xref: /illumos-gate/usr/src/uts/common/io/cxgbe/t4nex/t4_nexus.c (revision 8c69cc8fbe729fa7b091e901c4b50508ccc6bb33)
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 	/*
373 	 * Prepare the adapter for operation.
374 	 */
375 	rc = -t4_prep_adapter(sc, false);
376 	if (rc != 0) {
377 		cxgb_printf(dip, CE_WARN, "failed to prepare adapter: %d", rc);
378 		goto done;
379 	}
380 
381 	/*
382 	 * Enable BAR1 access.
383 	 */
384 	sc->doorbells |= DOORBELL_KDB;
385 	rc = ddi_regs_map_setup(dip, 2, &sc->reg1p, 0, 0, &da1, &sc->reg1h);
386 	if (rc != DDI_SUCCESS) {
387 		cxgb_printf(dip, CE_WARN,
388 		    "failed to map BAR1 device registers: %d", rc);
389 		goto done;
390 	} else {
391 		if (is_t5(sc->params.chip)) {
392 			sc->doorbells |= DOORBELL_UDB;
393 			if (prp->wc) {
394 				/*
395 				 * Enable write combining on BAR2.  This is the
396 				 * userspace doorbell BAR and is split into 128B
397 				 * (UDBS_SEG_SIZE) doorbell regions, each associated
398 				 * with an egress queue.  The first 64B has the doorbell
399 				 * and the second 64B can be used to submit a tx work
400 				 * request with an implicit doorbell.
401 				 */
402 				sc->doorbells &= ~DOORBELL_UDB;
403 				sc->doorbells |= (DOORBELL_WCWR |
404 				    DOORBELL_UDBWC);
405 				t4_write_reg(sc, A_SGE_STAT_CFG,
406 				    V_STATSOURCE_T5(7) | V_STATMODE(0));
407 			}
408 		}
409 	}
410 
411 	/*
412 	 * Do this really early.  Note that minor number = instance.
413 	 */
414 	(void) snprintf(name, sizeof (name), "%s,%d", T4_NEXUS_NAME, instance);
415 	rc = ddi_create_minor_node(dip, name, S_IFCHR, instance,
416 	    DDI_NT_NEXUS, 0);
417 	if (rc != DDI_SUCCESS) {
418 		cxgb_printf(dip, CE_WARN,
419 		    "failed to create device node: %d", rc);
420 		rc = DDI_SUCCESS; /* carry on */
421 	}
422 
423 	/* Do this early. Memory window is required for loading config file. */
424 	setup_memwin(sc);
425 
426 	/* Prepare the firmware for operation */
427 	rc = prep_firmware(sc);
428 	if (rc != 0)
429 		goto done; /* error message displayed already */
430 
431 	rc = adap__pre_init_tweaks(sc);
432 	if (rc != 0)
433 		goto done;
434 
435 	rc = get_params__pre_init(sc);
436 	if (rc != 0)
437 		goto done; /* error message displayed already */
438 
439 	t4_sge_init(sc);
440 
441 	if (sc->flags & MASTER_PF) {
442 		/* get basic stuff going */
443 		rc = -t4_fw_initialize(sc, sc->mbox);
444 		if (rc != 0) {
445 			cxgb_printf(sc->dip, CE_WARN,
446 			    "early init failed: %d.\n", rc);
447 			goto done;
448 		}
449 	}
450 
451 	rc = get_params__post_init(sc);
452 	if (rc != 0)
453 		goto done; /* error message displayed already */
454 
455 	rc = set_params__post_init(sc);
456 	if (rc != 0)
457 		goto done; /* error message displayed already */
458 
459 	/*
460 	 * TODO: This is the place to call t4_set_filter_mode()
461 	 */
462 
463 	/* tweak some settings */
464 	t4_write_reg(sc, A_TP_SHIFT_CNT, V_SYNSHIFTMAX(6) | V_RXTSHIFTMAXR1(4) |
465 	    V_RXTSHIFTMAXR2(15) | V_PERSHIFTBACKOFFMAX(8) | V_PERSHIFTMAX(8) |
466 	    V_KEEPALIVEMAXR1(4) | V_KEEPALIVEMAXR2(9));
467 	t4_write_reg(sc, A_ULP_RX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
468 
469 	/*
470 	 * Work-around for bug 2619
471 	 * Set DisableVlan field in TP_RSS_CONFIG_VRT register so that the
472 	 * VLAN tag extraction is disabled.
473 	 */
474 	t4_set_reg_field(sc, A_TP_RSS_CONFIG_VRT, F_DISABLEVLAN, F_DISABLEVLAN);
475 
476 	/* Store filter mode */
477 	t4_read_indirect(sc, A_TP_PIO_ADDR, A_TP_PIO_DATA, &sc->filter_mode, 1,
478 	    A_TP_VLAN_PRI_MAP);
479 
480 	/*
481 	 * First pass over all the ports - allocate VIs and initialize some
482 	 * basic parameters like mac address, port type, etc.  We also figure
483 	 * out whether a port is 10G or 1G and use that information when
484 	 * calculating how many interrupts to attempt to allocate.
