 1970 } 1971 1972 /* rtnl lock is held, process context */ 1973 static int enic_stop(struct net_device *netdev) 1974 { 1975 struct enic *enic = netdev_priv(netdev); 1976 unsigned int i; 1977 int err; 1978 1979 for (i = 0; i < enic->intr_count; i++) { 1980 vnic_intr_mask(&enic->intr[i]); 1981 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */ 1982 } 1983 1984 enic_synchronize_irqs(enic); 1985 1986 del_timer_sync(&enic->notify_timer); 1987 enic_rfs_flw_tbl_free(enic); 1988 1989 enic_dev_disable(enic); 1990 1991 for (i = 0; i < enic->rq_count; i++) 1992 napi_disable(&enic->napi[i]); 1993 1994 netif_carrier_off(netdev); 1995 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) 1996 for (i = 0; i < enic->wq_count; i++) 1997 napi_disable(&enic->napi[enic_cq_wq(enic, i)]); 1998 netif_tx_disable(netdev); 1999 2000 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic)) 2001 enic_dev_del_station_addr(enic); 2002 2003 for (i = 0; i < enic->wq_count; i++) { 2004 err = vnic_wq_disable(&enic->wq[i]); 2005 if (err) 2006 return err; 2007 } 2008 for (i = 0; i < enic->rq_count; i++) { 2009 err = vnic_rq_disable(&enic->rq[i]); 2010 if (err) 2011 return err; 2012 } 2013 2014 enic_dev_notify_unset(enic); 2015 enic_unset_affinity_hint(enic); 2016 enic_free_intr(enic); 2017 2018 for (i = 0; i < enic->wq_count; i++) 2019 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf); 2020 for (i = 0; i < enic->rq_count; i++) 2021 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf); 2022 for (i = 0; i < enic->cq_count; i++) 2023 vnic_cq_clean(&enic->cq[i]); 2024 for (i = 0; i < enic->intr_count; i++) 2025 vnic_intr_clean(&enic->intr[i]); 2026 2027 return 0; 2028 } 2029 2030 static int _enic_change_mtu(struct net_device *netdev, int new_mtu) 2031 { 2032 bool running = netif_running(netdev); 2033 int err = 0; 2034 2035 ASSERT_RTNL(); 2036 if (running) { 2037 err = enic_stop(netdev); 2038 if (err) 2039 return err; 2040 } 2041 2042 netdev->mtu = new_mtu; 2043 2044 if (running) { 2045 err = enic_open(netdev); 2046 if (err) 2047 return err; 2048 } 2049 2050 return 0; 2051 } 2052 2053 static int enic_change_mtu(struct net_device *netdev, int new_mtu) 2054 { 2055 struct enic *enic = netdev_priv(netdev); 2056 2057 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) 2058 return -EOPNOTSUPP; 2059 2060 if (netdev->mtu > enic->port_mtu) 2061 netdev_warn(netdev, 2062 "interface MTU (%d) set higher than port MTU (%d)\n", 2063 netdev->mtu, enic->port_mtu); 2064 2065 return _enic_change_mtu(netdev, new_mtu); 2066 } 2067 2068 static void enic_change_mtu_work(struct work_struct *work) 2069 { 2070 struct enic *enic = container_of(work, struct enic, change_mtu_work); 2071 struct net_device *netdev = enic->netdev; 2072 int new_mtu = vnic_dev_mtu(enic->vdev); 2073 2074 rtnl_lock(); 2075 (void)_enic_change_mtu(netdev, new_mtu); 2076 rtnl_unlock(); 2077 2078 netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu); 2079 } 2080 2081 #ifdef CONFIG_NET_POLL_CONTROLLER 2082 static void enic_poll_controller(struct net_device *netdev) 2083 { 2084 struct enic *enic = netdev_priv(netdev); 2085 struct vnic_dev *vdev = enic->vdev; 2086 unsigned int i, intr; 2087 2088 switch (vnic_dev_get_intr_mode(vdev)) { 2089 case VNIC_DEV_INTR_MODE_MSIX: 2090 for (i = 0; i < enic->rq_count; i++) { 2091 intr = enic_msix_rq_intr(enic, i); 2092 enic_isr_msix(enic->msix_entry[intr].vector, 2093 &enic->napi[i]); 2094 } 2095 2096 for (i = 0; i < enic->wq_count; i++) { 2097 intr = enic_msix_wq_intr(enic, i); 2098 enic_isr_msix(enic->msix_entry[intr].vector, 2099 &enic->napi[enic_cq_wq(enic, i)]); 2100 } 2101 2102 break; 2103 case VNIC_DEV_INTR_MODE_MSI: 2104 enic_isr_msi(enic->pdev->irq, enic); 2105 break; 2106 case