enic: Clean up: remove enic_dev_stats_clear
[linux-2.6.git] / drivers / net / enic / enic_main.c
1 /*
2  * Copyright 2008-2010 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 <linux/rtnetlink.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 int enic_dev_fw_info(struct enic *enic,
149         struct vnic_devcmd_fw_info **fw_info)
150 {
151         int err;
152
153         spin_lock(&enic->devcmd_lock);
154         err = vnic_dev_fw_info(enic->vdev, fw_info);
155         spin_unlock(&enic->devcmd_lock);
156
157         return err;
158 }
159
160 static void enic_get_drvinfo(struct net_device *netdev,
161         struct ethtool_drvinfo *drvinfo)
162 {
163         struct enic *enic = netdev_priv(netdev);
164         struct vnic_devcmd_fw_info *fw_info;
165
166         enic_dev_fw_info(enic, &fw_info);
167
168         strncpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
169         strncpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
170         strncpy(drvinfo->fw_version, fw_info->fw_version,
171                 sizeof(drvinfo->fw_version));
172         strncpy(drvinfo->bus_info, pci_name(enic->pdev),
173                 sizeof(drvinfo->bus_info));
174 }
175
176 static void enic_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
177 {
178         unsigned int i;
179
180         switch (stringset) {
181         case ETH_SS_STATS:
182                 for (i = 0; i < enic_n_tx_stats; i++) {
183                         memcpy(data, enic_tx_stats[i].name, ETH_GSTRING_LEN);
184                         data += ETH_GSTRING_LEN;
185                 }
186                 for (i = 0; i < enic_n_rx_stats; i++) {
187                         memcpy(data, enic_rx_stats[i].name, ETH_GSTRING_LEN);
188                         data += ETH_GSTRING_LEN;
189                 }
190                 break;
191         }
192 }
193
194 static int enic_get_sset_count(struct net_device *netdev, int sset)
195 {
196         switch (sset) {
197         case ETH_SS_STATS:
198                 return enic_n_tx_stats + enic_n_rx_stats;
199         default:
200                 return -EOPNOTSUPP;
201         }
202 }
203
204 static int enic_dev_stats_dump(struct enic *enic, struct vnic_stats **vstats)
205 {
206         int err;
207
208         spin_lock(&enic->devcmd_lock);
209         err = vnic_dev_stats_dump(enic->vdev, vstats);
210         spin_unlock(&enic->devcmd_lock);
211
212         return err;
213 }
214
215 static void enic_get_ethtool_stats(struct net_device *netdev,
216         struct ethtool_stats *stats, u64 *data)
217 {
218         struct enic *enic = netdev_priv(netdev);
219         struct vnic_stats *vstats;
220         unsigned int i;
221
222         enic_dev_stats_dump(enic, &vstats);
223
224         for (i = 0; i < enic_n_tx_stats; i++)
225                 *(data++) = ((u64 *)&vstats->tx)[enic_tx_stats[i].offset];
226         for (i = 0; i < enic_n_rx_stats; i++)
227                 *(data++) = ((u64 *)&vstats->rx)[enic_rx_stats[i].offset];
228 }
229
230 static u32 enic_get_rx_csum(struct net_device *netdev)
231 {
232         struct enic *enic = netdev_priv(netdev);
233         return enic->csum_rx_enabled;
234 }
235
236 static int enic_set_rx_csum(struct net_device *netdev, u32 data)
237 {
238         struct enic *enic = netdev_priv(netdev);
239
240         if (data && !ENIC_SETTING(enic, RXCSUM))
241                 return -EINVAL;
242
243         enic->csum_rx_enabled = !!data;
244
245         return 0;
246 }
247
248 static int enic_set_tx_csum(struct net_device *netdev, u32 data)
249 {
250         struct enic *enic = netdev_priv(netdev);
251
252         if (data && !ENIC_SETTING(enic, TXCSUM))
253                 return -EINVAL;
254
255         if (data)
256                 netdev->features |= NETIF_F_HW_CSUM;
257         else
258                 netdev->features &= ~NETIF_F_HW_CSUM;
259
260         return 0;
261 }
262
263 static int enic_set_tso(struct net_device *netdev, u32 data)
264 {
265         struct enic *enic = netdev_priv(netdev);
266
267         if (data && !ENIC_SETTING(enic, TSO))
268                 return -EINVAL;
269
270         if (data)
271                 netdev->features |=
272                         NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN;
273         else
274                 netdev->features &=
275                         ~(NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN);
276
277         return 0;
278 }
279
280 static u32 enic_get_msglevel(struct net_device *netdev)
281 {
282         struct enic *enic = netdev_priv(netdev);
283         return enic->msg_enable;
284 }
285
286 static void enic_set_msglevel(struct net_device *netdev, u32 value)
287 {
288         struct enic *enic = netdev_priv(netdev);
289         enic->msg_enable = value;
290 }
291
292 static int enic_get_coalesce(struct net_device *netdev,
293         struct ethtool_coalesce *ecmd)
294 {
295         struct enic *enic = netdev_priv(netdev);
296
297         ecmd->tx_coalesce_usecs = enic->tx_coalesce_usecs;
298         ecmd->rx_coalesce_usecs = enic->rx_coalesce_usecs;
299
300         return 0;
301 }
302
303 static int enic_set_coalesce(struct net_device *netdev,
304         struct ethtool_coalesce *ecmd)
305 {
306         struct enic *enic = netdev_priv(netdev);
307         u32 tx_coalesce_usecs;
308         u32 rx_coalesce_usecs;
309
310         tx_coalesce_usecs = min_t(u32,
311                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
312                 ecmd->tx_coalesce_usecs);
313         rx_coalesce_usecs = min_t(u32,
314                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
315                 ecmd->rx_coalesce_usecs);
316
317         switch (vnic_dev_get_intr_mode(enic->vdev)) {
318         case VNIC_DEV_INTR_MODE_INTX:
319                 if (tx_coalesce_usecs != rx_coalesce_usecs)
320                         return -EINVAL;
321
322                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_INTX_WQ_RQ],
323                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
324                 break;
325         case VNIC_DEV_INTR_MODE_MSI:
326                 if (tx_coalesce_usecs != rx_coalesce_usecs)
327                         return -EINVAL;
328
329                 vnic_intr_coalescing_timer_set(&enic->intr[0],
330                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
331                 break;
332         case VNIC_DEV_INTR_MODE_MSIX:
333                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_MSIX_WQ],
334                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
335                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_MSIX_RQ],
336                         INTR_COALESCE_USEC_TO_HW(rx_coalesce_usecs));
337                 break;
338         default:
339                 break;
340         }
341
342         enic->tx_coalesce_usecs = tx_coalesce_usecs;
343         enic->rx_coalesce_usecs = rx_coalesce_usecs;
344
345         return 0;
346 }
347
348 static const struct ethtool_ops enic_ethtool_ops = {
349         .get_settings = enic_get_settings,
350         .get_drvinfo = enic_get_drvinfo,
351         .get_msglevel = enic_get_msglevel,
352         .set_msglevel = enic_set_msglevel,
353         .get_link = ethtool_op_get_link,
354         .get_strings = enic_get_strings,
355         .get_sset_count = enic_get_sset_count,
356         .get_ethtool_stats = enic_get_ethtool_stats,
357         .get_rx_csum = enic_get_rx_csum,
358         .set_rx_csum = enic_set_rx_csum,
359         .get_tx_csum = ethtool_op_get_tx_csum,
360         .set_tx_csum = enic_set_tx_csum,
361         .get_sg = ethtool_op_get_sg,
362         .set_sg = ethtool_op_set_sg,
363         .get_tso = ethtool_op_get_tso,
364         .set_tso = enic_set_tso,
365         .get_coalesce = enic_get_coalesce,
366         .set_coalesce = enic_set_coalesce,
367         .get_flags = ethtool_op_get_flags,
368 };
369
370 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
371 {
372         struct enic *enic = vnic_dev_priv(wq->vdev);
373
374         if (buf->sop)
375                 pci_unmap_single(enic->pdev, buf->dma_addr,
376                         buf->len, PCI_DMA_TODEVICE);
377         else
378                 pci_unmap_page(enic->pdev, buf->dma_addr,
379                         buf->len, PCI_DMA_TODEVICE);
380
381         if (buf->os_buf)
382                 dev_kfree_skb_any(buf->os_buf);
383 }
384
385 static void enic_wq_free_buf(struct vnic_wq *wq,
386         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
387 {
388         enic_free_wq_buf(wq, buf);
389 }
390
391 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
392         u8 type, u16 q_number, u16 completed_index, void *opaque)
393 {
394         struct enic *enic = vnic_dev_priv(vdev);
395
396         spin_lock(&enic->wq_lock[q_number]);
397
398         vnic_wq_service(&enic->wq[q_number], cq_desc,
399                 completed_index, enic_wq_free_buf,
400                 opaque);
401
402         if (netif_queue_stopped(enic->netdev) &&
403             vnic_wq_desc_avail(&enic->wq[q_number]) >=
404             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
405                 netif_wake_queue(enic->netdev);
406
407         spin_unlock(&enic->wq_lock[q_number]);
408
409         return 0;
410 }
411
412 static void enic_log_q_error(struct enic *enic)
413 {
414         unsigned int i;
415         u32 error_status;
416
417         for (i = 0; i < enic->wq_count; i++) {
418                 error_status = vnic_wq_error_status(&enic->wq[i]);
419                 if (error_status)
420                         netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
