xref: /illumos-gate/usr/src/uts/common/io/cxgbe/t4nex/t4_nexus.c (revision 8a2b682e57a046b828f37bcde1776f131ef4629f)
1 /*
2  * This file and its contents are supplied under the terms of the
3  * Common Development and Distribution License ("CDDL"), version 1.0.
4  * You may only use this file in accordance with the terms of version
5  * 1.0 of the CDDL.
6  *
7  * A full copy of the text of the CDDL should have accompanied this
8  * source. A copy of the CDDL is also available via the Internet at
9  * http://www.illumos.org/license/CDDL.
10  */
11 
12 /*
13  * This file is part of the Chelsio T4 support code.
14  *
15  * Copyright (C) 2010-2013 Chelsio Communications.  All rights reserved.
16  *
17  * This program is distributed in the hope that it will be useful, but WITHOUT
18  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
19  * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
20  * release for licensing terms and conditions.
21  */
22 
23 #include <sys/ddi.h>
24 #include <sys/sunddi.h>
25 #include <sys/sunndi.h>
26 #include <sys/modctl.h>
27 #include <sys/conf.h>
28 #include <sys/devops.h>
29 #include <sys/pci.h>
30 #include <sys/atomic.h>
31 #include <sys/types.h>
32 #include <sys/file.h>
33 #include <sys/errno.h>
34 #include <sys/open.h>
35 #include <sys/cred.h>
36 #include <sys/stat.h>
37 #include <sys/mkdev.h>
38 #include <sys/queue.h>
39 #include <sys/containerof.h>
40 
41 #include "version.h"
42 #include "common/common.h"
43 #include "common/t4_msg.h"
44 #include "common/t4_regs.h"
45 #include "firmware/t4_fw.h"
46 #include "firmware/t4_cfg.h"
47 #include "firmware/t5_fw.h"
48 #include "firmware/t5_cfg.h"
49 #include "firmware/t6_fw.h"
50 #include "firmware/t6_cfg.h"
51 #include "t4_l2t.h"
52 
53 static int t4_cb_open(dev_t *devp, int flag, int otyp, cred_t *credp);
54 static int t4_cb_close(dev_t dev, int flag, int otyp, cred_t *credp);
55 static int t4_cb_ioctl(dev_t dev, int cmd, intptr_t d, int mode, cred_t *credp,
56     int *rp);
57 struct cb_ops t4_cb_ops = {
58 	.cb_open =		t4_cb_open,
59 	.cb_close =		t4_cb_close,
60 	.cb_strategy =		nodev,
61 	.cb_print =		nodev,
62 	.cb_dump =		nodev,
63 	.cb_read =		nodev,
64 	.cb_write =		nodev,
65 	.cb_ioctl =		t4_cb_ioctl,
66 	.cb_devmap =		nodev,
67 	.cb_mmap =		nodev,
68 	.cb_segmap =		nodev,
69 	.cb_chpoll =		nochpoll,
70 	.cb_prop_op =		ddi_prop_op,
71 	.cb_flag =		D_MP,
72 	.cb_rev =		CB_REV,
73 	.cb_aread =		nodev,
74 	.cb_awrite =		nodev
75 };
76 
77 static int t4_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t op,
78     void *arg, void *result);
79 static int t4_bus_config(dev_info_t *dip, uint_t flags, ddi_bus_config_op_t op,
80     void *arg, dev_info_t **cdipp);
81 static int t4_bus_unconfig(dev_info_t *dip, uint_t flags,
82     ddi_bus_config_op_t op, void *arg);
83 struct bus_ops t4_bus_ops = {
84 	.busops_rev =		BUSO_REV,
85 	.bus_ctl =		t4_bus_ctl,
86 	.bus_prop_op =		ddi_bus_prop_op,
87 	.bus_config =		t4_bus_config,
88 	.bus_unconfig =		t4_bus_unconfig,
89 };
90 
91 static int t4_devo_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg,
92     void **rp);
93 static int t4_devo_probe(dev_info_t *dip);
94 static int t4_devo_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
95 static int t4_devo_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
96 static int t4_devo_quiesce(dev_info_t *dip);
97 struct dev_ops t4_dev_ops = {
98 	.devo_rev =		DEVO_REV,
99 	.devo_getinfo =		t4_devo_getinfo,
100 	.devo_identify =	nulldev,
101 	.devo_probe =		t4_devo_probe,
102 	.devo_attach =		t4_devo_attach,
103 	.devo_detach =		t4_devo_detach,
104 	.devo_reset =		nodev,
105 	.devo_cb_ops =		&t4_cb_ops,
106 	.devo_bus_ops =		&t4_bus_ops,
107 	.devo_quiesce =		&t4_devo_quiesce,
108 };
109 
110 static struct modldrv modldrv = {
111 	.drv_modops =		&mod_driverops,
112 	.drv_linkinfo =		"Chelsio T4 nexus " DRV_VERSION,
113 	.drv_dev_ops =		&t4_dev_ops
114 };
115 
116 static struct modlinkage modlinkage = {
117 	.ml_rev =		MODREV_1,
118 	.ml_linkage =		{&modldrv, NULL},
119 };
120 
121 void *t4_list;
122 
123 struct intrs_and_queues {
124 	int intr_type;		/* DDI_INTR_TYPE_* */
125 	int nirq;		/* Number of vectors */
126 	int intr_fwd;		/* Interrupts forwarded */
127 	int ntxq10g;		/* # of NIC txq's for each 10G port */
128 	int nrxq10g;		/* # of NIC rxq's for each 10G port */
129 	int ntxq1g;		/* # of NIC txq's for each 1G port */
130 	int nrxq1g;		/* # of NIC rxq's for each 1G port */
131 #ifdef TCP_OFFLOAD_ENABLE
132 	int nofldtxq10g;	/* # of TOE txq's for each 10G port */
133 	int nofldrxq10g;	/* # of TOE rxq's for each 10G port */
134 	int nofldtxq1g;		/* # of TOE txq's for each 1G port */
135 	int nofldrxq1g;		/* # of TOE rxq's for each 1G port */
136 #endif
137 };
138 
139 struct fw_info fi[3];
140 
141 static int cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss,
142     mblk_t *m);
143 static int fw_msg_not_handled(struct adapter *, const __be64 *);
144 int t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h);
145 static unsigned int getpf(struct adapter *sc);
146 static int prep_firmware(struct adapter *sc);
147 static int upload_config_file(struct adapter *sc, uint32_t *mt, uint32_t *ma);
148 static int partition_resources(struct adapter *sc);
149 static int adap__pre_init_tweaks(struct adapter *sc);
150 static int get_params__pre_init(struct adapter *sc);
151 static int get_params__post_init(struct adapter *sc);
152 static int set_params__post_init(struct adapter *);
153 static void setup_memwin(struct adapter *sc);
154 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
155     uint32_t *);
156 void memwin_info(struct adapter *, int, uint32_t *, uint32_t *);
157 uint32_t position_memwin(struct adapter *, int, uint32_t);
158 static int prop_lookup_int_array(struct adapter *sc, char *name, int *data,
159     uint_t count);
160 static int prop_lookup_int_array(struct adapter *sc, char *name, int *data,
161     uint_t count);
162 static int init_driver_props(struct adapter *sc, struct driver_properties *p);
163 static int remove_extra_props(struct adapter *sc, int n10g, int n1g);
164 static int cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g,
165     struct intrs_and_queues *iaq);
166 static int add_child_node(struct adapter *sc, int idx);
167 static int remove_child_node(struct adapter *sc, int idx);
168 static kstat_t *setup_kstats(struct adapter *sc);
169 static kstat_t *setup_wc_kstats(struct adapter *);
170 static int update_wc_kstats(kstat_t *, int);
171 #ifdef TCP_OFFLOAD_ENABLE
172 static int toe_capability(struct port_info *pi, int enable);
173 static int activate_uld(struct adapter *sc, int id, struct uld_softc *usc);
174 static int deactivate_uld(struct uld_softc *usc);
175 #endif
176 static kmutex_t t4_adapter_list_lock;
177 static SLIST_HEAD(, adapter) t4_adapter_list;
178 #ifdef TCP_OFFLOAD_ENABLE
179 static kmutex_t t4_uld_list_lock;
180 static SLIST_HEAD(, uld_info) t4_uld_list;
181 #endif
182 
183 int
184 _init(void)
185 {
186 	int rc;
187 
188 	rc = ddi_soft_state_init(&t4_list, sizeof (struct adapter), 0);
189 	if (rc != 0)
190 		return (rc);
191 
192 	rc = mod_install(&modlinkage);
193 	if (rc != 0)
194 		ddi_soft_state_fini(&t4_list);
195 
196 	mutex_init(&t4_adapter_list_lock, NULL, MUTEX_DRIVER, NULL);
197 	SLIST_INIT(&t4_adapter_list);
198 
199 #ifdef TCP_OFFLOAD_ENABLE
200 	mutex_init(&t4_uld_list_lock, NULL, MUTEX_DRIVER, NULL);
201 	SLIST_INIT(&t4_uld_list);
202 #endif
203 
204 	return (rc);
205 }
206 
207 int
208 _fini(void)
209 {
210 	int rc;
211 
212 	rc = mod_remove(&modlinkage);
213 	if (rc != 0)
214 		return (rc);
215 
216 	ddi_soft_state_fini(&t4_list);
217 	return (0);
218 }
219 
220 int
221 _info(struct modinfo *mi)
222 {
223 	return (mod_info(&modlinkage, mi));
224 }
225 
226 /* ARGSUSED */
227 static int
228 t4_devo_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **rp)
229 {
230 	struct adapter *sc;
231 	minor_t minor;
232 
233 	minor = getminor((dev_t)arg);	/* same as instance# in our case */
234 
235 	if (cmd == DDI_INFO_DEVT2DEVINFO) {
236 		sc = ddi_get_soft_state(t4_list, minor);
237 		if (sc == NULL)
238 			return (DDI_FAILURE);
239 
240 		ASSERT(sc->dev == (dev_t)arg);
241 		*rp = (void *)sc->dip;
242 	} else if (cmd == DDI_INFO_DEVT2INSTANCE)
243 		*rp = (void *) (unsigned long) minor;
244 	else
245 		ASSERT(0);
246 
247 	return (DDI_SUCCESS);
248 }
249 
250 static int
251 t4_devo_probe(dev_info_t *dip)
252 {
253 	int rc, id, *reg;
254 	uint_t n, pf;
255 
256 	id = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
257 	    "device-id", 0xffff);
258 	if (id == 0xffff)
259 		return (DDI_PROBE_DONTCARE);
260 
261 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
262 	    "reg", &reg, &n);
263 	if (rc != DDI_SUCCESS)
264 		return (DDI_PROBE_DONTCARE);
265 
266 	pf = PCI_REG_FUNC_G(reg[0]);
267 	ddi_prop_free(reg);
268 
269 	/* Prevent driver attachment on any PF except 0 on the FPGA */
270 	if (id == 0xa000 && pf != 0)
271 		return (DDI_PROBE_FAILURE);
272 
273 	return (DDI_PROBE_DONTCARE);
274 }
275 
276 static int
277 t4_devo_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
278 {
279 	struct adapter *sc = NULL;
280 	struct sge *s;
281 	int i, instance, rc = DDI_SUCCESS, rqidx, tqidx, q;
282 	int irq = 0, nxg, n100g, n40g, n25g, n10g, n1g;
283 #ifdef TCP_OFFLOAD_ENABLE
284 	int ofld_rqidx, ofld_tqidx;
285 #endif
286 	char name[16];
287 	struct driver_properties *prp;
288 	struct intrs_and_queues iaq;
289 	ddi_device_acc_attr_t da = {
290 		.devacc_attr_version = DDI_DEVICE_ATTR_V0,
291 		.devacc_attr_endian_flags = DDI_STRUCTURE_LE_ACC,
292 		.devacc_attr_dataorder = DDI_UNORDERED_OK_ACC
293 	};
294 	ddi_device_acc_attr_t da1 = {
295 		.devacc_attr_version = DDI_DEVICE_ATTR_V0,
296 		.devacc_attr_endian_flags = DDI_STRUCTURE_LE_ACC,
297 		.devacc_attr_dataorder = DDI_MERGING_OK_ACC
298 	};
299 
300 	if (cmd != DDI_ATTACH)
301 		return (DDI_FAILURE);
302 
303 	/*
304 	 * Allocate space for soft state.
305 	 */
306 	instance = ddi_get_instance(dip);
307 	rc = ddi_soft_state_zalloc(t4_list, instance);
308 	if (rc != DDI_SUCCESS) {
309 		cxgb_printf(dip, CE_WARN,
310 		    "failed to allocate soft state: %d", rc);
311 		return (DDI_FAILURE);
312 	}
313 
314 	sc = ddi_get_soft_state(t4_list, instance);
315 	sc->dip = dip;
316 	sc->dev = makedevice(ddi_driver_major(dip), instance);
317 	mutex_init(&sc->lock, NULL, MUTEX_DRIVER, NULL);
318 	cv_init(&sc->cv, NULL, CV_DRIVER, NULL);
319 	mutex_init(&sc->sfl_lock, NULL, MUTEX_DRIVER, NULL);
320 	TAILQ_INIT(&sc->sfl);
321 
322 	mutex_enter(&t4_adapter_list_lock);
323 	SLIST_INSERT_HEAD(&t4_adapter_list, sc, link);
324 	mutex_exit(&t4_adapter_list_lock);
325 
326 	sc->pf = getpf(sc);
327 	if (sc->pf > 8) {
328 		rc = EINVAL;
329 		cxgb_printf(dip, CE_WARN,
330 		    "failed to determine PCI PF# of device");
331 		goto done;
332 	}
333 	sc->mbox = sc->pf;
334 
335 	/* Initialize the driver properties */
336 	prp = &sc->props;
337 	(void)init_driver_props(sc, prp);
338 
339 	/*
340 	 * Enable access to the PCI config space.
341 	 */
342 	rc = pci_config_setup(dip, &sc->pci_regh);
343 	if (rc != DDI_SUCCESS) {
344 		cxgb_printf(dip, CE_WARN,
345 		    "failed to enable PCI config space access: %d", rc);
346 		goto done;
347 	}
348 
349 	/* TODO: Set max read request to 4K */
350 
351 	/*
352 	 * Enable MMIO access.
353 	 */
354 	rc = ddi_regs_map_setup(dip, 1, &sc->regp, 0, 0, &da, &sc->regh);
355 	if (rc != DDI_SUCCESS) {
356 		cxgb_printf(dip, CE_WARN,
357 		    "failed to map device registers: %d", rc);
358 		goto done;
359 	}
360 
361 	(void) memset(sc->chan_map, 0xff, sizeof (sc->chan_map));
362 
363 	/*
364 	 * Initialize cpl handler.
