projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'for-linus' of git://neil.brown.name/md
[deliverable/linux.git] / drivers / net / enic / enic_main.c
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>
37 #include <net/ip6_checksum.h>
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)
47
48 /* Supported devices */
49 static struct pci_device_id enic_id_table[] = {
50 { PCI_VDEVICE(CISCO, 0x0043) },
51 { 0, } /* end of table */
52 };
53
54 MODULE_DESCRIPTION(DRV_DESCRIPTION);
55 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
56 MODULE_LICENSE("GPL");
57 MODULE_VERSION(DRV_VERSION);
58 MODULE_DEVICE_TABLE(pci, enic_id_table);
59
60 struct 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
70 static 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
84 static 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
108 static const unsigned int enic_n_tx_stats = ARRAY_SIZE(enic_tx_stats);
109 static const unsigned int enic_n_rx_stats = ARRAY_SIZE(enic_rx_stats);
110
111 static 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
134 static 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
152 static 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
170 static int enic_get_sset_count(struct net_device *netdev, int sset)
171 {
172 switch (sset) {
173 case ETH_SS_STATS:
174 return enic_n_tx_stats + enic_n_rx_stats;
175 default:
176 return -EOPNOTSUPP;
177 }
178 }
179
180 static 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
197 static 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
203 static int enic_set_rx_csum(struct net_device *netdev, u32 data)
204 {
205 struct enic *enic = netdev_priv(netdev);
206
207 if (data && !ENIC_SETTING(enic, RXCSUM))
208 return -EINVAL;
209
210 enic->csum_rx_enabled = !!data;
211
212 return 0;
213 }
214
215 static int enic_set_tx_csum(struct net_device *netdev, u32 data)
216 {
217 struct enic *enic = netdev_priv(netdev);
218
219 if (data && !ENIC_SETTING(enic, TXCSUM))
220 return -EINVAL;
221
222 if (data)
223 netdev->features |= NETIF_F_HW_CSUM;
224 else
225 netdev->features &= ~NETIF_F_HW_CSUM;
226
227 return 0;
228 }
229
230 static int enic_set_tso(struct net_device *netdev, u32 data)
231 {
232 struct enic *enic = netdev_priv(netdev);
233
234 if (data && !ENIC_SETTING(enic, TSO))
235 return -EINVAL;
236
237 if (data)
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
247 static u32 enic_get_msglevel(struct net_device *netdev)
248 {
249 struct enic *enic = netdev_priv(netdev);
250 return enic->msg_enable;
251 }
252
253 static 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
259 static 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,
266 .get_sset_count = enic_get_sset_count,
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 .get_flags = ethtool_op_get_flags,
277 .set_flags = ethtool_op_set_flags,
278 };
279
280 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
281 {
282 struct enic *enic = vnic_dev_priv(wq->vdev);
283
284 if (buf->sop)
285 pci_unmap_single(enic->pdev, buf->dma_addr,
286 buf->len, PCI_DMA_TODEVICE);
287 else
288 pci_unmap_page(enic->pdev, buf->dma_addr,
289 buf->len, PCI_DMA_TODEVICE);
290
291 if (buf->os_buf)
292 dev_kfree_skb_any(buf->os_buf);
293 }
294
295 static void enic_wq_free_buf(struct vnic_wq *wq,
296 struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
297 {
298 enic_free_wq_buf(wq, buf);
299 }
300
301 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
302 u8 type, u16 q_number, u16 completed_index, void *opaque)
303 {
304 struct enic *enic = vnic_dev_priv(vdev);
305
306 spin_lock(&enic->wq_lock[q_number]);
307
308 vnic_wq_service(&enic->wq[q_number], cq_desc,
309 completed_index, enic_wq_free_buf,
310 opaque);
311
312 if (netif_queue_stopped(enic->netdev) &&
313 vnic_wq_desc_avail(&enic->wq[q_number]) >= MAX_SKB_FRAGS + 1)
314 netif_wake_queue(enic->netdev);
315
316 spin_unlock(&enic->wq_lock[q_number]);
317
318 return 0;
319 }
320
321 static void enic_log_q_error(struct enic *enic)
322 {
323 unsigned int i;
324 u32 error_status;
325
326 for (i = 0; i < enic->wq_count; i++) {
327 error_status = vnic_wq_error_status(&enic->wq[i]);
328 if (error_status)
329 printk(KERN_ERR PFX "%s: WQ[%d] error_status %d\n",
330 enic->netdev->name, i, error_status);
331 }
332
333 for (i = 0; i < enic->rq_count; i++) {
334 error_status = vnic_rq_error_status(&enic->rq[i]);
335 if (error_status)
336 printk(KERN_ERR PFX "%s: RQ[%d] error_status %d\n",
337 enic->netdev->name, i, error_status);
338 }
339 }
340
341 static void enic_link_check(struct enic *enic)
342 {
343 int link_status = vnic_dev_link_status(enic->vdev);
344 int carrier_ok = netif_carrier_ok(enic->netdev);
345
346 if (link_status && !carrier_ok) {
347 printk(KERN_INFO PFX "%s: Link UP\n", enic->netdev->name);
348 netif_carrier_on(enic->netdev);
349 } else if (!link_status && carrier_ok) {
350 printk(KERN_INFO PFX "%s: Link DOWN\n", enic->netdev->name);
351 netif_carrier_off(enic->netdev);
352 }
353 }
354
355 static void enic_mtu_check(struct enic *enic)
356 {
357 u32 mtu = vnic_dev_mtu(enic->vdev);
358
359 if (mtu != enic->port_mtu) {
360 if (mtu < enic->netdev->mtu)
361 printk(KERN_WARNING PFX
362 "%s: interface MTU (%d) set higher "
363 "than switch port MTU (%d)\n",
364 enic->netdev->name, enic->netdev->mtu, mtu);
365 enic->port_mtu = mtu;
366 }
367 }
368
369 static void enic_msglvl_check(struct enic *enic)
370 {
371 u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
372
373 if (msg_enable != enic->msg_enable) {
374 printk(KERN_INFO PFX "%s: msg lvl changed from 0x%x to 0x%x\n",
375 enic->netdev->name, enic->msg_enable, msg_enable);
376 enic->msg_enable = msg_enable;
377 }
378 }
379
380 static void enic_notify_check(struct enic *enic)
381 {
382 enic_msglvl_check(enic);
383 enic_mtu_check(enic);
384 enic_link_check(enic);
385 }
386
387 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
388
389 static irqreturn_t enic_isr_legacy(int irq, void *data)
390 {
391 struct net_device *netdev = data;
392 struct enic *enic = netdev_priv(netdev);
393 u32 pba;
394
395 vnic_intr_mask(&enic->intr[ENIC_INTX_WQ_RQ]);
396
397 pba = vnic_intr_legacy_pba(enic->legacy_pba);
398 if (!pba) {
399 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
400 return IRQ_NONE; /* not our interrupt */
401 }
402
403 if (ENIC_TEST_INTR(pba, ENIC_INTX_NOTIFY)) {
404 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_NOTIFY]);
405 enic_notify_check(enic);
406 }
407
408 if (ENIC_TEST_INTR(pba, ENIC_INTX_ERR)) {
409 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_ERR]);
410 enic_log_q_error(enic);
411 /* schedule recovery from WQ/RQ error */
412 schedule_work(&enic->reset);
413 return IRQ_HANDLED;
414 }
415
416 if (ENIC_TEST_INTR(pba, ENIC_INTX_WQ_RQ)) {
417 if (napi_schedule_prep(&enic->napi))
418 __napi_schedule(&enic->napi);
419 } else {
420 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
421 }
422
423 return IRQ_HANDLED;
424 }
425
426 static irqreturn_t enic_isr_msi(int irq, void *data)
427 {
428 struct enic *enic = data;
429
430 /* With MSI, there is no sharing of interrupts, so this is
431 * our interrupt and there is no need to ack it. The device
432 * is not providing per-vector masking, so the OS will not
433 * write to PCI config space to mask/unmask the interrupt.
