Merge tag 'trace-printf-v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/trace...
[drm/drm-misc.git] / drivers / net / ethernet / cisco / enic / enic_main.c
blob9913952ccb42f2017037a81a8e2c42daa7b53ec3
1 /*
2 * Copyright 2008-2010 Cisco Systems, Inc. All rights reserved.
3 * Copyright 2007 Nuova Systems, Inc. All rights reserved.
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.
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.
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 #include <net/netdev_queues.h>
51 #include "cq_enet_desc.h"
52 #include "vnic_dev.h"
53 #include "vnic_intr.h"
54 #include "vnic_stats.h"
55 #include "vnic_vic.h"
56 #include "enic_res.h"
57 #include "enic.h"
58 #include "enic_dev.h"
59 #include "enic_pp.h"
60 #include "enic_clsf.h"
62 #define ENIC_NOTIFY_TIMER_PERIOD (2 * HZ)
63 #define WQ_ENET_MAX_DESC_LEN (1 << WQ_ENET_LEN_BITS)
64 #define MAX_TSO (1 << 16)
65 #define ENIC_DESC_MAX_SPLITS (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
67 #define PCI_DEVICE_ID_CISCO_VIC_ENET 0x0043 /* ethernet vnic */
68 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN 0x0044 /* enet dynamic vnic */
69 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF 0x0071 /* enet SRIOV VF */
71 #define RX_COPYBREAK_DEFAULT 256
73 /* Supported devices */
74 static const struct pci_device_id enic_id_table[] = {
75 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
76 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
77 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
78 { 0, } /* end of table */
81 MODULE_DESCRIPTION(DRV_DESCRIPTION);
82 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
83 MODULE_LICENSE("GPL");
84 MODULE_DEVICE_TABLE(pci, enic_id_table);
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}
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}
106 /* This table helps the driver to pick different ranges for rx coalescing
107 * timer depending on the link speed.
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 */
115 static void enic_init_affinity_hint(struct enic *enic)
117 int numa_node = dev_to_node(&enic->pdev->dev);
118 int i;
120 for (i = 0; i < enic->intr_count; i++) {
121 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i) ||
122 (cpumask_available(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);
132 static void enic_free_affinity_hint(struct enic *enic)
134 int i;
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);
143 static void enic_set_affinity_hint(struct enic *enic)
145 int i;
146 int err;
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 !cpumask_available(enic->msix[i].affinity_mask) ||
152 cpumask_empty(enic->msix[i].affinity_mask))
153 continue;
154 err = irq_update_affinity_hint(enic->msix_entry[i].vector,
155 enic->msix[i].affinity_mask);
156 if (err)
157 netdev_warn(enic->netdev, "irq_update_affinity_hint failed, err %d\n",
158 err);
161 for (i = 0; i < enic->wq_count; i++) {
162 int wq_intr = enic_msix_wq_intr(enic, i);
164 if (cpumask_available(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,
172 static void enic_unset_affinity_hint(struct enic *enic)
174 int i;
176 for (i = 0; i < enic->intr_count; i++)
177 irq_update_affinity_hint(enic->msix_entry[i].vector, NULL);
180 static int enic_udp_tunnel_set_port(struct net_device *netdev,
181 unsigned int table, unsigned int entry,
182 struct udp_tunnel_info *ti)
184 struct enic *enic = netdev_priv(netdev);
185 int err;
187 spin_lock_bh(&enic->devcmd_lock);
189 err = vnic_dev_overlay_offload_cfg(enic->vdev,
190 OVERLAY_CFG_VXLAN_PORT_UPDATE,
191 ntohs(ti->port));
192 if (err)
193 goto error;
195 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
196 enic->vxlan.patch_level);
197 if (err)
198 goto error;
200 enic->vxlan.vxlan_udp_port_number = ntohs(ti->port);
201 error:
202 spin_unlock_bh(&enic->devcmd_lock);
204 return err;
207 static int enic_udp_tunnel_unset_port(struct net_device *netdev,
208 unsigned int table, unsigned int entry,
209 struct udp_tunnel_info *ti)
211 struct enic *enic = netdev_priv(netdev);
212 int err;
214 spin_lock_bh(&enic->devcmd_lock);
216 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
217 OVERLAY_OFFLOAD_DISABLE);
218 if (err)
219 goto unlock;
221 enic->vxlan.vxlan_udp_port_number = 0;
223 unlock:
224 spin_unlock_bh(&enic->devcmd_lock);
226 return err;
229 static const struct udp_tunnel_nic_info enic_udp_tunnels = {
230 .set_port = enic_udp_tunnel_set_port,
231 .unset_port = enic_udp_tunnel_unset_port,
232 .tables = {
233 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
235 }, enic_udp_tunnels_v4 = {
236 .set_port = enic_udp_tunnel_set_port,
237 .unset_port = enic_udp_tunnel_unset_port,
238 .flags = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
239 .tables = {
240 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
244 static netdev_features_t enic_features_check(struct sk_buff *skb,
245 struct net_device *dev,
246 netdev_features_t features)
248 const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb);
249 struct enic *enic = netdev_priv(dev);
250 struct udphdr *udph;
251 u16 port = 0;
252 u8 proto;
254 if (!skb->encapsulation)
255 return features;
257 features = vxlan_features_check(skb, features);
259 switch (vlan_get_protocol(skb)) {
260 case htons(ETH_P_IPV6):
261 if (!(enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6))
262 goto out;
263 proto = ipv6_hdr(skb)->nexthdr;
264 break;
265 case htons(ETH_P_IP):
266 proto = ip_hdr(skb)->protocol;
267 break;
268 default:
269 goto out;
272 switch (eth->h_proto) {
273 case ntohs(ETH_P_IPV6):
274 if (!(enic->vxlan.flags & ENIC_VXLAN_INNER_IPV6))
275 goto out;
276 fallthrough;
277 case ntohs(ETH_P_IP):
278 break;
279 default:
280 goto out;
284 if (proto == IPPROTO_UDP) {
285 udph = udp_hdr(skb);
286 port = be16_to_cpu(udph->dest);
289 /* HW supports offload of only one UDP port. Remove CSUM and GSO MASK
290 * for other UDP port tunnels
292 if (port != enic->vxlan.vxlan_udp_port_number)
293 goto out;
295 return features;
297 out:
298 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
301 int enic_is_dynamic(struct enic *enic)
303 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
306 int enic_sriov_enabled(struct enic *enic)
308 return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
311 static int enic_is_sriov_vf(struct enic *enic)
313 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
316 int enic_is_valid_vf(struct enic *enic, int vf)
318 #ifdef CONFIG_PCI_IOV
319 return vf >= 0 && vf < enic->num_vfs;
320 #else
321 return 0;
322 #endif
325 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
327 struct enic *enic = vnic_dev_priv(wq->vdev);
329 if (buf->sop)
330 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
331 DMA_TO_DEVICE);
332 else
333 dma_unmap_page(&enic->pdev->dev, buf->dma_addr, buf->len,
334 DMA_TO_DEVICE);
336 if (buf->os_buf)
337 dev_kfree_skb_any(buf->os_buf);
340 static void enic_wq_free_buf(struct vnic_wq *wq,
341 struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
343 struct enic *enic = vnic_dev_priv(wq->vdev);
345 enic->wq[wq->index].stats.cq_work++;
346 enic->wq[wq->index].stats.cq_bytes += buf->len;
347 enic_free_wq_buf(wq, buf);
350 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
351 u8 type, u16 q_number, u16 completed_index, void *opaque)
353 struct enic *enic = vnic_dev_priv(vdev);
355 spin_lock(&enic->wq[q_number].lock);
357 vnic_wq_service(&enic->wq[q_number].vwq, cq_desc,
358 completed_index, enic_wq_free_buf,
359 opaque);
361 if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
362 vnic_wq_desc_avail(&enic->wq[q_number].vwq) >=
363 (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)) {
364 netif_wake_subqueue(enic->netdev, q_number);
365 enic->wq[q_number].stats.wake++;
368 spin_unlock(&enic->wq[q_number].lock);
370 return 0;
373 static bool enic_log_q_error(struct enic *enic)
375 unsigned int i;
376 u32 error_status;
377 bool err = false;
379 for (i = 0; i < enic->wq_count; i++) {
380 error_status = vnic_wq_error_status(&enic->wq[i].vwq);
381 err |= error_status;
382 if (error_status)
383 netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
384 i, error_status);
387 for (i = 0; i < enic->rq_count; i++) {
388 error_status = vnic_rq_error_status(&enic->rq[i].vrq);
389 err |= error_status;
390 if (error_status)
391 netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
392 i, error_status);
395 return err;
398 static void enic_msglvl_check(struct enic *enic)
400 u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
402 if (msg_enable != enic->msg_enable) {
403 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
404 enic->msg_enable, msg_enable);
405 enic->msg_enable = msg_enable;
409 static void enic_mtu_check(struct enic *enic)
411 u32 mtu = vnic_dev_mtu(enic->vdev);
412 struct net_device *netdev = enic->netdev;
414 if (mtu && mtu != enic->port_mtu) {
415 enic->port_mtu = mtu;
416 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
417 mtu = max_t(int, ENIC_MIN_MTU,
418 min_t(int, ENIC_MAX_MTU, mtu));
419 if (mtu != netdev->mtu)
420 schedule_work(&enic->change_mtu_work);
421 } else {
422 if (mtu < netdev->mtu)
423 netdev_warn(netdev,
424 "interface MTU (%d) set higher "
425 "than switch port MTU (%d)\n",
426 netdev->mtu, mtu);
431 static void enic_link_check(struct enic *enic)
433 int link_status = vnic_dev_link_status(enic->vdev);
434 int carrier_ok = netif_carrier_ok(enic->netdev);
436 if (link_status && !carrier_ok) {
437 netdev_info(enic->netdev, "Link UP\n");
438 netif_carrier_on(enic->netdev);
439 } else if (!link_status && carrier_ok) {
440 netdev_info(enic->netdev, "Link DOWN\n");
441 netif_carrier_off(enic->netdev);
445 static void enic_notify_check(struct enic *enic)
447 enic_msglvl_check(enic);
448 enic_mtu_check(enic);
449 enic_link_check(enic);
452 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
454 static irqreturn_t enic_isr_legacy(int irq, void *data)
456 struct net_device *netdev = data;
457 struct enic *enic = netdev_priv(netdev);
458 unsigned int io_intr = ENIC_LEGACY_IO_INTR;
459 unsigned int err_intr = ENIC_LEGACY_ERR_INTR;
460 unsigned int notify_intr = ENIC_LEGACY_NOTIFY_INTR;
461 u32 pba;
463 vnic_intr_mask(&enic->intr[io_intr]);
465 pba = vnic_intr_legacy_pba(enic->legacy_pba);
466 if (!pba) {
467 vnic_intr_unmask(&enic->intr[io_intr]);
468 return IRQ_NONE; /* not our interrupt */
471 if (ENIC_TEST_INTR(pba, notify_intr)) {
472 enic_notify_check(enic);
473 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
476 if (ENIC_TEST_INTR(pba, err_intr)) {
477 vnic_intr_return_all_credits(&enic->intr[err_intr]);
478 enic_log_q_error(enic);
479 /* schedule recovery from WQ/RQ error */
480 schedule_work(&enic->reset);
481 return IRQ_HANDLED;
484 if (ENIC_TEST_INTR(pba, io_intr))
485 napi_schedule_irqoff(&enic->napi[0]);
486 else
487 vnic_intr_unmask(&enic->intr[io_intr]);
489 return IRQ_HANDLED;
492 static irqreturn_t enic_isr_msi(int irq, void *data)
494 struct enic *enic = data;
496 /* With MSI, there is no sharing of interrupts, so this is
497 * our interrupt and there is no need to ack it. The device
498 * is not providing per-vector masking, so the OS will not
499 * write to PCI config space to mask/unmask the interrupt.
