Merge tag 'chrome-platform-for-linus-4.13' of git://git.kernel.org/pub/scm/linux...
[linux/fpc-iii.git] / drivers / net / ethernet / cisco / enic / enic_main.c
blobd24ee1ad3be1dadb36022b721a81effeb7cf6bb4
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>
50 #include "cq_enet_desc.h"
51 #include "vnic_dev.h"
52 #include "vnic_intr.h"
53 #include "vnic_stats.h"
54 #include "vnic_vic.h"
55 #include "enic_res.h"
56 #include "enic.h"
57 #include "enic_dev.h"
58 #include "enic_pp.h"
59 #include "enic_clsf.h"
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_VERSION(DRV_VERSION);
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 (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 !enic->msix[i].affinity_mask ||
152 cpumask_empty(enic->msix[i].affinity_mask))
153 continue;
154 err = irq_set_affinity_hint(enic->msix_entry[i].vector,
155 enic->msix[i].affinity_mask);
156 if (err)
157 netdev_warn(enic->netdev, "irq_set_affinity_hint failed, err %d\n",
158 err);
161 for (i = 0; i < enic->wq_count; i++) {
162 int wq_intr = enic_msix_wq_intr(enic, i);
164 if (enic->msix[wq_intr].affinity_mask &&
165 !cpumask_empty(enic->msix[wq_intr].affinity_mask))
166 netif_set_xps_queue(enic->netdev,
167 enic->msix[wq_intr].affinity_mask,
172 static void enic_unset_affinity_hint(struct enic *enic)
174 int i;
176 for (i = 0; i < enic->intr_count; i++)
177 irq_set_affinity_hint(enic->msix_entry[i].vector, NULL);
180 static void enic_udp_tunnel_add(struct net_device *netdev,
181 struct udp_tunnel_info *ti)
183 struct enic *enic = netdev_priv(netdev);
184 __be16 port = ti->port;
185 int err;
187 spin_lock_bh(&enic->devcmd_lock);
189 if (ti->type != UDP_TUNNEL_TYPE_VXLAN) {
190 netdev_info(netdev, "udp_tnl: only vxlan tunnel offload supported");
191 goto error;
194 if (ti->sa_family != AF_INET) {
195 netdev_info(netdev, "vxlan: only IPv4 offload supported");
196 goto error;
199 if (enic->vxlan.vxlan_udp_port_number) {
200 if (ntohs(port) == enic->vxlan.vxlan_udp_port_number)
201 netdev_warn(netdev, "vxlan: udp port already offloaded");
202 else
203 netdev_info(netdev, "vxlan: offload supported for only one UDP port");
205 goto error;
208 err = vnic_dev_overlay_offload_cfg(enic->vdev,
209 OVERLAY_CFG_VXLAN_PORT_UPDATE,
210 ntohs(port));
211 if (err)
212 goto error;
214 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
215 enic->vxlan.patch_level);
216 if (err)
217 goto error;
219 enic->vxlan.vxlan_udp_port_number = ntohs(port);
221 netdev_info(netdev, "vxlan fw-vers-%d: offload enabled for udp port: %d, sa_family: %d ",
222 (int)enic->vxlan.patch_level, ntohs(port), ti->sa_family);
224 goto unlock;
226 error:
227 netdev_info(netdev, "failed to offload udp port: %d, sa_family: %d, type: %d",
228 ntohs(port), ti->sa_family, ti->type);
229 unlock:
230 spin_unlock_bh(&enic->devcmd_lock);
233 static void enic_udp_tunnel_del(struct net_device *netdev,
234 struct udp_tunnel_info *ti)
236 struct enic *enic = netdev_priv(netdev);
237 int err;
239 spin_lock_bh(&enic->devcmd_lock);
241 if ((ti->sa_family != AF_INET) ||
242 ((ntohs(ti->port) != enic->vxlan.vxlan_udp_port_number)) ||
243 (ti->type != UDP_TUNNEL_TYPE_VXLAN)) {
244 netdev_info(netdev, "udp_tnl: port:%d, sa_family: %d, type: %d not offloaded",
245 ntohs(ti->port), ti->sa_family, ti->type);
246 goto unlock;
249 err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
250 OVERLAY_OFFLOAD_DISABLE);
251 if (err) {
252 netdev_err(netdev, "vxlan: del offload udp port: %d failed",
253 ntohs(ti->port));
254 goto unlock;
257 enic->vxlan.vxlan_udp_port_number = 0;
259 netdev_info(netdev, "vxlan: del offload udp port %d, family %d\n",
260 ntohs(ti->port), ti->sa_family);
262 unlock:
263 spin_unlock_bh(&enic->devcmd_lock);
266 static netdev_features_t enic_features_check(struct sk_buff *skb,
267 struct net_device *dev,
268 netdev_features_t features)
270 const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb);
271 struct enic *enic = netdev_priv(dev);
272 struct udphdr *udph;
273 u16 port = 0;
274 u16 proto;
276 if (!skb->encapsulation)
277 return features;
279 features = vxlan_features_check(skb, features);
281 /* hardware only supports IPv4 vxlan tunnel */
282 if (vlan_get_protocol(skb) != htons(ETH_P_IP))
283 goto out;
285 /* hardware does not support offload of ipv6 inner pkt */
286 if (eth->h_proto != ntohs(ETH_P_IP))
287 goto out;
289 proto = ip_hdr(skb)->protocol;
291 if (proto == IPPROTO_UDP) {
292 udph = udp_hdr(skb);
293 port = be16_to_cpu(udph->dest);
296 /* HW supports offload of only one UDP port. Remove CSUM and GSO MASK
297 * for other UDP port tunnels
299 if (port != enic->vxlan.vxlan_udp_port_number)
300 goto out;
302 return features;
304 out:
305 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
308 int enic_is_dynamic(struct enic *enic)
310 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
313 int enic_sriov_enabled(struct enic *enic)
315 return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
318 static int enic_is_sriov_vf(struct enic *enic)
320 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
323 int enic_is_valid_vf(struct enic *enic, int vf)
325 #ifdef CONFIG_PCI_IOV
326 return vf >= 0 && vf < enic->num_vfs;
327 #else
328 return 0;
329 #endif
332 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
334 struct enic *enic = vnic_dev_priv(wq->vdev);
336 if (buf->sop)
337 pci_unmap_single(enic->pdev, buf->dma_addr,
338 buf->len, PCI_DMA_TODEVICE);
339 else
340 pci_unmap_page(enic->pdev, buf->dma_addr,
341 buf->len, PCI_DMA_TODEVICE);
343 if (buf->os_buf)
344 dev_kfree_skb_any(buf->os_buf);
347 static void enic_wq_free_buf(struct vnic_wq *wq,
348 struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
350 enic_free_wq_buf(wq, buf);
353 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
354 u8 type, u16 q_number, u16 completed_index, void *opaque)
356 struct enic *enic = vnic_dev_priv(vdev);
358 spin_lock(&enic->wq_lock[q_number]);
360 vnic_wq_service(&enic->wq[q_number], cq_desc,
361 completed_index, enic_wq_free_buf,
362 opaque);
364 if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
365 vnic_wq_desc_avail(&enic->wq[q_number]) >=
366 (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
367 netif_wake_subqueue(enic->netdev, q_number);
369 spin_unlock(&enic->wq_lock[q_number]);
371 return 0;
374 static bool enic_log_q_error(struct enic *enic)
376 unsigned int i;
377 u32 error_status;
378 bool err = false;
380 for (i = 0; i < enic->wq_count; i++) {
381 error_status = vnic_wq_error_status(&enic->wq[i]);
382 err |= error_status;
383 if (error_status)
384 netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
385 i, error_status);
388 for (i = 0; i < enic->rq_count; i++) {
389 error_status = vnic_rq_error_status(&enic->rq[i]);
390 err |= error_status;
391 if (error_status)
392 netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
393 i, error_status);
396 return err;
399 static void enic_msglvl_check(struct enic *enic)
401 u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
403 if (msg_enable != enic->msg_enable) {
404 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
405 enic->msg_enable, msg_enable);
406 enic->msg_enable = msg_enable;
410 static void enic_mtu_check(struct enic *enic)
412 u32 mtu = vnic_dev_mtu(enic->vdev);
413 struct net_device *netdev = enic->netdev;
415 if (mtu && mtu != enic->port_mtu) {
416 enic->port_mtu = mtu;
417 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
418 mtu = max_t(int, ENIC_MIN_MTU,
419 min_t(int, ENIC_MAX_MTU, mtu));
420 if (mtu != netdev->mtu)
421 schedule_work(&enic->change_mtu_work);
422 } else {
423 if (mtu < netdev->mtu)
424 netdev_warn(netdev,
425 "interface MTU (%d) set higher "
426 "than switch port MTU (%d)\n",
427 netdev->mtu, mtu);
432 static void enic_link_check(struct enic *enic)
434 int link_status = vnic_dev_link_status(enic->vdev);
435 int carrier_ok = netif_carrier_ok(enic->netdev);
437 if (link_status && !carrier_ok) {
438 netdev_info(enic->netdev, "Link UP\n");
439 netif_carrier_on(enic->netdev);
440 } else if (!link_status && carrier_ok) {
441 netdev_info(enic->netdev, "Link DOWN\n");
442 netif_carrier_off(enic->netdev);
446 static void enic_notify_check(struct enic *enic)
448 enic_msglvl_check(enic);
449 enic_mtu_check(enic);
450 enic_link_check(enic);
453 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
455 static irqreturn_t enic_isr_legacy(int irq, void *data)
457 struct net_device *netdev = data;
458 struct enic *enic = netdev_priv(netdev);
459 unsigned int io_intr = enic_legacy_io_intr();
460 unsigned int err_intr = enic_legacy_err_intr();
461 unsigned int notify_intr = enic_legacy_notify_intr();
462 u32 pba;
464 vnic_intr_mask(&enic->intr[io_intr]);
466 pba = vnic_intr_legacy_pba(enic->legacy_pba);
467 if (!pba) {
468 vnic_intr_unmask(&enic->intr[io_intr]);
469 return IRQ_NONE; /* not our interrupt */
472 if (ENIC_TEST_INTR(pba, notify_intr)) {
473 enic_notify_check(enic);
474 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
477 if (ENIC_TEST_INTR(pba, err_intr)) {
478 vnic_intr_return_all_credits(&enic->intr[err_intr]);
479 enic_log_q_error(enic);
480 /* schedule recovery from WQ/RQ error */
481 schedule_work(&enic->reset);
482 return IRQ_HANDLED;
485 if (ENIC_TEST_INTR(pba, io_intr))
486 napi_schedule_irqoff(&enic->napi[0]);
487 else
488 vnic_intr_unmask(&enic->intr[io_intr]);
490 return IRQ_HANDLED;
493 static irqreturn_t enic_isr_msi(int irq, void *data)
495 struct enic *enic = data;
497 /* With MSI, there is no sharing of interrupts, so this is
498 * our interrupt and there is no need to ack it. The device
499 * is not providing per-vector masking, so the OS will not
500 * write to PCI config space to mask/unmask the interrupt.