434 * We're using mask_on_assertion for MSI, so the device
435 * automatically masks the interrupt when the interrupt is
436 * generated. Later, when exiting polling, the interrupt
437 * will be unmasked (see enic_poll).
438 *
439 * Also, the device uses the same PCIe Traffic Class (TC)
440 * for Memory Write data and MSI, so there are no ordering
441 * issues; the MSI will always arrive at the Root Complex
442 * _after_ corresponding Memory Writes (i.e. descriptor
443 * writes).
444 */
445
446 napi_schedule(&enic->napi);
447
448 return IRQ_HANDLED;
449 }
450
451 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
452 {
453 struct enic *enic = data;
454
455 /* schedule NAPI polling for RQ cleanup */
456 napi_schedule(&enic->napi);
457
458 return IRQ_HANDLED;
459 }
460
461 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
462 {
463 struct enic *enic = data;
464 unsigned int wq_work_to_do = -1; /* no limit */
465 unsigned int wq_work_done;
466
467 wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
468 wq_work_to_do, enic_wq_service, NULL);
469
470 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_WQ],
471 wq_work_done,
472 1 /* unmask intr */,
473 1 /* reset intr timer */);
474
475 return IRQ_HANDLED;
476 }
477
478 static irqreturn_t enic_isr_msix_err(int irq, void *data)
479 {
480 struct enic *enic = data;
481
482 vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_ERR]);
483
484 enic_log_q_error(enic);
485
486 /* schedule recovery from WQ/RQ error */
487 schedule_work(&enic->reset);
488
489 return IRQ_HANDLED;
490 }
491
492 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
493 {
494 struct enic *enic = data;
495
496 vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_NOTIFY]);
497 enic_notify_check(enic);
498
499 return IRQ_HANDLED;
500 }
501
502 static inline void enic_queue_wq_skb_cont(struct enic *enic,
503 struct vnic_wq *wq, struct sk_buff *skb,
504 unsigned int len_left)
505 {
506 skb_frag_t *frag;
507
508 /* Queue additional data fragments */
509 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
510 len_left -= frag->size;
511 enic_queue_wq_desc_cont(wq, skb,
512 pci_map_page(enic->pdev, frag->page,
513 frag->page_offset, frag->size,
514 PCI_DMA_TODEVICE),
515 frag->size,
516 (len_left == 0)); /* EOP? */
517 }
518 }
519
520 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
521 struct vnic_wq *wq, struct sk_buff *skb,
522 int vlan_tag_insert, unsigned int vlan_tag)
523 {
524 unsigned int head_len = skb_headlen(skb);
525 unsigned int len_left = skb->len - head_len;
526 int eop = (len_left == 0);
527
528 /* Queue the main skb fragment */
529 enic_queue_wq_desc(wq, skb,
530 pci_map_single(enic->pdev, skb->data,
531 head_len, PCI_DMA_TODEVICE),
532 head_len,
533 vlan_tag_insert, vlan_tag,
534 eop);
535
536 if (!eop)
537 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
538 }
539
540 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
541 struct vnic_wq *wq, struct sk_buff *skb,
542 int vlan_tag_insert, unsigned int vlan_tag)
543 {
544 unsigned int head_len = skb_headlen(skb);
545 unsigned int len_left = skb->len - head_len;
546 unsigned int hdr_len = skb_transport_offset(skb);
547 unsigned int csum_offset = hdr_len + skb->csum_offset;
548 int eop = (len_left == 0);
549
550 /* Queue the main skb fragment */
551 enic_queue_wq_desc_csum_l4(wq, skb,
552 pci_map_single(enic->pdev, skb->data,
553 head_len, PCI_DMA_TODEVICE),
554 head_len,
555 csum_offset,
556 hdr_len,
557 vlan_tag_insert, vlan_tag,
558 eop);
559
560 if (!eop)
561 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
562 }
563
564 static inline void enic_queue_wq_skb_tso(struct enic *enic,
565 struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
566 int vlan_tag_insert, unsigned int vlan_tag)
567 {
568 unsigned int head_len = skb_headlen(skb);
569 unsigned int len_left = skb->len - head_len;
570 unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
571 int eop = (len_left == 0);
572
573 /* Preload TCP csum field with IP pseudo hdr calculated
574 * with IP length set to zero. HW will later add in length
575 * to each TCP segment resulting from the TSO.
