xref: /dpdk/drivers/net/cxgbe/cxgbe_main.c (revision 250c9eb3ca895127f21a729caf4a928eb2f04d2c)
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_atomic.h>
24 #include <rte_branch_prediction.h>
25 #include <rte_memory.h>
26 #include <rte_tailq.h>
27 #include <rte_eal.h>
28 #include <rte_alarm.h>
29 #include <rte_ether.h>
30 #include <rte_ethdev_driver.h>
31 #include <rte_ethdev_pci.h>
32 #include <rte_random.h>
33 #include <rte_dev.h>
34 #include <rte_kvargs.h>
35 
36 #include "common.h"
37 #include "t4_regs.h"
38 #include "t4_msg.h"
39 #include "cxgbe.h"
40 #include "clip_tbl.h"
41 
42 /**
43  * Allocate a chunk of memory. The allocated memory is cleared.
44  */
45 void *t4_alloc_mem(size_t size)
46 {
47 	return rte_zmalloc(NULL, size, 0);
48 }
49 
50 /**
51  * Free memory allocated through t4_alloc_mem().
52  */
53 void t4_free_mem(void *addr)
54 {
55 	rte_free(addr);
56 }
57 
58 /*
59  * Response queue handler for the FW event queue.
60  */
61 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
62 			  __rte_unused const struct pkt_gl *gl)
63 {
64 	u8 opcode = ((const struct rss_header *)rsp)->opcode;
65 
66 	rsp++;                                          /* skip RSS header */
67 
68 	/*
69 	 * FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
70 	 */
71 	if (unlikely(opcode == CPL_FW4_MSG &&
72 		     ((const struct cpl_fw4_msg *)rsp)->type ==
73 		      FW_TYPE_RSSCPL)) {
74 		rsp++;
75 		opcode = ((const struct rss_header *)rsp)->opcode;
76 		rsp++;
77 		if (opcode != CPL_SGE_EGR_UPDATE) {
78 			dev_err(q->adapter, "unexpected FW4/CPL %#x on FW event queue\n",
79 				opcode);
80 			goto out;
81 		}
82 	}
83 
84 	if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
85 		/* do nothing */
86 	} else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
87 		const struct cpl_fw6_msg *msg = (const void *)rsp;
88 
89 		t4_handle_fw_rpl(q->adapter, msg->data);
90 	} else if (opcode == CPL_ABORT_RPL_RSS) {
91 		const struct cpl_abort_rpl_rss *p = (const void *)rsp;
92 
93 		hash_del_filter_rpl(q->adapter, p);
94 	} else if (opcode == CPL_SET_TCB_RPL) {
95 		const struct cpl_set_tcb_rpl *p = (const void *)rsp;
96 
97 		filter_rpl(q->adapter, p);
98 	} else if (opcode == CPL_ACT_OPEN_RPL) {
99 		const struct cpl_act_open_rpl *p = (const void *)rsp;
100 
101 		hash_filter_rpl(q->adapter, p);
102 	} else {
103 		dev_err(adapter, "unexpected CPL %#x on FW event queue\n",
104 			opcode);
105 	}
106 out:
107 	return 0;
108 }
109 
110 /**
111  * Setup sge control queues to pass control information.
112  */
113 int setup_sge_ctrl_txq(struct adapter *adapter)
114 {
115 	struct sge *s = &adapter->sge;
116 	int err = 0, i = 0;
117 
118 	for_each_port(adapter, i) {
119 		char name[RTE_ETH_NAME_MAX_LEN];
120 		struct sge_ctrl_txq *q = &s->ctrlq[i];
121 
122 		q->q.size = 1024;
123 		err = t4_sge_alloc_ctrl_txq(adapter, q,
124 					    adapter->eth_dev,  i,
125 					    s->fw_evtq.cntxt_id,
126 					    rte_socket_id());
127 		if (err) {
128 			dev_err(adapter, "Failed to alloc ctrl txq. Err: %d",
129 				err);
130 			goto out;
131 		}
132 		snprintf(name, sizeof(name), "cxgbe_ctrl_pool_%d", i);
133 		q->mb_pool = rte_pktmbuf_pool_create(name, s->ctrlq[i].q.size,
134 						     RTE_CACHE_LINE_SIZE,
135 						     RTE_MBUF_PRIV_ALIGN,
136 						     RTE_MBUF_DEFAULT_BUF_SIZE,
137 						     SOCKET_ID_ANY);
138 		if (!q->mb_pool) {
139 			dev_err(adapter, "Can't create ctrl pool for port: %d",
140 				i);
141 			err = -ENOMEM;
142 			goto out;
143 		}
144 	}
145 	return 0;
146 out:
147 	t4_free_sge_resources(adapter);
148 	return err;
149 }
150 
151 /**
152  * cxgbe_poll_for_completion: Poll rxq for completion
153  * @q: rxq to poll
154  * @us: microseconds to delay
155  * @cnt: number of times to poll
156  * @c: completion to check for 'done' status
157  *
158  * Polls the rxq for reples until completion is done or the count
159  * expires.
160  */
161 int cxgbe_poll_for_completion(struct sge_rspq *q, unsigned int us,
162 			      unsigned int cnt, struct t4_completion *c)
163 {
164 	unsigned int i;
165 	unsigned int work_done, budget = 4;
166 
167 	if (!c)
168 		return -EINVAL;
169 
170 	for (i = 0; i < cnt; i++) {
171 		cxgbe_poll(q, NULL, budget, &work_done);
172 		t4_os_lock(&c->lock);
173 		if (c->done) {
174 			t4_os_unlock(&c->lock);
175 			return 0;
176 		}
177 		t4_os_unlock(&c->lock);
178 		udelay(us);
179 	}
180 	return -ETIMEDOUT;
181 }
182 
183 int setup_sge_fwevtq(struct adapter *adapter)
184 {
185 	struct sge *s = &adapter->sge;
186 	int err = 0;
187 	int msi_idx = 0;
188 
189 	err = t4_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->eth_dev,
190 			       msi_idx, NULL, fwevtq_handler, -1, NULL, 0,
191 			       rte_socket_id());
192 	return err;
193 }
194 
195 static int closest_timer(const struct sge *s, int time)
196 {
197 	unsigned int i, match = 0;
198 	int delta, min_delta = INT_MAX;
199 
200 	for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
201 		delta = time - s->timer_val[i];
202 		if (delta < 0)
203 			delta = -delta;
204 		if (delta < min_delta) {
205 			min_delta = delta;
206 			match = i;
207 		}
208 	}
209 	return match;
210 }
211 
212 static int closest_thres(const struct sge *s, int thres)
213 {
214 	unsigned int i, match = 0;
215 	int delta, min_delta = INT_MAX;
216 
217 	for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
218 		delta = thres - s->counter_val[i];
219 		if (delta < 0)
220 			delta = -delta;
221 		if (delta < min_delta) {
222 			min_delta = delta;
223 			match = i;
224 		}
225 	}
226 	return match;
227 }
228 
229 /**
230  * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
231  * @q: the Rx queue
232  * @us: the hold-off time in us, or 0 to disable timer
233  * @cnt: the hold-off packet count, or 0 to disable counter
234  *
235  * Sets an Rx queue's interrupt hold-off time and packet count.  At least
236  * one of the two needs to be enabled for the queue to generate interrupts.
237  */
238 int cxgb4_set_rspq_intr_params(struct sge_rspq *q, unsigned int us,
239 			       unsigned int cnt)
240 {
241 	struct adapter *adap = q->adapter;
242 	unsigned int timer_val;
243 
244 	if (cnt) {
245 		int err;
246 		u32 v, new_idx;
247 
248 		new_idx = closest_thres(&adap->sge, cnt);
249 		if (q->desc && q->pktcnt_idx != new_idx) {
250 			/* the queue has already been created, update it */
251 			v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
252 			    V_FW_PARAMS_PARAM_X(
253 			    FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
254 			    V_FW_PARAMS_PARAM_YZ(q->cntxt_id);
255 			err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
256 					    &v, &new_idx);
257 			if (err)
258 				return err;
259 		}
260 		q->pktcnt_idx = new_idx;
261 	}
262 
263 	timer_val = (us == 0) ? X_TIMERREG_RESTART_COUNTER :
264 				closest_timer(&adap->sge, us);
265 
266 	if ((us | cnt) == 0)
267 		q->intr_params = V_QINTR_TIMER_IDX(X_TIMERREG_UPDATE_CIDX);
268 	else
269 		q->intr_params = V_QINTR_TIMER_IDX(timer_val) |
270 				 V_QINTR_CNT_EN(cnt > 0);
271 	return 0;
272 }
273 
274 /**
275  * Allocate an active-open TID and set it to the supplied value.
276  */
277 int cxgbe_alloc_atid(struct tid_info *t, void *data)
278 {
279 	int atid = -1;
280 
281 	t4_os_lock(&t->atid_lock);
282 	if (t->afree) {
283 		union aopen_entry *p = t->afree;
284 
285 		atid = p - t->atid_tab;
286 		t->afree = p->next;
287 		p->data = data;
288 		t->atids_in_use++;
289 	}
290 	t4_os_unlock(&t->atid_lock);
291 	return atid;
292 }
293 
294 /**
295  * Release an active-open TID.
