projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
driver/net: enic: record q_number and rss_hash for skb
[deliverable/linux.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/in.h>
35 #include <linux/ip.h>
36 #include <linux/ipv6.h>
37 #include <linux/tcp.h>
38 #include <linux/rtnetlink.h>
39 #include <linux/prefetch.h>
40 #include <net/ip6_checksum.h>
41
42 #include "cq_enet_desc.h"
43 #include "vnic_dev.h"
44 #include "vnic_intr.h"
45 #include "vnic_stats.h"
46 #include "vnic_vic.h"
47 #include "enic_res.h"
48 #include "enic.h"
49 #include "enic_dev.h"
50 #include "enic_pp.h"
51
52 #define ENIC_NOTIFY_TIMER_PERIOD (2 * HZ)
53 #define WQ_ENET_MAX_DESC_LEN (1 << WQ_ENET_LEN_BITS)
54 #define MAX_TSO (1 << 16)
55 #define ENIC_DESC_MAX_SPLITS (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
56
57 #define PCI_DEVICE_ID_CISCO_VIC_ENET 0x0043 /* ethernet vnic */
58 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN 0x0044 /* enet dynamic vnic */
59 #define PCI_DEVICE_ID_CISCO_VIC_ENET_VF 0x0071 /* enet SRIOV VF */
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 { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
66 { 0, } /* end of table */
67 };
68
69 MODULE_DESCRIPTION(DRV_DESCRIPTION);
70 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
71 MODULE_LICENSE("GPL");
72 MODULE_VERSION(DRV_VERSION);
73 MODULE_DEVICE_TABLE(pci, enic_id_table);
74
75 int enic_is_dynamic(struct enic *enic)
76 {
77 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
78 }
79
80 int enic_sriov_enabled(struct enic *enic)
81 {
82 return (enic->priv_flags & ENIC_SRIOV_ENABLED) ? 1 : 0;
83 }
84
85 static int enic_is_sriov_vf(struct enic *enic)
86 {
87 return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
88 }
89
90 int enic_is_valid_vf(struct enic *enic, int vf)
91 {
92 #ifdef CONFIG_PCI_IOV
93 return vf >= 0 && vf < enic->num_vfs;
94 #else
95 return 0;
96 #endif
97 }
98
99 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
100 {
101 struct enic *enic = vnic_dev_priv(wq->vdev);
102
103 if (buf->sop)
104 pci_unmap_single(enic->pdev, buf->dma_addr,
105 buf->len, PCI_DMA_TODEVICE);
106 else
107 pci_unmap_page(enic->pdev, buf->dma_addr,
108 buf->len, PCI_DMA_TODEVICE);
109
110 if (buf->os_buf)
111 dev_kfree_skb_any(buf->os_buf);
112 }
113
114 static void enic_wq_free_buf(struct vnic_wq *wq,
115 struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
116 {
117 enic_free_wq_buf(wq, buf);
118 }
119
120 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
121 u8 type, u16 q_number, u16 completed_index, void *opaque)
122 {
123 struct enic *enic = vnic_dev_priv(vdev);
124
125 spin_lock(&enic->wq_lock[q_number]);
126
127 vnic_wq_service(&enic->wq[q_number], cq_desc,
128 completed_index, enic_wq_free_buf,
129 opaque);
130
131 if (netif_tx_queue_stopped(netdev_get_tx_queue(enic->netdev, q_number)) &&
132 vnic_wq_desc_avail(&enic->wq[q_number]) >=
133 (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
134 netif_wake_subqueue(enic->netdev, q_number);
135
136 spin_unlock(&enic->wq_lock[q_number]);
137
138 return 0;
139 }
140
141 static void enic_log_q_error(struct enic *enic)
142 {
143 unsigned int i;
144 u32 error_status;
145
146 for (i = 0; i < enic->wq_count; i++) {
147 error_status = vnic_wq_error_status(&enic->wq[i]);
148 if (error_status)
149 netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
150 i, error_status);
151 }
152
153 for (i = 0; i < enic->rq_count; i++) {
154 error_status = vnic_rq_error_status(&enic->rq[i]);
155 if (error_status)
156 netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
157 i, error_status);
158 }
159 }
160
161 static void enic_msglvl_check(struct enic *enic)
162 {
163 u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
164
165 if (msg_enable != enic->msg_enable) {
166 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
167 enic->msg_enable, msg_enable);
168 enic->msg_enable = msg_enable;
169 }
170 }
171
172 static void enic_mtu_check(struct enic *enic)
173 {
174 u32 mtu = vnic_dev_mtu(enic->vdev);
175 struct net_device *netdev = enic->netdev;
176
177 if (mtu && mtu != enic->port_mtu) {
178 enic->port_mtu = mtu;
179 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
180 mtu = max_t(int, ENIC_MIN_MTU,
181 min_t(int, ENIC_MAX_MTU, mtu));
182 if (mtu != netdev->mtu)
183 schedule_work(&enic->change_mtu_work);
184 } else {
185 if (mtu < netdev->mtu)
186 netdev_warn(netdev,
187 "interface MTU (%d) set higher "
188 "than switch port MTU (%d)\n",
189 netdev->mtu, mtu);
190 }
191 }
192 }
193
194 static void enic_link_check(struct enic *enic)
195 {
196 int link_status = vnic_dev_link_status(enic->vdev);
197 int carrier_ok = netif_carrier_ok(enic->netdev);
198
199 if (link_status && !carrier_ok) {
200 netdev_info(enic->netdev, "Link UP\n");
201 netif_carrier_on(enic->netdev);
202 } else if (!link_status && carrier_ok) {
203 netdev_info(enic->netdev, "Link DOWN\n");
204 netif_carrier_off(enic->netdev);
205 }
206 }
207
208 static void enic_notify_check(struct enic *enic)
209 {
210 enic_msglvl_check(enic);
211 enic_mtu_check(enic);
212 enic_link_check(enic);
213 }
214
215 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
216
217 static irqreturn_t enic_isr_legacy(int irq, void *data)
218 {
219 struct net_device *netdev = data;
220 struct enic *enic = netdev_priv(netdev);
221 unsigned int io_intr = enic_legacy_io_intr();
222 unsigned int err_intr = enic_legacy_err_intr();
223 unsigned int notify_intr = enic_legacy_notify_intr();
224 u32 pba;
225
226 vnic_intr_mask(&enic->intr[io_intr]);
227
228 pba = vnic_intr_legacy_pba(enic->legacy_pba);
229 if (!pba) {
230 vnic_intr_unmask(&enic->intr[io_intr]);
231 return IRQ_NONE; /* not our interrupt */
232 }
233
234 if (ENIC_TEST_INTR(pba, notify_intr)) {
235 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
236 enic_notify_check(enic);
237 }
238
239 if (ENIC_TEST_INTR(pba, err_intr)) {
240 vnic_intr_return_all_credits(&enic->intr[err_intr]);
241 enic_log_q_error(enic);
242 /* schedule recovery from WQ/RQ error */
243 schedule_work(&enic->reset);
244 return IRQ_HANDLED;
245 }
246
247 if (ENIC_TEST_INTR(pba, io_intr)) {
248 if (napi_schedule_prep(&enic->napi[0]))
249 __napi_schedule(&enic->napi[0]);
250 } else {
251 vnic_intr_unmask(&enic->intr[io_intr]);
252 }
253
254 return IRQ_HANDLED;
255 }
256
257 static irqreturn_t enic_isr_msi(int irq, void *data)
258 {
259 struct enic *enic = data;
260
261 /* With MSI, there is no sharing of interrupts, so this is
262 * our interrupt and there is no need to ack it. The device
263 * is not providing per-vector masking, so the OS will not
264 * write to PCI config space to mask/unmask the interrupt.
265 * We're using mask_on_assertion for MSI, so the device
266 * automatically masks the interrupt when the interrupt is
267 * generated. Later, when exiting polling, the interrupt
268 * will be unmasked (see enic_poll).
269 *
270 * Also, the device uses the same PCIe Traffic Class (TC)
271 * for Memory Write data and MSI, so there are no ordering
272 * issues; the MSI will always arrive at the Root Complex
273 * _after_ corresponding Memory Writes (i.e. descriptor
274 * writes).