500 * We're using mask_on_assertion for MSI, so the device
501 * automatically masks the interrupt when the interrupt is
502 * generated. Later, when exiting polling, the interrupt
503 * will be unmasked (see enic_poll).
505 * Also, the device uses the same PCIe Traffic Class (TC)
506 * for Memory Write data and MSI, so there are no ordering
507 * issues; the MSI will always arrive at the Root Complex
508 * _after_ corresponding Memory Writes (i.e. descriptor
509 * writes).
512 napi_schedule_irqoff(&enic->napi[0]);
514 return IRQ_HANDLED;
517 static irqreturn_t enic_isr_msix(int irq, void *data)
519 struct napi_struct *napi = data;
521 napi_schedule_irqoff(napi);
523 return IRQ_HANDLED;
526 static irqreturn_t enic_isr_msix_err(int irq, void *data)
528 struct enic *enic = data;
529 unsigned int intr = enic_msix_err_intr(enic);
531 vnic_intr_return_all_credits(&enic->intr[intr]);
533 if (enic_log_q_error(enic))
534 /* schedule recovery from WQ/RQ error */
535 schedule_work(&enic->reset);
537 return IRQ_HANDLED;
540 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
542 struct enic *enic = data;
543 unsigned int intr = enic_msix_notify_intr(enic);
545 enic_notify_check(enic);
546 vnic_intr_return_all_credits(&enic->intr[intr]);
548 return IRQ_HANDLED;
551 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq,
552 struct sk_buff *skb, unsigned int len_left,
553 int loopback)
555 const skb_frag_t *frag;
556 dma_addr_t dma_addr;
558 /* Queue additional data fragments */
559 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
560 len_left -= skb_frag_size(frag);
561 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 0,
562 skb_frag_size(frag),
563 DMA_TO_DEVICE);
564 if (unlikely(enic_dma_map_check(enic, dma_addr)))
565 return -ENOMEM;
566 enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag),
567 (len_left == 0), /* EOP? */
568 loopback);
571 return 0;
574 static int enic_queue_wq_skb_vlan(struct enic *enic, struct vnic_wq *wq,
575 struct sk_buff *skb, int vlan_tag_insert,
576 unsigned int vlan_tag, int loopback)
578 unsigned int head_len = skb_headlen(skb);
579 unsigned int len_left = skb->len - head_len;
580 int eop = (len_left == 0);
581 dma_addr_t dma_addr;
582 int err = 0;
584 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len,
585 DMA_TO_DEVICE);
586 if (unlikely(enic_dma_map_check(enic, dma_addr)))
587 return -ENOMEM;
589 /* Queue the main skb fragment. The fragments are no larger
590 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
591 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
592 * per fragment is queued.
594 enic_queue_wq_desc(wq, skb, dma_addr, head_len, vlan_tag_insert,
595 vlan_tag, eop, loopback);
597 if (!eop)
598 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
600 /* The enic_queue_wq_desc() above does not do HW checksum */
601 enic->wq[wq->index].stats.csum_none++;
602 enic->wq[wq->index].stats.packets++;
603 enic->wq[wq->index].stats.bytes += skb->len;
605 return err;
608 static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq,
609 struct sk_buff *skb, int vlan_tag_insert,
610 unsigned int vlan_tag, int loopback)
612 unsigned int head_len = skb_headlen(skb);
613 unsigned int len_left = skb->len - head_len;
614 unsigned int hdr_len = skb_checksum_start_offset(skb);
615 unsigned int csum_offset = hdr_len + skb->csum_offset;
616 int eop = (len_left == 0);
617 dma_addr_t dma_addr;
618 int err = 0;
620 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len,
621 DMA_TO_DEVICE);
622 if (unlikely(enic_dma_map_check(enic, dma_addr)))
623 return -ENOMEM;
625 /* Queue the main skb fragment. The fragments are no larger
626 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
627 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
628 * per fragment is queued.
630 enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len, csum_offset,
631 hdr_len, vlan_tag_insert, vlan_tag, eop,
632 loopback);
634 if (!eop)
635 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
637 enic->wq[wq->index].stats.csum_partial++;
638 enic->wq[wq->index].stats.packets++;
639 enic->wq[wq->index].stats.bytes += skb->len;
641 return err;
644 static void enic_preload_tcp_csum_encap(struct sk_buff *skb)
646 const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb);
648 switch (eth->h_proto) {
649 case ntohs(ETH_P_IP):
650 inner_ip_hdr(skb)->check = 0;
651 inner_tcp_hdr(skb)->check =
652 ~csum_tcpudp_magic(inner_ip_hdr(skb)->saddr,
653 inner_ip_hdr(skb)->daddr, 0,
654 IPPROTO_TCP, 0);
655 break;
656 case ntohs(ETH_P_IPV6):
657 inner_tcp_hdr(skb)->check =
658 ~csum_ipv6_magic(&inner_ipv6_hdr(skb)->saddr,
659 &inner_ipv6_hdr(skb)->daddr, 0,
660 IPPROTO_TCP, 0);
661 break;
662 default:
663 WARN_ONCE(1, "Non ipv4/ipv6 inner pkt for encap offload");
664 break;
668 static void enic_preload_tcp_csum(struct sk_buff *skb)
670 /* Preload TCP csum field with IP pseudo hdr calculated
671 * with IP length set to zero. HW will later add in length
672 * to each TCP segment resulting from the TSO.
675 if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
676 ip_hdr(skb)->check = 0;
677 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
678 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
679 } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
680 tcp_v6_gso_csum_prep(skb);
684 static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
685 struct sk_buff *skb, unsigned int mss,
686 int vlan_tag_insert, unsigned int vlan_tag,
687 int loopback)
689 unsigned int frag_len_left = skb_headlen(skb);
690 unsigned int len_left = skb->len - frag_len_left;
691 int eop = (len_left == 0);
692 unsigned int offset = 0;
693 unsigned int hdr_len;
694 dma_addr_t dma_addr;
695 unsigned int pkts;
696 unsigned int len;
697 skb_frag_t *frag;
699 if (skb->encapsulation) {
700 hdr_len = skb_inner_tcp_all_headers(skb);
701 enic_preload_tcp_csum_encap(skb);
702 enic->wq[wq->index].stats.encap_tso++;
703 } else {
704 hdr_len = skb_tcp_all_headers(skb);
705 enic_preload_tcp_csum(skb);
706 enic->wq[wq->index].stats.tso++;
709 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
710 * for the main skb fragment
712 while (frag_len_left) {
713 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
714 dma_addr = dma_map_single(&enic->pdev->dev,
715 skb->data + offset, len,
716 DMA_TO_DEVICE);
717 if (unlikely(enic_dma_map_check(enic, dma_addr)))
718 return -ENOMEM;
719 enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len,
720 vlan_tag_insert, vlan_tag,
721 eop && (len == frag_len_left), loopback);
722 frag_len_left -= len;
723 offset += len;
726 if (eop)
727 goto tso_out_stats;
729 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
730 * for additional data fragments
732 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
733 len_left -= skb_frag_size(frag);
734 frag_len_left = skb_frag_size(frag);
735 offset = 0;
737 while (frag_len_left) {
738 len = min(frag_len_left,
739 (unsigned int)WQ_ENET_MAX_DESC_LEN);
740 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
741 offset, len,
742 DMA_TO_DEVICE);
743 if (unlikely(enic_dma_map_check(enic, dma_addr)))
744 return -ENOMEM;
745 enic_queue_wq_desc_cont(wq, skb, dma_addr, len,
746 (len_left == 0) &&
747 (len == frag_len_left),/*EOP*/
748 loopback);
749 frag_len_left -= len;
750 offset += len;
754 tso_out_stats:
755 /* calculate how many packets tso sent */
756 len = skb->len - hdr_len;
757 pkts = len / mss;
758 if ((len % mss) > 0)
759 pkts++;
760 enic->wq[wq->index].stats.packets += pkts;
761 enic->wq[wq->index].stats.bytes += (len + (pkts * hdr_len));
763 return 0;
766 static inline int enic_queue_wq_skb_encap(struct enic *enic, struct vnic_wq *wq,
767 struct sk_buff *skb,
768 int vlan_tag_insert,
769 unsigned int vlan_tag, int loopback)
771 unsigned int head_len = skb_headlen(skb);
772 unsigned int len_left = skb->len - head_len;
773 /* Hardware will overwrite the checksum fields, calculating from
774 * scratch and ignoring the value placed by software.
775 * Offload mode = 00
776 * mss[2], mss[1], mss[0] bits are set
778 unsigned int mss_or_csum = 7;
779 int eop = (len_left == 0);
780 dma_addr_t dma_addr;
781 int err = 0;
783 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len,
784 DMA_TO_DEVICE);
785 if (unlikely(enic_dma_map_check(enic, dma_addr)))
786 return -ENOMEM;
788 enic_queue_wq_desc_ex(wq, skb, dma_addr, head_len, mss_or_csum, 0,
789 vlan_tag_insert, vlan_tag,
790 WQ_ENET_OFFLOAD_MODE_CSUM, eop, 1 /* SOP */, eop,
791 loopback);
792 if (!eop)
793 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
795 enic->wq[wq->index].stats.encap_csum++;
796 enic->wq[wq->index].stats.packets++;
797 enic->wq[wq->index].stats.bytes += skb->len;
799 return err;
802 static inline int enic_queue_wq_skb(struct enic *enic,
803 struct vnic_wq *wq, struct sk_buff *skb)
805 unsigned int mss = skb_shinfo(skb)->gso_size;
806 unsigned int vlan_tag = 0;
807 int vlan_tag_insert = 0;
808 int loopback = 0;
809 int err;
811 if (skb_vlan_tag_present(skb)) {
812 /* VLAN tag from trunking driver */
813 vlan_tag_insert = 1;
814 vlan_tag = skb_vlan_tag_get(skb);
815 enic->wq[wq->index].stats.add_vlan++;
816 } else if (enic->loop_enable) {
817 vlan_tag = enic->loop_tag;
818 loopback = 1;
821 if (mss)
822 err = enic_queue_wq_skb_tso(enic, wq, skb, mss,
823 vlan_tag_insert, vlan_tag,
824 loopback);
825 else if (skb->encapsulation)
826 err = enic_queue_wq_skb_encap(enic, wq, skb, vlan_tag_insert,
827 vlan_tag, loopback);
828 else if (skb->ip_summed == CHECKSUM_PARTIAL)
829 err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert,
830 vlan_tag, loopback);
831 else
832 err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert,
833 vlan_tag, loopback);
834 if (unlikely(err)) {
835 struct vnic_wq_buf *buf;
837 buf = wq->to_use->prev;
838 /* while not EOP of previous pkt && queue not empty.
839 * For all non EOP bufs, os_buf is NULL.
841 while (!buf->os_buf && (buf->next != wq->to_clean)) {
842 enic_free_wq_buf(wq, buf);
843 wq->ring.desc_avail++;
844 buf = buf->prev;
846 wq->to_use = buf->next;
847 dev_kfree_skb(skb);
849 return err;
852 /* netif_tx_lock held, process context with BHs disabled, or BH */
853 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
854 struct net_device *netdev)
856 struct enic *enic = netdev_priv(netdev);
857 struct vnic_wq *wq;
858 unsigned int txq_map;
859 struct netdev_queue *txq;
861 txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
862 wq = &enic->wq[txq_map].vwq;
864 if (skb->len <= 0) {
865 dev_kfree_skb_any(skb);
866 enic->wq[wq->index].stats.null_pkt++;
867 return NETDEV_TX_OK;
870 txq = netdev_get_tx_queue(netdev, txq_map);
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.