576 */
577
578 if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
579 ip_hdr(skb)->check = 0;
580 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
581 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
582 } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
583 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
584 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
585 }
586
587 /* Queue the main skb fragment */
588 enic_queue_wq_desc_tso(wq, skb,
589 pci_map_single(enic->pdev, skb->data,
590 head_len, PCI_DMA_TODEVICE),
591 head_len,
592 mss, hdr_len,
593 vlan_tag_insert, vlan_tag,
594 eop);
595
596 if (!eop)
597 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
598 }
599
600 static inline void enic_queue_wq_skb(struct enic *enic,
601 struct vnic_wq *wq, struct sk_buff *skb)
602 {
603 unsigned int mss = skb_shinfo(skb)->gso_size;
604 unsigned int vlan_tag = 0;
605 int vlan_tag_insert = 0;
606
607 if (enic->vlan_group && vlan_tx_tag_present(skb)) {
608 /* VLAN tag from trunking driver */
609 vlan_tag_insert = 1;
610 vlan_tag = vlan_tx_tag_get(skb);
611 }
612
613 if (mss)
614 enic_queue_wq_skb_tso(enic, wq, skb, mss,
615 vlan_tag_insert, vlan_tag);
616 else if (skb->ip_summed == CHECKSUM_PARTIAL)
617 enic_queue_wq_skb_csum_l4(enic, wq, skb,
618 vlan_tag_insert, vlan_tag);
619 else
620 enic_queue_wq_skb_vlan(enic, wq, skb,
621 vlan_tag_insert, vlan_tag);
622 }
623
624 /* netif_tx_lock held, process context with BHs disabled, or BH */
625 static int enic_hard_start_xmit(struct sk_buff *skb, struct net_device *netdev)
626 {
627 struct enic *enic = netdev_priv(netdev);
628 struct vnic_wq *wq = &enic->wq[0];
629 unsigned long flags;
630
631 if (skb->len <= 0) {
632 dev_kfree_skb(skb);
633 return NETDEV_TX_OK;
634 }
635
636 /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
637 * which is very likely. In the off chance it's going to take
638 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
639 */
640
641 if (skb_shinfo(skb)->gso_size == 0 &&
642 skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
643 skb_linearize(skb)) {
644 dev_kfree_skb(skb);
645 return NETDEV_TX_OK;
646 }
647
648 spin_lock_irqsave(&enic->wq_lock[0], flags);
649
650 if (vnic_wq_desc_avail(wq) < skb_shinfo(skb)->nr_frags + 1) {
651 netif_stop_queue(netdev);
652 /* This is a hard error, log it */
653 printk(KERN_ERR PFX "%s: BUG! Tx ring full when "
654 "queue awake!\n", netdev->name);
655 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
656 return NETDEV_TX_BUSY;
657 }
658
659 enic_queue_wq_skb(enic, wq, skb);
660
661 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + 1)
662 netif_stop_queue(netdev);
663
664 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
665
666 return NETDEV_TX_OK;
667 }
668
669 /* dev_base_lock rwlock held, nominally process context */
670 static struct net_device_stats *enic_get_stats(struct net_device *netdev)
671 {
672 struct enic *enic = netdev_priv(netdev);
673 struct net_device_stats *net_stats = &netdev->stats;
674 struct vnic_stats *stats;
675
676 spin_lock(&enic->devcmd_lock);
677 vnic_dev_stats_dump(enic->vdev, &stats);
678 spin_unlock(&enic->devcmd_lock);
679
680 net_stats->tx_packets = stats->tx.tx_frames_ok;
681 net_stats->tx_bytes = stats->tx.tx_bytes_ok;
682 net_stats->tx_errors = stats->tx.tx_errors;
683 net_stats->tx_dropped = stats->tx.tx_drops;
684
685 net_stats->rx_packets = stats->rx.rx_frames_ok;
686 net_stats->rx_bytes = stats->rx.rx_bytes_ok;
687 net_stats->rx_errors = stats->rx.rx_errors;
688 net_stats->multicast = stats->rx.rx_multicast_frames_ok;
689 net_stats->rx_crc_errors = enic->rq_bad_fcs;
690 net_stats->rx_dropped = stats->rx.rx_no_bufs;
691
692 return net_stats;
693 }
694
695 static void enic_reset_mcaddrs(struct enic *enic)
696 {
697 enic->mc_count = 0;
698 }
699
700 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
701 {
702 if (!is_valid_ether_addr(addr))
703 return -EADDRNOTAVAIL;
704
705 memcpy(netdev->dev_addr, addr, netdev->addr_len);
706
707 return 0;
708 }
709
710 /* netif_tx_lock held, BHs disabled */
711 static void enic_set_multicast_list(struct net_device *netdev)
712 {
713 struct enic *enic = netdev_priv(netdev);
714 struct dev_mc_list *list = netdev->mc_list;
715 int directed = 1;
716 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
717 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
718 int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
719 int allmulti = (netdev->flags & IFF_ALLMULTI) ||
720 (netdev->mc_count > ENIC_MULTICAST_PERFECT_FILTERS);
721 u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
722 unsigned int mc_count = netdev->mc_count;
723 unsigned int i, j;
724
725 if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
726 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
727
728 spin_lock(&enic->devcmd_lock);
729
730 vnic_dev_packet_filter(enic->vdev, directed,
731 multicast, broadcast, promisc, allmulti);
732
733 /* Is there an easier way? Trying to minimize to
734 * calls to add/del multicast addrs. We keep the
735 * addrs from the last call in enic->mc_addr and
736 * look for changes to add/del.
737 */
738
739 for (i = 0; list && i < mc_count; i++) {
740 memcpy(mc_addr[i], list->dmi_addr, ETH_ALEN);
741 list = list->next;
742 }
743
744 for (i = 0; i < enic->mc_count; i++) {
745 for (j = 0; j < mc_count; j++)
746 if (compare_ether_addr(enic->mc_addr[i],
747 mc_addr[j]) == 0)
748 break;
749 if (j == mc_count)
750 enic_del_multicast_addr(enic, enic->mc_addr[i]);
751 }
752
753 for (i = 0; i < mc_count; i++) {
754 for (j = 0; j < enic->mc_count; j++)
755 if (compare_ether_addr(mc_addr[i],
756 enic->mc_addr[j]) == 0)
757 break;
758 if (j == enic->mc_count)
759 enic_add_multicast_addr(enic, mc_addr[i]);
760 }
761
762 /* Save the list to compare against next time
763 */
764
765 for (i = 0; i < mc_count; i++)
766 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
767
768 enic->mc_count = mc_count;
769
770 spin_unlock(&enic->devcmd_lock);
771 }
772
773 /* rtnl lock is held */
774 static void enic_vlan_rx_register(struct net_device *netdev,
775 struct vlan_group *vlan_group)
776 {
777 struct enic *enic = netdev_priv(netdev);
778 enic->vlan_group = vlan_group;
779 }
780
781 /* rtnl lock is held */
782 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
783 {
784 struct enic *enic = netdev_priv(netdev);
785
786 spin_lock(&enic->devcmd_lock);
787 enic_add_vlan(enic, vid);
788 spin_unlock(&enic->devcmd_lock);
789 }
790
791 /* rtnl lock is held */
792 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
793 {
794 struct enic *enic = netdev_priv(netdev);
795
796 spin_lock(&enic->devcmd_lock);
797 enic_del_vlan(enic, vid);
798 spin_unlock(&enic->devcmd_lock);
799 }
800
