enic: Feature Add: Replace LRO with GRO
[linux-2.6.git] / drivers / net / enic / enic_main.c
1 /*
2  * Copyright 2008 Cisco Systems, Inc.  All rights reserved.
3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
4  *
5  * This program is free software; you may redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; version 2 of the License.
8  *
9  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
10  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
11  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
12  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
13  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
14  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
15  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
16  * SOFTWARE.
17  *
18  */
19
20 #include <linux/module.h>
21 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/init.h>
26 #include <linux/workqueue.h>
27 #include <linux/pci.h>
28 #include <linux/netdevice.h>
29 #include <linux/etherdevice.h>
30 #include <linux/if_ether.h>
31 #include <linux/if_vlan.h>
32 #include <linux/if_link.h>
33 #include <linux/ethtool.h>
34 #include <linux/in.h>
35 #include <linux/ip.h>
36 #include <linux/ipv6.h>
37 #include <linux/tcp.h>
38 #include <net/ip6_checksum.h>
39
40 #include "cq_enet_desc.h"
41 #include "vnic_dev.h"
42 #include "vnic_intr.h"
43 #include "vnic_stats.h"
44 #include "vnic_vic.h"
45 #include "enic_res.h"
46 #include "enic.h"
47
48 #define ENIC_NOTIFY_TIMER_PERIOD        (2 * HZ)
49 #define WQ_ENET_MAX_DESC_LEN            (1 << WQ_ENET_LEN_BITS)
50 #define MAX_TSO                         (1 << 16)
51 #define ENIC_DESC_MAX_SPLITS            (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
52
53 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
54 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
55
56 /* Supported devices */
57 static DEFINE_PCI_DEVICE_TABLE(enic_id_table) = {
58         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
59         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
60         { 0, }  /* end of table */
61 };
62
63 MODULE_DESCRIPTION(DRV_DESCRIPTION);
64 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
65 MODULE_LICENSE("GPL");
66 MODULE_VERSION(DRV_VERSION);
67 MODULE_DEVICE_TABLE(pci, enic_id_table);
68
69 struct enic_stat {
70         char name[ETH_GSTRING_LEN];
71         unsigned int offset;
72 };
73
74 #define ENIC_TX_STAT(stat)      \
75         { .name = #stat, .offset = offsetof(struct vnic_tx_stats, stat) / 8 }
76 #define ENIC_RX_STAT(stat)      \
77         { .name = #stat, .offset = offsetof(struct vnic_rx_stats, stat) / 8 }
78
79 static const struct enic_stat enic_tx_stats[] = {
80         ENIC_TX_STAT(tx_frames_ok),
81         ENIC_TX_STAT(tx_unicast_frames_ok),
82         ENIC_TX_STAT(tx_multicast_frames_ok),
83         ENIC_TX_STAT(tx_broadcast_frames_ok),
84         ENIC_TX_STAT(tx_bytes_ok),
85         ENIC_TX_STAT(tx_unicast_bytes_ok),
86         ENIC_TX_STAT(tx_multicast_bytes_ok),
87         ENIC_TX_STAT(tx_broadcast_bytes_ok),
88         ENIC_TX_STAT(tx_drops),
89         ENIC_TX_STAT(tx_errors),
90         ENIC_TX_STAT(tx_tso),
91 };
92
93 static const struct enic_stat enic_rx_stats[] = {
94         ENIC_RX_STAT(rx_frames_ok),
95         ENIC_RX_STAT(rx_frames_total),
96         ENIC_RX_STAT(rx_unicast_frames_ok),
97         ENIC_RX_STAT(rx_multicast_frames_ok),
98         ENIC_RX_STAT(rx_broadcast_frames_ok),
99         ENIC_RX_STAT(rx_bytes_ok),
100         ENIC_RX_STAT(rx_unicast_bytes_ok),
101         ENIC_RX_STAT(rx_multicast_bytes_ok),
102         ENIC_RX_STAT(rx_broadcast_bytes_ok),
103         ENIC_RX_STAT(rx_drop),
104         ENIC_RX_STAT(rx_no_bufs),
105         ENIC_RX_STAT(rx_errors),
106         ENIC_RX_STAT(rx_rss),
107         ENIC_RX_STAT(rx_crc_errors),
108         ENIC_RX_STAT(rx_frames_64),
109         ENIC_RX_STAT(rx_frames_127),
110         ENIC_RX_STAT(rx_frames_255),
111         ENIC_RX_STAT(rx_frames_511),
112         ENIC_RX_STAT(rx_frames_1023),
113         ENIC_RX_STAT(rx_frames_1518),
114         ENIC_RX_STAT(rx_frames_to_max),
115 };
116
117 static const unsigned int enic_n_tx_stats = ARRAY_SIZE(enic_tx_stats);
118 static const unsigned int enic_n_rx_stats = ARRAY_SIZE(enic_rx_stats);
119
120 static int enic_is_dynamic(struct enic *enic)
121 {
122         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
123 }
124
125 static int enic_get_settings(struct net_device *netdev,
126         struct ethtool_cmd *ecmd)
127 {
128         struct enic *enic = netdev_priv(netdev);
129
130         ecmd->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
131         ecmd->advertising = (ADVERTISED_10000baseT_Full | ADVERTISED_FIBRE);
132         ecmd->port = PORT_FIBRE;
133         ecmd->transceiver = XCVR_EXTERNAL;
134
135         if (netif_carrier_ok(netdev)) {
136                 ecmd->speed = vnic_dev_port_speed(enic->vdev);
137                 ecmd->duplex = DUPLEX_FULL;
138         } else {
139                 ecmd->speed = -1;
140                 ecmd->duplex = -1;
141         }
142
143         ecmd->autoneg = AUTONEG_DISABLE;
144
145         return 0;
146 }
147
148 static void enic_get_drvinfo(struct net_device *netdev,
149         struct ethtool_drvinfo *drvinfo)
150 {
151         struct enic *enic = netdev_priv(netdev);
152         struct vnic_devcmd_fw_info *fw_info;
153
154         spin_lock(&enic->devcmd_lock);
155         vnic_dev_fw_info(enic->vdev, &fw_info);
156         spin_unlock(&enic->devcmd_lock);
157
158         strncpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
159         strncpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
160         strncpy(drvinfo->fw_version, fw_info->fw_version,
161                 sizeof(drvinfo->fw_version));
162         strncpy(drvinfo->bus_info, pci_name(enic->pdev),
163                 sizeof(drvinfo->bus_info));
164 }
165
166 static void enic_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
167 {
168         unsigned int i;
169
170         switch (stringset) {
171         case ETH_SS_STATS:
172                 for (i = 0; i < enic_n_tx_stats; i++) {
173                         memcpy(data, enic_tx_stats[i].name, ETH_GSTRING_LEN);
174                         data += ETH_GSTRING_LEN;
175                 }
176                 for (i = 0; i < enic_n_rx_stats; i++) {
177                         memcpy(data, enic_rx_stats[i].name, ETH_GSTRING_LEN);
178                         data += ETH_GSTRING_LEN;
179                 }
180                 break;
181         }
182 }
183
184 static int enic_get_sset_count(struct net_device *netdev, int sset)
185 {
186         switch (sset) {
187         case ETH_SS_STATS:
188                 return enic_n_tx_stats + enic_n_rx_stats;
189         default:
190                 return -EOPNOTSUPP;
191         }
192 }
193
194 static void enic_get_ethtool_stats(struct net_device *netdev,
195         struct ethtool_stats *stats, u64 *data)
196 {
197         struct enic *enic = netdev_priv(netdev);
198         struct vnic_stats *vstats;
199         unsigned int i;
200
201         spin_lock(&enic->devcmd_lock);
202         vnic_dev_stats_dump(enic->vdev, &vstats);
203         spin_unlock(&enic->devcmd_lock);
204
205         for (i = 0; i < enic_n_tx_stats; i++)
206                 *(data++) = ((u64 *)&vstats->tx)[enic_tx_stats[i].offset];
207         for (i = 0; i < enic_n_rx_stats; i++)
208                 *(data++) = ((u64 *)&vstats->rx)[enic_rx_stats[i].offset];
209 }
210
211 static u32 enic_get_rx_csum(struct net_device *netdev)
212 {
213         struct enic *enic = netdev_priv(netdev);
214         return enic->csum_rx_enabled;
215 }
216
217 static int enic_set_rx_csum(struct net_device *netdev, u32 data)
218 {
219         struct enic *enic = netdev_priv(netdev);
220
221         if (data && !ENIC_SETTING(enic, RXCSUM))
222                 return -EINVAL;
223
224         enic->csum_rx_enabled = !!data;
225
226         return 0;
227 }
228
229 static int enic_set_tx_csum(struct net_device *netdev, u32 data)
230 {
231         struct enic *enic = netdev_priv(netdev);
232
233         if (data && !ENIC_SETTING(enic, TXCSUM))
234                 return -EINVAL;
235
236         if (data)
237                 netdev->features |= NETIF_F_HW_CSUM;
238         else
239                 netdev->features &= ~NETIF_F_HW_CSUM;
240
241         return 0;
242 }
243
244 static int enic_set_tso(struct net_device *netdev, u32 data)
245 {
246         struct enic *enic = netdev_priv(netdev);
247
248         if (data && !ENIC_SETTING(enic, TSO))
249                 return -EINVAL;
250
251         if (data)
252                 netdev->features |=
253                         NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN;
254         else
255                 netdev->features &=
256                         ~(NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN);
257
258         return 0;
259 }
260
261 static u32 enic_get_msglevel(struct net_device *netdev)
262 {
263         struct enic *enic = netdev_priv(netdev);
264         return enic->msg_enable;
265 }
266
267 static void enic_set_msglevel(struct net_device *netdev, u32 value)
268 {
269         struct enic *enic = netdev_priv(netdev);
270         enic->msg_enable = value;
271 }
272
273 static int enic_get_coalesce(struct net_device *netdev,
274         struct ethtool_coalesce *ecmd)