421                                 i, error_status);
422         }
423
424         for (i = 0; i < enic->rq_count; i++) {
425                 error_status = vnic_rq_error_status(&enic->rq[i]);
426                 if (error_status)
427                         netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
428                                 i, error_status);
429         }
430 }
431
432 static void enic_msglvl_check(struct enic *enic)
433 {
434         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
435
436         if (msg_enable != enic->msg_enable) {
437                 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
438                         enic->msg_enable, msg_enable);
439                 enic->msg_enable = msg_enable;
440         }
441 }
442
443 static void enic_mtu_check(struct enic *enic)
444 {
445         u32 mtu = vnic_dev_mtu(enic->vdev);
446         struct net_device *netdev = enic->netdev;
447
448         if (mtu && mtu != enic->port_mtu) {
449                 enic->port_mtu = mtu;
450                 if (mtu < netdev->mtu)
451                         netdev_warn(netdev,
452                                 "interface MTU (%d) set higher "
453                                 "than switch port MTU (%d)\n",
454                                 netdev->mtu, mtu);
455         }
456 }
457
458 static void enic_link_check(struct enic *enic)
459 {
460         int link_status = vnic_dev_link_status(enic->vdev);
461         int carrier_ok = netif_carrier_ok(enic->netdev);
462
463         if (link_status && !carrier_ok) {
464                 netdev_info(enic->netdev, "Link UP\n");
465                 netif_carrier_on(enic->netdev);
466         } else if (!link_status && carrier_ok) {
467                 netdev_info(enic->netdev, "Link DOWN\n");
468                 netif_carrier_off(enic->netdev);
469         }
470 }
471
472 static void enic_notify_check(struct enic *enic)
473 {
474         enic_msglvl_check(enic);
475         enic_mtu_check(enic);
476         enic_link_check(enic);
477 }
478
479 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
480
481 static irqreturn_t enic_isr_legacy(int irq, void *data)
482 {
483         struct net_device *netdev = data;
484         struct enic *enic = netdev_priv(netdev);
485         u32 pba;
486
487         vnic_intr_mask(&enic->intr[ENIC_INTX_WQ_RQ]);
488
489         pba = vnic_intr_legacy_pba(enic->legacy_pba);
490         if (!pba) {
491                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
492                 return IRQ_NONE;        /* not our interrupt */
493         }
494
495         if (ENIC_TEST_INTR(pba, ENIC_INTX_NOTIFY)) {
496                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_NOTIFY]);
497                 enic_notify_check(enic);
498         }
499
500         if (ENIC_TEST_INTR(pba, ENIC_INTX_ERR)) {
501                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_ERR]);
502                 enic_log_q_error(enic);
503                 /* schedule recovery from WQ/RQ error */
504                 schedule_work(&enic->reset);
505                 return IRQ_HANDLED;
506         }
507
508         if (ENIC_TEST_INTR(pba, ENIC_INTX_WQ_RQ)) {
509                 if (napi_schedule_prep(&enic->napi))
510                         __napi_schedule(&enic->napi);
511         } else {
512                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
513         }
514
515         return IRQ_HANDLED;
516 }
517
518 static irqreturn_t enic_isr_msi(int irq, void *data)
519 {
520         struct enic *enic = data;
521
522         /* With MSI, there is no sharing of interrupts, so this is
523          * our interrupt and there is no need to ack it.  The device
524          * is not providing per-vector masking, so the OS will not
525          * write to PCI config space to mask/unmask the interrupt.
526          * We're using mask_on_assertion for MSI, so the device
527          * automatically masks the interrupt when the interrupt is
528          * generated.  Later, when exiting polling, the interrupt
529          * will be unmasked (see enic_poll).
530          *
531          * Also, the device uses the same PCIe Traffic Class (TC)
532          * for Memory Write data and MSI, so there are no ordering
533          * issues; the MSI will always arrive at the Root Complex
534          * _after_ corresponding Memory Writes (i.e. descriptor
535          * writes).
536          */
537
538         napi_schedule(&enic->napi);
539
540         return IRQ_HANDLED;
541 }
542
543 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
544 {
545         struct enic *enic = data;
546
547         /* schedule NAPI polling for RQ cleanup */
548         napi_schedule(&enic->napi);
549
550         return IRQ_HANDLED;
551 }
552
553 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
554 {
555         struct enic *enic = data;
556         unsigned int wq_work_to_do = -1; /* no limit */
557         unsigned int wq_work_done;
558
559         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
560                 wq_work_to_do, enic_wq_service, NULL);
561
562         vnic_intr_return_credits(&enic->intr[ENIC_MSIX_WQ],
563                 wq_work_done,
564                 1 /* unmask intr */,
565                 1 /* reset intr timer */);
566
567         return IRQ_HANDLED;
568 }
569
570 static irqreturn_t enic_isr_msix_err(int irq, void *data)
571 {
572         struct enic *enic = data;
573
574         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_ERR]);
575
576         enic_log_q_error(enic);
577
578         /* schedule recovery from WQ/RQ error */
579         schedule_work(&enic->reset);
580
581         return IRQ_HANDLED;
582 }
583
584 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
585 {
586         struct enic *enic = data;
587
588         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_NOTIFY]);
589         enic_notify_check(enic);
590
591         return IRQ_HANDLED;
592 }
593
594 static inline void enic_queue_wq_skb_cont(struct enic *enic,
595         struct vnic_wq *wq, struct sk_buff *skb,
596         unsigned int len_left, int loopback)
597 {
598         skb_frag_t *frag;
599
600         /* Queue additional data fragments */
601         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
602                 len_left -= frag->size;
603                 enic_queue_wq_desc_cont(wq, skb,
604                         pci_map_page(enic->pdev, frag->page,
605                                 frag->page_offset, frag->size,
606                                 PCI_DMA_TODEVICE),
607                         frag->size,
608                         (len_left == 0),        /* EOP? */
609                         loopback);
610         }
611 }
612
613 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
614         struct vnic_wq *wq, struct sk_buff *skb,
615         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
616 {
617         unsigned int head_len = skb_headlen(skb);
618         unsigned int len_left = skb->len - head_len;
619         int eop = (len_left == 0);
620
621         /* Queue the main skb fragment. The fragments are no larger
622          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
623          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
624          * per fragment is queued.
625          */
626         enic_queue_wq_desc(wq, skb,
627                 pci_map_single(enic->pdev, skb->data,
628                         head_len, PCI_DMA_TODEVICE),
629                 head_len,
630                 vlan_tag_insert, vlan_tag,
631                 eop, loopback);
632
633         if (!eop)
634                 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
635 }
636
637 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
638         struct vnic_wq *wq, struct sk_buff *skb,
639         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
640 {
641         unsigned int head_len = skb_headlen(skb);
642         unsigned int len_left = skb->len - head_len;
643         unsigned int hdr_len = skb_transport_offset(skb);
644         unsigned int csum_offset = hdr_len + skb->csum_offset;
645         int eop = (len_left == 0);
646
647         /* Queue the main skb fragment. The fragments are no larger
648          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
649          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
650          * per fragment is queued.
651          */
652         enic_queue_wq_desc_csum_l4(wq, skb,
653                 pci_map_single(enic->pdev, skb->data,
654                         head_len, PCI_DMA_TODEVICE),
655                 head_len,
656                 csum_offset,
657                 hdr_len,
658                 vlan_tag_insert, vlan_tag,
659                 eop, loopback);
660
661         if (!eop)
662                 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
663 }
664
665 static inline void enic_queue_wq_skb_tso(struct enic *enic,
666         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
667         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
668 {
669         unsigned int frag_len_left = skb_headlen(skb);
670         unsigned int len_left = skb->len - frag_len_left;
671         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
672         int eop = (len_left == 0);
673         unsigned int len;
674         dma_addr_t dma_addr;
675         unsigned int offset = 0;
676         skb_frag_t *frag;
677
678         /* Preload TCP csum field with IP pseudo hdr calculated
679          * with IP length set to zero.  HW will later add in length
680          * to each TCP segment resulting from the TSO.
