xref: /dpdk/drivers/net/cxgbe/cxgbe_main.c (revision a599eb31f2e477674fc6176cdf989ee17432b552)
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_malloc.h>
62 #include <rte_random.h>
63 #include <rte_dev.h>
64 
65 #include "common.h"
66 #include "t4_regs.h"
67 #include "t4_msg.h"
68 #include "cxgbe.h"
69 
70 /*
71  * Response queue handler for the FW event queue.
72  */
73 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
74 			  __rte_unused const struct pkt_gl *gl)
75 {
76 	u8 opcode = ((const struct rss_header *)rsp)->opcode;
77 
78 	rsp++;                                          /* skip RSS header */
79 
80 	/*
81 	 * FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
82 	 */
83 	if (unlikely(opcode == CPL_FW4_MSG &&
84 		     ((const struct cpl_fw4_msg *)rsp)->type ==
85 		      FW_TYPE_RSSCPL)) {
86 		rsp++;
87 		opcode = ((const struct rss_header *)rsp)->opcode;
88 		rsp++;
89 		if (opcode != CPL_SGE_EGR_UPDATE) {
90 			dev_err(q->adapter, "unexpected FW4/CPL %#x on FW event queue\n",
91 				opcode);
92 			goto out;
93 		}
94 	}
95 
96 	if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
97 		/* do nothing */
98 	} else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
99 		const struct cpl_fw6_msg *msg = (const void *)rsp;
100 
101 		t4_handle_fw_rpl(q->adapter, msg->data);
102 	} else {
103 		dev_err(adapter, "unexpected CPL %#x on FW event queue\n",
104 			opcode);
105 	}
106 out:
107 	return 0;
108 }
109 
110 int setup_sge_fwevtq(struct adapter *adapter)
111 {
112 	struct sge *s = &adapter->sge;
113 	int err = 0;
114 	int msi_idx = 0;
115 
116 	err = t4_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->eth_dev,
117 			       msi_idx, NULL, fwevtq_handler, -1, NULL, 0,
118 			       rte_socket_id());
119 	return err;
120 }
121 
122 static int closest_timer(const struct sge *s, int time)
123 {
124 	unsigned int i, match = 0;
125 	int delta, min_delta = INT_MAX;
126 
127 	for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
128 		delta = time - s->timer_val[i];
129 		if (delta < 0)
130 			delta = -delta;
131 		if (delta < min_delta) {
132 			min_delta = delta;
133 			match = i;
134 		}
135 	}
136 	return match;
137 }
138 
139 static int closest_thres(const struct sge *s, int thres)
140 {
141 	unsigned int i, match = 0;
142 	int delta, min_delta = INT_MAX;
143 
144 	for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
145 		delta = thres - s->counter_val[i];
146 		if (delta < 0)
147 			delta = -delta;
148 		if (delta < min_delta) {
149 			min_delta = delta;
150 			match = i;
151 		}
152 	}
153 	return match;
154 }
155 
156 /**
157  * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
158  * @q: the Rx queue
159  * @us: the hold-off time in us, or 0 to disable timer
160  * @cnt: the hold-off packet count, or 0 to disable counter
161  *
162  * Sets an Rx queue's interrupt hold-off time and packet count.  At least
163  * one of the two needs to be enabled for the queue to generate interrupts.
164  */
165 int cxgb4_set_rspq_intr_params(struct sge_rspq *q, unsigned int us,
166 			       unsigned int cnt)
167 {
168 	struct adapter *adap = q->adapter;
169 	unsigned int timer_val;
170 
171 	if (cnt) {
172 		int err;
173 		u32 v, new_idx;
174 
175 		new_idx = closest_thres(&adap->sge, cnt);
176 		if (q->desc && q->pktcnt_idx != new_idx) {
177 			/* the queue has already been created, update it */
178 			v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
179 			    V_FW_PARAMS_PARAM_X(
180 			    FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
181 			    V_FW_PARAMS_PARAM_YZ(q->cntxt_id);
182 			err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
183 					    &v, &new_idx);
184 			if (err)
185 				return err;
186 		}
187 		q->pktcnt_idx = new_idx;
188 	}
189 
190 	timer_val = (us == 0) ? X_TIMERREG_RESTART_COUNTER :
191 				closest_timer(&adap->sge, us);
192 
193 	if ((us | cnt) == 0)
194 		q->intr_params = V_QINTR_TIMER_IDX(X_TIMERREG_UPDATE_CIDX);
195 	else
196 		q->intr_params = V_QINTR_TIMER_IDX(timer_val) |
197 				 V_QINTR_CNT_EN(cnt > 0);
198 	return 0;
199 }
200 
201 static inline bool is_x_1g_port(const struct link_config *lc)
202 {
203 	return (lc->supported & FW_PORT_CAP_SPEED_1G) != 0;
204 }
205 
206 static inline bool is_x_10g_port(const struct link_config *lc)
207 {
208 	unsigned int speeds, high_speeds;
209 
210 	speeds = V_FW_PORT_CAP_SPEED(G_FW_PORT_CAP_SPEED(lc->supported));
211 	high_speeds = speeds & ~(FW_PORT_CAP_SPEED_100M | FW_PORT_CAP_SPEED_1G);
212 
213 	return high_speeds != 0;
214 }
215 
216 inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
217 		      unsigned int us, unsigned int cnt,
218 		      unsigned int size, unsigned int iqe_size)
219 {
220 	q->adapter = adap;
221 	cxgb4_set_rspq_intr_params(q, us, cnt);
222 	q->iqe_len = iqe_size;
223 	q->size = size;
224 }
225 
226 int cfg_queue_count(struct rte_eth_dev *eth_dev)
227 {
228 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
229 	struct adapter *adap = pi->adapter;
230 	struct sge *s = &adap->sge;
231 	unsigned int max_queues = s->max_ethqsets / adap->params.nports;
232 
233 	if ((eth_dev->data->nb_rx_queues < 1) ||
234 	    (eth_dev->data->nb_tx_queues < 1))
235 		return -EINVAL;
236 
237 	if ((eth_dev->data->nb_rx_queues > max_queues) ||
238 	    (eth_dev->data->nb_tx_queues > max_queues))
239 		return -EINVAL;
240 
241 	if (eth_dev->data->nb_rx_queues > pi->rss_size)
242 		return -EINVAL;
243 
244 	/* We must configure RSS, since config has changed*/
245 	pi->flags &= ~PORT_RSS_DONE;
246 
247 	pi->n_rx_qsets = eth_dev->data->nb_rx_queues;
248 	pi->n_tx_qsets = eth_dev->data->nb_tx_queues;
249 
250 	return 0;
251 }
252 
253 void cfg_queues(struct rte_eth_dev *eth_dev)
254 {
255 	struct rte_config *config = rte_eal_get_configuration();
256 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
257 	struct adapter *adap = pi->adapter;
258 	struct sge *s = &adap->sge;
259 	unsigned int i, nb_ports = 0, qidx = 0;
260 	unsigned int q_per_port = 0;
261 
262 	if (!(adap->flags & CFG_QUEUES)) {
263 		for_each_port(adap, i) {
264 			struct port_info *tpi = adap2pinfo(adap, i);
265 
266 			nb_ports += (is_x_10g_port(&tpi->link_cfg)) ||
267 				     is_x_1g_port(&tpi->link_cfg) ? 1 : 0;
268 		}
269 
270 		/*
271 		 * We default up to # of cores queues per 1G/10G port.
