2 * Copyright 2008-2010 Cisco Systems, Inc. All rights reserved.
3 * Copyright 2007 Nuova Systems, Inc. All rights reserved.
5 * This program is free software; you may redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; version 2 of the License.
9 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
10 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
11 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
12 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
13 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
14 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
15 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
20 #include <linux/module.h>
21 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/init.h>
26 #include <linux/interrupt.h>
27 #include <linux/workqueue.h>
28 #include <linux/pci.h>
29 #include <linux/netdevice.h>
30 #include <linux/etherdevice.h>
32 #include <linux/if_ether.h>
33 #include <linux/if_vlan.h>
36 #include <linux/ipv6.h>
37 #include <linux/tcp.h>
38 #include <linux/rtnetlink.h>
39 #include <linux/prefetch.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/ktime.h>
42 #ifdef CONFIG_RFS_ACCEL
43 #include <linux/cpu_rmap.h>
45 #ifdef CONFIG_NET_RX_BUSY_POLL
46 #include <net/busy_poll.h>
49 #include "cq_enet_desc.h"
51 #include "vnic_intr.h"
52 #include "vnic_stats.h"
58 #include "enic_clsf.h"
60 #define ENIC_NOTIFY_TIMER_PERIOD (2 * HZ)
61 #define WQ_ENET_MAX_DESC_LEN (1 << WQ_ENET_LEN_BITS)
62 #define MAX_TSO (1 << 16)
63 #define ENIC_DESC_MAX_SPLITS (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
65 #define PCI_DEVICE_ID_CISCO_VIC_ENET 0x0043 /* ethernet vnic */
66 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN 0x0044 /* enet dynamic vnic */
67 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF 0x0071 /* enet SRIOV VF */
69 #define RX_COPYBREAK_DEFAULT 256
71 /* Supported devices */
72 static const struct pci_device_id enic_id_table[] = {
73 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
74 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
75 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
76 { 0, } /* end of table */
79 MODULE_DESCRIPTION(DRV_DESCRIPTION);
80 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
81 MODULE_LICENSE("GPL");
82 MODULE_VERSION(DRV_VERSION);
83 MODULE_DEVICE_TABLE(pci, enic_id_table);
85 #define ENIC_LARGE_PKT_THRESHOLD 1000
86 #define ENIC_MAX_COALESCE_TIMERS 10
87 /* Interrupt moderation table, which will be used to decide the
88 * coalescing timer values
89 * {rx_rate in Mbps, mapping percentage of the range}
91 struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
105 /* This table helps the driver to pick different ranges for rx coalescing
106 * timer depending on the link speed.
108 struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = {
109 {0, 0}, /* 0 - 4 Gbps */
110 {0, 3}, /* 4 - 10 Gbps */
111 {3, 6}, /* 10 - 40 Gbps */
114 int enic_is_dynamic(struct enic *enic)
116 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
119 int enic_sriov_enabled(struct enic *enic)
121 return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
124 static int enic_is_sriov_vf(struct enic *enic)
126 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
129 int enic_is_valid_vf(struct enic *enic, int vf)
131 #ifdef CONFIG_PCI_IOV
132 return vf >= 0 && vf < enic->num_vfs;
138 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
140 struct enic *enic = vnic_dev_priv(wq->vdev);
143 pci_unmap_single(enic->pdev, buf->dma_addr,
144 buf->len, PCI_DMA_TODEVICE);
146 pci_unmap_page(enic->pdev, buf->dma_addr,
147 buf->len, PCI_DMA_TODEVICE);
150 dev_kfree_skb_any(buf->os_buf);
153 static void enic_wq_free_buf(struct vnic_wq *wq,
154 struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
156 enic_free_wq_buf(wq, buf);
159 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
160 u8 type, u16 q_number, u16 completed_index, void *opaque)
162 struct enic *enic = vnic_dev_priv(vdev);
164 spin_lock(&enic->wq_lock[q_number]);
166 vnic_wq_service(&enic->wq[q_number], cq_desc,
167 completed_index, enic_wq_free_buf,
170 if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
171 vnic_wq_desc_avail(&enic->wq[q_number]) >=
172 (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
173 netif_wake_subqueue(enic->netdev, q_number);
175 spin_unlock(&enic->wq_lock[q_number]);
180 static void enic_log_q_error(struct enic *enic)
185 for (i = 0; i < enic->wq_count; i++) {
186 error_status = vnic_wq_error_status(&enic->wq[i]);
188 netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
192 for (i = 0; i < enic->rq_count; i++) {
193 error_status = vnic_rq_error_status(&enic->rq[i]);
195 netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
200 static void enic_msglvl_check(struct enic *enic)
202 u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
204 if (msg_enable != enic->msg_enable) {
205 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
206 enic->msg_enable, msg_enable);
207 enic->msg_enable = msg_enable;
211 static void enic_mtu_check(struct enic *enic)
213 u32 mtu = vnic_dev_mtu(enic->vdev);
214 struct net_device *netdev = enic->netdev;
216 if (mtu && mtu != enic->port_mtu) {
217 enic->port_mtu = mtu;
218 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
219 mtu = max_t(int, ENIC_MIN_MTU,
220 min_t(int, ENIC_MAX_MTU, mtu));
221 if (mtu != netdev->mtu)
222 schedule_work(&enic->change_mtu_work);
224 if (mtu < netdev->mtu)
226 "interface MTU (%d) set higher "
227 "than switch port MTU (%d)\n",
233 static void enic_link_check(struct enic *enic)
235 int link_status = vnic_dev_link_status(enic->vdev);
236 int carrier_ok = netif_carrier_ok(enic->netdev);
238 if (link_status && !carrier_ok) {
239 netdev_info(enic->netdev, "Link UP\n");
240 netif_carrier_on(enic->netdev);
241 } else if (!link_status && carrier_ok) {
242 netdev_info(enic->netdev, "Link DOWN\n");
243 netif_carrier_off(enic->netdev);
247 static void enic_notify_check(struct enic *enic)
249 enic_msglvl_check(enic);
250 enic_mtu_check(enic);
251 enic_link_check(enic);
254 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
256 static irqreturn_t enic_isr_legacy(int irq, void *data)
258 struct net_device *netdev = data;
259 struct enic *enic = netdev_priv(netdev);
260 unsigned int io_intr = enic_legacy_io_intr();
261 unsigned int err_intr = enic_legacy_err_intr();
262 unsigned int notify_intr = enic_legacy_notify_intr();
265 vnic_intr_mask(&enic->intr[io_intr]);
267 pba = vnic_intr_legacy_pba(enic->legacy_pba);
269 vnic_intr_unmask(&enic->intr[io_intr]);
270 return IRQ_NONE; /* not our interrupt */
273 if (ENIC_TEST_INTR(pba, notify_intr)) {
274 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
275 enic_notify_check(enic);
278 if (ENIC_TEST_INTR(pba, err_intr)) {
279 vnic_intr_return_all_credits(&enic->intr[err_intr]);
280 enic_log_q_error(enic);
281 /* schedule recovery from WQ/RQ error */
282 schedule_work(&enic->reset);
286 if (ENIC_TEST_INTR(pba, io_intr))
287 napi_schedule_irqoff(&enic->napi[0]);
289 vnic_intr_unmask(&enic->intr[io_intr]);
294 static irqreturn_t enic_isr_msi(int irq, void *data)
296 struct enic *enic = data;
298 /* With MSI, there is no sharing of interrupts, so this is
299 * our interrupt and there is no need to ack it. The device
300 * is not providing per-vector masking, so the OS will not
301 * write to PCI config space to mask/unmask the interrupt.
302 * We're using mask_on_assertion for MSI, so the device
303 * automatically masks the interrupt when the interrupt is
304 * generated. Later, when exiting polling, the interrupt
305 * will be unmasked (see enic_poll).
