5d9bf4c13440921af3b0a54f490c6b14c4be035a
[dpdk.git] / drivers / net / enic / enic_main.c
1 /*
2  * Copyright 2008-2014 Cisco Systems, Inc.  All rights reserved.
3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
4  *
5  * Copyright (c) 2014, Cisco Systems, Inc.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * 1. Redistributions of source code must retain the above copyright
13  * notice, this list of conditions and the following disclaimer.
14  *
15  * 2. Redistributions in binary form must reproduce the above copyright
16  * notice, this list of conditions and the following disclaimer in
17  * the documentation and/or other materials provided with the
18  * distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  *
33  */
34
35 #include <stdio.h>
36
37 #include <sys/stat.h>
38 #include <sys/mman.h>
39 #include <fcntl.h>
40 #include <libgen.h>
41
42 #include <rte_pci.h>
43 #include <rte_memzone.h>
44 #include <rte_malloc.h>
45 #include <rte_mbuf.h>
46 #include <rte_string_fns.h>
47 #include <rte_ethdev.h>
48
49 #include "enic_compat.h"
50 #include "enic.h"
51 #include "wq_enet_desc.h"
52 #include "rq_enet_desc.h"
53 #include "cq_enet_desc.h"
54 #include "vnic_enet.h"
55 #include "vnic_dev.h"
56 #include "vnic_wq.h"
57 #include "vnic_rq.h"
58 #include "vnic_cq.h"
59 #include "vnic_intr.h"
60 #include "vnic_nic.h"
61
62 static inline int enic_is_sriov_vf(struct enic *enic)
63 {
64         return enic->pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
65 }
66
67 static int is_zero_addr(uint8_t *addr)
68 {
69         return !(addr[0] |  addr[1] | addr[2] | addr[3] | addr[4] | addr[5]);
70 }
71
72 static int is_mcast_addr(uint8_t *addr)
73 {
74         return addr[0] & 1;
75 }
76
77 static int is_eth_addr_valid(uint8_t *addr)
78 {
79         return !is_mcast_addr(addr) && !is_zero_addr(addr);
80 }
81
82 static void
83 enic_rxmbuf_queue_release(__rte_unused struct enic *enic, struct vnic_rq *rq)
84 {
85         uint16_t i;
86
87         if (!rq || !rq->mbuf_ring) {
88                 dev_debug(enic, "Pointer to rq or mbuf_ring is NULL");
89                 return;
90         }
91
92         for (i = 0; i < rq->ring.desc_count; i++) {
93                 if (rq->mbuf_ring[i]) {
94                         rte_pktmbuf_free_seg(rq->mbuf_ring[i]);
95                         rq->mbuf_ring[i] = NULL;
96                 }
97         }
98 }
99
100 void enic_set_hdr_split_size(struct enic *enic, u16 split_hdr_size)
101 {
102         vnic_set_hdr_split_size(enic->vdev, split_hdr_size);
103 }
104
105 static void enic_free_wq_buf(struct vnic_wq_buf *buf)
106 {
107         struct rte_mbuf *mbuf = (struct rte_mbuf *)buf->mb;
108
109         rte_pktmbuf_free_seg(mbuf);
110         buf->mb = NULL;
111 }
112
113 static void enic_log_q_error(struct enic *enic)
114 {
115         unsigned int i;
116         u32 error_status;
117
118         for (i = 0; i < enic->wq_count; i++) {
119                 error_status = vnic_wq_error_status(&enic->wq[i]);
120                 if (error_status)
121                         dev_err(enic, "WQ[%d] error_status %d\n", i,
122                                 error_status);
123         }
124
125         for (i = 0; i < enic_vnic_rq_count(enic); i++) {
126                 if (!enic->rq[i].in_use)
127                         continue;
128                 error_status = vnic_rq_error_status(&enic->rq[i]);
129                 if (error_status)
130                         dev_err(enic, "RQ[%d] error_status %d\n", i,
131                                 error_status);
132         }
133 }
134
135 static void enic_clear_soft_stats(struct enic *enic)
136 {
137         struct enic_soft_stats *soft_stats = &enic->soft_stats;
138         rte_atomic64_clear(&soft_stats->rx_nombuf);
139         rte_atomic64_clear(&soft_stats->rx_packet_errors);
140 }
141
142 static void enic_init_soft_stats(struct enic *enic)
143 {
144         struct enic_soft_stats *soft_stats = &enic->soft_stats;
145         rte_atomic64_init(&soft_stats->rx_nombuf);
146         rte_atomic64_init(&soft_stats->rx_packet_errors);
147         enic_clear_soft_stats(enic);
148 }
149
150 void enic_dev_stats_clear(struct enic *enic)
151 {
152         if (vnic_dev_stats_clear(enic->vdev))
153                 dev_err(enic, "Error in clearing stats\n");
154         enic_clear_soft_stats(enic);
155 }
156
157 void enic_dev_stats_get(struct enic *enic, struct rte_eth_stats *r_stats)
158 {
159         struct vnic_stats *stats;
160         struct enic_soft_stats *soft_stats = &enic->soft_stats;
161         int64_t rx_truncated;
162         uint64_t rx_packet_errors;
163
164         if (vnic_dev_stats_dump(enic->vdev, &stats)) {
165                 dev_err(enic, "Error in getting stats\n");
166                 return;
167         }
168
169         /* The number of truncated packets can only be calculated by
170          * subtracting a hardware counter from error packets received by
171          * the driver. Note: this causes transient inaccuracies in the
172          * ipackets count. Also, the length of truncated packets are
173          * counted in ibytes even though truncated packets are dropped
174          * which can make ibytes be slightly higher than it should be.
