]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/net/enic/enic_main.c
Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[net-next-2.6.git] / drivers / net / enic / enic_main.c
CommitLineData
01f2e4ea
SF
1/*
2 * Copyright 2008 Cisco Systems, Inc. All rights reserved.
3 * Copyright 2007 Nuova Systems, Inc. All rights reserved.
4 *
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.
8 *
9 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
10 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
11 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
12 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
13 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
14 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
15 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
16 * SOFTWARE.
17 *
18 */
19
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/workqueue.h>
27#include <linux/pci.h>
28#include <linux/netdevice.h>
29#include <linux/etherdevice.h>
30#include <linux/if_ether.h>
31#include <linux/if_vlan.h>
32#include <linux/ethtool.h>
33#include <linux/in.h>
34#include <linux/ip.h>
35#include <linux/ipv6.h>
36#include <linux/tcp.h>
b7c6bfb7 37#include <net/ip6_checksum.h>
01f2e4ea
SF
38
39#include "cq_enet_desc.h"
40#include "vnic_dev.h"
41#include "vnic_intr.h"
42#include "vnic_stats.h"
43#include "enic_res.h"
44#include "enic.h"
45
46#define ENIC_NOTIFY_TIMER_PERIOD (2 * HZ)
01f2e4ea
SF
47
48/* Supported devices */
49static struct pci_device_id enic_id_table[] = {
50 { PCI_VDEVICE(CISCO, 0x0043) },
51 { 0, } /* end of table */
52};
53
54MODULE_DESCRIPTION(DRV_DESCRIPTION);
55MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
56MODULE_LICENSE("GPL");
57MODULE_VERSION(DRV_VERSION);
58MODULE_DEVICE_TABLE(pci, enic_id_table);
59
60struct enic_stat {
61 char name[ETH_GSTRING_LEN];
62 unsigned int offset;
63};
64
65#define ENIC_TX_STAT(stat) \
66 { .name = #stat, .offset = offsetof(struct vnic_tx_stats, stat) / 8 }
67#define ENIC_RX_STAT(stat) \
68 { .name = #stat, .offset = offsetof(struct vnic_rx_stats, stat) / 8 }
69
70static const struct enic_stat enic_tx_stats[] = {
71 ENIC_TX_STAT(tx_frames_ok),
72 ENIC_TX_STAT(tx_unicast_frames_ok),
73 ENIC_TX_STAT(tx_multicast_frames_ok),
74 ENIC_TX_STAT(tx_broadcast_frames_ok),
75 ENIC_TX_STAT(tx_bytes_ok),
76 ENIC_TX_STAT(tx_unicast_bytes_ok),
77 ENIC_TX_STAT(tx_multicast_bytes_ok),
78 ENIC_TX_STAT(tx_broadcast_bytes_ok),
79 ENIC_TX_STAT(tx_drops),
80 ENIC_TX_STAT(tx_errors),
81 ENIC_TX_STAT(tx_tso),
82};
83
84static const struct enic_stat enic_rx_stats[] = {
85 ENIC_RX_STAT(rx_frames_ok),
86 ENIC_RX_STAT(rx_frames_total),
87 ENIC_RX_STAT(rx_unicast_frames_ok),
88 ENIC_RX_STAT(rx_multicast_frames_ok),
89 ENIC_RX_STAT(rx_broadcast_frames_ok),
90 ENIC_RX_STAT(rx_bytes_ok),
91 ENIC_RX_STAT(rx_unicast_bytes_ok),
92 ENIC_RX_STAT(rx_multicast_bytes_ok),
93 ENIC_RX_STAT(rx_broadcast_bytes_ok),
94 ENIC_RX_STAT(rx_drop),
95 ENIC_RX_STAT(rx_no_bufs),
96 ENIC_RX_STAT(rx_errors),
97 ENIC_RX_STAT(rx_rss),
98 ENIC_RX_STAT(rx_crc_errors),
99 ENIC_RX_STAT(rx_frames_64),
100 ENIC_RX_STAT(rx_frames_127),
101 ENIC_RX_STAT(rx_frames_255),
102 ENIC_RX_STAT(rx_frames_511),
103 ENIC_RX_STAT(rx_frames_1023),
104 ENIC_RX_STAT(rx_frames_1518),
105 ENIC_RX_STAT(rx_frames_to_max),
106};
107
108static const unsigned int enic_n_tx_stats = ARRAY_SIZE(enic_tx_stats);
109static const unsigned int enic_n_rx_stats = ARRAY_SIZE(enic_rx_stats);
110
111static int enic_get_settings(struct net_device *netdev,
112 struct ethtool_cmd *ecmd)
113{
114 struct enic *enic = netdev_priv(netdev);
115
116 ecmd->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
117 ecmd->advertising = (ADVERTISED_10000baseT_Full | ADVERTISED_FIBRE);
118 ecmd->port = PORT_FIBRE;
119 ecmd->transceiver = XCVR_EXTERNAL;
120
121 if (netif_carrier_ok(netdev)) {
122 ecmd->speed = vnic_dev_port_speed(enic->vdev);
123 ecmd->duplex = DUPLEX_FULL;
124 } else {
125 ecmd->speed = -1;
126 ecmd->duplex = -1;
127 }
128
129 ecmd->autoneg = AUTONEG_DISABLE;
130
131 return 0;
132}
133
134static void enic_get_drvinfo(struct net_device *netdev,
135 struct ethtool_drvinfo *drvinfo)
136{
137 struct enic *enic = netdev_priv(netdev);
138 struct vnic_devcmd_fw_info *fw_info;
139
140 spin_lock(&enic->devcmd_lock);
141 vnic_dev_fw_info(enic->vdev, &fw_info);
142 spin_unlock(&enic->devcmd_lock);
143
144 strncpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
145 strncpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
146 strncpy(drvinfo->fw_version, fw_info->fw_version,
147 sizeof(drvinfo->fw_version));
148 strncpy(drvinfo->bus_info, pci_name(enic->pdev),
149 sizeof(drvinfo->bus_info));
150}
151
152static void enic_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
153{
154 unsigned int i;
155
156 switch (stringset) {
157 case ETH_SS_STATS:
158 for (i = 0; i < enic_n_tx_stats; i++) {
159 memcpy(data, enic_tx_stats[i].name, ETH_GSTRING_LEN);
160 data += ETH_GSTRING_LEN;
161 }
162 for (i = 0; i < enic_n_rx_stats; i++) {
163 memcpy(data, enic_rx_stats[i].name, ETH_GSTRING_LEN);
164 data += ETH_GSTRING_LEN;
165 }
166 break;
167 }
168}
169
25f0a061 170static int enic_get_sset_count(struct net_device *netdev, int sset)
01f2e4ea 171{
25f0a061
SF
172 switch (sset) {
173 case ETH_SS_STATS:
174 return enic_n_tx_stats + enic_n_rx_stats;
175 default:
176 return -EOPNOTSUPP;
177 }
01f2e4ea
SF
178}
179
180static void enic_get_ethtool_stats(struct net_device *netdev,
181 struct ethtool_stats *stats, u64 *data)
182{
183 struct enic *enic = netdev_priv(netdev);
184 struct vnic_stats *vstats;
185 unsigned int i;
186
187 spin_lock(&enic->devcmd_lock);
188 vnic_dev_stats_dump(enic->vdev, &vstats);
189 spin_unlock(&enic->devcmd_lock);
190
191 for (i = 0; i < enic_n_tx_stats; i++)
192 *(data++) = ((u64 *)&vstats->tx)[enic_tx_stats[i].offset];
193 for (i = 0; i < enic_n_rx_stats; i++)
194 *(data++) = ((u64 *)&vstats->rx)[enic_rx_stats[i].offset];
195}
196
197static u32 enic_get_rx_csum(struct net_device *netdev)
198{
199 struct enic *enic = netdev_priv(netdev);
200 return enic->csum_rx_enabled;
201}
202
203static int enic_set_rx_csum(struct net_device *netdev, u32 data)
204{
205 struct enic *enic = netdev_priv(netdev);
206
25f0a061
SF
207 if (data && !ENIC_SETTING(enic, RXCSUM))
208 return -EINVAL;
209
210 enic->csum_rx_enabled = !!data;
01f2e4ea
SF
211
212 return 0;
213}
214
215static int enic_set_tx_csum(struct net_device *netdev, u32 data)
216{
217 struct enic *enic = netdev_priv(netdev);