681          */
682
683         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
684                 ip_hdr(skb)->check = 0;
685                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
686                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
687         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
688                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
689                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
690         }
691
692         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
693          * for the main skb fragment
694          */
695         while (frag_len_left) {
696                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
697                 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
698                                 len, PCI_DMA_TODEVICE);
699                 enic_queue_wq_desc_tso(wq, skb,
700                         dma_addr,
701                         len,
702                         mss, hdr_len,
703                         vlan_tag_insert, vlan_tag,
704                         eop && (len == frag_len_left), loopback);
705                 frag_len_left -= len;
706                 offset += len;
707         }
708
709         if (eop)
710                 return;
711
712         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
713          * for additional data fragments
714          */
715         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
716                 len_left -= frag->size;
717                 frag_len_left = frag->size;
718                 offset = frag->page_offset;
719
720                 while (frag_len_left) {
721                         len = min(frag_len_left,
722                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
723                         dma_addr = pci_map_page(enic->pdev, frag->page,
724                                 offset, len,
725                                 PCI_DMA_TODEVICE);
726                         enic_queue_wq_desc_cont(wq, skb,
727                                 dma_addr,
728                                 len,
729                                 (len_left == 0) &&
730                                 (len == frag_len_left),         /* EOP? */
731                                 loopback);
732                         frag_len_left -= len;
733                         offset += len;
734                 }
735         }
736 }
737
738 static inline void enic_queue_wq_skb(struct enic *enic,
739         struct vnic_wq *wq, struct sk_buff *skb)
740 {
741         unsigned int mss = skb_shinfo(skb)->gso_size;
742         unsigned int vlan_tag = 0;
743         int vlan_tag_insert = 0;
744         int loopback = 0;
745
746         if (enic->vlan_group && vlan_tx_tag_present(skb)) {
747                 /* VLAN tag from trunking driver */
748                 vlan_tag_insert = 1;
749                 vlan_tag = vlan_tx_tag_get(skb);
750         } else if (enic->loop_enable) {
751                 vlan_tag = enic->loop_tag;
752                 loopback = 1;
753         }
754
755         if (mss)
756                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
757                         vlan_tag_insert, vlan_tag, loopback);
758         else if (skb->ip_summed == CHECKSUM_PARTIAL)
759                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
760                         vlan_tag_insert, vlan_tag, loopback);
761         else
762                 enic_queue_wq_skb_vlan(enic, wq, skb,
763                         vlan_tag_insert, vlan_tag, loopback);
764 }
765
766 /* netif_tx_lock held, process context with BHs disabled, or BH */
767 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
768         struct net_device *netdev)
769 {
770         struct enic *enic = netdev_priv(netdev);
771         struct vnic_wq *wq = &enic->wq[0];
772         unsigned long flags;
773
774         if (skb->len <= 0) {
775                 dev_kfree_skb(skb);
776                 return NETDEV_TX_OK;
777         }
778
779         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
780          * which is very likely.  In the off chance it's going to take
781          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
782          */
783
784         if (skb_shinfo(skb)->gso_size == 0 &&
785             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
786             skb_linearize(skb)) {
787                 dev_kfree_skb(skb);
788                 return NETDEV_TX_OK;
789         }
790
791         spin_lock_irqsave(&enic->wq_lock[0], flags);
792
793         if (vnic_wq_desc_avail(wq) <
794             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
795                 netif_stop_queue(netdev);
796                 /* This is a hard error, log it */
797                 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
798                 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
799                 return NETDEV_TX_BUSY;
800         }
801
802         enic_queue_wq_skb(enic, wq, skb);
803
804         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
805                 netif_stop_queue(netdev);
806
807         spin_unlock_irqrestore(&enic->wq_lock[0], flags);
808
809         return NETDEV_TX_OK;
810 }
811
812 /* dev_base_lock rwlock held, nominally process context */
813 static struct net_device_stats *enic_get_stats(struct net_device *netdev)
814 {
815         struct enic *enic = netdev_priv(netdev);
816         struct net_device_stats *net_stats = &netdev->stats;
817         struct vnic_stats *stats;
818
819         enic_dev_stats_dump(enic, &stats);
820
821         net_stats->tx_packets = stats->tx.tx_frames_ok;
822         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
823         net_stats->tx_errors = stats->tx.tx_errors;
824         net_stats->tx_dropped = stats->tx.tx_drops;
825
826         net_stats->rx_packets = stats->rx.rx_frames_ok;
827         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
828         net_stats->rx_errors = stats->rx.rx_errors;
829         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
830         net_stats->rx_over_errors = enic->rq_truncated_pkts;
831         net_stats->rx_crc_errors = enic->rq_bad_fcs;
832         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
833
834         return net_stats;
835 }
836
837 static void enic_reset_multicast_list(struct enic *enic)
838 {
839         enic->mc_count = 0;
840         enic->flags = 0;
841 }
842
843 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
844 {
845         struct enic *enic = netdev_priv(netdev);
846
847         if (enic_is_dynamic(enic)) {
848                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
849                         return -EADDRNOTAVAIL;
850         } else {
851                 if (!is_valid_ether_addr(addr))
852                         return -EADDRNOTAVAIL;
853         }
854
855         memcpy(netdev->dev_addr, addr, netdev->addr_len);
856
857         return 0;
858 }
859
860 static int enic_dev_add_station_addr(struct enic *enic)
861 {
862         int err = 0;
863
864         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
865                 spin_lock(&enic->devcmd_lock);
866                 err = vnic_dev_add_addr(enic->vdev, enic->netdev->dev_addr);
867                 spin_unlock(&enic->devcmd_lock);
868         }
869
870         return err;
871 }
872
873 static int enic_dev_del_station_addr(struct enic *enic)
874 {
875         int err = 0;
876
877         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
878                 spin_lock(&enic->devcmd_lock);
879                 err = vnic_dev_del_addr(enic->vdev, enic->netdev->dev_addr);
880                 spin_unlock(&enic->devcmd_lock);
881         }
882
883         return err;
884 }
885
886 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
887 {
888         struct enic *enic = netdev_priv(netdev);
889         struct sockaddr *saddr = p;
890         char *addr = saddr->sa_data;
891         int err;
892
893         if (netif_running(enic->netdev)) {
894                 err = enic_dev_del_station_addr(enic);
895                 if (err)
896                         return err;
897         }
898
899         err = enic_set_mac_addr(netdev, addr);
900         if (err)
901                 return err;
902
903         if (netif_running(enic->netdev)) {
904                 err = enic_dev_add_station_addr(enic);
905                 if (err)
906                         return err;
907         }
908
909         return err;
910 }
911
912 static int enic_set_mac_address(struct net_device *netdev, void *p)
913 {
914         struct sockaddr *saddr = p;
915
916         return enic_set_mac_addr(netdev, (char *)saddr->sa_data);
917 }
918
919 static int enic_dev_packet_filter(struct enic *enic, int directed,
920         int multicast, int broadcast, int promisc, int allmulti)
921 {
922         int err;
923
924         spin_lock(&enic->devcmd_lock);
925         err = vnic_dev_packet_filter(enic->vdev, directed,
926                 multicast, broadcast, promisc, allmulti);
927         spin_unlock(&enic->devcmd_lock);
928
929         return err;
930 }
931
932 static int enic_dev_add_multicast_addr(struct enic *enic, u8 *addr)
933 {
934         int err;
935
936         spin_lock(&enic->devcmd_lock);
937         err = vnic_dev_add_addr(enic->vdev, addr);
938         spin_unlock(&enic->devcmd_lock);
939
940         return err;
941 }
942
943 static int enic_dev_del_multicast_addr(struct enic *enic, u8 *addr)
944 {
945         int err;
946
947         spin_lock(&enic->devcmd_lock);
948         err = vnic_dev_del_addr(enic->vdev, addr);
949         spin_unlock(&enic->devcmd_lock);
950
951         return err;
952 }
953
954 /* netif_tx_lock held, BHs disabled */
955 static void enic_set_multicast_list(struct net_device *netdev)
956 {
957         struct enic *enic = netdev_priv(netdev);
958         struct netdev_hw_addr *ha;
959         int directed = 1;
960         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
961         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
962         int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
963         unsigned int mc_count = netdev_mc_count(netdev);
964         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
965                 mc_count > ENIC_MULTICAST_PERFECT_FILTERS;
966         unsigned int flags = netdev->flags | (allmulti ? IFF_ALLMULTI : 0);
967         u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
968         unsigned int i, j;
969
970         if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
971                 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
972
973         if (enic->flags != flags) {
974                 enic->flags = flags;
975                 enic_dev_packet_filter(enic, directed,
976                         multicast, broadcast, promisc, allmulti);
977         }
978
979         /* Is there an easier way?  Trying to minimize to
980          * calls to add/del multicast addrs.  We keep the
981          * addrs from the last call in enic->mc_addr and
982          * look for changes to add/del.