307 * Also, the device uses the same PCIe Traffic Class (TC)
308 * for Memory Write data and MSI, so there are no ordering
309 * issues; the MSI will always arrive at the Root Complex
310 * _after_ corresponding Memory Writes (i.e. descriptor
314 napi_schedule_irqoff(&enic->napi[0]);
319 static irqreturn_t enic_isr_msix(int irq, void *data)
321 struct napi_struct *napi = data;
323 napi_schedule_irqoff(napi);
328 static irqreturn_t enic_isr_msix_err(int irq, void *data)
330 struct enic *enic = data;
331 unsigned int intr = enic_msix_err_intr(enic);
333 vnic_intr_return_all_credits(&enic->intr[intr]);
335 enic_log_q_error(enic);
337 /* schedule recovery from WQ/RQ error */
338 schedule_work(&enic->reset);
343 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
345 struct enic *enic = data;
346 unsigned int intr = enic_msix_notify_intr(enic);
348 vnic_intr_return_all_credits(&enic->intr[intr]);
349 enic_notify_check(enic);
354 static inline void enic_queue_wq_skb_cont(struct enic *enic,
355 struct vnic_wq *wq, struct sk_buff *skb,
356 unsigned int len_left, int loopback)
358 const skb_frag_t *frag;
360 /* Queue additional data fragments */
361 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
362 len_left -= skb_frag_size(frag);
363 enic_queue_wq_desc_cont(wq, skb,
364 skb_frag_dma_map(&enic->pdev->dev,
365 frag, 0, skb_frag_size(frag),
368 (len_left == 0), /* EOP? */
373 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
374 struct vnic_wq *wq, struct sk_buff *skb,
375 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
377 unsigned int head_len = skb_headlen(skb);
378 unsigned int len_left = skb->len - head_len;
379 int eop = (len_left == 0);
381 /* Queue the main skb fragment. The fragments are no larger
382 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
383 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
384 * per fragment is queued.
386 enic_queue_wq_desc(wq, skb,
387 pci_map_single(enic->pdev, skb->data,
388 head_len, PCI_DMA_TODEVICE),
390 vlan_tag_insert, vlan_tag,
394 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
397 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
398 struct vnic_wq *wq, struct sk_buff *skb,
399 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
401 unsigned int head_len = skb_headlen(skb);
402 unsigned int len_left = skb->len - head_len;
403 unsigned int hdr_len = skb_checksum_start_offset(skb);
404 unsigned int csum_offset = hdr_len + skb->csum_offset;
405 int eop = (len_left == 0);
407 /* Queue the main skb fragment. The fragments are no larger
408 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
409 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
410 * per fragment is queued.
412 enic_queue_wq_desc_csum_l4(wq, skb,
413 pci_map_single(enic->pdev, skb->data,
414 head_len, PCI_DMA_TODEVICE),
418 vlan_tag_insert, vlan_tag,
422 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
425 static inline void enic_queue_wq_skb_tso(struct enic *enic,
426 struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
427 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
429 unsigned int frag_len_left = skb_headlen(skb);
430 unsigned int len_left = skb->len - frag_len_left;
431 unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
432 int eop = (len_left == 0);
435 unsigned int offset = 0;
438 /* Preload TCP csum field with IP pseudo hdr calculated
439 * with IP length set to zero. HW will later add in length
440 * to each TCP segment resulting from the TSO.
443 if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
444 ip_hdr(skb)->check = 0;
445 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
446 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
447 } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
448 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
449 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
452 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
453 * for the main skb fragment
455 while (frag_len_left) {
456 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
457 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
458 len, PCI_DMA_TODEVICE);
459 enic_queue_wq_desc_tso(wq, skb,
463 vlan_tag_insert, vlan_tag,
464 eop && (len == frag_len_left), loopback);
465 frag_len_left -= len;
472 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
473 * for additional data fragments
475 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
476 len_left -= skb_frag_size(frag);
477 frag_len_left = skb_frag_size(frag);
480 while (frag_len_left) {
481 len = min(frag_len_left,
482 (unsigned int)WQ_ENET_MAX_DESC_LEN);
483 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
486 enic_queue_wq_desc_cont(wq, skb,
490 (len == frag_len_left), /* EOP? */
492 frag_len_left -= len;
498 static inline void enic_queue_wq_skb(struct enic *enic,
499 struct vnic_wq *wq, struct sk_buff *skb)
501 unsigned int mss = skb_shinfo(skb)->gso_size;
502 unsigned int vlan_tag = 0;
503 int vlan_tag_insert = 0;
506 if (vlan_tx_tag_present(skb)) {
507 /* VLAN tag from trunking driver */
509 vlan_tag = vlan_tx_tag_get(skb);
510 } else if (enic->loop_enable) {
511 vlan_tag = enic->loop_tag;
516 enic_queue_wq_skb_tso(enic, wq, skb, mss,
517 vlan_tag_insert, vlan_tag, loopback);
518 else if (skb->ip_summed == CHECKSUM_PARTIAL)
519 enic_queue_wq_skb_csum_l4(enic, wq, skb,
520 vlan_tag_insert, vlan_tag, loopback);
522 enic_queue_wq_skb_vlan(enic, wq, skb,
523 vlan_tag_insert, vlan_tag, loopback);
526 /* netif_tx_lock held, process context with BHs disabled, or BH */
527 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
528 struct net_device *netdev)
530 struct enic *enic = netdev_priv(netdev);
532 unsigned int txq_map;
533 struct netdev_queue *txq;
536 dev_kfree_skb_any(skb);
540 txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
541 wq = &enic->wq[txq_map];
542 txq = netdev_get_tx_queue(netdev, txq_map);
544 /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
545 * which is very likely. In the off chance it's going to take
546 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
549 if (skb_shinfo(skb)->gso_size == 0 &&
550 skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
551 skb_linearize(skb)) {
552 dev_kfree_skb_any(skb);
556 spin_lock(&enic->wq_lock[txq_map]);
558 if (vnic_wq_desc_avail(wq) <
559 skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
560 netif_tx_stop_queue(txq);
561 /* This is a hard error, log it */
562 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
563 spin_unlock(&enic->wq_lock[txq_map]);
564 return NETDEV_TX_BUSY;
567 enic_queue_wq_skb(enic, wq, skb);
569 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
570 netif_tx_stop_queue(txq);
571 if (!skb->xmit_more || netif_xmit_stopped(txq))
572 vnic_wq_doorbell(wq);
574 spin_unlock(&enic->wq_lock[txq_map]);
579 /* dev_base_lock rwlock held, nominally process context */
580 static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
581 struct rtnl_link_stats64 *net_stats)
583 struct enic *enic = netdev_priv(netdev);
584 struct vnic_stats *stats;
586 enic_dev_stats_dump(enic, &stats);
588 net_stats->tx_packets = stats->tx.tx_frames_ok;
589 net_stats->tx_bytes = stats->tx.tx_bytes_ok;
590 net_stats->tx_errors = stats->tx.tx_errors;
591 net_stats->tx_dropped = stats->tx.tx_drops;
593 net_stats->rx_packets = stats->rx.rx_frames_ok;
594 net_stats->rx_bytes = stats->rx.rx_bytes_ok;
595 net_stats->rx_errors = stats->rx.rx_errors;
596 net_stats->multicast = stats->rx.rx_multicast_frames_ok;
597 net_stats->rx_over_errors = enic->rq_truncated_pkts;
598 net_stats->rx_crc_errors = enic->rq_bad_fcs;
599 net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
604 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
606 struct enic *enic = netdev_priv(netdev);
608 if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
609 unsigned int mc_count = netdev_mc_count(netdev);
611 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
612 ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
617 enic_dev_add_addr(enic, mc_addr);
623 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
625 struct enic *enic = netdev_priv(netdev);
627 enic_dev_del_addr(enic, mc_addr);
633 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
635 struct enic *enic = netdev_priv(netdev);
637 if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
638 unsigned int uc_count = netdev_uc_count(netdev);
640 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
641 ENIC_UNICAST_PERFECT_FILTERS, uc_count);
646 enic_dev_add_addr(enic, uc_addr);
652 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
654 struct enic *enic = netdev_priv(netdev);
656 enic_dev_del_addr(enic, uc_addr);
662 void enic_reset_addr_lists(struct enic *enic)
664 struct net_device *netdev = enic->netdev;
666 __dev_uc_unsync(netdev, NULL);
667 __dev_mc_unsync(netdev, NULL);