218
25f0a061
SF
219 if (data && !ENIC_SETTING(enic, TXCSUM))
220 return -EINVAL;
221
222 if (data)
01f2e4ea
SF
223 netdev->features |= NETIF_F_HW_CSUM;
224 else
225 netdev->features &= ~NETIF_F_HW_CSUM;
226
227 return 0;
228}
229
230static int enic_set_tso(struct net_device *netdev, u32 data)
231{
232 struct enic *enic = netdev_priv(netdev);
233
25f0a061
SF
234 if (data && !ENIC_SETTING(enic, TSO))
235 return -EINVAL;
236
237 if (data)
01f2e4ea
SF
238 netdev->features |=
239 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN;
240 else
241 netdev->features &=
242 ~(NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN);
243
244 return 0;
245}
246
247static u32 enic_get_msglevel(struct net_device *netdev)
248{
249 struct enic *enic = netdev_priv(netdev);
250 return enic->msg_enable;
251}
252
253static void enic_set_msglevel(struct net_device *netdev, u32 value)
254{
255 struct enic *enic = netdev_priv(netdev);
256 enic->msg_enable = value;
257}
258
259static struct ethtool_ops enic_ethtool_ops = {
260 .get_settings = enic_get_settings,
261 .get_drvinfo = enic_get_drvinfo,
262 .get_msglevel = enic_get_msglevel,
263 .set_msglevel = enic_set_msglevel,
264 .get_link = ethtool_op_get_link,
265 .get_strings = enic_get_strings,
25f0a061 266 .get_sset_count = enic_get_sset_count,
01f2e4ea
SF
267 .get_ethtool_stats = enic_get_ethtool_stats,
268 .get_rx_csum = enic_get_rx_csum,
269 .set_rx_csum = enic_set_rx_csum,
270 .get_tx_csum = ethtool_op_get_tx_csum,
271 .set_tx_csum = enic_set_tx_csum,
272 .get_sg = ethtool_op_get_sg,
273 .set_sg = ethtool_op_set_sg,
274 .get_tso = ethtool_op_get_tso,
275 .set_tso = enic_set_tso,
276};
277
278static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
279{
280 struct enic *enic = vnic_dev_priv(wq->vdev);
281
282 if (buf->sop)
283 pci_unmap_single(enic->pdev, buf->dma_addr,
284 buf->len, PCI_DMA_TODEVICE);
285 else
286 pci_unmap_page(enic->pdev, buf->dma_addr,
287 buf->len, PCI_DMA_TODEVICE);
288
289 if (buf->os_buf)
290 dev_kfree_skb_any(buf->os_buf);
291}
292
293static void enic_wq_free_buf(struct vnic_wq *wq,
294 struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
295{
296 enic_free_wq_buf(wq, buf);
297}
298
299static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
300 u8 type, u16 q_number, u16 completed_index, void *opaque)
301{
302 struct enic *enic = vnic_dev_priv(vdev);
303
304 spin_lock(&enic->wq_lock[q_number]);
305
306 vnic_wq_service(&enic->wq[q_number], cq_desc,
307 completed_index, enic_wq_free_buf,
308 opaque);
309
310 if (netif_queue_stopped(enic->netdev) &&
311 vnic_wq_desc_avail(&enic->wq[q_number]) >= MAX_SKB_FRAGS + 1)
312 netif_wake_queue(enic->netdev);
313
314 spin_unlock(&enic->wq_lock[q_number]);
315
316 return 0;
317}
318
319static void enic_log_q_error(struct enic *enic)
320{
321 unsigned int i;
322 u32 error_status;
323
324 for (i = 0; i < enic->wq_count; i++) {
325 error_status = vnic_wq_error_status(&enic->wq[i]);
326 if (error_status)
327 printk(KERN_ERR PFX "%s: WQ[%d] error_status %d\n",
328 enic->netdev->name, i, error_status);
329 }
330
331 for (i = 0; i < enic->rq_count; i++) {
332 error_status = vnic_rq_error_status(&enic->rq[i]);
333 if (error_status)
334 printk(KERN_ERR PFX "%s: RQ[%d] error_status %d\n",
335 enic->netdev->name, i, error_status);
336 }
337}
338
339static void enic_link_check(struct enic *enic)
340{
341 int link_status = vnic_dev_link_status(enic->vdev);
342 int carrier_ok = netif_carrier_ok(enic->netdev);
343
344 if (link_status && !carrier_ok) {
345 printk(KERN_INFO PFX "%s: Link UP\n", enic->netdev->name);
346 netif_carrier_on(enic->netdev);
347 } else if (!link_status && carrier_ok) {
348 printk(KERN_INFO PFX "%s: Link DOWN\n", enic->netdev->name);
349 netif_carrier_off(enic->netdev);
350 }
351}
352
353static void enic_mtu_check(struct enic *enic)
354{
355 u32 mtu = vnic_dev_mtu(enic->vdev);
356
357 if (mtu != enic->port_mtu) {
358 if (mtu < enic->netdev->mtu)
359 printk(KERN_WARNING PFX
360 "%s: interface MTU (%d) set higher "
361 "than switch port MTU (%d)\n",
362 enic->netdev->name, enic->netdev->mtu, mtu);
363 enic->port_mtu = mtu;
364 }
365}
366
367static void enic_msglvl_check(struct enic *enic)
368{
369 u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
370
371 if (msg_enable != enic->msg_enable) {
372 printk(KERN_INFO PFX "%s: msg lvl changed from 0x%x to 0x%x\n",
373 enic->netdev->name, enic->msg_enable, msg_enable);
374 enic->msg_enable = msg_enable;
375 }
376}
377
378static void enic_notify_check(struct enic *enic)
379{
380 enic_msglvl_check(enic);
381 enic_mtu_check(enic);
382 enic_link_check(enic);
383}
384
385#define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
386
387static irqreturn_t enic_isr_legacy(int irq, void *data)
388{
389 struct net_device *netdev = data;
390 struct enic *enic = netdev_priv(netdev);
391 u32 pba;
392
393 vnic_intr_mask(&enic->intr[ENIC_INTX_WQ_RQ]);
394
395 pba = vnic_intr_legacy_pba(enic->legacy_pba);
396 if (!pba) {
397 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
398 return IRQ_NONE; /* not our interrupt */
399 }
400
401 if (ENIC_TEST_INTR(pba, ENIC_INTX_NOTIFY))
402 enic_notify_check(enic);
403
404 if (ENIC_TEST_INTR(pba, ENIC_INTX_ERR)) {
405 enic_log_q_error(enic);
406 /* schedule recovery from WQ/RQ error */
407 schedule_work(&enic->reset);
408 return IRQ_HANDLED;
409 }
410
411 if (ENIC_TEST_INTR(pba, ENIC_INTX_WQ_RQ)) {
412 if (netif_rx_schedule_prep(netdev, &enic->napi))
413 __netif_rx_schedule(netdev, &enic->napi);
414 } else {
415 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
416 }
417
418 return IRQ_HANDLED;
419}
420
421static irqreturn_t enic_isr_msi(int irq, void *data)
422{
423 struct enic *enic = data;
424
425 /* With MSI, there is no sharing of interrupts, so this is
426 * our interrupt and there is no need to ack it. The device
427 * is not providing per-vector masking, so the OS will not
428 * write to PCI config space to mask/unmask the interrupt.
429 * We're using mask_on_assertion for MSI, so the device
430 * automatically masks the interrupt when the interrupt is
431 * generated. Later, when exiting polling, the interrupt
432 * will be unmasked (see enic_poll).
433 *
434 * Also, the device uses the same PCIe Traffic Class (TC)
435 * for Memory Write data and MSI, so there are no ordering
436 * issues; the MSI will always arrive at the Root Complex
437 * _after_ corresponding Memory Writes (i.e. descriptor
438 * writes).