VNIC_DEV_INTR_MODE_INTX: 2107 enic_isr_legacy(enic->pdev->irq, netdev); 2108 break; 2109 default: 2110 break; 2111 } 2112 } 2113 #endif 2114 2115 static int enic_dev_wait(struct vnic_dev *vdev, 2116 int (*start)(struct vnic_dev *, int), 2117 int (*finished)(struct vnic_dev *, int *), 2118 int arg) 2119 { 2120 unsigned long time; 2121 int done; 2122 int err; 2123 2124 err = start(vdev, arg); 2125 if (err) 2126 return err; 2127 2128 /* Wait for func to complete...2 seconds max 2129 */ 2130 2131 time = jiffies + (HZ * 2); 2132 do { 2133 2134 err = finished(vdev, &done); 2135 if (err) 2136 return err; 2137 2138 if (done) 2139 return 0; 2140 2141 schedule_timeout_uninterruptible(HZ / 10); 2142 2143 } while (time_after(time, jiffies)); 2144 2145 return -ETIMEDOUT; 2146 } 2147 2148 static int enic_dev_open(struct enic *enic) 2149 { 2150 int err; 2151 u32 flags = CMD_OPENF_IG_DESCCACHE; 2152 2153 err = enic_dev_wait(enic->vdev, vnic_dev_open, 2154 vnic_dev_open_done, flags); 2155 if (err) 2156 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n", 2157 err); 2158 2159 return err; 2160 } 2161 2162 static int enic_dev_soft_reset(struct enic *enic) 2163 { 2164 int err; 2165 2166 err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset, 2167 vnic_dev_soft_reset_done, 0); 2168 if (err) 2169 netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n", 2170 err); 2171 2172 return err; 2173 } 2174 2175 static int enic_dev_hang_reset(struct enic *enic) 2176 { 2177 int err; 2178 2179 err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset, 2180 vnic_dev_hang_reset_done, 0); 2181 if (err) 2182 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n", 2183 err); 2184 2185 return err; 2186 } 2187 2188 int __enic_set_rsskey(struct enic *enic) 2189 { 2190 union vnic_rss_key *rss_key_buf_va; 2191 dma_addr_t rss_key_buf_pa; 2192 int i, kidx, bidx, err; 2193 2194 rss_key_buf_va = dma_alloc_coherent(&enic->pdev->dev, 2195 sizeof(union vnic_rss_key), 2196 &rss_key_buf_pa, GFP_ATOMIC); 2197 if (!rss_key_buf_va) 2198 return -ENOMEM; 2199 2200 for (i = 0; i < ENIC_RSS_LEN; i++) { 2201 kidx = i / ENIC_RSS_BYTES_PER_KEY; 2202 bidx = i % ENIC_RSS_BYTES_PER_KEY; 2203 rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i]; 2204 } 2205 spin_lock_bh(&enic->devcmd_lock); 2206 err = enic_set_rss_key(enic, 2207 rss_key_buf_pa, 2208 sizeof(union vnic_rss_key)); 2209 spin_unlock_bh(&enic->devcmd_lock); 2210 2211 dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_key), 2212 rss_key_buf_va, rss_key_buf_pa); 2213 2214 return err; 2215 } 2216 2217 static int enic_set_rsskey(struct enic *enic) 2218 { 2219 netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN); 2220 2221 return __enic_set_rsskey(enic); 2222 } 2223 2224 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits) 2225 { 2226 dma_addr_t rss_cpu_buf_pa; 2227 union vnic_rss_cpu *rss_cpu_buf_va = NULL; 2228 unsigned int i; 2229 int err; 2230 2231 rss_cpu_buf_va = dma_alloc_coherent(&enic->pdev->dev, 2232 sizeof(union vnic_rss_cpu), 2233 &rss_cpu_buf_pa, GFP_ATOMIC); 2234 if (!rss_cpu_buf_va) 2235 return -ENOMEM; 2236 2237 for (i = 0; i < (1 << rss_hash_bits); i++) 2238 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count; 2239 2240 spin_lock_bh(&enic->devcmd_lock); 2241 err = enic_set_rss_cpu(enic, 2242 rss_cpu_buf_pa, 2243 sizeof(union vnic_rss_cpu)); 2244 spin_unlock_bh(&enic->devcmd_lock); 2245 2246 dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_cpu), 2247 rss_cpu_buf_va, rss_cpu_buf_pa); 2248 2249 return err; 2250 } 2251 2252 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu, 2253 u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable) 2254 { 2255 const u8 tso_ipid_split_en = 0; 2256 const u8 ig_vlan_strip_en = 1; 2257 int err; 2258 2259 /* Enable VLAN tag stripping. 