485 	 */
486 	n100g = n40g = n25g = n10g = n1g = 0;
487 	for_each_port(sc, i) {
488 		struct port_info *pi;
489 
490 		pi = kmem_zalloc(sizeof (*pi), KM_SLEEP);
491 		sc->port[i] = pi;
492 
493 		/* These must be set before t4_port_init */
494 		pi->adapter = sc;
495 		/* LINTED: E_ASSIGN_NARROW_CONV */
496 		pi->port_id = i;
497 	}
498 
499 	/* Allocate the vi and initialize parameters like mac addr */
500 	rc = -t4_port_init(sc, sc->mbox, sc->pf, 0);
501 	if (rc) {
502 		cxgb_printf(dip, CE_WARN,
503 			    "unable to initialize port: %d", rc);
504 		goto done;
505 	}
506 
507 	for_each_port(sc, i) {
508 		struct port_info *pi = sc->port[i];
509 
510 		mutex_init(&pi->lock, NULL, MUTEX_DRIVER, NULL);
511 		pi->mtu = ETHERMTU;
512 
513 		if (is_100G_port(pi)) {
514 			n100g++;
515 			pi->tmr_idx = prp->tmr_idx_10g;
516 			pi->pktc_idx = prp->pktc_idx_10g;
517 		} else if (is_40G_port(pi)) {
518 			n40g++;
519 			pi->tmr_idx = prp->tmr_idx_10g;
520 			pi->pktc_idx = prp->pktc_idx_10g;
521 		} else if (is_25G_port(pi)) {
522 			n25g++;
523 			pi->tmr_idx = prp->tmr_idx_10g;
524 			pi->pktc_idx = prp->pktc_idx_10g;
525 		} else if (is_10G_port(pi)) {
526 			n10g++;
527 			pi->tmr_idx = prp->tmr_idx_10g;
528 			pi->pktc_idx = prp->pktc_idx_10g;
529 		} else {
530 			n1g++;
531 			pi->tmr_idx = prp->tmr_idx_1g;
532 			pi->pktc_idx = prp->pktc_idx_1g;
533 		}
534 
535 		pi->xact_addr_filt = -1;
536 		t4_mc_init(pi);
537 
538 		setbit(&sc->registered_device_map, i);
539 	}
540 
541 	nxg = n10g + n25g + n40g + n100g;
542 	(void) remove_extra_props(sc, nxg, n1g);
543 
544 	if (sc->registered_device_map == 0) {
545 		cxgb_printf(dip, CE_WARN, "no usable ports");
546 		rc = DDI_FAILURE;
547 		goto done;
548 	}
549 
550 	rc = cfg_itype_and_nqueues(sc, nxg, n1g, &iaq);
551 	if (rc != 0)
552 		goto done; /* error message displayed already */
553 
554 	sc->intr_type = iaq.intr_type;
555 	sc->intr_count = iaq.nirq;
556 
557 	if (sc->props.multi_rings && (sc->intr_type != DDI_INTR_TYPE_MSIX)) {
558 		sc->props.multi_rings = 0;
559 		cxgb_printf(dip, CE_WARN,
560 		    "Multiple rings disabled as interrupt type is not MSI-X");
561 	}
562 
563 	if (sc->props.multi_rings && iaq.intr_fwd) {
564 		sc->props.multi_rings = 0;
565 		cxgb_printf(dip, CE_WARN,
566 		    "Multiple rings disabled as interrupts are forwarded");
567 	}
568 
569 	if (!sc->props.multi_rings) {
570 		iaq.ntxq10g = 1;
571 		iaq.ntxq1g = 1;
572 	}
573 	s = &sc->sge;
574 	s->nrxq = nxg * iaq.nrxq10g + n1g * iaq.nrxq1g;
575 	s->ntxq = nxg * iaq.ntxq10g + n1g * iaq.ntxq1g;
576 	s->neq = s->ntxq + s->nrxq;	/* the fl in an rxq is an eq */
577 #ifdef TCP_OFFLOAD_ENABLE
578 	/* control queues, 1 per port + 1 mgmtq */
579 	s->neq += sc->params.nports + 1;
580 #endif
581 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
582 	if (iaq.intr_fwd != 0)
583 		sc->flags |= INTR_FWD;
584 #ifdef TCP_OFFLOAD_ENABLE
585 	if (is_offload(sc) != 0) {
586 
587 		s->nofldrxq = nxg * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
588 		s->nofldtxq = nxg * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
589 		s->neq += s->nofldtxq + s->nofldrxq;
590 		s->niq += s->nofldrxq;
591 
592 		s->ofld_rxq = kmem_zalloc(s->nofldrxq *
593 		    sizeof (struct sge_ofld_rxq), KM_SLEEP);
594 		s->ofld_txq = kmem_zalloc(s->nofldtxq *
595 		    sizeof (struct sge_wrq), KM_SLEEP);
596 		s->ctrlq = kmem_zalloc(sc->params.nports *
597 		    sizeof (struct sge_wrq), KM_SLEEP);
598 
599 	}
600 #endif
601 	s->rxq = kmem_zalloc(s->nrxq * sizeof (struct sge_rxq), KM_SLEEP);
602 	s->txq = kmem_zalloc(s->ntxq * sizeof (struct sge_txq), KM_SLEEP);
603 	s->iqmap = kmem_zalloc(s->niq * sizeof (struct sge_iq *), KM_SLEEP);
604 	s->eqmap = kmem_zalloc(s->neq * sizeof (struct sge_eq *), KM_SLEEP);
605 
606 	sc->intr_handle = kmem_zalloc(sc->intr_count *
607 	    sizeof (ddi_intr_handle_t), KM_SLEEP);
608 
609 	/*
610 	 * Second pass over the ports.  This time we know the number of rx and
611 	 * tx queues that each port should get.
612 	 */
613 	rqidx = tqidx = 0;
614 #ifdef TCP_OFFLOAD_ENABLE
615 	ofld_rqidx = ofld_tqidx = 0;
616 #endif
617 	for_each_port(sc, i) {
618 		struct port_info *pi = sc->port[i];
619 
620 		if (pi == NULL)
621 			continue;
622 
623 		t4_mc_cb_init(pi);
624 		/* LINTED: E_ASSIGN_NARROW_CONV */
625 		pi->first_rxq = rqidx;
626 		/* LINTED: E_ASSIGN_NARROW_CONV */
627 		pi->nrxq = (is_10XG_port(pi)) ? iaq.nrxq10g
628 		    : iaq.nrxq1g;
629 		/* LINTED: E_ASSIGN_NARROW_CONV */
630 		pi->first_txq = tqidx;
631 		/* LINTED: E_ASSIGN_NARROW_CONV */
632 		pi->ntxq = (is_10XG_port(pi)) ? iaq.ntxq10g
633 		    : iaq.ntxq1g;
634 
635 		rqidx += pi->nrxq;
636 		tqidx += pi->ntxq;
637 
638 #ifdef TCP_OFFLOAD_ENABLE
639 		if (is_offload(sc) != 0) {
640 			/* LINTED: E_ASSIGN_NARROW_CONV */
641 			pi->first_ofld_rxq = ofld_rqidx;
642 			pi->nofldrxq = max(1, pi->nrxq / 4);
643 
644 			/* LINTED: E_ASSIGN_NARROW_CONV */
645 			pi->first_ofld_txq = ofld_tqidx;
646 			pi->nofldtxq = max(1, pi->ntxq / 2);
647 
648 			ofld_rqidx += pi->nofldrxq;
649 			ofld_tqidx += pi->nofldtxq;
650 		}
651 #endif
652 
653 		/*
654 		 * Enable hw checksumming and LSO for all ports by default.
655 		 * They can be disabled using ndd (hw_csum and hw_lso).
656 		 */
657 		pi->features |= (CXGBE_HW_CSUM | CXGBE_HW_LSO);
658 	}
659 
660 #ifdef TCP_OFFLOAD_ENABLE
661 		sc->l2t = t4_init_l2t(sc);
662 #endif
663 
664 	/*
665 	 * Setup Interrupts.