801 /* netif_tx_lock held, BHs disabled */
802 static void enic_tx_timeout(struct net_device *netdev)
803 {
804 struct enic *enic = netdev_priv(netdev);
805 schedule_work(&enic->reset);
806 }
807
808 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
809 {
810 struct enic *enic = vnic_dev_priv(rq->vdev);
811
812 if (!buf->os_buf)
813 return;
814
815 pci_unmap_single(enic->pdev, buf->dma_addr,
816 buf->len, PCI_DMA_FROMDEVICE);
817 dev_kfree_skb_any(buf->os_buf);
818 }
819
820 static inline struct sk_buff *enic_rq_alloc_skb(unsigned int size)
821 {
822 struct sk_buff *skb;
823
824 skb = dev_alloc_skb(size + NET_IP_ALIGN);
825
826 if (skb)
827 skb_reserve(skb, NET_IP_ALIGN);
828
829 return skb;
830 }
831
832 static int enic_rq_alloc_buf(struct vnic_rq *rq)
833 {
834 struct enic *enic = vnic_dev_priv(rq->vdev);
835 struct sk_buff *skb;
836 unsigned int len = enic->netdev->mtu + ETH_HLEN;
837 unsigned int os_buf_index = 0;
838 dma_addr_t dma_addr;
839
840 skb = enic_rq_alloc_skb(len);
841 if (!skb)
842 return -ENOMEM;
843
844 dma_addr = pci_map_single(enic->pdev, skb->data,
845 len, PCI_DMA_FROMDEVICE);
846
847 enic_queue_rq_desc(rq, skb, os_buf_index,
848 dma_addr, len);
849
850 return 0;
851 }
852
853 static int enic_get_skb_header(struct sk_buff *skb, void **iphdr,
854 void **tcph, u64 *hdr_flags, void *priv)
855 {
856 struct cq_enet_rq_desc *cq_desc = priv;
857 unsigned int ip_len;
858 struct iphdr *iph;
859
860 u8 type, color, eop, sop, ingress_port, vlan_stripped;
861 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
862 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
863 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
864 u8 packet_error;
865 u16 q_number, completed_index, bytes_written, vlan, checksum;
866 u32 rss_hash;
867
868 cq_enet_rq_desc_dec(cq_desc,
869 &type, &color, &q_number, &completed_index,
870 &ingress_port, &fcoe, &eop, &sop, &rss_type,
871 &csum_not_calc, &rss_hash, &bytes_written,
872 &packet_error, &vlan_stripped, &vlan, &checksum,
873 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
874 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
875 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
876 &fcs_ok);
877
878 if (!(ipv4 && tcp && !ipv4_fragment))
879 return -1;
880
881 skb_reset_network_header(skb);
882 iph = ip_hdr(skb);
883
884 ip_len = ip_hdrlen(skb);
885 skb_set_transport_header(skb, ip_len);
886
887 /* check if ip header and tcp header are complete */
888 if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb))
889 return -1;
890
891 *hdr_flags = LRO_IPV4 | LRO_TCP;
892 *tcph = tcp_hdr(skb);
893 *iphdr = iph;
894
895 return 0;
896 }
897
898 static void enic_rq_indicate_buf(struct vnic_rq *rq,
899 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
900 int skipped, void *opaque)
901 {
902 struct enic *enic = vnic_dev_priv(rq->vdev);
903 struct net_device *netdev = enic->netdev;
904 struct sk_buff *skb;
905
906 u8 type, color, eop, sop, ingress_port, vlan_stripped;
907 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
908 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
909 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
910 u8 packet_error;
911 u16 q_number, completed_index, bytes_written, vlan, checksum;
912 u32 rss_hash;
913
914 if (skipped)
915 return;
916
917 skb = buf->os_buf;
918 prefetch(skb->data - NET_IP_ALIGN);
919 pci_unmap_single(enic->pdev, buf->dma_addr,
920 buf->len, PCI_DMA_FROMDEVICE);
921
922 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
923 &type, &color, &q_number, &completed_index,
924 &ingress_port, &fcoe, &eop, &sop, &rss_type,
925 &csum_not_calc, &rss_hash, &bytes_written,
926 &packet_error, &vlan_stripped, &vlan, &checksum,
927 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
928 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
929 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
930 &fcs_ok);
931
932 if (packet_error) {
933
934 if (bytes_written > 0 && !fcs_ok)
935 enic->rq_bad_fcs++;
936
937 dev_kfree_skb_any(skb);
938
939 return;
940 }
941
942 if (eop && bytes_written > 0) {
943
944 /* Good receive
945 */
946
947 skb_put(skb, bytes_written);
948 skb->protocol = eth_type_trans(skb, netdev);
949
950 if (enic->csum_rx_enabled && !csum_not_calc) {
951 skb->csum = htons(checksum);
952 skb->ip_summed = CHECKSUM_COMPLETE;
953 }
954
955 skb->dev = netdev;
956
957 if (enic->vlan_group && vlan_stripped) {
958
959 if ((netdev->features & NETIF_F_LRO) && ipv4)
960 lro_vlan_hwaccel_receive_skb(&enic->lro_mgr,
961 skb, enic->vlan_group,
962 vlan, cq_desc);
963 else
964 vlan_hwaccel_receive_skb(skb,
965 enic->vlan_group, vlan);
966
967 } else {
968
969 if ((netdev->features & NETIF_F_LRO) && ipv4)
970 lro_receive_skb(&enic->lro_mgr, skb, cq_desc);
971 else
972 netif_receive_skb(skb);
973
974 }
975
976 } else {
977
978 /* Buffer overflow
979 */
980
981 dev_kfree_skb_any(skb);
982 }
983 }
984
985 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
986 u8 type, u16 q_number, u16 completed_index, void *opaque)
987 {
988 struct enic *enic = vnic_dev_priv(vdev);
989
990 vnic_rq_service(&enic->rq[q_number], cq_desc,
991 completed_index, VNIC_RQ_RETURN_DESC,
992 enic_rq_indicate_buf, opaque);
993
994 return 0;
995 }
996
997 static void enic_rq_drop_buf(struct vnic_rq *rq,
998 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
999 int skipped, void *opaque)
1000 {
1001 struct enic *enic = vnic_dev_priv(rq->vdev);
1002 struct sk_buff *skb = buf->os_buf;
1003
1004 if (skipped)
1005 return;
1006
1007 pci_unmap_single(enic->pdev, buf->dma_addr,
1008 buf->len, PCI_DMA_FROMDEVICE);
1009
1010 dev_kfree_skb_any(skb);
1011 }
1012
1013 static int enic_rq_service_drop(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1014 u8 type, u16 q_number, u16 completed_index, void *opaque)
1015 {
1016 struct enic *enic = vnic_dev_priv(vdev);
1017
1018 vnic_rq_service(&enic->rq[q_number], cq_desc,
1019 completed_index, VNIC_RQ_RETURN_DESC,
1020 enic_rq_drop_buf, opaque);
1021
1022 return 0;
1023 }
1024
1025 static int enic_poll(struct napi_struct *napi, int budget)
1026 {
1027 struct enic *enic = container_of(napi, struct enic, napi);
1028 struct net_device *netdev = enic->netdev;
1029 unsigned int rq_work_to_do = budget;
1030 unsigned int wq_work_to_do = -1; /* no limit */
1031 unsigned int work_done, rq_work_done, wq_work_done;
1032
1033 /* Service RQ (first) and WQ
1034 */
1035
1036 rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1037 rq_work_to_do, enic_rq_service, NULL);
1038
1039 wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1040 wq_work_to_do, enic_wq_service, NULL);
1041
1042 /* Accumulate intr event credits for this polling
1043 * cycle. An intr event is the completion of a
1044 * a WQ or RQ packet.
