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>
43 #include "cq_enet_desc.h"
45 #include "vnic_intr.h"
46 #include "vnic_stats.h"
53 #define ENIC_NOTIFY_TIMER_PERIOD (2 * HZ)
54 #define WQ_ENET_MAX_DESC_LEN (1 << WQ_ENET_LEN_BITS)
55 #define MAX_TSO (1 << 16)
56 #define ENIC_DESC_MAX_SPLITS (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
58 #define PCI_DEVICE_ID_CISCO_VIC_ENET 0x0043 /* ethernet vnic */
59 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN 0x0044 /* enet dynamic vnic */
60 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF 0x0071 /* enet SRIOV VF */
62 /* Supported devices */
63 static DEFINE_PCI_DEVICE_TABLE(enic_id_table) = {
64 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
65 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
66 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
67 { 0, } /* end of table */
70 MODULE_DESCRIPTION(DRV_DESCRIPTION);
71 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
72 MODULE_LICENSE("GPL");
73 MODULE_VERSION(DRV_VERSION);
74 MODULE_DEVICE_TABLE(pci, enic_id_table);
76 #define ENIC_LARGE_PKT_THRESHOLD 1000
77 #define ENIC_MAX_COALESCE_TIMERS 10
78 /* Interrupt moderation table, which will be used to decide the
79 * coalescing timer values
80 * {rx_rate in Mbps, mapping percentage of the range}
82 struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
96 /* This table helps the driver to pick different ranges for rx coalescing
97 * timer depending on the link speed.
99 struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = {
100 {0, 0}, /* 0 - 4 Gbps */
101 {0, 3}, /* 4 - 10 Gbps */
102 {3, 6}, /* 10 - 40 Gbps */
105 int enic_is_dynamic(struct enic *enic)
107 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
110 int enic_sriov_enabled(struct enic *enic)
112 return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
115 static int enic_is_sriov_vf(struct enic *enic)
117 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
120 int enic_is_valid_vf(struct enic *enic, int vf)
122 #ifdef CONFIG_PCI_IOV
123 return vf >= 0 && vf < enic->num_vfs;
129 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
131 struct enic *enic = vnic_dev_priv(wq->vdev);
134 pci_unmap_single(enic->pdev, buf->dma_addr,
135 buf->len, PCI_DMA_TODEVICE);
137 pci_unmap_page(enic->pdev, buf->dma_addr,
138 buf->len, PCI_DMA_TODEVICE);
141 dev_kfree_skb_any(buf->os_buf);
144 static void enic_wq_free_buf(struct vnic_wq *wq,
145 struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
147 enic_free_wq_buf(wq, buf);
150 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
151 u8 type, u16 q_number, u16 completed_index, void *opaque)
153 struct enic *enic = vnic_dev_priv(vdev);
155 spin_lock(&enic->wq_lock[q_number]);
157 vnic_wq_service(&enic->wq[q_number], cq_desc,
158 completed_index, enic_wq_free_buf,
161 if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
162 vnic_wq_desc_avail(&enic->wq[q_number]) >=
163 (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
164 netif_wake_subqueue(enic->netdev, q_number);
166 spin_unlock(&enic->wq_lock[q_number]);
171 static void enic_log_q_error(struct enic *enic)
176 for (i = 0; i < enic->wq_count; i++) {
177 error_status = vnic_wq_error_status(&enic->wq[i]);
179 netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
183 for (i = 0; i < enic->rq_count; i++) {
184 error_status = vnic_rq_error_status(&enic->rq[i]);
186 netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
191 static void enic_msglvl_check(struct enic *enic)
193 u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
195 if (msg_enable != enic->msg_enable) {
196 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
197 enic->msg_enable, msg_enable);
198 enic->msg_enable = msg_enable;
202 static void enic_mtu_check(struct enic *enic)
204 u32 mtu = vnic_dev_mtu(enic->vdev);
205 struct net_device *netdev = enic->netdev;
207 if (mtu && mtu != enic->port_mtu) {
208 enic->port_mtu = mtu;
209 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
210 mtu = max_t(int, ENIC_MIN_MTU,
211 min_t(int, ENIC_MAX_MTU, mtu));
212 if (mtu != netdev->mtu)
213 schedule_work(&enic->change_mtu_work);
215 if (mtu < netdev->mtu)
217 "interface MTU (%d) set higher "
218 "than switch port MTU (%d)\n",
224 static void enic_link_check(struct enic *enic)
226 int link_status = vnic_dev_link_status(enic->vdev);
227 int carrier_ok = netif_carrier_ok(enic->netdev);
229 if (link_status && !carrier_ok) {
230 netdev_info(enic->netdev, "Link UP\n");
231 netif_carrier_on(enic->netdev);
232 } else if (!link_status && carrier_ok) {
233 netdev_info(enic->netdev, "Link DOWN\n");
234 netif_carrier_off(enic->netdev);
238 static void enic_notify_check(struct enic *enic)
240 enic_msglvl_check(enic);
241 enic_mtu_check(enic);
242 enic_link_check(enic);
245 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
247 static irqreturn_t enic_isr_legacy(int irq, void *data)
249 struct net_device *netdev = data;
250 struct enic *enic = netdev_priv(netdev);
251 unsigned int io_intr = enic_legacy_io_intr();
252 unsigned int err_intr = enic_legacy_err_intr();
253 unsigned int notify_intr = enic_legacy_notify_intr();
256 vnic_intr_mask(&enic->intr[io_intr]);
258 pba = vnic_intr_legacy_pba(enic->legacy_pba);
260 vnic_intr_unmask(&enic->intr[io_intr]);
261 return IRQ_NONE; /* not our interrupt */
264 if (ENIC_TEST_INTR(pba, notify_intr)) {
265 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
266 enic_notify_check(enic);
269 if (ENIC_TEST_INTR(pba, err_intr)) {
270 vnic_intr_return_all_credits(&enic->intr[err_intr]);
271 enic_log_q_error(enic);
272 /* schedule recovery from WQ/RQ error */
273 schedule_work(&enic->reset);
277 if (ENIC_TEST_INTR(pba, io_intr)) {
278 if (napi_schedule_prep(&enic->napi[0]))
279 __napi_schedule(&enic->napi[0]);
281 vnic_intr_unmask(&enic->intr[io_intr]);
287 static irqreturn_t enic_isr_msi(int irq, void *data)
289 struct enic *enic = data;
291 /* With MSI, there is no sharing of interrupts, so this is
292 * our interrupt and there is no need to ack it. The device
293 * is not providing per-vector masking, so the OS will not
294 * write to PCI config space to mask/unmask the interrupt.
295 * We're using mask_on_assertion for MSI, so the device
296 * automatically masks the interrupt when the interrupt is
297 * generated. Later, when exiting polling, the interrupt
298 * will be unmasked (see enic_poll).