275 {
276         struct enic *enic = netdev_priv(netdev);
277
278         ecmd->tx_coalesce_usecs = enic->tx_coalesce_usecs;
279         ecmd->rx_coalesce_usecs = enic->rx_coalesce_usecs;
280
281         return 0;
282 }
283
284 static int enic_set_coalesce(struct net_device *netdev,
285         struct ethtool_coalesce *ecmd)
286 {
287         struct enic *enic = netdev_priv(netdev);
288         u32 tx_coalesce_usecs;
289         u32 rx_coalesce_usecs;
290
291         tx_coalesce_usecs = min_t(u32,
292                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
293                 ecmd->tx_coalesce_usecs);
294         rx_coalesce_usecs = min_t(u32,
295                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
296                 ecmd->rx_coalesce_usecs);
297
298         switch (vnic_dev_get_intr_mode(enic->vdev)) {
299         case VNIC_DEV_INTR_MODE_INTX:
300                 if (tx_coalesce_usecs != rx_coalesce_usecs)
301                         return -EINVAL;
302
303                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_INTX_WQ_RQ],
304                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
305                 break;
306         case VNIC_DEV_INTR_MODE_MSI:
307                 if (tx_coalesce_usecs != rx_coalesce_usecs)
308                         return -EINVAL;
309
310                 vnic_intr_coalescing_timer_set(&enic->intr[0],
311                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
312                 break;
313         case VNIC_DEV_INTR_MODE_MSIX:
314                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_MSIX_WQ],
315                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
316                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_MSIX_RQ],
317                         INTR_COALESCE_USEC_TO_HW(rx_coalesce_usecs));
318                 break;
319         default:
320                 break;
321         }
322
323         enic->tx_coalesce_usecs = tx_coalesce_usecs;
324         enic->rx_coalesce_usecs = rx_coalesce_usecs;
325
326         return 0;
327 }
328
329 static const struct ethtool_ops enic_ethtool_ops = {
330         .get_settings = enic_get_settings,
331         .get_drvinfo = enic_get_drvinfo,
332         .get_msglevel = enic_get_msglevel,
333         .set_msglevel = enic_set_msglevel,
334         .get_link = ethtool_op_get_link,
335         .get_strings = enic_get_strings,
336         .get_sset_count = enic_get_sset_count,
337         .get_ethtool_stats = enic_get_ethtool_stats,
338         .get_rx_csum = enic_get_rx_csum,
339         .set_rx_csum = enic_set_rx_csum,
340         .get_tx_csum = ethtool_op_get_tx_csum,
341         .set_tx_csum = enic_set_tx_csum,
342         .get_sg = ethtool_op_get_sg,
343         .set_sg = ethtool_op_set_sg,
344         .get_tso = ethtool_op_get_tso,
345         .set_tso = enic_set_tso,
346         .get_coalesce = enic_get_coalesce,
347         .set_coalesce = enic_set_coalesce,
348         .get_flags = ethtool_op_get_flags,
349         .set_flags = ethtool_op_set_flags,
350 };
351
352 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
353 {
354         struct enic *enic = vnic_dev_priv(wq->vdev);
355
356         if (buf->sop)
357                 pci_unmap_single(enic->pdev, buf->dma_addr,
358                         buf->len, PCI_DMA_TODEVICE);
359         else
360                 pci_unmap_page(enic->pdev, buf->dma_addr,
361                         buf->len, PCI_DMA_TODEVICE);
362
363         if (buf->os_buf)
364                 dev_kfree_skb_any(buf->os_buf);
365 }
366
367 static void enic_wq_free_buf(struct vnic_wq *wq,
368         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
369 {
370         enic_free_wq_buf(wq, buf);
371 }
372
373 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
374         u8 type, u16 q_number, u16 completed_index, void *opaque)
375 {
376         struct enic *enic = vnic_dev_priv(vdev);
377
378         spin_lock(&enic->wq_lock[q_number]);
379
380         vnic_wq_service(&enic->wq[q_number], cq_desc,
381                 completed_index, enic_wq_free_buf,
382                 opaque);
383
384         if (netif_queue_stopped(enic->netdev) &&
385             vnic_wq_desc_avail(&enic->wq[q_number]) >=
386             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
387                 netif_wake_queue(enic->netdev);
388
389         spin_unlock(&enic->wq_lock[q_number]);
390
391         return 0;
392 }
393
394 static void enic_log_q_error(struct enic *enic)
395 {
396         unsigned int i;
397         u32 error_status;
398
399         for (i = 0; i < enic->wq_count; i++) {
400                 error_status = vnic_wq_error_status(&enic->wq[i]);
401                 if (error_status)
402                         printk(KERN_ERR PFX "%s: WQ[%d] error_status %d\n",
403                                 enic->netdev->name, i, error_status);
404         }
405
406         for (i = 0; i < enic->rq_count; i++) {
407                 error_status = vnic_rq_error_status(&enic->rq[i]);
408                 if (error_status)
409                         printk(KERN_ERR PFX "%s: RQ[%d] error_status %d\n",
410                                 enic->netdev->name, i, error_status);
411         }
412 }
413
414 static void enic_link_check(struct enic *enic)
415 {
416         int link_status = vnic_dev_link_status(enic->vdev);
417         int carrier_ok = netif_carrier_ok(enic->netdev);
418
419         if (link_status && !carrier_ok) {
420                 printk(KERN_INFO PFX "%s: Link UP\n", enic->netdev->name);
421                 netif_carrier_on(enic->netdev);
422         } else if (!link_status && carrier_ok) {
423                 printk(KERN_INFO PFX "%s: Link DOWN\n", enic->netdev->name);
424                 netif_carrier_off(enic->netdev);
425         }
426 }
427
428 static void enic_mtu_check(struct enic *enic)
429 {
430         u32 mtu = vnic_dev_mtu(enic->vdev);
431
432         if (mtu && mtu != enic->port_mtu) {
433                 enic->port_mtu = mtu;
434                 if (mtu < enic->netdev->mtu)
435                         printk(KERN_WARNING PFX
436                                 "%s: interface MTU (%d) set higher "
437                                 "than switch port MTU (%d)\n",
438                                 enic->netdev->name, enic->netdev->mtu, mtu);
439         }
440 }
441
442 static void enic_msglvl_check(struct enic *enic)
443 {
444         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
445
446         if (msg_enable != enic->msg_enable) {
447                 printk(KERN_INFO PFX "%s: msg lvl changed from 0x%x to 0x%x\n",
448                         enic->netdev->name, enic->msg_enable, msg_enable);
449                 enic->msg_enable = msg_enable;
450         }
451 }
452
453 static void enic_notify_check(struct enic *enic)
454 {
455         enic_msglvl_check(enic);
456         enic_mtu_check(enic);
457         enic_link_check(enic);
458 }
459
460 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
461
462 static irqreturn_t enic_isr_legacy(int irq, void *data)
463 {
464         struct net_device *netdev = data;
465         struct enic *enic = netdev_priv(netdev);
466         u32 pba;
467
468         vnic_intr_mask(&enic->intr[ENIC_INTX_WQ_RQ]);
469
470         pba = vnic_intr_legacy_pba(enic->legacy_pba);
471         if (!pba) {
472                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
473                 return IRQ_NONE;        /* not our interrupt */
474         }
475
476         if (ENIC_TEST_INTR(pba, ENIC_INTX_NOTIFY)) {
477                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_NOTIFY]);
478                 enic_notify_check(enic);
479         }
480
481         if (ENIC_TEST_INTR(pba, ENIC_INTX_ERR)) {
482                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_ERR]);
483                 enic_log_q_error(enic);
484                 /* schedule recovery from WQ/RQ error */
485                 schedule_work(&enic->reset);
486                 return IRQ_HANDLED;
487         }
488
489         if (ENIC_TEST_INTR(pba, ENIC_INTX_WQ_RQ)) {
490                 if (napi_schedule_prep(&enic->napi))
491                         __napi_schedule(&enic->napi);
492         } else {
493                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
494         }
495
496         return IRQ_HANDLED;
497 }
498
499 static irqreturn_t enic_isr_msi(int irq, void *data)
500 {
501         struct enic *enic = data;
502
503         /* With MSI, there is no sharing of interrupts, so this is
504          * our interrupt and there is no need to ack it.  The device
505          * is not providing per-vector masking, so the OS will not
506          * write to PCI config space to mask/unmask the interrupt.
507          * We're using mask_on_assertion for MSI, so the device
508          * automatically masks the interrupt when the interrupt is
509          * generated.  Later, when exiting polling, the interrupt
510          * will be unmasked (see enic_poll).
511          *
512          * Also, the device uses the same PCIe Traffic Class (TC)
513          * for Memory Write data and MSI, so there are no ordering
514          * issues; the MSI will always arrive at the Root Complex
515          * _after_ corresponding Memory Writes (i.e. descriptor
516          * writes).