674 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
676 struct enic *enic = netdev_priv(netdev);
678 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
679 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
680 return -EADDRNOTAVAIL;
682 if (!is_valid_ether_addr(addr))
683 return -EADDRNOTAVAIL;
686 memcpy(netdev->dev_addr, addr, netdev->addr_len);
691 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
693 struct enic *enic = netdev_priv(netdev);
694 struct sockaddr *saddr = p;
695 char *addr = saddr->sa_data;
698 if (netif_running(enic->netdev)) {
699 err = enic_dev_del_station_addr(enic);
704 err = enic_set_mac_addr(netdev, addr);
708 if (netif_running(enic->netdev)) {
709 err = enic_dev_add_station_addr(enic);
717 static int enic_set_mac_address(struct net_device *netdev, void *p)
719 struct sockaddr *saddr = p;
720 char *addr = saddr->sa_data;
721 struct enic *enic = netdev_priv(netdev);
724 err = enic_dev_del_station_addr(enic);
728 err = enic_set_mac_addr(netdev, addr);
732 return enic_dev_add_station_addr(enic);
735 /* netif_tx_lock held, BHs disabled */
736 static void enic_set_rx_mode(struct net_device *netdev)
738 struct enic *enic = netdev_priv(netdev);
740 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
741 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
742 int promisc = (netdev->flags & IFF_PROMISC) ||
743 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
744 int allmulti = (netdev->flags & IFF_ALLMULTI) ||
745 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
746 unsigned int flags = netdev->flags |
747 (allmulti ? IFF_ALLMULTI : 0) |
748 (promisc ? IFF_PROMISC : 0);
750 if (enic->flags != flags) {
752 enic_dev_packet_filter(enic, directed,
753 multicast, broadcast, promisc, allmulti);
757 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
759 __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
763 /* netif_tx_lock held, BHs disabled */
764 static void enic_tx_timeout(struct net_device *netdev)
766 struct enic *enic = netdev_priv(netdev);
767 schedule_work(&enic->reset);
770 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
772 struct enic *enic = netdev_priv(netdev);
773 struct enic_port_profile *pp;
776 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
780 if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
781 if (vf == PORT_SELF_VF) {
782 memcpy(pp->vf_mac, mac, ETH_ALEN);
786 * For sriov vf's set the mac in hw
788 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
789 vnic_dev_set_mac_addr, mac);
790 return enic_dev_status_to_errno(err);
796 static int enic_set_vf_port(struct net_device *netdev, int vf,
797 struct nlattr *port[])
799 struct enic *enic = netdev_priv(netdev);
800 struct enic_port_profile prev_pp;
801 struct enic_port_profile *pp;
802 int err = 0, restore_pp = 1;
804 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
808 if (!port[IFLA_PORT_REQUEST])
811 memcpy(&prev_pp, pp, sizeof(*enic->pp));
812 memset(pp, 0, sizeof(*enic->pp));
814 pp->set |= ENIC_SET_REQUEST;
815 pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
817 if (port[IFLA_PORT_PROFILE]) {
818 pp->set |= ENIC_SET_NAME;
819 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
823 if (port[IFLA_PORT_INSTANCE_UUID]) {
824 pp->set |= ENIC_SET_INSTANCE;
825 memcpy(pp->instance_uuid,
826 nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
829 if (port[IFLA_PORT_HOST_UUID]) {
830 pp->set |= ENIC_SET_HOST;
831 memcpy(pp->host_uuid,
832 nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
835 if (vf == PORT_SELF_VF) {
836 /* Special case handling: mac came from IFLA_VF_MAC */
837 if (!is_zero_ether_addr(prev_pp.vf_mac))
838 memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
840 if (is_zero_ether_addr(netdev->dev_addr))
841 eth_hw_addr_random(netdev);
843 /* SR-IOV VF: get mac from adapter */
844 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
845 vnic_dev_get_mac_addr, pp->mac_addr);
847 netdev_err(netdev, "Error getting mac for vf %d\n", vf);
848 memcpy(pp, &prev_pp, sizeof(*pp));
849 return enic_dev_status_to_errno(err);
853 err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
856 /* Things are still the way they were: Implicit
857 * DISASSOCIATE failed
859 memcpy(pp, &prev_pp, sizeof(*pp));
861 memset(pp, 0, sizeof(*pp));
862 if (vf == PORT_SELF_VF)
863 memset(netdev->dev_addr, 0, ETH_ALEN);
866 /* Set flag to indicate that the port assoc/disassoc
867 * request has been sent out to fw
869 pp->set |= ENIC_PORT_REQUEST_APPLIED;
871 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
872 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
873 memset(pp->mac_addr, 0, ETH_ALEN);
874 if (vf == PORT_SELF_VF)
875 memset(netdev->dev_addr, 0, ETH_ALEN);
879 if (vf == PORT_SELF_VF)
880 memset(pp->vf_mac, 0, ETH_ALEN);
885 static int enic_get_vf_port(struct net_device *netdev, int vf,
888 struct enic *enic = netdev_priv(netdev);
889 u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
890 struct enic_port_profile *pp;
893 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
897 if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
900 err = enic_process_get_pp_request(enic, vf, pp->request, &response);
904 if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
905 nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
906 ((pp->set & ENIC_SET_NAME) &&
907 nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
908 ((pp->set & ENIC_SET_INSTANCE) &&
909 nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
910 pp->instance_uuid)) ||
911 ((pp->set & ENIC_SET_HOST) &&
912 nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
913 goto nla_put_failure;
920 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
922 struct enic *enic = vnic_dev_priv(rq->vdev);
927 pci_unmap_single(enic->pdev, buf->dma_addr,
928 buf->len, PCI_DMA_FROMDEVICE);
929 dev_kfree_skb_any(buf->os_buf);
933 static int enic_rq_alloc_buf(struct vnic_rq *rq)
935 struct enic *enic = vnic_dev_priv(rq->vdev);
936 struct net_device *netdev = enic->netdev;
938 unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
939 unsigned int os_buf_index = 0;
941 struct vnic_rq_buf *buf = rq->to_use;
944 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
949 skb = netdev_alloc_skb_ip_align(netdev, len);
953 dma_addr = pci_map_single(enic->pdev, skb->data,
954 len, PCI_DMA_FROMDEVICE);
956 enic_queue_rq_desc(rq, skb, os_buf_index,
962 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
965 if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
966 pkt_size->large_pkt_bytes_cnt += pkt_len;
968 pkt_size->small_pkt_bytes_cnt += pkt_len;
971 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
972 struct vnic_rq_buf *buf, u16 len)
974 struct enic *enic = netdev_priv(netdev);
975 struct sk_buff *new_skb;
977 if (len > enic->rx_copybreak)
979 new_skb = netdev_alloc_skb_ip_align(netdev, len);
982 pci_dma_sync_single_for_cpu(enic->pdev, buf->dma_addr, len,
984 memcpy(new_skb->data, (*skb)->data, len);
990 static void enic_rq_indicate_buf(struct vnic_rq *rq,
991 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
992 int skipped, void *opaque)
994 struct enic *enic = vnic_dev_priv(rq->vdev);
995 struct net_device *netdev = enic->netdev;
997 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
999 u8 type, color, eop, sop, ingress_port, vlan_stripped;
1000 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1001 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1002 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1004 u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1012 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1013 &type, &color, &q_number, &completed_index,
1014 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1015 &csum_not_calc, &rss_hash, &bytes_written,
1016 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1017 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1018 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1019 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1025 if (bytes_written > 0)
1027 else if (bytes_written == 0)
1028 enic->rq_truncated_pkts++;
1031 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1032 PCI_DMA_FROMDEVICE);
1033 dev_kfree_skb_any(skb);
1039 if (eop && bytes_written > 0) {
1044 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1046 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1047 PCI_DMA_FROMDEVICE);
1049 prefetch(skb->data - NET_IP_ALIGN);
1051 skb_put(skb, bytes_written);
1052 skb->protocol = eth_type_trans(skb, netdev);
1053 skb_record_rx_queue(skb, q_number);
1054 if (netdev->features & NETIF_F_RXHASH) {
1055 skb_set_hash(skb, rss_hash,
1057 (NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX |
1058 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6 |
1059 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4)) ?
