xref: /dpdk/drivers/net/cxgbe/cxgbe_main.c (revision a3a2e2c8f7de433e10b1548df65b20bf10086d9c)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2014-2017 Chelsio Communications.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Chelsio Communications nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/queue.h>
35 #include <stdio.h>
36 #include <errno.h>
37 #include <stdint.h>
38 #include <string.h>
39 #include <unistd.h>
40 #include <stdarg.h>
41 #include <inttypes.h>
42 #include <netinet/in.h>
43 
44 #include <rte_byteorder.h>
45 #include <rte_common.h>
46 #include <rte_cycles.h>
47 #include <rte_interrupts.h>
48 #include <rte_log.h>
49 #include <rte_debug.h>
50 #include <rte_pci.h>
51 #include <rte_atomic.h>
52 #include <rte_branch_prediction.h>
53 #include <rte_memory.h>
54 #include <rte_memzone.h>
55 #include <rte_tailq.h>
56 #include <rte_eal.h>
57 #include <rte_alarm.h>
58 #include <rte_ether.h>
59 #include <rte_ethdev.h>
60 #include <rte_ethdev_pci.h>
61 #include <rte_atomic.h>
62 #include <rte_malloc.h>
63 #include <rte_random.h>
64 #include <rte_dev.h>
65 
66 #include "common.h"
67 #include "t4_regs.h"
68 #include "t4_msg.h"
69 #include "cxgbe.h"
70 
71 /*
72  * Response queue handler for the FW event queue.
73  */
74 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
75 			  __rte_unused const struct pkt_gl *gl)
76 {
77 	u8 opcode = ((const struct rss_header *)rsp)->opcode;
78 
79 	rsp++;                                          /* skip RSS header */
80 
81 	/*
82 	 * FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
83 	 */
84 	if (unlikely(opcode == CPL_FW4_MSG &&
85 		     ((const struct cpl_fw4_msg *)rsp)->type ==
86 		      FW_TYPE_RSSCPL)) {
87 		rsp++;
88 		opcode = ((const struct rss_header *)rsp)->opcode;
89 		rsp++;
90 		if (opcode != CPL_SGE_EGR_UPDATE) {
91 			dev_err(q->adapter, "unexpected FW4/CPL %#x on FW event queue\n",
92 				opcode);
93 			goto out;
94 		}
95 	}
96 
97 	if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
98 		/* do nothing */
99 	} else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
100 		const struct cpl_fw6_msg *msg = (const void *)rsp;
101 
102 		t4_handle_fw_rpl(q->adapter, msg->data);
103 	} else {
104 		dev_err(adapter, "unexpected CPL %#x on FW event queue\n",
105 			opcode);
106 	}
107 out:
108 	return 0;
109 }
110 
111 int setup_sge_fwevtq(struct adapter *adapter)
112 {
113 	struct sge *s = &adapter->sge;
114 	int err = 0;
115 	int msi_idx = 0;
116 
117 	err = t4_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->eth_dev,
118 			       msi_idx, NULL, fwevtq_handler, -1, NULL, 0,
119 			       rte_socket_id());
120 	return err;
121 }
122 
123 static int closest_timer(const struct sge *s, int time)
124 {
125 	unsigned int i, match = 0;
126 	int delta, min_delta = INT_MAX;
127 
128 	for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
129 		delta = time - s->timer_val[i];
130 		if (delta < 0)
131 			delta = -delta;
132 		if (delta < min_delta) {
133 			min_delta = delta;
134 			match = i;
135 		}
136 	}
137 	return match;
138 }
139 
140 static int closest_thres(const struct sge *s, int thres)
141 {
142 	unsigned int i, match = 0;
143 	int delta, min_delta = INT_MAX;
144 
145 	for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
146 		delta = thres - s->counter_val[i];
147 		if (delta < 0)
148 			delta = -delta;
149 		if (delta < min_delta) {
150 			min_delta = delta;
151 			match = i;
152 		}
153 	}
154 	return match;
155 }
156 
157 /**
158  * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
159  * @q: the Rx queue
160  * @us: the hold-off time in us, or 0 to disable timer
161  * @cnt: the hold-off packet count, or 0 to disable counter
162  *
163  * Sets an Rx queue's interrupt hold-off time and packet count.  At least
164  * one of the two needs to be enabled for the queue to generate interrupts.
