?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ocfs2: Optimize inode allocation by remembering last group
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / net / enic / enic_main.c
blob03403a51f7eae0f95cb26a3517c2b4d22536a149
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 netdev->trans_start = jiffies;
665
666 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
667
668 return NETDEV_TX_OK;
669 }
670
671 /* dev_base_lock rwlock held, nominally process context */
672 static struct net_device_stats *enic_get_stats(struct net_device *netdev)
673 {
674 struct enic *enic = netdev_priv(netdev);
675 struct net_device_stats *net_stats = &netdev->stats;
676 struct vnic_stats *stats;
677
678 spin_lock(&enic->devcmd_lock);
679 vnic_dev_stats_dump(enic->vdev, &stats);
680 spin_unlock(&enic->devcmd_lock);
681
682 net_stats->tx_packets = stats->tx.tx_frames_ok;
683 net_stats->tx_bytes = stats->tx.tx_bytes_ok;
684 net_stats->tx_errors = stats->tx.tx_errors;
685 net_stats->tx_dropped = stats->tx.tx_drops;
686
687 net_stats->rx_packets = stats->rx.rx_frames_ok;
688 net_stats->rx_bytes = stats->rx.rx_bytes_ok;
689 net_stats->rx_errors = stats->rx.rx_errors;
690 net_stats->multicast = stats->rx.rx_multicast_frames_ok;
691 net_stats->rx_crc_errors = enic->rq_bad_fcs;
692 net_stats->rx_dropped = stats->rx.rx_no_bufs;
693
694 return net_stats;
695 }
696
697 static void enic_reset_mcaddrs(struct enic *enic)
698 {
699 enic->mc_count = 0;
700 }
701
702 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
703 {
704 if (!is_valid_ether_addr(addr))
705 return -EADDRNOTAVAIL;
706
707 memcpy(netdev->dev_addr, addr, netdev->addr_len);
708
709 return 0;
710 }
711
712 /* netif_tx_lock held, BHs disabled */
713 static void enic_set_multicast_list(struct net_device *netdev)
714 {
715 struct enic *enic = netdev_priv(netdev);
716 struct dev_mc_list *list = netdev->mc_list;
717 int directed = 1;
718 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
719 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
720 int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
721 int allmulti = (netdev->flags & IFF_ALLMULTI) ||
722 (netdev->mc_count > ENIC_MULTICAST_PERFECT_FILTERS);
723 u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
724 unsigned int mc_count = netdev->mc_count;
725 unsigned int i, j;
726
727 if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
728 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
729
730 spin_lock(&enic->devcmd_lock);
731
732 vnic_dev_packet_filter(enic->vdev, directed,
733 multicast, broadcast, promisc, allmulti);
734
735 /* Is there an easier way? Trying to minimize to
736 * calls to add/del multicast addrs. We keep the
737 * addrs from the last call in enic->mc_addr and
738 * look for changes to add/del.
739 */
740
741 for (i = 0; list && i < mc_count; i++) {
742 memcpy(mc_addr[i], list->dmi_addr, ETH_ALEN);
743 list = list->next;
744 }
745
746 for (i = 0; i < enic->mc_count; i++) {
747 for (j = 0; j < mc_count; j++)
748 if (compare_ether_addr(enic->mc_addr[i],
749 mc_addr[j]) == 0)
750 break;
751 if (j == mc_count)
752 enic_del_multicast_addr(enic, enic->mc_addr[i]);
753 }
754
755 for (i = 0; i < mc_count; i++) {
756 for (j = 0; j < enic->mc_count; j++)
757 if (compare_ether_addr(mc_addr[i],
758 enic->mc_addr[j]) == 0)
759 break;
760 if (j == enic->mc_count)
761 enic_add_multicast_addr(enic, mc_addr[i]);
762 }
763
764 /* Save the list to compare against next time
765 */
766
767 for (i = 0; i < mc_count; i++)
768 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
769
770 enic->mc_count = mc_count;
771
772 spin_unlock(&enic->devcmd_lock);
773 }
774
775 /* rtnl lock is held */
776 static void enic_vlan_rx_register(struct net_device *netdev,
777 struct vlan_group *vlan_group)
778 {
779 struct enic *enic = netdev_priv(netdev);
780 enic->vlan_group = vlan_group;
781 }
782
783 /* rtnl lock is held */
784 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
785 {
786 struct enic *enic = netdev_priv(netdev);
787
788 spin_lock(&enic->devcmd_lock);
789 enic_add_vlan(enic, vid);
790 spin_unlock(&enic->devcmd_lock);
791 }
792
793 /* rtnl lock is held */