1045 */
1046
1047 work_done = rq_work_done + wq_work_done;
1048
1049 if (work_done > 0)
1050 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1051 work_done,
1052 0 /* don't unmask intr */,
1053 0 /* don't reset intr timer */);
1054
1055 if (rq_work_done > 0) {
1056
1057 /* Replenish RQ
1058 */
1059
1060 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1061
1062 } else {
1063
1064 /* If no work done, flush all LROs and exit polling
1065 */
1066
1067 if (netdev->features & NETIF_F_LRO)
1068 lro_flush_all(&enic->lro_mgr);
1069
1070 napi_complete(napi);
1071 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
1072 }
1073
1074 return rq_work_done;
1075 }
1076
1077 static int enic_poll_msix(struct napi_struct *napi, int budget)
1078 {
1079 struct enic *enic = container_of(napi, struct enic, napi);
1080 struct net_device *netdev = enic->netdev;
1081 unsigned int work_to_do = budget;
1082 unsigned int work_done;
1083
1084 /* Service RQ
1085 */
1086
1087 work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1088 work_to_do, enic_rq_service, NULL);
1089
1090 if (work_done > 0) {
1091
1092 /* Replenish RQ
1093 */
1094
1095 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1096
1097 /* Return intr event credits for this polling
1098 * cycle. An intr event is the completion of a
1099 * RQ packet.
1100 */
1101
1102 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1103 work_done,
1104 0 /* don't unmask intr */,
1105 0 /* don't reset intr timer */);
1106 } else {
1107
1108 /* If no work done, flush all LROs and exit polling
1109 */
1110
1111 if (netdev->features & NETIF_F_LRO)
1112 lro_flush_all(&enic->lro_mgr);
1113
1114 napi_complete(napi);
1115 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1116 }
1117
1118 return work_done;
1119 }
1120
1121 static void enic_notify_timer(unsigned long data)
1122 {
1123 struct enic *enic = (struct enic *)data;
1124
1125 enic_notify_check(enic);
1126
1127 mod_timer(&enic->notify_timer,
1128 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1129 }
1130
1131 static void enic_free_intr(struct enic *enic)
1132 {
1133 struct net_device *netdev = enic->netdev;
1134 unsigned int i;
1135
1136 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1137 case VNIC_DEV_INTR_MODE_INTX:
1138 free_irq(enic->pdev->irq, netdev);
1139 break;
1140 case VNIC_DEV_INTR_MODE_MSI:
1141 free_irq(enic->pdev->irq, enic);
1142 break;
1143 case VNIC_DEV_INTR_MODE_MSIX:
1144 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1145 if (enic->msix[i].requested)
1146 free_irq(enic->msix_entry[i].vector,
1147 enic->msix[i].devid);
1148 break;
1149 default:
1150 break;
1151 }
1152 }
1153
1154 static int enic_request_intr(struct enic *enic)
1155 {
1156 struct net_device *netdev = enic->netdev;
1157 unsigned int i;
1158 int err = 0;
1159
1160 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1161
1162 case VNIC_DEV_INTR_MODE_INTX:
1163
1164 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1165 IRQF_SHARED, netdev->name, netdev);
1166 break;
1167
1168 case VNIC_DEV_INTR_MODE_MSI:
1169
1170 err = request_irq(enic->pdev->irq, enic_isr_msi,
1171 0, netdev->name, enic);
1172 break;
1173
1174 case VNIC_DEV_INTR_MODE_MSIX:
1175
1176 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1177 "%.11s-rx-0", netdev->name);
1178 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1179 enic->msix[ENIC_MSIX_RQ].devid = enic;
1180
1181 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1182 "%.11s-tx-0", netdev->name);
1183 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1184 enic->msix[ENIC_MSIX_WQ].devid = enic;
1185
1186 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1187 "%.11s-err", netdev->name);
1188 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1189 enic->msix[ENIC_MSIX_ERR].devid = enic;
1190
1191 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1192 "%.11s-notify", netdev->name);
1193 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1194 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1195
1196 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1197 err = request_irq(enic->msix_entry[i].vector,
1198 enic->msix[i].isr, 0,
1199 enic->msix[i].devname,
1200 enic->msix[i].devid);
1201 if (err) {
1202 enic_free_intr(enic);
1203 break;
1204 }
1205 enic->msix[i].requested = 1;
1206 }
1207
1208 break;
1209
1210 default:
1211 break;
1212 }
1213
1214 return err;
1215 }
1216
1217 static int enic_notify_set(struct enic *enic)
1218 {
1219 int err;
1220
1221 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1222 case VNIC_DEV_INTR_MODE_INTX:
1223 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1224 break;
1225 case VNIC_DEV_INTR_MODE_MSIX:
1226 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1227 break;
1228 default:
1229 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1230 break;
1231 }
1232
1233 return err;
1234 }
1235
1236 static void enic_notify_timer_start(struct enic *enic)
1237 {
1238 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1239 case VNIC_DEV_INTR_MODE_MSI:
1240 mod_timer(&enic->notify_timer, jiffies);
1241 break;
1242 default:
1243 /* Using intr for notification for INTx/MSI-X */
1244 break;
1245 };
1246 }
1247
1248 /* rtnl lock is held, process context */
1249 static int enic_open(struct net_device *netdev)
1250 {
1251 struct enic *enic = netdev_priv(netdev);
1252 unsigned int i;
1253 int err;
1254
1255 err = enic_request_intr(enic);
1256 if (err) {
1257 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1258 netdev->name);
1259 return err;
1260 }
1261
1262 err = enic_notify_set(enic);
1263 if (err) {
1264 printk(KERN_ERR PFX
1265 "%s: Failed to alloc notify buffer, aborting.\n",
1266 netdev->name);
1267 goto err_out_free_intr;
1268 }
1269
1270 for (i = 0; i < enic->rq_count; i++) {
1271 err = vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1272 if (err) {
1273 printk(KERN_ERR PFX
1274 "%s: Unable to alloc receive buffers.\n",