165  */
166 int cxgb4_set_rspq_intr_params(struct sge_rspq *q, unsigned int us,
167 			       unsigned int cnt)
168 {
169 	struct adapter *adap = q->adapter;
170 	unsigned int timer_val;
171 
172 	if (cnt) {
173 		int err;
174 		u32 v, new_idx;
175 
176 		new_idx = closest_thres(&adap->sge, cnt);
177 		if (q->desc && q->pktcnt_idx != new_idx) {
178 			/* the queue has already been created, update it */
179 			v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
180 			    V_FW_PARAMS_PARAM_X(
181 			    FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
182 			    V_FW_PARAMS_PARAM_YZ(q->cntxt_id);
183 			err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
184 					    &v, &new_idx);
185 			if (err)
186 				return err;
187 		}
188 		q->pktcnt_idx = new_idx;
189 	}
190 
191 	timer_val = (us == 0) ? X_TIMERREG_RESTART_COUNTER :
192 				closest_timer(&adap->sge, us);
193 
194 	if ((us | cnt) == 0)
195 		q->intr_params = V_QINTR_TIMER_IDX(X_TIMERREG_UPDATE_CIDX);
196 	else
197 		q->intr_params = V_QINTR_TIMER_IDX(timer_val) |
198 				 V_QINTR_CNT_EN(cnt > 0);
199 	return 0;
200 }
201 
202 static inline bool is_x_1g_port(const struct link_config *lc)
203 {
204 	return (lc->supported & FW_PORT_CAP_SPEED_1G) != 0;
205 }
206 
207 static inline bool is_x_10g_port(const struct link_config *lc)
208 {
209 	unsigned int speeds, high_speeds;
210 
211 	speeds = V_FW_PORT_CAP_SPEED(G_FW_PORT_CAP_SPEED(lc->supported));
212 	high_speeds = speeds & ~(FW_PORT_CAP_SPEED_100M | FW_PORT_CAP_SPEED_1G);
213 
214 	return high_speeds != 0;
215 }
216 
217 inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
218 		      unsigned int us, unsigned int cnt,
219 		      unsigned int size, unsigned int iqe_size)
220 {
221 	q->adapter = adap;
222 	cxgb4_set_rspq_intr_params(q, us, cnt);
223 	q->iqe_len = iqe_size;
224 	q->size = size;
225 }
226 
227 int cfg_queue_count(struct rte_eth_dev *eth_dev)
228 {
229 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
230 	struct adapter *adap = pi->adapter;
231 	struct sge *s = &adap->sge;
232 	unsigned int max_queues = s->max_ethqsets / adap->params.nports;
233 
234 	if ((eth_dev->data->nb_rx_queues < 1) ||
235 	    (eth_dev->data->nb_tx_queues < 1))
236 		return -EINVAL;
237 
238 	if ((eth_dev->data->nb_rx_queues > max_queues) ||
239 	    (eth_dev->data->nb_tx_queues > max_queues))
240 		return -EINVAL;
241 
242 	if (eth_dev->data->nb_rx_queues > pi->rss_size)
243 		return -EINVAL;
244 
245 	/* We must configure RSS, since config has changed*/
246 	pi->flags &= ~PORT_RSS_DONE;
247 
248 	pi->n_rx_qsets = eth_dev->data->nb_rx_queues;
249 	pi->n_tx_qsets = eth_dev->data->nb_tx_queues;
250 
251 	return 0;
252 }
253 
254 void cfg_queues(struct rte_eth_dev *eth_dev)
255 {
256 	struct rte_config *config = rte_eal_get_configuration();
257 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
258 	struct adapter *adap = pi->adapter;
259 	struct sge *s = &adap->sge;
260 	unsigned int i, nb_ports = 0, qidx = 0;
261 	unsigned int q_per_port = 0;
262 
263 	if (!(adap->flags & CFG_QUEUES)) {
264 		for_each_port(adap, i) {
265 			struct port_info *tpi = adap2pinfo(adap, i);
266 
267 			nb_ports += (is_x_10g_port(&tpi->link_cfg)) ||
268 				     is_x_1g_port(&tpi->link_cfg) ? 1 : 0;
269 		}
270 
271 		/*
272 		 * We default up to # of cores queues per 1G/10G port.
273 		 */
274 		if (nb_ports)
275 			q_per_port = (MAX_ETH_QSETS -
276 				     (adap->params.nports - nb_ports)) /
277 				     nb_ports;
278 
279 		if (q_per_port > config->lcore_count)
280 			q_per_port = config->lcore_count;
281 
282 		for_each_port(adap, i) {
283 			struct port_info *pi = adap2pinfo(adap, i);
284 
285 			pi->first_qset = qidx;
286 
287 			/* Initially n_rx_qsets == n_tx_qsets */
288 			pi->n_rx_qsets = (is_x_10g_port(&pi->link_cfg) ||
289 					  is_x_1g_port(&pi->link_cfg)) ?
290 					  q_per_port : 1;
291 			pi->n_tx_qsets = pi->n_rx_qsets;
292 
293 			if (pi->n_rx_qsets > pi->rss_size)
294 				pi->n_rx_qsets = pi->rss_size;
295 
296 			qidx += pi->n_rx_qsets;
297 		}
298 
299 		s->max_ethqsets = qidx;
300 
301 		for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
302 			struct sge_eth_rxq *r = &s->ethrxq[i];
303 
304 			init_rspq(adap, &r->rspq, 5, 32, 1024, 64);
305 			r->usembufs = 1;
306 			r->fl.size = (r->usembufs ? 1024 : 72);
307 		}
308 
309 		for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
310 			s->ethtxq[i].q.size = 1024;
311 
312 		init_rspq(adap, &adap->sge.fw_evtq, 0, 0, 1024, 64);
313 		adap->flags |= CFG_QUEUES;
314 	}
315 }
316 
317 void cxgbe_stats_get(struct port_info *pi, struct port_stats *stats)
318 {
319 	t4_get_port_stats_offset(pi->adapter, pi->tx_chan, stats,
320 				 &pi->stats_base);
321 }
322 
323 void cxgbe_stats_reset(struct port_info *pi)
324 {
325 	t4_clr_port_stats(pi->adapter, pi->tx_chan);
326 }
327 
328 static void setup_memwin(struct adapter *adap)
329 {
330 	u32 mem_win0_base;
331 
332 	/* For T5, only relative offset inside the PCIe BAR is passed */
333 	mem_win0_base = MEMWIN0_BASE;
334 
335 	/*
336 	 * Set up memory window for accessing adapter memory ranges.  (Read
337 	 * back MA register to ensure that changes propagate before we attempt
338 	 * to use the new values.)
339 	 */
340 	t4_write_reg(adap,
341 		     PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
342 					 MEMWIN_NIC),
343 		     mem_win0_base | V_BIR(0) |
344 		     V_WINDOW(ilog2(MEMWIN0_APERTURE) - X_WINDOW_SHIFT));
345 	t4_read_reg(adap,
346 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
347 					MEMWIN_NIC));
348 }
349 
350 static int init_rss(struct adapter *adap)
351 {
352 	unsigned int i;
353 	int err;
354 
355 	err = t4_init_rss_mode(adap, adap->mbox);
356 	if (err)
357 		return err;
358 
359 	for_each_port(adap, i) {
360 		struct port_info *pi = adap2pinfo(adap, i);
361 
362 		pi->rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
363 		if (!pi->rss)
364 			return -ENOMEM;
365 	}
366 	return 0;
367 }
368 
369 /**
370  * Dump basic information about the adapter.
