net/enic: support scattered Rx
[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_mempool_put(mbuf->pool, 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                 error_status = vnic_rq_error_status(&enic->rq[i]);
127                 if (error_status)
128                         dev_err(enic, "RQ[%d] error_status %d\n", i,
129                                 error_status);
130         }
131 }
132
133 static void enic_clear_soft_stats(struct enic *enic)
134 {
135         struct enic_soft_stats *soft_stats = &enic->soft_stats;
136         rte_atomic64_clear(&soft_stats->rx_nombuf);
137         rte_atomic64_clear(&soft_stats->rx_packet_errors);
138 }
139
140 static void enic_init_soft_stats(struct enic *enic)
141 {
142         struct enic_soft_stats *soft_stats = &enic->soft_stats;
143         rte_atomic64_init(&soft_stats->rx_nombuf);
144         rte_atomic64_init(&soft_stats->rx_packet_errors);
145         enic_clear_soft_stats(enic);
146 }
147
148 void enic_dev_stats_clear(struct enic *enic)
149 {
150         if (vnic_dev_stats_clear(enic->vdev))
151                 dev_err(enic, "Error in clearing stats\n");
152         enic_clear_soft_stats(enic);
153 }
154
155 void enic_dev_stats_get(struct enic *enic, struct rte_eth_stats *r_stats)
156 {
157         struct vnic_stats *stats;
158         struct enic_soft_stats *soft_stats = &enic->soft_stats;
159         int64_t rx_truncated;
160         uint64_t rx_packet_errors;
161
162         if (vnic_dev_stats_dump(enic->vdev, &stats)) {
163                 dev_err(enic, "Error in getting stats\n");
164                 return;
165         }
166
167         /* The number of truncated packets can only be calculated by
168          * subtracting a hardware counter from error packets received by
169          * the driver. Note: this causes transient inaccuracies in the
170          * ipackets count. Also, the length of truncated packets are
171          * counted in ibytes even though truncated packets are dropped
172          * which can make ibytes be slightly higher than it should be.
173          */
174         rx_packet_errors = rte_atomic64_read(&soft_stats->rx_packet_errors);
175         rx_truncated = rx_packet_errors - stats->rx.rx_errors;
176
177         r_stats->ipackets = stats->rx.rx_frames_ok - rx_truncated;
178         r_stats->opackets = stats->tx.tx_frames_ok;
179
180         r_stats->ibytes = stats->rx.rx_bytes_ok;
181         r_stats->obytes = stats->tx.tx_bytes_ok;
182
183         r_stats->ierrors = stats->rx.rx_errors + stats->rx.rx_drop;
184         r_stats->oerrors = stats->tx.tx_errors;
185
186         r_stats->imissed = stats->rx.rx_no_bufs + rx_truncated;
187
188         r_stats->rx_nombuf = rte_atomic64_read(&soft_stats->rx_nombuf);
189 }
190
191 void enic_del_mac_address(struct enic *enic)
192 {
193         if (vnic_dev_del_addr(enic->vdev, enic->mac_addr))
194                 dev_err(enic, "del mac addr failed\n");
195 }
196
197 void enic_set_mac_address(struct enic *enic, uint8_t *mac_addr)
198 {
199         int err;
200
201         if (!is_eth_addr_valid(mac_addr)) {
202                 dev_err(enic, "invalid mac address\n");
203                 return;
204         }
205
206         err = vnic_dev_del_addr(enic->vdev, mac_addr);
207         if (err) {
208                 dev_err(enic, "del mac addr failed\n");
209                 return;
210         }
211
212         ether_addr_copy((struct ether_addr *)mac_addr,
213                 (struct ether_addr *)enic->mac_addr);
214
215         err = vnic_dev_add_addr(enic->vdev, mac_addr);
216         if (err) {
217                 dev_err(enic, "add mac addr failed\n");
218                 return;
219         }
220 }
221
222 static void
223 enic_free_rq_buf(struct rte_mbuf **mbuf)
224 {
225         if (*mbuf == NULL)
226                 return;
227
228         rte_pktmbuf_free(*mbuf);
229         mbuf = NULL;
230 }
231
232 void enic_init_vnic_resources(struct enic *enic)
233 {
234         unsigned int error_interrupt_enable = 1;
235         unsigned int error_interrupt_offset = 0;
236         unsigned int index = 0;
237         unsigned int cq_idx;
238         struct vnic_rq *data_rq;
239
240         for (index = 0; index < enic->rq_count; index++) {
241                 vnic_rq_init(&enic->rq[enic_sop_rq(index)],
242                         enic_cq_rq(enic, index),
243                         error_interrupt_enable,
244                         error_interrupt_offset);
245