272 		 */
273 		if (nb_ports)
274 			q_per_port = (MAX_ETH_QSETS -
275 				     (adap->params.nports - nb_ports)) /
276 				     nb_ports;
277 
278 		if (q_per_port > config->lcore_count)
279 			q_per_port = config->lcore_count;
280 
281 		for_each_port(adap, i) {
282 			struct port_info *pi = adap2pinfo(adap, i);
283 
284 			pi->first_qset = qidx;
285 
286 			/* Initially n_rx_qsets == n_tx_qsets */
287 			pi->n_rx_qsets = (is_x_10g_port(&pi->link_cfg) ||
288 					  is_x_1g_port(&pi->link_cfg)) ?
289 					  q_per_port : 1;
290 			pi->n_tx_qsets = pi->n_rx_qsets;
291 
292 			if (pi->n_rx_qsets > pi->rss_size)
293 				pi->n_rx_qsets = pi->rss_size;
294 
295 			qidx += pi->n_rx_qsets;
296 		}
297 
298 		s->max_ethqsets = qidx;
299 
300 		for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
301 			struct sge_eth_rxq *r = &s->ethrxq[i];
302 
303 			init_rspq(adap, &r->rspq, 5, 32, 1024, 64);
304 			r->usembufs = 1;
305 			r->fl.size = (r->usembufs ? 1024 : 72);
306 		}
307 
308 		for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
309 			s->ethtxq[i].q.size = 1024;
310 
311 		init_rspq(adap, &adap->sge.fw_evtq, 0, 0, 1024, 64);
312 		adap->flags |= CFG_QUEUES;
313 	}
314 }
315 
316 void cxgbe_stats_get(struct port_info *pi, struct port_stats *stats)
317 {
318 	t4_get_port_stats_offset(pi->adapter, pi->tx_chan, stats,
319 				 &pi->stats_base);
320 }
321 
322 void cxgbe_stats_reset(struct port_info *pi)
323 {
324 	t4_clr_port_stats(pi->adapter, pi->tx_chan);
325 }
326 
327 static void setup_memwin(struct adapter *adap)
328 {
329 	u32 mem_win0_base;
330 
331 	/* For T5, only relative offset inside the PCIe BAR is passed */
332 	mem_win0_base = MEMWIN0_BASE;
333 
334 	/*
335 	 * Set up memory window for accessing adapter memory ranges.  (Read
336 	 * back MA register to ensure that changes propagate before we attempt
337 	 * to use the new values.)
338 	 */
339 	t4_write_reg(adap,
340 		     PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
341 					 MEMWIN_NIC),
342 		     mem_win0_base | V_BIR(0) |
343 		     V_WINDOW(ilog2(MEMWIN0_APERTURE) - X_WINDOW_SHIFT));
344 	t4_read_reg(adap,
345 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
346 					MEMWIN_NIC));
347 }
348 
349 static int init_rss(struct adapter *adap)
350 {
351 	unsigned int i;
352 	int err;
353 
354 	err = t4_init_rss_mode(adap, adap->mbox);
355 	if (err)
356 		return err;
357 
358 	for_each_port(adap, i) {
359 		struct port_info *pi = adap2pinfo(adap, i);
360 
361 		pi->rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
362 		if (!pi->rss)
363 			return -ENOMEM;
364 	}
365 	return 0;
366 }
367 
368 /**
369  * Dump basic information about the adapter.
370  */
371 static void print_adapter_info(struct adapter *adap)
372 {
373 	/**
374 	 * Hardware/Firmware/etc. Version/Revision IDs.