1275 netdev->name);
1276 goto err_out_notify_unset;
1277 }
1278 }
1279
1280 for (i = 0; i < enic->wq_count; i++)
1281 vnic_wq_enable(&enic->wq[i]);
1282 for (i = 0; i < enic->rq_count; i++)
1283 vnic_rq_enable(&enic->rq[i]);
1284
1285 enic_add_station_addr(enic);
1286 enic_set_multicast_list(netdev);
1287
1288 netif_wake_queue(netdev);
1289 napi_enable(&enic->napi);
1290 vnic_dev_enable(enic->vdev);
1291
1292 for (i = 0; i < enic->intr_count; i++)
1293 vnic_intr_unmask(&enic->intr[i]);
1294
1295 enic_notify_timer_start(enic);
1296
1297 return 0;
1298
1299 err_out_notify_unset:
1300 vnic_dev_notify_unset(enic->vdev);
1301 err_out_free_intr:
1302 enic_free_intr(enic);
1303
1304 return err;
1305 }
1306
1307 /* rtnl lock is held, process context */
1308 static int enic_stop(struct net_device *netdev)
1309 {
1310 struct enic *enic = netdev_priv(netdev);
1311 unsigned int i;
1312 int err;
1313
1314 del_timer_sync(&enic->notify_timer);
1315
1316 vnic_dev_disable(enic->vdev);
1317 napi_disable(&enic->napi);
1318 netif_stop_queue(netdev);
1319
1320 for (i = 0; i < enic->intr_count; i++)
1321 vnic_intr_mask(&enic->intr[i]);
1322
1323 for (i = 0; i < enic->wq_count; i++) {
1324 err = vnic_wq_disable(&enic->wq[i]);
1325 if (err)
1326 return err;
1327 }
1328 for (i = 0; i < enic->rq_count; i++) {
1329 err = vnic_rq_disable(&enic->rq[i]);
1330 if (err)
1331 return err;
1332 }
1333
1334 vnic_dev_notify_unset(enic->vdev);
1335 enic_free_intr(enic);
1336
1337 (void)vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1338 -1, enic_rq_service_drop, NULL);
1339 (void)vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1340 -1, enic_wq_service, NULL);
1341
1342 for (i = 0; i < enic->wq_count; i++)
1343 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1344 for (i = 0; i < enic->rq_count; i++)
1345 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1346 for (i = 0; i < enic->cq_count; i++)
1347 vnic_cq_clean(&enic->cq[i]);
1348 for (i = 0; i < enic->intr_count; i++)
1349 vnic_intr_clean(&enic->intr[i]);
1350
1351 return 0;
1352 }
1353
1354 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1355 {
1356 struct enic *enic = netdev_priv(netdev);
1357 int running = netif_running(netdev);
1358
1359 if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1360 return -EINVAL;
1361
1362 if (running)
1363 enic_stop(netdev);
1364
1365 netdev->mtu = new_mtu;
1366
1367 if (netdev->mtu > enic->port_mtu)
1368 printk(KERN_WARNING PFX
1369 "%s: interface MTU (%d) set higher "
1370 "than port MTU (%d)\n",
1371 netdev->name, netdev->mtu, enic->port_mtu);
1372
1373 if (running)
1374 enic_open(netdev);
1375
1376 return 0;
1377 }
1378
1379 #ifdef CONFIG_NET_POLL_CONTROLLER
1380 static void enic_poll_controller(struct net_device *netdev)
1381 {
1382 struct enic *enic = netdev_priv(netdev);
1383 struct vnic_dev *vdev = enic->vdev;
1384
1385 switch (vnic_dev_get_intr_mode(vdev)) {
1386 case VNIC_DEV_INTR_MODE_MSIX:
1387 enic_isr_msix_rq(enic->pdev->irq, enic);
1388 enic_isr_msix_wq(enic->pdev->irq, enic);
1389 break;
1390 case VNIC_DEV_INTR_MODE_MSI:
1391 enic_isr_msi(enic->pdev->irq, enic);
1392 break;
1393 case VNIC_DEV_INTR_MODE_INTX:
1394 enic_isr_legacy(enic->pdev->irq, netdev);
1395 break;
1396 default:
1397 break;
1398 }
1399 }
1400 #endif
1401
1402 static int enic_dev_wait(struct vnic_dev *vdev,
1403 int (*start)(struct vnic_dev *, int),
1404 int (*finished)(struct vnic_dev *, int *),
1405 int arg)
1406 {
1407 unsigned long time;
1408 int done;
1409 int err;
1410
1411 BUG_ON(in_interrupt());
1412
1413 err = start(vdev, arg);
1414 if (err)
1415 return err;
1416
1417 /* Wait for func to complete...2 seconds max
1418 */
1419
1420 time = jiffies + (HZ * 2);
1421 do {
1422
1423 err = finished(vdev, &done);
1424 if (err)
1425 return err;
1426
1427 if (done)
1428 return 0;
1429
1430 schedule_timeout_uninterruptible(HZ / 10);
1431
1432 } while (time_after(time, jiffies));
1433
1434 return -ETIMEDOUT;
1435 }
1436
1437 static int enic_dev_open(struct enic *enic)
1438 {
1439 int err;
1440
1441 err = enic_dev_wait(enic->vdev, vnic_dev_open,
1442 vnic_dev_open_done, 0);
1443 if (err)
1444 printk(KERN_ERR PFX
1445 "vNIC device open failed, err %d.\n", err);
1446
1447 return err;
1448 }
1449
1450 static int enic_dev_soft_reset(struct enic *enic)
1451 {
1452 int err;
1453
1454 err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1455 vnic_dev_soft_reset_done, 0);
1456 if (err)
1457 printk(KERN_ERR PFX
1458 "vNIC soft reset failed, err %d.\n", err);
1459
1460 return err;
1461 }
1462
1463 static int enic_set_niccfg(struct enic *enic)
1464 {
1465 const u8 rss_default_cpu = 0;
1466 const u8 rss_hash_type = 0;
1467 const u8 rss_hash_bits = 0;
1468 const u8 rss_base_cpu = 0;
1469 const u8 rss_enable = 0;
1470 const u8 tso_ipid_split_en = 0;
1471 const u8 ig_vlan_strip_en = 1;
1472
1473 /* Enable VLAN tag stripping. RSS not enabled (yet).
1474 */
1475
1476 return enic_set_nic_cfg(enic,
1477 rss_default_cpu, rss_hash_type,
1478 rss_hash_bits, rss_base_cpu,
1479 rss_enable, tso_ipid_split_en,
1480 ig_vlan_strip_en);
1481 }
1482
1483 static void enic_reset(struct work_struct *work)
1484 {
1485 struct enic *enic = container_of(work, struct enic, reset);
1486
1487 if (!netif_running(enic->netdev))
1488 return;
1489
1490 rtnl_lock();
1491
1492 spin_lock(&enic->devcmd_lock);
1493 vnic_dev_hang_notify(enic->vdev);
1494 spin_unlock(&enic->devcmd_lock);
1495
1496 enic_stop(enic->netdev);
1497 enic_dev_soft_reset(enic);
1498 vnic_dev_init(enic->vdev, 0);
1499 enic_reset_mcaddrs(enic);
1500 enic_init_vnic_resources(enic);
1501 enic_set_niccfg(enic);
1502 enic_open(enic->netdev);
1503
1504 rtnl_unlock();
1505 }
1506
1507 static int enic_set_intr_mode(struct enic *enic)
1508 {
1509 unsigned int n = ARRAY_SIZE(enic->rq);
1510 unsigned int m = ARRAY_SIZE(enic->wq);
1511 unsigned int i;
1512
1513 /* Set interrupt mode (INTx, MSI, MSI-X) depending
1514 * system capabilities.