246                 data_rq = &enic->rq[enic_data_rq(index)];
247                 if (data_rq->in_use)
248                         vnic_rq_init(data_rq,
249                                      enic_cq_rq(enic, index),
250                                      error_interrupt_enable,
251                                      error_interrupt_offset);
252
253                 cq_idx = enic_cq_rq(enic, index);
254                 vnic_cq_init(&enic->cq[cq_idx],
255                         0 /* flow_control_enable */,
256                         1 /* color_enable */,
257                         0 /* cq_head */,
258                         0 /* cq_tail */,
259                         1 /* cq_tail_color */,
260                         0 /* interrupt_enable */,
261                         1 /* cq_entry_enable */,
262                         0 /* cq_message_enable */,
263                         0 /* interrupt offset */,
264                         0 /* cq_message_addr */);
265         }
266
267         for (index = 0; index < enic->wq_count; index++) {
268                 vnic_wq_init(&enic->wq[index],
269                         enic_cq_wq(enic, index),
270                         error_interrupt_enable,
271                         error_interrupt_offset);
272
273                 cq_idx = enic_cq_wq(enic, index);
274                 vnic_cq_init(&enic->cq[cq_idx],
275                         0 /* flow_control_enable */,
276                         1 /* color_enable */,
277                         0 /* cq_head */,
278                         0 /* cq_tail */,
279                         1 /* cq_tail_color */,
280                         0 /* interrupt_enable */,
281                         0 /* cq_entry_enable */,
282                         1 /* cq_message_enable */,
283                         0 /* interrupt offset */,
284                         (u64)enic->wq[index].cqmsg_rz->phys_addr);
285         }
286
287         vnic_intr_init(&enic->intr,
288                 enic->config.intr_timer_usec,
289                 enic->config.intr_timer_type,
290                 /*mask_on_assertion*/1);
291 }
292
293
294 static int
295 enic_alloc_rx_queue_mbufs(struct enic *enic, struct vnic_rq *rq)
296 {
297         struct rte_mbuf *mb;
298         struct rq_enet_desc *rqd = rq->ring.descs;
299         unsigned i;
300         dma_addr_t dma_addr;
301
302         if (!rq->in_use)
303                 return 0;
304
305         dev_debug(enic, "queue %u, allocating %u rx queue mbufs\n", rq->index,
306                   rq->ring.desc_count);
307
308         for (i = 0; i < rq->ring.desc_count; i++, rqd++) {
309                 mb = rte_mbuf_raw_alloc(rq->mp);
310                 if (mb == NULL) {
311                         dev_err(enic, "RX mbuf alloc failed queue_id=%u\n",
312                         (unsigned)rq->index);
313                         return -ENOMEM;
314                 }
315
316                 dma_addr = (dma_addr_t)(mb->buf_physaddr
317                            + RTE_PKTMBUF_HEADROOM);
318                 rq_enet_desc_enc(rqd, dma_addr,
319                                 (rq->is_sop ? RQ_ENET_TYPE_ONLY_SOP
320                                 : RQ_ENET_TYPE_NOT_SOP),
321                                 mb->buf_len - RTE_PKTMBUF_HEADROOM);
322                 rq->mbuf_ring[i] = mb;
323         }
324
325         /* make sure all prior writes are complete before doing the PIO write */
326         rte_rmb();
327
328         /* Post all but the last buffer to VIC. */
329         rq->posted_index = rq->ring.desc_count - 1;
330
331         rq->rx_nb_hold = 0;
332
333         dev_debug(enic, "port=%u, qidx=%u, Write %u posted idx, %u sw held\n",
334                 enic->port_id, rq->index, rq->posted_index, rq->rx_nb_hold);
335         iowrite32(rq->posted_index, &rq->ctrl->posted_index);
336         rte_rmb();
337
338         return 0;
339
340 }
341
342 static void *
343 enic_alloc_consistent(__rte_unused void *priv, size_t size,
344         dma_addr_t *dma_handle, u8 *name)
345 {
346         void *vaddr;
347         const struct rte_memzone *rz;
348         *dma_handle = 0;
349
350         rz = rte_memzone_reserve_aligned((const char *)name,
351                                          size, SOCKET_ID_ANY, 0, ENIC_ALIGN);
352         if (!rz) {
353                 pr_err("%s : Failed to allocate memory requested for %s\n",
354                         __func__, name);
355                 return NULL;
356         }
357
