ixgbe: Correctly name and handle MSI-X other interrupt
[pandora-kernel.git] / drivers / net / ethernet / cisco / 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/interrupt.h>
27 #include <linux/workqueue.h>
28 #include <linux/pci.h>
29 #include <linux/netdevice.h>
30 #include <linux/etherdevice.h>
31 #include <linux/if.h>
32 #include <linux/if_ether.h>
33 #include <linux/if_vlan.h>
34 #include <linux/ethtool.h>
35 #include <linux/in.h>
36 #include <linux/ip.h>
37 #include <linux/ipv6.h>
38 #include <linux/tcp.h>
39 #include <linux/rtnetlink.h>
40 #include <linux/prefetch.h>
41 #include <net/ip6_checksum.h>
42
43 #include "cq_enet_desc.h"
44 #include "vnic_dev.h"
45 #include "vnic_intr.h"
46 #include "vnic_stats.h"
47 #include "vnic_vic.h"
48 #include "enic_res.h"
49 #include "enic.h"
50 #include "enic_dev.h"
51 #include "enic_pp.h"
52
53 #define ENIC_NOTIFY_TIMER_PERIOD        (2 * HZ)
54 #define WQ_ENET_MAX_DESC_LEN            (1 << WQ_ENET_LEN_BITS)
55 #define MAX_TSO                         (1 << 16)
56 #define ENIC_DESC_MAX_SPLITS            (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
57
58 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
59 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
60
61 /* Supported devices */
62 static DEFINE_PCI_DEVICE_TABLE(enic_id_table) = {
63         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
64         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
65         { 0, }  /* end of table */
66 };
67
68 MODULE_DESCRIPTION(DRV_DESCRIPTION);
69 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
70 MODULE_LICENSE("GPL");
71 MODULE_VERSION(DRV_VERSION);
72 MODULE_DEVICE_TABLE(pci, enic_id_table);
73
74 struct enic_stat {
75         char name[ETH_GSTRING_LEN];
76         unsigned int offset;
77 };
78
79 #define ENIC_TX_STAT(stat)      \
80         { .name = #stat, .offset = offsetof(struct vnic_tx_stats, stat) / 8 }
81 #define ENIC_RX_STAT(stat)      \
82         { .name = #stat, .offset = offsetof(struct vnic_rx_stats, stat) / 8 }
83
84 static const struct enic_stat enic_tx_stats[] = {
85         ENIC_TX_STAT(tx_frames_ok),
86         ENIC_TX_STAT(tx_unicast_frames_ok),
87         ENIC_TX_STAT(tx_multicast_frames_ok),
88         ENIC_TX_STAT(tx_broadcast_frames_ok),
89         ENIC_TX_STAT(tx_bytes_ok),
90         ENIC_TX_STAT(tx_unicast_bytes_ok),
91         ENIC_TX_STAT(tx_multicast_bytes_ok),
92         ENIC_TX_STAT(tx_broadcast_bytes_ok),
93         ENIC_TX_STAT(tx_drops),
94         ENIC_TX_STAT(tx_errors),
95         ENIC_TX_STAT(tx_tso),
96 };
97
98 static const struct enic_stat enic_rx_stats[] = {
99         ENIC_RX_STAT(rx_frames_ok),
100         ENIC_RX_STAT(rx_frames_total),
101         ENIC_RX_STAT(rx_unicast_frames_ok),
102         ENIC_RX_STAT(rx_multicast_frames_ok),
103         ENIC_RX_STAT(rx_broadcast_frames_ok),
104         ENIC_RX_STAT(rx_bytes_ok),
105         ENIC_RX_STAT(rx_unicast_bytes_ok),
106         ENIC_RX_STAT(rx_multicast_bytes_ok),
107         ENIC_RX_STAT(rx_broadcast_bytes_ok),
108         ENIC_RX_STAT(rx_drop),
109         ENIC_RX_STAT(rx_no_bufs),
110         ENIC_RX_STAT(rx_errors),
111         ENIC_RX_STAT(rx_rss),
112         ENIC_RX_STAT(rx_crc_errors),
113         ENIC_RX_STAT(rx_frames_64),
114         ENIC_RX_STAT(rx_frames_127),
115         ENIC_RX_STAT(rx_frames_255),
116         ENIC_RX_STAT(rx_frames_511),
117         ENIC_RX_STAT(rx_frames_1023),
118         ENIC_RX_STAT(rx_frames_1518),
119         ENIC_RX_STAT(rx_frames_to_max),
120 };
121
122 static const unsigned int enic_n_tx_stats = ARRAY_SIZE(enic_tx_stats);
123 static const unsigned int enic_n_rx_stats = ARRAY_SIZE(enic_rx_stats);
124
125 static int enic_is_dynamic(struct enic *enic)
126 {
127         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
128 }
129
130 static inline unsigned int enic_cq_rq(struct enic *enic, unsigned int rq)
131 {
132         return rq;
133 }
134
135 static inline unsigned int enic_cq_wq(struct enic *enic, unsigned int wq)
136 {
137         return enic->rq_count + wq;
138 }
139
140 static inline unsigned int enic_legacy_io_intr(void)
141 {
142         return 0;
143 }
144
145 static inline unsigned int enic_legacy_err_intr(void)
146 {
147         return 1;
148 }
149
150 static inline unsigned int enic_legacy_notify_intr(void)
151 {
152         return 2;
153 }
154
155 static inline unsigned int enic_msix_rq_intr(struct enic *enic, unsigned int rq)
156 {
157         return enic->cq[enic_cq_rq(enic, rq)].interrupt_offset;
158 }
159
160 static inline unsigned int enic_msix_wq_intr(struct enic *enic, unsigned int wq)
161 {
162         return enic->cq[enic_cq_wq(enic, wq)].interrupt_offset;
163 }
164
165 static inline unsigned int enic_msix_err_intr(struct enic *enic)
166 {
167         return enic->rq_count + enic->wq_count;
168 }
169
170 static inline unsigned int enic_msix_notify_intr(struct enic *enic)
171 {
172         return enic->rq_count + enic->wq_count + 1;
173 }
174
175 static int enic_get_settings(struct net_device *netdev,
176         struct ethtool_cmd *ecmd)
177 {
178         struct enic *enic = netdev_priv(netdev);
179
180         ecmd->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
181         ecmd->advertising = (ADVERTISED_10000baseT_Full | ADVERTISED_FIBRE);
182         ecmd->port = PORT_FIBRE;
183         ecmd->transceiver = XCVR_EXTERNAL;
184
185         if (netif_carrier_ok(netdev)) {
186                 ethtool_cmd_speed_set(ecmd, vnic_dev_port_speed(enic->vdev));
187                 ecmd->duplex = DUPLEX_FULL;
188         } else {
189                 ethtool_cmd_speed_set(ecmd, -1);
190                 ecmd->duplex = -1;
191         }
192
193         ecmd->autoneg = AUTONEG_DISABLE;
194
195         return 0;
196 }
197
198 static void enic_get_drvinfo(struct net_device *netdev,
199         struct ethtool_drvinfo *drvinfo)
200 {
201         struct enic *enic = netdev_priv(netdev);
202         struct vnic_devcmd_fw_info *fw_info;
203
204         enic_dev_fw_info(enic, &fw_info);
205
206         strncpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
207         strncpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
208         strncpy(drvinfo->fw_version, fw_info->fw_version,
209                 sizeof(drvinfo->fw_version));
210         strncpy(drvinfo->bus_info, pci_name(enic->pdev),
211                 sizeof(drvinfo->bus_info));
212 }
213
214 static void enic_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
215 {
216         unsigned int i;
217
218         switch (stringset) {
219         case ETH_SS_STATS:
220                 for (i = 0; i < enic_n_tx_stats; i++) {
221                         memcpy(data, enic_tx_stats[i].name, ETH_GSTRING_LEN);
222                         data += ETH_GSTRING_LEN;
223                 }
224                 for (i = 0; i < enic_n_rx_stats; i++) {
225                         memcpy(data, enic_rx_stats[i].name, ETH_GSTRING_LEN);
226                         data += ETH_GSTRING_LEN;
227                 }
228                 break;
229         }
230 }
231
232 static int enic_get_sset_count(struct net_device *netdev, int sset)
233 {
234         switch (sset) {
235         case ETH_SS_STATS:
236                 return enic_n_tx_stats + enic_n_rx_stats;
237         default:
238                 return -EOPNOTSUPP;
239         }
240 }
241
242 static void enic_get_ethtool_stats(struct net_device *netdev,
243         struct ethtool_stats *stats, u64 *data)
244 {
245         struct enic *enic = netdev_priv(netdev);
246         struct vnic_stats *vstats;
247         unsigned int i;
248
249         enic_dev_stats_dump(enic, &vstats);
250
251         for (i = 0; i < enic_n_tx_stats; i++)
252                 *(data++) = ((u64 *)&vstats->tx)[enic_tx_stats[i].offset];
253         for (i = 0; i < enic_n_rx_stats; i++)
254                 *(data++) = ((u64 *)&vstats->rx)[enic_rx_stats[i].offset];
255 }
256
257 static u32 enic_get_msglevel(struct net_device *netdev)
258 {
259         struct enic *enic = netdev_priv(netdev);
260         return enic->msg_enable;
261 }
262
263 static void enic_set_msglevel(struct net_device *netdev, u32 value)
264 {
265         struct enic *enic = netdev_priv(netdev);
266         enic->msg_enable = value;
267 }
268
269 static int enic_get_coalesce(struct net_device *netdev,
270         struct ethtool_coalesce *ecmd)
271 {
272         struct enic *enic = netdev_priv(netdev);
273
274         ecmd->tx_coalesce_usecs = enic->tx_coalesce_usecs;
275         ecmd->rx_coalesce_usecs = enic->rx_coalesce_usecs;
276
277         return 0;
278 }
279
280 static int enic_set_coalesce(struct net_device *netdev,
281         struct ethtool_coalesce *ecmd)
282 {
283         struct enic *enic = netdev_priv(netdev);
284         u32 tx_coalesce_usecs;
285         u32 rx_coalesce_usecs;
286         unsigned int i, intr;
287
288         tx_coalesce_usecs = min_t(u32, ecmd->tx_coalesce_usecs,
289                 vnic_dev_get_intr_coal_timer_max(enic->vdev));
290         rx_coalesce_usecs = min_t(u32, ecmd->rx_coalesce_usecs,
291                 vnic_dev_get_intr_coal_timer_max(enic->vdev));
292
293         switch (vnic_dev_get_intr_mode(enic->vdev)) {
294         case VNIC_DEV_INTR_MODE_INTX:
295                 if (tx_coalesce_usecs != rx_coalesce_usecs)
296                         return -EINVAL;
297
298                 intr = enic_legacy_io_intr();
299                 vnic_intr_coalescing_timer_set(&enic->intr[intr],
300                         tx_coalesce_usecs);
301                 break;
302         case VNIC_DEV_INTR_MODE_MSI:
303                 if (tx_coalesce_usecs != rx_coalesce_usecs)
304                         return -EINVAL;
305
306                 vnic_intr_coalescing_timer_set(&enic->intr[0],
307                         tx_coalesce_usecs);
308                 break;
309         case VNIC_DEV_INTR_MODE_MSIX:
310                 for (i = 0; i < enic->wq_count; i++) {
311                         intr = enic_msix_wq_intr(enic, i);
312                         vnic_intr_coalescing_timer_set(&enic->intr[intr],
313                                 tx_coalesce_usecs);
314                 }
315
316                 for (i = 0; i < enic->rq_count; i++) {
317                         intr = enic_msix_rq_intr(enic, i);
318                         vnic_intr_coalescing_timer_set(&enic->intr[intr],
319                                 rx_coalesce_usecs);
320                 }
321
322                 break;
323         default:
324                 break;