517          */
518
519         napi_schedule(&enic->napi);
520
521         return IRQ_HANDLED;
522 }
523
524 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
525 {
526         struct enic *enic = data;
527
528         /* schedule NAPI polling for RQ cleanup */
529         napi_schedule(&enic->napi);
530
531         return IRQ_HANDLED;
532 }
533
534 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
535 {
536         struct enic *enic = data;
537         unsigned int wq_work_to_do = -1; /* no limit */
538         unsigned int wq_work_done;
539
540         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
541                 wq_work_to_do, enic_wq_service, NULL);
542
543         vnic_intr_return_credits(&enic->intr[ENIC_MSIX_WQ],
544                 wq_work_done,
545                 1 /* unmask intr */,
546                 1 /* reset intr timer */);
547
548         return IRQ_HANDLED;
549 }
550
551 static irqreturn_t enic_isr_msix_err(int irq, void *data)
552 {
553         struct enic *enic = data;
554
555         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_ERR]);
556
557         enic_log_q_error(enic);
558
559         /* schedule recovery from WQ/RQ error */
560         schedule_work(&enic->reset);
561
562         return IRQ_HANDLED;
563 }
564
565 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
566 {
567         struct enic *enic = data;
568
569         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_NOTIFY]);
570         enic_notify_check(enic);
571
572         return IRQ_HANDLED;
573 }
574
575 static inline void enic_queue_wq_skb_cont(struct enic *enic,
576         struct vnic_wq *wq, struct sk_buff *skb,
577         unsigned int len_left)
578 {
579         skb_frag_t *frag;
580
581         /* Queue additional data fragments */
582         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
583                 len_left -= frag->size;
584                 enic_queue_wq_desc_cont(wq, skb,
585                         pci_map_page(enic->pdev, frag->page,
586                                 frag->page_offset, frag->size,
587                                 PCI_DMA_TODEVICE),
588                         frag->size,
589                         (len_left == 0));       /* EOP? */
590         }
591 }
592
593 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
594         struct vnic_wq *wq, struct sk_buff *skb,
595         int vlan_tag_insert, unsigned int vlan_tag)
596 {
597         unsigned int head_len = skb_headlen(skb);
598         unsigned int len_left = skb->len - head_len;
599         int eop = (len_left == 0);
600
601         /* Queue the main skb fragment. The fragments are no larger
602          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
603          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
604          * per fragment is queued.
605          */
606         enic_queue_wq_desc(wq, skb,
607                 pci_map_single(enic->pdev, skb->data,
608                         head_len, PCI_DMA_TODEVICE),
609                 head_len,
610                 vlan_tag_insert, vlan_tag,
611                 eop);
612
613         if (!eop)
614                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
615 }
616
617 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
618         struct vnic_wq *wq, struct sk_buff *skb,
619         int vlan_tag_insert, unsigned int vlan_tag)
620 {
621         unsigned int head_len = skb_headlen(skb);
622         unsigned int len_left = skb->len - head_len;
623         unsigned int hdr_len = skb_transport_offset(skb);
624         unsigned int csum_offset = hdr_len + skb->csum_offset;
625         int eop = (len_left == 0);
626
627         /* Queue the main skb fragment. The fragments are no larger
628          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
629          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
630          * per fragment is queued.
631          */
632         enic_queue_wq_desc_csum_l4(wq, skb,
633                 pci_map_single(enic->pdev, skb->data,
634                         head_len, PCI_DMA_TODEVICE),
635                 head_len,
636                 csum_offset,
637                 hdr_len,
638                 vlan_tag_insert, vlan_tag,
639                 eop);
640
641         if (!eop)
642                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
643 }
644
645 static inline void enic_queue_wq_skb_tso(struct enic *enic,
646         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
647         int vlan_tag_insert, unsigned int vlan_tag)
648 {
649         unsigned int frag_len_left = skb_headlen(skb);
650         unsigned int len_left = skb->len - frag_len_left;
651         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
652         int eop = (len_left == 0);
653         unsigned int len;
654         dma_addr_t dma_addr;
655         unsigned int offset = 0;
656         skb_frag_t *frag;
657
658         /* Preload TCP csum field with IP pseudo hdr calculated
659          * with IP length set to zero.  HW will later add in length
660          * to each TCP segment resulting from the TSO.
661          */
662
663         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
664                 ip_hdr(skb)->check = 0;
665                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
666                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
667         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
668                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
669                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
670         }
671
672         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
673          * for the main skb fragment
674          */
675         while (frag_len_left) {
676                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
677                 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
678                                 len, PCI_DMA_TODEVICE);
679                 enic_queue_wq_desc_tso(wq, skb,
680                         dma_addr,
681                         len,
682                         mss, hdr_len,
683                         vlan_tag_insert, vlan_tag,
684                         eop && (len == frag_len_left));
685                 frag_len_left -= len;
686                 offset += len;
687         }
688
689         if (eop)
690                 return;
691
692         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
693          * for additional data fragments
694          */
695         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
696                 len_left -= frag->size;
697                 frag_len_left = frag->size;
698                 offset = frag->page_offset;
699
700                 while (frag_len_left) {
701                         len = min(frag_len_left,
702                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
703                         dma_addr = pci_map_page(enic->pdev, frag->page,
704                                 offset, len,
705                                 PCI_DMA_TODEVICE);
706                         enic_queue_wq_desc_cont(wq, skb,
707                                 dma_addr,
708                                 len,
709                                 (len_left == 0) &&
710                                 (len == frag_len_left));        /* EOP? */
711                         frag_len_left -= len;
712                         offset += len;
713                 }
714         }
715 }
716
717 static inline void enic_queue_wq_skb(struct enic *enic,
718         struct vnic_wq *wq, struct sk_buff *skb)
719 {
720         unsigned int mss = skb_shinfo(skb)->gso_size;
721         unsigned int vlan_tag = 0;
722         int vlan_tag_insert = 0;
723
724         if (enic->vlan_group && vlan_tx_tag_present(skb)) {
725                 /* VLAN tag from trunking driver */
726                 vlan_tag_insert = 1;
727                 vlan_tag = vlan_tx_tag_get(skb);
728         }
729
730         if (mss)
731                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
732                         vlan_tag_insert, vlan_tag);
733         else if (skb->ip_summed == CHECKSUM_PARTIAL)
734                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
735                         vlan_tag_insert, vlan_tag);
736         else
737                 enic_queue_wq_skb_vlan(enic, wq, skb,
738                         vlan_tag_insert, vlan_tag);
739 }
740
741 /* netif_tx_lock held, process context with BHs disabled, or BH */
742 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
743         struct net_device *netdev)
744 {
745         struct enic *enic = netdev_priv(netdev);
746         struct vnic_wq *wq = &enic->wq[0];
747         unsigned long flags;
748
749         if (skb->len <= 0) {
750                 dev_kfree_skb(skb);
751                 return NETDEV_TX_OK;
752         }
753
754         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
755          * which is very likely.  In the off chance it's going to take
756          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
757          */
758
759         if (skb_shinfo(skb)->gso_size == 0 &&
760             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
761             skb_linearize(skb)) {
762                 dev_kfree_skb(skb);
763                 return NETDEV_TX_OK;
764         }
765
766         spin_lock_irqsave(&enic->wq_lock[0], flags);
767
768         if (vnic_wq_desc_avail(wq) <
769             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
770                 netif_stop_queue(netdev);
771                 /* This is a hard error, log it */
772                 printk(KERN_ERR PFX "%s: BUG! Tx ring full when "
773                         "queue awake!\n", netdev->name);
774                 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
775                 return NETDEV_TX_BUSY;
776         }
777
778         enic_queue_wq_skb(enic, wq, skb);
779
780         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
781                 netif_stop_queue(netdev);
782
783         spin_unlock_irqrestore(&enic->wq_lock[0], flags);
784
785         return NETDEV_TX_OK;
786 }
787
788 /* dev_base_lock rwlock held, nominally process context */
789 static struct net_device_stats *enic_get_stats(struct net_device *netdev)
790 {
791         struct enic *enic = netdev_priv(netdev);
792         struct net_device_stats *net_stats = &netdev->stats;
793         struct vnic_stats *stats;
794
795         spin_lock(&enic->devcmd_lock);
796         vnic_dev_stats_dump(enic->vdev, &stats);
797         spin_unlock(&enic->devcmd_lock);
798
799         net_stats->tx_packets = stats->tx.tx_frames_ok;
800         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
801         net_stats->tx_errors = stats->tx.tx_errors;
802         net_stats->tx_dropped = stats->tx.tx_drops;
803
804         net_stats->rx_packets = stats->rx.rx_frames_ok;
805         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
806         net_stats->rx_errors = stats->rx.rx_errors;
807         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
808         net_stats->rx_over_errors = enic->rq_truncated_pkts;
809         net_stats->rx_crc_errors = enic->rq_bad_fcs;
810         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
811
812         return net_stats;
813 }
814
815 static void enic_reset_mcaddrs(struct enic *enic)
816 {
817         enic->mc_count = 0;
818 }
819
820 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
821 {
822         struct enic *enic = netdev_priv(netdev);
823
824         if (enic_is_dynamic(enic)) {
825                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
826                         return -EADDRNOTAVAIL;
827         } else {
828                 if (!is_valid_ether_addr(addr))
829                         return -EADDRNOTAVAIL;
830         }
831
832         memcpy(netdev->dev_addr, addr, netdev->addr_len);
833
834         return 0;
835 }
836
837 static int enic_dev_add_station_addr(struct enic *enic)
838 {
839         int err = 0;
840
841         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
842                 spin_lock(&enic->devcmd_lock);
843                 err = vnic_dev_add_addr(enic->vdev, enic->netdev->dev_addr);
844                 spin_unlock(&enic->devcmd_lock);
845         }
846
847         return err;
848 }
849
850 static int enic_dev_del_station_addr(struct enic *enic)
851 {
852         int err = 0;
853
854         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
855                 spin_lock(&enic->devcmd_lock);
856                 err = vnic_dev_del_addr(enic->vdev, enic->netdev->dev_addr);
857                 spin_unlock(&enic->devcmd_lock);
858         }
859
860         return err;
861 }
862
863 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
864 {
865         struct enic *enic = netdev_priv(netdev);
866         struct sockaddr *saddr = p;
867         char *addr = saddr->sa_data;
868         int err;
869
870         if (netif_running(enic->netdev)) {
871                 err = enic_dev_del_station_addr(enic);
872                 if (err)
873                         return err;
874         }
875
876         err = enic_set_mac_addr(netdev, addr);
877         if (err)
878                 return err;
879
880         if (netif_running(enic->netdev)) {
881                 err = enic_dev_add_station_addr(enic);
882                 if (err)
883                         return err;
884         }
885
886         return err;
887 }
888
889 static int enic_set_mac_address(struct net_device *netdev, void *p)
890 {
891         return -EOPNOTSUPP;
892 }
893
894 /* netif_tx_lock held, BHs disabled */
895 static void enic_set_multicast_list(struct net_device *netdev)
896 {
897         struct enic *enic = netdev_priv(netdev);
898         struct netdev_hw_addr *ha;
899         int directed = 1;
900         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
901         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
902         int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
903         unsigned int mc_count = netdev_mc_count(netdev);
904         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
905                 mc_count > ENIC_MULTICAST_PERFECT_FILTERS;
906         unsigned int flags = netdev->flags | (allmulti ? IFF_ALLMULTI : 0);
907         u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
908         unsigned int i, j;
909
910         if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
911                 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
912
913         spin_lock(&enic->devcmd_lock);
914
915         if (enic->flags != flags) {
916                 enic->flags = flags;
917                 vnic_dev_packet_filter(enic->vdev, directed,
918                         multicast, broadcast, promisc, allmulti);
919         }
920
921         /* Is there an easier way?  Trying to minimize to
922          * calls to add/del multicast addrs.  We keep the
923          * addrs from the last call in enic->mc_addr and
924          * look for changes to add/del.