1060 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1063 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc) {
1064 skb->csum = htons(checksum);
1065 skb->ip_summed = CHECKSUM_COMPLETE;
1069 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1071 skb_mark_napi_id(skb, &enic->napi[rq->index]);
1072 if (enic_poll_busy_polling(rq) ||
1073 !(netdev->features & NETIF_F_GRO))
1074 netif_receive_skb(skb);
1076 napi_gro_receive(&enic->napi[q_number], skb);
1077 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1078 enic_intr_update_pkt_size(&cq->pkt_size_counter,
1085 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1086 PCI_DMA_FROMDEVICE);
1087 dev_kfree_skb_any(skb);
1092 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1093 u8 type, u16 q_number, u16 completed_index, void *opaque)
1095 struct enic *enic = vnic_dev_priv(vdev);
1097 vnic_rq_service(&enic->rq[q_number], cq_desc,
1098 completed_index, VNIC_RQ_RETURN_DESC,
1099 enic_rq_indicate_buf, opaque);
1104 static int enic_poll(struct napi_struct *napi, int budget)
1106 struct net_device *netdev = napi->dev;
1107 struct enic *enic = netdev_priv(netdev);
1108 unsigned int cq_rq = enic_cq_rq(enic, 0);
1109 unsigned int cq_wq = enic_cq_wq(enic, 0);
1110 unsigned int intr = enic_legacy_io_intr();
1111 unsigned int rq_work_to_do = budget;
1112 unsigned int wq_work_to_do = -1; /* no limit */
1113 unsigned int work_done, rq_work_done = 0, wq_work_done;
1116 wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1117 enic_wq_service, NULL);
1119 if (!enic_poll_lock_napi(&enic->rq[cq_rq])) {
1120 if (wq_work_done > 0)
1121 vnic_intr_return_credits(&enic->intr[intr],
1123 0 /* dont unmask intr */,
1124 0 /* dont reset intr timer */);
1125 return rq_work_done;
1129 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1130 rq_work_to_do, enic_rq_service, NULL);
1132 /* Accumulate intr event credits for this polling
1133 * cycle. An intr event is the completion of a
1134 * a WQ or RQ packet.
1137 work_done = rq_work_done + wq_work_done;
1140 vnic_intr_return_credits(&enic->intr[intr],
1142 0 /* don't unmask intr */,
1143 0 /* don't reset intr timer */);
1145 err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1147 /* Buffer allocation failed. Stay in polling
1148 * mode so we can try to fill the ring again.
1152 rq_work_done = rq_work_to_do;
1154 if (rq_work_done < rq_work_to_do) {
1156 /* Some work done, but not enough to stay in polling,
1160 napi_complete(napi);
1161 vnic_intr_unmask(&enic->intr[intr]);
1163 enic_poll_unlock_napi(&enic->rq[cq_rq]);
1165 return rq_work_done;
1168 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1170 unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1171 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1172 u32 timer = cq->tobe_rx_coal_timeval;
1174 if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1175 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1176 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1180 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1182 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1183 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1184 struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1190 ktime_t now = ktime_get();
1192 delta = ktime_us_delta(now, cq->prev_ts);
1193 if (delta < ENIC_AIC_TS_BREAK)
1197 traffic = pkt_size_counter->large_pkt_bytes_cnt +
1198 pkt_size_counter->small_pkt_bytes_cnt;
1199 /* The table takes Mbps
1200 * traffic *= 8 => bits
1201 * traffic *= (10^6 / delta) => bps
1202 * traffic /= 10^6 => Mbps
1204 * Combining, traffic *= (8 / delta)
1208 traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1210 for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1211 if (traffic < mod_table[index].rx_rate)
1213 range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1214 pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1215 rx_coal->small_pkt_range_start :
1216 rx_coal->large_pkt_range_start;
1217 timer = range_start + ((rx_coal->range_end - range_start) *
1218 mod_table[index].range_percent / 100);
1220 cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1222 pkt_size_counter->large_pkt_bytes_cnt = 0;
1223 pkt_size_counter->small_pkt_bytes_cnt = 0;
1226 #ifdef CONFIG_RFS_ACCEL
1227 static void enic_free_rx_cpu_rmap(struct enic *enic)
1229 free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1230 enic->netdev->rx_cpu_rmap = NULL;
1233 static void enic_set_rx_cpu_rmap(struct enic *enic)
1237 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1238 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1239 if (unlikely(!enic->netdev->rx_cpu_rmap))
1241 for (i = 0; i < enic->rq_count; i++) {
1242 res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1243 enic->msix_entry[i].vector);
1244 if (unlikely(res)) {
1245 enic_free_rx_cpu_rmap(enic);
1254 static void enic_free_rx_cpu_rmap(struct enic *enic)
1258 static void enic_set_rx_cpu_rmap(struct enic *enic)
1262 #endif /* CONFIG_RFS_ACCEL */
1264 #ifdef CONFIG_NET_RX_BUSY_POLL
1265 int enic_busy_poll(struct napi_struct *napi)
1267 struct net_device *netdev = napi->dev;
1268 struct enic *enic = netdev_priv(netdev);
1269 unsigned int rq = (napi - &enic->napi[0]);
1270 unsigned int cq = enic_cq_rq(enic, rq);
1271 unsigned int intr = enic_msix_rq_intr(enic, rq);
1272 unsigned int work_to_do = -1; /* clean all pkts possible */
1273 unsigned int work_done;
1275 if (!enic_poll_lock_poll(&enic->rq[rq]))
1276 return LL_FLUSH_BUSY;
1277 work_done = vnic_cq_service(&enic->cq[cq], work_to_do,
1278 enic_rq_service, NULL);
1281 vnic_intr_return_credits(&enic->intr[intr],
1283 vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1284 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1285 enic_calc_int_moderation(enic, &enic->rq[rq]);
1286 enic_poll_unlock_poll(&enic->rq[rq]);
1290 #endif /* CONFIG_NET_RX_BUSY_POLL */
1292 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1294 struct net_device *netdev = napi->dev;
1295 struct enic *enic = netdev_priv(netdev);
1296 unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1297 struct vnic_wq *wq = &enic->wq[wq_index];
1300 unsigned int wq_work_to_do = -1; /* clean all desc possible */
1301 unsigned int wq_work_done;
1302 unsigned int wq_irq;
1305 cq = enic_cq_wq(enic, wq_irq);
1306 intr = enic_msix_wq_intr(enic, wq_irq);
1307 wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1308 enic_wq_service, NULL);
1310 vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1311 0 /* don't unmask intr */,
1312 1 /* reset intr timer */);
1313 if (!wq_work_done) {
1314 napi_complete(napi);
1315 vnic_intr_unmask(&enic->intr[intr]);
1322 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1324 struct net_device *netdev = napi->dev;
1325 struct enic *enic = netdev_priv(netdev);
1326 unsigned int rq = (napi - &enic->napi[0]);
1327 unsigned int cq = enic_cq_rq(enic, rq);
1328 unsigned int intr = enic_msix_rq_intr(enic, rq);
1329 unsigned int work_to_do = budget;
1330 unsigned int work_done = 0;
1333 if (!enic_poll_lock_napi(&enic->rq[rq]))
1339 work_done = vnic_cq_service(&enic->cq[cq],
1340 work_to_do, enic_rq_service, NULL);
1342 /* Return intr event credits for this polling
1343 * cycle. An intr event is the completion of a
1348 vnic_intr_return_credits(&enic->intr[intr],
1350 0 /* don't unmask intr */,
1351 0 /* don't reset intr timer */);
1353 err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1355 /* Buffer allocation failed. Stay in polling mode
1356 * so we can try to fill the ring again.