175          */
176         rx_packet_errors = rte_atomic64_read(&soft_stats->rx_packet_errors);
177         rx_truncated = rx_packet_errors - stats->rx.rx_errors -
178                 stats->rx.rx_no_bufs;
179
180         r_stats->ipackets = stats->rx.rx_frames_ok - rx_truncated;
181         r_stats->opackets = stats->tx.tx_frames_ok;
182
183         r_stats->ibytes = stats->rx.rx_bytes_ok;
184         r_stats->obytes = stats->tx.tx_bytes_ok;
185
186         r_stats->ierrors = stats->rx.rx_errors + stats->rx.rx_drop;
187         r_stats->oerrors = stats->tx.tx_errors;
188
189         r_stats->imissed = stats->rx.rx_no_bufs + rx_truncated;
190
191         r_stats->rx_nombuf = rte_atomic64_read(&soft_stats->rx_nombuf);
192 }
193
194 void enic_del_mac_address(struct enic *enic)
195 {
196         if (vnic_dev_del_addr(enic->vdev, enic->mac_addr))
197                 dev_err(enic, "del mac addr failed\n");
198 }
199
200 void enic_set_mac_address(struct enic *enic, uint8_t *mac_addr)
201 {
202         int err;
203
204         if (!is_eth_addr_valid(mac_addr)) {
205                 dev_err(enic, "invalid mac address\n");
206                 return;
207         }
208
209         err = vnic_dev_del_addr(enic->vdev, enic->mac_addr);
210         if (err) {
211                 dev_err(enic, "del mac addr failed\n");
212                 return;
213         }
214
215         ether_addr_copy((struct ether_addr *)mac_addr,
216                 (struct ether_addr *)enic->mac_addr);
217
218         err = vnic_dev_add_addr(enic->vdev, mac_addr);
219         if (err) {
220                 dev_err(enic, "add mac addr failed\n");
221                 return;
222         }
223 }
224
225 static void
226 enic_free_rq_buf(struct rte_mbuf **mbuf)
227 {
228         if (*mbuf == NULL)
229                 return;
230
231         rte_pktmbuf_free(*mbuf);
232         mbuf = NULL;
233 }
234
235 void enic_init_vnic_resources(struct enic *enic)
236 {
237         unsigned int error_interrupt_enable = 1;
238         unsigned int error_interrupt_offset = 0;
239         unsigned int index = 0;
240         unsigned int cq_idx;
241         struct vnic_rq *data_rq;
242
243         for (index = 0; index < enic->rq_count; index++) {
244                 cq_idx = enic_cq_rq(enic, enic_sop_rq(index));
245
246                 vnic_rq_init(&enic->rq[enic_sop_rq(index)],
247                         cq_idx,
248                         error_interrupt_enable,
249                         error_interrupt_offset);
250
251                 data_rq = &enic->rq[enic_data_rq(index)];
252                 if (data_rq->in_use)
253                         vnic_rq_init(data_rq,
254                                      cq_idx,
255                                      error_interrupt_enable,
256                                      error_interrupt_offset);
257
258                 vnic_cq_init(&enic->cq[cq_idx],
259                         0 /* flow_control_enable */,
260                         1 /* color_enable */,
261                         0 /* cq_head */,
262                         0 /* cq_tail */,
263                         1 /* cq_tail_color */,
264                         0 /* interrupt_enable */,
265                         1 /* cq_entry_enable */,
266                         0 /* cq_message_enable */,
267                         0 /* interrupt offset */,
268                         0 /* cq_message_addr */);
269         }
270
271         for (index = 0; index < enic->wq_count; index++) {
272                 vnic_wq_init(&enic->wq[index],
273                         enic_cq_wq(enic, index),
274                         error_interrupt_enable,
275                         error_interrupt_offset);
276
277                 cq_idx = enic_cq_wq(enic, index);
278                 vnic_cq_init(&enic->cq[cq_idx],
279                         0 /* flow_control_enable */,
280                         1 /* color_enable */,
281                         0 /* cq_head */,
282                         0 /* cq_tail */,
283                         1 /* cq_tail_color */,
284                         0 /* interrupt_enable */,
285                         0 /* cq_entry_enable */,
286                         1 /* cq_message_enable */,
287                         0 /* interrupt offset */,
288                         (u64)enic->wq[index].cqmsg_rz->phys_addr);
289         }
290
291         vnic_intr_init(&enic->intr,
292                 enic->config.intr_timer_usec,
293                 enic->config.intr_timer_type,
294                 /*mask_on_assertion*/1);
295 }
296
297
298 static int
299 enic_alloc_rx_queue_mbufs(struct enic *enic, struct vnic_rq *rq)
300 {
301         struct rte_mbuf *mb;
302         struct rq_enet_desc *rqd = rq->ring.descs;
303         unsigned i;
304         dma_addr_t dma_addr;
305
306         if (!rq->in_use)
307                 return 0;
308
309         dev_debug(enic, "queue %u, allocating %u rx queue mbufs\n", rq->index,
310                   rq->ring.desc_count);
311
312         for (i = 0; i < rq->ring.desc_count; i++, rqd++) {
313                 mb = rte_mbuf_raw_alloc(rq->mp);
314                 if (mb == NULL) {