300 * Also, the device uses the same PCIe Traffic Class (TC)
301 * for Memory Write data and MSI, so there are no ordering
302 * issues; the MSI will always arrive at the Root Complex
303 * _after_ corresponding Memory Writes (i.e. descriptor
307 napi_schedule(&enic->napi[0]);
312 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
314 struct napi_struct *napi = data;
316 /* schedule NAPI polling for RQ cleanup */
322 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
324 struct enic *enic = data;
327 unsigned int wq_work_to_do = -1; /* no limit */
328 unsigned int wq_work_done;
331 wq_irq = (u32)irq - enic->msix_entry[enic_msix_wq_intr(enic, 0)].vector;
332 cq = enic_cq_wq(enic, wq_irq);
333 intr = enic_msix_wq_intr(enic, wq_irq);
335 wq_work_done = vnic_cq_service(&enic->cq[cq],
336 wq_work_to_do, enic_wq_service, NULL);
338 vnic_intr_return_credits(&enic->intr[intr],
341 1 /* reset intr timer */);
346 static irqreturn_t enic_isr_msix_err(int irq, void *data)
348 struct enic *enic = data;
349 unsigned int intr = enic_msix_err_intr(enic);
351 vnic_intr_return_all_credits(&enic->intr[intr]);
353 enic_log_q_error(enic);
355 /* schedule recovery from WQ/RQ error */
356 schedule_work(&enic->reset);
361 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
363 struct enic *enic = data;
364 unsigned int intr = enic_msix_notify_intr(enic);
366 vnic_intr_return_all_credits(&enic->intr[intr]);
367 enic_notify_check(enic);
372 static inline void enic_queue_wq_skb_cont(struct enic *enic,
373 struct vnic_wq *wq, struct sk_buff *skb,
374 unsigned int len_left, int loopback)
376 const skb_frag_t *frag;
378 /* Queue additional data fragments */
379 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
380 len_left -= skb_frag_size(frag);
381 enic_queue_wq_desc_cont(wq, skb,
382 skb_frag_dma_map(&enic->pdev->dev,
383 frag, 0, skb_frag_size(frag),
386 (len_left == 0), /* EOP? */
391 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
392 struct vnic_wq *wq, struct sk_buff *skb,
393 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
395 unsigned int head_len = skb_headlen(skb);
396 unsigned int len_left = skb->len - head_len;
397 int eop = (len_left == 0);
399 /* Queue the main skb fragment. The fragments are no larger
400 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
401 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
402 * per fragment is queued.
404 enic_queue_wq_desc(wq, skb,
405 pci_map_single(enic->pdev, skb->data,
406 head_len, PCI_DMA_TODEVICE),
408 vlan_tag_insert, vlan_tag,
412 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
415 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
416 struct vnic_wq *wq, struct sk_buff *skb,
417 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
419 unsigned int head_len = skb_headlen(skb);
420 unsigned int len_left = skb->len - head_len;
421 unsigned int hdr_len = skb_checksum_start_offset(skb);
422 unsigned int csum_offset = hdr_len + skb->csum_offset;
423 int eop = (len_left == 0);
425 /* Queue the main skb fragment. The fragments are no larger
426 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
427 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
428 * per fragment is queued.
430 enic_queue_wq_desc_csum_l4(wq, skb,
431 pci_map_single(enic->pdev, skb->data,
432 head_len, PCI_DMA_TODEVICE),
436 vlan_tag_insert, vlan_tag,
440 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
443 static inline void enic_queue_wq_skb_tso(struct enic *enic,
444 struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
445 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
447 unsigned int frag_len_left = skb_headlen(skb);
448 unsigned int len_left = skb->len - frag_len_left;
449 unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
450 int eop = (len_left == 0);
453 unsigned int offset = 0;
456 /* Preload TCP csum field with IP pseudo hdr calculated
457 * with IP length set to zero. HW will later add in length
458 * to each TCP segment resulting from the TSO.
461 if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
462 ip_hdr(skb)->check = 0;
463 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
464 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
465 } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
466 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
467 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
470 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
471 * for the main skb fragment
473 while (frag_len_left) {
474 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
475 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
476 len, PCI_DMA_TODEVICE);
477 enic_queue_wq_desc_tso(wq, skb,
481 vlan_tag_insert, vlan_tag,
482 eop && (len == frag_len_left), loopback);
483 frag_len_left -= len;
490 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
491 * for additional data fragments
493 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
494 len_left -= skb_frag_size(frag);
495 frag_len_left = skb_frag_size(frag);
498 while (frag_len_left) {
499 len = min(frag_len_left,
500 (unsigned int)WQ_ENET_MAX_DESC_LEN);
501 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
504 enic_queue_wq_desc_cont(wq, skb,
508 (len == frag_len_left), /* EOP? */
510 frag_len_left -= len;
516 static inline void enic_queue_wq_skb(struct enic *enic,
517 struct vnic_wq *wq, struct sk_buff *skb)
519 unsigned int mss = skb_shinfo(skb)->gso_size;
520 unsigned int vlan_tag = 0;
521 int vlan_tag_insert = 0;
524 if (vlan_tx_tag_present(skb)) {
525 /* VLAN tag from trunking driver */
527 vlan_tag = vlan_tx_tag_get(skb);
528 } else if (enic->loop_enable) {
529 vlan_tag = enic->loop_tag;
534 enic_queue_wq_skb_tso(enic, wq, skb, mss,
535 vlan_tag_insert, vlan_tag, loopback);
536 else if (skb->ip_summed == CHECKSUM_PARTIAL)
537 enic_queue_wq_skb_csum_l4(enic, wq, skb,
538 vlan_tag_insert, vlan_tag, loopback);
540 enic_queue_wq_skb_vlan(enic, wq, skb,
541 vlan_tag_insert, vlan_tag, loopback);
544 /* netif_tx_lock held, process context with BHs disabled, or BH */
545 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
546 struct net_device *netdev)
548 struct enic *enic = netdev_priv(netdev);
551 unsigned int txq_map;
554 dev_kfree_skb_any(skb);
558 txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
559 wq = &enic->wq[txq_map];
561 /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
562 * which is very likely. In the off chance it's going to take
563 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
566 if (skb_shinfo(skb)->gso_size == 0 &&
567 skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
568 skb_linearize(skb)) {
569 dev_kfree_skb_any(skb);
573 spin_lock_irqsave(&enic->wq_lock[txq_map], flags);
575 if (vnic_wq_desc_avail(wq) <
576 skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
577 netif_tx_stop_queue(netdev_get_tx_queue(netdev, txq_map));
578 /* This is a hard error, log it */
579 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
580 spin_unlock_irqrestore(&enic->wq_lock[txq_map], flags);
581 return NETDEV_TX_BUSY;
584 enic_queue_wq_skb(enic, wq, skb);
586 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
587 netif_tx_stop_queue(netdev_get_tx_queue(netdev, txq_map));
589 spin_unlock_irqrestore(&enic->wq_lock[txq_map], flags);
594 /* dev_base_lock rwlock held, nominally process context */
595 static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
596 struct rtnl_link_stats64 *net_stats)
598 struct enic *enic = netdev_priv(netdev);
599 struct vnic_stats *stats;
601 enic_dev_stats_dump(enic, &stats);
603 net_stats->tx_packets = stats->tx.tx_frames_ok;
604 net_stats->tx_bytes = stats->tx.tx_bytes_ok;
605 net_stats->tx_errors = stats->tx.tx_errors;
606 net_stats->tx_dropped = stats->tx.tx_drops;
608 net_stats->rx_packets = stats->rx.rx_frames_ok;
609 net_stats->rx_bytes = stats->rx.rx_bytes_ok;
610 net_stats->rx_errors = stats->rx.rx_errors;
611 net_stats->multicast = stats->rx.rx_multicast_frames_ok;
612 net_stats->rx_over_errors = enic->rq_truncated_pkts;
613 net_stats->rx_crc_errors = enic->rq_bad_fcs;
614 net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
619 void enic_reset_addr_lists(struct enic *enic)
626 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
628 struct enic *enic = netdev_priv(netdev);
630 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
631 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
632 return -EADDRNOTAVAIL;
634 if (!is_valid_ether_addr(addr))
635 return -EADDRNOTAVAIL;
638 memcpy(netdev->dev_addr, addr, netdev->addr_len);
643 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
645 struct enic *enic = netdev_priv(netdev);
646 struct sockaddr *saddr = p;
647 char *addr = saddr->sa_data;
650 if (netif_running(enic->netdev)) {
651 err = enic_dev_del_station_addr(enic);
656 err = enic_set_mac_addr(netdev, addr);
660 if (netif_running(enic->netdev)) {
661 err = enic_dev_add_station_addr(enic);
669 static int enic_set_mac_address(struct net_device *netdev, void *p)
671 struct sockaddr *saddr = p;
672 char *addr = saddr->sa_data;
673 struct enic *enic = netdev_priv(netdev);
676 err = enic_dev_del_station_addr(enic);
680 err = enic_set_mac_addr(netdev, addr);
684 return enic_dev_add_station_addr(enic);
687 static void enic_update_multicast_addr_list(struct enic *enic)
689 struct net_device *netdev = enic->netdev;
690 struct netdev_hw_addr *ha;
691 unsigned int mc_count = netdev_mc_count(netdev);
692 u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
695 if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS) {
696 netdev_warn(netdev, "Registering only %d out of %d "
697 "multicast addresses\n",
698 ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
699 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
702 /* Is there an easier way? Trying to minimize to
703 * calls to add/del multicast addrs. We keep the
704 * addrs from the last call in enic->mc_addr and
705 * look for changes to add/del.