371  */
372 static void print_adapter_info(struct adapter *adap)
373 {
374 	/**
375 	 * Hardware/Firmware/etc. Version/Revision IDs.
376 	 */
377 	t4_dump_version_info(adap);
378 }
379 
380 static void print_port_info(struct adapter *adap)
381 {
382 	int i;
383 	char buf[80];
384 	struct rte_pci_addr *loc = &adap->pdev->addr;
385 
386 	for_each_port(adap, i) {
387 		const struct port_info *pi = &adap->port[i];
388 		char *bufp = buf;
389 
390 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
391 			bufp += sprintf(bufp, "100M/");
392 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
393 			bufp += sprintf(bufp, "1G/");
394 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
395 			bufp += sprintf(bufp, "10G/");
396 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_25G)
397 			bufp += sprintf(bufp, "25G/");
398 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G)
399 			bufp += sprintf(bufp, "40G/");
400 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100G)
401 			bufp += sprintf(bufp, "100G/");
402 		if (bufp != buf)
403 			--bufp;
404 		sprintf(bufp, "BASE-%s",
405 			t4_get_port_type_description(
406 					(enum fw_port_type)pi->port_type));
407 
408 		dev_info(adap,
409 			 " " PCI_PRI_FMT " Chelsio rev %d %s %s\n",
410 			 loc->domain, loc->bus, loc->devid, loc->function,
411 			 CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
412 			 (adap->flags & USING_MSIX) ? " MSI-X" :
413 			 (adap->flags & USING_MSI) ? " MSI" : "");
414 	}
415 }
416 
417 static void configure_pcie_ext_tag(struct adapter *adapter)
418 {
419 	u16 v;
420 	int pos = t4_os_find_pci_capability(adapter, PCI_CAP_ID_EXP);
421 
422 	if (!pos)
423 		return;
424 
425 	if (pos > 0) {
426 		t4_os_pci_read_cfg2(adapter, pos + PCI_EXP_DEVCTL, &v);
427 		v |= PCI_EXP_DEVCTL_EXT_TAG;
428 		t4_os_pci_write_cfg2(adapter, pos + PCI_EXP_DEVCTL, v);
429 		if (is_t6(adapter->params.chip)) {
430 			t4_set_reg_field(adapter, A_PCIE_CFG2,
431 					 V_T6_TOTMAXTAG(M_T6_TOTMAXTAG),
432 					 V_T6_TOTMAXTAG(7));
433 			t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
434 					 V_T6_MINTAG(M_T6_MINTAG),
435 					 V_T6_MINTAG(8));
436 		} else {
437 			t4_set_reg_field(adapter, A_PCIE_CFG2,
438 					 V_TOTMAXTAG(M_TOTMAXTAG),
439 					 V_TOTMAXTAG(3));
440 			t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
441 					 V_MINTAG(M_MINTAG),
442 					 V_MINTAG(8));
443 		}
444 	}
445 }
446 
447 /*
448  * Tweak configuration based on system architecture, etc.  Most of these have
449  * defaults assigned to them by Firmware Configuration Files (if we're using
450  * them) but need to be explicitly set if we're using hard-coded
451  * initialization. So these are essentially common tweaks/settings for
452  * Configuration Files and hard-coded initialization ...
453  */
454 static int adap_init0_tweaks(struct adapter *adapter)
455 {
456 	u8 rx_dma_offset;
457 
458 	/*
459 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
460 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
461 	 * 64B Cache Line Size ...
462 	 */
463 	t4_fixup_host_params_compat(adapter, CXGBE_PAGE_SIZE, L1_CACHE_BYTES,
464 				    T5_LAST_REV);
465 
466 	/*
467 	 * Keep the chip default offset to deliver Ingress packets into our
468 	 * DMA buffers to zero
469 	 */
470 	rx_dma_offset = 0;
471 	t4_set_reg_field(adapter, A_SGE_CONTROL, V_PKTSHIFT(M_PKTSHIFT),
472 			 V_PKTSHIFT(rx_dma_offset));
473 
474 	t4_set_reg_field(adapter, A_SGE_FLM_CFG,
475 			 V_CREDITCNT(M_CREDITCNT) | M_CREDITCNTPACKING,
476 			 V_CREDITCNT(3) | V_CREDITCNTPACKING(1));
477 
478 	t4_set_reg_field(adapter, A_SGE_INGRESS_RX_THRESHOLD,
479 			 V_THRESHOLD_3(M_THRESHOLD_3), V_THRESHOLD_3(32U));
480 
481 	t4_set_reg_field(adapter, A_SGE_CONTROL2, V_IDMAARBROUNDROBIN(1U),
482 			 V_IDMAARBROUNDROBIN(1U));
483 
484 	/*
485 	 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
486 	 * adds the pseudo header itself.
487 	 */
488 	t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG,
489 			       F_CSUM_HAS_PSEUDO_HDR, 0);
490 
491 	return 0;
492 }
493 
494 /*
495  * Attempt to initialize the adapter via a Firmware Configuration File.
496  */
497 static int adap_init0_config(struct adapter *adapter, int reset)
498 {
499 	struct fw_caps_config_cmd caps_cmd;
500 	unsigned long mtype = 0, maddr = 0;
501 	u32 finiver, finicsum, cfcsum;
502 	int ret;
503 	int config_issued = 0;
504 	int cfg_addr;
505 	char config_name[20];
506 
507 	/*
508 	 * Reset device if necessary.