275 */
276
277 napi_schedule(&enic->napi[0]);
278
279 return IRQ_HANDLED;
280 }
281
282 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
283 {
284 struct napi_struct *napi = data;
285
286 /* schedule NAPI polling for RQ cleanup */
287 napi_schedule(napi);
288
289 return IRQ_HANDLED;
290 }
291
292 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
293 {
294 struct enic *enic = data;
295 unsigned int cq;
296 unsigned int intr;
297 unsigned int wq_work_to_do = -1; /* no limit */
298 unsigned int wq_work_done;
299 unsigned int wq_irq;
300
301 wq_irq = (u32)irq - enic->msix_entry[enic_msix_wq_intr(enic, 0)].vector;
302 cq = enic_cq_wq(enic, wq_irq);
303 intr = enic_msix_wq_intr(enic, wq_irq);
304
305 wq_work_done = vnic_cq_service(&enic->cq[cq],
306 wq_work_to_do, enic_wq_service, NULL);
307
308 vnic_intr_return_credits(&enic->intr[intr],
309 wq_work_done,
310 1 /* unmask intr */,
311 1 /* reset intr timer */);
312
313 return IRQ_HANDLED;
314 }
315
316 static irqreturn_t enic_isr_msix_err(int irq, void *data)
317 {
318 struct enic *enic = data;
319 unsigned int intr = enic_msix_err_intr(enic);
320
321 vnic_intr_return_all_credits(&enic->intr[intr]);
322
323 enic_log_q_error(enic);
324
325 /* schedule recovery from WQ/RQ error */
326 schedule_work(&enic->reset);
327
328 return IRQ_HANDLED;
329 }
330
331 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
332 {
333 struct enic *enic = data;
334 unsigned int intr = enic_msix_notify_intr(enic);
335
336 vnic_intr_return_all_credits(&enic->intr[intr]);
337 enic_notify_check(enic);
338
339 return IRQ_HANDLED;
340 }
341
342 static inline void enic_queue_wq_skb_cont(struct enic *enic,
343 struct vnic_wq *wq, struct sk_buff *skb,
344 unsigned int len_left, int loopback)
345 {
346 const skb_frag_t *frag;
347
348 /* Queue additional data fragments */
349 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
350 len_left -= skb_frag_size(frag);
351 enic_queue_wq_desc_cont(wq, skb,
352 skb_frag_dma_map(&enic->pdev->dev,
353 frag, 0, skb_frag_size(frag),
354 DMA_TO_DEVICE),
355 skb_frag_size(frag),
356 (len_left == 0), /* EOP? */
357 loopback);
358 }
359 }
360
361 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
362 struct vnic_wq *wq, struct sk_buff *skb,
363 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
364 {
365 unsigned int head_len = skb_headlen(skb);
366 unsigned int len_left = skb->len - head_len;
367 int eop = (len_left == 0);
368
369 /* Queue the main skb fragment. The fragments are no larger
370 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
371 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
372 * per fragment is queued.
373 */
374 enic_queue_wq_desc(wq, skb,
375 pci_map_single(enic->pdev, skb->data,
376 head_len, PCI_DMA_TODEVICE),
377 head_len,
378 vlan_tag_insert, vlan_tag,
379 eop, loopback);
380
381 if (!eop)
382 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
383 }
384
385 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
386 struct vnic_wq *wq, struct sk_buff *skb,
387 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
388 {
389 unsigned int head_len = skb_headlen(skb);
390 unsigned int len_left = skb->len - head_len;
391 unsigned int hdr_len = skb_checksum_start_offset(skb);
392 unsigned int csum_offset = hdr_len + skb->csum_offset;
393 int eop = (len_left == 0);
394
395 /* Queue the main skb fragment. The fragments are no larger
396 * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
397 * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
398 * per fragment is queued.
399 */
400 enic_queue_wq_desc_csum_l4(wq, skb,
401 pci_map_single(enic->pdev, skb->data,
402 head_len, PCI_DMA_TODEVICE),
403 head_len,
404 csum_offset,
405 hdr_len,
406 vlan_tag_insert, vlan_tag,
407 eop, loopback);
408
409 if (!eop)
410 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
411 }
412
413 static inline void enic_queue_wq_skb_tso(struct enic *enic,
414 struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
415 int vlan_tag_insert, unsigned int vlan_tag, int loopback)
416 {
417 unsigned int frag_len_left = skb_headlen(skb);
418 unsigned int len_left = skb->len - frag_len_left;
419 unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
420 int eop = (len_left == 0);
421 unsigned int len;
422 dma_addr_t dma_addr;
423 unsigned int offset = 0;
424 skb_frag_t *frag;
425
426 /* Preload TCP csum field with IP pseudo hdr calculated
427 * with IP length set to zero. HW will later add in length
428 * to each TCP segment resulting from the TSO.
429 */
430
431 if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
432 ip_hdr(skb)->check = 0;
433 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
434 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
435 } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
436 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
437 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
438 }
439
440 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
441 * for the main skb fragment
442 */
443 while (frag_len_left) {
444 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
445 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
446 len, PCI_DMA_TODEVICE);
447 enic_queue_wq_desc_tso(wq, skb,
448 dma_addr,
449 len,
450 mss, hdr_len,
451 vlan_tag_insert, vlan_tag,
452 eop && (len == frag_len_left), loopback);
453 frag_len_left -= len;
454 offset += len;
455 }
456
457 if (eop)
458 return;
459
460 /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
461 * for additional data fragments
462 */
463 for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
464 len_left -= skb_frag_size(frag);
465 frag_len_left = skb_frag_size(frag);
466 offset = 0;
467
468 while (frag_len_left) {
469 len = min(frag_len_left,
470 (unsigned int)WQ_ENET_MAX_DESC_LEN);
471 dma_addr = skb_frag_dma_map(&enic->pdev->dev, frag,
472 offset, len,
473 DMA_TO_DEVICE);
474 enic_queue_wq_desc_cont(wq, skb,
475 dma_addr,
476 len,
477 (len_left == 0) &&
478 (len == frag_len_left), /* EOP? */
479 loopback);
480 frag_len_left -= len;
481 offset += len;
482 }
483 }
484 }
485
486 static inline void enic_queue_wq_skb(struct enic *enic,
487 struct vnic_wq *wq, struct sk_buff *skb)
488 {
489 unsigned int mss = skb_shinfo(skb)->gso_size;
490 unsigned int vlan_tag = 0;
491 int vlan_tag_insert = 0;
492 int loopback = 0;
493
494 if (vlan_tx_tag_present(skb)) {
495 /* VLAN tag from trunking driver */
496 vlan_tag_insert = 1;
497 vlan_tag = vlan_tx_tag_get(skb);
498 } else if (enic->loop_enable) {
499 vlan_tag = enic->loop_tag;
500 loopback = 1;
501 }
502
503 if (mss)
504 enic_queue_wq_skb_tso(enic, wq, skb, mss,
505 vlan_tag_insert, vlan_tag, loopback);
506 else if (skb->ip_summed == CHECKSUM_PARTIAL)
507 enic_queue_wq_skb_csum_l4(enic, wq, skb,
508 vlan_tag_insert, vlan_tag, loopback);
509 else
510 enic_queue_wq_skb_vlan(enic, wq, skb,
511 vlan_tag_insert, vlan_tag, loopback);
512 }
513
514 /* netif_tx_lock held, process context with BHs disabled, or BH */
515 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
516 struct net_device *netdev)
517 {
518 struct enic *enic = netdev_priv(netdev);
519 struct vnic_wq *wq;
520 unsigned long flags;
521 unsigned int txq_map;
522
523 if (skb->len <= 0) {
524 dev_kfree_skb(skb);
525 return NETDEV_TX_OK;
526 }
527
528 txq_map = skb_get_queue_mapping(skb) % enic->wq_count;
529 wq = &enic->wq[txq_map];
530
531 /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
532 * which is very likely. In the off chance it's going to take
533 * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
534 */
535
536 if (skb_shinfo(skb)->gso_size == 0 &&
537 skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
538 skb_linearize(skb)) {
539 dev_kfree_skb(skb);
540 return NETDEV_TX_OK;
541 }
542
543 spin_lock_irqsave(&enic->wq_lock[txq_map], flags);
544
545 if (vnic_wq_desc_avail(wq) <
546 skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
547 netif_tx_stop_queue(netdev_get_tx_queue(netdev, txq_map));
548 /* This is a hard error, log it */
549 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
550 spin_unlock_irqrestore(&enic->wq_lock[txq_map], flags);
551 return NETDEV_TX_BUSY;
552 }
553
554 enic_queue_wq_skb(enic, wq, skb);
555
556 if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
557 netif_tx_stop_queue(netdev_get_tx_queue(netdev, txq_map));
558
559 spin_unlock_irqrestore(&enic->wq_lock[txq_map], flags);
560
561 return NETDEV_TX_OK;
562 }
563
564 /* dev_base_lock rwlock held, nominally process context */
565 static struct rtnl_link_stats64 *enic_get_stats(struct net_device *netdev,
566 struct rtnl_link_stats64 *net_stats)
567 {
568 struct enic *enic = netdev_priv(netdev);
569 struct vnic_stats *stats;
570
571 enic_dev_stats_dump(enic, &stats);
572
573 net_stats->tx_packets = stats->tx.tx_frames_ok;
574 net_stats->tx_bytes = stats->tx.tx_bytes_ok;
575 net_stats->tx_errors = stats->tx.tx_errors;
576 net_stats->tx_dropped = stats->tx.tx_drops;
577
578 net_stats->rx_packets = stats->rx.rx_frames_ok;
579 net_stats->rx_bytes = stats->rx.rx_bytes_ok;
580 net_stats->rx_errors = stats->rx.rx_errors;
581 net_stats->multicast = stats->rx.rx_multicast_frames_ok;
582 net_stats->rx_over_errors = enic->rq_truncated_pkts;
583 net_stats->rx_crc_errors = enic->rq_bad_fcs;
584 net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
585
586 return net_stats;
587 }
588
589 void enic_reset_addr_lists(struct enic *enic)
590 {
591 enic->mc_count = 0;
592 enic->uc_count = 0;
593 enic->flags = 0;
594 }
595
596 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
597 {
598 struct enic *enic = netdev_priv(netdev);
599
600 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic)) {
601 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
602 return -EADDRNOTAVAIL;
603 } else {
604 if (!is_valid_ether_addr(addr))
605 return -EADDRNOTAVAIL;
606 }
607
608 memcpy(netdev->dev_addr, addr, netdev->addr_len);
609
610 return 0;
611 }
612
613 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
614 {
615 struct enic *enic = netdev_priv(netdev);
616 struct sockaddr *saddr = p;
617 char *addr = saddr->sa_data;
618 int err;
619
620 if (netif_running(enic->netdev)) {
621 err = enic_dev_del_station_addr(enic);
622 if (err)
623 return err;
624 }
625
626 err = enic_set_mac_addr(netdev, addr);
627 if (err)
628 return err;
629
630 if (netif_running(enic->netdev)) {
631 err = enic_dev_add_station_addr(enic);
632 if (err)
633 return err;
634 }
635
636 return err;
637 }
638
639 static int enic_set_mac_address(struct net_device *netdev, void *p)
640 {
641 struct sockaddr *saddr = p;
642 char *addr = saddr->sa_data;
643 struct enic *enic = netdev_priv(netdev);
644 int err;
645
646 err = enic_dev_del_station_addr(enic);
647 if (err)
648 return err;
649
650 err = enic_set_mac_addr(netdev, addr);
651 if (err)
652 return err;
653
654 return enic_dev_add_station_addr(enic);
655 }
656
657 static void enic_update_multicast_addr_list(struct enic *enic)
658 {
659 struct net_device *netdev = enic->netdev;
660 struct netdev_hw_addr *ha;
661 unsigned int mc_count = netdev_mc_count(netdev);
662 u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
663 unsigned int i, j;
664
665 if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS) {
666 netdev_warn(netdev, "Registering only %d out of %d "
667 "multicast addresses\n",
668 ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
669 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
670 }
671
672 /* Is there an easier way? Trying to minimize to
673 * calls to add/del multicast addrs. We keep the
674 * addrs from the last call in enic->mc_addr and
675 * look for changes to add/del.