439 */
440
441 netif_rx_schedule(enic->netdev, &enic->napi);
442
443 return IRQ_HANDLED;
444}
445
446static irqreturn_t enic_isr_msix_rq(int irq, void *data)
447{
448 struct enic *enic = data;
449
450 /* schedule NAPI polling for RQ cleanup */
451 netif_rx_schedule(enic->netdev, &enic->napi);
452
453 return IRQ_HANDLED;
454}
455
456static irqreturn_t enic_isr_msix_wq(int irq, void *data)
457{
458 struct enic *enic = data;
459 unsigned int wq_work_to_do = -1; /* no limit */
460 unsigned int wq_work_done;
461
462 wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
463 wq_work_to_do, enic_wq_service, NULL);
464
465 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_WQ],
466 wq_work_done,
467 1 /* unmask intr */,
468 1 /* reset intr timer */);
469
470 return IRQ_HANDLED;
471}
472
473static irqreturn_t enic_isr_msix_err(int irq, void *data)
474{
475 struct enic *enic = data;
476
477 enic_log_q_error(enic);
478
479 /* schedule recovery from WQ/RQ error */
480 schedule_work(&enic->reset);
481
482 return IRQ_HANDLED;
483}
484
485static irqreturn_t enic_isr_msix_notify(int irq, void *data)
486{
487 struct enic *enic = data;
488
489 enic_notify_check(enic);
490 vnic_intr_unmask(&enic->intr[ENIC_MSIX_NOTIFY]);
491
492 return IRQ_HANDLED;
493}
494
495static inline void enic_queue_wq_skb_cont(struct enic *enic,
496 struct vnic_wq *wq, struct sk_buff *skb,
497 unsigned int len_left)
498{
499 skb_frag_t *frag;
500
501 /* Queue additional data fragments */
502 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
503 len_left -= frag->size;
504 enic_queue_wq_desc_cont(wq, skb,
505 pci_map_page(enic->pdev, frag->page,
506 frag->page_offset, frag->size,
507 PCI_DMA_TODEVICE),
508 frag->size,
509 (len_left == 0)); /* EOP? */
510 }
511}
512
513static inline void enic_queue_wq_skb_vlan(struct enic *enic,
514 struct vnic_wq *wq, struct sk_buff *skb,
515 int vlan_tag_insert, unsigned int vlan_tag)
516{
517 unsigned int head_len = skb_headlen(skb);
518 unsigned int len_left = skb->len - head_len;
519 int eop = (len_left == 0);
520
521 /* Queue the main skb fragment */
522 enic_queue_wq_desc(wq, skb,
523 pci_map_single(enic->pdev, skb->data,
524 head_len, PCI_DMA_TODEVICE),
525 head_len,
526 vlan_tag_insert, vlan_tag,
527 eop);
528
529 if (!eop)
530 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
531}
532
533static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
534 struct vnic_wq *wq, struct sk_buff *skb,
535 int vlan_tag_insert, unsigned int vlan_tag)
536{
537 unsigned int head_len = skb_headlen(skb);
538 unsigned int len_left = skb->len - head_len;
539 unsigned int hdr_len = skb_transport_offset(skb);
540 unsigned int csum_offset = hdr_len + skb->csum_offset;
541 int eop = (len_left == 0);
542
543 /* Queue the main skb fragment */
544 enic_queue_wq_desc_csum_l4(wq, skb,
545 pci_map_single(enic->pdev, skb->data,
546 head_len, PCI_DMA_TODEVICE),
547 head_len,
548 csum_offset,
549 hdr_len,
550 vlan_tag_insert, vlan_tag,
551 eop);
552
553 if (!eop)
554 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
555}
556
557static inline void enic_queue_wq_skb_tso(struct enic *enic,
558 struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
559 int vlan_tag_insert, unsigned int vlan_tag)
560{
561 unsigned int head_len = skb_headlen(skb);
562 unsigned int len_left = skb->len - head_len;
563 unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
564 int eop = (len_left == 0);
565
566 /* Preload TCP csum field with IP pseudo hdr calculated
567 * with IP length set to zero. HW will later add in length
568 * to each TCP segment resulting from the TSO.
569 */
570
571 if (skb->protocol == __constant_htons(ETH_P_IP)) {
572 ip_hdr(skb)->check = 0;
573 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
574 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
575 } else if (skb->protocol == __constant_htons(ETH_P_IPV6)) {
576 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
577 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
578 }
579
580 /* Queue the main skb fragment */
581 enic_queue_wq_desc_tso(wq, skb,
582 pci_map_single(enic->pdev, skb->data,
583 head_len, PCI_DMA_TODEVICE),
584 head_len,
585 mss, hdr_len,
586 vlan_tag_insert, vlan_tag,
587 eop);
588
589 if (!eop)
590 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
591}
592
593static inline void enic_queue_wq_skb(struct enic *enic,
594 struct vnic_wq *wq, struct sk_buff *skb)
595{
596 unsigned int mss = skb_shinfo(skb)->gso_size;
597 unsigned int vlan_tag = 0;
598 int vlan_tag_insert = 0;
599
600 if (enic->vlan_group && vlan_tx_tag_present(skb)) {
601 /* VLAN tag from trunking driver */
602 vlan_tag_insert = 1;
603 vlan_tag = vlan_tx_tag_get(skb);
604 }
605
606 if (mss)
607 enic_queue_wq_skb_tso(enic, wq, skb, mss,
608 vlan_tag_insert, vlan_tag);
609 else if (skb->ip_summed == CHECKSUM_PARTIAL)
610 enic_queue_wq_skb_csum_l4(enic, wq, skb,
611 vlan_tag_insert, vlan_tag);
612 else
613 enic_queue_wq_skb_vlan(enic, wq, skb,
614 vlan_tag_insert, vlan_tag);
615}
616
617/* netif_tx_lock held, process context with BHs disabled */
618static int enic_hard_start_xmit(struct sk_buff *skb, struct net_device *netdev)
619{
620 struct enic *enic = netdev_priv(netdev);
621 struct vnic_wq *wq = &enic->wq[0];
622 unsigned long flags;
623
624 if (skb->len <= 0) {
625 dev_kfree_skb(skb);
626 return NETDEV_TX_OK;
627 }
628
629 /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
630 * which is very likely. In the off chance it's going to take
631 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
632 */
633
634 if (skb_shinfo(skb)->gso_size == 0 &&
635 skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
636 skb_linearize(skb)) {
637 dev_kfree_skb(skb);
638 return NETDEV_TX_OK;
639 }
640
641 spin_lock_irqsave(&enic->wq_lock[0], flags);
642
643 if (vnic_wq_desc_avail(wq) < skb_shinfo(skb)->nr_frags + 1) {
644 netif_stop_queue(netdev);
645 /* This is a hard error, log it */
646 printk(KERN_ERR PFX "%s: BUG! Tx ring full when "
647 "queue awake!\n", netdev->name);
648 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
649 return NETDEV_TX_BUSY;
650 }
651
652 enic_queue_wq_skb(enic, wq, skb);
653
654 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + 1)
655 netif_stop_queue(netdev);
656
657 netdev->trans_start = jiffies;
658
659 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
660
661 return NETDEV_TX_OK;
662}
663
664/* dev_base_lock rwlock held, nominally process context */
665static struct net_device_stats *enic_get_stats(struct net_device *netdev)
666{
667 struct enic *enic = netdev_priv(netdev);
25f0a061 668 struct net_device_stats *net_stats = &netdev->stats;
01f2e4ea
SF
669 struct vnic_stats *stats;
670
671 spin_lock(&enic->devcmd_lock);
672 vnic_dev_stats_dump(enic->vdev, &stats);
673 spin_unlock(&enic->devcmd_lock);
674
25f0a061
SF
675 net_stats->tx_packets = stats->tx.tx_frames_ok;
676 net_stats->tx_bytes = stats->tx.tx_bytes_ok;
677 net_stats->tx_errors = stats->tx.tx_errors;
678 net_stats->tx_dropped = stats->tx.tx_drops;
01f2e4ea 679
25f0a061
SF
680 net_stats->rx_packets = stats->rx.rx_frames_ok;
681 net_stats->rx_bytes = stats->rx.rx_bytes_ok;
682 net_stats->rx_errors = stats->rx.rx_errors;
683 net_stats->multicast = stats->rx.rx_multicast_frames_ok;
684 net_stats->rx_crc_errors = stats->rx.rx_crc_errors;
685 net_stats->rx_dropped = stats->rx.rx_no_bufs;
01f2e4ea 686
25f0a061 687 return net_stats;
01f2e4ea
SF
688}
689
690static void enic_reset_mcaddrs(struct enic *enic)
691{
692 enic->mc_count = 0;
693}
694
695static int enic_set_mac_addr(struct net_device *netdev, char *addr)
696{
697 if (!is_valid_ether_addr(addr))
698 return -EADDRNOTAVAIL;
699
700 memcpy(netdev->dev_addr, addr, netdev->addr_len);
701
702 return 0;
703}
704
705/* netif_tx_lock held, BHs disabled */
706static void enic_set_multicast_list(struct net_device *netdev)
707{
708 struct enic *enic = netdev_priv(netdev);
709 struct dev_mc_list *list = netdev->mc_list;
710 int directed = 1;
711 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
712 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
713 int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
714 int allmulti = (netdev->flags & IFF_ALLMULTI) ||
715 (netdev->mc_count > ENIC_MULTICAST_PERFECT_FILTERS);
716 u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
717 unsigned int mc_count = netdev->mc_count;
718 unsigned int i, j;
719
720 if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
721 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
722
723 spin_lock(&enic->devcmd_lock);
724
725 vnic_dev_packet_filter(enic->vdev, directed,
726 multicast, broadcast, promisc, allmulti);
727
728 /* Is there an easier way? Trying to minimize to
729 * calls to add/del multicast addrs. We keep the
730 * addrs from the last call in enic->mc_addr and
731 * look for changes to add/del.
