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 if (napi_schedule_prep(&enic->napi[0]))
288 __napi_schedule(&enic->napi[0]);
290 vnic_intr_unmask(&enic->intr[io_intr]);
296 static irqreturn_t enic_isr_msi(int irq, void *data)
298 struct enic *enic = data;
300 /* With MSI, there is no sharing of interrupts, so this is
301 * our interrupt and there is no need to ack it. The device
302 * is not providing per-vector masking, so the OS will not
303 * write to PCI config space to mask/unmask the interrupt.
304 * We're using mask_on_assertion for MSI, so the device
305 * automatically masks the interrupt when the interrupt is
306 * generated. Later, when exiting polling, the interrupt
307 * will be unmasked (see enic_poll).
309 * Also, the device uses the same PCIe Traffic Class (TC)
310 * for Memory Write data and MSI, so there are no ordering
311 * issues; the MSI will always arrive at the Root Complex
312 * _after_ corresponding Memory Writes (i.e. descriptor
316 napi_schedule(&enic->napi[0]);
321 static irqreturn_t enic_isr_msix(int irq, void *data)
323 struct napi_struct *napi = data;
330 static irqreturn_t enic_isr_msix_err(int irq, void *data)
332 struct enic *enic = data;
333 unsigned int intr = enic_msix_err_intr(enic);
335 vnic_intr_return_all_credits(&enic->intr[intr]);
337 enic_log_q_error(enic);
339 /* schedule recovery from WQ/RQ error */
340 schedule_work(&enic->reset);
345 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
347 struct enic *enic = data;
348 unsigned int intr = enic_msix_notify_intr(enic);
350 vnic_intr_return_all_credits(&enic->intr[intr]);
351 enic_notify_check(enic);
356 static inline void enic_queue_wq_skb_cont(struct enic *enic,
357 struct vnic_wq *wq, struct sk_buff *skb,
358 unsigned int len_left, int loopback)
360 const skb_frag_t *frag;
362 /* Queue additional data fragments */
363 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
364 len_left -= skb_frag_size(frag);
365 enic_queue_wq_desc_cont(wq, skb,
366 skb_frag_dma_map(&enic->pdev->dev,
367 frag, 0, skb_frag_size(frag),
370 (len_left == 0), /* EOP? */
375 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
376 struct vnic_wq *wq, struct sk_buff *skb,
377 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
379 unsigned int head_len = skb_headlen(skb);
380 unsigned int len_left = skb->len - head_len;
381 int eop = (len_left == 0);
383 /* Queue the main skb fragment. The fragments are no larger
384 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
385 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
386 * per fragment is queued.
388 enic_queue_wq_desc(wq, skb,
389 pci_map_single(enic->pdev, skb->data,
390 head_len, PCI_DMA_TODEVICE),
392 vlan_tag_insert, vlan_tag,
396 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
399 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
400 struct vnic_wq *wq, struct sk_buff *skb,
401 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
403 unsigned int head_len = skb_headlen(skb);
404 unsigned int len_left = skb->len - head_len;
405 unsigned int hdr_len = skb_checksum_start_offset(skb);
406 unsigned int csum_offset = hdr_len + skb->csum_offset;
407 int eop = (len_left == 0);
409 /* Queue the main skb fragment. The fragments are no larger
410 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
411 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
412 * per fragment is queued.
414 enic_queue_wq_desc_csum_l4(wq, skb,
415 pci_map_single(enic->pdev, skb->data,
416 head_len, PCI_DMA_TODEVICE),
420 vlan_tag_insert, vlan_tag,
424 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
427 static inline void enic_queue_wq_skb_tso(struct enic *enic,
428 struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
429 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
431 unsigned int frag_len_left = skb_headlen(skb);
432 unsigned int len_left = skb->len - frag_len_left;
433 unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
434 int eop = (len_left == 0);
437 unsigned int offset = 0;
440 /* Preload TCP csum field with IP pseudo hdr calculated
441 * with IP length set to zero. HW will later add in length
442 * to each TCP segment resulting from the TSO.
445 if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
446 ip_hdr(skb)->check = 0;
447 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
448 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
449 } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
450 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
451 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
454 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
455 * for the main skb fragment
457 while (frag_len_left) {
458 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
459 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
460 len, PCI_DMA_TODEVICE);
461 enic_queue_wq_desc_tso(wq, skb,
465 vlan_tag_insert, vlan_tag,
466 eop && (len == frag_len_left), loopback);
467 frag_len_left -= len;
474 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
475 * for additional data fragments
477 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
478 len_left -= skb_frag_size(frag);
479 frag_len_left = skb_frag_size(frag);
482 while (frag_len_left) {
483 len = min(frag_len_left,
484 (unsigned int)WQ_ENET_MAX_DESC_LEN);
485 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
488 enic_queue_wq_desc_cont(wq, skb,
492 (len == frag_len_left), /* EOP? */
494 frag_len_left -= len;
500 static inline void enic_queue_wq_skb(struct enic *enic,
501 struct vnic_wq *wq, struct sk_buff *skb)
503 unsigned int mss = skb_shinfo(skb)->gso_size;
504 unsigned int vlan_tag = 0;
505 int vlan_tag_insert = 0;
508 if (vlan_tx_tag_present(skb)) {
509 /* VLAN tag from trunking driver */
511 vlan_tag = vlan_tx_tag_get(skb);
512 } else if (enic->loop_enable) {
513 vlan_tag = enic->loop_tag;
518 enic_queue_wq_skb_tso(enic, wq, skb, mss,
519 vlan_tag_insert, vlan_tag, loopback);
520 else if (skb->ip_summed == CHECKSUM_PARTIAL)
521 enic_queue_wq_skb_csum_l4(enic, wq, skb,
522 vlan_tag_insert, vlan_tag, loopback);
524 enic_queue_wq_skb_vlan(enic, wq, skb,
525 vlan_tag_insert, vlan_tag, loopback);
528 /* netif_tx_lock held, process context with BHs disabled, or BH */
529 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
530 struct net_device *netdev)
532 struct enic *enic = netdev_priv(netdev);
535 unsigned int txq_map;
538 dev_kfree_skb_any(skb);
542 txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
543 wq = &enic->wq[txq_map];
545 /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
546 * which is very likely. In the off chance it's going to take
547 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
550 if (skb_shinfo(skb)->gso_size == 0 &&
551 skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
552 skb_linearize(skb)) {
553 dev_kfree_skb_any(skb);
557 spin_lock_irqsave(&enic->wq_lock[txq_map], flags);
559 if (vnic_wq_desc_avail(wq) <
560 skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
561 netif_tx_stop_queue(netdev_get_tx_queue(netdev, txq_map));
562 /* This is a hard error, log it */
563 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
564 spin_unlock_irqrestore(&enic->wq_lock[txq_map], flags);
565 return NETDEV_TX_BUSY;
568 enic_queue_wq_skb(enic, wq, skb);
570 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
571 netif_tx_stop_queue(netdev_get_tx_queue(netdev, txq_map));
573 spin_unlock_irqrestore(&enic->wq_lock[txq_map], flags);