501 * We're using mask_on_assertion for MSI, so the device
502 * automatically masks the interrupt when the interrupt is
503 * generated. Later, when exiting polling, the interrupt
504 * will be unmasked (see enic_poll).
506 * Also, the device uses the same PCIe Traffic Class (TC)
507 * for Memory Write data and MSI, so there are no ordering
508 * issues; the MSI will always arrive at the Root Complex
509 * _after_ corresponding Memory Writes (i.e. descriptor
510 * writes).
513 napi_schedule_irqoff(&enic->napi[0]);
515 return IRQ_HANDLED;
518 static irqreturn_t enic_isr_msix(int irq, void *data)
520 struct napi_struct *napi = data;
522 napi_schedule_irqoff(napi);
524 return IRQ_HANDLED;
527 static irqreturn_t enic_isr_msix_err(int irq, void *data)
529 struct enic *enic = data;
530 unsigned int intr = enic_msix_err_intr(enic);
532 vnic_intr_return_all_credits(&enic->intr[intr]);
534 if (enic_log_q_error(enic))
535 /* schedule recovery from WQ/RQ error */
536 schedule_work(&enic->reset);
538 return IRQ_HANDLED;
541 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
543 struct enic *enic = data;
544 unsigned int intr = enic_msix_notify_intr(enic);
546 enic_notify_check(enic);
547 vnic_intr_return_all_credits(&enic->intr[intr]);
549 return IRQ_HANDLED;
552 static int enic_queue_wq_skb_cont(struct enic *enic, struct vnic_wq *wq,
553 struct sk_buff *skb, unsigned int len_left,
554 int loopback)
556 const skb_frag_t *frag;
557 dma_addr_t dma_addr;
559 /* Queue additional data fragments */
560 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
561 len_left -= skb_frag_size(frag);
562 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag, 0,
563 skb_frag_size(frag),
564 DMA_TO_DEVICE);
565 if (unlikely(enic_dma_map_check(enic, dma_addr)))
566 return -ENOMEM;
567 enic_queue_wq_desc_cont(wq, skb, dma_addr, skb_frag_size(frag),
568 (len_left == 0), /* EOP? */
569 loopback);
572 return 0;
575 static int enic_queue_wq_skb_vlan(struct enic *enic, struct vnic_wq *wq,
576 struct sk_buff *skb, int vlan_tag_insert,
577 unsigned int vlan_tag, int loopback)
579 unsigned int head_len = skb_headlen(skb);
580 unsigned int len_left = skb->len - head_len;
581 int eop = (len_left == 0);
582 dma_addr_t dma_addr;
583 int err = 0;
585 dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
586 PCI_DMA_TODEVICE);
587 if (unlikely(enic_dma_map_check(enic, dma_addr)))
588 return -ENOMEM;
590 /* Queue the main skb fragment. The fragments are no larger
591 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
592 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
593 * per fragment is queued.
595 enic_queue_wq_desc(wq, skb, dma_addr, head_len, vlan_tag_insert,
596 vlan_tag, eop, loopback);
598 if (!eop)
599 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
601 return err;
604 static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq,
605 struct sk_buff *skb, int vlan_tag_insert,
606 unsigned int vlan_tag, int loopback)
608 unsigned int head_len = skb_headlen(skb);
609 unsigned int len_left = skb->len - head_len;
610 unsigned int hdr_len = skb_checksum_start_offset(skb);
611 unsigned int csum_offset = hdr_len + skb->csum_offset;
612 int eop = (len_left == 0);
613 dma_addr_t dma_addr;
614 int err = 0;
616 dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
617 PCI_DMA_TODEVICE);
618 if (unlikely(enic_dma_map_check(enic, dma_addr)))
619 return -ENOMEM;
621 /* Queue the main skb fragment. The fragments are no larger
622 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
623 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
624 * per fragment is queued.
626 enic_queue_wq_desc_csum_l4(wq, skb, dma_addr, head_len, csum_offset,
627 hdr_len, vlan_tag_insert, vlan_tag, eop,
628 loopback);
630 if (!eop)
631 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
633 return err;
636 static void enic_preload_tcp_csum_encap(struct sk_buff *skb)
638 if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
639 inner_ip_hdr(skb)->check = 0;
640 inner_tcp_hdr(skb)->check =
641 ~csum_tcpudp_magic(inner_ip_hdr(skb)->saddr,
642 inner_ip_hdr(skb)->daddr, 0,
643 IPPROTO_TCP, 0);
647 static void enic_preload_tcp_csum(struct sk_buff *skb)
649 /* Preload TCP csum field with IP pseudo hdr calculated
650 * with IP length set to zero. HW will later add in length
651 * to each TCP segment resulting from the TSO.
654 if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
655 ip_hdr(skb)->check = 0;
656 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
657 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
658 } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
659 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
660 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
664 static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
665 struct sk_buff *skb, unsigned int mss,
666 int vlan_tag_insert, unsigned int vlan_tag,
667 int loopback)
669 unsigned int frag_len_left = skb_headlen(skb);
670 unsigned int len_left = skb->len - frag_len_left;
671 int eop = (len_left == 0);
672 unsigned int offset = 0;
673 unsigned int hdr_len;
674 dma_addr_t dma_addr;
675 unsigned int len;
676 skb_frag_t *frag;
678 if (skb->encapsulation) {
679 hdr_len = skb_inner_transport_header(skb) - skb->data;
680 hdr_len += inner_tcp_hdrlen(skb);
681 enic_preload_tcp_csum_encap(skb);
682 } else {
683 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
684 enic_preload_tcp_csum(skb);
687 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
688 * for the main skb fragment
690 while (frag_len_left) {
691 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
692 dma_addr = pci_map_single(enic->pdev, skb->data + offset, len,
693 PCI_DMA_TODEVICE);
694 if (unlikely(enic_dma_map_check(enic, dma_addr)))
695 return -ENOMEM;
696 enic_queue_wq_desc_tso(wq, skb, dma_addr, len, mss, hdr_len,
697 vlan_tag_insert, vlan_tag,
698 eop && (len == frag_len_left), loopback);
699 frag_len_left -= len;
700 offset += len;
703 if (eop)
704 return 0;
706 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
707 * for additional data fragments
709 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
710 len_left -= skb_frag_size(frag);
711 frag_len_left = skb_frag_size(frag);
712 offset = 0;
714 while (frag_len_left) {
715 len = min(frag_len_left,
716 (unsigned int)WQ_ENET_MAX_DESC_LEN);
717 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
718 offset, len,
719 DMA_TO_DEVICE);
720 if (unlikely(enic_dma_map_check(enic, dma_addr)))
721 return -ENOMEM;
722 enic_queue_wq_desc_cont(wq, skb, dma_addr, len,
723 (len_left == 0) &&
724 (len == frag_len_left),/*EOP*/
725 loopback);
726 frag_len_left -= len;
727 offset += len;
731 return 0;
734 static inline int enic_queue_wq_skb_encap(struct enic *enic, struct vnic_wq *wq,
735 struct sk_buff *skb,
736 int vlan_tag_insert,
737 unsigned int vlan_tag, int loopback)
739 unsigned int head_len = skb_headlen(skb);
740 unsigned int len_left = skb->len - head_len;
741 /* Hardware will overwrite the checksum fields, calculating from
742 * scratch and ignoring the value placed by software.
743 * Offload mode = 00
744 * mss[2], mss[1], mss[0] bits are set
746 unsigned int mss_or_csum = 7;
747 int eop = (len_left == 0);
748 dma_addr_t dma_addr;
749 int err = 0;
751 dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
752 PCI_DMA_TODEVICE);
753 if (unlikely(enic_dma_map_check(enic, dma_addr)))
754 return -ENOMEM;
756 enic_queue_wq_desc_ex(wq, skb, dma_addr, head_len, mss_or_csum, 0,
757 vlan_tag_insert, vlan_tag,
758 WQ_ENET_OFFLOAD_MODE_CSUM, eop, 1 /* SOP */, eop,
759 loopback);
760 if (!eop)
761 err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
763 return err;
766 static inline void enic_queue_wq_skb(struct enic *enic,
767 struct vnic_wq *wq, struct sk_buff *skb)
769 unsigned int mss = skb_shinfo(skb)->gso_size;
770 unsigned int vlan_tag = 0;
771 int vlan_tag_insert = 0;
772 int loopback = 0;
773 int err;
775 if (skb_vlan_tag_present(skb)) {
776 /* VLAN tag from trunking driver */
777 vlan_tag_insert = 1;
778 vlan_tag = skb_vlan_tag_get(skb);
779 } else if (enic->loop_enable) {
780 vlan_tag = enic->loop_tag;
781 loopback = 1;
784 if (mss)
785 err = enic_queue_wq_skb_tso(enic, wq, skb, mss,
786 vlan_tag_insert, vlan_tag,
787 loopback);
788 else if (skb->encapsulation)
789 err = enic_queue_wq_skb_encap(enic, wq, skb, vlan_tag_insert,
790 vlan_tag, loopback);
791 else if (skb->ip_summed == CHECKSUM_PARTIAL)
792 err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert,
793 vlan_tag, loopback);
794 else
795 err = enic_queue_wq_skb_vlan(enic, wq, skb, vlan_tag_insert,
796 vlan_tag, loopback);
797 if (unlikely(err)) {
798 struct vnic_wq_buf *buf;
800 buf = wq->to_use->prev;
801 /* while not EOP of previous pkt && queue not empty.
802 * For all non EOP bufs, os_buf is NULL.