365 	 */
366 	for (i = 0; i < ARRAY_SIZE(sc->cpl_handler); i++) {
367 		sc->cpl_handler[i] = cpl_not_handled;
368 	}
369 
370 	for (i = 0; i < ARRAY_SIZE(sc->fw_msg_handler); i++) {
371 		sc->fw_msg_handler[i] = fw_msg_not_handled;
372 	}
373 
374 	for (i = 0; i < NCHAN; i++) {
375 		(void) snprintf(name, sizeof (name), "%s-%d",
376 				"reclaim", i);
377 		sc->tq[i] = ddi_taskq_create(sc->dip,
378 		   name, 1, TASKQ_DEFAULTPRI, 0);
379 
380 		if (sc->tq[i] == NULL) {
381 			cxgb_printf(dip, CE_WARN,
382 				   "failed to create task queues");
383 			rc = DDI_FAILURE;
384 			goto done;
385 		}
386 	}
387 
388 	/*
389 	 * Prepare the adapter for operation.
390 	 */
391 	rc = -t4_prep_adapter(sc, false);
392 	if (rc != 0) {
393 		cxgb_printf(dip, CE_WARN, "failed to prepare adapter: %d", rc);
394 		goto done;
395 	}
396 
397 	/*
398 	 * Enable BAR1 access.
399 	 */
400 	sc->doorbells |= DOORBELL_KDB;
401 	rc = ddi_regs_map_setup(dip, 2, &sc->reg1p, 0, 0, &da1, &sc->reg1h);
402 	if (rc != DDI_SUCCESS) {
403 		cxgb_printf(dip, CE_WARN,
404 		    "failed to map BAR1 device registers: %d", rc);
405 		goto done;
406 	} else {
407 		if (is_t5(sc->params.chip)) {
408 			sc->doorbells |= DOORBELL_UDB;
409 			if (prp->wc) {
410 				/*
411 				 * Enable write combining on BAR2.  This is the
412 				 * userspace doorbell BAR and is split into 128B
413 				 * (UDBS_SEG_SIZE) doorbell regions, each associated
414 				 * with an egress queue.  The first 64B has the doorbell
415 				 * and the second 64B can be used to submit a tx work
416 				 * request with an implicit doorbell.
417 				 */
418 				sc->doorbells &= ~DOORBELL_UDB;
419 				sc->doorbells |= (DOORBELL_WCWR |
420 				    DOORBELL_UDBWC);
421 				t4_write_reg(sc, A_SGE_STAT_CFG,
422 				    V_STATSOURCE_T5(7) | V_STATMODE(0));
423 			}
424 		}
425 	}
426 
427 	/*
428 	 * Do this really early.  Note that minor number = instance.
429 	 */
430 	(void) snprintf(name, sizeof (name), "%s,%d", T4_NEXUS_NAME, instance);
431 	rc = ddi_create_minor_node(dip, name, S_IFCHR, instance,
432 	    DDI_NT_NEXUS, 0);
433 	if (rc != DDI_SUCCESS) {
434 		cxgb_printf(dip, CE_WARN,
435 		    "failed to create device node: %d", rc);
436 		rc = DDI_SUCCESS; /* carry on */
437 	}
438 
439 	/* Do this early. Memory window is required for loading config file. */
440 	setup_memwin(sc);
441 
442 	/* Prepare the firmware for operation */
443 	rc = prep_firmware(sc);
444 	if (rc != 0)
445 		goto done; /* error message displayed already */
446 
447 	rc = adap__pre_init_tweaks(sc);
448 	if (rc != 0)
449 		goto done;
450 
451 	rc = get_params__pre_init(sc);
452 	if (rc != 0)
453 		goto done; /* error message displayed already */
454 
455 	t4_sge_init(sc);
456 
457 	if (sc->flags & MASTER_PF) {
458 		/* get basic stuff going */
459 		rc = -t4_fw_initialize(sc, sc->mbox);
460 		if (rc != 0) {
461 			cxgb_printf(sc->dip, CE_WARN,
462 			    "early init failed: %d.\n", rc);
463 			goto done;
464 		}
465 	}
466 
467 	rc = get_params__post_init(sc);
468 	if (rc != 0)
469 		goto done; /* error message displayed already */
470 
471 	rc = set_params__post_init(sc);
472 	if (rc != 0)
473 		goto done; /* error message displayed already */
474 
475 	/*
476 	 * TODO: This is the place to call t4_set_filter_mode()
477 	 */
478 
479 	/* tweak some settings */
480 	t4_write_reg(sc, A_TP_SHIFT_CNT, V_SYNSHIFTMAX(6) | V_RXTSHIFTMAXR1(4) |
481 	    V_RXTSHIFTMAXR2(15) | V_PERSHIFTBACKOFFMAX(8) | V_PERSHIFTMAX(8) |
482 	    V_KEEPALIVEMAXR1(4) | V_KEEPALIVEMAXR2(9));
483 	t4_write_reg(sc, A_ULP_RX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
484 
485 	/*
486 	 * Work-around for bug 2619
487 	 * Set DisableVlan field in TP_RSS_CONFIG_VRT register so that the
488 	 * VLAN tag extraction is disabled.
489 	 */
490 	t4_set_reg_field(sc, A_TP_RSS_CONFIG_VRT, F_DISABLEVLAN, F_DISABLEVLAN);
491 
492 	/* Store filter mode */
493 	t4_read_indirect(sc, A_TP_PIO_ADDR, A_TP_PIO_DATA, &sc->filter_mode, 1,
494 	    A_TP_VLAN_PRI_MAP);
495 
496 	/*
497 	 * First pass over all the ports - allocate VIs and initialize some
498 	 * basic parameters like mac address, port type, etc.  We also figure
499 	 * out whether a port is 10G or 1G and use that information when
500 	 * calculating how many interrupts to attempt to allocate.
501 	 */
502 	n100g = n40g = n25g = n10g = n1g = 0;
503 	for_each_port(sc, i) {
504 		struct port_info *pi;
505 
506 		pi = kmem_zalloc(sizeof (*pi), KM_SLEEP);
507 		sc->port[i] = pi;
508 
509 		/* These must be set before t4_port_init */
510 		pi->adapter = sc;
511 		/* LINTED: E_ASSIGN_NARROW_CONV */
512 		pi->port_id = i;
513 	}
514 
515 	/* Allocate the vi and initialize parameters like mac addr */
516 	rc = -t4_port_init(sc, sc->mbox, sc->pf, 0);
517 	if (rc) {
518 		cxgb_printf(dip, CE_WARN,
519 			    "unable to initialize port: %d", rc);
520 		goto done;
521 	}
522 
523 	for_each_port(sc, i) {
524 		struct port_info *pi = sc->port[i];
525 
526 		mutex_init(&pi->lock, NULL, MUTEX_DRIVER, NULL);
527 		pi->mtu = ETHERMTU;
528 
529 		if (is_100G_port(pi)) {
530 			n100g++;
531 			pi->tmr_idx = prp->tmr_idx_10g;
532 			pi->pktc_idx = prp->pktc_idx_10g;
533 		} else if (is_40G_port(pi)) {
534 			n40g++;
535 			pi->tmr_idx = prp->tmr_idx_10g;
536 			pi->pktc_idx = prp->pktc_idx_10g;
537 		} else if (is_25G_port(pi)) {
538 			n25g++;
539 			pi->tmr_idx = prp->tmr_idx_10g;
540 			pi->pktc_idx = prp->pktc_idx_10g;
541 		} else if (is_10G_port(pi)) {
542 			n10g++;
543 			pi->tmr_idx = prp->tmr_idx_10g;
544 			pi->pktc_idx = prp->pktc_idx_10g;
545 		} else {
546 			n1g++;
547 			pi->tmr_idx = prp->tmr_idx_1g;
548 			pi->pktc_idx = prp->pktc_idx_1g;
549 		}
550 
551 		pi->xact_addr_filt = -1;
552 		t4_mc_init(pi);
553 
554 		setbit(&sc->registered_device_map, i);
555 	}
556 
557 	nxg = n10g + n25g + n40g + n100g;
558 	(void) remove_extra_props(sc, nxg, n1g);
559 
560 	if (sc->registered_device_map == 0) {
561 		cxgb_printf(dip, CE_WARN, "no usable ports");
562 		rc = DDI_FAILURE;
563 		goto done;
564 	}
565 
566 	rc = cfg_itype_and_nqueues(sc, nxg, n1g, &iaq);
567 	if (rc != 0)
568 		goto done; /* error message displayed already */
569 
570 	sc->intr_type = iaq.intr_type;
571 	sc->intr_count = iaq.nirq;
572 
573 	if (sc->props.multi_rings && (sc->intr_type != DDI_INTR_TYPE_MSIX)) {
574 		sc->props.multi_rings = 0;
575 		cxgb_printf(dip, CE_WARN,
576 		    "Multiple rings disabled as interrupt type is not MSI-X");
577 	}
578 
579 	if (sc->props.multi_rings && iaq.intr_fwd) {
580 		sc->props.multi_rings = 0;
581 		cxgb_printf(dip, CE_WARN,
582 		    "Multiple rings disabled as interrupts are forwarded");
583 	}
584 
585 	if (!sc->props.multi_rings) {
586 		iaq.ntxq10g = 1;
587 		iaq.ntxq1g = 1;
588 	}
589 	s = &sc->sge;
590 	s->nrxq = nxg * iaq.nrxq10g + n1g * iaq.nrxq1g;
591 	s->ntxq = nxg * iaq.ntxq10g + n1g * iaq.ntxq1g;
592 	s->neq = s->ntxq + s->nrxq;	/* the fl in an rxq is an eq */
593 #ifdef TCP_OFFLOAD_ENABLE
594 	/* control queues, 1 per port + 1 mgmtq */
595 	s->neq += sc->params.nports + 1;
596 #endif
597 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
598 	if (iaq.intr_fwd != 0)
599 		sc->flags |= INTR_FWD;
600 #ifdef TCP_OFFLOAD_ENABLE
601 	if (is_offload(sc) != 0) {
602 
603 		s->nofldrxq = nxg * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
604 		s->nofldtxq = nxg * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
605 		s->neq += s->nofldtxq + s->nofldrxq;
606 		s->niq += s->nofldrxq;
607 
608 		s->ofld_rxq = kmem_zalloc(s->nofldrxq *
609 		    sizeof (struct sge_ofld_rxq), KM_SLEEP);
610 		s->ofld_txq = kmem_zalloc(s->nofldtxq *
611 		    sizeof (struct sge_wrq), KM_SLEEP);
612 		s->ctrlq = kmem_zalloc(sc->params.nports *
613 		    sizeof (struct sge_wrq), KM_SLEEP);
614 
615 	}
616 #endif
617 	s->rxq = kmem_zalloc(s->nrxq * sizeof (struct sge_rxq), KM_SLEEP);
618 	s->txq = kmem_zalloc(s->ntxq * sizeof (struct sge_txq), KM_SLEEP);
619 	s->iqmap = kmem_zalloc(s->niq * sizeof (struct sge_iq *), KM_SLEEP);
620 	s->eqmap = kmem_zalloc(s->neq * sizeof (struct sge_eq *), KM_SLEEP);
621 
622 	sc->intr_handle = kmem_zalloc(sc->intr_count *
623 	    sizeof (ddi_intr_handle_t), KM_SLEEP);
624 
625 	/*
626 	 * Second pass over the ports.  This time we know the number of rx and
627 	 * tx queues that each port should get.
628 	 */
629 	rqidx = tqidx = 0;
630 #ifdef TCP_OFFLOAD_ENABLE
631 	ofld_rqidx = ofld_tqidx = 0;
632 #endif
633 	for_each_port(sc, i) {
634 		struct port_info *pi = sc->port[i];
635 
636 		if (pi == NULL)
637 			continue;
638 
639 		t4_mc_cb_init(pi);
640 		/* LINTED: E_ASSIGN_NARROW_CONV */
641 		pi->first_rxq = rqidx;
642 		/* LINTED: E_ASSIGN_NARROW_CONV */
643 		pi->nrxq = (is_10XG_port(pi)) ? iaq.nrxq10g
644 		    : iaq.nrxq1g;
645 		/* LINTED: E_ASSIGN_NARROW_CONV */
646 		pi->first_txq = tqidx;
647 		/* LINTED: E_ASSIGN_NARROW_CONV */
648 		pi->ntxq = (is_10XG_port(pi)) ? iaq.ntxq10g
649 		    : iaq.ntxq1g;
650 
651 		rqidx += pi->nrxq;
652 		tqidx += pi->ntxq;
653 
654 #ifdef TCP_OFFLOAD_ENABLE
655 		if (is_offload(sc) != 0) {
656 			/* LINTED: E_ASSIGN_NARROW_CONV */
657 			pi->first_ofld_rxq = ofld_rqidx;
658 			pi->nofldrxq = max(1, pi->nrxq / 4);
659 
660 			/* LINTED: E_ASSIGN_NARROW_CONV */
661 			pi->first_ofld_txq = ofld_tqidx;
662 			pi->nofldtxq = max(1, pi->ntxq / 2);
663 
664 			ofld_rqidx += pi->nofldrxq;
665 			ofld_tqidx += pi->nofldtxq;
666 		}
667 #endif
668 
669 		/*
670 		 * Enable hw checksumming and LSO for all ports by default.
671 		 * They can be disabled using ndd (hw_csum and hw_lso).
672 		 */
673 		pi->features |= (CXGBE_HW_CSUM | CXGBE_HW_LSO);
674 	}
675 
676 #ifdef TCP_OFFLOAD_ENABLE
677 		sc->l2t = t4_init_l2t(sc);
678 #endif
679 
680 	/*
681 	 * Setup Interrupts.
