Merge branch 'core-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6] / 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                 enic_notify_check(enic);
405
406         if (ENIC_TEST_INTR(pba, ENIC_INTX_ERR)) {
407                 enic_log_q_error(enic);
408                 /* schedule recovery from WQ/RQ error */
409                 schedule_work(&enic->reset);
410                 return IRQ_HANDLED;
411         }
412
413         if (ENIC_TEST_INTR(pba, ENIC_INTX_WQ_RQ)) {
414                 if (netif_rx_schedule_prep(&enic->napi))
415                         __netif_rx_schedule(&enic->napi);
416         } else {
417                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
418         }
419
420         return IRQ_HANDLED;
421 }
422
423 static irqreturn_t enic_isr_msi(int irq, void *data)
424 {
425         struct enic *enic = data;
426
427         /* With MSI, there is no sharing of interrupts, so this is
428          * our interrupt and there is no need to ack it.  The device
429          * is not providing per-vector masking, so the OS will not
430          * write to PCI config space to mask/unmask the interrupt.
431          * We're using mask_on_assertion for MSI, so the device
432          * automatically masks the interrupt when the interrupt is
433          * generated.  Later, when exiting polling, the interrupt
434          * will be unmasked (see enic_poll).
435          *
436          * Also, the device uses the same PCIe Traffic Class (TC)
437          * for Memory Write data and MSI, so there are no ordering
438          * issues; the MSI will always arrive at the Root Complex
439          * _after_ corresponding Memory Writes (i.e. descriptor
440          * writes).
441          */
442
443         netif_rx_schedule(&enic->napi);
444
445         return IRQ_HANDLED;
446 }
447
448 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
449 {
450         struct enic *enic = data;
451
452         /* schedule NAPI polling for RQ cleanup */
453         netif_rx_schedule(&enic->napi);
454
455         return IRQ_HANDLED;
456 }
457
458 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
459 {
460         struct enic *enic = data;
461         unsigned int wq_work_to_do = -1; /* no limit */
462         unsigned int wq_work_done;
463
464         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
465                 wq_work_to_do, enic_wq_service, NULL);
466
467         vnic_intr_return_credits(&enic->intr[ENIC_MSIX_WQ],
468                 wq_work_done,
469                 1 /* unmask intr */,
470                 1 /* reset intr timer */);
471
472         return IRQ_HANDLED;
473 }
474
475 static irqreturn_t enic_isr_msix_err(int irq, void *data)
476 {
477         struct enic *enic = data;
478
479         enic_log_q_error(enic);
480
481         /* schedule recovery from WQ/RQ error */
482         schedule_work(&enic->reset);
483
484         return IRQ_HANDLED;
485 }
486
487 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
488 {
489         struct enic *enic = data;
490
491         enic_notify_check(enic);
492         vnic_intr_unmask(&enic->intr[ENIC_MSIX_NOTIFY]);
493
494         return IRQ_HANDLED;
495 }
496
497 static inline void enic_queue_wq_skb_cont(struct enic *enic,
498         struct vnic_wq *wq, struct sk_buff *skb,
499         unsigned int len_left)
500 {
501         skb_frag_t *frag;
502
503         /* Queue additional data fragments */
504         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
505                 len_left -= frag->size;
506                 enic_queue_wq_desc_cont(wq, skb,
507                         pci_map_page(enic->pdev, frag->page,
508                                 frag->page_offset, frag->size,
509                                 PCI_DMA_TODEVICE),
510                         frag->size,
511                         (len_left == 0));       /* EOP? */
512         }
513 }
514
515 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
516         struct vnic_wq *wq, struct sk_buff *skb,
517         int vlan_tag_insert, unsigned int vlan_tag)
518 {
519         unsigned int head_len = skb_headlen(skb);
520         unsigned int len_left = skb->len - head_len;
521         int eop = (len_left == 0);
522
523         /* Queue the main skb fragment */
524         enic_queue_wq_desc(wq, skb,
525                 pci_map_single(enic->pdev, skb->data,
526                         head_len, PCI_DMA_TODEVICE),
527                 head_len,
528                 vlan_tag_insert, vlan_tag,
529                 eop);
530
531         if (!eop)
532                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
533 }
534
535 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
536         struct vnic_wq *wq, struct sk_buff *skb,
537         int vlan_tag_insert, unsigned int vlan_tag)
538 {
539         unsigned int head_len = skb_headlen(skb);
540         unsigned int len_left = skb->len - head_len;
541         unsigned int hdr_len = skb_transport_offset(skb);
542         unsigned int csum_offset = hdr_len + skb->csum_offset;
543         int eop = (len_left == 0);
544
545         /* Queue the main skb fragment */
546         enic_queue_wq_desc_csum_l4(wq, skb,
547                 pci_map_single(enic->pdev, skb->data,
548                         head_len, PCI_DMA_TODEVICE),
549                 head_len,
550                 csum_offset,
551                 hdr_len,
552                 vlan_tag_insert, vlan_tag,
553                 eop);
554
555         if (!eop)
556                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
557 }
558
559 static inline void enic_queue_wq_skb_tso(struct enic *enic,
560         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
561         int vlan_tag_insert, unsigned int vlan_tag)
562 {
563         unsigned int head_len = skb_headlen(skb);
564         unsigned int len_left = skb->len - head_len;
565         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
566         int eop = (len_left == 0);
567
568         /* Preload TCP csum field with IP pseudo hdr calculated
569          * with IP length set to zero.  HW will later add in length
570          * to each TCP segment resulting from the TSO.
