net/cxgbe: improve latency for slow traffic
[dpdk.git] / drivers / net / cxgbe / cxgbe_ethdev.c
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2014-2017 Chelsio Communications.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Chelsio Communications nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33
34 #include <sys/queue.h>
35 #include <stdio.h>
36 #include <errno.h>
37 #include <stdint.h>
38 #include <string.h>
39 #include <unistd.h>
40 #include <stdarg.h>
41 #include <inttypes.h>
42 #include <netinet/in.h>
43
44 #include <rte_byteorder.h>
45 #include <rte_common.h>
46 #include <rte_cycles.h>
47 #include <rte_interrupts.h>
48 #include <rte_log.h>
49 #include <rte_debug.h>
50 #include <rte_pci.h>
51 #include <rte_atomic.h>
52 #include <rte_branch_prediction.h>
53 #include <rte_memory.h>
54 #include <rte_memzone.h>
55 #include <rte_tailq.h>
56 #include <rte_eal.h>
57 #include <rte_alarm.h>
58 #include <rte_ether.h>
59 #include <rte_ethdev.h>
60 #include <rte_ethdev_pci.h>
61 #include <rte_atomic.h>
62 #include <rte_malloc.h>
63 #include <rte_random.h>
64 #include <rte_dev.h>
65
66 #include "cxgbe.h"
67
68 /*
69  * Macros needed to support the PCI Device ID Table ...
70  */
71 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
72         static const struct rte_pci_id cxgb4_pci_tbl[] = {
73 #define CH_PCI_DEVICE_ID_FUNCTION 0x4
74
75 #define PCI_VENDOR_ID_CHELSIO 0x1425
76
77 #define CH_PCI_ID_TABLE_ENTRY(devid) \
78                 { RTE_PCI_DEVICE(PCI_VENDOR_ID_CHELSIO, (devid)) }
79
80 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
81                 { .vendor_id = 0, } \
82         }
83
84 /*
85  *... and the PCI ID Table itself ...
86  */
87 #include "t4_pci_id_tbl.h"
88
89 static uint16_t cxgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
90                                 uint16_t nb_pkts)
91 {
92         struct sge_eth_txq *txq = (struct sge_eth_txq *)tx_queue;
93         uint16_t pkts_sent, pkts_remain;
94         uint16_t total_sent = 0;
95         int ret = 0;
96
97         CXGBE_DEBUG_TX(adapter, "%s: txq = %p; tx_pkts = %p; nb_pkts = %d\n",
98                        __func__, txq, tx_pkts, nb_pkts);
99
100         t4_os_lock(&txq->txq_lock);
101         /* free up desc from already completed tx */
102         reclaim_completed_tx(&txq->q);
103         while (total_sent < nb_pkts) {
104                 pkts_remain = nb_pkts - total_sent;
105
106                 for (pkts_sent = 0; pkts_sent < pkts_remain; pkts_sent++) {
107                         ret = t4_eth_xmit(txq, tx_pkts[total_sent + pkts_sent],
108                                           nb_pkts);
109                         if (ret < 0)
110                                 break;
111                 }
112                 if (!pkts_sent)
113                         break;
114                 total_sent += pkts_sent;
115                 /* reclaim as much as possible */
116                 reclaim_completed_tx(&txq->q);
117         }
118
119         t4_os_unlock(&txq->txq_lock);
120         return total_sent;
121 }
122
123 static uint16_t cxgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
124                                 uint16_t nb_pkts)
125 {
126         struct sge_eth_rxq *rxq = (struct sge_eth_rxq *)rx_queue;
127         unsigned int work_done;
128
129         CXGBE_DEBUG_RX(adapter, "%s: rxq->rspq.cntxt_id = %u; nb_pkts = %d\n",
130                        __func__, rxq->rspq.cntxt_id, nb_pkts);
131
132         if (cxgbe_poll(&rxq->rspq, rx_pkts, (unsigned int)nb_pkts, &work_done))
133                 dev_err(adapter, "error in cxgbe poll\n");
134
135         CXGBE_DEBUG_RX(adapter, "%s: work_done = %u\n", __func__, work_done);
136         return work_done;
137 }
138
139 static void cxgbe_dev_info_get(struct rte_eth_dev *eth_dev,
140                                struct rte_eth_dev_info *device_info)
141 {
142         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
143         struct adapter *adapter = pi->adapter;
144         int max_queues = adapter->sge.max_ethqsets / adapter->params.nports;
145
146         static const struct rte_eth_desc_lim cxgbe_desc_lim = {
147                 .nb_max = CXGBE_MAX_RING_DESC_SIZE,
148                 .nb_min = CXGBE_MIN_RING_DESC_SIZE,
149                 .nb_align = 1,
150         };
151
152         device_info->pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
153
154         device_info->min_rx_bufsize = CXGBE_MIN_RX_BUFSIZE;
155         device_info->max_rx_pktlen = CXGBE_MAX_RX_PKTLEN;
156         device_info->max_rx_queues = max_queues;
157         device_info->max_tx_queues = max_queues;
158         device_info->max_mac_addrs = 1;
159         /* XXX: For now we support one MAC/port */
160         device_info->max_vfs = adapter->params.arch.vfcount;
161         device_info->max_vmdq_pools = 0; /* XXX: For now no support for VMDQ */
162
163         device_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP |
164                                        DEV_RX_OFFLOAD_IPV4_CKSUM |
165                                        DEV_RX_OFFLOAD_UDP_CKSUM |
166                                        DEV_RX_OFFLOAD_TCP_CKSUM;
167
168         device_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT |
169                                        DEV_TX_OFFLOAD_IPV4_CKSUM |
170                                        DEV_TX_OFFLOAD_UDP_CKSUM |
171                                        DEV_TX_OFFLOAD_TCP_CKSUM |
172                                        DEV_TX_OFFLOAD_TCP_TSO;
173
174         device_info->reta_size = pi->rss_size;
175
176         device_info->rx_desc_lim = cxgbe_desc_lim;
177         device_info->tx_desc_lim = cxgbe_desc_lim;
178         device_info->speed_capa = ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G;
179 }
180
181 static void cxgbe_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
182 {
183         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
184         struct adapter *adapter = pi->adapter;
185
186         t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
187                       1, -1, 1, -1, false);
188 }
189
190 static void cxgbe_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
191 {
192         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
193         struct adapter *adapter = pi->adapter;
194
195         t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
196                       0, -1, 1, -1, false);
197 }
198
199 static void cxgbe_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
200 {
201         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
202         struct adapter *adapter = pi->adapter;
203
204         /* TODO: address filters ?? */
205
206         t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
207                       -1, 1, 1, -1, false);
208 }
209
210 static void cxgbe_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
211 {
212         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
213         struct adapter *adapter = pi->adapter;
214
215         /* TODO: address filters ?? */
216
217         t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
218                       -1, 0, 1, -1, false);
219 }
220
221 static int cxgbe_dev_link_update(struct rte_eth_dev *eth_dev,
222                                  __rte_unused int wait_to_complete)
223 {
224         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
225         struct adapter *adapter = pi->adapter;
226         struct sge *s = &adapter->sge;
227         struct rte_eth_link *old_link = &eth_dev->data->dev_link;
228         unsigned int work_done, budget = 4;
229
230         cxgbe_poll(&s->fw_evtq, NULL, budget, &work_done);
231         if (old_link->link_status == pi->link_cfg.link_ok)
232                 return -1;  /* link not changed */
233
234         eth_dev->data->dev_link.link_status = pi->link_cfg.link_ok;
235         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
236         eth_dev->data->dev_link.link_speed = pi->link_cfg.speed;
237
238         /* link has changed */
239         return 0;
240 }
241
242 static int cxgbe_dev_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
243 {
244         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
245         struct adapter *adapter = pi->adapter;
246         struct rte_eth_dev_info dev_info;
247         int err;
248         uint16_t new_mtu = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
249
250         cxgbe_dev_info_get(eth_dev, &dev_info);
251
252         /* Must accommodate at least ETHER_MIN_MTU */
253         if ((new_mtu < ETHER_MIN_MTU) || (new_mtu > dev_info.max_rx_pktlen))
254                 return -EINVAL;
255
256         /* set to jumbo mode if needed */
257         if (new_mtu > ETHER_MAX_LEN)
258                 eth_dev->data->dev_conf.rxmode.jumbo_frame = 1;
259         else
260                 eth_dev->data->dev_conf.rxmode.jumbo_frame = 0;
261
262         err = t4_set_rxmode(adapter, adapter->mbox, pi->viid, new_mtu, -1, -1,
263                             -1, -1, true);
264         if (!err)
265                 eth_dev->data->dev_conf.rxmode.max_rx_pkt_len = new_mtu;
266
267         return err;
268 }
269
270 static int cxgbe_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
271                                     uint16_t tx_queue_id);
272 static int cxgbe_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
273                                     uint16_t tx_queue_id);
274 static void cxgbe_dev_tx_queue_release(void *q);
275 static void cxgbe_dev_rx_queue_release(void *q);
276
277 /*
278  * Stop device.
279  */
280 static void cxgbe_dev_close(struct rte_eth_dev *eth_dev)
281 {
282         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
283         struct adapter *adapter = pi->adapter;
284         int i, dev_down = 0;
285
286         CXGBE_FUNC_TRACE();
287
288         if (!(adapter->flags & FULL_INIT_DONE))
289                 return;
290
291         cxgbe_down(pi);
292
293         /*
294          *  We clear queues only if both tx and rx path of the port
295          *  have been disabled
296          */
297         t4_sge_eth_clear_queues(pi);
298
299         /*  See if all ports are down */
300         for_each_port(adapter, i) {
301                 pi = adap2pinfo(adapter, i);
302                 /*
303                  * Skip first port of the adapter since it will be closed
304                  * by DPDK
305                  */
306                 if (i == 0)
307                         continue;
308                 dev_down += (pi->eth_dev->data->dev_started == 0) ? 1 : 0;
309         }
310
311         /* If rest of the ports are stopped, then free up resources */
312         if (dev_down == (adapter->params.nports - 1))
313                 cxgbe_close(adapter);
314 }
315
316 /* Start the device.
317  * It returns 0 on success.
318  */
319 static int cxgbe_dev_start(struct rte_eth_dev *eth_dev)
320 {
321         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
322         struct adapter *adapter = pi->adapter;
323         int err = 0, i;
324
325         CXGBE_FUNC_TRACE();
326
327         /*
328          * If we don't have a connection to the firmware there's nothing we
329          * can do.
330          */
331         if (!(adapter->flags & FW_OK)) {
332                 err = -ENXIO;
333                 goto out;
334         }
335
336         if (!(adapter->flags & FULL_INIT_DONE)) {
337                 err = cxgbe_up(adapter);
338                 if (err < 0)
339                         goto out;
340         }
341
342         err = setup_rss(pi);
343         if (err)
344                 goto out;
345
346         for (i = 0; i < pi->n_tx_qsets; i++) {
347                 err = cxgbe_dev_tx_queue_start(eth_dev, i);
348                 if (err)
349                         goto out;
350         }
351
352         for (i = 0; i < pi->n_rx_qsets; i++) {
353                 err = cxgbe_dev_rx_queue_start(eth_dev, i);
354                 if (err)
355                         goto out;
356         }
357
358         err = link_start(pi);
359         if (err)
360                 goto out;
361
362 out:
363         return err;
364 }
365
366 /*
367  * Stop device: disable rx and tx functions to allow for reconfiguring.
368  */
369 static void cxgbe_dev_stop(struct rte_eth_dev *eth_dev)
370 {
371         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
372         struct adapter *adapter = pi->adapter;
373
374         CXGBE_FUNC_TRACE();
375
376         if (!(adapter->flags & FULL_INIT_DONE))
377                 return;
378
379         cxgbe_down(pi);
380
381         /*
382          *  We clear queues only if both tx and rx path of the port
383          *  have been disabled
384          */
385         t4_sge_eth_clear_queues(pi);
386 }
387
388 static int cxgbe_dev_configure(struct rte_eth_dev *eth_dev)
389 {
390         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
391         struct adapter *adapter = pi->adapter;
392         int err;
393
394         CXGBE_FUNC_TRACE();
395
396         if (!(adapter->flags & FW_QUEUE_BOUND)) {
397                 err = setup_sge_fwevtq(adapter);
398                 if (err)
399                         return err;
400                 adapter->flags |= FW_QUEUE_BOUND;
401         }
402
403         err = cfg_queue_count(eth_dev);
404         if (err)
405                 return err;
406
407         return 0;
408 }
409
410 static int cxgbe_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
411                                     uint16_t tx_queue_id)
412 {
413         int ret;
414         struct sge_eth_txq *txq = (struct sge_eth_txq *)
415                                   (eth_dev->data->tx_queues[tx_queue_id]);
416
417         dev_debug(NULL, "%s: tx_queue_id = %d\n", __func__, tx_queue_id);
418
419         ret = t4_sge_eth_txq_start(txq);
420         if (ret == 0)
421                 eth_dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
422
423         return ret;
424 }
425
426 static int cxgbe_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
427                                    uint16_t tx_queue_id)
428 {
429         int ret;
430         struct sge_eth_txq *txq = (struct sge_eth_txq *)
431                                   (eth_dev->data->tx_queues[tx_queue_id]);
432
433         dev_debug(NULL, "%s: tx_queue_id = %d\n", __func__, tx_queue_id);
434
435         ret = t4_sge_eth_txq_stop(txq);
436         if (ret == 0)
437                 eth_dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
438
439         return ret;
440 }
441
442 static int cxgbe_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
443                                     uint16_t queue_idx, uint16_t nb_desc,
444                                     unsigned int socket_id,
445                                     const struct rte_eth_txconf *tx_conf)
446 {
447         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
448         struct adapter *adapter = pi->adapter;
449         struct sge *s = &adapter->sge;
450         struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset + queue_idx];
451         int err = 0;
452         unsigned int temp_nb_desc;
453
454         RTE_SET_USED(tx_conf);
455
456         dev_debug(adapter, "%s: eth_dev->data->nb_tx_queues = %d; queue_idx = %d; nb_desc = %d; socket_id = %d; pi->first_qset = %u\n",
457                   __func__, eth_dev->data->nb_tx_queues, queue_idx, nb_desc,
458                   socket_id, pi->first_qset);
459
460         /*  Free up the existing queue  */
461         if (eth_dev->data->tx_queues[queue_idx]) {
462                 cxgbe_dev_tx_queue_release(eth_dev->data->tx_queues[queue_idx]);
463                 eth_dev->data->tx_queues[queue_idx] = NULL;
464         }
465
466         eth_dev->data->tx_queues[queue_idx] = (void *)txq;
467
468         /* Sanity Checking
469          *
470          * nb_desc should be > 1023 and <= CXGBE_MAX_RING_DESC_SIZE
471          */
472         temp_nb_desc = nb_desc;
473         if (nb_desc < CXGBE_MIN_RING_DESC_SIZE) {
474                 dev_warn(adapter, "%s: number of descriptors must be >= %d. Using default [%d]\n",
475                          __func__, CXGBE_MIN_RING_DESC_SIZE,
476                          CXGBE_DEFAULT_TX_DESC_SIZE);
477                 temp_nb_desc = CXGBE_DEFAULT_TX_DESC_SIZE;
478         } else if (nb_desc > CXGBE_MAX_RING_DESC_SIZE) {
479                 dev_err(adapter, "%s: number of descriptors must be between %d and %d inclusive. Default [%d]\n",
480                         __func__, CXGBE_MIN_RING_DESC_SIZE,
481                         CXGBE_MAX_RING_DESC_SIZE, CXGBE_DEFAULT_TX_DESC_SIZE);
482                 return -(EINVAL);
483         }
484
485         txq->q.size = temp_nb_desc;
486
487         err = t4_sge_alloc_eth_txq(adapter, txq, eth_dev, queue_idx,
488                                    s->fw_evtq.cntxt_id, socket_id);
489
490         dev_debug(adapter, "%s: txq->q.cntxt_id= %d err = %d\n",
491                   __func__, txq->q.cntxt_id, err);
492
493         return err;
494 }
495
496 static void cxgbe_dev_tx_queue_release(void *q)
497 {
498         struct sge_eth_txq *txq = (struct sge_eth_txq *)q;
499
500         if (txq) {
501                 struct port_info *pi = (struct port_info *)
502                                        (txq->eth_dev->data->dev_private);
503                 struct adapter *adap = pi->adapter;
504
505                 dev_debug(adapter, "%s: pi->port_id = %d; tx_queue_id = %d\n",
506                           __func__, pi->port_id, txq->q.cntxt_id);
507
508                 t4_sge_eth_txq_release(adap, txq);
509         }
510 }
511
512 static int cxgbe_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
513                                     uint16_t rx_queue_id)
514 {
515         int ret;
516         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
517         struct adapter *adap = pi->adapter;
518         struct sge_rspq *q;
519
520         dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
521                   __func__, pi->port_id, rx_queue_id);
522
523         q = eth_dev->data->rx_queues[rx_queue_id];
524
525         ret = t4_sge_eth_rxq_start(adap, q);
526         if (ret == 0)
527                 eth_dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
528
529         return ret;
530 }
531
532 static int cxgbe_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
533                                    uint16_t rx_queue_id)
534 {
535         int ret;
536         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
537         struct adapter *adap = pi->adapter;
538         struct sge_rspq *q;
539
540         dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
541                   __func__, pi->port_id, rx_queue_id);
542
543         q = eth_dev->data->rx_queues[rx_queue_id];
544         ret = t4_sge_eth_rxq_stop(adap, q);
545         if (ret == 0)
546                 eth_dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
547
548         return ret;
549 }
550
551 static int cxgbe_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
552                                     uint16_t queue_idx, uint16_t nb_desc,
553                                     unsigned int socket_id,
554                                     const struct rte_eth_rxconf *rx_conf,
555                                     struct rte_mempool *mp)
556 {
557         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
558         struct adapter *adapter = pi->adapter;
559         struct sge *s = &adapter->sge;
560         struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset + queue_idx];
561         int err = 0;
562         int msi_idx = 0;
563         unsigned int temp_nb_desc;
564         struct rte_eth_dev_info dev_info;
565         unsigned int pkt_len = eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
566
567         RTE_SET_USED(rx_conf);
568
569         dev_debug(adapter, "%s: eth_dev->data->nb_rx_queues = %d; queue_idx = %d; nb_desc = %d; socket_id = %d; mp = %p\n",
570                   __func__, eth_dev->data->nb_rx_queues, queue_idx, nb_desc,
571                   socket_id, mp);
572
573         cxgbe_dev_info_get(eth_dev, &dev_info);
574
575         /* Must accommodate at least ETHER_MIN_MTU */
576         if ((pkt_len < dev_info.min_rx_bufsize) ||
577             (pkt_len > dev_info.max_rx_pktlen)) {
578                 dev_err(adap, "%s: max pkt len must be > %d and <= %d\n",
579                         __func__, dev_info.min_rx_bufsize,
580                         dev_info.max_rx_pktlen);
581                 return -EINVAL;
582         }
583
584         /*  Free up the existing queue  */
585         if (eth_dev->data->rx_queues[queue_idx]) {
586                 cxgbe_dev_rx_queue_release(eth_dev->data->rx_queues[queue_idx]);
587                 eth_dev->data->rx_queues[queue_idx] = NULL;
588         }
589
590         eth_dev->data->rx_queues[queue_idx] = (void *)rxq;
591
592         /* Sanity Checking
593          *
594          * nb_desc should be > 0 and <= CXGBE_MAX_RING_DESC_SIZE
595          */
596         temp_nb_desc = nb_desc;
597         if (nb_desc < CXGBE_MIN_RING_DESC_SIZE) {
598                 dev_warn(adapter, "%s: number of descriptors must be >= %d. Using default [%d]\n",
599                          __func__, CXGBE_MIN_RING_DESC_SIZE,
600                          CXGBE_DEFAULT_RX_DESC_SIZE);
601                 temp_nb_desc = CXGBE_DEFAULT_RX_DESC_SIZE;
602         } else if (nb_desc > CXGBE_MAX_RING_DESC_SIZE) {
603                 dev_err(adapter, "%s: number of descriptors must be between %d and %d inclusive. Default [%d]\n",
604                         __func__, CXGBE_MIN_RING_DESC_SIZE,
605                         CXGBE_MAX_RING_DESC_SIZE, CXGBE_DEFAULT_RX_DESC_SIZE);
606                 return -(EINVAL);
607         }
608
609         rxq->rspq.size = temp_nb_desc;
610         if ((&rxq->fl) != NULL)
611                 rxq->fl.size = temp_nb_desc;
612
613         /* Set to jumbo mode if necessary */
614         if (pkt_len > ETHER_MAX_LEN)
615                 eth_dev->data->dev_conf.rxmode.jumbo_frame = 1;
616         else
617                 eth_dev->data->dev_conf.rxmode.jumbo_frame = 0;
618
619         err = t4_sge_alloc_rxq(adapter, &rxq->rspq, false, eth_dev, msi_idx,
620                                &rxq->fl, t4_ethrx_handler,
621                                t4_get_tp_ch_map(adapter, pi->tx_chan), mp,
622                                queue_idx, socket_id);
623
624         dev_debug(adapter, "%s: err = %d; port_id = %d; cntxt_id = %u\n",
625                   __func__, err, pi->port_id, rxq->rspq.cntxt_id);
626         return err;
627 }
628
629 static void cxgbe_dev_rx_queue_release(void *q)
630 {
631         struct sge_eth_rxq *rxq = (struct sge_eth_rxq *)q;
632         struct sge_rspq *rq = &rxq->rspq;
633
634         if (rq) {
635                 struct port_info *pi = (struct port_info *)
636                                        (rq->eth_dev->data->dev_private);
637                 struct adapter *adap = pi->adapter;
638
639                 dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
640                           __func__, pi->port_id, rxq->rspq.cntxt_id);
641
642                 t4_sge_eth_rxq_release(adap, rxq);
643         }
644 }
645
646 /*
647  * Get port statistics.
648  */
649 static void cxgbe_dev_stats_get(struct rte_eth_dev *eth_dev,
650                                 struct rte_eth_stats *eth_stats)
651 {
652         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
653         struct adapter *adapter = pi->adapter;
654         struct sge *s = &adapter->sge;
655         struct port_stats ps;
656         unsigned int i;
657
658         cxgbe_stats_get(pi, &ps);
659
660         /* RX Stats */
661         eth_stats->imissed  = ps.rx_ovflow0 + ps.rx_ovflow1 +
662                               ps.rx_ovflow2 + ps.rx_ovflow3 +
663                               ps.rx_trunc0 + ps.rx_trunc1 +
664                               ps.rx_trunc2 + ps.rx_trunc3;
665         eth_stats->ierrors  = ps.rx_symbol_err + ps.rx_fcs_err +
666                               ps.rx_jabber + ps.rx_too_long + ps.rx_runt +
667                               ps.rx_len_err;
668
669         /* TX Stats */
670         eth_stats->opackets = ps.tx_frames;
671         eth_stats->obytes   = ps.tx_octets;
672         eth_stats->oerrors  = ps.tx_error_frames;
673
674         for (i = 0; i < pi->n_rx_qsets; i++) {
675                 struct sge_eth_rxq *rxq =
676                         &s->ethrxq[pi->first_qset + i];
677
678                 eth_stats->q_ipackets[i] = rxq->stats.pkts;
679                 eth_stats->q_ibytes[i] = rxq->stats.rx_bytes;
680                 eth_stats->ipackets += eth_stats->q_ipackets[i];
681                 eth_stats->ibytes += eth_stats->q_ibytes[i];
682         }
683
684         for (i = 0; i < pi->n_tx_qsets; i++) {
685                 struct sge_eth_txq *txq =
686                         &s->ethtxq[pi->first_qset + i];
687
688                 eth_stats->q_opackets[i] = txq->stats.pkts;
689                 eth_stats->q_obytes[i] = txq->stats.tx_bytes;
690                 eth_stats->q_errors[i] = txq->stats.mapping_err;
691         }
692 }
693
694 /*
695  * Reset port statistics.
696  */
697 static void cxgbe_dev_stats_reset(struct rte_eth_dev *eth_dev)
698 {
699         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
700         struct adapter *adapter = pi->adapter;
701         struct sge *s = &adapter->sge;
702         unsigned int i;
703
704         cxgbe_stats_reset(pi);
705         for (i = 0; i < pi->n_rx_qsets; i++) {
706                 struct sge_eth_rxq *rxq =
707                         &s->ethrxq[pi->first_qset + i];
708
709                 rxq->stats.pkts = 0;
710                 rxq->stats.rx_bytes = 0;
711         }
712         for (i = 0; i < pi->n_tx_qsets; i++) {
713                 struct sge_eth_txq *txq =
714                         &s->ethtxq[pi->first_qset + i];
715
716                 txq->stats.pkts = 0;
717                 txq->stats.tx_bytes = 0;
718                 txq->stats.mapping_err = 0;
719         }
720 }
721
722 static int cxgbe_flow_ctrl_get(struct rte_eth_dev *eth_dev,
723                                struct rte_eth_fc_conf *fc_conf)
724 {
725         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
726         struct link_config *lc = &pi->link_cfg;
727         int rx_pause, tx_pause;
728
729         fc_conf->autoneg = lc->fc & PAUSE_AUTONEG;
730         rx_pause = lc->fc & PAUSE_RX;
731         tx_pause = lc->fc & PAUSE_TX;
732
733         if (rx_pause && tx_pause)
734                 fc_conf->mode = RTE_FC_FULL;
735         else if (rx_pause)
736                 fc_conf->mode = RTE_FC_RX_PAUSE;
737         else if (tx_pause)
738                 fc_conf->mode = RTE_FC_TX_PAUSE;
739         else
740                 fc_conf->mode = RTE_FC_NONE;
741         return 0;
742 }
743
744 static int cxgbe_flow_ctrl_set(struct rte_eth_dev *eth_dev,
745                                struct rte_eth_fc_conf *fc_conf)
746 {
747         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
748         struct adapter *adapter = pi->adapter;
749         struct link_config *lc = &pi->link_cfg;
750
751         if (lc->supported & FW_PORT_CAP_ANEG) {
752                 if (fc_conf->autoneg)
753                         lc->requested_fc |= PAUSE_AUTONEG;
754                 else
755                         lc->requested_fc &= ~PAUSE_AUTONEG;
756         }
757
758         if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
759             (fc_conf->mode & RTE_FC_RX_PAUSE))
760                 lc->requested_fc |= PAUSE_RX;
761         else
762                 lc->requested_fc &= ~PAUSE_RX;
763
764         if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
765             (fc_conf->mode & RTE_FC_TX_PAUSE))
766                 lc->requested_fc |= PAUSE_TX;
767         else
768                 lc->requested_fc &= ~PAUSE_TX;
769
770         return t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
771                              &pi->link_cfg);
772 }
773
774 static const uint32_t *
775 cxgbe_dev_supported_ptypes_get(struct rte_eth_dev *eth_dev)
776 {
777         static const uint32_t ptypes[] = {
778                 RTE_PTYPE_L3_IPV4,
779                 RTE_PTYPE_L3_IPV6,
780                 RTE_PTYPE_UNKNOWN
781         };
782
783         if (eth_dev->rx_pkt_burst == cxgbe_recv_pkts)
784                 return ptypes;
785         return NULL;
786 }
787
788 static int cxgbe_get_eeprom_length(struct rte_eth_dev *dev)
789 {
790         RTE_SET_USED(dev);
791         return EEPROMSIZE;
792 }
793
794 /**
795  * eeprom_ptov - translate a physical EEPROM address to virtual
796  * @phys_addr: the physical EEPROM address
797  * @fn: the PCI function number
798  * @sz: size of function-specific area
799  *
800  * Translate a physical EEPROM address to virtual.  The first 1K is
801  * accessed through virtual addresses starting at 31K, the rest is
802  * accessed through virtual addresses starting at 0.
803  *
804  * The mapping is as follows:
805  * [0..1K) -> [31K..32K)
806  * [1K..1K+A) -> [31K-A..31K)
807  * [1K+A..ES) -> [0..ES-A-1K)
808  *
809  * where A = @fn * @sz, and ES = EEPROM size.
810  */
811 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
812 {
813         fn *= sz;
814         if (phys_addr < 1024)
815                 return phys_addr + (31 << 10);
816         if (phys_addr < 1024 + fn)
817                 return fn + phys_addr - 1024;
818         if (phys_addr < EEPROMSIZE)
819                 return phys_addr - 1024 - fn;
820         if (phys_addr < EEPROMVSIZE)
821                 return phys_addr - 1024;
822         return -EINVAL;
823 }
824
825 /* The next two routines implement eeprom read/write from physical addresses.
826  */
827 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
828 {
829         int vaddr = eeprom_ptov(phys_addr, adap->pf, EEPROMPFSIZE);
830
831         if (vaddr >= 0)
832                 vaddr = t4_seeprom_read(adap, vaddr, v);
833         return vaddr < 0 ? vaddr : 0;
834 }
835
836 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
837 {
838         int vaddr = eeprom_ptov(phys_addr, adap->pf, EEPROMPFSIZE);
839
840         if (vaddr >= 0)
841                 vaddr = t4_seeprom_write(adap, vaddr, v);
842         return vaddr < 0 ? vaddr : 0;
843 }
844
845 #define EEPROM_MAGIC 0x38E2F10C
846
847 static int cxgbe_get_eeprom(struct rte_eth_dev *dev,
848                             struct rte_dev_eeprom_info *e)
849 {
850         struct port_info *pi = (struct port_info *)(dev->data->dev_private);
851         struct adapter *adapter = pi->adapter;
852         u32 i, err = 0;
853         u8 *buf = rte_zmalloc(NULL, EEPROMSIZE, 0);
854
855         if (!buf)
856                 return -ENOMEM;
857
858         e->magic = EEPROM_MAGIC;
859         for (i = e->offset & ~3; !err && i < e->offset + e->length; i += 4)
860                 err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
861
862         if (!err)
863                 rte_memcpy(e->data, buf + e->offset, e->length);
864         rte_free(buf);
865         return err;
866 }
867
868 static int cxgbe_set_eeprom(struct rte_eth_dev *dev,
869                             struct rte_dev_eeprom_info *eeprom)
870 {
871         struct port_info *pi = (struct port_info *)(dev->data->dev_private);
872         struct adapter *adapter = pi->adapter;
873         u8 *buf;
874         int err = 0;
875         u32 aligned_offset, aligned_len, *p;
876
877         if (eeprom->magic != EEPROM_MAGIC)
878                 return -EINVAL;
879
880         aligned_offset = eeprom->offset & ~3;
881         aligned_len = (eeprom->length + (eeprom->offset & 3) + 3) & ~3;
882
883         if (adapter->pf > 0) {
884                 u32 start = 1024 + adapter->pf * EEPROMPFSIZE;
885
886                 if (aligned_offset < start ||
887                     aligned_offset + aligned_len > start + EEPROMPFSIZE)
888                         return -EPERM;
889         }
890
891         if (aligned_offset != eeprom->offset || aligned_len != eeprom->length) {
892                 /* RMW possibly needed for first or last words.
893                  */
894                 buf = rte_zmalloc(NULL, aligned_len, 0);
895                 if (!buf)
896                         return -ENOMEM;
897                 err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
898                 if (!err && aligned_len > 4)
899                         err = eeprom_rd_phys(adapter,
900                                              aligned_offset + aligned_len - 4,
901                                              (u32 *)&buf[aligned_len - 4]);
902                 if (err)
903                         goto out;
904                 rte_memcpy(buf + (eeprom->offset & 3), eeprom->data,
905                            eeprom->length);
906         } else {
907                 buf = eeprom->data;
908         }
909
910         err = t4_seeprom_wp(adapter, false);
911         if (err)
912                 goto out;
913
914         for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
915                 err = eeprom_wr_phys(adapter, aligned_offset, *p);
916                 aligned_offset += 4;
917         }
918
919         if (!err)
920                 err = t4_seeprom_wp(adapter, true);
921 out:
922         if (buf != eeprom->data)
923                 rte_free(buf);
924         return err;
925 }
926
927 static int cxgbe_get_regs_len(struct rte_eth_dev *eth_dev)
928 {
929         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
930         struct adapter *adapter = pi->adapter;
931
932         return t4_get_regs_len(adapter) / sizeof(uint32_t);
933 }
934
935 static int cxgbe_get_regs(struct rte_eth_dev *eth_dev,
936                           struct rte_dev_reg_info *regs)
937 {
938         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
939         struct adapter *adapter = pi->adapter;
940
941         regs->version = CHELSIO_CHIP_VERSION(adapter->params.chip) |
942                 (CHELSIO_CHIP_RELEASE(adapter->params.chip) << 10) |
943                 (1 << 16);
944
945         if (regs->data == NULL) {
946                 regs->length = cxgbe_get_regs_len(eth_dev);
947                 regs->width = sizeof(uint32_t);
948
949                 return 0;
950         }
951
952         t4_get_regs(adapter, regs->data, (regs->length * sizeof(uint32_t)));
953
954         return 0;
955 }
956
957 static const struct eth_dev_ops cxgbe_eth_dev_ops = {
958         .dev_start              = cxgbe_dev_start,
959         .dev_stop               = cxgbe_dev_stop,
960         .dev_close              = cxgbe_dev_close,
961         .promiscuous_enable     = cxgbe_dev_promiscuous_enable,
962         .promiscuous_disable    = cxgbe_dev_promiscuous_disable,
963         .allmulticast_enable    = cxgbe_dev_allmulticast_enable,
964         .allmulticast_disable   = cxgbe_dev_allmulticast_disable,
965         .dev_configure          = cxgbe_dev_configure,
966         .dev_infos_get          = cxgbe_dev_info_get,
967         .dev_supported_ptypes_get = cxgbe_dev_supported_ptypes_get,
968         .link_update            = cxgbe_dev_link_update,
969         .mtu_set                = cxgbe_dev_mtu_set,
970         .tx_queue_setup         = cxgbe_dev_tx_queue_setup,
971         .tx_queue_start         = cxgbe_dev_tx_queue_start,
972         .tx_queue_stop          = cxgbe_dev_tx_queue_stop,
973         .tx_queue_release       = cxgbe_dev_tx_queue_release,
974         .rx_queue_setup         = cxgbe_dev_rx_queue_setup,
975         .rx_queue_start         = cxgbe_dev_rx_queue_start,
976         .rx_queue_stop          = cxgbe_dev_rx_queue_stop,
977         .rx_queue_release       = cxgbe_dev_rx_queue_release,
978         .stats_get              = cxgbe_dev_stats_get,
979         .stats_reset            = cxgbe_dev_stats_reset,
980         .flow_ctrl_get          = cxgbe_flow_ctrl_get,
981         .flow_ctrl_set          = cxgbe_flow_ctrl_set,
982         .get_eeprom_length      = cxgbe_get_eeprom_length,
983         .get_eeprom             = cxgbe_get_eeprom,
984         .set_eeprom             = cxgbe_set_eeprom,
985         .get_reg                = cxgbe_get_regs,
986 };
987
988 /*
989  * Initialize driver
990  * It returns 0 on success.
991  */
992 static int eth_cxgbe_dev_init(struct rte_eth_dev *eth_dev)
993 {
994         struct rte_pci_device *pci_dev;
995         struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
996         struct adapter *adapter = NULL;
997         char name[RTE_ETH_NAME_MAX_LEN];
998         int err = 0;
999
1000         CXGBE_FUNC_TRACE();
1001
1002         eth_dev->dev_ops = &cxgbe_eth_dev_ops;
1003         eth_dev->rx_pkt_burst = &cxgbe_recv_pkts;
1004         eth_dev->tx_pkt_burst = &cxgbe_xmit_pkts;
1005
1006         /* for secondary processes, we don't initialise any further as primary
1007          * has already done this work.
1008          */
1009         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1010                 return 0;
1011
1012         pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1013
1014         snprintf(name, sizeof(name), "cxgbeadapter%d", eth_dev->data->port_id);
1015         adapter = rte_zmalloc(name, sizeof(*adapter), 0);
1016         if (!adapter)
1017                 return -1;
1018
1019         adapter->use_unpacked_mode = 1;
1020         adapter->regs = (void *)pci_dev->mem_resource[0].addr;
1021         if (!adapter->regs) {
1022                 dev_err(adapter, "%s: cannot map device registers\n", __func__);
1023                 err = -ENOMEM;
1024                 goto out_free_adapter;
1025         }
1026         adapter->pdev = pci_dev;
1027         adapter->eth_dev = eth_dev;
1028         pi->adapter = adapter;
1029
1030         err = cxgbe_probe(adapter);
1031         if (err) {
1032                 dev_err(adapter, "%s: cxgbe probe failed with err %d\n",
1033                         __func__, err);
1034                 goto out_free_adapter;
1035         }
1036
1037         return 0;
1038
1039 out_free_adapter:
1040         rte_free(adapter);
1041         return err;
1042 }
1043
1044 static int eth_cxgbe_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1045         struct rte_pci_device *pci_dev)
1046 {
1047         return rte_eth_dev_pci_generic_probe(pci_dev,
1048                 sizeof(struct port_info), eth_cxgbe_dev_init);
1049 }
1050
1051 static int eth_cxgbe_pci_remove(struct rte_pci_device *pci_dev)
1052 {
1053         return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1054 }
1055
1056 static struct rte_pci_driver rte_cxgbe_pmd = {
1057         .id_table = cxgb4_pci_tbl,
1058         .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1059         .probe = eth_cxgbe_pci_probe,
1060         .remove = eth_cxgbe_pci_remove,
1061 };
1062
1063 RTE_PMD_REGISTER_PCI(net_cxgbe, rte_cxgbe_pmd);
1064 RTE_PMD_REGISTER_PCI_TABLE(net_cxgbe, cxgb4_pci_tbl);
1065 RTE_PMD_REGISTER_KMOD_DEP(net_cxgbe, "* igb_uio | uio_pci_generic | vfio-pci");