325         }
326
327         enic->tx_coalesce_usecs = tx_coalesce_usecs;
328         enic->rx_coalesce_usecs = rx_coalesce_usecs;
329
330         return 0;
331 }
332
333 static const struct ethtool_ops enic_ethtool_ops = {
334         .get_settings = enic_get_settings,
335         .get_drvinfo = enic_get_drvinfo,
336         .get_msglevel = enic_get_msglevel,
337         .set_msglevel = enic_set_msglevel,
338         .get_link = ethtool_op_get_link,
339         .get_strings = enic_get_strings,
340         .get_sset_count = enic_get_sset_count,
341         .get_ethtool_stats = enic_get_ethtool_stats,
342         .get_coalesce = enic_get_coalesce,
343         .set_coalesce = enic_set_coalesce,
344 };
345
346 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
347 {
348         struct enic *enic = vnic_dev_priv(wq->vdev);
349
350         if (buf->sop)
351                 pci_unmap_single(enic->pdev, buf->dma_addr,
352                         buf->len, PCI_DMA_TODEVICE);
353         else
354                 pci_unmap_page(enic->pdev, buf->dma_addr,
355                         buf->len, PCI_DMA_TODEVICE);
356
357         if (buf->os_buf)
358                 dev_kfree_skb_any(buf->os_buf);
359 }
360
361 static void enic_wq_free_buf(struct vnic_wq *wq,
362         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
363 {
364         enic_free_wq_buf(wq, buf);
365 }
366
367 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
368         u8 type, u16 q_number, u16 completed_index, void *opaque)
369 {
370         struct enic *enic = vnic_dev_priv(vdev);
371
372         spin_lock(&enic->wq_lock[q_number]);
373
374         vnic_wq_service(&enic->wq[q_number], cq_desc,
375                 completed_index, enic_wq_free_buf,
376                 opaque);
377
378         if (netif_queue_stopped(enic->netdev) &&
379             vnic_wq_desc_avail(&enic->wq[q_number]) >=
380             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
381                 netif_wake_queue(enic->netdev);
382
383         spin_unlock(&enic->wq_lock[q_number]);
384
385         return 0;
386 }
387
388 static void enic_log_q_error(struct enic *enic)
389 {
390         unsigned int i;
391         u32 error_status;
392
393         for (i = 0; i < enic->wq_count; i++) {
394                 error_status = vnic_wq_error_status(&enic->wq[i]);
395                 if (error_status)
396                         netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
397                                 i, error_status);
398         }
399
400         for (i = 0; i < enic->rq_count; i++) {
401                 error_status = vnic_rq_error_status(&enic->rq[i]);
402                 if (error_status)
403                         netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
404                                 i, error_status);
405         }
406 }
407
408 static void enic_msglvl_check(struct enic *enic)
409 {
410         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
411
412         if (msg_enable != enic->msg_enable) {
413                 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
414                         enic->msg_enable, msg_enable);
415                 enic->msg_enable = msg_enable;
416         }
417 }
418
419 static void enic_mtu_check(struct enic *enic)
420 {
421         u32 mtu = vnic_dev_mtu(enic->vdev);
422         struct net_device *netdev = enic->netdev;
423
424         if (mtu && mtu != enic->port_mtu) {
425                 enic->port_mtu = mtu;
426                 if (enic_is_dynamic(enic)) {
427                         mtu = max_t(int, ENIC_MIN_MTU,
428                                 min_t(int, ENIC_MAX_MTU, mtu));
429                         if (mtu != netdev->mtu)
430                                 schedule_work(&enic->change_mtu_work);
431                 } else {
432                         if (mtu < netdev->mtu)
433                                 netdev_warn(netdev,
434                                         "interface MTU (%d) set higher "
435                                         "than switch port MTU (%d)\n",
436                                         netdev->mtu, mtu);
437                 }
438         }
439 }
440
441 static void enic_link_check(struct enic *enic)
442 {
443         int link_status = vnic_dev_link_status(enic->vdev);
444         int carrier_ok = netif_carrier_ok(enic->netdev);
445
446         if (link_status && !carrier_ok) {
447                 netdev_info(enic->netdev, "Link UP\n");
448                 netif_carrier_on(enic->netdev);
449         } else if (!link_status && carrier_ok) {
450                 netdev_info(enic->netdev, "Link DOWN\n");
451                 netif_carrier_off(enic->netdev);
452         }
453 }
454
455 static void enic_notify_check(struct enic *enic)
456 {
457         enic_msglvl_check(enic);
458         enic_mtu_check(enic);
459         enic_link_check(enic);
460 }
461
462 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
463
464 static irqreturn_t enic_isr_legacy(int irq, void *data)
465 {
466         struct net_device *netdev = data;
467         struct enic *enic = netdev_priv(netdev);
468         unsigned int io_intr = enic_legacy_io_intr();
469         unsigned int err_intr = enic_legacy_err_intr();
470         unsigned int notify_intr = enic_legacy_notify_intr();
471         u32 pba;
472
473         vnic_intr_mask(&enic->intr[io_intr]);
474
475         pba = vnic_intr_legacy_pba(enic->legacy_pba);
476         if (!pba) {
477                 vnic_intr_unmask(&enic->intr[io_intr]);
478                 return IRQ_NONE;        /* not our interrupt */
479         }
480
481         if (ENIC_TEST_INTR(pba, notify_intr)) {
482                 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
483                 enic_notify_check(enic);
484         }
485
486         if (ENIC_TEST_INTR(pba, err_intr)) {
487                 vnic_intr_return_all_credits(&enic->intr[err_intr]);
488                 enic_log_q_error(enic);
489                 /* schedule recovery from WQ/RQ error */
490                 schedule_work(&enic->reset);
491                 return IRQ_HANDLED;
492         }
493
494         if (ENIC_TEST_INTR(pba, io_intr)) {
495                 if (napi_schedule_prep(&enic->napi[0]))
496                         __napi_schedule(&enic->napi[0]);
497         } else {
498                 vnic_intr_unmask(&enic->intr[io_intr]);
499         }
500
501         return IRQ_HANDLED;
502 }
503
504 static irqreturn_t enic_isr_msi(int irq, void *data)
505 {
506         struct enic *enic = data;
507
508         /* With MSI, there is no sharing of interrupts, so this is
509          * our interrupt and there is no need to ack it.  The device
510          * is not providing per-vector masking, so the OS will not
511          * write to PCI config space to mask/unmask the interrupt.
512          * We're using mask_on_assertion for MSI, so the device
513          * automatically masks the interrupt when the interrupt is
514          * generated.  Later, when exiting polling, the interrupt
515          * will be unmasked (see enic_poll).
516          *
517          * Also, the device uses the same PCIe Traffic Class (TC)
518          * for Memory Write data and MSI, so there are no ordering
519          * issues; the MSI will always arrive at the Root Complex
520          * _after_ corresponding Memory Writes (i.e. descriptor
521          * writes).
522          */
523
524         napi_schedule(&enic->napi[0]);
525
526         return IRQ_HANDLED;
527 }
528
529 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
530 {
531         struct napi_struct *napi = data;
532
533         /* schedule NAPI polling for RQ cleanup */
534         napi_schedule(napi);
535
536         return IRQ_HANDLED;
537 }
538
539 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
540 {
541         struct enic *enic = data;
542         unsigned int cq = enic_cq_wq(enic, 0);
543         unsigned int intr = enic_msix_wq_intr(enic, 0);
544         unsigned int wq_work_to_do = -1; /* no limit */
545         unsigned int wq_work_done;
546
547         wq_work_done = vnic_cq_service(&enic->cq[cq],
548                 wq_work_to_do, enic_wq_service, NULL);
549
550         vnic_intr_return_credits(&enic->intr[intr],
551                 wq_work_done,
552                 1 /* unmask intr */,
553                 1 /* reset intr timer */);
554
555         return IRQ_HANDLED;
556 }
557
558 static irqreturn_t enic_isr_msix_err(int irq, void *data)
559 {
560         struct enic *enic = data;
561         unsigned int intr = enic_msix_err_intr(enic);
562
563         vnic_intr_return_all_credits(&enic->intr[intr]);
564
565         enic_log_q_error(enic);
566
567         /* schedule recovery from WQ/RQ error */
568         schedule_work(&enic->reset);
569
570         return IRQ_HANDLED;
571 }
572
573 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
574 {
575         struct enic *enic = data;
576         unsigned int intr = enic_msix_notify_intr(enic);
577
578         vnic_intr_return_all_credits(&enic->intr[intr]);
579         enic_notify_check(enic);
580
581         return IRQ_HANDLED;
582 }
583
584 static inline void enic_queue_wq_skb_cont(struct enic *enic,
585         struct vnic_wq *wq, struct sk_buff *skb,
586         unsigned int len_left, int loopback)
587 {
588         skb_frag_t *frag;
589
590         /* Queue additional data fragments */
591         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
592                 len_left -= frag->size;
593                 enic_queue_wq_desc_cont(wq, skb,
594                         pci_map_page(enic->pdev, frag->page,
595                                 frag->page_offset, frag->size,
596                                 PCI_DMA_TODEVICE),
597                         frag->size,
598                         (len_left == 0),        /* EOP? */
599                         loopback);
600         }
601 }
602
603 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
604         struct vnic_wq *wq, struct sk_buff *skb,
605         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
606 {
607         unsigned int head_len = skb_headlen(skb);
608         unsigned int len_left = skb->len - head_len;
609         int eop = (len_left == 0);
610
611         /* Queue the main skb fragment. The fragments are no larger
612          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
613          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
614          * per fragment is queued.