315                         dev_err(enic, "RX mbuf alloc failed queue_id=%u\n",
316                         (unsigned)rq->index);
317                         return -ENOMEM;
318                 }
319
320                 mb->data_off = RTE_PKTMBUF_HEADROOM;
321                 dma_addr = (dma_addr_t)(mb->buf_physaddr
322                            + RTE_PKTMBUF_HEADROOM);
323                 rq_enet_desc_enc(rqd, dma_addr,
324                                 (rq->is_sop ? RQ_ENET_TYPE_ONLY_SOP
325                                 : RQ_ENET_TYPE_NOT_SOP),
326                                 mb->buf_len - RTE_PKTMBUF_HEADROOM);
327                 rq->mbuf_ring[i] = mb;
328         }
329
330         /* make sure all prior writes are complete before doing the PIO write */
331         rte_rmb();
332
333         /* Post all but the last buffer to VIC. */
334         rq->posted_index = rq->ring.desc_count - 1;
335
336         rq->rx_nb_hold = 0;
337
338         dev_debug(enic, "port=%u, qidx=%u, Write %u posted idx, %u sw held\n",
339                 enic->port_id, rq->index, rq->posted_index, rq->rx_nb_hold);
340         iowrite32(rq->posted_index, &rq->ctrl->posted_index);
341         iowrite32(0, &rq->ctrl->fetch_index);
342         rte_rmb();
343
344         return 0;
345
346 }
347
348 static void *
349 enic_alloc_consistent(void *priv, size_t size,
350         dma_addr_t *dma_handle, u8 *name)
351 {
352         void *vaddr;
353         const struct rte_memzone *rz;
354         *dma_handle = 0;
355         struct enic *enic = (struct enic *)priv;
356         struct enic_memzone_entry *mze;
357
358         rz = rte_memzone_reserve_aligned((const char *)name,
359                                          size, SOCKET_ID_ANY, 0, ENIC_ALIGN);
360         if (!rz) {
361                 pr_err("%s : Failed to allocate memory requested for %s\n",
362                         __func__, name);
363                 return NULL;
364         }
365
366         vaddr = rz->addr;
367         *dma_handle = (dma_addr_t)rz->phys_addr;
368
369         mze = rte_malloc("enic memzone entry",
370                          sizeof(struct enic_memzone_entry), 0);
371
372         if (!mze) {
373                 pr_err("%s : Failed to allocate memory for memzone list\n",
374                        __func__);
375                 rte_memzone_free(rz);
376         }
377
378         mze->rz = rz;
379
380         rte_spinlock_lock(&enic->memzone_list_lock);
381         LIST_INSERT_HEAD(&enic->memzone_list, mze, entries);
382         rte_spinlock_unlock(&enic->memzone_list_lock);
383
384         return vaddr;
385 }
386
387 static void
388 enic_free_consistent(void *priv,
389                      __rte_unused size_t size,
390                      void *vaddr,
391                      dma_addr_t dma_handle)
392 {
393         struct enic_memzone_entry *mze;
394         struct enic *enic = (struct enic *)priv;
395
396         rte_spinlock_lock(&enic->memzone_list_lock);
397         LIST_FOREACH(mze, &enic->memzone_list, entries) {
398                 if (mze->rz->addr == vaddr &&
399                     mze->rz->phys_addr == dma_handle)
400                         break;
401         }
402         if (mze == NULL) {
403                 rte_spinlock_unlock(&enic->memzone_list_lock);
404                 dev_warning(enic,
405                             "Tried to free memory, but couldn't find it in the memzone list\n");
406                 return;
407         }
408         LIST_REMOVE(mze, entries);
409         rte_spinlock_unlock(&enic->memzone_list_lock);
410         rte_memzone_free(mze->rz);
411         rte_free(mze);
412 }
413
414 int enic_link_update(struct enic *enic)
415 {
416         struct rte_eth_dev *eth_dev = enic->rte_dev;
417         int ret;
418         int link_status = 0;
419
420         link_status = enic_get_link_status(enic);
421         ret = (link_status == enic->link_status);
422         enic->link_status = link_status;
423         eth_dev->data->dev_link.link_status = link_status;
424         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
425         eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
426         return ret;
427 }
428
429 static void
430 enic_intr_handler(__rte_unused struct rte_intr_handle *handle,
431         void *arg)
432 {
433         struct rte_eth_dev *dev = (struct rte_eth_dev *)arg;
434         struct enic *enic = pmd_priv(dev);
435
436         vnic_intr_return_all_credits(&enic->intr);
437
438         enic_link_update(enic);
439         _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC);
440         enic_log_q_error(enic);
441 }
442
443 int enic_enable(struct enic *enic)
444 {
445         unsigned int index;
446         int err;
447         struct rte_eth_dev *eth_dev = enic->rte_dev;
448
449         eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
450         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
451
452         /* vnic notification of link status has already been turned on in
453          * enic_dev_init() which is called during probe time.  Here we are
454          * just turning on interrupt vector 0 if needed.