571          */
572
573         if (skb->protocol == __constant_htons(ETH_P_IP)) {
574                 ip_hdr(skb)->check = 0;
575                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
576                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
577         } else if (skb->protocol == __constant_htons(ETH_P_IPV6)) {
578                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
579                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
580         }
581
582         /* Queue the main skb fragment */
583         enic_queue_wq_desc_tso(wq, skb,
584                 pci_map_single(enic->pdev, skb->data,
585                         head_len, PCI_DMA_TODEVICE),
586                 head_len,
587                 mss, hdr_len,
588                 vlan_tag_insert, vlan_tag,
589                 eop);
590
591         if (!eop)
592                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
593 }
594
595 static inline void enic_queue_wq_skb(struct enic *enic,
596         struct vnic_wq *wq, struct sk_buff *skb)
597 {
598         unsigned int mss = skb_shinfo(skb)->gso_size;
599         unsigned int vlan_tag = 0;
600         int vlan_tag_insert = 0;
601
602         if (enic->vlan_group && vlan_tx_tag_present(skb)) {
603                 /* VLAN tag from trunking driver */
604                 vlan_tag_insert = 1;
605                 vlan_tag = vlan_tx_tag_get(skb);
606         }
607
608         if (mss)
609                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
610                         vlan_tag_insert, vlan_tag);
611         else if (skb->ip_summed == CHECKSUM_PARTIAL)
612                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
613                         vlan_tag_insert, vlan_tag);
614         else
615                 enic_queue_wq_skb_vlan(enic, wq, skb,
616                         vlan_tag_insert, vlan_tag);
617 }
618
619 /* netif_tx_lock held, process context with BHs disabled */
620 static int enic_hard_start_xmit(struct sk_buff *skb, struct net_device *netdev)
621 {
622         struct enic *enic = netdev_priv(netdev);
623         struct vnic_wq *wq = &enic->wq[0];
624         unsigned long flags;
625
626         if (skb->len <= 0) {
627                 dev_kfree_skb(skb);
628                 return NETDEV_TX_OK;
629         }
630
631         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
632          * which is very likely.  In the off chance it's going to take
633          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
634          */
635
636         if (skb_shinfo(skb)->gso_size == 0 &&
637             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
638             skb_linearize(skb)) {
639                 dev_kfree_skb(skb);
640                 return NETDEV_TX_OK;
641         }
642
643         spin_lock_irqsave(&enic->wq_lock[0], flags);
644
645         if (vnic_wq_desc_avail(wq) < skb_shinfo(skb)->nr_frags + 1) {
646                 netif_stop_queue(netdev);
647                 /* This is a hard error, log it */
648                 printk(KERN_ERR PFX "%s: BUG! Tx ring full when "
649                         "queue awake!\n", netdev->name);
650                 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
651                 return NETDEV_TX_BUSY;
652         }
653
654         enic_queue_wq_skb(enic, wq, skb);
655
656         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + 1)
657                 netif_stop_queue(netdev);
658
659         netdev->trans_start = jiffies;
660
661         spin_unlock_irqrestore(&enic->wq_lock[0], flags);
662
663         return NETDEV_TX_OK;
664 }
665
666 /* dev_base_lock rwlock held, nominally process context */
667 static struct net_device_stats *enic_get_stats(struct net_device *netdev)
668 {
669         struct enic *enic = netdev_priv(netdev);
670         struct net_device_stats *net_stats = &netdev->stats;
671         struct vnic_stats *stats;
672
673         spin_lock(&enic->devcmd_lock);
674         vnic_dev_stats_dump(enic->vdev, &stats);
675         spin_unlock(&enic->devcmd_lock);
676
677         net_stats->tx_packets = stats->tx.tx_frames_ok;
678         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
679         net_stats->tx_errors = stats->tx.tx_errors;
680         net_stats->tx_dropped = stats->tx.tx_drops;
681
682         net_stats->rx_packets = stats->rx.rx_frames_ok;
683         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
684         net_stats->rx_errors = stats->rx.rx_errors;
685         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
686         net_stats->rx_crc_errors = stats->rx.rx_crc_errors;
687         net_stats->rx_dropped = stats->rx.rx_no_bufs;
688
689         return net_stats;
690 }
691
692 static void enic_reset_mcaddrs(struct enic *enic)
693 {
694         enic->mc_count = 0;
695 }
696
697 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
698 {
699         if (!is_valid_ether_addr(addr))
700                 return -EADDRNOTAVAIL;
701
702         memcpy(netdev->dev_addr, addr, netdev->addr_len);
703
704         return 0;
705 }
706
707 /* netif_tx_lock held, BHs disabled */
708 static void enic_set_multicast_list(struct net_device *netdev)
709 {
710         struct enic *enic = netdev_priv(netdev);
711         struct dev_mc_list *list = netdev->mc_list;
712         int directed = 1;
713         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
714         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
715         int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
716         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
717             (netdev->mc_count > ENIC_MULTICAST_PERFECT_FILTERS);
718         u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
719         unsigned int mc_count = netdev->mc_count;
720         unsigned int i, j;
721
722         if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
723                 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
724
725         spin_lock(&enic->devcmd_lock);
726
727         vnic_dev_packet_filter(enic->vdev, directed,
728                 multicast, broadcast, promisc, allmulti);
729
730         /* Is there an easier way?  Trying to minimize to
731          * calls to add/del multicast addrs.  We keep the
732          * addrs from the last call in enic->mc_addr and
733          * look for changes to add/del.
