xref: /dpdk/drivers/net/cxgbe/cxgbe_main.c (revision 592ab76f9f0f41993bebb44da85c37750a93ece9)
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 <sys/stat.h>
8 #include <stdio.h>
9 #include <errno.h>
10 #include <stdint.h>
11 #include <string.h>
12 #include <unistd.h>
13 #include <stdarg.h>
14 #include <inttypes.h>
15 #include <fcntl.h>
16 #include <netinet/in.h>
17 
18 #include <rte_byteorder.h>
19 #include <rte_common.h>
20 #include <rte_cycles.h>
21 #include <rte_interrupts.h>
22 #include <rte_log.h>
23 #include <rte_debug.h>
24 #include <rte_pci.h>
25 #include <rte_branch_prediction.h>
26 #include <rte_memory.h>
27 #include <rte_tailq.h>
28 #include <rte_eal.h>
29 #include <rte_alarm.h>
30 #include <rte_ether.h>
31 #include <ethdev_driver.h>
32 #include <ethdev_pci.h>
33 #include <rte_random.h>
34 #include <rte_dev.h>
35 #include <rte_kvargs.h>
36 
37 #include "base/common.h"
38 #include "base/t4_regs.h"
39 #include "base/t4_msg.h"
40 #include "cxgbe.h"
41 #include "cxgbe_pfvf.h"
42 #include "clip_tbl.h"
43 #include "l2t.h"
44 #include "smt.h"
45 #include "mps_tcam.h"
46 
47 static const u16 cxgbe_filter_mode_features[] = {
48 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE |
49 	 F_PROTOCOL | F_PORT),
50 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE |
51 	 F_PROTOCOL | F_FCOE),
52 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_TOS |
53 	 F_PORT),
54 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_TOS |
55 	 F_FCOE),
56 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_PORT |
57 	 F_FCOE),
58 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_TOS |
59 	 F_PORT | F_FCOE),
60 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VLAN |
61 	 F_FCOE),
62 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VNIC_ID |
63 	 F_FCOE),
64 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VLAN |
65 	 F_FCOE),
66 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VNIC_ID |
67 	 F_FCOE),
68 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_VLAN | F_PORT |
69 	 F_FCOE),
70 	(F_FRAGMENTATION | F_MPSHITTYPE | F_MACMATCH | F_VNIC_ID | F_PORT |
71 	 F_FCOE),
72 	(F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_PROTOCOL | F_TOS |
73 	 F_PORT | F_FCOE),
74 	(F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VLAN | F_PORT),
75 	(F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VLAN | F_FCOE),
76 	(F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VNIC_ID | F_PORT),
77 	(F_FRAGMENTATION | F_MPSHITTYPE | F_ETHERTYPE | F_VNIC_ID | F_FCOE),
78 	(F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VLAN | F_PORT),
79 	(F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VLAN | F_FCOE),
80 	(F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VNIC_ID |
81 	 F_PORT),
82 	(F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VNIC_ID |
83 	 F_FCOE),
84 	(F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_VLAN | F_PORT |
85 	 F_FCOE),
86 	(F_FRAGMENTATION | F_MPSHITTYPE | F_PROTOCOL | F_VNIC_ID | F_PORT |
87 	 F_FCOE),
88 	(F_FRAGMENTATION | F_MPSHITTYPE | F_TOS | F_VLAN | F_PORT | F_FCOE),
89 	(F_FRAGMENTATION | F_MPSHITTYPE | F_TOS | F_VNIC_ID | F_PORT | F_FCOE),
90 	(F_FRAGMENTATION | F_MPSHITTYPE | F_VLAN | F_VNIC_ID | F_FCOE),
91 	(F_FRAGMENTATION | F_MACMATCH | F_ETHERTYPE | F_PROTOCOL | F_PORT |
92 	 F_FCOE),
93 	(F_FRAGMENTATION | F_MACMATCH | F_ETHERTYPE | F_TOS | F_PORT | F_FCOE),
94 	(F_FRAGMENTATION | F_MACMATCH | F_PROTOCOL | F_VLAN | F_PORT | F_FCOE),
95 	(F_FRAGMENTATION | F_MACMATCH | F_PROTOCOL | F_VNIC_ID | F_PORT |
96 	 F_FCOE),
97 	(F_FRAGMENTATION | F_MACMATCH | F_TOS | F_VLAN | F_PORT | F_FCOE),
98 	(F_FRAGMENTATION | F_MACMATCH | F_TOS | F_VNIC_ID | F_PORT | F_FCOE),
99 	(F_FRAGMENTATION | F_ETHERTYPE | F_VLAN | F_PORT | F_FCOE),
100 	(F_FRAGMENTATION | F_ETHERTYPE | F_VNIC_ID | F_PORT | F_FCOE),
101 	(F_FRAGMENTATION | F_PROTOCOL | F_TOS | F_VLAN | F_FCOE),
102 	(F_FRAGMENTATION | F_PROTOCOL | F_TOS | F_VNIC_ID | F_FCOE),
103 	(F_FRAGMENTATION | F_VLAN | F_VNIC_ID | F_PORT | F_FCOE),
104 	(F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_PROTOCOL | F_PORT |
105 	 F_FCOE),
106 	(F_MPSHITTYPE | F_MACMATCH | F_ETHERTYPE | F_TOS | F_PORT | F_FCOE),
107 	(F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VLAN | F_PORT),
108 	(F_MPSHITTYPE | F_MACMATCH | F_PROTOCOL | F_VNIC_ID | F_PORT),
109 	(F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VLAN | F_PORT),
110 	(F_MPSHITTYPE | F_MACMATCH | F_TOS | F_VNIC_ID | F_PORT),
111 	(F_MPSHITTYPE | F_ETHERTYPE | F_VLAN | F_PORT | F_FCOE),
112 	(F_MPSHITTYPE | F_ETHERTYPE | F_VNIC_ID | F_PORT | F_FCOE),
113 	(F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VLAN | F_PORT | F_FCOE),
114 	(F_MPSHITTYPE | F_PROTOCOL | F_TOS | F_VNIC_ID | F_PORT | F_FCOE),
115 	(F_MPSHITTYPE | F_VLAN | F_VNIC_ID | F_PORT),
116 };
117 
118 /**
119  * Allocate a chunk of memory. The allocated memory is cleared.
120  */
121 void *t4_alloc_mem(size_t size)
122 {
123 	return rte_zmalloc(NULL, size, 0);
124 }
125 
126 /**
127  * Free memory allocated through t4_alloc_mem().
128  */
129 void t4_free_mem(void *addr)
130 {
131 	rte_free(addr);
132 }
133 
134 /*
135  * Response queue handler for the FW event queue.
136  */
137 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
138 			  __rte_unused const struct pkt_gl *gl)
139 {
140 	u8 opcode = ((const struct rss_header *)rsp)->opcode;
141 
142 	rsp++;                                          /* skip RSS header */
143 
144 	/*
145 	 * FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
146 	 */
147 	if (unlikely(opcode == CPL_FW4_MSG &&
148 		     ((const struct cpl_fw4_msg *)rsp)->type ==
149 		      FW_TYPE_RSSCPL)) {
150 		rsp++;
151 		opcode = ((const struct rss_header *)rsp)->opcode;
152 		rsp++;
153 		if (opcode != CPL_SGE_EGR_UPDATE) {
154 			dev_err(q->adapter, "unexpected FW4/CPL %#x on FW event queue\n",
155 				opcode);
156 			goto out;
157 		}
158 	}
159 
160 	if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
161 		/* do nothing */
162 	} else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
163 		const struct cpl_fw6_msg *msg = (const void *)rsp;
164 
165 		t4_handle_fw_rpl(q->adapter, msg->data);
166 	} else if (opcode == CPL_ABORT_RPL_RSS) {
167 		const struct cpl_abort_rpl_rss *p = (const void *)rsp;
168 
169 		cxgbe_hash_del_filter_rpl(q->adapter, p);
170 	} else if (opcode == CPL_SET_TCB_RPL) {
171 		const struct cpl_set_tcb_rpl *p = (const void *)rsp;
172 
173 		cxgbe_filter_rpl(q->adapter, p);
174 	} else if (opcode == CPL_ACT_OPEN_RPL) {
175 		const struct cpl_act_open_rpl *p = (const void *)rsp;
176 
177 		cxgbe_hash_filter_rpl(q->adapter, p);
178 	} else if (opcode == CPL_L2T_WRITE_RPL) {
179 		const struct cpl_l2t_write_rpl *p = (const void *)rsp;
180 
181 		cxgbe_do_l2t_write_rpl(q->adapter, p);
182 	} else if (opcode == CPL_SMT_WRITE_RPL) {
183 		const struct cpl_smt_write_rpl *p = (const void *)rsp;
184 
185 		cxgbe_do_smt_write_rpl(q->adapter, p);
186 	} else {
187 		dev_err(adapter, "unexpected CPL %#x on FW event queue\n",
188 			opcode);
189 	}
190 out:
191 	return 0;
192 }
193 
194 /**
195  * Setup sge control queues to pass control information.
196  */
197 int cxgbe_setup_sge_ctrl_txq(struct adapter *adapter)
198 {
199 	struct sge *s = &adapter->sge;
200 	int err = 0, i = 0;
201 
202 	for_each_port(adapter, i) {
203 		struct port_info *pi = adap2pinfo(adapter, i);
204 		char name[RTE_ETH_NAME_MAX_LEN];
205 		struct sge_ctrl_txq *q = &s->ctrlq[i];
206 
207 		q->q.size = 1024;
208 		err = t4_sge_alloc_ctrl_txq(adapter, q,
209 					    adapter->eth_dev,  i,
210 					    s->fw_evtq.cntxt_id,
211 					    rte_socket_id());
212 		if (err) {
213 			dev_err(adapter, "Failed to alloc ctrl txq. Err: %d",
214 				err);
215 			goto out;
216 		}
217 		snprintf(name, sizeof(name), "%s_ctrl_pool_%d",
218 			 pi->eth_dev->device->driver->name,
219 			 pi->eth_dev->data->port_id);
220 		q->mb_pool = rte_pktmbuf_pool_create(name, s->ctrlq[i].q.size,
221 						     RTE_CACHE_LINE_SIZE,
222 						     RTE_MBUF_PRIV_ALIGN,
223 						     RTE_MBUF_DEFAULT_BUF_SIZE,
224 						     SOCKET_ID_ANY);
225 		if (!q->mb_pool) {
226 			err = -rte_errno;
227 			dev_err(adapter,
228 				"Can't create ctrl pool for port %d. Err: %d\n",
229 				pi->eth_dev->data->port_id, err);
230 			goto out;
231 		}
232 	}
233 	return 0;
234 out:
235 	t4_free_sge_resources(adapter);
236 	return err;
237 }
238 
239 /**
240  * cxgbe_poll_for_completion: Poll rxq for completion
241  * @q: rxq to poll
242  * @ms: milliseconds to delay
243  * @cnt: number of times to poll
244  * @c: completion to check for 'done' status
245  *
246  * Polls the rxq for reples until completion is done or the count
247  * expires.