794 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
795 {
796 struct enic *enic = netdev_priv(netdev);
797
798 spin_lock(&enic->devcmd_lock);
799 enic_del_vlan(enic, vid);
800 spin_unlock(&enic->devcmd_lock);
801 }
802
803 /* netif_tx_lock held, BHs disabled */
804 static void enic_tx_timeout(struct net_device *netdev)
805 {
806 struct enic *enic = netdev_priv(netdev);
807 schedule_work(&enic->reset);
808 }
809
810 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
811 {
812 struct enic *enic = vnic_dev_priv(rq->vdev);
813
814 if (!buf->os_buf)
815 return;
816
817 pci_unmap_single(enic->pdev, buf->dma_addr,
818 buf->len, PCI_DMA_FROMDEVICE);
819 dev_kfree_skb_any(buf->os_buf);
820 }
821
822 static inline struct sk_buff *enic_rq_alloc_skb(unsigned int size)
823 {
824 struct sk_buff *skb;
825
826 skb = dev_alloc_skb(size + NET_IP_ALIGN);
827
828 if (skb)
829 skb_reserve(skb, NET_IP_ALIGN);
830
831 return skb;
832 }
833
834 static int enic_rq_alloc_buf(struct vnic_rq *rq)
835 {
836 struct enic *enic = vnic_dev_priv(rq->vdev);
837 struct sk_buff *skb;
838 unsigned int len = enic->netdev->mtu + ETH_HLEN;
839 unsigned int os_buf_index = 0;
840 dma_addr_t dma_addr;
841
842 skb = enic_rq_alloc_skb(len);
843 if (!skb)
844 return -ENOMEM;
845
846 dma_addr = pci_map_single(enic->pdev, skb->data,
847 len, PCI_DMA_FROMDEVICE);
848
849 enic_queue_rq_desc(rq, skb, os_buf_index,
850 dma_addr, len);
851
852 return 0;
853 }
854
855 static int enic_get_skb_header(struct sk_buff *skb, void **iphdr,
856 void **tcph, u64 *hdr_flags, void *priv)
857 {
858 struct cq_enet_rq_desc *cq_desc = priv;
859 unsigned int ip_len;
860 struct iphdr *iph;
861
862 u8 type, color, eop, sop, ingress_port, vlan_stripped;
863 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
864 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
865 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
866 u8 packet_error;
867 u16 q_number, completed_index, bytes_written, vlan, checksum;
868 u32 rss_hash;
869
870 cq_enet_rq_desc_dec(cq_desc,
871 &type, &color, &q_number, &completed_index,
872 &ingress_port, &fcoe, &eop, &sop, &rss_type,
873 &csum_not_calc, &rss_hash, &bytes_written,
874 &packet_error, &vlan_stripped, &vlan, &checksum,
875 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
876 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
877 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
878 &fcs_ok);
879
880 if (!(ipv4 && tcp && !ipv4_fragment))
881 return -1;
882
883 skb_reset_network_header(skb);
884 iph = ip_hdr(skb);
885
886 ip_len = ip_hdrlen(skb);
887 skb_set_transport_header(skb, ip_len);
888
889 /* check if ip header and tcp header are complete */
890 if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb))
891 return -1;
892
893 *hdr_flags = LRO_IPV4 | LRO_TCP;
894 *tcph = tcp_hdr(skb);
895 *iphdr = iph;
896
897 return 0;
898 }
899
900 static void enic_rq_indicate_buf(struct vnic_rq *rq,
901 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
902 int skipped, void *opaque)
903 {
904 struct enic *enic = vnic_dev_priv(rq->vdev);
905 struct net_device *netdev = enic->netdev;
906 struct sk_buff *skb;
907
908 u8 type, color, eop, sop, ingress_port, vlan_stripped;
909 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
910 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
911 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
912 u8 packet_error;
913 u16 q_number, completed_index, bytes_written, vlan, checksum;
914 u32 rss_hash;
915
916 if (skipped)
917 return;
918
919 skb = buf->os_buf;
920 prefetch(skb->data - NET_IP_ALIGN);
921 pci_unmap_single(enic->pdev, buf->dma_addr,
922 buf->len, PCI_DMA_FROMDEVICE);
923
924 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
925 &type, &color, &q_number, &completed_index,
926 &ingress_port, &fcoe, &eop, &sop, &rss_type,
927 &csum_not_calc, &rss_hash, &bytes_written,
928 &packet_error, &vlan_stripped, &vlan, &checksum,
929 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
930 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
931 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
932 &fcs_ok);
933
934 if (packet_error) {