509 	 */
510 	if (reset) {
511 		ret = t4_fw_reset(adapter, adapter->mbox,
512 				  F_PIORSTMODE | F_PIORST);
513 		if (ret < 0) {
514 			dev_warn(adapter, "Firmware reset failed, error %d\n",
515 				 -ret);
516 			goto bye;
517 		}
518 	}
519 
520 	cfg_addr = t4_flash_cfg_addr(adapter);
521 	if (cfg_addr < 0) {
522 		ret = cfg_addr;
523 		dev_warn(adapter, "Finding address for firmware config file in flash failed, error %d\n",
524 			 -ret);
525 		goto bye;
526 	}
527 
528 	strcpy(config_name, "On Flash");
529 	mtype = FW_MEMTYPE_CF_FLASH;
530 	maddr = cfg_addr;
531 
532 	/*
533 	 * Issue a Capability Configuration command to the firmware to get it
534 	 * to parse the Configuration File.  We don't use t4_fw_config_file()
535 	 * because we want the ability to modify various features after we've
536 	 * processed the configuration file ...
537 	 */
538 	memset(&caps_cmd, 0, sizeof(caps_cmd));
539 	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
540 					   F_FW_CMD_REQUEST | F_FW_CMD_READ);
541 	caps_cmd.cfvalid_to_len16 =
542 		cpu_to_be32(F_FW_CAPS_CONFIG_CMD_CFVALID |
543 			    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
544 			    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
545 			    FW_LEN16(caps_cmd));
546 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
547 			 &caps_cmd);
548 	/*
549 	 * If the CAPS_CONFIG failed with an ENOENT (for a Firmware
550 	 * Configuration File in FLASH), our last gasp effort is to use the
551 	 * Firmware Configuration File which is embedded in the firmware.  A
552 	 * very few early versions of the firmware didn't have one embedded
553 	 * but we can ignore those.
554 	 */
555 	if (ret == -ENOENT) {
556 		dev_info(adapter, "%s: Going for embedded config in firmware..\n",
557 			 __func__);
558 
559 		memset(&caps_cmd, 0, sizeof(caps_cmd));
560 		caps_cmd.op_to_write =
561 			cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
562 				    F_FW_CMD_REQUEST | F_FW_CMD_READ);
563 		caps_cmd.cfvalid_to_len16 = cpu_to_be32(FW_LEN16(caps_cmd));
564 		ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
565 				 sizeof(caps_cmd), &caps_cmd);
566 		strcpy(config_name, "Firmware Default");
567 	}
568 
569 	config_issued = 1;
570 	if (ret < 0)
571 		goto bye;
572 
573 	finiver = be32_to_cpu(caps_cmd.finiver);
574 	finicsum = be32_to_cpu(caps_cmd.finicsum);
575 	cfcsum = be32_to_cpu(caps_cmd.cfcsum);
576 	if (finicsum != cfcsum)
577 		dev_warn(adapter, "Configuration File checksum mismatch: [fini] csum=%#x, computed csum=%#x\n",
578 			 finicsum, cfcsum);
579 
580 	/*
581 	 * If we're a pure NIC driver then disable all offloading facilities.
582 	 * This will allow the firmware to optimize aspects of the hardware
583 	 * configuration which will result in improved performance.
584 	 */
585 	caps_cmd.niccaps &= cpu_to_be16(~(FW_CAPS_CONFIG_NIC_HASHFILTER |
586 					  FW_CAPS_CONFIG_NIC_ETHOFLD));
587 	caps_cmd.toecaps = 0;
588 	caps_cmd.iscsicaps = 0;
589 	caps_cmd.rdmacaps = 0;
590 	caps_cmd.fcoecaps = 0;
591 
592 	/*
593 	 * And now tell the firmware to use the configuration we just loaded.
594 	 */
595 	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
596 					   F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
597 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
598 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
599 			 NULL);
600 	if (ret < 0) {
601 		dev_warn(adapter, "Unable to finalize Firmware Capabilities %d\n",
602 			 -ret);
603 		goto bye;
604 	}
605 
606 	/*
607 	 * Tweak configuration based on system architecture, etc.
608 	 */
609 	ret = adap_init0_tweaks(adapter);
610 	if (ret < 0) {
611 		dev_warn(adapter, "Unable to do init0-tweaks %d\n", -ret);
612 		goto bye;
613 	}
614 
615 	/*
616 	 * And finally tell the firmware to initialize itself using the
617 	 * parameters from the Configuration File.
618 	 */
619 	ret = t4_fw_initialize(adapter, adapter->mbox);
620 	if (ret < 0) {
621 		dev_warn(adapter, "Initializing Firmware failed, error %d\n",
622 			 -ret);
623 		goto bye;
624 	}
625 
626 	/*
627 	 * Return successfully and note that we're operating with parameters
628 	 * not supplied by the driver, rather than from hard-wired
629 	 * initialization constants buried in the driver.
630 	 */
631 	dev_info(adapter,
632 		 "Successfully configured using Firmware Configuration File \"%s\", version %#x, computed checksum %#x\n",
633 		 config_name, finiver, cfcsum);
634 
635 	return 0;
636 
637 	/*
638 	 * Something bad happened.  Return the error ...  (If the "error"
639 	 * is that there's no Configuration File on the adapter we don't
640 	 * want to issue a warning since this is fairly common.)