248  */
249 int cxgbe_poll_for_completion(struct sge_rspq *q, unsigned int ms,
250 			      unsigned int cnt, struct t4_completion *c)
251 {
252 	unsigned int i;
253 	unsigned int work_done, budget = 32;
254 
255 	if (!c)
256 		return -EINVAL;
257 
258 	for (i = 0; i < cnt; i++) {
259 		cxgbe_poll(q, NULL, budget, &work_done);
260 		t4_os_lock(&c->lock);
261 		if (c->done) {
262 			t4_os_unlock(&c->lock);
263 			return 0;
264 		}
265 		t4_os_unlock(&c->lock);
266 		rte_delay_ms(ms);
267 	}
268 	return -ETIMEDOUT;
269 }
270 
271 int cxgbe_setup_sge_fwevtq(struct adapter *adapter)
272 {
273 	struct sge *s = &adapter->sge;
274 	int err = 0;
275 	int msi_idx = 0;
276 
277 	err = t4_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->eth_dev,
278 			       msi_idx, NULL, fwevtq_handler, -1, NULL, 0,
279 			       rte_socket_id());
280 	return err;
281 }
282 
283 static int closest_timer(const struct sge *s, int time)
284 {
285 	unsigned int i, match = 0;
286 	int delta, min_delta = INT_MAX;
287 
288 	for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
289 		delta = time - s->timer_val[i];
290 		if (delta < 0)
291 			delta = -delta;
292 		if (delta < min_delta) {
293 			min_delta = delta;
294 			match = i;
295 		}
296 	}
297 	return match;
298 }
299 
300 static int closest_thres(const struct sge *s, int thres)
301 {
302 	unsigned int i, match = 0;
303 	int delta, min_delta = INT_MAX;
304 
305 	for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
306 		delta = thres - s->counter_val[i];
307 		if (delta < 0)
308 			delta = -delta;
309 		if (delta < min_delta) {
310 			min_delta = delta;
311 			match = i;
312 		}
313 	}
314 	return match;
315 }
316 
317 /**
318  * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
319  * @q: the Rx queue
320  * @us: the hold-off time in us, or 0 to disable timer
321  * @cnt: the hold-off packet count, or 0 to disable counter
322  *
323  * Sets an Rx queue's interrupt hold-off time and packet count.  At least
324  * one of the two needs to be enabled for the queue to generate interrupts.
325  */
326 int cxgb4_set_rspq_intr_params(struct sge_rspq *q, unsigned int us,
327 			       unsigned int cnt)
328 {
329 	struct adapter *adap = q->adapter;
330 	unsigned int timer_val;
331 
332 	if (cnt) {
333 		int err;
334 		u32 v, new_idx;
335 
336 		new_idx = closest_thres(&adap->sge, cnt);
337 		if (q->desc && q->pktcnt_idx != new_idx) {
338 			/* the queue has already been created, update it */
339 			v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
340 			    V_FW_PARAMS_PARAM_X(
341 			    FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
342 			    V_FW_PARAMS_PARAM_YZ(q->cntxt_id);
343 			err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
344 					    &v, &new_idx);
345 			if (err)
346 				return err;
347 		}
348 		q->pktcnt_idx = new_idx;
349 	}
350 
351 	timer_val = (us == 0) ? X_TIMERREG_RESTART_COUNTER :
352 				closest_timer(&adap->sge, us);
353 
354 	if ((us | cnt) == 0)
355 		q->intr_params = V_QINTR_TIMER_IDX(X_TIMERREG_UPDATE_CIDX);
356 	else
357 		q->intr_params = V_QINTR_TIMER_IDX(timer_val) |
358 				 V_QINTR_CNT_EN(cnt > 0);
359 	return 0;
360 }
361 
362 /**
363  * Allocate an active-open TID and set it to the supplied value.
364  */
365 int cxgbe_alloc_atid(struct tid_info *t, void *data)
366 {
367 	int atid = -1;
368 
369 	t4_os_lock(&t->atid_lock);
370 	if (t->afree) {
371 		union aopen_entry *p = t->afree;
372 
373 		atid = p - t->atid_tab;
374 		t->afree = p->next;
375 		p->data = data;
376 		t->atids_in_use++;
377 	}
378 	t4_os_unlock(&t->atid_lock);
379 	return atid;
380 }
381 
382 /**
383  * Release an active-open TID.
384  */
385 void cxgbe_free_atid(struct tid_info *t, unsigned int atid)
386 {
387 	union aopen_entry *p = &t->atid_tab[atid];
388 
389 	t4_os_lock(&t->atid_lock);
390 	p->next = t->afree;
391 	t->afree = p;
392 	t->atids_in_use--;
393 	t4_os_unlock(&t->atid_lock);
394 }
395 
396 /**
397  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
398  */
399 static void mk_tid_release(struct rte_mbuf *mbuf, unsigned int tid)
400 {
401 	struct cpl_tid_release *req;
402 
403 	req = rte_pktmbuf_mtod(mbuf, struct cpl_tid_release *);
404 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
405 }
406 
407 /**
408  * Release a TID and inform HW.  If we are unable to allocate the release
409  * message we defer to a work queue.
410  */
411 void cxgbe_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
412 		      unsigned short family)
413 {
414 	struct rte_mbuf *mbuf;
415 	struct adapter *adap = container_of(t, struct adapter, tids);
416 
417 	WARN_ON(tid >= t->ntids);
418 
419 	if (t->tid_tab[tid]) {
420 		t->tid_tab[tid] = NULL;
421 		__atomic_sub_fetch(&t->conns_in_use, 1, __ATOMIC_RELAXED);
422 		if (t->hash_base && tid >= t->hash_base) {
423 			if (family == FILTER_TYPE_IPV4)
424 				__atomic_sub_fetch(&t->hash_tids_in_use, 1,
425 						   __ATOMIC_RELAXED);
426 		} else {
427 			if (family == FILTER_TYPE_IPV4)
428 				__atomic_sub_fetch(&t->tids_in_use, 1,
429 						   __ATOMIC_RELAXED);
430 		}
431 	}
432 
433 	mbuf = rte_pktmbuf_alloc((&adap->sge.ctrlq[chan])->mb_pool);
434 	if (mbuf) {
435 		mbuf->data_len = sizeof(struct cpl_tid_release);
436 		mbuf->pkt_len = mbuf->data_len;
437 		mk_tid_release(mbuf, tid);
438 		t4_mgmt_tx(&adap->sge.ctrlq[chan], mbuf);
439 	}
440 }
441 
442 /**
443  * Insert a TID.
444  */
445 void cxgbe_insert_tid(struct tid_info *t, void *data, unsigned int tid,
446 		      unsigned short family)
447 {
448 	t->tid_tab[tid] = data;
449 	if (t->hash_base && tid >= t->hash_base) {
450 		if (family == FILTER_TYPE_IPV4)
451 			__atomic_add_fetch(&t->hash_tids_in_use, 1,
452 					   __ATOMIC_RELAXED);
453 	} else {
454 		if (family == FILTER_TYPE_IPV4)
455 			__atomic_add_fetch(&t->tids_in_use, 1,
456 					   __ATOMIC_RELAXED);
457 	}
458 
459 	__atomic_add_fetch(&t->conns_in_use, 1, __ATOMIC_RELAXED);
460 }
461 
462 /**
463  * Free TID tables.
464  */
465 static void tid_free(struct tid_info *t)
466 {
467 	if (t->tid_tab) {
468 		rte_bitmap_free(t->ftid_bmap);
469 
470 		if (t->ftid_bmap_array)
471 			t4_os_free(t->ftid_bmap_array);
472 
473 		t4_os_free(t->tid_tab);
474 	}
475 
476 	memset(t, 0, sizeof(struct tid_info));
477 }
478 
479 /**
480  * Allocate and initialize the TID tables.  Returns 0 on success.
481  */
482 static int tid_init(struct tid_info *t)
483 {
484 	size_t size;
485 	unsigned int ftid_bmap_size;
486 	unsigned int natids = t->natids;
487 	unsigned int max_ftids = t->nftids;
488 
489 	ftid_bmap_size = rte_bitmap_get_memory_footprint(t->nftids);
490 	size = t->ntids * sizeof(*t->tid_tab) +
491 		max_ftids * sizeof(*t->ftid_tab) +
492 		natids * sizeof(*t->atid_tab);
493 
494 	t->tid_tab = t4_os_alloc(size);
495 	if (!t->tid_tab)
496 		return -ENOMEM;
497 
498 	t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
499 	t->ftid_tab = (struct filter_entry *)&t->atid_tab[t->natids];
500 	t->ftid_bmap_array = t4_os_alloc(ftid_bmap_size);
501 	if (!t->ftid_bmap_array) {
502 		tid_free(t);
503 		return -ENOMEM;
504 	}
505 
506 	t4_os_lock_init(&t->atid_lock);
507 	t4_os_lock_init(&t->ftid_lock);
508 
509 	t->afree = NULL;
510 	t->atids_in_use = 0;
511 	t->tids_in_use = 0;
512 	t->conns_in_use = 0;
513 
514 	/* Setup the free list for atid_tab and clear the stid bitmap. */
515 	if (natids) {
516 		while (--natids)
517 			t->atid_tab[natids - 1].next = &t->atid_tab[natids];
518 		t->afree = t->atid_tab;
519 	}
520 
521 	t->ftid_bmap = rte_bitmap_init(t->nftids, t->ftid_bmap_array,
522 				       ftid_bmap_size);
523 	if (!t->ftid_bmap) {
524 		tid_free(t);
525 		return -ENOMEM;
526 	}
527 
528 	return 0;
529 }
530 
531 static inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
532 		      unsigned int us, unsigned int cnt,
533 		      unsigned int size, unsigned int iqe_size)
534 {
535 	q->adapter = adap;
536 	cxgb4_set_rspq_intr_params(q, us, cnt);
537 	q->iqe_len = iqe_size;
538 	q->size = size;
539 }
540 
541 int cxgbe_cfg_queue_count(struct rte_eth_dev *eth_dev)
542 {
543 	struct port_info *temp_pi, *pi = eth_dev->data->dev_private;
544 	struct adapter *adap = pi->adapter;
545 	u16 first_txq = 0, first_rxq = 0;
546 	struct sge *s = &adap->sge;
547 	u16 i, max_rxqs, max_txqs;
548 
549 	max_rxqs = s->max_ethqsets;
550 	max_txqs = s->max_ethqsets;
551 	for_each_port(adap, i) {
552 		temp_pi = adap2pinfo(adap, i);
553 		if (i == pi->port_id)
554 			break;
555 
556 		if (max_rxqs <= temp_pi->n_rx_qsets ||
557 		    max_txqs <= temp_pi->n_tx_qsets)
558 			return -ENOMEM;
559 
560 		first_rxq += temp_pi->n_rx_qsets;
561 		first_txq += temp_pi->n_tx_qsets;
562 		max_rxqs -= temp_pi->n_rx_qsets;
563 		max_txqs -= temp_pi->n_tx_qsets;
564 	}
565 
566 	if ((eth_dev->data->nb_rx_queues < 1) ||
567 	    (eth_dev->data->nb_tx_queues < 1))
568 		return -EINVAL;
569 
570 	if (eth_dev->data->nb_rx_queues > max_rxqs ||
571 	    eth_dev->data->nb_tx_queues > max_txqs)
572 		return -EINVAL;
573 
574 	/* We must configure RSS, since config has changed*/
575 	pi->flags &= ~PORT_RSS_DONE;
576 
577 	pi->n_rx_qsets = eth_dev->data->nb_rx_queues;
578 	pi->n_tx_qsets = eth_dev->data->nb_tx_queues;
579 	pi->first_rxqset = first_rxq;
580 	pi->first_txqset = first_txq;
581 
582 	return 0;
583 }
584 
585 void cxgbe_cfg_queues_free(struct adapter *adap)
586 {
587 	if (adap->sge.ethtxq) {
588 		rte_free(adap->sge.ethtxq);
589 		adap->sge.ethtxq = NULL;
590 	}
591 
592 	if (adap->sge.ethrxq) {
593 		rte_free(adap->sge.ethrxq);
594 		adap->sge.ethrxq = NULL;
595 	}
596 
597 	adap->flags &= ~CFG_QUEUES;
598 }
599 
600 int cxgbe_cfg_queues(struct rte_eth_dev *eth_dev)
601 {
602 	struct port_info *pi = eth_dev->data->dev_private;
603 	struct adapter *adap = pi->adapter;
604 	struct sge *s = &adap->sge;
605 	u16 i;
606 
607 	if (!(adap->flags & CFG_QUEUES)) {
608 		s->ethrxq = rte_calloc_socket(NULL, s->max_ethqsets,
609 					      sizeof(struct sge_eth_rxq), 0,
610 					      rte_socket_id());
611 		if (!s->ethrxq)
612 			return -ENOMEM;
613 
614 		s->ethtxq = rte_calloc_socket(NULL, s->max_ethqsets,
615 					      sizeof(struct sge_eth_txq), 0,
616 					      rte_socket_id());
617 		if (!s->ethtxq) {
618 			rte_free(s->ethrxq);
619 			s->ethrxq = NULL;
620 			return -ENOMEM;
621 		}
622 
623 		for (i = 0; i < s->max_ethqsets; i++) {
624 			struct sge_eth_rxq *r = &s->ethrxq[i];
625 			struct sge_eth_txq *t = &s->ethtxq[i];
626 
627 			init_rspq(adap, &r->rspq, 5, 32, 1024, 64);
628 			r->fl.size = 1024;
629 
630 			t->q.size = 1024;
631 		}
632 
633 		init_rspq(adap, &adap->sge.fw_evtq, 0, 0, 1024, 64);
634 		adap->flags |= CFG_QUEUES;
635 	}
636 
637 	return 0;
638 }
639 
640 void cxgbe_stats_get(struct port_info *pi, struct port_stats *stats)
641 {
642 	t4_get_port_stats_offset(pi->adapter, pi->tx_chan, stats,
643 				 &pi->stats_base);
644 }
645 
646 void cxgbe_stats_reset(struct port_info *pi)
647 {
648 	t4_clr_port_stats(pi->adapter, pi->tx_chan);
649 }
650 
651 static void setup_memwin(struct adapter *adap)
652 {
653 	u32 mem_win0_base;
654 
655 	/* For T5, only relative offset inside the PCIe BAR is passed */
656 	mem_win0_base = MEMWIN0_BASE;
657 
658 	/*
659 	 * Set up memory window for accessing adapter memory ranges.  (Read
660 	 * back MA register to ensure that changes propagate before we attempt
661 	 * to use the new values.)