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 enic->wq[wq->index].stats.skb_linear_fail++;
882 return NETDEV_TX_OK;
885 spin_lock(&enic->wq[txq_map].lock);
887 if (vnic_wq_desc_avail(wq) <
888 skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
889 netif_tx_stop_queue(txq);
890 /* This is a hard error, log it */
891 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
892 spin_unlock(&enic->wq[txq_map].lock);
893 enic->wq[wq->index].stats.desc_full_awake++;
894 return NETDEV_TX_BUSY;
897 if (enic_queue_wq_skb(enic, wq, skb))
898 goto error;
900 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS) {
901 netif_tx_stop_queue(txq);
902 enic->wq[wq->index].stats.stopped++;
904 skb_tx_timestamp(skb);
905 if (!netdev_xmit_more() || netif_xmit_stopped(txq))
906 vnic_wq_doorbell(wq);
908 error:
909 spin_unlock(&enic->wq[txq_map].lock);
911 return NETDEV_TX_OK;
914 /* rcu_read_lock potentially held, nominally process context */
915 static void enic_get_stats(struct net_device *netdev,
916 struct rtnl_link_stats64 *net_stats)
918 struct enic *enic = netdev_priv(netdev);
919 struct vnic_stats *stats;
920 u64 pkt_truncated = 0;
921 u64 bad_fcs = 0;
922 int err;
923 int i;
925 err = enic_dev_stats_dump(enic, &stats);
926 /* return only when dma_alloc_coherent fails in vnic_dev_stats_dump
927 * For other failures, like devcmd failure, we return previously
928 * recorded stats.
930 if (err == -ENOMEM)
931 return;
933 net_stats->tx_packets = stats->tx.tx_frames_ok;
934 net_stats->tx_bytes = stats->tx.tx_bytes_ok;
935 net_stats->tx_errors = stats->tx.tx_errors;
936 net_stats->tx_dropped = stats->tx.tx_drops;
938 net_stats->rx_packets = stats->rx.rx_frames_ok;
939 net_stats->rx_bytes = stats->rx.rx_bytes_ok;
940 net_stats->rx_errors = stats->rx.rx_errors;
941 net_stats->multicast = stats->rx.rx_multicast_frames_ok;
943 for (i = 0; i < enic->rq_count; i++) {
944 struct enic_rq_stats *rqs = &enic->rq[i].stats;
946 if (!enic->rq[i].vrq.ctrl)
947 break;
948 pkt_truncated += rqs->pkt_truncated;
949 bad_fcs += rqs->bad_fcs;
951 net_stats->rx_over_errors = pkt_truncated;
952 net_stats->rx_crc_errors = bad_fcs;
953 net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
956 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
958 struct enic *enic = netdev_priv(netdev);
960 if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
961 unsigned int mc_count = netdev_mc_count(netdev);
963 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
964 ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
966 return -ENOSPC;
969 enic_dev_add_addr(enic, mc_addr);
970 enic->mc_count++;
972 return 0;
975 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
977 struct enic *enic = netdev_priv(netdev);
979 enic_dev_del_addr(enic, mc_addr);
980 enic->mc_count--;
982 return 0;
985 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
987 struct enic *enic = netdev_priv(netdev);
989 if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
990 unsigned int uc_count = netdev_uc_count(netdev);
992 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
993 ENIC_UNICAST_PERFECT_FILTERS, uc_count);
995 return -ENOSPC;
998 enic_dev_add_addr(enic, uc_addr);
999 enic->uc_count++;
1001 return 0;
1004 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
1006 struct enic *enic = netdev_priv(netdev);
1008 enic_dev_del_addr(enic, uc_addr);
1009 enic->uc_count--;
1011 return 0;
1014 void enic_reset_addr_lists(struct enic *enic)
1016 struct net_device *netdev = enic->netdev;
1018 __dev_uc_unsync(netdev, NULL);
1019 __dev_mc_unsync(netdev, NULL);
1021 enic->mc_count = 0;
1022 enic->uc_count = 0;
1023 enic->flags = 0;
1026 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
1028 struct enic *enic = netdev_priv(netdev);
1030 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
1031 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
1032 return -EADDRNOTAVAIL;
1033 } else {
1034 if (!is_valid_ether_addr(addr))
1035 return -EADDRNOTAVAIL;
1038 eth_hw_addr_set(netdev, addr);
1040 return 0;
1043 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
1045 struct enic *enic = netdev_priv(netdev);
1046 struct sockaddr *saddr = p;
1047 char *addr = saddr->sa_data;
1048 int err;
1050 if (netif_running(enic->netdev)) {
1051 err = enic_dev_del_station_addr(enic);
1052 if (err)
1053 return err;
1056 err = enic_set_mac_addr(netdev, addr);
1057 if (err)
1058 return err;
1060 if (netif_running(enic->netdev)) {
1061 err = enic_dev_add_station_addr(enic);
1062 if (err)
1063 return err;
1066 return err;
1069 static int enic_set_mac_address(struct net_device *netdev, void *p)
1071 struct sockaddr *saddr = p;
1072 char *addr = saddr->sa_data;
1073 struct enic *enic = netdev_priv(netdev);
1074 int err;
1076 err = enic_dev_del_station_addr(enic);
1077 if (err)
1078 return err;
1080 err = enic_set_mac_addr(netdev, addr);
1081 if (err)
1082 return err;
1084 return enic_dev_add_station_addr(enic);
1087 /* netif_tx_lock held, BHs disabled */
1088 static void enic_set_rx_mode(struct net_device *netdev)
1090 struct enic *enic = netdev_priv(netdev);
1091 int directed = 1;
1092 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
1093 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
1094 int promisc = (netdev->flags & IFF_PROMISC) ||
1095 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
1096 int allmulti = (netdev->flags & IFF_ALLMULTI) ||
1097 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
1098 unsigned int flags = netdev->flags |
1099 (allmulti ? IFF_ALLMULTI : 0) |
1100 (promisc ? IFF_PROMISC : 0);
1102 if (enic->flags != flags) {
1103 enic->flags = flags;
1104 enic_dev_packet_filter(enic, directed,
1105 multicast, broadcast, promisc, allmulti);
1108 if (!promisc) {
1109 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
1110 if (!allmulti)
1111 __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
1115 /* netif_tx_lock held, BHs disabled */
1116 static void enic_tx_timeout(struct net_device *netdev, unsigned int txqueue)
1118 struct enic *enic = netdev_priv(netdev);
1119 schedule_work(&enic->tx_hang_reset);
1122 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1124 struct enic *enic = netdev_priv(netdev);
1125 struct enic_port_profile *pp;
1126 int err;
1128 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1129 if (err)
1130 return err;
1132 if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
1133 if (vf == PORT_SELF_VF) {
1134 memcpy(pp->vf_mac, mac, ETH_ALEN);
1135 return 0;
1136 } else {
1138 * For sriov vf's set the mac in hw
1140 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1141 vnic_dev_set_mac_addr, mac);
1142 return enic_dev_status_to_errno(err);
1144 } else
1145 return -EINVAL;
1148 static int enic_set_vf_port(struct net_device *netdev, int vf,
1149 struct nlattr *port[])
1151 static const u8 zero_addr[ETH_ALEN] = {};
1152 struct enic *enic = netdev_priv(netdev);
1153 struct enic_port_profile prev_pp;
1154 struct enic_port_profile *pp;
1155 int err = 0, restore_pp = 1;
1157 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1158 if (err)
1159 return err;
1161 if (!port[IFLA_PORT_REQUEST])
1162 return -EOPNOTSUPP;
1164 memcpy(&prev_pp, pp, sizeof(*enic->pp));
1165 memset(pp, 0, sizeof(*enic->pp));
1167 pp->set |= ENIC_SET_REQUEST;
1168 pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1170 if (port[IFLA_PORT_PROFILE]) {
1171 if (nla_len(port[IFLA_PORT_PROFILE]) != PORT_PROFILE_MAX) {
1172 memcpy(pp, &prev_pp, sizeof(*pp));
1173 return -EINVAL;
1175 pp->set |= ENIC_SET_NAME;
1176 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
1177 PORT_PROFILE_MAX);
1180 if (port[IFLA_PORT_INSTANCE_UUID]) {
1181 if (nla_len(port[IFLA_PORT_INSTANCE_UUID]) != PORT_UUID_MAX) {
1182 memcpy(pp, &prev_pp, sizeof(*pp));
1183 return -EINVAL;
1185 pp->set |= ENIC_SET_INSTANCE;
1186 memcpy(pp->instance_uuid,
1187 nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
1190 if (port[IFLA_PORT_HOST_UUID]) {
1191 if (nla_len(port[IFLA_PORT_HOST_UUID]) != PORT_UUID_MAX) {
1192 memcpy(pp, &prev_pp, sizeof(*pp));
1193 return -EINVAL;
1195 pp->set |= ENIC_SET_HOST;
1196 memcpy(pp->host_uuid,
1197 nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
1200 if (vf == PORT_SELF_VF) {
1201 /* Special case handling: mac came from IFLA_VF_MAC */
1202 if (!is_zero_ether_addr(prev_pp.vf_mac))
1203 memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
1205 if (is_zero_ether_addr(netdev->dev_addr))
1206 eth_hw_addr_random(netdev);
1207 } else {
1208 /* SR-IOV VF: get mac from adapter */
1209 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1210 vnic_dev_get_mac_addr, pp->mac_addr);
1211 if (err) {
1212 netdev_err(netdev, "Error getting mac for vf %d\n", vf);
1213 memcpy(pp, &prev_pp, sizeof(*pp));
1214 return enic_dev_status_to_errno(err);
1218 err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
1219 if (err) {
1220 if (restore_pp) {
1221 /* Things are still the way they were: Implicit
1222 * DISASSOCIATE failed
1224 memcpy(pp, &prev_pp, sizeof(*pp));
1225 } else {
1226 memset(pp, 0, sizeof(*pp));
1227 if (vf == PORT_SELF_VF)
1228 eth_hw_addr_set(netdev, zero_addr);
1230 } else {
1231 /* Set flag to indicate that the port assoc/disassoc
1232 * request has been sent out to fw
1234 pp->set |= ENIC_PORT_REQUEST_APPLIED;
1236 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
1237 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
1238 eth_zero_addr(pp->mac_addr);
1239 if (vf == PORT_SELF_VF)
1240 eth_hw_addr_set(netdev, zero_addr);
1244 if (vf == PORT_SELF_VF)
1245 eth_zero_addr(pp->vf_mac);
1247 return err;
1250 static int enic_get_vf_port(struct net_device *netdev, int vf,
1251 struct sk_buff *skb)
1253 struct enic *enic = netdev_priv(netdev);
1254 u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1255 struct enic_port_profile *pp;
1256 int err;
1258 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1259 if (err)
1260 return err;
1262 if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
1263 return -ENODATA;
1265 err = enic_process_get_pp_request(enic, vf, pp->request, &response);
1266 if (err)
1267 return err;
1269 if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
1270 nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
1271 ((pp->set & ENIC_SET_NAME) &&
1272 nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
1273 ((pp->set & ENIC_SET_INSTANCE) &&
1274 nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1275 pp->instance_uuid)) ||
1276 ((pp->set & ENIC_SET_HOST) &&
1277 nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
1278 goto nla_put_failure;
1279 return 0;
1281 nla_put_failure:
1282 return -EMSGSIZE;
1285 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1287 struct enic *enic = vnic_dev_priv(rq->vdev);
1289 if (!buf->os_buf)
1290 return;
1292 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
1293 DMA_FROM_DEVICE);
1294 dev_kfree_skb_any(buf->os_buf);
1295 buf->os_buf = NULL;
1298 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1300 struct enic *enic = vnic_dev_priv(rq->vdev);
1301 struct net_device *netdev = enic->netdev;
1302 struct sk_buff *skb;
1303 unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
1304 unsigned int os_buf_index = 0;
1305 dma_addr_t dma_addr;
1306 struct vnic_rq_buf *buf = rq->to_use;
1308 if (buf->os_buf) {
1309 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
1310 buf->len);
1312 return 0;
1314 skb = netdev_alloc_skb_ip_align(netdev, len);
1315 if (!skb) {
1316 enic->rq[rq->index].stats.no_skb++;
1317 return -ENOMEM;
1320 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, len,
1321 DMA_FROM_DEVICE);
1322 if (unlikely(enic_dma_map_check(enic, dma_addr))) {
1323 dev_kfree_skb(skb);
1324 return -ENOMEM;
1327 enic_queue_rq_desc(rq, skb, os_buf_index,
1328 dma_addr, len);
1330 return 0;