804 while (!buf->os_buf && (buf->next != wq->to_clean)) {
805 enic_free_wq_buf(wq, buf);
806 wq->ring.desc_avail++;
807 buf = buf->prev;
809 wq->to_use = buf->next;
810 dev_kfree_skb(skb);
814 /* netif_tx_lock held, process context with BHs disabled, or BH */
815 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
816 struct net_device *netdev)
818 struct enic *enic = netdev_priv(netdev);
819 struct vnic_wq *wq;
820 unsigned int txq_map;
821 struct netdev_queue *txq;
823 if (skb->len <= 0) {
824 dev_kfree_skb_any(skb);
825 return NETDEV_TX_OK;
828 txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
829 wq = &enic->wq[txq_map];
830 txq = netdev_get_tx_queue(netdev, txq_map);
832 /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
833 * which is very likely. In the off chance it's going to take
834 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
837 if (skb_shinfo(skb)->gso_size == 0 &&
838 skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
839 skb_linearize(skb)) {
840 dev_kfree_skb_any(skb);
841 return NETDEV_TX_OK;
844 spin_lock(&enic->wq_lock[txq_map]);
846 if (vnic_wq_desc_avail(wq) <
847 skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
848 netif_tx_stop_queue(txq);
849 /* This is a hard error, log it */
850 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
851 spin_unlock(&enic->wq_lock[txq_map]);
852 return NETDEV_TX_BUSY;
855 enic_queue_wq_skb(enic, wq, skb);
857 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
858 netif_tx_stop_queue(txq);
859 if (!skb->xmit_more || netif_xmit_stopped(txq))
860 vnic_wq_doorbell(wq);
862 spin_unlock(&enic->wq_lock[txq_map]);
864 return NETDEV_TX_OK;
867 /* dev_base_lock rwlock held, nominally process context */
868 static void enic_get_stats(struct net_device *netdev,
869 struct rtnl_link_stats64 *net_stats)
871 struct enic *enic = netdev_priv(netdev);
872 struct vnic_stats *stats;
873 int err;
875 err = enic_dev_stats_dump(enic, &stats);
876 /* return only when pci_zalloc_consistent fails in vnic_dev_stats_dump
877 * For other failures, like devcmd failure, we return previously
878 * recorded stats.
880 if (err == -ENOMEM)
881 return;
883 net_stats->tx_packets = stats->tx.tx_frames_ok;
884 net_stats->tx_bytes = stats->tx.tx_bytes_ok;
885 net_stats->tx_errors = stats->tx.tx_errors;
886 net_stats->tx_dropped = stats->tx.tx_drops;
888 net_stats->rx_packets = stats->rx.rx_frames_ok;
889 net_stats->rx_bytes = stats->rx.rx_bytes_ok;
890 net_stats->rx_errors = stats->rx.rx_errors;
891 net_stats->multicast = stats->rx.rx_multicast_frames_ok;
892 net_stats->rx_over_errors = enic->rq_truncated_pkts;
893 net_stats->rx_crc_errors = enic->rq_bad_fcs;
894 net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
897 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
899 struct enic *enic = netdev_priv(netdev);
901 if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
902 unsigned int mc_count = netdev_mc_count(netdev);
904 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
905 ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
907 return -ENOSPC;
910 enic_dev_add_addr(enic, mc_addr);
911 enic->mc_count++;
913 return 0;
916 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
918 struct enic *enic = netdev_priv(netdev);
920 enic_dev_del_addr(enic, mc_addr);
921 enic->mc_count--;
923 return 0;
926 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
928 struct enic *enic = netdev_priv(netdev);
930 if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
931 unsigned int uc_count = netdev_uc_count(netdev);
933 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
934 ENIC_UNICAST_PERFECT_FILTERS, uc_count);
936 return -ENOSPC;
939 enic_dev_add_addr(enic, uc_addr);
940 enic->uc_count++;
942 return 0;
945 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
947 struct enic *enic = netdev_priv(netdev);
949 enic_dev_del_addr(enic, uc_addr);
950 enic->uc_count--;
952 return 0;
955 void enic_reset_addr_lists(struct enic *enic)
957 struct net_device *netdev = enic->netdev;
959 __dev_uc_unsync(netdev, NULL);
960 __dev_mc_unsync(netdev, NULL);
962 enic->mc_count = 0;
963 enic->uc_count = 0;
964 enic->flags = 0;
967 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
969 struct enic *enic = netdev_priv(netdev);
971 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
972 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
973 return -EADDRNOTAVAIL;
974 } else {
975 if (!is_valid_ether_addr(addr))
976 return -EADDRNOTAVAIL;
979 memcpy(netdev->dev_addr, addr, netdev->addr_len);
981 return 0;
984 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
986 struct enic *enic = netdev_priv(netdev);
987 struct sockaddr *saddr = p;
988 char *addr = saddr->sa_data;
989 int err;
991 if (netif_running(enic->netdev)) {
992 err = enic_dev_del_station_addr(enic);
993 if (err)
994 return err;
997 err = enic_set_mac_addr(netdev, addr);
998 if (err)
999 return err;
1001 if (netif_running(enic->netdev)) {
1002 err = enic_dev_add_station_addr(enic);
1003 if (err)
1004 return err;
1007 return err;
1010 static int enic_set_mac_address(struct net_device *netdev, void *p)
1012 struct sockaddr *saddr = p;
1013 char *addr = saddr->sa_data;
1014 struct enic *enic = netdev_priv(netdev);
1015 int err;
1017 err = enic_dev_del_station_addr(enic);
1018 if (err)
1019 return err;
1021 err = enic_set_mac_addr(netdev, addr);
1022 if (err)
1023 return err;
1025 return enic_dev_add_station_addr(enic);
1028 /* netif_tx_lock held, BHs disabled */
1029 static void enic_set_rx_mode(struct net_device *netdev)
1031 struct enic *enic = netdev_priv(netdev);
1032 int directed = 1;
1033 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
1034 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
1035 int promisc = (netdev->flags & IFF_PROMISC) ||
1036 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
1037 int allmulti = (netdev->flags & IFF_ALLMULTI) ||
1038 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
1039 unsigned int flags = netdev->flags |
1040 (allmulti ? IFF_ALLMULTI : 0) |
1041 (promisc ? IFF_PROMISC : 0);
1043 if (enic->flags != flags) {
1044 enic->flags = flags;
1045 enic_dev_packet_filter(enic, directed,
1046 multicast, broadcast, promisc, allmulti);
1049 if (!promisc) {
1050 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
1051 if (!allmulti)
1052 __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
1056 /* netif_tx_lock held, BHs disabled */
1057 static void enic_tx_timeout(struct net_device *netdev)
1059 struct enic *enic = netdev_priv(netdev);
1060 schedule_work(&enic->tx_hang_reset);
1063 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1065 struct enic *enic = netdev_priv(netdev);
1066 struct enic_port_profile *pp;
1067 int err;
1069 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1070 if (err)
1071 return err;
1073 if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
1074 if (vf == PORT_SELF_VF) {
1075 memcpy(pp->vf_mac, mac, ETH_ALEN);
1076 return 0;
1077 } else {
1079 * For sriov vf's set the mac in hw
1081 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1082 vnic_dev_set_mac_addr, mac);
1083 return enic_dev_status_to_errno(err);
1085 } else
1086 return -EINVAL;
1089 static int enic_set_vf_port(struct net_device *netdev, int vf,
1090 struct nlattr *port[])
1092 struct enic *enic = netdev_priv(netdev);
1093 struct enic_port_profile prev_pp;
1094 struct enic_port_profile *pp;
1095 int err = 0, restore_pp = 1;
1097 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1098 if (err)
1099 return err;
1101 if (!port[IFLA_PORT_REQUEST])
1102 return -EOPNOTSUPP;
1104 memcpy(&prev_pp, pp, sizeof(*enic->pp));
1105 memset(pp, 0, sizeof(*enic->pp));
1107 pp->set |= ENIC_SET_REQUEST;
1108 pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1110 if (port[IFLA_PORT_PROFILE]) {
1111 pp->set |= ENIC_SET_NAME;
1112 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
1113 PORT_PROFILE_MAX);
1116 if (port[IFLA_PORT_INSTANCE_UUID]) {
1117 pp->set |= ENIC_SET_INSTANCE;
1118 memcpy(pp->instance_uuid,
1119 nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
1122 if (port[IFLA_PORT_HOST_UUID]) {
1123 pp->set |= ENIC_SET_HOST;
1124 memcpy(pp->host_uuid,
1125 nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
1128 if (vf == PORT_SELF_VF) {
1129 /* Special case handling: mac came from IFLA_VF_MAC */
1130 if (!is_zero_ether_addr(prev_pp.vf_mac))
1131 memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
1133 if (is_zero_ether_addr(netdev->dev_addr))
1134 eth_hw_addr_random(netdev);
1135 } else {
1136 /* SR-IOV VF: get mac from adapter */
1137 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
1138 vnic_dev_get_mac_addr, pp->mac_addr);
1139 if (err) {
1140 netdev_err(netdev, "Error getting mac for vf %d\n", vf);
1141 memcpy(pp, &prev_pp, sizeof(*pp));
1142 return enic_dev_status_to_errno(err);
1146 err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
1147 if (err) {
1148 if (restore_pp) {
1149 /* Things are still the way they were: Implicit
1150 * DISASSOCIATE failed
1152 memcpy(pp, &prev_pp, sizeof(*pp));
1153 } else {
1154 memset(pp, 0, sizeof(*pp));
1155 if (vf == PORT_SELF_VF)
1156 eth_zero_addr(netdev->dev_addr);
1158 } else {
1159 /* Set flag to indicate that the port assoc/disassoc
1160 * request has been sent out to fw
1162 pp->set |= ENIC_PORT_REQUEST_APPLIED;
1164 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
1165 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
1166 eth_zero_addr(pp->mac_addr);
1167 if (vf == PORT_SELF_VF)
1168 eth_zero_addr(netdev->dev_addr);
1172 if (vf == PORT_SELF_VF)
1173 eth_zero_addr(pp->vf_mac);
1175 return err;
1178 static int enic_get_vf_port(struct net_device *netdev, int vf,
1179 struct sk_buff *skb)
1181 struct enic *enic = netdev_priv(netdev);
1182 u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1183 struct enic_port_profile *pp;
1184 int err;
1186 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
1187 if (err)
1188 return err;
1190 if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
1191 return -ENODATA;
1193 err = enic_process_get_pp_request(enic, vf, pp->request, &response);
1194 if (err)
1195 return err;
1197 if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
1198 nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
1199 ((pp->set & ENIC_SET_NAME) &&
1200 nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
1201 ((pp->set & ENIC_SET_INSTANCE) &&
1202 nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1203 pp->instance_uuid)) ||
1204 ((pp->set & ENIC_SET_HOST) &&
1205 nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
1206 goto nla_put_failure;
1207 return 0;
1209 nla_put_failure:
1210 return -EMSGSIZE;
1213 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1215 struct enic *enic = vnic_dev_priv(rq->vdev);
1217 if (!buf->os_buf)
1218 return;
1220 pci_unmap_single(enic->pdev, buf->dma_addr,
1221 buf->len, PCI_DMA_FROMDEVICE);
1222 dev_kfree_skb_any(buf->os_buf);
1223 buf->os_buf = NULL;
1226 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1228 struct enic *enic = vnic_dev_priv(rq->vdev);
1229 struct net_device *netdev = enic->netdev;
1230 struct sk_buff *skb;
1231 unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
1232 unsigned int os_buf_index = 0;
1233 dma_addr_t dma_addr;
1234 struct vnic_rq_buf *buf = rq->to_use;
1236 if (buf->os_buf) {
1237 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