375 	 */
376 	t4_dump_version_info(adap);
377 }
378 
379 static void print_port_info(struct adapter *adap)
380 {
381 	int i;
382 	char buf[80];
383 	struct rte_pci_addr *loc = &adap->pdev->addr;
384 
385 	for_each_port(adap, i) {
386 		const struct port_info *pi = &adap->port[i];
387 		char *bufp = buf;
388 
389 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
390 			bufp += sprintf(bufp, "100M/");
391 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
392 			bufp += sprintf(bufp, "1G/");
393 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
394 			bufp += sprintf(bufp, "10G/");
395 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_25G)
396 			bufp += sprintf(bufp, "25G/");
397 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G)
398 			bufp += sprintf(bufp, "40G/");
399 		if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100G)
400 			bufp += sprintf(bufp, "100G/");
401 		if (bufp != buf)
402 			--bufp;
403 		sprintf(bufp, "BASE-%s",
404 			t4_get_port_type_description(
405 					(enum fw_port_type)pi->port_type));
406 
407 		dev_info(adap,
408 			 " " PCI_PRI_FMT " Chelsio rev %d %s %s\n",
409 			 loc->domain, loc->bus, loc->devid, loc->function,
410 			 CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
411 			 (adap->flags & USING_MSIX) ? " MSI-X" :
412 			 (adap->flags & USING_MSI) ? " MSI" : "");
413 	}
414 }
415 
416 static void configure_pcie_ext_tag(struct adapter *adapter)
417 {
418 	u16 v;
419 	int pos = t4_os_find_pci_capability(adapter, PCI_CAP_ID_EXP);
420 
421 	if (!pos)
422 		return;
423 
424 	if (pos > 0) {
425 		t4_os_pci_read_cfg2(adapter, pos + PCI_EXP_DEVCTL, &v);
426 		v |= PCI_EXP_DEVCTL_EXT_TAG;
427 		t4_os_pci_write_cfg2(adapter, pos + PCI_EXP_DEVCTL, v);
428 		if (is_t6(adapter->params.chip)) {
429 			t4_set_reg_field(adapter, A_PCIE_CFG2,
430 					 V_T6_TOTMAXTAG(M_T6_TOTMAXTAG),
431 					 V_T6_TOTMAXTAG(7));
432 			t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
433 					 V_T6_MINTAG(M_T6_MINTAG),
434 					 V_T6_MINTAG(8));
435 		} else {
436 			t4_set_reg_field(adapter, A_PCIE_CFG2,
437 					 V_TOTMAXTAG(M_TOTMAXTAG),
438 					 V_TOTMAXTAG(3));
439 			t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
440 					 V_MINTAG(M_MINTAG),
441 					 V_MINTAG(8));
442 		}
443 	}
444 }
445 
446 /*
447  * Tweak configuration based on system architecture, etc.  Most of these have
448  * defaults assigned to them by Firmware Configuration Files (if we're using
449  * them) but need to be explicitly set if we're using hard-coded
450  * initialization. So these are essentially common tweaks/settings for
451  * Configuration Files and hard-coded initialization ...
452  */
453 static int adap_init0_tweaks(struct adapter *adapter)
454 {
455 	u8 rx_dma_offset;
456 
457 	/*
458 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
459 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
460 	 * 64B Cache Line Size ...
461 	 */
462 	t4_fixup_host_params_compat(adapter, CXGBE_PAGE_SIZE, L1_CACHE_BYTES,
463 				    T5_LAST_REV);
464 
465 	/*
466 	 * Keep the chip default offset to deliver Ingress packets into our
467 	 * DMA buffers to zero
468 	 */
469 	rx_dma_offset = 0;
470 	t4_set_reg_field(adapter, A_SGE_CONTROL, V_PKTSHIFT(M_PKTSHIFT),
471 			 V_PKTSHIFT(rx_dma_offset));
472 
473 	t4_set_reg_field(adapter, A_SGE_FLM_CFG,
474 			 V_CREDITCNT(M_CREDITCNT) | M_CREDITCNTPACKING,
475 			 V_CREDITCNT(3) | V_CREDITCNTPACKING(1));
476 
477 	t4_set_reg_field(adapter, A_SGE_INGRESS_RX_THRESHOLD,
478 			 V_THRESHOLD_3(M_THRESHOLD_3), V_THRESHOLD_3(32U));
479 
480 	t4_set_reg_field(adapter, A_SGE_CONTROL2, V_IDMAARBROUNDROBIN(1U),
481 			 V_IDMAARBROUNDROBIN(1U));
482 
483 	/*
484 	 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
485 	 * adds the pseudo header itself.
486 	 */
487 	t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG,
488 			       F_CSUM_HAS_PSEUDO_HDR, 0);
489 
490 	return 0;
491 }
492 
493 /*
494  * Attempt to initialize the adapter via a Firmware Configuration File.
495  */
496 static int adap_init0_config(struct adapter *adapter, int reset)
497 {
498 	struct fw_caps_config_cmd caps_cmd;
499 	unsigned long mtype = 0, maddr = 0;
500 	u32 finiver, finicsum, cfcsum;
501 	int ret;
502 	int config_issued = 0;
503 	int cfg_addr;
504 	char config_name[20];
505 
506 	/*
507 	 * Reset device if necessary.
508 	 */
509 	if (reset) {
510 		ret = t4_fw_reset(adapter, adapter->mbox,
511 				  F_PIORSTMODE | F_PIORST);
512 		if (ret < 0) {
513 			dev_warn(adapter, "Firmware reset failed, error %d\n",
514 				 -ret);
515 			goto bye;
516 		}
517 	}
518 
519 	cfg_addr = t4_flash_cfg_addr(adapter);
520 	if (cfg_addr < 0) {
521 		ret = cfg_addr;
522 		dev_warn(adapter, "Finding address for firmware config file in flash failed, error %d\n",
523 			 -ret);
524 		goto bye;
525 	}
526 
527 	strcpy(config_name, "On Flash");
528 	mtype = FW_MEMTYPE_CF_FLASH;
529 	maddr = cfg_addr;
530 
531 	/*
532 	 * Issue a Capability Configuration command to the firmware to get it
533 	 * to parse the Configuration File.  We don't use t4_fw_config_file()
534 	 * because we want the ability to modify various features after we've
535 	 * processed the configuration file ...
536 	 */
537 	memset(&caps_cmd, 0, sizeof(caps_cmd));
538 	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
539 					   F_FW_CMD_REQUEST | F_FW_CMD_READ);
540 	caps_cmd.cfvalid_to_len16 =
541 		cpu_to_be32(F_FW_CAPS_CONFIG_CMD_CFVALID |
542 			    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
543 			    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
544 			    FW_LEN16(caps_cmd));
545 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
546 			 &caps_cmd);
547 	/*
548 	 * If the CAPS_CONFIG failed with an ENOENT (for a Firmware
549 	 * Configuration File in FLASH), our last gasp effort is to use the
550 	 * Firmware Configuration File which is embedded in the firmware.  A
551 	 * very few early versions of the firmware didn't have one embedded
552 	 * but we can ignore those.