682 	 */
683 
684 	i = 0;
685 	rc = ddi_intr_alloc(dip, sc->intr_handle, sc->intr_type, 0,
686 	    sc->intr_count, &i, DDI_INTR_ALLOC_STRICT);
687 	if (rc != DDI_SUCCESS) {
688 		cxgb_printf(dip, CE_WARN,
689 		    "failed to allocate %d interrupt(s) of type %d: %d, %d",
690 		    sc->intr_count, sc->intr_type, rc, i);
691 		goto done;
692 	}
693 	ASSERT(sc->intr_count == i); /* allocation was STRICT */
694 	(void) ddi_intr_get_cap(sc->intr_handle[0], &sc->intr_cap);
695 	(void) ddi_intr_get_pri(sc->intr_handle[0], &sc->intr_pri);
696 	if (sc->intr_count == 1) {
697 		ASSERT(sc->flags & INTR_FWD);
698 		(void) ddi_intr_add_handler(sc->intr_handle[0], t4_intr_all, sc,
699 		    &s->fwq);
700 	} else {
701 		/* Multiple interrupts.  The first one is always error intr */
702 		(void) ddi_intr_add_handler(sc->intr_handle[0], t4_intr_err, sc,
703 		    NULL);
704 		irq++;
705 
706 		/* The second one is always the firmware event queue */
707 		(void) ddi_intr_add_handler(sc->intr_handle[1], t4_intr, sc,
708 		    &s->fwq);
709 		irq++;
710 		/*
711 		 * Note that if INTR_FWD is set then either the NIC rx
712 		 * queues or (exclusive or) the TOE rx queueus will be taking
713 		 * direct interrupts.
714 		 *
715 		 * There is no need to check for is_offload(sc) as nofldrxq
716 		 * will be 0 if offload is disabled.
717 		 */
718 		for_each_port(sc, i) {
719 			struct port_info *pi = sc->port[i];
720 			struct sge_rxq *rxq;
721 #ifdef TCP_OFFLOAD_ENABLE
722 			struct sge_ofld_rxq *ofld_rxq;
723 
724 			/*
725 			 * Skip over the NIC queues if they aren't taking direct
726 			 * interrupts.
727 			 */
728 			if ((sc->flags & INTR_FWD) &&
729 			    pi->nofldrxq > pi->nrxq)
730 				goto ofld_queues;
731 #endif
732 			rxq = &s->rxq[pi->first_rxq];
733 			for (q = 0; q < pi->nrxq; q++, rxq++) {
734 				(void) ddi_intr_add_handler(
735 				    sc->intr_handle[irq], t4_intr, sc,
736 				    &rxq->iq);
737 				irq++;
738 			}
739 
740 #ifdef TCP_OFFLOAD_ENABLE
741 			/*
742 			 * Skip over the offload queues if they aren't taking
743 			 * direct interrupts.
744 			 */
745 			if ((sc->flags & INTR_FWD))
746 				continue;
747 ofld_queues:
748 			ofld_rxq = &s->ofld_rxq[pi->first_ofld_rxq];
749 			for (q = 0; q < pi->nofldrxq; q++, ofld_rxq++) {
750 				(void) ddi_intr_add_handler(
751 				    sc->intr_handle[irq], t4_intr, sc,
752 				    &ofld_rxq->iq);
753 				irq++;
754 			}
755 #endif
756 		}
757 
758 	}
759 	sc->flags |= INTR_ALLOCATED;
760 
761 	ASSERT(rc == DDI_SUCCESS);
762 	ddi_report_dev(dip);
763 
764 	/*
765 	 * Hardware/Firmware/etc. Version/Revision IDs.
766 	 */
767 	t4_dump_version_info(sc);
768 
769 	if (n100g) {
770 		cxgb_printf(dip, CE_NOTE,
771 		    "%dx100G (%d rxq, %d txq total) %d %s.",
772 		    n100g, rqidx, tqidx, sc->intr_count,
773 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
774 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
775 		    "fixed interrupt");
776 	} else if (n40g) {
777 		cxgb_printf(dip, CE_NOTE,
778 		    "%dx40G (%d rxq, %d txq total) %d %s.",
779 		    n40g, rqidx, tqidx, sc->intr_count,
780 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
781 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
782 		    "fixed interrupt");
783 	} else if (n25g) {
784 		cxgb_printf(dip, CE_NOTE,
785 		    "%dx25G (%d rxq, %d txq total) %d %s.",
786 		    n25g, rqidx, tqidx, sc->intr_count,
787 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
788 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
789 		    "fixed interrupt");
790 	} else if (n10g && n1g) {
791 		cxgb_printf(dip, CE_NOTE,
792 		    "%dx10G %dx1G (%d rxq, %d txq total) %d %s.",
793 		    n10g, n1g, rqidx, tqidx, sc->intr_count,
794 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
795 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
796 		    "fixed interrupt");
797 	} else {
798 		cxgb_printf(dip, CE_NOTE,
799 		    "%dx%sG (%d rxq, %d txq per port) %d %s.",
800 		    n10g ? n10g : n1g,
801 		    n10g ? "10" : "1",
802 		    n10g ? iaq.nrxq10g : iaq.nrxq1g,
803 		    n10g ? iaq.ntxq10g : iaq.ntxq1g,
804 		    sc->intr_count,
805 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
806 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
807 		    "fixed interrupt");
808 	}
809 
810 	sc->ksp = setup_kstats(sc);
811 	sc->ksp_stat = setup_wc_kstats(sc);
812 	sc->params.drv_memwin = MEMWIN_NIC;
813 
814 done:
815 	if (rc != DDI_SUCCESS) {
816 		(void) t4_devo_detach(dip, DDI_DETACH);
817 
818 		/* rc may have errno style errors or DDI errors */
819 		rc = DDI_FAILURE;
820 	}
821 
822 	return (rc);
823 }
824 
825 static int
826 t4_devo_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
827 {
828 	int instance, i;
829 	struct adapter *sc;
830 	struct port_info *pi;
831 	struct sge *s;
832 
833 	if (cmd != DDI_DETACH)
834 		return (DDI_FAILURE);
835 
836 	instance = ddi_get_instance(dip);
837 	sc = ddi_get_soft_state(t4_list, instance);
838 	if (sc == NULL)
839 		return (DDI_SUCCESS);
840 
841 	if (sc->flags & FULL_INIT_DONE) {
842 		t4_intr_disable(sc);
843 		for_each_port(sc, i) {
844 			pi = sc->port[i];
845 			if (pi && pi->flags & PORT_INIT_DONE)
846 				(void) port_full_uninit(pi);
847 		}
848 		(void) adapter_full_uninit(sc);
849 	}
850 
851 	/* Safe to call no matter what */
852 	ddi_prop_remove_all(dip);
853 	ddi_remove_minor_node(dip, NULL);
854 
855 	for (i = 0; i < NCHAN; i++) {
856 		if (sc->tq[i]) {
857 			ddi_taskq_wait(sc->tq[i]);
858 			ddi_taskq_destroy(sc->tq[i]);
859 		}
860 	}
861 
862 	if (sc->ksp != NULL)
863 		kstat_delete(sc->ksp);
864 	if (sc->ksp_stat != NULL)
865 		kstat_delete(sc->ksp_stat);
866 
867 	s = &sc->sge;
868 	if (s->rxq != NULL)
869 		kmem_free(s->rxq, s->nrxq * sizeof (struct sge_rxq));
870 #ifdef TCP_OFFLOAD_ENABLE
871 	if (s->ofld_txq != NULL)
872 		kmem_free(s->ofld_txq, s->nofldtxq * sizeof (struct sge_wrq));
873 	if (s->ofld_rxq != NULL)
874 		kmem_free(s->ofld_rxq,
875 		    s->nofldrxq * sizeof (struct sge_ofld_rxq));
876 	if (s->ctrlq != NULL)
877 		kmem_free(s->ctrlq,
878 		    sc->params.nports * sizeof (struct sge_wrq));
879 #endif
880 	if (s->txq != NULL)
881 		kmem_free(s->txq, s->ntxq * sizeof (struct sge_txq));
882 	if (s->iqmap != NULL)
883 		kmem_free(s->iqmap, s->niq * sizeof (struct sge_iq *));
884 	if (s->eqmap != NULL)
885 		kmem_free(s->eqmap, s->neq * sizeof (struct sge_eq *));
886 
887 	if (s->rxbuf_cache != NULL)
888 		rxbuf_cache_destroy(s->rxbuf_cache);
889 
890 	if (sc->flags & INTR_ALLOCATED) {
891 		for (i = 0; i < sc->intr_count; i++) {
892 			(void) ddi_intr_remove_handler(sc->intr_handle[i]);
893 			(void) ddi_intr_free(sc->intr_handle[i]);
894 		}
895 		sc->flags &= ~INTR_ALLOCATED;
896 	}
897 
898 	if (sc->intr_handle != NULL) {
899 		kmem_free(sc->intr_handle,
900 		    sc->intr_count * sizeof (*sc->intr_handle));
901 	}
902 
903 	for_each_port(sc, i) {
904 		pi = sc->port[i];
905 		if (pi != NULL) {
906 			mutex_destroy(&pi->lock);
907 			kmem_free(pi, sizeof (*pi));
908 			clrbit(&sc->registered_device_map, i);
909 		}
910 	}
911 
912 	if (sc->flags & FW_OK)
913 		(void) t4_fw_bye(sc, sc->mbox);
914 
915 	if (sc->reg1h != NULL)
916 		ddi_regs_map_free(&sc->reg1h);
917 
918 	if (sc->regh != NULL)
919 		ddi_regs_map_free(&sc->regh);
920 
921 	if (sc->pci_regh != NULL)
922 		pci_config_teardown(&sc->pci_regh);
923 
924 	mutex_enter(&t4_adapter_list_lock);
925 	SLIST_REMOVE_HEAD(&t4_adapter_list, link);
926 	mutex_exit(&t4_adapter_list_lock);
927 
928 	mutex_destroy(&sc->lock);
929 	cv_destroy(&sc->cv);
930 	mutex_destroy(&sc->sfl_lock);
931 
932 #ifdef DEBUG
933 	bzero(sc, sizeof (*sc));
934 #endif
935 	ddi_soft_state_free(t4_list, instance);
936 
937 	return (DDI_SUCCESS);
938 }
939 
940 static int
941 t4_devo_quiesce(dev_info_t *dip)
942 {
943 	int instance;
944 	struct adapter *sc;
945 
946 	instance = ddi_get_instance(dip);
947 	sc = ddi_get_soft_state(t4_list, instance);
948 	if (sc == NULL)
949 		return (DDI_SUCCESS);
950 
951 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
952 	t4_intr_disable(sc);
953 	t4_write_reg(sc, A_PL_RST, F_PIORSTMODE | F_PIORST);
954 
955 	return (DDI_SUCCESS);
956 }
957 
958 static int
959 t4_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t op, void *arg,
960     void *result)
961 {
962 	char s[4];
963 	struct port_info *pi;
964 	dev_info_t *child = (dev_info_t *)arg;
965 
966 	switch (op) {
967 	case DDI_CTLOPS_REPORTDEV:
968 		pi = ddi_get_parent_data(rdip);
969 		pi->instance = ddi_get_instance(dip);
970 		pi->child_inst = ddi_get_instance(rdip);
971 		cmn_err(CE_CONT, "?%s%d is port %s on %s%d\n",
972 		    ddi_node_name(rdip), ddi_get_instance(rdip),
973 		    ddi_get_name_addr(rdip), ddi_driver_name(dip),
974 		    ddi_get_instance(dip));
975 		return (DDI_SUCCESS);
976 
977 	case DDI_CTLOPS_INITCHILD:
978 		pi = ddi_get_parent_data(child);
979 		if (pi == NULL)
980 			return (DDI_NOT_WELL_FORMED);
981 		(void) snprintf(s, sizeof (s), "%d", pi->port_id);
982 		ddi_set_name_addr(child, s);
983 		return (DDI_SUCCESS);
984 
985 	case DDI_CTLOPS_UNINITCHILD:
986 		ddi_set_name_addr(child, NULL);
987 		return (DDI_SUCCESS);
988 
989 	case DDI_CTLOPS_ATTACH:
990 	case DDI_CTLOPS_DETACH:
991 		return (DDI_SUCCESS);
992 
993 	default:
994 		return (ddi_ctlops(dip, rdip, op, arg, result));
995 	}
996 }
997 
998 static int
999 t4_bus_config(dev_info_t *dip, uint_t flags, ddi_bus_config_op_t op, void *arg,
1000     dev_info_t **cdipp)
1001 {
1002 	int instance, i;
1003 	struct adapter *sc;
1004 
1005 	instance = ddi_get_instance(dip);
1006 	sc = ddi_get_soft_state(t4_list, instance);
1007 
1008 	if (op == BUS_CONFIG_ONE) {
1009 		char *c;
1010 
1011 		/*
1012 		 * arg is something like "cxgb@0" where 0 is the port_id hanging
1013 		 * off this nexus.