676 */
677
678 i = 0;
679 netdev_for_each_mc_addr(ha, netdev) {
680 if (i == mc_count)
681 break;
682 memcpy(mc_addr[i++], ha->addr, ETH_ALEN);
683 }
684
685 for (i = 0; i < enic->mc_count; i++) {
686 for (j = 0; j < mc_count; j++)
687 if (ether_addr_equal(enic->mc_addr[i], mc_addr[j]))
688 break;
689 if (j == mc_count)
690 enic_dev_del_addr(enic, enic->mc_addr[i]);
691 }
692
693 for (i = 0; i < mc_count; i++) {
694 for (j = 0; j < enic->mc_count; j++)
695 if (ether_addr_equal(mc_addr[i], enic->mc_addr[j]))
696 break;
697 if (j == enic->mc_count)
698 enic_dev_add_addr(enic, mc_addr[i]);
699 }
700
701 /* Save the list to compare against next time
702 */
703
704 for (i = 0; i < mc_count; i++)
705 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
706
707 enic->mc_count = mc_count;
708 }
709
710 static void enic_update_unicast_addr_list(struct enic *enic)
711 {
712 struct net_device *netdev = enic->netdev;
713 struct netdev_hw_addr *ha;
714 unsigned int uc_count = netdev_uc_count(netdev);
715 u8 uc_addr[ENIC_UNICAST_PERFECT_FILTERS][ETH_ALEN];
716 unsigned int i, j;
717
718 if (uc_count > ENIC_UNICAST_PERFECT_FILTERS) {
719 netdev_warn(netdev, "Registering only %d out of %d "
720 "unicast addresses\n",
721 ENIC_UNICAST_PERFECT_FILTERS, uc_count);
722 uc_count = ENIC_UNICAST_PERFECT_FILTERS;
723 }
724
725 /* Is there an easier way? Trying to minimize to
726 * calls to add/del unicast addrs. We keep the
727 * addrs from the last call in enic->uc_addr and
728 * look for changes to add/del.
729 */
730
731 i = 0;
732 netdev_for_each_uc_addr(ha, netdev) {
733 if (i == uc_count)
734 break;
735 memcpy(uc_addr[i++], ha->addr, ETH_ALEN);
736 }
737
738 for (i = 0; i < enic->uc_count; i++) {
739 for (j = 0; j < uc_count; j++)
740 if (ether_addr_equal(enic->uc_addr[i], uc_addr[j]))
741 break;
742 if (j == uc_count)
743 enic_dev_del_addr(enic, enic->uc_addr[i]);
744 }
745
746 for (i = 0; i < uc_count; i++) {
747 for (j = 0; j < enic->uc_count; j++)
748 if (ether_addr_equal(uc_addr[i], enic->uc_addr[j]))
749 break;
750 if (j == enic->uc_count)
751 enic_dev_add_addr(enic, uc_addr[i]);
752 }
753
754 /* Save the list to compare against next time
755 */
756
757 for (i = 0; i < uc_count; i++)
758 memcpy(enic->uc_addr[i], uc_addr[i], ETH_ALEN);
759
760 enic->uc_count = uc_count;
761 }
762
763 /* netif_tx_lock held, BHs disabled */
764 static void enic_set_rx_mode(struct net_device *netdev)
765 {
766 struct enic *enic = netdev_priv(netdev);
767 int directed = 1;
768 int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
769 int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
770 int promisc = (netdev->flags & IFF_PROMISC) ||
771 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
772 int allmulti = (netdev->flags & IFF_ALLMULTI) ||
773 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
774 unsigned int flags = netdev->flags |
775 (allmulti ? IFF_ALLMULTI : 0) |
776 (promisc ? IFF_PROMISC : 0);
777
778 if (enic->flags != flags) {
779 enic->flags = flags;
780 enic_dev_packet_filter(enic, directed,
781 multicast, broadcast, promisc, allmulti);
782 }
783
784 if (!promisc) {
785 enic_update_unicast_addr_list(enic);
786 if (!allmulti)
787 enic_update_multicast_addr_list(enic);
788 }
789 }
790
791 /* netif_tx_lock held, BHs disabled */
792 static void enic_tx_timeout(struct net_device *netdev)
793 {
794 struct enic *enic = netdev_priv(netdev);
795 schedule_work(&enic->reset);
796 }
797
798 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
799 {
800 struct enic *enic = netdev_priv(netdev);
801 struct enic_port_profile *pp;
802 int err;
803
804 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
805 if (err)
806 return err;
807
808 if (is_valid_ether_addr(mac) || is_zero_ether_addr(mac)) {
809 if (vf == PORT_SELF_VF) {
810 memcpy(pp->vf_mac, mac, ETH_ALEN);
811 return 0;
812 } else {
813 /*
814 * For sriov vf's set the mac in hw
815 */
816 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
817 vnic_dev_set_mac_addr, mac);
818 return enic_dev_status_to_errno(err);
819 }
820 } else
821 return -EINVAL;
822 }
823
824 static int enic_set_vf_port(struct net_device *netdev, int vf,
825 struct nlattr *port[])
826 {
827 struct enic *enic = netdev_priv(netdev);
828 struct enic_port_profile prev_pp;
829 struct enic_port_profile *pp;
830 int err = 0, restore_pp = 1;
831
832 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
833 if (err)
834 return err;
835
836 if (!port[IFLA_PORT_REQUEST])
837 return -EOPNOTSUPP;
838
839 memcpy(&prev_pp, pp, sizeof(*enic->pp));
840 memset(pp, 0, sizeof(*enic->pp));
841
842 pp->set |= ENIC_SET_REQUEST;
843 pp->request = nla_get_u8(port[IFLA_PORT_REQUEST]);
844
845 if (port[IFLA_PORT_PROFILE]) {
846 pp->set |= ENIC_SET_NAME;
847 memcpy(pp->name, nla_data(port[IFLA_PORT_PROFILE]),
848 PORT_PROFILE_MAX);
849 }
850
851 if (port[IFLA_PORT_INSTANCE_UUID]) {
852 pp->set |= ENIC_SET_INSTANCE;
853 memcpy(pp->instance_uuid,
854 nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
855 }
856
857 if (port[IFLA_PORT_HOST_UUID]) {
858 pp->set |= ENIC_SET_HOST;
859 memcpy(pp->host_uuid,
860 nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
861 }
862
863 if (vf == PORT_SELF_VF) {
864 /* Special case handling: mac came from IFLA_VF_MAC */
865 if (!is_zero_ether_addr(prev_pp.vf_mac))
866 memcpy(pp->mac_addr, prev_pp.vf_mac, ETH_ALEN);
867
868 if (is_zero_ether_addr(netdev->dev_addr))
869 eth_hw_addr_random(netdev);
870 } else {
871 /* SR-IOV VF: get mac from adapter */
872 ENIC_DEVCMD_PROXY_BY_INDEX(vf, err, enic,
873 vnic_dev_get_mac_addr, pp->mac_addr);
874 if (err) {
875 netdev_err(netdev, "Error getting mac for vf %d\n", vf);
876 memcpy(pp, &prev_pp, sizeof(*pp));
877 return enic_dev_status_to_errno(err);
878 }
879 }
880
881 err = enic_process_set_pp_request(enic, vf, &prev_pp, &restore_pp);
882 if (err) {
883 if (restore_pp) {
884 /* Things are still the way they were: Implicit
885 * DISASSOCIATE failed
886 */
887 memcpy(pp, &prev_pp, sizeof(*pp));
888 } else {
889 memset(pp, 0, sizeof(*pp));
890 if (vf == PORT_SELF_VF)
891 memset(netdev->dev_addr, 0, ETH_ALEN);
892 }
893 } else {
894 /* Set flag to indicate that the port assoc/disassoc
895 * request has been sent out to fw
896 */
897 pp->set |= ENIC_PORT_REQUEST_APPLIED;
898
899 /* If DISASSOCIATE, clean up all assigned/saved macaddresses */
900 if (pp->request == PORT_REQUEST_DISASSOCIATE) {
901 memset(pp->mac_addr, 0, ETH_ALEN);
902 if (vf == PORT_SELF_VF)
903 memset(netdev->dev_addr, 0, ETH_ALEN);
904 }
905 }
906
907 if (vf == PORT_SELF_VF)
908 memset(pp->vf_mac, 0, ETH_ALEN);
909
910 return err;
911 }
912
913 static int enic_get_vf_port(struct net_device *netdev, int vf,
914 struct sk_buff *skb)
915 {
916 struct enic *enic = netdev_priv(netdev);
917 u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
918 struct enic_port_profile *pp;
919 int err;
920
921 ENIC_PP_BY_INDEX(enic, vf, pp, &err);
922 if (err)
923 return err;
924
925 if (!(pp->set & ENIC_PORT_REQUEST_APPLIED))