709 netdev_for_each_mc_addr(ha, netdev) {
712 memcpy(mc_addr[i++], ha->addr, ETH_ALEN);
715 for (i = 0; i < enic->mc_count; i++) {
716 for (j = 0; j < mc_count; j++)
717 if (ether_addr_equal(enic->mc_addr[i], mc_addr[j]))
720 enic_dev_del_addr(enic, enic->mc_addr[i]);
723 for (i = 0; i < mc_count; i++) {
724 for (j = 0; j < enic->mc_count; j++)
725 if (ether_addr_equal(mc_addr[i], enic->mc_addr[j]))
727 if (j == enic->mc_count)
728 enic_dev_add_addr(enic, mc_addr[i]);
731 /* Save the list to compare against next time
734 for (i = 0; i < mc_count; i++)
735 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
737 enic->mc_count = mc_count;
740 static void enic_update_unicast_addr_list(struct enic *enic)
742 struct net_device *netdev = enic->netdev;
743 struct netdev_hw_addr *ha;
744 unsigned int uc_count = netdev_uc_count(netdev);
745 u8 uc_addr[ENIC_UNICAST_PERFECT_FILTERS][ETH_ALEN];
748 if (uc_count > ENIC_UNICAST_PERFECT_FILTERS) {
749 netdev_warn(netdev, "Registering only %d out of %d "
750 "unicast addresses\n",
751 ENIC_UNICAST_PERFECT_FILTERS, uc_count);
752 uc_count = ENIC_UNICAST_PERFECT_FILTERS;
755 /* Is there an easier way? Trying to minimize to
756 * calls to add/del unicast addrs. We keep the
757 * addrs from the last call in enic->uc_addr and
758 * look for changes to add/del.
762 netdev_for_each_uc_addr(ha, netdev) {
765 memcpy(uc_addr[i++], ha->addr, ETH_ALEN);
768 for (i = 0; i < enic->uc_count; i++) {
769 for (j = 0; j < uc_count; j++)
770 if (ether_addr_equal(enic->uc_addr[i], uc_addr[j]))
773 enic_dev_del_addr(enic, enic->uc_addr[i]);
776 for (i = 0; i < uc_count; i++) {
777 for (j = 0; j < enic->uc_count; j++)
778 if (ether_addr_equal(uc_addr[i], enic->uc_addr[j]))
780 if (j == enic->uc_count)
781 enic_dev_add_addr(enic, uc_addr[i]);
784 /* Save the list to compare against next time
787 for (i = 0; i < uc_count; i++)
788 memcpy(enic->uc_addr[i], uc_addr[i], ETH_ALEN);
790 enic->uc_count = uc_count;
793 /* netif_tx_lock held, BHs disabled */
794 static void enic_set_rx_mode(struct net_device *netdev)
796 struct enic *enic = netdev_priv(netdev);
798 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
799 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
800 int promisc = (netdev->flags & IFF_PROMISC) ||
801 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
802 int allmulti = (netdev->flags & IFF_ALLMULTI) ||
803 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
804 unsigned int flags = netdev->flags |
805 (allmulti ? IFF_ALLMULTI : 0) |
806 (promisc ? IFF_PROMISC : 0);
808 if (enic->flags != flags) {
810 enic_dev_packet_filter(enic, directed,
811 multicast, broadcast, promisc, allmulti);
815 enic_update_unicast_addr_list(enic);
817 enic_update_multicast_addr_list(enic);
821 /* netif_tx_lock held, BHs disabled */
822 static void enic_tx_timeout(struct net_device *netdev)
824 struct enic *enic = netdev_priv(netdev);
825 schedule_work(&enic->reset);
828 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
830 struct enic *enic = netdev_priv(netdev);
831 struct enic_port_profile *pp;
834 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
838 if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
839 if (vf == PORT_SELF_VF) {
840 memcpy(pp->vf_mac, mac, ETH_ALEN);
844 * For sriov vf's set the mac in hw
846 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
847 vnic_dev_set_mac_addr, mac);
848 return enic_dev_status_to_errno(err);
854 static int enic_set_vf_port(struct net_device *netdev, int vf,
855 struct nlattr *port[])
857 struct enic *enic = netdev_priv(netdev);
858 struct enic_port_profile prev_pp;
859 struct enic_port_profile *pp;
860 int err = 0, restore_pp = 1;
862 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
866 if (!port[IFLA_PORT_REQUEST])
869 memcpy(&prev_pp, pp, sizeof(*enic->pp));
870 memset(pp, 0, sizeof(*enic->pp));
872 pp->set |= ENIC_SET_REQUEST;
873 pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
875 if (port[IFLA_PORT_PROFILE]) {
876 pp->set |= ENIC_SET_NAME;
877 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
881 if (port[IFLA_PORT_INSTANCE_UUID]) {
882 pp->set |= ENIC_SET_INSTANCE;
883 memcpy(pp->instance_uuid,
884 nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
887 if (port[IFLA_PORT_HOST_UUID]) {
888 pp->set |= ENIC_SET_HOST;
889 memcpy(pp->host_uuid,
890 nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
893 if (vf == PORT_SELF_VF) {
894 /* Special case handling: mac came from IFLA_VF_MAC */
895 if (!is_zero_ether_addr(prev_pp.vf_mac))
896 memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
898 if (is_zero_ether_addr(netdev->dev_addr))
899 eth_hw_addr_random(netdev);
901 /* SR-IOV VF: get mac from adapter */
902 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
903 vnic_dev_get_mac_addr, pp->mac_addr);
905 netdev_err(netdev, "Error getting mac for vf %d\n", vf);
906 memcpy(pp, &prev_pp, sizeof(*pp));
907 return enic_dev_status_to_errno(err);
911 err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
914 /* Things are still the way they were: Implicit
915 * DISASSOCIATE failed
917 memcpy(pp, &prev_pp, sizeof(*pp));
919 memset(pp, 0, sizeof(*pp));
920 if (vf == PORT_SELF_VF)
921 memset(netdev->dev_addr, 0, ETH_ALEN);
924 /* Set flag to indicate that the port assoc/disassoc
925 * request has been sent out to fw
927 pp->set |= ENIC_PORT_REQUEST_APPLIED;
929 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
930 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
931 memset(pp->mac_addr, 0, ETH_ALEN);
932 if (vf == PORT_SELF_VF)
933 memset(netdev->dev_addr, 0, ETH_ALEN);
937 if (vf == PORT_SELF_VF)
938 memset(pp->vf_mac, 0, ETH_ALEN);
943 static int enic_get_vf_port(struct net_device *netdev, int vf,
946 struct enic *enic = netdev_priv(netdev);
947 u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
948 struct enic_port_profile *pp;
951 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
955 if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
958 err = enic_process_get_pp_request(enic, vf, pp->request, &response);
962 if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
963 nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
964 ((pp->set & ENIC_SET_NAME) &&
965 nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
966 ((pp->set & ENIC_SET_INSTANCE) &&
967 nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
968 pp->instance_uuid)) ||
969 ((pp->set & ENIC_SET_HOST) &&
970 nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
971 goto nla_put_failure;
978 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
980 struct enic *enic = vnic_dev_priv(rq->vdev);
985 pci_unmap_single(enic->pdev, buf->dma_addr,
986 buf->len, PCI_DMA_FROMDEVICE);
987 dev_kfree_skb_any(buf->os_buf);
990 static int enic_rq_alloc_buf(struct vnic_rq *rq)
992 struct enic *enic = vnic_dev_priv(rq->vdev);
993 struct net_device *netdev = enic->netdev;
