xref: /dpdk/drivers/net/cxgbe/cxgbe_main.c (revision a99564c680dd33d1dc4931985fd769c86e5791e5)
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 "base/common.h"
37 #include "base/t4_regs.h"
38 #include "base/t4_msg.h"
39 #include "cxgbe.h"
40 #include "cxgbe_pfvf.h"
41 #include "clip_tbl.h"
42 #include "l2t.h"
43 #include "smt.h"
44 #include "mps_tcam.h"
45 
46 /**
47  * Allocate a chunk of memory. The allocated memory is cleared.
48  */
49 void *t4_alloc_mem(size_t size)
50 {
51 	return rte_zmalloc(NULL, size, 0);
52 }
53 
54 /**
55  * Free memory allocated through t4_alloc_mem().
56  */
57 void t4_free_mem(void *addr)
58 {
59 	rte_free(addr);
60 }
61 
62 /*
63  * Response queue handler for the FW event queue.
64  */
65 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
66 			  __rte_unused const struct pkt_gl *gl)
67 {
68 	u8 opcode = ((const struct rss_header *)rsp)->opcode;
69 
70 	rsp++;                                          /* skip RSS header */
71 
72 	/*
73 	 * FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
74 	 */
75 	if (unlikely(opcode == CPL_FW4_MSG &&
76 		     ((const struct cpl_fw4_msg *)rsp)->type ==
77 		      FW_TYPE_RSSCPL)) {
78 		rsp++;
79 		opcode = ((const struct rss_header *)rsp)->opcode;
80 		rsp++;
81 		if (opcode != CPL_SGE_EGR_UPDATE) {
82 			dev_err(q->adapter, "unexpected FW4/CPL %#x on FW event queue\n",
83 				opcode);
84 			goto out;
85 		}
86 	}
87 
88 	if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
89 		/* do nothing */
90 	} else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
91 		const struct cpl_fw6_msg *msg = (const void *)rsp;
92 
93 		t4_handle_fw_rpl(q->adapter, msg->data);
94 	} else if (opcode == CPL_ABORT_RPL_RSS) {
95 		const struct cpl_abort_rpl_rss *p = (const void *)rsp;
96 
97 		cxgbe_hash_del_filter_rpl(q->adapter, p);
98 	} else if (opcode == CPL_SET_TCB_RPL) {
99 		const struct cpl_set_tcb_rpl *p = (const void *)rsp;
100 
101 		cxgbe_filter_rpl(q->adapter, p);
102 	} else if (opcode == CPL_ACT_OPEN_RPL) {
103 		const struct cpl_act_open_rpl *p = (const void *)rsp;
104 
105 		cxgbe_hash_filter_rpl(q->adapter, p);
106 	} else if (opcode == CPL_L2T_WRITE_RPL) {
107 		const struct cpl_l2t_write_rpl *p = (const void *)rsp;
108 
109 		cxgbe_do_l2t_write_rpl(q->adapter, p);
110 	} else {
111 		dev_err(adapter, "unexpected CPL %#x on FW event queue\n",
112 			opcode);
113 	}
114 out:
115 	return 0;
116 }
117 
118 /**
119  * Setup sge control queues to pass control information.
120  */
121 int cxgbe_setup_sge_ctrl_txq(struct adapter *adapter)
122 {
123 	struct sge *s = &adapter->sge;
124 	int err = 0, i = 0;
125 
126 	for_each_port(adapter, i) {
127 		struct port_info *pi = adap2pinfo(adapter, i);
128 		char name[RTE_ETH_NAME_MAX_LEN];
129 		struct sge_ctrl_txq *q = &s->ctrlq[i];
130 
131 		q->q.size = 1024;
132 		err = t4_sge_alloc_ctrl_txq(adapter, q,
133 					    adapter->eth_dev,  i,
134 					    s->fw_evtq.cntxt_id,
135 					    rte_socket_id());
136 		if (err) {
137 			dev_err(adapter, "Failed to alloc ctrl txq. Err: %d",
138 				err);
139 			goto out;
140 		}
141 		snprintf(name, sizeof(name), "%s_ctrl_pool_%d",
142 			 pi->eth_dev->device->driver->name,
143 			 pi->eth_dev->data->port_id);
144 		q->mb_pool = rte_pktmbuf_pool_create(name, s->ctrlq[i].q.size,
145 						     RTE_CACHE_LINE_SIZE,
146 						     RTE_MBUF_PRIV_ALIGN,
147 						     RTE_MBUF_DEFAULT_BUF_SIZE,
148 						     SOCKET_ID_ANY);
149 		if (!q->mb_pool) {
150 			err = -rte_errno;
151 			dev_err(adapter,
152 				"Can't create ctrl pool for port %d. Err: %d\n",
153 				pi->eth_dev->data->port_id, err);
154 			goto out;
155 		}
156 	}
157 	return 0;
158 out:
159 	t4_free_sge_resources(adapter);
160 	return err;
161 }
162 
163 /**
164  * cxgbe_poll_for_completion: Poll rxq for completion
165  * @q: rxq to poll
166  * @ms: milliseconds to delay
167  * @cnt: number of times to poll
168  * @c: completion to check for 'done' status
169  *
170  * Polls the rxq for reples until completion is done or the count
171  * expires.
172  */
173 int cxgbe_poll_for_completion(struct sge_rspq *q, unsigned int ms,
174 			      unsigned int cnt, struct t4_completion *c)
175 {
176 	unsigned int i;
177 	unsigned int work_done, budget = 32;
178 
179 	if (!c)
180 		return -EINVAL;
181 
182 	for (i = 0; i < cnt; i++) {
183 		cxgbe_poll(q, NULL, budget, &work_done);
184 		t4_os_lock(&c->lock);
185 		if (c->done) {
186 			t4_os_unlock(&c->lock);
187 			return 0;
188 		}
189 		t4_os_unlock(&c->lock);
190 		rte_delay_ms(ms);
191 	}
192 	return -ETIMEDOUT;
193 }
194 
195 int cxgbe_setup_sge_fwevtq(struct adapter *adapter)
196 {
197 	struct sge *s = &adapter->sge;
198 	int err = 0;
199 	int msi_idx = 0;
200 
201 	err = t4_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->eth_dev,
202 			       msi_idx, NULL, fwevtq_handler, -1, NULL, 0,
203 			       rte_socket_id());
204 	return err;
205 }
206 
207 static int closest_timer(const struct sge *s, int time)
208 {
209 	unsigned int i, match = 0;
210 	int delta, min_delta = INT_MAX;
211 
212 	for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
213 		delta = time - s->timer_val[i];
214 		if (delta < 0)
215 			delta = -delta;
216 		if (delta < min_delta) {
217 			min_delta = delta;
218 			match = i;
219 		}
220 	}
221 	return match;
222 }
223 
224 static int closest_thres(const struct sge *s, int thres)
225 {
226 	unsigned int i, match = 0;
227 	int delta, min_delta = INT_MAX;
228 
229 	for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
230 		delta = thres - s->counter_val[i];
231 		if (delta < 0)
232 			delta = -delta;
233 		if (delta < min_delta) {
234 			min_delta = delta;
235 			match = i;
236 		}
237 	}
238 	return match;
239 }
240 
241 /**
242  * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
243  * @q: the Rx queue
244  * @us: the hold-off time in us, or 0 to disable timer
245  * @cnt: the hold-off packet count, or 0 to disable counter
246  *
247  * Sets an Rx queue's interrupt hold-off time and packet count.  At least
248  * one of the two needs to be enabled for the queue to generate interrupts.
249  */
250 int cxgb4_set_rspq_intr_params(struct sge_rspq *q, unsigned int us,
251 			       unsigned int cnt)
252 {
253 	struct adapter *adap = q->adapter;
254 	unsigned int timer_val;
255 
256 	if (cnt) {
257 		int err;
258 		u32 v, new_idx;
259 
260 		new_idx = closest_thres(&adap->sge, cnt);
261 		if (q->desc && q->pktcnt_idx != new_idx) {
262 			/* the queue has already been created, update it */
263 			v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
264 			    V_FW_PARAMS_PARAM_X(
265 			    FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
266 			    V_FW_PARAMS_PARAM_YZ(q->cntxt_id);
267 			err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
268 					    &v, &new_idx);
269 			if (err)
270 				return err;
271 		}
272 		q->pktcnt_idx = new_idx;
273 	}
274 
275 	timer_val = (us == 0) ? X_TIMERREG_RESTART_COUNTER :
276 				closest_timer(&adap->sge, us);
277 
278 	if ((us | cnt) == 0)
279 		q->intr_params = V_QINTR_TIMER_IDX(X_TIMERREG_UPDATE_CIDX);
280 	else
281 		q->intr_params = V_QINTR_TIMER_IDX(timer_val) |
282 				 V_QINTR_CNT_EN(cnt > 0);
283 	return 0;
284 }
285 
286 /**
287  * Allocate an active-open TID and set it to the supplied value.
288  */
289 int cxgbe_alloc_atid(struct tid_info *t, void *data)
290 {
291 	int atid = -1;
292 
293 	t4_os_lock(&t->atid_lock);
294 	if (t->afree) {
295 		union aopen_entry *p = t->afree;
296 
297 		atid = p - t->atid_tab;
298 		t->afree = p->next;
299 		p->data = data;
300 		t->atids_in_use++;
301 	}
302 	t4_os_unlock(&t->atid_lock);
303 	return atid;
304 }
305 
306 /**
307  * Release an active-open TID.