615          */
616         enic_queue_wq_desc(wq, skb,
617                 pci_map_single(enic->pdev, skb->data,
618                         head_len, PCI_DMA_TODEVICE),
619                 head_len,
620                 vlan_tag_insert, vlan_tag,
621                 eop, loopback);
622
623         if (!eop)
624                 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
625 }
626
627 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
628         struct vnic_wq *wq, struct sk_buff *skb,
629         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
630 {
631         unsigned int head_len = skb_headlen(skb);
632         unsigned int len_left = skb->len - head_len;
633         unsigned int hdr_len = skb_checksum_start_offset(skb);
634         unsigned int csum_offset = hdr_len + skb->csum_offset;
635         int eop = (len_left == 0);
636
637         /* Queue the main skb fragment. The fragments are no larger
638          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
639          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
640          * per fragment is queued.
641          */
642         enic_queue_wq_desc_csum_l4(wq, skb,
643                 pci_map_single(enic->pdev, skb->data,
644                         head_len, PCI_DMA_TODEVICE),
645                 head_len,
646                 csum_offset,
647                 hdr_len,
648                 vlan_tag_insert, vlan_tag,
649                 eop, loopback);
650
651         if (!eop)
652                 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
653 }
654
655 static inline void enic_queue_wq_skb_tso(struct enic *enic,
656         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
657         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
658 {
659         unsigned int frag_len_left = skb_headlen(skb);
660         unsigned int len_left = skb->len - frag_len_left;
661         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
662         int eop = (len_left == 0);
663         unsigned int len;
664         dma_addr_t dma_addr;
665         unsigned int offset = 0;
666         skb_frag_t *frag;
667
668         /* Preload TCP csum field with IP pseudo hdr calculated
669          * with IP length set to zero.  HW will later add in length
670          * to each TCP segment resulting from the TSO.
671          */
672
673         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
674                 ip_hdr(skb)->check = 0;
675                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
676                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
677         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
678                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
679                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
680         }
681
682         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
683          * for the main skb fragment
684          */
685         while (frag_len_left) {
686                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
687                 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
688                                 len, PCI_DMA_TODEVICE);
689                 enic_queue_wq_desc_tso(wq, skb,
690                         dma_addr,
691                         len,
692                         mss, hdr_len,
693                         vlan_tag_insert, vlan_tag,
694                         eop && (len == frag_len_left), loopback);
695                 frag_len_left -= len;
696                 offset += len;
697         }
698
699         if (eop)
700                 return;
701
702         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
703          * for additional data fragments
704          */
705         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
706                 len_left -= frag->size;
707                 frag_len_left = frag->size;
708                 offset = frag->page_offset;
709
710                 while (frag_len_left) {
711                         len = min(frag_len_left,
712                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
713                         dma_addr = pci_map_page(enic->pdev, frag->page,
714                                 offset, len,
715                                 PCI_DMA_TODEVICE);
716                         enic_queue_wq_desc_cont(wq, skb,
717                                 dma_addr,
718                                 len,
719                                 (len_left == 0) &&
720                                 (len == frag_len_left),         /* EOP? */
721                                 loopback);
722                         frag_len_left -= len;
723                         offset += len;
724                 }
725         }
726 }
727
728 static inline void enic_queue_wq_skb(struct enic *enic,
729         struct vnic_wq *wq, struct sk_buff *skb)
730 {
731         unsigned int mss = skb_shinfo(skb)->gso_size;
732         unsigned int vlan_tag = 0;
733         int vlan_tag_insert = 0;
734         int loopback = 0;
735
736         if (vlan_tx_tag_present(skb)) {
737                 /* VLAN tag from trunking driver */
738                 vlan_tag_insert = 1;
739                 vlan_tag = vlan_tx_tag_get(skb);
740         } else if (enic->loop_enable) {
741                 vlan_tag = enic->loop_tag;
742                 loopback = 1;
743         }
744
745         if (mss)
746                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
747                         vlan_tag_insert, vlan_tag, loopback);
748         else if (skb->ip_summed == CHECKSUM_PARTIAL)
749                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
750                         vlan_tag_insert, vlan_tag, loopback);
751         else
752                 enic_queue_wq_skb_vlan(enic, wq, skb,
753                         vlan_tag_insert, vlan_tag, loopback);
754 }
755
756 /* netif_tx_lock held, process context with BHs disabled, or BH */
757 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
758         struct net_device *netdev)
759 {
760         struct enic *enic = netdev_priv(netdev);
761         struct vnic_wq *wq = &enic->wq[0];
762         unsigned long flags;
763
764         if (skb->len <= 0) {
765                 dev_kfree_skb(skb);
766                 return NETDEV_TX_OK;
767         }
768
769         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
770          * which is very likely.  In the off chance it's going to take
771          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
772          */
773
774         if (skb_shinfo(skb)->gso_size == 0 &&
775             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
776             skb_linearize(skb)) {
777                 dev_kfree_skb(skb);
778                 return NETDEV_TX_OK;
779         }
780
781         spin_lock_irqsave(&enic->wq_lock[0], flags);
782
783         if (vnic_wq_desc_avail(wq) <
784             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
785                 netif_stop_queue(netdev);
786                 /* This is a hard error, log it */
787                 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
788                 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
789                 return NETDEV_TX_BUSY;
790         }
791
792         enic_queue_wq_skb(enic, wq, skb);
793
794         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
795                 netif_stop_queue(netdev);
796
797         spin_unlock_irqrestore(&enic->wq_lock[0], flags);
798
799         return NETDEV_TX_OK;
800 }
801
802 /* dev_base_lock rwlock held, nominally process context */
803 static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
804                                                 struct rtnl_link_stats64 *net_stats)
805 {
806         struct enic *enic = netdev_priv(netdev);
807         struct vnic_stats *stats;
808
809         enic_dev_stats_dump(enic, &stats);
810
811         net_stats->tx_packets = stats->tx.tx_frames_ok;
812         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
813         net_stats->tx_errors = stats->tx.tx_errors;
814         net_stats->tx_dropped = stats->tx.tx_drops;
815
816         net_stats->rx_packets = stats->rx.rx_frames_ok;
817         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
818         net_stats->rx_errors = stats->rx.rx_errors;
819         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
820         net_stats->rx_over_errors = enic->rq_truncated_pkts;
821         net_stats->rx_crc_errors = enic->rq_bad_fcs;
822         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
823
824         return net_stats;
825 }
826
827 void enic_reset_addr_lists(struct enic *enic)
828 {
829         enic->mc_count = 0;
830         enic->uc_count = 0;
831         enic->flags = 0;
832 }
833
834 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
835 {
836         struct enic *enic = netdev_priv(netdev);
837
838         if (enic_is_dynamic(enic)) {
839                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
840                         return -EADDRNOTAVAIL;
841         } else {
842                 if (!is_valid_ether_addr(addr))
843                         return -EADDRNOTAVAIL;
844         }
845
846         memcpy(netdev->dev_addr, addr, netdev->addr_len);
847
848         return 0;
849 }
850
851 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
852 {
853         struct enic *enic = netdev_priv(netdev);
854         struct sockaddr *saddr = p;
855         char *addr = saddr->sa_data;
856         int err;
857
858         if (netif_running(enic->netdev)) {
859                 err = enic_dev_del_station_addr(enic);
860                 if (err)
861                         return err;
862         }
863
864         err = enic_set_mac_addr(netdev, addr);
865         if (err)
866                 return err;
867
868         if (netif_running(enic->netdev)) {
869                 err = enic_dev_add_station_addr(enic);
870                 if (err)
871                         return err;
872         }
873
874         return err;
875 }
876
877 static int enic_set_mac_address(struct net_device *netdev, void *p)
878 {
879         struct sockaddr *saddr = p;
880         char *addr = saddr->sa_data;
881         struct enic *enic = netdev_priv(netdev);
882         int err;
883
884         err = enic_dev_del_station_addr(enic);
885         if (err)
886                 return err;
887
888         err = enic_set_mac_addr(netdev, addr);
889         if (err)
890                 return err;
891
892         return enic_dev_add_station_addr(enic);
893 }
894
895 static void enic_update_multicast_addr_list(struct enic *enic)
896 {
897         struct net_device *netdev = enic->netdev;
898         struct netdev_hw_addr *ha;
899         unsigned int mc_count = netdev_mc_count(netdev);
900         u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
901         unsigned int i, j;
902
903         if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS) {
904                 netdev_warn(netdev, "Registering only %d out of %d "
905                         "multicast addresses\n",
906                         ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
907                 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
908         }
909
910         /* Is there an easier way?  Trying to minimize to
911          * calls to add/del multicast addrs.  We keep the
912          * addrs from the last call in enic->mc_addr and
913          * look for changes to add/del.