662 	 */
663 	t4_write_reg(adap,
664 		     PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
665 					 MEMWIN_NIC),
666 		     mem_win0_base | V_BIR(0) |
667 		     V_WINDOW(ilog2(MEMWIN0_APERTURE) - X_WINDOW_SHIFT));
668 	t4_read_reg(adap,
669 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
670 					MEMWIN_NIC));
671 }
672 
673 int cxgbe_init_rss(struct adapter *adap)
674 {
675 	unsigned int i;
676 
677 	if (is_pf4(adap)) {
678 		int err;
679 
680 		err = t4_init_rss_mode(adap, adap->mbox);
681 		if (err)
682 			return err;
683 	}
684 
685 	for_each_port(adap, i) {
686 		struct port_info *pi = adap2pinfo(adap, i);
687 
688 		pi->rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
689 		if (!pi->rss)
690 			return -ENOMEM;
691 
692 		pi->rss_hf = CXGBE_RSS_HF_ALL;
693 	}
694 	return 0;
695 }
696 
697 /**
698  * Dump basic information about the adapter.
699  */
700 void cxgbe_print_adapter_info(struct adapter *adap)
701 {
702 	/**
703 	 * Hardware/Firmware/etc. Version/Revision IDs.
704 	 */
705 	t4_dump_version_info(adap);
706 }
707 
708 void cxgbe_print_port_info(struct adapter *adap)
709 {
710 	int i;
711 	char buf[80];
712 	struct rte_pci_addr *loc = &adap->pdev->addr;
713 
714 	for_each_port(adap, i) {
715 		const struct port_info *pi = adap2pinfo(adap, i);
716 		char *bufp = buf;
717 
718 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
719 			bufp += sprintf(bufp, "100M/");
720 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
721 			bufp += sprintf(bufp, "1G/");
722 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
723 			bufp += sprintf(bufp, "10G/");
724 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
725 			bufp += sprintf(bufp, "25G/");
726 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
727 			bufp += sprintf(bufp, "40G/");
728 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
729 			bufp += sprintf(bufp, "50G/");
730 		if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
731 			bufp += sprintf(bufp, "100G/");
732 		if (bufp != buf)
733 			--bufp;
734 		sprintf(bufp, "BASE-%s",
735 			t4_get_port_type_description(
736 				(enum fw_port_type)pi->link_cfg.port_type));
737 
738 		dev_info(adap,
739 			 " " PCI_PRI_FMT " Chelsio rev %d %s %s\n",
740 			 loc->domain, loc->bus, loc->devid, loc->function,
741 			 CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
742 			 (adap->flags & USING_MSIX) ? " MSI-X" :
743 			 (adap->flags & USING_MSI) ? " MSI" : "");
744 	}
745 }
746 
747 static int check_devargs_handler(const char *key, const char *value, void *p)
748 {
749 	if (!strncmp(key, CXGBE_DEVARG_CMN_KEEP_OVLAN, strlen(key)) ||
750 	    !strncmp(key, CXGBE_DEVARG_CMN_TX_MODE_LATENCY, strlen(key)) ||
751 	    !strncmp(key, CXGBE_DEVARG_VF_FORCE_LINK_UP, strlen(key))) {
752 		if (!strncmp(value, "1", 1)) {
753 			bool *dst_val = (bool *)p;
754 
755 			*dst_val = true;
756 		}
757 	}
758 
759 	if (!strncmp(key, CXGBE_DEVARG_PF_FILTER_MODE, strlen(key)) ||
760 	    !strncmp(key, CXGBE_DEVARG_PF_FILTER_MASK, strlen(key))) {
761 		u32 *dst_val = (u32 *)p;
762 		char *endptr = NULL;
763 		u32 arg_val;
764 
765 		arg_val = strtoul(value, &endptr, 16);
766 		if (errno || endptr == value)
767 			return -EINVAL;
768 
769 		*dst_val = arg_val;
770 	}
771 
772 	return 0;
773 }
774 
775 static int cxgbe_get_devargs(struct rte_devargs *devargs, const char *key,
776 			     void *p)
777 {
778 	struct rte_kvargs *kvlist;
779 	int ret = 0;
780 
781 	if (!devargs)
782 		return 0;
783 
784 	kvlist = rte_kvargs_parse(devargs->args, NULL);
785 	if (!kvlist)
786 		return 0;
787 
788 	if (!rte_kvargs_count(kvlist, key))
789 		goto out;
790 
791 	ret = rte_kvargs_process(kvlist, key, check_devargs_handler, p);
792 
793 out:
794 	rte_kvargs_free(kvlist);
795 
796 	return ret;
797 }
798 
799 static void cxgbe_get_devargs_int(struct adapter *adap, bool *dst,
800 				  const char *key, bool default_value)
801 {
802 	struct rte_pci_device *pdev = adap->pdev;
803 	int ret;
804 	bool devarg_value = default_value;
805 
806 	*dst = default_value;
807 	if (!pdev)
808 		return;
809 
810 	ret = cxgbe_get_devargs(pdev->device.devargs, key, &devarg_value);
811 	if (ret)
812 		return;
813 
814 	*dst = devarg_value;
815 }
816 
817 static void cxgbe_get_devargs_u32(struct adapter *adap, u32 *dst,
818 				  const char *key, u32 default_value)
819 {
820 	struct rte_pci_device *pdev = adap->pdev;
821 	u32 devarg_value = default_value;
822 	int ret;
823 
824 	*dst = default_value;
825 	if (!pdev)
826 		return;
827 
828 	ret = cxgbe_get_devargs(pdev->device.devargs, key, &devarg_value);
829 	if (ret)
830 		return;
831 
832 	*dst = devarg_value;
833 }
834 
835 void cxgbe_process_devargs(struct adapter *adap)
836 {
837 	cxgbe_get_devargs_int(adap, &adap->devargs.keep_ovlan,
838 			      CXGBE_DEVARG_CMN_KEEP_OVLAN, false);
839 	cxgbe_get_devargs_int(adap, &adap->devargs.tx_mode_latency,
840 			      CXGBE_DEVARG_CMN_TX_MODE_LATENCY, false);
841 	cxgbe_get_devargs_int(adap, &adap->devargs.force_link_up,
842 			      CXGBE_DEVARG_VF_FORCE_LINK_UP, false);
843 	cxgbe_get_devargs_u32(adap, &adap->devargs.filtermode,
844 			      CXGBE_DEVARG_PF_FILTER_MODE, 0);
845 	cxgbe_get_devargs_u32(adap, &adap->devargs.filtermask,
846 			      CXGBE_DEVARG_PF_FILTER_MASK, 0);
847 }
848 
849 static void configure_vlan_types(struct adapter *adapter)
850 {
851 	int i;
852 
853 	for_each_port(adapter, i) {
854 		/* OVLAN Type 0x88a8 */
855 		t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN0),
856 				 V_OVLAN_MASK(M_OVLAN_MASK) |
857 				 V_OVLAN_ETYPE(M_OVLAN_ETYPE),
858 				 V_OVLAN_MASK(M_OVLAN_MASK) |
859 				 V_OVLAN_ETYPE(0x88a8));
860 		/* OVLAN Type 0x9100 */
861 		t4_set_reg_field(adapter, MPS_PORT_RX_OVLAN_REG(i, A_RX_OVLAN1),
862 				 V_OVLAN_MASK(M_OVLAN_MASK) |
863 				 V_OVLAN_ETYPE(M_OVLAN_ETYPE),
864 				 V_OVLAN_MASK(M_OVLAN_MASK) |
865 				 V_OVLAN_ETYPE(0x9100));
866 
867 		/* IVLAN 0X8100 */
868 		t4_set_reg_field(adapter, MPS_PORT_RX_IVLAN(i),
869 				 V_IVLAN_ETYPE(M_IVLAN_ETYPE),
870 				 V_IVLAN_ETYPE(0x8100));
871 
872 		t4_set_reg_field(adapter, MPS_PORT_RX_CTL(i),
873 				 F_OVLAN_EN0 | F_OVLAN_EN1 |
874 				 F_IVLAN_EN,
875 				 F_OVLAN_EN0 | F_OVLAN_EN1 |
876 				 F_IVLAN_EN);
877 	}
878 
879 	t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG, V_RM_OVLAN(1),
880 			       V_RM_OVLAN(!adapter->devargs.keep_ovlan));
881 }
882 
883 static int cxgbe_get_filter_vnic_mode_from_devargs(u32 val)
884 {
885 	u32 vnic_mode;
886 
887 	vnic_mode = val & (CXGBE_DEVARGS_FILTER_MODE_PF_VF |
888 			   CXGBE_DEVARGS_FILTER_MODE_VLAN_OUTER);
889 	if (vnic_mode) {
890 		switch (vnic_mode) {
891 		case CXGBE_DEVARGS_FILTER_MODE_VLAN_OUTER:
892 			return CXGBE_FILTER_VNIC_MODE_OVLAN;
893 		case CXGBE_DEVARGS_FILTER_MODE_PF_VF:
894 			return CXGBE_FILTER_VNIC_MODE_PFVF;
895 		default:
896 			return -EINVAL;
897 		}
898 	}
899 
900 	return CXGBE_FILTER_VNIC_MODE_NONE;
901 }
902 
903 static int cxgbe_get_filter_mode_from_devargs(u32 val, bool closest_match)
904 {
905 	int vnic_mode, fmode = 0;
906 	bool found = false;
907 	u8 i;
908 
909 	if (val >= CXGBE_DEVARGS_FILTER_MODE_MAX) {
910 		pr_err("Unsupported flags set in filter mode. Must be < 0x%x\n",
911 		       CXGBE_DEVARGS_FILTER_MODE_MAX);
912 		return -ERANGE;
913 	}
914 
915 	vnic_mode = cxgbe_get_filter_vnic_mode_from_devargs(val);
916 	if (vnic_mode < 0) {
917 		pr_err("Unsupported Vnic-mode, more than 1 Vnic-mode selected\n");
918 		return vnic_mode;
919 	}
920 
921 	if (vnic_mode)
922 		fmode |= F_VNIC_ID;
923 	if (val & CXGBE_DEVARGS_FILTER_MODE_PHYSICAL_PORT)
924 		fmode |= F_PORT;
925 	if (val & CXGBE_DEVARGS_FILTER_MODE_ETHERNET_DSTMAC)
926 		fmode |= F_MACMATCH;
927 	if (val & CXGBE_DEVARGS_FILTER_MODE_ETHERNET_ETHTYPE)
928 		fmode |= F_ETHERTYPE;
929 	if (val & CXGBE_DEVARGS_FILTER_MODE_VLAN_INNER)
930 		fmode |= F_VLAN;
931 	if (val & CXGBE_DEVARGS_FILTER_MODE_IP_TOS)
932 		fmode |= F_TOS;
933 	if (val & CXGBE_DEVARGS_FILTER_MODE_IP_PROTOCOL)
934 		fmode |= F_PROTOCOL;
935 
936 	for (i = 0; i < ARRAY_SIZE(cxgbe_filter_mode_features); i++) {
937 		if ((cxgbe_filter_mode_features[i] & fmode) == fmode) {
938 			found = true;
939 			break;
940 		}
941 	}
942 
943 	if (!found)
944 		return -EINVAL;
945 
946 	return closest_match ? cxgbe_filter_mode_features[i] : fmode;
947 }
948 
949 static int configure_filter_mode_mask(struct adapter *adap)