1333 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1334 u32 pkt_len)
1336 if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1337 pkt_size->large_pkt_bytes_cnt += pkt_len;
1338 else
1339 pkt_size->small_pkt_bytes_cnt += pkt_len;
1342 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
1343 struct vnic_rq_buf *buf, u16 len)
1345 struct enic *enic = netdev_priv(netdev);
1346 struct sk_buff *new_skb;
1348 if (len > enic->rx_copybreak)
1349 return false;
1350 new_skb = netdev_alloc_skb_ip_align(netdev, len);
1351 if (!new_skb)
1352 return false;
1353 dma_sync_single_for_cpu(&enic->pdev->dev, buf->dma_addr, len,
1354 DMA_FROM_DEVICE);
1355 memcpy(new_skb->data, (*skb)->data, len);
1356 *skb = new_skb;
1358 return true;
1361 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1362 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1363 int skipped, void *opaque)
1365 struct enic *enic = vnic_dev_priv(rq->vdev);
1366 struct net_device *netdev = enic->netdev;
1367 struct sk_buff *skb;
1368 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1369 struct enic_rq_stats *rqstats = &enic->rq[rq->index].stats;
1371 u8 type, color, eop, sop, ingress_port, vlan_stripped;
1372 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1373 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1374 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1375 u8 packet_error;
1376 u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1377 u32 rss_hash;
1378 bool outer_csum_ok = true, encap = false;
1380 rqstats->packets++;
1381 if (skipped) {
1382 rqstats->desc_skip++;
1383 return;
1386 skb = buf->os_buf;
1388 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1389 &type, &color, &q_number, &completed_index,
1390 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1391 &csum_not_calc, &rss_hash, &bytes_written,
1392 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1393 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1394 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1395 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1396 &fcs_ok);
1398 if (packet_error) {
1400 if (!fcs_ok) {
1401 if (bytes_written > 0)
1402 rqstats->bad_fcs++;
1403 else if (bytes_written == 0)
1404 rqstats->pkt_truncated++;
1407 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
1408 DMA_FROM_DEVICE);
1409 dev_kfree_skb_any(skb);
1410 buf->os_buf = NULL;
1412 return;
1415 if (eop && bytes_written > 0) {
1417 /* Good receive
1419 rqstats->bytes += bytes_written;
1420 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1421 buf->os_buf = NULL;
1422 dma_unmap_single(&enic->pdev->dev, buf->dma_addr,
1423 buf->len, DMA_FROM_DEVICE);
1425 prefetch(skb->data - NET_IP_ALIGN);
1427 skb_put(skb, bytes_written);
1428 skb->protocol = eth_type_trans(skb, netdev);
1429 skb_record_rx_queue(skb, q_number);
1430 if ((netdev->features & NETIF_F_RXHASH) && rss_hash &&
1431 (type == 3)) {
1432 switch (rss_type) {
1433 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
1434 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
1435 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX:
1436 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4);
1437 rqstats->l4_rss_hash++;
1438 break;
1439 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4:
1440 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6:
1441 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX:
1442 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3);
1443 rqstats->l3_rss_hash++;
1444 break;
1447 if (enic->vxlan.vxlan_udp_port_number) {
1448 switch (enic->vxlan.patch_level) {
1449 case 0:
1450 if (fcoe) {
1451 encap = true;
1452 outer_csum_ok = fcoe_fc_crc_ok;
1454 break;
1455 case 2:
1456 if ((type == 7) &&
1457 (rss_hash & BIT(0))) {
1458 encap = true;
1459 outer_csum_ok = (rss_hash & BIT(1)) &&
1460 (rss_hash & BIT(2));
1462 break;
1466 /* Hardware does not provide whole packet checksum. It only
1467 * provides pseudo checksum. Since hw validates the packet
1468 * checksum but not provide us the checksum value. use
1469 * CHECSUM_UNNECESSARY.
1471 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is
1472 * inner csum_ok. outer_csum_ok is set by hw when outer udp
1473 * csum is correct or is zero.
1475 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc &&
1476 tcp_udp_csum_ok && outer_csum_ok &&
1477 (ipv4_csum_ok || ipv6)) {
1478 skb->ip_summed = CHECKSUM_UNNECESSARY;
1479 skb->csum_level = encap;
1480 if (encap)
1481 rqstats->csum_unnecessary_encap++;
1482 else
1483 rqstats->csum_unnecessary++;
1486 if (vlan_stripped) {
1487 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1488 rqstats->vlan_stripped++;
1490 skb_mark_napi_id(skb, &enic->napi[rq->index]);
1491 if (!(netdev->features & NETIF_F_GRO))
1492 netif_receive_skb(skb);
1493 else
1494 napi_gro_receive(&enic->napi[q_number], skb);
1495 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1496 enic_intr_update_pkt_size(&cq->pkt_size_counter,
1497 bytes_written);
1498 } else {
1500 /* Buffer overflow
1502 rqstats->pkt_truncated++;
1503 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
1504 DMA_FROM_DEVICE);
1505 dev_kfree_skb_any(skb);
1506 buf->os_buf = NULL;
1510 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1511 u8 type, u16 q_number, u16 completed_index, void *opaque)
1513 struct enic *enic = vnic_dev_priv(vdev);
1515 vnic_rq_service(&enic->rq[q_number].vrq, cq_desc,
1516 completed_index, VNIC_RQ_RETURN_DESC,
1517 enic_rq_indicate_buf, opaque);
1519 return 0;
1522 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1524 unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1525 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1526 u32 timer = cq->tobe_rx_coal_timeval;
1528 if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1529 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1530 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1534 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1536 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1537 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1538 struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1539 int index;
1540 u32 timer;
1541 u32 range_start;
1542 u32 traffic;
1543 u64 delta;
1544 ktime_t now = ktime_get();
1546 delta = ktime_us_delta(now, cq->prev_ts);
1547 if (delta < ENIC_AIC_TS_BREAK)
1548 return;
1549 cq->prev_ts = now;
1551 traffic = pkt_size_counter->large_pkt_bytes_cnt +
1552 pkt_size_counter->small_pkt_bytes_cnt;
1553 /* The table takes Mbps
1554 * traffic *= 8 => bits
1555 * traffic *= (10^6 / delta) => bps
1556 * traffic /= 10^6 => Mbps
1558 * Combining, traffic *= (8 / delta)
1561 traffic <<= 3;
1562 traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1564 for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1565 if (traffic < mod_table[index].rx_rate)
1566 break;
1567 range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1568 pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1569 rx_coal->small_pkt_range_start :
1570 rx_coal->large_pkt_range_start;
1571 timer = range_start + ((rx_coal->range_end - range_start) *
1572 mod_table[index].range_percent / 100);
1573 /* Damping */
1574 cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1576 pkt_size_counter->large_pkt_bytes_cnt = 0;
1577 pkt_size_counter->small_pkt_bytes_cnt = 0;
1580 static int enic_poll(struct napi_struct *napi, int budget)
1582 struct net_device *netdev = napi->dev;
1583 struct enic *enic = netdev_priv(netdev);
1584 unsigned int cq_rq = enic_cq_rq(enic, 0);
1585 unsigned int cq_wq = enic_cq_wq(enic, 0);
1586 unsigned int intr = ENIC_LEGACY_IO_INTR;
1587 unsigned int rq_work_to_do = budget;
1588 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET;
1589 unsigned int work_done, rq_work_done = 0, wq_work_done;
1590 int err;
1592 wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1593 enic_wq_service, NULL);
1595 if (budget > 0)
1596 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1597 rq_work_to_do, enic_rq_service, NULL);
1599 /* Accumulate intr event credits for this polling
1600 * cycle. An intr event is the completion of a
1601 * a WQ or RQ packet.
1604 work_done = rq_work_done + wq_work_done;
1606 if (work_done > 0)
1607 vnic_intr_return_credits(&enic->intr[intr],
1608 work_done,
1609 0 /* don't unmask intr */,
1610 0 /* don't reset intr timer */);
1612 err = vnic_rq_fill(&enic->rq[0].vrq, enic_rq_alloc_buf);
1614 /* Buffer allocation failed. Stay in polling
1615 * mode so we can try to fill the ring again.
1618 if (err)
1619 rq_work_done = rq_work_to_do;
1620 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1621 /* Call the function which refreshes the intr coalescing timer
1622 * value based on the traffic.
1624 enic_calc_int_moderation(enic, &enic->rq[0].vrq);
1626 if ((rq_work_done < budget) && napi_complete_done(napi, rq_work_done)) {
1628 /* Some work done, but not enough to stay in polling,
1629 * exit polling
1632 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1633 enic_set_int_moderation(enic, &enic->rq[0].vrq);
1634 vnic_intr_unmask(&enic->intr[intr]);
1635 enic->rq[0].stats.napi_complete++;
1636 } else {
1637 enic->rq[0].stats.napi_repoll++;
1640 return rq_work_done;
1643 #ifdef CONFIG_RFS_ACCEL
1644 static void enic_free_rx_cpu_rmap(struct enic *enic)
1646 free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1647 enic->netdev->rx_cpu_rmap = NULL;
1650 static void enic_set_rx_cpu_rmap(struct enic *enic)
1652 int i, res;
1654 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1655 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1656 if (unlikely(!enic->netdev->rx_cpu_rmap))
1657 return;
1658 for (i = 0; i < enic->rq_count; i++) {
1659 res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1660 enic->msix_entry[i].vector);
1661 if (unlikely(res)) {
1662 enic_free_rx_cpu_rmap(enic);
1663 return;
1669 #else
1671 static void enic_free_rx_cpu_rmap(struct enic *enic)
1675 static void enic_set_rx_cpu_rmap(struct enic *enic)
1679 #endif /* CONFIG_RFS_ACCEL */
1681 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1683 struct net_device *netdev = napi->dev;
1684 struct enic *enic = netdev_priv(netdev);
1685 unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1686 struct vnic_wq *wq = &enic->wq[wq_index].vwq;
1687 unsigned int cq;
1688 unsigned int intr;
1689 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET;
1690 unsigned int wq_work_done;
1691 unsigned int wq_irq;
1693 wq_irq = wq->index;
1694 cq = enic_cq_wq(enic, wq_irq);
1695 intr = enic_msix_wq_intr(enic, wq_irq);
1696 wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1697 enic_wq_service, NULL);
1699 vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1700 0 /* don't unmask intr */,
1701 1 /* reset intr timer */);
1702 if (!wq_work_done) {
1703 napi_complete(napi);
1704 vnic_intr_unmask(&enic->intr[intr]);
1705 return 0;
1708 return budget;
1711 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1713 struct net_device *netdev = napi->dev;
1714 struct enic *enic = netdev_priv(netdev);
1715 unsigned int rq = (napi - &enic->napi[0]);
1716 unsigned int cq = enic_cq_rq(enic, rq);
1717 unsigned int intr = enic_msix_rq_intr(enic, rq);
1718 unsigned int work_to_do = budget;
1719 unsigned int work_done = 0;
1720 int err;
1722 /* Service RQ
1725 if (budget > 0)
1726 work_done = vnic_cq_service(&enic->cq[cq],
1727 work_to_do, enic_rq_service, NULL);
1729 /* Return intr event credits for this polling
1730 * cycle. An intr event is the completion of a
1731 * RQ packet.
1734 if (work_done > 0)
1735 vnic_intr_return_credits(&enic->intr[intr],
1736 work_done,
1737 0 /* don't unmask intr */,
1738 0 /* don't reset intr timer */);
1740 err = vnic_rq_fill(&enic->rq[rq].vrq, enic_rq_alloc_buf);
1742 /* Buffer allocation failed. Stay in polling mode
1743 * so we can try to fill the ring again.