914          */
915
916         i = 0;
917         netdev_for_each_mc_addr(ha, netdev) {
918                 if (i == mc_count)
919                         break;
920                 memcpy(mc_addr[i++], ha->addr, ETH_ALEN);
921         }
922
923         for (i = 0; i < enic->mc_count; i++) {
924                 for (j = 0; j < mc_count; j++)
925                         if (compare_ether_addr(enic->mc_addr[i],
926                                 mc_addr[j]) == 0)
927                                 break;
928                 if (j == mc_count)
929                         enic_dev_del_addr(enic, enic->mc_addr[i]);
930         }
931
932         for (i = 0; i < mc_count; i++) {
933                 for (j = 0; j < enic->mc_count; j++)
934                         if (compare_ether_addr(mc_addr[i],
935                                 enic->mc_addr[j]) == 0)
936                                 break;
937                 if (j == enic->mc_count)
938                         enic_dev_add_addr(enic, mc_addr[i]);
939         }
940
941         /* Save the list to compare against next time
942          */
943
944         for (i = 0; i < mc_count; i++)
945                 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
946
947         enic->mc_count = mc_count;
948 }
949
950 static void enic_update_unicast_addr_list(struct enic *enic)
951 {
952         struct net_device *netdev = enic->netdev;
953         struct netdev_hw_addr *ha;
954         unsigned int uc_count = netdev_uc_count(netdev);
955         u8 uc_addr[ENIC_UNICAST_PERFECT_FILTERS][ETH_ALEN];
956         unsigned int i, j;
957
958         if (uc_count > ENIC_UNICAST_PERFECT_FILTERS) {
959                 netdev_warn(netdev, "Registering only %d out of %d "
960                         "unicast addresses\n",
961                         ENIC_UNICAST_PERFECT_FILTERS, uc_count);
962                 uc_count = ENIC_UNICAST_PERFECT_FILTERS;
963         }
964
965         /* Is there an easier way?  Trying to minimize to
966          * calls to add/del unicast addrs.  We keep the
967          * addrs from the last call in enic->uc_addr and
968          * look for changes to add/del.
969          */
970
971         i = 0;
972         netdev_for_each_uc_addr(ha, netdev) {
973                 if (i == uc_count)
974                         break;
975                 memcpy(uc_addr[i++], ha->addr, ETH_ALEN);
976         }
977
978         for (i = 0; i < enic->uc_count; i++) {
979                 for (j = 0; j < uc_count; j++)
980                         if (compare_ether_addr(enic->uc_addr[i],
981                                 uc_addr[j]) == 0)
982                                 break;
983                 if (j == uc_count)
984                         enic_dev_del_addr(enic, enic->uc_addr[i]);
985         }
986
987         for (i = 0; i < uc_count; i++) {
988                 for (j = 0; j < enic->uc_count; j++)
989                         if (compare_ether_addr(uc_addr[i],
990                                 enic->uc_addr[j]) == 0)
991                                 break;
992                 if (j == enic->uc_count)
993                         enic_dev_add_addr(enic, uc_addr[i]);
994         }
995
996         /* Save the list to compare against next time
997          */
998
999         for (i = 0; i < uc_count; i++)
1000                 memcpy(enic->uc_addr[i], uc_addr[i], ETH_ALEN);
1001
1002         enic->uc_count = uc_count;
1003 }
1004
1005 /* netif_tx_lock held, BHs disabled */
1006 static void enic_set_rx_mode(struct net_device *netdev)
1007 {
1008         struct enic *enic = netdev_priv(netdev);
1009         int directed = 1;
1010         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
1011         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
1012         int promisc = (netdev->flags & IFF_PROMISC) ||
1013                 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
1014         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
1015                 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
1016         unsigned int flags = netdev->flags |
1017                 (allmulti ? IFF_ALLMULTI : 0) |
1018                 (promisc ? IFF_PROMISC : 0);
1019
1020         if (enic->flags != flags) {
1021                 enic->flags = flags;
1022                 enic_dev_packet_filter(enic, directed,
1023                         multicast, broadcast, promisc, allmulti);
1024         }
1025
1026         if (!promisc) {
1027                 enic_update_unicast_addr_list(enic);
1028                 if (!allmulti)
1029                         enic_update_multicast_addr_list(enic);
1030         }
1031 }
1032
1033 /* netif_tx_lock held, BHs disabled */
1034 static void enic_tx_timeout(struct net_device *netdev)
1035 {
1036         struct enic *enic = netdev_priv(netdev);
1037         schedule_work(&enic->reset);
1038 }
1039
1040 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1041 {
1042         struct enic *enic = netdev_priv(netdev);
1043
1044         if (vf != PORT_SELF_VF)
1045                 return -EOPNOTSUPP;
1046
1047         /* Ignore the vf argument for now. We can assume the request
1048          * is coming on a vf.
1049          */
1050         if (is_valid_ether_addr(mac)) {
1051                 memcpy(enic->pp.vf_mac, mac, ETH_ALEN);
1052                 return 0;
1053         } else
1054                 return -EINVAL;
1055 }
1056
1057 static int enic_set_vf_port(struct net_device *netdev, int vf,
1058         struct nlattr *port[])
1059 {
1060         struct enic *enic = netdev_priv(netdev);
1061         struct enic_port_profile prev_pp;
1062         int err = 0, restore_pp = 1;
1063
1064         /* don't support VFs, yet */
1065         if (vf != PORT_SELF_VF)
1066                 return -EOPNOTSUPP;
1067
1068         if (!port[IFLA_PORT_REQUEST])
1069                 return -EOPNOTSUPP;
1070
1071         memcpy(&prev_pp, &enic->pp, sizeof(enic->pp));
1072         memset(&enic->pp, 0, sizeof(enic->pp));
1073
1074         enic->pp.set |= ENIC_SET_REQUEST;
1075         enic->pp.request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1076
1077         if (port[IFLA_PORT_PROFILE]) {
1078                 enic->pp.set |= ENIC_SET_NAME;
1079                 memcpy(enic->pp.name, nla_data(port[IFLA_PORT_PROFILE]),
1080                         PORT_PROFILE_MAX);
1081         }
1082
1083         if (port[IFLA_PORT_INSTANCE_UUID]) {
1084                 enic->pp.set |= ENIC_SET_INSTANCE;
1085                 memcpy(enic->pp.instance_uuid,
1086                         nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
1087         }
1088
1089         if (port[IFLA_PORT_HOST_UUID]) {
1090                 enic->pp.set |= ENIC_SET_HOST;
1091                 memcpy(enic->pp.host_uuid,
1092                         nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
1093         }
1094
1095         /* Special case handling: mac came from IFLA_VF_MAC */
1096         if (!is_zero_ether_addr(prev_pp.vf_mac))
1097                 memcpy(enic->pp.mac_addr, prev_pp.vf_mac, ETH_ALEN);
1098
1099                 if (is_zero_ether_addr(netdev->dev_addr))
1100                         random_ether_addr(netdev->dev_addr);
1101
1102         err = enic_process_set_pp_request(enic, &prev_pp, &restore_pp);
1103         if (err) {
1104                 if (restore_pp) {
1105                         /* Things are still the way they were: Implicit
1106                          * DISASSOCIATE failed
1107                          */
1108                         memcpy(&enic->pp, &prev_pp, sizeof(enic->pp));
1109                 } else {
1110                         memset(&enic->pp, 0, sizeof(enic->pp));
1111                         memset(netdev->dev_addr, 0, ETH_ALEN);
1112                 }
1113         } else {
1114                 /* Set flag to indicate that the port assoc/disassoc
1115                  * request has been sent out to fw
1116                  */
1117                 enic->pp.set |= ENIC_PORT_REQUEST_APPLIED;
1118
1119                 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
1120                 if (enic->pp.request == PORT_REQUEST_DISASSOCIATE) {
1121                         memset(enic->pp.mac_addr, 0, ETH_ALEN);
1122                         memset(netdev->dev_addr, 0, ETH_ALEN);
1123                 }
1124         }
1125
1126         memset(enic->pp.vf_mac, 0, ETH_ALEN);
1127
1128         return err;
1129 }
1130
1131 static int enic_get_vf_port(struct net_device *netdev, int vf,
1132         struct sk_buff *skb)
1133 {
1134         struct enic *enic = netdev_priv(netdev);
1135         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1136         int err;
1137
1138         if (!(enic->pp.set & ENIC_PORT_REQUEST_APPLIED))
1139                 return -ENODATA;
1140
1141         err = enic_process_get_pp_request(enic, enic->pp.request, &response);
1142         if (err)
1143                 return err;
1144
1145         NLA_PUT_U16(skb, IFLA_PORT_REQUEST, enic->pp.request);
1146         NLA_PUT_U16(skb, IFLA_PORT_RESPONSE, response);
1147         if (enic->pp.set & ENIC_SET_NAME)
1148                 NLA_PUT(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX,
1149                         enic->pp.name);
1150         if (enic->pp.set & ENIC_SET_INSTANCE)
1151                 NLA_PUT(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1152                         enic->pp.instance_uuid);
1153         if (enic->pp.set & ENIC_SET_HOST)
1154                 NLA_PUT(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX,
1155                         enic->pp.host_uuid);
1156
1157         return 0;
1158
1159 nla_put_failure:
1160         return -EMSGSIZE;
1161 }
1162
1163 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1164 {
1165         struct enic *enic = vnic_dev_priv(rq->vdev);
1166
1167         if (!buf->os_buf)
1168                 return;
1169
1170         pci_unmap_single(enic->pdev, buf->dma_addr,
1171                 buf->len, PCI_DMA_FROMDEVICE);
1172         dev_kfree_skb_any(buf->os_buf);
1173 }
1174
1175 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1176 {
1177         struct enic *enic = vnic_dev_priv(rq->vdev);
1178         struct net_device *netdev = enic->netdev;
1179         struct sk_buff *skb;
1180         unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
1181         unsigned int os_buf_index = 0;
1182         dma_addr_t dma_addr;
1183
1184         skb = netdev_alloc_skb_ip_align(netdev, len);
1185         if (!skb)
1186                 return -ENOMEM;
1187
1188         dma_addr = pci_map_single(enic->pdev, skb->data,
1189                 len, PCI_DMA_FROMDEVICE);
1190
1191         enic_queue_rq_desc(rq, skb, os_buf_index,
1192                 dma_addr, len);
1193
1194         return 0;
1195 }
1196
1197 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1198         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1199         int skipped, void *opaque)
1200 {
1201         struct enic *enic = vnic_dev_priv(rq->vdev);
1202         struct net_device *netdev = enic->netdev;
1203         struct sk_buff *skb;
1204
1205         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1206         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1207         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1208         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1209         u8 packet_error;
1210         u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1211         u32 rss_hash;
1212
1213         if (skipped)
1214                 return;
1215
1216         skb = buf->os_buf;
1217         prefetch(skb->data - NET_IP_ALIGN);
1218         pci_unmap_single(enic->pdev, buf->dma_addr,
1219                 buf->len, PCI_DMA_FROMDEVICE);
1220
1221         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1222                 &type, &color, &q_number, &completed_index,
1223                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1224                 &csum_not_calc, &rss_hash, &bytes_written,