296  */
297 void cxgbe_free_atid(struct tid_info *t, unsigned int atid)
298 {
299 	union aopen_entry *p = &t->atid_tab[atid];
300 
301 	t4_os_lock(&t->atid_lock);
302 	p->next = t->afree;
303 	t->afree = p;
304 	t->atids_in_use--;
305 	t4_os_unlock(&t->atid_lock);
306 }
307 
308 /**
309  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
310  */
311 static void mk_tid_release(struct rte_mbuf *mbuf, unsigned int tid)
312 {
313 	struct cpl_tid_release *req;
314 
315 	req = rte_pktmbuf_mtod(mbuf, struct cpl_tid_release *);
316 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
317 }
318 
319 /**
320  * Release a TID and inform HW.  If we are unable to allocate the release
321  * message we defer to a work queue.
322  */
323 void cxgbe_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
324 		      unsigned short family)
325 {
326 	struct rte_mbuf *mbuf;
327 	struct adapter *adap = container_of(t, struct adapter, tids);
328 
329 	WARN_ON(tid >= t->ntids);
330 
331 	if (t->tid_tab[tid]) {
332 		t->tid_tab[tid] = NULL;
333 		rte_atomic32_dec(&t->conns_in_use);
334 		if (t->hash_base && tid >= t->hash_base) {
335 			if (family == FILTER_TYPE_IPV4)
336 				rte_atomic32_dec(&t->hash_tids_in_use);
337 		} else {
338 			if (family == FILTER_TYPE_IPV4)
339 				rte_atomic32_dec(&t->tids_in_use);
340 		}
341 	}
342 
343 	mbuf = rte_pktmbuf_alloc((&adap->sge.ctrlq[chan])->mb_pool);
344 	if (mbuf) {
345 		mbuf->data_len = sizeof(struct cpl_tid_release);
346 		mbuf->pkt_len = mbuf->data_len;
347 		mk_tid_release(mbuf, tid);
348 		t4_mgmt_tx(&adap->sge.ctrlq[chan], mbuf);
349 	}
350 }
351 
352 /**
353  * Insert a TID.
354  */
355 void cxgbe_insert_tid(struct tid_info *t, void *data, unsigned int tid,
356 		      unsigned short family)
357 {
358 	t->tid_tab[tid] = data;
359 	if (t->hash_base && tid >= t->hash_base) {
360 		if (family == FILTER_TYPE_IPV4)
361 			rte_atomic32_inc(&t->hash_tids_in_use);
362 	} else {
363 		if (family == FILTER_TYPE_IPV4)
364 			rte_atomic32_inc(&t->tids_in_use);
365 	}
366 
367 	rte_atomic32_inc(&t->conns_in_use);
368 }
369 
370 /**
371  * Free TID tables.
372  */
373 static void tid_free(struct tid_info *t)
374 {
375 	if (t->tid_tab) {
376 		if (t->ftid_bmap)
377 			rte_bitmap_free(t->ftid_bmap);
378 
379 		if (t->ftid_bmap_array)
380 			t4_os_free(t->ftid_bmap_array);
381 
382 		t4_os_free(t->tid_tab);
383 	}
384 
385 	memset(t, 0, sizeof(struct tid_info));
386 }
387 
388 /**
389  * Allocate and initialize the TID tables.  Returns 0 on success.
390  */
391 static int tid_init(struct tid_info *t)
392 {
393 	size_t size;
394 	unsigned int ftid_bmap_size;
395 	unsigned int natids = t->natids;
396 	unsigned int max_ftids = t->nftids;
397 
398 	ftid_bmap_size = rte_bitmap_get_memory_footprint(t->nftids);
399 	size = t->ntids * sizeof(*t->tid_tab) +
400 		max_ftids * sizeof(*t->ftid_tab) +
401 		natids * sizeof(*t->atid_tab);
402 
403 	t->tid_tab = t4_os_alloc(size);
404 	if (!t->tid_tab)
405 		return -ENOMEM;
406 
407 	t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
408 	t->ftid_tab = (struct filter_entry *)&t->tid_tab[t->natids];
409 	t->ftid_bmap_array = t4_os_alloc(ftid_bmap_size);
410 	if (!t->ftid_bmap_array) {
411 		tid_free(t);
412 		return -ENOMEM;
413 	}
414 
415 	t4_os_lock_init(&t->atid_lock);
416 	t4_os_lock_init(&t->ftid_lock);
417 
418 	t->afree = NULL;
419 	t->atids_in_use = 0;
420 	rte_atomic32_init(&t->tids_in_use);
421 	rte_atomic32_set(&t->tids_in_use, 0);
422 	rte_atomic32_init(&t->conns_in_use);
423 	rte_atomic32_set(&t->conns_in_use, 0);
424 
425 	/* Setup the free list for atid_tab and clear the stid bitmap. */
426 	if (natids) {
427 		while (--natids)
428 			t->atid_tab[natids - 1].next = &t->atid_tab[natids];
429 		t->afree = t->atid_tab;
430 	}
431 
432 	t->ftid_bmap = rte_bitmap_init(t->nftids, t->ftid_bmap_array,
433 				       ftid_bmap_size);
434 	if (!t->ftid_bmap) {
435 		tid_free(t);
436 		return -ENOMEM;
437 	}
438 
439 	return 0;
440 }
441 
442 static inline bool is_x_1g_port(const struct link_config *lc)
443 {
444 	return (lc->pcaps & FW_PORT_CAP32_SPEED_1G) != 0;
445 }
446 
447 static inline bool is_x_10g_port(const struct link_config *lc)
448 {
449 	unsigned int speeds, high_speeds;
450 
451 	speeds = V_FW_PORT_CAP32_SPEED(G_FW_PORT_CAP32_SPEED(lc->pcaps));
452 	high_speeds = speeds &
453 		      ~(FW_PORT_CAP32_SPEED_100M | FW_PORT_CAP32_SPEED_1G);
454 
455 	return high_speeds != 0;
456 }
457 
458 inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
459 		      unsigned int us, unsigned int cnt,
460 		      unsigned int size, unsigned int iqe_size)
461 {
462 	q->adapter = adap;
463 	cxgb4_set_rspq_intr_params(q, us, cnt);
464 	q->iqe_len = iqe_size;
465 	q->size = size;
466 }
467 
468 int cfg_queue_count(struct rte_eth_dev *eth_dev)
469 {
470 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
471 	struct adapter *adap = pi->adapter;
472 	struct sge *s = &adap->sge;
473 	unsigned int max_queues = s->max_ethqsets / adap->params.nports;
474 
475 	if ((eth_dev->data->nb_rx_queues < 1) ||
476 	    (eth_dev->data->nb_tx_queues < 1))
477 		return -EINVAL;
478 
479 	if ((eth_dev->data->nb_rx_queues > max_queues) ||
480 	    (eth_dev->data->nb_tx_queues > max_queues))
481 		return -EINVAL;
482 
483 	if (eth_dev->data->nb_rx_queues > pi->rss_size)
484 		return -EINVAL;
485 
486 	/* We must configure RSS, since config has changed*/
487 	pi->flags &= ~PORT_RSS_DONE;
488 
489 	pi->n_rx_qsets = eth_dev->data->nb_rx_queues;
490 	pi->n_tx_qsets = eth_dev->data->nb_tx_queues;
491 
492 	return 0;
493 }
494 
495 void cfg_queues(struct rte_eth_dev *eth_dev)
496 {
497 	struct rte_config *config = rte_eal_get_configuration();
498 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
499 	struct adapter *adap = pi->adapter;
500 	struct sge *s = &adap->sge;
501 	unsigned int i, nb_ports = 0, qidx = 0;
502 	unsigned int q_per_port = 0;
503 
504 	if (!(adap->flags & CFG_QUEUES)) {
505 		for_each_port(adap, i) {
506 			struct port_info *tpi = adap2pinfo(adap, i);
507 
508 			nb_ports += (is_x_10g_port(&tpi->link_cfg)) ||
509 				     is_x_1g_port(&tpi->link_cfg) ? 1 : 0;
510 		}
511 
512 		/*
513 		 * We default up to # of cores queues per 1G/10G port.
514 		 */
515 		if (nb_ports)
516 			q_per_port = (s->max_ethqsets -
517 				     (adap->params.nports - nb_ports)) /
518 				     nb_ports;
519 
520 		if (q_per_port > config->lcore_count)
521 			q_per_port = config->lcore_count;
522 
523 		for_each_port(adap, i) {
524 			struct port_info *pi = adap2pinfo(adap, i);
525 
526 			pi->first_qset = qidx;
527 
528 			/* Initially n_rx_qsets == n_tx_qsets */
529 			pi->n_rx_qsets = (is_x_10g_port(&pi->link_cfg) ||
530 					  is_x_1g_port(&pi->link_cfg)) ?