925          */
926
927         i = 0;
928         netdev_for_each_mc_addr(ha, netdev) {
929                 if (i == mc_count)
930                         break;
931                 memcpy(mc_addr[i++], ha->addr, ETH_ALEN);
932         }
933
934         for (i = 0; i < enic->mc_count; i++) {
935                 for (j = 0; j < mc_count; j++)
936                         if (compare_ether_addr(enic->mc_addr[i],
937                                 mc_addr[j]) == 0)
938                                 break;
939                 if (j == mc_count)
940                         enic_del_multicast_addr(enic, enic->mc_addr[i]);
941         }
942
943         for (i = 0; i < mc_count; i++) {
944                 for (j = 0; j < enic->mc_count; j++)
945                         if (compare_ether_addr(mc_addr[i],
946                                 enic->mc_addr[j]) == 0)
947                                 break;
948                 if (j == enic->mc_count)
949                         enic_add_multicast_addr(enic, mc_addr[i]);
950         }
951
952         /* Save the list to compare against next time
953          */
954
955         for (i = 0; i < mc_count; i++)
956                 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
957
958         enic->mc_count = mc_count;
959
960         spin_unlock(&enic->devcmd_lock);
961 }
962
963 /* rtnl lock is held */
964 static void enic_vlan_rx_register(struct net_device *netdev,
965         struct vlan_group *vlan_group)
966 {
967         struct enic *enic = netdev_priv(netdev);
968         enic->vlan_group = vlan_group;
969 }
970
971 /* rtnl lock is held */
972 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
973 {
974         struct enic *enic = netdev_priv(netdev);
975
976         spin_lock(&enic->devcmd_lock);
977         enic_add_vlan(enic, vid);
978         spin_unlock(&enic->devcmd_lock);
979 }
980
981 /* rtnl lock is held */
982 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
983 {
984         struct enic *enic = netdev_priv(netdev);
985
986         spin_lock(&enic->devcmd_lock);
987         enic_del_vlan(enic, vid);
988         spin_unlock(&enic->devcmd_lock);
989 }
990
991 /* netif_tx_lock held, BHs disabled */
992 static void enic_tx_timeout(struct net_device *netdev)
993 {
994         struct enic *enic = netdev_priv(netdev);
995         schedule_work(&enic->reset);
996 }
997
998 static int enic_vnic_dev_deinit(struct enic *enic)
999 {
1000         int err;
1001
1002         spin_lock(&enic->devcmd_lock);
1003         err = vnic_dev_deinit(enic->vdev);
1004         spin_unlock(&enic->devcmd_lock);
1005
1006         return err;
1007 }
1008
1009 static int enic_dev_init_prov(struct enic *enic, struct vic_provinfo *vp)
1010 {
1011         int err;
1012
1013         spin_lock(&enic->devcmd_lock);
1014         err = vnic_dev_init_prov(enic->vdev,
1015                 (u8 *)vp, vic_provinfo_size(vp));
1016         spin_unlock(&enic->devcmd_lock);
1017
1018         return err;
1019 }
1020
1021 static int enic_dev_init_done(struct enic *enic, int *done, int *error)
1022 {
1023         int err;
1024
1025         spin_lock(&enic->devcmd_lock);
1026         err = vnic_dev_init_done(enic->vdev, done, error);
1027         spin_unlock(&enic->devcmd_lock);
1028
1029         return err;
1030 }
1031
1032 static int enic_set_port_profile(struct enic *enic, u8 *mac)
1033 {
1034         struct vic_provinfo *vp;
1035         u8 oui[3] = VIC_PROVINFO_CISCO_OUI;
1036         u8 *uuid;
1037         char uuid_str[38];
1038         static char *uuid_fmt = "%02X%02X%02X%02X-%02X%02X-%02X%02X-"
1039                 "%02X%02X-%02X%02X%02X%02X%0X%02X";
1040         int err;
1041
1042         err = enic_vnic_dev_deinit(enic);
1043         if (err)
1044                 return err;
1045
1046         switch (enic->pp.request) {
1047
1048         case PORT_REQUEST_ASSOCIATE:
1049
1050                 if (!(enic->pp.set & ENIC_SET_NAME) || !strlen(enic->pp.name))
1051                         return -EINVAL;
1052
1053                 if (!is_valid_ether_addr(mac))
1054                         return -EADDRNOTAVAIL;
1055
1056                 vp = vic_provinfo_alloc(GFP_KERNEL, oui,
1057                         VIC_PROVINFO_LINUX_TYPE);
1058                 if (!vp)
1059                         return -ENOMEM;
1060
1061                 vic_provinfo_add_tlv(vp,
1062                         VIC_LINUX_PROV_TLV_PORT_PROFILE_NAME_STR,
1063                         strlen(enic->pp.name) + 1, enic->pp.name);
1064
1065                 vic_provinfo_add_tlv(vp,
1066                         VIC_LINUX_PROV_TLV_CLIENT_MAC_ADDR,
1067                         ETH_ALEN, mac);
1068
1069                 if (enic->pp.set & ENIC_SET_INSTANCE) {
1070                         uuid = enic->pp.instance_uuid;
1071                         sprintf(uuid_str, uuid_fmt,
1072                                 uuid[0],  uuid[1],  uuid[2],  uuid[3],
1073                                 uuid[4],  uuid[5],  uuid[6],  uuid[7],
1074                                 uuid[8],  uuid[9],  uuid[10], uuid[11],
1075                                 uuid[12], uuid[13], uuid[14], uuid[15]);
1076                         vic_provinfo_add_tlv(vp,
1077                                 VIC_LINUX_PROV_TLV_CLIENT_UUID_STR,
1078                                 sizeof(uuid_str), uuid_str);
1079                 }
1080
1081                 if (enic->pp.set & ENIC_SET_HOST) {
1082                         uuid = enic->pp.host_uuid;
1083                         sprintf(uuid_str, uuid_fmt,
1084                                 uuid[0],  uuid[1],  uuid[2],  uuid[3],
1085                                 uuid[4],  uuid[5],  uuid[6],  uuid[7],
1086                                 uuid[8],  uuid[9],  uuid[10], uuid[11],
1087                                 uuid[12], uuid[13], uuid[14], uuid[15]);
1088                         vic_provinfo_add_tlv(vp,
1089                                 VIC_LINUX_PROV_TLV_HOST_UUID_STR,
1090                                 sizeof(uuid_str), uuid_str);
1091                 }
1092
1093                 err = enic_dev_init_prov(enic, vp);
1094                 vic_provinfo_free(vp);
1095                 if (err)
1096                         return err;
1097                 break;
1098
1099         case PORT_REQUEST_DISASSOCIATE:
1100                 break;
1101
1102         default:
1103                 return -EINVAL;
1104         }
1105
1106         enic->pp.set |= ENIC_SET_APPLIED;
1107         return 0;
1108 }
1109
1110 static int enic_set_vf_port(struct net_device *netdev, int vf,
1111         struct nlattr *port[])
1112 {
1113         struct enic *enic = netdev_priv(netdev);
1114
1115         memset(&enic->pp, 0, sizeof(enic->pp));
1116
1117         if (port[IFLA_PORT_REQUEST]) {
1118                 enic->pp.set |= ENIC_SET_REQUEST;
1119                 enic->pp.request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1120         }
1121
1122         if (port[IFLA_PORT_PROFILE]) {
1123                 enic->pp.set |= ENIC_SET_NAME;
1124                 memcpy(enic->pp.name, nla_data(port[IFLA_PORT_PROFILE]),
1125                         PORT_PROFILE_MAX);
1126         }
1127
1128         if (port[IFLA_PORT_INSTANCE_UUID]) {
1129                 enic->pp.set |= ENIC_SET_INSTANCE;
1130                 memcpy(enic->pp.instance_uuid,
1131                         nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
1132         }
1133
1134         if (port[IFLA_PORT_HOST_UUID]) {
1135                 enic->pp.set |= ENIC_SET_HOST;
1136                 memcpy(enic->pp.host_uuid,
1137                         nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
1138         }
1139
1140         /* don't support VFs, yet */
1141         if (vf != PORT_SELF_VF)
1142                 return -EOPNOTSUPP;
1143
1144         if (!(enic->pp.set & ENIC_SET_REQUEST))
1145                 return -EOPNOTSUPP;
1146
1147         if (enic->pp.request == PORT_REQUEST_ASSOCIATE) {
1148
1149                 /* If the interface mac addr hasn't been assigned,
1150                  * assign a random mac addr before setting port-
1151                  * profile.