935
936 if (bytes_written > 0 && !fcs_ok)
937 enic->rq_bad_fcs++;
938
939 dev_kfree_skb_any(skb);
940
941 return;
942 }
943
944 if (eop && bytes_written > 0) {
945
946 /* Good receive
947 */
948
949 skb_put(skb, bytes_written);
950 skb->protocol = eth_type_trans(skb, netdev);
951
952 if (enic->csum_rx_enabled && !csum_not_calc) {
953 skb->csum = htons(checksum);
954 skb->ip_summed = CHECKSUM_COMPLETE;
955 }
956
957 skb->dev = netdev;
958
959 if (enic->vlan_group && vlan_stripped) {
960
961 if ((netdev->features & NETIF_F_LRO) && ipv4)
962 lro_vlan_hwaccel_receive_skb(&enic->lro_mgr,
963 skb, enic->vlan_group,
964 vlan, cq_desc);
965 else
966 vlan_hwaccel_receive_skb(skb,
967 enic->vlan_group, vlan);
968
969 } else {
970
971 if ((netdev->features & NETIF_F_LRO) && ipv4)
972 lro_receive_skb(&enic->lro_mgr, skb, cq_desc);
973 else
974 netif_receive_skb(skb);
975
976 }
977
978 } else {
979
980 /* Buffer overflow
981 */
982
983 dev_kfree_skb_any(skb);
984 }
985 }
986
987 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
988 u8 type, u16 q_number, u16 completed_index, void *opaque)
989 {
990 struct enic *enic = vnic_dev_priv(vdev);
991
992 vnic_rq_service(&enic->rq[q_number], cq_desc,
993 completed_index, VNIC_RQ_RETURN_DESC,
994 enic_rq_indicate_buf, opaque);
995
996 return 0;
997 }
998
999 static void enic_rq_drop_buf(struct vnic_rq *rq,
1000 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1001 int skipped, void *opaque)
1002 {
1003 struct enic *enic = vnic_dev_priv(rq->vdev);
1004 struct sk_buff *skb = buf->os_buf;
1005
1006 if (skipped)
1007 return;
1008
1009 pci_unmap_single(enic->pdev, buf->dma_addr,
1010 buf->len, PCI_DMA_FROMDEVICE);
1011
1012 dev_kfree_skb_any(skb);
1013 }
1014
1015 static int enic_rq_service_drop(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1016 u8 type, u16 q_number, u16 completed_index, void *opaque)
1017 {
1018 struct enic *enic = vnic_dev_priv(vdev);
1019
1020 vnic_rq_service(&enic->rq[q_number], cq_desc,
1021 completed_index, VNIC_RQ_RETURN_DESC,
1022 enic_rq_drop_buf, opaque);
1023
1024 return 0;
1025 }
1026
1027 static int enic_poll(struct napi_struct *napi, int budget)
1028 {
1029 struct enic *enic = container_of(napi, struct enic, napi);
1030 struct net_device *netdev = enic->netdev;
1031 unsigned int rq_work_to_do = budget;
1032 unsigned int wq_work_to_do = -1; /* no limit */
1033 unsigned int work_done, rq_work_done, wq_work_done;
1034
1035 /* Service RQ (first) and WQ
1036 */
1037
1038 rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1039 rq_work_to_do, enic_rq_service, NULL);
1040
1041 wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1042 wq_work_to_do, enic_wq_service, NULL);
1043
1044 /* Accumulate intr event credits for this polling
1045 * cycle. An intr event is the completion of a
1046 * a WQ or RQ packet.
1047 */
1048
1049 work_done = rq_work_done + wq_work_done;
1050
1051 if (work_done > 0)
1052 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1053 work_done,
1054 0 /* don't unmask intr */,
1055 0 /* don't reset intr timer */);
1056
1057 if (rq_work_done > 0) {
1058
1059 /* Replenish RQ
1060 */
1061
1062 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1063
1064 } else {
1065
1066 /* If no work done, flush all LROs and exit polling
1067 */
1068
1069 if (netdev->features & NETIF_F_LRO)
1070 lro_flush_all(&enic->lro_mgr);
1071
1072 napi_complete(napi);
1073 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
1074 }
1075
1076 return rq_work_done;
1077 }
1078
1079 static int enic_poll_msix(struct napi_struct *napi, int budget)
1080 {
1081 struct enic *enic = container_of(napi, struct enic, napi);
1082 struct net_device *netdev = enic->netdev;
1083 unsigned int work_to_do = budget;
1084 unsigned int work_done;
1085
1086 /* Service RQ
1087 */
1088
1089 work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1090 work_to_do, enic_rq_service, NULL);
1091
1092 if (work_done > 0) {
1093
1094 /* Replenish RQ
1095 */
1096
1097 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1098
1099 /* Return intr event credits for this polling
1100 * cycle. An intr event is the completion of a
1101 * RQ packet.