308  */
309 void cxgbe_free_atid(struct tid_info *t, unsigned int atid)
310 {
311 	union aopen_entry *p = &t->atid_tab[atid];
312 
313 	t4_os_lock(&t->atid_lock);
314 	p->next = t->afree;
315 	t->afree = p;
316 	t->atids_in_use--;
317 	t4_os_unlock(&t->atid_lock);
318 }
319 
320 /**
321  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
322  */
323 static void mk_tid_release(struct rte_mbuf *mbuf, unsigned int tid)
324 {
325 	struct cpl_tid_release *req;
326 
327 	req = rte_pktmbuf_mtod(mbuf, struct cpl_tid_release *);
328 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
329 }
330 
331 /**
332  * Release a TID and inform HW.  If we are unable to allocate the release
333  * message we defer to a work queue.
334  */
335 void cxgbe_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
336 		      unsigned short family)
337 {
338 	struct rte_mbuf *mbuf;
339 	struct adapter *adap = container_of(t, struct adapter, tids);
340 
341 	WARN_ON(tid >= t->ntids);
342 
343 	if (t->tid_tab[tid]) {
344 		t->tid_tab[tid] = NULL;
345 		rte_atomic32_dec(&t->conns_in_use);
346 		if (t->hash_base && tid >= t->hash_base) {
347 			if (family == FILTER_TYPE_IPV4)
348 				rte_atomic32_dec(&t->hash_tids_in_use);
349 		} else {
350 			if (family == FILTER_TYPE_IPV4)
351 				rte_atomic32_dec(&t->tids_in_use);
352 		}
353 	}
354 
355 	mbuf = rte_pktmbuf_alloc((&adap->sge.ctrlq[chan])->mb_pool);
356 	if (mbuf) {
357 		mbuf->data_len = sizeof(struct cpl_tid_release);
358 		mbuf->pkt_len = mbuf->data_len;
359 		mk_tid_release(mbuf, tid);
360 		t4_mgmt_tx(&adap->sge.ctrlq[chan], mbuf);
361 	}
362 }
363 
364 /**
365  * Insert a TID.
366  */
367 void cxgbe_insert_tid(struct tid_info *t, void *data, unsigned int tid,
368 		      unsigned short family)
369 {
370 	t->tid_tab[tid] = data;
371 	if (t->hash_base && tid >= t->hash_base) {
372 		if (family == FILTER_TYPE_IPV4)
373 			rte_atomic32_inc(&t->hash_tids_in_use);
374 	} else {
375 		if (family == FILTER_TYPE_IPV4)
376 			rte_atomic32_inc(&t->tids_in_use);
377 	}
378 
379 	rte_atomic32_inc(&t->conns_in_use);
380 }
381 
382 /**
383  * Free TID tables.
384  */
385 static void tid_free(struct tid_info *t)
386 {
387 	if (t->tid_tab) {
388 		if (t->ftid_bmap)
389 			rte_bitmap_free(t->ftid_bmap);
390 
391 		if (t->ftid_bmap_array)
392 			t4_os_free(t->ftid_bmap_array);
393 
394 		t4_os_free(t->tid_tab);
395 	}
396 
397 	memset(t, 0, sizeof(struct tid_info));
398 }
399 
400 /**
401  * Allocate and initialize the TID tables.  Returns 0 on success.
402  */
403 static int tid_init(struct tid_info *t)
404 {
405 	size_t size;
406 	unsigned int ftid_bmap_size;
407 	unsigned int natids = t->natids;
408 	unsigned int max_ftids = t->nftids;
409 
410 	ftid_bmap_size = rte_bitmap_get_memory_footprint(t->nftids);
411 	size = t->ntids * sizeof(*t->tid_tab) +
412 		max_ftids * sizeof(*t->ftid_tab) +
413 		natids * sizeof(*t->atid_tab);
414 
415 	t->tid_tab = t4_os_alloc(size);
416 	if (!t->tid_tab)
417 		return -ENOMEM;
418 
419 	t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
420 	t->ftid_tab = (struct filter_entry *)&t->atid_tab[t->natids];
421 	t->ftid_bmap_array = t4_os_alloc(ftid_bmap_size);
422 	if (!t->ftid_bmap_array) {
423 		tid_free(t);
424 		return -ENOMEM;
425 	}
426 
427 	t4_os_lock_init(&t->atid_lock);
428 	t4_os_lock_init(&t->ftid_lock);
429 
430 	t->afree = NULL;
431 	t->atids_in_use = 0;
432 	rte_atomic32_init(&t->tids_in_use);
433 	rte_atomic32_set(&t->tids_in_use, 0);
434 	rte_atomic32_init(&t->conns_in_use);
435 	rte_atomic32_set(&t->conns_in_use, 0);
436 
437 	/* Setup the free list for atid_tab and clear the stid bitmap. */
438 	if (natids) {
439 		while (--natids)
440 			t->atid_tab[natids - 1].next = &t->atid_tab[natids];
441 		t->afree = t->atid_tab;
442 	}
443 
444 	t->ftid_bmap = rte_bitmap_init(t->nftids, t->ftid_bmap_array,
445 				       ftid_bmap_size);
446 	if (!t->ftid_bmap) {
447 		tid_free(t);
448 		return -ENOMEM;
449 	}
450 
451 	return 0;
452 }
453 
454 static inline bool is_x_1g_port(const struct link_config *lc)
455 {
456 	return (lc->pcaps & FW_PORT_CAP32_SPEED_1G) != 0;
457 }
458 
459 static inline bool is_x_10g_port(const struct link_config *lc)
460 {
461 	unsigned int speeds, high_speeds;
462 
463 	speeds = V_FW_PORT_CAP32_SPEED(G_FW_PORT_CAP32_SPEED(lc->pcaps));
464 	high_speeds = speeds &
465 		      ~(FW_PORT_CAP32_SPEED_100M | FW_PORT_CAP32_SPEED_1G);
466 
467 	return high_speeds != 0;
468 }
469 
470 static inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
471 		      unsigned int us, unsigned int cnt,
472 		      unsigned int size, unsigned int iqe_size)
473 {
474 	q->adapter = adap;
475 	cxgb4_set_rspq_intr_params(q, us, cnt);
476 	q->iqe_len = iqe_size;
477 	q->size = size;
478 }
479 
480 int cxgbe_cfg_queue_count(struct rte_eth_dev *eth_dev)
481 {
482 	struct port_info *pi = eth_dev->data->dev_private;
483 	struct adapter *adap = pi->adapter;
484 	struct sge *s = &adap->sge;
485 	unsigned int max_queues = s->max_ethqsets / adap->params.nports;
486 
487 	if ((eth_dev->data->nb_rx_queues < 1) ||
488 	    (eth_dev->data->nb_tx_queues < 1))
489 		return -EINVAL;
490 
491 	if ((eth_dev->data->nb_rx_queues > max_queues) ||
492 	    (eth_dev->data->nb_tx_queues > max_queues))
493 		return -EINVAL;
494 
495 	if (eth_dev->data->nb_rx_queues > pi->rss_size)
496 		return -EINVAL;
497 
498 	/* We must configure RSS, since config has changed*/
499 	pi->flags &= ~PORT_RSS_DONE;
500 
501 	pi->n_rx_qsets = eth_dev->data->nb_rx_queues;
502 	pi->n_tx_qsets = eth_dev->data->nb_tx_queues;
503 
504 	return 0;
505 }
506 
507 void cxgbe_cfg_queues(struct rte_eth_dev *eth_dev)
508 {
509 	struct port_info *pi = eth_dev->data->dev_private;
510 	struct adapter *adap = pi->adapter;
511 	struct sge *s = &adap->sge;
512 	unsigned int i, nb_ports = 0, qidx = 0;
513 	unsigned int q_per_port = 0;
514 
515 	if (!(adap->flags & CFG_QUEUES)) {
516 		for_each_port(adap, i) {
517 			struct port_info *tpi = adap2pinfo(adap, i);
518 
519 			nb_ports += (is_x_10g_port(&tpi->link_cfg)) ||
520 				     is_x_1g_port(&tpi->link_cfg) ? 1 : 0;
521 		}
522 
523 		/*
524 		 * We default up to # of cores queues per 1G/10G port.
525 		 */
526 		if (nb_ports)
527 			q_per_port = (s->max_ethqsets -
528 				     (adap->params.nports - nb_ports)) /
529 				     nb_ports;
530 
531 		if (q_per_port > rte_lcore_count())
532 			q_per_port = rte_lcore_count();
533 
534 		for_each_port(adap, i) {
535 			struct port_info *pi = adap2pinfo(adap, i);
536 
537 			pi->first_qset = qidx;
538 
539 			/* Initially n_rx_qsets == n_tx_qsets */
540 			pi->n_rx_qsets = (is_x_10g_port(&pi->link_cfg) ||
541 					  is_x_1g_port(&pi->link_cfg)) ?