734          */
735
736         for (i = 0; list && i < mc_count; i++) {
737                 memcpy(mc_addr[i], list->dmi_addr, ETH_ALEN);
738                 list = list->next;
739         }
740
741         for (i = 0; i < enic->mc_count; i++) {
742                 for (j = 0; j < mc_count; j++)
743                         if (compare_ether_addr(enic->mc_addr[i],
744                                 mc_addr[j]) == 0)
745                                 break;
746                 if (j == mc_count)
747                         enic_del_multicast_addr(enic, enic->mc_addr[i]);
748         }
749
750         for (i = 0; i < mc_count; i++) {
751                 for (j = 0; j < enic->mc_count; j++)
752                         if (compare_ether_addr(mc_addr[i],
753                                 enic->mc_addr[j]) == 0)
754                                 break;
755                 if (j == enic->mc_count)
756                         enic_add_multicast_addr(enic, mc_addr[i]);
757         }
758
759         /* Save the list to compare against next time
760          */
761
762         for (i = 0; i < mc_count; i++)
763                 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
764
765         enic->mc_count = mc_count;
766
767         spin_unlock(&enic->devcmd_lock);
768 }
769
770 /* rtnl lock is held */
771 static void enic_vlan_rx_register(struct net_device *netdev,
772         struct vlan_group *vlan_group)
773 {
774         struct enic *enic = netdev_priv(netdev);
775         enic->vlan_group = vlan_group;
776 }
777
778 /* rtnl lock is held */
779 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
780 {
781         struct enic *enic = netdev_priv(netdev);
782
783         spin_lock(&enic->devcmd_lock);
784         enic_add_vlan(enic, vid);
785         spin_unlock(&enic->devcmd_lock);
786 }
787
788 /* rtnl lock is held */
789 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
790 {
791         struct enic *enic = netdev_priv(netdev);
792
793         spin_lock(&enic->devcmd_lock);
794         enic_del_vlan(enic, vid);
795         spin_unlock(&enic->devcmd_lock);
796 }
797
798 /* netif_tx_lock held, BHs disabled */
799 static void enic_tx_timeout(struct net_device *netdev)
800 {
801         struct enic *enic = netdev_priv(netdev);
802         schedule_work(&enic->reset);
803 }
804
805 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
806 {
807         struct enic *enic = vnic_dev_priv(rq->vdev);
808
809         if (!buf->os_buf)
810                 return;
811
812         pci_unmap_single(enic->pdev, buf->dma_addr,
813                 buf->len, PCI_DMA_FROMDEVICE);
814         dev_kfree_skb_any(buf->os_buf);
815 }
816
817 static inline struct sk_buff *enic_rq_alloc_skb(unsigned int size)
818 {
819         struct sk_buff *skb;
820
821         skb = dev_alloc_skb(size + NET_IP_ALIGN);
822
823         if (skb)
824                 skb_reserve(skb, NET_IP_ALIGN);
825
826         return skb;
827 }
828
829 static int enic_rq_alloc_buf(struct vnic_rq *rq)
830 {
831         struct enic *enic = vnic_dev_priv(rq->vdev);
832         struct sk_buff *skb;
833         unsigned int len = enic->netdev->mtu + ETH_HLEN;
834         unsigned int os_buf_index = 0;
835         dma_addr_t dma_addr;
836
837         skb = enic_rq_alloc_skb(len);
838         if (!skb)
839                 return -ENOMEM;
840
841         dma_addr = pci_map_single(enic->pdev, skb->data,
842                 len, PCI_DMA_FROMDEVICE);
843
844         enic_queue_rq_desc(rq, skb, os_buf_index,
845                 dma_addr, len);
846
847         return 0;
848 }
849
850 static int enic_get_skb_header(struct sk_buff *skb, void **iphdr,
851         void **tcph, u64 *hdr_flags, void *priv)
852 {
853         struct cq_enet_rq_desc *cq_desc = priv;
854         unsigned int ip_len;
855         struct iphdr *iph;
856
857         u8 type, color, eop, sop, ingress_port, vlan_stripped;
858         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
859         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
860         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
861         u8 packet_error;
862         u16 q_number, completed_index, bytes_written, vlan, checksum;
863         u32 rss_hash;
864
865         cq_enet_rq_desc_dec(cq_desc,
866                 &type, &color, &q_number, &completed_index,
867                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
868                 &csum_not_calc, &rss_hash, &bytes_written,
869                 &packet_error, &vlan_stripped, &vlan, &checksum,
870                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
871                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
872                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
873                 &fcs_ok);
874
875         if (!(ipv4 && tcp && !ipv4_fragment))
876                 return -1;
877
878         skb_reset_network_header(skb);
879         iph = ip_hdr(skb);
880
881         ip_len = ip_hdrlen(skb);
882         skb_set_transport_header(skb, ip_len);
883
884         /* check if ip header and tcp header are complete */
885         if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb))
886                 return -1;
887
888         *hdr_flags = LRO_IPV4 | LRO_TCP;
889         *tcph = tcp_hdr(skb);
890         *iphdr = iph;
891
892         return 0;
893 }
894
895 static void enic_rq_indicate_buf(struct vnic_rq *rq,
896         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
897         int skipped, void *opaque)
898 {
899         struct enic *enic = vnic_dev_priv(rq->vdev);
900         struct net_device *netdev = enic->netdev;
901         struct sk_buff *skb;
902
903         u8 type, color, eop, sop, ingress_port, vlan_stripped;
904         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
905         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
906         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
907         u8 packet_error;
908         u16 q_number, completed_index, bytes_written, vlan, checksum;
909         u32 rss_hash;
910
911         if (skipped)
912                 return;
913
914         skb = buf->os_buf;
915         prefetch(skb->data - NET_IP_ALIGN);
916         pci_unmap_single(enic->pdev, buf->dma_addr,
917                 buf->len, PCI_DMA_FROMDEVICE);
918
919         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
920                 &type, &color, &q_number, &completed_index,
921                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
922                 &csum_not_calc, &rss_hash, &bytes_written,
923                 &packet_error, &vlan_stripped, &vlan, &checksum,
924                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
925                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
926                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
927                 &fcs_ok);
928
929         if (packet_error) {
930
931                 if (bytes_written > 0 && !fcs_ok) {
932                         if (net_ratelimit())
933                                 printk(KERN_ERR PFX
934                                         "%s: packet error: bad FCS\n",
935                                         netdev->name);
936                 }
937
938                 dev_kfree_skb_any(skb);
939
940                 return;
941         }
942
943         if (eop && bytes_written > 0) {
944
945                 /* Good receive
946                  */
947
948                 skb_put(skb, bytes_written);
949                 skb->protocol = eth_type_trans(skb, netdev);
950
951                 if (enic->csum_rx_enabled && !csum_not_calc) {
952                         skb->csum = htons(checksum);
953                         skb->ip_summed = CHECKSUM_COMPLETE;
954                 }