1225                 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1226                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1227                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1228                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1229                 &fcs_ok);
1230
1231         if (packet_error) {
1232
1233                 if (!fcs_ok) {
1234                         if (bytes_written > 0)
1235                                 enic->rq_bad_fcs++;
1236                         else if (bytes_written == 0)
1237                                 enic->rq_truncated_pkts++;
1238                 }
1239
1240                 dev_kfree_skb_any(skb);
1241
1242                 return;
1243         }
1244
1245         if (eop && bytes_written > 0) {
1246
1247                 /* Good receive
1248                  */
1249
1250                 skb_put(skb, bytes_written);
1251                 skb->protocol = eth_type_trans(skb, netdev);
1252
1253                 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc) {
1254                         skb->csum = htons(checksum);
1255                         skb->ip_summed = CHECKSUM_COMPLETE;
1256                 }
1257
1258                 skb->dev = netdev;
1259
1260                 if (vlan_stripped)
1261                         __vlan_hwaccel_put_tag(skb, vlan_tci);
1262
1263                 if (netdev->features & NETIF_F_GRO)
1264                         napi_gro_receive(&enic->napi[q_number], skb);
1265                 else
1266                         netif_receive_skb(skb);
1267         } else {
1268
1269                 /* Buffer overflow
1270                  */
1271
1272                 dev_kfree_skb_any(skb);
1273         }
1274 }
1275
1276 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1277         u8 type, u16 q_number, u16 completed_index, void *opaque)
1278 {
1279         struct enic *enic = vnic_dev_priv(vdev);
1280
1281         vnic_rq_service(&enic->rq[q_number], cq_desc,
1282                 completed_index, VNIC_RQ_RETURN_DESC,
1283                 enic_rq_indicate_buf, opaque);
1284
1285         return 0;
1286 }
1287
1288 static int enic_poll(struct napi_struct *napi, int budget)
1289 {
1290         struct net_device *netdev = napi->dev;
1291         struct enic *enic = netdev_priv(netdev);
1292         unsigned int cq_rq = enic_cq_rq(enic, 0);
1293         unsigned int cq_wq = enic_cq_wq(enic, 0);
1294         unsigned int intr = enic_legacy_io_intr();
1295         unsigned int rq_work_to_do = budget;
1296         unsigned int wq_work_to_do = -1; /* no limit */
1297         unsigned int  work_done, rq_work_done, wq_work_done;
1298         int err;
1299
1300         /* Service RQ (first) and WQ
1301          */
1302
1303         rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1304                 rq_work_to_do, enic_rq_service, NULL);
1305
1306         wq_work_done = vnic_cq_service(&enic->cq[cq_wq],
1307                 wq_work_to_do, enic_wq_service, NULL);
1308
1309         /* Accumulate intr event credits for this polling
1310          * cycle.  An intr event is the completion of a
1311          * a WQ or RQ packet.
1312          */
1313
1314         work_done = rq_work_done + wq_work_done;
1315
1316         if (work_done > 0)
1317                 vnic_intr_return_credits(&enic->intr[intr],
1318                         work_done,
1319                         0 /* don't unmask intr */,
1320                         0 /* don't reset intr timer */);
1321
1322         err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1323
1324         /* Buffer allocation failed. Stay in polling
1325          * mode so we can try to fill the ring again.
1326          */
1327
1328         if (err)
1329                 rq_work_done = rq_work_to_do;
1330
1331         if (rq_work_done < rq_work_to_do) {
1332
1333                 /* Some work done, but not enough to stay in polling,
1334                  * exit polling
1335                  */
1336
1337                 napi_complete(napi);
1338                 vnic_intr_unmask(&enic->intr[intr]);
1339         }
1340
1341         return rq_work_done;
1342 }
1343
1344 static int enic_poll_msix(struct napi_struct *napi, int budget)
1345 {
1346         struct net_device *netdev = napi->dev;
1347         struct enic *enic = netdev_priv(netdev);
1348         unsigned int rq = (napi - &enic->napi[0]);
1349         unsigned int cq = enic_cq_rq(enic, rq);
1350         unsigned int intr = enic_msix_rq_intr(enic, rq);
1351         unsigned int work_to_do = budget;
1352         unsigned int work_done;
1353         int err;
1354
1355         /* Service RQ
1356          */
1357
1358         work_done = vnic_cq_service(&enic->cq[cq],
1359                 work_to_do, enic_rq_service, NULL);
1360
1361         /* Return intr event credits for this polling
1362          * cycle.  An intr event is the completion of a
1363          * RQ packet.
1364          */
1365
1366         if (work_done > 0)
1367                 vnic_intr_return_credits(&enic->intr[intr],
1368                         work_done,
1369                         0 /* don't unmask intr */,
1370                         0 /* don't reset intr timer */);
1371
1372         err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1373
1374         /* Buffer allocation failed. Stay in polling mode
1375          * so we can try to fill the ring again.
1376          */
1377
1378         if (err)
1379                 work_done = work_to_do;
1380
1381         if (work_done < work_to_do) {
1382
1383                 /* Some work done, but not enough to stay in polling,
1384                  * exit polling
1385                  */
1386
1387                 napi_complete(napi);
1388                 vnic_intr_unmask(&enic->intr[intr]);
1389         }
1390
1391         return work_done;
1392 }
1393
1394 static void enic_notify_timer(unsigned long data)
1395 {
1396         struct enic *enic = (struct enic *)data;
1397
1398         enic_notify_check(enic);
1399
1400         mod_timer(&enic->notify_timer,
1401                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1402 }
1403
1404 static void enic_free_intr(struct enic *enic)
1405 {
1406         struct net_device *netdev = enic->netdev;
1407         unsigned int i;
1408
1409         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1410         case VNIC_DEV_INTR_MODE_INTX:
1411                 free_irq(enic->pdev->irq, netdev);
1412                 break;
1413         case VNIC_DEV_INTR_MODE_MSI:
1414                 free_irq(enic->pdev->irq, enic);
1415                 break;
1416         case VNIC_DEV_INTR_MODE_MSIX:
1417                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1418                         if (enic->msix[i].requested)
1419                                 free_irq(enic->msix_entry[i].vector,
1420                                         enic->msix[i].devid);
1421                 break;
1422         default:
1423                 break;
1424         }
1425 }
1426
1427 static int enic_request_intr(struct enic *enic)
1428 {
1429         struct net_device *netdev = enic->netdev;
1430         unsigned int i, intr;
1431         int err = 0;
1432
1433         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1434
1435         case VNIC_DEV_INTR_MODE_INTX:
1436
1437                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1438                         IRQF_SHARED, netdev->name, netdev);
1439                 break;
1440
1441         case VNIC_DEV_INTR_MODE_MSI:
1442
1443                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1444                         0, netdev->name, enic);
1445                 break;
1446
1447         case VNIC_DEV_INTR_MODE_MSIX:
1448
1449                 for (i = 0; i < enic->rq_count; i++) {
1450                         intr = enic_msix_rq_intr(enic, i);
1451                         sprintf(enic->msix[intr].devname,
1452                                 "%.11s-rx-%d", netdev->name, i);
1453                         enic->msix[intr].isr = enic_isr_msix_rq;
1454                         enic->msix[intr].devid = &enic->napi[i];
1455                 }
1456
1457                 for (i = 0; i < enic->wq_count; i++) {
1458                         intr = enic_msix_wq_intr(enic, i);
1459                         sprintf(enic->msix[intr].devname,
1460                                 "%.11s-tx-%d", netdev->name, i);
1461                         enic->msix[intr].isr = enic_isr_msix_wq;
1462                         enic->msix[intr].devid = enic;
1463                 }
1464
1465                 intr = enic_msix_err_intr(enic);
1466                 sprintf(enic->msix[intr].devname,
1467                         "%.11s-err", netdev->name);
1468                 enic->msix[intr].isr = enic_isr_msix_err;
1469                 enic->msix[intr].devid = enic;
1470
1471                 intr = enic_msix_notify_intr(enic);
1472                 sprintf(enic->msix[intr].devname,
1473                         "%.11s-notify", netdev->name);
1474                 enic->msix[intr].isr = enic_isr_msix_notify;
1475                 enic->msix[intr].devid = enic;
1476
1477                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1478                         enic->msix[i].requested = 0;
1479
1480                 for (i = 0; i < enic->intr_count; i++) {
1481                         err = request_irq(enic->msix_entry[i].vector,
1482                                 enic->msix[i].isr, 0,
1483                                 enic->msix[i].devname,
1484                                 enic->msix[i].devid);
1485                         if (err) {
1486                                 enic_free_intr(enic);
1487                                 break;
1488                         }
1489                         enic->msix[i].requested = 1;
1490                 }
1491
1492                 break;
1493
1494         default:
1495                 break;
1496         }
1497
1498         return err;
1499 }
1500
1501 static void enic_synchronize_irqs(struct enic *enic)
1502 {
1503         unsigned int i;
1504
1505         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1506         case VNIC_DEV_INTR_MODE_INTX:
1507         case VNIC_DEV_INTR_MODE_MSI:
1508                 synchronize_irq(enic->pdev->irq);
1509                 break;
1510         case VNIC_DEV_INTR_MODE_MSIX:
1511                 for (i = 0; i < enic->intr_count; i++)
1512                         synchronize_irq(enic->msix_entry[i].vector);
1513                 break;
1514         default:
1515                 break;
1516         }
1517 }
1518
1519 static int enic_dev_notify_set(struct enic *enic)
1520 {
1521         int err;
1522
1523         spin_lock(&enic->devcmd_lock);
1524         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1525         case VNIC_DEV_INTR_MODE_INTX:
1526                 err = vnic_dev_notify_set(enic->vdev,
1527                         enic_legacy_notify_intr());
1528                 break;
1529         case VNIC_DEV_INTR_MODE_MSIX:
1530                 err = vnic_dev_notify_set(enic->vdev,
1531                         enic_msix_notify_intr(enic));
1532                 break;
1533         default:
1534                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1535                 break;
1536         }
1537         spin_unlock(&enic->devcmd_lock);
1538
1539         return err;
1540 }
1541
1542 static void enic_notify_timer_start(struct enic *enic)
1543 {
1544         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1545         case VNIC_DEV_INTR_MODE_MSI:
1546                 mod_timer(&enic->notify_timer, jiffies);
1547                 break;
1548         default:
1549                 /* Using intr for notification for INTx/MSI-X */
1550                 break;
1551         }
1552 }
1553
1554 /* rtnl lock is held, process context */
1555 static int enic_open(struct net_device *netdev)
1556 {