1238 buf->len);
1240 return 0;
1242 skb = netdev_alloc_skb_ip_align(netdev, len);
1243 if (!skb)
1244 return -ENOMEM;
1246 dma_addr = pci_map_single(enic->pdev, skb->data, len,
1247 PCI_DMA_FROMDEVICE);
1248 if (unlikely(enic_dma_map_check(enic, dma_addr))) {
1249 dev_kfree_skb(skb);
1250 return -ENOMEM;
1253 enic_queue_rq_desc(rq, skb, os_buf_index,
1254 dma_addr, len);
1256 return 0;
1259 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1260 u32 pkt_len)
1262 if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1263 pkt_size->large_pkt_bytes_cnt += pkt_len;
1264 else
1265 pkt_size->small_pkt_bytes_cnt += pkt_len;
1268 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
1269 struct vnic_rq_buf *buf, u16 len)
1271 struct enic *enic = netdev_priv(netdev);
1272 struct sk_buff *new_skb;
1274 if (len > enic->rx_copybreak)
1275 return false;
1276 new_skb = netdev_alloc_skb_ip_align(netdev, len);
1277 if (!new_skb)
1278 return false;
1279 pci_dma_sync_single_for_cpu(enic->pdev, buf->dma_addr, len,
1280 DMA_FROM_DEVICE);
1281 memcpy(new_skb->data, (*skb)->data, len);
1282 *skb = new_skb;
1284 return true;
1287 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1288 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1289 int skipped, void *opaque)
1291 struct enic *enic = vnic_dev_priv(rq->vdev);
1292 struct net_device *netdev = enic->netdev;
1293 struct sk_buff *skb;
1294 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1296 u8 type, color, eop, sop, ingress_port, vlan_stripped;
1297 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1298 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1299 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1300 u8 packet_error;
1301 u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1302 u32 rss_hash;
1303 bool outer_csum_ok = true, encap = false;
1305 if (skipped)
1306 return;
1308 skb = buf->os_buf;
1310 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1311 &type, &color, &q_number, &completed_index,
1312 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1313 &csum_not_calc, &rss_hash, &bytes_written,
1314 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1315 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1316 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1317 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1318 &fcs_ok);
1320 if (packet_error) {
1322 if (!fcs_ok) {
1323 if (bytes_written > 0)
1324 enic->rq_bad_fcs++;
1325 else if (bytes_written == 0)
1326 enic->rq_truncated_pkts++;
1329 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1330 PCI_DMA_FROMDEVICE);
1331 dev_kfree_skb_any(skb);
1332 buf->os_buf = NULL;
1334 return;
1337 if (eop && bytes_written > 0) {
1339 /* Good receive
1342 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1343 buf->os_buf = NULL;
1344 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1345 PCI_DMA_FROMDEVICE);
1347 prefetch(skb->data - NET_IP_ALIGN);
1349 skb_put(skb, bytes_written);
1350 skb->protocol = eth_type_trans(skb, netdev);
1351 skb_record_rx_queue(skb, q_number);
1352 if ((netdev->features & NETIF_F_RXHASH) && rss_hash &&
1353 (type == 3)) {
1354 switch (rss_type) {
1355 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
1356 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
1357 case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6_EX:
1358 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L4);
1359 break;
1360 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv4:
1361 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6:
1362 case CQ_ENET_RQ_DESC_RSS_TYPE_IPv6_EX:
1363 skb_set_hash(skb, rss_hash, PKT_HASH_TYPE_L3);
1364 break;
1367 if (enic->vxlan.vxlan_udp_port_number) {
1368 switch (enic->vxlan.patch_level) {
1369 case 0:
1370 if (fcoe) {
1371 encap = true;
1372 outer_csum_ok = fcoe_fc_crc_ok;
1374 break;
1375 case 2:
1376 if ((type == 7) &&
1377 (rss_hash & BIT(0))) {
1378 encap = true;
1379 outer_csum_ok = (rss_hash & BIT(1)) &&
1380 (rss_hash & BIT(2));
1382 break;
1386 /* Hardware does not provide whole packet checksum. It only
1387 * provides pseudo checksum. Since hw validates the packet
1388 * checksum but not provide us the checksum value. use
1389 * CHECSUM_UNNECESSARY.
1391 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is
1392 * inner csum_ok. outer_csum_ok is set by hw when outer udp
1393 * csum is correct or is zero.
1395 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc &&
1396 tcp_udp_csum_ok && ipv4_csum_ok && outer_csum_ok) {
1397 skb->ip_summed = CHECKSUM_UNNECESSARY;
1398 skb->csum_level = encap;
1401 if (vlan_stripped)
1402 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1404 skb_mark_napi_id(skb, &enic->napi[rq->index]);
1405 if (!(netdev->features & NETIF_F_GRO))
1406 netif_receive_skb(skb);
1407 else
1408 napi_gro_receive(&enic->napi[q_number], skb);
1409 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1410 enic_intr_update_pkt_size(&cq->pkt_size_counter,
1411 bytes_written);
1412 } else {
1414 /* Buffer overflow
1417 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1418 PCI_DMA_FROMDEVICE);
1419 dev_kfree_skb_any(skb);
1420 buf->os_buf = NULL;
1424 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1425 u8 type, u16 q_number, u16 completed_index, void *opaque)
1427 struct enic *enic = vnic_dev_priv(vdev);
1429 vnic_rq_service(&enic->rq[q_number], cq_desc,
1430 completed_index, VNIC_RQ_RETURN_DESC,
1431 enic_rq_indicate_buf, opaque);
1433 return 0;
1436 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1438 unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1439 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1440 u32 timer = cq->tobe_rx_coal_timeval;
1442 if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1443 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1444 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1448 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1450 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1451 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1452 struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1453 int index;
1454 u32 timer;
1455 u32 range_start;
1456 u32 traffic;
1457 u64 delta;
1458 ktime_t now = ktime_get();
1460 delta = ktime_us_delta(now, cq->prev_ts);
1461 if (delta < ENIC_AIC_TS_BREAK)
1462 return;
1463 cq->prev_ts = now;
1465 traffic = pkt_size_counter->large_pkt_bytes_cnt +
1466 pkt_size_counter->small_pkt_bytes_cnt;
1467 /* The table takes Mbps
1468 * traffic *= 8 => bits
1469 * traffic *= (10^6 / delta) => bps
1470 * traffic /= 10^6 => Mbps
1472 * Combining, traffic *= (8 / delta)
1475 traffic <<= 3;
1476 traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1478 for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1479 if (traffic < mod_table[index].rx_rate)
1480 break;
1481 range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1482 pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1483 rx_coal->small_pkt_range_start :
1484 rx_coal->large_pkt_range_start;
1485 timer = range_start + ((rx_coal->range_end - range_start) *
1486 mod_table[index].range_percent / 100);
1487 /* Damping */
1488 cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1490 pkt_size_counter->large_pkt_bytes_cnt = 0;
1491 pkt_size_counter->small_pkt_bytes_cnt = 0;
1494 static int enic_poll(struct napi_struct *napi, int budget)
1496 struct net_device *netdev = napi->dev;
1497 struct enic *enic = netdev_priv(netdev);
1498 unsigned int cq_rq = enic_cq_rq(enic, 0);
1499 unsigned int cq_wq = enic_cq_wq(enic, 0);
1500 unsigned int intr = enic_legacy_io_intr();
1501 unsigned int rq_work_to_do = budget;
1502 unsigned int wq_work_to_do = -1; /* no limit */
1503 unsigned int work_done, rq_work_done = 0, wq_work_done;
1504 int err;
1506 wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1507 enic_wq_service, NULL);
1509 if (budget > 0)
1510 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1511 rq_work_to_do, enic_rq_service, NULL);
1513 /* Accumulate intr event credits for this polling
1514 * cycle. An intr event is the completion of a
1515 * a WQ or RQ packet.
1518 work_done = rq_work_done + wq_work_done;
1520 if (work_done > 0)
1521 vnic_intr_return_credits(&enic->intr[intr],
1522 work_done,
1523 0 /* don't unmask intr */,
1524 0 /* don't reset intr timer */);
1526 err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1528 /* Buffer allocation failed. Stay in polling
1529 * mode so we can try to fill the ring again.
1532 if (err)
1533 rq_work_done = rq_work_to_do;
1534 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1535 /* Call the function which refreshes the intr coalescing timer
1536 * value based on the traffic.
1538 enic_calc_int_moderation(enic, &enic->rq[0]);
1540 if ((rq_work_done < budget) && napi_complete_done(napi, rq_work_done)) {
1542 /* Some work done, but not enough to stay in polling,
1543 * exit polling
1546 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1547 enic_set_int_moderation(enic, &enic->rq[0]);
1548 vnic_intr_unmask(&enic->intr[intr]);
1551 return rq_work_done;
1554 #ifdef CONFIG_RFS_ACCEL
1555 static void enic_free_rx_cpu_rmap(struct enic *enic)
1557 free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1558 enic->netdev->rx_cpu_rmap = NULL;
1561 static void enic_set_rx_cpu_rmap(struct enic *enic)
1563 int i, res;
1565 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1566 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1567 if (unlikely(!enic->netdev->rx_cpu_rmap))
1568 return;
1569 for (i = 0; i < enic->rq_count; i++) {
1570 res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1571 enic->msix_entry[i].vector);
1572 if (unlikely(res)) {
1573 enic_free_rx_cpu_rmap(enic);
1574 return;
1580 #else
1582 static void enic_free_rx_cpu_rmap(struct enic *enic)
1586 static void enic_set_rx_cpu_rmap(struct enic *enic)
1590 #endif /* CONFIG_RFS_ACCEL */
1592 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1594 struct net_device *netdev = napi->dev;
1595 struct enic *enic = netdev_priv(netdev);
1596 unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1597 struct vnic_wq *wq = &enic->wq[wq_index];
1598 unsigned int cq;
1599 unsigned int intr;
1600 unsigned int wq_work_to_do = -1; /* clean all desc possible */
1601 unsigned int wq_work_done;
1602 unsigned int wq_irq;
1604 wq_irq = wq->index;
1605 cq = enic_cq_wq(enic, wq_irq);
1606 intr = enic_msix_wq_intr(enic, wq_irq);
1607 wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1608 enic_wq_service, NULL);
1610 vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1611 0 /* don't unmask intr */,
1612 1 /* reset intr timer */);
1613 if (!wq_work_done) {
1614 napi_complete(napi);
1615 vnic_intr_unmask(&enic->intr[intr]);
1616 return 0;
1619 return budget;
1622 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1624 struct net_device *netdev = napi->dev;
1625 struct enic *enic = netdev_priv(netdev);
1626 unsigned int rq = (napi - &enic->napi[0]);
1627 unsigned int cq = enic_cq_rq(enic, rq);
1628 unsigned int intr = enic_msix_rq_intr(enic, rq);
1629 unsigned int work_to_do = budget;
1630 unsigned int work_done = 0;
1631 int err;
1633 /* Service RQ
1636 if (budget > 0)
1637 work_done = vnic_cq_service(&enic->cq[cq],
1638 work_to_do, enic_rq_service, NULL);
1640 /* Return intr event credits for this polling
1641 * cycle. An intr event is the completion of a
1642 * RQ packet.