955
956                 skb->dev = netdev;
957
958                 if (enic->vlan_group && vlan_stripped) {
959
960                         if ((netdev->features & NETIF_F_LRO) && ipv4)
961                                 lro_vlan_hwaccel_receive_skb(&enic->lro_mgr,
962                                         skb, enic->vlan_group,
963                                         vlan, cq_desc);
964                         else
965                                 vlan_hwaccel_receive_skb(skb,
966                                         enic->vlan_group, vlan);
967
968                 } else {
969
970                         if ((netdev->features & NETIF_F_LRO) && ipv4)
971                                 lro_receive_skb(&enic->lro_mgr, skb, cq_desc);
972                         else
973                                 netif_receive_skb(skb);
974
975                 }
976
977         } else {
978
979                 /* Buffer overflow
980                  */
981
982                 dev_kfree_skb_any(skb);
983         }
984 }
985
986 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
987         u8 type, u16 q_number, u16 completed_index, void *opaque)
988 {
989         struct enic *enic = vnic_dev_priv(vdev);
990
991         vnic_rq_service(&enic->rq[q_number], cq_desc,
992                 completed_index, VNIC_RQ_RETURN_DESC,
993                 enic_rq_indicate_buf, opaque);
994
995         return 0;
996 }
997
998 static void enic_rq_drop_buf(struct vnic_rq *rq,
999         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1000         int skipped, void *opaque)
1001 {
1002         struct enic *enic = vnic_dev_priv(rq->vdev);
1003         struct sk_buff *skb = buf->os_buf;
1004
1005         if (skipped)
1006                 return;
1007
1008         pci_unmap_single(enic->pdev, buf->dma_addr,
1009                 buf->len, PCI_DMA_FROMDEVICE);
1010
1011         dev_kfree_skb_any(skb);
1012 }
1013
1014 static int enic_rq_service_drop(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1015         u8 type, u16 q_number, u16 completed_index, void *opaque)
1016 {
1017         struct enic *enic = vnic_dev_priv(vdev);
1018
1019         vnic_rq_service(&enic->rq[q_number], cq_desc,
1020                 completed_index, VNIC_RQ_RETURN_DESC,
1021                 enic_rq_drop_buf, opaque);
1022
1023         return 0;
1024 }
1025
1026 static int enic_poll(struct napi_struct *napi, int budget)
1027 {
1028         struct enic *enic = container_of(napi, struct enic, napi);
1029         struct net_device *netdev = enic->netdev;
1030         unsigned int rq_work_to_do = budget;
1031         unsigned int wq_work_to_do = -1; /* no limit */
1032         unsigned int  work_done, rq_work_done, wq_work_done;
1033
1034         /* Service RQ (first) and WQ
1035          */
1036
1037         rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1038                 rq_work_to_do, enic_rq_service, NULL);
1039
1040         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1041                 wq_work_to_do, enic_wq_service, NULL);
1042
1043         /* Accumulate intr event credits for this polling
1044          * cycle.  An intr event is the completion of a
1045          * a WQ or RQ packet.
1046          */
1047
1048         work_done = rq_work_done + wq_work_done;
1049
1050         if (work_done > 0)
1051                 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1052                         work_done,
1053                         0 /* don't unmask intr */,
1054                         0 /* don't reset intr timer */);
1055
1056         if (rq_work_done > 0) {
1057
1058                 /* Replenish RQ
1059                  */
1060
1061                 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1062
1063         } else {
1064
1065                 /* If no work done, flush all LROs and exit polling
1066                  */
1067
1068                 if (netdev->features & NETIF_F_LRO)
1069                         lro_flush_all(&enic->lro_mgr);
1070
1071                 netif_rx_complete(napi);
1072                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1073         }
1074
1075         return rq_work_done;
1076 }
1077
1078 static int enic_poll_msix(struct napi_struct *napi, int budget)
1079 {
1080         struct enic *enic = container_of(napi, struct enic, napi);
1081         struct net_device *netdev = enic->netdev;
1082         unsigned int work_to_do = budget;
1083         unsigned int work_done;
1084
1085         /* Service RQ
1086          */
1087
1088         work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1089                 work_to_do, enic_rq_service, NULL);
1090
1091         if (work_done > 0) {
1092
1093                 /* Replenish RQ
1094                  */
1095
1096                 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1097
1098                 /* Accumulate intr event credits for this polling
1099                  * cycle.  An intr event is the completion of a
1100                  * a WQ or RQ packet.
1101                  */
1102
1103                 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1104                         work_done,
1105                         0 /* don't unmask intr */,
1106                         0 /* don't reset intr timer */);
1107         } else {
1108
1109                 /* If no work done, flush all LROs and exit polling
1110                  */
1111
1112                 if (netdev->features & NETIF_F_LRO)
1113                         lro_flush_all(&enic->lro_mgr);
1114
1115                 netif_rx_complete(napi);
1116                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1117         }
1118
1119         return work_done;
1120 }
1121
1122 static void enic_notify_timer(unsigned long data)
1123 {
1124         struct enic *enic = (struct enic *)data;
1125
1126         enic_notify_check(enic);
1127
1128         mod_timer(&enic->notify_timer,
1129                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1130 }
1131
1132 static void enic_free_intr(struct enic *enic)
1133 {
1134         struct net_device *netdev = enic->netdev;
1135         unsigned int i;
1136
1137         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1138         case VNIC_DEV_INTR_MODE_INTX:
1139                 free_irq(enic->pdev->irq, netdev);
1140                 break;
1141         case VNIC_DEV_INTR_MODE_MSI:
1142                 free_irq(enic->pdev->irq, enic);
1143                 break;
1144         case VNIC_DEV_INTR_MODE_MSIX:
1145                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1146                         if (enic->msix[i].requested)
1147                                 free_irq(enic->msix_entry[i].vector,
1148                                         enic->msix[i].devid);
1149                 break;
1150         default:
1151                 break;
1152         }
1153 }
1154
1155 static int enic_request_intr(struct enic *enic)
1156 {
1157         struct net_device *netdev = enic->netdev;
1158         unsigned int i;
1159         int err = 0;
1160
1161         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1162
1163         case VNIC_DEV_INTR_MODE_INTX:
1164
1165                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1166                         IRQF_SHARED, netdev->name, netdev);
1167                 break;
1168
1169         case VNIC_DEV_INTR_MODE_MSI:
1170
1171                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1172                         0, netdev->name, enic);
1173                 break;
1174
1175         case VNIC_DEV_INTR_MODE_MSIX:
1176
1177                 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1178                         "%.11s-rx-0", netdev->name);