531 					  q_per_port : 1;
532 			pi->n_tx_qsets = pi->n_rx_qsets;
533 
534 			if (pi->n_rx_qsets > pi->rss_size)
535 				pi->n_rx_qsets = pi->rss_size;
536 
537 			qidx += pi->n_rx_qsets;
538 		}
539 
540 		for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
541 			struct sge_eth_rxq *r = &s->ethrxq[i];
542 
543 			init_rspq(adap, &r->rspq, 5, 32, 1024, 64);
544 			r->usembufs = 1;
545 			r->fl.size = (r->usembufs ? 1024 : 72);
546 		}
547 
548 		for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
549 			s->ethtxq[i].q.size = 1024;
550 
551 		init_rspq(adap, &adap->sge.fw_evtq, 0, 0, 1024, 64);
552 		adap->flags |= CFG_QUEUES;
553 	}
554 }
555 
556 void cxgbe_stats_get(struct port_info *pi, struct port_stats *stats)
557 {
558 	t4_get_port_stats_offset(pi->adapter, pi->tx_chan, stats,
559 				 &pi->stats_base);
560 }
561 
562 void cxgbe_stats_reset(struct port_info *pi)
563 {
564 	t4_clr_port_stats(pi->adapter, pi->tx_chan);
565 }
566 
567 static void setup_memwin(struct adapter *adap)
568 {
569 	u32 mem_win0_base;
570 
571 	/* For T5, only relative offset inside the PCIe BAR is passed */
572 	mem_win0_base = MEMWIN0_BASE;
573 
574 	/*
575 	 * Set up memory window for accessing adapter memory ranges.  (Read
576 	 * back MA register to ensure that changes propagate before we attempt
577 	 * to use the new values.)
578 	 */
579 	t4_write_reg(adap,
580 		     PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
581 					 MEMWIN_NIC),
582 		     mem_win0_base | V_BIR(0) |
583 		     V_WINDOW(ilog2(MEMWIN0_APERTURE) - X_WINDOW_SHIFT));
584 	t4_read_reg(adap,
585 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
586 					MEMWIN_NIC));
587 }
588 
589 int init_rss(struct adapter *adap)
590 {
591 	unsigned int i;
592 
593 	if (is_pf4(adap)) {
594 		int err;
595 
596 		err = t4_init_rss_mode(adap, adap->mbox);
597 		if (err)
598 			return err;
599 	}
600 
601 	for_each_port(adap, i) {
602 		struct port_info *pi = adap2pinfo(adap, i);
603 
604 		pi->rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
605 		if (!pi->rss)
606 			return -ENOMEM;
607 
608 		pi->rss_hf = CXGBE_RSS_HF_ALL;
609 	}
610 	return 0;
611 }
612 
613 /**
614  * Dump basic information about the adapter.
615  */
616 void print_adapter_info(struct adapter *adap)
617 {
618 	/**
619 	 * Hardware/Firmware/etc. Version/Revision IDs.
620 	 */
621 	t4_dump_version_info(adap);
622 }
623 
624 void print_port_info(struct adapter *adap)
625 {
626 	int i;
627 	char buf[80];
628 	struct rte_pci_addr *loc = &adap->pdev->addr;
629 
630 	for_each_port(adap, i) {
631 		const struct port_info *pi = adap2pinfo(adap, i);
632 		char *bufp = buf;
633 
634 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
635 			bufp += sprintf(bufp, "100M/");
636 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
637 			bufp += sprintf(bufp, "1G/");
638 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
639 			bufp += sprintf(bufp, "10G/");
640 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
641 			bufp += sprintf(bufp, "25G/");
642 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
643 			bufp += sprintf(bufp, "40G/");
644 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
645 			bufp += sprintf(bufp, "50G/");
646 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
647 			bufp += sprintf(bufp, "100G/");
648 		if (bufp != buf)
649 			--bufp;
650 		sprintf(bufp, "BASE-%s",
651 			t4_get_port_type_description(
652 					(enum fw_port_type)pi->port_type));
653 
654 		dev_info(adap,
655 			 " " PCI_PRI_FMT " Chelsio rev %d %s %s\n",
656 			 loc->domain, loc->bus, loc->devid, loc->function,
657 			 CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
658 			 (adap->flags & USING_MSIX) ? " MSI-X" :
659 			 (adap->flags & USING_MSI) ? " MSI" : "");
660 	}
661 }
662 
663 static int
664 check_devargs_handler(__rte_unused const char *key, const char *value,
665 		      __rte_unused void *opaque)
666 {
667 	if (strcmp(value, "1"))
668 		return -1;
669 
670 	return 0;
671 }
672 
673 int cxgbe_get_devargs(struct rte_devargs *devargs, const char *key)
674 {
675 	struct rte_kvargs *kvlist;
676 
677 	if (!devargs)
678 		return 0;
679 
680 	kvlist = rte_kvargs_parse(devargs->args, NULL);
681 	if (!kvlist)
682 		return 0;
683 
684 	if (!rte_kvargs_count(kvlist, key)) {
685 		rte_kvargs_free(kvlist);
686 		return 0;
687 	}
688 
689 	if (rte_kvargs_process(kvlist, key,
690 			       check_devargs_handler, NULL) < 0) {
691 		rte_kvargs_free(kvlist);
692 		return 0;
693 	}
694 	rte_kvargs_free(kvlist);
695 
696 	return 1;
697 }
698 
699 static void configure_vlan_types(struct adapter *adapter)
700 {
701 	struct rte_pci_device *pdev = adapter->pdev;
702 	int i;
703 
704 	for_each_port(adapter, i) {
705 		/* OVLAN Type 0x88a8 */
706 		t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN0),
707 				 V_OVLAN_MASK(M_OVLAN_MASK) |
708 				 V_OVLAN_ETYPE(M_OVLAN_ETYPE),
709 				 V_OVLAN_MASK(M_OVLAN_MASK) |
710 				 V_OVLAN_ETYPE(0x88a8));
711 		/* OVLAN Type 0x9100 */
712 		t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN1),
713 				 V_OVLAN_MASK(M_OVLAN_MASK) |
714 				 V_OVLAN_ETYPE(M_OVLAN_ETYPE),
715 				 V_OVLAN_MASK(M_OVLAN_MASK) |
716 				 V_OVLAN_ETYPE(0x9100));
717 		/* OVLAN Type 0x8100 */
718 		t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN2),
719 				 V_OVLAN_MASK(M_OVLAN_MASK) |
720 				 V_OVLAN_ETYPE(M_OVLAN_ETYPE),
721 				 V_OVLAN_MASK(M_OVLAN_MASK) |
722 				 V_OVLAN_ETYPE(0x8100));
723 
724 		/* IVLAN 0X8100 */
725 		t4_set_reg_field(adapter, MPS_PORT_RX_IVLAN(i),
726 				 V_IVLAN_ETYPE(M_IVLAN_ETYPE),
727 				 V_IVLAN_ETYPE(0x8100));
728 
729 		t4_set_reg_field(adapter, MPS_PORT_RX_CTL(i),
730 				 F_OVLAN_EN0 | F_OVLAN_EN1 |
731 				 F_OVLAN_EN2 | F_IVLAN_EN,
732 				 F_OVLAN_EN0 | F_OVLAN_EN1 |
733 				 F_OVLAN_EN2 | F_IVLAN_EN);
734 	}
735 
736 	if (cxgbe_get_devargs(pdev->device.devargs, CXGBE_DEVARG_KEEP_OVLAN))
737 		t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG,
738 				       V_RM_OVLAN(1), V_RM_OVLAN(0));
739 }
740 
741 static void configure_pcie_ext_tag(struct adapter *adapter)
742 {
743 	u16 v;
744 	int pos = t4_os_find_pci_capability(adapter, PCI_CAP_ID_EXP);
745 
746 	if (!pos)
747 		return;
748 
749 	if (pos > 0) {
750 		t4_os_pci_read_cfg2(adapter, pos + PCI_EXP_DEVCTL, &v);
751 		v |= PCI_EXP_DEVCTL_EXT_TAG;
752 		t4_os_pci_write_cfg2(adapter, pos + PCI_EXP_DEVCTL, v);
753 		if (is_t6(adapter->params.chip)) {
754 			t4_set_reg_field(adapter, A_PCIE_CFG2,
755 					 V_T6_TOTMAXTAG(M_T6_TOTMAXTAG),
756 					 V_T6_TOTMAXTAG(7));
757 			t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
758 					 V_T6_MINTAG(M_T6_MINTAG),
759 					 V_T6_MINTAG(8));
760 		} else {
761 			t4_set_reg_field(adapter, A_PCIE_CFG2,
762 					 V_TOTMAXTAG(M_TOTMAXTAG),
763 					 V_TOTMAXTAG(3));
764 			t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
765 					 V_MINTAG(M_MINTAG),
766 					 V_MINTAG(8));
767 		}
768 	}
769 }
770 
771 /* Figure out how many Queue Sets we can support */
772 void configure_max_ethqsets(struct adapter *adapter)
773 {
774 	unsigned int ethqsets;
775 
776 	/*
777 	 * We need to reserve an Ingress Queue for the Asynchronous Firmware
778 	 * Event Queue.