641 	 */
642 bye:
643 	if (config_issued && ret != -ENOENT)
644 		dev_warn(adapter, "\"%s\" configuration file error %d\n",
645 			 config_name, -ret);
646 
647 	dev_debug(adapter, "%s: returning ret = %d ..\n", __func__, ret);
648 	return ret;
649 }
650 
651 static int adap_init0(struct adapter *adap)
652 {
653 	int ret = 0;
654 	u32 v, port_vec;
655 	enum dev_state state;
656 	u32 params[7], val[7];
657 	int reset = 1;
658 	int mbox = adap->mbox;
659 
660 	/*
661 	 * Contact FW, advertising Master capability.
662 	 */
663 	ret = t4_fw_hello(adap, adap->mbox, adap->mbox, MASTER_MAY, &state);
664 	if (ret < 0) {
665 		dev_err(adap, "%s: could not connect to FW, error %d\n",
666 			__func__, -ret);
667 		goto bye;
668 	}
669 
670 	CXGBE_DEBUG_MBOX(adap, "%s: adap->mbox = %d; ret = %d\n", __func__,
671 			 adap->mbox, ret);
672 
673 	if (ret == mbox)
674 		adap->flags |= MASTER_PF;
675 
676 	if (state == DEV_STATE_INIT) {
677 		/*
678 		 * Force halt and reset FW because a previous instance may have
679 		 * exited abnormally without properly shutting down
680 		 */
681 		ret = t4_fw_halt(adap, adap->mbox, reset);
682 		if (ret < 0) {
683 			dev_err(adap, "Failed to halt. Exit.\n");
684 			goto bye;
685 		}
686 
687 		ret = t4_fw_restart(adap, adap->mbox, reset);
688 		if (ret < 0) {
689 			dev_err(adap, "Failed to restart. Exit.\n");
690 			goto bye;
691 		}
692 		state = (enum dev_state)((unsigned)state & ~DEV_STATE_INIT);
693 	}
694 
695 	t4_get_version_info(adap);
696 
697 	ret = t4_get_core_clock(adap, &adap->params.vpd);
698 	if (ret < 0) {
699 		dev_err(adap, "%s: could not get core clock, error %d\n",
700 			__func__, -ret);
701 		goto bye;
702 	}
703 
704 	/*
705 	 * If the firmware is initialized already (and we're not forcing a
706 	 * master initialization), note that we're living with existing
707 	 * adapter parameters.  Otherwise, it's time to try initializing the
708 	 * adapter ...
709 	 */
710 	if (state == DEV_STATE_INIT) {
711 		dev_info(adap, "Coming up as %s: Adapter already initialized\n",
712 			 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
713 	} else {
714 		dev_info(adap, "Coming up as MASTER: Initializing adapter\n");
715 
716 		ret = adap_init0_config(adap, reset);
717 		if (ret == -ENOENT) {
718 			dev_err(adap,
719 				"No Configuration File present on adapter. Using hard-wired configuration parameters.\n");
720 			goto bye;
721 		}
722 	}
723 	if (ret < 0) {
724 		dev_err(adap, "could not initialize adapter, error %d\n", -ret);
725 		goto bye;
726 	}
727 
728 	/* Find out what ports are available to us. */
729 	v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
730 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
731 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
732 	if (ret < 0) {
733 		dev_err(adap, "%s: failure in t4_query_params; error = %d\n",
734 			__func__, ret);
735 		goto bye;
736 	}
737 
738 	adap->params.nports = hweight32(port_vec);
739 	adap->params.portvec = port_vec;
740 
741 	dev_debug(adap, "%s: adap->params.nports = %u\n", __func__,
742 		  adap->params.nports);
743 
744 	/*
745 	 * Give the SGE code a chance to pull in anything that it needs ...
746 	 * Note that this must be called after we retrieve our VPD parameters
747 	 * in order to know how to convert core ticks to seconds, etc.
748 	 */
749 	ret = t4_sge_init(adap);
750 	if (ret < 0) {
751 		dev_err(adap, "t4_sge_init failed with error %d\n",
752 			-ret);
753 		goto bye;
754 	}
755 
756 	/*
757 	 * Grab some of our basic fundamental operating parameters.
758 	 */
759 #define FW_PARAM_DEV(param) \
760 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
761 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
762 
763 #define FW_PARAM_PFVF(param) \
764 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
765 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param) |  \
766 	 V_FW_PARAMS_PARAM_Y(0) | \
767 	 V_FW_PARAMS_PARAM_Z(0))
768 
769 	/* If we're running on newer firmware, let it know that we're
770 	 * prepared to deal with encapsulated CPL messages.  Older
771 	 * firmware won't understand this and we'll just get
772 	 * unencapsulated messages ...
773 	 */
774 	params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
775 	val[0] = 1;
776 	(void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
777 
778 	/*
779 	 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
780 	 * capability.  Earlier versions of the firmware didn't have the
781 	 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
782 	 * permission to use ULPTX MEMWRITE DSGL.
783 	 */
784 	if (is_t4(adap->params.chip)) {
785 		adap->params.ulptx_memwrite_dsgl = false;
786 	} else {
787 		params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
788 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
789 				      1, params, val);
790 		adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
791 	}
792 
793 	/*
794 	 * The MTU/MSS Table is initialized by now, so load their values.  If
795 	 * we're initializing the adapter, then we'll make any modifications
796 	 * we want to the MTU/MSS Table and also initialize the congestion
797 	 * parameters.
798 	 */
799 	t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
800 	if (state != DEV_STATE_INIT) {
801 		int i;
802 
803 		/*
804 		 * The default MTU Table contains values 1492 and 1500.