1179                 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1180                 enic->msix[ENIC_MSIX_RQ].devid = enic;
1181
1182                 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1183                         "%.11s-tx-0", netdev->name);
1184                 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1185                 enic->msix[ENIC_MSIX_WQ].devid = enic;
1186
1187                 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1188                         "%.11s-err", netdev->name);
1189                 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1190                 enic->msix[ENIC_MSIX_ERR].devid = enic;
1191
1192                 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1193                         "%.11s-notify", netdev->name);
1194                 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1195                 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1196
1197                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1198                         err = request_irq(enic->msix_entry[i].vector,
1199                                 enic->msix[i].isr, 0,
1200                                 enic->msix[i].devname,
1201                                 enic->msix[i].devid);
1202                         if (err) {
1203                                 enic_free_intr(enic);
1204                                 break;
1205                         }
1206                         enic->msix[i].requested = 1;
1207                 }
1208
1209                 break;
1210
1211         default:
1212                 break;
1213         }
1214
1215         return err;
1216 }
1217
1218 static int enic_notify_set(struct enic *enic)
1219 {
1220         int err;
1221
1222         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1223         case VNIC_DEV_INTR_MODE_INTX:
1224                 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1225                 break;
1226         case VNIC_DEV_INTR_MODE_MSIX:
1227                 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1228                 break;
1229         default:
1230                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1231                 break;
1232         }
1233
1234         return err;
1235 }
1236
1237 static void enic_notify_timer_start(struct enic *enic)
1238 {
1239         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1240         case VNIC_DEV_INTR_MODE_MSI:
1241                 mod_timer(&enic->notify_timer, jiffies);
1242                 break;
1243         default:
1244                 /* Using intr for notification for INTx/MSI-X */
1245                 break;
1246         };
1247 }
1248
1249 /* rtnl lock is held, process context */
1250 static int enic_open(struct net_device *netdev)
1251 {
1252         struct enic *enic = netdev_priv(netdev);
1253         unsigned int i;
1254         int err;
1255
1256         err = enic_request_intr(enic);
1257         if (err) {
1258                 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1259                         netdev->name);
1260                 return err;
1261         }
1262
1263         err = enic_notify_set(enic);
1264         if (err) {
1265                 printk(KERN_ERR PFX
1266                         "%s: Failed to alloc notify buffer, aborting.\n",
1267                         netdev->name);
1268                 goto err_out_free_intr;
1269         }
1270
1271         for (i = 0; i < enic->rq_count; i++) {
1272                 err = vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1273                 if (err) {
1274                         printk(KERN_ERR PFX
1275                                 "%s: Unable to alloc receive buffers.\n",
1276                                 netdev->name);
1277                         goto err_out_notify_unset;
1278                 }
1279         }
1280
1281         for (i = 0; i < enic->wq_count; i++)
1282                 vnic_wq_enable(&enic->wq[i]);
1283         for (i = 0; i < enic->rq_count; i++)
1284                 vnic_rq_enable(&enic->rq[i]);
1285
1286         enic_add_station_addr(enic);
1287         enic_set_multicast_list(netdev);
1288
1289         netif_wake_queue(netdev);
1290         napi_enable(&enic->napi);
1291         vnic_dev_enable(enic->vdev);
1292
1293         for (i = 0; i < enic->intr_count; i++)
1294                 vnic_intr_unmask(&enic->intr[i]);
1295
1296         enic_notify_timer_start(enic);
1297
1298         return 0;
1299
1300 err_out_notify_unset:
1301         vnic_dev_notify_unset(enic->vdev);
1302 err_out_free_intr:
1303         enic_free_intr(enic);
1304
1305         return err;
1306 }
1307
1308 /* rtnl lock is held, process context */
1309 static int enic_stop(struct net_device *netdev)
1310 {
1311         struct enic *enic = netdev_priv(netdev);
1312         unsigned int i;
1313         int err;
1314
1315         del_timer_sync(&enic->notify_timer);
1316
1317         vnic_dev_disable(enic->vdev);
1318         napi_disable(&enic->napi);
1319         netif_stop_queue(netdev);
1320
1321         for (i = 0; i < enic->intr_count; i++)
1322                 vnic_intr_mask(&enic->intr[i]);
1323
1324         for (i = 0; i < enic->wq_count; i++) {
1325                 err = vnic_wq_disable(&enic->wq[i]);
1326                 if (err)
1327                         return err;
1328         }
1329         for (i = 0; i < enic->rq_count; i++) {
1330                 err = vnic_rq_disable(&enic->rq[i]);
1331                 if (err)
1332                         return err;
1333         }
1334
1335         vnic_dev_notify_unset(enic->vdev);
1336         enic_free_intr(enic);
1337
1338         (void)vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1339                 -1, enic_rq_service_drop, NULL);
1340         (void)vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1341                 -1, enic_wq_service, NULL);
1342
1343         for (i = 0; i < enic->wq_count; i++)
1344                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1345         for (i = 0; i < enic->rq_count; i++)
1346                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1347         for (i = 0; i < enic->cq_count; i++)
1348                 vnic_cq_clean(&enic->cq[i]);
1349         for (i = 0; i < enic->intr_count; i++)
1350                 vnic_intr_clean(&enic->intr[i]);
1351
1352         return 0;
1353 }
1354
1355 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1356 {
1357         struct enic *enic = netdev_priv(netdev);
1358         int running = netif_running(netdev);
1359
1360         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1361                 return -EINVAL;
1362
1363         if (running)
1364                 enic_stop(netdev);
1365
1366         netdev->mtu = new_mtu;
1367
1368         if (netdev->mtu > enic->port_mtu)
1369                 printk(KERN_WARNING PFX
1370                         "%s: interface MTU (%d) set higher "
1371                         "than port MTU (%d)\n",
1372                         netdev->name, netdev->mtu, enic->port_mtu);
1373
1374         if (running)
1375                 enic_open(netdev);
1376
1377         return 0;
1378 }
1379
1380 #ifdef CONFIG_NET_POLL_CONTROLLER
1381 static void enic_poll_controller(struct net_device *netdev)
1382 {
1383         struct enic *enic = netdev_priv(netdev);
1384         struct vnic_dev *vdev = enic->vdev;
1385
1386         switch (vnic_dev_get_intr_mode(vdev)) {
1387         case VNIC_DEV_INTR_MODE_MSIX:
1388                 enic_isr_msix_rq(enic->pdev->irq, enic);
1389                 enic_isr_msix_wq(enic->pdev->irq, enic);
1390                 break;
1391         case VNIC_DEV_INTR_MODE_MSI:
1392                 enic_isr_msi(enic->pdev->irq, enic);