553 	 */
554 	if (ret == -ENOENT) {
555 		dev_info(adapter, "%s: Going for embedded config in firmware..\n",
556 			 __func__);
557 
558 		memset(&caps_cmd, 0, sizeof(caps_cmd));
559 		caps_cmd.op_to_write =
560 			cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
561 				    F_FW_CMD_REQUEST | F_FW_CMD_READ);
562 		caps_cmd.cfvalid_to_len16 = cpu_to_be32(FW_LEN16(caps_cmd));
563 		ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
564 				 sizeof(caps_cmd), &caps_cmd);
565 		strcpy(config_name, "Firmware Default");
566 	}
567 
568 	config_issued = 1;
569 	if (ret < 0)
570 		goto bye;
571 
572 	finiver = be32_to_cpu(caps_cmd.finiver);
573 	finicsum = be32_to_cpu(caps_cmd.finicsum);
574 	cfcsum = be32_to_cpu(caps_cmd.cfcsum);
575 	if (finicsum != cfcsum)
576 		dev_warn(adapter, "Configuration File checksum mismatch: [fini] csum=%#x, computed csum=%#x\n",
577 			 finicsum, cfcsum);
578 
579 	/*
580 	 * If we're a pure NIC driver then disable all offloading facilities.
581 	 * This will allow the firmware to optimize aspects of the hardware
582 	 * configuration which will result in improved performance.
583 	 */
584 	caps_cmd.niccaps &= cpu_to_be16(~(FW_CAPS_CONFIG_NIC_HASHFILTER |
585 					  FW_CAPS_CONFIG_NIC_ETHOFLD));
586 	caps_cmd.toecaps = 0;
587 	caps_cmd.iscsicaps = 0;
588 	caps_cmd.rdmacaps = 0;
589 	caps_cmd.fcoecaps = 0;
590 
591 	/*
592 	 * And now tell the firmware to use the configuration we just loaded.
593 	 */
594 	caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
595 					   F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
596 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
597 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
598 			 NULL);
599 	if (ret < 0) {
600 		dev_warn(adapter, "Unable to finalize Firmware Capabilities %d\n",
601 			 -ret);
602 		goto bye;
603 	}
604 
605 	/*
606 	 * Tweak configuration based on system architecture, etc.
607 	 */
608 	ret = adap_init0_tweaks(adapter);
609 	if (ret < 0) {
610 		dev_warn(adapter, "Unable to do init0-tweaks %d\n", -ret);
611 		goto bye;
612 	}
613 
614 	/*
615 	 * And finally tell the firmware to initialize itself using the
616 	 * parameters from the Configuration File.
617 	 */
618 	ret = t4_fw_initialize(adapter, adapter->mbox);
619 	if (ret < 0) {
620 		dev_warn(adapter, "Initializing Firmware failed, error %d\n",
621 			 -ret);
622 		goto bye;
623 	}
624 
625 	/*
626 	 * Return successfully and note that we're operating with parameters
627 	 * not supplied by the driver, rather than from hard-wired
628 	 * initialization constants buried in the driver.
629 	 */
630 	dev_info(adapter,
631 		 "Successfully configured using Firmware Configuration File \"%s\", version %#x, computed checksum %#x\n",
632 		 config_name, finiver, cfcsum);
633 
634 	return 0;
635 
636 	/*
637 	 * Something bad happened.  Return the error ...  (If the "error"
638 	 * is that there's no Configuration File on the adapter we don't
639 	 * want to issue a warning since this is fairly common.)
640 	 */
641 bye:
642 	if (config_issued && ret != -ENOENT)
643 		dev_warn(adapter, "\"%s\" configuration file error %d\n",
644 			 config_name, -ret);
645 
646 	dev_debug(adapter, "%s: returning ret = %d ..\n", __func__, ret);
647 	return ret;
648 }
649 
650 static int adap_init0(struct adapter *adap)
651 {
652 	int ret = 0;
653 	u32 v, port_vec;
654 	enum dev_state state;
655 	u32 params[7], val[7];
656 	int reset = 1;
657 	int mbox = adap->mbox;
658 
659 	/*
660 	 * Contact FW, advertising Master capability.
661 	 */
662 	ret = t4_fw_hello(adap, adap->mbox, adap->mbox, MASTER_MAY, &state);
663 	if (ret < 0) {
664 		dev_err(adap, "%s: could not connect to FW, error %d\n",
665 			__func__, -ret);
666 		goto bye;
667 	}
668 
669 	CXGBE_DEBUG_MBOX(adap, "%s: adap->mbox = %d; ret = %d\n", __func__,
670 			 adap->mbox, ret);
671 
672 	if (ret == mbox)
673 		adap->flags |= MASTER_PF;
674 
675 	if (state == DEV_STATE_INIT) {
676 		/*
677 		 * Force halt and reset FW because a previous instance may have
678 		 * exited abnormally without properly shutting down
679 		 */
680 		ret = t4_fw_halt(adap, adap->mbox, reset);
681 		if (ret < 0) {
682 			dev_err(adap, "Failed to halt. Exit.\n");
683 			goto bye;
684 		}
685 
686 		ret = t4_fw_restart(adap, adap->mbox, reset);
687 		if (ret < 0) {
688 			dev_err(adap, "Failed to restart. Exit.\n");
689 			goto bye;
690 		}
691 		state = (enum dev_state)((unsigned)state & ~DEV_STATE_INIT);
692 	}
693 
694 	t4_get_version_info(adap);
695 
696 	ret = t4_get_core_clock(adap, &adap->params.vpd);
697 	if (ret < 0) {
698 		dev_err(adap, "%s: could not get core clock, error %d\n",
699 			__func__, -ret);
700 		goto bye;
701 	}
702 
703 	/*
704 	 * If the firmware is initialized already (and we're not forcing a
705 	 * master initialization), note that we're living with existing
706 	 * adapter parameters.  Otherwise, it's time to try initializing the
707 	 * adapter ...