666 	 */
667 
668 	i = 0;
669 	rc = ddi_intr_alloc(dip, sc->intr_handle, sc->intr_type, 0,
670 	    sc->intr_count, &i, DDI_INTR_ALLOC_STRICT);
671 	if (rc != DDI_SUCCESS) {
672 		cxgb_printf(dip, CE_WARN,
673 		    "failed to allocate %d interrupt(s) of type %d: %d, %d",
674 		    sc->intr_count, sc->intr_type, rc, i);
675 		goto done;
676 	}
677 	ASSERT(sc->intr_count == i); /* allocation was STRICT */
678 	(void) ddi_intr_get_cap(sc->intr_handle[0], &sc->intr_cap);
679 	(void) ddi_intr_get_pri(sc->intr_handle[0], &sc->intr_pri);
680 	if (sc->intr_count == 1) {
681 		ASSERT(sc->flags & INTR_FWD);
682 		(void) ddi_intr_add_handler(sc->intr_handle[0], t4_intr_all, sc,
683 		    &s->fwq);
684 	} else {
685 		/* Multiple interrupts.  The first one is always error intr */
686 		(void) ddi_intr_add_handler(sc->intr_handle[0], t4_intr_err, sc,
687 		    NULL);
688 		irq++;
689 
690 		/* The second one is always the firmware event queue */
691 		(void) ddi_intr_add_handler(sc->intr_handle[1], t4_intr, sc,
692 		    &s->fwq);
693 		irq++;
694 		/*
695 		 * Note that if INTR_FWD is set then either the NIC rx
696 		 * queues or (exclusive or) the TOE rx queueus will be taking
697 		 * direct interrupts.
698 		 *
699 		 * There is no need to check for is_offload(sc) as nofldrxq
700 		 * will be 0 if offload is disabled.
701 		 */
702 		for_each_port(sc, i) {
703 			struct port_info *pi = sc->port[i];
704 			struct sge_rxq *rxq;
705 #ifdef TCP_OFFLOAD_ENABLE
706 			struct sge_ofld_rxq *ofld_rxq;
707 
708 			/*
709 			 * Skip over the NIC queues if they aren't taking direct
710 			 * interrupts.
711 			 */
712 			if ((sc->flags & INTR_FWD) &&
713 			    pi->nofldrxq > pi->nrxq)
714 				goto ofld_queues;
715 #endif
716 			rxq = &s->rxq[pi->first_rxq];
717 			for (q = 0; q < pi->nrxq; q++, rxq++) {
718 				(void) ddi_intr_add_handler(
719 				    sc->intr_handle[irq], t4_intr, sc,
720 				    &rxq->iq);
721 				irq++;
722 			}
723 
724 #ifdef TCP_OFFLOAD_ENABLE
725 			/*
726 			 * Skip over the offload queues if they aren't taking
727 			 * direct interrupts.
728 			 */
729 			if ((sc->flags & INTR_FWD))
730 				continue;
731 ofld_queues:
732 			ofld_rxq = &s->ofld_rxq[pi->first_ofld_rxq];
733 			for (q = 0; q < pi->nofldrxq; q++, ofld_rxq++) {
734 				(void) ddi_intr_add_handler(
735 				    sc->intr_handle[irq], t4_intr, sc,
736 				    &ofld_rxq->iq);
737 				irq++;
738 			}
739 #endif
740 		}
741 
742 	}
743 	sc->flags |= INTR_ALLOCATED;
744 
745 	ASSERT(rc == DDI_SUCCESS);
746 	ddi_report_dev(dip);
747 
748 	/*
749 	 * Hardware/Firmware/etc. Version/Revision IDs.
750 	 */
751 	t4_dump_version_info(sc);
752 
753 	if (n100g) {
754 		cxgb_printf(dip, CE_NOTE,
755 		    "%dx100G (%d rxq, %d txq total) %d %s.",
756 		    n100g, rqidx, tqidx, sc->intr_count,
757 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
758 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
759 		    "fixed interrupt");
760 	} else if (n40g) {
761 		cxgb_printf(dip, CE_NOTE,
762 		    "%dx40G (%d rxq, %d txq total) %d %s.",
763 		    n40g, rqidx, tqidx, sc->intr_count,
764 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
765 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
766 		    "fixed interrupt");
767 	} else if (n25g) {
768 		cxgb_printf(dip, CE_NOTE,
769 		    "%dx25G (%d rxq, %d txq total) %d %s.",
770 		    n25g, 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 (n10g && n1g) {
775 		cxgb_printf(dip, CE_NOTE,
776 		    "%dx10G %dx1G (%d rxq, %d txq total) %d %s.",
777 		    n10g, n1g, 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 {
782 		cxgb_printf(dip, CE_NOTE,
783 		    "%dx%sG (%d rxq, %d txq per port) %d %s.",
784 		    n10g ? n10g : n1g,
785 		    n10g ? "10" : "1",
786 		    n10g ? iaq.nrxq10g : iaq.nrxq1g,
787 		    n10g ? iaq.ntxq10g : iaq.ntxq1g,
788 		    sc->intr_count,
789 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
790 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
791 		    "fixed interrupt");
792 	}
793 
794 	sc->ksp = setup_kstats(sc);
795 	sc->ksp_stat = setup_wc_kstats(sc);
796 	sc->params.drv_memwin = MEMWIN_NIC;
797 
798 done:
799 	if (rc != DDI_SUCCESS) {
800 		(void) t4_devo_detach(dip, DDI_DETACH);
801 
802 		/* rc may have errno style errors or DDI errors */
803 		rc = DDI_FAILURE;
804 	}
805 
806 	return (rc);
807 }
808 
809 static int
810 t4_devo_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
811 {
812 	int instance, i;
813 	struct adapter *sc;
814 	struct port_info *pi;
815 	struct sge *s;
816 
817 	if (cmd != DDI_DETACH)
818 		return (DDI_FAILURE);
819 
820 	instance = ddi_get_instance(dip);
821 	sc = ddi_get_soft_state(t4_list, instance);