1014 		 */
1015 
1016 		c = arg;
1017 		while (*(c + 1))
1018 			c++;
1019 
1020 		/* There should be exactly 1 digit after '@' */
1021 		if (*(c - 1) != '@')
1022 			return (NDI_FAILURE);
1023 
1024 		i = *c - '0';
1025 
1026 		if (add_child_node(sc, i) != 0)
1027 			return (NDI_FAILURE);
1028 
1029 		flags |= NDI_ONLINE_ATTACH;
1030 
1031 	} else if (op == BUS_CONFIG_ALL || op == BUS_CONFIG_DRIVER) {
1032 		/* Allocate and bind all child device nodes */
1033 		for_each_port(sc, i)
1034 		    (void) add_child_node(sc, i);
1035 		flags |= NDI_ONLINE_ATTACH;
1036 	}
1037 
1038 	return (ndi_busop_bus_config(dip, flags, op, arg, cdipp, 0));
1039 }
1040 
1041 static int
1042 t4_bus_unconfig(dev_info_t *dip, uint_t flags, ddi_bus_config_op_t op,
1043     void *arg)
1044 {
1045 	int instance, i, rc;
1046 	struct adapter *sc;
1047 
1048 	instance = ddi_get_instance(dip);
1049 	sc = ddi_get_soft_state(t4_list, instance);
1050 
1051 	if (op == BUS_CONFIG_ONE || op == BUS_UNCONFIG_ALL ||
1052 	    op == BUS_UNCONFIG_DRIVER)
1053 		flags |= NDI_UNCONFIG;
1054 
1055 	rc = ndi_busop_bus_unconfig(dip, flags, op, arg);
1056 	if (rc != 0)
1057 		return (rc);
1058 
1059 	if (op == BUS_UNCONFIG_ONE) {
1060 		char *c;
1061 
1062 		c = arg;
1063 		while (*(c + 1))
1064 			c++;
1065 
1066 		if (*(c - 1) != '@')
1067 			return (NDI_SUCCESS);
1068 
1069 		i = *c - '0';
1070 
1071 		rc = remove_child_node(sc, i);
1072 
1073 	} else if (op == BUS_UNCONFIG_ALL || op == BUS_UNCONFIG_DRIVER) {
1074 
1075 		for_each_port(sc, i)
1076 		    (void) remove_child_node(sc, i);
1077 	}
1078 
1079 	return (rc);
1080 }
1081 
1082 /* ARGSUSED */
1083 static int
1084 t4_cb_open(dev_t *devp, int flag, int otyp, cred_t *credp)
1085 {
1086 	struct adapter *sc;
1087 
1088 	if (otyp != OTYP_CHR)
1089 		return (EINVAL);
1090 
1091 	sc = ddi_get_soft_state(t4_list, getminor(*devp));
1092 	if (sc == NULL)
1093 		return (ENXIO);
1094 
1095 	return (atomic_cas_uint(&sc->open, 0, EBUSY));
1096 }
1097 
1098 /* ARGSUSED */
1099 static int
1100 t4_cb_close(dev_t dev, int flag, int otyp, cred_t *credp)
1101 {
1102 	struct adapter *sc;
1103 
1104 	sc = ddi_get_soft_state(t4_list, getminor(dev));
1105 	if (sc == NULL)
1106 		return (EINVAL);
1107 
1108 	(void) atomic_swap_uint(&sc->open, 0);
1109 	return (0);
1110 }
1111 
1112 /* ARGSUSED */
1113 static int
1114 t4_cb_ioctl(dev_t dev, int cmd, intptr_t d, int mode, cred_t *credp, int *rp)
1115 {
1116 	int instance;
1117 	struct adapter *sc;
1118 	void *data = (void *)d;
1119 
1120 	if (crgetuid(credp) != 0)
1121 		return (EPERM);
1122 
1123 	instance = getminor(dev);
1124 	sc = ddi_get_soft_state(t4_list, instance);
1125 	if (sc == NULL)
1126 		return (EINVAL);
1127 
1128 	return (t4_ioctl(sc, cmd, data, mode));
1129 }
1130 
1131 static unsigned int
1132 getpf(struct adapter *sc)
1133 {
1134 	int rc, *data;
1135 	uint_t n, pf;
1136 
1137 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, sc->dip,
1138 	    DDI_PROP_DONTPASS, "reg", &data, &n);
1139 	if (rc != DDI_SUCCESS) {
1140 		cxgb_printf(sc->dip, CE_WARN,
1141 		    "failed to lookup \"reg\" property: %d", rc);
1142 		return (0xff);
1143 	}
1144 
1145 	pf = PCI_REG_FUNC_G(data[0]);
1146 	ddi_prop_free(data);
1147 
1148 	return (pf);
1149 }
1150 
1151 
1152 static struct fw_info *
1153 find_fw_info(int chip)
1154 {
1155 	u32 i;
1156 
1157 	fi[0].chip = CHELSIO_T4;
1158 	fi[0].fw_hdr.chip = FW_HDR_CHIP_T4;
1159 	fi[0].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T4));
1160 	fi[0].fw_hdr.intfver_nic = FW_INTFVER(T4, NIC);
1161 	fi[0].fw_hdr.intfver_vnic = FW_INTFVER(T4, VNIC);
1162 	fi[0].fw_hdr.intfver_ofld = FW_INTFVER(T4, OFLD);
1163 	fi[0].fw_hdr.intfver_ri = FW_INTFVER(T4, RI);
1164 	fi[0].fw_hdr.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU);
1165 	fi[0].fw_hdr.intfver_iscsi = FW_INTFVER(T4, ISCSI);
1166 	fi[0].fw_hdr.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU);
1167 	fi[0].fw_hdr.intfver_fcoe = FW_INTFVER(T4, FCOE);
1168 
1169 	fi[1].chip = CHELSIO_T5;
1170 	fi[1].fw_hdr.chip = FW_HDR_CHIP_T5;
1171 	fi[1].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T5));
1172 	fi[1].fw_hdr.intfver_nic = FW_INTFVER(T5, NIC);
1173 	fi[1].fw_hdr.intfver_vnic = FW_INTFVER(T5, VNIC);
1174 	fi[1].fw_hdr.intfver_ofld = FW_INTFVER(T5, OFLD);
1175 	fi[1].fw_hdr.intfver_ri = FW_INTFVER(T5, RI);
1176 	fi[1].fw_hdr.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU);
1177 	fi[1].fw_hdr.intfver_iscsi = FW_INTFVER(T5, ISCSI);
1178 	fi[1].fw_hdr.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU);
1179 	fi[1].fw_hdr.intfver_fcoe = FW_INTFVER(T5, FCOE);
1180 
1181 	fi[2].chip = CHELSIO_T6;
1182 	fi[2].fw_hdr.chip = FW_HDR_CHIP_T6;
1183 	fi[2].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T6));
1184 	fi[2].fw_hdr.intfver_nic = FW_INTFVER(T6, NIC);
1185 	fi[2].fw_hdr.intfver_vnic = FW_INTFVER(T6, VNIC);
1186 	fi[2].fw_hdr.intfver_ofld = FW_INTFVER(T6, OFLD);
1187 	fi[2].fw_hdr.intfver_ri = FW_INTFVER(T6, RI);
1188 	fi[2].fw_hdr.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU);
1189 	fi[2].fw_hdr.intfver_iscsi = FW_INTFVER(T6, ISCSI);
1190 	fi[2].fw_hdr.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU);
1191 	fi[2].fw_hdr.intfver_fcoe = FW_INTFVER(T6, FCOE);
1192 
1193 	for (i = 0; i < ARRAY_SIZE(fi); i++) {
1194 		if (fi[i].chip == chip)
1195 			return &fi[i];
1196 	}
1197 
1198 	return NULL;
1199 }
1200 
1201 /*
1202  * Install a compatible firmware (if required), establish contact with it,
1203  * become the master, and reset the device.
1204  */
1205 static int
1206 prep_firmware(struct adapter *sc)
1207 {
1208 	int rc;
1209 	int fw_size;
1210 	int reset = 1;
1211 	enum dev_state state;
1212 	unsigned char *fw_data;
1213 	struct fw_info *fw_info;
1214 	struct fw_hdr *card_fw;
1215 
1216 	struct driver_properties *p = &sc->props;
1217 
1218 	/* Contact firmware, request master */
1219 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MUST, &state);
1220 	if (rc < 0) {
1221 		rc = -rc;
1222 		cxgb_printf(sc->dip, CE_WARN,
1223 		    "failed to connect to the firmware: %d.", rc);
1224 		return (rc);
1225 	}
1226 
1227 	if (rc == sc->mbox)
1228 		sc->flags |= MASTER_PF;
1229 
1230 	/* We may need FW version info for later reporting */
1231 	t4_get_version_info(sc);
1232 	fw_info = find_fw_info(CHELSIO_CHIP_VERSION(sc->params.chip));
1233 	/* allocate memory to read the header of the firmware on the
1234 	 * card
1235 	 */
1236 	if (!fw_info) {
1237 		cxgb_printf(sc->dip, CE_WARN,
1238 			    "unable to look up firmware information for chip %d.\n",
1239 			    CHELSIO_CHIP_VERSION(sc->params.chip));
1240 		return EINVAL;
1241 	}
1242 	card_fw = kmem_zalloc(sizeof(*card_fw), KM_SLEEP);
1243 	if(!card_fw) {
1244 		cxgb_printf(sc->dip, CE_WARN,
1245 			    "Memory allocation for card FW header failed\n");
1246 		return ENOMEM;
1247 	}
1248 	switch(CHELSIO_CHIP_VERSION(sc->params.chip)) {
1249 	case CHELSIO_T4:
1250 		fw_data = t4fw_data;
1251 		fw_size = t4fw_size;
1252 		break;
1253 	case CHELSIO_T5:
1254 		fw_data = t5fw_data;
1255 		fw_size = t5fw_size;
1256 		break;
1257 	case CHELSIO_T6:
1258 		fw_data = t6fw_data;
1259 		fw_size = t6fw_size;
1260 		break;
1261 	default:
1262 		cxgb_printf(sc->dip, CE_WARN, "Adapter type not supported\n");
1263 		kmem_free(card_fw, sizeof(*card_fw));
1264 		return EINVAL;
1265 	}
1266 
1267 	rc = -t4_prep_fw(sc, fw_info, fw_data, fw_size, card_fw,
1268 			 p->t4_fw_install, state, &reset);
1269 
1270 	kmem_free(card_fw, sizeof(*card_fw));
1271 
1272 	if (rc != 0) {
1273 		cxgb_printf(sc->dip, CE_WARN,
1274 		    "failed to install firmware: %d", rc);
1275 		return (rc);
1276 	} else {
1277 		/* refresh */
1278 		(void) t4_check_fw_version(sc);
1279 	}
1280 
1281 	/* Reset device */
1282 	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
1283 	if (rc != 0) {
1284 		cxgb_printf(sc->dip, CE_WARN,
1285 		    "firmware reset failed: %d.", rc);
1286 		if (rc != ETIMEDOUT && rc != EIO)
1287 			(void) t4_fw_bye(sc, sc->mbox);
1288 		return (rc);
1289 	}
1290 
1291 	/* Partition adapter resources as specified in the config file. */
1292 	if (sc->flags & MASTER_PF) {
1293 		/* Handle default vs special T4 config file */
1294 
1295 		rc = partition_resources(sc);
1296 		if (rc != 0)
1297 			goto err;	/* error message displayed already */
1298 	}
1299 
1300 	sc->flags |= FW_OK;
1301 	return (0);
1302 err:
1303 	return (rc);
1304 
1305 }
1306 
1307 static const struct memwin t4_memwin[] = {
1308 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1309 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1310 	{ MEMWIN2_BASE, MEMWIN2_APERTURE }
1311 };
1312 
1313 static const struct memwin t5_memwin[] = {
1314 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1315 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1316 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
1317 };
1318 
1319 #define	FW_PARAM_DEV(param) \
1320 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
1321 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
1322 #define	FW_PARAM_PFVF(param) \
1323 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
1324 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
1325 
1326 /*
1327  * Verify that the memory range specified by the memtype/offset/len pair is
1328  * valid and lies entirely within the memtype specified.  The global address of
1329  * the start of the range is returned in addr.
1330  */
1331 int
1332 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
1333 	uint32_t *addr)
1334 {
1335 	uint32_t em, addr_len, maddr, mlen;
1336 
1337 	/* Memory can only be accessed in naturally aligned 4 byte units */
1338 	if (off & 3 || len & 3 || len == 0)
1339 		return (EINVAL);
1340 
1341 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
1342 	switch (mtype) {
1343 		case MEM_EDC0:
1344 			if (!(em & F_EDRAM0_ENABLE))
1345 				return (EINVAL);
1346 			addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
1347 			maddr = G_EDRAM0_BASE(addr_len) << 20;
1348 			mlen = G_EDRAM0_SIZE(addr_len) << 20;
1349 			break;
1350 		case MEM_EDC1:
1351 			if (!(em & F_EDRAM1_ENABLE))
1352 				return (EINVAL);
1353 			addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
1354 			maddr = G_EDRAM1_BASE(addr_len) << 20;
1355 			mlen = G_EDRAM1_SIZE(addr_len) << 20;
1356 			break;
1357 		case MEM_MC:
1358 			if (!(em & F_EXT_MEM_ENABLE))
1359 				return (EINVAL);
1360 			addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
1361 			maddr = G_EXT_MEM_BASE(addr_len) << 20;
1362 			mlen = G_EXT_MEM_SIZE(addr_len) << 20;
1363 			break;
1364 		case MEM_MC1:
1365 			if (is_t4(sc->params.chip) || !(em & F_EXT_MEM1_ENABLE))
1366 				return (EINVAL);
1367 			addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
1368 			maddr = G_EXT_MEM1_BASE(addr_len) << 20;
1369 			mlen = G_EXT_MEM1_SIZE(addr_len) << 20;
1370 			break;
1371 		default:
1372 			return (EINVAL);
1373 	}
1374 
1375 	if (mlen > 0 && off < mlen && off + len <= mlen) {
1376 		*addr = maddr + off;    /* global address */
1377 		return (0);
1378 	}
1379 
1380 	return (EFAULT);
1381 }
1382 
1383 void
1384 memwin_info(struct adapter *sc, int win, uint32_t *base, uint32_t *aperture)
1385 {
1386 	const struct memwin *mw;
1387 
1388 	if (is_t4(sc->params.chip)) {
1389 		mw = &t4_memwin[win];
1390 	} else {
1391 		mw = &t5_memwin[win];
1392 	}
1393 
1394 	if (base != NULL)
1395 		*base = mw->base;
1396 	if (aperture != NULL)
1397 		*aperture = mw->aperture;
1398 }
1399 
1400 /*
1401  * Upload configuration file to card's memory.
1402  */
1403 static int
1404 upload_config_file(struct adapter *sc, uint32_t *mt, uint32_t *ma)
1405 {
1406 	int rc = 0, cflen;
1407 	u_int i, n;
1408 	uint32_t param, val, addr, mtype, maddr;
1409 	uint32_t off, mw_base, mw_aperture;
1410 	const uint32_t *cfdata;
1411 
1412 	/* Figure out where the firmware wants us to upload it. */
1413 	param = FW_PARAM_DEV(CF);
1414 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
1415 	if (rc != 0) {
1416 		/* Firmwares without config file support will fail this way */
1417 		cxgb_printf(sc->dip, CE_WARN,
1418 		    "failed to query config file location: %d.\n", rc);
1419 		return (rc);
1420 	}
1421 	*mt = mtype = G_FW_PARAMS_PARAM_Y(val);
1422 	*ma = maddr = G_FW_PARAMS_PARAM_Z(val) << 16;
1423 
1424 	switch (CHELSIO_CHIP_VERSION(sc->params.chip)) {
1425 	case CHELSIO_T4:
1426 		cflen = t4cfg_size & ~3;
1427 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1428 		cfdata = (const uint32_t *)t4cfg_data;
1429 		break;
1430 	case CHELSIO_T5:
1431 		cflen = t5cfg_size & ~3;
1432 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1433 		cfdata = (const uint32_t *)t5cfg_data;
1434 		break;
1435 	case CHELSIO_T6:
1436 		cflen = t6cfg_size & ~3;
1437 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1438 		cfdata = (const uint32_t *)t6cfg_data;
1439 		break;
1440 	default:
1441 		cxgb_printf(sc->dip, CE_WARN,
1442 			    "Invalid Adapter detected\n");
1443 		return EINVAL;
1444 	}
1445 
1446 	if (cflen > FLASH_CFG_MAX_SIZE) {
1447 		cxgb_printf(sc->dip, CE_WARN,
1448 		    "config file too long (%d, max allowed is %d).  ",
1449 		    cflen, FLASH_CFG_MAX_SIZE);
1450 		return (EFBIG);
1451 	}
1452 
1453 	rc = validate_mt_off_len(sc, mtype, maddr, cflen, &addr);
1454 	if (rc != 0) {
1455 
1456 		cxgb_printf(sc->dip, CE_WARN,
1457 		    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
1458 		    "Will try to use the config on the card, if any.\n",
1459 		    __func__, mtype, maddr, cflen, rc);
1460 		return (EFAULT);
1461 	}
1462 
1463 	memwin_info(sc, 2, &mw_base, &mw_aperture);
1464 	while (cflen) {
1465 		off = position_memwin(sc, 2, addr);
1466 		n = min(cflen, mw_aperture - off);
1467 		for (i = 0; i < n; i += 4)
1468 			t4_write_reg(sc, mw_base + off + i, *cfdata++);
1469 		cflen -= n;
1470 		addr += n;
1471 	}
1472 
1473 	return (rc);
1474 }
1475 
1476 /*
1477  * Partition chip resources for use between various PFs, VFs, etc.  This is done
1478  * by uploading the firmware configuration file to the adapter and instructing
1479  * the firmware to process it.