995 unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
996 unsigned int os_buf_index = 0;
999 skb = netdev_alloc_skb_ip_align(netdev, len);
1003 dma_addr = pci_map_single(enic->pdev, skb->data,
1004 len, PCI_DMA_FROMDEVICE);
1006 enic_queue_rq_desc(rq, skb, os_buf_index,
1012 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
1015 if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
1016 pkt_size->large_pkt_bytes_cnt += pkt_len;
1018 pkt_size->small_pkt_bytes_cnt += pkt_len;
1021 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1022 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1023 int skipped, void *opaque)
1025 struct enic *enic = vnic_dev_priv(rq->vdev);
1026 struct net_device *netdev = enic->netdev;
1027 struct sk_buff *skb;
1028 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1030 u8 type, color, eop, sop, ingress_port, vlan_stripped;
1031 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1032 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1033 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1035 u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1042 prefetch(skb->data - NET_IP_ALIGN);
1043 pci_unmap_single(enic->pdev, buf->dma_addr,
1044 buf->len, PCI_DMA_FROMDEVICE);
1046 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1047 &type, &color, &q_number, &completed_index,
1048 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1049 &csum_not_calc, &rss_hash, &bytes_written,
1050 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1051 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1052 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1053 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1059 if (bytes_written > 0)
1061 else if (bytes_written == 0)
1062 enic->rq_truncated_pkts++;
1065 dev_kfree_skb_any(skb);
1070 if (eop && bytes_written > 0) {
1075 skb_put(skb, bytes_written);
1076 skb->protocol = eth_type_trans(skb, netdev);
1077 skb_record_rx_queue(skb, q_number);
1078 if (netdev->features & NETIF_F_RXHASH) {
1079 skb_set_hash(skb, rss_hash,
1081 (NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX |
1082 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6 |
1083 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4)) ?
1084 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1087 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc) {
1088 skb->csum = htons(checksum);
1089 skb->ip_summed = CHECKSUM_COMPLETE;
1093 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1095 if (netdev->features & NETIF_F_GRO)
1096 napi_gro_receive(&enic->napi[q_number], skb);
1098 netif_receive_skb(skb);
1099 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1100 enic_intr_update_pkt_size(&cq->pkt_size_counter,
1107 dev_kfree_skb_any(skb);
1111 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1112 u8 type, u16 q_number, u16 completed_index, void *opaque)
1114 struct enic *enic = vnic_dev_priv(vdev);
1116 vnic_rq_service(&enic->rq[q_number], cq_desc,
1117 completed_index, VNIC_RQ_RETURN_DESC,
1118 enic_rq_indicate_buf, opaque);
1123 static int enic_poll(struct napi_struct *napi, int budget)
1125 struct net_device *netdev = napi->dev;
1126 struct enic *enic = netdev_priv(netdev);
1127 unsigned int cq_rq = enic_cq_rq(enic, 0);
1128 unsigned int cq_wq = enic_cq_wq(enic, 0);
1129 unsigned int intr = enic_legacy_io_intr();
1130 unsigned int rq_work_to_do = budget;
1131 unsigned int wq_work_to_do = -1; /* no limit */
1132 unsigned int work_done, rq_work_done = 0, wq_work_done;
1135 /* Service RQ (first) and WQ
1139 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1140 rq_work_to_do, enic_rq_service, NULL);
1142 wq_work_done = vnic_cq_service(&enic->cq[cq_wq],
1143 wq_work_to_do, enic_wq_service, NULL);
1145 /* Accumulate intr event credits for this polling
1146 * cycle. An intr event is the completion of a
1147 * a WQ or RQ packet.
1150 work_done = rq_work_done + wq_work_done;
1153 vnic_intr_return_credits(&enic->intr[intr],
1155 0 /* don't unmask intr */,
1156 0 /* don't reset intr timer */);
1158 err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1160 /* Buffer allocation failed. Stay in polling
1161 * mode so we can try to fill the ring again.
1165 rq_work_done = rq_work_to_do;
1167 if (rq_work_done < rq_work_to_do) {
1169 /* Some work done, but not enough to stay in polling,
1173 napi_complete(napi);
1174 vnic_intr_unmask(&enic->intr[intr]);
1177 return rq_work_done;
1180 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1182 unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1183 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1184 u32 timer = cq->tobe_rx_coal_timeval;
1186 if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1187 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1188 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1192 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1194 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1195 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1196 struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1202 ktime_t now = ktime_get();
1204 delta = ktime_us_delta(now, cq->prev_ts);
1205 if (delta < ENIC_AIC_TS_BREAK)
1209 traffic = pkt_size_counter->large_pkt_bytes_cnt +
1210 pkt_size_counter->small_pkt_bytes_cnt;
1211 /* The table takes Mbps
1212 * traffic *= 8 => bits
1213 * traffic *= (10^6 / delta) => bps
1214 * traffic /= 10^6 => Mbps
1216 * Combining, traffic *= (8 / delta)
1222 for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1223 if (traffic < mod_table[index].rx_rate)
1225 range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1226 pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1227 rx_coal->small_pkt_range_start :
1228 rx_coal->large_pkt_range_start;
1229 timer = range_start + ((rx_coal->range_end - range_start) *
1230 mod_table[index].range_percent / 100);
1232 cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1234 pkt_size_counter->large_pkt_bytes_cnt = 0;
1235 pkt_size_counter->small_pkt_bytes_cnt = 0;
1238 static int enic_poll_msix(struct napi_struct *napi, int budget)
1240 struct net_device *netdev = napi->dev;
1241 struct enic *enic = netdev_priv(netdev);
1242 unsigned int rq = (napi - &enic->napi[0]);
1243 unsigned int cq = enic_cq_rq(enic, rq);
1244 unsigned int intr = enic_msix_rq_intr(enic, rq);
1245 unsigned int work_to_do = budget;
1246 unsigned int work_done = 0;
1253 work_done = vnic_cq_service(&enic->cq[cq],
1254 work_to_do, enic_rq_service, NULL);
1256 /* Return intr event credits for this polling
1257 * cycle. An intr event is the completion of a
1262 vnic_intr_return_credits(&enic->intr[intr],
1264 0 /* don't unmask intr */,
1265 0 /* don't reset intr timer */);
1267 err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1269 /* Buffer allocation failed. Stay in polling mode
1270 * so we can try to fill the ring again.