822 	if (sc == NULL)
823 		return (DDI_SUCCESS);
824 
825 	if (sc->flags & FULL_INIT_DONE) {
826 		t4_intr_disable(sc);
827 		for_each_port(sc, i) {
828 			pi = sc->port[i];
829 			if (pi && pi->flags & PORT_INIT_DONE)
830 				(void) port_full_uninit(pi);
831 		}
832 		(void) adapter_full_uninit(sc);
833 	}
834 
835 	/* Safe to call no matter what */
836 	ddi_prop_remove_all(dip);
837 	ddi_remove_minor_node(dip, NULL);
838 
839 	if (sc->ksp != NULL)
840 		kstat_delete(sc->ksp);
841 	if (sc->ksp_stat != NULL)
842 		kstat_delete(sc->ksp_stat);
843 
844 	s = &sc->sge;
845 	if (s->rxq != NULL)
846 		kmem_free(s->rxq, s->nrxq * sizeof (struct sge_rxq));
847 #ifdef TCP_OFFLOAD_ENABLE
848 	if (s->ofld_txq != NULL)
849 		kmem_free(s->ofld_txq, s->nofldtxq * sizeof (struct sge_wrq));
850 	if (s->ofld_rxq != NULL)
851 		kmem_free(s->ofld_rxq,
852 		    s->nofldrxq * sizeof (struct sge_ofld_rxq));
853 	if (s->ctrlq != NULL)
854 		kmem_free(s->ctrlq,
855 		    sc->params.nports * sizeof (struct sge_wrq));
856 #endif
857 	if (s->txq != NULL)
858 		kmem_free(s->txq, s->ntxq * sizeof (struct sge_txq));
859 	if (s->iqmap != NULL)
860 		kmem_free(s->iqmap, s->niq * sizeof (struct sge_iq *));
861 	if (s->eqmap != NULL)
862 		kmem_free(s->eqmap, s->neq * sizeof (struct sge_eq *));
863 
864 	if (s->rxbuf_cache != NULL)
865 		rxbuf_cache_destroy(s->rxbuf_cache);
866 
867 	if (sc->flags & INTR_ALLOCATED) {
868 		for (i = 0; i < sc->intr_count; i++) {
869 			(void) ddi_intr_remove_handler(sc->intr_handle[i]);
870 			(void) ddi_intr_free(sc->intr_handle[i]);
871 		}
872 		sc->flags &= ~INTR_ALLOCATED;
873 	}
874 
875 	if (sc->intr_handle != NULL) {
876 		kmem_free(sc->intr_handle,
877 		    sc->intr_count * sizeof (*sc->intr_handle));
878 	}
879 
880 	for_each_port(sc, i) {
881 		pi = sc->port[i];
882 		if (pi != NULL) {
883 			mutex_destroy(&pi->lock);
884 			kmem_free(pi, sizeof (*pi));
885 			clrbit(&sc->registered_device_map, i);
886 		}
887 	}
888 
889 	if (sc->flags & FW_OK)
890 		(void) t4_fw_bye(sc, sc->mbox);
891 
892 	if (sc->reg1h != NULL)
893 		ddi_regs_map_free(&sc->reg1h);
894 
895 	if (sc->regh != NULL)
896 		ddi_regs_map_free(&sc->regh);
897 
898 	if (sc->pci_regh != NULL)
899 		pci_config_teardown(&sc->pci_regh);
900 
901 	mutex_enter(&t4_adapter_list_lock);
902 	SLIST_REMOVE_HEAD(&t4_adapter_list, link);
903 	mutex_exit(&t4_adapter_list_lock);
904 
905 	mutex_destroy(&sc->lock);
906 	cv_destroy(&sc->cv);
907 	mutex_destroy(&sc->sfl_lock);
908 
909 #ifdef DEBUG
910 	bzero(sc, sizeof (*sc));
911 #endif
912 	ddi_soft_state_free(t4_list, instance);
913 
914 	return (DDI_SUCCESS);
915 }
916 
917 static int
918 t4_devo_quiesce(dev_info_t *dip)
919 {
920 	int instance;
921 	struct adapter *sc;
922 
923 	instance = ddi_get_instance(dip);
924 	sc = ddi_get_soft_state(t4_list, instance);
925 	if (sc == NULL)
926 		return (DDI_SUCCESS);
927 
928 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
929 	t4_intr_disable(sc);
930 	t4_write_reg(sc, A_PL_RST, F_PIORSTMODE | F_PIORST);
931 
932 	return (DDI_SUCCESS);
933 }
934 
935 static int
936 t4_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t op, void *arg,
937     void *result)
938 {
939 	char s[4];
940 	struct port_info *pi;
941 	dev_info_t *child = (dev_info_t *)arg;
942 
943 	switch (op) {
944 	case DDI_CTLOPS_REPORTDEV:
945 		pi = ddi_get_parent_data(rdip);
946 		pi->instance = ddi_get_instance(dip);
947 		pi->child_inst = ddi_get_instance(rdip);
948 		cmn_err(CE_CONT, "?%s%d is port %s on %s%d\n",
949 		    ddi_node_name(rdip), ddi_get_instance(rdip),
950 		    ddi_get_name_addr(rdip), ddi_driver_name(dip),
951 		    ddi_get_instance(dip));
952 		return (DDI_SUCCESS);
953 
954 	case DDI_CTLOPS_INITCHILD:
955 		pi = ddi_get_parent_data(child);
956 		if (pi == NULL)
957 			return (DDI_NOT_WELL_FORMED);
958 		(void) snprintf(s, sizeof (s), "%d", pi->port_id);
959 		ddi_set_name_addr(child, s);
960 		return (DDI_SUCCESS);
961 
962 	case DDI_CTLOPS_UNINITCHILD:
963 		ddi_set_name_addr(child, NULL);
964 		return (DDI_SUCCESS);
965 
966 	case DDI_CTLOPS_ATTACH:
967 	case DDI_CTLOPS_DETACH:
968 		return (DDI_SUCCESS);
969 
970 	default:
971 		return (ddi_ctlops(dip, rdip, op, arg, result));
972 	}
973 }
974 
975 static int
976 t4_bus_config(dev_info_t *dip, uint_t flags, ddi_bus_config_op_t op, void *arg,
977     dev_info_t **cdipp)
978 {
979 	int instance, i;
980 	struct adapter *sc;
981 
982 	instance = ddi_get_instance(dip);
983 	sc = ddi_get_soft_state(t4_list, instance);
984 
985 	if (op == BUS_CONFIG_ONE) {
986 		char *c;
987 
988 		/*
989 		 * arg is something like "cxgb@0" where 0 is the port_id hanging
990 		 * off this nexus.