1645 if (work_done > 0)
1646 vnic_intr_return_credits(&enic->intr[intr],
1647 work_done,
1648 0 /* don't unmask intr */,
1649 0 /* don't reset intr timer */);
1651 err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1653 /* Buffer allocation failed. Stay in polling mode
1654 * so we can try to fill the ring again.
1657 if (err)
1658 work_done = work_to_do;
1659 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1660 /* Call the function which refreshes the intr coalescing timer
1661 * value based on the traffic.
1663 enic_calc_int_moderation(enic, &enic->rq[rq]);
1665 if ((work_done < budget) && napi_complete_done(napi, work_done)) {
1667 /* Some work done, but not enough to stay in polling,
1668 * exit polling
1671 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1672 enic_set_int_moderation(enic, &enic->rq[rq]);
1673 vnic_intr_unmask(&enic->intr[intr]);
1676 return work_done;
1679 static void enic_notify_timer(unsigned long data)
1681 struct enic *enic = (struct enic *)data;
1683 enic_notify_check(enic);
1685 mod_timer(&enic->notify_timer,
1686 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1689 static void enic_free_intr(struct enic *enic)
1691 struct net_device *netdev = enic->netdev;
1692 unsigned int i;
1694 enic_free_rx_cpu_rmap(enic);
1695 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1696 case VNIC_DEV_INTR_MODE_INTX:
1697 free_irq(enic->pdev->irq, netdev);
1698 break;
1699 case VNIC_DEV_INTR_MODE_MSI:
1700 free_irq(enic->pdev->irq, enic);
1701 break;
1702 case VNIC_DEV_INTR_MODE_MSIX:
1703 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1704 if (enic->msix[i].requested)
1705 free_irq(enic->msix_entry[i].vector,
1706 enic->msix[i].devid);
1707 break;
1708 default:
1709 break;
1713 static int enic_request_intr(struct enic *enic)
1715 struct net_device *netdev = enic->netdev;
1716 unsigned int i, intr;
1717 int err = 0;
1719 enic_set_rx_cpu_rmap(enic);
1720 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1722 case VNIC_DEV_INTR_MODE_INTX:
1724 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1725 IRQF_SHARED, netdev->name, netdev);
1726 break;
1728 case VNIC_DEV_INTR_MODE_MSI:
1730 err = request_irq(enic->pdev->irq, enic_isr_msi,
1731 0, netdev->name, enic);
1732 break;
1734 case VNIC_DEV_INTR_MODE_MSIX:
1736 for (i = 0; i < enic->rq_count; i++) {
1737 intr = enic_msix_rq_intr(enic, i);
1738 snprintf(enic->msix[intr].devname,
1739 sizeof(enic->msix[intr].devname),
1740 "%s-rx-%u", netdev->name, i);
1741 enic->msix[intr].isr = enic_isr_msix;
1742 enic->msix[intr].devid = &enic->napi[i];
1745 for (i = 0; i < enic->wq_count; i++) {
1746 int wq = enic_cq_wq(enic, i);
1748 intr = enic_msix_wq_intr(enic, i);
1749 snprintf(enic->msix[intr].devname,
1750 sizeof(enic->msix[intr].devname),
1751 "%s-tx-%u", netdev->name, i);
1752 enic->msix[intr].isr = enic_isr_msix;
1753 enic->msix[intr].devid = &enic->napi[wq];
1756 intr = enic_msix_err_intr(enic);
1757 snprintf(enic->msix[intr].devname,
1758 sizeof(enic->msix[intr].devname),
1759 "%s-err", netdev->name);
1760 enic->msix[intr].isr = enic_isr_msix_err;
1761 enic->msix[intr].devid = enic;
1763 intr = enic_msix_notify_intr(enic);
1764 snprintf(enic->msix[intr].devname,
1765 sizeof(enic->msix[intr].devname),
1766 "%s-notify", netdev->name);
1767 enic->msix[intr].isr = enic_isr_msix_notify;
1768 enic->msix[intr].devid = enic;
1770 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1771 enic->msix[i].requested = 0;
1773 for (i = 0; i < enic->intr_count; i++) {
1774 err = request_irq(enic->msix_entry[i].vector,
1775 enic->msix[i].isr, 0,
1776 enic->msix[i].devname,
1777 enic->msix[i].devid);
1778 if (err) {
1779 enic_free_intr(enic);
1780 break;
1782 enic->msix[i].requested = 1;
1785 break;
1787 default:
1788 break;
1791 return err;
1794 static void enic_synchronize_irqs(struct enic *enic)
1796 unsigned int i;
1798 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1799 case VNIC_DEV_INTR_MODE_INTX:
1800 case VNIC_DEV_INTR_MODE_MSI:
1801 synchronize_irq(enic->pdev->irq);
1802 break;
1803 case VNIC_DEV_INTR_MODE_MSIX:
1804 for (i = 0; i < enic->intr_count; i++)
1805 synchronize_irq(enic->msix_entry[i].vector);
1806 break;
1807 default:
1808 break;
1812 static void enic_set_rx_coal_setting(struct enic *enic)
1814 unsigned int speed;
1815 int index = -1;
1816 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1818 /* 1. Read the link speed from fw
1819 * 2. Pick the default range for the speed
1820 * 3. Update it in enic->rx_coalesce_setting
1822 speed = vnic_dev_port_speed(enic->vdev);
1823 if (ENIC_LINK_SPEED_10G < speed)
1824 index = ENIC_LINK_40G_INDEX;
1825 else if (ENIC_LINK_SPEED_4G < speed)
1826 index = ENIC_LINK_10G_INDEX;
1827 else
1828 index = ENIC_LINK_4G_INDEX;
1830 rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1831 rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1832 rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1834 /* Start with the value provided by UCSM */
1835 for (index = 0; index < enic->rq_count; index++)
1836 enic->cq[index].cur_rx_coal_timeval =
1837 enic->config.intr_timer_usec;
1839 rx_coal->use_adaptive_rx_coalesce = 1;
1842 static int enic_dev_notify_set(struct enic *enic)
1844 int err;
1846 spin_lock_bh(&enic->devcmd_lock);
1847 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1848 case VNIC_DEV_INTR_MODE_INTX:
1849 err = vnic_dev_notify_set(enic->vdev,
1850 enic_legacy_notify_intr());
1851 break;
1852 case VNIC_DEV_INTR_MODE_MSIX:
1853 err = vnic_dev_notify_set(enic->vdev,
1854 enic_msix_notify_intr(enic));
1855 break;
1856 default:
1857 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1858 break;
1860 spin_unlock_bh(&enic->devcmd_lock);
1862 return err;
1865 static void enic_notify_timer_start(struct enic *enic)
1867 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1868 case VNIC_DEV_INTR_MODE_MSI:
1869 mod_timer(&enic->notify_timer, jiffies);
1870 break;
1871 default:
1872 /* Using intr for notification for INTx/MSI-X */
1873 break;
1877 /* rtnl lock is held, process context */
1878 static int enic_open(struct net_device *netdev)
1880 struct enic *enic = netdev_priv(netdev);
1881 unsigned int i;
1882 int err;
1884 err = enic_request_intr(enic);
1885 if (err) {
1886 netdev_err(netdev, "Unable to request irq.\n");
1887 return err;
1889 enic_init_affinity_hint(enic);
1890 enic_set_affinity_hint(enic);
1892 err = enic_dev_notify_set(enic);
1893 if (err) {
1894 netdev_err(netdev,
1895 "Failed to alloc notify buffer, aborting.\n");
1896 goto err_out_free_intr;
1899 for (i = 0; i < enic->rq_count; i++) {
1900 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1901 /* Need at least one buffer on ring to get going */
1902 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1903 netdev_err(netdev, "Unable to alloc receive buffers\n");
1904 err = -ENOMEM;
1905 goto err_out_free_rq;
1909 for (i = 0; i < enic->wq_count; i++)
1910 vnic_wq_enable(&enic->wq[i]);
1911 for (i = 0; i < enic->rq_count; i++)
1912 vnic_rq_enable(&enic->rq[i]);
1914 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1915 enic_dev_add_station_addr(enic);
1917 enic_set_rx_mode(netdev);
1919 netif_tx_wake_all_queues(netdev);
1921 for (i = 0; i < enic->rq_count; i++)
1922 napi_enable(&enic->napi[i]);
1924 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1925 for (i = 0; i < enic->wq_count; i++)
1926 napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
1927 enic_dev_enable(enic);
1929 for (i = 0; i < enic->intr_count; i++)
1930 vnic_intr_unmask(&enic->intr[i]);
1932 enic_notify_timer_start(enic);
1933 enic_rfs_flw_tbl_init(enic);
1935 return 0;
1937 err_out_free_rq:
1938 for (i = 0; i < enic->rq_count; i++)
1939 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1940 enic_dev_notify_unset(enic);
1941 err_out_free_intr:
1942 enic_unset_affinity_hint(enic);
1943 enic_free_intr(enic);
1945 return err;
1948 /* rtnl lock is held, process context */
1949 static int enic_stop(struct net_device *netdev)
1951 struct enic *enic = netdev_priv(netdev);
1952 unsigned int i;
1953 int err;
1955 for (i = 0; i < enic->intr_count; i++) {
1956 vnic_intr_mask(&enic->intr[i]);
1957 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1960 enic_synchronize_irqs(enic);
1962 del_timer_sync(&enic->notify_timer);
1963 enic_rfs_flw_tbl_free(enic);
1965 enic_dev_disable(enic);
1967 for (i = 0; i < enic->rq_count; i++)
1968 napi_disable(&enic->napi[i]);
1970 netif_carrier_off(netdev);
1971 netif_tx_disable(netdev);
1972 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1973 for (i = 0; i < enic->wq_count; i++)
1974 napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
1976 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1977 enic_dev_del_station_addr(enic);
1979 for (i = 0; i < enic->wq_count; i++) {
1980 err = vnic_wq_disable(&enic->wq[i]);
1981 if (err)
1982 return err;
1984 for (i = 0; i < enic->rq_count; i++) {
1985 err = vnic_rq_disable(&enic->rq[i]);
1986 if (err)
1987 return err;
1990 enic_dev_notify_unset(enic);
1991 enic_unset_affinity_hint(enic);
1992 enic_free_intr(enic);