1274 work_done = work_to_do;
1275 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1276 /* Call the function which refreshes
1277 * the intr coalescing timer value based on
1278 * the traffic. This is supported only in
1279 * the case of MSI-x mode
1281 enic_calc_int_moderation(enic, &enic->rq[rq]);
1283 if (work_done < work_to_do) {
1285 /* Some work done, but not enough to stay in polling,
1289 napi_complete(napi);
1290 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1291 enic_set_int_moderation(enic, &enic->rq[rq]);
1292 vnic_intr_unmask(&enic->intr[intr]);
1298 static void enic_notify_timer(unsigned long data)
1300 struct enic *enic = (struct enic *)data;
1302 enic_notify_check(enic);
1304 mod_timer(&enic->notify_timer,
1305 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1308 static void enic_free_intr(struct enic *enic)
1310 struct net_device *netdev = enic->netdev;
1313 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1314 case VNIC_DEV_INTR_MODE_INTX:
1315 free_irq(enic->pdev->irq, netdev);
1317 case VNIC_DEV_INTR_MODE_MSI:
1318 free_irq(enic->pdev->irq, enic);
1320 case VNIC_DEV_INTR_MODE_MSIX:
1321 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1322 if (enic->msix[i].requested)
1323 free_irq(enic->msix_entry[i].vector,
1324 enic->msix[i].devid);
1331 static int enic_request_intr(struct enic *enic)
1333 struct net_device *netdev = enic->netdev;
1334 unsigned int i, intr;
1337 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1339 case VNIC_DEV_INTR_MODE_INTX:
1341 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1342 IRQF_SHARED, netdev->name, netdev);
1345 case VNIC_DEV_INTR_MODE_MSI:
1347 err = request_irq(enic->pdev->irq, enic_isr_msi,
1348 0, netdev->name, enic);
1351 case VNIC_DEV_INTR_MODE_MSIX:
1353 for (i = 0; i < enic->rq_count; i++) {
1354 intr = enic_msix_rq_intr(enic, i);
1355 snprintf(enic->msix[intr].devname,
1356 sizeof(enic->msix[intr].devname),
1357 "%.11s-rx-%d", netdev->name, i);
1358 enic->msix[intr].isr = enic_isr_msix_rq;
1359 enic->msix[intr].devid = &enic->napi[i];
1362 for (i = 0; i < enic->wq_count; i++) {
1363 intr = enic_msix_wq_intr(enic, i);
1364 snprintf(enic->msix[intr].devname,
1365 sizeof(enic->msix[intr].devname),
1366 "%.11s-tx-%d", netdev->name, i);
1367 enic->msix[intr].isr = enic_isr_msix_wq;
1368 enic->msix[intr].devid = enic;
1371 intr = enic_msix_err_intr(enic);
1372 snprintf(enic->msix[intr].devname,
1373 sizeof(enic->msix[intr].devname),
1374 "%.11s-err", netdev->name);
1375 enic->msix[intr].isr = enic_isr_msix_err;
1376 enic->msix[intr].devid = enic;
1378 intr = enic_msix_notify_intr(enic);
1379 snprintf(enic->msix[intr].devname,
1380 sizeof(enic->msix[intr].devname),
1381 "%.11s-notify", netdev->name);
1382 enic->msix[intr].isr = enic_isr_msix_notify;
1383 enic->msix[intr].devid = enic;
1385 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1386 enic->msix[i].requested = 0;
1388 for (i = 0; i < enic->intr_count; i++) {
1389 err = request_irq(enic->msix_entry[i].vector,
1390 enic->msix[i].isr, 0,
1391 enic->msix[i].devname,
1392 enic->msix[i].devid);
1394 enic_free_intr(enic);
1397 enic->msix[i].requested = 1;
1409 static void enic_synchronize_irqs(struct enic *enic)
1413 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1414 case VNIC_DEV_INTR_MODE_INTX:
1415 case VNIC_DEV_INTR_MODE_MSI:
1416 synchronize_irq(enic->pdev->irq);
1418 case VNIC_DEV_INTR_MODE_MSIX:
1419 for (i = 0; i < enic->intr_count; i++)
1420 synchronize_irq(enic->msix_entry[i].vector);
1427 static void enic_set_rx_coal_setting(struct enic *enic)
1431 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1433 /* If intr mode is not MSIX, do not do adaptive coalescing */
1434 if (VNIC_DEV_INTR_MODE_MSIX != vnic_dev_get_intr_mode(enic->vdev)) {
1435 netdev_info(enic->netdev, "INTR mode is not MSIX, Not initializing adaptive coalescing");
1439 /* 1. Read the link speed from fw
1440 * 2. Pick the default range for the speed
1441 * 3. Update it in enic->rx_coalesce_setting
1443 speed = vnic_dev_port_speed(enic->vdev);
1444 if (ENIC_LINK_SPEED_10G < speed)
1445 index = ENIC_LINK_40G_INDEX;
1446 else if (ENIC_LINK_SPEED_4G < speed)
1447 index = ENIC_LINK_10G_INDEX;
1449 index = ENIC_LINK_4G_INDEX;
1451 rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1452 rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1453 rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1455 /* Start with the value provided by UCSM */
1456 for (index = 0; index < enic->rq_count; index++)
1457 enic->cq[index].cur_rx_coal_timeval =
1458 enic->config.intr_timer_usec;
1460 rx_coal->use_adaptive_rx_coalesce = 1;
1463 static int enic_dev_notify_set(struct enic *enic)
1467 spin_lock(&enic->devcmd_lock);
1468 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1469 case VNIC_DEV_INTR_MODE_INTX:
1470 err = vnic_dev_notify_set(enic->vdev,
1471 enic_legacy_notify_intr());
1473 case VNIC_DEV_INTR_MODE_MSIX:
1474 err = vnic_dev_notify_set(enic->vdev,
1475 enic_msix_notify_intr(enic));
1478 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1481 spin_unlock(&enic->devcmd_lock);
1486 static void enic_notify_timer_start(struct enic *enic)
1488 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1489 case VNIC_DEV_INTR_MODE_MSI:
1490 mod_timer(&enic->notify_timer, jiffies);
1493 /* Using intr for notification for INTx/MSI-X */
1498 /* rtnl lock is held, process context */
1499 static int enic_open(struct net_device *netdev)
1501 struct enic *enic = netdev_priv(netdev);
1505 err = enic_request_intr(enic);
1507 netdev_err(netdev, "Unable to request irq.\n");
1511 err = enic_dev_notify_set(enic);
1514 "Failed to alloc notify buffer, aborting.\n");
1515 goto err_out_free_intr;
1518 for (i = 0; i < enic->rq_count; i++) {
1519 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1520 /* Need at least one buffer on ring to get going */
1521 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1522 netdev_err(netdev, "Unable to alloc receive buffers\n");
1524 goto err_out_notify_unset;
1528 for (i = 0; i < enic->wq_count; i++)
1529 vnic_wq_enable(&enic->wq[i]);
1530 for (i = 0; i < enic->rq_count; i++)
1531 vnic_rq_enable(&enic->rq[i]);
1533 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1534 enic_dev_add_station_addr(enic);
1536 enic_set_rx_mode(netdev);
1538 netif_tx_wake_all_queues(netdev);
1540 for (i = 0; i < enic->rq_count; i++)
1541 napi_enable(&enic->napi[i]);
1543 enic_dev_enable(enic);
1545 for (i = 0; i < enic->intr_count; i++)
1546 vnic_intr_unmask(&enic->intr[i]);
1548 enic_notify_timer_start(enic);
1552 err_out_notify_unset:
1553 enic_dev_notify_unset(enic);
1555 enic_free_intr(enic);
1560 /* rtnl lock is held, process context */
1561 static int enic_stop(struct net_device *netdev)
1563 struct enic *enic = netdev_priv(netdev);
1567 for (i = 0; i < enic->intr_count; i++) {
1568 vnic_intr_mask(&enic->intr[i]);
1569 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1572 enic_synchronize_irqs(enic);
1574 del_timer_sync(&enic->notify_timer);
1576 enic_dev_disable(enic);
1578 for (i = 0; i < enic->rq_count; i++)
1579 napi_disable(&enic->napi[i]);
1581 netif_carrier_off(netdev);
1582 netif_tx_disable(netdev);
1584 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1585 enic_dev_del_station_addr(enic);
1587 for (i = 0; i < enic->wq_count; i++) {