1360 work_done = work_to_do;
1361 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1362 /* Call the function which refreshes
1363 * the intr coalescing timer value based on
1364 * the traffic. This is supported only in
1365 * the case of MSI-x mode
1367 enic_calc_int_moderation(enic, &enic->rq[rq]);
1369 if (work_done < work_to_do) {
1371 /* Some work done, but not enough to stay in polling,
1375 napi_complete(napi);
1376 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1377 enic_set_int_moderation(enic, &enic->rq[rq]);
1378 vnic_intr_unmask(&enic->intr[intr]);
1380 enic_poll_unlock_napi(&enic->rq[rq]);
1385 static void enic_notify_timer(unsigned long data)
1387 struct enic *enic = (struct enic *)data;
1389 enic_notify_check(enic);
1391 mod_timer(&enic->notify_timer,
1392 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1395 static void enic_free_intr(struct enic *enic)
1397 struct net_device *netdev = enic->netdev;
1400 enic_free_rx_cpu_rmap(enic);
1401 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1402 case VNIC_DEV_INTR_MODE_INTX:
1403 free_irq(enic->pdev->irq, netdev);
1405 case VNIC_DEV_INTR_MODE_MSI:
1406 free_irq(enic->pdev->irq, enic);
1408 case VNIC_DEV_INTR_MODE_MSIX:
1409 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1410 if (enic->msix[i].requested)
1411 free_irq(enic->msix_entry[i].vector,
1412 enic->msix[i].devid);
1419 static int enic_request_intr(struct enic *enic)
1421 struct net_device *netdev = enic->netdev;
1422 unsigned int i, intr;
1425 enic_set_rx_cpu_rmap(enic);
1426 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1428 case VNIC_DEV_INTR_MODE_INTX:
1430 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1431 IRQF_SHARED, netdev->name, netdev);
1434 case VNIC_DEV_INTR_MODE_MSI:
1436 err = request_irq(enic->pdev->irq, enic_isr_msi,
1437 0, netdev->name, enic);
1440 case VNIC_DEV_INTR_MODE_MSIX:
1442 for (i = 0; i < enic->rq_count; i++) {
1443 intr = enic_msix_rq_intr(enic, i);
1444 snprintf(enic->msix[intr].devname,
1445 sizeof(enic->msix[intr].devname),
1446 "%.11s-rx-%d", netdev->name, i);
1447 enic->msix[intr].isr = enic_isr_msix;
1448 enic->msix[intr].devid = &enic->napi[i];
1451 for (i = 0; i < enic->wq_count; i++) {
1452 int wq = enic_cq_wq(enic, i);
1454 intr = enic_msix_wq_intr(enic, i);
1455 snprintf(enic->msix[intr].devname,
1456 sizeof(enic->msix[intr].devname),
1457 "%.11s-tx-%d", netdev->name, i);
1458 enic->msix[intr].isr = enic_isr_msix;
1459 enic->msix[intr].devid = &enic->napi[wq];
1462 intr = enic_msix_err_intr(enic);
1463 snprintf(enic->msix[intr].devname,
1464 sizeof(enic->msix[intr].devname),
1465 "%.11s-err", netdev->name);
1466 enic->msix[intr].isr = enic_isr_msix_err;
1467 enic->msix[intr].devid = enic;
1469 intr = enic_msix_notify_intr(enic);
1470 snprintf(enic->msix[intr].devname,
1471 sizeof(enic->msix[intr].devname),
1472 "%.11s-notify", netdev->name);
1473 enic->msix[intr].isr = enic_isr_msix_notify;
1474 enic->msix[intr].devid = enic;
1476 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1477 enic->msix[i].requested = 0;
1479 for (i = 0; i < enic->intr_count; i++) {
1480 err = request_irq(enic->msix_entry[i].vector,
1481 enic->msix[i].isr, 0,
1482 enic->msix[i].devname,
1483 enic->msix[i].devid);
1485 enic_free_intr(enic);
1488 enic->msix[i].requested = 1;
1500 static void enic_synchronize_irqs(struct enic *enic)
1504 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1505 case VNIC_DEV_INTR_MODE_INTX:
1506 case VNIC_DEV_INTR_MODE_MSI:
1507 synchronize_irq(enic->pdev->irq);
1509 case VNIC_DEV_INTR_MODE_MSIX:
1510 for (i = 0; i < enic->intr_count; i++)
1511 synchronize_irq(enic->msix_entry[i].vector);
1518 static void enic_set_rx_coal_setting(struct enic *enic)
1522 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1524 /* If intr mode is not MSIX, do not do adaptive coalescing */
1525 if (VNIC_DEV_INTR_MODE_MSIX != vnic_dev_get_intr_mode(enic->vdev)) {
1526 netdev_info(enic->netdev, "INTR mode is not MSIX, Not initializing adaptive coalescing");
1530 /* 1. Read the link speed from fw
1531 * 2. Pick the default range for the speed
1532 * 3. Update it in enic->rx_coalesce_setting
1534 speed = vnic_dev_port_speed(enic->vdev);
1535 if (ENIC_LINK_SPEED_10G < speed)
1536 index = ENIC_LINK_40G_INDEX;
1537 else if (ENIC_LINK_SPEED_4G < speed)
1538 index = ENIC_LINK_10G_INDEX;
1540 index = ENIC_LINK_4G_INDEX;
1542 rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1543 rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1544 rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1546 /* Start with the value provided by UCSM */
1547 for (index = 0; index < enic->rq_count; index++)
1548 enic->cq[index].cur_rx_coal_timeval =
1549 enic->config.intr_timer_usec;
1551 rx_coal->use_adaptive_rx_coalesce = 1;
1554 static int enic_dev_notify_set(struct enic *enic)
1558 spin_lock_bh(&enic->devcmd_lock);
1559 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1560 case VNIC_DEV_INTR_MODE_INTX:
1561 err = vnic_dev_notify_set(enic->vdev,
1562 enic_legacy_notify_intr());
1564 case VNIC_DEV_INTR_MODE_MSIX:
1565 err = vnic_dev_notify_set(enic->vdev,
1566 enic_msix_notify_intr(enic));
1569 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1572 spin_unlock_bh(&enic->devcmd_lock);
1577 static void enic_notify_timer_start(struct enic *enic)
1579 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1580 case VNIC_DEV_INTR_MODE_MSI:
1581 mod_timer(&enic->notify_timer, jiffies);
1584 /* Using intr for notification for INTx/MSI-X */
1589 /* rtnl lock is held, process context */
1590 static int enic_open(struct net_device *netdev)
1592 struct enic *enic = netdev_priv(netdev);
1596 err = enic_request_intr(enic);
1598 netdev_err(netdev, "Unable to request irq.\n");
1602 err = enic_dev_notify_set(enic);
1605 "Failed to alloc notify buffer, aborting.\n");
1606 goto err_out_free_intr;
1609 for (i = 0; i < enic->rq_count; i++) {
1610 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1611 /* Need at least one buffer on ring to get going */
1612 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1613 netdev_err(netdev, "Unable to alloc receive buffers\n");
1615 goto err_out_notify_unset;
1619 for (i = 0; i < enic->wq_count; i++)
1620 vnic_wq_enable(&enic->wq[i]);
1621 for (i = 0; i < enic->rq_count; i++)
1622 vnic_rq_enable(&enic->rq[i]);
1624 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1625 enic_dev_add_station_addr(enic);
1627 enic_set_rx_mode(netdev);
1629 netif_tx_wake_all_queues(netdev);
1631 for (i = 0; i < enic->rq_count; i++) {
1632 enic_busy_poll_init_lock(&enic->rq[i]);
1633 napi_enable(&enic->napi[i]);
1635 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1636 for (i = 0; i < enic->wq_count; i++)
1637 napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
1638 enic_dev_enable(enic);
1640 for (i = 0; i < enic->intr_count; i++)
1641 vnic_intr_unmask(&enic->intr[i]);
1643 enic_notify_timer_start(enic);
1644 enic_rfs_flw_tbl_init(enic);
1648 err_out_notify_unset:
1649 enic_dev_notify_unset(enic);
1651 enic_free_intr(enic);
1656 /* rtnl lock is held, process context */
1657 static int enic_stop(struct net_device *netdev)
1659 struct enic *enic = netdev_priv(netdev);
1663 for (i = 0; i < enic->intr_count; i++) {
1664 vnic_intr_mask(&enic->intr[i]);
1665 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1668 enic_synchronize_irqs(enic);
1670 del_timer_sync(&enic->notify_timer);
1671 enic_rfs_flw_tbl_free(enic);
1673 enic_dev_disable(enic);
1675 for (i = 0; i < enic->rq_count; i++) {
1676 napi_disable(&enic->napi[i]);
1678 while (!enic_poll_lock_napi(&enic->rq[i]))
1683 netif_carrier_off(netdev);
1684 netif_tx_disable(netdev);