1994 for (i = 0; i < enic->wq_count; i++)
1995 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1996 for (i = 0; i < enic->rq_count; i++)
1997 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1998 for (i = 0; i < enic->cq_count; i++)
1999 vnic_cq_clean(&enic->cq[i]);
2000 for (i = 0; i < enic->intr_count; i++)
2001 vnic_intr_clean(&enic->intr[i]);
2003 return 0;
2006 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
2008 struct enic *enic = netdev_priv(netdev);
2009 int running = netif_running(netdev);
2011 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2012 return -EOPNOTSUPP;
2014 if (running)
2015 enic_stop(netdev);
2017 netdev->mtu = new_mtu;
2019 if (netdev->mtu > enic->port_mtu)
2020 netdev_warn(netdev,
2021 "interface MTU (%d) set higher than port MTU (%d)\n",
2022 netdev->mtu, enic->port_mtu);
2024 if (running)
2025 enic_open(netdev);
2027 return 0;
2030 static void enic_change_mtu_work(struct work_struct *work)
2032 struct enic *enic = container_of(work, struct enic, change_mtu_work);
2033 struct net_device *netdev = enic->netdev;
2034 int new_mtu = vnic_dev_mtu(enic->vdev);
2035 int err;
2036 unsigned int i;
2038 new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
2040 rtnl_lock();
2042 /* Stop RQ */
2043 del_timer_sync(&enic->notify_timer);
2045 for (i = 0; i < enic->rq_count; i++)
2046 napi_disable(&enic->napi[i]);
2048 vnic_intr_mask(&enic->intr[0]);
2049 enic_synchronize_irqs(enic);
2050 err = vnic_rq_disable(&enic->rq[0]);
2051 if (err) {
2052 rtnl_unlock();
2053 netdev_err(netdev, "Unable to disable RQ.\n");
2054 return;
2056 vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
2057 vnic_cq_clean(&enic->cq[0]);
2058 vnic_intr_clean(&enic->intr[0]);
2060 /* Fill RQ with new_mtu-sized buffers */
2061 netdev->mtu = new_mtu;
2062 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
2063 /* Need at least one buffer on ring to get going */
2064 if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
2065 rtnl_unlock();
2066 netdev_err(netdev, "Unable to alloc receive buffers.\n");
2067 return;
2070 /* Start RQ */
2071 vnic_rq_enable(&enic->rq[0]);
2072 napi_enable(&enic->napi[0]);
2073 vnic_intr_unmask(&enic->intr[0]);
2074 enic_notify_timer_start(enic);
2076 rtnl_unlock();
2078 netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
2081 #ifdef CONFIG_NET_POLL_CONTROLLER
2082 static void enic_poll_controller(struct net_device *netdev)
2084 struct enic *enic = netdev_priv(netdev);
2085 struct vnic_dev *vdev = enic->vdev;
2086 unsigned int i, intr;
2088 switch (vnic_dev_get_intr_mode(vdev)) {
2089 case VNIC_DEV_INTR_MODE_MSIX:
2090 for (i = 0; i < enic->rq_count; i++) {
2091 intr = enic_msix_rq_intr(enic, i);
2092 enic_isr_msix(enic->msix_entry[intr].vector,
2093 &enic->napi[i]);
2096 for (i = 0; i < enic->wq_count; i++) {
2097 intr = enic_msix_wq_intr(enic, i);
2098 enic_isr_msix(enic->msix_entry[intr].vector,
2099 &enic->napi[enic_cq_wq(enic, i)]);
2102 break;
2103 case VNIC_DEV_INTR_MODE_MSI:
2104 enic_isr_msi(enic->pdev->irq, enic);
2105 break;
2106 case VNIC_DEV_INTR_MODE_INTX:
2107 enic_isr_legacy(enic->pdev->irq, netdev);
2108 break;
2109 default:
2110 break;
2113 #endif
2115 static int enic_dev_wait(struct vnic_dev *vdev,
2116 int (*start)(struct vnic_dev *, int),
2117 int (*finished)(struct vnic_dev *, int *),
2118 int arg)
2120 unsigned long time;
2121 int done;
2122 int err;
2124 BUG_ON(in_interrupt());
2126 err = start(vdev, arg);
2127 if (err)
2128 return err;
2130 /* Wait for func to complete...2 seconds max
2133 time = jiffies + (HZ * 2);
2134 do {
2136 err = finished(vdev, &done);
2137 if (err)
2138 return err;
2140 if (done)
2141 return 0;
2143 schedule_timeout_uninterruptible(HZ / 10);
2145 } while (time_after(time, jiffies));
2147 return -ETIMEDOUT;
2150 static int enic_dev_open(struct enic *enic)
2152 int err;
2154 err = enic_dev_wait(enic->vdev, vnic_dev_open,
2155 vnic_dev_open_done, 0);
2156 if (err)
2157 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
2158 err);
2160 return err;
2163 static int enic_dev_soft_reset(struct enic *enic)
2165 int err;
2167 err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
2168 vnic_dev_soft_reset_done, 0);
2169 if (err)
2170 netdev_err(enic->netdev, "vNIC soft reset failed, err %d\n",
2171 err);
2173 return err;
2176 static int enic_dev_hang_reset(struct enic *enic)
2178 int err;
2180 err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
2181 vnic_dev_hang_reset_done, 0);
2182 if (err)
2183 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
2184 err);
2186 return err;
2189 int __enic_set_rsskey(struct enic *enic)
2191 union vnic_rss_key *rss_key_buf_va;
2192 dma_addr_t rss_key_buf_pa;
2193 int i, kidx, bidx, err;
2195 rss_key_buf_va = pci_zalloc_consistent(enic->pdev,
2196 sizeof(union vnic_rss_key),
2197 &rss_key_buf_pa);
2198 if (!rss_key_buf_va)
2199 return -ENOMEM;
2201 for (i = 0; i < ENIC_RSS_LEN; i++) {
2202 kidx = i / ENIC_RSS_BYTES_PER_KEY;
2203 bidx = i % ENIC_RSS_BYTES_PER_KEY;
2204 rss_key_buf_va->key[kidx].b[bidx] = enic->rss_key[i];
2206 spin_lock_bh(&enic->devcmd_lock);
2207 err = enic_set_rss_key(enic,
2208 rss_key_buf_pa,
2209 sizeof(union vnic_rss_key));
2210 spin_unlock_bh(&enic->devcmd_lock);
2212 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
2213 rss_key_buf_va, rss_key_buf_pa);
2215 return err;
2218 static int enic_set_rsskey(struct enic *enic)
2220 netdev_rss_key_fill(enic->rss_key, ENIC_RSS_LEN);
2222 return __enic_set_rsskey(enic);
2225 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
2227 dma_addr_t rss_cpu_buf_pa;
2228 union vnic_rss_cpu *rss_cpu_buf_va = NULL;
2229 unsigned int i;
2230 int err;
2232 rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
2233 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
2234 if (!rss_cpu_buf_va)
2235 return -ENOMEM;
2237 for (i = 0; i < (1 << rss_hash_bits); i++)
2238 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
2240 spin_lock_bh(&enic->devcmd_lock);
2241 err = enic_set_rss_cpu(enic,
2242 rss_cpu_buf_pa,
2243 sizeof(union vnic_rss_cpu));
2244 spin_unlock_bh(&enic->devcmd_lock);
2246 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
2247 rss_cpu_buf_va, rss_cpu_buf_pa);
2249 return err;
2252 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
2253 u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
2255 const u8 tso_ipid_split_en = 0;
2256 const u8 ig_vlan_strip_en = 1;
2257 int err;
2259 /* Enable VLAN tag stripping.
2262 spin_lock_bh(&enic->devcmd_lock);
2263 err = enic_set_nic_cfg(enic,
2264 rss_default_cpu, rss_hash_type,
2265 rss_hash_bits, rss_base_cpu,
2266 rss_enable, tso_ipid_split_en,
2267 ig_vlan_strip_en);
2268 spin_unlock_bh(&enic->devcmd_lock);
2270 return err;
2273 static int enic_set_rss_nic_cfg(struct enic *enic)
2275 struct device *dev = enic_get_dev(enic);
2276 const u8 rss_default_cpu = 0;
2277 const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
2278 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
2279 NIC_CFG_RSS_HASH_TYPE_IPV6 |
2280 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2281 const u8 rss_hash_bits = 7;
2282 const u8 rss_base_cpu = 0;
2283 u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2285 if (rss_enable) {
2286 if (!enic_set_rsskey(enic)) {
2287 if (enic_set_rsscpu(enic, rss_hash_bits)) {
2288 rss_enable = 0;
2289 dev_warn(dev, "RSS disabled, "
2290 "Failed to set RSS cpu indirection table.");
2292 } else {
2293 rss_enable = 0;
2294 dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2298 return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2299 rss_hash_bits, rss_base_cpu, rss_enable);
2302 static void enic_reset(struct work_struct *work)
2304 struct enic *enic = container_of(work, struct enic, reset);
2306 if (!netif_running(enic->netdev))
2307 return;
2309 rtnl_lock();
2311 spin_lock(&enic->enic_api_lock);
2312 enic_stop(enic->netdev);
2313 enic_dev_soft_reset(enic);
2314 enic_reset_addr_lists(enic);
2315 enic_init_vnic_resources(enic);
2316 enic_set_rss_nic_cfg(enic);
2317 enic_dev_set_ig_vlan_rewrite_mode(enic);
2318 enic_open(enic->netdev);
2319 spin_unlock(&enic->enic_api_lock);
2320 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2322 rtnl_unlock();
2325 static void enic_tx_hang_reset(struct work_struct *work)
2327 struct enic *enic = container_of(work, struct enic, tx_hang_reset);
2329 rtnl_lock();
2331 spin_lock(&enic->enic_api_lock);
2332 enic_dev_hang_notify(enic);
2333 enic_stop(enic->netdev);
2334 enic_dev_hang_reset(enic);
2335 enic_reset_addr_lists(enic);
2336 enic_init_vnic_resources(enic);
2337 enic_set_rss_nic_cfg(enic);
2338 enic_dev_set_ig_vlan_rewrite_mode(enic);
2339 enic_open(enic->netdev);
2340 spin_unlock(&enic->enic_api_lock);
2341 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2343 rtnl_unlock();
2346 static int enic_set_intr_mode(struct enic *enic)
2348 unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2349 unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
2350 unsigned int i;
2352 /* Set interrupt mode (INTx, MSI, MSI-X) depending
2353 * on system capabilities.