2260 */ 2261 2262 spin_lock_bh(&enic->devcmd_lock); 2263 err = enic_set_nic_cfg(enic, 2264 rss_default_cpu, rss_hash_type, 2265 rss_hash_bits, rss_base_cpu, 2266 rss_enable, tso_ipid_split_en, 2267 ig_vlan_strip_en); 2268 spin_unlock_bh(&enic->devcmd_lock); 2269 2270 return err; 2271 } 2272 2273 static int enic_set_rss_nic_cfg(struct enic *enic) 2274 { 2275 struct device *dev = enic_get_dev(enic); 2276 const u8 rss_default_cpu = 0; 2277 const u8 rss_hash_bits = 7; 2278 const u8 rss_base_cpu = 0; 2279 u8 rss_hash_type; 2280 int res; 2281 u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1); 2282 2283 spin_lock_bh(&enic->devcmd_lock); 2284 res = vnic_dev_capable_rss_hash_type(enic->vdev, &rss_hash_type); 2285 spin_unlock_bh(&enic->devcmd_lock); 2286 if (res) { 2287 /* defaults for old adapters 2288 */ 2289 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 | 2290 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 | 2291 NIC_CFG_RSS_HASH_TYPE_IPV6 | 2292 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6; 2293 } 2294 2295 if (rss_enable) { 2296 if (!enic_set_rsskey(enic)) { 2297 if (enic_set_rsscpu(enic, rss_hash_bits)) { 2298 rss_enable = 0; 2299 dev_warn(dev, "RSS disabled, " 2300 "Failed to set RSS cpu indirection table."); 2301 } 2302 } else { 2303 rss_enable = 0; 2304 dev_warn(dev, "RSS disabled, Failed to set RSS key.\n"); 2305 } 2306 } 2307 2308 return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type, 2309 rss_hash_bits, rss_base_cpu, rss_enable); 2310 } 2311 2312 static void enic_set_api_busy(struct enic *enic, bool busy) 2313 { 2314 spin_lock(&enic->enic_api_lock); 2315 enic->enic_api_busy = busy; 2316 spin_unlock(&enic->enic_api_lock); 2317 } 2318 2319 static void enic_reset(struct work_struct *work) 2320 { 2321 struct enic *enic = container_of(work, struct enic, reset); 2322 2323 if (!netif_running(enic->netdev)) 2324 return; 2325 2326 rtnl_lock(); 2327 2328 /* Stop any activity from infiniband */ 2329 enic_set_api_busy(enic, true); 2330 2331 enic_stop(enic->netdev); 2332 enic_dev_soft_reset(enic); 2333 enic_reset_addr_lists(enic); 2334 enic_init_vnic_resources(enic); 2335 enic_set_rss_nic_cfg(enic); 2336 enic_dev_set_ig_vlan_rewrite_mode(enic); 2337 enic_open(enic->netdev); 2338 2339 /* Allow infiniband to fiddle with the device again */ 2340 enic_set_api_busy(enic, false); 2341 2342 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev); 2343 2344 rtnl_unlock(); 2345 } 2346 2347 static void enic_tx_hang_reset(struct work_struct *work) 2348 { 2349 struct enic *enic = container_of(work, struct enic, tx_hang_reset); 2350 2351 rtnl_lock(); 2352 2353 /* Stop any activity from infiniband */ 2354 enic_set_api_busy(enic, true); 2355 2356 enic_dev_hang_notify(enic); 2357 enic_stop(enic->netdev); 2358 enic_dev_hang_reset(enic); 2359 enic_reset_addr_lists(enic); 2360 enic_init_vnic_resources(enic); 2361 enic_set_rss_nic_cfg(enic); 2362 enic_dev_set_ig_vlan_rewrite_mode(enic); 2363 enic_open(enic->netdev); 2364 2365 /* Allow infiniband to fiddle with the device again */ 2366 enic_set_api_busy(enic, false); 2367 2368 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev); 2369 2370 rtnl_unlock(); 2371 } 2372 2373 static int enic_set_intr_mode(struct enic *enic) 2374 { 2375 unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX); 2376 unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX); 2377 unsigned int i; 2378 2379 /* Set interrupt mode (INTx, MSI, MSI-X) depending 2380 * on system capabilities. 