578 /* dev_base_lock rwlock held, nominally process context */
579 static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
580 struct rtnl_link_stats64 *net_stats)
582 struct enic *enic = netdev_priv(netdev);
583 struct vnic_stats *stats;
585 enic_dev_stats_dump(enic, &stats);
587 net_stats->tx_packets = stats->tx.tx_frames_ok;
588 net_stats->tx_bytes = stats->tx.tx_bytes_ok;
589 net_stats->tx_errors = stats->tx.tx_errors;
590 net_stats->tx_dropped = stats->tx.tx_drops;
592 net_stats->rx_packets = stats->rx.rx_frames_ok;
593 net_stats->rx_bytes = stats->rx.rx_bytes_ok;
594 net_stats->rx_errors = stats->rx.rx_errors;
595 net_stats->multicast = stats->rx.rx_multicast_frames_ok;
596 net_stats->rx_over_errors = enic->rq_truncated_pkts;
597 net_stats->rx_crc_errors = enic->rq_bad_fcs;
598 net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
603 static int enic_mc_sync(struct net_device *netdev, const u8 *mc_addr)
605 struct enic *enic = netdev_priv(netdev);
607 if (enic->mc_count == ENIC_MULTICAST_PERFECT_FILTERS) {
608 unsigned int mc_count = netdev_mc_count(netdev);
610 netdev_warn(netdev, "Registering only %d out of %d multicast addresses\n",
611 ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
616 enic_dev_add_addr(enic, mc_addr);
622 static int enic_mc_unsync(struct net_device *netdev, const u8 *mc_addr)
624 struct enic *enic = netdev_priv(netdev);
626 enic_dev_del_addr(enic, mc_addr);
632 static int enic_uc_sync(struct net_device *netdev, const u8 *uc_addr)
634 struct enic *enic = netdev_priv(netdev);
636 if (enic->uc_count == ENIC_UNICAST_PERFECT_FILTERS) {
637 unsigned int uc_count = netdev_uc_count(netdev);
639 netdev_warn(netdev, "Registering only %d out of %d unicast addresses\n",
640 ENIC_UNICAST_PERFECT_FILTERS, uc_count);
645 enic_dev_add_addr(enic, uc_addr);
651 static int enic_uc_unsync(struct net_device *netdev, const u8 *uc_addr)
653 struct enic *enic = netdev_priv(netdev);
655 enic_dev_del_addr(enic, uc_addr);
661 void enic_reset_addr_lists(struct enic *enic)
663 struct net_device *netdev = enic->netdev;
665 __dev_uc_unsync(netdev, NULL);
666 __dev_mc_unsync(netdev, NULL);
673 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
675 struct enic *enic = netdev_priv(netdev);
677 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
678 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
679 return -EADDRNOTAVAIL;
681 if (!is_valid_ether_addr(addr))
682 return -EADDRNOTAVAIL;
685 memcpy(netdev->dev_addr, addr, netdev->addr_len);
690 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
692 struct enic *enic = netdev_priv(netdev);
693 struct sockaddr *saddr = p;
694 char *addr = saddr->sa_data;
697 if (netif_running(enic->netdev)) {
698 err = enic_dev_del_station_addr(enic);
703 err = enic_set_mac_addr(netdev, addr);
707 if (netif_running(enic->netdev)) {
708 err = enic_dev_add_station_addr(enic);
716 static int enic_set_mac_address(struct net_device *netdev, void *p)
718 struct sockaddr *saddr = p;
719 char *addr = saddr->sa_data;
720 struct enic *enic = netdev_priv(netdev);
723 err = enic_dev_del_station_addr(enic);
727 err = enic_set_mac_addr(netdev, addr);
731 return enic_dev_add_station_addr(enic);
734 /* netif_tx_lock held, BHs disabled */
735 static void enic_set_rx_mode(struct net_device *netdev)
737 struct enic *enic = netdev_priv(netdev);
739 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
740 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
741 int promisc = (netdev->flags & IFF_PROMISC) ||
742 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
743 int allmulti = (netdev->flags & IFF_ALLMULTI) ||
744 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
745 unsigned int flags = netdev->flags |
746 (allmulti ? IFF_ALLMULTI : 0) |
747 (promisc ? IFF_PROMISC : 0);
749 if (enic->flags != flags) {
751 enic_dev_packet_filter(enic, directed,
752 multicast, broadcast, promisc, allmulti);
756 __dev_uc_sync(netdev, enic_uc_sync, enic_uc_unsync);
758 __dev_mc_sync(netdev, enic_mc_sync, enic_mc_unsync);
762 /* netif_tx_lock held, BHs disabled */
763 static void enic_tx_timeout(struct net_device *netdev)
765 struct enic *enic = netdev_priv(netdev);
766 schedule_work(&enic->reset);
769 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
771 struct enic *enic = netdev_priv(netdev);
772 struct enic_port_profile *pp;
775 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
779 if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
780 if (vf == PORT_SELF_VF) {
781 memcpy(pp->vf_mac, mac, ETH_ALEN);
785 * For sriov vf's set the mac in hw
787 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
788 vnic_dev_set_mac_addr, mac);
789 return enic_dev_status_to_errno(err);
795 static int enic_set_vf_port(struct net_device *netdev, int vf,
796 struct nlattr *port[])
798 struct enic *enic = netdev_priv(netdev);
799 struct enic_port_profile prev_pp;
800 struct enic_port_profile *pp;
801 int err = 0, restore_pp = 1;
803 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
807 if (!port[IFLA_PORT_REQUEST])
810 memcpy(&prev_pp, pp, sizeof(*enic->pp));
811 memset(pp, 0, sizeof(*enic->pp));
813 pp->set |= ENIC_SET_REQUEST;
814 pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
816 if (port[IFLA_PORT_PROFILE]) {
817 pp->set |= ENIC_SET_NAME;
818 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
822 if (port[IFLA_PORT_INSTANCE_UUID]) {
823 pp->set |= ENIC_SET_INSTANCE;
824 memcpy(pp->instance_uuid,
825 nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
828 if (port[IFLA_PORT_HOST_UUID]) {
829 pp->set |= ENIC_SET_HOST;
830 memcpy(pp->host_uuid,
831 nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
834 if (vf == PORT_SELF_VF) {
835 /* Special case handling: mac came from IFLA_VF_MAC */
836 if (!is_zero_ether_addr(prev_pp.vf_mac))
837 memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
839 if (is_zero_ether_addr(netdev->dev_addr))
840 eth_hw_addr_random(netdev);
842 /* SR-IOV VF: get mac from adapter */
843 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
844 vnic_dev_get_mac_addr, pp->mac_addr);
846 netdev_err(netdev, "Error getting mac for vf %d\n", vf);
847 memcpy(pp, &prev_pp, sizeof(*pp));
848 return enic_dev_status_to_errno(err);
852 err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
855 /* Things are still the way they were: Implicit
856 * DISASSOCIATE failed
858 memcpy(pp, &prev_pp, sizeof(*pp));
860 memset(pp, 0, sizeof(*pp));
861 if (vf == PORT_SELF_VF)
862 memset(netdev->dev_addr, 0, ETH_ALEN);
865 /* Set flag to indicate that the port assoc/disassoc
866 * request has been sent out to fw
868 pp->set |= ENIC_PORT_REQUEST_APPLIED;
870 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
871 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
872 memset(pp->mac_addr, 0, ETH_ALEN);
873 if (vf == PORT_SELF_VF)
874 memset(netdev->dev_addr, 0, ETH_ALEN);
878 if (vf == PORT_SELF_VF)
879 memset(pp->vf_mac, 0, ETH_ALEN);
884 static int enic_get_vf_port(struct net_device *netdev, int vf,
887 struct enic *enic = netdev_priv(netdev);
888 u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
889 struct enic_port_profile *pp;
892 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
896 if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
899 err = enic_process_get_pp_request(enic, vf, pp->request, &response);
903 if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
904 nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
905 ((pp->set & ENIC_SET_NAME) &&