542 					  q_per_port : 1;
543 			pi->n_tx_qsets = pi->n_rx_qsets;
544 
545 			if (pi->n_rx_qsets > pi->rss_size)
546 				pi->n_rx_qsets = pi->rss_size;
547 
548 			qidx += pi->n_rx_qsets;
549 		}
550 
551 		for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
552 			struct sge_eth_rxq *r = &s->ethrxq[i];
553 
554 			init_rspq(adap, &r->rspq, 5, 32, 1024, 64);
555 			r->usembufs = 1;
556 			r->fl.size = (r->usembufs ? 1024 : 72);
557 		}
558 
559 		for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
560 			s->ethtxq[i].q.size = 1024;
561 
562 		init_rspq(adap, &adap->sge.fw_evtq, 0, 0, 1024, 64);
563 		adap->flags |= CFG_QUEUES;
564 	}
565 }
566 
567 void cxgbe_stats_get(struct port_info *pi, struct port_stats *stats)
568 {
569 	t4_get_port_stats_offset(pi->adapter, pi->tx_chan, stats,
570 				 &pi->stats_base);
571 }
572 
573 void cxgbe_stats_reset(struct port_info *pi)
574 {
575 	t4_clr_port_stats(pi->adapter, pi->tx_chan);
576 }
577 
578 static void setup_memwin(struct adapter *adap)
579 {
580 	u32 mem_win0_base;
581 
582 	/* For T5, only relative offset inside the PCIe BAR is passed */
583 	mem_win0_base = MEMWIN0_BASE;
584 
585 	/*
586 	 * Set up memory window for accessing adapter memory ranges.  (Read
587 	 * back MA register to ensure that changes propagate before we attempt
588 	 * to use the new values.)
589 	 */
590 	t4_write_reg(adap,
591 		     PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
592 					 MEMWIN_NIC),
593 		     mem_win0_base | V_BIR(0) |
594 		     V_WINDOW(ilog2(MEMWIN0_APERTURE) - X_WINDOW_SHIFT));
595 	t4_read_reg(adap,
596 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
597 					MEMWIN_NIC));
598 }
599 
600 int cxgbe_init_rss(struct adapter *adap)
601 {
602 	unsigned int i;
603 
604 	if (is_pf4(adap)) {
605 		int err;
606 
607 		err = t4_init_rss_mode(adap, adap->mbox);
608 		if (err)
609 			return err;
610 	}
611 
612 	for_each_port(adap, i) {
613 		struct port_info *pi = adap2pinfo(adap, i);
614 
615 		pi->rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
616 		if (!pi->rss)
617 			return -ENOMEM;
618 
619 		pi->rss_hf = CXGBE_RSS_HF_ALL;
620 	}
621 	return 0;
622 }
623 
624 /**
625  * Dump basic information about the adapter.
626  */
627 void cxgbe_print_adapter_info(struct adapter *adap)
628 {
629 	/**
630 	 * Hardware/Firmware/etc. Version/Revision IDs.
631 	 */
632 	t4_dump_version_info(adap);
633 }
634 
635 void cxgbe_print_port_info(struct adapter *adap)
636 {
637 	int i;
638 	char buf[80];
639 	struct rte_pci_addr *loc = &adap->pdev->addr;
640 
641 	for_each_port(adap, i) {
642 		const struct port_info *pi = adap2pinfo(adap, i);
643 		char *bufp = buf;
644 
645 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
646 			bufp += sprintf(bufp, "100M/");
647 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
648 			bufp += sprintf(bufp, "1G/");
649 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
650 			bufp += sprintf(bufp, "10G/");
651 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
652 			bufp += sprintf(bufp, "25G/");
653 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
654 			bufp += sprintf(bufp, "40G/");
655 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
656 			bufp += sprintf(bufp, "50G/");
657 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
658 			bufp += sprintf(bufp, "100G/");
659 		if (bufp != buf)
660 			--bufp;
661 		sprintf(bufp, "BASE-%s",
662 			t4_get_port_type_description(
663 					(enum fw_port_type)pi->port_type));
664 
665 		dev_info(adap,
666 			 " " PCI_PRI_FMT " Chelsio rev %d %s %s\n",
667 			 loc->domain, loc->bus, loc->devid, loc->function,
668 			 CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
669 			 (adap->flags & USING_MSIX) ? " MSI-X" :
670 			 (adap->flags & USING_MSI) ? " MSI" : "");
671 	}
672 }
673 
674 static int check_devargs_handler(const char *key, const char *value, void *p)
675 {
676 	if (!strncmp(key, CXGBE_DEVARG_CMN_KEEP_OVLAN, strlen(key)) ||
677 	    !strncmp(key, CXGBE_DEVARG_CMN_TX_MODE_LATENCY, strlen(key)) ||
678 	    !strncmp(key, CXGBE_DEVARG_VF_FORCE_LINK_UP, strlen(key))) {
679 		if (!strncmp(value, "1", 1)) {
680 			bool *dst_val = (bool *)p;
681 
682 			*dst_val = true;
683 		}
684 	}
685 
686 	return 0;
687 }
688 
689 static int cxgbe_get_devargs(struct rte_devargs *devargs, const char *key,
690 			     void *p)
691 {
692 	struct rte_kvargs *kvlist;
693 	int ret = 0;
694 
695 	if (!devargs)
696 		return 0;
697 
698 	kvlist = rte_kvargs_parse(devargs->args, NULL);
699 	if (!kvlist)
700 		return 0;
701 
702 	if (!rte_kvargs_count(kvlist, key))
703 		goto out;
704 
705 	ret = rte_kvargs_process(kvlist, key, check_devargs_handler, p);
706 
707 out:
708 	rte_kvargs_free(kvlist);
709 
710 	return ret;
711 }
712 
713 static void cxgbe_get_devargs_int(struct adapter *adap, bool *dst,
714 				  const char *key, bool default_value)
715 {
716 	struct rte_pci_device *pdev = adap->pdev;
717 	int ret;
718 	bool devarg_value = default_value;
719 
720 	*dst = default_value;
721 	if (!pdev)
722 		return;
723 
724 	ret = cxgbe_get_devargs(pdev->device.devargs, key, &devarg_value);
725 	if (ret)
726 		return;
727 
728 	*dst = devarg_value;
729 }
730 
731 void cxgbe_process_devargs(struct adapter *adap)
732 {
733 	cxgbe_get_devargs_int(adap, &adap->devargs.keep_ovlan,
734 			      CXGBE_DEVARG_CMN_KEEP_OVLAN, false);
735 	cxgbe_get_devargs_int(adap, &adap->devargs.tx_mode_latency,
736 			      CXGBE_DEVARG_CMN_TX_MODE_LATENCY, false);
737 	cxgbe_get_devargs_int(adap, &adap->devargs.force_link_up,
738 			      CXGBE_DEVARG_VF_FORCE_LINK_UP, false);
739 }
740 
741 static void configure_vlan_types(struct adapter *adapter)
742 {
743 	int i;
744 
745 	for_each_port(adapter, i) {
746 		/* OVLAN Type 0x88a8 */
747 		t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN0),
748 				 V_OVLAN_MASK(M_OVLAN_MASK) |
749 				 V_OVLAN_ETYPE(M_OVLAN_ETYPE),
750 				 V_OVLAN_MASK(M_OVLAN_MASK) |
751 				 V_OVLAN_ETYPE(0x88a8));
752 		/* OVLAN Type 0x9100 */
753 		t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN1),
754 				 V_OVLAN_MASK(M_OVLAN_MASK) |
755 				 V_OVLAN_ETYPE(M_OVLAN_ETYPE),
756 				 V_OVLAN_MASK(M_OVLAN_MASK) |
757 				 V_OVLAN_ETYPE(0x9100));
758 
759 		/* IVLAN 0X8100 */
760 		t4_set_reg_field(adapter, MPS_PORT_RX_IVLAN(i),
761 				 V_IVLAN_ETYPE(M_IVLAN_ETYPE),
762 				 V_IVLAN_ETYPE(0x8100));
763 
764 		t4_set_reg_field(adapter, MPS_PORT_RX_CTL(i),
765 				 F_OVLAN_EN0 | F_OVLAN_EN1 |
766 				 F_IVLAN_EN,
767 				 F_OVLAN_EN0 | F_OVLAN_EN1 |
768 				 F_IVLAN_EN);
769 	}
770 
771 	t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG, V_RM_OVLAN(1),
772 			       V_RM_OVLAN(!adapter->devargs.keep_ovlan));
773 }
774 
775 static void configure_pcie_ext_tag(struct adapter *adapter)
776 {
777 	u16 v;
778 	int pos = t4_os_find_pci_capability(adapter, PCI_CAP_ID_EXP);
779 
780 	if (!pos)
781 		return;
782 
783 	if (pos > 0) {
784 		t4_os_pci_read_cfg2(adapter, pos + PCI_EXP_DEVCTL, &v);
785 		v |= PCI_EXP_DEVCTL_EXT_TAG;
786 		t4_os_pci_write_cfg2(adapter, pos + PCI_EXP_DEVCTL, v);
787 		if (is_t6(adapter->params.chip)) {
788 			t4_set_reg_field(adapter, A_PCIE_CFG2,
789 					 V_T6_TOTMAXTAG(M_T6_TOTMAXTAG),
790 					 V_T6_TOTMAXTAG(7));
791 			t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
792 					 V_T6_MINTAG(M_T6_MINTAG),
793 					 V_T6_MINTAG(8));
794 		} else {
795 			t4_set_reg_field(adapter, A_PCIE_CFG2,
796 					 V_TOTMAXTAG(M_TOTMAXTAG),
797 					 V_TOTMAXTAG(3));
798 			t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
799 					 V_MINTAG(M_MINTAG),
800 					 V_MINTAG(8));
801 		}
802 	}
803 }
804 
805 /* Figure out how many Queue Sets we can support */
806 void cxgbe_configure_max_ethqsets(struct adapter *adapter)
807 {
808 	unsigned int ethqsets;
809 
810 	/*
811 	 * We need to reserve an Ingress Queue for the Asynchronous Firmware
812 	 * Event Queue.
813 	 *
814 	 * For each Queue Set, we'll need the ability to allocate two Egress
815 	 * Contexts -- one for the Ingress Queue Free List and one for the TX
816 	 * Ethernet Queue.