950 {
951 	u32 params[2], val[2], nparams = 0;
952 	int ret;
953 
954 	if (!adap->devargs.filtermode && !adap->devargs.filtermask)
955 		return 0;
956 
957 	if (!adap->devargs.filtermode || !adap->devargs.filtermask) {
958 		pr_err("Unsupported, Provide both filtermode and filtermask devargs\n");
959 		return -EINVAL;
960 	}
961 
962 	if (adap->devargs.filtermask & ~adap->devargs.filtermode) {
963 		pr_err("Unsupported, filtermask (0x%x) must be subset of filtermode (0x%x)\n",
964 		       adap->devargs.filtermask, adap->devargs.filtermode);
965 
966 		return -EINVAL;
967 	}
968 
969 	params[0] = CXGBE_FW_PARAM_DEV(FILTER) |
970 		    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_FILTER_MODE_MASK);
971 
972 	ret = cxgbe_get_filter_mode_from_devargs(adap->devargs.filtermode,
973 						 true);
974 	if (ret < 0) {
975 		pr_err("Unsupported filtermode devargs combination:0x%x\n",
976 		       adap->devargs.filtermode);
977 		return ret;
978 	}
979 
980 	val[0] = V_FW_PARAMS_PARAM_FILTER_MODE(ret);
981 
982 	ret = cxgbe_get_filter_mode_from_devargs(adap->devargs.filtermask,
983 						 false);
984 	if (ret < 0) {
985 		pr_err("Unsupported filtermask devargs combination:0x%x\n",
986 		       adap->devargs.filtermask);
987 		return ret;
988 	}
989 
990 	val[0] |= V_FW_PARAMS_PARAM_FILTER_MASK(ret);
991 
992 	nparams++;
993 
994 	ret = cxgbe_get_filter_vnic_mode_from_devargs(adap->devargs.filtermode);
995 	if (ret < 0)
996 		return ret;
997 
998 	if (ret) {
999 		params[1] = CXGBE_FW_PARAM_DEV(FILTER) |
1000 			    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_FILTER_VNIC_MODE);
1001 
1002 		val[1] = ret - 1;
1003 
1004 		nparams++;
1005 	}
1006 
1007 	return t4_set_params(adap, adap->mbox, adap->pf, 0, nparams,
1008 			     params, val);
1009 }
1010 
1011 #define CXGBE_FW_CONFIG_PATH_T5 "/lib/firmware/cxgb4/t5-config.txt"
1012 #define CXGBE_FW_CONFIG_PATH_T6 "/lib/firmware/cxgb4/t6-config.txt"
1013 
1014 /*
1015  * Load firmware configuration from file in /lib/firmware/cxgb4/ path,
1016  * if it is present.
1017  */
1018 static int cxgbe_load_fw_config_from_filesystem(struct adapter *adap,
1019 						const char **config_name,
1020 						u32 *mem_type, u32 *mem_addr)
1021 {
1022 	u32 param, val, mtype, maddr;
1023 	const char *fw_cfg_path;
1024 	char *fw_cfg = NULL;
1025 	struct stat st;
1026 	int ret, fd;
1027 
1028 	switch (CHELSIO_CHIP_VERSION(adap->params.chip)) {
1029 	case CHELSIO_T5:
1030 		fw_cfg_path = CXGBE_FW_CONFIG_PATH_T5;
1031 		break;
1032 	case CHELSIO_T6:
1033 		fw_cfg_path = CXGBE_FW_CONFIG_PATH_T6;
1034 		break;
1035 	default:
1036 		return -ENOENT;
1037 	}
1038 
1039 	ret = open(fw_cfg_path, O_RDONLY);
1040 	if (ret < 0) {
1041 		dev_debug(adap, "Couldn't open FW config file\n");
1042 		return ret;
1043 	}
1044 
1045 	fd = ret;
1046 
1047 	ret = fstat(fd, &st);
1048 	if (ret < 0) {
1049 		dev_debug(adap, "Couldn't get FW config file size\n");
1050 		goto out_err;
1051 	}
1052 
1053 	if (st.st_size >= FLASH_CFG_MAX_SIZE) {
1054 		dev_debug(adap, "FW config file size >= max(%u)\n",
1055 			  FLASH_CFG_MAX_SIZE);
1056 		ret = -ENOMEM;
1057 		goto out_err;
1058 	}
1059 
1060 	fw_cfg = rte_zmalloc(NULL, st.st_size, 0);
1061 	if (fw_cfg == NULL) {
1062 		ret = -ENOMEM;
1063 		goto out_err;
1064 	}
1065 
1066 	if (read(fd, fw_cfg, st.st_size) != st.st_size) {
1067 		dev_debug(adap, "Couldn't read FW config file data\n");
1068 		ret = -EIO;
1069 		goto out_err;
1070 	}
1071 
1072 	close(fd);
1073 
1074 	/* Send it to FW to verify and update to new configuration */
1075 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
1076 		V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF);
1077 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &param, &val);
1078 	if (ret < 0) {
1079 		dev_debug(adap, "FW config param query failed: %d\n", ret);
1080 		goto out_free;
1081 	}
1082 
1083 	mtype = val >> 8;
1084 	maddr = (val & 0xff) << 16;
1085 
1086 	t4_os_lock(&adap->win0_lock);
1087 	ret = t4_memory_rw(adap, MEMWIN_NIC, mtype, maddr, st.st_size,
1088 			   fw_cfg, T4_MEMORY_WRITE);
1089 	t4_os_unlock(&adap->win0_lock);
1090 	if (ret < 0) {
1091 		dev_debug(adap, "FW config file update failed: %d\n", ret);
1092 		goto out_free;
1093 	}
1094 
1095 	rte_free(fw_cfg);
1096 
1097 	*mem_type = mtype;
1098 	*mem_addr = maddr;
1099 	*config_name = fw_cfg_path;
1100 	return 0;
1101 
1102 out_err:
1103 	close(fd);
1104 out_free:
1105 	rte_free(fw_cfg);
1106 	return ret;
1107 }
1108 
1109 static int cxgbe_load_fw_config(struct adapter *adap)
1110 {
1111 	u32 finiver, finicsum, cfcsum, mtype, maddr, param, val;
1112 	struct fw_caps_config_cmd caps_cmd = { 0 };
1113 	const char *config_name = NULL;
1114 	int ret;
1115 
1116 	ret = cxgbe_load_fw_config_from_filesystem(adap, &config_name,
1117 						   &mtype, &maddr);
1118 	if (ret < 0) {
1119 		config_name = "On Flash";
1120 
1121 		ret = t4_flash_cfg_addr(adap);
1122 		if (ret < 0) {
1123 			dev_warn(adap,
1124 				 "Finding address for FW config file in flash failed: %d\n",
1125 				 ret);
1126 			goto out_default_config;
1127 		}
1128 
1129 		mtype = FW_MEMTYPE_CF_FLASH;
1130 		maddr = ret;
1131 	}
1132 
1133 	/* Enable HASH filter region when support is available. */
1134 	val = 1;
1135 	param = CXGBE_FW_PARAM_DEV(HASHFILTER_WITH_OFLD);
1136 	t4_set_params(adap, adap->mbox, adap->pf, 0, 1, &param, &val);
1137 
1138 	/*
1139 	 * Issue a Capability Configuration command to the firmware to get it
1140 	 * to parse the Configuration File.
1141 	 */
1142 	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1143 					   F_FW_CMD_REQUEST | F_FW_CMD_READ);
1144 	caps_cmd.cfvalid_to_len16 =
1145 		cpu_to_be32(F_FW_CAPS_CONFIG_CMD_CFVALID |
1146 			    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
1147 			    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
1148 			    FW_LEN16(caps_cmd));
1149 	ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
1150 			 &caps_cmd);
1151 
1152 out_default_config:
1153 	/*
1154 	 * If the CAPS_CONFIG failed with an ENOENT (for a Firmware
1155 	 * Configuration File in filesystem or FLASH), our last gasp
1156 	 * effort is to use the Firmware Configuration File which is
1157 	 * embedded in the firmware.
1158 	 */
1159 	if (ret == -ENOENT) {
1160 		config_name = "Firmware Default";
1161 
1162 		memset(&caps_cmd, 0, sizeof(caps_cmd));
1163 		caps_cmd.op_to_write =
1164 			cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1165 				    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1166 		caps_cmd.cfvalid_to_len16 = cpu_to_be32(FW_LEN16(caps_cmd));
1167 		ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
1168 				 &caps_cmd);
1169 	}
1170 
1171 	if (ret < 0) {
1172 		dev_info(adap,
1173 			 "Failed to configure using %s Firmware Configuration file: %d\n",
1174 			 config_name, ret);
1175 		return ret;
1176 	}
1177 
1178 	finiver = be32_to_cpu(caps_cmd.finiver);
1179 	finicsum = be32_to_cpu(caps_cmd.finicsum);
1180 	cfcsum = be32_to_cpu(caps_cmd.cfcsum);
1181 	if (finicsum != cfcsum)
1182 		dev_warn(adap,
1183 			 "Configuration File checksum mismatch: [fini] csum=0x%x, computed csum=0x%x\n",
1184 			 finicsum, cfcsum);
1185 
1186 	/*
1187 	 * If we're a pure NIC driver then disable all offloading facilities.
1188 	 * This will allow the firmware to optimize aspects of the hardware
1189 	 * configuration which will result in improved performance.
1190 	 */
1191 	caps_cmd.niccaps &= cpu_to_be16(~FW_CAPS_CONFIG_NIC_ETHOFLD);
1192 	caps_cmd.toecaps = 0;
1193 	caps_cmd.iscsicaps = 0;
1194 	caps_cmd.rdmacaps = 0;
1195 	caps_cmd.fcoecaps = 0;
1196 	caps_cmd.cryptocaps = 0;
1197 
1198 	/*
1199 	 * And now tell the firmware to use the configuration we just loaded.
1200 	 */
1201 	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1202 					   F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
1203 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
1204 	ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
1205 			 NULL);
1206 	if (ret < 0) {
1207 		dev_warn(adap, "Unable to finalize Firmware Capabilities %d\n",
1208 			 ret);
1209 		return ret;
1210 	}
1211 
1212 	/*
1213 	 * Return successfully and note that we're operating with parameters
1214 	 * not supplied by the driver, rather than from hard-wired
1215 	 * initialization constants buried in the driver.