906 nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
907 ((pp->set & ENIC_SET_INSTANCE) &&
908 nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
909 pp->instance_uuid)) ||
910 ((pp->set & ENIC_SET_HOST) &&
911 nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
912 goto nla_put_failure;
919 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
921 struct enic *enic = vnic_dev_priv(rq->vdev);
926 pci_unmap_single(enic->pdev, buf->dma_addr,
927 buf->len, PCI_DMA_FROMDEVICE);
928 dev_kfree_skb_any(buf->os_buf);
932 static int enic_rq_alloc_buf(struct vnic_rq *rq)
934 struct enic *enic = vnic_dev_priv(rq->vdev);
935 struct net_device *netdev = enic->netdev;
937 unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
938 unsigned int os_buf_index = 0;
940 struct vnic_rq_buf *buf = rq->to_use;
943 enic_queue_rq_desc(rq, buf->os_buf, os_buf_index, buf->dma_addr,
948 skb = netdev_alloc_skb_ip_align(netdev, len);
952 dma_addr = pci_map_single(enic->pdev, skb->data,
953 len, PCI_DMA_FROMDEVICE);
955 enic_queue_rq_desc(rq, skb, os_buf_index,
961 static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
964 if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
965 pkt_size->large_pkt_bytes_cnt += pkt_len;
967 pkt_size->small_pkt_bytes_cnt += pkt_len;
970 static bool enic_rxcopybreak(struct net_device *netdev, struct sk_buff **skb,
971 struct vnic_rq_buf *buf, u16 len)
973 struct enic *enic = netdev_priv(netdev);
974 struct sk_buff *new_skb;
976 if (len > enic->rx_copybreak)
978 new_skb = netdev_alloc_skb_ip_align(netdev, len);
981 pci_dma_sync_single_for_cpu(enic->pdev, buf->dma_addr, len,
983 memcpy(new_skb->data, (*skb)->data, len);
989 static void enic_rq_indicate_buf(struct vnic_rq *rq,
990 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
991 int skipped, void *opaque)
993 struct enic *enic = vnic_dev_priv(rq->vdev);
994 struct net_device *netdev = enic->netdev;
996 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
998 u8 type, color, eop, sop, ingress_port, vlan_stripped;
999 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1000 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1001 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1003 u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1011 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1012 &type, &color, &q_number, &completed_index,
1013 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1014 &csum_not_calc, &rss_hash, &bytes_written,
1015 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1016 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1017 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1018 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1024 if (bytes_written > 0)
1026 else if (bytes_written == 0)
1027 enic->rq_truncated_pkts++;
1030 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1031 PCI_DMA_FROMDEVICE);
1032 dev_kfree_skb_any(skb);
1038 if (eop && bytes_written > 0) {
1043 if (!enic_rxcopybreak(netdev, &skb, buf, bytes_written)) {
1045 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1046 PCI_DMA_FROMDEVICE);
1048 prefetch(skb->data - NET_IP_ALIGN);
1050 skb_put(skb, bytes_written);
1051 skb->protocol = eth_type_trans(skb, netdev);
1052 skb_record_rx_queue(skb, q_number);
1053 if (netdev->features & NETIF_F_RXHASH) {
1054 skb_set_hash(skb, rss_hash,
1056 (NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX |
1057 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6 |
1058 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4)) ?
1059 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1062 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc) {
1063 skb->csum = htons(checksum);
1064 skb->ip_summed = CHECKSUM_COMPLETE;
1068 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1070 skb_mark_napi_id(skb, &enic->napi[rq->index]);
1071 if (enic_poll_busy_polling(rq) ||
1072 !(netdev->features & NETIF_F_GRO))
1073 netif_receive_skb(skb);
1075 napi_gro_receive(&enic->napi[q_number], skb);
1076 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1077 enic_intr_update_pkt_size(&cq->pkt_size_counter,
1084 pci_unmap_single(enic->pdev, buf->dma_addr, buf->len,
1085 PCI_DMA_FROMDEVICE);
1086 dev_kfree_skb_any(skb);
1091 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1092 u8 type, u16 q_number, u16 completed_index, void *opaque)
1094 struct enic *enic = vnic_dev_priv(vdev);
1096 vnic_rq_service(&enic->rq[q_number], cq_desc,
1097 completed_index, VNIC_RQ_RETURN_DESC,
1098 enic_rq_indicate_buf, opaque);
1103 static int enic_poll(struct napi_struct *napi, int budget)
1105 struct net_device *netdev = napi->dev;
1106 struct enic *enic = netdev_priv(netdev);
1107 unsigned int cq_rq = enic_cq_rq(enic, 0);
1108 unsigned int cq_wq = enic_cq_wq(enic, 0);
1109 unsigned int intr = enic_legacy_io_intr();
1110 unsigned int rq_work_to_do = budget;
1111 unsigned int wq_work_to_do = -1; /* no limit */
1112 unsigned int work_done, rq_work_done = 0, wq_work_done;
1115 wq_work_done = vnic_cq_service(&enic->cq[cq_wq], wq_work_to_do,
1116 enic_wq_service, NULL);
1118 if (!enic_poll_lock_napi(&enic->rq[cq_rq])) {
1119 if (wq_work_done > 0)
1120 vnic_intr_return_credits(&enic->intr[intr],
1122 0 /* dont unmask intr */,
1123 0 /* dont reset intr timer */);
1124 return rq_work_done;
1128 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1129 rq_work_to_do, enic_rq_service, NULL);
1131 /* Accumulate intr event credits for this polling
1132 * cycle. An intr event is the completion of a
1133 * a WQ or RQ packet.
1136 work_done = rq_work_done + wq_work_done;
1139 vnic_intr_return_credits(&enic->intr[intr],
1141 0 /* don't unmask intr */,
1142 0 /* don't reset intr timer */);
1144 err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1146 /* Buffer allocation failed. Stay in polling
1147 * mode so we can try to fill the ring again.
1151 rq_work_done = rq_work_to_do;
1153 if (rq_work_done < rq_work_to_do) {
1155 /* Some work done, but not enough to stay in polling,
1159 napi_complete(napi);
1160 vnic_intr_unmask(&enic->intr[intr]);
1162 enic_poll_unlock_napi(&enic->rq[cq_rq]);
1164 return rq_work_done;
1167 static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
1169 unsigned int intr = enic_msix_rq_intr(enic, rq->index);
1170 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1171 u32 timer = cq->tobe_rx_coal_timeval;
1173 if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
1174 vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
1175 cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
1179 static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
1181 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1182 struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
1183 struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
1189 ktime_t now = ktime_get();
1191 delta = ktime_us_delta(now, cq->prev_ts);
1192 if (delta < ENIC_AIC_TS_BREAK)
1196 traffic = pkt_size_counter->large_pkt_bytes_cnt +
1197 pkt_size_counter->small_pkt_bytes_cnt;
1198 /* The table takes Mbps
1199 * traffic *= 8 => bits
1200 * traffic *= (10^6 / delta) => bps
1201 * traffic /= 10^6 => Mbps
1203 * Combining, traffic *= (8 / delta)
1207 traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
1209 for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
1210 if (traffic < mod_table[index].rx_rate)
1212 range_start = (pkt_size_counter->small_pkt_bytes_cnt >
1213 pkt_size_counter->large_pkt_bytes_cnt << 1) ?