1557         struct enic *enic = netdev_priv(netdev);
1558         unsigned int i;
1559         int err;
1560
1561         err = enic_request_intr(enic);
1562         if (err) {
1563                 netdev_err(netdev, "Unable to request irq.\n");
1564                 return err;
1565         }
1566
1567         err = enic_dev_notify_set(enic);
1568         if (err) {
1569                 netdev_err(netdev,
1570                         "Failed to alloc notify buffer, aborting.\n");
1571                 goto err_out_free_intr;
1572         }
1573
1574         for (i = 0; i < enic->rq_count; i++) {
1575                 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1576                 /* Need at least one buffer on ring to get going */
1577                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1578                         netdev_err(netdev, "Unable to alloc receive buffers\n");
1579                         err = -ENOMEM;
1580                         goto err_out_notify_unset;
1581                 }
1582         }
1583
1584         for (i = 0; i < enic->wq_count; i++)
1585                 vnic_wq_enable(&enic->wq[i]);
1586         for (i = 0; i < enic->rq_count; i++)
1587                 vnic_rq_enable(&enic->rq[i]);
1588
1589         if (enic_is_dynamic(enic) && !is_zero_ether_addr(enic->pp.mac_addr))
1590                 enic_dev_add_addr(enic, enic->pp.mac_addr);
1591         else
1592                 enic_dev_add_station_addr(enic);
1593         enic_set_rx_mode(netdev);
1594
1595         netif_wake_queue(netdev);
1596
1597         for (i = 0; i < enic->rq_count; i++)
1598                 napi_enable(&enic->napi[i]);
1599
1600         enic_dev_enable(enic);
1601
1602         for (i = 0; i < enic->intr_count; i++)
1603                 vnic_intr_unmask(&enic->intr[i]);
1604
1605         enic_notify_timer_start(enic);
1606
1607         return 0;
1608
1609 err_out_notify_unset:
1610         enic_dev_notify_unset(enic);
1611 err_out_free_intr:
1612         enic_free_intr(enic);
1613
1614         return err;
1615 }
1616
1617 /* rtnl lock is held, process context */
1618 static int enic_stop(struct net_device *netdev)
1619 {
1620         struct enic *enic = netdev_priv(netdev);
1621         unsigned int i;
1622         int err;
1623
1624         for (i = 0; i < enic->intr_count; i++) {
1625                 vnic_intr_mask(&enic->intr[i]);
1626                 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1627         }
1628
1629         enic_synchronize_irqs(enic);
1630
1631         del_timer_sync(&enic->notify_timer);
1632
1633         enic_dev_disable(enic);
1634
1635         for (i = 0; i < enic->rq_count; i++)
1636                 napi_disable(&enic->napi[i]);
1637
1638         netif_carrier_off(netdev);
1639         netif_tx_disable(netdev);
1640         if (enic_is_dynamic(enic) && !is_zero_ether_addr(enic->pp.mac_addr))
1641                 enic_dev_del_addr(enic, enic->pp.mac_addr);
1642         else
1643                 enic_dev_del_station_addr(enic);
1644
1645         for (i = 0; i < enic->wq_count; i++) {
1646                 err = vnic_wq_disable(&enic->wq[i]);
1647                 if (err)
1648                         return err;
1649         }
1650         for (i = 0; i < enic->rq_count; i++) {
1651                 err = vnic_rq_disable(&enic->rq[i]);
1652                 if (err)
1653                         return err;
1654         }
1655
1656         enic_dev_notify_unset(enic);
1657         enic_free_intr(enic);
1658
1659         for (i = 0; i < enic->wq_count; i++)
1660                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1661         for (i = 0; i < enic->rq_count; i++)
1662                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1663         for (i = 0; i < enic->cq_count; i++)
1664                 vnic_cq_clean(&enic->cq[i]);
1665         for (i = 0; i < enic->intr_count; i++)
1666                 vnic_intr_clean(&enic->intr[i]);
1667
1668         return 0;
1669 }
1670
1671 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1672 {
1673         struct enic *enic = netdev_priv(netdev);
1674         int running = netif_running(netdev);
1675
1676         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1677                 return -EINVAL;
1678
1679         if (enic_is_dynamic(enic))
1680                 return -EOPNOTSUPP;
1681
1682         if (running)
1683                 enic_stop(netdev);
1684
1685         netdev->mtu = new_mtu;
1686
1687         if (netdev->mtu > enic->port_mtu)
1688                 netdev_warn(netdev,
1689                         "interface MTU (%d) set higher than port MTU (%d)\n",
1690                         netdev->mtu, enic->port_mtu);
1691
1692         if (running)
1693                 enic_open(netdev);
1694
1695         return 0;
1696 }
1697
1698 static void enic_change_mtu_work(struct work_struct *work)
1699 {
1700         struct enic *enic = container_of(work, struct enic, change_mtu_work);
1701         struct net_device *netdev = enic->netdev;
1702         int new_mtu = vnic_dev_mtu(enic->vdev);
1703         int err;
1704         unsigned int i;
1705
1706         new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
1707
1708         rtnl_lock();
1709
1710         /* Stop RQ */
1711         del_timer_sync(&enic->notify_timer);
1712
1713         for (i = 0; i < enic->rq_count; i++)
1714                 napi_disable(&enic->napi[i]);
1715
1716         vnic_intr_mask(&enic->intr[0]);
1717         enic_synchronize_irqs(enic);
1718         err = vnic_rq_disable(&enic->rq[0]);
1719         if (err) {
1720                 netdev_err(netdev, "Unable to disable RQ.\n");
1721                 return;
1722         }
1723         vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
1724         vnic_cq_clean(&enic->cq[0]);
1725         vnic_intr_clean(&enic->intr[0]);
1726
1727         /* Fill RQ with new_mtu-sized buffers */
1728         netdev->mtu = new_mtu;
1729         vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1730         /* Need at least one buffer on ring to get going */
1731         if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
1732                 netdev_err(netdev, "Unable to alloc receive buffers.\n");
1733                 return;
1734         }
1735
1736         /* Start RQ */
1737         vnic_rq_enable(&enic->rq[0]);
1738         napi_enable(&enic->napi[0]);
1739         vnic_intr_unmask(&enic->intr[0]);
1740         enic_notify_timer_start(enic);
1741
1742         rtnl_unlock();
1743
1744         netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1745 }
1746
1747 #ifdef CONFIG_NET_POLL_CONTROLLER
1748 static void enic_poll_controller(struct net_device *netdev)
1749 {
1750         struct enic *enic = netdev_priv(netdev);
1751         struct vnic_dev *vdev = enic->vdev;
1752         unsigned int i, intr;
1753
1754         switch (vnic_dev_get_intr_mode(vdev)) {
1755         case VNIC_DEV_INTR_MODE_MSIX:
1756                 for (i = 0; i < enic->rq_count; i++) {
1757                         intr = enic_msix_rq_intr(enic, i);
1758                         enic_isr_msix_rq(enic->msix_entry[intr].vector,
1759                                 &enic->napi[i]);
1760                 }
1761
1762                 for (i = 0; i < enic->wq_count; i++) {
1763                         intr = enic_msix_wq_intr(enic, i);
1764                         enic_isr_msix_wq(enic->msix_entry[intr].vector, enic);
1765                 }
1766
1767                 break;
1768         case VNIC_DEV_INTR_MODE_MSI:
1769                 enic_isr_msi(enic->pdev->irq, enic);
1770                 break;
1771         case VNIC_DEV_INTR_MODE_INTX:
1772                 enic_isr_legacy(enic->pdev->irq, netdev);
1773                 break;
1774         default:
1775                 break;
1776         }
1777 }
1778 #endif
1779
1780 static int enic_dev_wait(struct vnic_dev *vdev,
1781         int (*start)(struct vnic_dev *, int),
1782         int (*finished)(struct vnic_dev *, int *),
1783         int arg)
1784 {
1785         unsigned long time;
1786         int done;
1787         int err;
1788
1789         BUG_ON(in_interrupt());
1790
1791         err = start(vdev, arg);
1792         if (err)
1793                 return err;
1794
1795         /* Wait for func to complete...2 seconds max
1796          */
1797
1798         time = jiffies + (HZ * 2);
1799         do {
1800
1801                 err = finished(vdev, &done);
1802                 if (err)
1803                         return err;
1804
1805                 if (done)
1806                         return 0;
1807
1808                 schedule_timeout_uninterruptible(HZ / 10);
1809
1810         } while (time_after(time, jiffies));
1811
1812         return -ETIMEDOUT;
1813 }
1814
1815 static int enic_dev_open(struct enic *enic)
1816 {
1817         int err;
1818
1819         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1820                 vnic_dev_open_done, 0);
1821         if (err)
1822                 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1823                         err);
1824
1825         return err;
1826 }
1827
1828 static int enic_dev_hang_reset(struct enic *enic)
1829 {
1830         int err;
1831
1832         err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1833                 vnic_dev_hang_reset_done, 0);
1834         if (err)
1835                 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1836                         err);
1837
1838         return err;
1839 }
1840
1841 static int enic_set_rsskey(struct enic *enic)
1842 {
1843         dma_addr_t rss_key_buf_pa;
1844         union vnic_rss_key *rss_key_buf_va = NULL;
1845         union vnic_rss_key rss_key = {
1846                 .key[0].b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101},
1847                 .key[1].b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101},
1848                 .key[2].b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115},
1849                 .key[3].b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108},
1850         };
1851         int err;
1852
1853         rss_key_buf_va = pci_alloc_consistent(enic->pdev,
1854                 sizeof(union vnic_rss_key), &rss_key_buf_pa);
1855         if (!rss_key_buf_va)
1856                 return -ENOMEM;
1857
1858         memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
1859
1860         spin_lock(&enic->devcmd_lock);
1861         err = enic_set_rss_key(enic,
1862                 rss_key_buf_pa,
1863                 sizeof(union vnic_rss_key));
1864         spin_unlock(&enic->devcmd_lock);
1865
1866         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
1867                 rss_key_buf_va, rss_key_buf_pa);
1868
1869         return err;
1870 }
1871
1872 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
1873 {
1874         dma_addr_t rss_cpu_buf_pa;
1875         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
1876         unsigned int i;
1877         int err;
1878
1879         rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
1880                 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
1881         if (!rss_cpu_buf_va)
1882                 return -ENOMEM;
1883
1884         for (i = 0; i < (1 << rss_hash_bits); i++)
1885                 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
1886
1887         spin_lock(&enic->devcmd_lock);
1888         err = enic_set_rss_cpu(enic,
1889                 rss_cpu_buf_pa,
1890                 sizeof(union vnic_rss_cpu));
1891         spin_unlock(&enic->devcmd_lock);
1892
1893         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
1894                 rss_cpu_buf_va, rss_cpu_buf_pa);
1895
1896         return err;
1897 }
1898
1899 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
1900         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
1901 {
1902         const u8 tso_ipid_split_en = 0;
1903         const u8 ig_vlan_strip_en = 1;
1904         int err;
1905
1906         /* Enable VLAN tag stripping.