455          */
456         if (eth_dev->data->dev_conf.intr_conf.lsc)
457                 vnic_dev_notify_set(enic->vdev, 0);
458
459         if (enic_clsf_init(enic))
460                 dev_warning(enic, "Init of hash table for clsf failed."\
461                         "Flow director feature will not work\n");
462
463         for (index = 0; index < enic->rq_count; index++) {
464                 err = enic_alloc_rx_queue_mbufs(enic,
465                         &enic->rq[enic_sop_rq(index)]);
466                 if (err) {
467                         dev_err(enic, "Failed to alloc sop RX queue mbufs\n");
468                         return err;
469                 }
470                 err = enic_alloc_rx_queue_mbufs(enic,
471                         &enic->rq[enic_data_rq(index)]);
472                 if (err) {
473                         /* release the allocated mbufs for the sop rq*/
474                         enic_rxmbuf_queue_release(enic,
475                                 &enic->rq[enic_sop_rq(index)]);
476
477                         dev_err(enic, "Failed to alloc data RX queue mbufs\n");
478                         return err;
479                 }
480         }
481
482         for (index = 0; index < enic->wq_count; index++)
483                 enic_start_wq(enic, index);
484         for (index = 0; index < enic->rq_count; index++)
485                 enic_start_rq(enic, index);
486
487         vnic_dev_add_addr(enic->vdev, enic->mac_addr);
488
489         vnic_dev_enable_wait(enic->vdev);
490
491         /* Register and enable error interrupt */
492         rte_intr_callback_register(&(enic->pdev->intr_handle),
493                 enic_intr_handler, (void *)enic->rte_dev);
494
495         rte_intr_enable(&(enic->pdev->intr_handle));
496         vnic_intr_unmask(&enic->intr);
497
498         return 0;
499 }
500
501 int enic_alloc_intr_resources(struct enic *enic)
502 {
503         int err;
504
505         dev_info(enic, "vNIC resources used:  "\
506                 "wq %d rq %d cq %d intr %d\n",
507                 enic->wq_count, enic_vnic_rq_count(enic),
508                 enic->cq_count, enic->intr_count);
509
510         err = vnic_intr_alloc(enic->vdev, &enic->intr, 0);
511         if (err)
512                 enic_free_vnic_resources(enic);
513
514         return err;
515 }
516
517 void enic_free_rq(void *rxq)
518 {
519         struct vnic_rq *rq_sop, *rq_data;
520         struct enic *enic;
521
522         if (rxq == NULL)
523                 return;
524
525         rq_sop = (struct vnic_rq *)rxq;
526         enic = vnic_dev_priv(rq_sop->vdev);
527         rq_data = &enic->rq[rq_sop->data_queue_idx];
528
529         enic_rxmbuf_queue_release(enic, rq_sop);
530         if (rq_data->in_use)
531                 enic_rxmbuf_queue_release(enic, rq_data);
532
533         rte_free(rq_sop->mbuf_ring);
534         if (rq_data->in_use)
535                 rte_free(rq_data->mbuf_ring);
536
537         rq_sop->mbuf_ring = NULL;
538         rq_data->mbuf_ring = NULL;
539
540         vnic_rq_free(rq_sop);
541         if (rq_data->in_use)
542                 vnic_rq_free(rq_data);
543
544         vnic_cq_free(&enic->cq[rq_sop->index]);
545 }
546
547 void enic_start_wq(struct enic *enic, uint16_t queue_idx)
548 {
549         struct rte_eth_dev *eth_dev = enic->rte_dev;
550         vnic_wq_enable(&enic->wq[queue_idx]);
551         eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
552 }
553
554 int enic_stop_wq(struct enic *enic, uint16_t queue_idx)
555 {
556         struct rte_eth_dev *eth_dev = enic->rte_dev;
557         int ret;
558
559         ret = vnic_wq_disable(&enic->wq[queue_idx]);
560         if (ret)
561                 return ret;
562
563         eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
564         return 0;
565 }
566
567 void enic_start_rq(struct enic *enic, uint16_t queue_idx)
568 {
569         struct vnic_rq *rq_sop = &enic->rq[enic_sop_rq(queue_idx)];
570         struct vnic_rq *rq_data = &enic->rq[rq_sop->data_queue_idx];
571         struct rte_eth_dev *eth_dev = enic->rte_dev;
572
573         if (rq_data->in_use)
574                 vnic_rq_enable(rq_data);
575         rte_mb();
576         vnic_rq_enable(rq_sop);
577         eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
578 }
579
580 int enic_stop_rq(struct enic *enic, uint16_t queue_idx)
581 {
582         int ret1 = 0, ret2 = 0;
583         struct rte_eth_dev *eth_dev = enic->rte_dev;
584         struct vnic_rq *rq_sop = &enic->rq[enic_sop_rq(queue_idx)];
585         struct vnic_rq *rq_data = &enic->rq[rq_sop->data_queue_idx];
586
587         ret2 = vnic_rq_disable(rq_sop);
588         rte_mb();
589         if (rq_data->in_use)
590                 ret1 = vnic_rq_disable(rq_data);
591
592         if (ret2)
593                 return ret2;
594         else if (ret1)
595                 return ret1;
596
597         eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
598         return 0;
599 }
600
601 int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
602         unsigned int socket_id, struct rte_mempool *mp,
603         uint16_t nb_desc)
604 {
605         int rc;
606         uint16_t sop_queue_idx = enic_sop_rq(queue_idx);
607         uint16_t data_queue_idx = enic_data_rq(queue_idx);
608         struct vnic_rq *rq_sop = &enic->rq[sop_queue_idx];
609         struct vnic_rq *rq_data = &enic->rq[data_queue_idx];
610         unsigned int mbuf_size, mbufs_per_pkt;
611         unsigned int nb_sop_desc, nb_data_desc;
612         uint16_t min_sop, max_sop, min_data, max_data;
613
614         rq_sop->is_sop = 1;
615         rq_sop->data_queue_idx = data_queue_idx;
616         rq_data->is_sop = 0;
617         rq_data->data_queue_idx = 0;
618         rq_sop->socket_id = socket_id;
619         rq_sop->mp = mp;
620         rq_data->socket_id = socket_id;
621         rq_data->mp = mp;
622         rq_sop->in_use = 1;
623
624         mbuf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
625                                RTE_PKTMBUF_HEADROOM);
626
627         if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
628                 dev_info(enic, "Scatter rx mode enabled\n");