1214 rx_coal->small_pkt_range_start :
1215 rx_coal->large_pkt_range_start;
1216 timer = range_start + ((rx_coal->range_end - range_start) *
1217 mod_table[index].range_percent / 100);
1219 cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
1221 pkt_size_counter->large_pkt_bytes_cnt = 0;
1222 pkt_size_counter->small_pkt_bytes_cnt = 0;
1225 #ifdef CONFIG_RFS_ACCEL
1226 static void enic_free_rx_cpu_rmap(struct enic *enic)
1228 free_irq_cpu_rmap(enic->netdev->rx_cpu_rmap);
1229 enic->netdev->rx_cpu_rmap = NULL;
1232 static void enic_set_rx_cpu_rmap(struct enic *enic)
1236 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX) {
1237 enic->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(enic->rq_count);
1238 if (unlikely(!enic->netdev->rx_cpu_rmap))
1240 for (i = 0; i < enic->rq_count; i++) {
1241 res = irq_cpu_rmap_add(enic->netdev->rx_cpu_rmap,
1242 enic->msix_entry[i].vector);
1243 if (unlikely(res)) {
1244 enic_free_rx_cpu_rmap(enic);
1253 static void enic_free_rx_cpu_rmap(struct enic *enic)
1257 static void enic_set_rx_cpu_rmap(struct enic *enic)
1261 #endif /* CONFIG_RFS_ACCEL */
1263 #ifdef CONFIG_NET_RX_BUSY_POLL
1264 int enic_busy_poll(struct napi_struct *napi)
1266 struct net_device *netdev = napi->dev;
1267 struct enic *enic = netdev_priv(netdev);
1268 unsigned int rq = (napi - &enic->napi[0]);
1269 unsigned int cq = enic_cq_rq(enic, rq);
1270 unsigned int intr = enic_msix_rq_intr(enic, rq);
1271 unsigned int work_to_do = -1; /* clean all pkts possible */
1272 unsigned int work_done;
1274 if (!enic_poll_lock_poll(&enic->rq[rq]))
1275 return LL_FLUSH_BUSY;
1276 work_done = vnic_cq_service(&enic->cq[cq], work_to_do,
1277 enic_rq_service, NULL);
1280 vnic_intr_return_credits(&enic->intr[intr],
1282 vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1283 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1284 enic_calc_int_moderation(enic, &enic->rq[rq]);
1285 enic_poll_unlock_poll(&enic->rq[rq]);
1289 #endif /* CONFIG_NET_RX_BUSY_POLL */
1291 static int enic_poll_msix_wq(struct napi_struct *napi, int budget)
1293 struct net_device *netdev = napi->dev;
1294 struct enic *enic = netdev_priv(netdev);
1295 unsigned int wq_index = (napi - &enic->napi[0]) - enic->rq_count;
1296 struct vnic_wq *wq = &enic->wq[wq_index];
1299 unsigned int wq_work_to_do = -1; /* clean all desc possible */
1300 unsigned int wq_work_done;
1301 unsigned int wq_irq;
1304 cq = enic_cq_wq(enic, wq_irq);
1305 intr = enic_msix_wq_intr(enic, wq_irq);
1306 wq_work_done = vnic_cq_service(&enic->cq[cq], wq_work_to_do,
1307 enic_wq_service, NULL);
1309 vnic_intr_return_credits(&enic->intr[intr], wq_work_done,
1310 0 /* don't unmask intr */,
1311 1 /* reset intr timer */);
1312 if (!wq_work_done) {
1313 napi_complete(napi);
1314 vnic_intr_unmask(&enic->intr[intr]);
1321 static int enic_poll_msix_rq(struct napi_struct *napi, int budget)
1323 struct net_device *netdev = napi->dev;
1324 struct enic *enic = netdev_priv(netdev);
1325 unsigned int rq = (napi - &enic->napi[0]);
1326 unsigned int cq = enic_cq_rq(enic, rq);
1327 unsigned int intr = enic_msix_rq_intr(enic, rq);
1328 unsigned int work_to_do = budget;
1329 unsigned int work_done = 0;
1332 if (!enic_poll_lock_napi(&enic->rq[rq]))
1338 work_done = vnic_cq_service(&enic->cq[cq],
1339 work_to_do, enic_rq_service, NULL);
1341 /* Return intr event credits for this polling
1342 * cycle. An intr event is the completion of a
1347 vnic_intr_return_credits(&enic->intr[intr],
1349 0 /* don't unmask intr */,
1350 0 /* don't reset intr timer */);
1352 err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1354 /* Buffer allocation failed. Stay in polling mode
1355 * so we can try to fill the ring again.
1359 work_done = work_to_do;
1360 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1361 /* Call the function which refreshes
1362 * the intr coalescing timer value based on
1363 * the traffic. This is supported only in
1364 * the case of MSI-x mode
1366 enic_calc_int_moderation(enic, &enic->rq[rq]);
1368 if (work_done < work_to_do) {
1370 /* Some work done, but not enough to stay in polling,
1374 napi_complete(napi);
1375 if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
1376 enic_set_int_moderation(enic, &enic->rq[rq]);
1377 vnic_intr_unmask(&enic->intr[intr]);
1379 enic_poll_unlock_napi(&enic->rq[rq]);
1384 static void enic_notify_timer(unsigned long data)
1386 struct enic *enic = (struct enic *)data;
1388 enic_notify_check(enic);
1390 mod_timer(&enic->notify_timer,
1391 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1394 static void enic_free_intr(struct enic *enic)
1396 struct net_device *netdev = enic->netdev;
1399 enic_free_rx_cpu_rmap(enic);
1400 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1401 case VNIC_DEV_INTR_MODE_INTX:
1402 free_irq(enic->pdev->irq, netdev);
1404 case VNIC_DEV_INTR_MODE_MSI:
1405 free_irq(enic->pdev->irq, enic);
1407 case VNIC_DEV_INTR_MODE_MSIX:
1408 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1409 if (enic->msix[i].requested)
1410 free_irq(enic->msix_entry[i].vector,
1411 enic->msix[i].devid);
1418 static int enic_request_intr(struct enic *enic)
1420 struct net_device *netdev = enic->netdev;
1421 unsigned int i, intr;
1424 enic_set_rx_cpu_rmap(enic);
1425 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1427 case VNIC_DEV_INTR_MODE_INTX:
1429 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1430 IRQF_SHARED, netdev->name, netdev);
1433 case VNIC_DEV_INTR_MODE_MSI:
1435 err = request_irq(enic->pdev->irq, enic_isr_msi,
1436 0, netdev->name, enic);
1439 case VNIC_DEV_INTR_MODE_MSIX:
1441 for (i = 0; i < enic->rq_count; i++) {
1442 intr = enic_msix_rq_intr(enic, i);
1443 snprintf(enic->msix[intr].devname,
1444 sizeof(enic->msix[intr].devname),
1445 "%.11s-rx-%d", netdev->name, i);
1446 enic->msix[intr].isr = enic_isr_msix;
1447 enic->msix[intr].devid = &enic->napi[i];
1450 for (i = 0; i < enic->wq_count; i++) {
1451 int wq = enic_cq_wq(enic, i);
1453 intr = enic_msix_wq_intr(enic, i);
1454 snprintf(enic->msix[intr].devname,
1455 sizeof(enic->msix[intr].devname),
1456 "%.11s-tx-%d", netdev->name, i);
1457 enic->msix[intr].isr = enic_isr_msix;
1458 enic->msix[intr].devid = &enic->napi[wq];
1461 intr = enic_msix_err_intr(enic);
1462 snprintf(enic->msix[intr].devname,
1463 sizeof(enic->msix[intr].devname),
1464 "%.11s-err", netdev->name);