1152                  */
1153
1154                 if (is_zero_ether_addr(netdev->dev_addr))
1155                         random_ether_addr(netdev->dev_addr);
1156         }
1157
1158         return enic_set_port_profile(enic, netdev->dev_addr);
1159 }
1160
1161 static int enic_get_vf_port(struct net_device *netdev, int vf,
1162         struct sk_buff *skb)
1163 {
1164         struct enic *enic = netdev_priv(netdev);
1165         int err, error, done;
1166         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1167
1168         if (!(enic->pp.set & ENIC_SET_APPLIED))
1169                 return -ENODATA;
1170
1171         err = enic_dev_init_done(enic, &done, &error);
1172         if (err)
1173                 error = err;
1174
1175         switch (error) {
1176         case ERR_SUCCESS:
1177                 if (!done)
1178                         response = PORT_PROFILE_RESPONSE_INPROGRESS;
1179                 break;
1180         case ERR_EINVAL:
1181                 response = PORT_PROFILE_RESPONSE_INVALID;
1182                 break;
1183         case ERR_EBADSTATE:
1184                 response = PORT_PROFILE_RESPONSE_BADSTATE;
1185                 break;
1186         case ERR_ENOMEM:
1187                 response = PORT_PROFILE_RESPONSE_INSUFFICIENT_RESOURCES;
1188                 break;
1189         default:
1190                 response = PORT_PROFILE_RESPONSE_ERROR;
1191                 break;
1192         }
1193
1194         NLA_PUT_U16(skb, IFLA_PORT_REQUEST, enic->pp.request);
1195         NLA_PUT_U16(skb, IFLA_PORT_RESPONSE, response);
1196         if (enic->pp.set & ENIC_SET_NAME)
1197                 NLA_PUT(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX,
1198                         enic->pp.name);
1199         if (enic->pp.set & ENIC_SET_INSTANCE)
1200                 NLA_PUT(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1201                         enic->pp.instance_uuid);
1202         if (enic->pp.set & ENIC_SET_HOST)
1203                 NLA_PUT(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX,
1204                         enic->pp.host_uuid);
1205
1206         return 0;
1207
1208 nla_put_failure:
1209         return -EMSGSIZE;
1210 }
1211
1212 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1213 {
1214         struct enic *enic = vnic_dev_priv(rq->vdev);
1215
1216         if (!buf->os_buf)
1217                 return;
1218
1219         pci_unmap_single(enic->pdev, buf->dma_addr,
1220                 buf->len, PCI_DMA_FROMDEVICE);
1221         dev_kfree_skb_any(buf->os_buf);
1222 }
1223
1224 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1225 {
1226         struct enic *enic = vnic_dev_priv(rq->vdev);
1227         struct net_device *netdev = enic->netdev;
1228         struct sk_buff *skb;
1229         unsigned int len = netdev->mtu + ETH_HLEN;
1230         unsigned int os_buf_index = 0;
1231         dma_addr_t dma_addr;
1232
1233         skb = netdev_alloc_skb_ip_align(netdev, len);
1234         if (!skb)
1235                 return -ENOMEM;
1236
1237         dma_addr = pci_map_single(enic->pdev, skb->data,
1238                 len, PCI_DMA_FROMDEVICE);
1239
1240         enic_queue_rq_desc(rq, skb, os_buf_index,
1241                 dma_addr, len);
1242
1243         return 0;
1244 }
1245
1246 static int enic_rq_alloc_buf_a1(struct vnic_rq *rq)
1247 {
1248         struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
1249
1250         if (vnic_rq_posting_soon(rq)) {
1251
1252                 /* SW workaround for A0 HW erratum: if we're just about
1253                  * to write posted_index, insert a dummy desc
1254                  * of type resvd
1255                  */
1256
1257                 rq_enet_desc_enc(desc, 0, RQ_ENET_TYPE_RESV2, 0);
1258                 vnic_rq_post(rq, 0, 0, 0, 0);
1259         } else {
1260                 return enic_rq_alloc_buf(rq);
1261         }
1262
1263         return 0;
1264 }
1265
1266 static int enic_set_rq_alloc_buf(struct enic *enic)
1267 {
1268         enum vnic_dev_hw_version hw_ver;
1269         int err;
1270
1271         err = vnic_dev_hw_version(enic->vdev, &hw_ver);
1272         if (err)
1273                 return err;
1274
1275         switch (hw_ver) {
1276         case VNIC_DEV_HW_VER_A1:
1277                 enic->rq_alloc_buf = enic_rq_alloc_buf_a1;
1278                 break;
1279         case VNIC_DEV_HW_VER_A2:
1280         case VNIC_DEV_HW_VER_UNKNOWN:
1281                 enic->rq_alloc_buf = enic_rq_alloc_buf;
1282                 break;
1283         default:
1284                 return -ENODEV;
1285         }
1286
1287         return 0;
1288 }
1289
1290 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1291         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1292         int skipped, void *opaque)
1293 {
1294         struct enic *enic = vnic_dev_priv(rq->vdev);
1295         struct net_device *netdev = enic->netdev;
1296         struct sk_buff *skb;
1297
1298         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1299         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1300         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1301         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1302         u8 packet_error;
1303         u16 q_number, completed_index, bytes_written, vlan, checksum;
1304         u32 rss_hash;
1305
1306         if (skipped)
1307                 return;
1308
1309         skb = buf->os_buf;
1310         prefetch(skb->data - NET_IP_ALIGN);
1311         pci_unmap_single(enic->pdev, buf->dma_addr,
1312                 buf->len, PCI_DMA_FROMDEVICE);
1313
1314         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1315                 &type, &color, &q_number, &completed_index,
1316                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1317                 &csum_not_calc, &rss_hash, &bytes_written,
1318                 &packet_error, &vlan_stripped, &vlan, &checksum,
1319                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1320                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1321                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1322                 &fcs_ok);
1323
1324         if (packet_error) {
1325
1326                 if (!fcs_ok) {
1327                         if (bytes_written > 0)
1328                                 enic->rq_bad_fcs++;
1329                         else if (bytes_written == 0)
1330                                 enic->rq_truncated_pkts++;
1331                 }
1332
1333                 dev_kfree_skb_any(skb);
1334
1335                 return;
1336         }
1337
1338         if (eop && bytes_written > 0) {
1339
1340                 /* Good receive
1341                  */
1342
1343                 skb_put(skb, bytes_written);
1344                 skb->protocol = eth_type_trans(skb, netdev);
1345
1346                 if (enic->csum_rx_enabled && !csum_not_calc) {
1347                         skb->csum = htons(checksum);
1348                         skb->ip_summed = CHECKSUM_COMPLETE;
1349                 }
1350
1351                 skb->dev = netdev;
1352
1353                 if (enic->vlan_group && vlan_stripped) {
1354
1355                         if (netdev->features & NETIF_F_GRO)
1356                                 vlan_gro_receive(&enic->napi, enic->vlan_group,
1357                                         vlan, skb);
1358                         else
1359                                 vlan_hwaccel_receive_skb(skb,
1360                                         enic->vlan_group, vlan);
1361
1362                 } else {
1363
1364                         if (netdev->features & NETIF_F_GRO)
1365                                 napi_gro_receive(&enic->napi, skb);
1366                         else
1367                                 netif_receive_skb(skb);
1368
1369                 }
1370
1371         } else {
1372
1373                 /* Buffer overflow
1374                  */
1375
1376                 dev_kfree_skb_any(skb);
1377         }
1378 }
1379
1380 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1381         u8 type, u16 q_number, u16 completed_index, void *opaque)
1382 {
1383         struct enic *enic = vnic_dev_priv(vdev);
1384
1385         vnic_rq_service(&enic->rq[q_number], cq_desc,
1386                 completed_index, VNIC_RQ_RETURN_DESC,
1387                 enic_rq_indicate_buf, opaque);
1388
1389         return 0;
1390 }
1391
1392 static int enic_poll(struct napi_struct *napi, int budget)
1393 {
1394         struct enic *enic = container_of(napi, struct enic, napi);
1395         unsigned int rq_work_to_do = budget;
1396         unsigned int wq_work_to_do = -1; /* no limit */
1397         unsigned int  work_done, rq_work_done, wq_work_done;
1398         int err;
1399
1400         /* Service RQ (first) and WQ
1401          */
1402
1403         rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1404                 rq_work_to_do, enic_rq_service, NULL);
1405
1406         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1407                 wq_work_to_do, enic_wq_service, NULL);
1408
1409         /* Accumulate intr event credits for this polling
1410          * cycle.  An intr event is the completion of a
1411          * a WQ or RQ packet.
1412          */
1413
1414         work_done = rq_work_done + wq_work_done;
1415
1416         if (work_done > 0)
1417                 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1418                         work_done,
1419                         0 /* don't unmask intr */,
1420                         0 /* don't reset intr timer */);
1421
1422         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1423
1424         /* Buffer allocation failed. Stay in polling
1425          * mode so we can try to fill the ring again.
1426          */
1427
1428         if (err)
1429                 rq_work_done = rq_work_to_do;
1430
1431         if (rq_work_done < rq_work_to_do) {
1432
1433                 /* Some work done, but not enough to stay in polling,
1434                  * exit polling
1435                  */
1436
1437                 napi_complete(napi);
1438                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
1439         }
1440
1441         return rq_work_done;
1442 }
1443
1444 static int enic_poll_msix(struct napi_struct *napi, int budget)
1445 {
1446         struct enic *enic = container_of(napi, struct enic, napi);
1447         unsigned int work_to_do = budget;
1448         unsigned int work_done;
1449         int err;
1450
1451         /* Service RQ
1452          */
1453
1454         work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1455                 work_to_do, enic_rq_service, NULL);
1456
1457         /* Return intr event credits for this polling
1458          * cycle.  An intr event is the completion of a
1459          * RQ packet.