1393                 break;
1394         case VNIC_DEV_INTR_MODE_INTX:
1395                 enic_isr_legacy(enic->pdev->irq, netdev);
1396                 break;
1397         default:
1398                 break;
1399         }
1400 }
1401 #endif
1402
1403 static int enic_dev_wait(struct vnic_dev *vdev,
1404         int (*start)(struct vnic_dev *, int),
1405         int (*finished)(struct vnic_dev *, int *),
1406         int arg)
1407 {
1408         unsigned long time;
1409         int done;
1410         int err;
1411
1412         BUG_ON(in_interrupt());
1413
1414         err = start(vdev, arg);
1415         if (err)
1416                 return err;
1417
1418         /* Wait for func to complete...2 seconds max
1419          */
1420
1421         time = jiffies + (HZ * 2);
1422         do {
1423
1424                 err = finished(vdev, &done);
1425                 if (err)
1426                         return err;
1427
1428                 if (done)
1429                         return 0;
1430
1431                 schedule_timeout_uninterruptible(HZ / 10);
1432
1433         } while (time_after(time, jiffies));
1434
1435         return -ETIMEDOUT;
1436 }
1437
1438 static int enic_dev_open(struct enic *enic)
1439 {
1440         int err;
1441
1442         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1443                 vnic_dev_open_done, 0);
1444         if (err)
1445                 printk(KERN_ERR PFX
1446                         "vNIC device open failed, err %d.\n", err);
1447
1448         return err;
1449 }
1450
1451 static int enic_dev_soft_reset(struct enic *enic)
1452 {
1453         int err;
1454
1455         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1456                 vnic_dev_soft_reset_done, 0);
1457         if (err)
1458                 printk(KERN_ERR PFX
1459                         "vNIC soft reset failed, err %d.\n", err);
1460
1461         return err;
1462 }
1463
1464 static void enic_reset(struct work_struct *work)
1465 {
1466         struct enic *enic = container_of(work, struct enic, reset);
1467
1468         if (!netif_running(enic->netdev))
1469                 return;
1470
1471         rtnl_lock();
1472
1473         spin_lock(&enic->devcmd_lock);
1474         vnic_dev_hang_notify(enic->vdev);
1475         spin_unlock(&enic->devcmd_lock);
1476
1477         enic_stop(enic->netdev);
1478         enic_dev_soft_reset(enic);
1479         enic_reset_mcaddrs(enic);
1480         enic_init_vnic_resources(enic);
1481         enic_open(enic->netdev);
1482
1483         rtnl_unlock();
1484 }
1485
1486 static int enic_set_intr_mode(struct enic *enic)
1487 {
1488         unsigned int n = ARRAY_SIZE(enic->rq);
1489         unsigned int m = ARRAY_SIZE(enic->wq);
1490         unsigned int i;
1491
1492         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1493          * system capabilities.
1494          *
1495          * Try MSI-X first
1496          *
1497          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1498          * (the second to last INTR is used for WQ/RQ errors)
1499          * (the last INTR is used for notifications)
1500          */
1501
1502         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1503         for (i = 0; i < n + m + 2; i++)
1504                 enic->msix_entry[i].entry = i;
1505
1506         if (enic->config.intr_mode < 1 &&
1507             enic->rq_count >= n &&
1508             enic->wq_count >= m &&
1509             enic->cq_count >= n + m &&
1510             enic->intr_count >= n + m + 2 &&
1511             !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1512
1513                 enic->rq_count = n;
1514                 enic->wq_count = m;
1515                 enic->cq_count = n + m;
1516                 enic->intr_count = n + m + 2;
1517
1518                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1519
1520                 return 0;
1521         }
1522
1523         /* Next try MSI
1524          *
1525          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1526          */
1527
1528         if (enic->config.intr_mode < 2 &&
1529             enic->rq_count >= 1 &&
1530             enic->wq_count >= 1 &&
1531             enic->cq_count >= 2 &&
1532             enic->intr_count >= 1 &&
1533             !pci_enable_msi(enic->pdev)) {
1534
1535                 enic->rq_count = 1;
1536                 enic->wq_count = 1;
1537                 enic->cq_count = 2;
1538                 enic->intr_count = 1;
1539
1540                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1541
1542                 return 0;
1543         }
1544
1545         /* Next try INTx
1546          *
1547          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1548          * (the first INTR is used for WQ/RQ)
1549          * (the second INTR is used for WQ/RQ errors)
1550          * (the last INTR is used for notifications)
1551          */
1552
1553         if (enic->config.intr_mode < 3 &&
1554             enic->rq_count >= 1 &&
1555             enic->wq_count >= 1 &&
1556             enic->cq_count >= 2 &&
1557             enic->intr_count >= 3) {
1558
1559                 enic->rq_count = 1;
1560                 enic->wq_count = 1;
1561                 enic->cq_count = 2;
1562                 enic->intr_count = 3;
1563
1564                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1565
1566                 return 0;
1567         }
1568
1569         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1570
1571         return -EINVAL;
1572 }
1573
1574 static void enic_clear_intr_mode(struct enic *enic)
1575 {
1576         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1577         case VNIC_DEV_INTR_MODE_MSIX:
1578                 pci_disable_msix(enic->pdev);
1579                 break;
1580         case VNIC_DEV_INTR_MODE_MSI:
1581                 pci_disable_msi(enic->pdev);
1582                 break;
1583         default:
1584                 break;
1585         }
1586
1587         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1588 }
1589
1590 static void enic_iounmap(struct enic *enic)
1591 {
1592         if (enic->bar0.vaddr)
1593                 iounmap(enic->bar0.vaddr);
1594 }
1595
1596 static const struct net_device_ops enic_netdev_ops = {
1597         .ndo_open               = enic_open,
1598         .ndo_stop               = enic_stop,
1599         .ndo_start_xmit         = enic_hard_start_xmit,
1600         .ndo_get_stats          = enic_get_stats,
1601         .ndo_validate_addr      = eth_validate_addr,
1602         .ndo_set_mac_address    = eth_mac_addr,
1603         .ndo_set_multicast_list = enic_set_multicast_list,
1604         .ndo_change_mtu         = enic_change_mtu,
1605         .ndo_vlan_rx_register   = enic_vlan_rx_register,
1606         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
1607         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
1608         .ndo_tx_timeout         = enic_tx_timeout,
1609 #ifdef CONFIG_NET_POLL_CONTROLLER
1610         .ndo_poll_controller    = enic_poll_controller,
1611 #endif
1612 };
1613
1614 static int __devinit enic_probe(struct pci_dev *pdev,
1615         const struct pci_device_id *ent)
1616 {
1617         struct net_device *netdev;
1618         struct enic *enic;
1619         int using_dac = 0;
1620         unsigned int i;
1621         int err;
1622
1623         const u8 rss_default_cpu = 0;
1624         const u8 rss_hash_type = 0;
1625         const u8 rss_hash_bits = 0;
1626         const u8 rss_base_cpu = 0;
1627         const u8 rss_enable = 0;
1628         const u8 tso_ipid_split_en = 0;
1629         const u8 ig_vlan_strip_en = 1;
1630
1631         /* Allocate net device structure and initialize.  Private
1632          * instance data is initialized to zero.