1465 enic->msix[intr].isr = enic_isr_msix_err;
1466 enic->msix[intr].devid = enic;
1468 intr = enic_msix_notify_intr(enic);
1469 snprintf(enic->msix[intr].devname,
1470 sizeof(enic->msix[intr].devname),
1471 "%.11s-notify", netdev->name);
1472 enic->msix[intr].isr = enic_isr_msix_notify;
1473 enic->msix[intr].devid = enic;
1475 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1476 enic->msix[i].requested = 0;
1478 for (i = 0; i < enic->intr_count; i++) {
1479 err = request_irq(enic->msix_entry[i].vector,
1480 enic->msix[i].isr, 0,
1481 enic->msix[i].devname,
1482 enic->msix[i].devid);
1484 enic_free_intr(enic);
1487 enic->msix[i].requested = 1;
1499 static void enic_synchronize_irqs(struct enic *enic)
1503 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1504 case VNIC_DEV_INTR_MODE_INTX:
1505 case VNIC_DEV_INTR_MODE_MSI:
1506 synchronize_irq(enic->pdev->irq);
1508 case VNIC_DEV_INTR_MODE_MSIX:
1509 for (i = 0; i < enic->intr_count; i++)
1510 synchronize_irq(enic->msix_entry[i].vector);
1517 static void enic_set_rx_coal_setting(struct enic *enic)
1521 struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
1523 /* If intr mode is not MSIX, do not do adaptive coalescing */
1524 if (VNIC_DEV_INTR_MODE_MSIX != vnic_dev_get_intr_mode(enic->vdev)) {
1525 netdev_info(enic->netdev, "INTR mode is not MSIX, Not initializing adaptive coalescing");
1529 /* 1. Read the link speed from fw
1530 * 2. Pick the default range for the speed
1531 * 3. Update it in enic->rx_coalesce_setting
1533 speed = vnic_dev_port_speed(enic->vdev);
1534 if (ENIC_LINK_SPEED_10G < speed)
1535 index = ENIC_LINK_40G_INDEX;
1536 else if (ENIC_LINK_SPEED_4G < speed)
1537 index = ENIC_LINK_10G_INDEX;
1539 index = ENIC_LINK_4G_INDEX;
1541 rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
1542 rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
1543 rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
1545 /* Start with the value provided by UCSM */
1546 for (index = 0; index < enic->rq_count; index++)
1547 enic->cq[index].cur_rx_coal_timeval =
1548 enic->config.intr_timer_usec;
1550 rx_coal->use_adaptive_rx_coalesce = 1;
1553 static int enic_dev_notify_set(struct enic *enic)
1557 spin_lock_bh(&enic->devcmd_lock);
1558 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1559 case VNIC_DEV_INTR_MODE_INTX:
1560 err = vnic_dev_notify_set(enic->vdev,
1561 enic_legacy_notify_intr());
1563 case VNIC_DEV_INTR_MODE_MSIX:
1564 err = vnic_dev_notify_set(enic->vdev,
1565 enic_msix_notify_intr(enic));
1568 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1571 spin_unlock_bh(&enic->devcmd_lock);
1576 static void enic_notify_timer_start(struct enic *enic)
1578 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1579 case VNIC_DEV_INTR_MODE_MSI:
1580 mod_timer(&enic->notify_timer, jiffies);
1583 /* Using intr for notification for INTx/MSI-X */
1588 /* rtnl lock is held, process context */
1589 static int enic_open(struct net_device *netdev)
1591 struct enic *enic = netdev_priv(netdev);
1595 err = enic_request_intr(enic);
1597 netdev_err(netdev, "Unable to request irq.\n");
1601 err = enic_dev_notify_set(enic);
1604 "Failed to alloc notify buffer, aborting.\n");
1605 goto err_out_free_intr;
1608 for (i = 0; i < enic->rq_count; i++) {
1609 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1610 /* Need at least one buffer on ring to get going */
1611 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1612 netdev_err(netdev, "Unable to alloc receive buffers\n");
1614 goto err_out_notify_unset;
1618 for (i = 0; i < enic->wq_count; i++)
1619 vnic_wq_enable(&enic->wq[i]);
1620 for (i = 0; i < enic->rq_count; i++)
1621 vnic_rq_enable(&enic->rq[i]);
1623 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1624 enic_dev_add_station_addr(enic);
1626 enic_set_rx_mode(netdev);
1628 netif_tx_wake_all_queues(netdev);
1630 for (i = 0; i < enic->rq_count; i++) {
1631 enic_busy_poll_init_lock(&enic->rq[i]);
1632 napi_enable(&enic->napi[i]);
1634 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1635 for (i = 0; i < enic->wq_count; i++)
1636 napi_enable(&enic->napi[enic_cq_wq(enic, i)]);
1637 enic_dev_enable(enic);
1639 for (i = 0; i < enic->intr_count; i++)
1640 vnic_intr_unmask(&enic->intr[i]);
1642 enic_notify_timer_start(enic);
1643 enic_rfs_flw_tbl_init(enic);
1647 err_out_notify_unset:
1648 enic_dev_notify_unset(enic);
1650 enic_free_intr(enic);
1655 /* rtnl lock is held, process context */
1656 static int enic_stop(struct net_device *netdev)
1658 struct enic *enic = netdev_priv(netdev);
1662 for (i = 0; i < enic->intr_count; i++) {
1663 vnic_intr_mask(&enic->intr[i]);
1664 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1667 enic_synchronize_irqs(enic);
1669 del_timer_sync(&enic->notify_timer);
1670 enic_rfs_flw_tbl_free(enic);
1672 enic_dev_disable(enic);
1674 for (i = 0; i < enic->rq_count; i++) {
1675 napi_disable(&enic->napi[i]);
1677 while (!enic_poll_lock_napi(&enic->rq[i]))
1682 netif_carrier_off(netdev);
1683 netif_tx_disable(netdev);
1684 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
1685 for (i = 0; i < enic->wq_count; i++)
1686 napi_disable(&enic->napi[enic_cq_wq(enic, i)]);
1688 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1689 enic_dev_del_station_addr(enic);
1691 for (i = 0; i < enic->wq_count; i++) {
1692 err = vnic_wq_disable(&enic->wq[i]);
1696 for (i = 0; i < enic->rq_count; i++) {
1697 err = vnic_rq_disable(&enic->rq[i]);
1702 enic_dev_notify_unset(enic);
1703 enic_free_intr(enic);
1705 for (i = 0; i < enic->wq_count; i++)
1706 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1707 for (i = 0; i < enic->rq_count; i++)
1708 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1709 for (i = 0; i < enic->cq_count; i++)
1710 vnic_cq_clean(&enic->cq[i]);
1711 for (i = 0; i < enic->intr_count; i++)
1712 vnic_intr_clean(&enic->intr[i]);
1717 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1719 struct enic *enic = netdev_priv(netdev);
1720 int running = netif_running(netdev);
1722 if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1725 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
1731 netdev->mtu = new_mtu;
1733 if (netdev->mtu > enic->port_mtu)
1735 "interface MTU (%d) set higher than port MTU (%d)\n",
1736 netdev->mtu, enic->port_mtu);
1744 static void enic_change_mtu_work(struct work_struct *work)