629                 /* ceil((mtu + ETHER_HDR_LEN + 4)/mbuf_size) */
630                 mbufs_per_pkt = ((enic->config.mtu + ETHER_HDR_LEN + 4) +
631                                  (mbuf_size - 1)) / mbuf_size;
632         } else {
633                 dev_info(enic, "Scatter rx mode disabled\n");
634                 mbufs_per_pkt = 1;
635         }
636
637         if (mbufs_per_pkt > 1) {
638                 dev_info(enic, "Scatter rx mode in use\n");
639                 rq_data->in_use = 1;
640         } else {
641                 dev_info(enic, "Scatter rx mode not being used\n");
642                 rq_data->in_use = 0;
643         }
644
645         /* number of descriptors have to be a multiple of 32 */
646         nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
647         nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
648
649         rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
650         rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
651
652         if (mbufs_per_pkt > 1) {
653                 min_sop = 64;
654                 max_sop = ((enic->config.rq_desc_count /
655                             (mbufs_per_pkt - 1)) & ~0x1F);
656                 min_data = min_sop * (mbufs_per_pkt - 1);
657                 max_data = enic->config.rq_desc_count;
658         } else {
659                 min_sop = 64;
660                 max_sop = enic->config.rq_desc_count;
661                 min_data = 0;
662                 max_data = 0;
663         }
664
665         if (nb_desc < (min_sop + min_data)) {
666                 dev_warning(enic,
667                             "Number of rx descs too low, adjusting to minimum\n");
668                 nb_sop_desc = min_sop;
669                 nb_data_desc = min_data;
670         } else if (nb_desc > (max_sop + max_data)) {
671                 dev_warning(enic,
672                             "Number of rx_descs too high, adjusting to maximum\n");
673                 nb_sop_desc = max_sop;
674                 nb_data_desc = max_data;
675         }
676         if (mbufs_per_pkt > 1) {
677                 dev_info(enic, "For mtu %d and mbuf size %d valid rx descriptor range is %d to %d\n",
678                          enic->config.mtu, mbuf_size, min_sop + min_data,
679                          max_sop + max_data);
680         }
681         dev_info(enic, "Using %d rx descriptors (sop %d, data %d)\n",
682                  nb_sop_desc + nb_data_desc, nb_sop_desc, nb_data_desc);
683
684         /* Allocate sop queue resources */
685         rc = vnic_rq_alloc(enic->vdev, rq_sop, sop_queue_idx,
686                 nb_sop_desc, sizeof(struct rq_enet_desc));
687         if (rc) {
688                 dev_err(enic, "error in allocation of sop rq\n");
689                 goto err_exit;
690         }
691         nb_sop_desc = rq_sop->ring.desc_count;
692
693         if (rq_data->in_use) {
694                 /* Allocate data queue resources */
695                 rc = vnic_rq_alloc(enic->vdev, rq_data, data_queue_idx,
696                                    nb_data_desc,
697                                    sizeof(struct rq_enet_desc));
698                 if (rc) {
699                         dev_err(enic, "error in allocation of data rq\n");
700                         goto err_free_rq_sop;
701                 }
702                 nb_data_desc = rq_data->ring.desc_count;
703         }
704         rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
705                            socket_id, nb_sop_desc + nb_data_desc,
706                            sizeof(struct cq_enet_rq_desc));
707         if (rc) {
708                 dev_err(enic, "error in allocation of cq for rq\n");
709                 goto err_free_rq_data;
710         }
711
712         /* Allocate the mbuf rings */
713         rq_sop->mbuf_ring = (struct rte_mbuf **)
714                 rte_zmalloc_socket("rq->mbuf_ring",
715                                    sizeof(struct rte_mbuf *) * nb_sop_desc,
716                                    RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
717         if (rq_sop->mbuf_ring == NULL)
718                 goto err_free_cq;
719
720         if (rq_data->in_use) {
721                 rq_data->mbuf_ring = (struct rte_mbuf **)
722                         rte_zmalloc_socket("rq->mbuf_ring",
723                                 sizeof(struct rte_mbuf *) * nb_data_desc,
724                                 RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
725                 if (rq_data->mbuf_ring == NULL)
726                         goto err_free_sop_mbuf;
727         }
728
729         return 0;
730
731 err_free_sop_mbuf:
732         rte_free(rq_sop->mbuf_ring);
733 err_free_cq:
734         /* cleanup on error */
735         vnic_cq_free(&enic->cq[queue_idx]);
736 err_free_rq_data:
737         if (rq_data->in_use)
738                 vnic_rq_free(rq_data);
739 err_free_rq_sop:
740         vnic_rq_free(rq_sop);
741 err_exit:
742         return -ENOMEM;
743 }
744
745 void enic_free_wq(void *txq)
746 {
747         struct vnic_wq *wq;
748         struct enic *enic;
749
750         if (txq == NULL)
751                 return;
752
753         wq = (struct vnic_wq *)txq;
754         enic = vnic_dev_priv(wq->vdev);
755         rte_memzone_free(wq->cqmsg_rz);
756         vnic_wq_free(wq);
757         vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
758 }
759
760 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
761         unsigned int socket_id, uint16_t nb_desc)
762 {
763         int err;
764         struct vnic_wq *wq = &enic->wq[queue_idx];
765         unsigned int cq_index = enic_cq_wq(enic, queue_idx);
766         char name[NAME_MAX];
767         static int instance;
768
769         wq->socket_id = socket_id;
770         if (nb_desc) {
771                 if (nb_desc > enic->config.wq_desc_count) {
772                         dev_warning(enic,
773                                 "WQ %d - number of tx desc in cmd line (%d)"\
774                                 "is greater than that in the UCSM/CIMC adapter"\
775                                 "policy.  Applying the value in the adapter "\