1907         */
1908
1909         spin_lock(&enic->devcmd_lock);
1910         err = enic_set_nic_cfg(enic,
1911                 rss_default_cpu, rss_hash_type,
1912                 rss_hash_bits, rss_base_cpu,
1913                 rss_enable, tso_ipid_split_en,
1914                 ig_vlan_strip_en);
1915         spin_unlock(&enic->devcmd_lock);
1916
1917         return err;
1918 }
1919
1920 static int enic_set_rss_nic_cfg(struct enic *enic)
1921 {
1922         struct device *dev = enic_get_dev(enic);
1923         const u8 rss_default_cpu = 0;
1924         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1925                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1926                 NIC_CFG_RSS_HASH_TYPE_IPV6 |
1927                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1928         const u8 rss_hash_bits = 7;
1929         const u8 rss_base_cpu = 0;
1930         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1931
1932         if (rss_enable) {
1933                 if (!enic_set_rsskey(enic)) {
1934                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
1935                                 rss_enable = 0;
1936                                 dev_warn(dev, "RSS disabled, "
1937                                         "Failed to set RSS cpu indirection table.");
1938                         }
1939                 } else {
1940                         rss_enable = 0;
1941                         dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
1942                 }
1943         }
1944
1945         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1946                 rss_hash_bits, rss_base_cpu, rss_enable);
1947 }
1948
1949 static void enic_reset(struct work_struct *work)
1950 {
1951         struct enic *enic = container_of(work, struct enic, reset);
1952
1953         if (!netif_running(enic->netdev))
1954                 return;
1955
1956         rtnl_lock();
1957
1958         enic_dev_hang_notify(enic);
1959         enic_stop(enic->netdev);
1960         enic_dev_hang_reset(enic);
1961         enic_reset_addr_lists(enic);
1962         enic_init_vnic_resources(enic);
1963         enic_set_rss_nic_cfg(enic);
1964         enic_dev_set_ig_vlan_rewrite_mode(enic);
1965         enic_open(enic->netdev);
1966
1967         rtnl_unlock();
1968 }
1969
1970 static int enic_set_intr_mode(struct enic *enic)
1971 {
1972         unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
1973         unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
1974         unsigned int i;
1975
1976         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1977          * on system capabilities.
1978          *
1979          * Try MSI-X first
1980          *
1981          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1982          * (the second to last INTR is used for WQ/RQ errors)
1983          * (the last INTR is used for notifications)
1984          */
1985
1986         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1987         for (i = 0; i < n + m + 2; i++)
1988                 enic->msix_entry[i].entry = i;
1989
1990         /* Use multiple RQs if RSS is enabled
1991          */
1992
1993         if (ENIC_SETTING(enic, RSS) &&
1994             enic->config.intr_mode < 1 &&
1995             enic->rq_count >= n &&
1996             enic->wq_count >= m &&
1997             enic->cq_count >= n + m &&
1998             enic->intr_count >= n + m + 2) {
1999
2000                 if (!pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
2001
2002                         enic->rq_count = n;
2003                         enic->wq_count = m;
2004                         enic->cq_count = n + m;
2005                         enic->intr_count = n + m + 2;
2006
2007                         vnic_dev_set_intr_mode(enic->vdev,
2008                                 VNIC_DEV_INTR_MODE_MSIX);
2009
2010                         return 0;
2011                 }
2012         }
2013
2014         if (enic->config.intr_mode < 1 &&
2015             enic->rq_count >= 1 &&
2016             enic->wq_count >= m &&
2017             enic->cq_count >= 1 + m &&
2018             enic->intr_count >= 1 + m + 2) {
2019                 if (!pci_enable_msix(enic->pdev, enic->msix_entry, 1 + m + 2)) {
2020
2021                         enic->rq_count = 1;
2022                         enic->wq_count = m;
2023                         enic->cq_count = 1 + m;
2024                         enic->intr_count = 1 + m + 2;
2025
2026                         vnic_dev_set_intr_mode(enic->vdev,
2027                                 VNIC_DEV_INTR_MODE_MSIX);
2028
2029                         return 0;
2030                 }
2031         }
2032
2033         /* Next try MSI
2034          *
2035          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2036          */
2037
2038         if (enic->config.intr_mode < 2 &&
2039             enic->rq_count >= 1 &&
2040             enic->wq_count >= 1 &&
2041             enic->cq_count >= 2 &&
2042             enic->intr_count >= 1 &&
2043             !pci_enable_msi(enic->pdev)) {
2044
2045                 enic->rq_count = 1;
2046                 enic->wq_count = 1;
2047                 enic->cq_count = 2;
2048                 enic->intr_count = 1;
2049
2050                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2051
2052                 return 0;
2053         }
2054
2055         /* Next try INTx
2056          *
2057          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2058          * (the first INTR is used for WQ/RQ)
2059          * (the second INTR is used for WQ/RQ errors)
2060          * (the last INTR is used for notifications)
2061          */
2062
2063         if (enic->config.intr_mode < 3 &&
2064             enic->rq_count >= 1 &&
2065             enic->wq_count >= 1 &&
2066             enic->cq_count >= 2 &&
2067             enic->intr_count >= 3) {
2068
2069                 enic->rq_count = 1;
2070                 enic->wq_count = 1;
2071                 enic->cq_count = 2;
2072                 enic->intr_count = 3;
2073
2074                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2075
2076                 return 0;
2077         }
2078
2079         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2080
2081         return -EINVAL;
2082 }
2083
2084 static void enic_clear_intr_mode(struct enic *enic)
2085 {
2086         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2087         case VNIC_DEV_INTR_MODE_MSIX:
2088                 pci_disable_msix(enic->pdev);
2089                 break;
2090         case VNIC_DEV_INTR_MODE_MSI:
2091                 pci_disable_msi(enic->pdev);
2092                 break;
2093         default:
2094                 break;
2095         }
2096
2097         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2098 }
2099
2100 static const struct net_device_ops enic_netdev_dynamic_ops = {
2101         .ndo_open               = enic_open,
2102         .ndo_stop               = enic_stop,
2103         .ndo_start_xmit         = enic_hard_start_xmit,
2104         .ndo_get_stats64        = enic_get_stats,
2105         .ndo_validate_addr      = eth_validate_addr,
2106         .ndo_set_rx_mode        = enic_set_rx_mode,
2107         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2108         .ndo_change_mtu         = enic_change_mtu,
2109         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2110         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2111         .ndo_tx_timeout         = enic_tx_timeout,
2112         .ndo_set_vf_port        = enic_set_vf_port,
2113         .ndo_get_vf_port        = enic_get_vf_port,
2114         .ndo_set_vf_mac         = enic_set_vf_mac,
2115 #ifdef CONFIG_NET_POLL_CONTROLLER
2116         .ndo_poll_controller    = enic_poll_controller,
2117 #endif
2118 };
2119
2120 static const struct net_device_ops enic_netdev_ops = {
2121         .ndo_open               = enic_open,
2122         .ndo_stop               = enic_stop,
2123         .ndo_start_xmit         = enic_hard_start_xmit,
2124         .ndo_get_stats64        = enic_get_stats,
2125         .ndo_validate_addr      = eth_validate_addr,
2126         .ndo_set_mac_address    = enic_set_mac_address,
2127         .ndo_set_rx_mode        = enic_set_rx_mode,
2128         .ndo_change_mtu         = enic_change_mtu,
2129         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2130         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2131         .ndo_tx_timeout         = enic_tx_timeout,
2132 #ifdef CONFIG_NET_POLL_CONTROLLER
2133         .ndo_poll_controller    = enic_poll_controller,
2134 #endif
2135 };
2136
2137 static void enic_dev_deinit(struct enic *enic)
2138 {
2139         unsigned int i;
2140
2141         for (i = 0; i < enic->rq_count; i++)
2142                 netif_napi_del(&enic->napi[i]);
2143
2144         enic_free_vnic_resources(enic);
2145         enic_clear_intr_mode(enic);
2146 }
2147
2148 static int enic_dev_init(struct enic *enic)
2149 {
2150         struct device *dev = enic_get_dev(enic);
2151         struct net_device *netdev = enic->netdev;
2152         unsigned int i;
2153         int err;
2154
2155         /* Get interrupt coalesce timer info */
2156         err = enic_dev_intr_coal_timer_info(enic);
2157         if (err) {
2158                 dev_warn(dev, "Using default conversion factor for "
2159                         "interrupt coalesce timer\n");
2160                 vnic_dev_intr_coal_timer_info_default(enic->vdev);
2161         }
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_rss_nic_cfg(enic);
2200         if (err) {
2201                 dev_err(dev, "Failed to config nic, aborting\n");
2202                 goto err_out_free_vnic_resources;
2203         }
2204
2205         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2206         default:
2207                 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2208                 break;
2209         case VNIC_DEV_INTR_MODE_MSIX:
2210                 for (i = 0; i < enic->rq_count; i++)
2211                         netif_napi_add(netdev, &enic->napi[i],
2212                                 enic_poll_msix, 64);
2213                 break;
2214         }
2215
2216         return 0;
2217
2218 err_out_free_vnic_resources:
2219         enic_clear_intr_mode(enic);
2220         enic_free_vnic_resources(enic);
2221
2222         return err;
2223 }
2224
2225 static void enic_iounmap(struct enic *enic)
2226 {
2227         unsigned int i;
2228
2229         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2230                 if (enic->bar[i].vaddr)
2231                         iounmap(enic->bar[i].vaddr);
2232 }
2233
2234 static int __devinit enic_probe(struct pci_dev *pdev,
2235         const struct pci_device_id *ent)
2236 {
2237         struct device *dev = &pdev->dev;
2238         struct net_device *netdev;
2239         struct enic *enic;
2240         int using_dac = 0;
2241         unsigned int i;
2242         int err;
2243
2244         /* Allocate net device structure and initialize.  Private
2245          * instance data is initialized to zero.