779 	 *
780 	 * For each Queue Set, we'll need the ability to allocate two Egress
781 	 * Contexts -- one for the Ingress Queue Free List and one for the TX
782 	 * Ethernet Queue.
783 	 */
784 	if (is_pf4(adapter)) {
785 		struct pf_resources *pfres = &adapter->params.pfres;
786 
787 		ethqsets = pfres->niqflint - 1;
788 		if (pfres->neq < ethqsets * 2)
789 			ethqsets = pfres->neq / 2;
790 	} else {
791 		struct vf_resources *vfres = &adapter->params.vfres;
792 
793 		ethqsets = vfres->niqflint - 1;
794 		if (vfres->nethctrl != ethqsets)
795 			ethqsets = min(vfres->nethctrl, ethqsets);
796 		if (vfres->neq < ethqsets * 2)
797 			ethqsets = vfres->neq / 2;
798 	}
799 
800 	if (ethqsets > MAX_ETH_QSETS)
801 		ethqsets = MAX_ETH_QSETS;
802 	adapter->sge.max_ethqsets = ethqsets;
803 }
804 
805 /*
806  * Tweak configuration based on system architecture, etc.  Most of these have
807  * defaults assigned to them by Firmware Configuration Files (if we're using
808  * them) but need to be explicitly set if we're using hard-coded
809  * initialization. So these are essentially common tweaks/settings for
810  * Configuration Files and hard-coded initialization ...
811  */
812 static int adap_init0_tweaks(struct adapter *adapter)
813 {
814 	u8 rx_dma_offset;
815 
816 	/*
817 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
818 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
819 	 * 64B Cache Line Size ...
820 	 */
821 	t4_fixup_host_params_compat(adapter, CXGBE_PAGE_SIZE, L1_CACHE_BYTES,
822 				    T5_LAST_REV);
823 
824 	/*
825 	 * Keep the chip default offset to deliver Ingress packets into our
826 	 * DMA buffers to zero
827 	 */
828 	rx_dma_offset = 0;
829 	t4_set_reg_field(adapter, A_SGE_CONTROL, V_PKTSHIFT(M_PKTSHIFT),
830 			 V_PKTSHIFT(rx_dma_offset));
831 
832 	t4_set_reg_field(adapter, A_SGE_FLM_CFG,
833 			 V_CREDITCNT(M_CREDITCNT) | M_CREDITCNTPACKING,
834 			 V_CREDITCNT(3) | V_CREDITCNTPACKING(1));
835 
836 	t4_set_reg_field(adapter, A_SGE_INGRESS_RX_THRESHOLD,
837 			 V_THRESHOLD_3(M_THRESHOLD_3), V_THRESHOLD_3(32U));
838 
839 	t4_set_reg_field(adapter, A_SGE_CONTROL2, V_IDMAARBROUNDROBIN(1U),
840 			 V_IDMAARBROUNDROBIN(1U));
841 
842 	/*
843 	 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
844 	 * adds the pseudo header itself.
845 	 */
846 	t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG,
847 			       F_CSUM_HAS_PSEUDO_HDR, 0);
848 
849 	return 0;
850 }
851 
852 /*
853  * Attempt to initialize the adapter via a Firmware Configuration File.
854  */
855 static int adap_init0_config(struct adapter *adapter, int reset)
856 {
857 	struct fw_caps_config_cmd caps_cmd;
858 	unsigned long mtype = 0, maddr = 0;
859 	u32 finiver, finicsum, cfcsum;
860 	int ret;
861 	int config_issued = 0;
862 	int cfg_addr;
863 	char config_name[20];
864 
865 	/*
866 	 * Reset device if necessary.
867 	 */
868 	if (reset) {
869 		ret = t4_fw_reset(adapter, adapter->mbox,
870 				  F_PIORSTMODE | F_PIORST);
871 		if (ret < 0) {
872 			dev_warn(adapter, "Firmware reset failed, error %d\n",
873 				 -ret);
874 			goto bye;
875 		}
876 	}
877 
878 	cfg_addr = t4_flash_cfg_addr(adapter);
879 	if (cfg_addr < 0) {
880 		ret = cfg_addr;
881 		dev_warn(adapter, "Finding address for firmware config file in flash failed, error %d\n",
882 			 -ret);
883 		goto bye;
884 	}
885 
886 	strcpy(config_name, "On Flash");
887 	mtype = FW_MEMTYPE_CF_FLASH;
888 	maddr = cfg_addr;
889 
890 	/*
891 	 * Issue a Capability Configuration command to the firmware to get it
892 	 * to parse the Configuration File.  We don't use t4_fw_config_file()
893 	 * because we want the ability to modify various features after we've
894 	 * processed the configuration file ...
895 	 */
896 	memset(&caps_cmd, 0, sizeof(caps_cmd));
897 	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
898 					   F_FW_CMD_REQUEST | F_FW_CMD_READ);
899 	caps_cmd.cfvalid_to_len16 =
900 		cpu_to_be32(F_FW_CAPS_CONFIG_CMD_CFVALID |
901 			    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
902 			    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
903 			    FW_LEN16(caps_cmd));
904 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
905 			 &caps_cmd);
906 	/*
907 	 * If the CAPS_CONFIG failed with an ENOENT (for a Firmware
908 	 * Configuration File in FLASH), our last gasp effort is to use the
909 	 * Firmware Configuration File which is embedded in the firmware.  A
910 	 * very few early versions of the firmware didn't have one embedded
911 	 * but we can ignore those.
912 	 */
913 	if (ret == -ENOENT) {
914 		dev_info(adapter, "%s: Going for embedded config in firmware..\n",
915 			 __func__);
916 
917 		memset(&caps_cmd, 0, sizeof(caps_cmd));
918 		caps_cmd.op_to_write =
919 			cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
920 				    F_FW_CMD_REQUEST | F_FW_CMD_READ);
921 		caps_cmd.cfvalid_to_len16 = cpu_to_be32(FW_LEN16(caps_cmd));
922 		ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
923 				 sizeof(caps_cmd), &caps_cmd);
924 		strcpy(config_name, "Firmware Default");
925 	}
926 
927 	config_issued = 1;
928 	if (ret < 0)
929 		goto bye;
930 
931 	finiver = be32_to_cpu(caps_cmd.finiver);
932 	finicsum = be32_to_cpu(caps_cmd.finicsum);
933 	cfcsum = be32_to_cpu(caps_cmd.cfcsum);
934 	if (finicsum != cfcsum)
935 		dev_warn(adapter, "Configuration File checksum mismatch: [fini] csum=%#x, computed csum=%#x\n",
936 			 finicsum, cfcsum);
937 
938 	/*
939 	 * If we're a pure NIC driver then disable all offloading facilities.
940 	 * This will allow the firmware to optimize aspects of the hardware
941 	 * configuration which will result in improved performance.
942 	 */
943 	caps_cmd.niccaps &= cpu_to_be16(~FW_CAPS_CONFIG_NIC_ETHOFLD);
944 	caps_cmd.toecaps = 0;
945 	caps_cmd.iscsicaps = 0;
946 	caps_cmd.rdmacaps = 0;
947 	caps_cmd.fcoecaps = 0;
948 
949 	/*
950 	 * And now tell the firmware to use the configuration we just loaded.
951 	 */
952 	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
953 					   F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
954 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
955 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
956 			 NULL);
957 	if (ret < 0) {
958 		dev_warn(adapter, "Unable to finalize Firmware Capabilities %d\n",
959 			 -ret);
960 		goto bye;
961 	}
962 
963 	/*
964 	 * Tweak configuration based on system architecture, etc.
965 	 */
966 	ret = adap_init0_tweaks(adapter);
967 	if (ret < 0) {
968 		dev_warn(adapter, "Unable to do init0-tweaks %d\n", -ret);
969 		goto bye;
970 	}
971 
972 	/*
973 	 * And finally tell the firmware to initialize itself using the
974 	 * parameters from the Configuration File.
975 	 */
976 	ret = t4_fw_initialize(adapter, adapter->mbox);
977 	if (ret < 0) {
978 		dev_warn(adapter, "Initializing Firmware failed, error %d\n",
979 			 -ret);
980 		goto bye;
981 	}
982 
983 	/*
984 	 * Return successfully and note that we're operating with parameters
985 	 * not supplied by the driver, rather than from hard-wired
986 	 * initialization constants buried in the driver.
987 	 */
988 	dev_info(adapter,
989 		 "Successfully configured using Firmware Configuration File \"%s\", version %#x, computed checksum %#x\n",
990 		 config_name, finiver, cfcsum);
991 
992 	return 0;
993 
994 	/*
995 	 * Something bad happened.  Return the error ...  (If the "error"
996 	 * is that there's no Configuration File on the adapter we don't
997 	 * want to issue a warning since this is fairly common.)