2381 * 2382 * Try MSI-X first 2383 * 2384 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs 2385 * (the second to last INTR is used for WQ/RQ errors) 2386 * (the last INTR is used for notifications) 2387 */ 2388 2389 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2); 2390 for (i = 0; i < n + m + 2; i++) 2391 enic->msix_entry[i].entry = i; 2392 2393 /* Use multiple RQs if RSS is enabled 2394 */ 2395 2396 if (ENIC_SETTING(enic, RSS) && 2397 enic->config.intr_mode < 1 && 2398 enic->rq_count >= n && 2399 enic->wq_count >= m && 2400 enic->cq_count >= n + m && 2401 enic->intr_count >= n + m + 2) { 2402 2403 if (pci_enable_msix_range(enic->pdev, enic->msix_entry, 2404 n + m + 2, n + m + 2) > 0) { 2405 2406 enic->rq_count = n; 2407 enic->wq_count = m; 2408 enic->cq_count = n + m; 2409 enic->intr_count = n + m + 2; 2410 2411 vnic_dev_set_intr_mode(enic->vdev, 2412 VNIC_DEV_INTR_MODE_MSIX); 2413 2414 return 0; 2415 } 2416 } 2417 2418 if (enic->config.intr_mode < 1 && 2419 enic->rq_count >= 1 && 2420 enic->wq_count >= m && 2421 enic->cq_count >= 1 + m && 2422 enic->intr_count >= 1 + m + 2) { 2423 if (pci_enable_msix_range(enic->pdev, enic->msix_entry, 2424 1 + m + 2, 1 + m + 2) > 0) { 2425 2426 enic->rq_count = 1; 2427 enic->wq_count = m; 2428 enic->cq_count = 1 + m; 2429 enic->intr_count = 1 + m + 2; 2430 2431 vnic_dev_set_intr_mode(enic->vdev, 2432 VNIC_DEV_INTR_MODE_MSIX); 2433 2434 return 0; 2435 } 2436 } 2437 2438 /* Next try MSI 2439 * 2440 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR 2441 */ 2442 2443 if (enic->config.intr_mode < 2 && 2444 enic->rq_count >= 1 && 2445 enic->wq_count >= 1 && 2446 enic->cq_count >= 2 && 2447 enic->intr_count >= 1 && 2448 !pci_enable_msi(enic->pdev)) { 2449 2450 enic->rq_count = 1; 2451 enic->wq_count = 1; 2452 enic->cq_count = 2; 2453 enic->intr_count = 1; 2454 2455 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI); 2456 2457 return 0; 2458 } 2459 2460 /* Next try INTx 2461 * 2462 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs 2463 * (the first INTR is used for WQ/RQ) 2464 * (the second INTR is used for WQ/RQ errors) 2465 * (the last INTR is used for notifications) 2466 */ 2467 2468 if (enic->config.intr_mode < 3 && 2469 enic->rq_count >= 1 && 2470 enic->wq_count >= 1 && 2471 enic->cq_count >= 2 && 2472 enic->intr_count >= 3) { 2473 2474 enic->rq_count = 1; 2475 enic->wq_count = 1; 2476 enic->cq_count = 2; 2477 enic->intr_count = 3; 2478 2479 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX); 2480 2481 return 0; 2482 } 2483 2484 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN); 2485 2486 return -EINVAL; 2487 } 2488 2489 static void enic_clear_intr_mode(struct enic *enic) 2490 { 2491 switch (vnic_dev_get_intr_mode(enic->vdev)) { 2492 case VNIC_DEV_INTR_MODE_MSIX: 2493 pci_disable_msix(enic->pdev); 2494 break; 2495 case VNIC_DEV_INTR_MODE_MSI: 2496 pci_disable_msi(enic->pdev); 2497 break; 2498 default: 2499 break; 2500 } 2501 2502 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN); 2503 } 2504 2505 static const struct net_device_ops enic_netdev_dynamic_ops = { 2506 .ndo_open = enic_open, 2507 .ndo_stop = enic_stop, 2508 .ndo_start_xmit = enic_hard_start_xmit, 2509 .ndo_get_stats64 = enic_get_stats, 2510 .ndo_validate_addr = eth_validate_addr, 2511 .ndo_set_rx_mode = enic_set_rx_mode, 2512 .ndo_set_mac_address = enic_set_mac_address_dynamic, 2513 .ndo_change_mtu = enic_change_mtu, 2514 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid, 2515 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid, 2516 .ndo_tx_timeout = enic_tx_timeout, 2517 .ndo_set_vf_port = enic_set_vf_port, 2518 .ndo_get_vf_port = enic_get_vf_port, 2519 .ndo_set_vf_mac = enic_set_vf_mac, 2520 #ifdef CONFIG_NET_POLL_CONTROLLER 2521 .ndo_poll_controller = enic_poll_controller, 2522 #endif 2523 #ifdef CONFIG_RFS_ACCEL 2524 .ndo_rx_flow_steer = enic_rx_flow_steer, 2525 #endif 2526 .ndo_features_check = enic_features_check, 2527 }; 2528 2529 static const struct net_device_ops enic_netdev_ops = { 2530 .ndo_open = enic_open, 2531 .ndo_stop = enic_stop, 2532 .ndo_start_xmit = enic_hard_start_xmit, 2533 .ndo_get_stats64 = enic_get_stats, 2534 .ndo_validate_addr = eth_validate_addr, 2535 .ndo_set_mac_address = enic_set_mac_address, 2536 .ndo_set_rx_mode = enic_set_rx_mode, 2537 .ndo_change_mtu = enic_change_mtu, 2538 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid, 2539 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid, 2540 .ndo_tx_timeout = enic_tx_timeout, 2541 .ndo_set_vf_port = enic_set_vf_port, 2542 .ndo_get_vf_port = enic_get_vf_port, 2543 .ndo_set_vf_mac = enic_set_vf_mac, 2544 #ifdef CONFIG_NET_POLL_CONTROLLER 2545 .ndo_poll_controller = enic_poll_controller, 2546 #endif 2547 #ifdef CONFIG_RFS_ACCEL 2548 .ndo_rx_flow_steer = enic_rx_flow_steer, 2549 #endif 2550 .ndo_features_check = enic_features_check, 2551 }; 2552 2553 static void enic_dev_deinit(struct enic *enic) 2554 { 2555 unsigned int i; 2556 2557 for (i = 0; i < enic->rq_count; i++) 2558 __netif_napi_del(&enic->napi[i]); 2559 2560 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) 2561 for (i = 0; i < enic->wq_count; i++) 2562 __netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]); 2563 2564 /* observe RCU grace period after __netif_napi_del() calls */ 2565 synchronize_net(); 2566 2567 enic_free_vnic_resources(enic); 2568 enic_clear_intr_mode(enic); 2569 enic_free_affinity_hint(enic); 2570 } 2571 2572 static void enic_kdump_kernel_config(struct enic *enic) 2573 { 2574 if (is_kdump_kernel()) { 2575 dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n"); 2576 enic->rq_count = 1; 2577 enic->wq_count = 1; 2578 enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS; 2579 enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS; 2580 enic->config.mtu = min_t(u16, 1500, enic->config.mtu); 2581 } 2582 } 2583 2584 static int enic_dev_init(struct enic *enic) 2585 { 2586 struct device *dev = enic_get_dev(enic); 2587 struct net_device *netdev = enic->netdev; 2588 unsigned int i; 2589 int err; 2590 2591 /* Get interrupt coalesce timer info */ 2592 err = enic_dev_intr_coal_timer_info(enic); 2593 if (err) { 2594 dev_warn(dev, "Using default conversion factor for " 2595 "interrupt coalesce timer\n"); 2596 vnic_dev_intr_coal_timer_info_default(enic->vdev); 2597 } 2598 2599 /* Get vNIC configuration 2600 */ 2601 2602 err = enic_get_vnic_config(enic); 2603 if (err) { 2604 dev_err(dev, "Get vNIC configuration failed, aborting\n"); 2605 return err; 2606 } 2607 2608 /* Get available resource counts 2609 */ 2610 2611 enic_get_res_counts(enic); 2612 2613 /* modify resource count if we are in kdump_kernel 2614 */ 2615 enic_kdump_kernel_config(enic); 2616 2617 /* Set interrupt mode based on resource counts and system 2618 * capabilities 2619 */ 2620 2621 err = enic_set_intr_mode(enic); 2622 if (err) { 2623 dev_err(dev, "Failed to set intr mode based on resource " 2624 "counts and system capabilities, aborting\n"); 2625 return err; 2626 } 2627 2628 /* Allocate and configure vNIC resources 2629 */ 2630 2631 err = enic_alloc_vnic_resources(enic); 2632 if (err) { 2633 dev_err(dev, "Failed to alloc vNIC resources, aborting\n"); 2634 goto err_out_free_vnic_resources; 2635 } 2636 2637 enic_init_vnic_resources(enic); 2638 2639 err = enic_set_rss_nic_cfg(enic); 2640 if (err) { 2641 dev_err(dev, "Failed to config nic, aborting\n"); 2642 goto err_out_free_vnic_resources; 2643 } 2644 2645 switch (vnic_dev_get_intr_mode(enic->vdev)) { 2646 default: 2647 netif_napi_add(netdev, &enic->napi[0], enic_poll); 2648 break; 2649 case VNIC_DEV_INTR_MODE_MSIX: 2650 for (i = 0; i < enic->rq_count; i++) { 2651 netif_napi_add(netdev, &enic->napi[i], 2652 enic_poll_msix_rq); 2653 } 2654 for (i = 0; i < enic->wq_count; i++) 2655 netif_napi_add(netdev, 2656 &enic->napi[enic_cq_wq(enic, i)], 2657 enic_poll_msix_wq); 2658 break; 2659 } 2660 2661 return 0; 2662 2663 err_out_free_vnic_resources: 2664 enic_free_affinity_hint(enic); 2665 enic_clear_intr_mode(enic); 2666 enic_free_vnic_resources(enic); 2667 2668 return err; 2669 } 2670 2671 static void enic_iounmap(struct enic *enic) 2672 { 2673 unsigned int i; 2674 2675 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) 2676 if (enic->bar[i].vaddr) 2677 iounmap(enic->bar[i].vaddr); 2678 } 2679 2680 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 2681 { 2682 struct device *dev = &pdev->dev; 2683 struct net_device *netdev; 2684 struct enic *enic; 2685 int using_dac = 0; 2686 unsigned int i; 2687 int err; 2688 #ifdef CONFIG_PCI_IOV 2689 int pos = 0; 2690 #endif 2691 int num_pps = 1; 2692 2693 /* Allocate net device structure and initialize. Private 2694 * instance data is initialized to zero. 2695 */ 2696 2697 netdev = alloc_etherdev_mqs(sizeof(struct enic), 2698 ENIC_RQ_MAX, ENIC_WQ_MAX); 2699 if (!netdev) 2700 return -ENOMEM; 2701 2702 pci_set_drvdata(pdev, netdev); 2703 2704 SET_NETDEV_DEV(netdev, &pdev->dev); 2705 2706 enic = netdev_priv(netdev); 2707 enic->netdev = netdev; 2708 enic->pdev = pdev; 2709 2710 /* Setup PCI resources 2711 */ 2712 2713 err = pci_enable_device_mem(pdev); 2714 if (err) { 2715 dev_err(dev, "Cannot enable PCI device, aborting\n"); 2716 goto err_out_free_netdev; 2717 } 2718 2719 err = pci_request_regions(pdev, DRV_NAME); 2720 if (err) { 2721 dev_err(dev, "Cannot request PCI regions, aborting\n"); 2722 goto err_out_disable_device; 2723 } 2724 2725 pci_set_master(pdev); 2726 2727 /* Query PCI controller on system for DMA addressing 2728 * limitation for the device. Try 47-bit first, and 2729 * fail to 32-bit. 2730 */ 2731 2732 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(47)); 2733 if (err) { 2734 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); 2735 if (err) { 2736 dev_err(dev, "No usable DMA configuration, aborting\n"); 2737 goto err_out_release_regions; 2738 } 2739 } else { 2740 using_dac = 1; 2741 } 2742 2743 /* Map vNIC resources from BAR0-5 2744 */ 2745 2746 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) { 2747 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM)) 2748 continue; 2749 enic->bar[i].len = pci_resource_len(pdev, i); 2750 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len); 2751 if (!enic->bar[i].vaddr) { 2752 dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i); 2753 err = -ENODEV; 2754 goto err_out_iounmap; 2755 } 2756 enic->bar[i].bus_addr = pci_resource_start(pdev, i); 2757 } 2758 2759 /* Register vNIC device 2760 */ 2761 2762 enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar, 2763 ARRAY_SIZE(enic->bar)); 2764 if (!enic->vdev) { 2765 dev_err(dev, "vNIC registration failed, aborting\n"); 2766 err = -ENODEV; 2767 goto err_out_iounmap; 2768 } 2769 2770 err = vnic_devcmd_init(enic->vdev); 2771 2772 if (err) 2773 goto err_out_vnic_unregister; 2774 2775 #ifdef CONFIG_PCI_IOV 2776 /* Get number of subvnics */ 2777 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV); 2778 if (pos) { 2779 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF, 2780 &enic->num_vfs); 2781 if (enic->num_vfs) { 2782 err = pci_enable_sriov(pdev, enic->num_vfs); 2783 if (err) { 2784 dev_err(dev, "SRIOV enable failed, aborting." 