732 */
733
734 for (i = 0; list && i < mc_count; i++) {
735 memcpy(mc_addr[i], list->dmi_addr, ETH_ALEN);
736 list = list->next;
737 }
738
739 for (i = 0; i < enic->mc_count; i++) {
740 for (j = 0; j < mc_count; j++)
741 if (compare_ether_addr(enic->mc_addr[i],
742 mc_addr[j]) == 0)
743 break;
744 if (j == mc_count)
745 enic_del_multicast_addr(enic, enic->mc_addr[i]);
746 }
747
748 for (i = 0; i < mc_count; i++) {
749 for (j = 0; j < enic->mc_count; j++)
750 if (compare_ether_addr(mc_addr[i],
751 enic->mc_addr[j]) == 0)
752 break;
753 if (j == enic->mc_count)
754 enic_add_multicast_addr(enic, mc_addr[i]);
755 }
756
757 /* Save the list to compare against next time
758 */
759
760 for (i = 0; i < mc_count; i++)
761 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
762
763 enic->mc_count = mc_count;
764
765 spin_unlock(&enic->devcmd_lock);
766}
767
768/* rtnl lock is held */
769static void enic_vlan_rx_register(struct net_device *netdev,
770 struct vlan_group *vlan_group)
771{
772 struct enic *enic = netdev_priv(netdev);
773 enic->vlan_group = vlan_group;
774}
775
776/* rtnl lock is held */
777static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
778{
779 struct enic *enic = netdev_priv(netdev);
780
781 spin_lock(&enic->devcmd_lock);
782 enic_add_vlan(enic, vid);
783 spin_unlock(&enic->devcmd_lock);
784}
785
786/* rtnl lock is held */
787static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
788{
789 struct enic *enic = netdev_priv(netdev);
790
791 spin_lock(&enic->devcmd_lock);
792 enic_del_vlan(enic, vid);
793 spin_unlock(&enic->devcmd_lock);
794}
795
796/* netif_tx_lock held, BHs disabled */
797static void enic_tx_timeout(struct net_device *netdev)
798{
799 struct enic *enic = netdev_priv(netdev);
800 schedule_work(&enic->reset);
801}
802
803static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
804{
805 struct enic *enic = vnic_dev_priv(rq->vdev);
806
807 if (!buf->os_buf)
808 return;
809
810 pci_unmap_single(enic->pdev, buf->dma_addr,
811 buf->len, PCI_DMA_FROMDEVICE);
812 dev_kfree_skb_any(buf->os_buf);
813}
814
815static inline struct sk_buff *enic_rq_alloc_skb(unsigned int size)
816{
817 struct sk_buff *skb;
818
819 skb = dev_alloc_skb(size + NET_IP_ALIGN);
820
821 if (skb)
822 skb_reserve(skb, NET_IP_ALIGN);
823
824 return skb;
825}
826
827static int enic_rq_alloc_buf(struct vnic_rq *rq)
828{
829 struct enic *enic = vnic_dev_priv(rq->vdev);
830 struct sk_buff *skb;
831 unsigned int len = enic->netdev->mtu + ETH_HLEN;
832 unsigned int os_buf_index = 0;
833 dma_addr_t dma_addr;
834
835 skb = enic_rq_alloc_skb(len);
836 if (!skb)
837 return -ENOMEM;
838
839 dma_addr = pci_map_single(enic->pdev, skb->data,
840 len, PCI_DMA_FROMDEVICE);
841
842 enic_queue_rq_desc(rq, skb, os_buf_index,
843 dma_addr, len);
844
845 return 0;
846}
847
848static int enic_get_skb_header(struct sk_buff *skb, void **iphdr,
849 void **tcph, u64 *hdr_flags, void *priv)
850{
851 struct cq_enet_rq_desc *cq_desc = priv;
852 unsigned int ip_len;
853 struct iphdr *iph;
854
855 u8 type, color, eop, sop, ingress_port, vlan_stripped;
856 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
857 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
858 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
859 u8 packet_error;
860 u16 q_number, completed_index, bytes_written, vlan, checksum;
861 u32 rss_hash;
862
863 cq_enet_rq_desc_dec(cq_desc,
864 &type, &color, &q_number, &completed_index,
865 &ingress_port, &fcoe, &eop, &sop, &rss_type,
866 &csum_not_calc, &rss_hash, &bytes_written,
867 &packet_error, &vlan_stripped, &vlan, &checksum,
868 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
869 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
870 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
871 &fcs_ok);
872
873 if (!(ipv4 && tcp && !ipv4_fragment))
874 return -1;
875
876 skb_reset_network_header(skb);
877 iph = ip_hdr(skb);
878
879 ip_len = ip_hdrlen(skb);
880 skb_set_transport_header(skb, ip_len);
881
882 /* check if ip header and tcp header are complete */
883 if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb))
884 return -1;
885
886 *hdr_flags = LRO_IPV4 | LRO_TCP;
887 *tcph = tcp_hdr(skb);
888 *iphdr = iph;
889
890 return 0;
891}
892
893static void enic_rq_indicate_buf(struct vnic_rq *rq,
894 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
895 int skipped, void *opaque)
896{
897 struct enic *enic = vnic_dev_priv(rq->vdev);
898 struct sk_buff *skb;
899
900 u8 type, color, eop, sop, ingress_port, vlan_stripped;
901 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
902 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
903 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
904 u8 packet_error;
905 u16 q_number, completed_index, bytes_written, vlan, checksum;
906 u32 rss_hash;
907
908 if (skipped)
909 return;
910
911 skb = buf->os_buf;
912 prefetch(skb->data - NET_IP_ALIGN);
913 pci_unmap_single(enic->pdev, buf->dma_addr,
914 buf->len, PCI_DMA_FROMDEVICE);
915
916 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
917 &type, &color, &q_number, &completed_index,
918 &ingress_port, &fcoe, &eop, &sop, &rss_type,
919 &csum_not_calc, &rss_hash, &bytes_written,
920 &packet_error, &vlan_stripped, &vlan, &checksum,
921 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
922 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
923 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
924 &fcs_ok);
925
926 if (packet_error) {
927
928 if (bytes_written > 0 && !fcs_ok) {
929 if (net_ratelimit())
930 printk(KERN_ERR PFX
931 "%s: packet error: bad FCS\n",
932 enic->netdev->name);
933 }
934
935 dev_kfree_skb_any(skb);
936
937 return;
938 }
939
940 if (eop && bytes_written > 0) {
941
942 /* Good receive
943 */
944
945 skb_put(skb, bytes_written);
946 skb->protocol = eth_type_trans(skb, enic->netdev);
947
948 if (enic->csum_rx_enabled && !csum_not_calc) {
949 skb->csum = htons(checksum);
950 skb->ip_summed = CHECKSUM_COMPLETE;
951 }
952
953 skb->dev = enic->netdev;
01f2e4ea
SF
954
955 if (enic->vlan_group && vlan_stripped) {
956
d9c3c57f 957 if (ENIC_SETTING(enic, LRO) && ipv4)
01f2e4ea
SF
958 lro_vlan_hwaccel_receive_skb(&enic->lro_mgr,
959 skb, enic->vlan_group,
960 vlan, cq_desc);
961 else
962 vlan_hwaccel_receive_skb(skb,
963 enic->vlan_group, vlan);
964
965 } else {
966
d9c3c57f 967 if (ENIC_SETTING(enic, LRO) && ipv4)
01f2e4ea
SF
968 lro_receive_skb(&enic->lro_mgr, skb, cq_desc);
969 else
970 netif_receive_skb(skb);
971
972 }
973
974 } else {
975
976 /* Buffer overflow
977 */
978
979 dev_kfree_skb_any(skb);
980 }
981}
982
983static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