1588 err = vnic_wq_disable(&enic->wq[i]);
1592 for (i = 0; i < enic->rq_count; i++) {
1593 err = vnic_rq_disable(&enic->rq[i]);
1598 enic_dev_notify_unset(enic);
1599 enic_free_intr(enic);
1601 for (i = 0; i < enic->wq_count; i++)
1602 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1603 for (i = 0; i < enic->rq_count; i++)
1604 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1605 for (i = 0; i < enic->cq_count; i++)
1606 vnic_cq_clean(&enic->cq[i]);
1607 for (i = 0; i < enic->intr_count; i++)
1608 vnic_intr_clean(&enic->intr[i]);
1613 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1615 struct enic *enic = netdev_priv(netdev);
1616 int running = netif_running(netdev);
1618 if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1621 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
1627 netdev->mtu = new_mtu;
1629 if (netdev->mtu > enic->port_mtu)
1631 "interface MTU (%d) set higher than port MTU (%d)\n",
1632 netdev->mtu, enic->port_mtu);
1640 static void enic_change_mtu_work(struct work_struct *work)
1642 struct enic *enic = container_of(work, struct enic, change_mtu_work);
1643 struct net_device *netdev = enic->netdev;
1644 int new_mtu = vnic_dev_mtu(enic->vdev);
1648 new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
1653 del_timer_sync(&enic->notify_timer);
1655 for (i = 0; i < enic->rq_count; i++)
1656 napi_disable(&enic->napi[i]);
1658 vnic_intr_mask(&enic->intr[0]);
1659 enic_synchronize_irqs(enic);
1660 err = vnic_rq_disable(&enic->rq[0]);
1663 netdev_err(netdev, "Unable to disable RQ.\n");
1666 vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
1667 vnic_cq_clean(&enic->cq[0]);
1668 vnic_intr_clean(&enic->intr[0]);
1670 /* Fill RQ with new_mtu-sized buffers */
1671 netdev->mtu = new_mtu;
1672 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1673 /* Need at least one buffer on ring to get going */
1674 if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
1676 netdev_err(netdev, "Unable to alloc receive buffers.\n");
1681 vnic_rq_enable(&enic->rq[0]);
1682 napi_enable(&enic->napi[0]);
1683 vnic_intr_unmask(&enic->intr[0]);
1684 enic_notify_timer_start(enic);
1688 netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1691 #ifdef CONFIG_NET_POLL_CONTROLLER
1692 static void enic_poll_controller(struct net_device *netdev)
1694 struct enic *enic = netdev_priv(netdev);
1695 struct vnic_dev *vdev = enic->vdev;
1696 unsigned int i, intr;
1698 switch (vnic_dev_get_intr_mode(vdev)) {
1699 case VNIC_DEV_INTR_MODE_MSIX:
1700 for (i = 0; i < enic->rq_count; i++) {
1701 intr = enic_msix_rq_intr(enic, i);
1702 enic_isr_msix_rq(enic->msix_entry[intr].vector,
1706 for (i = 0; i < enic->wq_count; i++) {
1707 intr = enic_msix_wq_intr(enic, i);
1708 enic_isr_msix_wq(enic->msix_entry[intr].vector, enic);
1712 case VNIC_DEV_INTR_MODE_MSI:
1713 enic_isr_msi(enic->pdev->irq, enic);
1715 case VNIC_DEV_INTR_MODE_INTX:
1716 enic_isr_legacy(enic->pdev->irq, netdev);
1724 static int enic_dev_wait(struct vnic_dev *vdev,
1725 int (*start)(struct vnic_dev *, int),
1726 int (*finished)(struct vnic_dev *, int *),
1733 BUG_ON(in_interrupt());
1735 err = start(vdev, arg);
1739 /* Wait for func to complete...2 seconds max
1742 time = jiffies + (HZ * 2);
1745 err = finished(vdev, &done);
1752 schedule_timeout_uninterruptible(HZ / 10);
1754 } while (time_after(time, jiffies));
1759 static int enic_dev_open(struct enic *enic)
1763 err = enic_dev_wait(enic->vdev, vnic_dev_open,
1764 vnic_dev_open_done, 0);
1766 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1772 static int enic_dev_hang_reset(struct enic *enic)
1776 err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1777 vnic_dev_hang_reset_done, 0);
1779 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1785 static int enic_set_rsskey(struct enic *enic)
1787 dma_addr_t rss_key_buf_pa;
1788 union vnic_rss_key *rss_key_buf_va = NULL;
1789 union vnic_rss_key rss_key = {
1790 .key[0].b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101},
1791 .key[1].b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101},
1792 .key[2].b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115},
1793 .key[3].b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108},
1797 rss_key_buf_va = pci_alloc_consistent(enic->pdev,
1798 sizeof(union vnic_rss_key), &rss_key_buf_pa);
1799 if (!rss_key_buf_va)
1802 memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
1804 spin_lock(&enic->devcmd_lock);
1805 err = enic_set_rss_key(enic,
1807 sizeof(union vnic_rss_key));
1808 spin_unlock(&enic->devcmd_lock);
1810 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
1811 rss_key_buf_va, rss_key_buf_pa);
1816 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
1818 dma_addr_t rss_cpu_buf_pa;
1819 union vnic_rss_cpu *rss_cpu_buf_va = NULL;
1823 rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
1824 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
1825 if (!rss_cpu_buf_va)
1828 for (i = 0; i < (1 << rss_hash_bits); i++)
1829 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
1831 spin_lock(&enic->devcmd_lock);
1832 err = enic_set_rss_cpu(enic,
1834 sizeof(union vnic_rss_cpu));
1835 spin_unlock(&enic->devcmd_lock);
1837 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
1838 rss_cpu_buf_va, rss_cpu_buf_pa);
1843 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
1844 u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
1846 const u8 tso_ipid_split_en = 0;
1847 const u8 ig_vlan_strip_en = 1;
1850 /* Enable VLAN tag stripping.
1853 spin_lock(&enic->devcmd_lock);
1854 err = enic_set_nic_cfg(enic,
1855 rss_default_cpu, rss_hash_type,
1856 rss_hash_bits, rss_base_cpu,
1857 rss_enable, tso_ipid_split_en,
1859 spin_unlock(&enic->devcmd_lock);
1864 static int enic_set_rss_nic_cfg(struct enic *enic)
1866 struct device *dev = enic_get_dev(enic);
1867 const u8 rss_default_cpu = 0;
1868 const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1869 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1870 NIC_CFG_RSS_HASH_TYPE_IPV6 |
1871 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1872 const u8 rss_hash_bits = 7;
1873 const u8 rss_base_cpu = 0;
1874 u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1877 if (!enic_set_rsskey(enic)) {
1878 if (enic_set_rsscpu(enic, rss_hash_bits)) {
1880 dev_warn(dev, "RSS disabled, "
1881 "Failed to set RSS cpu indirection table.");
1885 dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
1889 return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1890 rss_hash_bits, rss_base_cpu, rss_enable);
1893 static void enic_reset(struct work_struct *work)
1895 struct enic *enic = container_of(work, struct enic, reset);
1897 if (!netif_running(enic->netdev))
1902 spin_lock(&enic->enic_api_lock);
1903 enic_dev_hang_notify(enic);
1904 enic_stop(enic->netdev);
1905 enic_dev_hang_reset(enic);
1906 enic_reset_addr_lists(enic);
1907 enic_init_vnic_resources(enic);
1908 enic_set_rss_nic_cfg(enic);
1909 enic_dev_set_ig_vlan_rewrite_mode(enic);
1910 enic_open(enic->netdev);
1911 spin_unlock(&enic->enic_api_lock);
1912 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
1917 static int enic_set_intr_mode(struct enic *enic)
1919 unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
1920 unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
1923 /* Set interrupt mode (INTx, MSI, MSI-X) depending
1924 * on system capabilities.