708 	 */
709 	if (state == DEV_STATE_INIT) {
710 		dev_info(adap, "Coming up as %s: Adapter already initialized\n",
711 			 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
712 	} else {
713 		dev_info(adap, "Coming up as MASTER: Initializing adapter\n");
714 
715 		ret = adap_init0_config(adap, reset);
716 		if (ret == -ENOENT) {
717 			dev_err(adap,
718 				"No Configuration File present on adapter. Using hard-wired configuration parameters.\n");
719 			goto bye;
720 		}
721 	}
722 	if (ret < 0) {
723 		dev_err(adap, "could not initialize adapter, error %d\n", -ret);
724 		goto bye;
725 	}
726 
727 	/* Find out what ports are available to us. */
728 	v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
729 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
730 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
731 	if (ret < 0) {
732 		dev_err(adap, "%s: failure in t4_query_params; error = %d\n",
733 			__func__, ret);
734 		goto bye;
735 	}
736 
737 	adap->params.nports = hweight32(port_vec);
738 	adap->params.portvec = port_vec;
739 
740 	dev_debug(adap, "%s: adap->params.nports = %u\n", __func__,
741 		  adap->params.nports);
742 
743 	/*
744 	 * Give the SGE code a chance to pull in anything that it needs ...
745 	 * Note that this must be called after we retrieve our VPD parameters
746 	 * in order to know how to convert core ticks to seconds, etc.
747 	 */
748 	ret = t4_sge_init(adap);
749 	if (ret < 0) {
750 		dev_err(adap, "t4_sge_init failed with error %d\n",
751 			-ret);
752 		goto bye;
753 	}
754 
755 	/*
756 	 * Grab some of our basic fundamental operating parameters.
757 	 */
758 #define FW_PARAM_DEV(param) \
759 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
760 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
761 
762 #define FW_PARAM_PFVF(param) \
763 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
764 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param) |  \
765 	 V_FW_PARAMS_PARAM_Y(0) | \
766 	 V_FW_PARAMS_PARAM_Z(0))
767 
768 	/* If we're running on newer firmware, let it know that we're
769 	 * prepared to deal with encapsulated CPL messages.  Older
770 	 * firmware won't understand this and we'll just get
771 	 * unencapsulated messages ...
772 	 */
773 	params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
774 	val[0] = 1;
775 	(void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
776 
777 	/*
778 	 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
779 	 * capability.  Earlier versions of the firmware didn't have the
780 	 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
781 	 * permission to use ULPTX MEMWRITE DSGL.
782 	 */
783 	if (is_t4(adap->params.chip)) {
784 		adap->params.ulptx_memwrite_dsgl = false;
785 	} else {
786 		params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
787 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
788 				      1, params, val);
789 		adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
790 	}
791 
792 	/*
793 	 * The MTU/MSS Table is initialized by now, so load their values.  If
794 	 * we're initializing the adapter, then we'll make any modifications
795 	 * we want to the MTU/MSS Table and also initialize the congestion
796 	 * parameters.
797 	 */
798 	t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
799 	if (state != DEV_STATE_INIT) {
800 		int i;
801 
802 		/*
803 		 * The default MTU Table contains values 1492 and 1500.
804 		 * However, for TCP, it's better to have two values which are
805 		 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
806 		 * This allows us to have a TCP Data Payload which is a
807 		 * multiple of 8 regardless of what combination of TCP Options
808 		 * are in use (always a multiple of 4 bytes) which is
809 		 * important for performance reasons.  For instance, if no
810 		 * options are in use, then we have a 20-byte IP header and a
811 		 * 20-byte TCP header.  In this case, a 1500-byte MSS would
812 		 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
813 		 * which is not a multiple of 8.  So using an MSS of 1488 in
814 		 * this case results in a TCP Data Payload of 1448 bytes which
815 		 * is a multiple of 8.  On the other hand, if 12-byte TCP Time
816 		 * Stamps have been negotiated, then an MTU of 1500 bytes
817 		 * results in a TCP Data Payload of 1448 bytes which, as
818 		 * above, is a multiple of 8 bytes ...
819 		 */
820 		for (i = 0; i < NMTUS; i++)
821 			if (adap->params.mtus[i] == 1492) {
822 				adap->params.mtus[i] = 1488;
823 				break;
824 			}
825 
826 		t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
827 			     adap->params.b_wnd);
828 	}
829 	t4_init_sge_params(adap);
830 	t4_init_tp_params(adap);
831 	configure_pcie_ext_tag(adap);
832 
833 	adap->params.drv_memwin = MEMWIN_NIC;
834 	adap->flags |= FW_OK;
835 	dev_debug(adap, "%s: returning zero..\n", __func__);
836 	return 0;
837 
838 	/*
839 	 * Something bad happened.  If a command timed out or failed with EIO
840 	 * FW does not operate within its spec or something catastrophic
841 	 * happened to HW/FW, stop issuing commands.
842 	 */
843 bye:
844 	if (ret != -ETIMEDOUT && ret != -EIO)
845 		t4_fw_bye(adap, adap->mbox);
846 	return ret;
847 }
848 
849 /**
850  * t4_os_portmod_changed - handle port module changes
851  * @adap: the adapter associated with the module change
852  * @port_id: the port index whose module status has changed
853  *
854  * This is the OS-dependent handler for port module changes.  It is
855  * invoked when a port module is removed or inserted for any OS-specific
856  * processing.
857  */
858 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
859 {
860 	static const char * const mod_str[] = {
861 		NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
862 	};
863 
864 	const struct port_info *pi = &adap->port[port_id];
865 
866 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
867 		dev_info(adap, "Port%d: port module unplugged\n", pi->port_id);
868 	else if (pi->mod_type < ARRAY_SIZE(mod_str))
869 		dev_info(adap, "Port%d: %s port module inserted\n", pi->port_id,
870 			 mod_str[pi->mod_type]);
871 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
872 		dev_info(adap, "Port%d: unsupported port module inserted\n",
873 			 pi->port_id);
874 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
875 		dev_info(adap, "Port%d: unknown port module inserted\n",
876 			 pi->port_id);
877 	else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
878 		dev_info(adap, "Port%d: transceiver module error\n",
879 			 pi->port_id);
880 	else
881 		dev_info(adap, "Port%d: unknown module type %d inserted\n",
882 			 pi->port_id, pi->mod_type);
883 }
884 
885 /**
886  * link_start - enable a port
887  * @dev: the port to enable
888  *
889  * Performs the MAC and PHY actions needed to enable a port.
890  */
891 int link_start(struct port_info *pi)
892 {
893 	struct adapter *adapter = pi->adapter;
894 	int ret;
895 	unsigned int mtu;
896 
897 	mtu = pi->eth_dev->data->dev_conf.rxmode.max_rx_pkt_len -
898 	      (ETHER_HDR_LEN + ETHER_CRC_LEN);
899 
900 	/*
901 	 * We do not set address filters and promiscuity here, the stack does
902 	 * that step explicitly.