984 u8 type, u16 q_number, u16 completed_index, void *opaque)
985{
986 struct enic *enic = vnic_dev_priv(vdev);
987
988 vnic_rq_service(&enic->rq[q_number], cq_desc,
989 completed_index, VNIC_RQ_RETURN_DESC,
990 enic_rq_indicate_buf, opaque);
991
992 return 0;
993}
994
995static void enic_rq_drop_buf(struct vnic_rq *rq,
996 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
997 int skipped, void *opaque)
998{
999 struct enic *enic = vnic_dev_priv(rq->vdev);
1000 struct sk_buff *skb = buf->os_buf;
1001
1002 if (skipped)
1003 return;
1004
1005 pci_unmap_single(enic->pdev, buf->dma_addr,
1006 buf->len, PCI_DMA_FROMDEVICE);
1007
1008 dev_kfree_skb_any(skb);
1009}
1010
1011static int enic_rq_service_drop(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1012 u8 type, u16 q_number, u16 completed_index, void *opaque)
1013{
1014 struct enic *enic = vnic_dev_priv(vdev);
1015
1016 vnic_rq_service(&enic->rq[q_number], cq_desc,
1017 completed_index, VNIC_RQ_RETURN_DESC,
1018 enic_rq_drop_buf, opaque);
1019
1020 return 0;
1021}
1022
1023static int enic_poll(struct napi_struct *napi, int budget)
1024{
1025 struct enic *enic = container_of(napi, struct enic, napi);
1026 struct net_device *netdev = enic->netdev;
1027 unsigned int rq_work_to_do = budget;
1028 unsigned int wq_work_to_do = -1; /* no limit */
1029 unsigned int work_done, rq_work_done, wq_work_done;
1030
1031 /* Service RQ (first) and WQ
1032 */
1033
1034 rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1035 rq_work_to_do, enic_rq_service, NULL);
1036
1037 wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1038 wq_work_to_do, enic_wq_service, NULL);
1039
1040 /* Accumulate intr event credits for this polling
1041 * cycle. An intr event is the completion of a
1042 * a WQ or RQ packet.
1043 */
1044
1045 work_done = rq_work_done + wq_work_done;
1046
1047 if (work_done > 0)
1048 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1049 work_done,
1050 0 /* don't unmask intr */,
1051 0 /* don't reset intr timer */);
1052
1053 if (rq_work_done > 0) {
1054
1055 /* Replenish RQ
1056 */
1057
1058 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1059
1060 } else {
1061
1062 /* If no work done, flush all LROs and exit polling
1063 */
1064
1065 if (ENIC_SETTING(enic, LRO))
1066 lro_flush_all(&enic->lro_mgr);
1067
1068 netif_rx_complete(netdev, napi);
1069 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1070 }
1071
1072 return rq_work_done;
1073}
1074
1075static int enic_poll_msix(struct napi_struct *napi, int budget)
1076{
1077 struct enic *enic = container_of(napi, struct enic, napi);
1078 struct net_device *netdev = enic->netdev;
1079 unsigned int work_to_do = budget;
1080 unsigned int work_done;
1081
1082 /* Service RQ
1083 */
1084
1085 work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1086 work_to_do, enic_rq_service, NULL);
1087
1088 if (work_done > 0) {
1089
1090 /* Replenish RQ
1091 */
1092
1093 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1094
1095 /* Accumulate intr event credits for this polling
1096 * cycle. An intr event is the completion of a
1097 * a WQ or RQ packet.
1098 */
1099
1100 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1101 work_done,
1102 0 /* don't unmask intr */,
1103 0 /* don't reset intr timer */);
1104 } else {
1105
1106 /* If no work done, flush all LROs and exit polling
1107 */
1108
1109 if (ENIC_SETTING(enic, LRO))
1110 lro_flush_all(&enic->lro_mgr);
1111
1112 netif_rx_complete(netdev, napi);
1113 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1114 }
1115
1116 return work_done;
1117}
1118
1119static void enic_notify_timer(unsigned long data)
1120{
1121 struct enic *enic = (struct enic *)data;
1122
1123 enic_notify_check(enic);
1124
25f0a061
SF
1125 mod_timer(&enic->notify_timer,
1126 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
01f2e4ea
SF
1127}
1128
1129static void enic_free_intr(struct enic *enic)
1130{
1131 struct net_device *netdev = enic->netdev;
1132 unsigned int i;
1133
1134 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1135 case VNIC_DEV_INTR_MODE_INTX:
01f2e4ea
SF
1136 free_irq(enic->pdev->irq, netdev);
1137 break;
8f4d248c
SF
1138 case VNIC_DEV_INTR_MODE_MSI:
1139 free_irq(enic->pdev->irq, enic);
1140 break;
01f2e4ea
SF
1141 case VNIC_DEV_INTR_MODE_MSIX:
1142 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1143 if (enic->msix[i].requested)
1144 free_irq(enic->msix_entry[i].vector,
1145 enic->msix[i].devid);
1146 break;
1147 default:
1148 break;
1149 }
1150}
1151
1152static int enic_request_intr(struct enic *enic)
1153{
1154 struct net_device *netdev = enic->netdev;
1155 unsigned int i;
1156 int err = 0;
1157
1158 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1159
1160 case VNIC_DEV_INTR_MODE_INTX:
1161
1162 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1163 IRQF_SHARED, netdev->name, netdev);
1164 break;
1165
1166 case VNIC_DEV_INTR_MODE_MSI:
1167
1168 err = request_irq(enic->pdev->irq, enic_isr_msi,
1169 0, netdev->name, enic);
1170 break;
1171
1172 case VNIC_DEV_INTR_MODE_MSIX:
1173
1174 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
8f4d248c 1175 "%.11s-rx-0", netdev->name);
01f2e4ea
SF
1176 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1177 enic->msix[ENIC_MSIX_RQ].devid = enic;
1178
1179 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
8f4d248c 1180 "%.11s-tx-0", netdev->name);
01f2e4ea
SF
1181 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1182 enic->msix[ENIC_MSIX_WQ].devid = enic;
1183
1184 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1185 "%.11s-err", netdev->name);
1186 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1187 enic->msix[ENIC_MSIX_ERR].devid = enic;
1188
1189 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1190 "%.11s-notify", netdev->name);
1191 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1192 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1193
1194 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1195 err = request_irq(enic->msix_entry[i].vector,
1196 enic->msix[i].isr, 0,
1197 enic->msix[i].devname,
1198 enic->msix[i].devid);
1199 if (err) {
1200 enic_free_intr(enic);
1201 break;
1202 }
1203 enic->msix[i].requested = 1;
1204 }
1205
1206 break;
1207
1208 default:
1209 break;
1210 }
1211
1212 return err;
1213}
1214
1215static int enic_notify_set(struct enic *enic)
1216{
1217 int err;
1218
1219 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1220 case VNIC_DEV_INTR_MODE_INTX:
1221 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1222 break;
1223 case VNIC_DEV_INTR_MODE_MSIX:
1224 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1225 break;