1685 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1686 for (i = 0; i < enic->wq_count; i++)
1687 napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
1689 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1690 enic_dev_del_station_addr(enic);
1692 for (i = 0; i < enic->wq_count; i++) {
1693 err = vnic_wq_disable(&enic->wq[i]);
1697 for (i = 0; i < enic->rq_count; i++) {
1698 err = vnic_rq_disable(&enic->rq[i]);
1703 enic_dev_notify_unset(enic);
1704 enic_free_intr(enic);
1706 for (i = 0; i < enic->wq_count; i++)
1707 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1708 for (i = 0; i < enic->rq_count; i++)
1709 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1710 for (i = 0; i < enic->cq_count; i++)
1711 vnic_cq_clean(&enic->cq[i]);
1712 for (i = 0; i < enic->intr_count; i++)
1713 vnic_intr_clean(&enic->intr[i]);
1718 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1720 struct enic *enic = netdev_priv(netdev);
1721 int running = netif_running(netdev);
1723 if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1726 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
1732 netdev->mtu = new_mtu;
1734 if (netdev->mtu > enic->port_mtu)
1736 "interface MTU (%d) set higher than port MTU (%d)\n",
1737 netdev->mtu, enic->port_mtu);
1745 static void enic_change_mtu_work(struct work_struct *work)
1747 struct enic *enic = container_of(work, struct enic, change_mtu_work);
1748 struct net_device *netdev = enic->netdev;
1749 int new_mtu = vnic_dev_mtu(enic->vdev);
1753 new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
1758 del_timer_sync(&enic->notify_timer);
1760 for (i = 0; i < enic->rq_count; i++)
1761 napi_disable(&enic->napi[i]);
1763 vnic_intr_mask(&enic->intr[0]);
1764 enic_synchronize_irqs(enic);
1765 err = vnic_rq_disable(&enic->rq[0]);
1768 netdev_err(netdev, "Unable to disable RQ.\n");
1771 vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
1772 vnic_cq_clean(&enic->cq[0]);
1773 vnic_intr_clean(&enic->intr[0]);
1775 /* Fill RQ with new_mtu-sized buffers */
1776 netdev->mtu = new_mtu;
1777 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1778 /* Need at least one buffer on ring to get going */
1779 if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
1781 netdev_err(netdev, "Unable to alloc receive buffers.\n");
1786 vnic_rq_enable(&enic->rq[0]);
1787 napi_enable(&enic->napi[0]);
1788 vnic_intr_unmask(&enic->intr[0]);
1789 enic_notify_timer_start(enic);
1793 netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1796 #ifdef CONFIG_NET_POLL_CONTROLLER
1797 static void enic_poll_controller(struct net_device *netdev)
1799 struct enic *enic = netdev_priv(netdev);
1800 struct vnic_dev *vdev = enic->vdev;
1801 unsigned int i, intr;
1803 switch (vnic_dev_get_intr_mode(vdev)) {
1804 case VNIC_DEV_INTR_MODE_MSIX:
1805 for (i = 0; i < enic->rq_count; i++) {
1806 intr = enic_msix_rq_intr(enic, i);
1807 enic_isr_msix(enic->msix_entry[intr].vector,
1811 for (i = 0; i < enic->wq_count; i++) {
1812 intr = enic_msix_wq_intr(enic, i);
1813 enic_isr_msix(enic->msix_entry[intr].vector,
1814 &enic->napi[enic_cq_wq(enic, i)]);
1818 case VNIC_DEV_INTR_MODE_MSI:
1819 enic_isr_msi(enic->pdev->irq, enic);
1821 case VNIC_DEV_INTR_MODE_INTX:
1822 enic_isr_legacy(enic->pdev->irq, netdev);
1830 static int enic_dev_wait(struct vnic_dev *vdev,
1831 int (*start)(struct vnic_dev *, int),
1832 int (*finished)(struct vnic_dev *, int *),
1839 BUG_ON(in_interrupt());
1841 err = start(vdev, arg);
1845 /* Wait for func to complete...2 seconds max
1848 time = jiffies + (HZ * 2);
1851 err = finished(vdev, &done);
1858 schedule_timeout_uninterruptible(HZ / 10);
1860 } while (time_after(time, jiffies));
1865 static int enic_dev_open(struct enic *enic)
1869 err = enic_dev_wait(enic->vdev, vnic_dev_open,
1870 vnic_dev_open_done, 0);
1872 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1878 static int enic_dev_hang_reset(struct enic *enic)
1882 err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1883 vnic_dev_hang_reset_done, 0);
1885 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1891 static int enic_set_rsskey(struct enic *enic)
1893 dma_addr_t rss_key_buf_pa;
1894 union vnic_rss_key *rss_key_buf_va = NULL;
1895 union vnic_rss_key rss_key = {
1896 .key[0].b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101},
1897 .key[1].b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101},
1898 .key[2].b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115},
1899 .key[3].b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108},
1903 rss_key_buf_va = pci_alloc_consistent(enic->pdev,
1904 sizeof(union vnic_rss_key), &rss_key_buf_pa);
1905 if (!rss_key_buf_va)
1908 memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
1910 spin_lock_bh(&enic->devcmd_lock);
1911 err = enic_set_rss_key(enic,
1913 sizeof(union vnic_rss_key));
1914 spin_unlock_bh(&enic->devcmd_lock);
1916 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
1917 rss_key_buf_va, rss_key_buf_pa);
1922 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
1924 dma_addr_t rss_cpu_buf_pa;
1925 union vnic_rss_cpu *rss_cpu_buf_va = NULL;
1929 rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
1930 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
1931 if (!rss_cpu_buf_va)
1934 for (i = 0; i < (1 << rss_hash_bits); i++)
1935 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
1937 spin_lock_bh(&enic->devcmd_lock);
1938 err = enic_set_rss_cpu(enic,
1940 sizeof(union vnic_rss_cpu));
1941 spin_unlock_bh(&enic->devcmd_lock);
1943 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
1944 rss_cpu_buf_va, rss_cpu_buf_pa);
1949 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
1950 u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
1952 const u8 tso_ipid_split_en = 0;
1953 const u8 ig_vlan_strip_en = 1;
1956 /* Enable VLAN tag stripping.
1959 spin_lock_bh(&enic->devcmd_lock);
1960 err = enic_set_nic_cfg(enic,
1961 rss_default_cpu, rss_hash_type,
1962 rss_hash_bits, rss_base_cpu,
1963 rss_enable, tso_ipid_split_en,
1965 spin_unlock_bh(&enic->devcmd_lock);
1970 static int enic_set_rss_nic_cfg(struct enic *enic)
1972 struct device *dev = enic_get_dev(enic);
1973 const u8 rss_default_cpu = 0;
1974 const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1975 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1976 NIC_CFG_RSS_HASH_TYPE_IPV6 |
1977 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1978 const u8 rss_hash_bits = 7;
1979 const u8 rss_base_cpu = 0;
1980 u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1983 if (!enic_set_rsskey(enic)) {
1984 if (enic_set_rsscpu(enic, rss_hash_bits)) {
1986 dev_warn(dev, "RSS disabled, "
1987 "Failed to set RSS cpu indirection table.");
1991 dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
1995 return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1996 rss_hash_bits, rss_base_cpu, rss_enable);
1999 static void enic_reset(struct work_struct *work)
2001 struct enic *enic = container_of(work, struct enic, reset);
2003 if (!netif_running(enic->netdev))
2008 spin_lock(&enic->enic_api_lock);
2009 enic_dev_hang_notify(enic);
2010 enic_stop(enic->netdev);
2011 enic_dev_hang_reset(enic);
2012 enic_reset_addr_lists(enic);
2013 enic_init_vnic_resources(enic);
2014 enic_set_rss_nic_cfg(enic);
2015 enic_dev_set_ig_vlan_rewrite_mode(enic);
2016 enic_open(enic->netdev);
2017 spin_unlock(&enic->enic_api_lock);
2018 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2023 static int enic_set_intr_mode(struct enic *enic)
2025 unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2026 unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
2029 /* Set interrupt mode (INTx, MSI, MSI-X) depending
2030 * on system capabilities.