983          */
984
985         i = 0;
986         netdev_for_each_mc_addr(ha, netdev) {
987                 if (i == mc_count)
988                         break;
989                 memcpy(mc_addr[i++], ha->addr, ETH_ALEN);
990         }
991
992         for (i = 0; i < enic->mc_count; i++) {
993                 for (j = 0; j < mc_count; j++)
994                         if (compare_ether_addr(enic->mc_addr[i],
995                                 mc_addr[j]) == 0)
996                                 break;
997                 if (j == mc_count)
998                         enic_dev_del_multicast_addr(enic, enic->mc_addr[i]);
999         }
1000
1001         for (i = 0; i < mc_count; i++) {
1002                 for (j = 0; j < enic->mc_count; j++)
1003                         if (compare_ether_addr(mc_addr[i],
1004                                 enic->mc_addr[j]) == 0)
1005                                 break;
1006                 if (j == enic->mc_count)
1007                         enic_dev_add_multicast_addr(enic, mc_addr[i]);
1008         }
1009
1010         /* Save the list to compare against next time
1011          */
1012
1013         for (i = 0; i < mc_count; i++)
1014                 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
1015
1016         enic->mc_count = mc_count;
1017 }
1018
1019 /* rtnl lock is held */
1020 static void enic_vlan_rx_register(struct net_device *netdev,
1021         struct vlan_group *vlan_group)
1022 {
1023         struct enic *enic = netdev_priv(netdev);
1024         enic->vlan_group = vlan_group;
1025 }
1026
1027 /* rtnl lock is held */
1028 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1029 {
1030         struct enic *enic = netdev_priv(netdev);
1031
1032         spin_lock(&enic->devcmd_lock);
1033         enic_add_vlan(enic, vid);
1034         spin_unlock(&enic->devcmd_lock);
1035 }
1036
1037 /* rtnl lock is held */
1038 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1039 {
1040         struct enic *enic = netdev_priv(netdev);
1041
1042         spin_lock(&enic->devcmd_lock);
1043         enic_del_vlan(enic, vid);
1044         spin_unlock(&enic->devcmd_lock);
1045 }
1046
1047 /* netif_tx_lock held, BHs disabled */
1048 static void enic_tx_timeout(struct net_device *netdev)
1049 {
1050         struct enic *enic = netdev_priv(netdev);
1051         schedule_work(&enic->reset);
1052 }
1053
1054 static int enic_vnic_dev_deinit(struct enic *enic)
1055 {
1056         int err;
1057
1058         spin_lock(&enic->devcmd_lock);
1059         err = vnic_dev_deinit(enic->vdev);
1060         spin_unlock(&enic->devcmd_lock);
1061
1062         return err;
1063 }
1064
1065 static int enic_dev_init_prov(struct enic *enic, struct vic_provinfo *vp)
1066 {
1067         int err;
1068
1069         spin_lock(&enic->devcmd_lock);
1070         err = vnic_dev_init_prov(enic->vdev,
1071                 (u8 *)vp, vic_provinfo_size(vp));
1072         spin_unlock(&enic->devcmd_lock);
1073
1074         return err;
1075 }
1076
1077 static int enic_dev_init_done(struct enic *enic, int *done, int *error)
1078 {
1079         int err;
1080
1081         spin_lock(&enic->devcmd_lock);
1082         err = vnic_dev_init_done(enic->vdev, done, error);
1083         spin_unlock(&enic->devcmd_lock);
1084
1085         return err;
1086 }
1087
1088 static int enic_set_port_profile(struct enic *enic, u8 *mac)
1089 {
1090         struct vic_provinfo *vp;
1091         u8 oui[3] = VIC_PROVINFO_CISCO_OUI;
1092         char uuid_str[38];
1093         int err;
1094
1095         err = enic_vnic_dev_deinit(enic);
1096         if (err)
1097                 return err;
1098
1099         switch (enic->pp.request) {
1100
1101         case PORT_REQUEST_ASSOCIATE:
1102
1103                 if (!(enic->pp.set & ENIC_SET_NAME) || !strlen(enic->pp.name))
1104                         return -EINVAL;
1105
1106                 if (!is_valid_ether_addr(mac))
1107                         return -EADDRNOTAVAIL;
1108
1109                 vp = vic_provinfo_alloc(GFP_KERNEL, oui,
1110                         VIC_PROVINFO_LINUX_TYPE);
1111                 if (!vp)
1112                         return -ENOMEM;
1113
1114                 vic_provinfo_add_tlv(vp,
1115                         VIC_LINUX_PROV_TLV_PORT_PROFILE_NAME_STR,
1116                         strlen(enic->pp.name) + 1, enic->pp.name);
1117
1118                 vic_provinfo_add_tlv(vp,
1119                         VIC_LINUX_PROV_TLV_CLIENT_MAC_ADDR,
1120                         ETH_ALEN, mac);
1121
1122                 if (enic->pp.set & ENIC_SET_INSTANCE) {
1123                         sprintf(uuid_str, "%pUB", enic->pp.instance_uuid);
1124                         vic_provinfo_add_tlv(vp,
1125                                 VIC_LINUX_PROV_TLV_CLIENT_UUID_STR,
1126                                 sizeof(uuid_str), uuid_str);
1127                 }
1128
1129                 if (enic->pp.set & ENIC_SET_HOST) {
1130                         sprintf(uuid_str, "%pUB", enic->pp.host_uuid);
1131                         vic_provinfo_add_tlv(vp,
1132                                 VIC_LINUX_PROV_TLV_HOST_UUID_STR,
1133                                 sizeof(uuid_str), uuid_str);
1134                 }
1135
1136                 err = enic_dev_init_prov(enic, vp);
1137                 vic_provinfo_free(vp);
1138                 if (err)
1139                         return err;
1140                 break;
1141
1142         case PORT_REQUEST_DISASSOCIATE:
1143                 break;
1144
1145         default:
1146                 return -EINVAL;
1147         }
1148
1149         enic->pp.set |= ENIC_SET_APPLIED;
1150         return 0;
1151 }
1152
1153 static int enic_set_vf_port(struct net_device *netdev, int vf,
1154         struct nlattr *port[])
1155 {
1156         struct enic *enic = netdev_priv(netdev);
1157
1158         memset(&enic->pp, 0, sizeof(enic->pp));
1159
1160         if (port[IFLA_PORT_REQUEST]) {
1161                 enic->pp.set |= ENIC_SET_REQUEST;
1162                 enic->pp.request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1163         }
1164
1165         if (port[IFLA_PORT_PROFILE]) {
1166                 enic->pp.set |= ENIC_SET_NAME;
1167                 memcpy(enic->pp.name, nla_data(port[IFLA_PORT_PROFILE]),
1168                         PORT_PROFILE_MAX);
1169         }
1170
1171         if (port[IFLA_PORT_INSTANCE_UUID]) {
1172                 enic->pp.set |= ENIC_SET_INSTANCE;
1173                 memcpy(enic->pp.instance_uuid,
1174                         nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
1175         }
1176
1177         if (port[IFLA_PORT_HOST_UUID]) {
1178                 enic->pp.set |= ENIC_SET_HOST;
1179                 memcpy(enic->pp.host_uuid,
1180                         nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
1181         }
1182
1183         /* don't support VFs, yet */
1184         if (vf != PORT_SELF_VF)
1185                 return -EOPNOTSUPP;
1186
1187         if (!(enic->pp.set & ENIC_SET_REQUEST))
1188                 return -EOPNOTSUPP;
1189
1190         if (enic->pp.request == PORT_REQUEST_ASSOCIATE) {
1191
1192                 /* If the interface mac addr hasn't been assigned,
1193                  * assign a random mac addr before setting port-
1194                  * profile.
1195                  */
1196
1197                 if (is_zero_ether_addr(netdev->dev_addr))
1198                         random_ether_addr(netdev->dev_addr);
1199         }
1200
1201         return enic_set_port_profile(enic, netdev->dev_addr);
1202 }
1203
1204 static int enic_get_vf_port(struct net_device *netdev, int vf,
1205         struct sk_buff *skb)
1206 {
1207         struct enic *enic = netdev_priv(netdev);
1208         int err, error, done;
1209         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1210
1211         if (!(enic->pp.set & ENIC_SET_APPLIED))
1212                 return -ENODATA;
1213
1214         err = enic_dev_init_done(enic, &done, &error);
1215         if (err)
1216                 error = err;
1217
1218         switch (error) {
1219         case ERR_SUCCESS:
1220                 if (!done)
1221                         response = PORT_PROFILE_RESPONSE_INPROGRESS;
1222                 break;
1223         case ERR_EINVAL:
1224                 response = PORT_PROFILE_RESPONSE_INVALID;
1225                 break;
1226         case ERR_EBADSTATE:
1227                 response = PORT_PROFILE_RESPONSE_BADSTATE;
1228                 break;
1229         case ERR_ENOMEM:
1230                 response = PORT_PROFILE_RESPONSE_INSUFFICIENT_RESOURCES;
1231                 break;
1232         default:
1233                 response = PORT_PROFILE_RESPONSE_ERROR;
1234                 break;
1235         }
1236
1237         NLA_PUT_U16(skb, IFLA_PORT_REQUEST, enic->pp.request);
1238         NLA_PUT_U16(skb, IFLA_PORT_RESPONSE, response);
1239         if (enic->pp.set & ENIC_SET_NAME)
1240                 NLA_PUT(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX,
1241                         enic->pp.name);
1242         if (enic->pp.set & ENIC_SET_INSTANCE)
1243                 NLA_PUT(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1244                         enic->pp.instance_uuid);
1245         if (enic->pp.set & ENIC_SET_HOST)
1246                 NLA_PUT(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX,
1247                         enic->pp.host_uuid);
1248
1249         return 0;
1250
1251 nla_put_failure:
1252         return -EMSGSIZE;
1253 }
1254