2355 * Try MSI-X first
2357 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2358 * (the second to last INTR is used for WQ/RQ errors)
2359 * (the last INTR is used for notifications)
2362 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2363 for (i = 0; i < n + m + 2; i++)
2364 enic->msix_entry[i].entry = i;
2366 /* Use multiple RQs if RSS is enabled
2369 if (ENIC_SETTING(enic, RSS) &&
2370 enic->config.intr_mode < 1 &&
2371 enic->rq_count >= n &&
2372 enic->wq_count >= m &&
2373 enic->cq_count >= n + m &&
2374 enic->intr_count >= n + m + 2) {
2376 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2377 n + m + 2, n + m + 2) > 0) {
2379 enic->rq_count = n;
2380 enic->wq_count = m;
2381 enic->cq_count = n + m;
2382 enic->intr_count = n + m + 2;
2384 vnic_dev_set_intr_mode(enic->vdev,
2385 VNIC_DEV_INTR_MODE_MSIX);
2387 return 0;
2391 if (enic->config.intr_mode < 1 &&
2392 enic->rq_count >= 1 &&
2393 enic->wq_count >= m &&
2394 enic->cq_count >= 1 + m &&
2395 enic->intr_count >= 1 + m + 2) {
2396 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2397 1 + m + 2, 1 + m + 2) > 0) {
2399 enic->rq_count = 1;
2400 enic->wq_count = m;
2401 enic->cq_count = 1 + m;
2402 enic->intr_count = 1 + m + 2;
2404 vnic_dev_set_intr_mode(enic->vdev,
2405 VNIC_DEV_INTR_MODE_MSIX);
2407 return 0;
2411 /* Next try MSI
2413 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2416 if (enic->config.intr_mode < 2 &&
2417 enic->rq_count >= 1 &&
2418 enic->wq_count >= 1 &&
2419 enic->cq_count >= 2 &&
2420 enic->intr_count >= 1 &&
2421 !pci_enable_msi(enic->pdev)) {
2423 enic->rq_count = 1;
2424 enic->wq_count = 1;
2425 enic->cq_count = 2;
2426 enic->intr_count = 1;
2428 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2430 return 0;
2433 /* Next try INTx
2435 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2436 * (the first INTR is used for WQ/RQ)
2437 * (the second INTR is used for WQ/RQ errors)
2438 * (the last INTR is used for notifications)
2441 if (enic->config.intr_mode < 3 &&
2442 enic->rq_count >= 1 &&
2443 enic->wq_count >= 1 &&
2444 enic->cq_count >= 2 &&
2445 enic->intr_count >= 3) {
2447 enic->rq_count = 1;
2448 enic->wq_count = 1;
2449 enic->cq_count = 2;
2450 enic->intr_count = 3;
2452 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2454 return 0;
2457 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2459 return -EINVAL;
2462 static void enic_clear_intr_mode(struct enic *enic)
2464 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2465 case VNIC_DEV_INTR_MODE_MSIX:
2466 pci_disable_msix(enic->pdev);
2467 break;
2468 case VNIC_DEV_INTR_MODE_MSI:
2469 pci_disable_msi(enic->pdev);
2470 break;
2471 default:
2472 break;
2475 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2478 static const struct net_device_ops enic_netdev_dynamic_ops = {
2479 .ndo_open = enic_open,
2480 .ndo_stop = enic_stop,
2481 .ndo_start_xmit = enic_hard_start_xmit,
2482 .ndo_get_stats64 = enic_get_stats,
2483 .ndo_validate_addr = eth_validate_addr,
2484 .ndo_set_rx_mode = enic_set_rx_mode,
2485 .ndo_set_mac_address = enic_set_mac_address_dynamic,
2486 .ndo_change_mtu = enic_change_mtu,
2487 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2488 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2489 .ndo_tx_timeout = enic_tx_timeout,
2490 .ndo_set_vf_port = enic_set_vf_port,
2491 .ndo_get_vf_port = enic_get_vf_port,
2492 .ndo_set_vf_mac = enic_set_vf_mac,
2493 #ifdef CONFIG_NET_POLL_CONTROLLER
2494 .ndo_poll_controller = enic_poll_controller,
2495 #endif
2496 #ifdef CONFIG_RFS_ACCEL
2497 .ndo_rx_flow_steer = enic_rx_flow_steer,
2498 #endif
2499 .ndo_udp_tunnel_add = enic_udp_tunnel_add,
2500 .ndo_udp_tunnel_del = enic_udp_tunnel_del,
2501 .ndo_features_check = enic_features_check,
2504 static const struct net_device_ops enic_netdev_ops = {
2505 .ndo_open = enic_open,
2506 .ndo_stop = enic_stop,
2507 .ndo_start_xmit = enic_hard_start_xmit,
2508 .ndo_get_stats64 = enic_get_stats,
2509 .ndo_validate_addr = eth_validate_addr,
2510 .ndo_set_mac_address = enic_set_mac_address,
2511 .ndo_set_rx_mode = enic_set_rx_mode,
2512 .ndo_change_mtu = enic_change_mtu,
2513 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2514 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2515 .ndo_tx_timeout = enic_tx_timeout,
2516 .ndo_set_vf_port = enic_set_vf_port,
2517 .ndo_get_vf_port = enic_get_vf_port,
2518 .ndo_set_vf_mac = enic_set_vf_mac,
2519 #ifdef CONFIG_NET_POLL_CONTROLLER
2520 .ndo_poll_controller = enic_poll_controller,
2521 #endif
2522 #ifdef CONFIG_RFS_ACCEL
2523 .ndo_rx_flow_steer = enic_rx_flow_steer,
2524 #endif
2525 .ndo_udp_tunnel_add = enic_udp_tunnel_add,
2526 .ndo_udp_tunnel_del = enic_udp_tunnel_del,
2527 .ndo_features_check = enic_features_check,
2530 static void enic_dev_deinit(struct enic *enic)
2532 unsigned int i;
2534 for (i = 0; i < enic->rq_count; i++) {
2535 napi_hash_del(&enic->napi[i]);
2536 netif_napi_del(&enic->napi[i]);
2538 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2539 for (i = 0; i < enic->wq_count; i++)
2540 netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2542 enic_free_vnic_resources(enic);
2543 enic_clear_intr_mode(enic);
2544 enic_free_affinity_hint(enic);
2547 static void enic_kdump_kernel_config(struct enic *enic)
2549 if (is_kdump_kernel()) {
2550 dev_info(enic_get_dev(enic), "Running from within kdump kernel. Using minimal resources\n");
2551 enic->rq_count = 1;
2552 enic->wq_count = 1;
2553 enic->config.rq_desc_count = ENIC_MIN_RQ_DESCS;
2554 enic->config.wq_desc_count = ENIC_MIN_WQ_DESCS;
2555 enic->config.mtu = min_t(u16, 1500, enic->config.mtu);
2559 static int enic_dev_init(struct enic *enic)
2561 struct device *dev = enic_get_dev(enic);
2562 struct net_device *netdev = enic->netdev;
2563 unsigned int i;
2564 int err;
2566 /* Get interrupt coalesce timer info */
2567 err = enic_dev_intr_coal_timer_info(enic);
2568 if (err) {
2569 dev_warn(dev, "Using default conversion factor for "
2570 "interrupt coalesce timer\n");
2571 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2574 /* Get vNIC configuration
2577 err = enic_get_vnic_config(enic);
2578 if (err) {
2579 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2580 return err;
2583 /* Get available resource counts
2586 enic_get_res_counts(enic);
2588 /* modify resource count if we are in kdump_kernel
2590 enic_kdump_kernel_config(enic);
2592 /* Set interrupt mode based on resource counts and system
2593 * capabilities
2596 err = enic_set_intr_mode(enic);
2597 if (err) {
2598 dev_err(dev, "Failed to set intr mode based on resource "
2599 "counts and system capabilities, aborting\n");
2600 return err;
2603 /* Allocate and configure vNIC resources
2606 err = enic_alloc_vnic_resources(enic);
2607 if (err) {
2608 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2609 goto err_out_free_vnic_resources;
2612 enic_init_vnic_resources(enic);
2614 err = enic_set_rss_nic_cfg(enic);
2615 if (err) {
2616 dev_err(dev, "Failed to config nic, aborting\n");
2617 goto err_out_free_vnic_resources;
2620 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2621 default:
2622 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2623 break;
2624 case VNIC_DEV_INTR_MODE_MSIX:
2625 for (i = 0; i < enic->rq_count; i++) {
2626 netif_napi_add(netdev, &enic->napi[i],
2627 enic_poll_msix_rq, NAPI_POLL_WEIGHT);
2629 for (i = 0; i < enic->wq_count; i++)
2630 netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],
2631 enic_poll_msix_wq, NAPI_POLL_WEIGHT);
2632 break;
2635 return 0;
2637 err_out_free_vnic_resources:
2638 enic_free_affinity_hint(enic);
2639 enic_clear_intr_mode(enic);
2640 enic_free_vnic_resources(enic);
2642 return err;
2645 static void enic_iounmap(struct enic *enic)
2647 unsigned int i;
2649 for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2650 if (enic->bar[i].vaddr)
2651 iounmap(enic->bar[i].vaddr);
2654 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2656 struct device *dev = &pdev->dev;
2657 struct net_device *netdev;
2658 struct enic *enic;
2659 int using_dac = 0;
2660 unsigned int i;
2661 int err;
2662 #ifdef CONFIG_PCI_IOV
2663 int pos = 0;
2664 #endif
2665 int num_pps = 1;
2667 /* Allocate net device structure and initialize. Private
2668 * instance data is initialized to zero.
2671 netdev = alloc_etherdev_mqs(sizeof(struct enic),
2672 ENIC_RQ_MAX, ENIC_WQ_MAX);
2673 if (!netdev)
2674 return -ENOMEM;
2676 pci_set_drvdata(pdev, netdev);
2678 SET_NETDEV_DEV(netdev, &pdev->dev);
2680 enic = netdev_priv(netdev);
2681 enic->netdev = netdev;
2682 enic->pdev = pdev;
2684 /* Setup PCI resources
2687 err = pci_enable_device_mem(pdev);
2688 if (err) {
2689 dev_err(dev, "Cannot enable PCI device, aborting\n");
2690 goto err_out_free_netdev;
2693 err = pci_request_regions(pdev, DRV_NAME);
2694 if (err) {
2695 dev_err(dev, "Cannot request PCI regions, aborting\n");
2696 goto err_out_disable_device;
2699 pci_set_master(pdev);
2701 /* Query PCI controller on system for DMA addressing
2702 * limitation for the device. Try 64-bit first, and
2703 * fail to 32-bit.