358         vaddr = rz->addr;
359         *dma_handle = (dma_addr_t)rz->phys_addr;
360
361         return vaddr;
362 }
363
364 static void
365 enic_free_consistent(__rte_unused struct rte_pci_device *hwdev,
366         __rte_unused size_t size,
367         __rte_unused void *vaddr,
368         __rte_unused dma_addr_t dma_handle)
369 {
370         /* Nothing to be done */
371 }
372
373 static void
374 enic_intr_handler(__rte_unused struct rte_intr_handle *handle,
375         void *arg)
376 {
377         struct enic *enic = pmd_priv((struct rte_eth_dev *)arg);
378
379         vnic_intr_return_all_credits(&enic->intr);
380
381         enic_log_q_error(enic);
382 }
383
384 int enic_enable(struct enic *enic)
385 {
386         unsigned int index;
387         int err;
388         struct rte_eth_dev *eth_dev = enic->rte_dev;
389
390         eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
391         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
392         vnic_dev_notify_set(enic->vdev, -1); /* No Intr for notify */
393
394         if (enic_clsf_init(enic))
395                 dev_warning(enic, "Init of hash table for clsf failed."\
396                         "Flow director feature will not work\n");
397
398         for (index = 0; index < enic->rq_count; index++) {
399                 err = enic_alloc_rx_queue_mbufs(enic,
400                         &enic->rq[enic_sop_rq(index)]);
401                 if (err) {
402                         dev_err(enic, "Failed to alloc sop RX queue mbufs\n");
403                         return err;
404                 }
405                 err = enic_alloc_rx_queue_mbufs(enic,
406                         &enic->rq[enic_data_rq(index)]);
407                 if (err) {
408                         /* release the allocated mbufs for the sop rq*/
409                         enic_rxmbuf_queue_release(enic,
410                                 &enic->rq[enic_sop_rq(index)]);
411
412                         dev_err(enic, "Failed to alloc data RX queue mbufs\n");
413                         return err;
414                 }
415         }
416
417         for (index = 0; index < enic->wq_count; index++)
418                 enic_start_wq(enic, index);
419         for (index = 0; index < enic->rq_count; index++)
420                 enic_start_rq(enic, index);
421
422         vnic_dev_enable_wait(enic->vdev);
423
424         /* Register and enable error interrupt */
425         rte_intr_callback_register(&(enic->pdev->intr_handle),
426                 enic_intr_handler, (void *)enic->rte_dev);
427
428         rte_intr_enable(&(enic->pdev->intr_handle));
429         vnic_intr_unmask(&enic->intr);
430
431         return 0;
432 }
433
434 int enic_alloc_intr_resources(struct enic *enic)
435 {
436         int err;
437
438         dev_info(enic, "vNIC resources used:  "\
439                 "wq %d rq %d cq %d intr %d\n",
440                 enic->wq_count, enic_vnic_rq_count(enic),
441                 enic->cq_count, enic->intr_count);
442
443         err = vnic_intr_alloc(enic->vdev, &enic->intr, 0);
444         if (err)
445                 enic_free_vnic_resources(enic);
446
447         return err;
448 }
449
450 void enic_free_rq(void *rxq)
451 {
452         struct vnic_rq *rq_sop, *rq_data;
453         struct enic *enic;
454
455         if (rxq == NULL)
456                 return;
457
458         rq_sop = (struct vnic_rq *)rxq;
459         enic = vnic_dev_priv(rq_sop->vdev);
460         rq_data = &enic->rq[rq_sop->data_queue_idx];
461
462         enic_rxmbuf_queue_release(enic, rq_sop);
463         if (rq_data->in_use)
464                 enic_rxmbuf_queue_release(enic, rq_data);
465
466         rte_free(rq_sop->mbuf_ring);
467         if (rq_data->in_use)
468                 rte_free(rq_data->mbuf_ring);
469
470         rq_sop->mbuf_ring = NULL;
471         rq_data->mbuf_ring = NULL;
472
473         vnic_rq_free(rq_sop);
474         if (rq_data->in_use)
475                 vnic_rq_free(rq_data);
476
477         vnic_cq_free(&enic->cq[rq_sop->index]);
478 }
479
480 void enic_start_wq(struct enic *enic, uint16_t queue_idx)
481 {
482         vnic_wq_enable(&enic->wq[queue_idx]);
483 }
484
485 int enic_stop_wq(struct enic *enic, uint16_t queue_idx)
486 {
487         return vnic_wq_disable(&enic->wq[queue_idx]);
488 }
489
490 void enic_start_rq(struct enic *enic, uint16_t queue_idx)
491 {
492         struct vnic_rq *rq_sop = &enic->rq[enic_sop_rq(queue_idx)];
493         struct vnic_rq *rq_data = &enic->rq[rq_sop->data_queue_idx];
