net/enic: fix memory freeing
[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(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         struct enic *enic = (struct enic *)priv;
350         struct enic_memzone_entry *mze;
351
352         rz = rte_memzone_reserve_aligned((const char *)name,
353                                          size, SOCKET_ID_ANY, 0, ENIC_ALIGN);
354         if (!rz) {
355                 pr_err("%s : Failed to allocate memory requested for %s\n",
356                         __func__, name);
357                 return NULL;
358         }
359
360         vaddr = rz->addr;
361         *dma_handle = (dma_addr_t)rz->phys_addr;
362
363         mze = rte_malloc("enic memzone entry",
364                          sizeof(struct enic_memzone_entry), 0);
365
366         if (!mze) {
367                 pr_err("%s : Failed to allocate memory for memzone list\n",
368                        __func__);
369                 rte_memzone_free(rz);
370         }
371
372         mze->rz = rz;
373
374         rte_spinlock_lock(&enic->memzone_list_lock);
375         LIST_INSERT_HEAD(&enic->memzone_list, mze, entries);
376         rte_spinlock_unlock(&enic->memzone_list_lock);
377
378         return vaddr;
379 }
380
381 static void
382 enic_free_consistent(void *priv,
383                      __rte_unused size_t size,
384                      void *vaddr,
385                      dma_addr_t dma_handle)
386 {
387         struct enic_memzone_entry *mze;
388         struct enic *enic = (struct enic *)priv;
389
390         rte_spinlock_lock(&enic->memzone_list_lock);
391         LIST_FOREACH(mze, &enic->memzone_list, entries) {
392                 if (mze->rz->addr == vaddr &&
393                     mze->rz->phys_addr == dma_handle)
394                         break;
395         }
396         if (mze == NULL) {
397                 rte_spinlock_unlock(&enic->memzone_list_lock);
398                 dev_warning(enic,
399                             "Tried to free memory, but couldn't find it in the memzone list\n");
400                 return;
401         }
402         LIST_REMOVE(mze, entries);
403         rte_spinlock_unlock(&enic->memzone_list_lock);
404         rte_memzone_free(mze->rz);
405         rte_free(mze);
406 }
407
408 static void
409 enic_intr_handler(__rte_unused struct rte_intr_handle *handle,
410         void *arg)
411 {
412         struct enic *enic = pmd_priv((struct rte_eth_dev *)arg);
413
414         vnic_intr_return_all_credits(&enic->intr);
415
416         enic_log_q_error(enic);
417 }
418
419 int enic_enable(struct enic *enic)
420 {
421         unsigned int index;
422         int err;
423         struct rte_eth_dev *eth_dev = enic->rte_dev;
424
425         eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
426         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
427         vnic_dev_notify_set(enic->vdev, -1); /* No Intr for notify */
428
429         if (enic_clsf_init(enic))
430                 dev_warning(enic, "Init of hash table for clsf failed."\
431                         "Flow director feature will not work\n");
432
433         for (index = 0; index < enic->rq_count; index++) {
434                 err = enic_alloc_rx_queue_mbufs(enic,
435                         &enic->rq[enic_sop_rq(index)]);
436                 if (err) {
437                         dev_err(enic, "Failed to alloc sop RX queue mbufs\n");
438                         return err;
439                 }
440                 err = enic_alloc_rx_queue_mbufs(enic,
441                         &enic->rq[enic_data_rq(index)]);
442                 if (err) {
443                         /* release the allocated mbufs for the sop rq*/
444                         enic_rxmbuf_queue_release(enic,
445                                 &enic->rq[enic_sop_rq(index)]);
446
447                         dev_err(enic, "Failed to alloc data RX queue mbufs\n");
448                         return err;
449                 }
450         }
451
452         for (index = 0; index < enic->wq_count; index++)