1102 */
1103
1104 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1105 work_done,
1106 0 /* don't unmask intr */,
1107 0 /* don't reset intr timer */);
1108 } else {
1109
1110 /* If no work done, flush all LROs and exit polling
1111 */
1112
1113 if (netdev->features & NETIF_F_LRO)
1114 lro_flush_all(&enic->lro_mgr);
1115
1116 napi_complete(napi);
1117 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1118 }
1119
1120 return work_done;
1121 }
1122
1123 static void enic_notify_timer(unsigned long data)
1124 {
1125 struct enic *enic = (struct enic *)data;
1126
1127 enic_notify_check(enic);
1128
1129 mod_timer(&enic->notify_timer,
1130 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1131 }
1132
1133 static void enic_free_intr(struct enic *enic)
1134 {
1135 struct net_device *netdev = enic->netdev;
1136 unsigned int i;
1137
1138 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1139 case VNIC_DEV_INTR_MODE_INTX:
1140 free_irq(enic->pdev->irq, netdev);
1141 break;
1142 case VNIC_DEV_INTR_MODE_MSI:
1143 free_irq(enic->pdev->irq, enic);
1144 break;
1145 case VNIC_DEV_INTR_MODE_MSIX:
1146 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1147 if (enic->msix[i].requested)
1148 free_irq(enic->msix_entry[i].vector,
1149 enic->msix[i].devid);
1150 break;
1151 default:
1152 break;
1153 }
1154 }
1155
1156 static int enic_request_intr(struct enic *enic)
1157 {
1158 struct net_device *netdev = enic->netdev;
1159 unsigned int i;
1160 int err = 0;
1161
1162 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1163
1164 case VNIC_DEV_INTR_MODE_INTX:
1165
1166 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1167 IRQF_SHARED, netdev->name, netdev);
1168 break;
1169
1170 case VNIC_DEV_INTR_MODE_MSI:
1171
1172 err = request_irq(enic->pdev->irq, enic_isr_msi,
1173 0, netdev->name, enic);
1174 break;
1175
1176 case VNIC_DEV_INTR_MODE_MSIX:
1177
1178 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1179 "%.11s-rx-0", netdev->name);
1180 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1181 enic->msix[ENIC_MSIX_RQ].devid = enic;
1182
1183 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1184 "%.11s-tx-0", netdev->name);
1185 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1186 enic->msix[ENIC_MSIX_WQ].devid = enic;
1187
1188 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1189 "%.11s-err", netdev->name);
1190 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1191 enic->msix[ENIC_MSIX_ERR].devid = enic;
1192
1193 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1194 "%.11s-notify", netdev->name);
1195 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1196 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1197
1198 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1199 err = request_irq(enic->msix_entry[i].vector,
1200 enic->msix[i].isr, 0,
1201 enic->msix[i].devname,
1202 enic->msix[i].devid);
1203 if (err) {
1204 enic_free_intr(enic);
1205 break;
1206 }
1207 enic->msix[i].requested = 1;
1208 }
1209
1210 break;
1211
1212 default:
1213 break;
1214 }
1215
1216 return err;
1217 }
1218
1219 static int enic_notify_set(struct enic *enic)
1220 {
1221 int err;
1222
1223 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1224 case VNIC_DEV_INTR_MODE_INTX:
1225 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1226 break;
1227 case VNIC_DEV_INTR_MODE_MSIX:
1228 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1229 break;
1230 default:
1231 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1232 break;
1233 }
1234
1235 return err;
1236 }
1237
1238 static void enic_notify_timer_start(struct enic *enic)
1239 {
1240 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1241 case VNIC_DEV_INTR_MODE_MSI:
1242 mod_timer(&enic->notify_timer, jiffies);
1243 break;
1244 default:
1245 /* Using intr for notification for INTx/MSI-X */
1246 break;
1247 };
1248 }
1249
1250 /* rtnl lock is held, process context */
1251 static int enic_open(struct net_device *netdev)
1252 {
1253 struct enic *enic = netdev_priv(netdev);
1254 unsigned int i;
1255 int err;
1256
1257 err = enic_request_intr(enic);
1258 if (err) {
1259 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1260 netdev->name);
1261 return err;
1262 }
1263
1264 err = enic_notify_set(enic);
1265 if (err) {
1266 printk(KERN_ERR PFX
1267 "%s: Failed to alloc notify buffer, aborting.\n",
1268 netdev->name);
1269 goto err_out_free_intr;
1270 }
1271
1272 for (i = 0; i < enic->rq_count; i++) {
1273 err = vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1274 if (err) {
1275 printk(KERN_ERR PFX
1276 "%s: Unable to alloc receive buffers.\n",
1277 netdev->name);
1278 goto err_out_notify_unset;
1279 }
1280 }
1281
1282 for (i = 0; i < enic->wq_count; i++)
1283 vnic_wq_enable(&enic->wq[i]);