1226 default:
1227 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1228 break;
1229 }
1230
1231 return err;
1232}
1233
1234static void enic_notify_timer_start(struct enic *enic)
1235{
1236 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1237 case VNIC_DEV_INTR_MODE_MSI:
1238 mod_timer(&enic->notify_timer, jiffies);
1239 break;
1240 default:
1241 /* Using intr for notification for INTx/MSI-X */
1242 break;
1243 };
1244}
1245
1246/* rtnl lock is held, process context */
1247static int enic_open(struct net_device *netdev)
1248{
1249 struct enic *enic = netdev_priv(netdev);
1250 unsigned int i;
1251 int err;
1252
4b75a442
SF
1253 err = enic_request_intr(enic);
1254 if (err) {
1255 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1256 netdev->name);
1257 return err;
1258 }
1259
1260 err = enic_notify_set(enic);
1261 if (err) {
1262 printk(KERN_ERR PFX
1263 "%s: Failed to alloc notify buffer, aborting.\n",
1264 netdev->name);
1265 goto err_out_free_intr;
1266 }
1267
01f2e4ea
SF
1268 for (i = 0; i < enic->rq_count; i++) {
1269 err = vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1270 if (err) {
1271 printk(KERN_ERR PFX
1272 "%s: Unable to alloc receive buffers.\n",
1273 netdev->name);
4b75a442 1274 goto err_out_notify_unset;
01f2e4ea
SF
1275 }
1276 }
1277
1278 for (i = 0; i < enic->wq_count; i++)
1279 vnic_wq_enable(&enic->wq[i]);
1280 for (i = 0; i < enic->rq_count; i++)
1281 vnic_rq_enable(&enic->rq[i]);
1282
1283 enic_add_station_addr(enic);
1284 enic_set_multicast_list(netdev);
1285
1286 netif_wake_queue(netdev);
1287 napi_enable(&enic->napi);
1288 vnic_dev_enable(enic->vdev);
1289
1290 for (i = 0; i < enic->intr_count; i++)
1291 vnic_intr_unmask(&enic->intr[i]);
1292
1293 enic_notify_timer_start(enic);
1294
1295 return 0;
4b75a442
SF
1296
1297err_out_notify_unset:
1298 vnic_dev_notify_unset(enic->vdev);
1299err_out_free_intr:
1300 enic_free_intr(enic);
1301
1302 return err;
01f2e4ea
SF
1303}
1304
1305/* rtnl lock is held, process context */
1306static int enic_stop(struct net_device *netdev)
1307{
1308 struct enic *enic = netdev_priv(netdev);
1309 unsigned int i;
1310 int err;
1311
1312 del_timer_sync(&enic->notify_timer);
1313
1314 vnic_dev_disable(enic->vdev);
1315 napi_disable(&enic->napi);
1316 netif_stop_queue(netdev);
1317
1318 for (i = 0; i < enic->intr_count; i++)
1319 vnic_intr_mask(&enic->intr[i]);
1320
1321 for (i = 0; i < enic->wq_count; i++) {
1322 err = vnic_wq_disable(&enic->wq[i]);
1323 if (err)
1324 return err;
1325 }
1326 for (i = 0; i < enic->rq_count; i++) {
1327 err = vnic_rq_disable(&enic->rq[i]);
1328 if (err)
1329 return err;
1330 }
1331
4b75a442
SF
1332 vnic_dev_notify_unset(enic->vdev);
1333 enic_free_intr(enic);
1334
01f2e4ea
SF
1335 (void)vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1336 -1, enic_rq_service_drop, NULL);
1337 (void)vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1338 -1, enic_wq_service, NULL);
1339
1340 for (i = 0; i < enic->wq_count; i++)
1341 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1342 for (i = 0; i < enic->rq_count; i++)
1343 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1344 for (i = 0; i < enic->cq_count; i++)
1345 vnic_cq_clean(&enic->cq[i]);
1346 for (i = 0; i < enic->intr_count; i++)
1347 vnic_intr_clean(&enic->intr[i]);
1348
1349 return 0;
1350}
1351
1352static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1353{
1354 struct enic *enic = netdev_priv(netdev);
1355 int running = netif_running(netdev);
1356
25f0a061
SF
1357 if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1358 return -EINVAL;
1359
01f2e4ea
SF
1360 if (running)
1361 enic_stop(netdev);
1362
01f2e4ea
SF
1363 netdev->mtu = new_mtu;
1364
1365 if (netdev->mtu > enic->port_mtu)
1366 printk(KERN_WARNING PFX
1367 "%s: interface MTU (%d) set higher "
1368 "than port MTU (%d)\n",
1369 netdev->name, netdev->mtu, enic->port_mtu);
1370
1371 if (running)
1372 enic_open(netdev);
1373
1374 return 0;
1375}
1376
1377#ifdef CONFIG_NET_POLL_CONTROLLER
1378static void enic_poll_controller(struct net_device *netdev)
1379{
1380 struct enic *enic = netdev_priv(netdev);
1381 struct vnic_dev *vdev = enic->vdev;
1382
1383 switch (vnic_dev_get_intr_mode(vdev)) {
1384 case VNIC_DEV_INTR_MODE_MSIX:
1385 enic_isr_msix_rq(enic->pdev->irq, enic);
1386 enic_isr_msix_wq(enic->pdev->irq, enic);
1387 break;
1388 case VNIC_DEV_INTR_MODE_MSI:
1389 enic_isr_msi(enic->pdev->irq, enic);
1390 break;
1391 case VNIC_DEV_INTR_MODE_INTX:
1392 enic_isr_legacy(enic->pdev->irq, netdev);
1393 break;
1394 default:
1395 break;
1396 }
1397}
1398#endif
1399
1400static int enic_dev_wait(struct vnic_dev *vdev,
1401 int (*start)(struct vnic_dev *, int),
1402 int (*finished)(struct vnic_dev *, int *),
1403 int arg)
1404{
1405 unsigned long time;
1406 int done;
1407 int err;
1408
1409 BUG_ON(in_interrupt());
1410
1411 err = start(vdev, arg);
1412 if (err)
1413 return err;
1414
1415 /* Wait for func to complete...2 seconds max
1416 */
1417
1418 time = jiffies + (HZ * 2);
1419 do {
1420
1421 err = finished(vdev, &done);
1422 if (err)
1423 return err;
1424
1425 if (done)
1426 return 0;
1427
1428 schedule_timeout_uninterruptible(HZ / 10);
1429
1430 } while (time_after(time, jiffies));
1431
1432 return -ETIMEDOUT;
1433}
1434
1435static int enic_dev_open(struct enic *enic)
1436{
1437 int err;
1438
1439 err = enic_dev_wait(enic->vdev, vnic_dev_open,
1440 vnic_dev_open_done, 0);
1441 if (err)
1442 printk(KERN_ERR PFX
1443 "vNIC device open failed, err %d.\n", err);
1444
1445 return err;
1446}
1447
1448static int enic_dev_soft_reset(struct enic *enic)
1449{
1450 int err;
1451
1452 err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1453 vnic_dev_soft_reset_done, 0);
1454 if (err)
1455 printk(KERN_ERR PFX
1456 "vNIC soft reset failed, err %d.\n", err);
1457
1458 return err;
1459}
1460
1461static void enic_reset(struct work_struct *work)
1462{
1463 struct enic *enic = container_of(work, struct enic, reset);
1464
1465 if (!netif_running(enic->netdev))
1466 return;
1467
1468 rtnl_lock();
1469
1470 spin_lock(&enic->devcmd_lock);
1471 vnic_dev_hang_notify(enic->vdev);
1472 spin_unlock(&enic->devcmd_lock);
1473
1474 enic_stop(enic->netdev);
1475 enic_dev_soft_reset(enic);
1476 enic_reset_mcaddrs(enic);
1477 enic_init_vnic_resources(enic);
1478 enic_open(enic->netdev);
1479
1480 rtnl_unlock();
1481}
1482
1483static int enic_set_intr_mode(struct enic *enic)
1484{
1485 unsigned int n = ARRAY_SIZE(enic->rq);
1486 unsigned int m = ARRAY_SIZE(enic->wq);
1487 unsigned int i;
1488
1489 /* Set interrupt mode (INTx, MSI, MSI-X) depending
1490 * system capabilities.