926 return -ENODATA;
927
928 err = enic_process_get_pp_request(enic, vf, pp->request, &response);
929 if (err)
930 return err;
931
932 if (nla_put_u16(skb, IFLA_PORT_REQUEST, pp->request) ||
933 nla_put_u16(skb, IFLA_PORT_RESPONSE, response) ||
934 ((pp->set & ENIC_SET_NAME) &&
935 nla_put(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX, pp->name)) ||
936 ((pp->set & ENIC_SET_INSTANCE) &&
937 nla_put(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
938 pp->instance_uuid)) ||
939 ((pp->set & ENIC_SET_HOST) &&
940 nla_put(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX, pp->host_uuid)))
941 goto nla_put_failure;
942 return 0;
943
944 nla_put_failure:
945 return -EMSGSIZE;
946 }
947
948 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
949 {
950 struct enic *enic = vnic_dev_priv(rq->vdev);
951
952 if (!buf->os_buf)
953 return;
954
955 pci_unmap_single(enic->pdev, buf->dma_addr,
956 buf->len, PCI_DMA_FROMDEVICE);
957 dev_kfree_skb_any(buf->os_buf);
958 }
959
960 static int enic_rq_alloc_buf(struct vnic_rq *rq)
961 {
962 struct enic *enic = vnic_dev_priv(rq->vdev);
963 struct net_device *netdev = enic->netdev;
964 struct sk_buff *skb;
965 unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
966 unsigned int os_buf_index = 0;
967 dma_addr_t dma_addr;
968
969 skb = netdev_alloc_skb_ip_align(netdev, len);
970 if (!skb)
971 return -ENOMEM;
972
973 dma_addr = pci_map_single(enic->pdev, skb->data,
974 len, PCI_DMA_FROMDEVICE);
975
976 enic_queue_rq_desc(rq, skb, os_buf_index,
977 dma_addr, len);
978
979 return 0;
980 }
981
982 static void enic_rq_indicate_buf(struct vnic_rq *rq,
983 struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
984 int skipped, void *opaque)
985 {
986 struct enic *enic = vnic_dev_priv(rq->vdev);
987 struct net_device *netdev = enic->netdev;
988 struct sk_buff *skb;
989
990 u8 type, color, eop, sop, ingress_port, vlan_stripped;
991 u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
992 u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
993 u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
994 u8 packet_error;
995 u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
996 u32 rss_hash;
997
998 if (skipped)
999 return;
1000
1001 skb = buf->os_buf;
1002 prefetch(skb->data - NET_IP_ALIGN);
1003 pci_unmap_single(enic->pdev, buf->dma_addr,
1004 buf->len, PCI_DMA_FROMDEVICE);
1005
1006 cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1007 &type, &color, &q_number, &completed_index,
1008 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1009 &csum_not_calc, &rss_hash, &bytes_written,
1010 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1011 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1012 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1013 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1014 &fcs_ok);
1015
1016 if (packet_error) {
1017
1018 if (!fcs_ok) {
1019 if (bytes_written > 0)
1020 enic->rq_bad_fcs++;
1021 else if (bytes_written == 0)
1022 enic->rq_truncated_pkts++;
1023 }
1024
1025 dev_kfree_skb_any(skb);
1026
1027 return;
1028 }
1029
1030 if (eop && bytes_written > 0) {
1031
1032 /* Good receive
1033 */
1034
1035 skb_put(skb, bytes_written);
1036 skb->protocol = eth_type_trans(skb, netdev);
1037 skb_record_rx_queue(skb, q_number);
1038 if (netdev->features & NETIF_F_RXHASH) {
1039 skb->rxhash = rss_hash;
1040 if (rss_type & (NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX |
1041 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6 |
1042 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4))
1043 skb->l4_rxhash = true;
1044 }
1045
1046 if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc) {
1047 skb->csum = htons(checksum);
1048 skb->ip_summed = CHECKSUM_COMPLETE;
1049 }
1050
1051 if (vlan_stripped)
1052 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1053
1054 if (netdev->features & NETIF_F_GRO)
1055 napi_gro_receive(&enic->napi[q_number], skb);
1056 else
1057 netif_receive_skb(skb);
1058 } else {
1059
1060 /* Buffer overflow
1061 */
1062
1063 dev_kfree_skb_any(skb);
1064 }
1065 }
1066
1067 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1068 u8 type, u16 q_number, u16 completed_index, void *opaque)
1069 {
1070 struct enic *enic = vnic_dev_priv(vdev);
1071
1072 vnic_rq_service(&enic->rq[q_number], cq_desc,
1073 completed_index, VNIC_RQ_RETURN_DESC,
1074 enic_rq_indicate_buf, opaque);
1075
1076 return 0;
1077 }
1078
1079 static int enic_poll(struct napi_struct *napi, int budget)
1080 {
1081 struct net_device *netdev = napi->dev;
1082 struct enic *enic = netdev_priv(netdev);
1083 unsigned int cq_rq = enic_cq_rq(enic, 0);
1084 unsigned int cq_wq = enic_cq_wq(enic, 0);
1085 unsigned int intr = enic_legacy_io_intr();
1086 unsigned int rq_work_to_do = budget;
1087 unsigned int wq_work_to_do = -1; /* no limit */
1088 unsigned int work_done, rq_work_done, wq_work_done;
1089 int err;
1090
1091 /* Service RQ (first) and WQ
1092 */
1093
1094 rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1095 rq_work_to_do, enic_rq_service, NULL);
1096
1097 wq_work_done = vnic_cq_service(&enic->cq[cq_wq],
1098 wq_work_to_do, enic_wq_service, NULL);
1099
1100 /* Accumulate intr event credits for this polling
1101 * cycle. An intr event is the completion of a
1102 * a WQ or RQ packet.
1103 */
1104
1105 work_done = rq_work_done + wq_work_done;
1106
1107 if (work_done > 0)
1108 vnic_intr_return_credits(&enic->intr[intr],
1109 work_done,
1110 0 /* don't unmask intr */,
1111 0 /* don't reset intr timer */);
1112
1113 err = vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1114
1115 /* Buffer allocation failed. Stay in polling
1116 * mode so we can try to fill the ring again.
1117 */
1118
1119 if (err)
1120 rq_work_done = rq_work_to_do;
1121
1122 if (rq_work_done < rq_work_to_do) {
1123
1124 /* Some work done, but not enough to stay in polling,
1125 * exit polling
1126 */
1127
1128 napi_complete(napi);
1129 vnic_intr_unmask(&enic->intr[intr]);
1130 }
1131
1132 return rq_work_done;
1133 }
1134
1135 static int enic_poll_msix(struct napi_struct *napi, int budget)
1136 {
1137 struct net_device *netdev = napi->dev;
1138 struct enic *enic = netdev_priv(netdev);
1139 unsigned int rq = (napi - &enic->napi[0]);
1140 unsigned int cq = enic_cq_rq(enic, rq);
1141 unsigned int intr = enic_msix_rq_intr(enic, rq);
1142 unsigned int work_to_do = budget;
1143 unsigned int work_done;
1144 int err;
1145
1146 /* Service RQ
1147 */
1148
1149 work_done = vnic_cq_service(&enic->cq[cq],
1150 work_to_do, enic_rq_service, NULL);
1151
1152 /* Return intr event credits for this polling
1153 * cycle. An intr event is the completion of a
1154 * RQ packet.
1155 */
1156
1157 if (work_done > 0)
1158 vnic_intr_return_credits(&enic->intr[intr],
1159 work_done,
1160 0 /* don't unmask intr */,
1161 0 /* don't reset intr timer */);
1162
1163 err = vnic_rq_fill(&enic->rq[rq], enic_rq_alloc_buf);
1164
1165 /* Buffer allocation failed. Stay in polling mode
1166 * so we can try to fill the ring again.