1480  */
1481 static int
1482 partition_resources(struct adapter *sc)
1483 {
1484 	int rc;
1485 	struct fw_caps_config_cmd caps;
1486 	uint32_t mtype, maddr, finicsum, cfcsum;
1487 
1488 	rc = upload_config_file(sc, &mtype, &maddr);
1489 	if (rc != 0) {
1490 		mtype = FW_MEMTYPE_CF_FLASH;
1491 		maddr = t4_flash_cfg_addr(sc);
1492 	}
1493 
1494 	bzero(&caps, sizeof (caps));
1495 	caps.op_to_write = BE_32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1496 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1497 	caps.cfvalid_to_len16 = BE_32(F_FW_CAPS_CONFIG_CMD_CFVALID |
1498 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
1499 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) | FW_LEN16(caps));
1500 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), &caps);
1501 	if (rc != 0) {
1502 		cxgb_printf(sc->dip, CE_WARN,
1503 		    "failed to pre-process config file: %d.\n", rc);
1504 		return (rc);
1505 	}
1506 
1507 	finicsum = ntohl(caps.finicsum);
1508 	cfcsum = ntohl(caps.cfcsum);
1509 	if (finicsum != cfcsum) {
1510 		cxgb_printf(sc->dip, CE_WARN,
1511 		    "WARNING: config file checksum mismatch: %08x %08x\n",
1512 		    finicsum, cfcsum);
1513 	}
1514 	sc->cfcsum = cfcsum;
1515 
1516 	/* TODO: Need to configure this correctly */
1517 	caps.toecaps = htons(FW_CAPS_CONFIG_TOE);
1518 	caps.iscsicaps = 0;
1519 	caps.rdmacaps = 0;
1520 	caps.fcoecaps = 0;
1521 	/* TODO: Disable VNIC cap for now */
1522 	caps.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
1523 
1524 	caps.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1525 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
1526 	caps.cfvalid_to_len16 = htonl(FW_LEN16(caps));
1527 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), NULL);
1528 	if (rc != 0) {
1529 		cxgb_printf(sc->dip, CE_WARN,
1530 		    "failed to process config file: %d.\n", rc);
1531 		return (rc);
1532 	}
1533 
1534 	return (0);
1535 }
1536 
1537 /*
1538  * Tweak configuration based on module parameters, etc.  Most of these have
1539  * defaults assigned to them by Firmware Configuration Files (if we're using
1540  * them) but need to be explicitly set if we're using hard-coded
1541  * initialization.  But even in the case of using Firmware Configuration
1542  * Files, we'd like to expose the ability to change these via module
1543  * parameters so these are essentially common tweaks/settings for
1544  * Configuration Files and hard-coded initialization ...
1545  */
1546 static int
1547 adap__pre_init_tweaks(struct adapter *sc)
1548 {
1549 	int rx_dma_offset = 2; /* Offset of RX packets into DMA buffers */
1550 
1551 	/*
1552 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
1553 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
1554 	 * 64B Cache Line Size ...
1555 	 */
1556 	(void) t4_fixup_host_params_compat(sc, PAGE_SIZE, CACHE_LINE, T5_LAST_REV);
1557 
1558 	t4_set_reg_field(sc, A_SGE_CONTROL,
1559 			 V_PKTSHIFT(M_PKTSHIFT), V_PKTSHIFT(rx_dma_offset));
1560 
1561 	return 0;
1562 }
1563 /*
1564  * Retrieve parameters that are needed (or nice to have) prior to calling
1565  * t4_sge_init and t4_fw_initialize.
1566  */
1567 static int
1568 get_params__pre_init(struct adapter *sc)
1569 {
1570 	int rc;
1571 	uint32_t param[2], val[2];
1572 	struct fw_devlog_cmd cmd;
1573 	struct devlog_params *dlog = &sc->params.devlog;
1574 
1575 	/*
1576 	 * Grab the raw VPD parameters.
1577 	 */
1578 	rc = -t4_get_raw_vpd_params(sc, &sc->params.vpd);
1579 	if (rc != 0) {
1580 		cxgb_printf(sc->dip, CE_WARN,
1581 		    "failed to query VPD parameters (pre_init): %d.\n", rc);
1582 		return (rc);
1583 	}
1584 
1585 	param[0] = FW_PARAM_DEV(PORTVEC);
1586 	param[1] = FW_PARAM_DEV(CCLK);
1587 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
1588 	if (rc != 0) {
1589 		cxgb_printf(sc->dip, CE_WARN,
1590 		    "failed to query parameters (pre_init): %d.\n", rc);
1591 		return (rc);
1592 	}
1593 
1594 	sc->params.portvec = val[0];
1595 	sc->params.nports = 0;
1596 	while (val[0]) {
1597 		sc->params.nports++;
1598 		val[0] &= val[0] - 1;
1599 	}
1600 
1601 	sc->params.vpd.cclk = val[1];
1602 
1603 	/* Read device log parameters. */
1604 	bzero(&cmd, sizeof (cmd));
1605 	cmd.op_to_write = htonl(V_FW_CMD_OP(FW_DEVLOG_CMD) |
1606 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1607 	cmd.retval_len16 = htonl(FW_LEN16(cmd));
1608 	rc = -t4_wr_mbox(sc, sc->mbox, &cmd, sizeof (cmd), &cmd);
1609 	if (rc != 0) {
1610 		cxgb_printf(sc->dip, CE_WARN,
1611 		    "failed to get devlog parameters: %d.\n", rc);
1612 		bzero(dlog, sizeof (*dlog));
1613 		rc = 0;	/* devlog isn't critical for device operation */
1614 	} else {
1615 		val[0] = ntohl(cmd.memtype_devlog_memaddr16_devlog);
1616 		dlog->memtype = G_FW_DEVLOG_CMD_MEMTYPE_DEVLOG(val[0]);
1617 		dlog->start = G_FW_DEVLOG_CMD_MEMADDR16_DEVLOG(val[0]) << 4;
1618 		dlog->size = ntohl(cmd.memsize_devlog);
1619 	}
1620 
1621 	return (rc);
1622 }
1623 
1624 /*
1625  * Retrieve various parameters that are of interest to the driver.  The device
1626  * has been initialized by the firmware at this point.
1627  */
1628 static int
1629 get_params__post_init(struct adapter *sc)
1630 {
1631 	int rc;
1632 	uint32_t param[7], val[7];
1633 	struct fw_caps_config_cmd caps;
1634 
1635 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
1636 	param[1] = FW_PARAM_PFVF(EQ_START);
1637 	param[2] = FW_PARAM_PFVF(FILTER_START);
1638 	param[3] = FW_PARAM_PFVF(FILTER_END);
1639 	param[4] = FW_PARAM_PFVF(L2T_START);
1640 	param[5] = FW_PARAM_PFVF(L2T_END);
1641 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
1642 	if (rc != 0) {
1643 		cxgb_printf(sc->dip, CE_WARN,
1644 		    "failed to query parameters (post_init): %d.\n", rc);
1645 		return (rc);
1646 	}
1647 
1648 	/* LINTED: E_ASSIGN_NARROW_CONV */
1649 	sc->sge.iq_start = val[0];
1650 	sc->sge.eq_start = val[1];
1651 	sc->tids.ftid_base = val[2];
1652 	sc->tids.nftids = val[3] - val[2] + 1;
1653 	sc->vres.l2t.start = val[4];
1654 	sc->vres.l2t.size = val[5] - val[4] + 1;
1655 
1656 	/* get capabilites */
1657 	bzero(&caps, sizeof (caps));
1658 	caps.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1659 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1660 	caps.cfvalid_to_len16 = htonl(FW_LEN16(caps));
1661 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), &caps);
1662 	if (rc != 0) {
1663 		cxgb_printf(sc->dip, CE_WARN,
1664 		    "failed to get card capabilities: %d.\n", rc);
1665 		return (rc);
1666 	}
1667 
1668 	if (caps.toecaps != 0) {
1669 		/* query offload-related parameters */
1670 		param[0] = FW_PARAM_DEV(NTID);
1671 		param[1] = FW_PARAM_PFVF(SERVER_START);
1672 		param[2] = FW_PARAM_PFVF(SERVER_END);
1673 		param[3] = FW_PARAM_PFVF(TDDP_START);
1674 		param[4] = FW_PARAM_PFVF(TDDP_END);
1675 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
1676 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
1677 		if (rc != 0) {
1678 			cxgb_printf(sc->dip, CE_WARN,
1679 			    "failed to query TOE parameters: %d.\n", rc);
1680 			return (rc);
1681 		}
1682 		sc->tids.ntids = val[0];
1683 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
1684 		sc->tids.stid_base = val[1];
1685 		sc->tids.nstids = val[2] - val[1] + 1;
1686 		sc->vres.ddp.start = val[3];
1687 		sc->vres.ddp.size = val[4] - val[3] + 1;
1688 		sc->params.ofldq_wr_cred = val[5];
1689 		sc->params.offload = 1;
1690 	}
1691 
1692 	/* These are finalized by FW initialization, load their values now */
1693 	val[0] = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
1694 	sc->params.tp.tre = G_TIMERRESOLUTION(val[0]);
1695 	sc->params.tp.dack_re = G_DELAYEDACKRESOLUTION(val[0]);
1696 	t4_read_mtu_tbl(sc, sc->params.mtus, NULL);
1697 
1698 	return (rc);
1699 }
1700 
1701 static int
1702 set_params__post_init(struct adapter *sc)
1703 {
1704 	uint32_t param, val;
1705 
1706 	/* ask for encapsulated CPLs */
1707 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
1708 	val = 1;
1709 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
1710 
1711 	return (0);
1712 }
1713 
1714 /* TODO: verify */
1715 static void
1716 setup_memwin(struct adapter *sc)
1717 {
1718 	pci_regspec_t *data;
1719 	int rc;
1720 	uint_t n;
1721 	uintptr_t bar0;
1722 	uintptr_t mem_win0_base, mem_win1_base, mem_win2_base;
1723 	uintptr_t mem_win2_aperture;
1724 
1725 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, sc->dip,
1726 	    DDI_PROP_DONTPASS, "assigned-addresses", (int **)&data, &n);
1727 	if (rc != DDI_SUCCESS) {
1728 		cxgb_printf(sc->dip, CE_WARN,
1729 		    "failed to lookup \"assigned-addresses\" property: %d", rc);
1730 		return;
1731 	}
1732 	n /= sizeof (*data);
1733 
1734 	bar0 = ((uint64_t)data[0].pci_phys_mid << 32) | data[0].pci_phys_low;
1735 	ddi_prop_free(data);
1736 
1737 	if (is_t4(sc->params.chip)) {
1738 		mem_win0_base = bar0 + MEMWIN0_BASE;
1739 		mem_win1_base = bar0 + MEMWIN1_BASE;
1740 		mem_win2_base = bar0 + MEMWIN2_BASE;
1741 		mem_win2_aperture = MEMWIN2_APERTURE;
1742 	} else {
1743 		/* For T5, only relative offset inside the PCIe BAR is passed */
1744 		mem_win0_base = MEMWIN0_BASE;
1745 		mem_win1_base = MEMWIN1_BASE;
1746 		mem_win2_base = MEMWIN2_BASE_T5;
1747 		mem_win2_aperture = MEMWIN2_APERTURE_T5;
1748 	}
1749 
1750 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 0),
1751 	    mem_win0_base | V_BIR(0) |
1752 	    V_WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
1753 
1754 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 1),
1755 	    mem_win1_base | V_BIR(0) |
1756 	    V_WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
1757 
1758 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2),
1759 	    mem_win2_base | V_BIR(0) |
1760 	    V_WINDOW(ilog2(mem_win2_aperture) - 10));
1761 
1762 	/* flush */
1763 	(void)t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
1764 }
1765 
1766 /*
1767  * Positions the memory window such that it can be used to access the specified
1768  * address in the chip's address space.  The return value is the offset of addr
1769  * from the start of the window.
1770  */
1771 uint32_t
1772 position_memwin(struct adapter *sc, int n, uint32_t addr)
1773 {
1774 	uint32_t start, pf;
1775 	uint32_t reg;
1776 
1777 	if (addr & 3) {
1778 		cxgb_printf(sc->dip, CE_WARN,
1779 		    "addr (0x%x) is not at a 4B boundary.\n", addr);
1780 		return (EFAULT);
1781 	}
1782 
1783 	if (is_t4(sc->params.chip)) {
1784 		pf = 0;
1785 		start = addr & ~0xf;    /* start must be 16B aligned */
1786 	} else {
1787 		pf = V_PFNUM(sc->pf);
1788 		start = addr & ~0x7f;   /* start must be 128B aligned */
1789 	}
1790 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, n);
1791 
1792 	t4_write_reg(sc, reg, start | pf);
1793 	(void) t4_read_reg(sc, reg);
1794 
1795 	return (addr - start);
1796 }
1797 
1798 
1799 /*
1800  * Reads the named property and fills up the "data" array (which has at least
1801  * "count" elements).  We first try and lookup the property for our dev_t and
1802  * then retry with DDI_DEV_T_ANY if it's not found.
1803  *
1804  * Returns non-zero if the property was found and "data" has been updated.