1255 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1256 {
1257         struct enic *enic = vnic_dev_priv(rq->vdev);
1258
1259         if (!buf->os_buf)
1260                 return;
1261
1262         pci_unmap_single(enic->pdev, buf->dma_addr,
1263                 buf->len, PCI_DMA_FROMDEVICE);
1264         dev_kfree_skb_any(buf->os_buf);
1265 }
1266
1267 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1268 {
1269         struct enic *enic = vnic_dev_priv(rq->vdev);
1270         struct net_device *netdev = enic->netdev;
1271         struct sk_buff *skb;
1272         unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
1273         unsigned int os_buf_index = 0;
1274         dma_addr_t dma_addr;
1275
1276         skb = netdev_alloc_skb_ip_align(netdev, len);
1277         if (!skb)
1278                 return -ENOMEM;
1279
1280         dma_addr = pci_map_single(enic->pdev, skb->data,
1281                 len, PCI_DMA_FROMDEVICE);
1282
1283         enic_queue_rq_desc(rq, skb, os_buf_index,
1284                 dma_addr, len);
1285
1286         return 0;
1287 }
1288
1289 static int enic_rq_alloc_buf_a1(struct vnic_rq *rq)
1290 {
1291         struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
1292
1293         if (vnic_rq_posting_soon(rq)) {
1294
1295                 /* SW workaround for A0 HW erratum: if we're just about
1296                  * to write posted_index, insert a dummy desc
1297                  * of type resvd
1298                  */
1299
1300                 rq_enet_desc_enc(desc, 0, RQ_ENET_TYPE_RESV2, 0);
1301                 vnic_rq_post(rq, 0, 0, 0, 0);
1302         } else {
1303                 return enic_rq_alloc_buf(rq);
1304         }
1305
1306         return 0;
1307 }
1308
1309 static int enic_dev_hw_version(struct enic *enic,
1310         enum vnic_dev_hw_version *hw_ver)
1311 {
1312         int err;
1313
1314         spin_lock(&enic->devcmd_lock);
1315         err = vnic_dev_hw_version(enic->vdev, hw_ver);
1316         spin_unlock(&enic->devcmd_lock);
1317
1318         return err;
1319 }
1320
1321 static int enic_set_rq_alloc_buf(struct enic *enic)
1322 {
1323         enum vnic_dev_hw_version hw_ver;
1324         int err;
1325
1326         err = enic_dev_hw_version(enic, &hw_ver);
1327         if (err)
1328                 return err;
1329
1330         switch (hw_ver) {
1331         case VNIC_DEV_HW_VER_A1:
1332                 enic->rq_alloc_buf = enic_rq_alloc_buf_a1;
1333                 break;
1334         case VNIC_DEV_HW_VER_A2:
1335         case VNIC_DEV_HW_VER_UNKNOWN:
1336                 enic->rq_alloc_buf = enic_rq_alloc_buf;
1337                 break;
1338         default:
1339                 return -ENODEV;
1340         }
1341
1342         return 0;
1343 }
1344
1345 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1346         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1347         int skipped, void *opaque)
1348 {
1349         struct enic *enic = vnic_dev_priv(rq->vdev);
1350         struct net_device *netdev = enic->netdev;
1351         struct sk_buff *skb;
1352
1353         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1354         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1355         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1356         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1357         u8 packet_error;
1358         u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1359         u32 rss_hash;
1360
1361         if (skipped)
1362                 return;
1363
1364         skb = buf->os_buf;
1365         prefetch(skb->data - NET_IP_ALIGN);
1366         pci_unmap_single(enic->pdev, buf->dma_addr,
1367                 buf->len, PCI_DMA_FROMDEVICE);
1368
1369         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1370                 &type, &color, &q_number, &completed_index,
1371                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1372                 &csum_not_calc, &rss_hash, &bytes_written,
1373                 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1374                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1375                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1376                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1377                 &fcs_ok);
1378
1379         if (packet_error) {
1380
1381                 if (!fcs_ok) {
1382                         if (bytes_written > 0)
1383                                 enic->rq_bad_fcs++;
1384                         else if (bytes_written == 0)
1385                                 enic->rq_truncated_pkts++;
1386                 }
1387
1388                 dev_kfree_skb_any(skb);
1389
1390                 return;
1391         }
1392
1393         if (eop && bytes_written > 0) {
1394
1395                 /* Good receive
1396                  */
1397
1398                 skb_put(skb, bytes_written);
1399                 skb->protocol = eth_type_trans(skb, netdev);
1400
1401                 if (enic->csum_rx_enabled && !csum_not_calc) {
1402                         skb->csum = htons(checksum);
1403                         skb->ip_summed = CHECKSUM_COMPLETE;
1404                 }
1405
1406                 skb->dev = netdev;
1407
1408                 if (enic->vlan_group && vlan_stripped &&
1409                         (vlan_tci & CQ_ENET_RQ_DESC_VLAN_TCI_VLAN_MASK)) {
1410
1411                         if (netdev->features & NETIF_F_GRO)
1412                                 vlan_gro_receive(&enic->napi, enic->vlan_group,
1413                                         vlan_tci, skb);
1414                         else
1415                                 vlan_hwaccel_receive_skb(skb,
1416                                         enic->vlan_group, vlan_tci);
1417
1418                 } else {
1419
1420                         if (netdev->features & NETIF_F_GRO)
1421                                 napi_gro_receive(&enic->napi, skb);
1422                         else
1423                                 netif_receive_skb(skb);
1424
1425                 }
1426
1427         } else {
1428
1429                 /* Buffer overflow
1430                  */
1431
1432                 dev_kfree_skb_any(skb);
1433         }
1434 }
1435
1436 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1437         u8 type, u16 q_number, u16 completed_index, void *opaque)
1438 {
1439         struct enic *enic = vnic_dev_priv(vdev);
1440
1441         vnic_rq_service(&enic->rq[q_number], cq_desc,
1442                 completed_index, VNIC_RQ_RETURN_DESC,
1443                 enic_rq_indicate_buf, opaque);
1444
1445         return 0;
1446 }
1447
1448 static int enic_poll(struct napi_struct *napi, int budget)
1449 {
1450         struct enic *enic = container_of(napi, struct enic, napi);
1451         unsigned int rq_work_to_do = budget;
1452         unsigned int wq_work_to_do = -1; /* no limit */
1453         unsigned int  work_done, rq_work_done, wq_work_done;
1454         int err;
1455
1456         /* Service RQ (first) and WQ
1457          */
1458
1459         rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1460                 rq_work_to_do, enic_rq_service, NULL);
1461
1462         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1463                 wq_work_to_do, enic_wq_service, NULL);
1464
1465         /* Accumulate intr event credits for this polling
1466          * cycle.  An intr event is the completion of a
1467          * a WQ or RQ packet.
1468          */
1469
1470         work_done = rq_work_done + wq_work_done;
1471
1472         if (work_done > 0)
1473                 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1474                         work_done,
1475                         0 /* don't unmask intr */,
1476                         0 /* don't reset intr timer */);
1477
1478         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1479
1480         /* Buffer allocation failed. Stay in polling
1481          * mode so we can try to fill the ring again.
1482          */
1483
1484         if (err)
1485                 rq_work_done = rq_work_to_do;
1486
1487         if (rq_work_done < rq_work_to_do) {
1488
1489                 /* Some work done, but not enough to stay in polling,
1490                  * exit polling
1491                  */
1492
1493                 napi_complete(napi);
1494                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
1495         }
1496
1497         return rq_work_done;
1498 }
1499
1500 static int enic_poll_msix(struct napi_struct *napi, int budget)
1501 {
1502         struct enic *enic = container_of(napi, struct enic, napi);
1503         unsigned int work_to_do = budget;
1504         unsigned int work_done;
1505         int err;
1506
1507         /* Service RQ
1508          */
1509
1510         work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1511                 work_to_do, enic_rq_service, NULL);
1512
1513         /* Return intr event credits for this polling
1514          * cycle.  An intr event is the completion of a
1515          * RQ packet.
1516          */
1517
1518         if (work_done > 0)
1519                 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1520                         work_done,
1521                         0 /* don't unmask intr */,
1522                         0 /* don't reset intr timer */);
1523
1524         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1525
1526         /* Buffer allocation failed. Stay in polling mode
1527          * so we can try to fill the ring again.