1216 	 */
1217 	dev_info(adap,
1218 		 "Successfully configured using Firmware Configuration File \"%s\", version: 0x%x, computed csum: 0x%x\n",
1219 		 config_name, finiver, cfcsum);
1220 	return 0;
1221 }
1222 
1223 static void configure_pcie_ext_tag(struct adapter *adapter)
1224 {
1225 	u16 v;
1226 	int pos = t4_os_find_pci_capability(adapter, PCI_CAP_ID_EXP);
1227 
1228 	if (!pos)
1229 		return;
1230 
1231 	if (pos > 0) {
1232 		t4_os_pci_read_cfg2(adapter, pos + PCI_EXP_DEVCTL, &v);
1233 		v |= PCI_EXP_DEVCTL_EXT_TAG;
1234 		t4_os_pci_write_cfg2(adapter, pos + PCI_EXP_DEVCTL, v);
1235 		if (is_t6(adapter->params.chip)) {
1236 			t4_set_reg_field(adapter, A_PCIE_CFG2,
1237 					 V_T6_TOTMAXTAG(M_T6_TOTMAXTAG),
1238 					 V_T6_TOTMAXTAG(7));
1239 			t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
1240 					 V_T6_MINTAG(M_T6_MINTAG),
1241 					 V_T6_MINTAG(8));
1242 		} else {
1243 			t4_set_reg_field(adapter, A_PCIE_CFG2,
1244 					 V_TOTMAXTAG(M_TOTMAXTAG),
1245 					 V_TOTMAXTAG(3));
1246 			t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
1247 					 V_MINTAG(M_MINTAG),
1248 					 V_MINTAG(8));
1249 		}
1250 	}
1251 }
1252 
1253 /* Figure out how many Queue Sets we can support */
1254 void cxgbe_configure_max_ethqsets(struct adapter *adapter)
1255 {
1256 	unsigned int ethqsets, reserved;
1257 
1258 	/* We need to reserve an Ingress Queue for the Asynchronous Firmware
1259 	 * Event Queue and 1 Control Queue per port.
1260 	 *
1261 	 * For each Queue Set, we'll need the ability to allocate two Egress
1262 	 * Contexts -- one for the Ingress Queue Free List and one for the TX
1263 	 * Ethernet Queue.
1264 	 */
1265 	reserved = max(adapter->params.nports, 1);
1266 	if (is_pf4(adapter)) {
1267 		struct pf_resources *pfres = &adapter->params.pfres;
1268 
1269 		ethqsets = min(pfres->niqflint, pfres->nethctrl);
1270 		if (ethqsets > (pfres->neq / 2))
1271 			ethqsets = pfres->neq / 2;
1272 	} else {
1273 		struct vf_resources *vfres = &adapter->params.vfres;
1274 
1275 		ethqsets = min(vfres->niqflint, vfres->nethctrl);
1276 		if (ethqsets > (vfres->neq / 2))
1277 			ethqsets = vfres->neq / 2;
1278 	}
1279 
1280 	ethqsets -= reserved;
1281 	adapter->sge.max_ethqsets = ethqsets;
1282 }
1283 
1284 /*
1285  * Tweak configuration based on system architecture, etc.  Most of these have
1286  * defaults assigned to them by Firmware Configuration Files (if we're using
1287  * them) but need to be explicitly set if we're using hard-coded
1288  * initialization. So these are essentially common tweaks/settings for
1289  * Configuration Files and hard-coded initialization ...
1290  */
1291 static int adap_init0_tweaks(struct adapter *adapter)
1292 {
1293 	u8 rx_dma_offset;
1294 
1295 	/*
1296 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
1297 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
1298 	 * 64B Cache Line Size ...
1299 	 */
1300 	t4_fixup_host_params_compat(adapter, CXGBE_PAGE_SIZE, L1_CACHE_BYTES,
1301 				    T5_LAST_REV);
1302 
1303 	/*
1304 	 * Keep the chip default offset to deliver Ingress packets into our
1305 	 * DMA buffers to zero
1306 	 */
1307 	rx_dma_offset = 0;
1308 	t4_set_reg_field(adapter, A_SGE_CONTROL, V_PKTSHIFT(M_PKTSHIFT),
1309 			 V_PKTSHIFT(rx_dma_offset));
1310 
1311 	t4_set_reg_field(adapter, A_SGE_FLM_CFG,
1312 			 V_CREDITCNT(M_CREDITCNT) | M_CREDITCNTPACKING,
1313 			 V_CREDITCNT(3) | V_CREDITCNTPACKING(1));
1314 
1315 	t4_set_reg_field(adapter, A_SGE_INGRESS_RX_THRESHOLD,
1316 			 V_THRESHOLD_3(M_THRESHOLD_3), V_THRESHOLD_3(32U));
1317 
1318 	t4_set_reg_field(adapter, A_SGE_CONTROL2, V_IDMAARBROUNDROBIN(1U),
1319 			 V_IDMAARBROUNDROBIN(1U));
1320 
1321 	/*
1322 	 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
1323 	 * adds the pseudo header itself.
1324 	 */
1325 	t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG,
1326 			       F_CSUM_HAS_PSEUDO_HDR, 0);
1327 
1328 	return 0;
1329 }
1330 
1331 /*
1332  * Attempt to initialize the adapter via a Firmware Configuration File.
1333  */
1334 static int adap_init0_config(struct adapter *adapter, int reset)
1335 {
1336 	int ret;
1337 
1338 	/*
1339 	 * Reset device if necessary.
1340 	 */
1341 	if (reset) {
1342 		ret = t4_fw_reset(adapter, adapter->mbox,
1343 				  F_PIORSTMODE | F_PIORST);
1344 		if (ret < 0) {
1345 			dev_warn(adapter, "Firmware reset failed, error %d\n",
1346 				 -ret);
1347 			goto bye;
1348 		}
1349 	}
1350 
1351 	ret = cxgbe_load_fw_config(adapter);
1352 	if (ret < 0)
1353 		goto bye;
1354 
1355 	/*
1356 	 * Tweak configuration based on system architecture, etc.
1357 	 */
1358 	ret = adap_init0_tweaks(adapter);
1359 	if (ret < 0) {
1360 		dev_warn(adapter, "Unable to do init0-tweaks %d\n", -ret);
1361 		goto bye;
1362 	}
1363 
1364 	/*
1365 	 * And finally tell the firmware to initialize itself using the
1366 	 * parameters from the Configuration File.
1367 	 */
1368 	ret = t4_fw_initialize(adapter, adapter->mbox);
1369 	if (ret < 0) {
1370 		dev_warn(adapter, "Initializing Firmware failed, error %d\n",
1371 			 -ret);
1372 		goto bye;
1373 	}
1374 
1375 	return 0;
1376 
1377 bye:
1378 	dev_debug(adapter, "%s: returning ret = %d ..\n", __func__, ret);
1379 	return ret;
1380 }
1381 
1382 static int adap_init0(struct adapter *adap)
1383 {
1384 	struct fw_caps_config_cmd caps_cmd;
1385 	int ret = 0;
1386 	u32 v, port_vec;
1387 	enum dev_state state;
1388 	u32 params[7], val[7];
1389 	int reset = 1;
1390 	int mbox = adap->mbox;
1391 
1392 	/*
1393 	 * Contact FW, advertising Master capability.
1394 	 */
1395 	ret = t4_fw_hello(adap, adap->mbox, adap->mbox, MASTER_MAY, &state);
1396 	if (ret < 0) {
1397 		dev_err(adap, "%s: could not connect to FW, error %d\n",
1398 			__func__, -ret);
1399 		goto bye;
1400 	}
1401 
1402 	CXGBE_DEBUG_MBOX(adap, "%s: adap->mbox = %d; ret = %d\n", __func__,
1403 			 adap->mbox, ret);
1404 
1405 	if (ret == mbox)
1406 		adap->flags |= MASTER_PF;
1407 
1408 	if (state == DEV_STATE_INIT) {
1409 		/*
1410 		 * Force halt and reset FW because a previous instance may have
1411 		 * exited abnormally without properly shutting down
1412 		 */
1413 		ret = t4_fw_halt(adap, adap->mbox, reset);
1414 		if (ret < 0) {
1415 			dev_err(adap, "Failed to halt. Exit.\n");
1416 			goto bye;
1417 		}
1418 
1419 		ret = t4_fw_restart(adap, adap->mbox, reset);
1420 		if (ret < 0) {
1421 			dev_err(adap, "Failed to restart. Exit.\n");
1422 			goto bye;
1423 		}
1424 		state = (enum dev_state)((unsigned)state & ~DEV_STATE_INIT);
1425 	}
1426 
1427 	t4_get_version_info(adap);
1428 
1429 	ret = t4_get_core_clock(adap, &adap->params.vpd);
1430 	if (ret < 0) {
1431 		dev_err(adap, "%s: could not get core clock, error %d\n",
1432 			__func__, -ret);
1433 		goto bye;
1434 	}
1435 
1436 	/*
1437 	 * If the firmware is initialized already (and we're not forcing a
1438 	 * master initialization), note that we're living with existing
1439 	 * adapter parameters.  Otherwise, it's time to try initializing the
1440 	 * adapter ...
1441 	 */
1442 	if (state == DEV_STATE_INIT) {
1443 		dev_info(adap, "Coming up as %s: Adapter already initialized\n",
1444 			 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
1445 	} else {
1446 		dev_info(adap, "Coming up as MASTER: Initializing adapter\n");
1447 
1448 		ret = adap_init0_config(adap, reset);
1449 		if (ret == -ENOENT) {
1450 			dev_err(adap,
1451 				"No Configuration File present on adapter. Using hard-wired configuration parameters.\n");
1452 			goto bye;
1453 		}
1454 	}
1455 	if (ret < 0) {
1456 		dev_err(adap, "could not initialize adapter, error %d\n", -ret);
1457 		goto bye;
1458 	}
1459 
1460 	/* Now that we've successfully configured and initialized the adapter
1461 	 * (or found it already initialized), we can ask the Firmware what
1462 	 * resources it has provisioned for us.
1463 	 */
1464 	ret = t4_get_pfres(adap);
1465 	if (ret) {
1466 		dev_err(adap->pdev_dev,
1467 			"Unable to retrieve resource provisioning info\n");
1468 		goto bye;
1469 	}
1470 
1471 	/* Find out what ports are available to us. */
1472 	v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
1473 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
1474 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
1475 	if (ret < 0) {
1476 		dev_err(adap, "%s: failure in t4_query_params; error = %d\n",
1477 			__func__, ret);
1478 		goto bye;
1479 	}
1480 
1481 	adap->params.nports = hweight32(port_vec);
1482 	adap->params.portvec = port_vec;
1483 
1484 	dev_debug(adap, "%s: adap->params.nports = %u\n", __func__,
1485 		  adap->params.nports);
1486 
1487 	/*
1488 	 * Give the SGE code a chance to pull in anything that it needs ...
1489 	 * Note that this must be called after we retrieve our VPD parameters
1490 	 * in order to know how to convert core ticks to seconds, etc.
1491 	 */
1492 	ret = t4_sge_init(adap);
1493 	if (ret < 0) {
1494 		dev_err(adap, "t4_sge_init failed with error %d\n",
1495 			-ret);
1496 		goto bye;
1497 	}
1498 
1499 	/*
1500 	 * Grab some of our basic fundamental operating parameters.
1501 	 */
1502 	params[0] = CXGBE_FW_PARAM_PFVF(L2T_START);
1503 	params[1] = CXGBE_FW_PARAM_PFVF(L2T_END);
1504 	params[2] = CXGBE_FW_PARAM_PFVF(FILTER_START);
1505 	params[3] = CXGBE_FW_PARAM_PFVF(FILTER_END);
1506 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 4, params, val);
1507 	if (ret < 0)
1508 		goto bye;
1509 	adap->l2t_start = val[0];
1510 	adap->l2t_end = val[1];
1511 	adap->tids.ftid_base = val[2];
1512 	adap->tids.nftids = val[3] - val[2] + 1;
1513 
1514 	params[0] = CXGBE_FW_PARAM_PFVF(CLIP_START);
1515 	params[1] = CXGBE_FW_PARAM_PFVF(CLIP_END);
1516 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
1517 	if (ret < 0)
1518 		goto bye;
1519 	adap->clipt_start = val[0];
1520 	adap->clipt_end = val[1];
1521 
1522 	/*
1523 	 * Get device capabilities so we can determine what resources we need
1524 	 * to manage.