1167 */
1168
1169 if (err)
1170 work_done = work_to_do;
1171
1172 if (work_done < work_to_do) {
1173
1174 /* Some work done, but not enough to stay in polling,
1175 * exit polling
1176 */
1177
1178 napi_complete(napi);
1179 vnic_intr_unmask(&enic->intr[intr]);
1180 }
1181
1182 return work_done;
1183 }
1184
1185 static void enic_notify_timer(unsigned long data)
1186 {
1187 struct enic *enic = (struct enic *)data;
1188
1189 enic_notify_check(enic);
1190
1191 mod_timer(&enic->notify_timer,
1192 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1193 }
1194
1195 static void enic_free_intr(struct enic *enic)
1196 {
1197 struct net_device *netdev = enic->netdev;
1198 unsigned int i;
1199
1200 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1201 case VNIC_DEV_INTR_MODE_INTX:
1202 free_irq(enic->pdev->irq, netdev);
1203 break;
1204 case VNIC_DEV_INTR_MODE_MSI:
1205 free_irq(enic->pdev->irq, enic);
1206 break;
1207 case VNIC_DEV_INTR_MODE_MSIX:
1208 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1209 if (enic->msix[i].requested)
1210 free_irq(enic->msix_entry[i].vector,
1211 enic->msix[i].devid);
1212 break;
1213 default:
1214 break;
1215 }
1216 }
1217
1218 static int enic_request_intr(struct enic *enic)
1219 {
1220 struct net_device *netdev = enic->netdev;
1221 unsigned int i, intr;
1222 int err = 0;
1223
1224 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1225
1226 case VNIC_DEV_INTR_MODE_INTX:
1227
1228 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1229 IRQF_SHARED, netdev->name, netdev);
1230 break;
1231
1232 case VNIC_DEV_INTR_MODE_MSI:
1233
1234 err = request_irq(enic->pdev->irq, enic_isr_msi,
1235 0, netdev->name, enic);
1236 break;
1237
1238 case VNIC_DEV_INTR_MODE_MSIX:
1239
1240 for (i = 0; i < enic->rq_count; i++) {
1241 intr = enic_msix_rq_intr(enic, i);
1242 snprintf(enic->msix[intr].devname,
1243 sizeof(enic->msix[intr].devname),
1244 "%.11s-rx-%d", netdev->name, i);
1245 enic->msix[intr].isr = enic_isr_msix_rq;
1246 enic->msix[intr].devid = &enic->napi[i];
1247 }
1248
1249 for (i = 0; i < enic->wq_count; i++) {
1250 intr = enic_msix_wq_intr(enic, i);
1251 snprintf(enic->msix[intr].devname,
1252 sizeof(enic->msix[intr].devname),
1253 "%.11s-tx-%d", netdev->name, i);
1254 enic->msix[intr].isr = enic_isr_msix_wq;
1255 enic->msix[intr].devid = enic;
1256 }
1257
1258 intr = enic_msix_err_intr(enic);
1259 snprintf(enic->msix[intr].devname,
1260 sizeof(enic->msix[intr].devname),
1261 "%.11s-err", netdev->name);
1262 enic->msix[intr].isr = enic_isr_msix_err;
1263 enic->msix[intr].devid = enic;
1264
1265 intr = enic_msix_notify_intr(enic);
1266 snprintf(enic->msix[intr].devname,
1267 sizeof(enic->msix[intr].devname),
1268 "%.11s-notify", netdev->name);
1269 enic->msix[intr].isr = enic_isr_msix_notify;
1270 enic->msix[intr].devid = enic;
1271
1272 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1273 enic->msix[i].requested = 0;
1274
1275 for (i = 0; i < enic->intr_count; i++) {
1276 err = request_irq(enic->msix_entry[i].vector,
1277 enic->msix[i].isr, 0,
1278 enic->msix[i].devname,
1279 enic->msix[i].devid);
1280 if (err) {
1281 enic_free_intr(enic);
1282 break;
1283 }
1284 enic->msix[i].requested = 1;
1285 }
1286
1287 break;
1288
1289 default:
1290 break;
1291 }
1292
1293 return err;
1294 }
1295
1296 static void enic_synchronize_irqs(struct enic *enic)
1297 {
1298 unsigned int i;
1299
1300 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1301 case VNIC_DEV_INTR_MODE_INTX:
1302 case VNIC_DEV_INTR_MODE_MSI:
1303 synchronize_irq(enic->pdev->irq);
1304 break;
1305 case VNIC_DEV_INTR_MODE_MSIX:
1306 for (i = 0; i < enic->intr_count; i++)
1307 synchronize_irq(enic->msix_entry[i].vector);
1308 break;
1309 default:
1310 break;
1311 }
1312 }
1313
1314 static int enic_dev_notify_set(struct enic *enic)
1315 {
1316 int err;
1317
1318 spin_lock(&enic->devcmd_lock);
1319 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1320 case VNIC_DEV_INTR_MODE_INTX:
1321 err = vnic_dev_notify_set(enic->vdev,
1322 enic_legacy_notify_intr());
1323 break;
1324 case VNIC_DEV_INTR_MODE_MSIX:
1325 err = vnic_dev_notify_set(enic->vdev,
1326 enic_msix_notify_intr(enic));
1327 break;
1328 default:
1329 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1330 break;
1331 }
1332 spin_unlock(&enic->devcmd_lock);
1333
1334 return err;
1335 }
1336
1337 static void enic_notify_timer_start(struct enic *enic)
1338 {
1339 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1340 case VNIC_DEV_INTR_MODE_MSI:
1341 mod_timer(&enic->notify_timer, jiffies);
1342 break;
1343 default:
1344 /* Using intr for notification for INTx/MSI-X */
1345 break;
1346 }
1347 }
1348
1349 /* rtnl lock is held, process context */
1350 static int enic_open(struct net_device *netdev)
1351 {
1352 struct enic *enic = netdev_priv(netdev);
1353 unsigned int i;
1354 int err;
1355
1356 err = enic_request_intr(enic);
1357 if (err) {
1358 netdev_err(netdev, "Unable to request irq.\n");
1359 return err;
1360 }
1361
1362 err = enic_dev_notify_set(enic);
1363 if (err) {
1364 netdev_err(netdev,
1365 "Failed to alloc notify buffer, aborting.\n");
1366 goto err_out_free_intr;
1367 }
1368
1369 for (i = 0; i < enic->rq_count; i++) {
1370 vnic_rq_fill(&enic->rq[i], enic_rq_alloc_buf);
1371 /* Need at least one buffer on ring to get going */
1372 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1373 netdev_err(netdev, "Unable to alloc receive buffers\n");
1374 err = -ENOMEM;
1375 goto err_out_notify_unset;
1376 }
1377 }
1378
1379 for (i = 0; i < enic->wq_count; i++)
1380 vnic_wq_enable(&enic->wq[i]);
1381 for (i = 0; i < enic->rq_count; i++)
1382 vnic_rq_enable(&enic->rq[i]);
1383
1384 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1385 enic_dev_add_station_addr(enic);
1386
1387 enic_set_rx_mode(netdev);
1388
1389 netif_tx_wake_all_queues(netdev);
1390
1391 for (i = 0; i < enic->rq_count; i++)
1392 napi_enable(&enic->napi[i]);
1393
1394 enic_dev_enable(enic);
1395
1396 for (i = 0; i < enic->intr_count; i++)
1397 vnic_intr_unmask(&enic->intr[i]);
1398
1399 enic_notify_timer_start(enic);
1400
1401 return 0;
1402
1403 err_out_notify_unset:
1404 enic_dev_notify_unset(enic);
1405 err_out_free_intr:
1406 enic_free_intr(enic);
1407
1408 return err;
1409 }
1410
1411 /* rtnl lock is held, process context */
1412 static int enic_stop(struct net_device *netdev)
1413 {
1414 struct enic *enic = netdev_priv(netdev);
1415 unsigned int i;
1416 int err;
1417
1418 for (i = 0; i < enic->intr_count; i++) {
1419 vnic_intr_mask(&enic->intr[i]);
1420 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
1421 }
1422
1423 enic_synchronize_irqs(enic);
1424
1425 del_timer_sync(&enic->notify_timer);
1426
1427 enic_dev_disable(enic);
1428
1429 for (i = 0; i < enic->rq_count; i++)
1430 napi_disable(&enic->napi[i]);
1431
1432 netif_carrier_off(netdev);
1433 netif_tx_disable(netdev);
1434
1435 if (!enic_is_dynamic(enic) && !enic_is_sriov_vf(enic))
1436 enic_dev_del_station_addr(enic);
1437
1438 for (i = 0; i < enic->wq_count; i++) {
1439 err = vnic_wq_disable(&enic->wq[i]);
1440 if (err)
1441 return err;
1442 }
1443 for (i = 0; i < enic->rq_count; i++) {
1444 err = vnic_rq_disable(&enic->rq[i]);
1445 if (err)
1446 return err;
1447 }
1448
1449 enic_dev_notify_unset(enic);
1450 enic_free_intr(enic);
1451
1452 for (i = 0; i < enic->wq_count; i++)
1453 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1454 for (i = 0; i < enic->rq_count; i++)
1455 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1456 for (i = 0; i < enic->cq_count; i++)
1457 vnic_cq_clean(&enic->cq[i]);
1458 for (i = 0; i < enic->intr_count; i++)
1459 vnic_intr_clean(&enic->intr[i]);
1460
1461 return 0;
1462 }
1463
1464 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1465 {
1466 struct enic *enic = netdev_priv(netdev);
1467 int running = netif_running(netdev);
1468
1469 if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1470 return -EINVAL;
1471
1472 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
1473 return -EOPNOTSUPP;
1474
1475 if (running)
1476 enic_stop(netdev);
1477
1478 netdev->mtu = new_mtu;
1479