991 		 */
992 
993 		c = arg;
994 		while (*(c + 1))
995 			c++;
996 
997 		/* There should be exactly 1 digit after '@' */
998 		if (*(c - 1) != '@')
999 			return (NDI_FAILURE);
1000 
1001 		i = *c - '0';
1002 
1003 		if (add_child_node(sc, i) != 0)
1004 			return (NDI_FAILURE);
1005 
1006 		flags |= NDI_ONLINE_ATTACH;
1007 
1008 	} else if (op == BUS_CONFIG_ALL || op == BUS_CONFIG_DRIVER) {
1009 		/* Allocate and bind all child device nodes */
1010 		for_each_port(sc, i)
1011 		    (void) add_child_node(sc, i);
1012 		flags |= NDI_ONLINE_ATTACH;
1013 	}
1014 
1015 	return (ndi_busop_bus_config(dip, flags, op, arg, cdipp, 0));
1016 }
1017 
1018 static int
1019 t4_bus_unconfig(dev_info_t *dip, uint_t flags, ddi_bus_config_op_t op,
1020     void *arg)
1021 {
1022 	int instance, i, rc;
1023 	struct adapter *sc;
1024 
1025 	instance = ddi_get_instance(dip);
1026 	sc = ddi_get_soft_state(t4_list, instance);
1027 
1028 	if (op == BUS_CONFIG_ONE || op == BUS_UNCONFIG_ALL ||
1029 	    op == BUS_UNCONFIG_DRIVER)
1030 		flags |= NDI_UNCONFIG;
1031 
1032 	rc = ndi_busop_bus_unconfig(dip, flags, op, arg);
1033 	if (rc != 0)
1034 		return (rc);
1035 
1036 	if (op == BUS_UNCONFIG_ONE) {
1037 		char *c;
1038 
1039 		c = arg;
1040 		while (*(c + 1))
1041 			c++;
1042 
1043 		if (*(c - 1) != '@')
1044 			return (NDI_SUCCESS);
1045 
1046 		i = *c - '0';
1047 
1048 		rc = remove_child_node(sc, i);
1049 
1050 	} else if (op == BUS_UNCONFIG_ALL || op == BUS_UNCONFIG_DRIVER) {
1051 
1052 		for_each_port(sc, i)
1053 		    (void) remove_child_node(sc, i);
1054 	}
1055 
1056 	return (rc);
1057 }
1058 
1059 /* ARGSUSED */
1060 static int
1061 t4_cb_open(dev_t *devp, int flag, int otyp, cred_t *credp)
1062 {
1063 	struct adapter *sc;
1064 
1065 	if (otyp != OTYP_CHR)
1066 		return (EINVAL);
1067 
1068 	sc = ddi_get_soft_state(t4_list, getminor(*devp));
1069 	if (sc == NULL)
1070 		return (ENXIO);
1071 
1072 	return (atomic_cas_uint(&sc->open, 0, EBUSY));
1073 }
1074 
1075 /* ARGSUSED */
1076 static int
1077 t4_cb_close(dev_t dev, int flag, int otyp, cred_t *credp)
1078 {
1079 	struct adapter *sc;
1080 
1081 	sc = ddi_get_soft_state(t4_list, getminor(dev));
1082 	if (sc == NULL)
1083 		return (EINVAL);
1084 
1085 	(void) atomic_swap_uint(&sc->open, 0);
1086 	return (0);
1087 }
1088 
1089 /* ARGSUSED */
1090 static int
1091 t4_cb_ioctl(dev_t dev, int cmd, intptr_t d, int mode, cred_t *credp, int *rp)
1092 {
1093 	int instance;
1094 	struct adapter *sc;
1095 	void *data = (void *)d;
1096 
1097 	if (crgetuid(credp) != 0)
1098 		return (EPERM);
1099 
1100 	instance = getminor(dev);
1101 	sc = ddi_get_soft_state(t4_list, instance);
1102 	if (sc == NULL)
1103 		return (EINVAL);
1104 
1105 	return (t4_ioctl(sc, cmd, data, mode));
1106 }
1107 
1108 static unsigned int
1109 getpf(struct adapter *sc)
1110 {
1111 	int rc, *data;
1112 	uint_t n, pf;
1113 
1114 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, sc->dip,
1115 	    DDI_PROP_DONTPASS, "reg", &data, &n);
1116 	if (rc != DDI_SUCCESS) {
1117 		cxgb_printf(sc->dip, CE_WARN,
1118 		    "failed to lookup \"reg\" property: %d", rc);
1119 		return (0xff);
1120 	}
1121 
1122 	pf = PCI_REG_FUNC_G(data[0]);
1123 	ddi_prop_free(data);
1124 
1125 	return (pf);
1126 }
1127 
1128 
1129 static struct fw_info *
1130 find_fw_info(int chip)
1131 {
1132 	u32 i;
1133 
1134 	fi[0].chip = CHELSIO_T4;
1135 	fi[0].fw_hdr.chip = FW_HDR_CHIP_T4;
1136 	fi[0].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T4));
1137 	fi[0].fw_hdr.intfver_nic = FW_INTFVER(T4, NIC);
1138 	fi[0].fw_hdr.intfver_vnic = FW_INTFVER(T4, VNIC);
1139 	fi[0].fw_hdr.intfver_ofld = FW_INTFVER(T4, OFLD);
1140 	fi[0].fw_hdr.intfver_ri = FW_INTFVER(T4, RI);
1141 	fi[0].fw_hdr.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU);
1142 	fi[0].fw_hdr.intfver_iscsi = FW_INTFVER(T4, ISCSI);
1143 	fi[0].fw_hdr.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU);
1144 	fi[0].fw_hdr.intfver_fcoe = FW_INTFVER(T4, FCOE);
1145 
1146 	fi[1].chip = CHELSIO_T5;
1147 	fi[1].fw_hdr.chip = FW_HDR_CHIP_T5;
1148 	fi[1].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T5));
1149 	fi[1].fw_hdr.intfver_nic = FW_INTFVER(T5, NIC);
1150 	fi[1].fw_hdr.intfver_vnic = FW_INTFVER(T5, VNIC);
1151 	fi[1].fw_hdr.intfver_ofld = FW_INTFVER(T5, OFLD);
1152 	fi[1].fw_hdr.intfver_ri = FW_INTFVER(T5, RI);
1153 	fi[1].fw_hdr.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU);
1154 	fi[1].fw_hdr.intfver_iscsi = FW_INTFVER(T5, ISCSI);
1155 	fi[1].fw_hdr.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU);
1156 	fi[1].fw_hdr.intfver_fcoe = FW_INTFVER(T5, FCOE);
1157 
1158 	fi[2].chip = CHELSIO_T6;
1159 	fi[2].fw_hdr.chip = FW_HDR_CHIP_T6;
1160 	fi[2].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T6));
1161 	fi[2].fw_hdr.intfver_nic = FW_INTFVER(T6, NIC);
1162 	fi[2].fw_hdr.intfver_vnic = FW_INTFVER(T6, VNIC);
1163 	fi[2].fw_hdr.intfver_ofld = FW_INTFVER(T6, OFLD);
1164 	fi[2].fw_hdr.intfver_ri = FW_INTFVER(T6, RI);
1165 	fi[2].fw_hdr.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU);
1166 	fi[2].fw_hdr.intfver_iscsi = FW_INTFVER(T6, ISCSI);
1167 	fi[2].fw_hdr.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU);
1168 	fi[2].fw_hdr.intfver_fcoe = FW_INTFVER(T6, FCOE);
1169 
1170 	for (i = 0; i < ARRAY_SIZE(fi); i++) {
1171 		if (fi[i].chip == chip)
1172 			return &fi[i];
1173 	}
1174 
1175 	return NULL;
1176 }
1177 
1178 /*
1179  * Install a compatible firmware (if required), establish contact with it,
1180  * become the master, and reset the device.