1928 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1929 * (the second to last INTR is used for WQ/RQ errors)
1930 * (the last INTR is used for notifications)
1933 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1934 for (i = 0; i < n + m + 2; i++)
1935 enic->msix_entry[i].entry = i;
1937 /* Use multiple RQs if RSS is enabled
1940 if (ENIC_SETTING(enic, RSS) &&
1941 enic->config.intr_mode < 1 &&
1942 enic->rq_count >= n &&
1943 enic->wq_count >= m &&
1944 enic->cq_count >= n + m &&
1945 enic->intr_count >= n + m + 2) {
1947 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
1948 n + m + 2, n + m + 2) > 0) {
1952 enic->cq_count = n + m;
1953 enic->intr_count = n + m + 2;
1955 vnic_dev_set_intr_mode(enic->vdev,
1956 VNIC_DEV_INTR_MODE_MSIX);
1962 if (enic->config.intr_mode < 1 &&
1963 enic->rq_count >= 1 &&
1964 enic->wq_count >= m &&
1965 enic->cq_count >= 1 + m &&
1966 enic->intr_count >= 1 + m + 2) {
1967 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
1968 1 + m + 2, 1 + m + 2) > 0) {
1972 enic->cq_count = 1 + m;
1973 enic->intr_count = 1 + m + 2;
1975 vnic_dev_set_intr_mode(enic->vdev,
1976 VNIC_DEV_INTR_MODE_MSIX);
1984 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1987 if (enic->config.intr_mode < 2 &&
1988 enic->rq_count >= 1 &&
1989 enic->wq_count >= 1 &&
1990 enic->cq_count >= 2 &&
1991 enic->intr_count >= 1 &&
1992 !pci_enable_msi(enic->pdev)) {
1997 enic->intr_count = 1;
1999 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2006 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2007 * (the first INTR is used for WQ/RQ)
2008 * (the second INTR is used for WQ/RQ errors)
2009 * (the last INTR is used for notifications)
2012 if (enic->config.intr_mode < 3 &&
2013 enic->rq_count >= 1 &&
2014 enic->wq_count >= 1 &&
2015 enic->cq_count >= 2 &&
2016 enic->intr_count >= 3) {
2021 enic->intr_count = 3;
2023 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2028 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2033 static void enic_clear_intr_mode(struct enic *enic)
2035 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2036 case VNIC_DEV_INTR_MODE_MSIX:
2037 pci_disable_msix(enic->pdev);
2039 case VNIC_DEV_INTR_MODE_MSI:
2040 pci_disable_msi(enic->pdev);
2046 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2049 static const struct net_device_ops enic_netdev_dynamic_ops = {
2050 .ndo_open = enic_open,
2051 .ndo_stop = enic_stop,
2052 .ndo_start_xmit = enic_hard_start_xmit,
2053 .ndo_get_stats64 = enic_get_stats,
2054 .ndo_validate_addr = eth_validate_addr,
2055 .ndo_set_rx_mode = enic_set_rx_mode,
2056 .ndo_set_mac_address = enic_set_mac_address_dynamic,
2057 .ndo_change_mtu = enic_change_mtu,
2058 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2059 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2060 .ndo_tx_timeout = enic_tx_timeout,
2061 .ndo_set_vf_port = enic_set_vf_port,
2062 .ndo_get_vf_port = enic_get_vf_port,
2063 .ndo_set_vf_mac = enic_set_vf_mac,
2064 #ifdef CONFIG_NET_POLL_CONTROLLER
2065 .ndo_poll_controller = enic_poll_controller,
2069 static const struct net_device_ops enic_netdev_ops = {
2070 .ndo_open = enic_open,
2071 .ndo_stop = enic_stop,
2072 .ndo_start_xmit = enic_hard_start_xmit,
2073 .ndo_get_stats64 = enic_get_stats,
2074 .ndo_validate_addr = eth_validate_addr,
2075 .ndo_set_mac_address = enic_set_mac_address,
2076 .ndo_set_rx_mode = enic_set_rx_mode,
2077 .ndo_change_mtu = enic_change_mtu,
2078 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2079 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2080 .ndo_tx_timeout = enic_tx_timeout,
2081 .ndo_set_vf_port = enic_set_vf_port,
2082 .ndo_get_vf_port = enic_get_vf_port,
2083 .ndo_set_vf_mac = enic_set_vf_mac,
2084 #ifdef CONFIG_NET_POLL_CONTROLLER
2085 .ndo_poll_controller = enic_poll_controller,
2089 static void enic_dev_deinit(struct enic *enic)
2093 for (i = 0; i < enic->rq_count; i++)
2094 netif_napi_del(&enic->napi[i]);
2096 enic_free_vnic_resources(enic);
2097 enic_clear_intr_mode(enic);
2100 static int enic_dev_init(struct enic *enic)
2102 struct device *dev = enic_get_dev(enic);
2103 struct net_device *netdev = enic->netdev;
2107 /* Get interrupt coalesce timer info */
2108 err = enic_dev_intr_coal_timer_info(enic);
2110 dev_warn(dev, "Using default conversion factor for "
2111 "interrupt coalesce timer\n");
2112 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2115 /* Get vNIC configuration
2118 err = enic_get_vnic_config(enic);
2120 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2124 /* Get available resource counts
2127 enic_get_res_counts(enic);
2129 /* Set interrupt mode based on resource counts and system
2133 err = enic_set_intr_mode(enic);
2135 dev_err(dev, "Failed to set intr mode based on resource "
2136 "counts and system capabilities, aborting\n");
2140 /* Allocate and configure vNIC resources
2143 err = enic_alloc_vnic_resources(enic);
2145 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2146 goto err_out_free_vnic_resources;
2149 enic_init_vnic_resources(enic);
2151 err = enic_set_rss_nic_cfg(enic);
2153 dev_err(dev, "Failed to config nic, aborting\n");
2154 goto err_out_free_vnic_resources;
2157 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2159 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2161 case VNIC_DEV_INTR_MODE_MSIX:
2162 for (i = 0; i < enic->rq_count; i++)
2163 netif_napi_add(netdev, &enic->napi[i],
2164 enic_poll_msix, 64);
2170 err_out_free_vnic_resources:
2171 enic_clear_intr_mode(enic);
2172 enic_free_vnic_resources(enic);
2177 static void enic_iounmap(struct enic *enic)
2181 for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2182 if (enic->bar[i].vaddr)
2183 iounmap(enic->bar[i].vaddr);
2186 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2188 struct device *dev = &pdev->dev;
2189 struct net_device *netdev;
2194 #ifdef CONFIG_PCI_IOV
2199 /* Allocate net device structure and initialize. Private
2200 * instance data is initialized to zero.