1746 struct enic *enic = container_of(work, struct enic, change_mtu_work);
1747 struct net_device *netdev = enic->netdev;
1748 int new_mtu = vnic_dev_mtu(enic->vdev);
1752 new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
1757 del_timer_sync(&enic->notify_timer);
1759 for (i = 0; i < enic->rq_count; i++)
1760 napi_disable(&enic->napi[i]);
1762 vnic_intr_mask(&enic->intr[0]);
1763 enic_synchronize_irqs(enic);
1764 err = vnic_rq_disable(&enic->rq[0]);
1767 netdev_err(netdev, "Unable to disable RQ.\n");
1770 vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
1771 vnic_cq_clean(&enic->cq[0]);
1772 vnic_intr_clean(&enic->intr[0]);
1774 /* Fill RQ with new_mtu-sized buffers */
1775 netdev->mtu = new_mtu;
1776 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1777 /* Need at least one buffer on ring to get going */
1778 if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
1780 netdev_err(netdev, "Unable to alloc receive buffers.\n");
1785 vnic_rq_enable(&enic->rq[0]);
1786 napi_enable(&enic->napi[0]);
1787 vnic_intr_unmask(&enic->intr[0]);
1788 enic_notify_timer_start(enic);
1792 netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1795 #ifdef CONFIG_NET_POLL_CONTROLLER
1796 static void enic_poll_controller(struct net_device *netdev)
1798 struct enic *enic = netdev_priv(netdev);
1799 struct vnic_dev *vdev = enic->vdev;
1800 unsigned int i, intr;
1802 switch (vnic_dev_get_intr_mode(vdev)) {
1803 case VNIC_DEV_INTR_MODE_MSIX:
1804 for (i = 0; i < enic->rq_count; i++) {
1805 intr = enic_msix_rq_intr(enic, i);
1806 enic_isr_msix(enic->msix_entry[intr].vector,
1810 for (i = 0; i < enic->wq_count; i++) {
1811 intr = enic_msix_wq_intr(enic, i);
1812 enic_isr_msix(enic->msix_entry[intr].vector,
1813 &enic->napi[enic_cq_wq(enic, i)]);
1817 case VNIC_DEV_INTR_MODE_MSI:
1818 enic_isr_msi(enic->pdev->irq, enic);
1820 case VNIC_DEV_INTR_MODE_INTX:
1821 enic_isr_legacy(enic->pdev->irq, netdev);
1829 static int enic_dev_wait(struct vnic_dev *vdev,
1830 int (*start)(struct vnic_dev *, int),
1831 int (*finished)(struct vnic_dev *, int *),
1838 BUG_ON(in_interrupt());
1840 err = start(vdev, arg);
1844 /* Wait for func to complete...2 seconds max
1847 time = jiffies + (HZ * 2);
1850 err = finished(vdev, &done);
1857 schedule_timeout_uninterruptible(HZ / 10);
1859 } while (time_after(time, jiffies));
1864 static int enic_dev_open(struct enic *enic)
1868 err = enic_dev_wait(enic->vdev, vnic_dev_open,
1869 vnic_dev_open_done, 0);
1871 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1877 static int enic_dev_hang_reset(struct enic *enic)
1881 err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1882 vnic_dev_hang_reset_done, 0);
1884 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1890 static int enic_set_rsskey(struct enic *enic)
1892 dma_addr_t rss_key_buf_pa;
1893 union vnic_rss_key *rss_key_buf_va = NULL;
1894 union vnic_rss_key rss_key = {
1895 .key[0].b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101},
1896 .key[1].b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101},
1897 .key[2].b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115},
1898 .key[3].b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108},
1902 rss_key_buf_va = pci_alloc_consistent(enic->pdev,
1903 sizeof(union vnic_rss_key), &rss_key_buf_pa);
1904 if (!rss_key_buf_va)
1907 memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
1909 spin_lock_bh(&enic->devcmd_lock);
1910 err = enic_set_rss_key(enic,
1912 sizeof(union vnic_rss_key));
1913 spin_unlock_bh(&enic->devcmd_lock);
1915 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
1916 rss_key_buf_va, rss_key_buf_pa);
1921 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
1923 dma_addr_t rss_cpu_buf_pa;
1924 union vnic_rss_cpu *rss_cpu_buf_va = NULL;
1928 rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
1929 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
1930 if (!rss_cpu_buf_va)
1933 for (i = 0; i < (1 << rss_hash_bits); i++)
1934 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
1936 spin_lock_bh(&enic->devcmd_lock);
1937 err = enic_set_rss_cpu(enic,
1939 sizeof(union vnic_rss_cpu));
1940 spin_unlock_bh(&enic->devcmd_lock);
1942 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
1943 rss_cpu_buf_va, rss_cpu_buf_pa);
1948 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
1949 u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
1951 const u8 tso_ipid_split_en = 0;
1952 const u8 ig_vlan_strip_en = 1;
1955 /* Enable VLAN tag stripping.
1958 spin_lock_bh(&enic->devcmd_lock);
1959 err = enic_set_nic_cfg(enic,
1960 rss_default_cpu, rss_hash_type,
1961 rss_hash_bits, rss_base_cpu,
1962 rss_enable, tso_ipid_split_en,
1964 spin_unlock_bh(&enic->devcmd_lock);
1969 static int enic_set_rss_nic_cfg(struct enic *enic)
1971 struct device *dev = enic_get_dev(enic);
1972 const u8 rss_default_cpu = 0;
1973 const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1974 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1975 NIC_CFG_RSS_HASH_TYPE_IPV6 |
1976 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1977 const u8 rss_hash_bits = 7;
1978 const u8 rss_base_cpu = 0;
1979 u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1982 if (!enic_set_rsskey(enic)) {
1983 if (enic_set_rsscpu(enic, rss_hash_bits)) {
1985 dev_warn(dev, "RSS disabled, "
1986 "Failed to set RSS cpu indirection table.");
1990 dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
1994 return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1995 rss_hash_bits, rss_base_cpu, rss_enable);
1998 static void enic_reset(struct work_struct *work)
2000 struct enic *enic = container_of(work, struct enic, reset);
2002 if (!netif_running(enic->netdev))
2007 spin_lock(&enic->enic_api_lock);
2008 enic_dev_hang_notify(enic);
2009 enic_stop(enic->netdev);
2010 enic_dev_hang_reset(enic);
2011 enic_reset_addr_lists(enic);
2012 enic_init_vnic_resources(enic);
2013 enic_set_rss_nic_cfg(enic);
2014 enic_dev_set_ig_vlan_rewrite_mode(enic);
2015 enic_open(enic->netdev);
2016 spin_unlock(&enic->enic_api_lock);
2017 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2022 static int enic_set_intr_mode(struct enic *enic)
2024 unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2025 unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
2028 /* Set interrupt mode (INTx, MSI, MSI-X) depending
2029 * on system capabilities.