494
495         if (rq_data->in_use)
496                 vnic_rq_enable(rq_data);
497         rte_mb();
498         vnic_rq_enable(rq_sop);
499
500 }
501
502 int enic_stop_rq(struct enic *enic, uint16_t queue_idx)
503 {
504         int ret1 = 0, ret2 = 0;
505
506         struct vnic_rq *rq_sop = &enic->rq[enic_sop_rq(queue_idx)];
507         struct vnic_rq *rq_data = &enic->rq[rq_sop->data_queue_idx];
508
509         ret2 = vnic_rq_disable(rq_sop);
510         rte_mb();
511         if (rq_data->in_use)
512                 ret1 = vnic_rq_disable(rq_data);
513
514         if (ret2)
515                 return ret2;
516         else
517                 return ret1;
518 }
519
520 int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
521         unsigned int socket_id, struct rte_mempool *mp,
522         uint16_t nb_desc)
523 {
524         int rc;
525         uint16_t sop_queue_idx = enic_sop_rq(queue_idx);
526         uint16_t data_queue_idx = enic_data_rq(queue_idx);
527         struct vnic_rq *rq_sop = &enic->rq[sop_queue_idx];
528         struct vnic_rq *rq_data = &enic->rq[data_queue_idx];
529         unsigned int mbuf_size, mbufs_per_pkt;
530         unsigned int nb_sop_desc, nb_data_desc;
531         uint16_t min_sop, max_sop, min_data, max_data;
532
533         rq_sop->is_sop = 1;
534         rq_sop->data_queue_idx = data_queue_idx;
535         rq_data->is_sop = 0;
536         rq_data->data_queue_idx = 0;
537         rq_sop->socket_id = socket_id;
538         rq_sop->mp = mp;
539         rq_data->socket_id = socket_id;
540         rq_data->mp = mp;
541         rq_sop->in_use = 1;
542
543         mbuf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
544                                RTE_PKTMBUF_HEADROOM);
545
546         if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
547                 dev_info(enic, "Scatter rx mode enabled\n");
548                 /* ceil((mtu + ETHER_HDR_LEN + 4)/mbuf_size) */
549                 mbufs_per_pkt = ((enic->config.mtu + ETHER_HDR_LEN + 4) +
550                                  (mbuf_size - 1)) / mbuf_size;
551         } else {
552                 dev_info(enic, "Scatter rx mode disabled\n");
553                 mbufs_per_pkt = 1;
554         }
555
556         if (mbufs_per_pkt > 1) {
557                 dev_info(enic, "Scatter rx mode in use\n");
558                 rq_data->in_use = 1;
559         } else {
560                 dev_info(enic, "Scatter rx mode not being used\n");
561                 rq_data->in_use = 0;
562         }
563
564         /* number of descriptors have to be a multiple of 32 */
565         nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
566         nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
567
568         rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
569         rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
570
571         if (mbufs_per_pkt > 1) {
572                 min_sop = 64;
573                 max_sop = ((enic->config.rq_desc_count /
574                             (mbufs_per_pkt - 1)) & ~0x1F);
575                 min_data = min_sop * (mbufs_per_pkt - 1);
576                 max_data = enic->config.rq_desc_count;
577         } else {
578                 min_sop = 64;
579                 max_sop = enic->config.rq_desc_count;
580                 min_data = 0;
581                 max_data = 0;
582         }
583
584         if (nb_desc < (min_sop + min_data)) {
585                 dev_warning(enic,
586                             "Number of rx descs too low, adjusting to minimum\n");
587                 nb_sop_desc = min_sop;
588                 nb_data_desc = min_data;
589         } else if (nb_desc > (max_sop + max_data)) {
590                 dev_warning(enic,
591                             "Number of rx_descs too high, adjusting to maximum\n");
592                 nb_sop_desc = max_sop;
593                 nb_data_desc = max_data;
594         }
595         if (mbufs_per_pkt > 1) {
596                 dev_info(enic, "For mtu %d and mbuf size %d valid rx descriptor range is %d to %d\n",
597                          enic->config.mtu, mbuf_size, min_sop + min_data,
598                          max_sop + max_data);
599         }
600         dev_info(enic, "Using %d rx descriptors (sop %d, data %d)\n",
601                  nb_sop_desc + nb_data_desc, nb_sop_desc, nb_data_desc);
602
603         /* Allocate sop queue resources */