1284 for (i = 0; i < enic->rq_count; i++)
1285 vnic_rq_enable(&enic->rq[i]);
1286
1287 enic_add_station_addr(enic);
1288 enic_set_multicast_list(netdev);
1289
1290 netif_wake_queue(netdev);
1291 napi_enable(&enic->napi);
1292 vnic_dev_enable(enic->vdev);
1293
1294 for (i = 0; i < enic->intr_count; i++)
1295 vnic_intr_unmask(&enic->intr[i]);
1296
1297 enic_notify_timer_start(enic);
1298
1299 return 0;
1300
1301 err_out_notify_unset:
1302 vnic_dev_notify_unset(enic->vdev);
1303 err_out_free_intr:
1304 enic_free_intr(enic);
1305
1306 return err;
1307 }
1308
1309 /* rtnl lock is held, process context */
1310 static int enic_stop(struct net_device *netdev)
1311 {
1312 struct enic *enic = netdev_priv(netdev);
1313 unsigned int i;
1314 int err;
1315
1316 del_timer_sync(&enic->notify_timer);
1317
1318 vnic_dev_disable(enic->vdev);
1319 napi_disable(&enic->napi);
1320 netif_stop_queue(netdev);
1321
1322 for (i = 0; i < enic->intr_count; i++)
1323 vnic_intr_mask(&enic->intr[i]);
1324
1325 for (i = 0; i < enic->wq_count; i++) {
1326 err = vnic_wq_disable(&enic->wq[i]);
1327 if (err)
1328 return err;
1329 }
1330 for (i = 0; i < enic->rq_count; i++) {
1331 err = vnic_rq_disable(&enic->rq[i]);
1332 if (err)
1333 return err;
1334 }
1335
1336 vnic_dev_notify_unset(enic->vdev);
1337 enic_free_intr(enic);
1338
1339 (void)vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1340 -1, enic_rq_service_drop, NULL);
1341 (void)vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1342 -1, enic_wq_service, NULL);
1343
1344 for (i = 0; i < enic->wq_count; i++)
1345 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1346 for (i = 0; i < enic->rq_count; i++)
1347 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1348 for (i = 0; i < enic->cq_count; i++)
1349 vnic_cq_clean(&enic->cq[i]);
1350 for (i = 0; i < enic->intr_count; i++)
1351 vnic_intr_clean(&enic->intr[i]);
1352
1353 return 0;
1354 }
1355
1356 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1357 {
1358 struct enic *enic = netdev_priv(netdev);
1359 int running = netif_running(netdev);
1360
1361 if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1362 return -EINVAL;
1363
1364 if (running)
1365 enic_stop(netdev);
1366
1367 netdev->mtu = new_mtu;
1368
1369 if (netdev->mtu > enic->port_mtu)
1370 printk(KERN_WARNING PFX
1371 "%s: interface MTU (%d) set higher "
1372 "than port MTU (%d)\n",
1373 netdev->name, netdev->mtu, enic->port_mtu);
1374
1375 if (running)
1376 enic_open(netdev);
1377
1378 return 0;
1379 }
1380
1381 #ifdef CONFIG_NET_POLL_CONTROLLER
1382 static void enic_poll_controller(struct net_device *netdev)
1383 {
1384 struct enic *enic = netdev_priv(netdev);
1385 struct vnic_dev *vdev = enic->vdev;
1386
1387 switch (vnic_dev_get_intr_mode(vdev)) {
1388 case VNIC_DEV_INTR_MODE_MSIX:
1389 enic_isr_msix_rq(enic->pdev->irq, enic);
1390 enic_isr_msix_wq(enic->pdev->irq, enic);
1391 break;
1392 case VNIC_DEV_INTR_MODE_MSI:
1393 enic_isr_msi(enic->pdev->irq, enic);
1394 break;
1395 case VNIC_DEV_INTR_MODE_INTX:
1396 enic_isr_legacy(enic->pdev->irq, netdev);
1397 break;
1398 default:
1399 break;
1400 }
1401 }
1402 #endif
1403
1404 static int enic_dev_wait(struct vnic_dev *vdev,
1405 int (*start)(struct vnic_dev *, int),
1406 int (*finished)(struct vnic_dev *, int *),
1407 int arg)
1408 {
1409 unsigned long time;
1410 int done;
1411 int err;
1412
1413 BUG_ON(in_interrupt());
1414
1415 err = start(vdev, arg);
1416 if (err)
1417 return err;
1418
1419 /* Wait for func to complete...2 seconds max
1420 */
1421
1422 time = jiffies + (HZ * 2);
1423 do {
1424
1425 err = finished(vdev, &done);
1426 if (err)
1427 return err;
1428
1429 if (done)
1430 return 0;
1431
1432 schedule_timeout_uninterruptible(HZ / 10);
1433
1434 } while (time_after(time, jiffies));
1435
1436 return -ETIMEDOUT;
1437 }
1438
1439 static int enic_dev_open(struct enic *enic)
1440 {
1441 int err;
1442
1443 err = enic_dev_wait(enic->vdev, vnic_dev_open,
1444 vnic_dev_open_done, 0);
1445 if (err)
1446 printk(KERN_ERR PFX
1447 "vNIC device open failed, err %d.\n", err);
1448
1449 return err;
1450 }
1451
1452 static int enic_dev_soft_reset(struct enic *enic)
1453 {
1454 int err;
1455
1456 err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1457 vnic_dev_soft_reset_done, 0);
1458 if (err)
1459 printk(KERN_ERR PFX
1460 "vNIC soft reset failed, err %d.\n", err);
1461
1462 return err;
1463 }
1464
1465 static int enic_set_niccfg(struct enic *enic)
1466 {
1467 const u8 rss_default_cpu = 0;
1468 const u8 rss_hash_type = 0;
1469 const u8 rss_hash_bits = 0;
1470 const u8 rss_base_cpu = 0;
1471 const u8 rss_enable = 0;
1472 const u8 tso_ipid_split_en = 0;
1473 const u8 ig_vlan_strip_en = 1;
1474
1475 /* Enable VLAN tag stripping. RSS not enabled (yet).