776                                 "policy (%d)\n",
777                                 queue_idx, nb_desc, enic->config.wq_desc_count);
778                 } else if (nb_desc != enic->config.wq_desc_count) {
779                         enic->config.wq_desc_count = nb_desc;
780                         dev_info(enic,
781                                 "TX Queues - effective number of descs:%d\n",
782                                 nb_desc);
783                 }
784         }
785
786         /* Allocate queue resources */
787         err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
788                 enic->config.wq_desc_count,
789                 sizeof(struct wq_enet_desc));
790         if (err) {
791                 dev_err(enic, "error in allocation of wq\n");
792                 return err;
793         }
794
795         err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
796                 socket_id, enic->config.wq_desc_count,
797                 sizeof(struct cq_enet_wq_desc));
798         if (err) {
799                 vnic_wq_free(wq);
800                 dev_err(enic, "error in allocation of cq for wq\n");
801         }
802
803         /* setup up CQ message */
804         snprintf((char *)name, sizeof(name),
805                  "vnic_cqmsg-%s-%d-%d", enic->bdf_name, queue_idx,
806                 instance++);
807
808         wq->cqmsg_rz = rte_memzone_reserve_aligned((const char *)name,
809                                                    sizeof(uint32_t),
810                                                    SOCKET_ID_ANY, 0,
811                                                    ENIC_ALIGN);
812         if (!wq->cqmsg_rz)
813                 return -ENOMEM;
814
815         return err;
816 }
817
818 int enic_disable(struct enic *enic)
819 {
820         unsigned int i;
821         int err;
822
823         vnic_intr_mask(&enic->intr);
824         (void)vnic_intr_masked(&enic->intr); /* flush write */
825         rte_intr_disable(&enic->pdev->intr_handle);
826         rte_intr_callback_unregister(&enic->pdev->intr_handle,
827                                      enic_intr_handler,
828                                      (void *)enic->rte_dev);
829
830         vnic_dev_disable(enic->vdev);
831
832         enic_clsf_destroy(enic);
833
834         if (!enic_is_sriov_vf(enic))
835                 vnic_dev_del_addr(enic->vdev, enic->mac_addr);
836
837         for (i = 0; i < enic->wq_count; i++) {
838                 err = vnic_wq_disable(&enic->wq[i]);
839                 if (err)
840                         return err;
841         }
842         for (i = 0; i < enic_vnic_rq_count(enic); i++) {
843                 if (enic->rq[i].in_use) {
844                         err = vnic_rq_disable(&enic->rq[i]);
845                         if (err)
846                                 return err;
847                 }
848         }
849
850         /* If we were using interrupts, set the interrupt vector to -1
851          * to disable interrupts.  We are not disabling link notifcations,
852          * though, as we want the polling of link status to continue working.
853          */
854         if (enic->rte_dev->data->dev_conf.intr_conf.lsc)
855                 vnic_dev_notify_set(enic->vdev, -1);
856
857         vnic_dev_set_reset_flag(enic->vdev, 1);
858
859         for (i = 0; i < enic->wq_count; i++)
860                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
861
862         for (i = 0; i < enic_vnic_rq_count(enic); i++)
863                 if (enic->rq[i].in_use)
864                         vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
865         for (i = 0; i < enic->cq_count; i++)
866                 vnic_cq_clean(&enic->cq[i]);
867         vnic_intr_clean(&enic->intr);
868
869         return 0;
870 }
871
872 static int enic_dev_wait(struct vnic_dev *vdev,
873         int (*start)(struct vnic_dev *, int),
874         int (*finished)(struct vnic_dev *, int *),
875         int arg)
876 {
877         int done;
878         int err;
879         int i;
880
881         err = start(vdev, arg);
882         if (err)
883                 return err;
884
885         /* Wait for func to complete...2 seconds max */
886         for (i = 0; i < 2000; i++) {
887                 err = finished(vdev, &done);
888                 if (err)
889                         return err;
890                 if (done)
891                         return 0;
892                 usleep(1000);
893         }
894         return -ETIMEDOUT;
895 }
896
897 static int enic_dev_open(struct enic *enic)
898 {
899         int err;
900
901         err = enic_dev_wait(enic->vdev, vnic_dev_open,
902                 vnic_dev_open_done, 0);
903         if (err)
904                 dev_err(enic_get_dev(enic),
905                         "vNIC device open failed, err %d\n", err);
906
907         return err;
908 }
909
910 static int enic_set_rsskey(struct enic *enic)
911 {
912         dma_addr_t rss_key_buf_pa;
913         union vnic_rss_key *rss_key_buf_va = NULL;
914         static union vnic_rss_key rss_key = {
915                 .key = {
916                         [0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
917                         [1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
918                         [2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
919                         [3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
920                 }
921         };
922         int err;
923         u8 name[NAME_MAX];
924
925         snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
926         rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
927                 &rss_key_buf_pa, name);
928         if (!rss_key_buf_va)
929                 return -ENOMEM;
930
931         rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
932
933         err = enic_set_rss_key(enic,
934                 rss_key_buf_pa,
935                 sizeof(union vnic_rss_key));
936
937         enic_free_consistent(enic, sizeof(union vnic_rss_key),
938                 rss_key_buf_va, rss_key_buf_pa);