903 	 */
904 	ret = t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu, -1, -1,
905 			    -1, 1, true);
906 	if (ret == 0) {
907 		ret = t4_change_mac(adapter, adapter->mbox, pi->viid,
908 				    pi->xact_addr_filt,
909 				    (u8 *)&pi->eth_dev->data->mac_addrs[0],
910 				    true, true);
911 		if (ret >= 0) {
912 			pi->xact_addr_filt = ret;
913 			ret = 0;
914 		}
915 	}
916 	if (ret == 0)
917 		ret = t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
918 				    &pi->link_cfg);
919 	if (ret == 0) {
920 		/*
921 		 * Enabling a Virtual Interface can result in an interrupt
922 		 * during the processing of the VI Enable command and, in some
923 		 * paths, result in an attempt to issue another command in the
924 		 * interrupt context.  Thus, we disable interrupts during the
925 		 * course of the VI Enable command ...
926 		 */
927 		ret = t4_enable_vi_params(adapter, adapter->mbox, pi->viid,
928 					  true, true, false);
929 	}
930 	return ret;
931 }
932 
933 /**
934  * cxgb4_write_rss - write the RSS table for a given port
935  * @pi: the port
936  * @queues: array of queue indices for RSS
937  *
938  * Sets up the portion of the HW RSS table for the port's VI to distribute
939  * packets to the Rx queues in @queues.
940  */
941 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
942 {
943 	u16 *rss;
944 	int i, err;
945 	struct adapter *adapter = pi->adapter;
946 	const struct sge_eth_rxq *rxq;
947 
948 	/*  Should never be called before setting up sge eth rx queues */
949 	BUG_ON(!(adapter->flags & FULL_INIT_DONE));
950 
951 	rxq = &adapter->sge.ethrxq[pi->first_qset];
952 	rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
953 	if (!rss)
954 		return -ENOMEM;
955 
956 	/* map the queue indices to queue ids */
957 	for (i = 0; i < pi->rss_size; i++, queues++)
958 		rss[i] = rxq[*queues].rspq.abs_id;
959 
960 	err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
961 				  pi->rss_size, rss, pi->rss_size);
962 	/*
963 	 * If Tunnel All Lookup isn't specified in the global RSS
964 	 * Configuration, then we need to specify a default Ingress
965 	 * Queue for any ingress packets which aren't hashed.  We'll
966 	 * use our first ingress queue ...
967 	 */
968 	if (!err)
969 		err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
970 				       F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
971 				       F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
972 				       F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
973 				       F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN |
974 				       F_FW_RSS_VI_CONFIG_CMD_UDPEN,
975 				       rss[0]);
976 	rte_free(rss);
977 	return err;
978 }
979 
980 /**
981  * setup_rss - configure RSS
982  * @adapter: the adapter
983  *
984  * Sets up RSS to distribute packets to multiple receive queues.  We
985  * configure the RSS CPU lookup table to distribute to the number of HW
986  * receive queues, and the response queue lookup table to narrow that
987  * down to the response queues actually configured for each port.
988  * We always configure the RSS mapping for all ports since the mapping
989  * table has plenty of entries.
990  */
991 int setup_rss(struct port_info *pi)
992 {
993 	int j, err;
994 	struct adapter *adapter = pi->adapter;
995 
996 	dev_debug(adapter, "%s:  pi->rss_size = %u; pi->n_rx_qsets = %u\n",
997 		  __func__, pi->rss_size, pi->n_rx_qsets);
998 
999 	if (!(pi->flags & PORT_RSS_DONE)) {
1000 		if (adapter->flags & FULL_INIT_DONE) {
1001 			/* Fill default values with equal distribution */
1002 			for (j = 0; j < pi->rss_size; j++)
1003 				pi->rss[j] = j % pi->n_rx_qsets;
1004 
1005 			err = cxgb4_write_rss(pi, pi->rss);
1006 			if (err)
1007 				return err;
1008 			pi->flags |= PORT_RSS_DONE;
1009 		}
1010 	}
1011 	return 0;
1012 }
1013 
1014 /*
1015  * Enable NAPI scheduling and interrupt generation for all Rx queues.
1016  */
1017 static void enable_rx(struct adapter *adap, struct sge_rspq *q)
1018 {
1019 	/* 0-increment GTS to start the timer and enable interrupts */
1020 	t4_write_reg(adap, MYPF_REG(A_SGE_PF_GTS),
1021 		     V_SEINTARM(q->intr_params) |
1022 		     V_INGRESSQID(q->cntxt_id));
1023 }
1024 
1025 void cxgbe_enable_rx_queues(struct port_info *pi)
1026 {
1027 	struct adapter *adap = pi->adapter;
1028 	struct sge *s = &adap->sge;
1029 	unsigned int i;
1030 
1031 	for (i = 0; i < pi->n_rx_qsets; i++)
1032 		enable_rx(adap, &s->ethrxq[pi->first_qset + i].rspq);
1033 }
1034 
1035 /**
1036  * fw_caps_to_speed_caps - translate Firmware Port Caps to Speed Caps.
1037  * @port_type: Firmware Port Type
1038  * @fw_caps: Firmware Port Capabilities
1039  * @speed_caps: Device Info Speed Capabilities
1040  *
1041  * Translate a Firmware Port Capabilities specification to Device Info
1042  * Speed Capabilities.