1476 */
1477
1478 return enic_set_nic_cfg(enic,
1479 rss_default_cpu, rss_hash_type,
1480 rss_hash_bits, rss_base_cpu,
1481 rss_enable, tso_ipid_split_en,
1482 ig_vlan_strip_en);
1483 }
1484
1485 static void enic_reset(struct work_struct *work)
1486 {
1487 struct enic *enic = container_of(work, struct enic, reset);
1488
1489 if (!netif_running(enic->netdev))
1490 return;
1491
1492 rtnl_lock();
1493
1494 spin_lock(&enic->devcmd_lock);
1495 vnic_dev_hang_notify(enic->vdev);
1496 spin_unlock(&enic->devcmd_lock);
1497
1498 enic_stop(enic->netdev);
1499 enic_dev_soft_reset(enic);
1500 vnic_dev_init(enic->vdev, 0);
1501 enic_reset_mcaddrs(enic);
1502 enic_init_vnic_resources(enic);
1503 enic_set_niccfg(enic);
1504 enic_open(enic->netdev);
1505
1506 rtnl_unlock();
1507 }
1508
1509 static int enic_set_intr_mode(struct enic *enic)
1510 {
1511 unsigned int n = ARRAY_SIZE(enic->rq);
1512 unsigned int m = ARRAY_SIZE(enic->wq);
1513 unsigned int i;
1514
1515 /* Set interrupt mode (INTx, MSI, MSI-X) depending
1516 * system capabilities.
1517 *
1518 * Try MSI-X first
1519 *
1520 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1521 * (the second to last INTR is used for WQ/RQ errors)
1522 * (the last INTR is used for notifications)
1523 */
1524
1525 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1526 for (i = 0; i < n + m + 2; i++)
1527 enic->msix_entry[i].entry = i;
1528
1529 if (enic->config.intr_mode < 1 &&
1530 enic->rq_count >= n &&
1531 enic->wq_count >= m &&
1532 enic->cq_count >= n + m &&
1533 enic->intr_count >= n + m + 2 &&
1534 !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1535
1536 enic->rq_count = n;
1537 enic->wq_count = m;
1538 enic->cq_count = n + m;
1539 enic->intr_count = n + m + 2;
1540
1541 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1542
1543 return 0;
1544 }
1545
1546 /* Next try MSI
1547 *
1548 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1549 */
1550
1551 if (enic->config.intr_mode < 2 &&
1552 enic->rq_count >= 1 &&
1553 enic->wq_count >= 1 &&
1554 enic->cq_count >= 2 &&
1555 enic->intr_count >= 1 &&
1556 !pci_enable_msi(enic->pdev)) {
1557
1558 enic->rq_count = 1;
1559 enic->wq_count = 1;
1560 enic->cq_count = 2;
1561 enic->intr_count = 1;
1562
1563 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1564
1565 return 0;
1566 }
1567
1568 /* Next try INTx
1569 *
1570 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1571 * (the first INTR is used for WQ/RQ)
1572 * (the second INTR is used for WQ/RQ errors)
1573 * (the last INTR is used for notifications)
1574 */
1575
1576 if (enic->config.intr_mode < 3 &&
1577 enic->rq_count >= 1 &&
1578 enic->wq_count >= 1 &&
1579 enic->cq_count >= 2 &&
1580 enic->intr_count >= 3) {
1581
1582 enic->rq_count = 1;
1583 enic->wq_count = 1;
1584 enic->cq_count = 2;
1585 enic->intr_count = 3;
1586
1587 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1588
1589 return 0;
1590 }
1591
1592 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1593
1594 return -EINVAL;
1595 }
1596
1597 static void enic_clear_intr_mode(struct enic *enic)
1598 {
1599 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1600 case VNIC_DEV_INTR_MODE_MSIX:
1601 pci_disable_msix(enic->pdev);
1602 break;
1603 case VNIC_DEV_INTR_MODE_MSI:
1604 pci_disable_msi(enic->pdev);
1605 break;
1606 default:
1607 break;
1608 }
1609
1610 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1611 }
1612
1613 static void enic_iounmap(struct enic *enic)
1614 {
1615 if (enic->bar0.vaddr)
1616 iounmap(enic->bar0.vaddr);
1617 }
1618
1619 static const struct net_device_ops enic_netdev_ops = {
1620 .ndo_open = enic_open,
1621 .ndo_stop = enic_stop,
1622 .ndo_start_xmit = enic_hard_start_xmit,
1623 .ndo_get_stats = enic_get_stats,
1624 .ndo_validate_addr = eth_validate_addr,
1625 .ndo_set_mac_address = eth_mac_addr,
1626 .ndo_set_multicast_list = enic_set_multicast_list,
1627 .ndo_change_mtu = enic_change_mtu,
1628 .ndo_vlan_rx_register = enic_vlan_rx_register,
1629 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
1630 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
1631 .ndo_tx_timeout = enic_tx_timeout,
1632 #ifdef CONFIG_NET_POLL_CONTROLLER
1633 .ndo_poll_controller = enic_poll_controller,
1634 #endif
1635 };
1636
1637 static int __devinit enic_probe(struct pci_dev *pdev,
1638 const struct pci_device_id *ent)
1639 {
1640 struct net_device *netdev;
1641 struct enic *enic;
1642 int using_dac = 0;
1643 unsigned int i;
1644 int err;
1645
1646 /* Allocate net device structure and initialize. Private
1647 * instance data is initialized to zero.