1525 	 */
1526 	memset(&caps_cmd, 0, sizeof(caps_cmd));
1527 	caps_cmd.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1528 				     F_FW_CMD_REQUEST | F_FW_CMD_READ);
1529 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
1530 	ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
1531 			 &caps_cmd);
1532 	if (ret < 0)
1533 		goto bye;
1534 
1535 	if ((caps_cmd.niccaps & cpu_to_be16(FW_CAPS_CONFIG_NIC_HASHFILTER)) &&
1536 	    is_t6(adap->params.chip)) {
1537 		if (cxgbe_init_hash_filter(adap) < 0)
1538 			goto bye;
1539 	}
1540 
1541 	/* See if FW supports FW_FILTER2 work request */
1542 	if (is_t4(adap->params.chip)) {
1543 		adap->params.filter2_wr_support = 0;
1544 	} else {
1545 		params[0] = CXGBE_FW_PARAM_DEV(FILTER2_WR);
1546 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1547 				      1, params, val);
1548 		adap->params.filter2_wr_support = (ret == 0 && val[0] != 0);
1549 	}
1550 
1551 	/* Check if FW supports returning vin.
1552 	 * If this is not supported, driver will interpret
1553 	 * these values from viid.
1554 	 */
1555 	params[0] = CXGBE_FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
1556 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1557 			      1, params, val);
1558 	adap->params.viid_smt_extn_support = (ret == 0 && val[0] != 0);
1559 
1560 	/* query tid-related parameters */
1561 	params[0] = CXGBE_FW_PARAM_DEV(NTID);
1562 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
1563 			      params, val);
1564 	if (ret < 0)
1565 		goto bye;
1566 	adap->tids.ntids = val[0];
1567 	adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
1568 
1569 	/* If we're running on newer firmware, let it know that we're
1570 	 * prepared to deal with encapsulated CPL messages.  Older
1571 	 * firmware won't understand this and we'll just get
1572 	 * unencapsulated messages ...
1573 	 */
1574 	params[0] = CXGBE_FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
1575 	val[0] = 1;
1576 	(void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1577 
1578 	/*
1579 	 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
1580 	 * capability.  Earlier versions of the firmware didn't have the
1581 	 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
1582 	 * permission to use ULPTX MEMWRITE DSGL.
1583 	 */
1584 	if (is_t4(adap->params.chip)) {
1585 		adap->params.ulptx_memwrite_dsgl = false;
1586 	} else {
1587 		params[0] = CXGBE_FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
1588 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1589 				      1, params, val);
1590 		adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
1591 	}
1592 
1593 	/* Query for max number of packets that can be coalesced for Tx */
1594 	params[0] = CXGBE_FW_PARAM_PFVF(MAX_PKTS_PER_ETH_TX_PKTS_WR);
1595 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1596 	if (!ret && val[0] > 0)
1597 		adap->params.max_tx_coalesce_num = val[0];
1598 	else
1599 		adap->params.max_tx_coalesce_num = ETH_COALESCE_PKT_NUM;
1600 
1601 	params[0] = CXGBE_FW_PARAM_DEV(VI_ENABLE_INGRESS_AFTER_LINKUP);
1602 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
1603 	adap->params.vi_enable_rx = (ret == 0 && val[0] != 0);
1604 
1605 	/* Read the RAW MPS entries. In T6, the last 2 TCAM entries
1606 	 * are reserved for RAW MAC addresses (rawf = 2, one per port).
1607 	 */
1608 	if (CHELSIO_CHIP_VERSION(adap->params.chip) > CHELSIO_T5) {
1609 		params[0] = CXGBE_FW_PARAM_PFVF(RAWF_START);
1610 		params[1] = CXGBE_FW_PARAM_PFVF(RAWF_END);
1611 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
1612 				      params, val);
1613 		if (ret == 0) {
1614 			adap->params.rawf_start = val[0];
1615 			adap->params.rawf_size = val[1] - val[0] + 1;
1616 		}
1617 	}
1618 
1619 	/*
1620 	 * The MTU/MSS Table is initialized by now, so load their values.  If
1621 	 * we're initializing the adapter, then we'll make any modifications
1622 	 * we want to the MTU/MSS Table and also initialize the congestion
1623 	 * parameters.
1624 	 */
1625 	t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
1626 	if (state != DEV_STATE_INIT) {
1627 		int i;
1628 
1629 		/*
1630 		 * The default MTU Table contains values 1492 and 1500.
1631 		 * However, for TCP, it's better to have two values which are
1632 		 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
1633 		 * This allows us to have a TCP Data Payload which is a
1634 		 * multiple of 8 regardless of what combination of TCP Options
1635 		 * are in use (always a multiple of 4 bytes) which is
1636 		 * important for performance reasons.  For instance, if no
1637 		 * options are in use, then we have a 20-byte IP header and a
1638 		 * 20-byte TCP header.  In this case, a 1500-byte MSS would
1639 		 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
1640 		 * which is not a multiple of 8.  So using an MSS of 1488 in
1641 		 * this case results in a TCP Data Payload of 1448 bytes which
1642 		 * is a multiple of 8.  On the other hand, if 12-byte TCP Time
1643 		 * Stamps have been negotiated, then an MTU of 1500 bytes
1644 		 * results in a TCP Data Payload of 1448 bytes which, as
1645 		 * above, is a multiple of 8 bytes ...
1646 		 */
1647 		for (i = 0; i < NMTUS; i++)
1648 			if (adap->params.mtus[i] == 1492) {
1649 				adap->params.mtus[i] = 1488;
1650 				break;
1651 			}
1652 
1653 		t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
1654 			     adap->params.b_wnd);
1655 	}
1656 	t4_init_sge_params(adap);
1657 	ret = configure_filter_mode_mask(adap);
1658 	if (ret < 0)
1659 		goto bye;
1660 	t4_init_tp_params(adap);
1661 	configure_pcie_ext_tag(adap);
1662 	configure_vlan_types(adap);
1663 	cxgbe_configure_max_ethqsets(adap);
1664 
1665 	adap->params.drv_memwin = MEMWIN_NIC;
1666 	adap->flags |= FW_OK;
1667 	dev_debug(adap, "%s: returning zero..\n", __func__);
1668 	return 0;
1669 
1670 	/*
1671 	 * Something bad happened.  If a command timed out or failed with EIO
1672 	 * FW does not operate within its spec or something catastrophic
1673 	 * happened to HW/FW, stop issuing commands.
1674 	 */
1675 bye:
1676 	if (ret != -ETIMEDOUT && ret != -EIO)
1677 		t4_fw_bye(adap, adap->mbox);
1678 	return ret;
1679 }
1680 
1681 /**
1682  * t4_os_portmod_changed - handle port module changes
1683  * @adap: the adapter associated with the module change
1684  * @port_id: the port index whose module status has changed
1685  *
1686  * This is the OS-dependent handler for port module changes.  It is
1687  * invoked when a port module is removed or inserted for any OS-specific
1688  * processing.
1689  */
1690 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
1691 {
1692 	static const char * const mod_str[] = {
1693 		NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
1694 	};
1695 
1696 	const struct port_info *pi = adap2pinfo(adap, port_id);
1697 
1698 	if (pi->link_cfg.mod_type == FW_PORT_MOD_TYPE_NONE)
1699 		dev_info(adap, "Port%d: port module unplugged\n", pi->port_id);
1700 	else if (pi->link_cfg.mod_type < ARRAY_SIZE(mod_str))
1701 		dev_info(adap, "Port%d: %s port module inserted\n", pi->port_id,
1702 			 mod_str[pi->link_cfg.mod_type]);
1703 	else if (pi->link_cfg.mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
1704 		dev_info(adap, "Port%d: unsupported port module inserted\n",
1705 			 pi->port_id);
1706 	else if (pi->link_cfg.mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
1707 		dev_info(adap, "Port%d: unknown port module inserted\n",
1708 			 pi->port_id);
1709 	else if (pi->link_cfg.mod_type == FW_PORT_MOD_TYPE_ERROR)
1710 		dev_info(adap, "Port%d: transceiver module error\n",
1711 			 pi->port_id);
1712 	else
1713 		dev_info(adap, "Port%d: unknown module type %d inserted\n",
1714 			 pi->port_id, pi->link_cfg.mod_type);
1715 }
1716 
1717 void t4_os_link_changed(struct adapter *adap, int port_id)
1718 {
1719 	struct port_info *pi = adap2pinfo(adap, port_id);
1720 
1721 	/* If link status has not changed or if firmware doesn't
1722 	 * support enabling/disabling VI's Rx path during runtime,
1723 	 * then return.
1724 	 */
1725 	if (adap->params.vi_enable_rx == 0 ||
1726 	    pi->vi_en_rx == pi->link_cfg.link_ok)
1727 		return;
1728 
1729 	/* Don't enable VI Rx path, if link has been administratively
1730 	 * turned off.
1731 	 */
1732 	if (pi->vi_en_tx == 0 && pi->vi_en_rx == 0)
1733 		return;
1734 
1735 	/* When link goes down, disable the port's Rx path to drop
1736 	 * Rx traffic closer to the wire, instead of processing it
1737 	 * further in the Rx pipeline. The Rx path will be re-enabled
1738 	 * once the link up message comes in firmware event queue.
1739 	 */
1740 	pi->vi_en_rx = pi->link_cfg.link_ok;
1741 	t4_enable_vi(adap, adap->mbox, pi->viid, pi->vi_en_rx, pi->vi_en_tx);
1742 }
1743 
1744 bool cxgbe_force_linkup(struct adapter *adap)
1745 {
1746 	if (is_pf4(adap))
1747 		return false;	/* force_linkup not required for pf driver */
1748 
1749 	return adap->devargs.force_link_up;
1750 }
1751 
1752 /**
1753  * link_start - enable a port
1754  * @dev: the port to enable
1755  *
1756  * Performs the MAC and PHY actions needed to enable a port.
1757  */
1758 int cxgbe_link_start(struct port_info *pi)
1759 {
1760 	struct adapter *adapter = pi->adapter;
1761 	u64 conf_offloads;
1762 	unsigned int mtu;
1763 	int ret;
1764 
1765 	mtu = pi->eth_dev->data->mtu;
1766 
1767 	conf_offloads = pi->eth_dev->data->dev_conf.rxmode.offloads;
1768 
1769 	/*
1770 	 * We do not set address filters and promiscuity here, the stack does
1771 	 * that step explicitly.
1772 	 */
1773 	ret = t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu, -1, -1, -1,
1774 			    !!(conf_offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP),
1775 			    true);
1776 	if (ret == 0) {
1777 		ret = cxgbe_mpstcam_modify(pi, (int)pi->xact_addr_filt,
1778 				(u8 *)&pi->eth_dev->data->mac_addrs[0]);
1779 		if (ret >= 0) {
1780 			pi->xact_addr_filt = ret;
1781 			ret = 0;
1782 		}
1783 	}
1784 	if (ret == 0 && is_pf4(adapter))
1785 		ret = t4_link_l1cfg(pi, pi->link_cfg.admin_caps);
1786 	if (ret == 0) {
1787 		/* Disable VI Rx until link up message is received in
1788 		 * firmware event queue, if firmware supports enabling/
1789 		 * disabling VI Rx at runtime.