1460          */
1461
1462         if (work_done > 0)
1463                 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1464                         work_done,
1465                         0 /* don't unmask intr */,
1466                         0 /* don't reset intr timer */);
1467
1468         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1469
1470         /* Buffer allocation failed. Stay in polling mode
1471          * so we can try to fill the ring again.
1472          */
1473
1474         if (err)
1475                 work_done = work_to_do;
1476
1477         if (work_done < work_to_do) {
1478
1479                 /* Some work done, but not enough to stay in polling,
1480                  * exit polling
1481                  */
1482
1483                 napi_complete(napi);
1484                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1485         }
1486
1487         return work_done;
1488 }
1489
1490 static void enic_notify_timer(unsigned long data)
1491 {
1492         struct enic *enic = (struct enic *)data;
1493
1494         enic_notify_check(enic);
1495
1496         mod_timer(&enic->notify_timer,
1497                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1498 }
1499
1500 static void enic_free_intr(struct enic *enic)
1501 {
1502         struct net_device *netdev = enic->netdev;
1503         unsigned int i;
1504
1505         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1506         case VNIC_DEV_INTR_MODE_INTX:
1507                 free_irq(enic->pdev->irq, netdev);
1508                 break;
1509         case VNIC_DEV_INTR_MODE_MSI:
1510                 free_irq(enic->pdev->irq, enic);
1511                 break;
1512         case VNIC_DEV_INTR_MODE_MSIX:
1513                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1514                         if (enic->msix[i].requested)
1515                                 free_irq(enic->msix_entry[i].vector,
1516                                         enic->msix[i].devid);
1517                 break;
1518         default:
1519                 break;
1520         }
1521 }
1522
1523 static int enic_request_intr(struct enic *enic)
1524 {
1525         struct net_device *netdev = enic->netdev;
1526         unsigned int i;
1527         int err = 0;
1528
1529         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1530
1531         case VNIC_DEV_INTR_MODE_INTX:
1532
1533                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1534                         IRQF_SHARED, netdev->name, netdev);
1535                 break;
1536
1537         case VNIC_DEV_INTR_MODE_MSI:
1538
1539                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1540                         0, netdev->name, enic);
1541                 break;
1542
1543         case VNIC_DEV_INTR_MODE_MSIX:
1544
1545                 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1546                         "%.11s-rx-0", netdev->name);
1547                 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1548                 enic->msix[ENIC_MSIX_RQ].devid = enic;
1549
1550                 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1551                         "%.11s-tx-0", netdev->name);
1552                 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1553                 enic->msix[ENIC_MSIX_WQ].devid = enic;
1554
1555                 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1556                         "%.11s-err", netdev->name);
1557                 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1558                 enic->msix[ENIC_MSIX_ERR].devid = enic;
1559
1560                 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1561                         "%.11s-notify", netdev->name);
1562                 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1563                 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1564
1565                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1566                         err = request_irq(enic->msix_entry[i].vector,
1567                                 enic->msix[i].isr, 0,
1568                                 enic->msix[i].devname,
1569                                 enic->msix[i].devid);
1570                         if (err) {
1571                                 enic_free_intr(enic);
1572                                 break;
1573                         }
1574                         enic->msix[i].requested = 1;
1575                 }
1576
1577                 break;
1578
1579         default:
1580                 break;
1581         }
1582
1583         return err;
1584 }
1585
1586 static void enic_synchronize_irqs(struct enic *enic)
1587 {
1588         unsigned int i;
1589
1590         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1591         case VNIC_DEV_INTR_MODE_INTX:
1592         case VNIC_DEV_INTR_MODE_MSI:
1593                 synchronize_irq(enic->pdev->irq);
1594                 break;
1595         case VNIC_DEV_INTR_MODE_MSIX:
1596                 for (i = 0; i < enic->intr_count; i++)
1597                         synchronize_irq(enic->msix_entry[i].vector);
1598                 break;
1599         default:
1600                 break;
1601         }
1602 }
1603
1604 static int enic_notify_set(struct enic *enic)
1605 {
1606         int err;
1607
1608         spin_lock(&enic->devcmd_lock);
1609         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1610         case VNIC_DEV_INTR_MODE_INTX:
1611                 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1612                 break;
1613         case VNIC_DEV_INTR_MODE_MSIX:
1614                 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1615                 break;
1616         default:
1617                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1618                 break;
1619         }
1620         spin_unlock(&enic->devcmd_lock);
1621
1622         return err;
1623 }
1624
1625 static void enic_notify_timer_start(struct enic *enic)
1626 {
1627         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1628         case VNIC_DEV_INTR_MODE_MSI:
1629                 mod_timer(&enic->notify_timer, jiffies);
1630                 break;
1631         default:
1632                 /* Using intr for notification for INTx/MSI-X */
1633                 break;
1634         };
1635 }
1636
1637 /* rtnl lock is held, process context */
1638 static int enic_open(struct net_device *netdev)
1639 {
1640         struct enic *enic = netdev_priv(netdev);
1641         unsigned int i;
1642         int err;
1643
1644         err = enic_request_intr(enic);
1645         if (err) {
1646                 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1647                         netdev->name);
1648                 return err;
1649         }
1650
1651         err = enic_notify_set(enic);
1652         if (err) {
1653                 printk(KERN_ERR PFX
1654                         "%s: Failed to alloc notify buffer, aborting.\n",
1655                         netdev->name);
1656                 goto err_out_free_intr;
1657         }
1658
1659         for (i = 0; i < enic->rq_count; i++) {
1660                 vnic_rq_fill(&enic->rq[i], enic->rq_alloc_buf);
1661                 /* Need at least one buffer on ring to get going */
1662                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1663                         printk(KERN_ERR PFX
1664                                 "%s: Unable to alloc receive buffers.\n",
1665                                 netdev->name);
1666                         err = -ENOMEM;
1667                         goto err_out_notify_unset;
1668                 }
1669         }
1670
1671         for (i = 0; i < enic->wq_count; i++)
1672                 vnic_wq_enable(&enic->wq[i]);
1673         for (i = 0; i < enic->rq_count; i++)
1674                 vnic_rq_enable(&enic->rq[i]);
1675
1676         enic_dev_add_station_addr(enic);
1677         enic_set_multicast_list(netdev);
1678
1679         netif_wake_queue(netdev);
1680         napi_enable(&enic->napi);
1681         spin_lock(&enic->devcmd_lock);
1682         vnic_dev_enable(enic->vdev);
1683         spin_unlock(&enic->devcmd_lock);
1684
1685         for (i = 0; i < enic->intr_count; i++)
1686                 vnic_intr_unmask(&enic->intr[i]);
1687
1688         enic_notify_timer_start(enic);
1689
1690         return 0;
1691
1692 err_out_notify_unset:
1693         spin_lock(&enic->devcmd_lock);
1694         vnic_dev_notify_unset(enic->vdev);
1695         spin_unlock(&enic->devcmd_lock);
1696 err_out_free_intr:
1697         enic_free_intr(enic);
1698
1699         return err;
1700 }
1701
1702 /* rtnl lock is held, process context */
1703 static int enic_stop(struct net_device *netdev)
1704 {
1705         struct enic *enic = netdev_priv(netdev);
1706         unsigned int i;
1707         int err;
1708
1709         for (i = 0; i < enic->intr_count; i++)
1710                 vnic_intr_mask(&enic->intr[i]);
1711
1712         enic_synchronize_irqs(enic);
1713
1714         del_timer_sync(&enic->notify_timer);
1715
1716         spin_lock(&enic->devcmd_lock);
1717         vnic_dev_disable(enic->vdev);
1718         spin_unlock(&enic->devcmd_lock);
1719         napi_disable(&enic->napi);
1720         netif_carrier_off(netdev);
1721         netif_tx_disable(netdev);
1722
1723         enic_dev_del_station_addr(enic);
1724
1725         for (i = 0; i < enic->wq_count; i++) {
1726                 err = vnic_wq_disable(&enic->wq[i]);
1727                 if (err)
1728                         return err;
1729         }
1730         for (i = 0; i < enic->rq_count; i++) {
1731                 err = vnic_rq_disable(&enic->rq[i]);
1732                 if (err)
1733                         return err;
1734         }
1735
1736         spin_lock(&enic->devcmd_lock);
1737         vnic_dev_notify_unset(enic->vdev);
1738         spin_unlock(&enic->devcmd_lock);
1739         enic_free_intr(enic);
1740
1741         for (i = 0; i < enic->wq_count; i++)
1742                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1743         for (i = 0; i < enic->rq_count; i++)
1744                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1745         for (i = 0; i < enic->cq_count; i++)
1746                 vnic_cq_clean(&enic->cq[i]);
1747         for (i = 0; i < enic->intr_count; i++)
1748                 vnic_intr_clean(&enic->intr[i]);
1749
1750         return 0;
1751 }
1752
1753 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1754 {
1755         struct enic *enic = netdev_priv(netdev);
1756         int running = netif_running(netdev);
1757
1758         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1759                 return -EINVAL;
1760
1761         if (running)
1762                 enic_stop(netdev);
1763
1764         netdev->mtu = new_mtu;
1765
1766         if (netdev->mtu > enic->port_mtu)
1767                 printk(KERN_WARNING PFX
1768                         "%s: interface MTU (%d) set higher "
1769                         "than port MTU (%d)\n",
1770                         netdev->name, netdev->mtu, enic->port_mtu);
1771
1772         if (running)
1773                 enic_open(netdev);
1774
1775         return 0;
1776 }
1777
1778 #ifdef CONFIG_NET_POLL_CONTROLLER
1779 static void enic_poll_controller(struct net_device *netdev)
1780 {
1781         struct enic *enic = netdev_priv(netdev);
1782         struct vnic_dev *vdev = enic->vdev;
1783
1784         switch (vnic_dev_get_intr_mode(vdev)) {
1785         case VNIC_DEV_INTR_MODE_MSIX:
1786                 enic_isr_msix_rq(enic->pdev->irq, enic);
1787                 enic_isr_msix_wq(enic->pdev->irq, enic);
1788                 break;
1789         case VNIC_DEV_INTR_MODE_MSI:
1790                 enic_isr_msi(enic->pdev->irq, enic);
1791                 break;
1792         case VNIC_DEV_INTR_MODE_INTX:
1793                 enic_isr_legacy(enic->pdev->irq, netdev);
1794                 break;
1795         default:
1796                 break;
1797         }
1798 }
1799 #endif
1800
1801 static int enic_dev_wait(struct vnic_dev *vdev,
1802         int (*start)(struct vnic_dev *, int),
1803         int (*finished)(struct vnic_dev *, int *),
1804         int arg)
1805 {
1806         unsigned long time;
1807         int done;
1808         int err;
1809
1810         BUG_ON(in_interrupt());
1811
1812         err = start(vdev, arg);
1813         if (err)
1814                 return err;
1815
1816         /* Wait for func to complete...2 seconds max
1817          */
1818
1819         time = jiffies + (HZ * 2);
1820         do {
1821
1822                 err = finished(vdev, &done);
1823                 if (err)
1824                         return err;
1825
1826                 if (done)
1827                         return 0;
1828
1829                 schedule_timeout_uninterruptible(HZ / 10);
1830
1831         } while (time_after(time, jiffies));
1832
1833         return -ETIMEDOUT;
1834 }
1835
1836 static int enic_dev_open(struct enic *enic)
1837 {
1838         int err;
1839
1840         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1841                 vnic_dev_open_done, 0);
1842         if (err)
1843                 printk(KERN_ERR PFX
1844                         "vNIC device open failed, err %d.\n", err);
1845
1846         return err;
1847 }
1848
1849 static int enic_dev_soft_reset(struct enic *enic)
1850 {
1851         int err;
1852
1853         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1854                 vnic_dev_soft_reset_done, 0);
1855         if (err)
1856                 printk(KERN_ERR PFX
1857                         "vNIC soft reset failed, err %d.\n", err);
1858
1859         return err;
1860 }
1861
1862 static int enic_set_niccfg(struct enic *enic)
1863 {
1864         const u8 rss_default_cpu = 0;
1865         const u8 rss_hash_type = 0;
1866         const u8 rss_hash_bits = 0;
1867         const u8 rss_base_cpu = 0;
1868         const u8 rss_enable = 0;
1869         const u8 tso_ipid_split_en = 0;
1870         const u8 ig_vlan_strip_en = 1;
1871
1872         /* Enable VLAN tag stripping.  RSS not enabled (yet).
1873          */
1874
1875         return enic_set_nic_cfg(enic,
1876                 rss_default_cpu, rss_hash_type,
1877                 rss_hash_bits, rss_base_cpu,
1878                 rss_enable, tso_ipid_split_en,
1879                 ig_vlan_strip_en);
1880 }
1881
1882 static void enic_reset(struct work_struct *work)
1883 {
1884         struct enic *enic = container_of(work, struct enic, reset);
1885
1886         if (!netif_running(enic->netdev))
1887                 return;
1888
1889         rtnl_lock();
1890
1891         spin_lock(&enic->devcmd_lock);
1892         vnic_dev_hang_notify(enic->vdev);
1893         spin_unlock(&enic->devcmd_lock);
1894
1895         enic_stop(enic->netdev);
1896         enic_dev_soft_reset(enic);
1897         vnic_dev_init(enic->vdev, 0);
1898         enic_reset_mcaddrs(enic);
1899         enic_init_vnic_resources(enic);
1900         enic_set_niccfg(enic);
1901         enic_open(enic->netdev);
1902
1903         rtnl_unlock();
1904 }
1905
1906 static int enic_set_intr_mode(struct enic *enic)
1907 {
1908         unsigned int n = 1;
1909         unsigned int m = 1;
1910         unsigned int i;
1911
1912         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1913          * system capabilities.
1914          *
1915          * Try MSI-X first
1916          *
1917          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1918          * (the second to last INTR is used for WQ/RQ errors)
1919          * (the last INTR is used for notifications)
1920          */
1921
1922         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1923         for (i = 0; i < n + m + 2; i++)
1924                 enic->msix_entry[i].entry = i;
1925
1926         if (enic->config.intr_mode < 1 &&
1927             enic->rq_count >= n &&
1928             enic->wq_count >= m &&
1929             enic->cq_count >= n + m &&
1930             enic->intr_count >= n + m + 2 &&
1931             !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1932
1933                 enic->rq_count = n;
1934                 enic->wq_count = m;
1935                 enic->cq_count = n + m;
1936                 enic->intr_count = n + m + 2;
1937
1938                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1939
1940                 return 0;
1941         }
1942
1943         /* Next try MSI
1944          *
1945          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1946          */
1947
1948         if (enic->config.intr_mode < 2 &&
1949             enic->rq_count >= 1 &&
1950             enic->wq_count >= 1 &&
1951             enic->cq_count >= 2 &&
1952             enic->intr_count >= 1 &&
1953             !pci_enable_msi(enic->pdev)) {
1954
1955                 enic->rq_count = 1;
1956                 enic->wq_count = 1;
1957                 enic->cq_count = 2;
1958                 enic->intr_count = 1;
1959
1960                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1961
1962                 return 0;
1963         }
1964
1965         /* Next try INTx
1966          *
1967          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1968          * (the first INTR is used for WQ/RQ)
1969          * (the second INTR is used for WQ/RQ errors)
1970          * (the last INTR is used for notifications)
1971          */
1972
1973         if (enic->config.intr_mode < 3 &&
1974             enic->rq_count >= 1 &&
1975             enic->wq_count >= 1 &&
1976             enic->cq_count >= 2 &&
1977             enic->intr_count >= 3) {
1978
1979                 enic->rq_count = 1;
1980                 enic->wq_count = 1;
1981                 enic->cq_count = 2;
1982                 enic->intr_count = 3;
1983
1984                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1985
1986                 return 0;
1987         }
1988
1989         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1990
1991         return -EINVAL;
1992 }
1993
1994 static void enic_clear_intr_mode(struct enic *enic)
1995 {
1996         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1997         case VNIC_DEV_INTR_MODE_MSIX:
1998                 pci_disable_msix(enic->pdev);
1999                 break;
2000         case VNIC_DEV_INTR_MODE_MSI:
2001                 pci_disable_msi(enic->pdev);
2002                 break;
2003         default:
2004                 break;
2005         }
2006
2007         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2008 }
2009
2010 static const struct net_device_ops enic_netdev_dynamic_ops = {
2011         .ndo_open               = enic_open,
2012         .ndo_stop               = enic_stop,
2013         .ndo_start_xmit         = enic_hard_start_xmit,
2014         .ndo_get_stats          = enic_get_stats,
2015         .ndo_validate_addr      = eth_validate_addr,
2016         .ndo_set_multicast_list = enic_set_multicast_list,
2017         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2018         .ndo_change_mtu         = enic_change_mtu,
2019         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2020         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2021         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2022         .ndo_tx_timeout         = enic_tx_timeout,
2023         .ndo_set_vf_port        = enic_set_vf_port,
2024         .ndo_get_vf_port        = enic_get_vf_port,
2025 #ifdef CONFIG_NET_POLL_CONTROLLER
2026         .ndo_poll_controller    = enic_poll_controller,
2027 #endif
2028 };
2029
2030 static const struct net_device_ops enic_netdev_ops = {
2031         .ndo_open               = enic_open,
2032         .ndo_stop               = enic_stop,
2033         .ndo_start_xmit         = enic_hard_start_xmit,
2034         .ndo_get_stats          = enic_get_stats,
2035         .ndo_validate_addr      = eth_validate_addr,
2036         .ndo_set_multicast_list = enic_set_multicast_list,
2037         .ndo_set_mac_address    = enic_set_mac_address,
2038         .ndo_change_mtu         = enic_change_mtu,
2039         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2040         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2041         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2042         .ndo_tx_timeout         = enic_tx_timeout,
2043 #ifdef CONFIG_NET_POLL_CONTROLLER
2044         .ndo_poll_controller    = enic_poll_controller,
2045 #endif
2046 };
2047
2048 void enic_dev_deinit(struct enic *enic)
2049 {
2050         netif_napi_del(&enic->napi);
2051         enic_free_vnic_resources(enic);
2052         enic_clear_intr_mode(enic);
2053 }
2054
2055 int enic_dev_init(struct enic *enic)
2056 {
2057         struct net_device *netdev = enic->netdev;
2058         int err;
2059
2060         /* Get vNIC configuration
2061          */
2062
2063         err = enic_get_vnic_config(enic);
2064         if (err) {
2065                 printk(KERN_ERR PFX
2066                         "Get vNIC configuration failed, aborting.\n");
2067                 return err;
2068         }
2069
2070         /* Get available resource counts
2071          */
2072
2073         enic_get_res_counts(enic);
2074
2075         /* Set interrupt mode based on resource counts and system
2076          * capabilities
2077          */
2078
2079         err = enic_set_intr_mode(enic);
2080         if (err) {
2081                 printk(KERN_ERR PFX
2082                         "Failed to set intr mode based on resource "
2083                         "counts and system capabilities, aborting.\n");
2084                 return err;
2085         }
2086
2087         /* Allocate and configure vNIC resources
2088          */
2089
2090         err = enic_alloc_vnic_resources(enic);
2091         if (err) {
2092                 printk(KERN_ERR PFX
2093                         "Failed to alloc vNIC resources, aborting.\n");
2094                 goto err_out_free_vnic_resources;
2095         }
2096
2097         enic_init_vnic_resources(enic);
2098
2099         err = enic_set_rq_alloc_buf(enic);
2100         if (err) {
2101                 printk(KERN_ERR PFX
2102                         "Failed to set RQ buffer allocator, aborting.\n");
2103                 goto err_out_free_vnic_resources;
2104         }
2105
2106         err = enic_set_niccfg(enic);
2107         if (err) {
2108                 printk(KERN_ERR PFX
2109                         "Failed to config nic, aborting.\n");
2110                 goto err_out_free_vnic_resources;
2111         }
2112
2113         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2114         default:
2115                 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
2116                 break;
2117         case VNIC_DEV_INTR_MODE_MSIX:
2118                 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
2119                 break;
2120         }
2121
2122         return 0;
2123
2124 err_out_free_vnic_resources:
2125         enic_clear_intr_mode(enic);
2126         enic_free_vnic_resources(enic);
2127
2128         return err;
2129 }
2130
2131 static void enic_iounmap(struct enic *enic)
2132 {
2133         unsigned int i;
2134
2135         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2136                 if (enic->bar[i].vaddr)
2137                         iounmap(enic->bar[i].vaddr);
2138 }
2139
2140 static int __devinit enic_probe(struct pci_dev *pdev,
2141         const struct pci_device_id *ent)
2142 {
2143         struct net_device *netdev;
2144         struct enic *enic;
2145         int using_dac = 0;
2146         unsigned int i;
2147         int err;
2148
2149         /* Allocate net device structure and initialize.  Private
2150          * instance data is initialized to zero.