2034 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2035 * (the second to last INTR is used for WQ/RQ errors)
2036 * (the last INTR is used for notifications)
2039 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2040 for (i = 0; i < n + m + 2; i++)
2041 enic->msix_entry[i].entry = i;
2043 /* Use multiple RQs if RSS is enabled
2046 if (ENIC_SETTING(enic, RSS) &&
2047 enic->config.intr_mode < 1 &&
2048 enic->rq_count >= n &&
2049 enic->wq_count >= m &&
2050 enic->cq_count >= n + m &&
2051 enic->intr_count >= n + m + 2) {
2053 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2054 n + m + 2, n + m + 2) > 0) {
2058 enic->cq_count = n + m;
2059 enic->intr_count = n + m + 2;
2061 vnic_dev_set_intr_mode(enic->vdev,
2062 VNIC_DEV_INTR_MODE_MSIX);
2068 if (enic->config.intr_mode < 1 &&
2069 enic->rq_count >= 1 &&
2070 enic->wq_count >= m &&
2071 enic->cq_count >= 1 + m &&
2072 enic->intr_count >= 1 + m + 2) {
2073 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2074 1 + m + 2, 1 + m + 2) > 0) {
2078 enic->cq_count = 1 + m;
2079 enic->intr_count = 1 + m + 2;
2081 vnic_dev_set_intr_mode(enic->vdev,
2082 VNIC_DEV_INTR_MODE_MSIX);
2090 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2093 if (enic->config.intr_mode < 2 &&
2094 enic->rq_count >= 1 &&
2095 enic->wq_count >= 1 &&
2096 enic->cq_count >= 2 &&
2097 enic->intr_count >= 1 &&
2098 !pci_enable_msi(enic->pdev)) {
2103 enic->intr_count = 1;
2105 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2112 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2113 * (the first INTR is used for WQ/RQ)
2114 * (the second INTR is used for WQ/RQ errors)
2115 * (the last INTR is used for notifications)
2118 if (enic->config.intr_mode < 3 &&
2119 enic->rq_count >= 1 &&
2120 enic->wq_count >= 1 &&
2121 enic->cq_count >= 2 &&
2122 enic->intr_count >= 3) {
2127 enic->intr_count = 3;
2129 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2134 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2139 static void enic_clear_intr_mode(struct enic *enic)
2141 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2142 case VNIC_DEV_INTR_MODE_MSIX:
2143 pci_disable_msix(enic->pdev);
2145 case VNIC_DEV_INTR_MODE_MSI:
2146 pci_disable_msi(enic->pdev);
2152 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2155 static const struct net_device_ops enic_netdev_dynamic_ops = {
2156 .ndo_open = enic_open,
2157 .ndo_stop = enic_stop,
2158 .ndo_start_xmit = enic_hard_start_xmit,
2159 .ndo_get_stats64 = enic_get_stats,
2160 .ndo_validate_addr = eth_validate_addr,
2161 .ndo_set_rx_mode = enic_set_rx_mode,
2162 .ndo_set_mac_address = enic_set_mac_address_dynamic,
2163 .ndo_change_mtu = enic_change_mtu,
2164 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2165 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2166 .ndo_tx_timeout = enic_tx_timeout,
2167 .ndo_set_vf_port = enic_set_vf_port,
2168 .ndo_get_vf_port = enic_get_vf_port,
2169 .ndo_set_vf_mac = enic_set_vf_mac,
2170 #ifdef CONFIG_NET_POLL_CONTROLLER
2171 .ndo_poll_controller = enic_poll_controller,
2173 #ifdef CONFIG_RFS_ACCEL
2174 .ndo_rx_flow_steer = enic_rx_flow_steer,
2176 #ifdef CONFIG_NET_RX_BUSY_POLL
2177 .ndo_busy_poll = enic_busy_poll,
2181 static const struct net_device_ops enic_netdev_ops = {
2182 .ndo_open = enic_open,
2183 .ndo_stop = enic_stop,
2184 .ndo_start_xmit = enic_hard_start_xmit,
2185 .ndo_get_stats64 = enic_get_stats,
2186 .ndo_validate_addr = eth_validate_addr,
2187 .ndo_set_mac_address = enic_set_mac_address,
2188 .ndo_set_rx_mode = enic_set_rx_mode,
2189 .ndo_change_mtu = enic_change_mtu,
2190 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2191 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2192 .ndo_tx_timeout = enic_tx_timeout,
2193 .ndo_set_vf_port = enic_set_vf_port,
2194 .ndo_get_vf_port = enic_get_vf_port,
2195 .ndo_set_vf_mac = enic_set_vf_mac,
2196 #ifdef CONFIG_NET_POLL_CONTROLLER
2197 .ndo_poll_controller = enic_poll_controller,
2199 #ifdef CONFIG_RFS_ACCEL
2200 .ndo_rx_flow_steer = enic_rx_flow_steer,
2202 #ifdef CONFIG_NET_RX_BUSY_POLL
2203 .ndo_busy_poll = enic_busy_poll,
2207 static void enic_dev_deinit(struct enic *enic)
2211 for (i = 0; i < enic->rq_count; i++) {
2212 napi_hash_del(&enic->napi[i]);
2213 netif_napi_del(&enic->napi[i]);
2215 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2216 for (i = 0; i < enic->wq_count; i++)
2217 netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2219 enic_free_vnic_resources(enic);
2220 enic_clear_intr_mode(enic);
2223 static int enic_dev_init(struct enic *enic)
2225 struct device *dev = enic_get_dev(enic);
2226 struct net_device *netdev = enic->netdev;
2230 /* Get interrupt coalesce timer info */
2231 err = enic_dev_intr_coal_timer_info(enic);
2233 dev_warn(dev, "Using default conversion factor for "
2234 "interrupt coalesce timer\n");
2235 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2238 /* Get vNIC configuration
2241 err = enic_get_vnic_config(enic);
2243 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2247 /* Get available resource counts
2250 enic_get_res_counts(enic);
2252 /* Set interrupt mode based on resource counts and system
2256 err = enic_set_intr_mode(enic);
2258 dev_err(dev, "Failed to set intr mode based on resource "
2259 "counts and system capabilities, aborting\n");
2263 /* Allocate and configure vNIC resources
2266 err = enic_alloc_vnic_resources(enic);
2268 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2269 goto err_out_free_vnic_resources;
2272 enic_init_vnic_resources(enic);
2274 err = enic_set_rss_nic_cfg(enic);
2276 dev_err(dev, "Failed to config nic, aborting\n");
2277 goto err_out_free_vnic_resources;
2280 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2282 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2283 napi_hash_add(&enic->napi[0]);
2285 case VNIC_DEV_INTR_MODE_MSIX:
2286 for (i = 0; i < enic->rq_count; i++) {
2287 netif_napi_add(netdev, &enic->napi[i],
2288 enic_poll_msix_rq, NAPI_POLL_WEIGHT);
2289 napi_hash_add(&enic->napi[i]);
2291 for (i = 0; i < enic->wq_count; i++)
2292 netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],
2293 enic_poll_msix_wq, NAPI_POLL_WEIGHT);
2299 err_out_free_vnic_resources:
2300 enic_clear_intr_mode(enic);
2301 enic_free_vnic_resources(enic);
2306 static void enic_iounmap(struct enic *enic)
2310 for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2311 if (enic->bar[i].vaddr)
2312 iounmap(enic->bar[i].vaddr);
2315 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2317 struct device *dev = &pdev->dev;
2318 struct net_device *netdev;
2323 #ifdef CONFIG_PCI_IOV
2328 /* Allocate net device structure and initialize. Private
2329 * instance data is initialized to zero.