1181  */
1182 static int
1183 prep_firmware(struct adapter *sc)
1184 {
1185 	int rc;
1186 	int fw_size;
1187 	int reset = 1;
1188 	enum dev_state state;
1189 	unsigned char *fw_data;
1190 	struct fw_info *fw_info;
1191 	struct fw_hdr *card_fw;
1192 
1193 	struct driver_properties *p = &sc->props;
1194 
1195 	/* Contact firmware, request master */
1196 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MUST, &state);
1197 	if (rc < 0) {
1198 		rc = -rc;
1199 		cxgb_printf(sc->dip, CE_WARN,
1200 		    "failed to connect to the firmware: %d.", rc);
1201 		return (rc);
1202 	}
1203 
1204 	if (rc == sc->mbox)
1205 		sc->flags |= MASTER_PF;
1206 
1207 	/* We may need FW version info for later reporting */
1208 	t4_get_version_info(sc);
1209 	fw_info = find_fw_info(CHELSIO_CHIP_VERSION(sc->params.chip));
1210 	/* allocate memory to read the header of the firmware on the
1211 	 * card
1212 	 */
1213 	if (!fw_info) {
1214 		cxgb_printf(sc->dip, CE_WARN,
1215 			    "unable to look up firmware information for chip %d.\n",
1216 			    CHELSIO_CHIP_VERSION(sc->params.chip));
1217 		return EINVAL;
1218 	}
1219 	card_fw = kmem_zalloc(sizeof(*card_fw), KM_SLEEP);
1220 	if(!card_fw) {
1221 		cxgb_printf(sc->dip, CE_WARN,
1222 			    "Memory allocation for card FW header failed\n");
1223 		return ENOMEM;
1224 	}
1225 	switch(CHELSIO_CHIP_VERSION(sc->params.chip)) {
1226 	case CHELSIO_T4:
1227 		fw_data = t4fw_data;
1228 		fw_size = t4fw_size;
1229 		break;
1230 	case CHELSIO_T5:
1231 		fw_data = t5fw_data;
1232 		fw_size = t5fw_size;
1233 		break;
1234 	case CHELSIO_T6:
1235 		fw_data = t6fw_data;
1236 		fw_size = t6fw_size;
1237 		break;
1238 	default:
1239 		cxgb_printf(sc->dip, CE_WARN, "Adapter type not supported\n");
1240 		kmem_free(card_fw, sizeof(*card_fw));
1241 		return EINVAL;
1242 	}
1243 
1244 	rc = -t4_prep_fw(sc, fw_info, fw_data, fw_size, card_fw,
1245 			 p->t4_fw_install, state, &reset);
1246 
1247 	kmem_free(card_fw, sizeof(*card_fw));
1248 
1249 	if (rc != 0) {
1250 		cxgb_printf(sc->dip, CE_WARN,
1251 		    "failed to install firmware: %d", rc);
1252 		return (rc);
1253 	} else {
1254 		/* refresh */
1255 		(void) t4_check_fw_version(sc);
1256 	}
1257 
1258 	/* Reset device */
1259 	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
1260 	if (rc != 0) {
1261 		cxgb_printf(sc->dip, CE_WARN,
1262 		    "firmware reset failed: %d.", rc);
1263 		if (rc != ETIMEDOUT && rc != EIO)
1264 			(void) t4_fw_bye(sc, sc->mbox);
1265 		return (rc);
1266 	}
1267 
1268 	/* Partition adapter resources as specified in the config file. */
1269 	if (sc->flags & MASTER_PF) {
1270 		/* Handle default vs special T4 config file */
1271 
1272 		rc = partition_resources(sc);
1273 		if (rc != 0)
1274 			goto err;	/* error message displayed already */
1275 	}
1276 
1277 	sc->flags |= FW_OK;
1278 	return (0);
1279 err:
1280 	return (rc);
1281 
1282 }
1283 
1284 static const struct memwin t4_memwin[] = {
1285 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1286 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1287 	{ MEMWIN2_BASE, MEMWIN2_APERTURE }
1288 };
1289 
1290 static const struct memwin t5_memwin[] = {
1291 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1292 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1293 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
1294 };
1295 
1296 #define	FW_PARAM_DEV(param) \
1297 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
1298 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
1299 #define	FW_PARAM_PFVF(param) \
1300 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
1301 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
1302 
1303 /*
1304  * Verify that the memory range specified by the memtype/offset/len pair is
1305  * valid and lies entirely within the memtype specified.  The global address of
1306  * the start of the range is returned in addr.
1307  */
1308 int
1309 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
1310 	uint32_t *addr)
1311 {
1312 	uint32_t em, addr_len, maddr, mlen;
1313 
1314 	/* Memory can only be accessed in naturally aligned 4 byte units */
1315 	if (off & 3 || len & 3 || len == 0)
1316 		return (EINVAL);
1317 
1318 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
1319 	switch (mtype) {
1320 		case MEM_EDC0:
1321 			if (!(em & F_EDRAM0_ENABLE))
1322 				return (EINVAL);
1323 			addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
1324 			maddr = G_EDRAM0_BASE(addr_len) << 20;
1325 			mlen = G_EDRAM0_SIZE(addr_len) << 20;
1326 			break;
1327 		case MEM_EDC1:
1328 			if (!(em & F_EDRAM1_ENABLE))
1329 				return (EINVAL);
1330 			addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
1331 			maddr = G_EDRAM1_BASE(addr_len) << 20;
1332 			mlen = G_EDRAM1_SIZE(addr_len) << 20;
1333 			break;
1334 		case MEM_MC:
1335 			if (!(em & F_EXT_MEM_ENABLE))
1336 				return (EINVAL);
1337 			addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
1338 			maddr = G_EXT_MEM_BASE(addr_len) << 20;
1339 			mlen = G_EXT_MEM_SIZE(addr_len) << 20;
1340 			break;
1341 		case MEM_MC1:
1342 			if (is_t4(sc->params.chip) || !(em & F_EXT_MEM1_ENABLE))
1343 				return (EINVAL);
1344 			addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
1345 			maddr = G_EXT_MEM1_BASE(addr_len) << 20;
1346 			mlen = G_EXT_MEM1_SIZE(addr_len) << 20;
1347 			break;
1348 		default:
1349 			return (EINVAL);
1350 	}
1351 
1352 	if (mlen > 0 && off < mlen && off + len <= mlen) {
1353 		*addr = maddr + off;    /* global address */
1354 		return (0);
1355 	}
1356 
1357 	return (EFAULT);
1358 }
1359 
1360 void
1361 memwin_info(struct adapter *sc, int win, uint32_t *base, uint32_t *aperture)
1362 {
1363 	const struct memwin *mw;
1364 
1365 	if (is_t4(sc->params.chip)) {
1366 		mw = &t4_memwin[win];
1367 	} else {
1368 		mw = &t5_memwin[win];
1369 	}
1370 
1371 	if (base != NULL)
1372 		*base = mw->base;
1373 	if (aperture != NULL)
1374 		*aperture = mw->aperture;
1375 }
1376 
1377 /*
1378  * Upload configuration file to card's memory.