1746 if (err)
1747 work_done = work_to_do;
1748 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1749 /* Call the function which refreshes the intr coalescing timer
1750 * value based on the traffic.
1752 enic_calc_int_moderation(enic, &enic->rq[rq].vrq);
1754 if ((work_done < budget) && napi_complete_done(napi, work_done)) {
1756 /* Some work done, but not enough to stay in polling,
1757 * exit polling
1760 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1761 enic_set_int_moderation(enic, &enic->rq[rq].vrq);
1762 vnic_intr_unmask(&enic->intr[intr]);
1763 enic->rq[rq].stats.napi_complete++;
1764 } else {
1765 enic->rq[rq].stats.napi_repoll++;
1768 return work_done;
1771 static void enic_notify_timer(struct timer_list *t)
1773 struct enic *enic = from_timer(enic, t, notify_timer);
1775 enic_notify_check(enic);
1777 mod_timer(&enic->notify_timer,
1778 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1781 static void enic_free_intr(struct enic *enic)
1783 struct net_device *netdev = enic->netdev;
1784 unsigned int i;
1786 enic_free_rx_cpu_rmap(enic);
1787 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1788 case VNIC_DEV_INTR_MODE_INTX:
1789 free_irq(enic->pdev->irq, netdev);
1790 break;
1791 case VNIC_DEV_INTR_MODE_MSI:
1792 free_irq(enic->pdev->irq, enic);
1793 break;
1794 case VNIC_DEV_INTR_MODE_MSIX:
1795 for (i = 0; i < enic->intr_count; i++)
1796 if (enic->msix[i].requested)
1797 free_irq(enic->msix_entry[i].vector,
1798 enic->msix[i].devid);
1799 break;
1800 default:
1801 break;
1805 static int enic_request_intr(struct enic *enic)
1807 struct net_device *netdev = enic->netdev;
1808 unsigned int i, intr;
1809 int err = 0;
1811 enic_set_rx_cpu_rmap(enic);
1812 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1814 case VNIC_DEV_INTR_MODE_INTX:
1816 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1817 IRQF_SHARED, netdev->name, netdev);
1818 break;
1820 case VNIC_DEV_INTR_MODE_MSI:
1822 err = request_irq(enic->pdev->irq, enic_isr_msi,
1823 0, netdev->name, enic);
1824 break;
1826 case VNIC_DEV_INTR_MODE_MSIX:
1828 for (i = 0; i < enic->rq_count; i++) {
1829 intr = enic_msix_rq_intr(enic, i);
1830 snprintf(enic->msix[intr].devname,
1831 sizeof(enic->msix[intr].devname),
1832 "%s-rx-%u", netdev->name, i);
1833 enic->msix[intr].isr = enic_isr_msix;
1834 enic->msix[intr].devid = &enic->napi[i];
1837 for (i = 0; i < enic->wq_count; i++) {
1838 int wq = enic_cq_wq(enic, i);
1840 intr = enic_msix_wq_intr(enic, i);
1841 snprintf(enic->msix[intr].devname,
1842 sizeof(enic->msix[intr].devname),
1843 "%s-tx-%u", netdev->name, i);
1844 enic->msix[intr].isr = enic_isr_msix;
1845 enic->msix[intr].devid = &enic->napi[wq];
1848 intr = enic_msix_err_intr(enic);
1849 snprintf(enic->msix[intr].devname,
1850 sizeof(enic->msix[intr].devname),
1851 "%s-err", netdev->name);
1852 enic->msix[intr].isr = enic_isr_msix_err;
1853 enic->msix[intr].devid = enic;
1855 intr = enic_msix_notify_intr(enic);
1856 snprintf(enic->msix[intr].devname,
1857 sizeof(enic->msix[intr].devname),
1858 "%s-notify", netdev->name);
1859 enic->msix[intr].isr = enic_isr_msix_notify;
1860 enic->msix[intr].devid = enic;
1862 for (i = 0; i < enic->intr_count; i++)
1863 enic->msix[i].requested = 0;
1865 for (i = 0; i < enic->intr_count; i++) {
1866 err = request_irq(enic->msix_entry[i].vector,
1867 enic->msix[i].isr, 0,
1868 enic->msix[i].devname,
1869 enic->msix[i].devid);
1870 if (err) {
1871 enic_free_intr(enic);
1872 break;
1874 enic->msix[i].requested = 1;
1877 break;
1879 default:
1880 break;
1883 return err;
1886 static void enic_synchronize_irqs(struct enic *enic)
1888 unsigned int i;
1890 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1891 case VNIC_DEV_INTR_MODE_INTX:
1892 case VNIC_DEV_INTR_MODE_MSI:
1893 synchronize_irq(enic->pdev->irq);
1894 break;
1895 case VNIC_DEV_INTR_MODE_MSIX:
1896 for (i = 0; i < enic->intr_count; i++)
1897 synchronize_irq(enic->msix_entry[i].vector);
1898 break;
1899 default:
1900 break;
1904 static void enic_set_rx_coal_setting(struct enic *enic)
1906 unsigned int speed;
1907 int index = -1;
1908 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1910 /* 1. Read the link speed from fw
1911 * 2. Pick the default range for the speed
1912 * 3. Update it in enic->rx_coalesce_setting
1914 speed = vnic_dev_port_speed(enic->vdev);
1915 if (ENIC_LINK_SPEED_10G < speed)
1916 index = ENIC_LINK_40G_INDEX;
1917 else if (ENIC_LINK_SPEED_4G < speed)
1918 index = ENIC_LINK_10G_INDEX;
1919 else
1920 index = ENIC_LINK_4G_INDEX;
1922 rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1923 rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1924 rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1926 /* Start with the value provided by UCSM */
1927 for (index = 0; index < enic->rq_count; index++)
1928 enic->cq[index].cur_rx_coal_timeval =
1929 enic->config.intr_timer_usec;
1931 rx_coal->use_adaptive_rx_coalesce = 1;
1934 static int enic_dev_notify_set(struct enic *enic)
1936 int err;
1938 spin_lock_bh(&enic->devcmd_lock);
1939 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1940 case VNIC_DEV_INTR_MODE_INTX:
1941 err = vnic_dev_notify_set(enic->vdev, ENIC_LEGACY_NOTIFY_INTR);
1942 break;
1943 case VNIC_DEV_INTR_MODE_MSIX:
1944 err = vnic_dev_notify_set(enic->vdev,
1945 enic_msix_notify_intr(enic));
1946 break;
1947 default:
1948 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1949 break;
1951 spin_unlock_bh(&enic->devcmd_lock);
1953 return err;
1956 static void enic_notify_timer_start(struct enic *enic)
1958 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1959 case VNIC_DEV_INTR_MODE_MSI:
1960 mod_timer(&enic->notify_timer, jiffies);
1961 break;
1962 default:
1963 /* Using intr for notification for INTx/MSI-X */
1964 break;
1968 /* rtnl lock is held, process context */
1969 static int enic_open(struct net_device *netdev)
1971 struct enic *enic = netdev_priv(netdev);
1972 unsigned int i;
1973 int err, ret;
1975 err = enic_request_intr(enic);
1976 if (err) {
1977 netdev_err(netdev, "Unable to request irq.\n");
1978 return err;
1980 enic_init_affinity_hint(enic);
1981 enic_set_affinity_hint(enic);
1983 err = enic_dev_notify_set(enic);
1984 if (err) {
1985 netdev_err(netdev,
1986 "Failed to alloc notify buffer, aborting.\n");
1987 goto err_out_free_intr;
1990 for (i = 0; i < enic->rq_count; i++) {
1991 /* enable rq before updating rq desc */
1992 vnic_rq_enable(&enic->rq[i].vrq);
1993 vnic_rq_fill(&enic->rq[i].vrq, enic_rq_alloc_buf);
1994 /* Need at least one buffer on ring to get going */
1995 if (vnic_rq_desc_used(&enic->rq[i].vrq) == 0) {
1996 netdev_err(netdev, "Unable to alloc receive buffers\n");
1997 err = -ENOMEM;
1998 goto err_out_free_rq;
2002 for (i = 0; i < enic->wq_count; i++)
2003 vnic_wq_enable(&enic->wq[i].vwq);
2005 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
2006 enic_dev_add_station_addr(enic);
2008 enic_set_rx_mode(netdev);
2010 netif_tx_wake_all_queues(netdev);
2012 for (i = 0; i < enic->rq_count; i++)
2013 napi_enable(&enic->napi[i]);
2015 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2016 for (i = 0; i < enic->wq_count; i++)
2017 napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
2018 enic_dev_enable(enic);
2020 for (i = 0; i < enic->intr_count; i++)
2021 vnic_intr_unmask(&enic->intr[i]);
2023 enic_notify_timer_start(enic);
2024 enic_rfs_timer_start(enic);
2026 return 0;
2028 err_out_free_rq:
2029 for (i = 0; i < enic->rq_count; i++) {
2030 ret = vnic_rq_disable(&enic->rq[i].vrq);
2031 if (!ret)
2032 vnic_rq_clean(&enic->rq[i].vrq, enic_free_rq_buf);
2034 enic_dev_notify_unset(enic);
2035 err_out_free_intr:
2036 enic_unset_affinity_hint(enic);
2037 enic_free_intr(enic);
2039 return err;
2042 /* rtnl lock is held, process context */
2043 static int enic_stop(struct net_device *netdev)
2045 struct enic *enic = netdev_priv(netdev);
2046 unsigned int i;
2047 int err;
2049 for (i = 0; i < enic->intr_count; i++) {
2050 vnic_intr_mask(&enic->intr[i]);
2051 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
2054 enic_synchronize_irqs(enic);
2056 del_timer_sync(&enic->notify_timer);
2057 enic_rfs_flw_tbl_free(enic);
2059 enic_dev_disable(enic);
2061 for (i = 0; i < enic->rq_count; i++)
2062 napi_disable(&enic->napi[i]);
2064 netif_carrier_off(netdev);
2065 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2066 for (i = 0; i < enic->wq_count; i++)
2067 napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
2068 netif_tx_disable(netdev);
2070 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
2071 enic_dev_del_station_addr(enic);
2073 for (i = 0; i < enic->wq_count; i++) {
2074 err = vnic_wq_disable(&enic->wq[i].vwq);
2075 if (err)
2076 return err;
2078 for (i = 0; i < enic->rq_count; i++) {
2079 err = vnic_rq_disable(&enic->rq[i].vrq);
2080 if (err)
2081 return err;
2084 enic_dev_notify_unset(enic);
2085 enic_unset_affinity_hint(enic);
2086 enic_free_intr(enic);
2088 for (i = 0; i < enic->wq_count; i++)
2089 vnic_wq_clean(&enic->wq[i].vwq, enic_free_wq_buf);
2090 for (i = 0; i < enic->rq_count; i++)
2091 vnic_rq_clean(&enic->rq[i].vrq, enic_free_rq_buf);
2092 for (i = 0; i < enic->cq_count; i++)
2093 vnic_cq_clean(&enic->cq[i]);
2094 for (i = 0; i < enic->intr_count; i++)
2095 vnic_intr_clean(&enic->intr[i]);
2097 return 0;
2100 static int _enic_change_mtu(struct net_device *netdev, int new_mtu)
2102 bool running = netif_running(netdev);
2103 int err = 0;
2105 ASSERT_RTNL();
2106 if (running) {
2107 err = enic_stop(netdev);
2108 if (err)
2109 return err;
2112 WRITE_ONCE(netdev->mtu, new_mtu);
2114 if (running) {
2115 err = enic_open(netdev);
2116 if (err)
2117 return err;
2120 return 0;
2123 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
2125 struct enic *enic = netdev_priv(netdev);
2127 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2128 return -EOPNOTSUPP;
2130 if (netdev->mtu > enic->port_mtu)
2131 netdev_warn(netdev,
2132 "interface MTU (%d) set higher than port MTU (%d)\n",
2133 netdev->mtu, enic->port_mtu);
2135 return _enic_change_mtu(netdev, new_mtu);
2138 static void enic_change_mtu_work(struct work_struct *work)
2140 struct enic *enic = container_of(work, struct enic, change_mtu_work);
2141 struct net_device *netdev = enic->netdev;
2142 int new_mtu = vnic_dev_mtu(enic->vdev);
2144 rtnl_lock();
2145 (void)_enic_change_mtu(netdev, new_mtu);
2146 rtnl_unlock();
2148 netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
2151 #ifdef CONFIG_NET_POLL_CONTROLLER
2152 static void enic_poll_controller(struct net_device *netdev)
2154 struct enic *enic = netdev_priv(netdev);
2155 struct vnic_dev *vdev = enic->vdev;
2156 unsigned int i, intr;
2158 switch (vnic_dev_get_intr_mode(vdev)) {
2159 case VNIC_DEV_INTR_MODE_MSIX:
2160 for (i = 0; i < enic->rq_count; i++) {
2161 intr = enic_msix_rq_intr(enic, i);
2162 enic_isr_msix(enic->msix_entry[intr].vector,
2163 &enic->napi[i]);
2166 for (i = 0; i < enic->wq_count; i++) {
2167 intr = enic_msix_wq_intr(enic, i);
2168 enic_isr_msix(enic->msix_entry[intr].vector,
2169 &enic->napi[enic_cq_wq(enic, i)]);
2172 break;
2173 case VNIC_DEV_INTR_MODE_MSI:
2174 enic_isr_msi(enic->pdev->irq, enic);
2175 break;
2176 case VNIC_DEV_INTR_MODE_INTX:
2177 enic_isr_legacy(enic->pdev->irq, netdev);
2178 break;
2179 default:
2180 break;
2183 #endif
2185 static int enic_dev_wait(struct vnic_dev *vdev,
2186 int (*start)(struct vnic_dev *, int),
2187 int (*finished)(struct vnic_dev *, int *),
2188 int arg)
2190 unsigned long time;
2191 int done;
2192 int err;
2194 err = start(vdev, arg);
2195 if (err)
2196 return err;
2198 /* Wait for func to complete...2 seconds max
2201 time = jiffies + (HZ * 2);
2202 do {
2204 err = finished(vdev, &done);
2205 if (err)
2206 return err;
2208 if (done)
2209 return 0;
2211 schedule_timeout_uninterruptible(HZ / 10);
2213 } while (time_after(time, jiffies));
2215 return -ETIMEDOUT;
2218 static int enic_dev_open(struct enic *enic)
2220 int err;
2221 u32 flags = CMD_OPENF_IG_DESCCACHE;
2223 err = enic_dev_wait(enic->vdev, vnic_dev_open,
2224 vnic_dev_open_done, flags);
2225 if (err)
2226 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
2227 err);
2229 return err;
2232 static int enic_dev_soft_reset(struct enic *enic)
2234 int err;
2236 err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
2237 vnic_dev_soft_reset_done, 0);
2238 if (err)
2239 netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n",
2240 err);
2242 return err;
2245 static int enic_dev_hang_reset(struct enic *enic)
2247 int err;
2249 err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
2250 vnic_dev_hang_reset_done, 0);
2251 if (err)
2252 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
2253 err);
2255 return err;
2258 int __enic_set_rsskey(struct enic *enic)
2260 union vnic_rss_key *rss_key_buf_va;
2261 dma_addr_t rss_key_buf_pa;
2262 int i, kidx, bidx, err;
2264 rss_key_buf_va = dma_alloc_coherent(&enic->pdev->dev,
2265 sizeof(union vnic_rss_key),
2266 &rss_key_buf_pa, GFP_ATOMIC);
2267 if (!rss_key_buf_va)
2268 return -ENOMEM;
2270 for (i = 0; i < ENIC_RSS_LEN; i++) {
2271 kidx = i / ENIC_RSS_BYTES_PER_KEY;
2272 bidx = i % ENIC_RSS_BYTES_PER_KEY;
2273 rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i];
2275 spin_lock_bh(&enic->devcmd_lock);
2276 err = enic_set_rss_key(enic,
2277 rss_key_buf_pa,
2278 sizeof(union vnic_rss_key));
2279 spin_unlock_bh(&enic->devcmd_lock);
2281 dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_key),
2282 rss_key_buf_va, rss_key_buf_pa);
2284 return err;
2287 static int enic_set_rsskey(struct enic *enic)
2289 netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN);
2291 return __enic_set_rsskey(enic);
2294 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
2296 dma_addr_t rss_cpu_buf_pa;
2297 union vnic_rss_cpu *rss_cpu_buf_va = NULL;
2298 unsigned int i;
2299 int err;
2301 rss_cpu_buf_va = dma_alloc_coherent(&enic->pdev->dev,
2302 sizeof(union vnic_rss_cpu),
2303 &rss_cpu_buf_pa, GFP_ATOMIC);
2304 if (!rss_cpu_buf_va)
2305 return -ENOMEM;
2307 for (i = 0; i < (1 << rss_hash_bits); i++)
2308 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
2310 spin_lock_bh(&enic->devcmd_lock);
2311 err = enic_set_rss_cpu(enic,
2312 rss_cpu_buf_pa,
2313 sizeof(union vnic_rss_cpu));
2314 spin_unlock_bh(&enic->devcmd_lock);
2316 dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_cpu),
2317 rss_cpu_buf_va, rss_cpu_buf_pa);
2319 return err;
2322 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
2323 u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
2325 const u8 tso_ipid_split_en = 0;
2326 const u8 ig_vlan_strip_en = 1;
2327 int err;
2329 /* Enable VLAN tag stripping.
2332 spin_lock_bh(&enic->devcmd_lock);
2333 err = enic_set_nic_cfg(enic,
2334 rss_default_cpu, rss_hash_type,
2335 rss_hash_bits, rss_base_cpu,
2336 rss_enable, tso_ipid_split_en,
2337 ig_vlan_strip_en);
2338 spin_unlock_bh(&enic->devcmd_lock);
2340 return err;
2343 static int enic_set_rss_nic_cfg(struct enic *enic)
2345 struct device *dev = enic_get_dev(enic);
2346 const u8 rss_default_cpu = 0;
2347 const u8 rss_hash_bits = 7;
2348 const u8 rss_base_cpu = 0;
2349 u8 rss_hash_type;
2350 int res;
2351 u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2353 spin_lock_bh(&enic->devcmd_lock);
2354 res = vnic_dev_capable_rss_hash_type(enic->vdev, &rss_hash_type);
2355 spin_unlock_bh(&enic->devcmd_lock);
2356 if (res) {
2357 /* defaults for old adapters
2359 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
2360 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
2361 NIC_CFG_RSS_HASH_TYPE_IPV6 |
2362 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2365 if (rss_enable) {
2366 if (!enic_set_rsskey(enic)) {
2367 if (enic_set_rsscpu(enic, rss_hash_bits)) {
2368 rss_enable = 0;
2369 dev_warn(dev, "RSS disabled, "
2370 "Failed to set RSS cpu indirection table.");
2372 } else {
2373 rss_enable = 0;
2374 dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2378 return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2379 rss_hash_bits, rss_base_cpu, rss_enable);
2382 static void enic_set_api_busy(struct enic *enic, bool busy)
2384 spin_lock(&enic->enic_api_lock);
2385 enic->enic_api_busy = busy;
2386 spin_unlock(&enic->enic_api_lock);
2389 static void enic_reset(struct work_struct *work)
2391 struct enic *enic = container_of(work, struct enic, reset);
2393 if (!netif_running(enic->netdev))
2394 return;
2396 rtnl_lock();
2398 /* Stop any activity from infiniband */
2399 enic_set_api_busy(enic, true);
2401 enic_stop(enic->netdev);
2402 enic_dev_soft_reset(enic);
2403 enic_reset_addr_lists(enic);
2404 enic_init_vnic_resources(enic);
2405 enic_set_rss_nic_cfg(enic);
2406 enic_dev_set_ig_vlan_rewrite_mode(enic);
2407 enic_open(enic->netdev);
2409 /* Allow infiniband to fiddle with the device again */
2410 enic_set_api_busy(enic, false);
2412 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2414 rtnl_unlock();
2417 static void enic_tx_hang_reset(struct work_struct *work)
2419 struct enic *enic = container_of(work, struct enic, tx_hang_reset);
2421 rtnl_lock();
2423 /* Stop any activity from infiniband */
2424 enic_set_api_busy(enic, true);
2426 enic_dev_hang_notify(enic);
2427 enic_stop(enic->netdev);
2428 enic_dev_hang_reset(enic);
2429 enic_reset_addr_lists(enic);
2430 enic_init_vnic_resources(enic);
2431 enic_set_rss_nic_cfg(enic);
2432 enic_dev_set_ig_vlan_rewrite_mode(enic);
2433 enic_open(enic->netdev);
2435 /* Allow infiniband to fiddle with the device again */
2436 enic_set_api_busy(enic, false);
2438 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2440 rtnl_unlock();
2443 static int enic_set_intr_mode(struct enic *enic)
2445 unsigned int i;
2446 int num_intr;
2448 /* Set interrupt mode (INTx, MSI, MSI-X) depending
2449 * on system capabilities.