817 	 */
818 	if (is_pf4(adapter)) {
819 		struct pf_resources *pfres = &adapter->params.pfres;
820 
821 		ethqsets = pfres->niqflint - 1;
822 		if (pfres->neq < ethqsets * 2)
823 			ethqsets = pfres->neq / 2;
824 	} else {
825 		struct vf_resources *vfres = &adapter->params.vfres;
826 
827 		ethqsets = vfres->niqflint - 1;
828 		if (vfres->nethctrl != ethqsets)
829 			ethqsets = min(vfres->nethctrl, ethqsets);
830 		if (vfres->neq < ethqsets * 2)
831 			ethqsets = vfres->neq / 2;
832 	}
833 
834 	if (ethqsets > MAX_ETH_QSETS)
835 		ethqsets = MAX_ETH_QSETS;
836 	adapter->sge.max_ethqsets = ethqsets;
837 }
838 
839 /*
840  * Tweak configuration based on system architecture, etc.  Most of these have
841  * defaults assigned to them by Firmware Configuration Files (if we're using
842  * them) but need to be explicitly set if we're using hard-coded
843  * initialization. So these are essentially common tweaks/settings for
844  * Configuration Files and hard-coded initialization ...
845  */
846 static int adap_init0_tweaks(struct adapter *adapter)
847 {
848 	u8 rx_dma_offset;
849 
850 	/*
851 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
852 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
853 	 * 64B Cache Line Size ...
854 	 */
855 	t4_fixup_host_params_compat(adapter, CXGBE_PAGE_SIZE, L1_CACHE_BYTES,
856 				    T5_LAST_REV);
857 
858 	/*
859 	 * Keep the chip default offset to deliver Ingress packets into our
860 	 * DMA buffers to zero
861 	 */
862 	rx_dma_offset = 0;
863 	t4_set_reg_field(adapter, A_SGE_CONTROL, V_PKTSHIFT(M_PKTSHIFT),
864 			 V_PKTSHIFT(rx_dma_offset));
865 
866 	t4_set_reg_field(adapter, A_SGE_FLM_CFG,
867 			 V_CREDITCNT(M_CREDITCNT) | M_CREDITCNTPACKING,
868 			 V_CREDITCNT(3) | V_CREDITCNTPACKING(1));
869 
870 	t4_set_reg_field(adapter, A_SGE_INGRESS_RX_THRESHOLD,
871 			 V_THRESHOLD_3(M_THRESHOLD_3), V_THRESHOLD_3(32U));
872 
873 	t4_set_reg_field(adapter, A_SGE_CONTROL2, V_IDMAARBROUNDROBIN(1U),
874 			 V_IDMAARBROUNDROBIN(1U));
875 
876 	/*
877 	 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
878 	 * adds the pseudo header itself.
879 	 */
880 	t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG,
881 			       F_CSUM_HAS_PSEUDO_HDR, 0);
882 
883 	return 0;
884 }
885 
886 /*
887  * Attempt to initialize the adapter via a Firmware Configuration File.
888  */
889 static int adap_init0_config(struct adapter *adapter, int reset)
890 {
891 	struct fw_caps_config_cmd caps_cmd;
892 	unsigned long mtype = 0, maddr = 0;
893 	u32 finiver, finicsum, cfcsum;
894 	int ret;
895 	int config_issued = 0;
896 	int cfg_addr;
897 	char config_name[20];
898 
899 	/*
900 	 * Reset device if necessary.
901 	 */
902 	if (reset) {
903 		ret = t4_fw_reset(adapter, adapter->mbox,
904 				  F_PIORSTMODE | F_PIORST);
905 		if (ret < 0) {
906 			dev_warn(adapter, "Firmware reset failed, error %d\n",
907 				 -ret);
908 			goto bye;
909 		}
910 	}
911 
912 	cfg_addr = t4_flash_cfg_addr(adapter);
913 	if (cfg_addr < 0) {
914 		ret = cfg_addr;
915 		dev_warn(adapter, "Finding address for firmware config file in flash failed, error %d\n",
916 			 -ret);
917 		goto bye;
918 	}
919 
920 	strcpy(config_name, "On Flash");
921 	mtype = FW_MEMTYPE_CF_FLASH;
922 	maddr = cfg_addr;
923 
924 	/*
925 	 * Issue a Capability Configuration command to the firmware to get it
926 	 * to parse the Configuration File.  We don't use t4_fw_config_file()
927 	 * because we want the ability to modify various features after we've
928 	 * processed the configuration file ...
929 	 */
930 	memset(&caps_cmd, 0, sizeof(caps_cmd));
931 	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
932 					   F_FW_CMD_REQUEST | F_FW_CMD_READ);
933 	caps_cmd.cfvalid_to_len16 =
934 		cpu_to_be32(F_FW_CAPS_CONFIG_CMD_CFVALID |
935 			    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
936 			    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
937 			    FW_LEN16(caps_cmd));
938 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
939 			 &caps_cmd);
940 	/*
941 	 * If the CAPS_CONFIG failed with an ENOENT (for a Firmware
942 	 * Configuration File in FLASH), our last gasp effort is to use the
943 	 * Firmware Configuration File which is embedded in the firmware.  A
944 	 * very few early versions of the firmware didn't have one embedded
945 	 * but we can ignore those.
946 	 */
947 	if (ret == -ENOENT) {
948 		dev_info(adapter, "%s: Going for embedded config in firmware..\n",
949 			 __func__);
950 
951 		memset(&caps_cmd, 0, sizeof(caps_cmd));
952 		caps_cmd.op_to_write =
953 			cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
954 				    F_FW_CMD_REQUEST | F_FW_CMD_READ);
955 		caps_cmd.cfvalid_to_len16 = cpu_to_be32(FW_LEN16(caps_cmd));
956 		ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
957 				 sizeof(caps_cmd), &caps_cmd);
958 		strcpy(config_name, "Firmware Default");
959 	}
960 
961 	config_issued = 1;
962 	if (ret < 0)
963 		goto bye;
964 
965 	finiver = be32_to_cpu(caps_cmd.finiver);
966 	finicsum = be32_to_cpu(caps_cmd.finicsum);
967 	cfcsum = be32_to_cpu(caps_cmd.cfcsum);
968 	if (finicsum != cfcsum)
969 		dev_warn(adapter, "Configuration File checksum mismatch: [fini] csum=%#x, computed csum=%#x\n",
970 			 finicsum, cfcsum);
971 
972 	/*
973 	 * If we're a pure NIC driver then disable all offloading facilities.
974 	 * This will allow the firmware to optimize aspects of the hardware
975 	 * configuration which will result in improved performance.
976 	 */
977 	caps_cmd.niccaps &= cpu_to_be16(~FW_CAPS_CONFIG_NIC_ETHOFLD);
978 	caps_cmd.toecaps = 0;
979 	caps_cmd.iscsicaps = 0;
980 	caps_cmd.rdmacaps = 0;
981 	caps_cmd.fcoecaps = 0;
982 
983 	/*
984 	 * And now tell the firmware to use the configuration we just loaded.
985 	 */
986 	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
987 					   F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
988 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
989 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
990 			 NULL);
991 	if (ret < 0) {
992 		dev_warn(adapter, "Unable to finalize Firmware Capabilities %d\n",
993 			 -ret);
994 		goto bye;
995 	}
996 
997 	/*
998 	 * Tweak configuration based on system architecture, etc.
999 	 */
1000 	ret = adap_init0_tweaks(adapter);
1001 	if (ret < 0) {
1002 		dev_warn(adapter, "Unable to do init0-tweaks %d\n", -ret);
1003 		goto bye;
1004 	}
1005 
1006 	/*
1007 	 * And finally tell the firmware to initialize itself using the
1008 	 * parameters from the Configuration File.
1009 	 */
1010 	ret = t4_fw_initialize(adapter, adapter->mbox);
1011 	if (ret < 0) {
1012 		dev_warn(adapter, "Initializing Firmware failed, error %d\n",
1013 			 -ret);
1014 		goto bye;
1015 	}
1016 
1017 	/*
1018 	 * Return successfully and note that we're operating with parameters
1019 	 * not supplied by the driver, rather than from hard-wired
1020 	 * initialization constants buried in the driver.
1021 	 */
1022 	dev_info(adapter,
1023 		 "Successfully configured using Firmware Configuration File \"%s\", version %#x, computed checksum %#x\n",
1024 		 config_name, finiver, cfcsum);
1025 
1026 	return 0;
1027 
1028 	/*
1029 	 * Something bad happened.  Return the error ...  (If the "error"
1030 	 * is that there's no Configuration File on the adapter we don't
1031 	 * want to issue a warning since this is fairly common.)
1032 	 */
1033 bye:
1034 	if (config_issued && ret != -ENOENT)
1035 		dev_warn(adapter, "\"%s\" configuration file error %d\n",
1036 			 config_name, -ret);
1037 
1038 	dev_debug(adapter, "%s: returning ret = %d ..\n", __func__, ret);
1039 	return ret;
1040 }
1041 
1042 static int adap_init0(struct adapter *adap)
1043 {
1044 	struct fw_caps_config_cmd caps_cmd;
1045 	int ret = 0;
1046 	u32 v, port_vec;
1047 	enum dev_state state;
1048 	u32 params[7], val[7];
1049 	int reset = 1;
1050 	int mbox = adap->mbox;
1051 
1052 	/*
1053 	 * Contact FW, advertising Master capability.