998 	 */
999 bye:
1000 	if (config_issued && ret != -ENOENT)
1001 		dev_warn(adapter, "\"%s\" configuration file error %d\n",
1002 			 config_name, -ret);
1003 
1004 	dev_debug(adapter, "%s: returning ret = %d ..\n", __func__, ret);
1005 	return ret;
1006 }
1007 
1008 static int adap_init0(struct adapter *adap)
1009 {
1010 	struct fw_caps_config_cmd caps_cmd;
1011 	int ret = 0;
1012 	u32 v, port_vec;
1013 	enum dev_state state;
1014 	u32 params[7], val[7];
1015 	int reset = 1;
1016 	int mbox = adap->mbox;
1017 
1018 	/*
1019 	 * Contact FW, advertising Master capability.
1020 	 */
1021 	ret = t4_fw_hello(adap, adap->mbox, adap->mbox, MASTER_MAY, &state);
1022 	if (ret < 0) {
1023 		dev_err(adap, "%s: could not connect to FW, error %d\n",
1024 			__func__, -ret);
1025 		goto bye;
1026 	}
1027 
1028 	CXGBE_DEBUG_MBOX(adap, "%s: adap->mbox = %d; ret = %d\n", __func__,
1029 			 adap->mbox, ret);
1030 
1031 	if (ret == mbox)
1032 		adap->flags |= MASTER_PF;
1033 
1034 	if (state == DEV_STATE_INIT) {
1035 		/*
1036 		 * Force halt and reset FW because a previous instance may have
1037 		 * exited abnormally without properly shutting down
1038 		 */
1039 		ret = t4_fw_halt(adap, adap->mbox, reset);
1040 		if (ret < 0) {
1041 			dev_err(adap, "Failed to halt. Exit.\n");
1042 			goto bye;
1043 		}
1044 
1045 		ret = t4_fw_restart(adap, adap->mbox, reset);
1046 		if (ret < 0) {
1047 			dev_err(adap, "Failed to restart. Exit.\n");
1048 			goto bye;
1049 		}
1050 		state = (enum dev_state)((unsigned)state & ~DEV_STATE_INIT);
1051 	}
1052 
1053 	t4_get_version_info(adap);
1054 
1055 	ret = t4_get_core_clock(adap, &adap->params.vpd);
1056 	if (ret < 0) {
1057 		dev_err(adap, "%s: could not get core clock, error %d\n",
1058 			__func__, -ret);
1059 		goto bye;
1060 	}
1061 
1062 	/*
1063 	 * If the firmware is initialized already (and we're not forcing a
1064 	 * master initialization), note that we're living with existing
1065 	 * adapter parameters.  Otherwise, it's time to try initializing the
1066 	 * adapter ...
1067 	 */
1068 	if (state == DEV_STATE_INIT) {
1069 		dev_info(adap, "Coming up as %s: Adapter already initialized\n",
1070 			 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
1071 	} else {
1072 		dev_info(adap, "Coming up as MASTER: Initializing adapter\n");
1073 
1074 		ret = adap_init0_config(adap, reset);
1075 		if (ret == -ENOENT) {
1076 			dev_err(adap,
1077 				"No Configuration File present on adapter. Using hard-wired configuration parameters.\n");
1078 			goto bye;
1079 		}
1080 	}
1081 	if (ret < 0) {
1082 		dev_err(adap, "could not initialize adapter, error %d\n", -ret);
1083 		goto bye;
1084 	}
1085 
1086 	/* Now that we've successfully configured and initialized the adapter
1087 	 * (or found it already initialized), we can ask the Firmware what
1088 	 * resources it has provisioned for us.
1089 	 */
1090 	ret = t4_get_pfres(adap);
1091 	if (ret) {
1092 		dev_err(adap->pdev_dev,
1093 			"Unable to retrieve resource provisioning info\n");
1094 		goto bye;
1095 	}
1096 
1097 	/* Find out what ports are available to us. */
1098 	v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
1099 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
1100 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
1101 	if (ret < 0) {
1102 		dev_err(adap, "%s: failure in t4_query_params; error = %d\n",
1103 			__func__, ret);
1104 		goto bye;
1105 	}
1106 
1107 	adap->params.nports = hweight32(port_vec);
1108 	adap->params.portvec = port_vec;
1109 
1110 	dev_debug(adap, "%s: adap->params.nports = %u\n", __func__,
1111 		  adap->params.nports);
1112 
1113 	/*
1114 	 * Give the SGE code a chance to pull in anything that it needs ...
1115 	 * Note that this must be called after we retrieve our VPD parameters
1116 	 * in order to know how to convert core ticks to seconds, etc.
1117 	 */
1118 	ret = t4_sge_init(adap);
1119 	if (ret < 0) {
1120 		dev_err(adap, "t4_sge_init failed with error %d\n",
1121 			-ret);
1122 		goto bye;
1123 	}
1124 
1125 	/*
1126 	 * Grab some of our basic fundamental operating parameters.
1127 	 */
1128 #define FW_PARAM_DEV(param) \
1129 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
1130 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
1131 
1132 #define FW_PARAM_PFVF(param) \
1133 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
1134 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param) |  \
1135 	 V_FW_PARAMS_PARAM_Y(0) | \
1136 	 V_FW_PARAMS_PARAM_Z(0))
1137 
1138 	params[0] = FW_PARAM_PFVF(FILTER_START);
1139 	params[1] = FW_PARAM_PFVF(FILTER_END);
1140 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
1141 	if (ret < 0)
1142 		goto bye;
1143 	adap->tids.ftid_base = val[0];
1144 	adap->tids.nftids = val[1] - val[0] + 1;
1145 
1146 	params[0] = FW_PARAM_PFVF(CLIP_START);
1147 	params[1] = FW_PARAM_PFVF(CLIP_END);
1148 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
1149 	if (ret < 0)
1150 		goto bye;
1151 	adap->clipt_start = val[0];
1152 	adap->clipt_end = val[1];
1153 
1154 	/*
1155 	 * Get device capabilities so we can determine what resources we need
1156 	 * to manage.
1157 	 */
1158 	memset(&caps_cmd, 0, sizeof(caps_cmd));
1159 	caps_cmd.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1160 				     F_FW_CMD_REQUEST | F_FW_CMD_READ);
1161 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
1162 	ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
1163 			 &caps_cmd);
1164 	if (ret < 0)
1165 		goto bye;
1166 
1167 	if ((caps_cmd.niccaps & cpu_to_be16(FW_CAPS_CONFIG_NIC_HASHFILTER)) &&
1168 	    is_t6(adap->params.chip)) {
1169 		if (init_hash_filter(adap) < 0)
1170 			goto bye;
1171 	}
1172 
1173 	/* query tid-related parameters */
1174 	params[0] = FW_PARAM_DEV(NTID);
1175 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
1176 			      params, val);
1177 	if (ret < 0)
1178 		goto bye;
1179 	adap->tids.ntids = val[0];
1180 	adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
1181 
1182 	/* If we're running on newer firmware, let it know that we're
1183 	 * prepared to deal with encapsulated CPL messages.  Older
1184 	 * firmware won't understand this and we'll just get
1185 	 * unencapsulated messages ...
1186 	 */
1187 	params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
1188 	val[0] = 1;
1189 	(void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1190 
1191 	/*
1192 	 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
1193 	 * capability.  Earlier versions of the firmware didn't have the
1194 	 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
1195 	 * permission to use ULPTX MEMWRITE DSGL.
1196 	 */
1197 	if (is_t4(adap->params.chip)) {
1198 		adap->params.ulptx_memwrite_dsgl = false;
1199 	} else {
1200 		params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
1201 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1202 				      1, params, val);
1203 		adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
1204 	}
1205 
1206 	/*
1207 	 * The MTU/MSS Table is initialized by now, so load their values.  If
1208 	 * we're initializing the adapter, then we'll make any modifications
1209 	 * we want to the MTU/MSS Table and also initialize the congestion
1210 	 * parameters.
1211 	 */
1212 	t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
1213 	if (state != DEV_STATE_INIT) {
1214 		int i;
1215 
1216 		/*
1217 		 * The default MTU Table contains values 1492 and 1500.
1218 		 * However, for TCP, it's better to have two values which are
1219 		 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
1220 		 * This allows us to have a TCP Data Payload which is a
1221 		 * multiple of 8 regardless of what combination of TCP Options
1222 		 * are in use (always a multiple of 4 bytes) which is
1223 		 * important for performance reasons.  For instance, if no
1224 		 * options are in use, then we have a 20-byte IP header and a
1225 		 * 20-byte TCP header.  In this case, a 1500-byte MSS would
1226 		 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
1227 		 * which is not a multiple of 8.  So using an MSS of 1488 in
1228 		 * this case results in a TCP Data Payload of 1448 bytes which
1229 		 * is a multiple of 8.  On the other hand, if 12-byte TCP Time
1230 		 * Stamps have been negotiated, then an MTU of 1500 bytes
1231 		 * results in a TCP Data Payload of 1448 bytes which, as
1232 		 * above, is a multiple of 8 bytes ...