939
940         return err;
941 }
942
943 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
944 {
945         dma_addr_t rss_cpu_buf_pa;
946         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
947         int i;
948         int err;
949         u8 name[NAME_MAX];
950
951         snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
952         rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
953                 &rss_cpu_buf_pa, name);
954         if (!rss_cpu_buf_va)
955                 return -ENOMEM;
956
957         for (i = 0; i < (1 << rss_hash_bits); i++)
958                 (*rss_cpu_buf_va).cpu[i / 4].b[i % 4] =
959                         enic_sop_rq(i % enic->rq_count);
960
961         err = enic_set_rss_cpu(enic,
962                 rss_cpu_buf_pa,
963                 sizeof(union vnic_rss_cpu));
964
965         enic_free_consistent(enic, sizeof(union vnic_rss_cpu),
966                 rss_cpu_buf_va, rss_cpu_buf_pa);
967
968         return err;
969 }
970
971 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
972         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
973 {
974         const u8 tso_ipid_split_en = 0;
975         int err;
976
977         /* Enable VLAN tag stripping */
978
979         err = enic_set_nic_cfg(enic,
980                 rss_default_cpu, rss_hash_type,
981                 rss_hash_bits, rss_base_cpu,
982                 rss_enable, tso_ipid_split_en,
983                 enic->ig_vlan_strip_en);
984
985         return err;
986 }
987
988 int enic_set_rss_nic_cfg(struct enic *enic)
989 {
990         const u8 rss_default_cpu = 0;
991         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
992             NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
993             NIC_CFG_RSS_HASH_TYPE_IPV6 |
994             NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
995         const u8 rss_hash_bits = 7;
996         const u8 rss_base_cpu = 0;
997         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
998
999         if (rss_enable) {
1000                 if (!enic_set_rsskey(enic)) {
1001                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
1002                                 rss_enable = 0;
1003                                 dev_warning(enic, "RSS disabled, "\
1004                                         "Failed to set RSS cpu indirection table.");
1005                         }
1006                 } else {
1007                         rss_enable = 0;
1008                         dev_warning(enic,
1009                                 "RSS disabled, Failed to set RSS key.\n");
1010                 }
1011         }
1012
1013         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1014                 rss_hash_bits, rss_base_cpu, rss_enable);
1015 }
1016
1017 int enic_setup_finish(struct enic *enic)
1018 {
1019         int ret;
1020
1021         enic_init_soft_stats(enic);
1022
1023         ret = enic_set_rss_nic_cfg(enic);
1024         if (ret) {
1025                 dev_err(enic, "Failed to config nic, aborting.\n");
1026                 return -1;
1027         }
1028
1029         /* Default conf */
1030         vnic_dev_packet_filter(enic->vdev,
1031                 1 /* directed  */,
1032                 1 /* multicast */,
1033                 1 /* broadcast */,
1034                 0 /* promisc   */,
1035                 1 /* allmulti  */);
1036
1037         enic->promisc = 0;
1038         enic->allmulti = 1;
1039
1040         return 0;
1041 }
1042
1043 void enic_add_packet_filter(struct enic *enic)
1044 {
1045         /* Args -> directed, multicast, broadcast, promisc, allmulti */
1046         vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
1047                 enic->promisc, enic->allmulti);
1048 }
1049
1050 int enic_get_link_status(struct enic *enic)
1051 {
1052         return vnic_dev_link_status(enic->vdev);
1053 }
1054
1055 static void enic_dev_deinit(struct enic *enic)
1056 {
1057         struct rte_eth_dev *eth_dev = enic->rte_dev;
1058
1059         /* stop link status checking */
1060         vnic_dev_notify_unset(enic->vdev);
1061
1062         rte_free(eth_dev->data->mac_addrs);
1063 }
1064
1065
1066 int enic_set_vnic_res(struct enic *enic)
1067 {
1068         struct rte_eth_dev *eth_dev = enic->rte_dev;
1069         int rc = 0;
1070
1071         /* With Rx scatter support, two RQs are now used per RQ used by
1072          * the application.
1073          */
1074         if (enic->conf_rq_count < eth_dev->data->nb_rx_queues) {
1075                 dev_err(dev, "Not enough Receive queues. Requested:%u which uses %d RQs on VIC, Configured:%u\n",
1076                         eth_dev->data->nb_rx_queues,
1077                         eth_dev->data->nb_rx_queues * 2, enic->conf_rq_count);
1078                 rc = -EINVAL;
1079         }
1080         if (enic->conf_wq_count < eth_dev->data->nb_tx_queues) {
1081                 dev_err(dev, "Not enough Transmit queues. Requested:%u, Configured:%u\n",
1082                         eth_dev->data->nb_tx_queues, enic->conf_wq_count);
1083                 rc = -EINVAL;
1084         }
1085
1086         if (enic->conf_cq_count < (eth_dev->data->nb_rx_queues +
1087                                    eth_dev->data->nb_tx_queues)) {
1088                 dev_err(dev, "Not enough Completion queues. Required:%u, Configured:%u\n",
1089                         (eth_dev->data->nb_rx_queues +
1090                          eth_dev->data->nb_tx_queues), enic->conf_cq_count);
1091                 rc = -EINVAL;
1092         }
1093
1094         if (rc == 0) {
1095                 enic->rq_count = eth_dev->data->nb_rx_queues;
1096                 enic->wq_count = eth_dev->data->nb_tx_queues;
1097                 enic->cq_count = enic->rq_count + enic->wq_count;
1098         }
1099
1100         return rc;
1101 }
1102
1103 /* The Cisco NIC can send and receive packets up to a max packet size
1104  * determined by the NIC type and firmware. There is also an MTU
1105  * configured into the NIC via the CIMC/UCSM management interface
1106  * which can be overridden by this function (up to the max packet size).