1491 *
1492 * Try MSI-X first
1493 *
1494 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1495 * (the second to last INTR is used for WQ/RQ errors)
1496 * (the last INTR is used for notifications)
1497 */
1498
1499 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1500 for (i = 0; i < n + m + 2; i++)
1501 enic->msix_entry[i].entry = i;
1502
1503 if (enic->config.intr_mode < 1 &&
1504 enic->rq_count >= n &&
1505 enic->wq_count >= m &&
1506 enic->cq_count >= n + m &&
1507 enic->intr_count >= n + m + 2 &&
1508 !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1509
1510 enic->rq_count = n;
1511 enic->wq_count = m;
1512 enic->cq_count = n + m;
1513 enic->intr_count = n + m + 2;
1514
1515 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1516
1517 return 0;
1518 }
1519
1520 /* Next try MSI
1521 *
1522 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1523 */
1524
1525 if (enic->config.intr_mode < 2 &&
1526 enic->rq_count >= 1 &&
1527 enic->wq_count >= 1 &&
1528 enic->cq_count >= 2 &&
1529 enic->intr_count >= 1 &&
1530 !pci_enable_msi(enic->pdev)) {
1531
1532 enic->rq_count = 1;
1533 enic->wq_count = 1;
1534 enic->cq_count = 2;
1535 enic->intr_count = 1;
1536
1537 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1538
1539 return 0;
1540 }
1541
1542 /* Next try INTx
1543 *
1544 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1545 * (the first INTR is used for WQ/RQ)
1546 * (the second INTR is used for WQ/RQ errors)
1547 * (the last INTR is used for notifications)
1548 */
1549
1550 if (enic->config.intr_mode < 3 &&
1551 enic->rq_count >= 1 &&
1552 enic->wq_count >= 1 &&
1553 enic->cq_count >= 2 &&
1554 enic->intr_count >= 3) {
1555
1556 enic->rq_count = 1;
1557 enic->wq_count = 1;
1558 enic->cq_count = 2;
1559 enic->intr_count = 3;
1560
1561 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1562
1563 return 0;
1564 }
1565
1566 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1567
1568 return -EINVAL;
1569}
1570
1571static void enic_clear_intr_mode(struct enic *enic)
1572{
1573 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1574 case VNIC_DEV_INTR_MODE_MSIX:
1575 pci_disable_msix(enic->pdev);
1576 break;
1577 case VNIC_DEV_INTR_MODE_MSI:
1578 pci_disable_msi(enic->pdev);
1579 break;
1580 default:
1581 break;
1582 }
1583
1584 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1585}
1586
1587static void enic_iounmap(struct enic *enic)
1588{
1589 if (enic->bar0.vaddr)
1590 iounmap(enic->bar0.vaddr);
1591}
1592
afe29f7a
SH
1593static const struct net_device_ops enic_netdev_ops = {
1594 .ndo_open = enic_open,
1595 .ndo_stop = enic_stop,
1596 .ndo_get_stats = enic_get_stats,
1597 .ndo_validate_addr = eth_validate_addr,
1598 .ndo_set_multicast_list = enic_set_multicast_list,
1599 .ndo_change_mtu = enic_change_mtu,
1600 .ndo_vlan_rx_register = enic_vlan_rx_register,
1601 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
1602 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
1603 .ndo_tx_timeout = enic_tx_timeout,
1604#ifdef CONFIG_NET_POLL_CONTROLLER
1605 .ndo_poll_controller = enic_poll_controller,
1606#endif
1607};
1608
01f2e4ea
SF
1609static int __devinit enic_probe(struct pci_dev *pdev,
1610 const struct pci_device_id *ent)
1611{
1612 struct net_device *netdev;
1613 struct enic *enic;
1614 int using_dac = 0;
1615 unsigned int i;
1616 int err;
1617
1618 const u8 rss_default_cpu = 0;
1619 const u8 rss_hash_type = 0;
1620 const u8 rss_hash_bits = 0;
1621 const u8 rss_base_cpu = 0;
1622 const u8 rss_enable = 0;
1623 const u8 tso_ipid_split_en = 0;
1624 const u8 ig_vlan_strip_en = 1;
1625
1626 /* Allocate net device structure and initialize. Private
1627 * instance data is initialized to zero.
1628 */
1629
1630 netdev = alloc_etherdev(sizeof(struct enic));
1631 if (!netdev) {
1632 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
1633 return -ENOMEM;
1634 }
1635
01f2e4ea
SF
1636 pci_set_drvdata(pdev, netdev);
1637
1638 SET_NETDEV_DEV(netdev, &pdev->dev);
1639
1640 enic = netdev_priv(netdev);
1641 enic->netdev = netdev;
1642 enic->pdev = pdev;
1643
1644 /* Setup PCI resources
1645 */
1646
1647 err = pci_enable_device(pdev);
1648 if (err) {
1649 printk(KERN_ERR PFX
4b75a442 1650 "Cannot enable PCI device, aborting.\n");
01f2e4ea
SF
1651 goto err_out_free_netdev;
1652 }
1653
1654 err = pci_request_regions(pdev, DRV_NAME);
1655 if (err) {
1656 printk(KERN_ERR PFX
4b75a442 1657 "Cannot request PCI regions, aborting.\n");
01f2e4ea
SF
1658 goto err_out_disable_device;
1659 }
1660
1661 pci_set_master(pdev);
1662
1663 /* Query PCI controller on system for DMA addressing
1664 * limitation for the device. Try 40-bit first, and
1665 * fail to 32-bit.
1666 */
1667
1668 err = pci_set_dma_mask(pdev, DMA_40BIT_MASK);
1669 if (err) {
1670 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1671 if (err) {
1672 printk(KERN_ERR PFX
4b75a442 1673 "No usable DMA configuration, aborting.\n");
01f2e4ea
SF
1674 goto err_out_release_regions;
1675 }
1676 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
1677 if (err) {
1678 printk(KERN_ERR PFX
4b75a442
SF
1679 "Unable to obtain 32-bit DMA "
1680 "for consistent allocations, aborting.\n");
01f2e4ea
SF
1681 goto err_out_release_regions;
1682 }
1683 } else {
1684 err = pci_set_consistent_dma_mask(pdev, DMA_40BIT_MASK);
1685 if (err) {
1686 printk(KERN_ERR PFX
4b75a442
SF
1687 "Unable to obtain 40-bit DMA "
1688 "for consistent allocations, aborting.\n");
01f2e4ea
SF
1689 goto err_out_release_regions;
1690 }
1691 using_dac = 1;
1692 }
1693
1694 /* Map vNIC resources from BAR0
1695 */
1696
1697 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
1698 printk(KERN_ERR PFX
4b75a442 1699 "BAR0 not memory-map'able, aborting.\n");
01f2e4ea
SF
1700 err = -ENODEV;
1701 goto err_out_release_regions;
1702 }
1703
1704 enic->bar0.vaddr = pci_iomap(pdev, 0, enic->bar0.len);
1705 enic->bar0.bus_addr = pci_resource_start(pdev, 0);
1706 enic->bar0.len = pci_resource_len(pdev, 0);
1707
1708 if (!enic->bar0.vaddr) {
1709 printk(KERN_ERR PFX
4b75a442 1710 "Cannot memory-map BAR0 res hdr, aborting.\n");
01f2e4ea
SF
1711 err = -ENODEV;
1712 goto err_out_release_regions;
1713 }
1714
1715 /* Register vNIC device
1716 */
1717
1718 enic->vdev = vnic_dev_register(NULL, enic, pdev, &enic->bar0);
1719 if (!enic->vdev) {
1720 printk(KERN_ERR PFX
4b75a442 1721 "vNIC registration failed, aborting.\n");
01f2e4ea
SF
1722 err = -ENODEV;
1723 goto err_out_iounmap;
1724 }
1725
1726 /* Issue device open to get device in known state
1727 */
1728
1729 err = enic_dev_open(enic);
1730 if (err) {
1731 printk(KERN_ERR PFX
4b75a442 1732 "vNIC dev open failed, aborting.\n");
01f2e4ea
SF
1733 goto err_out_vnic_unregister;
1734 }
1735
1736 /* Issue device init to initialize the vnic-to-switch link.