805 		 * However, for TCP, it's better to have two values which are
806 		 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
807 		 * This allows us to have a TCP Data Payload which is a
808 		 * multiple of 8 regardless of what combination of TCP Options
809 		 * are in use (always a multiple of 4 bytes) which is
810 		 * important for performance reasons.  For instance, if no
811 		 * options are in use, then we have a 20-byte IP header and a
812 		 * 20-byte TCP header.  In this case, a 1500-byte MSS would
813 		 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
814 		 * which is not a multiple of 8.  So using an MSS of 1488 in
815 		 * this case results in a TCP Data Payload of 1448 bytes which
816 		 * is a multiple of 8.  On the other hand, if 12-byte TCP Time
817 		 * Stamps have been negotiated, then an MTU of 1500 bytes
818 		 * results in a TCP Data Payload of 1448 bytes which, as
819 		 * above, is a multiple of 8 bytes ...
820 		 */
821 		for (i = 0; i < NMTUS; i++)
822 			if (adap->params.mtus[i] == 1492) {
823 				adap->params.mtus[i] = 1488;
824 				break;
825 			}
826 
827 		t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
828 			     adap->params.b_wnd);
829 	}
830 	t4_init_sge_params(adap);
831 	t4_init_tp_params(adap);
832 	configure_pcie_ext_tag(adap);
833 
834 	adap->params.drv_memwin = MEMWIN_NIC;
835 	adap->flags |= FW_OK;
836 	dev_debug(adap, "%s: returning zero..\n", __func__);
837 	return 0;
838 
839 	/*
840 	 * Something bad happened.  If a command timed out or failed with EIO
841 	 * FW does not operate within its spec or something catastrophic
842 	 * happened to HW/FW, stop issuing commands.
843 	 */
844 bye:
845 	if (ret != -ETIMEDOUT && ret != -EIO)
846 		t4_fw_bye(adap, adap->mbox);
847 	return ret;
848 }
849 
850 /**
851  * t4_os_portmod_changed - handle port module changes
852  * @adap: the adapter associated with the module change
853  * @port_id: the port index whose module status has changed
854  *
855  * This is the OS-dependent handler for port module changes.  It is
856  * invoked when a port module is removed or inserted for any OS-specific
857  * processing.
858  */
859 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
860 {
861 	static const char * const mod_str[] = {
862 		NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
863 	};
864 
865 	const struct port_info *pi = &adap->port[port_id];
866 
867 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
868 		dev_info(adap, "Port%d: port module unplugged\n", pi->port_id);
869 	else if (pi->mod_type < ARRAY_SIZE(mod_str))
870 		dev_info(adap, "Port%d: %s port module inserted\n", pi->port_id,
871 			 mod_str[pi->mod_type]);
872 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
873 		dev_info(adap, "Port%d: unsupported port module inserted\n",
874 			 pi->port_id);
875 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
876 		dev_info(adap, "Port%d: unknown port module inserted\n",
877 			 pi->port_id);
878 	else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
879 		dev_info(adap, "Port%d: transceiver module error\n",
880 			 pi->port_id);
881 	else
882 		dev_info(adap, "Port%d: unknown module type %d inserted\n",
883 			 pi->port_id, pi->mod_type);
884 }
885 
886 /**
887  * link_start - enable a port
888  * @dev: the port to enable
889  *
890  * Performs the MAC and PHY actions needed to enable a port.
891  */
892 int link_start(struct port_info *pi)
893 {
894 	struct adapter *adapter = pi->adapter;
895 	int ret;
896 	unsigned int mtu;
897 
898 	mtu = pi->eth_dev->data->dev_conf.rxmode.max_rx_pkt_len -
899 	      (ETHER_HDR_LEN + ETHER_CRC_LEN);
900 
901 	/*
902 	 * We do not set address filters and promiscuity here, the stack does
903 	 * that step explicitly.
904 	 */
905 	ret = t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu, -1, -1,
906 			    -1, 1, true);
907 	if (ret == 0) {
908 		ret = t4_change_mac(adapter, adapter->mbox, pi->viid,
909 				    pi->xact_addr_filt,
910 				    (u8 *)&pi->eth_dev->data->mac_addrs[0],
911 				    true, true);
912 		if (ret >= 0) {
913 			pi->xact_addr_filt = ret;
914 			ret = 0;
915 		}
916 	}
917 	if (ret == 0)
918 		ret = t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
919 				    &pi->link_cfg);
920 	if (ret == 0) {
921 		/*
922 		 * Enabling a Virtual Interface can result in an interrupt
923 		 * during the processing of the VI Enable command and, in some
924 		 * paths, result in an attempt to issue another command in the
925 		 * interrupt context.  Thus, we disable interrupts during the
926 		 * course of the VI Enable command ...
927 		 */
928 		ret = t4_enable_vi_params(adapter, adapter->mbox, pi->viid,
929 					  true, true, false);
930 	}
931 	return ret;
932 }
933 
934 /**
935  * cxgb4_write_rss - write the RSS table for a given port
936  * @pi: the port
937  * @queues: array of queue indices for RSS
938  *
939  * Sets up the portion of the HW RSS table for the port's VI to distribute
940  * packets to the Rx queues in @queues.
941  */
942 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
943 {
944 	u16 *rss;
945 	int i, err;
946 	struct adapter *adapter = pi->adapter;
947 	const struct sge_eth_rxq *rxq;
948 
949 	/*  Should never be called before setting up sge eth rx queues */
950 	BUG_ON(!(adapter->flags & FULL_INIT_DONE));
951 
952 	rxq = &adapter->sge.ethrxq[pi->first_qset];
953 	rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
954 	if (!rss)
955 		return -ENOMEM;
956 
957 	/* map the queue indices to queue ids */
958 	for (i = 0; i < pi->rss_size; i++, queues++)
959 		rss[i] = rxq[*queues].rspq.abs_id;
960 
961 	err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
962 				  pi->rss_size, rss, pi->rss_size);
963 	/*
964 	 * If Tunnel All Lookup isn't specified in the global RSS
965 	 * Configuration, then we need to specify a default Ingress
966 	 * Queue for any ingress packets which aren't hashed.  We'll
967 	 * use our first ingress queue ...