1233 		 */
1234 		for (i = 0; i < NMTUS; i++)
1235 			if (adap->params.mtus[i] == 1492) {
1236 				adap->params.mtus[i] = 1488;
1237 				break;
1238 			}
1239 
1240 		t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
1241 			     adap->params.b_wnd);
1242 	}
1243 	t4_init_sge_params(adap);
1244 	t4_init_tp_params(adap);
1245 	configure_pcie_ext_tag(adap);
1246 	configure_vlan_types(adap);
1247 	configure_max_ethqsets(adap);
1248 
1249 	adap->params.drv_memwin = MEMWIN_NIC;
1250 	adap->flags |= FW_OK;
1251 	dev_debug(adap, "%s: returning zero..\n", __func__);
1252 	return 0;
1253 
1254 	/*
1255 	 * Something bad happened.  If a command timed out or failed with EIO
1256 	 * FW does not operate within its spec or something catastrophic
1257 	 * happened to HW/FW, stop issuing commands.
1258 	 */
1259 bye:
1260 	if (ret != -ETIMEDOUT && ret != -EIO)
1261 		t4_fw_bye(adap, adap->mbox);
1262 	return ret;
1263 }
1264 
1265 /**
1266  * t4_os_portmod_changed - handle port module changes
1267  * @adap: the adapter associated with the module change
1268  * @port_id: the port index whose module status has changed
1269  *
1270  * This is the OS-dependent handler for port module changes.  It is
1271  * invoked when a port module is removed or inserted for any OS-specific
1272  * processing.
1273  */
1274 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
1275 {
1276 	static const char * const mod_str[] = {
1277 		NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
1278 	};
1279 
1280 	const struct port_info *pi = adap2pinfo(adap, port_id);
1281 
1282 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
1283 		dev_info(adap, "Port%d: port module unplugged\n", pi->port_id);
1284 	else if (pi->mod_type < ARRAY_SIZE(mod_str))
1285 		dev_info(adap, "Port%d: %s port module inserted\n", pi->port_id,
1286 			 mod_str[pi->mod_type]);
1287 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
1288 		dev_info(adap, "Port%d: unsupported port module inserted\n",
1289 			 pi->port_id);
1290 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
1291 		dev_info(adap, "Port%d: unknown port module inserted\n",
1292 			 pi->port_id);
1293 	else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
1294 		dev_info(adap, "Port%d: transceiver module error\n",
1295 			 pi->port_id);
1296 	else
1297 		dev_info(adap, "Port%d: unknown module type %d inserted\n",
1298 			 pi->port_id, pi->mod_type);
1299 }
1300 
1301 inline bool force_linkup(struct adapter *adap)
1302 {
1303 	struct rte_pci_device *pdev = adap->pdev;
1304 
1305 	if (is_pf4(adap))
1306 		return false;	/* force_linkup not required for pf driver*/
1307 	if (!cxgbe_get_devargs(pdev->device.devargs,
1308 			       CXGBE_DEVARG_FORCE_LINK_UP))
1309 		return false;
1310 	return true;
1311 }
1312 
1313 /**
1314  * link_start - enable a port
1315  * @dev: the port to enable
1316  *
1317  * Performs the MAC and PHY actions needed to enable a port.
1318  */
1319 int link_start(struct port_info *pi)
1320 {
1321 	struct adapter *adapter = pi->adapter;
1322 	int ret;
1323 	unsigned int mtu;
1324 
1325 	mtu = pi->eth_dev->data->dev_conf.rxmode.max_rx_pkt_len -
1326 	      (ETHER_HDR_LEN + ETHER_CRC_LEN);
1327 
1328 	/*
1329 	 * We do not set address filters and promiscuity here, the stack does
1330 	 * that step explicitly.
1331 	 */
1332 	ret = t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu, -1, -1,
1333 			    -1, 1, true);
1334 	if (ret == 0) {
1335 		ret = t4_change_mac(adapter, adapter->mbox, pi->viid,
1336 				    pi->xact_addr_filt,
1337 				    (u8 *)&pi->eth_dev->data->mac_addrs[0],
1338 				    true, true);
1339 		if (ret >= 0) {
1340 			pi->xact_addr_filt = ret;
1341 			ret = 0;
1342 		}
1343 	}
1344 	if (ret == 0 && is_pf4(adapter))
1345 		ret = t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
1346 				    &pi->link_cfg);
1347 	if (ret == 0) {
1348 		/*
1349 		 * Enabling a Virtual Interface can result in an interrupt
1350 		 * during the processing of the VI Enable command and, in some
1351 		 * paths, result in an attempt to issue another command in the
1352 		 * interrupt context.  Thus, we disable interrupts during the
1353 		 * course of the VI Enable command ...
1354 		 */
1355 		ret = t4_enable_vi_params(adapter, adapter->mbox, pi->viid,
1356 					  true, true, false);
1357 	}
1358 
1359 	if (ret == 0 && force_linkup(adapter))
1360 		pi->eth_dev->data->dev_link.link_status = ETH_LINK_UP;
1361 	return ret;
1362 }
1363 
1364 /**
1365  * cxgbe_write_rss_conf - flash the RSS configuration for a given port
1366  * @pi: the port
1367  * @rss_hf: Hash configuration to apply
1368  */
1369 int cxgbe_write_rss_conf(const struct port_info *pi, uint64_t rss_hf)
1370 {
1371 	struct adapter *adapter = pi->adapter;
1372 	const struct sge_eth_rxq *rxq;
1373 	u64 flags = 0;
1374 	u16 rss;
1375 	int err;
1376 
1377 	/*  Should never be called before setting up sge eth rx queues */
1378 	if (!(adapter->flags & FULL_INIT_DONE)) {
1379 		dev_err(adap, "%s No RXQs available on port %d\n",
1380 			__func__, pi->port_id);
1381 		return -EINVAL;
1382 	}
1383 
1384 	/* Don't allow unsupported hash functions */
1385 	if (rss_hf & ~CXGBE_RSS_HF_ALL)
1386 		return -EINVAL;
1387 
1388 	if (rss_hf & CXGBE_RSS_HF_IPV4_MASK)
1389 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
1390 
1391 	if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
1392 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
1393 
1394 	if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
1395 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
1396 			 F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1397 
1398 	if (rss_hf & CXGBE_RSS_HF_IPV6_MASK)
1399 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
1400 
1401 	if (rss_hf & CXGBE_RSS_HF_TCP_IPV6_MASK)
1402 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1403 			 F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
1404 
1405 	if (rss_hf & CXGBE_RSS_HF_UDP_IPV6_MASK)
1406 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1407 			 F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
1408 			 F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1409 
1410 	rxq = &adapter->sge.ethrxq[pi->first_qset];
1411 	rss = rxq[0].rspq.abs_id;
1412 
1413 	/* If Tunnel All Lookup isn't specified in the global RSS
1414 	 * Configuration, then we need to specify a default Ingress
1415 	 * Queue for any ingress packets which aren't hashed.  We'll
1416 	 * use our first ingress queue ...
1417 	 */
1418 	err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
1419 			       flags, rss);
1420 	return err;
1421 }
1422 
1423 /**
1424  * cxgbe_write_rss - write the RSS table for a given port
1425  * @pi: the port
1426  * @queues: array of queue indices for RSS
1427  *
1428  * Sets up the portion of the HW RSS table for the port's VI to distribute
1429  * packets to the Rx queues in @queues.
1430  */
1431 int cxgbe_write_rss(const struct port_info *pi, const u16 *queues)
1432 {
1433 	u16 *rss;
1434 	int i, err;
1435 	struct adapter *adapter = pi->adapter;
1436 	const struct sge_eth_rxq *rxq;
1437 
1438 	/*  Should never be called before setting up sge eth rx queues */
1439 	BUG_ON(!(adapter->flags & FULL_INIT_DONE));
1440 
1441 	rxq = &adapter->sge.ethrxq[pi->first_qset];
1442 	rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
1443 	if (!rss)
1444 		return -ENOMEM;
1445 
1446 	/* map the queue indices to queue ids */
1447 	for (i = 0; i < pi->rss_size; i++, queues++)
1448 		rss[i] = rxq[*queues].rspq.abs_id;
1449 
1450 	err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
1451 				  pi->rss_size, rss, pi->rss_size);
1452 	rte_free(rss);
1453 	return err;
1454 }
1455 
1456 /**
1457  * setup_rss - configure RSS
1458  * @adapter: the adapter
1459  *
1460  * Sets up RSS to distribute packets to multiple receive queues.  We
1461  * configure the RSS CPU lookup table to distribute to the number of HW
1462  * receive queues, and the response queue lookup table to narrow that
1463  * down to the response queues actually configured for each port.
1464  * We always configure the RSS mapping for all ports since the mapping
1465  * table has plenty of entries.