2706 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2707 if (err) {
2708 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2709 if (err) {
2710 dev_err(dev, "No usable DMA configuration, aborting\n");
2711 goto err_out_release_regions;
2713 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2714 if (err) {
2715 dev_err(dev, "Unable to obtain %u-bit DMA "
2716 "for consistent allocations, aborting\n", 32);
2717 goto err_out_release_regions;
2719 } else {
2720 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2721 if (err) {
2722 dev_err(dev, "Unable to obtain %u-bit DMA "
2723 "for consistent allocations, aborting\n", 64);
2724 goto err_out_release_regions;
2726 using_dac = 1;
2729 /* Map vNIC resources from BAR0-5
2732 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2733 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2734 continue;
2735 enic->bar[i].len = pci_resource_len(pdev, i);
2736 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2737 if (!enic->bar[i].vaddr) {
2738 dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2739 err = -ENODEV;
2740 goto err_out_iounmap;
2742 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2745 /* Register vNIC device
2748 enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2749 ARRAY_SIZE(enic->bar));
2750 if (!enic->vdev) {
2751 dev_err(dev, "vNIC registration failed, aborting\n");
2752 err = -ENODEV;
2753 goto err_out_iounmap;
2756 err = vnic_devcmd_init(enic->vdev);
2758 if (err)
2759 goto err_out_vnic_unregister;
2761 #ifdef CONFIG_PCI_IOV
2762 /* Get number of subvnics */
2763 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2764 if (pos) {
2765 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2766 &enic->num_vfs);
2767 if (enic->num_vfs) {
2768 err = pci_enable_sriov(pdev, enic->num_vfs);
2769 if (err) {
2770 dev_err(dev, "SRIOV enable failed, aborting."
2771 " pci_enable_sriov() returned %d\n",
2772 err);
2773 goto err_out_vnic_unregister;
2775 enic->priv_flags |= ENIC_SRIOV_ENABLED;
2776 num_pps = enic->num_vfs;
2779 #endif
2781 /* Allocate structure for port profiles */
2782 enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2783 if (!enic->pp) {
2784 err = -ENOMEM;
2785 goto err_out_disable_sriov_pp;
2788 /* Issue device open to get device in known state
2791 err = enic_dev_open(enic);
2792 if (err) {
2793 dev_err(dev, "vNIC dev open failed, aborting\n");
2794 goto err_out_disable_sriov;
2797 /* Setup devcmd lock
2800 spin_lock_init(&enic->devcmd_lock);
2801 spin_lock_init(&enic->enic_api_lock);
2804 * Set ingress vlan rewrite mode before vnic initialization
2807 err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2808 if (err) {
2809 dev_err(dev,
2810 "Failed to set ingress vlan rewrite mode, aborting.\n");
2811 goto err_out_dev_close;
2814 /* Issue device init to initialize the vnic-to-switch link.
2815 * We'll start with carrier off and wait for link UP
2816 * notification later to turn on carrier. We don't need
2817 * to wait here for the vnic-to-switch link initialization
2818 * to complete; link UP notification is the indication that
2819 * the process is complete.
2822 netif_carrier_off(netdev);
2824 /* Do not call dev_init for a dynamic vnic.
2825 * For a dynamic vnic, init_prov_info will be
2826 * called later by an upper layer.
2829 if (!enic_is_dynamic(enic)) {
2830 err = vnic_dev_init(enic->vdev, 0);
2831 if (err) {
2832 dev_err(dev, "vNIC dev init failed, aborting\n");
2833 goto err_out_dev_close;
2837 err = enic_dev_init(enic);
2838 if (err) {
2839 dev_err(dev, "Device initialization failed, aborting\n");
2840 goto err_out_dev_close;
2843 netif_set_real_num_tx_queues(netdev, enic->wq_count);
2844 netif_set_real_num_rx_queues(netdev, enic->rq_count);
2846 /* Setup notification timer, HW reset task, and wq locks
2849 init_timer(&enic->notify_timer);
2850 enic->notify_timer.function = enic_notify_timer;
2851 enic->notify_timer.data = (unsigned long)enic;
2853 enic_set_rx_coal_setting(enic);
2854 INIT_WORK(&enic->reset, enic_reset);
2855 INIT_WORK(&enic->tx_hang_reset, enic_tx_hang_reset);
2856 INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2858 for (i = 0; i < enic->wq_count; i++)
2859 spin_lock_init(&enic->wq_lock[i]);
2861 /* Register net device
2864 enic->port_mtu = enic->config.mtu;
2865 (void)enic_change_mtu(netdev, enic->port_mtu);
2867 err = enic_set_mac_addr(netdev, enic->mac_addr);
2868 if (err) {
2869 dev_err(dev, "Invalid MAC address, aborting\n");
2870 goto err_out_dev_deinit;
2873 enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2874 /* rx coalesce time already got initialized. This gets used
2875 * if adaptive coal is turned off
2877 enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2879 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2880 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2881 else
2882 netdev->netdev_ops = &enic_netdev_ops;
2884 netdev->watchdog_timeo = 2 * HZ;
2885 enic_set_ethtool_ops(netdev);
2887 netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2888 if (ENIC_SETTING(enic, LOOP)) {
2889 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2890 enic->loop_enable = 1;
2891 enic->loop_tag = enic->config.loop_tag;
2892 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2894 if (ENIC_SETTING(enic, TXCSUM))
2895 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2896 if (ENIC_SETTING(enic, TSO))
2897 netdev->hw_features |= NETIF_F_TSO |
2898 NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2899 if (ENIC_SETTING(enic, RSS))
2900 netdev->hw_features |= NETIF_F_RXHASH;
2901 if (ENIC_SETTING(enic, RXCSUM))
2902 netdev->hw_features |= NETIF_F_RXCSUM;
2903 if (ENIC_SETTING(enic, VXLAN)) {
2904 u64 patch_level;
2906 netdev->hw_enc_features |= NETIF_F_RXCSUM |
2907 NETIF_F_TSO |
2908 NETIF_F_TSO_ECN |
2909 NETIF_F_GSO_UDP_TUNNEL |
2910 NETIF_F_HW_CSUM |
2911 NETIF_F_GSO_UDP_TUNNEL_CSUM;
2912 netdev->hw_features |= netdev->hw_enc_features;
2913 /* get bit mask from hw about supported offload bit level
2914 * BIT(0) = fw supports patch_level 0
2915 * fcoe bit = encap
2916 * fcoe_fc_crc_ok = outer csum ok
2917 * BIT(1) = always set by fw
2918 * BIT(2) = fw supports patch_level 2
2919 * BIT(0) in rss_hash = encap
2920 * BIT(1,2) in rss_hash = outer_ip_csum_ok/
2921 * outer_tcp_csum_ok
2922 * used in enic_rq_indicate_buf
2924 err = vnic_dev_get_supported_feature_ver(enic->vdev,
2925 VIC_FEATURE_VXLAN,
2926 &patch_level);
2927 if (err)
2928 patch_level = 0;
2929 /* mask bits that are supported by driver
2931 patch_level &= BIT_ULL(0) | BIT_ULL(2);
2932 patch_level = fls(patch_level);
2933 patch_level = patch_level ? patch_level - 1 : 0;
2934 enic->vxlan.patch_level = patch_level;
2937 netdev->features |= netdev->hw_features;
2938 netdev->vlan_features |= netdev->features;
2940 #ifdef CONFIG_RFS_ACCEL
2941 netdev->hw_features |= NETIF_F_NTUPLE;
2942 #endif
2944 if (using_dac)
2945 netdev->features |= NETIF_F_HIGHDMA;
2947 netdev->priv_flags |= IFF_UNICAST_FLT;
2949 /* MTU range: 68 - 9000 */
2950 netdev->min_mtu = ENIC_MIN_MTU;
2951 netdev->max_mtu = ENIC_MAX_MTU;
2953 err = register_netdev(netdev);
2954 if (err) {
2955 dev_err(dev, "Cannot register net device, aborting\n");
2956 goto err_out_dev_deinit;
2958 enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
2960 return 0;
2962 err_out_dev_deinit:
2963 enic_dev_deinit(enic);
2964 err_out_dev_close:
2965 vnic_dev_close(enic->vdev);
2966 err_out_disable_sriov:
2967 kfree(enic->pp);
2968 err_out_disable_sriov_pp:
2969 #ifdef CONFIG_PCI_IOV
2970 if (enic_sriov_enabled(enic)) {
2971 pci_disable_sriov(pdev);
2972 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2974 #endif
2975 err_out_vnic_unregister:
2976 vnic_dev_unregister(enic->vdev);
2977 err_out_iounmap:
2978 enic_iounmap(enic);
2979 err_out_release_regions:
2980 pci_release_regions(pdev);
2981 err_out_disable_device:
2982 pci_disable_device(pdev);
2983 err_out_free_netdev:
2984 free_netdev(netdev);
2986 return err;
2989 static void enic_remove(struct pci_dev *pdev)
2991 struct net_device *netdev = pci_get_drvdata(pdev);
2993 if (netdev) {
2994 struct enic *enic = netdev_priv(netdev);
2996 cancel_work_sync(&enic->reset);
2997 cancel_work_sync(&enic->change_mtu_work);
2998 unregister_netdev(netdev);
2999 enic_dev_deinit(enic);
3000 vnic_dev_close(enic->vdev);
3001 #ifdef CONFIG_PCI_IOV
3002 if (enic_sriov_enabled(enic)) {
3003 pci_disable_sriov(pdev);
3004 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
3006 #endif
3007 kfree(enic->pp);
3008 vnic_dev_unregister(enic->vdev);
3009 enic_iounmap(enic);
3010 pci_release_regions(pdev);
3011 pci_disable_device(pdev);
3012 free_netdev(netdev);
3016 static struct pci_driver enic_driver = {
3017 .name = DRV_NAME,
3018 .id_table = enic_id_table,
3019 .probe = enic_probe,
3020 .remove = enic_remove,
3023 static int __init enic_init_module(void)
3025 pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
3027 return pci_register_driver(&enic_driver);
3030 static void __exit enic_cleanup_module(void)
3032 pci_unregister_driver(&enic_driver);
3035 module_init(enic_init_module);
3036 module_exit(enic_cleanup_module);