2332 netdev = alloc_etherdev_mqs(sizeof(struct enic),
2333 ENIC_RQ_MAX, ENIC_WQ_MAX);
2337 pci_set_drvdata(pdev, netdev);
2339 SET_NETDEV_DEV(netdev, &pdev->dev);
2341 enic = netdev_priv(netdev);
2342 enic->netdev = netdev;
2345 /* Setup PCI resources
2348 err = pci_enable_device_mem(pdev);
2350 dev_err(dev, "Cannot enable PCI device, aborting\n");
2351 goto err_out_free_netdev;
2354 err = pci_request_regions(pdev, DRV_NAME);
2356 dev_err(dev, "Cannot request PCI regions, aborting\n");
2357 goto err_out_disable_device;
2360 pci_set_master(pdev);
2362 /* Query PCI controller on system for DMA addressing
2363 * limitation for the device. Try 64-bit first, and
2367 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2369 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2371 dev_err(dev, "No usable DMA configuration, aborting\n");
2372 goto err_out_release_regions;
2374 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2376 dev_err(dev, "Unable to obtain %u-bit DMA "
2377 "for consistent allocations, aborting\n", 32);
2378 goto err_out_release_regions;
2381 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2383 dev_err(dev, "Unable to obtain %u-bit DMA "
2384 "for consistent allocations, aborting\n", 64);
2385 goto err_out_release_regions;
2390 /* Map vNIC resources from BAR0-5
2393 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2394 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2396 enic->bar[i].len = pci_resource_len(pdev, i);
2397 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2398 if (!enic->bar[i].vaddr) {
2399 dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2401 goto err_out_iounmap;
2403 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2406 /* Register vNIC device
2409 enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2410 ARRAY_SIZE(enic->bar));
2412 dev_err(dev, "vNIC registration failed, aborting\n");
2414 goto err_out_iounmap;
2417 #ifdef CONFIG_PCI_IOV
2418 /* Get number of subvnics */
2419 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2421 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2423 if (enic->num_vfs) {
2424 err = pci_enable_sriov(pdev, enic->num_vfs);
2426 dev_err(dev, "SRIOV enable failed, aborting."
2427 " pci_enable_sriov() returned %d\n",
2429 goto err_out_vnic_unregister;
2431 enic->priv_flags |= ENIC_SRIOV_ENABLED;
2432 num_pps = enic->num_vfs;
2437 /* Allocate structure for port profiles */
2438 enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2441 goto err_out_disable_sriov_pp;
2444 /* Issue device open to get device in known state
2447 err = enic_dev_open(enic);
2449 dev_err(dev, "vNIC dev open failed, aborting\n");
2450 goto err_out_disable_sriov;
2453 /* Setup devcmd lock
2456 spin_lock_init(&enic->devcmd_lock);
2457 spin_lock_init(&enic->enic_api_lock);
2460 * Set ingress vlan rewrite mode before vnic initialization
2463 err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2466 "Failed to set ingress vlan rewrite mode, aborting.\n");
2467 goto err_out_dev_close;
2470 /* Issue device init to initialize the vnic-to-switch link.
2471 * We'll start with carrier off and wait for link UP
2472 * notification later to turn on carrier. We don't need
2473 * to wait here for the vnic-to-switch link initialization
2474 * to complete; link UP notification is the indication that
2475 * the process is complete.
2478 netif_carrier_off(netdev);
2480 /* Do not call dev_init for a dynamic vnic.
2481 * For a dynamic vnic, init_prov_info will be
2482 * called later by an upper layer.
2485 if (!enic_is_dynamic(enic)) {
2486 err = vnic_dev_init(enic->vdev, 0);
2488 dev_err(dev, "vNIC dev init failed, aborting\n");
2489 goto err_out_dev_close;
2493 err = enic_dev_init(enic);
2495 dev_err(dev, "Device initialization failed, aborting\n");
2496 goto err_out_dev_close;
2499 netif_set_real_num_tx_queues(netdev, enic->wq_count);
2500 netif_set_real_num_rx_queues(netdev, enic->rq_count);
2502 /* Setup notification timer, HW reset task, and wq locks
2505 init_timer(&enic->notify_timer);
2506 enic->notify_timer.function = enic_notify_timer;
2507 enic->notify_timer.data = (unsigned long)enic;
2509 enic_set_rx_coal_setting(enic);
2510 INIT_WORK(&enic->reset, enic_reset);
2511 INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2513 for (i = 0; i < enic->wq_count; i++)
2514 spin_lock_init(&enic->wq_lock[i]);
2516 /* Register net device
2519 enic->port_mtu = enic->config.mtu;
2520 (void)enic_change_mtu(netdev, enic->port_mtu);
2522 err = enic_set_mac_addr(netdev, enic->mac_addr);
2524 dev_err(dev, "Invalid MAC address, aborting\n");
2525 goto err_out_dev_deinit;
2528 enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2529 /* rx coalesce time already got initialized. This gets used
2530 * if adaptive coal is turned off
2532 enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2534 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2535 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2537 netdev->netdev_ops = &enic_netdev_ops;
2539 netdev->watchdog_timeo = 2 * HZ;
2540 enic_set_ethtool_ops(netdev);
2542 netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2543 if (ENIC_SETTING(enic, LOOP)) {
2544 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2545 enic->loop_enable = 1;
2546 enic->loop_tag = enic->config.loop_tag;
2547 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2549 if (ENIC_SETTING(enic, TXCSUM))
2550 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2551 if (ENIC_SETTING(enic, TSO))
2552 netdev->hw_features |= NETIF_F_TSO |
2553 NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2554 if (ENIC_SETTING(enic, RSS))
2555 netdev->hw_features |= NETIF_F_RXHASH;
2556 if (ENIC_SETTING(enic, RXCSUM))
2557 netdev->hw_features |= NETIF_F_RXCSUM;
2559 netdev->features |= netdev->hw_features;
2561 #ifdef CONFIG_RFS_ACCEL
2562 netdev->hw_features |= NETIF_F_NTUPLE;
2566 netdev->features |= NETIF_F_HIGHDMA;
2568 netdev->priv_flags |= IFF_UNICAST_FLT;
2570 err = register_netdev(netdev);
2572 dev_err(dev, "Cannot register net device, aborting\n");
2573 goto err_out_dev_deinit;
2575 enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
2580 enic_dev_deinit(enic);
2582 vnic_dev_close(enic->vdev);
2583 err_out_disable_sriov:
2585 err_out_disable_sriov_pp:
2586 #ifdef CONFIG_PCI_IOV
2587 if (enic_sriov_enabled(enic)) {
2588 pci_disable_sriov(pdev);
2589 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2591 err_out_vnic_unregister:
2593 vnic_dev_unregister(enic->vdev);
2596 err_out_release_regions:
2597 pci_release_regions(pdev);
2598 err_out_disable_device:
2599 pci_disable_device(pdev);
2600 err_out_free_netdev:
2601 free_netdev(netdev);
2606 static void enic_remove(struct pci_dev *pdev)
2608 struct net_device *netdev = pci_get_drvdata(pdev);
2611 struct enic *enic = netdev_priv(netdev);
2613 cancel_work_sync(&enic->reset);
2614 cancel_work_sync(&enic->change_mtu_work);
2615 unregister_netdev(netdev);
2616 enic_dev_deinit(enic);
2617 vnic_dev_close(enic->vdev);
2618 #ifdef CONFIG_PCI_IOV
2619 if (enic_sriov_enabled(enic)) {
2620 pci_disable_sriov(pdev);
2621 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2625 vnic_dev_unregister(enic->vdev);
2627 pci_release_regions(pdev);
2628 pci_disable_device(pdev);
2629 free_netdev(netdev);
2633 static struct pci_driver enic_driver = {
2635 .id_table = enic_id_table,
2636 .probe = enic_probe,
2637 .remove = enic_remove,
2640 static int __init enic_init_module(void)
2642 pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2644 return pci_register_driver(&enic_driver);
2647 static void __exit enic_cleanup_module(void)
2649 pci_unregister_driver(&enic_driver);
2652 module_init(enic_init_module);
2653 module_exit(enic_cleanup_module);