604         rc = vnic_rq_alloc(enic->vdev, rq_sop, sop_queue_idx,
605                 nb_sop_desc, sizeof(struct rq_enet_desc));
606         if (rc) {
607                 dev_err(enic, "error in allocation of sop rq\n");
608                 goto err_exit;
609         }
610         nb_sop_desc = rq_sop->ring.desc_count;
611
612         if (rq_data->in_use) {
613                 /* Allocate data queue resources */
614                 rc = vnic_rq_alloc(enic->vdev, rq_data, data_queue_idx,
615                                    nb_data_desc,
616                                    sizeof(struct rq_enet_desc));
617                 if (rc) {
618                         dev_err(enic, "error in allocation of data rq\n");
619                         goto err_free_rq_sop;
620                 }
621                 nb_data_desc = rq_data->ring.desc_count;
622         }
623         rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
624                            socket_id, nb_sop_desc + nb_data_desc,
625                            sizeof(struct cq_enet_rq_desc));
626         if (rc) {
627                 dev_err(enic, "error in allocation of cq for rq\n");
628                 goto err_free_rq_data;
629         }
630
631         /* Allocate the mbuf rings */
632         rq_sop->mbuf_ring = (struct rte_mbuf **)
633                 rte_zmalloc_socket("rq->mbuf_ring",
634                                    sizeof(struct rte_mbuf *) * nb_sop_desc,
635                                    RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
636         if (rq_sop->mbuf_ring == NULL)
637                 goto err_free_cq;
638
639         if (rq_data->in_use) {
640                 rq_data->mbuf_ring = (struct rte_mbuf **)
641                         rte_zmalloc_socket("rq->mbuf_ring",
642                                 sizeof(struct rte_mbuf *) * nb_data_desc,
643                                 RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
644                 if (rq_data->mbuf_ring == NULL)
645                         goto err_free_sop_mbuf;
646         }
647
648         return 0;
649
650 err_free_sop_mbuf:
651         rte_free(rq_sop->mbuf_ring);
652 err_free_cq:
653         /* cleanup on error */
654         vnic_cq_free(&enic->cq[queue_idx]);
655 err_free_rq_data:
656         if (rq_data->in_use)
657                 vnic_rq_free(rq_data);
658 err_free_rq_sop:
659         vnic_rq_free(rq_sop);
660 err_exit:
661         return -ENOMEM;
662 }
663
664 void enic_free_wq(void *txq)
665 {
666         struct vnic_wq *wq;
667         struct enic *enic;
668
669         if (txq == NULL)
670                 return;
671
672         wq = (struct vnic_wq *)txq;
673         enic = vnic_dev_priv(wq->vdev);
674         rte_memzone_free(wq->cqmsg_rz);
675         vnic_wq_free(wq);
676         vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
677 }
678
679 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
680         unsigned int socket_id, uint16_t nb_desc)
681 {
682         int err;
683         struct vnic_wq *wq = &enic->wq[queue_idx];
684         unsigned int cq_index = enic_cq_wq(enic, queue_idx);
685         char name[NAME_MAX];
686         static int instance;
687
688         wq->socket_id = socket_id;
689         if (nb_desc) {
690                 if (nb_desc > enic->config.wq_desc_count) {
691                         dev_warning(enic,
692                                 "WQ %d - number of tx desc in cmd line (%d)"\
693                                 "is greater than that in the UCSM/CIMC adapter"\
694                                 "policy.  Applying the value in the adapter "\
695                                 "policy (%d)\n",
696                                 queue_idx, nb_desc, enic->config.wq_desc_count);
697                 } else if (nb_desc != enic->config.wq_desc_count) {
698                         enic->config.wq_desc_count = nb_desc;
699                         dev_info(enic,
700                                 "TX Queues - effective number of descs:%d\n",
701                                 nb_desc);
702                 }
703         }
704
705         /* Allocate queue resources */
706         err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
707                 enic->config.wq_desc_count,
708                 sizeof(struct wq_enet_desc));
709         if (err) {
710                 dev_err(enic, "error in allocation of wq\n");
711                 return err;
712         }
713
714         err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