1379  */
1380 static int
1381 upload_config_file(struct adapter *sc, uint32_t *mt, uint32_t *ma)
1382 {
1383 	int rc = 0, cflen;
1384 	u_int i, n;
1385 	uint32_t param, val, addr, mtype, maddr;
1386 	uint32_t off, mw_base, mw_aperture;
1387 	const uint32_t *cfdata;
1388 
1389 	/* Figure out where the firmware wants us to upload it. */
1390 	param = FW_PARAM_DEV(CF);
1391 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
1392 	if (rc != 0) {
1393 		/* Firmwares without config file support will fail this way */
1394 		cxgb_printf(sc->dip, CE_WARN,
1395 		    "failed to query config file location: %d.\n", rc);
1396 		return (rc);
1397 	}
1398 	*mt = mtype = G_FW_PARAMS_PARAM_Y(val);
1399 	*ma = maddr = G_FW_PARAMS_PARAM_Z(val) << 16;
1400 
1401 	switch (CHELSIO_CHIP_VERSION(sc->params.chip)) {
1402 	case CHELSIO_T4:
1403 		cflen = t4cfg_size & ~3;
1404 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1405 		cfdata = (const uint32_t *)t4cfg_data;
1406 		break;
1407 	case CHELSIO_T5:
1408 		cflen = t5cfg_size & ~3;
1409 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1410 		cfdata = (const uint32_t *)t5cfg_data;
1411 		break;
1412 	case CHELSIO_T6:
1413 		cflen = t6cfg_size & ~3;
1414 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1415 		cfdata = (const uint32_t *)t6cfg_data;
1416 		break;
1417 	default:
1418 		cxgb_printf(sc->dip, CE_WARN,
1419 			    "Invalid Adapter detected\n");
1420 		return EINVAL;
1421 	}
1422 
1423 	if (cflen > FLASH_CFG_MAX_SIZE) {
1424 		cxgb_printf(sc->dip, CE_WARN,
1425 		    "config file too long (%d, max allowed is %d).  ",
1426 		    cflen, FLASH_CFG_MAX_SIZE);
1427 		return (EFBIG);
1428 	}
1429 
1430 	rc = validate_mt_off_len(sc, mtype, maddr, cflen, &addr);
1431 	if (rc != 0) {
1432 
1433 		cxgb_printf(sc->dip, CE_WARN,
1434 		    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
1435 		    "Will try to use the config on the card, if any.\n",
1436 		    __func__, mtype, maddr, cflen, rc);
1437 		return (EFAULT);
1438 	}
1439 
1440 	memwin_info(sc, 2, &mw_base, &mw_aperture);
1441 	while (cflen) {
1442 		off = position_memwin(sc, 2, addr);
1443 		n = min(cflen, mw_aperture - off);
1444 		for (i = 0; i < n; i += 4)
1445 			t4_write_reg(sc, mw_base + off + i, *cfdata++);
1446 		cflen -= n;
1447 		addr += n;
1448 	}
1449 
1450 	return (rc);
1451 }
1452 
1453 /*
1454  * Partition chip resources for use between various PFs, VFs, etc.  This is done
1455  * by uploading the firmware configuration file to the adapter and instructing
1456  * the firmware to process it.
1457  */
1458 static int
1459 partition_resources(struct adapter *sc)
1460 {
1461 	int rc;
1462 	struct fw_caps_config_cmd caps;
1463 	uint32_t mtype, maddr, finicsum, cfcsum;
1464 
1465 	rc = upload_config_file(sc, &mtype, &maddr);
1466 	if (rc != 0) {
1467 		mtype = FW_MEMTYPE_CF_FLASH;
1468 		maddr = t4_flash_cfg_addr(sc);
1469 	}
1470 
1471 	bzero(&caps, sizeof (caps));
1472 	caps.op_to_write = BE_32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1473 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1474 	caps.cfvalid_to_len16 = BE_32(F_FW_CAPS_CONFIG_CMD_CFVALID |
1475 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
1476 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) | FW_LEN16(caps));
1477 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), &caps);
1478 	if (rc != 0) {
1479 		cxgb_printf(sc->dip, CE_WARN,
1480 		    "failed to pre-process config file: %d.\n", rc);
1481 		return (rc);
1482 	}
1483 
1484 	finicsum = ntohl(caps.finicsum);
1485 	cfcsum = ntohl(caps.cfcsum);
1486 	if (finicsum != cfcsum) {
1487 		cxgb_printf(sc->dip, CE_WARN,
1488 		    "WARNING: config file checksum mismatch: %08x %08x\n",
1489 		    finicsum, cfcsum);
1490 	}
1491 	sc->cfcsum = cfcsum;
1492 
1493 	/* TODO: Need to configure this correctly */
1494 	caps.toecaps = htons(FW_CAPS_CONFIG_TOE);
1495 	caps.iscsicaps = 0;
1496 	caps.rdmacaps = 0;
1497 	caps.fcoecaps = 0;
1498 	/* TODO: Disable VNIC cap for now */
1499 	caps.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
1500 
1501 	caps.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1502 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
1503 	caps.cfvalid_to_len16 = htonl(FW_LEN16(caps));
1504 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), NULL);
1505 	if (rc != 0) {
1506 		cxgb_printf(sc->dip, CE_WARN,
1507 		    "failed to process config file: %d.\n", rc);
1508 		return (rc);
1509 	}
1510 
1511 	return (0);
1512 }
1513 
1514 /*
1515  * Tweak configuration based on module parameters, etc.  Most of these have
1516  * defaults assigned to them by Firmware Configuration Files (if we're using
1517  * them) but need to be explicitly set if we're using hard-coded
1518  * initialization.  But even in the case of using Firmware Configuration
1519  * Files, we'd like to expose the ability to change these via module
1520  * parameters so these are essentially common tweaks/settings for
1521  * Configuration Files and hard-coded initialization ...
1522  */
1523 static int
1524 adap__pre_init_tweaks(struct adapter *sc)
1525 {
1526 	int rx_dma_offset = 2; /* Offset of RX packets into DMA buffers */
1527 
1528 	/*
1529 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
1530 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
1531 	 * 64B Cache Line Size ...
1532 	 */
1533 	(void) t4_fixup_host_params_compat(sc, PAGE_SIZE, CACHE_LINE, T5_LAST_REV);
1534 
1535 	t4_set_reg_field(sc, A_SGE_CONTROL,
1536 			 V_PKTSHIFT(M_PKTSHIFT), V_PKTSHIFT(rx_dma_offset));
1537 
1538 	return 0;
1539 }
1540 /*
1541  * Retrieve parameters that are needed (or nice to have) prior to calling
1542  * t4_sge_init and t4_fw_initialize.
1543  */
1544 static int
1545 get_params__pre_init(struct adapter *sc)
1546 {
1547 	int rc;
1548 	uint32_t param[2], val[2];
1549 	struct fw_devlog_cmd cmd;
1550 	struct devlog_params *dlog = &sc->params.devlog;
1551 
1552 	/*
1553 	 * Grab the raw VPD parameters.