1054 	 */
1055 	ret = t4_fw_hello(adap, adap->mbox, adap->mbox, MASTER_MAY, &state);
1056 	if (ret < 0) {
1057 		dev_err(adap, "%s: could not connect to FW, error %d\n",
1058 			__func__, -ret);
1059 		goto bye;
1060 	}
1061 
1062 	CXGBE_DEBUG_MBOX(adap, "%s: adap->mbox = %d; ret = %d\n", __func__,
1063 			 adap->mbox, ret);
1064 
1065 	if (ret == mbox)
1066 		adap->flags |= MASTER_PF;
1067 
1068 	if (state == DEV_STATE_INIT) {
1069 		/*
1070 		 * Force halt and reset FW because a previous instance may have
1071 		 * exited abnormally without properly shutting down
1072 		 */
1073 		ret = t4_fw_halt(adap, adap->mbox, reset);
1074 		if (ret < 0) {
1075 			dev_err(adap, "Failed to halt. Exit.\n");
1076 			goto bye;
1077 		}
1078 
1079 		ret = t4_fw_restart(adap, adap->mbox, reset);
1080 		if (ret < 0) {
1081 			dev_err(adap, "Failed to restart. Exit.\n");
1082 			goto bye;
1083 		}
1084 		state = (enum dev_state)((unsigned)state & ~DEV_STATE_INIT);
1085 	}
1086 
1087 	t4_get_version_info(adap);
1088 
1089 	ret = t4_get_core_clock(adap, &adap->params.vpd);
1090 	if (ret < 0) {
1091 		dev_err(adap, "%s: could not get core clock, error %d\n",
1092 			__func__, -ret);
1093 		goto bye;
1094 	}
1095 
1096 	/*
1097 	 * If the firmware is initialized already (and we're not forcing a
1098 	 * master initialization), note that we're living with existing
1099 	 * adapter parameters.  Otherwise, it's time to try initializing the
1100 	 * adapter ...
1101 	 */
1102 	if (state == DEV_STATE_INIT) {
1103 		dev_info(adap, "Coming up as %s: Adapter already initialized\n",
1104 			 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
1105 	} else {
1106 		dev_info(adap, "Coming up as MASTER: Initializing adapter\n");
1107 
1108 		ret = adap_init0_config(adap, reset);
1109 		if (ret == -ENOENT) {
1110 			dev_err(adap,
1111 				"No Configuration File present on adapter. Using hard-wired configuration parameters.\n");
1112 			goto bye;
1113 		}
1114 	}
1115 	if (ret < 0) {
1116 		dev_err(adap, "could not initialize adapter, error %d\n", -ret);
1117 		goto bye;
1118 	}
1119 
1120 	/* Now that we've successfully configured and initialized the adapter
1121 	 * (or found it already initialized), we can ask the Firmware what
1122 	 * resources it has provisioned for us.
1123 	 */
1124 	ret = t4_get_pfres(adap);
1125 	if (ret) {
1126 		dev_err(adap->pdev_dev,
1127 			"Unable to retrieve resource provisioning info\n");
1128 		goto bye;
1129 	}
1130 
1131 	/* Find out what ports are available to us. */
1132 	v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
1133 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
1134 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
1135 	if (ret < 0) {
1136 		dev_err(adap, "%s: failure in t4_query_params; error = %d\n",
1137 			__func__, ret);
1138 		goto bye;
1139 	}
1140 
1141 	adap->params.nports = hweight32(port_vec);
1142 	adap->params.portvec = port_vec;
1143 
1144 	dev_debug(adap, "%s: adap->params.nports = %u\n", __func__,
1145 		  adap->params.nports);
1146 
1147 	/*
1148 	 * Give the SGE code a chance to pull in anything that it needs ...
1149 	 * Note that this must be called after we retrieve our VPD parameters
1150 	 * in order to know how to convert core ticks to seconds, etc.
1151 	 */
1152 	ret = t4_sge_init(adap);
1153 	if (ret < 0) {
1154 		dev_err(adap, "t4_sge_init failed with error %d\n",
1155 			-ret);
1156 		goto bye;
1157 	}
1158 
1159 	/*
1160 	 * Grab some of our basic fundamental operating parameters.
1161 	 */
1162 	params[0] = CXGBE_FW_PARAM_PFVF(L2T_START);
1163 	params[1] = CXGBE_FW_PARAM_PFVF(L2T_END);
1164 	params[2] = CXGBE_FW_PARAM_PFVF(FILTER_START);
1165 	params[3] = CXGBE_FW_PARAM_PFVF(FILTER_END);
1166 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 4, params, val);
1167 	if (ret < 0)
1168 		goto bye;
1169 	adap->l2t_start = val[0];
1170 	adap->l2t_end = val[1];
1171 	adap->tids.ftid_base = val[2];
1172 	adap->tids.nftids = val[3] - val[2] + 1;
1173 
1174 	params[0] = CXGBE_FW_PARAM_PFVF(CLIP_START);
1175 	params[1] = CXGBE_FW_PARAM_PFVF(CLIP_END);
1176 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
1177 	if (ret < 0)
1178 		goto bye;
1179 	adap->clipt_start = val[0];
1180 	adap->clipt_end = val[1];
1181 
1182 	/*
1183 	 * Get device capabilities so we can determine what resources we need
1184 	 * to manage.
1185 	 */
1186 	memset(&caps_cmd, 0, sizeof(caps_cmd));
1187 	caps_cmd.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1188 				     F_FW_CMD_REQUEST | F_FW_CMD_READ);
1189 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
1190 	ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
1191 			 &caps_cmd);
1192 	if (ret < 0)
1193 		goto bye;
1194 
1195 	if ((caps_cmd.niccaps & cpu_to_be16(FW_CAPS_CONFIG_NIC_HASHFILTER)) &&
1196 	    is_t6(adap->params.chip)) {
1197 		if (cxgbe_init_hash_filter(adap) < 0)
1198 			goto bye;
1199 	}
1200 
1201 	/* See if FW supports FW_FILTER2 work request */
1202 	if (is_t4(adap->params.chip)) {
1203 		adap->params.filter2_wr_support = 0;
1204 	} else {
1205 		params[0] = CXGBE_FW_PARAM_DEV(FILTER2_WR);
1206 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1207 				      1, params, val);
1208 		adap->params.filter2_wr_support = (ret == 0 && val[0] != 0);
1209 	}
1210 
1211 	/* Check if FW supports returning vin.
1212 	 * If this is not supported, driver will interpret
1213 	 * these values from viid.
1214 	 */
1215 	params[0] = CXGBE_FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
1216 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1217 			      1, params, val);
1218 	adap->params.viid_smt_extn_support = (ret == 0 && val[0] != 0);
1219 
1220 	/* query tid-related parameters */
1221 	params[0] = CXGBE_FW_PARAM_DEV(NTID);
1222 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
1223 			      params, val);
1224 	if (ret < 0)
1225 		goto bye;
1226 	adap->tids.ntids = val[0];
1227 	adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
1228 
1229 	/* If we're running on newer firmware, let it know that we're
1230 	 * prepared to deal with encapsulated CPL messages.  Older
1231 	 * firmware won't understand this and we'll just get
1232 	 * unencapsulated messages ...
1233 	 */
1234 	params[0] = CXGBE_FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
1235 	val[0] = 1;
1236 	(void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1237 
1238 	/*
1239 	 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
1240 	 * capability.  Earlier versions of the firmware didn't have the
1241 	 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
1242 	 * permission to use ULPTX MEMWRITE DSGL.
1243 	 */
1244 	if (is_t4(adap->params.chip)) {
1245 		adap->params.ulptx_memwrite_dsgl = false;
1246 	} else {
1247 		params[0] = CXGBE_FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
1248 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1249 				      1, params, val);
1250 		adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
1251 	}
1252 
1253 	/* Query for max number of packets that can be coalesced for Tx */
1254 	params[0] = CXGBE_FW_PARAM_PFVF(MAX_PKTS_PER_ETH_TX_PKTS_WR);
1255 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1256 	if (!ret && val[0] > 0)
1257 		adap->params.max_tx_coalesce_num = val[0];
1258 	else
1259 		adap->params.max_tx_coalesce_num = ETH_COALESCE_PKT_NUM;
1260 
1261 	/*
1262 	 * The MTU/MSS Table is initialized by now, so load their values.  If
1263 	 * we're initializing the adapter, then we'll make any modifications
1264 	 * we want to the MTU/MSS Table and also initialize the congestion
1265 	 * parameters.
1266 	 */
1267 	t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
1268 	if (state != DEV_STATE_INIT) {
1269 		int i;
1270 
1271 		/*
1272 		 * The default MTU Table contains values 1492 and 1500.
1273 		 * However, for TCP, it's better to have two values which are
1274 		 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
1275 		 * This allows us to have a TCP Data Payload which is a
1276 		 * multiple of 8 regardless of what combination of TCP Options
1277 		 * are in use (always a multiple of 4 bytes) which is
1278 		 * important for performance reasons.  For instance, if no
1279 		 * options are in use, then we have a 20-byte IP header and a
1280 		 * 20-byte TCP header.  In this case, a 1500-byte MSS would
1281 		 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
1282 		 * which is not a multiple of 8.  So using an MSS of 1488 in
1283 		 * this case results in a TCP Data Payload of 1448 bytes which
1284 		 * is a multiple of 8.  On the other hand, if 12-byte TCP Time
1285 		 * Stamps have been negotiated, then an MTU of 1500 bytes
1286 		 * results in a TCP Data Payload of 1448 bytes which, as
1287 		 * above, is a multiple of 8 bytes ...