1805  */
1806 static int
1807 prop_lookup_int_array(struct adapter *sc, char *name, int *data, uint_t count)
1808 {
1809 	dev_info_t *dip = sc->dip;
1810 	dev_t dev = sc->dev;
1811 	int rc, *d;
1812 	uint_t i, n;
1813 
1814 	rc = ddi_prop_lookup_int_array(dev, dip, DDI_PROP_DONTPASS,
1815 	    name, &d, &n);
1816 	if (rc == DDI_PROP_SUCCESS)
1817 		goto found;
1818 
1819 	if (rc != DDI_PROP_NOT_FOUND) {
1820 		cxgb_printf(dip, CE_WARN,
1821 		    "failed to lookup property %s for minor %d: %d.",
1822 		    name, getminor(dev), rc);
1823 		return (0);
1824 	}
1825 
1826 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
1827 	    name, &d, &n);
1828 	if (rc == DDI_PROP_SUCCESS)
1829 		goto found;
1830 
1831 	if (rc != DDI_PROP_NOT_FOUND) {
1832 		cxgb_printf(dip, CE_WARN,
1833 		    "failed to lookup property %s: %d.", name, rc);
1834 		return (0);
1835 	}
1836 
1837 	return (0);
1838 
1839 found:
1840 	if (n > count) {
1841 		cxgb_printf(dip, CE_NOTE,
1842 		    "property %s has too many elements (%d), ignoring extras",
1843 		    name, n);
1844 	}
1845 
1846 	for (i = 0; i < n && i < count; i++)
1847 		data[i] = d[i];
1848 	ddi_prop_free(d);
1849 
1850 	return (1);
1851 }
1852 
1853 static int
1854 prop_lookup_int(struct adapter *sc, char *name, int defval)
1855 {
1856 	int rc;
1857 
1858 	rc = ddi_prop_get_int(sc->dev, sc->dip, DDI_PROP_DONTPASS, name, -1);
1859 	if (rc != -1)
1860 		return (rc);
1861 
1862 	return (ddi_prop_get_int(DDI_DEV_T_ANY, sc->dip, DDI_PROP_DONTPASS,
1863 	    name, defval));
1864 }
1865 
1866 static int
1867 init_driver_props(struct adapter *sc, struct driver_properties *p)
1868 {
1869 	dev_t dev = sc->dev;
1870 	dev_info_t *dip = sc->dip;
1871 	int i, *data;
1872 	uint_t tmr[SGE_NTIMERS] = {5, 10, 20, 50, 100, 200};
1873 	uint_t cnt[SGE_NCOUNTERS] = {1, 8, 16, 32}; /* 63 max */
1874 
1875 	/*
1876 	 * Holdoff timer
1877 	 */
1878 	data = &p->timer_val[0];
1879 	for (i = 0; i < SGE_NTIMERS; i++)
1880 		data[i] = tmr[i];
1881 	(void) prop_lookup_int_array(sc, "holdoff-timer-values", data,
1882 	    SGE_NTIMERS);
1883 	for (i = 0; i < SGE_NTIMERS; i++) {
1884 		int limit = 200U;
1885 		if (data[i] > limit) {
1886 			cxgb_printf(dip, CE_WARN,
1887 			    "holdoff timer %d is too high (%d), lowered to %d.",
1888 			    i, data[i], limit);
1889 			data[i] = limit;
1890 		}
1891 	}
1892 	(void) ddi_prop_update_int_array(dev, dip, "holdoff-timer-values",
1893 	    data, SGE_NTIMERS);
1894 
1895 	/*
1896 	 * Holdoff packet counter
1897 	 */
1898 	data = &p->counter_val[0];
1899 	for (i = 0; i < SGE_NCOUNTERS; i++)
1900 		data[i] = cnt[i];
1901 	(void) prop_lookup_int_array(sc, "holdoff-pkt-counter-values", data,
1902 	    SGE_NCOUNTERS);
1903 	for (i = 0; i < SGE_NCOUNTERS; i++) {
1904 		int limit = M_THRESHOLD_0;
1905 		if (data[i] > limit) {
1906 			cxgb_printf(dip, CE_WARN,
1907 			    "holdoff pkt-counter %d is too high (%d), "
1908 			    "lowered to %d.", i, data[i], limit);
1909 			data[i] = limit;
1910 		}
1911 	}
1912 	(void) ddi_prop_update_int_array(dev, dip, "holdoff-pkt-counter-values",
1913 	    data, SGE_NCOUNTERS);
1914 
1915 	/*
1916 	 * Maximum # of tx and rx queues to use for each
1917 	 * 100G, 40G, 25G, 10G and 1G port.
1918 	 */
1919 	p->max_ntxq_10g = prop_lookup_int(sc, "max-ntxq-10G-port", 8);
1920 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-10G-port",
1921 	    p->max_ntxq_10g);
1922 
1923 	p->max_nrxq_10g = prop_lookup_int(sc, "max-nrxq-10G-port", 8);
1924 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-10G-port",
1925 	    p->max_nrxq_10g);
1926 
1927 	p->max_ntxq_1g = prop_lookup_int(sc, "max-ntxq-1G-port", 2);
1928 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-1G-port",
1929 	    p->max_ntxq_1g);
1930 
1931 	p->max_nrxq_1g = prop_lookup_int(sc, "max-nrxq-1G-port", 2);
1932 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-1G-port",
1933 	    p->max_nrxq_1g);
1934 
1935 #ifdef TCP_OFFLOAD_ENABLE
1936 	p->max_nofldtxq_10g = prop_lookup_int(sc, "max-nofldtxq-10G-port", 8);
1937 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-10G-port",
1938 	    p->max_nofldtxq_10g);
1939 
1940 	p->max_nofldrxq_10g = prop_lookup_int(sc, "max-nofldrxq-10G-port", 2);
1941 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-10G-port",
1942 	    p->max_nofldrxq_10g);
1943 
1944 	p->max_nofldtxq_1g = prop_lookup_int(sc, "max-nofldtxq-1G-port", 2);
1945 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-1G-port",
1946 	    p->max_nofldtxq_1g);
1947 
1948 	p->max_nofldrxq_1g = prop_lookup_int(sc, "max-nofldrxq-1G-port", 1);
1949 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-1G-port",
1950 	    p->max_nofldrxq_1g);
1951 #endif
1952 
1953 	/*
1954 	 * Holdoff parameters for 10G and 1G ports.
1955 	 */
1956 	p->tmr_idx_10g = prop_lookup_int(sc, "holdoff-timer-idx-10G", 0);
1957 	(void) ddi_prop_update_int(dev, dip, "holdoff-timer-idx-10G",
1958 	    p->tmr_idx_10g);
1959 
1960 	p->pktc_idx_10g = prop_lookup_int(sc, "holdoff-pktc-idx-10G", 2);
1961 	(void) ddi_prop_update_int(dev, dip, "holdoff-pktc-idx-10G",
1962 	    p->pktc_idx_10g);
1963 
1964 	p->tmr_idx_1g = prop_lookup_int(sc, "holdoff-timer-idx-1G", 0);
1965 	(void) ddi_prop_update_int(dev, dip, "holdoff-timer-idx-1G",
1966 	    p->tmr_idx_1g);
1967 
1968 	p->pktc_idx_1g = prop_lookup_int(sc, "holdoff-pktc-idx-1G", 2);
1969 	(void) ddi_prop_update_int(dev, dip, "holdoff-pktc-idx-1G",
1970 	    p->pktc_idx_1g);
1971 
1972 	/*
1973 	 * Size (number of entries) of each tx and rx queue.
1974 	 */
1975 	i = prop_lookup_int(sc, "qsize-txq", TX_EQ_QSIZE);
1976 	p->qsize_txq = max(i, 128);
1977 	if (p->qsize_txq != i) {
1978 		cxgb_printf(dip, CE_WARN,
1979 		    "using %d instead of %d as the tx queue size",
1980 		    p->qsize_txq, i);
1981 	}
1982 	(void) ddi_prop_update_int(dev, dip, "qsize-txq", p->qsize_txq);
1983 
1984 	i = prop_lookup_int(sc, "qsize-rxq", RX_IQ_QSIZE);
1985 	p->qsize_rxq = max(i, 128);
1986 	while (p->qsize_rxq & 7)
1987 		p->qsize_rxq--;
1988 	if (p->qsize_rxq != i) {
1989 		cxgb_printf(dip, CE_WARN,
1990 		    "using %d instead of %d as the rx queue size",
1991 		    p->qsize_rxq, i);
1992 	}
1993 	(void) ddi_prop_update_int(dev, dip, "qsize-rxq", p->qsize_rxq);
1994 
1995 	/*
1996 	 * Interrupt types allowed.
1997 	 * Bits 0, 1, 2 = INTx, MSI, MSI-X respectively.  See sys/ddi_intr.h
1998 	 */
1999 	p->intr_types = prop_lookup_int(sc, "interrupt-types",
2000 	    DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI | DDI_INTR_TYPE_FIXED);
2001 	(void) ddi_prop_update_int(dev, dip, "interrupt-types", p->intr_types);
2002 
2003 	/*
2004 	 * Forwarded interrupt queues.  Create this property to force the driver
2005 	 * to use forwarded interrupt queues.
2006 	 */
2007 	if (ddi_prop_exists(dev, dip, DDI_PROP_DONTPASS,
2008 	    "interrupt-forwarding") != 0 ||
2009 	    ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2010 	    "interrupt-forwarding") != 0) {
2011 		UNIMPLEMENTED();
2012 		(void) ddi_prop_create(dev, dip, DDI_PROP_CANSLEEP,
2013 		    "interrupt-forwarding", NULL, 0);
2014 	}
2015 
2016 	/*
2017 	 * Write combining
2018 	 * 0 to disable, 1 to enable
2019 	 */
2020 	p->wc = prop_lookup_int(sc, "write-combine", 1);
2021 	cxgb_printf(dip, CE_WARN, "write-combine: using of %d", p->wc);
2022 	if (p->wc != 0 && p->wc != 1) {
2023 		cxgb_printf(dip, CE_WARN,
2024 		    "write-combine: using 1 instead of %d", p->wc);
2025 		p->wc = 1;
2026 	}
2027 	(void) ddi_prop_update_int(dev, dip, "write-combine", p->wc);
2028 
2029 	p->t4_fw_install = prop_lookup_int(sc, "t4_fw_install", 1);
2030 	if (p->t4_fw_install != 0 && p->t4_fw_install != 2)
2031 		p->t4_fw_install = 1;
2032 	(void) ddi_prop_update_int(dev, dip, "t4_fw_install", p->t4_fw_install);
2033 
2034 	/* Multiple Rings */
2035 	p->multi_rings = prop_lookup_int(sc, "multi-rings", 1);
2036 	if (p->multi_rings != 0 && p->multi_rings != 1) {
2037 		cxgb_printf(dip, CE_NOTE,
2038 			   "multi-rings: using value 1 instead of %d", p->multi_rings);
2039 		p->multi_rings = 1;
2040 	}
2041 
2042 	(void) ddi_prop_update_int(dev, dip, "multi-rings", p->multi_rings);
2043 
2044 	return (0);
2045 }
2046 
2047 static int
2048 remove_extra_props(struct adapter *sc, int n10g, int n1g)
2049 {
2050 	if (n10g == 0) {
2051 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-ntxq-10G-port");
2052 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-nrxq-10G-port");
2053 		(void) ddi_prop_remove(sc->dev, sc->dip,
2054 		    "holdoff-timer-idx-10G");
2055 		(void) ddi_prop_remove(sc->dev, sc->dip,
2056 		    "holdoff-pktc-idx-10G");
2057 	}
2058 
2059 	if (n1g == 0) {
2060 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-ntxq-1G-port");
2061 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-nrxq-1G-port");
2062 		(void) ddi_prop_remove(sc->dev, sc->dip,
2063 		    "holdoff-timer-idx-1G");
2064 		(void) ddi_prop_remove(sc->dev, sc->dip, "holdoff-pktc-idx-1G");
2065 	}
2066 
2067 	return (0);
2068 }
2069 
2070 static int
2071 cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g,
2072     struct intrs_and_queues *iaq)
2073 {
2074 	struct driver_properties *p = &sc->props;
2075 	int rc, itype, itypes, navail, nc, nrxq10g, nrxq1g, n;
2076 	int nofldrxq10g = 0, nofldrxq1g = 0;
2077 
2078 	bzero(iaq, sizeof (*iaq));
2079 	nc = ncpus;	/* our snapshot of the number of CPUs */
2080 	iaq->ntxq10g = min(nc, p->max_ntxq_10g);
2081 	iaq->ntxq1g = min(nc, p->max_ntxq_1g);
2082 	iaq->nrxq10g = nrxq10g = min(nc, p->max_nrxq_10g);
2083 	iaq->nrxq1g = nrxq1g = min(nc, p->max_nrxq_1g);
2084 #ifdef TCP_OFFLOAD_ENABLE
2085 	iaq->nofldtxq10g = min(nc, p->max_nofldtxq_10g);
2086 	iaq->nofldtxq1g = min(nc, p->max_nofldtxq_1g);
2087 	iaq->nofldrxq10g = nofldrxq10g = min(nc, p->max_nofldrxq_10g);
2088 	iaq->nofldrxq1g = nofldrxq1g = min(nc, p->max_nofldrxq_1g);
2089 #endif
2090 
2091 	rc = ddi_intr_get_supported_types(sc->dip, &itypes);
2092 	if (rc != DDI_SUCCESS) {
2093 		cxgb_printf(sc->dip, CE_WARN,
2094 		    "failed to determine supported interrupt types: %d", rc);
2095 		return (rc);
2096 	}
2097 
2098 	for (itype = DDI_INTR_TYPE_MSIX; itype; itype >>= 1) {
2099 		ASSERT(itype == DDI_INTR_TYPE_MSIX ||
2100 		    itype == DDI_INTR_TYPE_MSI ||
2101 		    itype == DDI_INTR_TYPE_FIXED);
2102 
2103 		if ((itype & itypes & p->intr_types) == 0)
2104 			continue;	/* not supported or not allowed */
2105 
2106 		navail = 0;
2107 		rc = ddi_intr_get_navail(sc->dip, itype, &navail);
2108 		if (rc != DDI_SUCCESS || navail == 0) {
2109 			cxgb_printf(sc->dip, CE_WARN,
2110 			    "failed to get # of interrupts for type %d: %d",
2111 			    itype, rc);
2112 			continue;	/* carry on */
2113 		}
2114 
2115 		iaq->intr_type = itype;
2116 		if (navail == 0)
2117 			continue;
2118 
2119 		/*
2120 		 * Best option: an interrupt vector for errors, one for the
2121 		 * firmware event queue, and one each for each rxq (NIC as well
2122 		 * as offload).
2123 		 */
2124 		iaq->nirq = T4_EXTRA_INTR;
2125 		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2126 		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2127 
2128 		if (iaq->nirq <= navail &&
2129 		    (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq))) {
2130 			iaq->intr_fwd = 0;
2131 			goto allocate;
2132 		}
2133 
2134 		/*
2135 		 * Second best option: an interrupt vector for errors, one for
2136 		 * the firmware event queue, and one each for either NIC or
2137 		 * offload rxq's.