1648 */
1649
1650 netdev = alloc_etherdev(sizeof(struct enic));
1651 if (!netdev) {
1652 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
1653 return -ENOMEM;
1654 }
1655
1656 pci_set_drvdata(pdev, netdev);
1657
1658 SET_NETDEV_DEV(netdev, &pdev->dev);
1659
1660 enic = netdev_priv(netdev);
1661 enic->netdev = netdev;
1662 enic->pdev = pdev;
1663
1664 /* Setup PCI resources
1665 */
1666
1667 err = pci_enable_device(pdev);
1668 if (err) {
1669 printk(KERN_ERR PFX
1670 "Cannot enable PCI device, aborting.\n");
1671 goto err_out_free_netdev;
1672 }
1673
1674 err = pci_request_regions(pdev, DRV_NAME);
1675 if (err) {
1676 printk(KERN_ERR PFX
1677 "Cannot request PCI regions, aborting.\n");
1678 goto err_out_disable_device;
1679 }
1680
1681 pci_set_master(pdev);
1682
1683 /* Query PCI controller on system for DMA addressing
1684 * limitation for the device. Try 40-bit first, and
1685 * fail to 32-bit.
1686 */
1687
1688 err = pci_set_dma_mask(pdev, DMA_40BIT_MASK);
1689 if (err) {
1690 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1691 if (err) {
1692 printk(KERN_ERR PFX
1693 "No usable DMA configuration, aborting.\n");
1694 goto err_out_release_regions;
1695 }
1696 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
1697 if (err) {
1698 printk(KERN_ERR PFX
1699 "Unable to obtain 32-bit DMA "
1700 "for consistent allocations, aborting.\n");
1701 goto err_out_release_regions;
1702 }
1703 } else {
1704 err = pci_set_consistent_dma_mask(pdev, DMA_40BIT_MASK);
1705 if (err) {
1706 printk(KERN_ERR PFX
1707 "Unable to obtain 40-bit DMA "
1708 "for consistent allocations, aborting.\n");
1709 goto err_out_release_regions;
1710 }
1711 using_dac = 1;
1712 }
1713
1714 /* Map vNIC resources from BAR0
1715 */
1716
1717 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
1718 printk(KERN_ERR PFX
1719 "BAR0 not memory-map'able, aborting.\n");
1720 err = -ENODEV;
1721 goto err_out_release_regions;
1722 }
1723
1724 enic->bar0.vaddr = pci_iomap(pdev, 0, enic->bar0.len);
1725 enic->bar0.bus_addr = pci_resource_start(pdev, 0);
1726 enic->bar0.len = pci_resource_len(pdev, 0);
1727
1728 if (!enic->bar0.vaddr) {
1729 printk(KERN_ERR PFX
1730 "Cannot memory-map BAR0 res hdr, aborting.\n");
1731 err = -ENODEV;
1732 goto err_out_release_regions;
1733 }
1734
1735 /* Register vNIC device
1736 */
1737
1738 enic->vdev = vnic_dev_register(NULL, enic, pdev, &enic->bar0);
1739 if (!enic->vdev) {
1740 printk(KERN_ERR PFX
1741 "vNIC registration failed, aborting.\n");
1742 err = -ENODEV;
1743 goto err_out_iounmap;
1744 }
1745
1746 /* Issue device open to get device in known state
1747 */
1748
1749 err = enic_dev_open(enic);
1750 if (err) {
1751 printk(KERN_ERR PFX
1752 "vNIC dev open failed, aborting.\n");
1753 goto err_out_vnic_unregister;
1754 }
1755
1756 /* Issue device init to initialize the vnic-to-switch link.
1757 * We'll start with carrier off and wait for link UP
1758 * notification later to turn on carrier. We don't need
1759 * to wait here for the vnic-to-switch link initialization
1760 * to complete; link UP notification is the indication that
1761 * the process is complete.