1288 		 */
1289 		for (i = 0; i < NMTUS; i++)
1290 			if (adap->params.mtus[i] == 1492) {
1291 				adap->params.mtus[i] = 1488;
1292 				break;
1293 			}
1294 
1295 		t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
1296 			     adap->params.b_wnd);
1297 	}
1298 	t4_init_sge_params(adap);
1299 	t4_init_tp_params(adap);
1300 	configure_pcie_ext_tag(adap);
1301 	configure_vlan_types(adap);
1302 	cxgbe_configure_max_ethqsets(adap);
1303 
1304 	adap->params.drv_memwin = MEMWIN_NIC;
1305 	adap->flags |= FW_OK;
1306 	dev_debug(adap, "%s: returning zero..\n", __func__);
1307 	return 0;
1308 
1309 	/*
1310 	 * Something bad happened.  If a command timed out or failed with EIO
1311 	 * FW does not operate within its spec or something catastrophic
1312 	 * happened to HW/FW, stop issuing commands.
1313 	 */
1314 bye:
1315 	if (ret != -ETIMEDOUT && ret != -EIO)
1316 		t4_fw_bye(adap, adap->mbox);
1317 	return ret;
1318 }
1319 
1320 /**
1321  * t4_os_portmod_changed - handle port module changes
1322  * @adap: the adapter associated with the module change
1323  * @port_id: the port index whose module status has changed
1324  *
1325  * This is the OS-dependent handler for port module changes.  It is
1326  * invoked when a port module is removed or inserted for any OS-specific
1327  * processing.
1328  */
1329 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
1330 {
1331 	static const char * const mod_str[] = {
1332 		NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
1333 	};
1334 
1335 	const struct port_info *pi = adap2pinfo(adap, port_id);
1336 
1337 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
1338 		dev_info(adap, "Port%d: port module unplugged\n", pi->port_id);
1339 	else if (pi->mod_type < ARRAY_SIZE(mod_str))
1340 		dev_info(adap, "Port%d: %s port module inserted\n", pi->port_id,
1341 			 mod_str[pi->mod_type]);
1342 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
1343 		dev_info(adap, "Port%d: unsupported port module inserted\n",
1344 			 pi->port_id);
1345 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
1346 		dev_info(adap, "Port%d: unknown port module inserted\n",
1347 			 pi->port_id);
1348 	else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
1349 		dev_info(adap, "Port%d: transceiver module error\n",
1350 			 pi->port_id);
1351 	else
1352 		dev_info(adap, "Port%d: unknown module type %d inserted\n",
1353 			 pi->port_id, pi->mod_type);
1354 }
1355 
1356 bool cxgbe_force_linkup(struct adapter *adap)
1357 {
1358 	if (is_pf4(adap))
1359 		return false;	/* force_linkup not required for pf driver */
1360 
1361 	return adap->devargs.force_link_up;
1362 }
1363 
1364 /**
1365  * link_start - enable a port
1366  * @dev: the port to enable
1367  *
1368  * Performs the MAC and PHY actions needed to enable a port.
1369  */
1370 int cxgbe_link_start(struct port_info *pi)
1371 {
1372 	struct adapter *adapter = pi->adapter;
1373 	u64 conf_offloads;
1374 	unsigned int mtu;
1375 	int ret;
1376 
1377 	mtu = pi->eth_dev->data->dev_conf.rxmode.max_rx_pkt_len -
1378 	      (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN);
1379 
1380 	conf_offloads = pi->eth_dev->data->dev_conf.rxmode.offloads;
1381 
1382 	/*
1383 	 * We do not set address filters and promiscuity here, the stack does
1384 	 * that step explicitly.
1385 	 */
1386 	ret = t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu, -1, -1, -1,
1387 			    !!(conf_offloads & DEV_RX_OFFLOAD_VLAN_STRIP),
1388 			    true);
1389 	if (ret == 0) {
1390 		ret = cxgbe_mpstcam_modify(pi, (int)pi->xact_addr_filt,
1391 				(u8 *)&pi->eth_dev->data->mac_addrs[0]);
1392 		if (ret >= 0) {
1393 			pi->xact_addr_filt = ret;
1394 			ret = 0;
1395 		}
1396 	}
1397 	if (ret == 0 && is_pf4(adapter))
1398 		ret = t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
1399 				    &pi->link_cfg);
1400 	if (ret == 0) {
1401 		/*
1402 		 * Enabling a Virtual Interface can result in an interrupt
1403 		 * during the processing of the VI Enable command and, in some
1404 		 * paths, result in an attempt to issue another command in the
1405 		 * interrupt context.  Thus, we disable interrupts during the
1406 		 * course of the VI Enable command ...
1407 		 */
1408 		ret = t4_enable_vi_params(adapter, adapter->mbox, pi->viid,
1409 					  true, true, false);
1410 	}
1411 
1412 	if (ret == 0 && cxgbe_force_linkup(adapter))
1413 		pi->eth_dev->data->dev_link.link_status = ETH_LINK_UP;
1414 	return ret;
1415 }
1416 
1417 /**
1418  * cxgbe_write_rss_conf - flash the RSS configuration for a given port
1419  * @pi: the port
1420  * @rss_hf: Hash configuration to apply
1421  */
1422 int cxgbe_write_rss_conf(const struct port_info *pi, uint64_t rss_hf)
1423 {
1424 	struct adapter *adapter = pi->adapter;
1425 	const struct sge_eth_rxq *rxq;
1426 	u64 flags = 0;
1427 	u16 rss;
1428 	int err;
1429 
1430 	/*  Should never be called before setting up sge eth rx queues */
1431 	if (!(adapter->flags & FULL_INIT_DONE)) {
1432 		dev_err(adap, "%s No RXQs available on port %d\n",
1433 			__func__, pi->port_id);
1434 		return -EINVAL;
1435 	}
1436 
1437 	/* Don't allow unsupported hash functions */
1438 	if (rss_hf & ~CXGBE_RSS_HF_ALL)
1439 		return -EINVAL;
1440 
1441 	if (rss_hf & CXGBE_RSS_HF_IPV4_MASK)
1442 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
1443 
1444 	if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
1445 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
1446 
1447 	if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
1448 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
1449 			 F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1450 
1451 	if (rss_hf & CXGBE_RSS_HF_IPV6_MASK)
1452 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
1453 
1454 	if (rss_hf & CXGBE_RSS_HF_TCP_IPV6_MASK)
1455 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1456 			 F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
1457 
1458 	if (rss_hf & CXGBE_RSS_HF_UDP_IPV6_MASK)
1459 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1460 			 F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
1461 			 F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1462 
1463 	rxq = &adapter->sge.ethrxq[pi->first_qset];
1464 	rss = rxq[0].rspq.abs_id;
1465 
1466 	/* If Tunnel All Lookup isn't specified in the global RSS
1467 	 * Configuration, then we need to specify a default Ingress
1468 	 * Queue for any ingress packets which aren't hashed.  We'll
1469 	 * use our first ingress queue ...
1470 	 */
1471 	err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
1472 			       flags, rss);
1473 	return err;
1474 }
1475 
1476 /**
1477  * cxgbe_write_rss - write the RSS table for a given port
1478  * @pi: the port
1479  * @queues: array of queue indices for RSS
1480  *
1481  * Sets up the portion of the HW RSS table for the port's VI to distribute
1482  * packets to the Rx queues in @queues.
1483  */
1484 int cxgbe_write_rss(const struct port_info *pi, const u16 *queues)
1485 {
1486 	u16 *rss;
1487 	int i, err;
1488 	struct adapter *adapter = pi->adapter;
1489 	const struct sge_eth_rxq *rxq;
1490 
1491 	/*  Should never be called before setting up sge eth rx queues */
1492 	BUG_ON(!(adapter->flags & FULL_INIT_DONE));
1493 
1494 	rxq = &adapter->sge.ethrxq[pi->first_qset];
1495 	rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
1496 	if (!rss)
1497 		return -ENOMEM;
1498 
1499 	/* map the queue indices to queue ids */
1500 	for (i = 0; i < pi->rss_size; i++, queues++)
1501 		rss[i] = rxq[*queues].rspq.abs_id;
1502 
1503 	err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
1504 				  pi->rss_size, rss, pi->rss_size);
1505 	rte_free(rss);
1506 	return err;
1507 }
1508 
1509 /**
1510  * setup_rss - configure RSS
1511  * @adapter: the adapter
1512  *
1513  * Sets up RSS to distribute packets to multiple receive queues.  We
1514  * configure the RSS CPU lookup table to distribute to the number of HW
1515  * receive queues, and the response queue lookup table to narrow that
1516  * down to the response queues actually configured for each port.
1517  * We always configure the RSS mapping for all ports since the mapping
1518  * table has plenty of entries.
1519  */
1520 int cxgbe_setup_rss(struct port_info *pi)
1521 {
1522 	int j, err;
1523 	struct adapter *adapter = pi->adapter;
1524 
1525 	dev_debug(adapter, "%s:  pi->rss_size = %u; pi->n_rx_qsets = %u\n",
1526 		  __func__, pi->rss_size, pi->n_rx_qsets);
1527 
1528 	if (!(pi->flags & PORT_RSS_DONE)) {
1529 		if (adapter->flags & FULL_INIT_DONE) {
1530 			/* Fill default values with equal distribution */
1531 			for (j = 0; j < pi->rss_size; j++)
1532 				pi->rss[j] = j % pi->n_rx_qsets;
1533 
1534 			err = cxgbe_write_rss(pi, pi->rss);
1535 			if (err)
1536 				return err;
1537 
1538 			err = cxgbe_write_rss_conf(pi, pi->rss_hf);
1539 			if (err)
1540 				return err;
1541 			pi->flags |= PORT_RSS_DONE;
1542 		}
1543 	}
1544 	return 0;
1545 }
1546 
1547 /*
1548  * Enable NAPI scheduling and interrupt generation for all Rx queues.