715                 socket_id, enic->config.wq_desc_count,
716                 sizeof(struct cq_enet_wq_desc));
717         if (err) {
718                 vnic_wq_free(wq);
719                 dev_err(enic, "error in allocation of cq for wq\n");
720         }
721
722         /* setup up CQ message */
723         snprintf((char *)name, sizeof(name),
724                  "vnic_cqmsg-%s-%d-%d", enic->bdf_name, queue_idx,
725                 instance++);
726
727         wq->cqmsg_rz = rte_memzone_reserve_aligned((const char *)name,
728                                                    sizeof(uint32_t),
729                                                    SOCKET_ID_ANY, 0,
730                                                    ENIC_ALIGN);
731         if (!wq->cqmsg_rz)
732                 return -ENOMEM;
733
734         return err;
735 }
736
737 int enic_disable(struct enic *enic)
738 {
739         unsigned int i;
740         int err;
741
742         vnic_intr_mask(&enic->intr);
743         (void)vnic_intr_masked(&enic->intr); /* flush write */
744
745         vnic_dev_disable(enic->vdev);
746
747         enic_clsf_destroy(enic);
748
749         if (!enic_is_sriov_vf(enic))
750                 vnic_dev_del_addr(enic->vdev, enic->mac_addr);
751
752         for (i = 0; i < enic->wq_count; i++) {
753                 err = vnic_wq_disable(&enic->wq[i]);
754                 if (err)
755                         return err;
756         }
757         for (i = 0; i < enic_vnic_rq_count(enic); i++) {
758                 if (enic->rq[i].in_use) {
759                         err = vnic_rq_disable(&enic->rq[i]);
760                         if (err)
761                                 return err;
762                 }
763         }
764
765         vnic_dev_set_reset_flag(enic->vdev, 1);
766         vnic_dev_notify_unset(enic->vdev);
767
768         for (i = 0; i < enic->wq_count; i++)
769                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
770
771         for (i = 0; i < enic_vnic_rq_count(enic); i++)
772                 if (enic->rq[i].in_use)
773                         vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
774         for (i = 0; i < enic->cq_count; i++)
775                 vnic_cq_clean(&enic->cq[i]);
776         vnic_intr_clean(&enic->intr);
777
778         return 0;
779 }
780
781 static int enic_dev_wait(struct vnic_dev *vdev,
782         int (*start)(struct vnic_dev *, int),
783         int (*finished)(struct vnic_dev *, int *),
784         int arg)
785 {
786         int done;
787         int err;
788         int i;
789
790         err = start(vdev, arg);
791         if (err)
792                 return err;
793
794         /* Wait for func to complete...2 seconds max */
795         for (i = 0; i < 2000; i++) {
796                 err = finished(vdev, &done);
797                 if (err)
798                         return err;
799                 if (done)
800                         return 0;
801                 usleep(1000);
802         }
803         return -ETIMEDOUT;
804 }
805
806 static int enic_dev_open(struct enic *enic)
807 {
808         int err;
809
810         err = enic_dev_wait(enic->vdev, vnic_dev_open,
811                 vnic_dev_open_done, 0);
812         if (err)
813                 dev_err(enic_get_dev(enic),
814                         "vNIC device open failed, err %d\n", err);
815
816         return err;
817 }
818
819 static int enic_set_rsskey(struct enic *enic)
820 {
821         dma_addr_t rss_key_buf_pa;
822         union vnic_rss_key *rss_key_buf_va = NULL;
823         static union vnic_rss_key rss_key = {
824                 .key = {
825                         [0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
826                         [1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
827                         [2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
828                         [3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
829                 }
830         };
831         int err;
832         u8 name[NAME_MAX];
833
834         snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
835         rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
836                 &rss_key_buf_pa, name);
837         if (!rss_key_buf_va)
838                 return -ENOMEM;
839
840         rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
841
842         err = enic_set_rss_key(enic,
843                 rss_key_buf_pa,