1528          */
1529
1530         if (err)
1531                 work_done = work_to_do;
1532
1533         if (work_done < work_to_do) {
1534
1535                 /* Some work done, but not enough to stay in polling,
1536                  * exit polling
1537                  */
1538
1539                 napi_complete(napi);
1540                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1541         }
1542
1543         return work_done;
1544 }
1545
1546 static void enic_notify_timer(unsigned long data)
1547 {
1548         struct enic *enic = (struct enic *)data;
1549
1550         enic_notify_check(enic);
1551
1552         mod_timer(&enic->notify_timer,
1553                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1554 }
1555
1556 static void enic_free_intr(struct enic *enic)
1557 {
1558         struct net_device *netdev = enic->netdev;
1559         unsigned int i;
1560
1561         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1562         case VNIC_DEV_INTR_MODE_INTX:
1563                 free_irq(enic->pdev->irq, netdev);
1564                 break;
1565         case VNIC_DEV_INTR_MODE_MSI:
1566                 free_irq(enic->pdev->irq, enic);
1567                 break;
1568         case VNIC_DEV_INTR_MODE_MSIX:
1569                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1570                         if (enic->msix[i].requested)
1571                                 free_irq(enic->msix_entry[i].vector,
1572                                         enic->msix[i].devid);
1573                 break;
1574         default:
1575                 break;
1576         }
1577 }
1578
1579 static int enic_request_intr(struct enic *enic)
1580 {
1581         struct net_device *netdev = enic->netdev;
1582         unsigned int i;
1583         int err = 0;
1584
1585         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1586
1587         case VNIC_DEV_INTR_MODE_INTX:
1588
1589                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1590                         IRQF_SHARED, netdev->name, netdev);
1591                 break;
1592
1593         case VNIC_DEV_INTR_MODE_MSI:
1594
1595                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1596                         0, netdev->name, enic);
1597                 break;
1598
1599         case VNIC_DEV_INTR_MODE_MSIX:
1600
1601                 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1602                         "%.11s-rx-0", netdev->name);
1603                 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1604                 enic->msix[ENIC_MSIX_RQ].devid = enic;
1605
1606                 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1607                         "%.11s-tx-0", netdev->name);
1608                 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1609                 enic->msix[ENIC_MSIX_WQ].devid = enic;
1610
1611                 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1612                         "%.11s-err", netdev->name);
1613                 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1614                 enic->msix[ENIC_MSIX_ERR].devid = enic;
1615
1616                 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1617                         "%.11s-notify", netdev->name);
1618                 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1619                 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1620
1621                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1622                         err = request_irq(enic->msix_entry[i].vector,
1623                                 enic->msix[i].isr, 0,
1624                                 enic->msix[i].devname,
1625                                 enic->msix[i].devid);
1626                         if (err) {
1627                                 enic_free_intr(enic);
1628                                 break;
1629                         }
1630                         enic->msix[i].requested = 1;
1631                 }
1632
1633                 break;
1634
1635         default:
1636                 break;
1637         }
1638
1639         return err;
1640 }
1641
1642 static void enic_synchronize_irqs(struct enic *enic)
1643 {
1644         unsigned int i;
1645
1646         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1647         case VNIC_DEV_INTR_MODE_INTX:
1648         case VNIC_DEV_INTR_MODE_MSI:
1649                 synchronize_irq(enic->pdev->irq);
1650                 break;
1651         case VNIC_DEV_INTR_MODE_MSIX:
1652                 for (i = 0; i < enic->intr_count; i++)
1653                         synchronize_irq(enic->msix_entry[i].vector);
1654                 break;
1655         default:
1656                 break;
1657         }
1658 }
1659
1660 static int enic_dev_notify_set(struct enic *enic)
1661 {
1662         int err;
1663
1664         spin_lock(&enic->devcmd_lock);
1665         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1666         case VNIC_DEV_INTR_MODE_INTX:
1667                 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1668                 break;
1669         case VNIC_DEV_INTR_MODE_MSIX:
1670                 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1671                 break;
1672         default:
1673                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1674                 break;
1675         }
1676         spin_unlock(&enic->devcmd_lock);
1677
1678         return err;
1679 }
1680
1681 static int enic_dev_notify_unset(struct enic *enic)
1682 {
1683         int err;
1684
1685         spin_lock(&enic->devcmd_lock);
1686         err = vnic_dev_notify_unset(enic->vdev);
1687         spin_unlock(&enic->devcmd_lock);
1688
1689         return err;
1690 }
1691
1692 static int enic_dev_enable(struct enic *enic)
1693 {
1694         int err;
1695
1696         spin_lock(&enic->devcmd_lock);
1697         err = vnic_dev_enable(enic->vdev);
1698         spin_unlock(&enic->devcmd_lock);
1699
1700         return err;
1701 }
1702
1703 static int enic_dev_disable(struct enic *enic)
1704 {
1705         int err;
1706
1707         spin_lock(&enic->devcmd_lock);
1708         err = vnic_dev_disable(enic->vdev);
1709         spin_unlock(&enic->devcmd_lock);
1710
1711         return err;
1712 }
1713
1714 static void enic_notify_timer_start(struct enic *enic)
1715 {
1716         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1717         case VNIC_DEV_INTR_MODE_MSI:
1718                 mod_timer(&enic->notify_timer, jiffies);
1719                 break;
1720         default:
1721                 /* Using intr for notification for INTx/MSI-X */
1722                 break;
1723         };
1724 }
1725
1726 /* rtnl lock is held, process context */
1727 static int enic_open(struct net_device *netdev)
1728 {
1729         struct enic *enic = netdev_priv(netdev);
1730         unsigned int i;
1731         int err;
1732
1733         err = enic_request_intr(enic);
1734         if (err) {
1735                 netdev_err(netdev, "Unable to request irq.\n");
1736                 return err;
1737         }
1738
1739         err = enic_dev_notify_set(enic);
1740         if (err) {
1741                 netdev_err(netdev,
1742                         "Failed to alloc notify buffer, aborting.\n");
1743                 goto err_out_free_intr;
1744         }
1745
1746         for (i = 0; i < enic->rq_count; i++) {
1747                 vnic_rq_fill(&enic->rq[i], enic->rq_alloc_buf);
1748                 /* Need at least one buffer on ring to get going */
1749                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1750                         netdev_err(netdev, "Unable to alloc receive buffers\n");
1751                         err = -ENOMEM;
1752                         goto err_out_notify_unset;
1753                 }
1754         }
1755
1756         for (i = 0; i < enic->wq_count; i++)
1757                 vnic_wq_enable(&enic->wq[i]);
1758         for (i = 0; i < enic->rq_count; i++)
1759                 vnic_rq_enable(&enic->rq[i]);
1760
1761         enic_dev_add_station_addr(enic);
1762         enic_set_multicast_list(netdev);
1763
1764         netif_wake_queue(netdev);
1765         napi_enable(&enic->napi);
1766         enic_dev_enable(enic);
1767
1768         for (i = 0; i < enic->intr_count; i++)
1769                 vnic_intr_unmask(&enic->intr[i]);
1770
1771         enic_notify_timer_start(enic);
1772
1773         return 0;
1774
1775 err_out_notify_unset:
1776         enic_dev_notify_unset(enic);
1777 err_out_free_intr:
1778         enic_free_intr(enic);
1779
1780         return err;
1781 }
1782
1783 /* rtnl lock is held, process context */
1784 static int enic_stop(struct net_device *netdev)
1785 {
1786         struct enic *enic = netdev_priv(netdev);
1787         unsigned int i;
1788         int err;
1789
1790         for (i = 0; i < enic->intr_count; i++) {
1791                 vnic_intr_mask(&enic->intr[i]);
1792                 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1793         }
1794
1795         enic_synchronize_irqs(enic);
1796
1797         del_timer_sync(&enic->notify_timer);
1798
1799         enic_dev_disable(enic);
1800         napi_disable(&enic->napi);
1801         netif_carrier_off(netdev);
1802         netif_tx_disable(netdev);
1803         enic_dev_del_station_addr(enic);
1804
1805         for (i = 0; i < enic->wq_count; i++) {
1806                 err = vnic_wq_disable(&enic->wq[i]);
1807                 if (err)
1808                         return err;
1809         }
1810         for (i = 0; i < enic->rq_count; i++) {
1811                 err = vnic_rq_disable(&enic->rq[i]);
1812                 if (err)
1813                         return err;
1814         }
1815
1816         enic_dev_notify_unset(enic);
1817         enic_free_intr(enic);
1818
1819         for (i = 0; i < enic->wq_count; i++)
1820                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1821         for (i = 0; i < enic->rq_count; i++)
1822                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1823         for (i = 0; i < enic->cq_count; i++)
1824                 vnic_cq_clean(&enic->cq[i]);
1825         for (i = 0; i < enic->intr_count; i++)
1826                 vnic_intr_clean(&enic->intr[i]);
1827
1828         return 0;
1829 }
1830
1831 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1832 {
1833         struct enic *enic = netdev_priv(netdev);
1834         int running = netif_running(netdev);
1835
1836         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1837                 return -EINVAL;
1838
1839         if (running)
1840                 enic_stop(netdev);
1841
1842         netdev->mtu = new_mtu;
1843
1844         if (netdev->mtu > enic->port_mtu)
1845                 netdev_warn(netdev,
1846                         "interface MTU (%d) set higher than port MTU (%d)\n",
1847                         netdev->mtu, enic->port_mtu);
1848
1849         if (running)
1850                 enic_open(netdev);
1851
1852         return 0;
1853 }
1854
1855 #ifdef CONFIG_NET_POLL_CONTROLLER
1856 static void enic_poll_controller(struct net_device *netdev)
1857 {
1858         struct enic *enic = netdev_priv(netdev);
1859         struct vnic_dev *vdev = enic->vdev;
1860
1861         switch (vnic_dev_get_intr_mode(vdev)) {
1862         case VNIC_DEV_INTR_MODE_MSIX:
1863                 enic_isr_msix_rq(enic->pdev->irq, enic);
1864                 enic_isr_msix_wq(enic->pdev->irq, enic);
1865                 break;
1866         case VNIC_DEV_INTR_MODE_MSI:
1867                 enic_isr_msi(enic->pdev->irq, enic);
1868                 break;
1869         case VNIC_DEV_INTR_MODE_INTX:
1870                 enic_isr_legacy(enic->pdev->irq, netdev);
1871                 break;
1872         default:
1873                 break;
1874         }
1875 }
1876 #endif
1877
1878 static int enic_dev_wait(struct vnic_dev *vdev,
1879         int (*start)(struct vnic_dev *, int),
1880         int (*finished)(struct vnic_dev *, int *),
1881         int arg)
1882 {
1883         unsigned long time;
1884         int done;
1885         int err;
1886
1887         BUG_ON(in_interrupt());
1888
1889         err = start(vdev, arg);
1890         if (err)
1891                 return err;
1892
1893         /* Wait for func to complete...2 seconds max
1894          */
1895
1896         time = jiffies + (HZ * 2);
1897         do {
1898
1899                 err = finished(vdev, &done);
1900                 if (err)
1901                         return err;
1902
1903                 if (done)
1904                         return 0;
1905
1906                 schedule_timeout_uninterruptible(HZ / 10);
1907
1908         } while (time_after(time, jiffies));
1909
1910         return -ETIMEDOUT;
1911 }
1912
1913 static int enic_dev_open(struct enic *enic)
1914 {
1915         int err;
1916
1917         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1918                 vnic_dev_open_done, 0);
1919         if (err)
1920                 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1921                         err);
1922
1923         return err;
1924 }
1925
1926 static int enic_dev_hang_reset(struct enic *enic)
1927 {
1928         int err;
1929
1930         err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1931                 vnic_dev_hang_reset_done, 0);
1932         if (err)
1933                 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1934                         err);
1935
1936         return err;
1937 }
1938
1939 static int enic_set_niccfg(struct enic *enic)
1940 {
1941         const u8 rss_default_cpu = 0;
1942         const u8 rss_hash_type = 0;
1943         const u8 rss_hash_bits = 0;
1944         const u8 rss_base_cpu = 0;
1945         const u8 rss_enable = 0;
1946         const u8 tso_ipid_split_en = 0;
1947         const u8 ig_vlan_strip_en = 1;
1948         int err;
1949
1950         /* Enable VLAN tag stripping.  RSS not enabled (yet).
1951          */
1952
1953         spin_lock(&enic->devcmd_lock);
1954         err = enic_set_nic_cfg(enic,
1955                 rss_default_cpu, rss_hash_type,
1956                 rss_hash_bits, rss_base_cpu,
1957                 rss_enable, tso_ipid_split_en,
1958                 ig_vlan_strip_en);
1959         spin_unlock(&enic->devcmd_lock);
1960
1961         return err;
1962 }
1963
1964 static int enic_dev_hang_notify(struct enic *enic)
1965 {
1966         int err;
1967
1968         spin_lock(&enic->devcmd_lock);
1969         err = vnic_dev_hang_notify(enic->vdev);
1970         spin_unlock(&enic->devcmd_lock);
1971
1972         return err;
1973 }
1974
1975 int enic_dev_set_ig_vlan_rewrite_mode(struct enic *enic)
1976 {
1977         int err;
1978
1979         spin_lock(&enic->devcmd_lock);
1980         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1981                 IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN);
1982         spin_unlock(&enic->devcmd_lock);
1983
1984         return err;
1985 }
1986
1987 static void enic_reset(struct work_struct *work)
1988 {
1989         struct enic *enic = container_of(work, struct enic, reset);
1990
1991         if (!netif_running(enic->netdev))
1992                 return;
1993
1994         rtnl_lock();
1995
1996         enic_dev_hang_notify(enic);
1997         enic_stop(enic->netdev);
1998         enic_dev_hang_reset(enic);
1999         enic_reset_multicast_list(enic);
2000         enic_init_vnic_resources(enic);
2001         enic_set_niccfg(enic);
2002         enic_dev_set_ig_vlan_rewrite_mode(enic);
2003         enic_open(enic->netdev);
2004
2005         rtnl_unlock();
2006 }
2007
2008 static int enic_set_intr_mode(struct enic *enic)
2009 {
2010         unsigned int n = 1;
2011         unsigned int m = 1;
2012         unsigned int i;
2013
2014         /* Set interrupt mode (INTx, MSI, MSI-X) depending
2015          * system capabilities.
2016          *
2017          * Try MSI-X first
2018          *
2019          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2020          * (the second to last INTR is used for WQ/RQ errors)
2021          * (the last INTR is used for notifications)
2022          */
2023
2024         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2025         for (i = 0; i < n + m + 2; i++)
2026                 enic->msix_entry[i].entry = i;
2027
2028         if (enic->config.intr_mode < 1 &&
2029             enic->rq_count >= n &&
2030             enic->wq_count >= m &&
2031             enic->cq_count >= n + m &&
2032             enic->intr_count >= n + m + 2 &&
2033             !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
2034
2035                 enic->rq_count = n;
2036                 enic->wq_count = m;
2037                 enic->cq_count = n + m;
2038                 enic->intr_count = n + m + 2;
2039
2040                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
2041
2042                 return 0;
2043         }
2044
2045         /* Next try MSI
2046          *
2047          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2048          */
2049
2050         if (enic->config.intr_mode < 2 &&
2051             enic->rq_count >= 1 &&
2052             enic->wq_count >= 1 &&
2053             enic->cq_count >= 2 &&
2054             enic->intr_count >= 1 &&
2055             !pci_enable_msi(enic->pdev)) {
2056
2057                 enic->rq_count = 1;
2058                 enic->wq_count = 1;
2059                 enic->cq_count = 2;
2060                 enic->intr_count = 1;
2061
2062                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2063
2064                 return 0;
2065         }
2066
2067         /* Next try INTx
2068          *
2069          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2070          * (the first INTR is used for WQ/RQ)
2071          * (the second INTR is used for WQ/RQ errors)
2072          * (the last INTR is used for notifications)
2073          */
2074
2075         if (enic->config.intr_mode < 3 &&
2076             enic->rq_count >= 1 &&
2077             enic->wq_count >= 1 &&
2078             enic->cq_count >= 2 &&
2079             enic->intr_count >= 3) {
2080
2081                 enic->rq_count = 1;
2082                 enic->wq_count = 1;
2083                 enic->cq_count = 2;
2084                 enic->intr_count = 3;
2085
2086                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2087
2088                 return 0;
2089         }
2090
2091         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2092
2093         return -EINVAL;
2094 }
2095
2096 static void enic_clear_intr_mode(struct enic *enic)
2097 {
2098         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2099         case VNIC_DEV_INTR_MODE_MSIX:
2100                 pci_disable_msix(enic->pdev);
2101                 break;
2102         case VNIC_DEV_INTR_MODE_MSI:
2103                 pci_disable_msi(enic->pdev);
2104                 break;
2105         default:
2106                 break;
2107         }
2108
2109         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2110 }
2111
2112 static const struct net_device_ops enic_netdev_dynamic_ops = {
2113         .ndo_open               = enic_open,
2114         .ndo_stop               = enic_stop,
2115         .ndo_start_xmit         = enic_hard_start_xmit,
2116         .ndo_get_stats          = enic_get_stats,
2117         .ndo_validate_addr      = eth_validate_addr,
2118         .ndo_set_multicast_list = enic_set_multicast_list,
2119         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2120         .ndo_change_mtu         = enic_change_mtu,
2121         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2122         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2123         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2124         .ndo_tx_timeout         = enic_tx_timeout,
2125         .ndo_set_vf_port        = enic_set_vf_port,
2126         .ndo_get_vf_port        = enic_get_vf_port,
2127 #ifdef CONFIG_NET_POLL_CONTROLLER
2128         .ndo_poll_controller    = enic_poll_controller,
2129 #endif
2130 };
2131
2132 static const struct net_device_ops enic_netdev_ops = {
2133         .ndo_open               = enic_open,
2134         .ndo_stop               = enic_stop,
2135         .ndo_start_xmit         = enic_hard_start_xmit,
2136         .ndo_get_stats          = enic_get_stats,
2137         .ndo_validate_addr      = eth_validate_addr,
2138         .ndo_set_mac_address    = enic_set_mac_address,
2139         .ndo_set_multicast_list = enic_set_multicast_list,
2140         .ndo_change_mtu         = enic_change_mtu,
2141         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2142         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2143         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2144         .ndo_tx_timeout         = enic_tx_timeout,
2145 #ifdef CONFIG_NET_POLL_CONTROLLER
2146         .ndo_poll_controller    = enic_poll_controller,
2147 #endif
2148 };
2149
2150 void enic_dev_deinit(struct enic *enic)
2151 {
2152         netif_napi_del(&enic->napi);
2153         enic_free_vnic_resources(enic);
2154         enic_clear_intr_mode(enic);
2155 }
2156
2157 int enic_dev_init(struct enic *enic)
2158 {
2159         struct device *dev = enic_get_dev(enic);
2160         struct net_device *netdev = enic->netdev;
2161         int err;
2162
2163         /* Get vNIC configuration
2164          */
2165
2166         err = enic_get_vnic_config(enic);
2167         if (err) {
2168                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2169                 return err;
2170         }
2171
2172         /* Get available resource counts
2173          */
2174
2175         enic_get_res_counts(enic);
2176
2177         /* Set interrupt mode based on resource counts and system
2178          * capabilities
2179          */
2180
2181         err = enic_set_intr_mode(enic);
2182         if (err) {
2183                 dev_err(dev, "Failed to set intr mode based on resource "
2184                         "counts and system capabilities, aborting\n");
2185                 return err;
2186         }
2187
2188         /* Allocate and configure vNIC resources
2189          */
2190
2191         err = enic_alloc_vnic_resources(enic);
2192         if (err) {
2193                 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2194                 goto err_out_free_vnic_resources;
2195         }
2196
2197         enic_init_vnic_resources(enic);
2198
2199         err = enic_set_rq_alloc_buf(enic);
2200         if (err) {
2201                 dev_err(dev, "Failed to set RQ buffer allocator, aborting\n");
2202                 goto err_out_free_vnic_resources;
2203         }
2204
2205         err = enic_set_niccfg(enic);
2206         if (err) {
2207                 dev_err(dev, "Failed to config nic, aborting\n");
2208                 goto err_out_free_vnic_resources;
2209         }
2210
2211         err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2212         if (err) {
2213                 netdev_err(netdev,
2214                         "Failed to set ingress vlan rewrite mode, aborting.\n");
2215                 goto err_out_free_vnic_resources;
2216         }
2217
2218         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2219         default:
2220                 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
2221                 break;
2222         case VNIC_DEV_INTR_MODE_MSIX:
2223                 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
2224                 break;
2225         }
2226
2227         return 0;
2228
2229 err_out_free_vnic_resources:
2230         enic_clear_intr_mode(enic);
2231         enic_free_vnic_resources(enic);
2232
2233         return err;
2234 }
2235
2236 static void enic_iounmap(struct enic *enic)
2237 {
2238         unsigned int i;
2239
2240         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2241                 if (enic->bar[i].vaddr)
2242                         iounmap(enic->bar[i].vaddr);
2243 }
2244
2245 static int __devinit enic_probe(struct pci_dev *pdev,
2246         const struct pci_device_id *ent)
2247 {
2248         struct device *dev = &pdev->dev;
2249         struct net_device *netdev;
2250         struct enic *enic;
2251         int using_dac = 0;
2252         unsigned int i;
2253         int err;
2254
2255         /* Allocate net device structure and initialize.  Private
2256          * instance data is initialized to zero.