1737 * We'll start with carrier off and wait for link UP
1738 * notification later to turn on carrier. We don't need
1739 * to wait here for the vnic-to-switch link initialization
1740 * to complete; link UP notification is the indication that
1741 * the process is complete.
1742 */
1743
1744 netif_carrier_off(netdev);
1745
1746 err = vnic_dev_init(enic->vdev, 0);
1747 if (err) {
1748 printk(KERN_ERR PFX
4b75a442 1749 "vNIC dev init failed, aborting.\n");
01f2e4ea
SF
1750 goto err_out_dev_close;
1751 }
1752
1753 /* Get vNIC configuration
1754 */
1755
1756 err = enic_get_vnic_config(enic);
1757 if (err) {
1758 printk(KERN_ERR PFX
4b75a442 1759 "Get vNIC configuration failed, aborting.\n");
01f2e4ea
SF
1760 goto err_out_dev_close;
1761 }
1762
1763 /* Get available resource counts
1764 */
1765
1766 enic_get_res_counts(enic);
1767
1768 /* Set interrupt mode based on resource counts and system
1769 * capabilities
1770 */
1771
1772 err = enic_set_intr_mode(enic);
1773 if (err) {
1774 printk(KERN_ERR PFX
4b75a442 1775 "Failed to set intr mode, aborting.\n");
01f2e4ea
SF
1776 goto err_out_dev_close;
1777 }
1778
1779 /* Allocate and configure vNIC resources
1780 */
1781
1782 err = enic_alloc_vnic_resources(enic);
1783 if (err) {
1784 printk(KERN_ERR PFX
4b75a442 1785 "Failed to alloc vNIC resources, aborting.\n");
01f2e4ea
SF
1786 goto err_out_free_vnic_resources;
1787 }
1788
1789 enic_init_vnic_resources(enic);
1790
1791 /* Enable VLAN tag stripping. RSS not enabled (yet).
1792 */
1793
1794 err = enic_set_nic_cfg(enic,
1795 rss_default_cpu, rss_hash_type,
1796 rss_hash_bits, rss_base_cpu,
1797 rss_enable, tso_ipid_split_en,
1798 ig_vlan_strip_en);
1799 if (err) {
1800 printk(KERN_ERR PFX
4b75a442 1801 "Failed to config nic, aborting.\n");
01f2e4ea
SF
1802 goto err_out_free_vnic_resources;
1803 }
1804
1805 /* Setup notification timer, HW reset task, and locks
1806 */
1807
1808 init_timer(&enic->notify_timer);
1809 enic->notify_timer.function = enic_notify_timer;
1810 enic->notify_timer.data = (unsigned long)enic;
1811
1812 INIT_WORK(&enic->reset, enic_reset);
1813
1814 for (i = 0; i < enic->wq_count; i++)
1815 spin_lock_init(&enic->wq_lock[i]);
1816
1817 spin_lock_init(&enic->devcmd_lock);
1818
1819 /* Register net device
1820 */
1821
1822 enic->port_mtu = enic->config.mtu;
1823 (void)enic_change_mtu(netdev, enic->port_mtu);
1824
1825 err = enic_set_mac_addr(netdev, enic->mac_addr);
1826 if (err) {
1827 printk(KERN_ERR PFX
4b75a442
SF
1828 "Invalid MAC address, aborting.\n");
1829 goto err_out_free_vnic_resources;
01f2e4ea
SF
1830 }
1831
afe29f7a 1832 netdev->netdev_ops = &enic_netdev_ops;
01f2e4ea 1833 netdev->hard_start_xmit = enic_hard_start_xmit;
01f2e4ea
SF
1834 netdev->watchdog_timeo = 2 * HZ;
1835 netdev->ethtool_ops = &enic_ethtool_ops;
01f2e4ea
SF
1836
1837 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1838 default:
1839 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
1840 break;
1841 case VNIC_DEV_INTR_MODE_MSIX:
1842 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
1843 break;
1844 }
1845
1846 netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
1847 if (ENIC_SETTING(enic, TXCSUM))
1848 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1849 if (ENIC_SETTING(enic, TSO))
1850 netdev->features |= NETIF_F_TSO |
1851 NETIF_F_TSO6 | NETIF_F_TSO_ECN;
1852 if (using_dac)
1853 netdev->features |= NETIF_F_HIGHDMA;
1854
1855
1856 enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
1857
1858 if (ENIC_SETTING(enic, LRO)) {
1859 enic->lro_mgr.max_aggr = ENIC_LRO_MAX_AGGR;
1860 enic->lro_mgr.max_desc = ENIC_LRO_MAX_DESC;
1861 enic->lro_mgr.lro_arr = enic->lro_desc;
1862 enic->lro_mgr.get_skb_header = enic_get_skb_header;
1863 enic->lro_mgr.features = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
1864 enic->lro_mgr.dev = netdev;
1865 enic->lro_mgr.ip_summed = CHECKSUM_COMPLETE;
1866 enic->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1867 }
1868
1869 err = register_netdev(netdev);
1870 if (err) {
1871 printk(KERN_ERR PFX
4b75a442
SF
1872 "Cannot register net device, aborting.\n");
1873 goto err_out_free_vnic_resources;
01f2e4ea
SF
1874 }
1875
1876 return 0;
1877
01f2e4ea
SF
1878err_out_free_vnic_resources:
1879 enic_free_vnic_resources(enic);
01f2e4ea
SF
1880err_out_dev_close:
1881 vnic_dev_close(enic->vdev);
1882err_out_vnic_unregister:
1883 enic_clear_intr_mode(enic);
1884 vnic_dev_unregister(enic->vdev);
1885err_out_iounmap:
1886 enic_iounmap(enic);
1887err_out_release_regions:
1888 pci_release_regions(pdev);
1889err_out_disable_device:
1890 pci_disable_device(pdev);
1891err_out_free_netdev:
1892 pci_set_drvdata(pdev, NULL);
1893 free_netdev(netdev);
1894
1895 return err;
1896}
1897
1898static void __devexit enic_remove(struct pci_dev *pdev)
1899{
1900 struct net_device *netdev = pci_get_drvdata(pdev);
1901
1902 if (netdev) {
1903 struct enic *enic = netdev_priv(netdev);
1904
1905 flush_scheduled_work();
1906 unregister_netdev(netdev);
01f2e4ea 1907 enic_free_vnic_resources(enic);
01f2e4ea
SF
1908 vnic_dev_close(enic->vdev);
1909 enic_clear_intr_mode(enic);
1910 vnic_dev_unregister(enic->vdev);
1911 enic_iounmap(enic);
1912 pci_release_regions(pdev);
1913 pci_disable_device(pdev);
1914 pci_set_drvdata(pdev, NULL);
1915 free_netdev(netdev);
1916 }
1917}
1918
1919static struct pci_driver enic_driver = {
1920 .name = DRV_NAME,
1921 .id_table = enic_id_table,
1922 .probe = enic_probe,
1923 .remove = __devexit_p(enic_remove),
1924};
1925
1926static int __init enic_init_module(void)
1927{
1928 printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
1929
1930 return pci_register_driver(&enic_driver);
1931}
1932
1933static void __exit enic_cleanup_module(void)
1934{
1935 pci_unregister_driver(&enic_driver);
1936}
1937
1938module_init(enic_init_module);
1939module_exit(enic_cleanup_module);