844                 sizeof(union vnic_rss_key));
845
846         enic_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
847                 rss_key_buf_va, rss_key_buf_pa);
848
849         return err;
850 }
851
852 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
853 {
854         dma_addr_t rss_cpu_buf_pa;
855         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
856         int i;
857         int err;
858         u8 name[NAME_MAX];
859
860         snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
861         rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
862                 &rss_cpu_buf_pa, name);
863         if (!rss_cpu_buf_va)
864                 return -ENOMEM;
865
866         for (i = 0; i < (1 << rss_hash_bits); i++)
867                 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
868
869         err = enic_set_rss_cpu(enic,
870                 rss_cpu_buf_pa,
871                 sizeof(union vnic_rss_cpu));
872
873         enic_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
874                 rss_cpu_buf_va, rss_cpu_buf_pa);
875
876         return err;
877 }
878
879 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
880         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
881 {
882         const u8 tso_ipid_split_en = 0;
883         int err;
884
885         /* Enable VLAN tag stripping */
886
887         err = enic_set_nic_cfg(enic,
888                 rss_default_cpu, rss_hash_type,
889                 rss_hash_bits, rss_base_cpu,
890                 rss_enable, tso_ipid_split_en,
891                 enic->ig_vlan_strip_en);
892
893         return err;
894 }
895
896 int enic_set_rss_nic_cfg(struct enic *enic)
897 {
898         const u8 rss_default_cpu = 0;
899         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
900             NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
901             NIC_CFG_RSS_HASH_TYPE_IPV6 |
902             NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
903         const u8 rss_hash_bits = 7;
904         const u8 rss_base_cpu = 0;
905         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
906
907         if (rss_enable) {
908                 if (!enic_set_rsskey(enic)) {
909                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
910                                 rss_enable = 0;
911                                 dev_warning(enic, "RSS disabled, "\
912                                         "Failed to set RSS cpu indirection table.");
913                         }
914                 } else {
915                         rss_enable = 0;
916                         dev_warning(enic,
917                                 "RSS disabled, Failed to set RSS key.\n");
918                 }
919         }
920
921         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
922                 rss_hash_bits, rss_base_cpu, rss_enable);
923 }
924
925 int enic_setup_finish(struct enic *enic)
926 {
927         int ret;
928
929         enic_init_soft_stats(enic);
930
931         ret = enic_set_rss_nic_cfg(enic);
932         if (ret) {
933                 dev_err(enic, "Failed to config nic, aborting.\n");
934                 return -1;
935         }
936
937         vnic_dev_add_addr(enic->vdev, enic->mac_addr);
938
939         /* Default conf */
940         vnic_dev_packet_filter(enic->vdev,
941                 1 /* directed  */,
942                 1 /* multicast */,
943                 1 /* broadcast */,
944                 0 /* promisc   */,
945                 1 /* allmulti  */);
946
947         enic->promisc = 0;
948         enic->allmulti = 1;
949
950         return 0;
951 }
952
953 void enic_add_packet_filter(struct enic *enic)
954 {
955         /* Args -> directed, multicast, broadcast, promisc, allmulti */
956         vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
957                 enic->promisc, enic->allmulti);
958 }
959
960 int enic_get_link_status(struct enic *enic)
961 {
962         return vnic_dev_link_status(enic->vdev);
963 }
964
965 static void enic_dev_deinit(struct enic *enic)
966 {
967         struct rte_eth_dev *eth_dev = enic->rte_dev;
968
969         rte_free(eth_dev->data->mac_addrs);
970 }
971
972
973 int enic_set_vnic_res(struct enic *enic)
974 {
975         struct rte_eth_dev *eth_dev = enic->rte_dev;
976         int rc = 0;
977
978         /* With Rx scatter support, two RQs are now used per RQ used by
979          * the application.