1790 		 */
1791 		pi->vi_en_rx = adapter->params.vi_enable_rx ? 0 : 1;
1792 		pi->vi_en_tx = 1;
1793 		ret = t4_enable_vi_params(adapter, adapter->mbox, pi->viid,
1794 					  pi->vi_en_rx, pi->vi_en_tx, false);
1795 	}
1796 
1797 	if (ret == 0 && cxgbe_force_linkup(adapter))
1798 		pi->eth_dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
1799 	return ret;
1800 }
1801 
1802 /**
1803  * cxgbe_write_rss_conf - flash the RSS configuration for a given port
1804  * @pi: the port
1805  * @rss_hf: Hash configuration to apply
1806  */
1807 int cxgbe_write_rss_conf(const struct port_info *pi, uint64_t rss_hf)
1808 {
1809 	struct adapter *adapter = pi->adapter;
1810 	const struct sge_eth_rxq *rxq;
1811 	u64 flags = 0;
1812 	u16 rss;
1813 	int err;
1814 
1815 	/*  Should never be called before setting up sge eth rx queues */
1816 	if (!(adapter->flags & FULL_INIT_DONE)) {
1817 		dev_err(adap, "%s No RXQs available on port %d\n",
1818 			__func__, pi->port_id);
1819 		return -EINVAL;
1820 	}
1821 
1822 	/* Don't allow unsupported hash functions */
1823 	if (rss_hf & ~CXGBE_RSS_HF_ALL)
1824 		return -EINVAL;
1825 
1826 	if (rss_hf & CXGBE_RSS_HF_IPV4_MASK)
1827 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
1828 
1829 	if (rss_hf & RTE_ETH_RSS_NONFRAG_IPV4_TCP)
1830 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
1831 
1832 	if (rss_hf & RTE_ETH_RSS_NONFRAG_IPV4_UDP)
1833 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
1834 			 F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1835 
1836 	if (rss_hf & CXGBE_RSS_HF_IPV6_MASK)
1837 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
1838 
1839 	if (rss_hf & CXGBE_RSS_HF_TCP_IPV6_MASK)
1840 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1841 			 F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
1842 
1843 	if (rss_hf & CXGBE_RSS_HF_UDP_IPV6_MASK)
1844 		flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
1845 			 F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
1846 			 F_FW_RSS_VI_CONFIG_CMD_UDPEN;
1847 
1848 	rxq = &adapter->sge.ethrxq[pi->first_rxqset];
1849 	rss = rxq[0].rspq.abs_id;
1850 
1851 	/* If Tunnel All Lookup isn't specified in the global RSS
1852 	 * Configuration, then we need to specify a default Ingress
1853 	 * Queue for any ingress packets which aren't hashed.  We'll
1854 	 * use our first ingress queue ...
1855 	 */
1856 	err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
1857 			       flags, rss);
1858 	return err;
1859 }
1860 
1861 /**
1862  * cxgbe_write_rss - write the RSS table for a given port
1863  * @pi: the port
1864  * @queues: array of queue indices for RSS
1865  *
1866  * Sets up the portion of the HW RSS table for the port's VI to distribute
1867  * packets to the Rx queues in @queues.
1868  */
1869 int cxgbe_write_rss(const struct port_info *pi, const u16 *queues)
1870 {
1871 	u16 *rss;
1872 	int i, err;
1873 	struct adapter *adapter = pi->adapter;
1874 	const struct sge_eth_rxq *rxq;
1875 
1876 	/*  Should never be called before setting up sge eth rx queues */
1877 	BUG_ON(!(adapter->flags & FULL_INIT_DONE));
1878 
1879 	rxq = &adapter->sge.ethrxq[pi->first_rxqset];
1880 	rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
1881 	if (!rss)
1882 		return -ENOMEM;
1883 
1884 	/* map the queue indices to queue ids */
1885 	for (i = 0; i < pi->rss_size; i++, queues++)
1886 		rss[i] = rxq[*queues].rspq.abs_id;
1887 
1888 	err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
1889 				  pi->rss_size, rss, pi->rss_size);
1890 	rte_free(rss);
1891 	return err;
1892 }
1893 
1894 /**
1895  * setup_rss - configure RSS
1896  * @adapter: the adapter
1897  *
1898  * Sets up RSS to distribute packets to multiple receive queues.  We
1899  * configure the RSS CPU lookup table to distribute to the number of HW
1900  * receive queues, and the response queue lookup table to narrow that
1901  * down to the response queues actually configured for each port.
1902  * We always configure the RSS mapping for all ports since the mapping
1903  * table has plenty of entries.
1904  */
1905 int cxgbe_setup_rss(struct port_info *pi)
1906 {
1907 	int j, err;
1908 	struct adapter *adapter = pi->adapter;
1909 
1910 	dev_debug(adapter, "%s:  pi->rss_size = %u; pi->n_rx_qsets = %u\n",
1911 		  __func__, pi->rss_size, pi->n_rx_qsets);
1912 
1913 	if (!(pi->flags & PORT_RSS_DONE)) {
1914 		if (adapter->flags & FULL_INIT_DONE) {
1915 			/* Fill default values with equal distribution */
1916 			for (j = 0; j < pi->rss_size; j++)
1917 				pi->rss[j] = j % pi->n_rx_qsets;
1918 
1919 			err = cxgbe_write_rss(pi, pi->rss);
1920 			if (err)
1921 				return err;
1922 
1923 			err = cxgbe_write_rss_conf(pi, pi->rss_hf);
1924 			if (err)
1925 				return err;
1926 			pi->flags |= PORT_RSS_DONE;
1927 		}
1928 	}
1929 	return 0;
1930 }
1931 
1932 /*
1933  * Enable NAPI scheduling and interrupt generation for all Rx queues.
1934  */
1935 static void enable_rx(struct adapter *adap, struct sge_rspq *q)
1936 {
1937 	/* 0-increment GTS to start the timer and enable interrupts */
1938 	t4_write_reg(adap, is_pf4(adap) ? MYPF_REG(A_SGE_PF_GTS) :
1939 					  T4VF_SGE_BASE_ADDR + A_SGE_VF_GTS,
1940 		     V_SEINTARM(q->intr_params) |
1941 		     V_INGRESSQID(q->cntxt_id));
1942 }
1943 
1944 void cxgbe_enable_rx_queues(struct port_info *pi)
1945 {
1946 	struct adapter *adap = pi->adapter;
1947 	struct sge *s = &adap->sge;
1948 	unsigned int i;
1949 
1950 	for (i = 0; i < pi->n_rx_qsets; i++)
1951 		enable_rx(adap, &s->ethrxq[pi->first_rxqset + i].rspq);
1952 }
1953 
1954 /**
1955  * fw_caps_to_speed_caps - translate Firmware Port Caps to Speed Caps.
1956  * @port_type: Firmware Port Type
1957  * @fw_caps: Firmware Port Capabilities
1958  * @speed_caps: Device Info Speed Capabilities
1959  *
1960  * Translate a Firmware Port Capabilities specification to Device Info
1961  * Speed Capabilities.
1962  */
1963 static void fw_caps_to_speed_caps(enum fw_port_type port_type,
1964 				  unsigned int fw_caps,
1965 				  u32 *speed_caps)
1966 {
1967 #define SET_SPEED(__speed_name) \
1968 	do { \
1969 		*speed_caps |= RTE_ETH_LINK_ ## __speed_name; \
1970 	} while (0)
1971 
1972 #define FW_CAPS_TO_SPEED(__fw_name) \
1973 	do { \
1974 		if (fw_caps & FW_PORT_CAP32_ ## __fw_name) \
1975 			SET_SPEED(__fw_name); \
1976 	} while (0)
1977 
1978 	switch (port_type) {
1979 	case FW_PORT_TYPE_BT_SGMII:
1980 	case FW_PORT_TYPE_BT_XFI:
1981 	case FW_PORT_TYPE_BT_XAUI:
1982 		FW_CAPS_TO_SPEED(SPEED_100M);
1983 		FW_CAPS_TO_SPEED(SPEED_1G);
1984 		FW_CAPS_TO_SPEED(SPEED_10G);
1985 		break;
1986 
1987 	case FW_PORT_TYPE_KX4:
1988 	case FW_PORT_TYPE_KX:
1989 	case FW_PORT_TYPE_FIBER_XFI:
1990 	case FW_PORT_TYPE_FIBER_XAUI:
1991 	case FW_PORT_TYPE_SFP:
1992 	case FW_PORT_TYPE_QSFP_10G:
1993 	case FW_PORT_TYPE_QSA:
1994 		FW_CAPS_TO_SPEED(SPEED_1G);
1995 		FW_CAPS_TO_SPEED(SPEED_10G);
1996 		break;
1997 
1998 	case FW_PORT_TYPE_KR:
1999 		SET_SPEED(SPEED_10G);
2000 		break;
2001 
2002 	case FW_PORT_TYPE_BP_AP:
2003 	case FW_PORT_TYPE_BP4_AP:
2004 		SET_SPEED(SPEED_1G);
2005 		SET_SPEED(SPEED_10G);
2006 		break;
2007 
2008 	case FW_PORT_TYPE_BP40_BA:
2009 	case FW_PORT_TYPE_QSFP:
2010 		SET_SPEED(SPEED_40G);
2011 		break;
2012 
2013 	case FW_PORT_TYPE_CR_QSFP:
2014 	case FW_PORT_TYPE_SFP28:
2015 	case FW_PORT_TYPE_KR_SFP28:
2016 		FW_CAPS_TO_SPEED(SPEED_1G);
2017 		FW_CAPS_TO_SPEED(SPEED_10G);
2018 		FW_CAPS_TO_SPEED(SPEED_25G);
2019 		break;
2020 
2021 	case FW_PORT_TYPE_CR2_QSFP:
2022 		SET_SPEED(SPEED_50G);
2023 		break;
2024 
2025 	case FW_PORT_TYPE_KR4_100G:
2026 	case FW_PORT_TYPE_CR4_QSFP:
2027 		FW_CAPS_TO_SPEED(SPEED_25G);
2028 		FW_CAPS_TO_SPEED(SPEED_40G);
2029 		FW_CAPS_TO_SPEED(SPEED_50G);
2030 		FW_CAPS_TO_SPEED(SPEED_100G);
2031 		break;
2032 
2033 	default:
2034 		break;
2035 	}
2036 
2037 #undef FW_CAPS_TO_SPEED
2038 #undef SET_SPEED
2039 }
2040 
2041 /**
2042  * cxgbe_get_speed_caps - Fetch supported speed capabilities
2043  * @pi: Underlying port's info
2044  * @speed_caps: Device Info speed capabilities
2045  *
2046  * Fetch supported speed capabilities of the underlying port.
2047  */
2048 void cxgbe_get_speed_caps(struct port_info *pi, u32 *speed_caps)
2049 {
2050 	*speed_caps = 0;
2051 
2052 	fw_caps_to_speed_caps(pi->link_cfg.port_type, pi->link_cfg.pcaps,
2053 			      speed_caps);
2054 
2055 	if (!(pi->link_cfg.pcaps & FW_PORT_CAP32_ANEG))
2056 		*speed_caps |= RTE_ETH_LINK_SPEED_FIXED;
2057 }
2058 
2059 /**
2060  * cxgbe_set_link_status - Set device link up or down.
2061  * @pi: Underlying port's info
2062  * @status: 0 - down, 1 - up
2063  *
2064  * Set the device link up or down.
2065  */
2066 int cxgbe_set_link_status(struct port_info *pi, bool status)
2067 {
2068 	struct adapter *adapter = pi->adapter;
2069 	int err = 0;
2070 
2071 	/* Wait for link up message from firmware to enable Rx path,
2072 	 * if firmware supports enabling/disabling VI Rx at runtime.
2073 	 */
2074 	pi->vi_en_rx = adapter->params.vi_enable_rx ? 0 : status;
2075 	pi->vi_en_tx = status;
2076 	err = t4_enable_vi(adapter, adapter->mbox, pi->viid, pi->vi_en_rx,
2077 			   pi->vi_en_tx);
2078 	if (err) {
2079 		dev_err(adapter, "%s: disable_vi failed: %d\n", __func__, err);
2080 		return err;
2081 	}
2082 
2083 	if (!status)
2084 		t4_reset_link_config(adapter, pi->pidx);
2085 
2086 	return 0;
2087 }
2088 
2089 /**
2090  * cxgb_up - enable the adapter
2091  * @adap: adapter being enabled
2092  *
2093  * Called when the first port is enabled, this function performs the
2094  * actions necessary to make an adapter operational, such as completing
2095  * the initialization of HW modules, and enabling interrupts.