1480 if (netdev->mtu > enic->port_mtu)
1481 netdev_warn(netdev,
1482 "interface MTU (%d) set higher than port MTU (%d)\n",
1483 netdev->mtu, enic->port_mtu);
1484
1485 if (running)
1486 enic_open(netdev);
1487
1488 return 0;
1489 }
1490
1491 static void enic_change_mtu_work(struct work_struct *work)
1492 {
1493 struct enic *enic = container_of(work, struct enic, change_mtu_work);
1494 struct net_device *netdev = enic->netdev;
1495 int new_mtu = vnic_dev_mtu(enic->vdev);
1496 int err;
1497 unsigned int i;
1498
1499 new_mtu = max_t(int, ENIC_MIN_MTU, min_t(int, ENIC_MAX_MTU, new_mtu));
1500
1501 rtnl_lock();
1502
1503 /* Stop RQ */
1504 del_timer_sync(&enic->notify_timer);
1505
1506 for (i = 0; i < enic->rq_count; i++)
1507 napi_disable(&enic->napi[i]);
1508
1509 vnic_intr_mask(&enic->intr[0]);
1510 enic_synchronize_irqs(enic);
1511 err = vnic_rq_disable(&enic->rq[0]);
1512 if (err) {
1513 rtnl_unlock();
1514 netdev_err(netdev, "Unable to disable RQ.\n");
1515 return;
1516 }
1517 vnic_rq_clean(&enic->rq[0], enic_free_rq_buf);
1518 vnic_cq_clean(&enic->cq[0]);
1519 vnic_intr_clean(&enic->intr[0]);
1520
1521 /* Fill RQ with new_mtu-sized buffers */
1522 netdev->mtu = new_mtu;
1523 vnic_rq_fill(&enic->rq[0], enic_rq_alloc_buf);
1524 /* Need at least one buffer on ring to get going */
1525 if (vnic_rq_desc_used(&enic->rq[0]) == 0) {
1526 rtnl_unlock();
1527 netdev_err(netdev, "Unable to alloc receive buffers.\n");
1528 return;
1529 }
1530
1531 /* Start RQ */
1532 vnic_rq_enable(&enic->rq[0]);
1533 napi_enable(&enic->napi[0]);
1534 vnic_intr_unmask(&enic->intr[0]);
1535 enic_notify_timer_start(enic);
1536
1537 rtnl_unlock();
1538
1539 netdev_info(netdev, "interface MTU set as %d\n", netdev->mtu);
1540 }
1541
1542 #ifdef CONFIG_NET_POLL_CONTROLLER
1543 static void enic_poll_controller(struct net_device *netdev)
1544 {
1545 struct enic *enic = netdev_priv(netdev);
1546 struct vnic_dev *vdev = enic->vdev;
1547 unsigned int i, intr;
1548
1549 switch (vnic_dev_get_intr_mode(vdev)) {
1550 case VNIC_DEV_INTR_MODE_MSIX:
1551 for (i = 0; i < enic->rq_count; i++) {
1552 intr = enic_msix_rq_intr(enic, i);
1553 enic_isr_msix_rq(enic->msix_entry[intr].vector,
1554 &enic->napi[i]);
1555 }
1556
1557 for (i = 0; i < enic->wq_count; i++) {
1558 intr = enic_msix_wq_intr(enic, i);
1559 enic_isr_msix_wq(enic->msix_entry[intr].vector, enic);
1560 }
1561
1562 break;
1563 case VNIC_DEV_INTR_MODE_MSI:
1564 enic_isr_msi(enic->pdev->irq, enic);
1565 break;
1566 case VNIC_DEV_INTR_MODE_INTX:
1567 enic_isr_legacy(enic->pdev->irq, netdev);
1568 break;
1569 default:
1570 break;
1571 }
1572 }
1573 #endif
1574
1575 static int enic_dev_wait(struct vnic_dev *vdev,
1576 int (*start)(struct vnic_dev *, int),
1577 int (*finished)(struct vnic_dev *, int *),
1578 int arg)
1579 {
1580 unsigned long time;
1581 int done;
1582 int err;
1583
1584 BUG_ON(in_interrupt());
1585
1586 err = start(vdev, arg);
1587 if (err)
1588 return err;
1589
1590 /* Wait for func to complete...2 seconds max
1591 */
1592
1593 time = jiffies + (HZ * 2);
1594 do {
1595
1596 err = finished(vdev, &done);
1597 if (err)
1598 return err;
1599
1600 if (done)
1601 return 0;
1602
1603 schedule_timeout_uninterruptible(HZ / 10);
1604
1605 } while (time_after(time, jiffies));
1606
1607 return -ETIMEDOUT;
1608 }
1609
1610 static int enic_dev_open(struct enic *enic)
1611 {
1612 int err;
1613
1614 err = enic_dev_wait(enic->vdev, vnic_dev_open,
1615 vnic_dev_open_done, 0);
1616 if (err)
1617 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
1618 err);
1619
1620 return err;
1621 }
1622
1623 static int enic_dev_hang_reset(struct enic *enic)
1624 {
1625 int err;
1626
1627 err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1628 vnic_dev_hang_reset_done, 0);
1629 if (err)
1630 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
1631 err);
1632
1633 return err;
1634 }
1635
1636 static int enic_set_rsskey(struct enic *enic)
1637 {
1638 dma_addr_t rss_key_buf_pa;
1639 union vnic_rss_key *rss_key_buf_va = NULL;
1640 union vnic_rss_key rss_key = {
1641 .key[0].b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101},
1642 .key[1].b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101},
1643 .key[2].b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115},
1644 .key[3].b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108},
1645 };
1646 int err;
1647
1648 rss_key_buf_va = pci_alloc_consistent(enic->pdev,
1649 sizeof(union vnic_rss_key), &rss_key_buf_pa);
1650 if (!rss_key_buf_va)
1651 return -ENOMEM;
1652
1653 memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
1654
1655 spin_lock(&enic->devcmd_lock);
1656 err = enic_set_rss_key(enic,
1657 rss_key_buf_pa,
1658 sizeof(union vnic_rss_key));
1659 spin_unlock(&enic->devcmd_lock);
1660
1661 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
1662 rss_key_buf_va, rss_key_buf_pa);
1663
1664 return err;
1665 }
1666
1667 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
1668 {
1669 dma_addr_t rss_cpu_buf_pa;
1670 union vnic_rss_cpu *rss_cpu_buf_va = NULL;
1671 unsigned int i;
1672 int err;
1673
1674 rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
1675 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
1676 if (!rss_cpu_buf_va)
1677 return -ENOMEM;
1678
1679 for (i = 0; i < (1 << rss_hash_bits); i++)
1680 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
1681
1682 spin_lock(&enic->devcmd_lock);
1683 err = enic_set_rss_cpu(enic,
1684 rss_cpu_buf_pa,
1685 sizeof(union vnic_rss_cpu));
1686 spin_unlock(&enic->devcmd_lock);
1687
1688 pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
1689 rss_cpu_buf_va, rss_cpu_buf_pa);
1690
1691 return err;
1692 }
1693
1694 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
1695 u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
1696 {
1697 const u8 tso_ipid_split_en = 0;
1698 const u8 ig_vlan_strip_en = 1;
1699 int err;
1700
1701 /* Enable VLAN tag stripping.
1702 */
1703
1704 spin_lock(&enic->devcmd_lock);
1705 err = enic_set_nic_cfg(enic,
1706 rss_default_cpu, rss_hash_type,
1707 rss_hash_bits, rss_base_cpu,
1708 rss_enable, tso_ipid_split_en,
1709 ig_vlan_strip_en);
1710 spin_unlock(&enic->devcmd_lock);
1711
1712 return err;
1713 }
1714
1715 static int enic_set_rss_nic_cfg(struct enic *enic)
1716 {
1717 struct device *dev = enic_get_dev(enic);
1718 const u8 rss_default_cpu = 0;
1719 const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1720 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1721 NIC_CFG_RSS_HASH_TYPE_IPV6 |
1722 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1723 const u8 rss_hash_bits = 7;
1724 const u8 rss_base_cpu = 0;
1725 u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1726
1727 if (rss_enable) {
1728 if (!enic_set_rsskey(enic)) {
1729 if (enic_set_rsscpu(enic, rss_hash_bits)) {
1730 rss_enable = 0;
1731 dev_warn(dev, "RSS disabled, "
1732 "Failed to set RSS cpu indirection table.");
1733 }
1734 } else {
1735 rss_enable = 0;
1736 dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
1737 }
1738 }
1739
1740 return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1741 rss_hash_bits, rss_base_cpu, rss_enable);
1742 }
1743
1744 static void enic_reset(struct work_struct *work)
1745 {
1746 struct enic *enic = container_of(work, struct enic, reset);
1747
1748 if (!netif_running(enic->netdev))
1749 return;
1750
1751 rtnl_lock();
1752
1753 spin_lock(&enic->enic_api_lock);
1754 enic_dev_hang_notify(enic);
1755 enic_stop(enic->netdev);
1756 enic_dev_hang_reset(enic);
1757 enic_reset_addr_lists(enic);
1758 enic_init_vnic_resources(enic);
1759 enic_set_rss_nic_cfg(enic);
1760 enic_dev_set_ig_vlan_rewrite_mode(enic);
1761 enic_open(enic->netdev);
1762 spin_unlock(&enic->enic_api_lock);
1763 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
1764
1765 rtnl_unlock();
1766 }
1767
1768 static int enic_set_intr_mode(struct enic *enic)
1769 {
1770 unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
1771 unsigned int m = min_t(unsigned int, enic->wq_count, ENIC_WQ_MAX);
1772 unsigned int i;
1773
1774 /* Set interrupt mode (INTx, MSI, MSI-X) depending
1775 * on system capabilities.