2785 " pci_enable_sriov() returned %d\n", 2786 err); 2787 goto err_out_vnic_unregister; 2788 } 2789 enic->priv_flags |= ENIC_SRIOV_ENABLED; 2790 num_pps = enic->num_vfs; 2791 } 2792 } 2793 #endif 2794 2795 /* Allocate structure for port profiles */ 2796 enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL); 2797 if (!enic->pp) { 2798 err = -ENOMEM; 2799 goto err_out_disable_sriov_pp; 2800 } 2801 2802 /* Issue device open to get device in known state 2803 */ 2804 2805 err = enic_dev_open(enic); 2806 if (err) { 2807 dev_err(dev, "vNIC dev open failed, aborting\n"); 2808 goto err_out_disable_sriov; 2809 } 2810 2811 /* Setup devcmd lock 2812 */ 2813 2814 spin_lock_init(&enic->devcmd_lock); 2815 spin_lock_init(&enic->enic_api_lock); 2816 2817 /* 2818 * Set ingress vlan rewrite mode before vnic initialization 2819 */ 2820 2821 err = enic_dev_set_ig_vlan_rewrite_mode(enic); 2822 if (err) { 2823 dev_err(dev, 2824 "Failed to set ingress vlan rewrite mode, aborting.\n"); 2825 goto err_out_dev_close; 2826 } 2827 2828 /* Issue device init to initialize the vnic-to-switch link. 2829 * We'll start with carrier off and wait for link UP 2830 * notification later to turn on carrier. We don't need 2831 * to wait here for the vnic-to-switch link initialization 2832 * to complete; link UP notification is the indication that 2833 * the process is complete. 2834 */ 2835 2836 netif_carrier_off(netdev); 2837 2838 /* Do not call dev_init for a dynamic vnic. 2839 * For a dynamic vnic, init_prov_info will be 2840 * called later by an upper layer. 2841 */ 2842 2843 if (!enic_is_dynamic(enic)) { 2844 err = vnic_dev_init(enic->vdev, 0); 2845 if (err) { 2846 dev_err(dev, "vNIC dev init failed, aborting\n"); 2847 goto err_out_dev_close; 2848 } 2849 } 2850 2851 err = enic_dev_init(enic); 2852 if (err) { 2853 dev_err(dev, "Device initialization failed, aborting\n"); 2854 goto err_out_dev_close; 2855 } 2856 2857 netif_set_real_num_tx_queues(netdev, enic->wq_count); 2858 netif_set_real_num_rx_queues(netdev, enic->rq_count); 2859 2860 /* Setup notification timer, HW reset task, and wq locks 2861 */ 2862 2863 timer_setup(&enic->notify_timer, enic_notify_timer, 0); 2864 2865 enic_rfs_flw_tbl_init(enic); 2866 enic_set_rx_coal_setting(enic); 2867 INIT_WORK(&enic->reset, enic_reset); 2868 INIT_WORK(&enic->tx_hang_reset, enic_tx_hang_reset); 2869 INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work); 2870 2871 for (i = 0; i < enic->wq_count; i++) 2872 spin_lock_init(&enic->wq_lock[i]); 2873 2874 /* Register net device 2875 */ 2876 2877 enic->port_mtu = enic->config.mtu; 2878 2879 err = enic_set_mac_addr(netdev, enic->mac_addr); 2880 if (err) { 2881 dev_err(dev, "Invalid MAC address, aborting\n"); 2882 goto err_out_dev_deinit; 2883 } 2884 2885 enic->tx_coalesce_usecs = enic->config.intr_timer_usec; 2886 /* rx coalesce time already got initialized. This gets used 2887 * if adaptive coal is turned off 2888 */ 2889 enic->rx_coalesce_usecs = enic->tx_coalesce_usecs; 2890 2891 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) 2892 netdev->netdev_ops = &enic_netdev_dynamic_ops; 2893 else 2894 netdev->netdev_ops = &enic_netdev_ops; 2895 2896 netdev->watchdog_timeo = 2 * HZ; 2897 enic_set_ethtool_ops(netdev); 2898 2899 netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX; 2900 if (ENIC_SETTING(enic, LOOP)) { 2901 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX; 2902 enic->loop_enable = 1; 2903 enic->loop_tag = enic->config.loop_tag; 2904 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag); 2905 } 2906 if (ENIC_SETTING(enic, TXCSUM)) 2907 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM; 2908 if (ENIC_SETTING(enic, TSO)) 2909 netdev->hw_features |= NETIF_F_TSO | 2910 NETIF_F_TSO6 | NETIF_F_TSO_ECN; 2911 if (ENIC_SETTING(enic, RSS)) 2912 netdev->hw_features |= NETIF_F_RXHASH; 2913 if (ENIC_SETTING(enic, RXCSUM)) 2914 netdev->hw_features |= NETIF_F_RXCSUM; 2915 