1633          */
1634
1635         netdev = alloc_etherdev(sizeof(struct enic));
1636         if (!netdev) {
1637                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
1638                 return -ENOMEM;
1639         }
1640
1641         pci_set_drvdata(pdev, netdev);
1642
1643         SET_NETDEV_DEV(netdev, &pdev->dev);
1644
1645         enic = netdev_priv(netdev);
1646         enic->netdev = netdev;
1647         enic->pdev = pdev;
1648
1649         /* Setup PCI resources
1650          */
1651
1652         err = pci_enable_device(pdev);
1653         if (err) {
1654                 printk(KERN_ERR PFX
1655                         "Cannot enable PCI device, aborting.\n");
1656                 goto err_out_free_netdev;
1657         }
1658
1659         err = pci_request_regions(pdev, DRV_NAME);
1660         if (err) {
1661                 printk(KERN_ERR PFX
1662                         "Cannot request PCI regions, aborting.\n");
1663                 goto err_out_disable_device;
1664         }
1665
1666         pci_set_master(pdev);
1667
1668         /* Query PCI controller on system for DMA addressing
1669          * limitation for the device.  Try 40-bit first, and
1670          * fail to 32-bit.
1671          */
1672
1673         err = pci_set_dma_mask(pdev, DMA_40BIT_MASK);
1674         if (err) {
1675                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1676                 if (err) {
1677                         printk(KERN_ERR PFX
1678                                 "No usable DMA configuration, aborting.\n");
1679                         goto err_out_release_regions;
1680                 }
1681                 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
1682                 if (err) {
1683                         printk(KERN_ERR PFX
1684                                 "Unable to obtain 32-bit DMA "
1685                                 "for consistent allocations, aborting.\n");
1686                         goto err_out_release_regions;
1687                 }
1688         } else {
1689                 err = pci_set_consistent_dma_mask(pdev, DMA_40BIT_MASK);
1690                 if (err) {
1691                         printk(KERN_ERR PFX
1692                                 "Unable to obtain 40-bit DMA "
1693                                 "for consistent allocations, aborting.\n");
1694                         goto err_out_release_regions;
1695                 }
1696                 using_dac = 1;
1697         }
1698
1699         /* Map vNIC resources from BAR0
1700          */
1701
1702         if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
1703                 printk(KERN_ERR PFX
1704                         "BAR0 not memory-map'able, aborting.\n");
1705                 err = -ENODEV;
1706                 goto err_out_release_regions;
1707         }
1708
1709         enic->bar0.vaddr = pci_iomap(pdev, 0, enic->bar0.len);
1710         enic->bar0.bus_addr = pci_resource_start(pdev, 0);
1711         enic->bar0.len = pci_resource_len(pdev, 0);
1712
1713         if (!enic->bar0.vaddr) {
1714                 printk(KERN_ERR PFX
1715                         "Cannot memory-map BAR0 res hdr, aborting.\n");
1716                 err = -ENODEV;
1717                 goto err_out_release_regions;
1718         }
1719
1720         /* Register vNIC device
1721          */
1722
1723         enic->vdev = vnic_dev_register(NULL, enic, pdev, &enic->bar0);
1724         if (!enic->vdev) {
1725                 printk(KERN_ERR PFX
1726                         "vNIC registration failed, aborting.\n");
1727                 err = -ENODEV;
1728                 goto err_out_iounmap;
1729         }
1730
1731         /* Issue device open to get device in known state
1732          */
1733
1734         err = enic_dev_open(enic);
1735         if (err) {
1736                 printk(KERN_ERR PFX
1737                         "vNIC dev open failed, aborting.\n");
1738                 goto err_out_vnic_unregister;
1739         }
1740
1741         /* Issue device init to initialize the vnic-to-switch link.
1742          * We'll start with carrier off and wait for link UP
1743          * notification later to turn on carrier.  We don't need
1744          * to wait here for the vnic-to-switch link initialization
1745          * to complete; link UP notification is the indication that
1746          * the process is complete.