1776 *
1777 * Try MSI-X first
1778 *
1779 * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1780 * (the second to last INTR is used for WQ/RQ errors)
1781 * (the last INTR is used for notifications)
1782 */
1783
1784 BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1785 for (i = 0; i < n + m + 2; i++)
1786 enic->msix_entry[i].entry = i;
1787
1788 /* Use multiple RQs if RSS is enabled
1789 */
1790
1791 if (ENIC_SETTING(enic, RSS) &&
1792 enic->config.intr_mode < 1 &&
1793 enic->rq_count >= n &&
1794 enic->wq_count >= m &&
1795 enic->cq_count >= n + m &&
1796 enic->intr_count >= n + m + 2) {
1797
1798 if (!pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1799
1800 enic->rq_count = n;
1801 enic->wq_count = m;
1802 enic->cq_count = n + m;
1803 enic->intr_count = n + m + 2;
1804
1805 vnic_dev_set_intr_mode(enic->vdev,
1806 VNIC_DEV_INTR_MODE_MSIX);
1807
1808 return 0;
1809 }
1810 }
1811
1812 if (enic->config.intr_mode < 1 &&
1813 enic->rq_count >= 1 &&
1814 enic->wq_count >= m &&
1815 enic->cq_count >= 1 + m &&
1816 enic->intr_count >= 1 + m + 2) {
1817 if (!pci_enable_msix(enic->pdev, enic->msix_entry, 1 + m + 2)) {
1818
1819 enic->rq_count = 1;
1820 enic->wq_count = m;
1821 enic->cq_count = 1 + m;
1822 enic->intr_count = 1 + m + 2;
1823
1824 vnic_dev_set_intr_mode(enic->vdev,
1825 VNIC_DEV_INTR_MODE_MSIX);
1826
1827 return 0;
1828 }
1829 }
1830
1831 /* Next try MSI
1832 *
1833 * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1834 */
1835
1836 if (enic->config.intr_mode < 2 &&
1837 enic->rq_count >= 1 &&
1838 enic->wq_count >= 1 &&
1839 enic->cq_count >= 2 &&
1840 enic->intr_count >= 1 &&
1841 !pci_enable_msi(enic->pdev)) {
1842
1843 enic->rq_count = 1;
1844 enic->wq_count = 1;
1845 enic->cq_count = 2;
1846 enic->intr_count = 1;
1847
1848 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1849
1850 return 0;
1851 }
1852
1853 /* Next try INTx
1854 *
1855 * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1856 * (the first INTR is used for WQ/RQ)
1857 * (the second INTR is used for WQ/RQ errors)
1858 * (the last INTR is used for notifications)
1859 */
1860
1861 if (enic->config.intr_mode < 3 &&
1862 enic->rq_count >= 1 &&
1863 enic->wq_count >= 1 &&
1864 enic->cq_count >= 2 &&
1865 enic->intr_count >= 3) {
1866
1867 enic->rq_count = 1;
1868 enic->wq_count = 1;
1869 enic->cq_count = 2;
1870 enic->intr_count = 3;
1871
1872 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1873
1874 return 0;
1875 }
1876
1877 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1878
1879 return -EINVAL;
1880 }
1881
1882 static void enic_clear_intr_mode(struct enic *enic)
1883 {
1884 switch (vnic_dev_get_intr_mode(enic->vdev)) {
1885 case VNIC_DEV_INTR_MODE_MSIX:
1886 pci_disable_msix(enic->pdev);
1887 break;
1888 case VNIC_DEV_INTR_MODE_MSI:
1889 pci_disable_msi(enic->pdev);
1890 break;
1891 default:
1892 break;
1893 }
1894
1895 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1896 }
1897
1898 static const struct net_device_ops enic_netdev_dynamic_ops = {
1899 .ndo_open = enic_open,
1900 .ndo_stop = enic_stop,
1901 .ndo_start_xmit = enic_hard_start_xmit,
1902 .ndo_get_stats64 = enic_get_stats,
1903 .ndo_validate_addr = eth_validate_addr,
1904 .ndo_set_rx_mode = enic_set_rx_mode,
1905 .ndo_set_mac_address = enic_set_mac_address_dynamic,
1906 .ndo_change_mtu = enic_change_mtu,
1907 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
1908 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
1909 .ndo_tx_timeout = enic_tx_timeout,
1910 .ndo_set_vf_port = enic_set_vf_port,
1911 .ndo_get_vf_port = enic_get_vf_port,
1912 .ndo_set_vf_mac = enic_set_vf_mac,
1913 #ifdef CONFIG_NET_POLL_CONTROLLER
1914 .ndo_poll_controller = enic_poll_controller,
1915 #endif
1916 };
1917
1918 static const struct net_device_ops enic_netdev_ops = {
1919 .ndo_open = enic_open,
1920 .ndo_stop = enic_stop,
1921 .ndo_start_xmit = enic_hard_start_xmit,
1922 .ndo_get_stats64 = enic_get_stats,
1923 .ndo_validate_addr = eth_validate_addr,
1924 .ndo_set_mac_address = enic_set_mac_address,
1925 .ndo_set_rx_mode = enic_set_rx_mode,
1926 .ndo_change_mtu = enic_change_mtu,
1927 .ndo_vlan_rx_add_vid = enic_vlan_rx_add_vid,
1928 .ndo_vlan_rx_kill_vid = enic_vlan_rx_kill_vid,
1929 .ndo_tx_timeout = enic_tx_timeout,
1930 .ndo_set_vf_port = enic_set_vf_port,
1931 .ndo_get_vf_port = enic_get_vf_port,
1932 .ndo_set_vf_mac = enic_set_vf_mac,
1933 #ifdef CONFIG_NET_POLL_CONTROLLER
1934 .ndo_poll_controller = enic_poll_controller,
1935 #endif
1936 };
1937
1938 static void enic_dev_deinit(struct enic *enic)
1939 {
1940 unsigned int i;
1941
1942 for (i = 0; i < enic->rq_count; i++)
1943 netif_napi_del(&enic->napi[i]);
1944
1945 enic_free_vnic_resources(enic);
1946 enic_clear_intr_mode(enic);
1947 }
1948
1949 static int enic_dev_init(struct enic *enic)
1950 {
1951 struct device *dev = enic_get_dev(enic);
1952 struct net_device *netdev = enic->netdev;
1953 unsigned int i;
1954 int err;
1955
1956 /* Get interrupt coalesce timer info */
1957 err = enic_dev_intr_coal_timer_info(enic);
1958 if (err) {
1959 dev_warn(dev, "Using default conversion factor for "
1960 "interrupt coalesce timer\n");
1961 vnic_dev_intr_coal_timer_info_default(enic->vdev);
1962 }
1963
1964 /* Get vNIC configuration
1965 */
1966
1967 err = enic_get_vnic_config(enic);
1968 if (err) {
1969 dev_err(dev, "Get vNIC configuration failed, aborting\n");
1970 return err;
1971 }
1972
1973 /* Get available resource counts
1974 */
1975
1976 enic_get_res_counts(enic);
1977
1978 /* Set interrupt mode based on resource counts and system
1979 * capabilities
1980 */
1981
1982 err = enic_set_intr_mode(enic);
1983 if (err) {
1984 dev_err(dev, "Failed to set intr mode based on resource "
1985 "counts and system capabilities, aborting\n");
1986 return err;
1987 }
1988
1989 /* Allocate and configure vNIC resources
1990 */
1991
1992 err = enic_alloc_vnic_resources(enic);
1993 if (err) {
1994 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
1995 goto err_out_free_vnic_resources;
1996 }
1997
1998 enic_init_vnic_resources(enic);
1999
2000 err = enic_set_rss_nic_cfg(enic);
2001 if (err) {
2002 dev_err(dev, "Failed to config nic, aborting\n");
2003 goto err_out_free_vnic_resources;
2004 }
2005
2006 switch (vnic_dev_get_intr_mode(enic->vdev)) {
2007 default:
2008 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2009 break;
2010 case VNIC_DEV_INTR_MODE_MSIX:
2011 for (i = 0; i < enic->rq_count; i++)
2012 netif_napi_add(netdev, &enic->napi[i],
2013 enic_poll_msix, 64);
2014 break;
2015 }
2016
2017 return 0;
2018
2019 err_out_free_vnic_resources:
2020 enic_clear_intr_mode(enic);
2021 enic_free_vnic_resources(enic);
2022
2023 return err;
2024 }
2025
2026 static void enic_iounmap(struct enic *enic)
2027 {
2028 unsigned int i;
2029
2030 for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2031 if (enic->bar[i].vaddr)
2032 iounmap(enic->bar[i].vaddr);
2033 }
2034
2035 static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2036 {
2037 struct device *dev = &pdev->dev;
2038 struct net_device *netdev;
2039 struct enic *enic;
2040 int using_dac = 0;
2041 unsigned int i;
2042 int err;
2043 #ifdef CONFIG_PCI_IOV
2044 int pos = 0;
2045 #endif
2046 int num_pps = 1;
2047
2048 /* Allocate net device structure and initialize. Private
2049 * instance data is initialized to zero.
2050 */
2051
2052 netdev = alloc_etherdev_mqs(sizeof(struct enic),
2053 ENIC_RQ_MAX, ENIC_WQ_MAX);
2054 if (!netdev)
2055 return -ENOMEM;
2056
2057 pci_set_drvdata(pdev, netdev);
2058
2059 SET_NETDEV_DEV(netdev, &pdev->dev);
2060
2061 enic = netdev_priv(netdev);
2062 enic->netdev = netdev;
2063 enic->pdev = pdev;
2064
2065 /* Setup PCI resources
2066 */
2067
2068 err = pci_enable_device_mem(pdev);
2069 if (err) {
2070 dev_err(dev, "Cannot enable PCI device, aborting\n");
2071 goto err_out_free_netdev;
2072 }
2073
2074 err = pci_request_regions(pdev, DRV_NAME);
2075 if (err) {
2076 dev_err(dev, "Cannot request PCI regions, aborting\n");
2077 goto err_out_disable_device;
2078 }
2079
2080 pci_set_master(pdev);
2081
2082 /* Query PCI controller on system for DMA addressing
2083 * limitation for the device. Try 40-bit first, and
2084 * fail to 32-bit.