1762 */
1763
1764 netif_carrier_off(netdev);
1765
1766 err = vnic_dev_init(enic->vdev, 0);
1767 if (err) {
1768 printk(KERN_ERR PFX
1769 "vNIC dev init failed, aborting.\n");
1770 goto err_out_dev_close;
1771 }
1772
1773 /* Get vNIC configuration
1774 */
1775
1776 err = enic_get_vnic_config(enic);
1777 if (err) {
1778 printk(KERN_ERR PFX
1779 "Get vNIC configuration failed, aborting.\n");
1780 goto err_out_dev_close;
1781 }
1782
1783 /* Get available resource counts
1784 */
1785
1786 enic_get_res_counts(enic);
1787
1788 /* Set interrupt mode based on resource counts and system
1789 * capabilities
1790 */
1791
1792 err = enic_set_intr_mode(enic);
1793 if (err) {
1794 printk(KERN_ERR PFX
1795 "Failed to set intr mode, aborting.\n");
1796 goto err_out_dev_close;
1797 }
1798
1799 /* Allocate and configure vNIC resources
1800 */
1801
1802 err = enic_alloc_vnic_resources(enic);
1803 if (err) {
1804 printk(KERN_ERR PFX
1805 "Failed to alloc vNIC resources, aborting.\n");
1806 goto err_out_free_vnic_resources;
1807 }
1808
1809 enic_init_vnic_resources(enic);
1810
1811 err = enic_set_niccfg(enic);
1812 if (err) {
1813 printk(KERN_ERR PFX
1814 "Failed to config nic, aborting.\n");
1815 goto err_out_free_vnic_resources;
1816 }
1817
1818 /* Setup notification timer, HW reset task, and locks
1819 */
1820
1821 init_timer(&enic->notify_timer);
1822 enic->notify_timer.function = enic_notify_timer;
1823 enic->notify_timer.data = (unsigned long)enic;
1824
1825 INIT_WORK(&enic->reset, enic_reset);
1826
1827 for (i = 0; i < enic->wq_count; i++)
1828 spin_lock_init(&enic->wq_lock[i]);
1829
1830 spin_lock_init(&enic->devcmd_lock);
1831
1832 /* Register net device
1833 */
1834
1835 enic->port_mtu = enic->config.mtu;
1836 (void)enic_change_mtu(netdev, enic->port_mtu);
1837
1838 err = enic_set_mac_addr(netdev, enic->mac_addr);
1839 if (err) {
1840 printk(KERN_ERR PFX
1841 "Invalid MAC address, aborting.\n");
1842 goto err_out_free_vnic_resources;
1843 }
1844
1845 netdev->netdev_ops = &enic_netdev_ops;
1846 netdev->watchdog_timeo = 2 * HZ;
1847 netdev->ethtool_ops = &enic_ethtool_ops;
1848
1849 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1850 default:
1851 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
1852 break;
1853 case VNIC_DEV_INTR_MODE_MSIX:
1854 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
1855 break;
1856 }
1857
1858 netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
1859 if (ENIC_SETTING(enic, TXCSUM))
1860 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1861 if (ENIC_SETTING(enic, TSO))
1862 netdev->features |= NETIF_F_TSO |
1863 NETIF_F_TSO6 | NETIF_F_TSO_ECN;
1864 if (ENIC_SETTING(enic, LRO))
1865 netdev->features |= NETIF_F_LRO;
1866 if (using_dac)
1867 netdev->features |= NETIF_F_HIGHDMA;
1868
1869 enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
1870
1871 enic->lro_mgr.max_aggr = ENIC_LRO_MAX_AGGR;
1872 enic->lro_mgr.max_desc = ENIC_LRO_MAX_DESC;
1873 enic->lro_mgr.lro_arr = enic->lro_desc;
1874 enic->lro_mgr.get_skb_header = enic_get_skb_header;
1875 enic->lro_mgr.features = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
1876 enic->lro_mgr.dev = netdev;
1877 enic->lro_mgr.ip_summed = CHECKSUM_COMPLETE;
1878 enic->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1879
1880 err = register_netdev(netdev);
1881 if (err) {
1882 printk(KERN_ERR PFX
1883 "Cannot register net device, aborting.\n");
1884 goto err_out_free_vnic_resources;
1885 }
1886
1887 return 0;
1888
1889 err_out_free_vnic_resources:
1890 enic_free_vnic_resources(enic);
1891 err_out_dev_close:
1892 vnic_dev_close(enic->vdev);
1893 err_out_vnic_unregister:
1894 enic_clear_intr_mode(enic);
1895 vnic_dev_unregister(enic->vdev);
1896 err_out_iounmap:
1897 enic_iounmap(enic);
1898 err_out_release_regions:
1899 pci_release_regions(pdev);
1900 err_out_disable_device:
1901 pci_disable_device(pdev);
1902 err_out_free_netdev:
1903 pci_set_drvdata(pdev, NULL);
1904 free_netdev(netdev);
1905
1906 return err;
1907 }
1908
1909 static void __devexit enic_remove(struct pci_dev *pdev)
1910 {
1911 struct net_device *netdev = pci_get_drvdata(pdev);
1912
1913 if (netdev) {
1914 struct enic *enic = netdev_priv(netdev);
1915
1916 flush_scheduled_work();
1917 unregister_netdev(netdev);
1918 enic_free_vnic_resources(enic);
1919 vnic_dev_close(enic->vdev);
1920 enic_clear_intr_mode(enic);
1921 vnic_dev_unregister(enic->vdev);
1922 enic_iounmap(enic);
1923 pci_release_regions(pdev);
1924 pci_disable_device(pdev);
1925 pci_set_drvdata(pdev, NULL);
1926 free_netdev(netdev);
1927 }
1928 }
1929
1930 static struct pci_driver enic_driver = {
1931 .name = DRV_NAME,
1932 .id_table = enic_id_table,
1933 .probe = enic_probe,
1934 .remove = __devexit_p(enic_remove),
1935 };
1936
1937 static int __init enic_init_module(void)
1938 {
1939 printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
1940
1941 return pci_register_driver(&enic_driver);
1942 }
1943
1944 static void __exit enic_cleanup_module(void)
1945 {
1946 pci_unregister_driver(&enic_driver);
1947 }
1948
1949 module_init(enic_init_module);
1950 module_exit(enic_cleanup_module);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree