net/hns3: fix rollback on RSS hash update
[dpdk.git] / drivers / net / cxgbe / cxgbe_main.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2014-2018 Chelsio Communications.
3  * All rights reserved.
4  */
5
6 #include <sys/queue.h>
7 #include <stdio.h>
8 #include <errno.h>
9 #include <stdint.h>
10 #include <string.h>
11 #include <unistd.h>
12 #include <stdarg.h>
13 #include <inttypes.h>
14 #include <netinet/in.h>
15
16 #include <rte_byteorder.h>
17 #include <rte_common.h>
18 #include <rte_cycles.h>
19 #include <rte_interrupts.h>
20 #include <rte_log.h>
21 #include <rte_debug.h>
22 #include <rte_pci.h>
23 #include <rte_branch_prediction.h>
24 #include <rte_memory.h>
25 #include <rte_tailq.h>
26 #include <rte_eal.h>
27 #include <rte_alarm.h>
28 #include <rte_ether.h>
29 #include <ethdev_driver.h>
30 #include <ethdev_pci.h>
31 #include <rte_random.h>
32 #include <rte_dev.h>
33 #include <rte_kvargs.h>
34
35 #include "base/common.h"
36 #include "base/t4_regs.h"
37 #include "base/t4_msg.h"
38 #include "cxgbe.h"
39 #include "cxgbe_pfvf.h"
40 #include "clip_tbl.h"
41 #include "l2t.h"
42 #include "smt.h"
43 #include "mps_tcam.h"
44
45 static const u16 cxgbe_filter_mode_features[] = {
46         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE |
47          F_PROTOCOL | F_PORT),
48         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE |
49          F_PROTOCOL | F_FCOE),
50         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_TOS |
51          F_PORT),
52         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_TOS |
53          F_FCOE),
54         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_PORT |
55          F_FCOE),
56         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_TOS |
57          F_PORT | F_FCOE),
58         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VLAN |
59          F_FCOE),
60         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VNIC_ID |
61          F_FCOE),
62         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VLAN |
63          F_FCOE),
64         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VNIC_ID |
65          F_FCOE),
66         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_VLAN | F_PORT |
67          F_FCOE),
68         (F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_VNIC_ID | F_PORT |
69          F_FCOE),
70         (F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_PROTOCOL | F_TOS |
71          F_PORT | F_FCOE),
72         (F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VLAN | F_PORT),
73         (F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VLAN | F_FCOE),
74         (F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VNIC_ID | F_PORT),
75         (F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VNIC_ID | F_FCOE),
76         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VLAN | F_PORT),
77         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VLAN | F_FCOE),
78         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VNIC_ID |
79          F_PORT),
80         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VNIC_ID |
81          F_FCOE),
82         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_VLAN | F_PORT |
83          F_FCOE),
84         (F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_VNIC_ID | F_PORT |
85          F_FCOE),
86         (F_FRAGMENTATION | F_MPSHITTYPE | F_TOS | F_VLAN | F_PORT | F_FCOE),
87         (F_FRAGMENTATION | F_MPSHITTYPE | F_TOS | F_VNIC_ID | F_PORT | F_FCOE),
88         (F_FRAGMENTATION | F_MPSHITTYPE | F_VLAN | F_VNIC_ID | F_FCOE),
89         (F_FRAGMENTATION | F_MACMATCH | F_ETHERTYPE | F_PROTOCOL | F_PORT |
90          F_FCOE),
91         (F_FRAGMENTATION | F_MACMATCH | F_ETHERTYPE | F_TOS | F_PORT | F_FCOE),
92         (F_FRAGMENTATION | F_MACMATCH | F_PROTOCOL | F_VLAN | F_PORT | F_FCOE),
93         (F_FRAGMENTATION | F_MACMATCH | F_PROTOCOL | F_VNIC_ID | F_PORT |
94          F_FCOE),
95         (F_FRAGMENTATION | F_MACMATCH | F_TOS | F_VLAN | F_PORT | F_FCOE),
96         (F_FRAGMENTATION | F_MACMATCH | F_TOS | F_VNIC_ID | F_PORT | F_FCOE),
97         (F_FRAGMENTATION | F_ETHERTYPE | F_VLAN | F_PORT | F_FCOE),
98         (F_FRAGMENTATION | F_ETHERTYPE | F_VNIC_ID | F_PORT | F_FCOE),
99         (F_FRAGMENTATION | F_PROTOCOL | F_TOS | F_VLAN | F_FCOE),
100         (F_FRAGMENTATION | F_PROTOCOL | F_TOS | F_VNIC_ID | F_FCOE),
101         (F_FRAGMENTATION | F_VLAN | F_VNIC_ID | F_PORT | F_FCOE),
102         (F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_PROTOCOL | F_PORT |
103          F_FCOE),
104         (F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_TOS | F_PORT | F_FCOE),
105         (F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VLAN | F_PORT),
106         (F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VNIC_ID | F_PORT),
107         (F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VLAN | F_PORT),
108         (F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VNIC_ID | F_PORT),
109         (F_MPSHITTYPE | F_ETHERTYPE | F_VLAN | F_PORT | F_FCOE),
110         (F_MPSHITTYPE | F_ETHERTYPE | F_VNIC_ID | F_PORT | F_FCOE),
111         (F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VLAN | F_PORT | F_FCOE),
112         (F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VNIC_ID | F_PORT | F_FCOE),
113         (F_MPSHITTYPE | F_VLAN | F_VNIC_ID | F_PORT),
114 };
115
116 /**
117  * Allocate a chunk of memory. The allocated memory is cleared.
118  */
119 void *t4_alloc_mem(size_t size)
120 {
121         return rte_zmalloc(NULL, size, 0);
122 }
123
124 /**
125  * Free memory allocated through t4_alloc_mem().
126  */
127 void t4_free_mem(void *addr)
128 {
129         rte_free(addr);
130 }
131
132 /*
133  * Response queue handler for the FW event queue.
134  */
135 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
136                           __rte_unused const struct pkt_gl *gl)
137 {
138         u8 opcode = ((const struct rss_header *)rsp)->opcode;
139
140         rsp++;                                          /* skip RSS header */
141
142         /*
143          * FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
144          */
145         if (unlikely(opcode == CPL_FW4_MSG &&
146                      ((const struct cpl_fw4_msg *)rsp)->type ==
147                       FW_TYPE_RSSCPL)) {
148                 rsp++;
149                 opcode = ((const struct rss_header *)rsp)->opcode;
150                 rsp++;
151                 if (opcode != CPL_SGE_EGR_UPDATE) {
152                         dev_err(q->adapter, "unexpected FW4/CPL %#x on FW event queue\n",
153                                 opcode);
154                         goto out;
155                 }
156         }
157
158         if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
159                 /* do nothing */
160         } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
161                 const struct cpl_fw6_msg *msg = (const void *)rsp;
162
163                 t4_handle_fw_rpl(q->adapter, msg->data);
164         } else if (opcode == CPL_ABORT_RPL_RSS) {
165                 const struct cpl_abort_rpl_rss *p = (const void *)rsp;
166
167                 cxgbe_hash_del_filter_rpl(q->adapter, p);
168         } else if (opcode == CPL_SET_TCB_RPL) {
169                 const struct cpl_set_tcb_rpl *p = (const void *)rsp;
170
171                 cxgbe_filter_rpl(q->adapter, p);
172         } else if (opcode == CPL_ACT_OPEN_RPL) {
173                 const struct cpl_act_open_rpl *p = (const void *)rsp;
174
175                 cxgbe_hash_filter_rpl(q->adapter, p);
176         } else if (opcode == CPL_L2T_WRITE_RPL) {
177                 const struct cpl_l2t_write_rpl *p = (const void *)rsp;
178
179                 cxgbe_do_l2t_write_rpl(q->adapter, p);
180         } else if (opcode == CPL_SMT_WRITE_RPL) {
181                 const struct cpl_smt_write_rpl *p = (const void *)rsp;
182
183                 cxgbe_do_smt_write_rpl(q->adapter, p);
184         } else {
185                 dev_err(adapter, "unexpected CPL %#x on FW event queue\n",
186                         opcode);
187         }
188 out:
189         return 0;
190 }
191
192 /**
193  * Setup sge control queues to pass control information.
194  */
195 int cxgbe_setup_sge_ctrl_txq(struct adapter *adapter)
196 {
197         struct sge *s = &adapter->sge;
198         int err = 0, i = 0;
199
200         for_each_port(adapter, i) {
201                 struct port_info *pi = adap2pinfo(adapter, i);
202                 char name[RTE_ETH_NAME_MAX_LEN];
203                 struct sge_ctrl_txq *q = &s->ctrlq[i];
204
205                 q->q.size = 1024;
206                 err = t4_sge_alloc_ctrl_txq(adapter, q,
207                                             adapter->eth_dev,  i,
208                                             s->fw_evtq.cntxt_id,
209                                             rte_socket_id());
210                 if (err) {
211                         dev_err(adapter, "Failed to alloc ctrl txq. Err: %d",
212                                 err);
213                         goto out;
214                 }
215                 snprintf(name, sizeof(name), "%s_ctrl_pool_%d",
216                          pi->eth_dev->device->driver->name,
217                          pi->eth_dev->data->port_id);
218                 q->mb_pool = rte_pktmbuf_pool_create(name, s->ctrlq[i].q.size,
219                                                      RTE_CACHE_LINE_SIZE,
220                                                      RTE_MBUF_PRIV_ALIGN,
221                                                      RTE_MBUF_DEFAULT_BUF_SIZE,
222                                                      SOCKET_ID_ANY);
223                 if (!q->mb_pool) {
224                         err = -rte_errno;
225                         dev_err(adapter,
226                                 "Can't create ctrl pool for port %d. Err: %d\n",
227                                 pi->eth_dev->data->port_id, err);
228                         goto out;
229                 }
230         }
231         return 0;
232 out:
233         t4_free_sge_resources(adapter);
234         return err;
235 }
236
237 /**
238  * cxgbe_poll_for_completion: Poll rxq for completion
239  * @q: rxq to poll
240  * @ms: milliseconds to delay
241  * @cnt: number of times to poll
242  * @c: completion to check for 'done' status
243  *
244  * Polls the rxq for reples until completion is done or the count
245  * expires.
246  */
247 int cxgbe_poll_for_completion(struct sge_rspq *q, unsigned int ms,
248                               unsigned int cnt, struct t4_completion *c)
249 {
250         unsigned int i;
251         unsigned int work_done, budget = 32;
252
253         if (!c)
254                 return -EINVAL;
255
256         for (i = 0; i < cnt; i++) {
257                 cxgbe_poll(q, NULL, budget, &work_done);
258                 t4_os_lock(&c->lock);
259                 if (c->done) {
260                         t4_os_unlock(&c->lock);
261                         return 0;
262                 }
263                 t4_os_unlock(&c->lock);
264                 rte_delay_ms(ms);
265         }
266         return -ETIMEDOUT;
267 }
268
269 int cxgbe_setup_sge_fwevtq(struct adapter *adapter)
270 {
271         struct sge *s = &adapter->sge;
272         int err = 0;
273         int msi_idx = 0;
274
275         err = t4_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->eth_dev,
276                                msi_idx, NULL, fwevtq_handler, -1, NULL, 0,
277                                rte_socket_id());
278         return err;
279 }
280
281 static int closest_timer(const struct sge *s, int time)
282 {
283         unsigned int i, match = 0;
284         int delta, min_delta = INT_MAX;
285
286         for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
287                 delta = time - s->timer_val[i];
288                 if (delta < 0)
289                         delta = -delta;
290                 if (delta < min_delta) {
291                         min_delta = delta;
292                         match = i;
293                 }
294         }
295         return match;
296 }
297
298 static int closest_thres(const struct sge *s, int thres)
299 {
300         unsigned int i, match = 0;
301         int delta, min_delta = INT_MAX;
302
303         for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
304                 delta = thres - s->counter_val[i];
305                 if (delta < 0)
306                         delta = -delta;
307                 if (delta < min_delta) {
308                         min_delta = delta;
309                         match = i;
310                 }
311         }
312         return match;
313 }
314
315 /**
316  * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
317  * @q: the Rx queue
318  * @us: the hold-off time in us, or 0 to disable timer
319  * @cnt: the hold-off packet count, or 0 to disable counter
320  *
321  * Sets an Rx queue's interrupt hold-off time and packet count.  At least
322  * one of the two needs to be enabled for the queue to generate interrupts.
323  */
324 int cxgb4_set_rspq_intr_params(struct sge_rspq *q, unsigned int us,
325                                unsigned int cnt)
326 {
327         struct adapter *adap = q->adapter;
328         unsigned int timer_val;
329
330         if (cnt) {
331                 int err;
332                 u32 v, new_idx;
333
334                 new_idx = closest_thres(&adap->sge, cnt);
335                 if (q->desc && q->pktcnt_idx != new_idx) {
336                         /* the queue has already been created, update it */
337                         v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
338                             V_FW_PARAMS_PARAM_X(
339                             FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
340                             V_FW_PARAMS_PARAM_YZ(q->cntxt_id);
341                         err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
342                                             &v, &new_idx);
343                         if (err)
344                                 return err;
345                 }
346                 q->pktcnt_idx = new_idx;
347         }
348
349         timer_val = (us == 0) ? X_TIMERREG_RESTART_COUNTER :
350                                 closest_timer(&adap->sge, us);
351
352         if ((us | cnt) == 0)
353                 q->intr_params = V_QINTR_TIMER_IDX(X_TIMERREG_UPDATE_CIDX);
354         else
355                 q->intr_params = V_QINTR_TIMER_IDX(timer_val) |
356                                  V_QINTR_CNT_EN(cnt > 0);
357         return 0;
358 }
359
360 /**
361  * Allocate an active-open TID and set it to the supplied value.
362  */
363 int cxgbe_alloc_atid(struct tid_info *t, void *data)
364 {
365         int atid = -1;
366
367         t4_os_lock(&t->atid_lock);
368         if (t->afree) {
369                 union aopen_entry *p = t->afree;
370
371                 atid = p - t->atid_tab;
372                 t->afree = p->next;
373                 p->data = data;
374                 t->atids_in_use++;
375         }
376         t4_os_unlock(&t->atid_lock);
377         return atid;
378 }
379
380 /**
381  * Release an active-open TID.
382  */
383 void cxgbe_free_atid(struct tid_info *t, unsigned int atid)
384 {
385         union aopen_entry *p = &t->atid_tab[atid];
386
387         t4_os_lock(&t->atid_lock);
388         p->next = t->afree;
389         t->afree = p;
390         t->atids_in_use--;
391         t4_os_unlock(&t->atid_lock);
392 }
393
394 /**
395  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
396  */
397 static void mk_tid_release(struct rte_mbuf *mbuf, unsigned int tid)
398 {
399         struct cpl_tid_release *req;
400
401         req = rte_pktmbuf_mtod(mbuf, struct cpl_tid_release *);
402         INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
403 }
404
405 /**
406  * Release a TID and inform HW.  If we are unable to allocate the release
407  * message we defer to a work queue.
408  */
409 void cxgbe_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
410                       unsigned short family)
411 {
412         struct rte_mbuf *mbuf;
413         struct adapter *adap = container_of(t, struct adapter, tids);
414
415         WARN_ON(tid >= t->ntids);
416
417         if (t->tid_tab[tid]) {
418                 t->tid_tab[tid] = NULL;
419                 __atomic_sub_fetch(&t->conns_in_use, 1, __ATOMIC_RELAXED);
420                 if (t->hash_base && tid >= t->hash_base) {
421                         if (family == FILTER_TYPE_IPV4)
422                                 __atomic_sub_fetch(&t->hash_tids_in_use, 1,
423                                                    __ATOMIC_RELAXED);
424                 } else {
425                         if (family == FILTER_TYPE_IPV4)
426                                 __atomic_sub_fetch(&t->tids_in_use, 1,
427                                                    __ATOMIC_RELAXED);
428                 }
429         }
430
431         mbuf = rte_pktmbuf_alloc((&adap->sge.ctrlq[chan])->mb_pool);
432         if (mbuf) {
433                 mbuf->data_len = sizeof(struct cpl_tid_release);
434                 mbuf->pkt_len = mbuf->data_len;
435                 mk_tid_release(mbuf, tid);
436                 t4_mgmt_tx(&adap->sge.ctrlq[chan], mbuf);
437         }
438 }
439
440 /**
441  * Insert a TID.
442  */
443 void cxgbe_insert_tid(struct tid_info *t, void *data, unsigned int tid,
444                       unsigned short family)
445 {
446         t->tid_tab[tid] = data;
447         if (t->hash_base && tid >= t->hash_base) {
448                 if (family == FILTER_TYPE_IPV4)
449                         __atomic_add_fetch(&t->hash_tids_in_use, 1,
450                                            __ATOMIC_RELAXED);
451         } else {
452                 if (family == FILTER_TYPE_IPV4)
453                         __atomic_add_fetch(&t->tids_in_use, 1,
454                                            __ATOMIC_RELAXED);
455         }
456
457         __atomic_add_fetch(&t->conns_in_use, 1, __ATOMIC_RELAXED);
458 }
459
460 /**
461  * Free TID tables.
462  */
463 static void tid_free(struct tid_info *t)
464 {
465         if (t->tid_tab) {
466                 rte_bitmap_free(t->ftid_bmap);
467
468                 if (t->ftid_bmap_array)
469                         t4_os_free(t->ftid_bmap_array);
470
471                 t4_os_free(t->tid_tab);
472         }
473
474         memset(t, 0, sizeof(struct tid_info));
475 }
476
477 /**
478  * Allocate and initialize the TID tables.  Returns 0 on success.
479  */
480 static int tid_init(struct tid_info *t)
481 {
482         size_t size;
483         unsigned int ftid_bmap_size;
484         unsigned int natids = t->natids;
485         unsigned int max_ftids = t->nftids;
486
487         ftid_bmap_size = rte_bitmap_get_memory_footprint(t->nftids);
488         size = t->ntids * sizeof(*t->tid_tab) +
489                 max_ftids * sizeof(*t->ftid_tab) +
490                 natids * sizeof(*t->atid_tab);
491
492         t->tid_tab = t4_os_alloc(size);
493         if (!t->tid_tab)
494                 return -ENOMEM;
495
496         t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
497         t->ftid_tab = (struct filter_entry *)&t->atid_tab[t->natids];
498         t->ftid_bmap_array = t4_os_alloc(ftid_bmap_size);
499         if (!t->ftid_bmap_array) {
500                 tid_free(t);
501                 return -ENOMEM;
502         }
503
504         t4_os_lock_init(&t->atid_lock);
505         t4_os_lock_init(&t->ftid_lock);
506
507         t->afree = NULL;
508         t->atids_in_use = 0;
509         t->tids_in_use = 0;
510         t->conns_in_use = 0;
511
512         /* Setup the free list for atid_tab and clear the stid bitmap. */
513         if (natids) {
514                 while (--natids)
515                         t->atid_tab[natids - 1].next = &t->atid_tab[natids];
516                 t->afree = t->atid_tab;
517         }
518
519         t->ftid_bmap = rte_bitmap_init(t->nftids, t->ftid_bmap_array,
520                                        ftid_bmap_size);
521         if (!t->ftid_bmap) {
522                 tid_free(t);
523                 return -ENOMEM;
524         }
525
526         return 0;
527 }
528
529 static inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
530                       unsigned int us, unsigned int cnt,
531                       unsigned int size, unsigned int iqe_size)
532 {
533         q->adapter = adap;
534         cxgb4_set_rspq_intr_params(q, us, cnt);
535         q->iqe_len = iqe_size;
536         q->size = size;
537 }
538
539 int cxgbe_cfg_queue_count(struct rte_eth_dev *eth_dev)
540 {
541         struct port_info *temp_pi, *pi = eth_dev->data->dev_private;
542         struct adapter *adap = pi->adapter;
543         u16 first_txq = 0, first_rxq = 0;
544         struct sge *s = &adap->sge;
545         u16 i, max_rxqs, max_txqs;
546
547         max_rxqs = s->max_ethqsets;
548         max_txqs = s->max_ethqsets;
549         for_each_port(adap, i) {
550                 temp_pi = adap2pinfo(adap, i);
551                 if (i == pi->port_id)
552                         break;
553
554                 if (max_rxqs <= temp_pi->n_rx_qsets ||
555                     max_txqs <= temp_pi->n_tx_qsets)
556                         return -ENOMEM;
557
558                 first_rxq += temp_pi->n_rx_qsets;
559                 first_txq += temp_pi->n_tx_qsets;
560                 max_rxqs -= temp_pi->n_rx_qsets;
561                 max_txqs -= temp_pi->n_tx_qsets;
562         }
563
564         if ((eth_dev->data->nb_rx_queues < 1) ||
565             (eth_dev->data->nb_tx_queues < 1))
566                 return -EINVAL;
567
568         if (eth_dev->data->nb_rx_queues > max_rxqs ||
569             eth_dev->data->nb_tx_queues > max_txqs)
570                 return -EINVAL;
571
572         /* We must configure RSS, since config has changed*/
573         pi->flags &= ~PORT_RSS_DONE;
574
575         pi->n_rx_qsets = eth_dev->data->nb_rx_queues;
576         pi->n_tx_qsets = eth_dev->data->nb_tx_queues;
577         pi->first_rxqset = first_rxq;
578         pi->first_txqset = first_txq;
579
580         return 0;
581 }
582
583 void cxgbe_cfg_queues_free(struct adapter *adap)
584 {
585         if (adap->sge.ethtxq) {
586                 rte_free(adap->sge.ethtxq);
587                 adap->sge.ethtxq = NULL;
588         }
589
590         if (adap->sge.ethrxq) {
591                 rte_free(adap->sge.ethrxq);
592                 adap->sge.ethrxq = NULL;
593         }
594
595         adap->flags &= ~CFG_QUEUES;
596 }
597
598 int cxgbe_cfg_queues(struct rte_eth_dev *eth_dev)
599 {
600         struct port_info *pi = eth_dev->data->dev_private;
601         struct adapter *adap = pi->adapter;
602         struct sge *s = &adap->sge;
603         u16 i;
604
605         if (!(adap->flags & CFG_QUEUES)) {
606                 s->ethrxq = rte_calloc_socket(NULL, s->max_ethqsets,
607                                               sizeof(struct sge_eth_rxq), 0,
608                                               rte_socket_id());
609                 if (!s->ethrxq)
610                         return -ENOMEM;
611
612                 s->ethtxq = rte_calloc_socket(NULL, s->max_ethqsets,
613                                               sizeof(struct sge_eth_txq), 0,
614                                               rte_socket_id());
615                 if (!s->ethtxq) {
616                         rte_free(s->ethrxq);
617                         s->ethrxq = NULL;
618                         return -ENOMEM;
619                 }
620
621                 for (i = 0; i < s->max_ethqsets; i++) {
622                         struct sge_eth_rxq *r = &s->ethrxq[i];
623                         struct sge_eth_txq *t = &s->ethtxq[i];
624
625                         init_rspq(adap, &r->rspq, 5, 32, 1024, 64);
626                         r->fl.size = 1024;
627
628                         t->q.size = 1024;
629                 }
630
631                 init_rspq(adap, &adap->sge.fw_evtq, 0, 0, 1024, 64);
632                 adap->flags |= CFG_QUEUES;
633         }
634
635         return 0;
636 }
637
638 void cxgbe_stats_get(struct port_info *pi, struct port_stats *stats)
639 {
640         t4_get_port_stats_offset(pi->adapter, pi->tx_chan, stats,
641                                  &pi->stats_base);
642 }
643
644 void cxgbe_stats_reset(struct port_info *pi)
645 {
646         t4_clr_port_stats(pi->adapter, pi->tx_chan);
647 }
648
649 static void setup_memwin(struct adapter *adap)
650 {
651         u32 mem_win0_base;
652
653         /* For T5, only relative offset inside the PCIe BAR is passed */
654         mem_win0_base = MEMWIN0_BASE;
655
656         /*
657          * Set up memory window for accessing adapter memory ranges.  (Read
658          * back MA register to ensure that changes propagate before we attempt
659          * to use the new values.)
660          */
661         t4_write_reg(adap,
662                      PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
663                                          MEMWIN_NIC),
664                      mem_win0_base | V_BIR(0) |
665                      V_WINDOW(ilog2(MEMWIN0_APERTURE) - X_WINDOW_SHIFT));
666         t4_read_reg(adap,
667                     PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
668                                         MEMWIN_NIC));
669 }
670
671 int cxgbe_init_rss(struct adapter *adap)
672 {
673         unsigned int i;
674
675         if (is_pf4(adap)) {
676                 int err;
677
678                 err = t4_init_rss_mode(adap, adap->mbox);
679                 if (err)
680                         return err;
681         }
682
683         for_each_port(adap, i) {
684                 struct port_info *pi = adap2pinfo(adap, i);
685
686                 pi->rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
687                 if (!pi->rss)
688                         return -ENOMEM;
689
690                 pi->rss_hf = CXGBE_RSS_HF_ALL;
691         }
692         return 0;
693 }
694
695 /**
696  * Dump basic information about the adapter.
697  */
698 void cxgbe_print_adapter_info(struct adapter *adap)
699 {
700         /**
701          * Hardware/Firmware/etc. Version/Revision IDs.
702          */
703         t4_dump_version_info(adap);
704 }
705
706 void cxgbe_print_port_info(struct adapter *adap)
707 {
708         int i;
709         char buf[80];
710         struct rte_pci_addr *loc = &adap->pdev->addr;
711
712         for_each_port(adap, i) {
713                 const struct port_info *pi = adap2pinfo(adap, i);
714                 char *bufp = buf;
715
716                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
717                         bufp += sprintf(bufp, "100M/");
718                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
719                         bufp += sprintf(bufp, "1G/");
720                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
721                         bufp += sprintf(bufp, "10G/");
722                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
723                         bufp += sprintf(bufp, "25G/");
724                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
725                         bufp += sprintf(bufp, "40G/");
726                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
727                         bufp += sprintf(bufp, "50G/");
728                 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
729                         bufp += sprintf(bufp, "100G/");
730                 if (bufp != buf)
731                         --bufp;
732                 sprintf(bufp, "BASE-%s",
733                         t4_get_port_type_description(
734                                 (enum fw_port_type)pi->link_cfg.port_type));
735
736                 dev_info(adap,
737                          " " PCI_PRI_FMT " Chelsio rev %d %s %s\n",
738                          loc->domain, loc->bus, loc->devid, loc->function,
739                          CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
740                          (adap->flags & USING_MSIX) ? " MSI-X" :
741                          (adap->flags & USING_MSI) ? " MSI" : "");
742         }
743 }
744
745 static int check_devargs_handler(const char *key, const char *value, void *p)
746 {
747         if (!strncmp(key, CXGBE_DEVARG_CMN_KEEP_OVLAN, strlen(key)) ||
748             !strncmp(key, CXGBE_DEVARG_CMN_TX_MODE_LATENCY, strlen(key)) ||
749             !strncmp(key, CXGBE_DEVARG_VF_FORCE_LINK_UP, strlen(key))) {
750                 if (!strncmp(value, "1", 1)) {
751                         bool *dst_val = (bool *)p;
752
753                         *dst_val = true;
754                 }
755         }
756
757         if (!strncmp(key, CXGBE_DEVARG_PF_FILTER_MODE, strlen(key)) ||
758             !strncmp(key, CXGBE_DEVARG_PF_FILTER_MASK, strlen(key))) {
759                 u32 *dst_val = (u32 *)p;
760                 char *endptr = NULL;
761                 u32 arg_val;
762
763                 arg_val = strtoul(value, &endptr, 16);
764                 if (errno || endptr == value)
765                         return -EINVAL;
766
767                 *dst_val = arg_val;
768         }
769
770         return 0;
771 }
772
773 static int cxgbe_get_devargs(struct rte_devargs *devargs, const char *key,
774                              void *p)
775 {
776         struct rte_kvargs *kvlist;
777         int ret = 0;
778
779         if (!devargs)
780                 return 0;
781
782         kvlist = rte_kvargs_parse(devargs->args, NULL);
783         if (!kvlist)
784                 return 0;
785
786         if (!rte_kvargs_count(kvlist, key))
787                 goto out;
788
789         ret = rte_kvargs_process(kvlist, key, check_devargs_handler, p);
790
791 out:
792         rte_kvargs_free(kvlist);
793
794         return ret;
795 }
796
797 static void cxgbe_get_devargs_int(struct adapter *adap, bool *dst,
798                                   const char *key, bool default_value)
799 {
800         struct rte_pci_device *pdev = adap->pdev;
801         int ret;
802         bool devarg_value = default_value;
803
804         *dst = default_value;
805         if (!pdev)
806                 return;
807
808         ret = cxgbe_get_devargs(pdev->device.devargs, key, &devarg_value);
809         if (ret)
810                 return;
811
812         *dst = devarg_value;
813 }
814
815 static void cxgbe_get_devargs_u32(struct adapter *adap, u32 *dst,
816                                   const char *key, u32 default_value)
817 {
818         struct rte_pci_device *pdev = adap->pdev;
819         u32 devarg_value = default_value;
820         int ret;
821
822         *dst = default_value;
823         if (!pdev)
824                 return;
825
826         ret = cxgbe_get_devargs(pdev->device.devargs, key, &devarg_value);
827         if (ret)
828                 return;
829
830         *dst = devarg_value;
831 }
832
833 void cxgbe_process_devargs(struct adapter *adap)
834 {
835         cxgbe_get_devargs_int(adap, &adap->devargs.keep_ovlan,
836                               CXGBE_DEVARG_CMN_KEEP_OVLAN, false);
837         cxgbe_get_devargs_int(adap, &adap->devargs.tx_mode_latency,
838                               CXGBE_DEVARG_CMN_TX_MODE_LATENCY, false);
839         cxgbe_get_devargs_int(adap, &adap->devargs.force_link_up,
840                               CXGBE_DEVARG_VF_FORCE_LINK_UP, false);
841         cxgbe_get_devargs_u32(adap, &adap->devargs.filtermode,
842                               CXGBE_DEVARG_PF_FILTER_MODE, 0);
843         cxgbe_get_devargs_u32(adap, &adap->devargs.filtermask,
844                               CXGBE_DEVARG_PF_FILTER_MASK, 0);
845 }
846
847 static void configure_vlan_types(struct adapter *adapter)
848 {
849         int i;
850
851         for_each_port(adapter, i) {
852                 /* OVLAN Type 0x88a8 */
853                 t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN0),
854                                  V_OVLAN_MASK(M_OVLAN_MASK) |
855                                  V_OVLAN_ETYPE(M_OVLAN_ETYPE),
856                                  V_OVLAN_MASK(M_OVLAN_MASK) |
857                                  V_OVLAN_ETYPE(0x88a8));
858                 /* OVLAN Type 0x9100 */
859                 t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN1),
860                                  V_OVLAN_MASK(M_OVLAN_MASK) |
861                                  V_OVLAN_ETYPE(M_OVLAN_ETYPE),
862                                  V_OVLAN_MASK(M_OVLAN_MASK) |
863                                  V_OVLAN_ETYPE(0x9100));
864
865                 /* IVLAN 0X8100 */
866                 t4_set_reg_field(adapter, MPS_PORT_RX_IVLAN(i),
867                                  V_IVLAN_ETYPE(M_IVLAN_ETYPE),
868                                  V_IVLAN_ETYPE(0x8100));
869
870                 t4_set_reg_field(adapter, MPS_PORT_RX_CTL(i),
871                                  F_OVLAN_EN0 | F_OVLAN_EN1 |
872                                  F_IVLAN_EN,
873                                  F_OVLAN_EN0 | F_OVLAN_EN1 |
874                                  F_IVLAN_EN);
875         }
876
877         t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG, V_RM_OVLAN(1),
878                                V_RM_OVLAN(!adapter->devargs.keep_ovlan));
879 }
880
881 static int cxgbe_get_filter_vnic_mode_from_devargs(u32 val)
882 {
883         u32 vnic_mode;
884
885         vnic_mode = val & (CXGBE_DEVARGS_FILTER_MODE_PF_VF |
886                            CXGBE_DEVARGS_FILTER_MODE_VLAN_OUTER);
887         if (vnic_mode) {
888                 switch (vnic_mode) {
889                 case CXGBE_DEVARGS_FILTER_MODE_VLAN_OUTER:
890                         return CXGBE_FILTER_VNIC_MODE_OVLAN;
891                 case CXGBE_DEVARGS_FILTER_MODE_PF_VF:
892                         return CXGBE_FILTER_VNIC_MODE_PFVF;
893                 default:
894                         return -EINVAL;
895                 }
896         }
897
898         return CXGBE_FILTER_VNIC_MODE_NONE;
899 }
900
901 static int cxgbe_get_filter_mode_from_devargs(u32 val, bool closest_match)
902 {
903         int vnic_mode, fmode = 0;
904         bool found = false;
905         u8 i;
906
907         if (val >= CXGBE_DEVARGS_FILTER_MODE_MAX) {
908                 pr_err("Unsupported flags set in filter mode. Must be < 0x%x\n",
909                        CXGBE_DEVARGS_FILTER_MODE_MAX);
910                 return -ERANGE;
911         }
912
913         vnic_mode = cxgbe_get_filter_vnic_mode_from_devargs(val);
914         if (vnic_mode < 0) {
915                 pr_err("Unsupported Vnic-mode, more than 1 Vnic-mode selected\n");
916                 return vnic_mode;
917         }
918
919         if (vnic_mode)
920                 fmode |= F_VNIC_ID;
921         if (val & CXGBE_DEVARGS_FILTER_MODE_PHYSICAL_PORT)
922                 fmode |= F_PORT;
923         if (val & CXGBE_DEVARGS_FILTER_MODE_ETHERNET_DSTMAC)
924                 fmode |= F_MACMATCH;
925         if (val & CXGBE_DEVARGS_FILTER_MODE_ETHERNET_ETHTYPE)
926                 fmode |= F_ETHERTYPE;
927         if (val & CXGBE_DEVARGS_FILTER_MODE_VLAN_INNER)
928                 fmode |= F_VLAN;
929         if (val & CXGBE_DEVARGS_FILTER_MODE_IP_TOS)
930                 fmode |= F_TOS;
931         if (val & CXGBE_DEVARGS_FILTER_MODE_IP_PROTOCOL)
932                 fmode |= F_PROTOCOL;
933
934         for (i = 0; i < ARRAY_SIZE(cxgbe_filter_mode_features); i++) {
935                 if ((cxgbe_filter_mode_features[i] & fmode) == fmode) {
936                         found = true;
937                         break;
938                 }
939         }
940
941         if (!found)
942                 return -EINVAL;
943
944         return closest_match ? cxgbe_filter_mode_features[i] : fmode;
945 }
946
947 static int configure_filter_mode_mask(struct adapter *adap)
948 {
949         u32 params[2], val[2], nparams = 0;
950         int ret;
951
952         if (!adap->devargs.filtermode && !adap->devargs.filtermask)
953                 return 0;
954
955         if (!adap->devargs.filtermode || !adap->devargs.filtermask) {
956                 pr_err("Unsupported, Provide both filtermode and filtermask devargs\n");
957                 return -EINVAL;
958         }
959
960         if (adap->devargs.filtermask & ~adap->devargs.filtermode) {
961                 pr_err("Unsupported, filtermask (0x%x) must be subset of filtermode (0x%x)\n",
962                        adap->devargs.filtermask, adap->devargs.filtermode);
963
964                 return -EINVAL;
965         }
966
967         params[0] = CXGBE_FW_PARAM_DEV(FILTER) |
968                     V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_FILTER_MODE_MASK);
969
970         ret = cxgbe_get_filter_mode_from_devargs(adap->devargs.filtermode,
971                                                  true);
972         if (ret < 0) {
973                 pr_err("Unsupported filtermode devargs combination:0x%x\n",
974                        adap->devargs.filtermode);
975                 return ret;
976         }
977
978         val[0] = V_FW_PARAMS_PARAM_FILTER_MODE(ret);
979
980         ret = cxgbe_get_filter_mode_from_devargs(adap->devargs.filtermask,
981                                                  false);
982         if (ret < 0) {
983                 pr_err("Unsupported filtermask devargs combination:0x%x\n",
984                        adap->devargs.filtermask);
985                 return ret;
986         }
987
988         val[0] |= V_FW_PARAMS_PARAM_FILTER_MASK(ret);
989
990         nparams++;
991
992         ret = cxgbe_get_filter_vnic_mode_from_devargs(adap->devargs.filtermode);
993         if (ret < 0)
994                 return ret;
995
996         if (ret) {
997                 params[1] = CXGBE_FW_PARAM_DEV(FILTER) |
998                             V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_FILTER_VNIC_MODE);
999
1000                 val[1] = ret - 1;
1001
1002                 nparams++;
1003         }
1004
1005         return t4_set_params(adap, adap->mbox, adap->pf, 0, nparams,
1006                              params, val);
1007 }
1008
1009 static void configure_pcie_ext_tag(struct adapter *adapter)
1010 {
1011         u16 v;
1012         int pos = t4_os_find_pci_capability(adapter, PCI_CAP_ID_EXP);
1013
1014         if (!pos)
1015                 return;
1016
1017         if (pos > 0) {
1018                 t4_os_pci_read_cfg2(adapter, pos + PCI_EXP_DEVCTL, &v);
1019                 v |= PCI_EXP_DEVCTL_EXT_TAG;
1020                 t4_os_pci_write_cfg2(adapter, pos + PCI_EXP_DEVCTL, v);
1021                 if (is_t6(adapter->params.chip)) {
1022                         t4_set_reg_field(adapter, A_PCIE_CFG2,
1023                                          V_T6_TOTMAXTAG(M_T6_TOTMAXTAG),
1024                                          V_T6_TOTMAXTAG(7));
1025                         t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
1026                                          V_T6_MINTAG(M_T6_MINTAG),
1027                                          V_T6_MINTAG(8));
1028                 } else {
1029                         t4_set_reg_field(adapter, A_PCIE_CFG2,
1030                                          V_TOTMAXTAG(M_TOTMAXTAG),
1031                                          V_TOTMAXTAG(3));
1032                         t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
1033                                          V_MINTAG(M_MINTAG),
1034                                          V_MINTAG(8));
1035                 }
1036         }
1037 }
1038
1039 /* Figure out how many Queue Sets we can support */
1040 void cxgbe_configure_max_ethqsets(struct adapter *adapter)
1041 {
1042         unsigned int ethqsets, reserved;
1043
1044         /* We need to reserve an Ingress Queue for the Asynchronous Firmware
1045          * Event Queue and 1 Control Queue per port.
1046          *
1047          * For each Queue Set, we'll need the ability to allocate two Egress
1048          * Contexts -- one for the Ingress Queue Free List and one for the TX
1049          * Ethernet Queue.
1050          */
1051         reserved = max(adapter->params.nports, 1);
1052         if (is_pf4(adapter)) {
1053                 struct pf_resources *pfres = &adapter->params.pfres;
1054
1055                 ethqsets = min(pfres->niqflint, pfres->nethctrl);
1056                 if (ethqsets > (pfres->neq / 2))
1057                         ethqsets = pfres->neq / 2;
1058         } else {
1059                 struct vf_resources *vfres = &adapter->params.vfres;
1060
1061                 ethqsets = min(vfres->niqflint, vfres->nethctrl);
1062                 if (ethqsets > (vfres->neq / 2))
1063                         ethqsets = vfres->neq / 2;
1064         }
1065
1066         ethqsets -= reserved;
1067         adapter->sge.max_ethqsets = ethqsets;
1068 }
1069
1070 /*
1071  * Tweak configuration based on system architecture, etc.  Most of these have
1072  * defaults assigned to them by Firmware Configuration Files (if we're using
1073  * them) but need to be explicitly set if we're using hard-coded
1074  * initialization. So these are essentially common tweaks/settings for
1075  * Configuration Files and hard-coded initialization ...
1076  */
1077 static int adap_init0_tweaks(struct adapter *adapter)
1078 {
1079         u8 rx_dma_offset;
1080
1081         /*
1082          * Fix up various Host-Dependent Parameters like Page Size, Cache
1083          * Line Size, etc.  The firmware default is for a 4KB Page Size and
1084          * 64B Cache Line Size ...
1085          */
1086         t4_fixup_host_params_compat(adapter, CXGBE_PAGE_SIZE, L1_CACHE_BYTES,
1087                                     T5_LAST_REV);
1088
1089         /*
1090          * Keep the chip default offset to deliver Ingress packets into our
1091          * DMA buffers to zero
1092          */
1093         rx_dma_offset = 0;
1094         t4_set_reg_field(adapter, A_SGE_CONTROL, V_PKTSHIFT(M_PKTSHIFT),
1095                          V_PKTSHIFT(rx_dma_offset));
1096
1097         t4_set_reg_field(adapter, A_SGE_FLM_CFG,
1098                          V_CREDITCNT(M_CREDITCNT) | M_CREDITCNTPACKING,
1099                          V_CREDITCNT(3) | V_CREDITCNTPACKING(1));
1100
1101         t4_set_reg_field(adapter, A_SGE_INGRESS_RX_THRESHOLD,
1102                          V_THRESHOLD_3(M_THRESHOLD_3), V_THRESHOLD_3(32U));
1103
1104         t4_set_reg_field(adapter, A_SGE_CONTROL2, V_IDMAARBROUNDROBIN(1U),
1105                          V_IDMAARBROUNDROBIN(1U));
1106
1107         /*
1108          * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
1109          * adds the pseudo header itself.
1110          */
1111         t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG,
1112                                F_CSUM_HAS_PSEUDO_HDR, 0);
1113
1114         return 0;
1115 }
1116
1117 /*
1118  * Attempt to initialize the adapter via a Firmware Configuration File.
1119  */
1120 static int adap_init0_config(struct adapter *adapter, int reset)
1121 {
1122         u32 finiver, finicsum, cfcsum, param, val;
1123         struct fw_caps_config_cmd caps_cmd;
1124         unsigned long mtype = 0, maddr = 0;
1125         u8 config_issued = 0;
1126         char config_name[20];
1127         int cfg_addr, ret;
1128
1129         /*
1130          * Reset device if necessary.
1131          */
1132         if (reset) {
1133                 ret = t4_fw_reset(adapter, adapter->mbox,
1134                                   F_PIORSTMODE | F_PIORST);
1135                 if (ret < 0) {
1136                         dev_warn(adapter, "Firmware reset failed, error %d\n",
1137                                  -ret);
1138                         goto bye;
1139                 }
1140         }
1141
1142         cfg_addr = t4_flash_cfg_addr(adapter);
1143         if (cfg_addr < 0) {
1144                 ret = cfg_addr;
1145                 dev_warn(adapter, "Finding address for firmware config file in flash failed, error %d\n",
1146                          -ret);
1147                 goto bye;
1148         }
1149
1150         strcpy(config_name, "On Flash");
1151         mtype = FW_MEMTYPE_CF_FLASH;
1152         maddr = cfg_addr;
1153
1154         /* Enable HASH filter region when support is available. */
1155         val = 1;
1156         param = CXGBE_FW_PARAM_DEV(HASHFILTER_WITH_OFLD);
1157         t4_set_params(adapter, adapter->mbox, adapter->pf, 0, 1,
1158                       &param, &val);
1159
1160         /*
1161          * Issue a Capability Configuration command to the firmware to get it
1162          * to parse the Configuration File.  We don't use t4_fw_config_file()
1163          * because we want the ability to modify various features after we've
1164          * processed the configuration file ...
1165          */
1166         memset(&caps_cmd, 0, sizeof(caps_cmd));
1167         caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1168                                            F_FW_CMD_REQUEST | F_FW_CMD_READ);
1169         caps_cmd.cfvalid_to_len16 =
1170                 cpu_to_be32(F_FW_CAPS_CONFIG_CMD_CFVALID |
1171                             V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
1172                             V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
1173                             FW_LEN16(caps_cmd));
1174         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
1175                          &caps_cmd);
1176         /*
1177          * If the CAPS_CONFIG failed with an ENOENT (for a Firmware
1178          * Configuration File in FLASH), our last gasp effort is to use the
1179          * Firmware Configuration File which is embedded in the firmware.  A
1180          * very few early versions of the firmware didn't have one embedded
1181          * but we can ignore those.
1182          */
1183         if (ret == -ENOENT) {
1184                 dev_info(adapter, "%s: Going for embedded config in firmware..\n",
1185                          __func__);
1186
1187                 memset(&caps_cmd, 0, sizeof(caps_cmd));
1188                 caps_cmd.op_to_write =
1189                         cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1190                                     F_FW_CMD_REQUEST | F_FW_CMD_READ);
1191                 caps_cmd.cfvalid_to_len16 = cpu_to_be32(FW_LEN16(caps_cmd));
1192                 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
1193                                  sizeof(caps_cmd), &caps_cmd);
1194                 strcpy(config_name, "Firmware Default");
1195         }
1196
1197         config_issued = 1;
1198         if (ret < 0)
1199                 goto bye;
1200
1201         finiver = be32_to_cpu(caps_cmd.finiver);
1202         finicsum = be32_to_cpu(caps_cmd.finicsum);
1203         cfcsum = be32_to_cpu(caps_cmd.cfcsum);
1204         if (finicsum != cfcsum)
1205                 dev_warn(adapter, "Configuration File checksum mismatch: [fini] csum=%#x, computed csum=%#x\n",
1206                          finicsum, cfcsum);
1207
1208         /*
1209          * If we're a pure NIC driver then disable all offloading facilities.
1210          * This will allow the firmware to optimize aspects of the hardware
1211          * configuration which will result in improved performance.
1212          */
1213         caps_cmd.niccaps &= cpu_to_be16(~FW_CAPS_CONFIG_NIC_ETHOFLD);
1214         caps_cmd.toecaps = 0;
1215         caps_cmd.iscsicaps = 0;
1216         caps_cmd.rdmacaps = 0;
1217         caps_cmd.fcoecaps = 0;
1218         caps_cmd.cryptocaps = 0;
1219
1220         /*
1221          * And now tell the firmware to use the configuration we just loaded.
1222          */
1223         caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1224                                            F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
1225         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
1226         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
1227                          NULL);
1228         if (ret < 0) {
1229                 dev_warn(adapter, "Unable to finalize Firmware Capabilities %d\n",
1230                          -ret);
1231                 goto bye;
1232         }
1233
1234         /*
1235          * Tweak configuration based on system architecture, etc.
1236          */
1237         ret = adap_init0_tweaks(adapter);
1238         if (ret < 0) {
1239                 dev_warn(adapter, "Unable to do init0-tweaks %d\n", -ret);
1240                 goto bye;
1241         }
1242
1243         /*
1244          * And finally tell the firmware to initialize itself using the
1245          * parameters from the Configuration File.
1246          */
1247         ret = t4_fw_initialize(adapter, adapter->mbox);
1248         if (ret < 0) {
1249                 dev_warn(adapter, "Initializing Firmware failed, error %d\n",
1250                          -ret);
1251                 goto bye;
1252         }
1253
1254         /*
1255          * Return successfully and note that we're operating with parameters
1256          * not supplied by the driver, rather than from hard-wired
1257          * initialization constants buried in the driver.
1258          */
1259         dev_info(adapter,
1260                  "Successfully configured using Firmware Configuration File \"%s\", version %#x, computed checksum %#x\n",
1261                  config_name, finiver, cfcsum);
1262
1263         return 0;
1264
1265         /*
1266          * Something bad happened.  Return the error ...  (If the "error"
1267          * is that there's no Configuration File on the adapter we don't
1268          * want to issue a warning since this is fairly common.)
1269          */
1270 bye:
1271         if (config_issued && ret != -ENOENT)
1272                 dev_warn(adapter, "\"%s\" configuration file error %d\n",
1273                          config_name, -ret);
1274
1275         dev_debug(adapter, "%s: returning ret = %d ..\n", __func__, ret);
1276         return ret;
1277 }
1278
1279 static int adap_init0(struct adapter *adap)
1280 {
1281         struct fw_caps_config_cmd caps_cmd;
1282         int ret = 0;
1283         u32 v, port_vec;
1284         enum dev_state state;
1285         u32 params[7], val[7];
1286         int reset = 1;
1287         int mbox = adap->mbox;
1288
1289         /*
1290          * Contact FW, advertising Master capability.
1291          */
1292         ret = t4_fw_hello(adap, adap->mbox, adap->mbox, MASTER_MAY, &state);
1293         if (ret < 0) {
1294                 dev_err(adap, "%s: could not connect to FW, error %d\n",
1295                         __func__, -ret);
1296                 goto bye;
1297         }
1298
1299         CXGBE_DEBUG_MBOX(adap, "%s: adap->mbox = %d; ret = %d\n", __func__,
1300                          adap->mbox, ret);
1301
1302         if (ret == mbox)
1303                 adap->flags |= MASTER_PF;
1304
1305         if (state == DEV_STATE_INIT) {
1306                 /*
1307                  * Force halt and reset FW because a previous instance may have
1308                  * exited abnormally without properly shutting down
1309                  */
1310                 ret = t4_fw_halt(adap, adap->mbox, reset);
1311                 if (ret < 0) {
1312                         dev_err(adap, "Failed to halt. Exit.\n");
1313                         goto bye;
1314                 }
1315
1316                 ret = t4_fw_restart(adap, adap->mbox, reset);
1317                 if (ret < 0) {
1318                         dev_err(adap, "Failed to restart. Exit.\n");
1319                         goto bye;
1320                 }
1321                 state = (enum dev_state)((unsigned)state & ~DEV_STATE_INIT);
1322         }
1323
1324         t4_get_version_info(adap);
1325
1326         ret = t4_get_core_clock(adap, &adap->params.vpd);
1327         if (ret < 0) {
1328                 dev_err(adap, "%s: could not get core clock, error %d\n",
1329                         __func__, -ret);
1330                 goto bye;
1331         }
1332
1333         /*
1334          * If the firmware is initialized already (and we're not forcing a
1335          * master initialization), note that we're living with existing
1336          * adapter parameters.  Otherwise, it's time to try initializing the
1337          * adapter ...
1338          */
1339         if (state == DEV_STATE_INIT) {
1340                 dev_info(adap, "Coming up as %s: Adapter already initialized\n",
1341                          adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
1342         } else {
1343                 dev_info(adap, "Coming up as MASTER: Initializing adapter\n");
1344
1345                 ret = adap_init0_config(adap, reset);
1346                 if (ret == -ENOENT) {
1347                         dev_err(adap,
1348                                 "No Configuration File present on adapter. Using hard-wired configuration parameters.\n");
1349                         goto bye;
1350                 }
1351         }
1352         if (ret < 0) {
1353                 dev_err(adap, "could not initialize adapter, error %d\n", -ret);
1354                 goto bye;
1355         }
1356
1357         /* Now that we've successfully configured and initialized the adapter
1358          * (or found it already initialized), we can ask the Firmware what
1359          * resources it has provisioned for us.
1360          */
1361         ret = t4_get_pfres(adap);
1362         if (ret) {
1363                 dev_err(adap->pdev_dev,
1364                         "Unable to retrieve resource provisioning info\n");
1365                 goto bye;
1366         }
1367
1368         /* Find out what ports are available to us. */
1369         v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
1370             V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
1371         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
1372         if (ret < 0) {
1373                 dev_err(adap, "%s: failure in t4_query_params; error = %d\n",
1374                         __func__, ret);
1375                 goto bye;
1376         }
1377
1378         adap->params.nports = hweight32(port_vec);
1379         adap->params.portvec = port_vec;
1380
1381         dev_debug(adap, "%s: adap->params.nports = %u\n", __func__,
1382                   adap->params.nports);
1383
1384         /*
1385          * Give the SGE code a chance to pull in anything that it needs ...
1386          * Note that this must be called after we retrieve our VPD parameters
1387          * in order to know how to convert core ticks to seconds, etc.
1388          */
1389         ret = t4_sge_init(adap);
1390         if (ret < 0) {
1391                 dev_err(adap, "t4_sge_init failed with error %d\n",
1392                         -ret);
1393                 goto bye;
1394         }
1395
1396         /*
1397          * Grab some of our basic fundamental operating parameters.
1398          */
1399         params[0] = CXGBE_FW_PARAM_PFVF(L2T_START);
1400         params[1] = CXGBE_FW_PARAM_PFVF(L2T_END);
1401         params[2] = CXGBE_FW_PARAM_PFVF(FILTER_START);
1402         params[3] = CXGBE_FW_PARAM_PFVF(FILTER_END);
1403         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 4, params, val);
1404         if (ret < 0)
1405                 goto bye;
1406         adap->l2t_start = val[0];
1407         adap->l2t_end = val[1];
1408         adap->tids.ftid_base = val[2];
1409         adap->tids.nftids = val[3] - val[2] + 1;
1410
1411         params[0] = CXGBE_FW_PARAM_PFVF(CLIP_START);
1412         params[1] = CXGBE_FW_PARAM_PFVF(CLIP_END);
1413         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
1414         if (ret < 0)
1415                 goto bye;
1416         adap->clipt_start = val[0];
1417         adap->clipt_end = val[1];
1418
1419         /*
1420          * Get device capabilities so we can determine what resources we need
1421          * to manage.
1422          */
1423         memset(&caps_cmd, 0, sizeof(caps_cmd));
1424         caps_cmd.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1425                                      F_FW_CMD_REQUEST | F_FW_CMD_READ);
1426         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
1427         ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
1428                          &caps_cmd);
1429         if (ret < 0)
1430                 goto bye;
1431
1432         if ((caps_cmd.niccaps & cpu_to_be16(FW_CAPS_CONFIG_NIC_HASHFILTER)) &&
1433             is_t6(adap->params.chip)) {
1434                 if (cxgbe_init_hash_filter(adap) < 0)
1435                         goto bye;
1436         }
1437
1438         /* See if FW supports FW_FILTER2 work request */
1439         if (is_t4(adap->params.chip)) {
1440                 adap->params.filter2_wr_support = 0;
1441         } else {
1442                 params[0] = CXGBE_FW_PARAM_DEV(FILTER2_WR);
1443                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1444                                       1, params, val);
1445                 adap->params.filter2_wr_support = (ret == 0 && val[0] != 0);
1446         }
1447
1448         /* Check if FW supports returning vin.
1449          * If this is not supported, driver will interpret
1450          * these values from viid.
1451          */
1452         params[0] = CXGBE_FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
1453         ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1454                               1, params, val);
1455         adap->params.viid_smt_extn_support = (ret == 0 && val[0] != 0);
1456
1457         /* query tid-related parameters */
1458         params[0] = CXGBE_FW_PARAM_DEV(NTID);
1459         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
1460                               params, val);
1461         if (ret < 0)
1462                 goto bye;
1463         adap->tids.ntids = val[0];
1464         adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
1465
1466         /* If we're running on newer firmware, let it know that we're
1467          * prepared to deal with encapsulated CPL messages.  Older
1468          * firmware won't understand this and we'll just get
1469          * unencapsulated messages ...
1470          */
1471         params[0] = CXGBE_FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
1472         val[0] = 1;
1473         (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1474
1475         /*
1476          * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
1477          * capability.  Earlier versions of the firmware didn't have the
1478          * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
1479          * permission to use ULPTX MEMWRITE DSGL.
1480          */
1481         if (is_t4(adap->params.chip)) {
1482                 adap->params.ulptx_memwrite_dsgl = false;
1483         } else {
1484                 params[0] = CXGBE_FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
1485                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1486                                       1, params, val);
1487                 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
1488         }
1489
1490         /* Query for max number of packets that can be coalesced for Tx */
1491         params[0] = CXGBE_FW_PARAM_PFVF(MAX_PKTS_PER_ETH_TX_PKTS_WR);
1492         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1493         if (!ret && val[0] > 0)
1494                 adap->params.max_tx_coalesce_num = val[0];
1495         else
1496                 adap->params.max_tx_coalesce_num = ETH_COALESCE_PKT_NUM;
1497
1498         params[0] = CXGBE_FW_PARAM_DEV(VI_ENABLE_INGRESS_AFTER_LINKUP);
1499         ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1500         adap->params.vi_enable_rx = (ret == 0 && val[0] != 0);
1501
1502         /* Read the RAW MPS entries. In T6, the last 2 TCAM entries
1503          * are reserved for RAW MAC addresses (rawf = 2, one per port).
1504          */
1505         if (CHELSIO_CHIP_VERSION(adap->params.chip) > CHELSIO_T5) {
1506                 params[0] = CXGBE_FW_PARAM_PFVF(RAWF_START);
1507                 params[1] = CXGBE_FW_PARAM_PFVF(RAWF_END);
1508                 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
1509                                       params, val);
1510                 if (ret == 0) {
1511                         adap->params.rawf_start = val[0];
1512                         adap->params.rawf_size = val[1] - val[0] + 1;
1513                 }
1514         }
1515
1516         /*
1517          * The MTU/MSS Table is initialized by now, so load their values.  If
1518          * we're initializing the adapter, then we'll make any modifications
1519          * we want to the MTU/MSS Table and also initialize the congestion
1520          * parameters.
1521          */
1522         t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
1523         if (state != DEV_STATE_INIT) {
1524                 int i;
1525
1526                 /*
1527                  * The default MTU Table contains values 1492 and 1500.
1528                  * However, for TCP, it's better to have two values which are
1529                  * a multiple of 8 +/- 4 bytes apart near this popular MTU.
1530                  * This allows us to have a TCP Data Payload which is a
1531                  * multiple of 8 regardless of what combination of TCP Options
1532                  * are in use (always a multiple of 4 bytes) which is
1533                  * important for performance reasons.  For instance, if no
1534                  * options are in use, then we have a 20-byte IP header and a
1535                  * 20-byte TCP header.  In this case, a 1500-byte MSS would
1536                  * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
1537                  * which is not a multiple of 8.  So using an MSS of 1488 in
1538                  * this case results in a TCP Data Payload of 1448 bytes which
1539                  * is a multiple of 8.  On the other hand, if 12-byte TCP Time
1540                  * Stamps have been negotiated, then an MTU of 1500 bytes
1541                  * results in a TCP Data Payload of 1448 bytes which, as
1542                  * above, is a multiple of 8 bytes ...
1543                  */
1544                 for (i = 0; i < NMTUS; i++)
1545                         if (adap->params.mtus[i] == 1492) {
1546                                 adap->params.mtus[i] = 1488;
1547                                 break;
1548                         }
1549
1550                 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
1551                              adap->params.b_wnd);
1552         }
1553         t4_init_sge_params(adap);
1554         ret = configure_filter_mode_mask(adap);
1555         if (ret < 0)
1556                 goto bye;
1557         t4_init_tp_params(adap);
1558         configure_pcie_ext_tag(adap);
1559         configure_vlan_types(adap);
1560         cxgbe_configure_max_ethqsets(adap);
1561
1562         adap->params.drv_memwin = MEMWIN_NIC;
1563         adap->flags |= FW_OK;
1564         dev_debug(adap, "%s: returning zero..\n", __func__);
1565         return 0;
1566
1567         /*
1568          * Something bad happened.  If a command timed out or failed with EIO
1569          * FW does not operate within its spec or something catastrophic
1570          * happened to HW/FW, stop issuing commands.
1571          */
1572 bye:
1573         if (ret != -ETIMEDOUT && ret != -EIO)
1574                 t4_fw_bye(adap, adap->mbox);
1575         return ret;
1576 }
1577
1578 /**
1579  * t4_os_portmod_changed - handle port module changes
1580  * @adap: the adapter associated with the module change
1581  * @port_id: the port index whose module status has changed
1582  *
1583  * This is the OS-dependent handler for port module changes.  It is
1584  * invoked when a port module is removed or inserted for any OS-specific
1585  * processing.
1586  */
1587 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
1588 {
1589         static const char * const mod_str[] = {
1590                 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
1591         };
1592
1593         const struct port_info *pi = adap2pinfo(adap, port_id);
1594
1595         if (pi->link_cfg.mod_type == FW_PORT_MOD_TYPE_NONE)
1596                 dev_info(adap, "Port%d: port module unplugged\n", pi->port_id);
1597         else if (pi->link_cfg.mod_type < ARRAY_SIZE(mod_str))
1598                 dev_info(adap, "Port%d: %s port module inserted\n", pi->port_id,
1599                          mod_str[pi->link_cfg.mod_type]);
1600         else if (pi->link_cfg.mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
1601                 dev_info(adap, "Port%d: unsupported port module inserted\n",
1602                          pi->port_id);
1603         else if (pi->link_cfg.mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
1604                 dev_info(adap, "Port%d: unknown port module inserted\n",
1605                          pi->port_id);
1606         else if (pi->link_cfg.mod_type == FW_PORT_MOD_TYPE_ERROR)
1607                 dev_info(adap, "Port%d: transceiver module error\n",
1608                          pi->port_id);
1609         else
1610                 dev_info(adap, "Port%d: unknown module type %d inserted\n",
1611                          pi->port_id, pi->link_cfg.mod_type);
1612 }
1613
1614 void t4_os_link_changed(struct adapter *adap, int port_id)
1615 {
1616         struct port_info *pi = adap2pinfo(adap, port_id);
1617
1618         /* If link status has not changed or if firmware doesn't
1619          * support enabling/disabling VI's Rx path during runtime,
1620          * then return.
1621          */
1622         if (adap->params.vi_enable_rx == 0 ||
1623             pi->vi_en_rx == pi->link_cfg.link_ok)
1624                 return;
1625
1626         /* Don't enable VI Rx path, if link has been administratively
1627          * turned off.
1628          */
1629         if (pi->vi_en_tx == 0 && pi->vi_en_rx == 0)
1630                 return;
1631
1632         /* When link goes down, disable the port's Rx path to drop
1633          * Rx traffic closer to the wire, instead of processing it
1634          * further in the Rx pipeline. The Rx path will be re-enabled
1635          * once the link up message comes in firmware event queue.
1636          */
1637         pi->vi_en_rx = pi->link_cfg.link_ok;
1638         t4_enable_vi(adap, adap->mbox, pi->viid, pi->vi_en_rx, pi->vi_en_tx);
1639 }
1640
1641 bool cxgbe_force_linkup(struct adapter *adap)
1642 {
1643         if (is_pf4(adap))
1644                 return false;   /* force_linkup not required for pf driver */
1645
1646         return adap->devargs.force_link_up;
1647 }
1648
1649 /**
1650  * link_start - enable a port
1651  * @dev: the port to enable
1652  *
1653  * Performs the MAC and PHY actions needed to enable a port.
1654  */
1655 int cxgbe_link_start(struct port_info *pi)
1656 {
1657         struct adapter *adapter = pi->adapter;
1658         u64 conf_offloads;
1659         unsigned int mtu;
1660         int ret;
1661
1662         mtu = pi->eth_dev->data->mtu;
1663
1664         conf_offloads = pi->eth_dev->data->dev_conf.rxmode.offloads;
1665
1666         /*
1667          * We do not set address filters and promiscuity here, the stack does
1668          * that step explicitly.
1669          */
1670         ret = t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu, -1, -1, -1,
1671                             !!(conf_offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP),
1672                             true);
1673         if (ret == 0) {
1674                 ret = cxgbe_mpstcam_modify(pi, (int)pi->xact_addr_filt,
1675                                 (u8 *)&pi->eth_dev->data->mac_addrs[0]);
1676                 if (ret >= 0) {
1677                         pi->xact_addr_filt = ret;
1678                         ret = 0;
1679                 }
1680         }
1681         if (ret == 0 && is_pf4(adapter))
1682                 ret = t4_link_l1cfg(pi, pi->link_cfg.admin_caps);
1683         if (ret == 0) {
1684                 /* Disable VI Rx until link up message is received in
1685                  * firmware event queue, if firmware supports enabling/
1686                  * disabling VI Rx at runtime.
1687                  */
1688                 pi->vi_en_rx = adapter->params.vi_enable_rx ? 0 : 1;
1689                 pi->vi_en_tx = 1;
1690                 ret = t4_enable_vi_params(adapter, adapter->mbox, pi->viid,
1691                                           pi->vi_en_rx, pi->vi_en_tx, false);
1692         }
1693
1694         if (ret == 0 && cxgbe_force_linkup(adapter))
1695                 pi->eth_dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
1696         return ret;
1697 }
1698
1699 /**
1700  * cxgbe_write_rss_conf - flash the RSS configuration for a given port
1701  * @pi: the port
1702  * @rss_hf: Hash configuration to apply
1703  */
1704 int cxgbe_write_rss_conf(const struct port_info *pi, uint64_t rss_hf)
1705 {
1706         struct adapter *adapter = pi->adapter;
1707         const struct sge_eth_rxq *rxq;
1708         u64 flags = 0;
1709         u16 rss;
1710         int err;
1711
1712         /*  Should never be called before setting up sge eth rx queues */
1713         if (!(adapter->flags & FULL_INIT_DONE)) {
1714                 dev_err(adap, "%s No RXQs available on port %d\n",
1715                         __func__, pi->port_id);
1716                 return -EINVAL;
1717         }
1718
1719         /* Don't allow unsupported hash functions */
1720         if (rss_hf & ~CXGBE_RSS_HF_ALL)
1721                 return -EINVAL;
1722
1723         if (rss_hf & CXGBE_RSS_HF_IPV4_MASK)
1724                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
1725
1726         if (rss_hf & RTE_ETH_RSS_NONFRAG_IPV4_TCP)
1727                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
1728
1729         if (rss_hf & RTE_ETH_RSS_NONFRAG_IPV4_UDP)
1730                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
1731                          F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1732
1733         if (rss_hf & CXGBE_RSS_HF_IPV6_MASK)
1734                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
1735
1736         if (rss_hf & CXGBE_RSS_HF_TCP_IPV6_MASK)
1737                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1738                          F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
1739
1740         if (rss_hf & CXGBE_RSS_HF_UDP_IPV6_MASK)
1741                 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1742                          F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
1743                          F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1744
1745         rxq = &adapter->sge.ethrxq[pi->first_rxqset];
1746         rss = rxq[0].rspq.abs_id;
1747
1748         /* If Tunnel All Lookup isn't specified in the global RSS
1749          * Configuration, then we need to specify a default Ingress
1750          * Queue for any ingress packets which aren't hashed.  We'll
1751          * use our first ingress queue ...
1752          */
1753         err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
1754                                flags, rss);
1755         return err;
1756 }
1757
1758 /**
1759  * cxgbe_write_rss - write the RSS table for a given port
1760  * @pi: the port
1761  * @queues: array of queue indices for RSS
1762  *
1763  * Sets up the portion of the HW RSS table for the port's VI to distribute
1764  * packets to the Rx queues in @queues.
1765  */
1766 int cxgbe_write_rss(const struct port_info *pi, const u16 *queues)
1767 {
1768         u16 *rss;
1769         int i, err;
1770         struct adapter *adapter = pi->adapter;
1771         const struct sge_eth_rxq *rxq;
1772
1773         /*  Should never be called before setting up sge eth rx queues */
1774         BUG_ON(!(adapter->flags & FULL_INIT_DONE));
1775
1776         rxq = &adapter->sge.ethrxq[pi->first_rxqset];
1777         rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
1778         if (!rss)
1779                 return -ENOMEM;
1780
1781         /* map the queue indices to queue ids */
1782         for (i = 0; i < pi->rss_size; i++, queues++)
1783                 rss[i] = rxq[*queues].rspq.abs_id;
1784
1785         err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
1786                                   pi->rss_size, rss, pi->rss_size);
1787         rte_free(rss);
1788         return err;
1789 }
1790
1791 /**
1792  * setup_rss - configure RSS
1793  * @adapter: the adapter
1794  *
1795  * Sets up RSS to distribute packets to multiple receive queues.  We
1796  * configure the RSS CPU lookup table to distribute to the number of HW
1797  * receive queues, and the response queue lookup table to narrow that
1798  * down to the response queues actually configured for each port.
1799  * We always configure the RSS mapping for all ports since the mapping
1800  * table has plenty of entries.
1801  */
1802 int cxgbe_setup_rss(struct port_info *pi)
1803 {
1804         int j, err;
1805         struct adapter *adapter = pi->adapter;
1806
1807         dev_debug(adapter, "%s:  pi->rss_size = %u; pi->n_rx_qsets = %u\n",
1808                   __func__, pi->rss_size, pi->n_rx_qsets);
1809
1810         if (!(pi->flags & PORT_RSS_DONE)) {
1811                 if (adapter->flags & FULL_INIT_DONE) {
1812                         /* Fill default values with equal distribution */
1813                         for (j = 0; j < pi->rss_size; j++)
1814                                 pi->rss[j] = j % pi->n_rx_qsets;
1815
1816                         err = cxgbe_write_rss(pi, pi->rss);
1817                         if (err)
1818                                 return err;
1819
1820                         err = cxgbe_write_rss_conf(pi, pi->rss_hf);
1821                         if (err)
1822                                 return err;
1823                         pi->flags |= PORT_RSS_DONE;
1824                 }
1825         }
1826         return 0;
1827 }
1828
1829 /*
1830  * Enable NAPI scheduling and interrupt generation for all Rx queues.
1831  */
1832 static void enable_rx(struct adapter *adap, struct sge_rspq *q)
1833 {
1834         /* 0-increment GTS to start the timer and enable interrupts */
1835         t4_write_reg(adap, is_pf4(adap) ? MYPF_REG(A_SGE_PF_GTS) :
1836                                           T4VF_SGE_BASE_ADDR + A_SGE_VF_GTS,
1837                      V_SEINTARM(q->intr_params) |
1838                      V_INGRESSQID(q->cntxt_id));
1839 }
1840
1841 void cxgbe_enable_rx_queues(struct port_info *pi)
1842 {
1843         struct adapter *adap = pi->adapter;
1844         struct sge *s = &adap->sge;
1845         unsigned int i;
1846
1847         for (i = 0; i < pi->n_rx_qsets; i++)
1848                 enable_rx(adap, &s->ethrxq[pi->first_rxqset + i].rspq);
1849 }
1850
1851 /**
1852  * fw_caps_to_speed_caps - translate Firmware Port Caps to Speed Caps.
1853  * @port_type: Firmware Port Type
1854  * @fw_caps: Firmware Port Capabilities
1855  * @speed_caps: Device Info Speed Capabilities
1856  *
1857  * Translate a Firmware Port Capabilities specification to Device Info
1858  * Speed Capabilities.
1859  */
1860 static void fw_caps_to_speed_caps(enum fw_port_type port_type,
1861                                   unsigned int fw_caps,
1862                                   u32 *speed_caps)
1863 {
1864 #define SET_SPEED(__speed_name) \
1865         do { \
1866                 *speed_caps |= RTE_ETH_LINK_ ## __speed_name; \
1867         } while (0)
1868
1869 #define FW_CAPS_TO_SPEED(__fw_name) \
1870         do { \
1871                 if (fw_caps & FW_PORT_CAP32_ ## __fw_name) \
1872                         SET_SPEED(__fw_name); \
1873         } while (0)
1874
1875         switch (port_type) {
1876         case FW_PORT_TYPE_BT_SGMII:
1877         case FW_PORT_TYPE_BT_XFI:
1878         case FW_PORT_TYPE_BT_XAUI:
1879                 FW_CAPS_TO_SPEED(SPEED_100M);
1880                 FW_CAPS_TO_SPEED(SPEED_1G);
1881                 FW_CAPS_TO_SPEED(SPEED_10G);
1882                 break;
1883
1884         case FW_PORT_TYPE_KX4:
1885         case FW_PORT_TYPE_KX:
1886         case FW_PORT_TYPE_FIBER_XFI:
1887         case FW_PORT_TYPE_FIBER_XAUI:
1888         case FW_PORT_TYPE_SFP:
1889         case FW_PORT_TYPE_QSFP_10G:
1890         case FW_PORT_TYPE_QSA:
1891                 FW_CAPS_TO_SPEED(SPEED_1G);
1892                 FW_CAPS_TO_SPEED(SPEED_10G);
1893                 break;
1894
1895         case FW_PORT_TYPE_KR:
1896                 SET_SPEED(SPEED_10G);
1897                 break;
1898
1899         case FW_PORT_TYPE_BP_AP:
1900         case FW_PORT_TYPE_BP4_AP:
1901                 SET_SPEED(SPEED_1G);
1902                 SET_SPEED(SPEED_10G);
1903                 break;
1904
1905         case FW_PORT_TYPE_BP40_BA:
1906         case FW_PORT_TYPE_QSFP:
1907                 SET_SPEED(SPEED_40G);
1908                 break;
1909
1910         case FW_PORT_TYPE_CR_QSFP:
1911         case FW_PORT_TYPE_SFP28:
1912         case FW_PORT_TYPE_KR_SFP28:
1913                 FW_CAPS_TO_SPEED(SPEED_1G);
1914                 FW_CAPS_TO_SPEED(SPEED_10G);
1915                 FW_CAPS_TO_SPEED(SPEED_25G);
1916                 break;
1917
1918         case FW_PORT_TYPE_CR2_QSFP:
1919                 SET_SPEED(SPEED_50G);
1920                 break;
1921
1922         case FW_PORT_TYPE_KR4_100G:
1923         case FW_PORT_TYPE_CR4_QSFP:
1924                 FW_CAPS_TO_SPEED(SPEED_25G);
1925                 FW_CAPS_TO_SPEED(SPEED_40G);
1926                 FW_CAPS_TO_SPEED(SPEED_50G);
1927                 FW_CAPS_TO_SPEED(SPEED_100G);
1928                 break;
1929
1930         default:
1931                 break;
1932         }
1933
1934 #undef FW_CAPS_TO_SPEED
1935 #undef SET_SPEED
1936 }
1937
1938 /**
1939  * cxgbe_get_speed_caps - Fetch supported speed capabilities
1940  * @pi: Underlying port's info
1941  * @speed_caps: Device Info speed capabilities
1942  *
1943  * Fetch supported speed capabilities of the underlying port.
1944  */
1945 void cxgbe_get_speed_caps(struct port_info *pi, u32 *speed_caps)
1946 {
1947         *speed_caps = 0;
1948
1949         fw_caps_to_speed_caps(pi->link_cfg.port_type, pi->link_cfg.pcaps,
1950                               speed_caps);
1951
1952         if (!(pi->link_cfg.pcaps & FW_PORT_CAP32_ANEG))
1953                 *speed_caps |= RTE_ETH_LINK_SPEED_FIXED;
1954 }
1955
1956 /**
1957  * cxgbe_set_link_status - Set device link up or down.
1958  * @pi: Underlying port's info
1959  * @status: 0 - down, 1 - up
1960  *
1961  * Set the device link up or down.
1962  */
1963 int cxgbe_set_link_status(struct port_info *pi, bool status)
1964 {
1965         struct adapter *adapter = pi->adapter;
1966         int err = 0;
1967
1968         /* Wait for link up message from firmware to enable Rx path,
1969          * if firmware supports enabling/disabling VI Rx at runtime.
1970          */
1971         pi->vi_en_rx = adapter->params.vi_enable_rx ? 0 : status;
1972         pi->vi_en_tx = status;
1973         err = t4_enable_vi(adapter, adapter->mbox, pi->viid, pi->vi_en_rx,
1974                            pi->vi_en_tx);
1975         if (err) {
1976                 dev_err(adapter, "%s: disable_vi failed: %d\n", __func__, err);
1977                 return err;
1978         }
1979
1980         if (!status)
1981                 t4_reset_link_config(adapter, pi->pidx);
1982
1983         return 0;
1984 }
1985
1986 /**
1987  * cxgb_up - enable the adapter
1988  * @adap: adapter being enabled
1989  *
1990  * Called when the first port is enabled, this function performs the
1991  * actions necessary to make an adapter operational, such as completing
1992  * the initialization of HW modules, and enabling interrupts.
1993  */
1994 int cxgbe_up(struct adapter *adap)
1995 {
1996         enable_rx(adap, &adap->sge.fw_evtq);
1997         t4_sge_tx_monitor_start(adap);
1998         if (is_pf4(adap))
1999                 t4_intr_enable(adap);
2000         adap->flags |= FULL_INIT_DONE;
2001
2002         /* TODO: deadman watchdog ?? */
2003         return 0;
2004 }
2005
2006 /*
2007  * Close the port
2008  */
2009 int cxgbe_down(struct port_info *pi)
2010 {
2011         return cxgbe_set_link_status(pi, false);
2012 }
2013
2014 /*
2015  * Release resources when all the ports have been stopped.
2016  */
2017 void cxgbe_close(struct adapter *adapter)
2018 {
2019         if (adapter->flags & FULL_INIT_DONE) {
2020                 tid_free(&adapter->tids);
2021                 t4_cleanup_mpstcam(adapter);
2022                 t4_cleanup_clip_tbl(adapter);
2023                 t4_cleanup_l2t(adapter);
2024                 t4_cleanup_smt(adapter);
2025                 if (is_pf4(adapter))
2026                         t4_intr_disable(adapter);
2027                 t4_sge_tx_monitor_stop(adapter);
2028                 t4_free_sge_resources(adapter);
2029                 adapter->flags &= ~FULL_INIT_DONE;
2030         }
2031
2032         cxgbe_cfg_queues_free(adapter);
2033
2034         if (is_pf4(adapter) && (adapter->flags & FW_OK))
2035                 t4_fw_bye(adapter, adapter->mbox);
2036 }
2037
2038 static void adap_smt_index(struct adapter *adapter, u32 *smt_start_idx,
2039                            u32 *smt_size)
2040 {
2041         u32 params[2], smt_val[2];
2042         int ret;
2043
2044         params[0] = CXGBE_FW_PARAM_PFVF(GET_SMT_START);
2045         params[1] = CXGBE_FW_PARAM_PFVF(GET_SMT_SIZE);
2046
2047         ret = t4_query_params(adapter, adapter->mbox, adapter->pf, 0,
2048                               2, params, smt_val);
2049
2050         /* if FW doesn't recognize this command then set it to default setting
2051          * which is start index as 0 and size as 256.
2052          */
2053         if (ret < 0) {
2054                 *smt_start_idx = 0;
2055                 *smt_size = SMT_SIZE;
2056         } else {
2057                 *smt_start_idx = smt_val[0];
2058                 /* smt size can be zero, if nsmt is not yet configured in
2059                  * the config file or set as zero, then configure all the
2060                  * remaining entries to this PF itself.
2061                  */
2062                 if (!smt_val[1])
2063                         *smt_size = SMT_SIZE - *smt_start_idx;
2064                 else
2065                         *smt_size = smt_val[1];
2066         }
2067 }
2068
2069 int cxgbe_probe(struct adapter *adapter)
2070 {
2071         u32 smt_start_idx, smt_size;
2072         struct port_info *pi;
2073         int func, i;
2074         int err = 0;
2075         u32 whoami;
2076         int chip;
2077
2078         whoami = t4_read_reg(adapter, A_PL_WHOAMI);
2079         chip = t4_get_chip_type(adapter,
2080                         CHELSIO_PCI_ID_VER(adapter->pdev->id.device_id));
2081         if (chip < 0)
2082                 return chip;
2083
2084         func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
2085                G_SOURCEPF(whoami) : G_T6_SOURCEPF(whoami);
2086
2087         adapter->mbox = func;
2088         adapter->pf = func;
2089
2090         t4_os_lock_init(&adapter->mbox_lock);
2091         TAILQ_INIT(&adapter->mbox_list);
2092         t4_os_lock_init(&adapter->win0_lock);
2093
2094         err = t4_prep_adapter(adapter);
2095         if (err)
2096                 return err;
2097
2098         setup_memwin(adapter);
2099         err = adap_init0(adapter);
2100         if (err) {
2101                 dev_err(adapter, "%s: Adapter initialization failed, error %d\n",
2102                         __func__, err);
2103                 goto out_free;
2104         }
2105
2106         if (!is_t4(adapter->params.chip)) {
2107                 /*
2108                  * The userspace doorbell BAR is split evenly into doorbell
2109                  * regions, each associated with an egress queue.  If this
2110                  * per-queue region is large enough (at least UDBS_SEG_SIZE)
2111                  * then it can be used to submit a tx work request with an
2112                  * implied doorbell.  Enable write combining on the BAR if
2113                  * there is room for such work requests.
2114                  */
2115                 int s_qpp, qpp, num_seg;
2116
2117                 s_qpp = (S_QUEUESPERPAGEPF0 +
2118                         (S_QUEUESPERPAGEPF1 - S_QUEUESPERPAGEPF0) *
2119                         adapter->pf);
2120                 qpp = 1 << ((t4_read_reg(adapter,
2121                                 A_SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp)
2122                                 & M_QUEUESPERPAGEPF0);
2123                 num_seg = CXGBE_PAGE_SIZE / UDBS_SEG_SIZE;
2124                 if (qpp > num_seg)
2125                         dev_warn(adapter, "Incorrect SGE EGRESS QUEUES_PER_PAGE configuration, continuing in debug mode\n");
2126
2127                 adapter->bar2 = (void *)adapter->pdev->mem_resource[2].addr;
2128                 if (!adapter->bar2) {
2129                         dev_err(adapter, "cannot map device bar2 region\n");
2130                         err = -ENOMEM;
2131                         goto out_free;
2132                 }
2133                 t4_write_reg(adapter, A_SGE_STAT_CFG, V_STATSOURCE_T5(7) |
2134                              V_STATMODE(0));
2135         }
2136
2137         for_each_port(adapter, i) {
2138                 const unsigned int numa_node = rte_socket_id();
2139                 char name[RTE_ETH_NAME_MAX_LEN];
2140                 struct rte_eth_dev *eth_dev;
2141
2142                 snprintf(name, sizeof(name), "%s_%d",
2143                          adapter->pdev->device.name, i);
2144
2145                 if (i == 0) {
2146                         /* First port is already allocated by DPDK */
2147                         eth_dev = adapter->eth_dev;
2148                         goto allocate_mac;
2149                 }
2150
2151                 /*
2152                  * now do all data allocation - for eth_dev structure,
2153                  * and internal (private) data for the remaining ports
2154                  */
2155
2156                 /* reserve an ethdev entry */
2157                 eth_dev = rte_eth_dev_allocate(name);
2158                 if (!eth_dev)
2159                         goto out_free;
2160
2161                 eth_dev->data->dev_private =
2162                         rte_zmalloc_socket(name, sizeof(struct port_info),
2163                                            RTE_CACHE_LINE_SIZE, numa_node);
2164                 if (!eth_dev->data->dev_private)
2165                         goto out_free;
2166
2167 allocate_mac:
2168                 pi = eth_dev->data->dev_private;
2169                 adapter->port[i] = pi;
2170                 pi->eth_dev = eth_dev;
2171                 pi->adapter = adapter;
2172                 pi->xact_addr_filt = -1;
2173                 pi->port_id = i;
2174                 pi->pidx = i;
2175
2176                 pi->eth_dev->device = &adapter->pdev->device;
2177                 pi->eth_dev->dev_ops = adapter->eth_dev->dev_ops;
2178                 pi->eth_dev->tx_pkt_burst = adapter->eth_dev->tx_pkt_burst;
2179                 pi->eth_dev->rx_pkt_burst = adapter->eth_dev->rx_pkt_burst;
2180
2181                 rte_eth_copy_pci_info(pi->eth_dev, adapter->pdev);
2182
2183                 pi->eth_dev->data->mac_addrs = rte_zmalloc(name,
2184                                                         RTE_ETHER_ADDR_LEN, 0);
2185                 if (!pi->eth_dev->data->mac_addrs) {
2186                         dev_err(adapter, "%s: Mem allocation failed for storing mac addr, aborting\n",
2187                                 __func__);
2188                         err = -1;
2189                         goto out_free;
2190                 }
2191
2192                 if (i > 0) {
2193                         /* First port will be notified by upper layer */
2194                         rte_eth_dev_probing_finish(eth_dev);
2195                 }
2196         }
2197
2198         if (adapter->flags & FW_OK) {
2199                 err = t4_port_init(adapter, adapter->mbox, adapter->pf, 0);
2200                 if (err) {
2201                         dev_err(adapter, "%s: t4_port_init failed with err %d\n",
2202                                 __func__, err);
2203                         goto out_free;
2204                 }
2205         }
2206
2207         err = cxgbe_cfg_queues(adapter->eth_dev);
2208         if (err)
2209                 goto out_free;
2210
2211         cxgbe_print_adapter_info(adapter);
2212         cxgbe_print_port_info(adapter);
2213
2214         adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
2215                                           adapter->clipt_end);
2216         if (!adapter->clipt) {
2217                 /* We tolerate a lack of clip_table, giving up some
2218                  * functionality
2219                  */
2220                 dev_warn(adapter, "could not allocate CLIP. Continuing\n");
2221         }
2222
2223         adap_smt_index(adapter, &smt_start_idx, &smt_size);
2224         adapter->smt = t4_init_smt(smt_start_idx, smt_size);
2225         if (!adapter->smt)
2226                 dev_warn(adapter, "could not allocate SMT, continuing\n");
2227
2228         adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
2229         if (!adapter->l2t) {
2230                 /* We tolerate a lack of L2T, giving up some functionality */
2231                 dev_warn(adapter, "could not allocate L2T. Continuing\n");
2232         }
2233
2234         if (tid_init(&adapter->tids) < 0) {
2235                 /* Disable filtering support */
2236                 dev_warn(adapter, "could not allocate TID table, "
2237                          "filter support disabled. Continuing\n");
2238         }
2239
2240         t4_os_lock_init(&adapter->flow_lock);
2241
2242         adapter->mpstcam = t4_init_mpstcam(adapter);
2243         if (!adapter->mpstcam)
2244                 dev_warn(adapter, "could not allocate mps tcam table."
2245                          " Continuing\n");
2246
2247         if (is_hashfilter(adapter)) {
2248                 if (t4_read_reg(adapter, A_LE_DB_CONFIG) & F_HASHEN) {
2249                         u32 hash_base, hash_reg;
2250
2251                         hash_reg = A_LE_DB_TID_HASHBASE;
2252                         hash_base = t4_read_reg(adapter, hash_reg);
2253                         adapter->tids.hash_base = hash_base / 4;
2254                 }
2255         } else {
2256                 /* Disable hash filtering support */
2257                 dev_warn(adapter,
2258                          "Maskless filter support disabled. Continuing\n");
2259         }
2260
2261         err = cxgbe_init_rss(adapter);
2262         if (err)
2263                 goto out_free;
2264
2265         return 0;
2266
2267 out_free:
2268         cxgbe_cfg_queues_free(adapter);
2269
2270         for_each_port(adapter, i) {
2271                 pi = adap2pinfo(adapter, i);
2272                 if (pi->viid != 0)
2273                         t4_free_vi(adapter, adapter->mbox, adapter->pf,
2274                                    0, pi->viid);
2275                 rte_eth_dev_release_port(pi->eth_dev);
2276         }
2277
2278         if (adapter->flags & FW_OK)
2279                 t4_fw_bye(adapter, adapter->mbox);
2280         return -err;
2281 }