968 	 */
969 	if (!err)
970 		err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
971 				       F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
972 				       F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
973 				       F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
974 				       F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN |
975 				       F_FW_RSS_VI_CONFIG_CMD_UDPEN,
976 				       rss[0]);
977 	rte_free(rss);
978 	return err;
979 }
980 
981 /**
982  * setup_rss - configure RSS
983  * @adapter: the adapter
984  *
985  * Sets up RSS to distribute packets to multiple receive queues.  We
986  * configure the RSS CPU lookup table to distribute to the number of HW
987  * receive queues, and the response queue lookup table to narrow that
988  * down to the response queues actually configured for each port.
989  * We always configure the RSS mapping for all ports since the mapping
990  * table has plenty of entries.
991  */
992 int setup_rss(struct port_info *pi)
993 {
994 	int j, err;
995 	struct adapter *adapter = pi->adapter;
996 
997 	dev_debug(adapter, "%s:  pi->rss_size = %u; pi->n_rx_qsets = %u\n",
998 		  __func__, pi->rss_size, pi->n_rx_qsets);
999 
1000 	if (!(pi->flags & PORT_RSS_DONE)) {
1001 		if (adapter->flags & FULL_INIT_DONE) {
1002 			/* Fill default values with equal distribution */
1003 			for (j = 0; j < pi->rss_size; j++)
1004 				pi->rss[j] = j % pi->n_rx_qsets;
1005 
1006 			err = cxgb4_write_rss(pi, pi->rss);
1007 			if (err)
1008 				return err;
1009 			pi->flags |= PORT_RSS_DONE;
1010 		}
1011 	}
1012 	return 0;
1013 }
1014 
1015 /*
1016  * Enable NAPI scheduling and interrupt generation for all Rx queues.
1017  */
1018 static void enable_rx(struct adapter *adap, struct sge_rspq *q)
1019 {
1020 	/* 0-increment GTS to start the timer and enable interrupts */
1021 	t4_write_reg(adap, MYPF_REG(A_SGE_PF_GTS),
1022 		     V_SEINTARM(q->intr_params) |
1023 		     V_INGRESSQID(q->cntxt_id));
1024 }
1025 
1026 void cxgbe_enable_rx_queues(struct port_info *pi)
1027 {
1028 	struct adapter *adap = pi->adapter;
1029 	struct sge *s = &adap->sge;
1030 	unsigned int i;
1031 
1032 	for (i = 0; i < pi->n_rx_qsets; i++)
1033 		enable_rx(adap, &s->ethrxq[pi->first_qset + i].rspq);
1034 }
1035 
1036 /**
1037  * cxgb_up - enable the adapter
1038  * @adap: adapter being enabled
1039  *
1040  * Called when the first port is enabled, this function performs the
1041  * actions necessary to make an adapter operational, such as completing
1042  * the initialization of HW modules, and enabling interrupts.
1043  */
1044 int cxgbe_up(struct adapter *adap)
1045 {
1046 	enable_rx(adap, &adap->sge.fw_evtq);
1047 	t4_sge_tx_monitor_start(adap);
1048 	t4_intr_enable(adap);
1049 	adap->flags |= FULL_INIT_DONE;
1050 
1051 	/* TODO: deadman watchdog ?? */
1052 	return 0;
1053 }
1054 
1055 /*
1056  * Close the port
1057  */
1058 int cxgbe_down(struct port_info *pi)
1059 {
1060 	struct adapter *adapter = pi->adapter;
1061 	int err = 0;
1062 
1063 	err = t4_enable_vi(adapter, adapter->mbox, pi->viid, false, false);
1064 	if (err) {
1065 		dev_err(adapter, "%s: disable_vi failed: %d\n", __func__, err);
1066 		return err;
1067 	}
1068 
1069 	t4_reset_link_config(adapter, pi->port_id);
1070 	return 0;
1071 }
1072 
1073 /*
1074  * Release resources when all the ports have been stopped.
1075  */
1076 void cxgbe_close(struct adapter *adapter)
1077 {
1078 	struct port_info *pi;
1079 	int i;
1080 
1081 	if (adapter->flags & FULL_INIT_DONE) {
1082 		t4_intr_disable(adapter);
1083 		t4_sge_tx_monitor_stop(adapter);
1084 		t4_free_sge_resources(adapter);
1085 		for_each_port(adapter, i) {
1086 			pi = adap2pinfo(adapter, i);
1087 			if (pi->viid != 0)
1088 				t4_free_vi(adapter, adapter->mbox,
1089 					   adapter->pf, 0, pi->viid);
1090 			rte_free(pi->eth_dev->data->mac_addrs);
1091 		}
1092 		adapter->flags &= ~FULL_INIT_DONE;
1093 	}
1094 
1095 	if (adapter->flags & FW_OK)
1096 		t4_fw_bye(adapter, adapter->mbox);
1097 }
1098 
1099 int cxgbe_probe(struct adapter *adapter)
1100 {
1101 	struct port_info *pi;
1102 	int chip;
1103 	int func, i;
1104 	int err = 0;
1105 	u32 whoami;
1106 
1107 	whoami = t4_read_reg(adapter, A_PL_WHOAMI);
1108 	chip = t4_get_chip_type(adapter,
1109 			CHELSIO_PCI_ID_VER(adapter->pdev->id.device_id));
1110 	if (chip < 0)
1111 		return chip;
1112 
1113 	func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
1114 	       G_SOURCEPF(whoami) : G_T6_SOURCEPF(whoami);
1115 
1116 	adapter->mbox = func;
1117 	adapter->pf = func;
1118 
1119 	t4_os_lock_init(&adapter->mbox_lock);
1120 	TAILQ_INIT(&adapter->mbox_list);
1121 
1122 	err = t4_prep_adapter(adapter);
1123 	if (err)
1124 		return err;
1125 
1126 	setup_memwin(adapter);
1127 	err = adap_init0(adapter);
1128 	if (err) {
1129 		dev_err(adapter, "%s: Adapter initialization failed, error %d\n",
1130 			__func__, err);
1131 		goto out_free;
1132 	}
1133 
1134 	if (!is_t4(adapter->params.chip)) {
1135 		/*
1136 		 * The userspace doorbell BAR is split evenly into doorbell
1137 		 * regions, each associated with an egress queue.  If this
1138 		 * per-queue region is large enough (at least UDBS_SEG_SIZE)
1139 		 * then it can be used to submit a tx work request with an
1140 		 * implied doorbell.  Enable write combining on the BAR if
1141 		 * there is room for such work requests.