if (ENIC_SETTING(enic, VXLAN)) { 2916 u64 patch_level; 2917 u64 a1 = 0; 2918 2919 netdev->hw_enc_features |= NETIF_F_RXCSUM | 2920 NETIF_F_TSO | 2921 NETIF_F_TSO6 | 2922 NETIF_F_TSO_ECN | 2923 NETIF_F_GSO_UDP_TUNNEL | 2924 NETIF_F_HW_CSUM | 2925 NETIF_F_GSO_UDP_TUNNEL_CSUM; 2926 netdev->hw_features |= netdev->hw_enc_features; 2927 /* get bit mask from hw about supported offload bit level 2928 * BIT(0) = fw supports patch_level 0 2929 * fcoe bit = encap 2930 * fcoe_fc_crc_ok = outer csum ok 2931 * BIT(1) = always set by fw 2932 * BIT(2) = fw supports patch_level 2 2933 * BIT(0) in rss_hash = encap 2934 * BIT(1,2) in rss_hash = outer_ip_csum_ok/ 2935 * outer_tcp_csum_ok 2936 * used in enic_rq_indicate_buf 2937 */ 2938 err = vnic_dev_get_supported_feature_ver(enic->vdev, 2939 VIC_FEATURE_VXLAN, 2940 &patch_level, &a1); 2941 if (err) 2942 patch_level = 0; 2943 enic->vxlan.flags = (u8)a1; 2944 /* mask bits that are supported by driver 2945 */ 2946 patch_level &= BIT_ULL(0) | BIT_ULL(2); 2947 patch_level = fls(patch_level); 2948 patch_level = patch_level ? patch_level - 1 : 0; 2949 enic->vxlan.patch_level = patch_level; 2950 2951 if (vnic_dev_get_res_count(enic->vdev, RES_TYPE_WQ) == 1 || 2952 enic->vxlan.flags & ENIC_VXLAN_MULTI_WQ) { 2953 netdev->udp_tunnel_nic_info = &enic_udp_tunnels_v4; 2954 if (enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6) 2955 netdev->udp_tunnel_nic_info = &enic_udp_tunnels; 2956 } 2957 } 2958 2959 netdev->features |= netdev->hw_features; 2960 netdev->vlan_features |= netdev->features; 2961 2962 #ifdef CONFIG_RFS_ACCEL 2963 netdev->hw_features |= NETIF_F_NTUPLE; 2964 #endif 2965 2966 if (using_dac) 2967 netdev->features |= NETIF_F_HIGHDMA; 2968 2969 netdev->priv_flags |= IFF_UNICAST_FLT; 2970 2971 /* MTU range: 68 - 9000 */ 2972 netdev->min_mtu = ENIC_MIN_MTU; 2973 netdev->max_mtu = ENIC_MAX_MTU; 2974 netdev->mtu = enic->port_mtu; 2975 2976 err = register_netdev(netdev); 2977 if (err) { 2978 dev_err(dev, "Cannot register net device, aborting\n"); 2979 goto err_out_dev_deinit; 2980 } 2981 enic->rx_copybreak = RX_COPYBREAK_DEFAULT; 2982 2983 return 0; 2984 2985 err_out_dev_deinit: 2986 enic_dev_deinit(enic); 2987 err_out_dev_close: 2988 vnic_dev_close(enic->vdev); 2989 err_out_disable_sriov: 2990 kfree(enic->pp); 2991 err_out_disable_sriov_pp: 2992 #ifdef CONFIG_PCI_IOV 2993 if (enic_sriov_enabled(enic)) { 2994 pci_disable_sriov(pdev); 2995 enic->priv_flags &= ~ENIC_SRIOV_ENABLED; 2996 } 2997 #endif 2998 err_out_vnic_unregister: 2999 vnic_dev_unregister(enic->vdev); 3000 err_out_iounmap: 3001 enic_iounmap(enic); 3002 err_out_release_regions: 3003 pci_release_regions(pdev); 3004 err_out_disable_device: 3005 pci_disable_device(pdev); 3006 err_out_free_netdev: 3007 free_netdev(netdev); 3008 3009 return err; 3010 } 3011 3012 static void enic_remove(struct pci_dev *pdev) 3013 { 3014 struct net_device *netdev = pci_get_drvdata(pdev); 3015 3016 if (netdev) { 3017 struct enic *enic = netdev_priv(netdev); 3018 3019 cancel_work_sync(&enic->reset); 3020 cancel_work_sync(&enic->change_mtu_work); 3021 unregister_netdev(netdev); 3022 enic_dev_deinit(enic); 3023 vnic_dev_close(enic->vdev); 3024 #ifdef CONFIG_PCI_IOV 3025 if (enic_sriov_enabled(enic)) { 3026 pci_disable_sriov(pdev); 3027 enic->priv_flags &= ~ENIC_SRIOV_ENABLED; 3028 } 3029 #endif 3030 kfree(enic->pp); 3031 vnic_dev_unregister(enic->vdev); 3032 enic_iounmap(enic); 3033 pci_release_regions(pdev); 3034 pci_disable_device(pdev); 3035 free_netdev(netdev); 3036 } 3037 } 3038 3039 static struct pci_driver enic_driver = { 3040 .name = DRV_NAME, 3041 .id_table = enic_id_table, 3042 .probe = enic_probe, 3043 .remove = enic_remove, 3044 }; 3045 3046 module_pci_driver(enic_driver); 3047