2257          */
2258
2259         netdev = alloc_etherdev(sizeof(struct enic));
2260         if (!netdev) {
2261                 pr_err("Etherdev alloc failed, aborting\n");
2262                 return -ENOMEM;
2263         }
2264
2265         pci_set_drvdata(pdev, netdev);
2266
2267         SET_NETDEV_DEV(netdev, &pdev->dev);
2268
2269         enic = netdev_priv(netdev);
2270         enic->netdev = netdev;
2271         enic->pdev = pdev;
2272
2273         /* Setup PCI resources
2274          */
2275
2276         err = pci_enable_device_mem(pdev);
2277         if (err) {
2278                 dev_err(dev, "Cannot enable PCI device, aborting\n");
2279                 goto err_out_free_netdev;
2280         }
2281
2282         err = pci_request_regions(pdev, DRV_NAME);
2283         if (err) {
2284                 dev_err(dev, "Cannot request PCI regions, aborting\n");
2285                 goto err_out_disable_device;
2286         }
2287
2288         pci_set_master(pdev);
2289
2290         /* Query PCI controller on system for DMA addressing
2291          * limitation for the device.  Try 40-bit first, and
2292          * fail to 32-bit.
2293          */
2294
2295         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
2296         if (err) {
2297                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2298                 if (err) {
2299                         dev_err(dev, "No usable DMA configuration, aborting\n");
2300                         goto err_out_release_regions;
2301                 }
2302                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2303                 if (err) {
2304                         dev_err(dev, "Unable to obtain %u-bit DMA "
2305                                 "for consistent allocations, aborting\n", 32);
2306                         goto err_out_release_regions;
2307                 }
2308         } else {
2309                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
2310                 if (err) {
2311                         dev_err(dev, "Unable to obtain %u-bit DMA "
2312                                 "for consistent allocations, aborting\n", 40);
2313                         goto err_out_release_regions;
2314                 }
2315                 using_dac = 1;
2316         }
2317
2318         /* Map vNIC resources from BAR0-5
2319          */
2320
2321         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2322                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2323                         continue;
2324                 enic->bar[i].len = pci_resource_len(pdev, i);
2325                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2326                 if (!enic->bar[i].vaddr) {
2327                         dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2328                         err = -ENODEV;
2329                         goto err_out_iounmap;
2330                 }
2331                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2332         }
2333
2334         /* Register vNIC device
2335          */
2336
2337         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2338                 ARRAY_SIZE(enic->bar));
2339         if (!enic->vdev) {
2340                 dev_err(dev, "vNIC registration failed, aborting\n");
2341                 err = -ENODEV;
2342                 goto err_out_iounmap;
2343         }
2344
2345         /* Issue device open to get device in known state
2346          */
2347
2348         err = enic_dev_open(enic);
2349         if (err) {
2350                 dev_err(dev, "vNIC dev open failed, aborting\n");
2351                 goto err_out_vnic_unregister;
2352         }
2353
2354         /* Issue device init to initialize the vnic-to-switch link.
2355          * We'll start with carrier off and wait for link UP
2356          * notification later to turn on carrier.  We don't need
2357          * to wait here for the vnic-to-switch link initialization
2358          * to complete; link UP notification is the indication that
2359          * the process is complete.
2360          */
2361
2362         netif_carrier_off(netdev);
2363
2364         /* Do not call dev_init for a dynamic vnic.
2365          * For a dynamic vnic, init_prov_info will be
2366          * called later by an upper layer.
2367          */
2368
2369         if (!enic_is_dynamic(enic)) {
2370                 err = vnic_dev_init(enic->vdev, 0);
2371                 if (err) {
2372                         dev_err(dev, "vNIC dev init failed, aborting\n");
2373                         goto err_out_dev_close;
2374                 }
2375         }
2376
2377         /* Setup devcmd lock
2378          */
2379
2380         spin_lock_init(&enic->devcmd_lock);
2381
2382         err = enic_dev_init(enic);
2383         if (err) {
2384                 dev_err(dev, "Device initialization failed, aborting\n");
2385                 goto err_out_dev_close;
2386         }
2387
2388         /* Setup notification timer, HW reset task, and wq locks
2389          */
2390
2391         init_timer(&enic->notify_timer);
2392         enic->notify_timer.function = enic_notify_timer;
2393         enic->notify_timer.data = (unsigned long)enic;
2394
2395         INIT_WORK(&enic->reset, enic_reset);
2396
2397         for (i = 0; i < enic->wq_count; i++)
2398                 spin_lock_init(&enic->wq_lock[i]);
2399
2400         /* Register net device
2401          */
2402
2403         enic->port_mtu = enic->config.mtu;
2404         (void)enic_change_mtu(netdev, enic->port_mtu);
2405
2406         err = enic_set_mac_addr(netdev, enic->mac_addr);
2407         if (err) {
2408                 dev_err(dev, "Invalid MAC address, aborting\n");
2409                 goto err_out_dev_deinit;
2410         }
2411
2412         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2413         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2414
2415         if (enic_is_dynamic(enic))
2416                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2417         else
2418                 netdev->netdev_ops = &enic_netdev_ops;
2419
2420         netdev->watchdog_timeo = 2 * HZ;
2421         netdev->ethtool_ops = &enic_ethtool_ops;
2422
2423         netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
2424         if (ENIC_SETTING(enic, LOOP)) {
2425                 netdev->features &= ~NETIF_F_HW_VLAN_TX;
2426                 enic->loop_enable = 1;
2427                 enic->loop_tag = enic->config.loop_tag;
2428                 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2429         }
2430         if (ENIC_SETTING(enic, TXCSUM))
2431                 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2432         if (ENIC_SETTING(enic, TSO))
2433                 netdev->features |= NETIF_F_TSO |
2434                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2435         if (ENIC_SETTING(enic, LRO))
2436                 netdev->features |= NETIF_F_GRO;
2437         if (using_dac)
2438                 netdev->features |= NETIF_F_HIGHDMA;
2439
2440         enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
2441
2442         err = register_netdev(netdev);
2443         if (err) {
2444                 dev_err(dev, "Cannot register net device, aborting\n");
2445                 goto err_out_dev_deinit;
2446         }
2447
2448         return 0;
2449
2450 err_out_dev_deinit:
2451         enic_dev_deinit(enic);
2452 err_out_dev_close:
2453         vnic_dev_close(enic->vdev);
2454 err_out_vnic_unregister:
2455         vnic_dev_unregister(enic->vdev);
2456 err_out_iounmap:
2457         enic_iounmap(enic);
2458 err_out_release_regions:
2459         pci_release_regions(pdev);
2460 err_out_disable_device:
2461         pci_disable_device(pdev);
2462 err_out_free_netdev:
2463         pci_set_drvdata(pdev, NULL);
2464         free_netdev(netdev);
2465
2466         return err;
2467 }
2468
2469 static void __devexit enic_remove(struct pci_dev *pdev)
2470 {
2471         struct net_device *netdev = pci_get_drvdata(pdev);
2472
2473         if (netdev) {
2474                 struct enic *enic = netdev_priv(netdev);
2475
2476                 flush_scheduled_work();
2477                 unregister_netdev(netdev);
2478                 enic_dev_deinit(enic);
2479                 vnic_dev_close(enic->vdev);
2480                 vnic_dev_unregister(enic->vdev);
2481                 enic_iounmap(enic);
2482                 pci_release_regions(pdev);
2483                 pci_disable_device(pdev);
2484                 pci_set_drvdata(pdev, NULL);
2485                 free_netdev(netdev);
2486         }
2487 }
2488
2489 static struct pci_driver enic_driver = {
2490         .name = DRV_NAME,
2491         .id_table = enic_id_table,
2492         .probe = enic_probe,
2493         .remove = __devexit_p(enic_remove),
2494 };
2495
2496 static int __init enic_init_module(void)
2497 {
2498         pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2499
2500         return pci_register_driver(&enic_driver);
2501 }
2502
2503 static void __exit enic_cleanup_module(void)
2504 {
2505         pci_unregister_driver(&enic_driver);
2506 }
2507
2508 module_init(enic_init_module);
2509 module_exit(enic_cleanup_module);