2033 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2034 * (the second to last INTR is used for WQ/RQ errors)
2035 * (the last INTR is used for notifications)
2038 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2039 for (i = 0; i < n + m + 2; i++)
2040 enic->msix_entry[i].entry = i;
2042 /* Use multiple RQs if RSS is enabled
2045 if (ENIC_SETTING(enic, RSS) &&
2046 enic->config.intr_mode < 1 &&
2047 enic->rq_count >= n &&
2048 enic->wq_count >= m &&
2049 enic->cq_count >= n + m &&
2050 enic->intr_count >= n + m + 2) {
2052 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2053 n + m + 2, n + m + 2) > 0) {
2057 enic->cq_count = n + m;
2058 enic->intr_count = n + m + 2;
2060 vnic_dev_set_intr_mode(enic->vdev,
2061 VNIC_DEV_INTR_MODE_MSIX);
2067 if (enic->config.intr_mode < 1 &&
2068 enic->rq_count >= 1 &&
2069 enic->wq_count >= m &&
2070 enic->cq_count >= 1 + m &&
2071 enic->intr_count >= 1 + m + 2) {
2072 if (pci_enable_msix_range(enic->pdev, enic->msix_entry,
2073 1 + m + 2, 1 + m + 2) > 0) {
2077 enic->cq_count = 1 + m;
2078 enic->intr_count = 1 + m + 2;
2080 vnic_dev_set_intr_mode(enic->vdev,
2081 VNIC_DEV_INTR_MODE_MSIX);
2089 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2092 if (enic->config.intr_mode < 2 &&
2093 enic->rq_count >= 1 &&
2094 enic->wq_count >= 1 &&
2095 enic->cq_count >= 2 &&
2096 enic->intr_count >= 1 &&
2097 !pci_enable_msi(enic->pdev)) {
2102 enic->intr_count = 1;
2104 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2111 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2112 * (the first INTR is used for WQ/RQ)
2113 * (the second INTR is used for WQ/RQ errors)
2114 * (the last INTR is used for notifications)
2117 if (enic->config.intr_mode < 3 &&
2118 enic->rq_count >= 1 &&
2119 enic->wq_count >= 1 &&
2120 enic->cq_count >= 2 &&
2121 enic->intr_count >= 3) {
2126 enic->intr_count = 3;
2128 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2133 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2138 static void enic_clear_intr_mode(struct enic *enic)
2140 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2141 case VNIC_DEV_INTR_MODE_MSIX:
2142 pci_disable_msix(enic->pdev);
2144 case VNIC_DEV_INTR_MODE_MSI:
2145 pci_disable_msi(enic->pdev);
2151 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2154 static const struct net_device_ops enic_netdev_dynamic_ops = {
2155 .ndo_open = enic_open,
2156 .ndo_stop = enic_stop,
2157 .ndo_start_xmit = enic_hard_start_xmit,
2158 .ndo_get_stats64 = enic_get_stats,
2159 .ndo_validate_addr = eth_validate_addr,
2160 .ndo_set_rx_mode = enic_set_rx_mode,
2161 .ndo_set_mac_address = enic_set_mac_address_dynamic,
2162 .ndo_change_mtu = enic_change_mtu,
2163 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2164 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2165 .ndo_tx_timeout = enic_tx_timeout,
2166 .ndo_set_vf_port = enic_set_vf_port,
2167 .ndo_get_vf_port = enic_get_vf_port,
2168 .ndo_set_vf_mac = enic_set_vf_mac,
2169 #ifdef CONFIG_NET_POLL_CONTROLLER
2170 .ndo_poll_controller = enic_poll_controller,
2172 #ifdef CONFIG_RFS_ACCEL
2173 .ndo_rx_flow_steer = enic_rx_flow_steer,
2175 #ifdef CONFIG_NET_RX_BUSY_POLL
2176 .ndo_busy_poll = enic_busy_poll,
2180 static const struct net_device_ops enic_netdev_ops = {
2181 .ndo_open = enic_open,
2182 .ndo_stop = enic_stop,
2183 .ndo_start_xmit = enic_hard_start_xmit,
2184 .ndo_get_stats64 = enic_get_stats,
2185 .ndo_validate_addr = eth_validate_addr,
2186 .ndo_set_mac_address = enic_set_mac_address,
2187 .ndo_set_rx_mode = enic_set_rx_mode,
2188 .ndo_change_mtu = enic_change_mtu,
2189 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
2190 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
2191 .ndo_tx_timeout = enic_tx_timeout,
2192 .ndo_set_vf_port = enic_set_vf_port,
2193 .ndo_get_vf_port = enic_get_vf_port,
2194 .ndo_set_vf_mac = enic_set_vf_mac,
2195 #ifdef CONFIG_NET_POLL_CONTROLLER
2196 .ndo_poll_controller = enic_poll_controller,
2198 #ifdef CONFIG_RFS_ACCEL
2199 .ndo_rx_flow_steer = enic_rx_flow_steer,
2201 #ifdef CONFIG_NET_RX_BUSY_POLL
2202 .ndo_busy_poll = enic_busy_poll,
2206 static void enic_dev_deinit(struct enic *enic)
2210 for (i = 0; i < enic->rq_count; i++) {
2211 napi_hash_del(&enic->napi[i]);
2212 netif_napi_del(&enic->napi[i]);
2214 if (vnic_dev_get_intr_mode(enic->vdev) == VNIC_DEV_INTR_MODE_MSIX)
2215 for (i = 0; i < enic->wq_count; i++)
2216 netif_napi_del(&enic->napi[enic_cq_wq(enic, i)]);
2218 enic_free_vnic_resources(enic);
2219 enic_clear_intr_mode(enic);
2222 static int enic_dev_init(struct enic *enic)
2224 struct device *dev = enic_get_dev(enic);
2225 struct net_device *netdev = enic->netdev;
2229 /* Get interrupt coalesce timer info */
2230 err = enic_dev_intr_coal_timer_info(enic);
2232 dev_warn(dev, "Using default conversion factor for "
2233 "interrupt coalesce timer\n");
2234 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2237 /* Get vNIC configuration
2240 err = enic_get_vnic_config(enic);
2242 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2246 /* Get available resource counts
2249 enic_get_res_counts(enic);
2251 /* Set interrupt mode based on resource counts and system
2255 err = enic_set_intr_mode(enic);
2257 dev_err(dev, "Failed to set intr mode based on resource "
2258 "counts and system capabilities, aborting\n");
2262 /* Allocate and configure vNIC resources
2265 err = enic_alloc_vnic_resources(enic);
2267 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2268 goto err_out_free_vnic_resources;
2271 enic_init_vnic_resources(enic);
2273 err = enic_set_rss_nic_cfg(enic);
2275 dev_err(dev, "Failed to config nic, aborting\n");
2276 goto err_out_free_vnic_resources;
2279 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2281 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2282 napi_hash_add(&enic->napi[0]);
2284 case VNIC_DEV_INTR_MODE_MSIX:
2285 for (i = 0; i < enic->rq_count; i++) {
2286 netif_napi_add(netdev, &enic->napi[i],
2287 enic_poll_msix_rq, NAPI_POLL_WEIGHT);
2288 napi_hash_add(&enic->napi[i]);
2290 for (i = 0; i < enic->wq_count; i++)
2291 netif_napi_add(netdev, &enic->napi[enic_cq_wq(enic, i)],
2292 enic_poll_msix_wq, NAPI_POLL_WEIGHT);
2298 err_out_free_vnic_resources:
2299 enic_clear_intr_mode(enic);
2300 enic_free_vnic_resources(enic);
2305 static void enic_iounmap(struct enic *enic)
2309 for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2310 if (enic->bar[i].vaddr)
2311 iounmap(enic->bar[i].vaddr);
2314 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2316 struct device *dev = &pdev->dev;
2317 struct net_device *netdev;
2322 #ifdef CONFIG_PCI_IOV
2327 /* Allocate net device structure and initialize. Private
2328 * instance data is initialized to zero.