2096  */
2097 int cxgbe_up(struct adapter *adap)
2098 {
2099 	enable_rx(adap, &adap->sge.fw_evtq);
2100 	t4_sge_tx_monitor_start(adap);
2101 	if (is_pf4(adap))
2102 		t4_intr_enable(adap);
2103 	adap->flags |= FULL_INIT_DONE;
2104 
2105 	/* TODO: deadman watchdog ?? */
2106 	return 0;
2107 }
2108 
2109 /*
2110  * Close the port
2111  */
2112 int cxgbe_down(struct port_info *pi)
2113 {
2114 	return cxgbe_set_link_status(pi, false);
2115 }
2116 
2117 /*
2118  * Release resources when all the ports have been stopped.
2119  */
2120 void cxgbe_close(struct adapter *adapter)
2121 {
2122 	if (adapter->flags & FULL_INIT_DONE) {
2123 		tid_free(&adapter->tids);
2124 		t4_cleanup_mpstcam(adapter);
2125 		t4_cleanup_clip_tbl(adapter);
2126 		t4_cleanup_l2t(adapter);
2127 		t4_cleanup_smt(adapter);
2128 		if (is_pf4(adapter))
2129 			t4_intr_disable(adapter);
2130 		t4_sge_tx_monitor_stop(adapter);
2131 		t4_free_sge_resources(adapter);
2132 		adapter->flags &= ~FULL_INIT_DONE;
2133 	}
2134 
2135 	cxgbe_cfg_queues_free(adapter);
2136 
2137 	if (is_pf4(adapter) && (adapter->flags & FW_OK))
2138 		t4_fw_bye(adapter, adapter->mbox);
2139 }
2140 
2141 static void adap_smt_index(struct adapter *adapter, u32 *smt_start_idx,
2142 			   u32 *smt_size)
2143 {
2144 	u32 params[2], smt_val[2];
2145 	int ret;
2146 
2147 	params[0] = CXGBE_FW_PARAM_PFVF(GET_SMT_START);
2148 	params[1] = CXGBE_FW_PARAM_PFVF(GET_SMT_SIZE);
2149 
2150 	ret = t4_query_params(adapter, adapter->mbox, adapter->pf, 0,
2151 			      2, params, smt_val);
2152 
2153 	/* if FW doesn't recognize this command then set it to default setting
2154 	 * which is start index as 0 and size as 256.
2155 	 */
2156 	if (ret < 0) {
2157 		*smt_start_idx = 0;
2158 		*smt_size = SMT_SIZE;
2159 	} else {
2160 		*smt_start_idx = smt_val[0];
2161 		/* smt size can be zero, if nsmt is not yet configured in
2162 		 * the config file or set as zero, then configure all the
2163 		 * remaining entries to this PF itself.
2164 		 */
2165 		if (!smt_val[1])
2166 			*smt_size = SMT_SIZE - *smt_start_idx;
2167 		else
2168 			*smt_size = smt_val[1];
2169 	}
2170 }
2171 
2172 int cxgbe_probe(struct adapter *adapter)
2173 {
2174 	u32 smt_start_idx, smt_size;
2175 	struct port_info *pi;
2176 	int func, i;
2177 	int err = 0;
2178 	u32 whoami;
2179 	int chip;
2180 
2181 	whoami = t4_read_reg(adapter, A_PL_WHOAMI);
2182 	chip = t4_get_chip_type(adapter,
2183 			CHELSIO_PCI_ID_VER(adapter->pdev->id.device_id));
2184 	if (chip < 0)
2185 		return chip;
2186 
2187 	func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
2188 	       G_SOURCEPF(whoami) : G_T6_SOURCEPF(whoami);
2189 
2190 	adapter->mbox = func;
2191 	adapter->pf = func;
2192 
2193 	t4_os_lock_init(&adapter->mbox_lock);
2194 	TAILQ_INIT(&adapter->mbox_list);
2195 	t4_os_lock_init(&adapter->win0_lock);
2196 
2197 	err = t4_prep_adapter(adapter);
2198 	if (err)
2199 		return err;
2200 
2201 	setup_memwin(adapter);
2202 	err = adap_init0(adapter);
2203 	if (err) {
2204 		dev_err(adapter, "%s: Adapter initialization failed, error %d\n",
2205 			__func__, err);
2206 		goto out_free;
2207 	}
2208 
2209 	if (!is_t4(adapter->params.chip)) {
2210 		/*
2211 		 * The userspace doorbell BAR is split evenly into doorbell
2212 		 * regions, each associated with an egress queue.  If this
2213 		 * per-queue region is large enough (at least UDBS_SEG_SIZE)
2214 		 * then it can be used to submit a tx work request with an
2215 		 * implied doorbell.  Enable write combining on the BAR if
2216 		 * there is room for such work requests.
2217 		 */
2218 		int s_qpp, qpp, num_seg;
2219 
2220 		s_qpp = (S_QUEUESPERPAGEPF0 +
2221 			(S_QUEUESPERPAGEPF1 - S_QUEUESPERPAGEPF0) *
2222 			adapter->pf);
2223 		qpp = 1 << ((t4_read_reg(adapter,
2224 				A_SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp)
2225 				& M_QUEUESPERPAGEPF0);
2226 		num_seg = CXGBE_PAGE_SIZE / UDBS_SEG_SIZE;
2227 		if (qpp > num_seg)
2228 			dev_warn(adapter, "Incorrect SGE EGRESS QUEUES_PER_PAGE configuration, continuing in debug mode\n");
2229 
2230 		adapter->bar2 = (void *)adapter->pdev->mem_resource[2].addr;
2231 		if (!adapter->bar2) {
2232 			dev_err(adapter, "cannot map device bar2 region\n");
2233 			err = -ENOMEM;
2234 			goto out_free;
2235 		}
2236 		t4_write_reg(adapter, A_SGE_STAT_CFG, V_STATSOURCE_T5(7) |
2237 			     V_STATMODE(0));
2238 	}
2239 
2240 	for_each_port(adapter, i) {
2241 		const unsigned int numa_node = rte_socket_id();
2242 		char name[RTE_ETH_NAME_MAX_LEN];
2243 		struct rte_eth_dev *eth_dev;
2244 
2245 		snprintf(name, sizeof(name), "%s_%d",
2246 			 adapter->pdev->device.name, i);
2247 
2248 		if (i == 0) {
2249 			/* First port is already allocated by DPDK */
2250 			eth_dev = adapter->eth_dev;
2251 			goto allocate_mac;
2252 		}
2253 
2254 		/*
2255 		 * now do all data allocation - for eth_dev structure,
2256 		 * and internal (private) data for the remaining ports
2257 		 */
2258 
2259 		/* reserve an ethdev entry */
2260 		eth_dev = rte_eth_dev_allocate(name);
2261 		if (!eth_dev)
2262 			goto out_free;
2263 
2264 		eth_dev->data->dev_private =
2265 			rte_zmalloc_socket(name, sizeof(struct port_info),
2266 					   RTE_CACHE_LINE_SIZE, numa_node);
2267 		if (!eth_dev->data->dev_private)
2268 			goto out_free;
2269 
2270 allocate_mac:
2271 		pi = eth_dev->data->dev_private;
2272 		adapter->port[i] = pi;
2273 		pi->eth_dev = eth_dev;
2274 		pi->adapter = adapter;
2275 		pi->xact_addr_filt = -1;
2276 		pi->port_id = i;
2277 		pi->pidx = i;
2278 
2279 		pi->eth_dev->device = &adapter->pdev->device;
2280 		pi->eth_dev->dev_ops = adapter->eth_dev->dev_ops;
2281 		pi->eth_dev->tx_pkt_burst = adapter->eth_dev->tx_pkt_burst;
2282 		pi->eth_dev->rx_pkt_burst = adapter->eth_dev->rx_pkt_burst;
2283 
2284 		rte_eth_copy_pci_info(pi->eth_dev, adapter->pdev);
2285 
2286 		pi->eth_dev->data->mac_addrs = rte_zmalloc(name,
2287 							RTE_ETHER_ADDR_LEN, 0);
2288 		if (!pi->eth_dev->data->mac_addrs) {
2289 			dev_err(adapter, "%s: Mem allocation failed for storing mac addr, aborting\n",
2290 				__func__);
2291 			err = -1;
2292 			goto out_free;
2293 		}
2294 
2295 		if (i > 0) {
2296 			/* First port will be notified by upper layer */
2297 			rte_eth_dev_probing_finish(eth_dev);
2298 		}
2299 	}
2300 
2301 	if (adapter->flags & FW_OK) {
2302 		err = t4_port_init(adapter, adapter->mbox, adapter->pf, 0);
2303 		if (err) {
2304 			dev_err(adapter, "%s: t4_port_init failed with err %d\n",
2305 				__func__, err);
2306 			goto out_free;
2307 		}
2308 	}
2309 
2310 	err = cxgbe_cfg_queues(adapter->eth_dev);
2311 	if (err)
2312 		goto out_free;
2313 
2314 	cxgbe_print_adapter_info(adapter);
2315 	cxgbe_print_port_info(adapter);
2316 
2317 	adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
2318 					  adapter->clipt_end);
2319 	if (!adapter->clipt) {
2320 		/* We tolerate a lack of clip_table, giving up some
2321 		 * functionality
2322 		 */
2323 		dev_warn(adapter, "could not allocate CLIP. Continuing\n");
2324 	}
2325 
2326 	adap_smt_index(adapter, &smt_start_idx, &smt_size);
2327 	adapter->smt = t4_init_smt(smt_start_idx, smt_size);
2328 	if (!adapter->smt)
2329 		dev_warn(adapter, "could not allocate SMT, continuing\n");
2330 
2331 	adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
2332 	if (!adapter->l2t) {
2333 		/* We tolerate a lack of L2T, giving up some functionality */
2334 		dev_warn(adapter, "could not allocate L2T. Continuing\n");
2335 	}
2336 
2337 	if (tid_init(&adapter->tids) < 0) {
2338 		/* Disable filtering support */
2339 		dev_warn(adapter, "could not allocate TID table, "
2340 			 "filter support disabled. Continuing\n");
2341 	}
2342 
2343 	t4_os_lock_init(&adapter->flow_lock);
2344 
2345 	adapter->mpstcam = t4_init_mpstcam(adapter);
2346 	if (!adapter->mpstcam)
2347 		dev_warn(adapter, "could not allocate mps tcam table."
2348 			 " Continuing\n");
2349 
2350 	if (is_hashfilter(adapter)) {
2351 		if (t4_read_reg(adapter, A_LE_DB_CONFIG) & F_HASHEN) {
2352 			u32 hash_base, hash_reg;
2353 
2354 			hash_reg = A_LE_DB_TID_HASHBASE;
2355 			hash_base = t4_read_reg(adapter, hash_reg);
2356 			adapter->tids.hash_base = hash_base / 4;
2357 		}
2358 	} else {
2359 		/* Disable hash filtering support */
2360 		dev_warn(adapter,
2361 			 "Maskless filter support disabled. Continuing\n");
2362 	}
2363 
2364 	err = cxgbe_init_rss(adapter);
2365 	if (err)
2366 		goto out_free;
2367 
2368 	return 0;
2369 
2370 out_free:
2371 	cxgbe_cfg_queues_free(adapter);
2372 
2373 	for_each_port(adapter, i) {
2374 		pi = adap2pinfo(adapter, i);
2375 		if (pi->viid != 0)
2376 			t4_free_vi(adapter, adapter->mbox, adapter->pf,
2377 				   0, pi->viid);
2378 		rte_eth_dev_release_port(pi->eth_dev);
2379 	}
2380 
2381 	if (adapter->flags & FW_OK)
2382 		t4_fw_bye(adapter, adapter->mbox);
2383 	return -err;
2384 }
2385