1515 *
1516 * Try MSI-X first
1517 *
1518 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1519 * (the second to last INTR is used for WQ/RQ errors)
1520 * (the last INTR is used for notifications)
1521 */
1522
1523 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1524 for (i = 0; i < n + m + 2; i++)
1525 enic->msix_entry[i].entry = i;
1526
1527 if (enic->config.intr_mode < 1 &&
1528 enic->rq_count >= n &&
1529 enic->wq_count >= m &&
1530 enic->cq_count >= n + m &&
1531 enic->intr_count >= n + m + 2 &&
1532 !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1533
1534 enic->rq_count = n;
1535 enic->wq_count = m;
1536 enic->cq_count = n + m;
1537 enic->intr_count = n + m + 2;
1538
1539 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1540
1541 return 0;
1542 }
1543
1544 /* Next try MSI
1545 *
1546 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1547 */
1548
1549 if (enic->config.intr_mode < 2 &&
1550 enic->rq_count >= 1 &&
1551 enic->wq_count >= 1 &&
1552 enic->cq_count >= 2 &&
1553 enic->intr_count >= 1 &&
1554 !pci_enable_msi(enic->pdev)) {
1555
1556 enic->rq_count = 1;
1557 enic->wq_count = 1;
1558 enic->cq_count = 2;
1559 enic->intr_count = 1;
1560
1561 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1562
1563 return 0;
1564 }
1565
1566 /* Next try INTx
1567 *
1568 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1569 * (the first INTR is used for WQ/RQ)
1570 * (the second INTR is used for WQ/RQ errors)
1571 * (the last INTR is used for notifications)
1572 */
1573
1574 if (enic->config.intr_mode < 3 &&
1575 enic->rq_count >= 1 &&
1576 enic->wq_count >= 1 &&
1577 enic->cq_count >= 2 &&
1578 enic->intr_count >= 3) {
1579
1580 enic->rq_count = 1;
1581 enic->wq_count = 1;
1582 enic->cq_count = 2;
1583 enic->intr_count = 3;
1584
1585 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1586
1587 return 0;
1588 }
1589
1590 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1591
1592 return -EINVAL;
1593 }
1594
1595 static void enic_clear_intr_mode(struct enic *enic)
1596 {
1597 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1598 case VNIC_DEV_INTR_MODE_MSIX:
1599 pci_disable_msix(enic->pdev);
1600 break;
1601 case VNIC_DEV_INTR_MODE_MSI:
1602 pci_disable_msi(enic->pdev);
1603 break;
1604 default:
1605 break;
1606 }
1607
1608 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1609 }
1610
1611 static void enic_iounmap(struct enic *enic)
1612 {
1613 if (enic->bar0.vaddr)
1614 iounmap(enic->bar0.vaddr);
1615 }
1616
1617 static const struct net_device_ops enic_netdev_ops = {
1618 .ndo_open = enic_open,
1619 .ndo_stop = enic_stop,
1620 .ndo_start_xmit = enic_hard_start_xmit,
1621 .ndo_get_stats = enic_get_stats,
1622 .ndo_validate_addr = eth_validate_addr,
1623 .ndo_set_mac_address = eth_mac_addr,
1624 .ndo_set_multicast_list = enic_set_multicast_list,
1625 .ndo_change_mtu = enic_change_mtu,
1626 .ndo_vlan_rx_register = enic_vlan_rx_register,
1627 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
1628 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
1629 .ndo_tx_timeout = enic_tx_timeout,
1630 #ifdef CONFIG_NET_POLL_CONTROLLER
1631 .ndo_poll_controller = enic_poll_controller,
1632 #endif
1633 };
1634
1635 static int __devinit enic_probe(struct pci_dev *pdev,
1636 const struct pci_device_id *ent)
1637 {
1638 struct net_device *netdev;
1639 struct enic *enic;
1640 int using_dac = 0;
1641 unsigned int i;
1642 int err;
1643
1644 /* Allocate net device structure and initialize. Private
1645 * instance data is initialized to zero.
1646 */
1647
1648 netdev = alloc_etherdev(sizeof(struct enic));
1649 if (!netdev) {
1650 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
1651 return -ENOMEM;
1652 }
1653
1654 pci_set_drvdata(pdev, netdev);
1655
1656 SET_NETDEV_DEV(netdev, &pdev->dev);
1657
1658 enic = netdev_priv(netdev);
1659 enic->netdev = netdev;
1660 enic->pdev = pdev;
1661
1662 /* Setup PCI resources
1663 */
1664
1665 err = pci_enable_device(pdev);
1666 if (err) {
1667 printk(KERN_ERR PFX
1668 "Cannot enable PCI device, aborting.\n");
1669 goto err_out_free_netdev;
1670 }
1671
1672 err = pci_request_regions(pdev, DRV_NAME);
1673 if (err) {
1674 printk(KERN_ERR PFX
1675 "Cannot request PCI regions, aborting.\n");
1676 goto err_out_disable_device;
1677 }
1678
1679 pci_set_master(pdev);
1680
1681 /* Query PCI controller on system for DMA addressing
1682 * limitation for the device. Try 40-bit first, and
1683 * fail to 32-bit.
1684 */
1685
1686 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
1687 if (err) {
1688 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1689 if (err) {
1690 printk(KERN_ERR PFX
1691 "No usable DMA configuration, aborting.\n");
1692 goto err_out_release_regions;
1693 }
1694 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1695 if (err) {
1696 printk(KERN_ERR PFX
1697 "Unable to obtain 32-bit DMA "
1698 "for consistent allocations, aborting.\n");
1699 goto err_out_release_regions;
1700 }
1701 } else {
1702 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
1703 if (err) {
1704 printk(KERN_ERR PFX
1705 "Unable to obtain 40-bit DMA "
1706 "for consistent allocations, aborting.\n");
1707 goto err_out_release_regions;
1708 }
1709 using_dac = 1;
1710 }
1711
1712 /* Map vNIC resources from BAR0
1713 */
1714
1715 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
1716 printk(KERN_ERR PFX
1717 "BAR0 not memory-map'able, aborting.\n");
1718 err = -ENODEV;
1719 goto err_out_release_regions;
1720 }
1721
1722 enic->bar0.vaddr = pci_iomap(pdev, 0, enic->bar0.len);
1723 enic->bar0.bus_addr = pci_resource_start(pdev, 0);
1724 enic->bar0.len = pci_resource_len(pdev, 0);
1725
1726 if (!enic->bar0.vaddr) {
1727 printk(KERN_ERR PFX
1728 "Cannot memory-map BAR0 res hdr, aborting.\n");
1729 err = -ENODEV;
1730 goto err_out_release_regions;
1731 }
1732
1733 /* Register vNIC device
1734 */
1735
1736 enic->vdev = vnic_dev_register(NULL, enic, pdev, &enic->bar0);
1737 if (!enic->vdev) {
1738 printk(KERN_ERR PFX
1739 "vNIC registration failed, aborting.\n");
1740 err = -ENODEV;
1741 goto err_out_iounmap;
1742 }
1743
1744 /* Issue device open to get device in known state
1745 */
1746
1747 err = enic_dev_open(enic);
1748 if (err) {
1749 printk(KERN_ERR PFX
1750 "vNIC dev open failed, aborting.\n");
1751 goto err_out_vnic_unregister;
1752 }
1753
1754 /* Issue device init to initialize the vnic-to-switch link.