2331 netdev = alloc_etherdev_mqs(sizeof(struct enic),
2332 ENIC_RQ_MAX, ENIC_WQ_MAX);
2336 pci_set_drvdata(pdev, netdev);
2338 SET_NETDEV_DEV(netdev, &pdev->dev);
2340 enic = netdev_priv(netdev);
2341 enic->netdev = netdev;
2344 /* Setup PCI resources
2347 err = pci_enable_device_mem(pdev);
2349 dev_err(dev, "Cannot enable PCI device, aborting\n");
2350 goto err_out_free_netdev;
2353 err = pci_request_regions(pdev, DRV_NAME);
2355 dev_err(dev, "Cannot request PCI regions, aborting\n");
2356 goto err_out_disable_device;
2359 pci_set_master(pdev);
2361 /* Query PCI controller on system for DMA addressing
2362 * limitation for the device. Try 64-bit first, and
2366 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2368 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2370 dev_err(dev, "No usable DMA configuration, aborting\n");
2371 goto err_out_release_regions;
2373 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2375 dev_err(dev, "Unable to obtain %u-bit DMA "
2376 "for consistent allocations, aborting\n", 32);
2377 goto err_out_release_regions;
2380 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2382 dev_err(dev, "Unable to obtain %u-bit DMA "
2383 "for consistent allocations, aborting\n", 64);
2384 goto err_out_release_regions;
2389 /* Map vNIC resources from BAR0-5
2392 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2393 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2395 enic->bar[i].len = pci_resource_len(pdev, i);
2396 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2397 if (!enic->bar[i].vaddr) {
2398 dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2400 goto err_out_iounmap;
2402 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2405 /* Register vNIC device
2408 enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2409 ARRAY_SIZE(enic->bar));
2411 dev_err(dev, "vNIC registration failed, aborting\n");
2413 goto err_out_iounmap;
2416 #ifdef CONFIG_PCI_IOV
2417 /* Get number of subvnics */
2418 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2420 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2422 if (enic->num_vfs) {
2423 err = pci_enable_sriov(pdev, enic->num_vfs);
2425 dev_err(dev, "SRIOV enable failed, aborting."
2426 " pci_enable_sriov() returned %d\n",
2428 goto err_out_vnic_unregister;
2430 enic->priv_flags |= ENIC_SRIOV_ENABLED;
2431 num_pps = enic->num_vfs;
2436 /* Allocate structure for port profiles */
2437 enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2440 goto err_out_disable_sriov_pp;
2443 /* Issue device open to get device in known state
2446 err = enic_dev_open(enic);
2448 dev_err(dev, "vNIC dev open failed, aborting\n");
2449 goto err_out_disable_sriov;
2452 /* Setup devcmd lock
2455 spin_lock_init(&enic->devcmd_lock);
2456 spin_lock_init(&enic->enic_api_lock);
2459 * Set ingress vlan rewrite mode before vnic initialization
2462 err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2465 "Failed to set ingress vlan rewrite mode, aborting.\n");
2466 goto err_out_dev_close;
2469 /* Issue device init to initialize the vnic-to-switch link.
2470 * We'll start with carrier off and wait for link UP
2471 * notification later to turn on carrier. We don't need
2472 * to wait here for the vnic-to-switch link initialization
2473 * to complete; link UP notification is the indication that
2474 * the process is complete.
2477 netif_carrier_off(netdev);
2479 /* Do not call dev_init for a dynamic vnic.
2480 * For a dynamic vnic, init_prov_info will be
2481 * called later by an upper layer.
2484 if (!enic_is_dynamic(enic)) {
2485 err = vnic_dev_init(enic->vdev, 0);
2487 dev_err(dev, "vNIC dev init failed, aborting\n");
2488 goto err_out_dev_close;
2492 err = enic_dev_init(enic);
2494 dev_err(dev, "Device initialization failed, aborting\n");
2495 goto err_out_dev_close;
2498 netif_set_real_num_tx_queues(netdev, enic->wq_count);
2499 netif_set_real_num_rx_queues(netdev, enic->rq_count);
2501 /* Setup notification timer, HW reset task, and wq locks
2504 init_timer(&enic->notify_timer);
2505 enic->notify_timer.function = enic_notify_timer;
2506 enic->notify_timer.data = (unsigned long)enic;
2508 enic_set_rx_coal_setting(enic);
2509 INIT_WORK(&enic->reset, enic_reset);
2510 INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2512 for (i = 0; i < enic->wq_count; i++)
2513 spin_lock_init(&enic->wq_lock[i]);
2515 /* Register net device
2518 enic->port_mtu = enic->config.mtu;
2519 (void)enic_change_mtu(netdev, enic->port_mtu);
2521 err = enic_set_mac_addr(netdev, enic->mac_addr);
2523 dev_err(dev, "Invalid MAC address, aborting\n");
2524 goto err_out_dev_deinit;
2527 enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2528 /* rx coalesce time already got initialized. This gets used
2529 * if adaptive coal is turned off
2531 enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2533 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2534 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2536 netdev->netdev_ops = &enic_netdev_ops;
2538 netdev->watchdog_timeo = 2 * HZ;
2539 enic_set_ethtool_ops(netdev);
2541 netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2542 if (ENIC_SETTING(enic, LOOP)) {
2543 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2544 enic->loop_enable = 1;
2545 enic->loop_tag = enic->config.loop_tag;
2546 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2548 if (ENIC_SETTING(enic, TXCSUM))
2549 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2550 if (ENIC_SETTING(enic, TSO))
2551 netdev->hw_features |= NETIF_F_TSO |
2552 NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2553 if (ENIC_SETTING(enic, RSS))
2554 netdev->hw_features |= NETIF_F_RXHASH;
2555 if (ENIC_SETTING(enic, RXCSUM))
2556 netdev->hw_features |= NETIF_F_RXCSUM;
2558 netdev->features |= netdev->hw_features;
2560 #ifdef CONFIG_RFS_ACCEL
2561 netdev->hw_features |= NETIF_F_NTUPLE;
2565 netdev->features |= NETIF_F_HIGHDMA;
2567 netdev->priv_flags |= IFF_UNICAST_FLT;
2569 err = register_netdev(netdev);
2571 dev_err(dev, "Cannot register net device, aborting\n");
2572 goto err_out_dev_deinit;
2574 enic->rx_copybreak = RX_COPYBREAK_DEFAULT;
2579 enic_dev_deinit(enic);
2581 vnic_dev_close(enic->vdev);
2582 err_out_disable_sriov:
2584 err_out_disable_sriov_pp:
2585 #ifdef CONFIG_PCI_IOV
2586 if (enic_sriov_enabled(enic)) {
2587 pci_disable_sriov(pdev);
2588 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2590 err_out_vnic_unregister:
2592 vnic_dev_unregister(enic->vdev);
2595 err_out_release_regions:
2596 pci_release_regions(pdev);
2597 err_out_disable_device:
2598 pci_disable_device(pdev);
2599 err_out_free_netdev:
2600 free_netdev(netdev);
2605 static void enic_remove(struct pci_dev *pdev)
2607 struct net_device *netdev = pci_get_drvdata(pdev);
2610 struct enic *enic = netdev_priv(netdev);
2612 cancel_work_sync(&enic->reset);
2613 cancel_work_sync(&enic->change_mtu_work);
2614 unregister_netdev(netdev);
2615 enic_dev_deinit(enic);
2616 vnic_dev_close(enic->vdev);
2617 #ifdef CONFIG_PCI_IOV
2618 if (enic_sriov_enabled(enic)) {
2619 pci_disable_sriov(pdev);
2620 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2624 vnic_dev_unregister(enic->vdev);
2626 pci_release_regions(pdev);
2627 pci_disable_device(pdev);
2628 free_netdev(netdev);
2632 static struct pci_driver enic_driver = {
2634 .id_table = enic_id_table,
2635 .probe = enic_probe,
2636 .remove = enic_remove,
2639 static int __init enic_init_module(void)
2641 pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2643 return pci_register_driver(&enic_driver);
2646 static void __exit enic_cleanup_module(void)
2648 pci_unregister_driver(&enic_driver);
2651 module_init(enic_init_module);
2652 module_exit(enic_cleanup_module);