2151          */
2152
2153         netdev = alloc_etherdev(sizeof(struct enic));
2154         if (!netdev) {
2155                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
2156                 return -ENOMEM;
2157         }
2158
2159         pci_set_drvdata(pdev, netdev);
2160
2161         SET_NETDEV_DEV(netdev, &pdev->dev);
2162
2163         enic = netdev_priv(netdev);
2164         enic->netdev = netdev;
2165         enic->pdev = pdev;
2166
2167         /* Setup PCI resources
2168          */
2169
2170         err = pci_enable_device(pdev);
2171         if (err) {
2172                 printk(KERN_ERR PFX
2173                         "Cannot enable PCI device, aborting.\n");
2174                 goto err_out_free_netdev;
2175         }
2176
2177         err = pci_request_regions(pdev, DRV_NAME);
2178         if (err) {
2179                 printk(KERN_ERR PFX
2180                         "Cannot request PCI regions, aborting.\n");
2181                 goto err_out_disable_device;
2182         }
2183
2184         pci_set_master(pdev);
2185
2186         /* Query PCI controller on system for DMA addressing
2187          * limitation for the device.  Try 40-bit first, and
2188          * fail to 32-bit.
2189          */
2190
2191         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
2192         if (err) {
2193                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2194                 if (err) {
2195                         printk(KERN_ERR PFX
2196                                 "No usable DMA configuration, aborting.\n");
2197                         goto err_out_release_regions;
2198                 }
2199                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2200                 if (err) {
2201                         printk(KERN_ERR PFX
2202                                 "Unable to obtain 32-bit DMA "
2203                                 "for consistent allocations, aborting.\n");
2204                         goto err_out_release_regions;
2205                 }
2206         } else {
2207                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
2208                 if (err) {
2209                         printk(KERN_ERR PFX
2210                                 "Unable to obtain 40-bit DMA "
2211                                 "for consistent allocations, aborting.\n");
2212                         goto err_out_release_regions;
2213                 }
2214                 using_dac = 1;
2215         }
2216
2217         /* Map vNIC resources from BAR0-5
2218          */
2219
2220         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2221                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2222                         continue;
2223                 enic->bar[i].len = pci_resource_len(pdev, i);
2224                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2225                 if (!enic->bar[i].vaddr) {
2226                         printk(KERN_ERR PFX
2227                                 "Cannot memory-map BAR %d, aborting.\n", i);
2228                         err = -ENODEV;
2229                         goto err_out_iounmap;
2230                 }
2231                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2232         }
2233
2234         /* Register vNIC device
2235          */
2236
2237         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2238                 ARRAY_SIZE(enic->bar));
2239         if (!enic->vdev) {
2240                 printk(KERN_ERR PFX
2241                         "vNIC registration failed, aborting.\n");
2242                 err = -ENODEV;
2243                 goto err_out_iounmap;
2244         }
2245
2246         /* Issue device open to get device in known state
2247          */
2248
2249         err = enic_dev_open(enic);
2250         if (err) {
2251                 printk(KERN_ERR PFX
2252                         "vNIC dev open failed, aborting.\n");
2253                 goto err_out_vnic_unregister;
2254         }
2255
2256         /* Issue device init to initialize the vnic-to-switch link.
2257          * We'll start with carrier off and wait for link UP
2258          * notification later to turn on carrier.  We don't need
2259          * to wait here for the vnic-to-switch link initialization
2260          * to complete; link UP notification is the indication that
2261          * the process is complete.
2262          */
2263
2264         netif_carrier_off(netdev);
2265
2266         if (!enic_is_dynamic(enic)) {
2267                 err = vnic_dev_init(enic->vdev, 0);
2268                 if (err) {
2269                         printk(KERN_ERR PFX
2270                                 "vNIC dev init failed, aborting.\n");
2271                         goto err_out_dev_close;
2272                 }
2273         }
2274
2275         err = enic_dev_init(enic);
2276         if (err) {
2277                 printk(KERN_ERR PFX
2278                         "Device initialization failed, aborting.\n");
2279                 goto err_out_dev_close;
2280         }
2281
2282         /* Setup notification timer, HW reset task, and locks
2283          */
2284
2285         init_timer(&enic->notify_timer);
2286         enic->notify_timer.function = enic_notify_timer;
2287         enic->notify_timer.data = (unsigned long)enic;
2288
2289         INIT_WORK(&enic->reset, enic_reset);
2290
2291         for (i = 0; i < enic->wq_count; i++)
2292                 spin_lock_init(&enic->wq_lock[i]);
2293
2294         spin_lock_init(&enic->devcmd_lock);
2295
2296         /* Register net device
2297          */
2298
2299         enic->port_mtu = enic->config.mtu;
2300         (void)enic_change_mtu(netdev, enic->port_mtu);
2301
2302         err = enic_set_mac_addr(netdev, enic->mac_addr);
2303         if (err) {
2304                 printk(KERN_ERR PFX
2305                         "Invalid MAC address, aborting.\n");
2306                 goto err_out_dev_deinit;
2307         }
2308
2309         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2310         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2311
2312         if (enic_is_dynamic(enic))
2313                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2314         else
2315                 netdev->netdev_ops = &enic_netdev_ops;
2316
2317         netdev->watchdog_timeo = 2 * HZ;
2318         netdev->ethtool_ops = &enic_ethtool_ops;
2319
2320         netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
2321         if (ENIC_SETTING(enic, TXCSUM))
2322                 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2323         if (ENIC_SETTING(enic, TSO))
2324                 netdev->features |= NETIF_F_TSO |
2325                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2326         if (ENIC_SETTING(enic, LRO))
2327                 netdev->features |= NETIF_F_GRO;
2328         if (using_dac)
2329                 netdev->features |= NETIF_F_HIGHDMA;
2330
2331         enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
2332
2333         err = register_netdev(netdev);
2334         if (err) {
2335                 printk(KERN_ERR PFX
2336                         "Cannot register net device, aborting.\n");
2337                 goto err_out_dev_deinit;
2338         }
2339
2340         return 0;
2341
2342 err_out_dev_deinit:
2343         enic_dev_deinit(enic);
2344 err_out_dev_close:
2345         vnic_dev_close(enic->vdev);
2346 err_out_vnic_unregister:
2347         vnic_dev_unregister(enic->vdev);
2348 err_out_iounmap:
2349         enic_iounmap(enic);
2350 err_out_release_regions:
2351         pci_release_regions(pdev);
2352 err_out_disable_device:
2353         pci_disable_device(pdev);
2354 err_out_free_netdev:
2355         pci_set_drvdata(pdev, NULL);
2356         free_netdev(netdev);
2357
2358         return err;
2359 }
2360
2361 static void __devexit enic_remove(struct pci_dev *pdev)
2362 {
2363         struct net_device *netdev = pci_get_drvdata(pdev);
2364
2365         if (netdev) {
2366                 struct enic *enic = netdev_priv(netdev);
2367
2368                 flush_scheduled_work();
2369                 unregister_netdev(netdev);
2370                 enic_dev_deinit(enic);
2371                 vnic_dev_close(enic->vdev);
2372                 vnic_dev_unregister(enic->vdev);
2373                 enic_iounmap(enic);
2374                 pci_release_regions(pdev);
2375                 pci_disable_device(pdev);
2376                 pci_set_drvdata(pdev, NULL);
2377                 free_netdev(netdev);
2378         }
2379 }
2380
2381 static struct pci_driver enic_driver = {
2382         .name = DRV_NAME,
2383         .id_table = enic_id_table,
2384         .probe = enic_probe,
2385         .remove = __devexit_p(enic_remove),
2386 };
2387
2388 static int __init enic_init_module(void)
2389 {
2390         printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2391
2392         return pci_register_driver(&enic_driver);
2393 }
2394
2395 static void __exit enic_cleanup_module(void)
2396 {
2397         pci_unregister_driver(&enic_driver);
2398 }
2399
2400 module_init(enic_init_module);
2401 module_exit(enic_cleanup_module);