1549  */
1550 static void enable_rx(struct adapter *adap, struct sge_rspq *q)
1551 {
1552 	/* 0-increment GTS to start the timer and enable interrupts */
1553 	t4_write_reg(adap, is_pf4(adap) ? MYPF_REG(A_SGE_PF_GTS) :
1554 					  T4VF_SGE_BASE_ADDR + A_SGE_VF_GTS,
1555 		     V_SEINTARM(q->intr_params) |
1556 		     V_INGRESSQID(q->cntxt_id));
1557 }
1558 
1559 void cxgbe_enable_rx_queues(struct port_info *pi)
1560 {
1561 	struct adapter *adap = pi->adapter;
1562 	struct sge *s = &adap->sge;
1563 	unsigned int i;
1564 
1565 	for (i = 0; i < pi->n_rx_qsets; i++)
1566 		enable_rx(adap, &s->ethrxq[pi->first_qset + i].rspq);
1567 }
1568 
1569 /**
1570  * fw_caps_to_speed_caps - translate Firmware Port Caps to Speed Caps.
1571  * @port_type: Firmware Port Type
1572  * @fw_caps: Firmware Port Capabilities
1573  * @speed_caps: Device Info Speed Capabilities
1574  *
1575  * Translate a Firmware Port Capabilities specification to Device Info
1576  * Speed Capabilities.
1577  */
1578 static void fw_caps_to_speed_caps(enum fw_port_type port_type,
1579 				  unsigned int fw_caps,
1580 				  u32 *speed_caps)
1581 {
1582 #define SET_SPEED(__speed_name) \
1583 	do { \
1584 		*speed_caps |= ETH_LINK_ ## __speed_name; \
1585 	} while (0)
1586 
1587 #define FW_CAPS_TO_SPEED(__fw_name) \
1588 	do { \
1589 		if (fw_caps & FW_PORT_CAP32_ ## __fw_name) \
1590 			SET_SPEED(__fw_name); \
1591 	} while (0)
1592 
1593 	switch (port_type) {
1594 	case FW_PORT_TYPE_BT_SGMII:
1595 	case FW_PORT_TYPE_BT_XFI:
1596 	case FW_PORT_TYPE_BT_XAUI:
1597 		FW_CAPS_TO_SPEED(SPEED_100M);
1598 		FW_CAPS_TO_SPEED(SPEED_1G);
1599 		FW_CAPS_TO_SPEED(SPEED_10G);
1600 		break;
1601 
1602 	case FW_PORT_TYPE_KX4:
1603 	case FW_PORT_TYPE_KX:
1604 	case FW_PORT_TYPE_FIBER_XFI:
1605 	case FW_PORT_TYPE_FIBER_XAUI:
1606 	case FW_PORT_TYPE_SFP:
1607 	case FW_PORT_TYPE_QSFP_10G:
1608 	case FW_PORT_TYPE_QSA:
1609 		FW_CAPS_TO_SPEED(SPEED_1G);
1610 		FW_CAPS_TO_SPEED(SPEED_10G);
1611 		break;
1612 
1613 	case FW_PORT_TYPE_KR:
1614 		SET_SPEED(SPEED_10G);
1615 		break;
1616 
1617 	case FW_PORT_TYPE_BP_AP:
1618 	case FW_PORT_TYPE_BP4_AP:
1619 		SET_SPEED(SPEED_1G);
1620 		SET_SPEED(SPEED_10G);
1621 		break;
1622 
1623 	case FW_PORT_TYPE_BP40_BA:
1624 	case FW_PORT_TYPE_QSFP:
1625 		SET_SPEED(SPEED_40G);
1626 		break;
1627 
1628 	case FW_PORT_TYPE_CR_QSFP:
1629 	case FW_PORT_TYPE_SFP28:
1630 	case FW_PORT_TYPE_KR_SFP28:
1631 		FW_CAPS_TO_SPEED(SPEED_1G);
1632 		FW_CAPS_TO_SPEED(SPEED_10G);
1633 		FW_CAPS_TO_SPEED(SPEED_25G);
1634 		break;
1635 
1636 	case FW_PORT_TYPE_CR2_QSFP:
1637 		SET_SPEED(SPEED_50G);
1638 		break;
1639 
1640 	case FW_PORT_TYPE_KR4_100G:
1641 	case FW_PORT_TYPE_CR4_QSFP:
1642 		FW_CAPS_TO_SPEED(SPEED_25G);
1643 		FW_CAPS_TO_SPEED(SPEED_40G);
1644 		FW_CAPS_TO_SPEED(SPEED_50G);
1645 		FW_CAPS_TO_SPEED(SPEED_100G);
1646 		break;
1647 
1648 	default:
1649 		break;
1650 	}
1651 
1652 #undef FW_CAPS_TO_SPEED
1653 #undef SET_SPEED
1654 }
1655 
1656 /**
1657  * cxgbe_get_speed_caps - Fetch supported speed capabilities
1658  * @pi: Underlying port's info
1659  * @speed_caps: Device Info speed capabilities
1660  *
1661  * Fetch supported speed capabilities of the underlying port.
1662  */
1663 void cxgbe_get_speed_caps(struct port_info *pi, u32 *speed_caps)
1664 {
1665 	*speed_caps = 0;
1666 
1667 	fw_caps_to_speed_caps(pi->port_type, pi->link_cfg.pcaps,
1668 			      speed_caps);
1669 
1670 	if (!(pi->link_cfg.pcaps & FW_PORT_CAP32_ANEG))
1671 		*speed_caps |= ETH_LINK_SPEED_FIXED;
1672 }
1673 
1674 /**
1675  * cxgbe_set_link_status - Set device link up or down.
1676  * @pi: Underlying port's info
1677  * @status: 0 - down, 1 - up
1678  *
1679  * Set the device link up or down.
1680  */
1681 int cxgbe_set_link_status(struct port_info *pi, bool status)
1682 {
1683 	struct adapter *adapter = pi->adapter;
1684 	int err = 0;
1685 
1686 	err = t4_enable_vi(adapter, adapter->mbox, pi->viid, status, status);
1687 	if (err) {
1688 		dev_err(adapter, "%s: disable_vi failed: %d\n", __func__, err);
1689 		return err;
1690 	}
1691 
1692 	if (!status)
1693 		t4_reset_link_config(adapter, pi->pidx);
1694 
1695 	return 0;
1696 }
1697 
1698 /**
1699  * cxgb_up - enable the adapter
1700  * @adap: adapter being enabled
1701  *
1702  * Called when the first port is enabled, this function performs the
1703  * actions necessary to make an adapter operational, such as completing
1704  * the initialization of HW modules, and enabling interrupts.
1705  */
1706 int cxgbe_up(struct adapter *adap)
1707 {
1708 	enable_rx(adap, &adap->sge.fw_evtq);
1709 	t4_sge_tx_monitor_start(adap);
1710 	if (is_pf4(adap))
1711 		t4_intr_enable(adap);
1712 	adap->flags |= FULL_INIT_DONE;
1713 
1714 	/* TODO: deadman watchdog ?? */
1715 	return 0;
1716 }
1717 
1718 /*
1719  * Close the port
1720  */
1721 int cxgbe_down(struct port_info *pi)
1722 {
1723 	return cxgbe_set_link_status(pi, false);
1724 }
1725 
1726 /*
1727  * Release resources when all the ports have been stopped.
1728  */
1729 void cxgbe_close(struct adapter *adapter)
1730 {
1731 	struct port_info *pi;
1732 	int i;
1733 
1734 	if (adapter->flags & FULL_INIT_DONE) {
1735 		tid_free(&adapter->tids);
1736 		t4_cleanup_mpstcam(adapter);
1737 		t4_cleanup_clip_tbl(adapter);
1738 		t4_cleanup_l2t(adapter);
1739 		t4_cleanup_smt(adapter);
1740 		if (is_pf4(adapter))
1741 			t4_intr_disable(adapter);
1742 		t4_sge_tx_monitor_stop(adapter);
1743 		t4_free_sge_resources(adapter);
1744 		for_each_port(adapter, i) {
1745 			pi = adap2pinfo(adapter, i);
1746 			if (pi->viid != 0)
1747 				t4_free_vi(adapter, adapter->mbox,
1748 					   adapter->pf, 0, pi->viid);
1749 			rte_eth_dev_release_port(pi->eth_dev);
1750 		}
1751 		adapter->flags &= ~FULL_INIT_DONE;
1752 	}
1753 
1754 	if (is_pf4(adapter) && (adapter->flags & FW_OK))
1755 		t4_fw_bye(adapter, adapter->mbox);
1756 }
1757 
1758 static void adap_smt_index(struct adapter *adapter, u32 *smt_start_idx,
1759 			   u32 *smt_size)
1760 {
1761 	u32 params[2], smt_val[2];
1762 	int ret;
1763 
1764 	params[0] = CXGBE_FW_PARAM_PFVF(GET_SMT_START);
1765 	params[1] = CXGBE_FW_PARAM_PFVF(GET_SMT_SIZE);
1766 
1767 	ret = t4_query_params(adapter, adapter->mbox, adapter->pf, 0,
1768 			      2, params, smt_val);
1769 
1770 	/* if FW doesn't recognize this command then set it to default setting
1771 	 * which is start index as 0 and size as 256.