1466  */
1467 int setup_rss(struct port_info *pi)
1468 {
1469 	int j, err;
1470 	struct adapter *adapter = pi->adapter;
1471 
1472 	dev_debug(adapter, "%s:  pi->rss_size = %u; pi->n_rx_qsets = %u\n",
1473 		  __func__, pi->rss_size, pi->n_rx_qsets);
1474 
1475 	if (!(pi->flags & PORT_RSS_DONE)) {
1476 		if (adapter->flags & FULL_INIT_DONE) {
1477 			/* Fill default values with equal distribution */
1478 			for (j = 0; j < pi->rss_size; j++)
1479 				pi->rss[j] = j % pi->n_rx_qsets;
1480 
1481 			err = cxgbe_write_rss(pi, pi->rss);
1482 			if (err)
1483 				return err;
1484 
1485 			err = cxgbe_write_rss_conf(pi, pi->rss_hf);
1486 			if (err)
1487 				return err;
1488 			pi->flags |= PORT_RSS_DONE;
1489 		}
1490 	}
1491 	return 0;
1492 }
1493 
1494 /*
1495  * Enable NAPI scheduling and interrupt generation for all Rx queues.
1496  */
1497 static void enable_rx(struct adapter *adap, struct sge_rspq *q)
1498 {
1499 	/* 0-increment GTS to start the timer and enable interrupts */
1500 	t4_write_reg(adap, is_pf4(adap) ? MYPF_REG(A_SGE_PF_GTS) :
1501 					  T4VF_SGE_BASE_ADDR + A_SGE_VF_GTS,
1502 		     V_SEINTARM(q->intr_params) |
1503 		     V_INGRESSQID(q->cntxt_id));
1504 }
1505 
1506 void cxgbe_enable_rx_queues(struct port_info *pi)
1507 {
1508 	struct adapter *adap = pi->adapter;
1509 	struct sge *s = &adap->sge;
1510 	unsigned int i;
1511 
1512 	for (i = 0; i < pi->n_rx_qsets; i++)
1513 		enable_rx(adap, &s->ethrxq[pi->first_qset + i].rspq);
1514 }
1515 
1516 /**
1517  * fw_caps_to_speed_caps - translate Firmware Port Caps to Speed Caps.
1518  * @port_type: Firmware Port Type
1519  * @fw_caps: Firmware Port Capabilities
1520  * @speed_caps: Device Info Speed Capabilities
1521  *
1522  * Translate a Firmware Port Capabilities specification to Device Info
1523  * Speed Capabilities.
1524  */
1525 static void fw_caps_to_speed_caps(enum fw_port_type port_type,
1526 				  unsigned int fw_caps,
1527 				  u32 *speed_caps)
1528 {
1529 #define SET_SPEED(__speed_name) \
1530 	do { \
1531 		*speed_caps |= ETH_LINK_ ## __speed_name; \
1532 	} while (0)
1533 
1534 #define FW_CAPS_TO_SPEED(__fw_name) \
1535 	do { \
1536 		if (fw_caps & FW_PORT_CAP32_ ## __fw_name) \
1537 			SET_SPEED(__fw_name); \
1538 	} while (0)
1539 
1540 	switch (port_type) {
1541 	case FW_PORT_TYPE_BT_SGMII:
1542 	case FW_PORT_TYPE_BT_XFI:
1543 	case FW_PORT_TYPE_BT_XAUI:
1544 		FW_CAPS_TO_SPEED(SPEED_100M);
1545 		FW_CAPS_TO_SPEED(SPEED_1G);
1546 		FW_CAPS_TO_SPEED(SPEED_10G);
1547 		break;
1548 
1549 	case FW_PORT_TYPE_KX4:
1550 	case FW_PORT_TYPE_KX:
1551 	case FW_PORT_TYPE_FIBER_XFI:
1552 	case FW_PORT_TYPE_FIBER_XAUI:
1553 	case FW_PORT_TYPE_SFP:
1554 	case FW_PORT_TYPE_QSFP_10G:
1555 	case FW_PORT_TYPE_QSA:
1556 		FW_CAPS_TO_SPEED(SPEED_1G);
1557 		FW_CAPS_TO_SPEED(SPEED_10G);
1558 		break;
1559 
1560 	case FW_PORT_TYPE_KR:
1561 		SET_SPEED(SPEED_10G);
1562 		break;
1563 
1564 	case FW_PORT_TYPE_BP_AP:
1565 	case FW_PORT_TYPE_BP4_AP:
1566 		SET_SPEED(SPEED_1G);
1567 		SET_SPEED(SPEED_10G);
1568 		break;
1569 
1570 	case FW_PORT_TYPE_BP40_BA:
1571 	case FW_PORT_TYPE_QSFP:
1572 		SET_SPEED(SPEED_40G);
1573 		break;
1574 
1575 	case FW_PORT_TYPE_CR_QSFP:
1576 	case FW_PORT_TYPE_SFP28:
1577 	case FW_PORT_TYPE_KR_SFP28:
1578 		FW_CAPS_TO_SPEED(SPEED_1G);
1579 		FW_CAPS_TO_SPEED(SPEED_10G);
1580 		FW_CAPS_TO_SPEED(SPEED_25G);
1581 		break;
1582 
1583 	case FW_PORT_TYPE_CR2_QSFP:
1584 		SET_SPEED(SPEED_50G);
1585 		break;
1586 
1587 	case FW_PORT_TYPE_KR4_100G:
1588 	case FW_PORT_TYPE_CR4_QSFP:
1589 		FW_CAPS_TO_SPEED(SPEED_25G);
1590 		FW_CAPS_TO_SPEED(SPEED_40G);
1591 		FW_CAPS_TO_SPEED(SPEED_50G);
1592 		FW_CAPS_TO_SPEED(SPEED_100G);
1593 		break;
1594 
1595 	default:
1596 		break;
1597 	}
1598 
1599 #undef FW_CAPS_TO_SPEED
1600 #undef SET_SPEED
1601 }
1602 
1603 /**
1604  * cxgbe_get_speed_caps - Fetch supported speed capabilities
1605  * @pi: Underlying port's info
1606  * @speed_caps: Device Info speed capabilities
1607  *
1608  * Fetch supported speed capabilities of the underlying port.
1609  */
1610 void cxgbe_get_speed_caps(struct port_info *pi, u32 *speed_caps)
1611 {
1612 	*speed_caps = 0;
1613 
1614 	fw_caps_to_speed_caps(pi->port_type, pi->link_cfg.pcaps,
1615 			      speed_caps);
1616 
1617 	if (!(pi->link_cfg.pcaps & FW_PORT_CAP32_ANEG))
1618 		*speed_caps |= ETH_LINK_SPEED_FIXED;
1619 }
1620 
1621 /**
1622  * cxgbe_set_link_status - Set device link up or down.
1623  * @pi: Underlying port's info
1624  * @status: 0 - down, 1 - up
1625  *
1626  * Set the device link up or down.
1627  */
1628 int cxgbe_set_link_status(struct port_info *pi, bool status)
1629 {
1630 	struct adapter *adapter = pi->adapter;
1631 	int err = 0;
1632 
1633 	err = t4_enable_vi(adapter, adapter->mbox, pi->viid, status, status);
1634 	if (err) {
1635 		dev_err(adapter, "%s: disable_vi failed: %d\n", __func__, err);
1636 		return err;
1637 	}
1638 
1639 	if (!status)
1640 		t4_reset_link_config(adapter, pi->pidx);
1641 
1642 	return 0;
1643 }
1644 
1645 /**
1646  * cxgb_up - enable the adapter
1647  * @adap: adapter being enabled
1648  *
1649  * Called when the first port is enabled, this function performs the
1650  * actions necessary to make an adapter operational, such as completing
1651  * the initialization of HW modules, and enabling interrupts.
1652  */
1653 int cxgbe_up(struct adapter *adap)
1654 {
1655 	enable_rx(adap, &adap->sge.fw_evtq);
1656 	t4_sge_tx_monitor_start(adap);
1657 	if (is_pf4(adap))
1658 		t4_intr_enable(adap);
1659 	adap->flags |= FULL_INIT_DONE;
1660 
1661 	/* TODO: deadman watchdog ?? */
1662 	return 0;
1663 }
1664 
1665 /*
1666  * Close the port
1667  */
1668 int cxgbe_down(struct port_info *pi)
1669 {
1670 	return cxgbe_set_link_status(pi, false);
1671 }
1672 
1673 /*
1674  * Release resources when all the ports have been stopped.