1043  */
1044 static void fw_caps_to_speed_caps(enum fw_port_type port_type,
1045 				  unsigned int fw_caps,
1046 				  u32 *speed_caps)
1047 {
1048 #define SET_SPEED(__speed_name) \
1049 	do { \
1050 		*speed_caps |= ETH_LINK_ ## __speed_name; \
1051 	} while (0)
1052 
1053 #define FW_CAPS_TO_SPEED(__fw_name) \
1054 	do { \
1055 		if (fw_caps & FW_PORT_CAP_ ## __fw_name) \
1056 			SET_SPEED(__fw_name); \
1057 	} while (0)
1058 
1059 	switch (port_type) {
1060 	case FW_PORT_TYPE_BT_SGMII:
1061 	case FW_PORT_TYPE_BT_XFI:
1062 	case FW_PORT_TYPE_BT_XAUI:
1063 		FW_CAPS_TO_SPEED(SPEED_100M);
1064 		FW_CAPS_TO_SPEED(SPEED_1G);
1065 		FW_CAPS_TO_SPEED(SPEED_10G);
1066 		break;
1067 
1068 	case FW_PORT_TYPE_KX4:
1069 	case FW_PORT_TYPE_KX:
1070 	case FW_PORT_TYPE_FIBER_XFI:
1071 	case FW_PORT_TYPE_FIBER_XAUI:
1072 	case FW_PORT_TYPE_SFP:
1073 	case FW_PORT_TYPE_QSFP_10G:
1074 	case FW_PORT_TYPE_QSA:
1075 		FW_CAPS_TO_SPEED(SPEED_1G);
1076 		FW_CAPS_TO_SPEED(SPEED_10G);
1077 		break;
1078 
1079 	case FW_PORT_TYPE_KR:
1080 		SET_SPEED(SPEED_10G);
1081 		break;
1082 
1083 	case FW_PORT_TYPE_BP_AP:
1084 	case FW_PORT_TYPE_BP4_AP:
1085 		SET_SPEED(SPEED_1G);
1086 		SET_SPEED(SPEED_10G);
1087 		break;
1088 
1089 	case FW_PORT_TYPE_BP40_BA:
1090 	case FW_PORT_TYPE_QSFP:
1091 		SET_SPEED(SPEED_40G);
1092 		break;
1093 
1094 	case FW_PORT_TYPE_CR_QSFP:
1095 	case FW_PORT_TYPE_SFP28:
1096 	case FW_PORT_TYPE_KR_SFP28:
1097 		FW_CAPS_TO_SPEED(SPEED_1G);
1098 		FW_CAPS_TO_SPEED(SPEED_10G);
1099 		FW_CAPS_TO_SPEED(SPEED_25G);
1100 		break;
1101 
1102 	case FW_PORT_TYPE_CR2_QSFP:
1103 		SET_SPEED(SPEED_50G);
1104 		break;
1105 
1106 	case FW_PORT_TYPE_KR4_100G:
1107 	case FW_PORT_TYPE_CR4_QSFP:
1108 		FW_CAPS_TO_SPEED(SPEED_25G);
1109 		FW_CAPS_TO_SPEED(SPEED_40G);
1110 		FW_CAPS_TO_SPEED(SPEED_100G);
1111 		break;
1112 
1113 	default:
1114 		break;
1115 	}
1116 
1117 #undef FW_CAPS_TO_SPEED
1118 #undef SET_SPEED
1119 }
1120 
1121 /**
1122  * cxgbe_get_speed_caps - Fetch supported speed capabilities
1123  * @pi: Underlying port's info
1124  * @speed_caps: Device Info speed capabilities
1125  *
1126  * Fetch supported speed capabilities of the underlying port.
1127  */
1128 void cxgbe_get_speed_caps(struct port_info *pi, u32 *speed_caps)
1129 {
1130 	*speed_caps = 0;
1131 
1132 	fw_caps_to_speed_caps(pi->port_type, pi->link_cfg.supported,
1133 			      speed_caps);
1134 
1135 	if (!(pi->link_cfg.supported & FW_PORT_CAP_ANEG))
1136 		*speed_caps |= ETH_LINK_SPEED_FIXED;
1137 }
1138 
1139 /**
1140  * cxgb_up - enable the adapter
1141  * @adap: adapter being enabled
1142  *
1143  * Called when the first port is enabled, this function performs the
1144  * actions necessary to make an adapter operational, such as completing
1145  * the initialization of HW modules, and enabling interrupts.
1146  */
1147 int cxgbe_up(struct adapter *adap)
1148 {
1149 	enable_rx(adap, &adap->sge.fw_evtq);
1150 	t4_sge_tx_monitor_start(adap);
1151 	t4_intr_enable(adap);
1152 	adap->flags |= FULL_INIT_DONE;
1153 
1154 	/* TODO: deadman watchdog ?? */
1155 	return 0;
1156 }
1157 
1158 /*
1159  * Close the port
1160  */
1161 int cxgbe_down(struct port_info *pi)
1162 {
1163 	struct adapter *adapter = pi->adapter;
1164 	int err = 0;
1165 
1166 	err = t4_enable_vi(adapter, adapter->mbox, pi->viid, false, false);
1167 	if (err) {
1168 		dev_err(adapter, "%s: disable_vi failed: %d\n", __func__, err);
1169 		return err;
1170 	}
1171 
1172 	t4_reset_link_config(adapter, pi->port_id);
1173 	return 0;
1174 }
1175 
1176 /*
1177  * Release resources when all the ports have been stopped.