1747          */
1748
1749         netif_carrier_off(netdev);
1750
1751         err = vnic_dev_init(enic->vdev, 0);
1752         if (err) {
1753                 printk(KERN_ERR PFX
1754                         "vNIC dev init failed, aborting.\n");
1755                 goto err_out_dev_close;
1756         }
1757
1758         /* Get vNIC configuration
1759          */
1760
1761         err = enic_get_vnic_config(enic);
1762         if (err) {
1763                 printk(KERN_ERR PFX
1764                         "Get vNIC configuration failed, aborting.\n");
1765                 goto err_out_dev_close;
1766         }
1767
1768         /* Get available resource counts
1769          */
1770
1771         enic_get_res_counts(enic);
1772
1773         /* Set interrupt mode based on resource counts and system
1774          * capabilities
1775          */
1776
1777         err = enic_set_intr_mode(enic);
1778         if (err) {
1779                 printk(KERN_ERR PFX
1780                         "Failed to set intr mode, aborting.\n");
1781                 goto err_out_dev_close;
1782         }
1783
1784         /* Allocate and configure vNIC resources
1785          */
1786
1787         err = enic_alloc_vnic_resources(enic);
1788         if (err) {
1789                 printk(KERN_ERR PFX
1790                         "Failed to alloc vNIC resources, aborting.\n");
1791                 goto err_out_free_vnic_resources;
1792         }
1793
1794         enic_init_vnic_resources(enic);
1795
1796         /* Enable VLAN tag stripping.  RSS not enabled (yet).
1797          */
1798
1799         err = enic_set_nic_cfg(enic,
1800                 rss_default_cpu, rss_hash_type,
1801                 rss_hash_bits, rss_base_cpu,
1802                 rss_enable, tso_ipid_split_en,
1803                 ig_vlan_strip_en);
1804         if (err) {
1805                 printk(KERN_ERR PFX
1806                         "Failed to config nic, aborting.\n");
1807                 goto err_out_free_vnic_resources;
1808         }
1809
1810         /* Setup notification timer, HW reset task, and locks
1811          */
1812
1813         init_timer(&enic->notify_timer);
1814         enic->notify_timer.function = enic_notify_timer;
1815         enic->notify_timer.data = (unsigned long)enic;
1816
1817         INIT_WORK(&enic->reset, enic_reset);
1818
1819         for (i = 0; i < enic->wq_count; i++)
1820                 spin_lock_init(&enic->wq_lock[i]);
1821
1822         spin_lock_init(&enic->devcmd_lock);
1823
1824         /* Register net device
1825          */
1826
1827         enic->port_mtu = enic->config.mtu;
1828         (void)enic_change_mtu(netdev, enic->port_mtu);
1829
1830         err = enic_set_mac_addr(netdev, enic->mac_addr);
1831         if (err) {
1832                 printk(KERN_ERR PFX
1833                         "Invalid MAC address, aborting.\n");
1834                 goto err_out_free_vnic_resources;
1835         }
1836
1837         netdev->netdev_ops = &enic_netdev_ops;
1838         netdev->watchdog_timeo = 2 * HZ;
1839         netdev->ethtool_ops = &enic_ethtool_ops;
1840
1841         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1842         default:
1843                 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
1844                 break;
1845         case VNIC_DEV_INTR_MODE_MSIX:
1846                 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
1847                 break;
1848         }
1849
1850         netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
1851         if (ENIC_SETTING(enic, TXCSUM))
1852                 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1853         if (ENIC_SETTING(enic, TSO))
1854                 netdev->features |= NETIF_F_TSO |
1855                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
1856         if (ENIC_SETTING(enic, LRO))
1857                 netdev->features |= NETIF_F_LRO;
1858         if (using_dac)
1859                 netdev->features |= NETIF_F_HIGHDMA;
1860
1861
1862         enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
1863
1864         enic->lro_mgr.max_aggr = ENIC_LRO_MAX_AGGR;
1865         enic->lro_mgr.max_desc = ENIC_LRO_MAX_DESC;
1866         enic->lro_mgr.lro_arr = enic->lro_desc;
1867         enic->lro_mgr.get_skb_header = enic_get_skb_header;
1868         enic->lro_mgr.features  = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
1869         enic->lro_mgr.dev = netdev;
1870         enic->lro_mgr.ip_summed = CHECKSUM_COMPLETE;
1871         enic->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1872
1873
1874         err = register_netdev(netdev);
1875         if (err) {
1876                 printk(KERN_ERR PFX
1877                         "Cannot register net device, aborting.\n");
1878                 goto err_out_free_vnic_resources;
1879         }
1880
1881         return 0;
1882
1883 err_out_free_vnic_resources:
1884         enic_free_vnic_resources(enic);
1885 err_out_dev_close:
1886         vnic_dev_close(enic->vdev);
1887 err_out_vnic_unregister:
1888         enic_clear_intr_mode(enic);
1889         vnic_dev_unregister(enic->vdev);
1890 err_out_iounmap:
1891         enic_iounmap(enic);
1892 err_out_release_regions:
1893         pci_release_regions(pdev);
1894 err_out_disable_device:
1895         pci_disable_device(pdev);
1896 err_out_free_netdev:
1897         pci_set_drvdata(pdev, NULL);
1898         free_netdev(netdev);
1899
1900         return err;
1901 }
1902
1903 static void __devexit enic_remove(struct pci_dev *pdev)
1904 {
1905         struct net_device *netdev = pci_get_drvdata(pdev);
1906
1907         if (netdev) {
1908                 struct enic *enic = netdev_priv(netdev);
1909
1910                 flush_scheduled_work();
1911                 unregister_netdev(netdev);
1912                 enic_free_vnic_resources(enic);
1913                 vnic_dev_close(enic->vdev);
1914                 enic_clear_intr_mode(enic);
1915                 vnic_dev_unregister(enic->vdev);
1916                 enic_iounmap(enic);
1917                 pci_release_regions(pdev);
1918                 pci_disable_device(pdev);
1919                 pci_set_drvdata(pdev, NULL);
1920                 free_netdev(netdev);
1921         }
1922 }
1923
1924 static struct pci_driver enic_driver = {
1925         .name = DRV_NAME,
1926         .id_table = enic_id_table,
1927         .probe = enic_probe,
1928         .remove = __devexit_p(enic_remove),
1929 };
1930
1931 static int __init enic_init_module(void)
1932 {
1933         printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
1934
1935         return pci_register_driver(&enic_driver);
1936 }
1937
1938 static void __exit enic_cleanup_module(void)
1939 {
1940         pci_unregister_driver(&enic_driver);
1941 }
1942
1943 module_init(enic_init_module);
1944 module_exit(enic_cleanup_module);