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
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>
32 #include <linux/if_ether.h>
33 #include <linux/if_vlan.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>
46 #include <linux/crash_dump.h>
47 #include <net/busy_poll.h>
48 #include <net/vxlan.h>
50 #include "cq_enet_desc.h"
52 #include "vnic_intr.h"
53 #include "vnic_stats.h"
59 #include "enic_clsf.h"
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)
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 */
70 #define RX_COPYBREAK_DEFAULT 256
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 */
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);
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}
91 static struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
105 /* This table helps the driver to pick different ranges for rx coalescing
106 * timer depending on the link speed.
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 */
114 static void enic_init_affinity_hint(struct enic *enic)
116 int numa_node = dev_to_node(&enic->pdev->dev);
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)))
124 if (zalloc_cpumask_var(&enic->msix[i].affinity_mask,
126 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
127 enic->msix[i].affinity_mask);
131 static void enic_free_affinity_hint(struct enic *enic)
135 for (i = 0; i < enic->intr_count; i++) {
136 if (enic_is_err_intr(enic, i) || enic_is_notify_intr(enic, i))
138 free_cpumask_var(enic->msix[i].affinity_mask);
142 static void enic_set_affinity_hint(struct enic *enic)
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))
153 err = irq_set_affinity_hint(enic->msix_entry[i].vector,
154 enic->msix[i].affinity_mask);
156 netdev_warn(enic->netdev, "irq_set_affinity_hint failed, err %d\n",
160 for (i = 0; i < enic->wq_count; i++) {
161 int wq_intr = enic_msix_wq_intr(enic, i);
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,
171 static void enic_unset_affinity_hint(struct enic *enic)
175 for (i = 0; i < enic->intr_count; i++)
176 irq_set_affinity_hint(enic->msix_entry[i].vector, NULL);
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)
183 struct enic *enic = netdev_priv(netdev);
186 spin_lock_bh(&enic->devcmd_lock);
188 err = vnic_dev_overlay_offload_cfg(enic->vdev,
189 OVERLAY_CFG_VXLAN_PORT_UPDATE,
194 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
195 enic->vxlan.patch_level);
199 enic->vxlan.vxlan_udp_port_number = ntohs(ti->port);
201 spin_unlock_bh(&enic->devcmd_lock);
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)
210 struct enic *enic = netdev_priv(netdev);
213 spin_lock_bh(&enic->devcmd_lock);
215 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
216 OVERLAY_OFFLOAD_DISABLE);
220 enic->vxlan.vxlan_udp_port_number = 0;
223 spin_unlock_bh(&enic->devcmd_lock);
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,
232 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
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,
239 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
243 static netdev_features_t enic_features_check(struct sk_buff *skb,
244 struct net_device *dev,
245 netdev_features_t features)
247 const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb);
248 struct enic *enic = netdev_priv(dev);
253 if (!skb->encapsulation)
256 features = vxlan_features_check(skb, features);
258 switch (vlan_get_protocol(skb)) {
259 case htons(ETH_P_IPV6):
260 if (!(enic->vxlan.flags & ENIC_VXLAN_OUTER_IPV6))
262 proto = ipv6_hdr(skb)->nexthdr;
264 case htons(ETH_P_IP):
265 proto = ip_hdr(skb)->protocol;
271 switch (eth->h_proto) {
272 case ntohs(ETH_P_IPV6):
273 if (!(enic->vxlan.flags & ENIC_VXLAN_INNER_IPV6))
276 case ntohs(ETH_P_IP):
283 if (proto == IPPROTO_UDP) {
285 port = be16_to_cpu(udph->dest);
288 /* HW supports offload of only one UDP port. Remove CSUM and GSO MASK
289 * for other UDP port tunnels
291 if (port != enic->vxlan.vxlan_udp_port_number)
297 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
300 int enic_is_dynamic(struct enic *enic)
302 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
305 int enic_sriov_enabled(struct enic *enic)
307 return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
310 static int enic_is_sriov_vf(struct enic *enic)
312 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
315 int enic_is_valid_vf(struct enic *enic, int vf)
317 #ifdef CONFIG_PCI_IOV
318 return vf >= 0 && vf < enic->num_vfs;
324 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
326 struct enic *enic = vnic_dev_priv(wq->vdev);
329 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
332 dma_unmap_page(&enic->pdev->dev, buf->dma_addr, buf->len,
336 dev_kfree_skb_any(buf->os_buf);
339 static void enic_wq_free_buf(struct vnic_wq *wq,
340 struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
342 enic_free_wq_buf(wq, buf);
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)
348 struct enic *enic = vnic_dev_priv(vdev);
350 spin_lock(&enic->wq_lock[q_number]);
352 vnic_wq_service(&enic->wq[q_number], cq_desc,
353 completed_index, enic_wq_free_buf,
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);
361 spin_unlock(&enic->wq_lock[q_number]);
366 static bool enic_log_q_error(struct enic *enic)
372 for (i = 0; i < enic->wq_count; i++) {
373 error_status = vnic_wq_error_status(&enic->wq[i]);
376 netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
380 for (i = 0; i < enic->rq_count; i++) {
381 error_status = vnic_rq_error_status(&enic->rq[i]);
384 netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
391 static void enic_msglvl_check(struct enic *enic)
393 u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
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;
402 static void enic_mtu_check(struct enic *enic)
404 u32 mtu = vnic_dev_mtu(enic->vdev);
405 struct net_device *netdev = enic->netdev;
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);
415 if (mtu < netdev->mtu)
417 "interface MTU (%d) set higher "
418 "than switch port MTU (%d)\n",
424 static void enic_link_check(struct enic *enic)
426 int link_status = vnic_dev_link_status(enic->vdev);
427 int carrier_ok = netif_carrier_ok(enic->netdev);
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);
438 static void enic_notify_check(struct enic *enic)
440 enic_msglvl_check(enic);
441 enic_mtu_check(enic);
442 enic_link_check(enic);
445 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
447 static irqreturn_t enic_isr_legacy(int irq, void *data)
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();
456 vnic_intr_mask(&enic->intr[io_intr]);
458 pba = vnic_intr_legacy_pba(enic->legacy_pba);
460 vnic_intr_unmask(&enic->intr[io_intr]);
461 return IRQ_NONE; /* not our interrupt */
464 if (ENIC_TEST_INTR(pba, notify_intr)) {
465 enic_notify_check(enic);
466 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
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);
477 if (ENIC_TEST_INTR(pba, io_intr))
478 napi_schedule_irqoff(&enic->napi[0]);
480 vnic_intr_unmask(&enic->intr[io_intr]);
485 static irqreturn_t enic_isr_msi(int irq, void *data)
487 struct enic *enic = data;
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).
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
505 napi_schedule_irqoff(&enic->napi[0]);
510 static irqreturn_t enic_isr_msix(int irq, void *data)
512 struct napi_struct *napi = data;
514 napi_schedule_irqoff(napi);
519 static irqreturn_t enic_isr_msix_err(int irq, void *data)
521 struct enic *enic = data;
522 unsigned int intr = enic_msix_err_intr(enic);
524 vnic_intr_return_all_credits(&enic->intr[intr]);
526 if (enic_log_q_error(enic))
527 /* schedule recovery from WQ/RQ error */
528 schedule_work(&enic->reset);
533 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
535 struct enic *enic = data;
536 unsigned int intr = enic_msix_notify_intr(enic);
538 enic_notify_check(enic);
539 vnic_intr_return_all_credits(&enic->intr[intr]);
544 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq,
545 struct sk_buff *skb, unsigned int len_left,
548 const skb_frag_t *frag;
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,
557 if (unlikely(enic_dma_map_check(enic, dma_addr)))
559 enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag),
560 (len_left == 0), /* EOP? */
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)
571 unsigned int head_len = skb_headlen(skb);
572 unsigned int len_left = skb->len - head_len;
573 int eop = (len_left == 0);
577 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len,
579 if (unlikely(enic_dma_map_check(enic, dma_addr)))
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.
587 enic_queue_wq_desc(wq, skb, dma_addr, head_len, vlan_tag_insert,
588 vlan_tag, eop, loopback);
591 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
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)
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);
608 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len,
610 if (unlikely(enic_dma_map_check(enic, dma_addr)))
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.
618 enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len, csum_offset,
619 hdr_len, vlan_tag_insert, vlan_tag, eop,
623 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
628 static void enic_preload_tcp_csum_encap(struct sk_buff *skb)
630 const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb);
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,
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,
647 WARN_ONCE(1, "Non ipv4/ipv6 inner pkt for encap offload");
652 static void enic_preload_tcp_csum(struct sk_buff *skb)
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.
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);
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,
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;
682 if (skb->encapsulation) {
683 hdr_len = skb_inner_transport_header(skb) - skb->data;
684 hdr_len += inner_tcp_hdrlen(skb);
685 enic_preload_tcp_csum_encap(skb);
687 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
688 enic_preload_tcp_csum(skb);
691 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
692 * for the main skb fragment
694 while (frag_len_left) {
695 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
696 dma_addr = dma_map_single(&enic->pdev->dev,
697 skb->data + offset, len,
699 if (unlikely(enic_dma_map_check(enic, dma_addr)))
701 enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len,
702 vlan_tag_insert, vlan_tag,
703 eop && (len == frag_len_left), loopback);
704 frag_len_left -= len;
711 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
712 * for additional data fragments
714 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
715 len_left -= skb_frag_size(frag);
716 frag_len_left = skb_frag_size(frag);
719 while (frag_len_left) {
720 len = min(frag_len_left,
721 (unsigned int)WQ_ENET_MAX_DESC_LEN);
722 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
725 if (unlikely(enic_dma_map_check(enic, dma_addr)))
727 enic_queue_wq_desc_cont(wq, skb, dma_addr, len,
729 (len == frag_len_left),/*EOP*/
731 frag_len_left -= len;
739 static inline int enic_queue_wq_skb_encap(struct enic *enic, struct vnic_wq *wq,
742 unsigned int vlan_tag, int loopback)
744 unsigned int head_len = skb_headlen(skb);
745 unsigned int len_left = skb->len - head_len;
746 /* Hardware will overwrite the checksum fields, calculating from
747 * scratch and ignoring the value placed by software.
749 * mss[2], mss[1], mss[0] bits are set
751 unsigned int mss_or_csum = 7;
752 int eop = (len_left == 0);
756 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, head_len,
758 if (unlikely(enic_dma_map_check(enic, dma_addr)))
761 enic_queue_wq_desc_ex(wq, skb, dma_addr, head_len, mss_or_csum, 0,
762 vlan_tag_insert, vlan_tag,
763 WQ_ENET_OFFLOAD_MODE_CSUM, eop, 1 /* SOP */, eop,
766 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
771 static inline int enic_queue_wq_skb(struct enic *enic,
772 struct vnic_wq *wq, struct sk_buff *skb)
774 unsigned int mss = skb_shinfo(skb)->gso_size;
775 unsigned int vlan_tag = 0;
776 int vlan_tag_insert = 0;
780 if (skb_vlan_tag_present(skb)) {
781 /* VLAN tag from trunking driver */
783 vlan_tag = skb_vlan_tag_get(skb);
784 } else if (enic->loop_enable) {
785 vlan_tag = enic->loop_tag;
790 err = enic_queue_wq_skb_tso(enic, wq, skb, mss,
791 vlan_tag_insert, vlan_tag,
793 else if (skb->encapsulation)
794 err = enic_queue_wq_skb_encap(enic, wq, skb, vlan_tag_insert,
796 else if (skb->ip_summed == CHECKSUM_PARTIAL)
797 err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert,
800 err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert,
803 struct vnic_wq_buf *buf;
805 buf = wq->to_use->prev;
806 /* while not EOP of previous pkt && queue not empty.
807 * For all non EOP bufs, os_buf is NULL.
809 while (!buf->os_buf && (buf->next != wq->to_clean)) {
810 enic_free_wq_buf(wq, buf);
811 wq->ring.desc_avail++;
814 wq->to_use = buf->next;
820 /* netif_tx_lock held, process context with BHs disabled, or BH */
821 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
822 struct net_device *netdev)
824 struct enic *enic = netdev_priv(netdev);
826 unsigned int txq_map;
827 struct netdev_queue *txq;
830 dev_kfree_skb_any(skb);
834 txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
835 wq = &enic->wq[txq_map];
836 txq = netdev_get_tx_queue(netdev, txq_map);
838 /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
839 * which is very likely. In the off chance it's going to take
840 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
843 if (skb_shinfo(skb)->gso_size == 0 &&
844 skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
845 skb_linearize(skb)) {
846 dev_kfree_skb_any(skb);
850 spin_lock(&enic->wq_lock[txq_map]);
852 if (vnic_wq_desc_avail(wq) <
853 skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
854 netif_tx_stop_queue(txq);
855 /* This is a hard error, log it */
856 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
857 spin_unlock(&enic->wq_lock[txq_map]);
858 return NETDEV_TX_BUSY;
861 if (enic_queue_wq_skb(enic, wq, skb))
864 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
865 netif_tx_stop_queue(txq);
866 skb_tx_timestamp(skb);
867 if (!netdev_xmit_more() || netif_xmit_stopped(txq))
868 vnic_wq_doorbell(wq);
871 spin_unlock(&enic->wq_lock[txq_map]);
876 /* dev_base_lock rwlock held, nominally process context */
877 static void enic_get_stats(struct net_device *netdev,
878 struct rtnl_link_stats64 *net_stats)
880 struct enic *enic = netdev_priv(netdev);
881 struct vnic_stats *stats;
884 err = enic_dev_stats_dump(enic, &stats);
885 /* return only when pci_zalloc_consistent fails in vnic_dev_stats_dump
886 * For other failures, like devcmd failure, we return previously
892 net_stats->tx_packets = stats->tx.tx_frames_ok;
893 net_stats->tx_bytes = stats->tx.tx_bytes_ok;
894 net_stats->tx_errors = stats->tx.tx_errors;
895 net_stats->tx_dropped = stats->tx.tx_drops;
897 net_stats->rx_packets = stats->rx.rx_frames_ok;
898 net_stats->rx_bytes = stats->rx.rx_bytes_ok;
899 net_stats->rx_errors = stats->rx.rx_errors;
900 net_stats->multicast = stats->rx.rx_multicast_frames_ok;
901 net_stats->rx_over_errors = enic->rq_truncated_pkts;
902 net_stats->rx_crc_errors = enic->rq_bad_fcs;
903 net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
906 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
908 struct enic *enic = netdev_priv(netdev);
910 if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
911 unsigned int mc_count = netdev_mc_count(netdev);
913 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
914 ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
919 enic_dev_add_addr(enic, mc_addr);
925 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
927 struct enic *enic = netdev_priv(netdev);
929 enic_dev_del_addr(enic, mc_addr);
935 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
937 struct enic *enic = netdev_priv(netdev);
939 if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
940 unsigned int uc_count = netdev_uc_count(netdev);
942 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
943 ENIC_UNICAST_PERFECT_FILTERS, uc_count);
948 enic_dev_add_addr(enic, uc_addr);
954 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
956 struct enic *enic = netdev_priv(netdev);
958 enic_dev_del_addr(enic, uc_addr);
964 void enic_reset_addr_lists(struct enic *enic)
966 struct net_device *netdev = enic->netdev;
968 __dev_uc_unsync(netdev, NULL);
969 __dev_mc_unsync(netdev, NULL);
976 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
978 struct enic *enic = netdev_priv(netdev);
980 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
981 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
982 return -EADDRNOTAVAIL;
984 if (!is_valid_ether_addr(addr))
985 return -EADDRNOTAVAIL;
988 memcpy(netdev->dev_addr, addr, netdev->addr_len);
993 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
995 struct enic *enic = netdev_priv(netdev);
996 struct sockaddr *saddr = p;
997 char *addr = saddr->sa_data;
1000 if (netif_running(enic->netdev)) {
1001 err = enic_dev_del_station_addr(enic);
1006 err = enic_set_mac_addr(netdev, addr);
1010 if (netif_running(enic->netdev)) {
1011 err = enic_dev_add_station_addr(enic);
1019 static int enic_set_mac_address(struct net_device *netdev, void *p)
1021 struct sockaddr *saddr = p;
1022 char *addr = saddr->sa_data;
1023 struct enic *enic = netdev_priv(netdev);
1026 err = enic_dev_del_station_addr(enic);
1030 err = enic_set_mac_addr(netdev, addr);
1034 return enic_dev_add_station_addr(enic);
1037 /* netif_tx_lock held, BHs disabled */
1038 static void enic_set_rx_mode(struct net_device *netdev)
1040 struct enic *enic = netdev_priv(netdev);
1042 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
1043 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
1044 int promisc = (netdev->flags & IFF_PROMISC) ||
1045 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
1046 int allmulti = (netdev->flags & IFF_ALLMULTI) ||
1047 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
1048 unsigned int flags = netdev->flags |
1049 (allmulti ? IFF_ALLMULTI : 0) |
1050 (promisc ? IFF_PROMISC : 0);
1052 if (enic->flags != flags) {
1053 enic->flags = flags;
1054 enic_dev_packet_filter(enic, directed,
1055 multicast, broadcast, promisc, allmulti);
1059 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
1061 __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
1065 /* netif_tx_lock held, BHs disabled */
1066 static void enic_tx_timeout(struct net_device *netdev, unsigned int txqueue)
1068 struct enic *enic = netdev_priv(netdev);
1069 schedule_work(&enic->tx_hang_reset);
1072 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1074 struct enic *enic = netdev_priv(netdev);
1075 struct enic_port_profile *pp;
1078 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1082 if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
1083 if (vf == PORT_SELF_VF) {
1084 memcpy(pp->vf_mac, mac, ETH_ALEN);
1088 * For sriov vf's set the mac in hw
1090 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1091 vnic_dev_set_mac_addr, mac);
1092 return enic_dev_status_to_errno(err);
1098 static int enic_set_vf_port(struct net_device *netdev, int vf,
1099 struct nlattr *port[])
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;
1106 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1110 if (!port[IFLA_PORT_REQUEST])
1113 memcpy(&prev_pp, pp, sizeof(*enic->pp));
1114 memset(pp, 0, sizeof(*enic->pp));
1116 pp->set |= ENIC_SET_REQUEST;
1117 pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1119 if (port[IFLA_PORT_PROFILE]) {
1120 pp->set |= ENIC_SET_NAME;
1121 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
1125 if (port[IFLA_PORT_INSTANCE_UUID]) {
1126 pp->set |= ENIC_SET_INSTANCE;
1127 memcpy(pp->instance_uuid,
1128 nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
1131 if (port[IFLA_PORT_HOST_UUID]) {
1132 pp->set |= ENIC_SET_HOST;
1133 memcpy(pp->host_uuid,
1134 nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
1137 if (vf == PORT_SELF_VF) {
1138 /* Special case handling: mac came from IFLA_VF_MAC */
1139 if (!is_zero_ether_addr(prev_pp.vf_mac))
1140 memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
1142 if (is_zero_ether_addr(netdev->dev_addr))
1143 eth_hw_addr_random(netdev);
1145 /* SR-IOV VF: get mac from adapter */
1146 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1147 vnic_dev_get_mac_addr, pp->mac_addr);
1149 netdev_err(netdev, "Error getting mac for vf %d\n", vf);
1150 memcpy(pp, &prev_pp, sizeof(*pp));
1151 return enic_dev_status_to_errno(err);
1155 err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
1158 /* Things are still the way they were: Implicit
1159 * DISASSOCIATE failed
1161 memcpy(pp, &prev_pp, sizeof(*pp));
1163 memset(pp, 0, sizeof(*pp));
1164 if (vf == PORT_SELF_VF)
1165 eth_zero_addr(netdev->dev_addr);
1168 /* Set flag to indicate that the port assoc/disassoc
1169 * request has been sent out to fw
1171 pp->set |= ENIC_PORT_REQUEST_APPLIED;
1173 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
1174 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
1175 eth_zero_addr(pp->mac_addr);
1176 if (vf == PORT_SELF_VF)
1177 eth_zero_addr(netdev->dev_addr);
1181 if (vf == PORT_SELF_VF)
1182 eth_zero_addr(pp->vf_mac);
1187 static int enic_get_vf_port(struct net_device *netdev, int vf,
1188 struct sk_buff *skb)
1190 struct enic *enic = netdev_priv(netdev);
1191 u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1192 struct enic_port_profile *pp;
1195 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1199 if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
1202 err = enic_process_get_pp_request(enic, vf, pp->request, &response);
1206 if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
1207 nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
1208 ((pp->set & ENIC_SET_NAME) &&
1209 nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
1210 ((pp->set & ENIC_SET_INSTANCE) &&
1211 nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1212 pp->instance_uuid)) ||
1213 ((pp->set & ENIC_SET_HOST) &&
1214 nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
1215 goto nla_put_failure;
1222 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1224 struct enic *enic = vnic_dev_priv(rq->vdev);
1229 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
1231 dev_kfree_skb_any(buf->os_buf);
1235 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1237 struct enic *enic = vnic_dev_priv(rq->vdev);
1238 struct net_device *netdev = enic->netdev;
1239 struct sk_buff *skb;
1240 unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
1241 unsigned int os_buf_index = 0;
1242 dma_addr_t dma_addr;
1243 struct vnic_rq_buf *buf = rq->to_use;
1246 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
1251 skb = netdev_alloc_skb_ip_align(netdev, len);
1255 dma_addr = dma_map_single(&enic->pdev->dev, skb->data, len,
1257 if (unlikely(enic_dma_map_check(enic, dma_addr))) {
1262 enic_queue_rq_desc(rq, skb, os_buf_index,
1268 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1271 if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1272 pkt_size->large_pkt_bytes_cnt += pkt_len;
1274 pkt_size->small_pkt_bytes_cnt += pkt_len;
1277 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
1278 struct vnic_rq_buf *buf, u16 len)
1280 struct enic *enic = netdev_priv(netdev);
1281 struct sk_buff *new_skb;
1283 if (len > enic->rx_copybreak)
1285 new_skb = netdev_alloc_skb_ip_align(netdev, len);
1288 dma_sync_single_for_cpu(&enic->pdev->dev, buf->dma_addr, len,
1290 memcpy(new_skb->data, (*skb)->data, len);
1296 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1297 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1298 int skipped, void *opaque)
1300 struct enic *enic = vnic_dev_priv(rq->vdev);
1301 struct net_device *netdev = enic->netdev;
1302 struct sk_buff *skb;
1303 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1305 u8 type, color, eop, sop, ingress_port, vlan_stripped;
1306 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1307 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1308 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1310 u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1312 bool outer_csum_ok = true, encap = false;
1319 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1320 &type, &color, &q_number, &completed_index,
1321 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1322 &csum_not_calc, &rss_hash, &bytes_written,
1323 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1324 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1325 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1326 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1332 if (bytes_written > 0)
1334 else if (bytes_written == 0)
1335 enic->rq_truncated_pkts++;
1338 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
1340 dev_kfree_skb_any(skb);
1346 if (eop && bytes_written > 0) {
1351 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1353 dma_unmap_single(&enic->pdev->dev, buf->dma_addr,
1354 buf->len, DMA_FROM_DEVICE);
1356 prefetch(skb->data - NET_IP_ALIGN);
1358 skb_put(skb, bytes_written);
1359 skb->protocol = eth_type_trans(skb, netdev);
1360 skb_record_rx_queue(skb, q_number);
1361 if ((netdev->features & NETIF_F_RXHASH) && rss_hash &&
1364 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
1365 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
1366 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX:
1367 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4);
1369 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4:
1370 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6:
1371 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX:
1372 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3);
1376 if (enic->vxlan.vxlan_udp_port_number) {
1377 switch (enic->vxlan.patch_level) {
1381 outer_csum_ok = fcoe_fc_crc_ok;
1386 (rss_hash & BIT(0))) {
1388 outer_csum_ok = (rss_hash & BIT(1)) &&
1389 (rss_hash & BIT(2));
1395 /* Hardware does not provide whole packet checksum. It only
1396 * provides pseudo checksum. Since hw validates the packet
1397 * checksum but not provide us the checksum value. use
1398 * CHECSUM_UNNECESSARY.
1400 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is
1401 * inner csum_ok. outer_csum_ok is set by hw when outer udp
1402 * csum is correct or is zero.
1404 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc &&
1405 tcp_udp_csum_ok && outer_csum_ok &&
1406 (ipv4_csum_ok || ipv6)) {
1407 skb->ip_summed = CHECKSUM_UNNECESSARY;
1408 skb->csum_level = encap;
1412 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1414 skb_mark_napi_id(skb, &enic->napi[rq->index]);
1415 if (!(netdev->features & NETIF_F_GRO))
1416 netif_receive_skb(skb);
1418 napi_gro_receive(&enic->napi[q_number], skb);
1419 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1420 enic_intr_update_pkt_size(&cq->pkt_size_counter,
1427 dma_unmap_single(&enic->pdev->dev, buf->dma_addr, buf->len,
1429 dev_kfree_skb_any(skb);
1434 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1435 u8 type, u16 q_number, u16 completed_index, void *opaque)
1437 struct enic *enic = vnic_dev_priv(vdev);
1439 vnic_rq_service(&enic->rq[q_number], cq_desc,
1440 completed_index, VNIC_RQ_RETURN_DESC,
1441 enic_rq_indicate_buf, opaque);
1446 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1448 unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1449 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1450 u32 timer = cq->tobe_rx_coal_timeval;
1452 if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1453 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1454 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1458 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1460 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1461 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1462 struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1468 ktime_t now = ktime_get();
1470 delta = ktime_us_delta(now, cq->prev_ts);
1471 if (delta < ENIC_AIC_TS_BREAK)
1475 traffic = pkt_size_counter->large_pkt_bytes_cnt +
1476 pkt_size_counter->small_pkt_bytes_cnt;
1477 /* The table takes Mbps
1478 * traffic *= 8 => bits
1479 * traffic *= (10^6 / delta) => bps
1480 * traffic /= 10^6 => Mbps
1482 * Combining, traffic *= (8 / delta)
1486 traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1488 for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1489 if (traffic < mod_table[index].rx_rate)
1491 range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1492 pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1493 rx_coal->small_pkt_range_start :
1494 rx_coal->large_pkt_range_start;
1495 timer = range_start + ((rx_coal->range_end - range_start) *
1496 mod_table[index].range_percent / 100);
1498 cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1500 pkt_size_counter->large_pkt_bytes_cnt = 0;
1501 pkt_size_counter->small_pkt_bytes_cnt = 0;
1504 static int enic_poll(struct napi_struct *napi, int budget)
1506 struct net_device *netdev = napi->dev;
1507 struct enic *enic = netdev_priv(netdev);
1508 unsigned int cq_rq = enic_cq_rq(enic, 0);
1509 unsigned int cq_wq = enic_cq_wq(enic, 0);
1510 unsigned int intr = enic_legacy_io_intr();
1511 unsigned int rq_work_to_do = budget;
1512 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET;
1513 unsigned int work_done, rq_work_done = 0, wq_work_done;
1516 wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1517 enic_wq_service, NULL);
1520 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1521 rq_work_to_do, enic_rq_service, NULL);
1523 /* Accumulate intr event credits for this polling
1524 * cycle. An intr event is the completion of a
1525 * a WQ or RQ packet.
1528 work_done = rq_work_done + wq_work_done;
1531 vnic_intr_return_credits(&enic->intr[intr],
1533 0 /* don't unmask intr */,
1534 0 /* don't reset intr timer */);
1536 err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1538 /* Buffer allocation failed. Stay in polling
1539 * mode so we can try to fill the ring again.
1543 rq_work_done = rq_work_to_do;
1544 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1545 /* Call the function which refreshes the intr coalescing timer
1546 * value based on the traffic.
1548 enic_calc_int_moderation(enic, &enic->rq[0]);
1550 if ((rq_work_done < budget) && napi_complete_done(napi, rq_work_done)) {
1552 /* Some work done, but not enough to stay in polling,
1556 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1557 enic_set_int_moderation(enic, &enic->rq[0]);
1558 vnic_intr_unmask(&enic->intr[intr]);
1561 return rq_work_done;
1564 #ifdef CONFIG_RFS_ACCEL
1565 static void enic_free_rx_cpu_rmap(struct enic *enic)
1567 free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1568 enic->netdev->rx_cpu_rmap = NULL;
1571 static void enic_set_rx_cpu_rmap(struct enic *enic)
1575 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1576 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1577 if (unlikely(!enic->netdev->rx_cpu_rmap))
1579 for (i = 0; i < enic->rq_count; i++) {
1580 res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1581 enic->msix_entry[i].vector);
1582 if (unlikely(res)) {
1583 enic_free_rx_cpu_rmap(enic);
1592 static void enic_free_rx_cpu_rmap(struct enic *enic)
1596 static void enic_set_rx_cpu_rmap(struct enic *enic)
1600 #endif /* CONFIG_RFS_ACCEL */
1602 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1604 struct net_device *netdev = napi->dev;
1605 struct enic *enic = netdev_priv(netdev);
1606 unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1607 struct vnic_wq *wq = &enic->wq[wq_index];
1610 unsigned int wq_work_to_do = ENIC_WQ_NAPI_BUDGET;
1611 unsigned int wq_work_done;
1612 unsigned int wq_irq;
1615 cq = enic_cq_wq(enic, wq_irq);
1616 intr = enic_msix_wq_intr(enic, wq_irq);
1617 wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1618 enic_wq_service, NULL);
1620 vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1621 0 /* don't unmask intr */,
1622 1 /* reset intr timer */);
1623 if (!wq_work_done) {
1624 napi_complete(napi);
1625 vnic_intr_unmask(&enic->intr[intr]);
1632 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1634 struct net_device *netdev = napi->dev;
1635 struct enic *enic = netdev_priv(netdev);
1636 unsigned int rq = (napi - &enic->napi[0]);
1637 unsigned int cq = enic_cq_rq(enic, rq);
1638 unsigned int intr = enic_msix_rq_intr(enic, rq);
1639 unsigned int work_to_do = budget;
1640 unsigned int work_done = 0;
1647 work_done = vnic_cq_service(&enic->cq[cq],
1648 work_to_do, enic_rq_service, NULL);
1650 /* Return intr event credits for this polling
1651 * cycle. An intr event is the completion of a
1656 vnic_intr_return_credits(&enic->intr[intr],
1658 0 /* don't unmask intr */,
1659 0 /* don't reset intr timer */);
1661 err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1663 /* Buffer allocation failed. Stay in polling mode
1664 * so we can try to fill the ring again.
1668 work_done = work_to_do;
1669 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1670 /* Call the function which refreshes the intr coalescing timer
1671 * value based on the traffic.
1673 enic_calc_int_moderation(enic, &enic->rq[rq]);
1675 if ((work_done < budget) && napi_complete_done(napi, work_done)) {
1677 /* Some work done, but not enough to stay in polling,
1681 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1682 enic_set_int_moderation(enic, &enic->rq[rq]);
1683 vnic_intr_unmask(&enic->intr[intr]);
1689 static void enic_notify_timer(struct timer_list *t)
1691 struct enic *enic = from_timer(enic, t, notify_timer);
1693 enic_notify_check(enic);
1695 mod_timer(&enic->notify_timer,
1696 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1699 static void enic_free_intr(struct enic *enic)
1701 struct net_device *netdev = enic->netdev;
1704 enic_free_rx_cpu_rmap(enic);
1705 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1706 case VNIC_DEV_INTR_MODE_INTX:
1707 free_irq(enic->pdev->irq, netdev);
1709 case VNIC_DEV_INTR_MODE_MSI:
1710 free_irq(enic->pdev->irq, enic);
1712 case VNIC_DEV_INTR_MODE_MSIX:
1713 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1714 if (enic->msix[i].requested)
1715 free_irq(enic->msix_entry[i].vector,
1716 enic->msix[i].devid);
1723 static int enic_request_intr(struct enic *enic)
1725 struct net_device *netdev = enic->netdev;
1726 unsigned int i, intr;
1729 enic_set_rx_cpu_rmap(enic);
1730 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1732 case VNIC_DEV_INTR_MODE_INTX:
1734 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1735 IRQF_SHARED, netdev->name, netdev);
1738 case VNIC_DEV_INTR_MODE_MSI:
1740 err = request_irq(enic->pdev->irq, enic_isr_msi,
1741 0, netdev->name, enic);
1744 case VNIC_DEV_INTR_MODE_MSIX:
1746 for (i = 0; i < enic->rq_count; i++) {
1747 intr = enic_msix_rq_intr(enic, i);
1748 snprintf(enic->msix[intr].devname,
1749 sizeof(enic->msix[intr].devname),
1750 "%s-rx-%u", netdev->name, i);
1751 enic->msix[intr].isr = enic_isr_msix;
1752 enic->msix[intr].devid = &enic->napi[i];
1755 for (i = 0; i < enic->wq_count; i++) {
1756 int wq = enic_cq_wq(enic, i);
1758 intr = enic_msix_wq_intr(enic, i);
1759 snprintf(enic->msix[intr].devname,
1760 sizeof(enic->msix[intr].devname),
1761 "%s-tx-%u", netdev->name, i);
1762 enic->msix[intr].isr = enic_isr_msix;
1763 enic->msix[intr].devid = &enic->napi[wq];
1766 intr = enic_msix_err_intr(enic);
1767 snprintf(enic->msix[intr].devname,
1768 sizeof(enic->msix[intr].devname),
1769 "%s-err", netdev->name);
1770 enic->msix[intr].isr = enic_isr_msix_err;
1771 enic->msix[intr].devid = enic;
1773 intr = enic_msix_notify_intr(enic);
1774 snprintf(enic->msix[intr].devname,
1775 sizeof(enic->msix[intr].devname),
1776 "%s-notify", netdev->name);
1777 enic->msix[intr].isr = enic_isr_msix_notify;
1778 enic->msix[intr].devid = enic;
1780 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1781 enic->msix[i].requested = 0;
1783 for (i = 0; i < enic->intr_count; i++) {
1784 err = request_irq(enic->msix_entry[i].vector,
1785 enic->msix[i].isr, 0,
1786 enic->msix[i].devname,
1787 enic->msix[i].devid);
1789 enic_free_intr(enic);
1792 enic->msix[i].requested = 1;
1804 static void enic_synchronize_irqs(struct enic *enic)
1808 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1809 case VNIC_DEV_INTR_MODE_INTX:
1810 case VNIC_DEV_INTR_MODE_MSI:
1811 synchronize_irq(enic->pdev->irq);
1813 case VNIC_DEV_INTR_MODE_MSIX:
1814 for (i = 0; i < enic->intr_count; i++)
1815 synchronize_irq(enic->msix_entry[i].vector);
1822 static void enic_set_rx_coal_setting(struct enic *enic)
1826 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1828 /* 1. Read the link speed from fw
1829 * 2. Pick the default range for the speed
1830 * 3. Update it in enic->rx_coalesce_setting
1832 speed = vnic_dev_port_speed(enic->vdev);
1833 if (ENIC_LINK_SPEED_10G < speed)
1834 index = ENIC_LINK_40G_INDEX;
1835 else if (ENIC_LINK_SPEED_4G < speed)
1836 index = ENIC_LINK_10G_INDEX;
1838 index = ENIC_LINK_4G_INDEX;
1840 rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1841 rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1842 rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1844 /* Start with the value provided by UCSM */
1845 for (index = 0; index < enic->rq_count; index++)
1846 enic->cq[index].cur_rx_coal_timeval =
1847 enic->config.intr_timer_usec;
1849 rx_coal->use_adaptive_rx_coalesce = 1;
1852 static int enic_dev_notify_set(struct enic *enic)
1856 spin_lock_bh(&enic->devcmd_lock);
1857 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1858 case VNIC_DEV_INTR_MODE_INTX:
1859 err = vnic_dev_notify_set(enic->vdev,
1860 enic_legacy_notify_intr());
1862 case VNIC_DEV_INTR_MODE_MSIX:
1863 err = vnic_dev_notify_set(enic->vdev,
1864 enic_msix_notify_intr(enic));
1867 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1870 spin_unlock_bh(&enic->devcmd_lock);
1875 static void enic_notify_timer_start(struct enic *enic)
1877 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1878 case VNIC_DEV_INTR_MODE_MSI:
1879 mod_timer(&enic->notify_timer, jiffies);
1882 /* Using intr for notification for INTx/MSI-X */
1887 /* rtnl lock is held, process context */
1888 static int enic_open(struct net_device *netdev)
1890 struct enic *enic = netdev_priv(netdev);
1894 err = enic_request_intr(enic);
1896 netdev_err(netdev, "Unable to request irq.\n");
1899 enic_init_affinity_hint(enic);
1900 enic_set_affinity_hint(enic);
1902 err = enic_dev_notify_set(enic);
1905 "Failed to alloc notify buffer, aborting.\n");
1906 goto err_out_free_intr;
1909 for (i = 0; i < enic->rq_count; i++) {
1910 /* enable rq before updating rq desc */
1911 vnic_rq_enable(&enic->rq[i]);
1912 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1913 /* Need at least one buffer on ring to get going */
1914 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1915 netdev_err(netdev, "Unable to alloc receive buffers\n");
1917 goto err_out_free_rq;
1921 for (i = 0; i < enic->wq_count; i++)
1922 vnic_wq_enable(&enic->wq[i]);
1924 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1925 enic_dev_add_station_addr(enic);
1927 enic_set_rx_mode(netdev);
1929 netif_tx_wake_all_queues(netdev);
1931 for (i = 0; i < enic->rq_count; i++)
1932 napi_enable(&enic->napi[i]);
1934 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1935 for (i = 0; i < enic->wq_count; i++)
1936 napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
1937 enic_dev_enable(enic);
1939 for (i = 0; i < enic->intr_count; i++)
1940 vnic_intr_unmask(&enic->intr[i]);
1942 enic_notify_timer_start(enic);
1943 enic_rfs_timer_start(enic);
1948 for (i = 0; i < enic->rq_count; i++) {
1949 ret = vnic_rq_disable(&enic->rq[i]);
1951 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1953 enic_dev_notify_unset(enic);
1955 enic_unset_affinity_hint(enic);
1956 enic_free_intr(enic);
1961 /* rtnl lock is held, process context */
1962 static int enic_stop(struct net_device *netdev)
1964 struct enic *enic = netdev_priv(netdev);
1968 for (i = 0; i < enic->intr_count; i++) {
1969 vnic_intr_mask(&enic->intr[i]);
1970 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1973 enic_synchronize_irqs(enic);
1975 del_timer_sync(&enic->notify_timer);
1976 enic_rfs_flw_tbl_free(enic);
1978 enic_dev_disable(enic);
1980 for (i = 0; i < enic->rq_count; i++)
1981 napi_disable(&enic->napi[i]);
1983 netif_carrier_off(netdev);
1984 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1985 for (i = 0; i < enic->wq_count; i++)
1986 napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
1987 netif_tx_disable(netdev);
1989 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1990 enic_dev_del_station_addr(enic);
1992 for (i = 0; i < enic->wq_count; i++) {
1993 err = vnic_wq_disable(&enic->wq[i]);
1997 for (i = 0; i < enic->rq_count; i++) {
1998 err = vnic_rq_disable(&enic->rq[i]);
2003 enic_dev_notify_unset(enic);
2004 enic_unset_affinity_hint(enic);
2005 enic_free_intr(enic);
2007 for (i = 0; i < enic->wq_count; i++)
2008 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
2009 for (i = 0; i < enic->rq_count; i++)
2010 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
2011 for (i = 0; i < enic->cq_count; i++)
2012 vnic_cq_clean(&enic->cq[i]);
2013 for (i = 0; i < enic->intr_count; i++)
2014 vnic_intr_clean(&enic->intr[i]);
2019 static int _enic_change_mtu(struct net_device *netdev, int new_mtu)
2021 bool running = netif_running(netdev);
2026 err = enic_stop(netdev);
2031 netdev->mtu = new_mtu;
2034 err = enic_open(netdev);
2042 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
2044 struct enic *enic = netdev_priv(netdev);
2046 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2049 if (netdev->mtu > enic->port_mtu)
2051 "interface MTU (%d) set higher than port MTU (%d)\n",
2052 netdev->mtu, enic->port_mtu);
2054 return _enic_change_mtu(netdev, new_mtu);
2057 static void enic_change_mtu_work(struct work_struct *work)
2059 struct enic *enic = container_of(work, struct enic, change_mtu_work);
2060 struct net_device *netdev = enic->netdev;
2061 int new_mtu = vnic_dev_mtu(enic->vdev);
2064 (void)_enic_change_mtu(netdev, new_mtu);
2067 netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
2070 #ifdef CONFIG_NET_POLL_CONTROLLER
2071 static void enic_poll_controller(struct net_device *netdev)
2073 struct enic *enic = netdev_priv(netdev);
2074 struct vnic_dev *vdev = enic->vdev;
2075 unsigned int i, intr;
2077 switch (vnic_dev_get_intr_mode(vdev)) {
2078 case VNIC_DEV_INTR_MODE_MSIX:
2079 for (i = 0; i < enic->rq_count; i++) {
2080 intr = enic_msix_rq_intr(enic, i);
2081 enic_isr_msix(enic->msix_entry[intr].vector,
2085 for (i = 0; i < enic->wq_count; i++) {
2086 intr = enic_msix_wq_intr(enic, i);
2087 enic_isr_msix(enic->msix_entry[intr].vector,
2088 &enic->napi[enic_cq_wq(enic, i)]);
2092 case VNIC_DEV_INTR_MODE_MSI:
2093 enic_isr_msi(enic->pdev->irq, enic);
2095 case VNIC_DEV_INTR_MODE_INTX:
2096 enic_isr_legacy(enic->pdev->irq, netdev);
2104 static int enic_dev_wait(struct vnic_dev *vdev,
2105 int (*start)(struct vnic_dev *, int),
2106 int (*finished)(struct vnic_dev *, int *),
2113 err = start(vdev, arg);
2117 /* Wait for func to complete...2 seconds max
2120 time = jiffies + (HZ * 2);
2123 err = finished(vdev, &done);
2130 schedule_timeout_uninterruptible(HZ / 10);
2132 } while (time_after(time, jiffies));
2137 static int enic_dev_open(struct enic *enic)
2140 u32 flags = CMD_OPENF_IG_DESCCACHE;
2142 err = enic_dev_wait(enic->vdev, vnic_dev_open,
2143 vnic_dev_open_done, flags);
2145 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
2151 static int enic_dev_soft_reset(struct enic *enic)
2155 err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
2156 vnic_dev_soft_reset_done, 0);
2158 netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n",
2164 static int enic_dev_hang_reset(struct enic *enic)
2168 err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
2169 vnic_dev_hang_reset_done, 0);
2171 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
2177 int __enic_set_rsskey(struct enic *enic)
2179 union vnic_rss_key *rss_key_buf_va;
2180 dma_addr_t rss_key_buf_pa;
2181 int i, kidx, bidx, err;
2183 rss_key_buf_va = dma_alloc_coherent(&enic->pdev->dev,
2184 sizeof(union vnic_rss_key),
2185 &rss_key_buf_pa, GFP_ATOMIC);
2186 if (!rss_key_buf_va)
2189 for (i = 0; i < ENIC_RSS_LEN; i++) {
2190 kidx = i / ENIC_RSS_BYTES_PER_KEY;
2191 bidx = i % ENIC_RSS_BYTES_PER_KEY;
2192 rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i];
2194 spin_lock_bh(&enic->devcmd_lock);
2195 err = enic_set_rss_key(enic,
2197 sizeof(union vnic_rss_key));
2198 spin_unlock_bh(&enic->devcmd_lock);
2200 dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_key),
2201 rss_key_buf_va, rss_key_buf_pa);
2206 static int enic_set_rsskey(struct enic *enic)
2208 netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN);
2210 return __enic_set_rsskey(enic);
2213 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
2215 dma_addr_t rss_cpu_buf_pa;
2216 union vnic_rss_cpu *rss_cpu_buf_va = NULL;
2220 rss_cpu_buf_va = dma_alloc_coherent(&enic->pdev->dev,
2221 sizeof(union vnic_rss_cpu),
2222 &rss_cpu_buf_pa, GFP_ATOMIC);
2223 if (!rss_cpu_buf_va)
2226 for (i = 0; i < (1 << rss_hash_bits); i++)
2227 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
2229 spin_lock_bh(&enic->devcmd_lock);
2230 err = enic_set_rss_cpu(enic,
2232 sizeof(union vnic_rss_cpu));
2233 spin_unlock_bh(&enic->devcmd_lock);
2235 dma_free_coherent(&enic->pdev->dev, sizeof(union vnic_rss_cpu),
2236 rss_cpu_buf_va, rss_cpu_buf_pa);
2241 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
2242 u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
2244 const u8 tso_ipid_split_en = 0;
2245 const u8 ig_vlan_strip_en = 1;
2248 /* Enable VLAN tag stripping.
2251 spin_lock_bh(&enic->devcmd_lock);
2252 err = enic_set_nic_cfg(enic,
2253 rss_default_cpu, rss_hash_type,
2254 rss_hash_bits, rss_base_cpu,
2255 rss_enable, tso_ipid_split_en,
2257 spin_unlock_bh(&enic->devcmd_lock);
2262 static int enic_set_rss_nic_cfg(struct enic *enic)
2264 struct device *dev = enic_get_dev(enic);
2265 const u8 rss_default_cpu = 0;
2266 const u8 rss_hash_bits = 7;
2267 const u8 rss_base_cpu = 0;
2270 u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2272 spin_lock_bh(&enic->devcmd_lock);
2273 res = vnic_dev_capable_rss_hash_type(enic->vdev, &rss_hash_type);
2274 spin_unlock_bh(&enic->devcmd_lock);
2276 /* defaults for old adapters
2278 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
2279 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
2280 NIC_CFG_RSS_HASH_TYPE_IPV6 |
2281 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2285 if (!enic_set_rsskey(enic)) {
2286 if (enic_set_rsscpu(enic, rss_hash_bits)) {
2288 dev_warn(dev, "RSS disabled, "
2289 "Failed to set RSS cpu indirection table.");
2293 dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2297 return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2298 rss_hash_bits, rss_base_cpu, rss_enable);
2301 static void enic_set_api_busy(struct enic *enic, bool busy)
2303 spin_lock(&enic->enic_api_lock);
2304 enic->enic_api_busy = busy;
2305 spin_unlock(&enic->enic_api_lock);
2308 static void enic_reset(struct work_struct *work)
2310 struct enic *enic = container_of(work, struct enic, reset);
2312 if (!netif_running(enic->netdev))
2317 /* Stop any activity from infiniband */
2318 enic_set_api_busy(enic, true);
2320 enic_stop(enic->netdev);
2321 enic_dev_soft_reset(enic);
2322 enic_reset_addr_lists(enic);
2323 enic_init_vnic_resources(enic);
2324 enic_set_rss_nic_cfg(enic);
2325 enic_dev_set_ig_vlan_rewrite_mode(enic);
2326 enic_open(enic->netdev);
2328 /* Allow infiniband to fiddle with the device again */
2329 enic_set_api_busy(enic, false);
2331 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2336 static void enic_tx_hang_reset(struct work_struct *work)
2338 struct enic *enic = container_of(work, struct enic, tx_hang_reset);
2342 /* Stop any activity from infiniband */
2343 enic_set_api_busy(enic, true);
2345 enic_dev_hang_notify(enic);
2346 enic_stop(enic->netdev);
2347 enic_dev_hang_reset(enic);
2348 enic_reset_addr_lists(enic);
2349 enic_init_vnic_resources(enic);
2350 enic_set_rss_nic_cfg(enic);
2351 enic_dev_set_ig_vlan_rewrite_mode(enic);
2352 enic_open(enic->netdev);
2354 /* Allow infiniband to fiddle with the device again */
2355 enic_set_api_busy(enic, false);
2357 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2362 static int enic_set_intr_mode(struct enic *enic)
2364 unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2365 unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
2368 /* Set interrupt mode (INTx, MSI, MSI-X) depending
2369 * on system capabilities.
2373 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2374 * (the second to last INTR is used for WQ/RQ errors)
2375 * (the last INTR is used for notifications)
2378 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2379 for (i = 0; i < n + m + 2; i++)
2380 enic->msix_entry[i].entry = i;
2382 /* Use multiple RQs if RSS is enabled
2385 if (ENIC_SETTING(enic, RSS) &&
2386 enic->config.intr_mode < 1 &&
2387 enic->rq_count >= n &&
2388 enic->wq_count >= m &&
2389 enic->cq_count >= n + m &&
2390 enic->intr_count >= n + m + 2) {
2392 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2393 n + m + 2, n + m + 2) > 0) {
2397 enic->cq_count = n + m;
2398 enic->intr_count = n + m + 2;
2400 vnic_dev_set_intr_mode(enic->vdev,
2401 VNIC_DEV_INTR_MODE_MSIX);
2407 if (enic->config.intr_mode < 1 &&
2408 enic->rq_count >= 1 &&
2409 enic->wq_count >= m &&
2410 enic->cq_count >= 1 + m &&
2411 enic->intr_count >= 1 + m + 2) {
2412 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2413 1 + m + 2, 1 + m + 2) > 0) {
2417 enic->cq_count = 1 + m;
2418 enic->intr_count = 1 + m + 2;
2420 vnic_dev_set_intr_mode(enic->vdev,
2421 VNIC_DEV_INTR_MODE_MSIX);
2429 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2432 if (enic->config.intr_mode < 2 &&
2433 enic->rq_count >= 1 &&
2434 enic->wq_count >= 1 &&
2435 enic->cq_count >= 2 &&
2436 enic->intr_count >= 1 &&
2437 !pci_enable_msi(enic->pdev)) {
2442 enic->intr_count = 1;
2444 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2451 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2452 * (the first INTR is used for WQ/RQ)
2453 * (the second INTR is used for WQ/RQ errors)
2454 * (the last INTR is used for notifications)
2457 if (enic->config.intr_mode < 3 &&
2458 enic->rq_count >= 1 &&
2459 enic->wq_count >= 1 &&
2460 enic->cq_count >= 2 &&
2461 enic->intr_count >= 3) {
2466 enic->intr_count = 3;
2468 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2473 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2478 static void enic_clear_intr_mode(struct enic *enic)
2480 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2481 case VNIC_DEV_INTR_MODE_MSIX:
2482 pci_disable_msix(enic->pdev);
2484 case VNIC_DEV_INTR_MODE_MSI:
2485 pci_disable_msi(enic->pdev);
2491 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2494 static const struct net_device_ops enic_netdev_dynamic_ops = {
2495 .ndo_open = enic_open,
2496 .ndo_stop = enic_stop,
2497 .ndo_start_xmit = enic_hard_start_xmit,
2498 .ndo_get_stats64 = enic_get_stats,
2499 .ndo_validate_addr = eth_validate_addr,
2500 .ndo_set_rx_mode = enic_set_rx_mode,
2501 .ndo_set_mac_address = enic_set_mac_address_dynamic,
2502 .ndo_change_mtu = enic_change_mtu,
2503 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2504 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2505 .ndo_tx_timeout = enic_tx_timeout,
2506 .ndo_set_vf_port = enic_set_vf_port,
2507 .ndo_get_vf_port = enic_get_vf_port,
2508 .ndo_set_vf_mac = enic_set_vf_mac,
2509 #ifdef CONFIG_NET_POLL_CONTROLLER
2510 .ndo_poll_controller = enic_poll_controller,
2512 #ifdef CONFIG_RFS_ACCEL
2513 .ndo_rx_flow_steer = enic_rx_flow_steer,
2515 .ndo_features_check = enic_features_check,
2518 static const struct net_device_ops enic_netdev_ops = {
2519 .ndo_open = enic_open,
2520 .ndo_stop = enic_stop,
2521 .ndo_start_xmit = enic_hard_start_xmit,
2522 .ndo_get_stats64 = enic_get_stats,
2523 .ndo_validate_addr = eth_validate_addr,
2524 .ndo_set_mac_address = enic_set_mac_address,
2525 .ndo_set_rx_mode = enic_set_rx_mode,
2526 .ndo_change_mtu = enic_change_mtu,
2527 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2528 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2529 .ndo_tx_timeout = enic_tx_timeout,
2530 .ndo_set_vf_port = enic_set_vf_port,
2531 .ndo_get_vf_port = enic_get_vf_port,
2532 .ndo_set_vf_mac = enic_set_vf_mac,
2533 #ifdef CONFIG_NET_POLL_CONTROLLER
2534 .ndo_poll_controller = enic_poll_controller,
2536 #ifdef CONFIG_RFS_ACCEL
2537 .ndo_rx_flow_steer = enic_rx_flow_steer,
2539 .ndo_features_check = enic_features_check,
2542 static void enic_dev_deinit(struct enic *enic)
2546 for (i = 0; i < enic->rq_count; i++)
2547 __netif_napi_del(&enic->napi[i]);
2549 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2550 for (i = 0; i < enic->wq_count; i++)
2551 __netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2553 /* observe RCU grace period after __netif_napi_del() calls */
2556 enic_free_vnic_resources(enic);
2557 enic_clear_intr_mode(enic);
2558 enic_free_affinity_hint(enic);
2561 static void enic_kdump_kernel_config(struct enic *enic)
2563 if (is_kdump_kernel()) {
2564 dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n");
2567 enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS;
2568 enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS;
2569 enic->config.mtu = min_t(u16, 1500, enic->config.mtu);
2573 static int enic_dev_init(struct enic *enic)
2575 struct device *dev = enic_get_dev(enic);
2576 struct net_device *netdev = enic->netdev;
2580 /* Get interrupt coalesce timer info */
2581 err = enic_dev_intr_coal_timer_info(enic);
2583 dev_warn(dev, "Using default conversion factor for "
2584 "interrupt coalesce timer\n");
2585 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2588 /* Get vNIC configuration
2591 err = enic_get_vnic_config(enic);
2593 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2597 /* Get available resource counts
2600 enic_get_res_counts(enic);
2602 /* modify resource count if we are in kdump_kernel
2604 enic_kdump_kernel_config(enic);
2606 /* Set interrupt mode based on resource counts and system
2610 err = enic_set_intr_mode(enic);
2612 dev_err(dev, "Failed to set intr mode based on resource "
2613 "counts and system capabilities, aborting\n");
2617 /* Allocate and configure vNIC resources
2620 err = enic_alloc_vnic_resources(enic);
2622 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2623 goto err_out_free_vnic_resources;
2626 enic_init_vnic_resources(enic);
2628 err = enic_set_rss_nic_cfg(enic);
2630 dev_err(dev, "Failed to config nic, aborting\n");
2631 goto err_out_free_vnic_resources;
2634 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2636 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2638 case VNIC_DEV_INTR_MODE_MSIX:
2639 for (i = 0; i < enic->rq_count; i++) {
2640 netif_napi_add(netdev, &enic->napi[i],
2641 enic_poll_msix_rq, NAPI_POLL_WEIGHT);
2643 for (i = 0; i < enic->wq_count; i++)
2644 netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],
2645 enic_poll_msix_wq, NAPI_POLL_WEIGHT);
2651 err_out_free_vnic_resources:
2652 enic_free_affinity_hint(enic);
2653 enic_clear_intr_mode(enic);
2654 enic_free_vnic_resources(enic);
2659 static void enic_iounmap(struct enic *enic)
2663 for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2664 if (enic->bar[i].vaddr)
2665 iounmap(enic->bar[i].vaddr);
2668 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2670 struct device *dev = &pdev->dev;
2671 struct net_device *netdev;
2676 #ifdef CONFIG_PCI_IOV
2681 /* Allocate net device structure and initialize. Private
2682 * instance data is initialized to zero.
2685 netdev = alloc_etherdev_mqs(sizeof(struct enic),
2686 ENIC_RQ_MAX, ENIC_WQ_MAX);
2690 pci_set_drvdata(pdev, netdev);
2692 SET_NETDEV_DEV(netdev, &pdev->dev);
2694 enic = netdev_priv(netdev);
2695 enic->netdev = netdev;
2698 /* Setup PCI resources
2701 err = pci_enable_device_mem(pdev);
2703 dev_err(dev, "Cannot enable PCI device, aborting\n");
2704 goto err_out_free_netdev;
2707 err = pci_request_regions(pdev, DRV_NAME);
2709 dev_err(dev, "Cannot request PCI regions, aborting\n");
2710 goto err_out_disable_device;
2713 pci_set_master(pdev);
2715 /* Query PCI controller on system for DMA addressing
2716 * limitation for the device. Try 47-bit first, and
2720 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(47));
2722 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2724 dev_err(dev, "No usable DMA configuration, aborting\n");
2725 goto err_out_release_regions;
2727 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
2729 dev_err(dev, "Unable to obtain %u-bit DMA "
2730 "for consistent allocations, aborting\n", 32);
2731 goto err_out_release_regions;
2734 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(47));
2736 dev_err(dev, "Unable to obtain %u-bit DMA "
2737 "for consistent allocations, aborting\n", 47);
2738 goto err_out_release_regions;
2743 /* Map vNIC resources from BAR0-5
2746 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2747 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
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);
2754 goto err_out_iounmap;
2756 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2759 /* Register vNIC device
2762 enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2763 ARRAY_SIZE(enic->bar));
2765 dev_err(dev, "vNIC registration failed, aborting\n");
2767 goto err_out_iounmap;
2770 err = vnic_devcmd_init(enic->vdev);
2773 goto err_out_vnic_unregister;
2775 #ifdef CONFIG_PCI_IOV
2776 /* Get number of subvnics */
2777 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2779 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2781 if (enic->num_vfs) {
2782 err = pci_enable_sriov(pdev, enic->num_vfs);
2784 dev_err(dev, "SRIOV enable failed, aborting."
2785 " pci_enable_sriov() returned %d\n",
2787 goto err_out_vnic_unregister;
2789 enic->priv_flags |= ENIC_SRIOV_ENABLED;
2790 num_pps = enic->num_vfs;
2795 /* Allocate structure for port profiles */
2796 enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2799 goto err_out_disable_sriov_pp;
2802 /* Issue device open to get device in known state
2805 err = enic_dev_open(enic);
2807 dev_err(dev, "vNIC dev open failed, aborting\n");
2808 goto err_out_disable_sriov;
2811 /* Setup devcmd lock
2814 spin_lock_init(&enic->devcmd_lock);
2815 spin_lock_init(&enic->enic_api_lock);
2818 * Set ingress vlan rewrite mode before vnic initialization
2821 err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2824 "Failed to set ingress vlan rewrite mode, aborting.\n");
2825 goto err_out_dev_close;
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.
2836 netif_carrier_off(netdev);
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.
2843 if (!enic_is_dynamic(enic)) {
2844 err = vnic_dev_init(enic->vdev, 0);
2846 dev_err(dev, "vNIC dev init failed, aborting\n");
2847 goto err_out_dev_close;
2851 err = enic_dev_init(enic);
2853 dev_err(dev, "Device initialization failed, aborting\n");
2854 goto err_out_dev_close;
2857 netif_set_real_num_tx_queues(netdev, enic->wq_count);
2858 netif_set_real_num_rx_queues(netdev, enic->rq_count);
2860 /* Setup notification timer, HW reset task, and wq locks
2863 timer_setup(&enic->notify_timer, enic_notify_timer, 0);
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);
2871 for (i = 0; i < enic->wq_count; i++)
2872 spin_lock_init(&enic->wq_lock[i]);
2874 /* Register net device
2877 enic->port_mtu = enic->config.mtu;
2879 err = enic_set_mac_addr(netdev, enic->mac_addr);
2881 dev_err(dev, "Invalid MAC address, aborting\n");
2882 goto err_out_dev_deinit;
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
2889 enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2891 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2892 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2894 netdev->netdev_ops = &enic_netdev_ops;
2896 netdev->watchdog_timeo = 2 * HZ;
2897 enic_set_ethtool_ops(netdev);
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);
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)) {
2919 netdev->hw_enc_features |= NETIF_F_RXCSUM |
2923 NETIF_F_GSO_UDP_TUNNEL |
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
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/
2936 * used in enic_rq_indicate_buf
2938 err = vnic_dev_get_supported_feature_ver(enic->vdev,
2943 enic->vxlan.flags = (u8)a1;
2944 /* mask bits that are supported by driver
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;
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;
2959 netdev->features |= netdev->hw_features;
2960 netdev->vlan_features |= netdev->features;
2962 #ifdef CONFIG_RFS_ACCEL
2963 netdev->hw_features |= NETIF_F_NTUPLE;
2967 netdev->features |= NETIF_F_HIGHDMA;
2969 netdev->priv_flags |= IFF_UNICAST_FLT;
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;
2976 err = register_netdev(netdev);
2978 dev_err(dev, "Cannot register net device, aborting\n");
2979 goto err_out_dev_deinit;
2981 enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
2986 enic_dev_deinit(enic);
2988 vnic_dev_close(enic->vdev);
2989 err_out_disable_sriov:
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;
2998 err_out_vnic_unregister:
2999 vnic_dev_unregister(enic->vdev);
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);
3012 static void enic_remove(struct pci_dev *pdev)
3014 struct net_device *netdev = pci_get_drvdata(pdev);
3017 struct enic *enic = netdev_priv(netdev);
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;
3031 vnic_dev_unregister(enic->vdev);
3033 pci_release_regions(pdev);
3034 pci_disable_device(pdev);
3035 free_netdev(netdev);
3039 static struct pci_driver enic_driver = {
3041 .id_table = enic_id_table,
3042 .probe = enic_probe,
3043 .remove = enic_remove,
3046 module_pci_driver(enic_driver);