2451 * Try MSI-X first
2454 if (enic->config.intr_mode < 1 &&
2455 enic->intr_avail >= ENIC_MSIX_MIN_INTR) {
2456 for (i = 0; i < enic->intr_avail; i++)
2457 enic->msix_entry[i].entry = i;
2459 num_intr = pci_enable_msix_range(enic->pdev, enic->msix_entry,
2460 ENIC_MSIX_MIN_INTR,
2461 enic->intr_avail);
2462 if (num_intr > 0) {
2463 vnic_dev_set_intr_mode(enic->vdev,
2464 VNIC_DEV_INTR_MODE_MSIX);
2465 enic->intr_avail = num_intr;
2466 return 0;
2470 /* Next try MSI
2472 * We need 1 INTR
2475 if (enic->config.intr_mode < 2 &&
2476 enic->intr_avail >= 1 &&
2477 !pci_enable_msi(enic->pdev)) {
2478 enic->intr_avail = 1;
2479 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2480 return 0;
2483 /* Next try INTx
2485 * We need 3 INTRs
2486 * (the first INTR is used for WQ/RQ)
2487 * (the second INTR is used for WQ/RQ errors)
2488 * (the last INTR is used for notifications)
2491 if (enic->config.intr_mode < 3 &&
2492 enic->intr_avail >= 3) {
2493 enic->intr_avail = 3;
2494 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2495 return 0;
2498 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2500 return -EINVAL;
2503 static void enic_clear_intr_mode(struct enic *enic)
2505 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2506 case VNIC_DEV_INTR_MODE_MSIX:
2507 pci_disable_msix(enic->pdev);
2508 break;
2509 case VNIC_DEV_INTR_MODE_MSI:
2510 pci_disable_msi(enic->pdev);
2511 break;
2512 default:
2513 break;
2516 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2519 static int enic_adjust_resources(struct enic *enic)
2521 unsigned int max_queues;
2522 unsigned int rq_default;
2523 unsigned int rq_avail;
2524 unsigned int wq_avail;
2526 if (enic->rq_avail < 1 || enic->wq_avail < 1 || enic->cq_avail < 2) {
2527 dev_err(enic_get_dev(enic),
2528 "Not enough resources available rq: %d wq: %d cq: %d\n",
2529 enic->rq_avail, enic->wq_avail,
2530 enic->cq_avail);
2531 return -ENOSPC;
2534 if (is_kdump_kernel()) {
2535 dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n");
2536 enic->rq_avail = 1;
2537 enic->wq_avail = 1;
2538 enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS;
2539 enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS;
2540 enic->config.mtu = min_t(u16, 1500, enic->config.mtu);
2543 /* if RSS isn't set, then we can only use one RQ */
2544 if (!ENIC_SETTING(enic, RSS))
2545 enic->rq_avail = 1;
2547 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2548 case VNIC_DEV_INTR_MODE_INTX:
2549 case VNIC_DEV_INTR_MODE_MSI:
2550 enic->rq_count = 1;
2551 enic->wq_count = 1;
2552 enic->cq_count = 2;
2553 enic->intr_count = enic->intr_avail;
2554 break;
2555 case VNIC_DEV_INTR_MODE_MSIX:
2556 /* Adjust the number of wqs/rqs/cqs/interrupts that will be
2557 * used based on which resource is the most constrained
2559 wq_avail = min(enic->wq_avail, ENIC_WQ_MAX);
2560 rq_default = netif_get_num_default_rss_queues();
2561 rq_avail = min3(enic->rq_avail, ENIC_RQ_MAX, rq_default);
2562 max_queues = min(enic->cq_avail,
2563 enic->intr_avail - ENIC_MSIX_RESERVED_INTR);
2564 if (wq_avail + rq_avail <= max_queues) {
2565 enic->rq_count = rq_avail;
2566 enic->wq_count = wq_avail;
2567 } else {
2568 /* recalculate wq/rq count */
2569 if (rq_avail < wq_avail) {
2570 enic->rq_count = min(rq_avail, max_queues / 2);
2571 enic->wq_count = max_queues - enic->rq_count;
2572 } else {
2573 enic->wq_count = min(wq_avail, max_queues / 2);
2574 enic->rq_count = max_queues - enic->wq_count;
2577 enic->cq_count = enic->rq_count + enic->wq_count;
2578 enic->intr_count = enic->cq_count + ENIC_MSIX_RESERVED_INTR;
2580 break;
2581 default:
2582 dev_err(enic_get_dev(enic), "Unknown interrupt mode\n");
2583 return -EINVAL;
2586 return 0;
2589 static void enic_get_queue_stats_rx(struct net_device *dev, int idx,
2590 struct netdev_queue_stats_rx *rxs)
2592 struct enic *enic = netdev_priv(dev);
2593 struct enic_rq_stats *rqstats = &enic->rq[idx].stats;
2595 rxs->bytes = rqstats->bytes;
2596 rxs->packets = rqstats->packets;
2597 rxs->hw_drops = rqstats->bad_fcs + rqstats->pkt_truncated;
2598 rxs->hw_drop_overruns = rqstats->pkt_truncated;
2599 rxs->csum_unnecessary = rqstats->csum_unnecessary +
2600 rqstats->csum_unnecessary_encap;
2603 static void enic_get_queue_stats_tx(struct net_device *dev, int idx,
2604 struct netdev_queue_stats_tx *txs)
2606 struct enic *enic = netdev_priv(dev);
2607 struct enic_wq_stats *wqstats = &enic->wq[idx].stats;
2609 txs->bytes = wqstats->bytes;
2610 txs->packets = wqstats->packets;
2611 txs->csum_none = wqstats->csum_none;
2612 txs->needs_csum = wqstats->csum_partial + wqstats->encap_csum +
2613 wqstats->tso;
2614 txs->hw_gso_packets = wqstats->tso;
2615 txs->stop = wqstats->stopped;
2616 txs->wake = wqstats->wake;
2619 static void enic_get_base_stats(struct net_device *dev,
2620 struct netdev_queue_stats_rx *rxs,
2621 struct netdev_queue_stats_tx *txs)
2623 rxs->bytes = 0;
2624 rxs->packets = 0;
2625 rxs->hw_drops = 0;
2626 rxs->hw_drop_overruns = 0;
2627 rxs->csum_unnecessary = 0;
2628 txs->bytes = 0;
2629 txs->packets = 0;
2630 txs->csum_none = 0;
2631 txs->needs_csum = 0;
2632 txs->hw_gso_packets = 0;
2633 txs->stop = 0;
2634 txs->wake = 0;
2637 static const struct net_device_ops enic_netdev_dynamic_ops = {
2638 .ndo_open = enic_open,
2639 .ndo_stop = enic_stop,
2640 .ndo_start_xmit = enic_hard_start_xmit,
2641 .ndo_get_stats64 = enic_get_stats,
2642 .ndo_validate_addr = eth_validate_addr,
2643 .ndo_set_rx_mode = enic_set_rx_mode,
2644 .ndo_set_mac_address = enic_set_mac_address_dynamic,
2645 .ndo_change_mtu = enic_change_mtu,
2646 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2647 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2648 .ndo_tx_timeout = enic_tx_timeout,
2649 .ndo_set_vf_port = enic_set_vf_port,
2650 .ndo_get_vf_port = enic_get_vf_port,
2651 .ndo_set_vf_mac = enic_set_vf_mac,
2652 #ifdef CONFIG_NET_POLL_CONTROLLER
2653 .ndo_poll_controller = enic_poll_controller,
2654 #endif
2655 #ifdef CONFIG_RFS_ACCEL
2656 .ndo_rx_flow_steer = enic_rx_flow_steer,
2657 #endif
2658 .ndo_features_check = enic_features_check,
2661 static const struct net_device_ops enic_netdev_ops = {
2662 .ndo_open = enic_open,
2663 .ndo_stop = enic_stop,
2664 .ndo_start_xmit = enic_hard_start_xmit,
2665 .ndo_get_stats64 = enic_get_stats,
2666 .ndo_validate_addr = eth_validate_addr,
2667 .ndo_set_mac_address = enic_set_mac_address,
2668 .ndo_set_rx_mode = enic_set_rx_mode,
2669 .ndo_change_mtu = enic_change_mtu,
2670 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2671 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2672 .ndo_tx_timeout = enic_tx_timeout,
2673 .ndo_set_vf_port = enic_set_vf_port,
2674 .ndo_get_vf_port = enic_get_vf_port,
2675 .ndo_set_vf_mac = enic_set_vf_mac,
2676 #ifdef CONFIG_NET_POLL_CONTROLLER
2677 .ndo_poll_controller = enic_poll_controller,
2678 #endif
2679 #ifdef CONFIG_RFS_ACCEL
2680 .ndo_rx_flow_steer = enic_rx_flow_steer,
2681 #endif
2682 .ndo_features_check = enic_features_check,
2685 static const struct netdev_stat_ops enic_netdev_stat_ops = {
2686 .get_queue_stats_rx = enic_get_queue_stats_rx,
2687 .get_queue_stats_tx = enic_get_queue_stats_tx,
2688 .get_base_stats = enic_get_base_stats,
2691 static void enic_free_enic_resources(struct enic *enic)
2693 kfree(enic->wq);
2694 enic->wq = NULL;
2696 kfree(enic->rq);
2697 enic->rq = NULL;
2699 kfree(enic->cq);
2700 enic->cq = NULL;
2702 kfree(enic->napi);
2703 enic->napi = NULL;
2705 kfree(enic->msix_entry);
2706 enic->msix_entry = NULL;
2708 kfree(enic->msix);
2709 enic->msix = NULL;
2711 kfree(enic->intr);
2712 enic->intr = NULL;
2715 static int enic_alloc_enic_resources(struct enic *enic)
2717 enic->wq = kcalloc(enic->wq_avail, sizeof(struct enic_wq), GFP_KERNEL);
2718 if (!enic->wq)
2719 goto free_queues;
2721 enic->rq = kcalloc(enic->rq_avail, sizeof(struct enic_rq), GFP_KERNEL);
2722 if (!enic->rq)
2723 goto free_queues;
2725 enic->cq = kcalloc(enic->cq_avail, sizeof(struct vnic_cq), GFP_KERNEL);
2726 if (!enic->cq)
2727 goto free_queues;
2729 enic->napi = kcalloc(enic->wq_avail + enic->rq_avail,
2730 sizeof(struct napi_struct), GFP_KERNEL);
2731 if (!enic->napi)
2732 goto free_queues;
2734 enic->msix_entry = kcalloc(enic->intr_avail, sizeof(struct msix_entry),
2735 GFP_KERNEL);
2736 if (!enic->msix_entry)
2737 goto free_queues;
2739 enic->msix = kcalloc(enic->intr_avail, sizeof(struct enic_msix_entry),
2740 GFP_KERNEL);
2741 if (!enic->msix)
2742 goto free_queues;
2744 enic->intr = kcalloc(enic->intr_avail, sizeof(struct vnic_intr),
2745 GFP_KERNEL);
2746 if (!enic->intr)
2747 goto free_queues;
2749 return 0;
2751 free_queues:
2752 enic_free_enic_resources(enic);
2753 return -ENOMEM;
2756 static void enic_dev_deinit(struct enic *enic)
2758 unsigned int i;
2760 for (i = 0; i < enic->rq_count; i++)
2761 __netif_napi_del(&enic->napi[i]);
2763 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2764 for (i = 0; i < enic->wq_count; i++)
2765 __netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2767 /* observe RCU grace period after __netif_napi_del() calls */
2768 synchronize_net();
2770 enic_free_vnic_resources(enic);
2771 enic_clear_intr_mode(enic);
2772 enic_free_affinity_hint(enic);
2773 enic_free_enic_resources(enic);
2776 static int enic_dev_init(struct enic *enic)
2778 struct device *dev = enic_get_dev(enic);
2779 struct net_device *netdev = enic->netdev;
2780 unsigned int i;
2781 int err;
2783 /* Get interrupt coalesce timer info */
2784 err = enic_dev_intr_coal_timer_info(enic);
2785 if (err) {
2786 dev_warn(dev, "Using default conversion factor for "
2787 "interrupt coalesce timer\n");
2788 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2791 /* Get vNIC configuration
2794 err = enic_get_vnic_config(enic);
2795 if (err) {
2796 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2797 return err;
2800 /* Get available resource counts
2803 enic_get_res_counts(enic);
2805 err = enic_alloc_enic_resources(enic);
2806 if (err) {
2807 dev_err(dev, "Failed to allocate enic resources\n");
2808 return err;
2811 /* Set interrupt mode based on system capabilities */
2813 err = enic_set_intr_mode(enic);
2814 if (err) {
2815 dev_err(dev, "Failed to set intr mode based on resource "
2816 "counts and system capabilities, aborting\n");
2817 goto err_out_free_vnic_resources;
2820 /* Adjust resource counts based on most constrained resources */
2821 err = enic_adjust_resources(enic);
2822 if (err) {
2823 dev_err(dev, "Failed to adjust resources\n");
2824 goto err_out_free_vnic_resources;
2827 /* Allocate and configure vNIC resources
2830 err = enic_alloc_vnic_resources(enic);
2831 if (err) {
2832 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2833 goto err_out_free_vnic_resources;
2836 enic_init_vnic_resources(enic);
2838 err = enic_set_rss_nic_cfg(enic);
2839 if (err) {
2840 dev_err(dev, "Failed to config nic, aborting\n");
2841 goto err_out_free_vnic_resources;
2844 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2845 default:
2846 netif_napi_add(netdev, &enic->napi[0], enic_poll);
2847 break;
2848 case VNIC_DEV_INTR_MODE_MSIX:
2849 for (i = 0; i < enic->rq_count; i++) {
2850 netif_napi_add(netdev, &enic->napi[i],
2851 enic_poll_msix_rq);
2853 for (i = 0; i < enic->wq_count; i++)
2854 netif_napi_add(netdev,
2855 &enic->napi[enic_cq_wq(enic, i)],
2856 enic_poll_msix_wq);
2857 break;
2860 return 0;
2862 err_out_free_vnic_resources:
2863 enic_free_affinity_hint(enic);
2864 enic_clear_intr_mode(enic);
2865 enic_free_vnic_resources(enic);
2866 enic_free_enic_resources(enic);
2868 return err;
2871 static void enic_iounmap(struct enic *enic)
2873 unsigned int i;
2875 for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2876 if (enic->bar[i].vaddr)
2877 iounmap(enic->bar[i].vaddr);
2880 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2882 struct device *dev = &pdev->dev;
2883 struct net_device *netdev;
2884 struct enic *enic;
2885 int using_dac = 0;
2886 unsigned int i;
2887 int err;
2888 #ifdef CONFIG_PCI_IOV
2889 int pos = 0;
2890 #endif
2891 int num_pps = 1;
2893 /* Allocate net device structure and initialize. Private
2894 * instance data is initialized to zero.