2246          */
2247
2248         netdev = alloc_etherdev(sizeof(struct enic));
2249         if (!netdev) {
2250                 pr_err("Etherdev alloc failed, aborting\n");
2251                 return -ENOMEM;
2252         }
2253
2254         pci_set_drvdata(pdev, netdev);
2255
2256         SET_NETDEV_DEV(netdev, &pdev->dev);
2257
2258         enic = netdev_priv(netdev);
2259         enic->netdev = netdev;
2260         enic->pdev = pdev;
2261
2262         /* Setup PCI resources
2263          */
2264
2265         err = pci_enable_device_mem(pdev);
2266         if (err) {
2267                 dev_err(dev, "Cannot enable PCI device, aborting\n");
2268                 goto err_out_free_netdev;
2269         }
2270
2271         err = pci_request_regions(pdev, DRV_NAME);
2272         if (err) {
2273                 dev_err(dev, "Cannot request PCI regions, aborting\n");
2274                 goto err_out_disable_device;
2275         }
2276
2277         pci_set_master(pdev);
2278
2279         /* Query PCI controller on system for DMA addressing
2280          * limitation for the device.  Try 40-bit first, and
2281          * fail to 32-bit.
2282          */
2283
2284         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
2285         if (err) {
2286                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2287                 if (err) {
2288                         dev_err(dev, "No usable DMA configuration, aborting\n");
2289                         goto err_out_release_regions;
2290                 }
2291                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2292                 if (err) {
2293                         dev_err(dev, "Unable to obtain %u-bit DMA "
2294                                 "for consistent allocations, aborting\n", 32);
2295                         goto err_out_release_regions;
2296                 }
2297         } else {
2298                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
2299                 if (err) {
2300                         dev_err(dev, "Unable to obtain %u-bit DMA "
2301                                 "for consistent allocations, aborting\n", 40);
2302                         goto err_out_release_regions;
2303                 }
2304                 using_dac = 1;
2305         }
2306
2307         /* Map vNIC resources from BAR0-5
2308          */
2309
2310         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2311                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2312                         continue;
2313                 enic->bar[i].len = pci_resource_len(pdev, i);
2314                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2315                 if (!enic->bar[i].vaddr) {
2316                         dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2317                         err = -ENODEV;
2318                         goto err_out_iounmap;
2319                 }
2320                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2321         }
2322
2323         /* Register vNIC device
2324          */
2325
2326         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2327                 ARRAY_SIZE(enic->bar));
2328         if (!enic->vdev) {
2329                 dev_err(dev, "vNIC registration failed, aborting\n");
2330                 err = -ENODEV;
2331                 goto err_out_iounmap;
2332         }
2333
2334         /* Issue device open to get device in known state
2335          */
2336
2337         err = enic_dev_open(enic);
2338         if (err) {
2339                 dev_err(dev, "vNIC dev open failed, aborting\n");
2340                 goto err_out_vnic_unregister;
2341         }
2342
2343         /* Setup devcmd lock
2344          */
2345
2346         spin_lock_init(&enic->devcmd_lock);
2347
2348         /*
2349          * Set ingress vlan rewrite mode before vnic initialization
2350          */
2351
2352         err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2353         if (err) {
2354                 dev_err(dev,
2355                         "Failed to set ingress vlan rewrite mode, aborting.\n");
2356                 goto err_out_dev_close;
2357         }
2358
2359         /* Issue device init to initialize the vnic-to-switch link.
2360          * We'll start with carrier off and wait for link UP
2361          * notification later to turn on carrier.  We don't need
2362          * to wait here for the vnic-to-switch link initialization
2363          * to complete; link UP notification is the indication that
2364          * the process is complete.
2365          */
2366
2367         netif_carrier_off(netdev);
2368
2369         /* Do not call dev_init for a dynamic vnic.
2370          * For a dynamic vnic, init_prov_info will be
2371          * called later by an upper layer.
2372          */
2373
2374         if (!enic_is_dynamic(enic)) {
2375                 err = vnic_dev_init(enic->vdev, 0);
2376                 if (err) {
2377                         dev_err(dev, "vNIC dev init failed, aborting\n");
2378                         goto err_out_dev_close;
2379                 }
2380         }
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         INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2397
2398         for (i = 0; i < enic->wq_count; i++)
2399                 spin_lock_init(&enic->wq_lock[i]);
2400
2401         /* Register net device
2402          */
2403
2404         enic->port_mtu = enic->config.mtu;
2405         (void)enic_change_mtu(netdev, enic->port_mtu);
2406
2407         err = enic_set_mac_addr(netdev, enic->mac_addr);
2408         if (err) {
2409                 dev_err(dev, "Invalid MAC address, aborting\n");
2410                 goto err_out_dev_deinit;
2411         }
2412
2413         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2414         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2415
2416         if (enic_is_dynamic(enic))
2417                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2418         else
2419                 netdev->netdev_ops = &enic_netdev_ops;
2420
2421         netdev->watchdog_timeo = 2 * HZ;
2422         netdev->ethtool_ops = &enic_ethtool_ops;
2423
2424         netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
2425         if (ENIC_SETTING(enic, LOOP)) {
2426                 netdev->features &= ~NETIF_F_HW_VLAN_TX;
2427                 enic->loop_enable = 1;
2428                 enic->loop_tag = enic->config.loop_tag;
2429                 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2430         }
2431         if (ENIC_SETTING(enic, TXCSUM))
2432                 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2433         if (ENIC_SETTING(enic, TSO))
2434                 netdev->hw_features |= NETIF_F_TSO |
2435                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2436         if (ENIC_SETTING(enic, RXCSUM))
2437                 netdev->hw_features |= NETIF_F_RXCSUM;
2438
2439         netdev->features |= netdev->hw_features;
2440
2441         if (using_dac)
2442                 netdev->features |= NETIF_F_HIGHDMA;
2443
2444         netdev->priv_flags |= IFF_UNICAST_FLT;
2445
2446         err = register_netdev(netdev);
2447         if (err) {
2448                 dev_err(dev, "Cannot register net device, aborting\n");
2449                 goto err_out_dev_deinit;
2450         }
2451
2452         return 0;
2453
2454 err_out_dev_deinit:
2455         enic_dev_deinit(enic);
2456 err_out_dev_close:
2457         vnic_dev_close(enic->vdev);
2458 err_out_vnic_unregister:
2459         vnic_dev_unregister(enic->vdev);
2460 err_out_iounmap:
2461         enic_iounmap(enic);
2462 err_out_release_regions:
2463         pci_release_regions(pdev);
2464 err_out_disable_device:
2465         pci_disable_device(pdev);
2466 err_out_free_netdev:
2467         pci_set_drvdata(pdev, NULL);
2468         free_netdev(netdev);
2469
2470         return err;
2471 }
2472
2473 static void __devexit enic_remove(struct pci_dev *pdev)
2474 {
2475         struct net_device *netdev = pci_get_drvdata(pdev);
2476
2477         if (netdev) {
2478                 struct enic *enic = netdev_priv(netdev);
2479
2480                 cancel_work_sync(&enic->reset);
2481                 cancel_work_sync(&enic->change_mtu_work);
2482                 unregister_netdev(netdev);
2483                 enic_dev_deinit(enic);
2484                 vnic_dev_close(enic->vdev);
2485                 vnic_dev_unregister(enic->vdev);
2486                 enic_iounmap(enic);
2487                 pci_release_regions(pdev);
2488                 pci_disable_device(pdev);
2489                 pci_set_drvdata(pdev, NULL);
2490                 free_netdev(netdev);
2491         }
2492 }
2493
2494 static struct pci_driver enic_driver = {
2495         .name = DRV_NAME,
2496         .id_table = enic_id_table,
2497         .probe = enic_probe,
2498         .remove = __devexit_p(enic_remove),
2499 };
2500
2501 static int __init enic_init_module(void)
2502 {
2503         pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2504
2505         return pci_register_driver(&enic_driver);
2506 }
2507
2508 static void __exit enic_cleanup_module(void)
2509 {
2510         pci_unregister_driver(&enic_driver);
2511 }
2512
2513 module_init(enic_init_module);
2514 module_exit(enic_cleanup_module);