1178  */
1179 void cxgbe_close(struct adapter *adapter)
1180 {
1181 	struct port_info *pi;
1182 	int i;
1183 
1184 	if (adapter->flags & FULL_INIT_DONE) {
1185 		t4_intr_disable(adapter);
1186 		t4_sge_tx_monitor_stop(adapter);
1187 		t4_free_sge_resources(adapter);
1188 		for_each_port(adapter, i) {
1189 			pi = adap2pinfo(adapter, i);
1190 			if (pi->viid != 0)
1191 				t4_free_vi(adapter, adapter->mbox,
1192 					   adapter->pf, 0, pi->viid);
1193 			rte_free(pi->eth_dev->data->mac_addrs);
1194 		}
1195 		adapter->flags &= ~FULL_INIT_DONE;
1196 	}
1197 
1198 	if (adapter->flags & FW_OK)
1199 		t4_fw_bye(adapter, adapter->mbox);
1200 }
1201 
1202 int cxgbe_probe(struct adapter *adapter)
1203 {
1204 	struct port_info *pi;
1205 	int chip;
1206 	int func, i;
1207 	int err = 0;
1208 	u32 whoami;
1209 
1210 	whoami = t4_read_reg(adapter, A_PL_WHOAMI);
1211 	chip = t4_get_chip_type(adapter,
1212 			CHELSIO_PCI_ID_VER(adapter->pdev->id.device_id));
1213 	if (chip < 0)
1214 		return chip;
1215 
1216 	func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
1217 	       G_SOURCEPF(whoami) : G_T6_SOURCEPF(whoami);
1218 
1219 	adapter->mbox = func;
1220 	adapter->pf = func;
1221 
1222 	t4_os_lock_init(&adapter->mbox_lock);
1223 	TAILQ_INIT(&adapter->mbox_list);
1224 
1225 	err = t4_prep_adapter(adapter);
1226 	if (err)
1227 		return err;
1228 
1229 	setup_memwin(adapter);
1230 	err = adap_init0(adapter);
1231 	if (err) {
1232 		dev_err(adapter, "%s: Adapter initialization failed, error %d\n",
1233 			__func__, err);
1234 		goto out_free;
1235 	}
1236 
1237 	if (!is_t4(adapter->params.chip)) {
1238 		/*
1239 		 * The userspace doorbell BAR is split evenly into doorbell
1240 		 * regions, each associated with an egress queue.  If this
1241 		 * per-queue region is large enough (at least UDBS_SEG_SIZE)
1242 		 * then it can be used to submit a tx work request with an
1243 		 * implied doorbell.  Enable write combining on the BAR if
1244 		 * there is room for such work requests.
1245 		 */
1246 		int s_qpp, qpp, num_seg;
1247 
1248 		s_qpp = (S_QUEUESPERPAGEPF0 +
1249 			(S_QUEUESPERPAGEPF1 - S_QUEUESPERPAGEPF0) *
1250 			adapter->pf);
1251 		qpp = 1 << ((t4_read_reg(adapter,
1252 				A_SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp)
1253 				& M_QUEUESPERPAGEPF0);
1254 		num_seg = CXGBE_PAGE_SIZE / UDBS_SEG_SIZE;
1255 		if (qpp > num_seg)
1256 			dev_warn(adapter, "Incorrect SGE EGRESS QUEUES_PER_PAGE configuration, continuing in debug mode\n");
1257 
1258 		adapter->bar2 = (void *)adapter->pdev->mem_resource[2].addr;
1259 		if (!adapter->bar2) {
1260 			dev_err(adapter, "cannot map device bar2 region\n");
1261 			err = -ENOMEM;
1262 			goto out_free;
1263 		}
1264 		t4_write_reg(adapter, A_SGE_STAT_CFG, V_STATSOURCE_T5(7) |
1265 			     V_STATMODE(0));
1266 	}
1267 
1268 	for_each_port(adapter, i) {
1269 		char name[RTE_ETH_NAME_MAX_LEN];
1270 		struct rte_eth_dev_data *data = NULL;
1271 		const unsigned int numa_node = rte_socket_id();
1272 
1273 		pi = &adapter->port[i];
1274 		pi->adapter = adapter;
1275 		pi->xact_addr_filt = -1;
1276 		pi->port_id = i;
1277 
1278 		snprintf(name, sizeof(name), "cxgbe%d",
1279 			 adapter->eth_dev->data->port_id + i);
1280 
1281 		if (i == 0) {
1282 			/* First port is already allocated by DPDK */
1283 			pi->eth_dev = adapter->eth_dev;
1284 			goto allocate_mac;
1285 		}
1286 
1287 		/*
1288 		 * now do all data allocation - for eth_dev structure,
1289 		 * and internal (private) data for the remaining ports
1290 		 */
1291 
1292 		/* reserve an ethdev entry */
1293 		pi->eth_dev = rte_eth_dev_allocate(name);
1294 		if (!pi->eth_dev)
1295 			goto out_free;
1296 
1297 		data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1298 		if (!data)
1299 			goto out_free;
1300 
1301 		data->port_id = adapter->eth_dev->data->port_id + i;
1302 
1303 		pi->eth_dev->data = data;
1304 
1305 allocate_mac:
1306 		pi->eth_dev->device = &adapter->pdev->device;
1307 		pi->eth_dev->data->dev_private = pi;
1308 		pi->eth_dev->dev_ops = adapter->eth_dev->dev_ops;
1309 		pi->eth_dev->tx_pkt_burst = adapter->eth_dev->tx_pkt_burst;
1310 		pi->eth_dev->rx_pkt_burst = adapter->eth_dev->rx_pkt_burst;
1311 
1312 		rte_eth_copy_pci_info(pi->eth_dev, adapter->pdev);
1313 
1314 		pi->eth_dev->data->mac_addrs = rte_zmalloc(name,
1315 							   ETHER_ADDR_LEN, 0);
1316 		if (!pi->eth_dev->data->mac_addrs) {
1317 			dev_err(adapter, "%s: Mem allocation failed for storing mac addr, aborting\n",
1318 				__func__);
1319 			err = -1;
1320 			goto out_free;
1321 		}
1322 	}
1323 
1324 	if (adapter->flags & FW_OK) {
1325 		err = t4_port_init(adapter, adapter->mbox, adapter->pf, 0);
1326 		if (err) {
1327 			dev_err(adapter, "%s: t4_port_init failed with err %d\n",
1328 				__func__, err);
1329 			goto out_free;
1330 		}
1331 	}
1332 
1333 	cfg_queues(adapter->eth_dev);
1334 
1335 	print_adapter_info(adapter);
1336 	print_port_info(adapter);
1337 
1338 	err = init_rss(adapter);
1339 	if (err)
1340 		goto out_free;
1341 
1342 	return 0;
1343 
1344 out_free:
1345 	for_each_port(adapter, i) {
1346 		pi = adap2pinfo(adapter, i);
1347 		if (pi->viid != 0)
1348 			t4_free_vi(adapter, adapter->mbox, adapter->pf,
1349 				   0, pi->viid);
1350 		/* Skip first port since it'll be de-allocated by DPDK */
1351 		if (i == 0)
1352 			continue;
1353 		if (pi->eth_dev->data)
1354 			rte_free(pi->eth_dev->data);
1355 	}
1356 
1357 	if (adapter->flags & FW_OK)
1358 		t4_fw_bye(adapter, adapter->mbox);
1359 	return -err;
1360 }
1361