2897 netdev = alloc_etherdev_mqs(sizeof(struct enic),
2898 ENIC_RQ_MAX, ENIC_WQ_MAX);
2899 if (!netdev)
2900 return -ENOMEM;
2902 pci_set_drvdata(pdev, netdev);
2904 SET_NETDEV_DEV(netdev, &pdev->dev);
2906 enic = netdev_priv(netdev);
2907 enic->netdev = netdev;
2908 enic->pdev = pdev;
2910 /* Setup PCI resources
2913 err = pci_enable_device_mem(pdev);
2914 if (err) {
2915 dev_err(dev, "Cannot enable PCI device, aborting\n");
2916 goto err_out_free_netdev;
2919 err = pci_request_regions(pdev, DRV_NAME);
2920 if (err) {
2921 dev_err(dev, "Cannot request PCI regions, aborting\n");
2922 goto err_out_disable_device;
2925 pci_set_master(pdev);
2927 /* Query PCI controller on system for DMA addressing
2928 * limitation for the device. Try 47-bit first, and
2929 * fail to 32-bit.
2932 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(47));
2933 if (err) {
2934 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2935 if (err) {
2936 dev_err(dev, "No usable DMA configuration, aborting\n");
2937 goto err_out_release_regions;
2939 } else {
2940 using_dac = 1;
2943 /* Map vNIC resources from BAR0-5
2946 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2947 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2948 continue;
2949 enic->bar[i].len = pci_resource_len(pdev, i);
2950 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2951 if (!enic->bar[i].vaddr) {
2952 dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2953 err = -ENODEV;
2954 goto err_out_iounmap;
2956 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2959 /* Register vNIC device
2962 enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2963 ARRAY_SIZE(enic->bar));
2964 if (!enic->vdev) {
2965 dev_err(dev, "vNIC registration failed, aborting\n");
2966 err = -ENODEV;
2967 goto err_out_iounmap;
2970 err = vnic_devcmd_init(enic->vdev);
2972 if (err)
2973 goto err_out_vnic_unregister;
2975 #ifdef CONFIG_PCI_IOV
2976 /* Get number of subvnics */
2977 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2978 if (pos) {
2979 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2980 &enic->num_vfs);
2981 if (enic->num_vfs) {
2982 err = pci_enable_sriov(pdev, enic->num_vfs);
2983 if (err) {
2984 dev_err(dev, "SRIOV enable failed, aborting."
2985 " pci_enable_sriov() returned %d\n",
2986 err);
2987 goto err_out_vnic_unregister;
2989 enic->priv_flags |= ENIC_SRIOV_ENABLED;
2990 num_pps = enic->num_vfs;
2993 #endif
2995 /* Allocate structure for port profiles */
2996 enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2997 if (!enic->pp) {
2998 err = -ENOMEM;
2999 goto err_out_disable_sriov_pp;
3002 /* Issue device open to get device in known state
3005 err = enic_dev_open(enic);
3006 if (err) {
3007 dev_err(dev, "vNIC dev open failed, aborting\n");
3008 goto err_out_disable_sriov;
3011 /* Setup devcmd lock
3014 spin_lock_init(&enic->devcmd_lock);
3015 spin_lock_init(&enic->enic_api_lock);
3018 * Set ingress vlan rewrite mode before vnic initialization
3021 err = enic_dev_set_ig_vlan_rewrite_mode(enic);
3022 if (err) {
3023 dev_err(dev,
3024 "Failed to set ingress vlan rewrite mode, aborting.\n");
3025 goto err_out_dev_close;
3028 /* Issue device init to initialize the vnic-to-switch link.
3029 * We'll start with carrier off and wait for link UP
3030 * notification later to turn on carrier. We don't need
3031 * to wait here for the vnic-to-switch link initialization
3032 * to complete; link UP notification is the indication that
3033 * the process is complete.
3036 netif_carrier_off(netdev);
3038 /* Do not call dev_init for a dynamic vnic.
3039 * For a dynamic vnic, init_prov_info will be
3040 * called later by an upper layer.
3043 if (!enic_is_dynamic(enic)) {
3044 err = vnic_dev_init(enic->vdev, 0);
3045 if (err) {
3046 dev_err(dev, "vNIC dev init failed, aborting\n");
3047 goto err_out_dev_close;
3051 err = enic_dev_init(enic);
3052 if (err) {
3053 dev_err(dev, "Device initialization failed, aborting\n");
3054 goto err_out_dev_close;
3057 netif_set_real_num_tx_queues(netdev, enic->wq_count);
3058 netif_set_real_num_rx_queues(netdev, enic->rq_count);
3060 /* Setup notification timer, HW reset task, and wq locks
3063 timer_setup(&enic->notify_timer, enic_notify_timer, 0);
3065 enic_rfs_flw_tbl_init(enic);
3066 enic_set_rx_coal_setting(enic);
3067 INIT_WORK(&enic->reset, enic_reset);
3068 INIT_WORK(&enic->tx_hang_reset, enic_tx_hang_reset);
3069 INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
3071 for (i = 0; i < enic->wq_count; i++)
3072 spin_lock_init(&enic->wq[i].lock);
3074 /* Register net device
3077 enic->port_mtu = enic->config.mtu;
3079 err = enic_set_mac_addr(netdev, enic->mac_addr);
3080 if (err) {
3081 dev_err(dev, "Invalid MAC address, aborting\n");
3082 goto err_out_dev_deinit;
3085 enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
3086 /* rx coalesce time already got initialized. This gets used
3087 * if adaptive coal is turned off
3089 enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
3091 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
3092 netdev->netdev_ops = &enic_netdev_dynamic_ops;
3093 else
3094 netdev->netdev_ops = &enic_netdev_ops;
3095 netdev->stat_ops = &enic_netdev_stat_ops;
3097 netdev->watchdog_timeo = 2 * HZ;
3098 enic_set_ethtool_ops(netdev);
3100 netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3101 if (ENIC_SETTING(enic, LOOP)) {
3102 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
3103 enic->loop_enable = 1;
3104 enic->loop_tag = enic->config.loop_tag;
3105 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
3107 if (ENIC_SETTING(enic, TXCSUM))
3108 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
3109 if (ENIC_SETTING(enic, TSO))
3110 netdev->hw_features |= NETIF_F_TSO |
3111 NETIF_F_TSO6 | NETIF_F_TSO_ECN;
3112 if (ENIC_SETTING(enic, RSS))
3113 netdev->hw_features |= NETIF_F_RXHASH;
3114 if (ENIC_SETTING(enic, RXCSUM))
3115 netdev->hw_features |= NETIF_F_RXCSUM;
3116 if (ENIC_SETTING(enic, VXLAN)) {
3117 u64 patch_level;
3118 u64 a1 = 0;
3120 netdev->hw_enc_features |= NETIF_F_RXCSUM |
3121 NETIF_F_TSO |
3122 NETIF_F_TSO6 |
3123 NETIF_F_TSO_ECN |
3124 NETIF_F_GSO_UDP_TUNNEL |
3125 NETIF_F_HW_CSUM |
3126 NETIF_F_GSO_UDP_TUNNEL_CSUM;
3127 netdev->hw_features |= netdev->hw_enc_features;
3128 /* get bit mask from hw about supported offload bit level
3129 * BIT(0) = fw supports patch_level 0
3130 * fcoe bit = encap
3131 * fcoe_fc_crc_ok = outer csum ok
3132 * BIT(1) = always set by fw
3133 * BIT(2) = fw supports patch_level 2
3134 * BIT(0) in rss_hash = encap
3135 * BIT(1,2) in rss_hash = outer_ip_csum_ok/
3136 * outer_tcp_csum_ok
3137 * used in enic_rq_indicate_buf
3139 err = vnic_dev_get_supported_feature_ver(enic->vdev,
3140 VIC_FEATURE_VXLAN,
3141 &patch_level, &a1);
3142 if (err)
3143 patch_level = 0;
3144 enic->vxlan.flags = (u8)a1;
3145 /* mask bits that are supported by driver
3147 patch_level &= BIT_ULL(0) | BIT_ULL(2);
3148 patch_level = fls(patch_level);
3149 patch_level = patch_level ? patch_level - 1 : 0;
3150 enic->vxlan.patch_level = patch_level;
3152 if (vnic_dev_get_res_count(enic->vdev, RES_TYPE_WQ) == 1 ||
3153 enic->vxlan.flags & ENIC_VXLAN_MULTI_WQ) {
3154 netdev->udp_tunnel_nic_info = &enic_udp_tunnels_v4;
3155 if (enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6)
3156 netdev->udp_tunnel_nic_info = &enic_udp_tunnels;
3160 netdev->features |= netdev->hw_features;
3161 netdev->vlan_features |= netdev->features;
3163 #ifdef CONFIG_RFS_ACCEL
3164 netdev->hw_features |= NETIF_F_NTUPLE;
3165 #endif
3167 if (using_dac)
3168 netdev->features |= NETIF_F_HIGHDMA;
3170 netdev->priv_flags |= IFF_UNICAST_FLT;
3172 /* MTU range: 68 - 9000 */
3173 netdev->min_mtu = ENIC_MIN_MTU;
3174 netdev->max_mtu = ENIC_MAX_MTU;
3175 netdev->mtu = enic->port_mtu;
3177 err = register_netdev(netdev);
3178 if (err) {
3179 dev_err(dev, "Cannot register net device, aborting\n");
3180 goto err_out_dev_deinit;
3182 enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
3184 return 0;
3186 err_out_dev_deinit:
3187 enic_dev_deinit(enic);
3188 err_out_dev_close:
3189 vnic_dev_close(enic->vdev);
3190 err_out_disable_sriov:
3191 kfree(enic->pp);
3192 err_out_disable_sriov_pp:
3193 #ifdef CONFIG_PCI_IOV
3194 if (enic_sriov_enabled(enic)) {
3195 pci_disable_sriov(pdev);
3196 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
3198 #endif
3199 err_out_vnic_unregister:
3200 vnic_dev_unregister(enic->vdev);
3201 err_out_iounmap:
3202 enic_iounmap(enic);
3203 err_out_release_regions:
3204 pci_release_regions(pdev);
3205 err_out_disable_device:
3206 pci_disable_device(pdev);
3207 err_out_free_netdev:
3208 free_netdev(netdev);
3210 return err;
3213 static void enic_remove(struct pci_dev *pdev)
3215 struct net_device *netdev = pci_get_drvdata(pdev);
3217 if (netdev) {
3218 struct enic *enic = netdev_priv(netdev);
3220 cancel_work_sync(&enic->reset);
3221 cancel_work_sync(&enic->change_mtu_work);
3222 unregister_netdev(netdev);
3223 enic_dev_deinit(enic);
3224 vnic_dev_close(enic->vdev);
3225 #ifdef CONFIG_PCI_IOV
3226 if (enic_sriov_enabled(enic)) {
3227 pci_disable_sriov(pdev);
3228 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
3230 #endif
3231 kfree(enic->pp);
3232 vnic_dev_unregister(enic->vdev);
3233 enic_iounmap(enic);
3234 pci_release_regions(pdev);
3235 pci_disable_device(pdev);
3236 free_netdev(netdev);
3240 static struct pci_driver enic_driver = {
3241 .name = DRV_NAME,
3242 .id_table = enic_id_table,
3243 .probe = enic_probe,
3244 .remove = enic_remove,
3247 module_pci_driver(enic_driver);