1675  */
1676 void cxgbe_close(struct adapter *adapter)
1677 {
1678 	struct port_info *pi;
1679 	int i;
1680 
1681 	if (adapter->flags & FULL_INIT_DONE) {
1682 		if (is_pf4(adapter))
1683 			t4_intr_disable(adapter);
1684 		tid_free(&adapter->tids);
1685 		t4_cleanup_clip_tbl(adapter);
1686 		t4_sge_tx_monitor_stop(adapter);
1687 		t4_free_sge_resources(adapter);
1688 		for_each_port(adapter, i) {
1689 			pi = adap2pinfo(adapter, i);
1690 			if (pi->viid != 0)
1691 				t4_free_vi(adapter, adapter->mbox,
1692 					   adapter->pf, 0, pi->viid);
1693 			rte_free(pi->eth_dev->data->mac_addrs);
1694 			/* Skip first port since it'll be freed by DPDK stack */
1695 			if (i) {
1696 				rte_free(pi->eth_dev->data->dev_private);
1697 				rte_eth_dev_release_port(pi->eth_dev);
1698 			}
1699 		}
1700 		adapter->flags &= ~FULL_INIT_DONE;
1701 	}
1702 
1703 	if (is_pf4(adapter) && (adapter->flags & FW_OK))
1704 		t4_fw_bye(adapter, adapter->mbox);
1705 }
1706 
1707 int cxgbe_probe(struct adapter *adapter)
1708 {
1709 	struct port_info *pi;
1710 	int chip;
1711 	int func, i;
1712 	int err = 0;
1713 	u32 whoami;
1714 
1715 	whoami = t4_read_reg(adapter, A_PL_WHOAMI);
1716 	chip = t4_get_chip_type(adapter,
1717 			CHELSIO_PCI_ID_VER(adapter->pdev->id.device_id));
1718 	if (chip < 0)
1719 		return chip;
1720 
1721 	func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
1722 	       G_SOURCEPF(whoami) : G_T6_SOURCEPF(whoami);
1723 
1724 	adapter->mbox = func;
1725 	adapter->pf = func;
1726 
1727 	t4_os_lock_init(&adapter->mbox_lock);
1728 	TAILQ_INIT(&adapter->mbox_list);
1729 	t4_os_lock_init(&adapter->win0_lock);
1730 
1731 	err = t4_prep_adapter(adapter);
1732 	if (err)
1733 		return err;
1734 
1735 	setup_memwin(adapter);
1736 	err = adap_init0(adapter);
1737 	if (err) {
1738 		dev_err(adapter, "%s: Adapter initialization failed, error %d\n",
1739 			__func__, err);
1740 		goto out_free;
1741 	}
1742 
1743 	if (!is_t4(adapter->params.chip)) {
1744 		/*
1745 		 * The userspace doorbell BAR is split evenly into doorbell
1746 		 * regions, each associated with an egress queue.  If this
1747 		 * per-queue region is large enough (at least UDBS_SEG_SIZE)
1748 		 * then it can be used to submit a tx work request with an
1749 		 * implied doorbell.  Enable write combining on the BAR if
1750 		 * there is room for such work requests.
1751 		 */
1752 		int s_qpp, qpp, num_seg;
1753 
1754 		s_qpp = (S_QUEUESPERPAGEPF0 +
1755 			(S_QUEUESPERPAGEPF1 - S_QUEUESPERPAGEPF0) *
1756 			adapter->pf);
1757 		qpp = 1 << ((t4_read_reg(adapter,
1758 				A_SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp)
1759 				& M_QUEUESPERPAGEPF0);
1760 		num_seg = CXGBE_PAGE_SIZE / UDBS_SEG_SIZE;
1761 		if (qpp > num_seg)
1762 			dev_warn(adapter, "Incorrect SGE EGRESS QUEUES_PER_PAGE configuration, continuing in debug mode\n");
1763 
1764 		adapter->bar2 = (void *)adapter->pdev->mem_resource[2].addr;
1765 		if (!adapter->bar2) {
1766 			dev_err(adapter, "cannot map device bar2 region\n");
1767 			err = -ENOMEM;
1768 			goto out_free;
1769 		}
1770 		t4_write_reg(adapter, A_SGE_STAT_CFG, V_STATSOURCE_T5(7) |
1771 			     V_STATMODE(0));
1772 	}
1773 
1774 	for_each_port(adapter, i) {
1775 		const unsigned int numa_node = rte_socket_id();
1776 		char name[RTE_ETH_NAME_MAX_LEN];
1777 		struct rte_eth_dev *eth_dev;
1778 
1779 		snprintf(name, sizeof(name), "%s_%d",
1780 			 adapter->pdev->device.name, i);
1781 
1782 		if (i == 0) {
1783 			/* First port is already allocated by DPDK */
1784 			eth_dev = adapter->eth_dev;
1785 			goto allocate_mac;
1786 		}
1787 
1788 		/*
1789 		 * now do all data allocation - for eth_dev structure,
1790 		 * and internal (private) data for the remaining ports
1791 		 */
1792 
1793 		/* reserve an ethdev entry */
1794 		eth_dev = rte_eth_dev_allocate(name);
1795 		if (!eth_dev)
1796 			goto out_free;
1797 
1798 		eth_dev->data->dev_private =
1799 			rte_zmalloc_socket(name, sizeof(struct port_info),
1800 					   RTE_CACHE_LINE_SIZE, numa_node);
1801 		if (!eth_dev->data->dev_private)
1802 			goto out_free;
1803 
1804 allocate_mac:
1805 		pi = (struct port_info *)eth_dev->data->dev_private;
1806 		adapter->port[i] = pi;
1807 		pi->eth_dev = eth_dev;
1808 		pi->adapter = adapter;
1809 		pi->xact_addr_filt = -1;
1810 		pi->port_id = i;
1811 		pi->pidx = i;
1812 
1813 		pi->eth_dev->device = &adapter->pdev->device;
1814 		pi->eth_dev->dev_ops = adapter->eth_dev->dev_ops;
1815 		pi->eth_dev->tx_pkt_burst = adapter->eth_dev->tx_pkt_burst;
1816 		pi->eth_dev->rx_pkt_burst = adapter->eth_dev->rx_pkt_burst;
1817 
1818 		rte_eth_copy_pci_info(pi->eth_dev, adapter->pdev);
1819 
1820 		pi->eth_dev->data->mac_addrs = rte_zmalloc(name,
1821 							   ETHER_ADDR_LEN, 0);
1822 		if (!pi->eth_dev->data->mac_addrs) {
1823 			dev_err(adapter, "%s: Mem allocation failed for storing mac addr, aborting\n",
1824 				__func__);
1825 			err = -1;
1826 			goto out_free;
1827 		}
1828 
1829 		if (i > 0) {
1830 			/* First port will be notified by upper layer */
1831 			rte_eth_dev_probing_finish(eth_dev);
1832 		}
1833 	}
1834 
1835 	if (adapter->flags & FW_OK) {
1836 		err = t4_port_init(adapter, adapter->mbox, adapter->pf, 0);
1837 		if (err) {
1838 			dev_err(adapter, "%s: t4_port_init failed with err %d\n",
1839 				__func__, err);
1840 			goto out_free;
1841 		}
1842 	}
1843 
1844 	cfg_queues(adapter->eth_dev);
1845 
1846 	print_adapter_info(adapter);
1847 	print_port_info(adapter);
1848 
1849 	adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
1850 					  adapter->clipt_end);
1851 	if (!adapter->clipt) {
1852 		/* We tolerate a lack of clip_table, giving up some
1853 		 * functionality
1854 		 */
1855 		dev_warn(adapter, "could not allocate CLIP. Continuing\n");
1856 	}
1857 
1858 	if (tid_init(&adapter->tids) < 0) {
1859 		/* Disable filtering support */
1860 		dev_warn(adapter, "could not allocate TID table, "
1861 			 "filter support disabled. Continuing\n");
1862 	}
1863 
1864 	if (is_hashfilter(adapter)) {
1865 		if (t4_read_reg(adapter, A_LE_DB_CONFIG) & F_HASHEN) {
1866 			u32 hash_base, hash_reg;
1867 
1868 			hash_reg = A_LE_DB_TID_HASHBASE;
1869 			hash_base = t4_read_reg(adapter, hash_reg);
1870 			adapter->tids.hash_base = hash_base / 4;
1871 		}
1872 	} else {
1873 		/* Disable hash filtering support */
1874 		dev_warn(adapter,
1875 			 "Maskless filter support disabled. Continuing\n");
1876 	}
1877 
1878 	err = init_rss(adapter);
1879 	if (err)
1880 		goto out_free;
1881 
1882 	return 0;
1883 
1884 out_free:
1885 	for_each_port(adapter, i) {
1886 		pi = adap2pinfo(adapter, i);
1887 		if (pi->viid != 0)
1888 			t4_free_vi(adapter, adapter->mbox, adapter->pf,
1889 				   0, pi->viid);
1890 		/* Skip first port since it'll be de-allocated by DPDK */
1891 		if (i == 0)
1892 			continue;
1893 		if (pi->eth_dev) {
1894 			if (pi->eth_dev->data->dev_private)
1895 				rte_free(pi->eth_dev->data->dev_private);
1896 			rte_eth_dev_release_port(pi->eth_dev);
1897 		}
1898 	}
1899 
1900 	if (adapter->flags & FW_OK)
1901 		t4_fw_bye(adapter, adapter->mbox);
1902 	return -err;
1903 }
1904