2085 */
2086
2087 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
2088 if (err) {
2089 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2090 if (err) {
2091 dev_err(dev, "No usable DMA configuration, aborting\n");
2092 goto err_out_release_regions;
2093 }
2094 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2095 if (err) {
2096 dev_err(dev, "Unable to obtain %u-bit DMA "
2097 "for consistent allocations, aborting\n", 32);
2098 goto err_out_release_regions;
2099 }
2100 } else {
2101 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
2102 if (err) {
2103 dev_err(dev, "Unable to obtain %u-bit DMA "
2104 "for consistent allocations, aborting\n", 40);
2105 goto err_out_release_regions;
2106 }
2107 using_dac = 1;
2108 }
2109
2110 /* Map vNIC resources from BAR0-5
2111 */
2112
2113 for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2114 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2115 continue;
2116 enic->bar[i].len = pci_resource_len(pdev, i);
2117 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2118 if (!enic->bar[i].vaddr) {
2119 dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2120 err = -ENODEV;
2121 goto err_out_iounmap;
2122 }
2123 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2124 }
2125
2126 /* Register vNIC device
2127 */
2128
2129 enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2130 ARRAY_SIZE(enic->bar));
2131 if (!enic->vdev) {
2132 dev_err(dev, "vNIC registration failed, aborting\n");
2133 err = -ENODEV;
2134 goto err_out_iounmap;
2135 }
2136
2137 #ifdef CONFIG_PCI_IOV
2138 /* Get number of subvnics */
2139 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
2140 if (pos) {
2141 pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
2142 &enic->num_vfs);
2143 if (enic->num_vfs) {
2144 err = pci_enable_sriov(pdev, enic->num_vfs);
2145 if (err) {
2146 dev_err(dev, "SRIOV enable failed, aborting."
2147 " pci_enable_sriov() returned %d\n",
2148 err);
2149 goto err_out_vnic_unregister;
2150 }
2151 enic->priv_flags |= ENIC_SRIOV_ENABLED;
2152 num_pps = enic->num_vfs;
2153 }
2154 }
2155 #endif
2156
2157 /* Allocate structure for port profiles */
2158 enic->pp = kcalloc(num_pps, sizeof(*enic->pp), GFP_KERNEL);
2159 if (!enic->pp) {
2160 err = -ENOMEM;
2161 goto err_out_disable_sriov_pp;
2162 }
2163
2164 /* Issue device open to get device in known state
2165 */
2166
2167 err = enic_dev_open(enic);
2168 if (err) {
2169 dev_err(dev, "vNIC dev open failed, aborting\n");
2170 goto err_out_disable_sriov;
2171 }
2172
2173 /* Setup devcmd lock
2174 */
2175
2176 spin_lock_init(&enic->devcmd_lock);
2177 spin_lock_init(&enic->enic_api_lock);
2178
2179 /*
2180 * Set ingress vlan rewrite mode before vnic initialization
2181 */
2182
2183 err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2184 if (err) {
2185 dev_err(dev,
2186 "Failed to set ingress vlan rewrite mode, aborting.\n");
2187 goto err_out_dev_close;
2188 }
2189
2190 /* Issue device init to initialize the vnic-to-switch link.
2191 * We'll start with carrier off and wait for link UP
2192 * notification later to turn on carrier. We don't need
2193 * to wait here for the vnic-to-switch link initialization
2194 * to complete; link UP notification is the indication that
2195 * the process is complete.
2196 */
2197
2198 netif_carrier_off(netdev);
2199
2200 /* Do not call dev_init for a dynamic vnic.
2201 * For a dynamic vnic, init_prov_info will be
2202 * called later by an upper layer.
2203 */
2204
2205 if (!enic_is_dynamic(enic)) {
2206 err = vnic_dev_init(enic->vdev, 0);
2207 if (err) {
2208 dev_err(dev, "vNIC dev init failed, aborting\n");
2209 goto err_out_dev_close;
2210 }
2211 }
2212
2213 err = enic_dev_init(enic);
2214 if (err) {
2215 dev_err(dev, "Device initialization failed, aborting\n");
2216 goto err_out_dev_close;
2217 }
2218
2219 netif_set_real_num_tx_queues(netdev, enic->wq_count);
2220 netif_set_real_num_rx_queues(netdev, enic->rq_count);
2221
2222 /* Setup notification timer, HW reset task, and wq locks
2223 */
2224
2225 init_timer(&enic->notify_timer);
2226 enic->notify_timer.function = enic_notify_timer;
2227 enic->notify_timer.data = (unsigned long)enic;
2228
2229 INIT_WORK(&enic->reset, enic_reset);
2230 INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
2231
2232 for (i = 0; i < enic->wq_count; i++)
2233 spin_lock_init(&enic->wq_lock[i]);
2234
2235 /* Register net device
2236 */
2237
2238 enic->port_mtu = enic->config.mtu;
2239 (void)enic_change_mtu(netdev, enic->port_mtu);
2240
2241 err = enic_set_mac_addr(netdev, enic->mac_addr);
2242 if (err) {
2243 dev_err(dev, "Invalid MAC address, aborting\n");
2244 goto err_out_dev_deinit;
2245 }
2246
2247 enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2248 enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2249
2250 if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))
2251 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2252 else
2253 netdev->netdev_ops = &enic_netdev_ops;
2254
2255 netdev->watchdog_timeo = 2 * HZ;
2256 enic_set_ethtool_ops(netdev);
2257
2258 netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2259 if (ENIC_SETTING(enic, LOOP)) {
2260 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2261 enic->loop_enable = 1;
2262 enic->loop_tag = enic->config.loop_tag;
2263 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2264 }
2265 if (ENIC_SETTING(enic, TXCSUM))
2266 netdev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2267 if (ENIC_SETTING(enic, TSO))
2268 netdev->hw_features |= NETIF_F_TSO |
2269 NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2270 if (ENIC_SETTING(enic, RSS))
2271 netdev->hw_features |= NETIF_F_RXHASH;
2272 if (ENIC_SETTING(enic, RXCSUM))
2273 netdev->hw_features |= NETIF_F_RXCSUM;
2274
2275 netdev->features |= netdev->hw_features;
2276
2277 if (using_dac)
2278 netdev->features |= NETIF_F_HIGHDMA;
2279
2280 netdev->priv_flags |= IFF_UNICAST_FLT;
2281
2282 err = register_netdev(netdev);
2283 if (err) {
2284 dev_err(dev, "Cannot register net device, aborting\n");
2285 goto err_out_dev_deinit;
2286 }
2287
2288 return 0;
2289
2290 err_out_dev_deinit:
2291 enic_dev_deinit(enic);
2292 err_out_dev_close:
2293 vnic_dev_close(enic->vdev);
2294 err_out_disable_sriov:
2295 kfree(enic->pp);
2296 err_out_disable_sriov_pp:
2297 #ifdef CONFIG_PCI_IOV
2298 if (enic_sriov_enabled(enic)) {
2299 pci_disable_sriov(pdev);
2300 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2301 }
2302 err_out_vnic_unregister:
2303 #endif
2304 vnic_dev_unregister(enic->vdev);
2305 err_out_iounmap:
2306 enic_iounmap(enic);
2307 err_out_release_regions:
2308 pci_release_regions(pdev);
2309 err_out_disable_device:
2310 pci_disable_device(pdev);
2311 err_out_free_netdev:
2312 pci_set_drvdata(pdev, NULL);
2313 free_netdev(netdev);
2314
2315 return err;
2316 }
2317
2318 static void enic_remove(struct pci_dev *pdev)
2319 {
2320 struct net_device *netdev = pci_get_drvdata(pdev);
2321
2322 if (netdev) {
2323 struct enic *enic = netdev_priv(netdev);
2324
2325 cancel_work_sync(&enic->reset);
2326 cancel_work_sync(&enic->change_mtu_work);
2327 unregister_netdev(netdev);
2328 enic_dev_deinit(enic);
2329 vnic_dev_close(enic->vdev);
2330 #ifdef CONFIG_PCI_IOV
2331 if (enic_sriov_enabled(enic)) {
2332 pci_disable_sriov(pdev);
2333 enic->priv_flags &= ~ENIC_SRIOV_ENABLED;
2334 }
2335 #endif
2336 kfree(enic->pp);
2337 vnic_dev_unregister(enic->vdev);
2338 enic_iounmap(enic);
2339 pci_release_regions(pdev);
2340 pci_disable_device(pdev);
2341 pci_set_drvdata(pdev, NULL);
2342 free_netdev(netdev);
2343 }
2344 }
2345
2346 static struct pci_driver enic_driver = {
2347 .name = DRV_NAME,
2348 .id_table = enic_id_table,
2349 .probe = enic_probe,
2350 .remove = enic_remove,
2351 };
2352
2353 static int __init enic_init_module(void)
2354 {
2355 pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2356
2357 return pci_register_driver(&enic_driver);
2358 }
2359
2360 static void __exit enic_cleanup_module(void)
2361 {
2362 pci_unregister_driver(&enic_driver);
2363 }
2364
2365 module_init(enic_init_module);
2366 module_exit(enic_cleanup_module);
Linux kernel - Deliverables
This page took 0.080933 seconds and 5 git commands to generate.