1107  * Depending on the network setup, doing so may cause packet drops
1108  * and unexpected behavior.
1109  */
1110 int enic_set_mtu(struct enic *enic, uint16_t new_mtu)
1111 {
1112         uint16_t old_mtu;       /* previous setting */
1113         uint16_t config_mtu;    /* Value configured into NIC via CIMC/UCSM */
1114         struct rte_eth_dev *eth_dev = enic->rte_dev;
1115
1116         old_mtu = eth_dev->data->mtu;
1117         config_mtu = enic->config.mtu;
1118
1119         /* only works with Rx scatter disabled */
1120         if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter)
1121                 return -ENOTSUP;
1122
1123         if (new_mtu > enic->max_mtu) {
1124                 dev_err(enic,
1125                         "MTU not updated: requested (%u) greater than max (%u)\n",
1126                         new_mtu, enic->max_mtu);
1127                 return -EINVAL;
1128         }
1129         if (new_mtu < ENIC_MIN_MTU) {
1130                 dev_info(enic,
1131                         "MTU not updated: requested (%u) less than min (%u)\n",
1132                         new_mtu, ENIC_MIN_MTU);
1133                 return -EINVAL;
1134         }
1135         if (new_mtu > config_mtu)
1136                 dev_warning(enic,
1137                         "MTU (%u) is greater than value configured in NIC (%u)\n",
1138                         new_mtu, config_mtu);
1139
1140         /* update the mtu */
1141         eth_dev->data->mtu = new_mtu;
1142
1143         dev_info(enic, "MTU changed from %u to %u\n",  old_mtu, new_mtu);
1144         return 0;
1145 }
1146
1147 static int enic_dev_init(struct enic *enic)
1148 {
1149         int err;
1150         struct rte_eth_dev *eth_dev = enic->rte_dev;
1151
1152         vnic_dev_intr_coal_timer_info_default(enic->vdev);
1153
1154         /* Get vNIC configuration
1155         */
1156         err = enic_get_vnic_config(enic);
1157         if (err) {
1158                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
1159                 return err;
1160         }
1161
1162         eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN, 0);
1163         if (!eth_dev->data->mac_addrs) {
1164                 dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1165                 return -1;
1166         }
1167         ether_addr_copy((struct ether_addr *) enic->mac_addr,
1168                 &eth_dev->data->mac_addrs[0]);
1169
1170
1171         /* Get available resource counts
1172         */
1173         enic_get_res_counts(enic);
1174
1175         vnic_dev_set_reset_flag(enic->vdev, 0);
1176
1177         /* set up link status checking */
1178         vnic_dev_notify_set(enic->vdev, -1); /* No Intr for notify */
1179
1180         return 0;
1181
1182 }
1183
1184 int enic_probe(struct enic *enic)
1185 {
1186         struct rte_pci_device *pdev = enic->pdev;
1187         int err = -1;
1188
1189         dev_debug(enic, " Initializing ENIC PMD\n");
1190
1191         enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1192         enic->bar0.len = pdev->mem_resource[0].len;
1193
1194         /* Register vNIC device */
1195         enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1196         if (!enic->vdev) {
1197                 dev_err(enic, "vNIC registration failed, aborting\n");
1198                 goto err_out;
1199         }
1200
1201         LIST_INIT(&enic->memzone_list);
1202         rte_spinlock_init(&enic->memzone_list_lock);
1203
1204         vnic_register_cbacks(enic->vdev,
1205                 enic_alloc_consistent,
1206                 enic_free_consistent);
1207
1208         /* Issue device open to get device in known state */
1209         err = enic_dev_open(enic);
1210         if (err) {
1211                 dev_err(enic, "vNIC dev open failed, aborting\n");
1212                 goto err_out_unregister;
1213         }
1214
1215         /* Set ingress vlan rewrite mode before vnic initialization */
1216         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1217                 IG_VLAN_REWRITE_MODE_PASS_THRU);
1218         if (err) {
1219                 dev_err(enic,
1220                         "Failed to set ingress vlan rewrite mode, aborting.\n");
1221                 goto err_out_dev_close;
1222         }
1223
1224         /* Issue device init to initialize the vnic-to-switch link.
1225          * We'll start with carrier off and wait for link UP
1226          * notification later to turn on carrier.  We don't need
1227          * to wait here for the vnic-to-switch link initialization
1228          * to complete; link UP notification is the indication that
1229          * the process is complete.
1230          */
1231
1232         err = vnic_dev_init(enic->vdev, 0);
1233         if (err) {
1234                 dev_err(enic, "vNIC dev init failed, aborting\n");
1235                 goto err_out_dev_close;
1236         }
1237
1238         err = enic_dev_init(enic);
1239         if (err) {
1240                 dev_err(enic, "Device initialization failed, aborting\n");
1241                 goto err_out_dev_close;
1242         }
1243
1244         return 0;
1245
1246 err_out_dev_close:
1247         vnic_dev_close(enic->vdev);
1248 err_out_unregister:
1249         vnic_dev_unregister(enic->vdev);
1250 err_out:
1251         return err;
1252 }
1253
1254 void enic_remove(struct enic *enic)
1255 {
1256         enic_dev_deinit(enic);
1257         vnic_dev_close(enic->vdev);
1258         vnic_dev_unregister(enic->vdev);
1259 }