980          */
981         if (enic->rq_count < (eth_dev->data->nb_rx_queues * 2)) {
982                 dev_err(dev, "Not enough Receive queues. Requested:%u which uses %d RQs on VIC, Configured:%u\n",
983                         eth_dev->data->nb_rx_queues,
984                         eth_dev->data->nb_rx_queues * 2, enic->rq_count);
985                 rc = -EINVAL;
986         }
987         if (enic->wq_count < eth_dev->data->nb_tx_queues) {
988                 dev_err(dev, "Not enough Transmit queues. Requested:%u, Configured:%u\n",
989                         eth_dev->data->nb_tx_queues, enic->wq_count);
990                 rc = -EINVAL;
991         }
992
993         if (enic->cq_count < (enic->rq_count + enic->wq_count)) {
994                 dev_err(dev, "Not enough Completion queues. Required:%u, Configured:%u\n",
995                         enic->rq_count + enic->wq_count, enic->cq_count);
996                 rc = -EINVAL;
997         }
998
999         if (rc == 0) {
1000                 enic->rq_count = eth_dev->data->nb_rx_queues;
1001                 enic->wq_count = eth_dev->data->nb_tx_queues;
1002                 enic->cq_count = enic->rq_count + enic->wq_count;
1003         }
1004
1005         return rc;
1006 }
1007
1008 static int enic_dev_init(struct enic *enic)
1009 {
1010         int err;
1011         struct rte_eth_dev *eth_dev = enic->rte_dev;
1012
1013         vnic_dev_intr_coal_timer_info_default(enic->vdev);
1014
1015         /* Get vNIC configuration
1016         */
1017         err = enic_get_vnic_config(enic);
1018         if (err) {
1019                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
1020                 return err;
1021         }
1022
1023         eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN, 0);
1024         if (!eth_dev->data->mac_addrs) {
1025                 dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1026                 return -1;
1027         }
1028         ether_addr_copy((struct ether_addr *) enic->mac_addr,
1029                 &eth_dev->data->mac_addrs[0]);
1030
1031
1032         /* Get available resource counts
1033         */
1034         enic_get_res_counts(enic);
1035
1036         vnic_dev_set_reset_flag(enic->vdev, 0);
1037
1038         return 0;
1039
1040 }
1041
1042 int enic_probe(struct enic *enic)
1043 {
1044         struct rte_pci_device *pdev = enic->pdev;
1045         int err = -1;
1046
1047         dev_debug(enic, " Initializing ENIC PMD version %s\n", DRV_VERSION);
1048
1049         enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1050         enic->bar0.len = pdev->mem_resource[0].len;
1051
1052         /* Register vNIC device */
1053         enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1054         if (!enic->vdev) {
1055                 dev_err(enic, "vNIC registration failed, aborting\n");
1056                 goto err_out;
1057         }
1058
1059         vnic_register_cbacks(enic->vdev,
1060                 enic_alloc_consistent,
1061                 enic_free_consistent);
1062
1063         /* Issue device open to get device in known state */
1064         err = enic_dev_open(enic);
1065         if (err) {
1066                 dev_err(enic, "vNIC dev open failed, aborting\n");
1067                 goto err_out_unregister;
1068         }
1069
1070         /* Set ingress vlan rewrite mode before vnic initialization */
1071         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1072                 IG_VLAN_REWRITE_MODE_PASS_THRU);
1073         if (err) {
1074                 dev_err(enic,
1075                         "Failed to set ingress vlan rewrite mode, aborting.\n");
1076                 goto err_out_dev_close;
1077         }
1078
1079         /* Issue device init to initialize the vnic-to-switch link.
1080          * We'll start with carrier off and wait for link UP
1081          * notification later to turn on carrier.  We don't need
1082          * to wait here for the vnic-to-switch link initialization
1083          * to complete; link UP notification is the indication that
1084          * the process is complete.
1085          */
1086
1087         err = vnic_dev_init(enic->vdev, 0);
1088         if (err) {
1089                 dev_err(enic, "vNIC dev init failed, aborting\n");
1090                 goto err_out_dev_close;
1091         }
1092
1093         err = enic_dev_init(enic);
1094         if (err) {
1095                 dev_err(enic, "Device initialization failed, aborting\n");
1096                 goto err_out_dev_close;
1097         }
1098
1099         return 0;
1100
1101 err_out_dev_close:
1102         vnic_dev_close(enic->vdev);
1103 err_out_unregister:
1104         vnic_dev_unregister(enic->vdev);
1105 err_out:
1106         return err;
1107 }
1108
1109 void enic_remove(struct enic *enic)
1110 {
1111         enic_dev_deinit(enic);
1112         vnic_dev_close(enic->vdev);
1113         vnic_dev_unregister(enic->vdev);
1114 }