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