453                 enic_start_wq(enic, index);
454         for (index = 0; index < enic->rq_count; index++)
455                 enic_start_rq(enic, index);
456
457         vnic_dev_enable_wait(enic->vdev);
458
459         /* Register and enable error interrupt */
460         rte_intr_callback_register(&(enic->pdev->intr_handle),
461                 enic_intr_handler, (void *)enic->rte_dev);
462
463         rte_intr_enable(&(enic->pdev->intr_handle));
464         vnic_intr_unmask(&enic->intr);
465
466         return 0;
467 }
468
469 int enic_alloc_intr_resources(struct enic *enic)
470 {
471         int err;
472
473         dev_info(enic, "vNIC resources used:  "\
474                 "wq %d rq %d cq %d intr %d\n",
475                 enic->wq_count, enic_vnic_rq_count(enic),
476                 enic->cq_count, enic->intr_count);
477
478         err = vnic_intr_alloc(enic->vdev, &enic->intr, 0);
479         if (err)
480                 enic_free_vnic_resources(enic);
481
482         return err;
483 }
484
485 void enic_free_rq(void *rxq)
486 {
487         struct vnic_rq *rq_sop, *rq_data;
488         struct enic *enic;
489
490         if (rxq == NULL)
491                 return;
492
493         rq_sop = (struct vnic_rq *)rxq;
494         enic = vnic_dev_priv(rq_sop->vdev);
495         rq_data = &enic->rq[rq_sop->data_queue_idx];
496
497         enic_rxmbuf_queue_release(enic, rq_sop);
498         if (rq_data->in_use)
499                 enic_rxmbuf_queue_release(enic, rq_data);
500
501         rte_free(rq_sop->mbuf_ring);
502         if (rq_data->in_use)
503                 rte_free(rq_data->mbuf_ring);
504
505         rq_sop->mbuf_ring = NULL;
506         rq_data->mbuf_ring = NULL;
507
508         vnic_rq_free(rq_sop);
509         if (rq_data->in_use)
510                 vnic_rq_free(rq_data);
511
512         vnic_cq_free(&enic->cq[rq_sop->index]);
513 }
514
515 void enic_start_wq(struct enic *enic, uint16_t queue_idx)
516 {
517         vnic_wq_enable(&enic->wq[queue_idx]);
518 }
519
520 int enic_stop_wq(struct enic *enic, uint16_t queue_idx)
521 {
522         return vnic_wq_disable(&enic->wq[queue_idx]);
523 }
524
525 void enic_start_rq(struct enic *enic, uint16_t queue_idx)
526 {
527         struct vnic_rq *rq_sop = &enic->rq[enic_sop_rq(queue_idx)];
528         struct vnic_rq *rq_data = &enic->rq[rq_sop->data_queue_idx];
529
530         if (rq_data->in_use)
531                 vnic_rq_enable(rq_data);
532         rte_mb();
533         vnic_rq_enable(rq_sop);
534
535 }
536
537 int enic_stop_rq(struct enic *enic, uint16_t queue_idx)
538 {
539         int ret1 = 0, ret2 = 0;
540
541         struct vnic_rq *rq_sop = &enic->rq[enic_sop_rq(queue_idx)];
542         struct vnic_rq *rq_data = &enic->rq[rq_sop->data_queue_idx];
543
544         ret2 = vnic_rq_disable(rq_sop);
545         rte_mb();
546         if (rq_data->in_use)
547                 ret1 = vnic_rq_disable(rq_data);
548
549         if (ret2)
550                 return ret2;
551         else
552                 return ret1;
553 }
554
555 int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
556         unsigned int socket_id, struct rte_mempool *mp,
557         uint16_t nb_desc)
558 {
559         int rc;
560         uint16_t sop_queue_idx = enic_sop_rq(queue_idx);
561         uint16_t data_queue_idx = enic_data_rq(queue_idx);
562         struct vnic_rq *rq_sop = &enic->rq[sop_queue_idx];
563         struct vnic_rq *rq_data = &enic->rq[data_queue_idx];
564         unsigned int mbuf_size, mbufs_per_pkt;
565         unsigned int nb_sop_desc, nb_data_desc;
566         uint16_t min_sop, max_sop, min_data, max_data;
567
568         rq_sop->is_sop = 1;
569         rq_sop->data_queue_idx = data_queue_idx;
570         rq_data->is_sop = 0;
571         rq_data->data_queue_idx = 0;
572         rq_sop->socket_id = socket_id;
573         rq_sop->mp = mp;
574         rq_data->socket_id = socket_id;
575         rq_data->mp = mp;
576         rq_sop->in_use = 1;
577
578         mbuf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
579                                RTE_PKTMBUF_HEADROOM);
580
581         if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
582                 dev_info(enic, "Scatter rx mode enabled\n");
583                 /* ceil((mtu + ETHER_HDR_LEN + 4)/mbuf_size) */
584                 mbufs_per_pkt = ((enic->config.mtu + ETHER_HDR_LEN + 4) +
585                                  (mbuf_size - 1)) / mbuf_size;
586         } else {
587                 dev_info(enic, "Scatter rx mode disabled\n");
588                 mbufs_per_pkt = 1;
589         }
590
591         if (mbufs_per_pkt > 1) {
592                 dev_info(enic, "Scatter rx mode in use\n");
593                 rq_data->in_use = 1;
594         } else {
595                 dev_info(enic, "Scatter rx mode not being used\n");
596                 rq_data->in_use = 0;
597         }
598
599         /* number of descriptors have to be a multiple of 32 */
600         nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
601         nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
602
603         rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
604         rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
605
606         if (mbufs_per_pkt > 1) {
607                 min_sop = 64;
608                 max_sop = ((enic->config.rq_desc_count /
609                             (mbufs_per_pkt - 1)) & ~0x1F);
610                 min_data = min_sop * (mbufs_per_pkt - 1);
611                 max_data = enic->config.rq_desc_count;
612         } else {
613                 min_sop = 64;
614                 max_sop = enic->config.rq_desc_count;
615                 min_data = 0;
616                 max_data = 0;
617         }
618
619         if (nb_desc < (min_sop + min_data)) {
620                 dev_warning(enic,
621                             "Number of rx descs too low, adjusting to minimum\n");
622                 nb_sop_desc = min_sop;
623                 nb_data_desc = min_data;
624         } else if (nb_desc > (max_sop + max_data)) {
625                 dev_warning(enic,
626                             "Number of rx_descs too high, adjusting to maximum\n");
627                 nb_sop_desc = max_sop;
628                 nb_data_desc = max_data;
629         }
630         if (mbufs_per_pkt > 1) {
631                 dev_info(enic, "For mtu %d and mbuf size %d valid rx descriptor range is %d to %d\n",
632                          enic->config.mtu, mbuf_size, min_sop + min_data,
633                          max_sop + max_data);
634         }
635         dev_info(enic, "Using %d rx descriptors (sop %d, data %d)\n",
636                  nb_sop_desc + nb_data_desc, nb_sop_desc, nb_data_desc);
637
638         /* Allocate sop queue resources */
639         rc = vnic_rq_alloc(enic->vdev, rq_sop, sop_queue_idx,
640                 nb_sop_desc, sizeof(struct rq_enet_desc));
641         if (rc) {
642                 dev_err(enic, "error in allocation of sop rq\n");
643                 goto err_exit;
644         }
645         nb_sop_desc = rq_sop->ring.desc_count;
646
647         if (rq_data->in_use) {
648                 /* Allocate data queue resources */
649                 rc = vnic_rq_alloc(enic->vdev, rq_data, data_queue_idx,
650                                    nb_data_desc,
651                                    sizeof(struct rq_enet_desc));
652                 if (rc) {
653                         dev_err(enic, "error in allocation of data rq\n");
654                         goto err_free_rq_sop;
655                 }
656                 nb_data_desc = rq_data->ring.desc_count;
657         }
658         rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
659                            socket_id, nb_sop_desc + nb_data_desc,
660                            sizeof(struct cq_enet_rq_desc));
661         if (rc) {
662                 dev_err(enic, "error in allocation of cq for rq\n");
663                 goto err_free_rq_data;
664         }
665
666         /* Allocate the mbuf rings */
667         rq_sop->mbuf_ring = (struct rte_mbuf **)
668                 rte_zmalloc_socket("rq->mbuf_ring",
669                                    sizeof(struct rte_mbuf *) * nb_sop_desc,
670                                    RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
671         if (rq_sop->mbuf_ring == NULL)
672                 goto err_free_cq;
673
674         if (rq_data->in_use) {
675                 rq_data->mbuf_ring = (struct rte_mbuf **)
676                         rte_zmalloc_socket("rq->mbuf_ring",
677                                 sizeof(struct rte_mbuf *) * nb_data_desc,
678                                 RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
679                 if (rq_data->mbuf_ring == NULL)
680                         goto err_free_sop_mbuf;
681         }
682
683         return 0;
684
685 err_free_sop_mbuf:
686         rte_free(rq_sop->mbuf_ring);
687 err_free_cq:
688         /* cleanup on error */
689         vnic_cq_free(&enic->cq[queue_idx]);
690 err_free_rq_data:
691         if (rq_data->in_use)
692                 vnic_rq_free(rq_data);
693 err_free_rq_sop:
694         vnic_rq_free(rq_sop);
695 err_exit:
696         return -ENOMEM;
697 }
698
699 void enic_free_wq(void *txq)
700 {
701         struct vnic_wq *wq;
702         struct enic *enic;
703
704         if (txq == NULL)
705                 return;
706
707         wq = (struct vnic_wq *)txq;
708         enic = vnic_dev_priv(wq->vdev);
709         rte_memzone_free(wq->cqmsg_rz);
710         vnic_wq_free(wq);
711         vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
712 }
713
714 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
715         unsigned int socket_id, uint16_t nb_desc)
716 {
717         int err;
718         struct vnic_wq *wq = &enic->wq[queue_idx];
719         unsigned int cq_index = enic_cq_wq(enic, queue_idx);
720         char name[NAME_MAX];
721         static int instance;
722
723         wq->socket_id = socket_id;
724         if (nb_desc) {
725                 if (nb_desc > enic->config.wq_desc_count) {
726                         dev_warning(enic,
727                                 "WQ %d - number of tx desc in cmd line (%d)"\
728                                 "is greater than that in the UCSM/CIMC adapter"\
729                                 "policy.  Applying the value in the adapter "\
730                                 "policy (%d)\n",
731                                 queue_idx, nb_desc, enic->config.wq_desc_count);
732                 } else if (nb_desc != enic->config.wq_desc_count) {
733                         enic->config.wq_desc_count = nb_desc;
734                         dev_info(enic,
735                                 "TX Queues - effective number of descs:%d\n",
736                                 nb_desc);
737                 }
738         }
739
740         /* Allocate queue resources */
741         err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
742                 enic->config.wq_desc_count,
743                 sizeof(struct wq_enet_desc));
744         if (err) {
745                 dev_err(enic, "error in allocation of wq\n");
746                 return err;
747         }
748
749         err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
750                 socket_id, enic->config.wq_desc_count,
751                 sizeof(struct cq_enet_wq_desc));
752         if (err) {
753                 vnic_wq_free(wq);
754                 dev_err(enic, "error in allocation of cq for wq\n");
755         }
756
757         /* setup up CQ message */
758         snprintf((char *)name, sizeof(name),
759                  "vnic_cqmsg-%s-%d-%d", enic->bdf_name, queue_idx,
760                 instance++);
761
762         wq->cqmsg_rz = rte_memzone_reserve_aligned((const char *)name,
763                                                    sizeof(uint32_t),
764                                                    SOCKET_ID_ANY, 0,
765                                                    ENIC_ALIGN);
766         if (!wq->cqmsg_rz)
767                 return -ENOMEM;
768
769         return err;
770 }
771
772 int enic_disable(struct enic *enic)
773 {
774         unsigned int i;
775         int err;
776
777         vnic_intr_mask(&enic->intr);
778         (void)vnic_intr_masked(&enic->intr); /* flush write */
779
780         vnic_dev_disable(enic->vdev);
781
782         enic_clsf_destroy(enic);
783
784         if (!enic_is_sriov_vf(enic))
785                 vnic_dev_del_addr(enic->vdev, enic->mac_addr);
786
787         for (i = 0; i < enic->wq_count; i++) {
788                 err = vnic_wq_disable(&enic->wq[i]);
789                 if (err)
790                         return err;
791         }
792         for (i = 0; i < enic_vnic_rq_count(enic); i++) {
793                 if (enic->rq[i].in_use) {
794                         err = vnic_rq_disable(&enic->rq[i]);
795                         if (err)
796                                 return err;
797                 }
798         }
799
800         vnic_dev_set_reset_flag(enic->vdev, 1);
801         vnic_dev_notify_unset(enic->vdev);
802
803         for (i = 0; i < enic->wq_count; i++)
804                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
805
806         for (i = 0; i < enic_vnic_rq_count(enic); i++)
807                 if (enic->rq[i].in_use)
808                         vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
809         for (i = 0; i < enic->cq_count; i++)
810                 vnic_cq_clean(&enic->cq[i]);
811         vnic_intr_clean(&enic->intr);
812
813         return 0;
814 }
815
816 static int enic_dev_wait(struct vnic_dev *vdev,
817         int (*start)(struct vnic_dev *, int),
818         int (*finished)(struct vnic_dev *, int *),
819         int arg)
820 {
821         int done;
822         int err;
823         int i;
824
825         err = start(vdev, arg);
826         if (err)
827                 return err;
828
829         /* Wait for func to complete...2 seconds max */
830         for (i = 0; i < 2000; i++) {
831                 err = finished(vdev, &done);
832                 if (err)
833                         return err;
834                 if (done)
835                         return 0;
836                 usleep(1000);
837         }
838         return -ETIMEDOUT;
839 }
840
841 static int enic_dev_open(struct enic *enic)
842 {
843         int err;
844
845         err = enic_dev_wait(enic->vdev, vnic_dev_open,
846                 vnic_dev_open_done, 0);
847         if (err)
848                 dev_err(enic_get_dev(enic),
849                         "vNIC device open failed, err %d\n", err);
850
851         return err;
852 }
853
854 static int enic_set_rsskey(struct enic *enic)
855 {
856         dma_addr_t rss_key_buf_pa;
857         union vnic_rss_key *rss_key_buf_va = NULL;
858         static union vnic_rss_key rss_key = {
859                 .key = {
860                         [0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
861                         [1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
862                         [2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
863                         [3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
864                 }
865         };
866         int err;
867         u8 name[NAME_MAX];
868
869         snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
870         rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
871                 &rss_key_buf_pa, name);
872         if (!rss_key_buf_va)
873                 return -ENOMEM;
874
875         rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
876
877         err = enic_set_rss_key(enic,
878                 rss_key_buf_pa,
879                 sizeof(union vnic_rss_key));
880
881         enic_free_consistent(enic, sizeof(union vnic_rss_key),
882                 rss_key_buf_va, rss_key_buf_pa);
883
884         return err;
885 }
886
887 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
888 {
889         dma_addr_t rss_cpu_buf_pa;
890         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
891         int i;
892         int err;
893         u8 name[NAME_MAX];
894
895         snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
896         rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
897                 &rss_cpu_buf_pa, name);
898         if (!rss_cpu_buf_va)
899                 return -ENOMEM;
900
901         for (i = 0; i < (1 << rss_hash_bits); i++)
902                 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
903
904         err = enic_set_rss_cpu(enic,
905                 rss_cpu_buf_pa,
906                 sizeof(union vnic_rss_cpu));
907
908         enic_free_consistent(enic, sizeof(union vnic_rss_cpu),
909                 rss_cpu_buf_va, rss_cpu_buf_pa);
910
911         return err;
912 }
913
914 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
915         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
916 {
917         const u8 tso_ipid_split_en = 0;
918         int err;
919
920         /* Enable VLAN tag stripping */
921
922         err = enic_set_nic_cfg(enic,
923                 rss_default_cpu, rss_hash_type,
924                 rss_hash_bits, rss_base_cpu,
925                 rss_enable, tso_ipid_split_en,
926                 enic->ig_vlan_strip_en);
927
928         return err;
929 }
930
931 int enic_set_rss_nic_cfg(struct enic *enic)
932 {
933         const u8 rss_default_cpu = 0;
934         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
935             NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
936             NIC_CFG_RSS_HASH_TYPE_IPV6 |
937             NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
938         const u8 rss_hash_bits = 7;
939         const u8 rss_base_cpu = 0;
940         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
941
942         if (rss_enable) {
943                 if (!enic_set_rsskey(enic)) {
944                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
945                                 rss_enable = 0;
946                                 dev_warning(enic, "RSS disabled, "\
947                                         "Failed to set RSS cpu indirection table.");
948                         }
949                 } else {
950                         rss_enable = 0;
951                         dev_warning(enic,
952                                 "RSS disabled, Failed to set RSS key.\n");
953                 }
954         }
955
956         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
957                 rss_hash_bits, rss_base_cpu, rss_enable);
958 }
959
960 int enic_setup_finish(struct enic *enic)
961 {
962         int ret;
963
964         enic_init_soft_stats(enic);
965
966         ret = enic_set_rss_nic_cfg(enic);
967         if (ret) {
968                 dev_err(enic, "Failed to config nic, aborting.\n");
969                 return -1;
970         }
971
972         vnic_dev_add_addr(enic->vdev, enic->mac_addr);
973
974         /* Default conf */
975         vnic_dev_packet_filter(enic->vdev,
976                 1 /* directed  */,
977                 1 /* multicast */,
978                 1 /* broadcast */,
979                 0 /* promisc   */,
980                 1 /* allmulti  */);
981
982         enic->promisc = 0;
983         enic->allmulti = 1;
984
985         return 0;
986 }
987
988 void enic_add_packet_filter(struct enic *enic)
989 {
990         /* Args -> directed, multicast, broadcast, promisc, allmulti */
991         vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
992                 enic->promisc, enic->allmulti);
993 }
994
995 int enic_get_link_status(struct enic *enic)
996 {
997         return vnic_dev_link_status(enic->vdev);
998 }
999
1000 static void enic_dev_deinit(struct enic *enic)
1001 {
1002         struct rte_eth_dev *eth_dev = enic->rte_dev;
1003
1004         rte_free(eth_dev->data->mac_addrs);
1005 }
1006
1007
1008 int enic_set_vnic_res(struct enic *enic)
1009 {
1010         struct rte_eth_dev *eth_dev = enic->rte_dev;
1011         int rc = 0;
1012
1013         /* With Rx scatter support, two RQs are now used per RQ used by
1014          * the application.