1755 * We'll start with carrier off and wait for link UP
1756 * notification later to turn on carrier. We don't need
1757 * to wait here for the vnic-to-switch link initialization
1758 * to complete; link UP notification is the indication that
1759 * the process is complete.
1760 */
1761
1762 netif_carrier_off(netdev);
1763
1764 err = vnic_dev_init(enic->vdev, 0);
1765 if (err) {
1766 printk(KERN_ERR PFX
1767 "vNIC dev init failed, aborting.\n");
1768 goto err_out_dev_close;
1769 }
1770
1771 /* Get vNIC configuration
1772 */
1773
1774 err = enic_get_vnic_config(enic);
1775 if (err) {
1776 printk(KERN_ERR PFX
1777 "Get vNIC configuration failed, aborting.\n");
1778 goto err_out_dev_close;
1779 }
1780
1781 /* Get available resource counts
1782 */
1783
1784 enic_get_res_counts(enic);
1785
1786 /* Set interrupt mode based on resource counts and system
1787 * capabilities
1788 */
1789
1790 err = enic_set_intr_mode(enic);
1791 if (err) {
1792 printk(KERN_ERR PFX
1793 "Failed to set intr mode, aborting.\n");
1794 goto err_out_dev_close;
1795 }
1796
1797 /* Allocate and configure vNIC resources
1798 */
1799
1800 err = enic_alloc_vnic_resources(enic);
1801 if (err) {
1802 printk(KERN_ERR PFX
1803 "Failed to alloc vNIC resources, aborting.\n");
1804 goto err_out_free_vnic_resources;
1805 }
1806
1807 enic_init_vnic_resources(enic);
1808
1809 err = enic_set_niccfg(enic);
1810 if (err) {
1811 printk(KERN_ERR PFX
1812 "Failed to config nic, aborting.\n");
1813 goto err_out_free_vnic_resources;
1814 }
1815
1816 /* Setup notification timer, HW reset task, and locks
1817 */
1818
1819 init_timer(&enic->notify_timer);
1820 enic->notify_timer.function = enic_notify_timer;
1821 enic->notify_timer.data = (unsigned long)enic;
1822
1823 INIT_WORK(&enic->reset, enic_reset);
1824
1825 for (i = 0; i < enic->wq_count; i++)
1826 spin_lock_init(&enic->wq_lock[i]);
1827
1828 spin_lock_init(&enic->devcmd_lock);
1829
1830 /* Register net device
1831 */
1832
1833 enic->port_mtu = enic->config.mtu;
1834 (void)enic_change_mtu(netdev, enic->port_mtu);
1835
1836 err = enic_set_mac_addr(netdev, enic->mac_addr);
1837 if (err) {
1838 printk(KERN_ERR PFX
1839 "Invalid MAC address, aborting.\n");
1840 goto err_out_free_vnic_resources;
1841 }
1842
1843 netdev->netdev_ops = &enic_netdev_ops;
1844 netdev->watchdog_timeo = 2 * HZ;
1845 netdev->ethtool_ops = &enic_ethtool_ops;
1846
1847 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1848 default:
1849 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
1850 break;
1851 case VNIC_DEV_INTR_MODE_MSIX:
1852 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
1853 break;
1854 }
1855
1856 netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
1857 if (ENIC_SETTING(enic, TXCSUM))
1858 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1859 if (ENIC_SETTING(enic, TSO))
1860 netdev->features |= NETIF_F_TSO |
1861 NETIF_F_TSO6 | NETIF_F_TSO_ECN;
1862 if (ENIC_SETTING(enic, LRO))
1863 netdev->features |= NETIF_F_LRO;
1864 if (using_dac)
1865 netdev->features |= NETIF_F_HIGHDMA;
1866
1867 enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
1868
1869 enic->lro_mgr.max_aggr = ENIC_LRO_MAX_AGGR;
1870 enic->lro_mgr.max_desc = ENIC_LRO_MAX_DESC;
1871 enic->lro_mgr.lro_arr = enic->lro_desc;
1872 enic->lro_mgr.get_skb_header = enic_get_skb_header;
1873 enic->lro_mgr.features = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
1874 enic->lro_mgr.dev = netdev;
1875 enic->lro_mgr.ip_summed = CHECKSUM_COMPLETE;
1876 enic->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1877
1878 err = register_netdev(netdev);
1879 if (err) {
1880 printk(KERN_ERR PFX
1881 "Cannot register net device, aborting.\n");
1882 goto err_out_free_vnic_resources;
1883 }
1884
1885 return 0;
1886
1887 err_out_free_vnic_resources:
1888 enic_free_vnic_resources(enic);
1889 err_out_dev_close:
1890 vnic_dev_close(enic->vdev);
1891 err_out_vnic_unregister:
1892 enic_clear_intr_mode(enic);
1893 vnic_dev_unregister(enic->vdev);
1894 err_out_iounmap:
1895 enic_iounmap(enic);
1896 err_out_release_regions:
1897 pci_release_regions(pdev);
1898 err_out_disable_device:
1899 pci_disable_device(pdev);
1900 err_out_free_netdev:
1901 pci_set_drvdata(pdev, NULL);
1902 free_netdev(netdev);
1903
1904 return err;
1905 }
1906
1907 static void __devexit enic_remove(struct pci_dev *pdev)
1908 {
1909 struct net_device *netdev = pci_get_drvdata(pdev);
1910
1911 if (netdev) {
1912 struct enic *enic = netdev_priv(netdev);
1913
1914 flush_scheduled_work();
1915 unregister_netdev(netdev);
1916 enic_free_vnic_resources(enic);
1917 vnic_dev_close(enic->vdev);
1918 enic_clear_intr_mode(enic);
1919 vnic_dev_unregister(enic->vdev);
1920 enic_iounmap(enic);
1921 pci_release_regions(pdev);
1922 pci_disable_device(pdev);
1923 pci_set_drvdata(pdev, NULL);
1924 free_netdev(netdev);
1925 }
1926 }
1927
1928 static struct pci_driver enic_driver = {
1929 .name = DRV_NAME,
1930 .id_table = enic_id_table,
1931 .probe = enic_probe,
1932 .remove = __devexit_p(enic_remove),
1933 };
1934
1935 static int __init enic_init_module(void)
1936 {
1937 printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
1938
1939 return pci_register_driver(&enic_driver);
1940 }
1941
1942 static void __exit enic_cleanup_module(void)
1943 {
1944 pci_unregister_driver(&enic_driver);
1945 }
1946
1947 module_init(enic_init_module);
1948 module_exit(enic_cleanup_module);
Linux kernel - Deliverables
This page took 0.133832 seconds and 6 git commands to generate.