1142 		 */
1143 		int s_qpp, qpp, num_seg;
1144 
1145 		s_qpp = (S_QUEUESPERPAGEPF0 +
1146 			(S_QUEUESPERPAGEPF1 - S_QUEUESPERPAGEPF0) *
1147 			adapter->pf);
1148 		qpp = 1 << ((t4_read_reg(adapter,
1149 				A_SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp)
1150 				& M_QUEUESPERPAGEPF0);
1151 		num_seg = CXGBE_PAGE_SIZE / UDBS_SEG_SIZE;
1152 		if (qpp > num_seg)
1153 			dev_warn(adapter, "Incorrect SGE EGRESS QUEUES_PER_PAGE configuration, continuing in debug mode\n");
1154 
1155 		adapter->bar2 = (void *)adapter->pdev->mem_resource[2].addr;
1156 		if (!adapter->bar2) {
1157 			dev_err(adapter, "cannot map device bar2 region\n");
1158 			err = -ENOMEM;
1159 			goto out_free;
1160 		}
1161 		t4_write_reg(adapter, A_SGE_STAT_CFG, V_STATSOURCE_T5(7) |
1162 			     V_STATMODE(0));
1163 	}
1164 
1165 	for_each_port(adapter, i) {
1166 		char name[RTE_ETH_NAME_MAX_LEN];
1167 		struct rte_eth_dev_data *data = NULL;
1168 		const unsigned int numa_node = rte_socket_id();
1169 
1170 		pi = &adapter->port[i];
1171 		pi->adapter = adapter;
1172 		pi->xact_addr_filt = -1;
1173 		pi->port_id = i;
1174 
1175 		snprintf(name, sizeof(name), "cxgbe%d",
1176 			 adapter->eth_dev->data->port_id + i);
1177 
1178 		if (i == 0) {
1179 			/* First port is already allocated by DPDK */
1180 			pi->eth_dev = adapter->eth_dev;
1181 			goto allocate_mac;
1182 		}
1183 
1184 		/*
1185 		 * now do all data allocation - for eth_dev structure,
1186 		 * and internal (private) data for the remaining ports
1187 		 */
1188 
1189 		/* reserve an ethdev entry */
1190 		pi->eth_dev = rte_eth_dev_allocate(name);
1191 		if (!pi->eth_dev)
1192 			goto out_free;
1193 
1194 		data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1195 		if (!data)
1196 			goto out_free;
1197 
1198 		data->port_id = adapter->eth_dev->data->port_id + i;
1199 
1200 		pi->eth_dev->data = data;
1201 
1202 allocate_mac:
1203 		pi->eth_dev->device = &adapter->pdev->device;
1204 		pi->eth_dev->data->dev_private = pi;
1205 		pi->eth_dev->dev_ops = adapter->eth_dev->dev_ops;
1206 		pi->eth_dev->tx_pkt_burst = adapter->eth_dev->tx_pkt_burst;
1207 		pi->eth_dev->rx_pkt_burst = adapter->eth_dev->rx_pkt_burst;
1208 
1209 		rte_eth_copy_pci_info(pi->eth_dev, adapter->pdev);
1210 
1211 		pi->eth_dev->data->mac_addrs = rte_zmalloc(name,
1212 							   ETHER_ADDR_LEN, 0);
1213 		if (!pi->eth_dev->data->mac_addrs) {
1214 			dev_err(adapter, "%s: Mem allocation failed for storing mac addr, aborting\n",
1215 				__func__);
1216 			err = -1;
1217 			goto out_free;
1218 		}
1219 	}
1220 
1221 	if (adapter->flags & FW_OK) {
1222 		err = t4_port_init(adapter, adapter->mbox, adapter->pf, 0);
1223 		if (err) {
1224 			dev_err(adapter, "%s: t4_port_init failed with err %d\n",
1225 				__func__, err);
1226 			goto out_free;
1227 		}
1228 	}
1229 
1230 	cfg_queues(adapter->eth_dev);
1231 
1232 	print_adapter_info(adapter);
1233 	print_port_info(adapter);
1234 
1235 	err = init_rss(adapter);
1236 	if (err)
1237 		goto out_free;
1238 
1239 	return 0;
1240 
1241 out_free:
1242 	for_each_port(adapter, i) {
1243 		pi = adap2pinfo(adapter, i);
1244 		if (pi->viid != 0)
1245 			t4_free_vi(adapter, adapter->mbox, adapter->pf,
1246 				   0, pi->viid);
1247 		/* Skip first port since it'll be de-allocated by DPDK */
1248 		if (i == 0)
1249 			continue;
1250 		if (pi->eth_dev->data)
1251 			rte_free(pi->eth_dev->data);
1252 	}
1253 
1254 	if (adapter->flags & FW_OK)
1255 		t4_fw_bye(adapter, adapter->mbox);
1256 	return -err;
1257 }
1258