1772 	 */
1773 	if (ret < 0) {
1774 		*smt_start_idx = 0;
1775 		*smt_size = SMT_SIZE;
1776 	} else {
1777 		*smt_start_idx = smt_val[0];
1778 		/* smt size can be zero, if nsmt is not yet configured in
1779 		 * the config file or set as zero, then configure all the
1780 		 * remaining entries to this PF itself.
1781 		 */
1782 		if (!smt_val[1])
1783 			*smt_size = SMT_SIZE - *smt_start_idx;
1784 		else
1785 			*smt_size = smt_val[1];
1786 	}
1787 }
1788 
1789 int cxgbe_probe(struct adapter *adapter)
1790 {
1791 	u32 smt_start_idx, smt_size;
1792 	struct port_info *pi;
1793 	int func, i;
1794 	int err = 0;
1795 	u32 whoami;
1796 	int chip;
1797 
1798 	whoami = t4_read_reg(adapter, A_PL_WHOAMI);
1799 	chip = t4_get_chip_type(adapter,
1800 			CHELSIO_PCI_ID_VER(adapter->pdev->id.device_id));
1801 	if (chip < 0)
1802 		return chip;
1803 
1804 	func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
1805 	       G_SOURCEPF(whoami) : G_T6_SOURCEPF(whoami);
1806 
1807 	adapter->mbox = func;
1808 	adapter->pf = func;
1809 
1810 	t4_os_lock_init(&adapter->mbox_lock);
1811 	TAILQ_INIT(&adapter->mbox_list);
1812 	t4_os_lock_init(&adapter->win0_lock);
1813 
1814 	err = t4_prep_adapter(adapter);
1815 	if (err)
1816 		return err;
1817 
1818 	setup_memwin(adapter);
1819 	err = adap_init0(adapter);
1820 	if (err) {
1821 		dev_err(adapter, "%s: Adapter initialization failed, error %d\n",
1822 			__func__, err);
1823 		goto out_free;
1824 	}
1825 
1826 	if (!is_t4(adapter->params.chip)) {
1827 		/*
1828 		 * The userspace doorbell BAR is split evenly into doorbell
1829 		 * regions, each associated with an egress queue.  If this
1830 		 * per-queue region is large enough (at least UDBS_SEG_SIZE)
1831 		 * then it can be used to submit a tx work request with an
1832 		 * implied doorbell.  Enable write combining on the BAR if
1833 		 * there is room for such work requests.
1834 		 */
1835 		int s_qpp, qpp, num_seg;
1836 
1837 		s_qpp = (S_QUEUESPERPAGEPF0 +
1838 			(S_QUEUESPERPAGEPF1 - S_QUEUESPERPAGEPF0) *
1839 			adapter->pf);
1840 		qpp = 1 << ((t4_read_reg(adapter,
1841 				A_SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp)
1842 				& M_QUEUESPERPAGEPF0);
1843 		num_seg = CXGBE_PAGE_SIZE / UDBS_SEG_SIZE;
1844 		if (qpp > num_seg)
1845 			dev_warn(adapter, "Incorrect SGE EGRESS QUEUES_PER_PAGE configuration, continuing in debug mode\n");
1846 
1847 		adapter->bar2 = (void *)adapter->pdev->mem_resource[2].addr;
1848 		if (!adapter->bar2) {
1849 			dev_err(adapter, "cannot map device bar2 region\n");
1850 			err = -ENOMEM;
1851 			goto out_free;
1852 		}
1853 		t4_write_reg(adapter, A_SGE_STAT_CFG, V_STATSOURCE_T5(7) |
1854 			     V_STATMODE(0));
1855 	}
1856 
1857 	for_each_port(adapter, i) {
1858 		const unsigned int numa_node = rte_socket_id();
1859 		char name[RTE_ETH_NAME_MAX_LEN];
1860 		struct rte_eth_dev *eth_dev;
1861 
1862 		snprintf(name, sizeof(name), "%s_%d",
1863 			 adapter->pdev->device.name, i);
1864 
1865 		if (i == 0) {
1866 			/* First port is already allocated by DPDK */
1867 			eth_dev = adapter->eth_dev;
1868 			goto allocate_mac;
1869 		}
1870 
1871 		/*
1872 		 * now do all data allocation - for eth_dev structure,
1873 		 * and internal (private) data for the remaining ports
1874 		 */
1875 
1876 		/* reserve an ethdev entry */
1877 		eth_dev = rte_eth_dev_allocate(name);
1878 		if (!eth_dev)
1879 			goto out_free;
1880 
1881 		eth_dev->data->dev_private =
1882 			rte_zmalloc_socket(name, sizeof(struct port_info),
1883 					   RTE_CACHE_LINE_SIZE, numa_node);
1884 		if (!eth_dev->data->dev_private)
1885 			goto out_free;
1886 
1887 allocate_mac:
1888 		pi = eth_dev->data->dev_private;
1889 		adapter->port[i] = pi;
1890 		pi->eth_dev = eth_dev;
1891 		pi->adapter = adapter;
1892 		pi->xact_addr_filt = -1;
1893 		pi->port_id = i;
1894 		pi->pidx = i;
1895 
1896 		pi->eth_dev->device = &adapter->pdev->device;
1897 		pi->eth_dev->dev_ops = adapter->eth_dev->dev_ops;
1898 		pi->eth_dev->tx_pkt_burst = adapter->eth_dev->tx_pkt_burst;
1899 		pi->eth_dev->rx_pkt_burst = adapter->eth_dev->rx_pkt_burst;
1900 
1901 		rte_eth_copy_pci_info(pi->eth_dev, adapter->pdev);
1902 
1903 		pi->eth_dev->data->mac_addrs = rte_zmalloc(name,
1904 							RTE_ETHER_ADDR_LEN, 0);
1905 		if (!pi->eth_dev->data->mac_addrs) {
1906 			dev_err(adapter, "%s: Mem allocation failed for storing mac addr, aborting\n",
1907 				__func__);
1908 			err = -1;
1909 			goto out_free;
1910 		}
1911 
1912 		if (i > 0) {
1913 			/* First port will be notified by upper layer */
1914 			rte_eth_dev_probing_finish(eth_dev);
1915 		}
1916 	}
1917 
1918 	if (adapter->flags & FW_OK) {
1919 		err = t4_port_init(adapter, adapter->mbox, adapter->pf, 0);
1920 		if (err) {
1921 			dev_err(adapter, "%s: t4_port_init failed with err %d\n",
1922 				__func__, err);
1923 			goto out_free;
1924 		}
1925 	}
1926 
1927 	cxgbe_cfg_queues(adapter->eth_dev);
1928 
1929 	cxgbe_print_adapter_info(adapter);
1930 	cxgbe_print_port_info(adapter);
1931 
1932 	adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
1933 					  adapter->clipt_end);
1934 	if (!adapter->clipt) {
1935 		/* We tolerate a lack of clip_table, giving up some
1936 		 * functionality
1937 		 */
1938 		dev_warn(adapter, "could not allocate CLIP. Continuing\n");
1939 	}
1940 
1941 	adap_smt_index(adapter, &smt_start_idx, &smt_size);
1942 	adapter->smt = t4_init_smt(smt_start_idx, smt_size);
1943 	if (!adapter->smt)
1944 		dev_warn(adapter, "could not allocate SMT, continuing\n");
1945 
1946 	adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
1947 	if (!adapter->l2t) {
1948 		/* We tolerate a lack of L2T, giving up some functionality */
1949 		dev_warn(adapter, "could not allocate L2T. Continuing\n");
1950 	}
1951 
1952 	if (tid_init(&adapter->tids) < 0) {
1953 		/* Disable filtering support */
1954 		dev_warn(adapter, "could not allocate TID table, "
1955 			 "filter support disabled. Continuing\n");
1956 	}
1957 
1958 	t4_os_lock_init(&adapter->flow_lock);
1959 
1960 	adapter->mpstcam = t4_init_mpstcam(adapter);
1961 	if (!adapter->mpstcam)
1962 		dev_warn(adapter, "could not allocate mps tcam table."
1963 			 " Continuing\n");
1964 
1965 	if (is_hashfilter(adapter)) {
1966 		if (t4_read_reg(adapter, A_LE_DB_CONFIG) & F_HASHEN) {
1967 			u32 hash_base, hash_reg;
1968 
1969 			hash_reg = A_LE_DB_TID_HASHBASE;
1970 			hash_base = t4_read_reg(adapter, hash_reg);
1971 			adapter->tids.hash_base = hash_base / 4;
1972 		}
1973 	} else {
1974 		/* Disable hash filtering support */
1975 		dev_warn(adapter,
1976 			 "Maskless filter support disabled. Continuing\n");
1977 	}
1978 
1979 	err = cxgbe_init_rss(adapter);
1980 	if (err)
1981 		goto out_free;
1982 
1983 	return 0;
1984 
1985 out_free:
1986 	for_each_port(adapter, i) {
1987 		pi = adap2pinfo(adapter, i);
1988 		if (pi->viid != 0)
1989 			t4_free_vi(adapter, adapter->mbox, adapter->pf,
1990 				   0, pi->viid);
1991 		rte_eth_dev_release_port(pi->eth_dev);
1992 	}
1993 
1994 	if (adapter->flags & FW_OK)
1995 		t4_fw_bye(adapter, adapter->mbox);
1996 	return -err;
1997 }
1998