2138 		 */
2139 		iaq->nirq = T4_EXTRA_INTR;
2140 		iaq->nirq += n10g * max(nrxq10g, nofldrxq10g);
2141 		iaq->nirq += n1g * max(nrxq1g, nofldrxq1g);
2142 		if (iaq->nirq <= navail &&
2143 		    (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq))) {
2144 			iaq->intr_fwd = 1;
2145 			goto allocate;
2146 		}
2147 
2148 		/*
2149 		 * Next best option: an interrupt vector for errors, one for the
2150 		 * firmware event queue, and at least one per port.  At this
2151 		 * point we know we'll have to downsize nrxq or nofldrxq to fit
2152 		 * what's available to us.
2153 		 */
2154 		iaq->nirq = T4_EXTRA_INTR;
2155 		iaq->nirq += n10g + n1g;
2156 		if (iaq->nirq <= navail) {
2157 			int leftover = navail - iaq->nirq;
2158 
2159 			if (n10g > 0) {
2160 				int target = max(nrxq10g, nofldrxq10g);
2161 
2162 				n = 1;
2163 				while (n < target && leftover >= n10g) {
2164 					leftover -= n10g;
2165 					iaq->nirq += n10g;
2166 					n++;
2167 				}
2168 				iaq->nrxq10g = min(n, nrxq10g);
2169 #ifdef TCP_OFFLOAD_ENABLE
2170 				iaq->nofldrxq10g = min(n, nofldrxq10g);
2171 #endif
2172 			}
2173 
2174 			if (n1g > 0) {
2175 				int target = max(nrxq1g, nofldrxq1g);
2176 
2177 				n = 1;
2178 				while (n < target && leftover >= n1g) {
2179 					leftover -= n1g;
2180 					iaq->nirq += n1g;
2181 					n++;
2182 				}
2183 				iaq->nrxq1g = min(n, nrxq1g);
2184 #ifdef TCP_OFFLOAD_ENABLE
2185 				iaq->nofldrxq1g = min(n, nofldrxq1g);
2186 #endif
2187 			}
2188 
2189 			/* We have arrived at a minimum value required to enable
2190 			 * per queue irq(either NIC or offload). Thus for non-
2191 			 * offload case, we will get a vector per queue, while
2192 			 * offload case, we will get a vector per offload/NIC q.
2193 			 * Hence enable Interrupt forwarding only for offload
2194 			 * case.
2195 			 */
2196 #ifdef TCP_OFFLOAD_ENABLE
2197 			if (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq)) {
2198 				iaq->intr_fwd = 1;
2199 #else
2200 			if (itype != DDI_INTR_TYPE_MSI) {
2201 #endif
2202 				goto allocate;
2203 			}
2204 		}
2205 
2206 		/*
2207 		 * Least desirable option: one interrupt vector for everything.
2208 		 */
2209 		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2210 #ifdef TCP_OFFLOAD_ENABLE
2211 		iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2212 #endif
2213 		iaq->intr_fwd = 1;
2214 
2215 allocate:
2216 		return (0);
2217 	}
2218 
2219 	cxgb_printf(sc->dip, CE_WARN,
2220 	    "failed to find a usable interrupt type.  supported=%d, allowed=%d",
2221 	    itypes, p->intr_types);
2222 	return (DDI_FAILURE);
2223 }
2224 
2225 static int
2226 add_child_node(struct adapter *sc, int idx)
2227 {
2228 	int rc;
2229 	struct port_info *pi;
2230 
2231 	if (idx < 0 || idx >= sc->params.nports)
2232 		return (EINVAL);
2233 
2234 	pi = sc->port[idx];
2235 	if (pi == NULL)
2236 		return (ENODEV);	/* t4_port_init failed earlier */
2237 
2238 	PORT_LOCK(pi);
2239 	if (pi->dip != NULL) {
2240 		rc = 0;		/* EEXIST really, but then bus_config fails */
2241 		goto done;
2242 	}
2243 
2244 	rc = ndi_devi_alloc(sc->dip, T4_PORT_NAME, DEVI_SID_NODEID, &pi->dip);
2245 	if (rc != DDI_SUCCESS || pi->dip == NULL) {
2246 		rc = ENOMEM;
2247 		goto done;
2248 	}
2249 
2250 	(void) ddi_set_parent_data(pi->dip, pi);
2251 	(void) ndi_devi_bind_driver(pi->dip, 0);
2252 	rc = 0;
2253 done:
2254 	PORT_UNLOCK(pi);
2255 	return (rc);
2256 }
2257 
2258 static int
2259 remove_child_node(struct adapter *sc, int idx)
2260 {
2261 	int rc;
2262 	struct port_info *pi;
2263 
2264 	if (idx < 0 || idx >= sc->params.nports)
2265 		return (EINVAL);
2266 
2267 	pi = sc->port[idx];
2268 	if (pi == NULL)
2269 		return (ENODEV);
2270 
2271 	PORT_LOCK(pi);
2272 	if (pi->dip == NULL) {
2273 		rc = ENODEV;
2274 		goto done;
2275 	}
2276 
2277 	rc = ndi_devi_free(pi->dip);
2278 	if (rc == 0)
2279 		pi->dip = NULL;
2280 done:
2281 	PORT_UNLOCK(pi);
2282 	return (rc);
2283 }
2284 
2285 #define	KS_UINIT(x)	kstat_named_init(&kstatp->x, #x, KSTAT_DATA_ULONG)
2286 #define	KS_CINIT(x)	kstat_named_init(&kstatp->x, #x, KSTAT_DATA_CHAR)
2287 #define	KS_U_SET(x, y)	kstatp->x.value.ul = (y)
2288 #define	KS_C_SET(x, ...)	\
2289 			(void) snprintf(kstatp->x.value.c, 16,  __VA_ARGS__)
2290 
2291 /*
2292  * t4nex:X:config
2293  */
2294 struct t4_kstats {
2295 	kstat_named_t chip_ver;
2296 	kstat_named_t fw_vers;
2297 	kstat_named_t tp_vers;
2298 	kstat_named_t driver_version;
2299 	kstat_named_t serial_number;
2300 	kstat_named_t ec_level;
2301 	kstat_named_t id;
2302 	kstat_named_t bus_type;
2303 	kstat_named_t bus_width;
2304 	kstat_named_t bus_speed;
2305 	kstat_named_t core_clock;
2306 	kstat_named_t port_cnt;
2307 	kstat_named_t port_type;
2308 	kstat_named_t pci_vendor_id;
2309 	kstat_named_t pci_device_id;
2310 };
2311 static kstat_t *
2312 setup_kstats(struct adapter *sc)
2313 {
2314 	kstat_t *ksp;
2315 	struct t4_kstats *kstatp;
2316 	int ndata;
2317 	struct pci_params *p = &sc->params.pci;
2318 	struct vpd_params *v = &sc->params.vpd;
2319 	uint16_t pci_vendor, pci_device;
2320 
2321 	ndata = sizeof (struct t4_kstats) / sizeof (kstat_named_t);
2322 
2323 	ksp = kstat_create(T4_NEXUS_NAME, ddi_get_instance(sc->dip), "config",
2324 	    "nexus", KSTAT_TYPE_NAMED, ndata, 0);
2325 	if (ksp == NULL) {
2326 		cxgb_printf(sc->dip, CE_WARN, "failed to initialize kstats.");
2327 		return (NULL);
2328 	}
2329 
2330 	kstatp = (struct t4_kstats *)ksp->ks_data;
2331 
2332 	KS_UINIT(chip_ver);
2333 	KS_CINIT(fw_vers);
2334 	KS_CINIT(tp_vers);
2335 	KS_CINIT(driver_version);
2336 	KS_CINIT(serial_number);
2337 	KS_CINIT(ec_level);
2338 	KS_CINIT(id);
2339 	KS_CINIT(bus_type);
2340 	KS_CINIT(bus_width);
2341 	KS_CINIT(bus_speed);
2342 	KS_UINIT(core_clock);
2343 	KS_UINIT(port_cnt);
2344 	KS_CINIT(port_type);
2345 	KS_CINIT(pci_vendor_id);
2346 	KS_CINIT(pci_device_id);
2347 
2348 	KS_U_SET(chip_ver, sc->params.chip);
2349 	KS_C_SET(fw_vers, "%d.%d.%d.%d",
2350 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
2351 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
2352 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
2353 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
2354 	KS_C_SET(tp_vers, "%d.%d.%d.%d",
2355 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
2356 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
2357 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
2358 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
2359 	KS_C_SET(driver_version, DRV_VERSION);
2360 	KS_C_SET(serial_number, "%s", v->sn);
2361 	KS_C_SET(ec_level, "%s", v->ec);
2362 	KS_C_SET(id, "%s", v->id);
2363 	KS_C_SET(bus_type, "pci-express");
2364 	KS_C_SET(bus_width, "x%d lanes", p->width);
2365 	KS_C_SET(bus_speed, "%d", p->speed);
2366 	KS_U_SET(core_clock, v->cclk);
2367 	KS_U_SET(port_cnt, sc->params.nports);
2368 
2369 	t4_os_pci_read_cfg2(sc, PCI_CONF_VENID, &pci_vendor);
2370 	KS_C_SET(pci_vendor_id, "0x%x", pci_vendor);
2371 
2372 	t4_os_pci_read_cfg2(sc, PCI_CONF_DEVID, &pci_device);
2373 	KS_C_SET(pci_device_id, "0x%x", pci_device);
2374 
2375 	KS_C_SET(port_type, "%s/%s/%s/%s",
2376 		 print_port_speed(sc->port[0]),
2377 		 print_port_speed(sc->port[1]),
2378 		 print_port_speed(sc->port[2]),
2379 		 print_port_speed(sc->port[3]));
2380 
2381 	/* Do NOT set ksp->ks_update.  These kstats do not change. */
2382 
2383 	/* Install the kstat */
2384 	ksp->ks_private = (void *)sc;
2385 	kstat_install(ksp);
2386 
2387 	return (ksp);
2388 }
2389 
2390 /*
2391  * t4nex:X:stat
2392  */
2393 struct t4_wc_kstats {
2394 	kstat_named_t write_coal_success;
2395 	kstat_named_t write_coal_failure;
2396 };
2397 static kstat_t *
2398 setup_wc_kstats(struct adapter *sc)
2399 {
2400 	kstat_t *ksp;
2401 	struct t4_wc_kstats *kstatp;
2402 	int ndata;
2403 
2404 	ndata = sizeof(struct t4_wc_kstats) / sizeof(kstat_named_t);
2405 	ksp = kstat_create(T4_NEXUS_NAME, ddi_get_instance(sc->dip), "stats",
2406 	    "nexus", KSTAT_TYPE_NAMED, ndata, 0);
2407 	if (ksp == NULL) {
2408 		cxgb_printf(sc->dip, CE_WARN, "failed to initialize kstats.");
2409 		return (NULL);
2410 	}
2411 
2412 	kstatp = (struct t4_wc_kstats *)ksp->ks_data;
2413 
2414 	KS_UINIT(write_coal_success);
2415 	KS_UINIT(write_coal_failure);
2416 
2417 	ksp->ks_update = update_wc_kstats;
2418 	/* Install the kstat */
2419 	ksp->ks_private = (void *)sc;
2420 	kstat_install(ksp);
2421 
2422 	return (ksp);
2423 }
2424 
2425 static int
2426 update_wc_kstats(kstat_t *ksp, int rw)
2427 {
2428 	struct t4_wc_kstats *kstatp = (struct t4_wc_kstats *)ksp->ks_data;
2429 	struct adapter *sc = ksp->ks_private;
2430 	uint32_t wc_total, wc_success, wc_failure;
2431 
2432 	if (rw == KSTAT_WRITE)
2433 		return (0);
2434 
2435 	if (is_t5(sc->params.chip)) {
2436 		wc_total = t4_read_reg(sc, A_SGE_STAT_TOTAL);
2437 		wc_failure = t4_read_reg(sc, A_SGE_STAT_MATCH);
2438 		wc_success = wc_total - wc_failure;
2439 	} else {
2440 		wc_success = 0;
2441 		wc_failure = 0;
2442 	}
2443 
2444 	KS_U_SET(write_coal_success, wc_success);
2445 	KS_U_SET(write_coal_failure, wc_failure);
2446 
2447 	return (0);
2448 }
2449 
2450 int
2451 adapter_full_init(struct adapter *sc)
2452 {
2453 	int i, rc = 0;
2454 
2455 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2456 
2457 	rc = t4_setup_adapter_queues(sc);
2458 	if (rc != 0)
2459 		goto done;
2460 
2461 	if (sc->intr_cap & DDI_INTR_FLAG_BLOCK)
2462 		(void) ddi_intr_block_enable(sc->intr_handle, sc->intr_count);
2463 	else {
2464 		for (i = 0; i < sc->intr_count; i++)
2465 			(void) ddi_intr_enable(sc->intr_handle[i]);
2466 	}
2467 	t4_intr_enable(sc);
2468 	sc->flags |= FULL_INIT_DONE;
2469 
2470 #ifdef TCP_OFFLOAD_ENABLE
2471 	/* TODO: wrong place to enable TOE capability */
2472 	if (is_offload(sc) != 0) {
2473 		for_each_port(sc, i) {
2474 			struct port_info *pi = sc->port[i];
2475 			rc = toe_capability(pi, 1);
2476 			if (rc != 0) {
2477 				cxgb_printf(pi->dip, CE_WARN,
2478 				    "Failed to activate toe capability: %d",
2479 				    rc);
2480 				rc = 0;		/* not a fatal error */
2481 			}
2482 		}
2483 	}
2484 #endif
2485 
2486 done:
2487 	if (rc != 0)
2488 		(void) adapter_full_uninit(sc);
2489 
2490 	return (rc);
2491 }
2492 
2493 int
2494 adapter_full_uninit(struct adapter *sc)
2495 {
2496 	int i, rc = 0;
2497 
2498 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2499 
2500 	if (sc->intr_cap & DDI_INTR_FLAG_BLOCK)
2501 		(void) ddi_intr_block_disable(sc->intr_handle, sc->intr_count);
2502 	else {
2503 		for (i = 0; i < sc->intr_count; i++)
2504 			(void) ddi_intr_disable(sc->intr_handle[i]);
2505 	}
2506 
2507 	rc = t4_teardown_adapter_queues(sc);
2508 	if (rc != 0)
2509 		return (rc);
2510 
2511 	sc->flags &= ~FULL_INIT_DONE;
2512 
2513 	return (0);
2514 }
2515 
2516 int
2517 port_full_init(struct port_info *pi)
2518 {
2519 	struct adapter *sc = pi->adapter;
2520 	uint16_t *rss;
2521 	struct sge_rxq *rxq;
2522 	int rc, i;
2523 
2524 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2525 	ASSERT((pi->flags & PORT_INIT_DONE) == 0);
2526 
2527 	/*
2528 	 * Allocate tx/rx/fl queues for this port.