1015          */
1016         if (enic->rq_count < (eth_dev->data->nb_rx_queues * 2)) {
1017                 dev_err(dev, "Not enough Receive queues. Requested:%u which uses %d RQs on VIC, Configured:%u\n",
1018                         eth_dev->data->nb_rx_queues,
1019                         eth_dev->data->nb_rx_queues * 2, enic->rq_count);
1020                 rc = -EINVAL;
1021         }
1022         if (enic->wq_count < eth_dev->data->nb_tx_queues) {
1023                 dev_err(dev, "Not enough Transmit queues. Requested:%u, Configured:%u\n",
1024                         eth_dev->data->nb_tx_queues, enic->wq_count);
1025                 rc = -EINVAL;
1026         }
1027
1028         if (enic->cq_count < (enic->rq_count + enic->wq_count)) {
1029                 dev_err(dev, "Not enough Completion queues. Required:%u, Configured:%u\n",
1030                         enic->rq_count + enic->wq_count, enic->cq_count);
1031                 rc = -EINVAL;
1032         }
1033
1034         if (rc == 0) {
1035                 enic->rq_count = eth_dev->data->nb_rx_queues;
1036                 enic->wq_count = eth_dev->data->nb_tx_queues;
1037                 enic->cq_count = enic->rq_count + enic->wq_count;
1038         }
1039
1040         return rc;
1041 }
1042
1043 static int enic_dev_init(struct enic *enic)
1044 {
1045         int err;
1046         struct rte_eth_dev *eth_dev = enic->rte_dev;
1047
1048         vnic_dev_intr_coal_timer_info_default(enic->vdev);
1049
1050         /* Get vNIC configuration
1051         */
1052         err = enic_get_vnic_config(enic);
1053         if (err) {
1054                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
1055                 return err;
1056         }
1057
1058         eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN, 0);
1059         if (!eth_dev->data->mac_addrs) {
1060                 dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1061                 return -1;
1062         }
1063         ether_addr_copy((struct ether_addr *) enic->mac_addr,
1064                 &eth_dev->data->mac_addrs[0]);
1065
1066
1067         /* Get available resource counts
1068         */
1069         enic_get_res_counts(enic);
1070
1071         vnic_dev_set_reset_flag(enic->vdev, 0);
1072
1073         return 0;
1074
1075 }
1076
1077 int enic_probe(struct enic *enic)
1078 {
1079         struct rte_pci_device *pdev = enic->pdev;
1080         int err = -1;
1081
1082         dev_debug(enic, " Initializing ENIC PMD version %s\n", DRV_VERSION);
1083
1084         enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1085         enic->bar0.len = pdev->mem_resource[0].len;
1086
1087         /* Register vNIC device */
1088         enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1089         if (!enic->vdev) {
1090                 dev_err(enic, "vNIC registration failed, aborting\n");
1091                 goto err_out;
1092         }
1093
1094         LIST_INIT(&enic->memzone_list);
1095         rte_spinlock_init(&enic->memzone_list_lock);
1096
1097         vnic_register_cbacks(enic->vdev,
1098                 enic_alloc_consistent,
1099                 enic_free_consistent);
1100
1101         /* Issue device open to get device in known state */
1102         err = enic_dev_open(enic);
1103         if (err) {
1104                 dev_err(enic, "vNIC dev open failed, aborting\n");
1105                 goto err_out_unregister;
1106         }
1107
1108         /* Set ingress vlan rewrite mode before vnic initialization */
1109         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1110                 IG_VLAN_REWRITE_MODE_PASS_THRU);
1111         if (err) {
1112                 dev_err(enic,
1113                         "Failed to set ingress vlan rewrite mode, aborting.\n");
1114                 goto err_out_dev_close;
1115         }
1116
1117         /* Issue device init to initialize the vnic-to-switch link.
1118          * We'll start with carrier off and wait for link UP
1119          * notification later to turn on carrier.  We don't need
1120          * to wait here for the vnic-to-switch link initialization
1121          * to complete; link UP notification is the indication that
1122          * the process is complete.
1123          */
1124
1125         err = vnic_dev_init(enic->vdev, 0);
1126         if (err) {
1127                 dev_err(enic, "vNIC dev init failed, aborting\n");
1128                 goto err_out_dev_close;
1129         }
1130
1131         err = enic_dev_init(enic);
1132         if (err) {
1133                 dev_err(enic, "Device initialization failed, aborting\n");
1134                 goto err_out_dev_close;
1135         }
1136
1137         return 0;
1138
1139 err_out_dev_close:
1140         vnic_dev_close(enic->vdev);
1141 err_out_unregister:
1142         vnic_dev_unregister(enic->vdev);
1143 err_out:
1144         return err;
1145 }
1146
1147 void enic_remove(struct enic *enic)
1148 {
1149         enic_dev_deinit(enic);
1150         vnic_dev_close(enic->vdev);
1151         vnic_dev_unregister(enic->vdev);
1152 }