2203 netdev = alloc_etherdev_mqs(sizeof(struct enic),
2204 ENIC_RQ_MAX, ENIC_WQ_MAX);
2208 pci_set_drvdata(pdev, netdev);
2210 SET_NETDEV_DEV(netdev, &pdev->dev);
2212 enic = netdev_priv(netdev);
2213 enic->netdev = netdev;
2216 /* Setup PCI resources
2219 err = pci_enable_device_mem(pdev);
2221 dev_err(dev, "Cannot enable PCI device, aborting\n");
2222 goto err_out_free_netdev;
2225 err = pci_request_regions(pdev, DRV_NAME);
2227 dev_err(dev, "Cannot request PCI regions, aborting\n");
2228 goto err_out_disable_device;
2231 pci_set_master(pdev);
2233 /* Query PCI controller on system for DMA addressing
2234 * limitation for the device. Try 64-bit first, and
2238 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2240 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2242 dev_err(dev, "No usable DMA configuration, aborting\n");
2243 goto err_out_release_regions;
2245 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2247 dev_err(dev, "Unable to obtain %u-bit DMA "
2248 "for consistent allocations, aborting\n", 32);
2249 goto err_out_release_regions;
2252 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2254 dev_err(dev, "Unable to obtain %u-bit DMA "
2255 "for consistent allocations, aborting\n", 64);
2256 goto err_out_release_regions;
2261 /* Map vNIC resources from BAR0-5
2264 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2265 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2267 enic->bar[i].len = pci_resource_len(pdev, i);
2268 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2269 if (!enic->bar[i].vaddr) {
2270 dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2272 goto err_out_iounmap;
2274 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2277 /* Register vNIC device
2280 enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2281 ARRAY_SIZE(enic->bar));
2283 dev_err(dev, "vNIC registration failed, aborting\n");
2285 goto err_out_iounmap;
2288 #ifdef CONFIG_PCI_IOV
2289 /* Get number of subvnics */
2290 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2292 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2294 if (enic->num_vfs) {
2295 err = pci_enable_sriov(pdev, enic->num_vfs);
2297 dev_err(dev, "SRIOV enable failed, aborting."
2298 " pci_enable_sriov() returned %d\n",
2300 goto err_out_vnic_unregister;
2302 enic->priv_flags |= ENIC_SRIOV_ENABLED;
2303 num_pps = enic->num_vfs;
2308 /* Allocate structure for port profiles */
2309 enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2312 goto err_out_disable_sriov_pp;
2315 /* Issue device open to get device in known state
2318 err = enic_dev_open(enic);
2320 dev_err(dev, "vNIC dev open failed, aborting\n");
2321 goto err_out_disable_sriov;
2324 /* Setup devcmd lock
2327 spin_lock_init(&enic->devcmd_lock);
2328 spin_lock_init(&enic->enic_api_lock);
2331 * Set ingress vlan rewrite mode before vnic initialization
2334 err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2337 "Failed to set ingress vlan rewrite mode, aborting.\n");
2338 goto err_out_dev_close;
2341 /* Issue device init to initialize the vnic-to-switch link.
2342 * We'll start with carrier off and wait for link UP
2343 * notification later to turn on carrier. We don't need
2344 * to wait here for the vnic-to-switch link initialization
2345 * to complete; link UP notification is the indication that
2346 * the process is complete.
2349 netif_carrier_off(netdev);
2351 /* Do not call dev_init for a dynamic vnic.
2352 * For a dynamic vnic, init_prov_info will be
2353 * called later by an upper layer.
2356 if (!enic_is_dynamic(enic)) {
2357 err = vnic_dev_init(enic->vdev, 0);
2359 dev_err(dev, "vNIC dev init failed, aborting\n");
2360 goto err_out_dev_close;
2364 err = enic_dev_init(enic);
2366 dev_err(dev, "Device initialization failed, aborting\n");
2367 goto err_out_dev_close;
2370 netif_set_real_num_tx_queues(netdev, enic->wq_count);
2371 netif_set_real_num_rx_queues(netdev, enic->rq_count);
2373 /* Setup notification timer, HW reset task, and wq locks
2376 init_timer(&enic->notify_timer);
2377 enic->notify_timer.function = enic_notify_timer;
2378 enic->notify_timer.data = (unsigned long)enic;
2380 enic_set_rx_coal_setting(enic);
2381 INIT_WORK(&enic->reset, enic_reset);
2382 INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2384 for (i = 0; i < enic->wq_count; i++)
2385 spin_lock_init(&enic->wq_lock[i]);
2387 /* Register net device
2390 enic->port_mtu = enic->config.mtu;
2391 (void)enic_change_mtu(netdev, enic->port_mtu);
2393 err = enic_set_mac_addr(netdev, enic->mac_addr);
2395 dev_err(dev, "Invalid MAC address, aborting\n");
2396 goto err_out_dev_deinit;
2399 enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2400 /* rx coalesce time already got initialized. This gets used
2401 * if adaptive coal is turned off
2403 enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2405 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2406 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2408 netdev->netdev_ops = &enic_netdev_ops;
2410 netdev->watchdog_timeo = 2 * HZ;
2411 enic_set_ethtool_ops(netdev);
2413 netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2414 if (ENIC_SETTING(enic, LOOP)) {
2415 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2416 enic->loop_enable = 1;
2417 enic->loop_tag = enic->config.loop_tag;
2418 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2420 if (ENIC_SETTING(enic, TXCSUM))
2421 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2422 if (ENIC_SETTING(enic, TSO))
2423 netdev->hw_features |= NETIF_F_TSO |
2424 NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2425 if (ENIC_SETTING(enic, RSS))
2426 netdev->hw_features |= NETIF_F_RXHASH;
2427 if (ENIC_SETTING(enic, RXCSUM))
2428 netdev->hw_features |= NETIF_F_RXCSUM;
2430 netdev->features |= netdev->hw_features;
2433 netdev->features |= NETIF_F_HIGHDMA;
2435 netdev->priv_flags |= IFF_UNICAST_FLT;
2437 err = register_netdev(netdev);
2439 dev_err(dev, "Cannot register net device, aborting\n");
2440 goto err_out_dev_deinit;
2446 enic_dev_deinit(enic);
2448 vnic_dev_close(enic->vdev);
2449 err_out_disable_sriov:
2451 err_out_disable_sriov_pp:
2452 #ifdef CONFIG_PCI_IOV
2453 if (enic_sriov_enabled(enic)) {
2454 pci_disable_sriov(pdev);
2455 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2457 err_out_vnic_unregister:
2459 vnic_dev_unregister(enic->vdev);
2462 err_out_release_regions:
2463 pci_release_regions(pdev);
2464 err_out_disable_device:
2465 pci_disable_device(pdev);
2466 err_out_free_netdev:
2467 free_netdev(netdev);
2472 static void enic_remove(struct pci_dev *pdev)
2474 struct net_device *netdev = pci_get_drvdata(pdev);
2477 struct enic *enic = netdev_priv(netdev);
2479 cancel_work_sync(&enic->reset);
2480 cancel_work_sync(&enic->change_mtu_work);
2481 unregister_netdev(netdev);
2482 enic_dev_deinit(enic);
2483 vnic_dev_close(enic->vdev);
2484 #ifdef CONFIG_PCI_IOV
2485 if (enic_sriov_enabled(enic)) {
2486 pci_disable_sriov(pdev);
2487 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2491 vnic_dev_unregister(enic->vdev);
2493 pci_release_regions(pdev);
2494 pci_disable_device(pdev);
2495 free_netdev(netdev);
2499 static struct pci_driver enic_driver = {
2501 .id_table = enic_id_table,
2502 .probe = enic_probe,
2503 .remove = enic_remove,
2506 static int __init enic_init_module(void)
2508 pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2510 return pci_register_driver(&enic_driver);
2513 static void __exit enic_cleanup_module(void)
2515 pci_unregister_driver(&enic_driver);
2518 module_init(enic_init_module);
2519 module_exit(enic_cleanup_module);