2529 	 */
2530 	rc = t4_setup_port_queues(pi);
2531 	if (rc != 0)
2532 		goto done;	/* error message displayed already */
2533 
2534 	/*
2535 	 * Setup RSS for this port.
2536 	 */
2537 	rss = kmem_zalloc(pi->nrxq * sizeof (*rss), KM_SLEEP);
2538 	for_each_rxq(pi, i, rxq) {
2539 		rss[i] = rxq->iq.abs_id;
2540 	}
2541 	rc = -t4_config_rss_range(sc, sc->mbox, pi->viid, 0,
2542 	    pi->rss_size, rss, pi->nrxq);
2543 	kmem_free(rss, pi->nrxq * sizeof (*rss));
2544 	if (rc != 0) {
2545 		cxgb_printf(pi->dip, CE_WARN, "rss_config failed: %d", rc);
2546 		goto done;
2547 	}
2548 
2549 	pi->flags |= PORT_INIT_DONE;
2550 done:
2551 	if (rc != 0)
2552 		(void) port_full_uninit(pi);
2553 
2554 	return (rc);
2555 }
2556 
2557 /*
2558  * Idempotent.
2559  */
2560 int
2561 port_full_uninit(struct port_info *pi)
2562 {
2563 
2564 	ASSERT(pi->flags & PORT_INIT_DONE);
2565 
2566 	(void) t4_teardown_port_queues(pi);
2567 	pi->flags &= ~PORT_INIT_DONE;
2568 
2569 	return (0);
2570 }
2571 
2572 void
2573 enable_port_queues(struct port_info *pi)
2574 {
2575 	struct adapter *sc = pi->adapter;
2576 	int i;
2577 	struct sge_iq *iq;
2578 	struct sge_rxq *rxq;
2579 #ifdef TCP_OFFLOAD_ENABLE
2580 	struct sge_ofld_rxq *ofld_rxq;
2581 #endif
2582 
2583 	ASSERT(pi->flags & PORT_INIT_DONE);
2584 
2585 	/*
2586 	 * TODO: whatever was queued up after we set iq->state to IQS_DISABLED
2587 	 * back in disable_port_queues will be processed now, after an unbounded
2588 	 * delay.  This can't be good.
2589 	 */
2590 
2591 #ifdef TCP_OFFLOAD_ENABLE
2592 	for_each_ofld_rxq(pi, i, ofld_rxq) {
2593 		iq = &ofld_rxq->iq;
2594 		if (atomic_cas_uint(&iq->state, IQS_DISABLED, IQS_IDLE) !=
2595 		    IQS_DISABLED)
2596 			panic("%s: iq %p wasn't disabled", __func__,
2597 			    (void *)iq);
2598 		t4_write_reg(sc, MYPF_REG(A_SGE_PF_GTS),
2599 		    V_SEINTARM(iq->intr_params) | V_INGRESSQID(iq->cntxt_id));
2600 	}
2601 #endif
2602 
2603 	for_each_rxq(pi, i, rxq) {
2604 		iq = &rxq->iq;
2605 		if (atomic_cas_uint(&iq->state, IQS_DISABLED, IQS_IDLE) !=
2606 		    IQS_DISABLED)
2607 			panic("%s: iq %p wasn't disabled", __func__,
2608 			    (void *) iq);
2609 		t4_write_reg(sc, MYPF_REG(A_SGE_PF_GTS),
2610 		    V_SEINTARM(iq->intr_params) | V_INGRESSQID(iq->cntxt_id));
2611 	}
2612 }
2613 
2614 void
2615 disable_port_queues(struct port_info *pi)
2616 {
2617 	int i;
2618 	struct adapter *sc = pi->adapter;
2619 	struct sge_rxq *rxq;
2620 #ifdef TCP_OFFLOAD_ENABLE
2621 	struct sge_ofld_rxq *ofld_rxq;
2622 #endif
2623 
2624 	ASSERT(pi->flags & PORT_INIT_DONE);
2625 
2626 	/*
2627 	 * TODO: need proper implementation for all tx queues (ctrl, eth, ofld).
2628 	 */
2629 
2630 #ifdef TCP_OFFLOAD_ENABLE
2631 	for_each_ofld_rxq(pi, i, ofld_rxq) {
2632 		while (atomic_cas_uint(&ofld_rxq->iq.state, IQS_IDLE,
2633 		    IQS_DISABLED) != IQS_IDLE)
2634 			msleep(1);
2635 	}
2636 #endif
2637 
2638 	for_each_rxq(pi, i, rxq) {
2639 		while (atomic_cas_uint(&rxq->iq.state, IQS_IDLE,
2640 		    IQS_DISABLED) != IQS_IDLE)
2641 			msleep(1);
2642 	}
2643 
2644 	mutex_enter(&sc->sfl_lock);
2645 #ifdef TCP_OFFLOAD_ENABLE
2646 	for_each_ofld_rxq(pi, i, ofld_rxq)
2647 	    ofld_rxq->fl.flags |= FL_DOOMED;
2648 #endif
2649 	for_each_rxq(pi, i, rxq)
2650 	    rxq->fl.flags |= FL_DOOMED;
2651 	mutex_exit(&sc->sfl_lock);
2652 	/* TODO: need to wait for all fl's to be removed from sc->sfl */
2653 }
2654 
2655 void
2656 t4_fatal_err(struct adapter *sc)
2657 {
2658 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2659 	t4_intr_disable(sc);
2660 	cxgb_printf(sc->dip, CE_WARN,
2661 	    "encountered fatal error, adapter stopped.");
2662 }
2663 
2664 int
2665 t4_os_find_pci_capability(struct adapter *sc, int cap)
2666 {
2667 	uint16_t stat;
2668 	uint8_t cap_ptr, cap_id;
2669 
2670 	t4_os_pci_read_cfg2(sc, PCI_CONF_STAT, &stat);
2671 	if ((stat & PCI_STAT_CAP) == 0)
2672 		return (0); /* does not implement capabilities */
2673 
2674 	t4_os_pci_read_cfg1(sc, PCI_CONF_CAP_PTR, &cap_ptr);
2675 	while (cap_ptr) {
2676 		t4_os_pci_read_cfg1(sc, cap_ptr + PCI_CAP_ID, &cap_id);
2677 		if (cap_id == cap)
2678 			return (cap_ptr); /* found */
2679 		t4_os_pci_read_cfg1(sc, cap_ptr + PCI_CAP_NEXT_PTR, &cap_ptr);
2680 	}
2681 
2682 	return (0); /* not found */
2683 }
2684 
2685 void
2686 t4_os_portmod_changed(const struct adapter *sc, int idx)
2687 {
2688 	static const char *mod_str[] = {
2689 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
2690 	};
2691 	const struct port_info *pi = sc->port[idx];
2692 
2693 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
2694 		cxgb_printf(pi->dip, CE_NOTE, "transceiver unplugged.");
2695 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
2696 		cxgb_printf(pi->dip, CE_NOTE,
2697 		    "unknown transceiver inserted.\n");
2698 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
2699 		cxgb_printf(pi->dip, CE_NOTE,
2700 		    "unsupported transceiver inserted.\n");
2701 	else if (pi->mod_type > 0 && pi->mod_type < ARRAY_SIZE(mod_str))
2702 		cxgb_printf(pi->dip, CE_NOTE, "%s transceiver inserted.\n",
2703 		    mod_str[pi->mod_type]);
2704 	else
2705 		cxgb_printf(pi->dip, CE_NOTE, "transceiver (type %d) inserted.",
2706 		    pi->mod_type);
2707 }
2708 
2709 /* ARGSUSED */
2710 static int
2711 cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss, mblk_t *m)
2712 {
2713 	if (m != NULL)
2714 		freemsg(m);
2715 	return (0);
2716 }
2717 
2718 int
2719 t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
2720 {
2721 	uint_t *loc, new;
2722 
2723 	if (opcode >= ARRAY_SIZE(sc->cpl_handler))
2724 		return (EINVAL);
2725 
2726 	new = (uint_t)(unsigned long) (h ? h : cpl_not_handled);
2727 	loc = (uint_t *)&sc->cpl_handler[opcode];
2728 	(void) atomic_swap_uint(loc, new);
2729 
2730 	return (0);
2731 }
2732 
2733 static int
2734 fw_msg_not_handled(struct adapter *sc, const __be64 *data)
2735 {
2736 	struct cpl_fw6_msg *cpl;
2737 
2738 	cpl = __containerof((void *)data, struct cpl_fw6_msg, data);
2739 
2740 	cxgb_printf(sc->dip, CE_WARN, "%s fw_msg type %d", __func__, cpl->type);
2741 	return (0);
2742 }
2743 
2744 int
2745 t4_register_fw_msg_handler(struct adapter *sc, int type, fw_msg_handler_t h)
2746 {
2747 	fw_msg_handler_t *loc, new;
2748 
2749 	if (type >= ARRAY_SIZE(sc->fw_msg_handler))
2750 		return (EINVAL);
2751 
2752 	/*
2753 	 * These are dispatched by the handler for FW{4|6}_CPL_MSG using the CPL
2754 	 * handler dispatch table.  Reject any attempt to install a handler for
2755 	 * this subtype.
2756 	 */
2757 	if (type == FW_TYPE_RSSCPL || type == FW6_TYPE_RSSCPL)
2758 		return (EINVAL);
2759 
2760 	new = h ? h : fw_msg_not_handled;
2761 	loc = &sc->fw_msg_handler[type];
2762 	(void)atomic_swap_ptr(loc, (void *)new);
2763 
2764 	return (0);
2765 }
2766 
2767 #ifdef TCP_OFFLOAD_ENABLE
2768 static int
2769 toe_capability(struct port_info *pi, int enable)
2770 {
2771 	int rc;
2772 	struct adapter *sc = pi->adapter;
2773 
2774 	if (!is_offload(sc))
2775 		return (ENODEV);
2776 
2777 	if (enable != 0) {
2778 		if (isset(&sc->offload_map, pi->port_id) != 0)
2779 			return (0);
2780 
2781 		if (sc->offload_map == 0) {
2782 			rc = activate_uld(sc, ULD_TOM, &sc->tom);
2783 			if (rc != 0)
2784 				return (rc);
2785 		}
2786 
2787 		setbit(&sc->offload_map, pi->port_id);
2788 	} else {
2789 		if (!isset(&sc->offload_map, pi->port_id))
2790 			return (0);
2791 
2792 		clrbit(&sc->offload_map, pi->port_id);
2793 
2794 		if (sc->offload_map == 0) {
2795 			rc = deactivate_uld(&sc->tom);
2796 			if (rc != 0) {
2797 				setbit(&sc->offload_map, pi->port_id);
2798 				return (rc);
2799 			}
2800 		}
2801 	}
2802 
2803 	return (0);
2804 }
2805 
2806 /*
2807  * Add an upper layer driver to the global list.
2808  */
2809 int
2810 t4_register_uld(struct uld_info *ui)
2811 {
2812 	int rc = 0;
2813 	struct uld_info *u;
2814 
2815 	mutex_enter(&t4_uld_list_lock);
2816 	SLIST_FOREACH(u, &t4_uld_list, link) {
2817 		if (u->uld_id == ui->uld_id) {
2818 			rc = EEXIST;
2819 			goto done;
2820 		}
2821 	}
2822 
2823 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
2824 	ui->refcount = 0;
2825 done:
2826 	mutex_exit(&t4_uld_list_lock);
2827 	return (rc);
2828 }
2829 
2830 int
2831 t4_unregister_uld(struct uld_info *ui)
2832 {
2833 	int rc = EINVAL;
2834 	struct uld_info *u;
2835 
2836 	mutex_enter(&t4_uld_list_lock);
2837 
2838 	SLIST_FOREACH(u, &t4_uld_list, link) {
2839 		if (u == ui) {
2840 			if (ui->refcount > 0) {
2841 				rc = EBUSY;
2842 				goto done;
2843 			}
2844 
2845 			SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
2846 			rc = 0;
2847 			goto done;
2848 		}
2849 	}
2850 done:
2851 	mutex_exit(&t4_uld_list_lock);
2852 	return (rc);
2853 }
2854 
2855 static int
2856 activate_uld(struct adapter *sc, int id, struct uld_softc *usc)
2857 {
2858 	int rc = EAGAIN;
2859 	struct uld_info *ui;
2860 
2861 	mutex_enter(&t4_uld_list_lock);
2862 
2863 	SLIST_FOREACH(ui, &t4_uld_list, link) {
2864 		if (ui->uld_id == id) {
2865 			rc = ui->attach(sc, &usc->softc);
2866 			if (rc == 0) {
2867 				ASSERT(usc->softc != NULL);
2868 				ui->refcount++;
2869 				usc->uld = ui;
2870 			}
2871 			goto done;
2872 		}
2873 	}
2874 done:
2875 	mutex_exit(&t4_uld_list_lock);
2876 
2877 	return (rc);
2878 }
2879 
2880 static int
2881 deactivate_uld(struct uld_softc *usc)
2882 {
2883 	int rc;
2884 
2885 	mutex_enter(&t4_uld_list_lock);
2886 
2887 	if (usc->uld == NULL || usc->softc == NULL) {
2888 		rc = EINVAL;
2889 		goto done;
2890 	}
2891 
2892 	rc = usc->uld->detach(usc->softc);
2893 	if (rc == 0) {
2894 		ASSERT(usc->uld->refcount > 0);
2895 		usc->uld->refcount--;
2896 		usc->uld = NULL;
2897 		usc->softc = NULL;
2898 	}
2899 done:
2900 	mutex_exit(&t4_uld_list_lock);
2901 
2902 	return (rc);
2903 }
2904 
2905 void
2906 t4_iterate(void (*func)(int, void *), void *arg)
2907 {
2908 	struct adapter *sc;
2909 
2910 	mutex_enter(&t4_adapter_list_lock);
2911 	SLIST_FOREACH(sc, &t4_adapter_list, link) {
2912 		/*
2913 		 * func should not make any assumptions about what state sc is
2914 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
2915 		 */
2916 		func(ddi_get_instance(sc->dip), arg);
2917 	}
2918 	mutex_exit(&t4_adapter_list_lock);
2919 }
2920 
2921 #endif
2922