1554 	 */
1555 	rc = -t4_get_raw_vpd_params(sc, &sc->params.vpd);
1556 	if (rc != 0) {
1557 		cxgb_printf(sc->dip, CE_WARN,
1558 		    "failed to query VPD parameters (pre_init): %d.\n", rc);
1559 		return (rc);
1560 	}
1561 
1562 	param[0] = FW_PARAM_DEV(PORTVEC);
1563 	param[1] = FW_PARAM_DEV(CCLK);
1564 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
1565 	if (rc != 0) {
1566 		cxgb_printf(sc->dip, CE_WARN,
1567 		    "failed to query parameters (pre_init): %d.\n", rc);
1568 		return (rc);
1569 	}
1570 
1571 	sc->params.portvec = val[0];
1572 	sc->params.nports = 0;
1573 	while (val[0]) {
1574 		sc->params.nports++;
1575 		val[0] &= val[0] - 1;
1576 	}
1577 
1578 	sc->params.vpd.cclk = val[1];
1579 
1580 	/* Read device log parameters. */
1581 	bzero(&cmd, sizeof (cmd));
1582 	cmd.op_to_write = htonl(V_FW_CMD_OP(FW_DEVLOG_CMD) |
1583 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1584 	cmd.retval_len16 = htonl(FW_LEN16(cmd));
1585 	rc = -t4_wr_mbox(sc, sc->mbox, &cmd, sizeof (cmd), &cmd);
1586 	if (rc != 0) {
1587 		cxgb_printf(sc->dip, CE_WARN,
1588 		    "failed to get devlog parameters: %d.\n", rc);
1589 		bzero(dlog, sizeof (*dlog));
1590 		rc = 0;	/* devlog isn't critical for device operation */
1591 	} else {
1592 		val[0] = ntohl(cmd.memtype_devlog_memaddr16_devlog);
1593 		dlog->memtype = G_FW_DEVLOG_CMD_MEMTYPE_DEVLOG(val[0]);
1594 		dlog->start = G_FW_DEVLOG_CMD_MEMADDR16_DEVLOG(val[0]) << 4;
1595 		dlog->size = ntohl(cmd.memsize_devlog);
1596 	}
1597 
1598 	return (rc);
1599 }
1600 
1601 /*
1602  * Retrieve various parameters that are of interest to the driver.  The device
1603  * has been initialized by the firmware at this point.
1604  */
1605 static int
1606 get_params__post_init(struct adapter *sc)
1607 {
1608 	int rc;
1609 	uint32_t param[7], val[7];
1610 	struct fw_caps_config_cmd caps;
1611 
1612 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
1613 	param[1] = FW_PARAM_PFVF(EQ_START);
1614 	param[2] = FW_PARAM_PFVF(FILTER_START);
1615 	param[3] = FW_PARAM_PFVF(FILTER_END);
1616 	param[4] = FW_PARAM_PFVF(L2T_START);
1617 	param[5] = FW_PARAM_PFVF(L2T_END);
1618 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
1619 	if (rc != 0) {
1620 		cxgb_printf(sc->dip, CE_WARN,
1621 		    "failed to query parameters (post_init): %d.\n", rc);
1622 		return (rc);
1623 	}
1624 
1625 	/* LINTED: E_ASSIGN_NARROW_CONV */
1626 	sc->sge.iq_start = val[0];
1627 	sc->sge.eq_start = val[1];
1628 	sc->tids.ftid_base = val[2];
1629 	sc->tids.nftids = val[3] - val[2] + 1;
1630 	sc->vres.l2t.start = val[4];
1631 	sc->vres.l2t.size = val[5] - val[4] + 1;
1632 
1633 	/* get capabilites */
1634 	bzero(&caps, sizeof (caps));
1635 	caps.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1636 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1637 	caps.cfvalid_to_len16 = htonl(FW_LEN16(caps));
1638 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), &caps);
1639 	if (rc != 0) {
1640 		cxgb_printf(sc->dip, CE_WARN,
1641 		    "failed to get card capabilities: %d.\n", rc);
1642 		return (rc);
1643 	}
1644 
1645 	if (caps.toecaps != 0) {
1646 		/* query offload-related parameters */
1647 		param[0] = FW_PARAM_DEV(NTID);
1648 		param[1] = FW_PARAM_PFVF(SERVER_START);
1649 		param[2] = FW_PARAM_PFVF(SERVER_END);
1650 		param[3] = FW_PARAM_PFVF(TDDP_START);
1651 		param[4] = FW_PARAM_PFVF(TDDP_END);
1652 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
1653 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
1654 		if (rc != 0) {
1655 			cxgb_printf(sc->dip, CE_WARN,
1656 			    "failed to query TOE parameters: %d.\n", rc);
1657 			return (rc);
1658 		}
1659 		sc->tids.ntids = val[0];
1660 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
1661 		sc->tids.stid_base = val[1];
1662 		sc->tids.nstids = val[2] - val[1] + 1;
1663 		sc->vres.ddp.start = val[3];
1664 		sc->vres.ddp.size = val[4] - val[3] + 1;
1665 		sc->params.ofldq_wr_cred = val[5];
1666 		sc->params.offload = 1;
1667 	}
1668 
1669 	/* These are finalized by FW initialization, load their values now */
1670 	val[0] = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
1671 	sc->params.tp.tre = G_TIMERRESOLUTION(val[0]);
1672 	sc->params.tp.dack_re = G_DELAYEDACKRESOLUTION(val[0]);
1673 	t4_read_mtu_tbl(sc, sc->params.mtus, NULL);
1674 
1675 	return (rc);
1676 }
1677 
1678 static int
1679 set_params__post_init(struct adapter *sc)
1680 {
1681 	uint32_t param, val;
1682 
1683 	/* ask for encapsulated CPLs */
1684 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
1685 	val = 1;
1686 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
1687 
1688 	return (0);
1689 }
1690 
1691 /* TODO: verify */
1692 static void
1693 setup_memwin(struct adapter *sc)
1694 {
1695 	pci_regspec_t *data;
1696 	int rc;
1697 	uint_t n;
1698 	uintptr_t bar0;
1699 	uintptr_t mem_win0_base, mem_win1_base, mem_win2_base;
1700 	uintptr_t mem_win2_aperture;
1701 
1702 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, sc->dip,
1703 	    DDI_PROP_DONTPASS, "assigned-addresses", (int **)&data, &n);
1704 	if (rc != DDI_SUCCESS) {
1705 		cxgb_printf(sc->dip, CE_WARN,
1706 		    "failed to lookup \"assigned-addresses\" property: %d", rc);
1707 		return;
1708 	}
1709 	n /= sizeof (*data);
1710 
1711 	bar0 = ((uint64_t)data[0].pci_phys_mid << 32) | data[0].pci_phys_low;
1712 	ddi_prop_free(data);
1713 
1714 	if (is_t4(sc->params.chip)) {
1715 		mem_win0_base = bar0 + MEMWIN0_BASE;
1716 		mem_win1_base = bar0 + MEMWIN1_BASE;
1717 		mem_win2_base = bar0 + MEMWIN2_BASE;
1718 		mem_win2_aperture = MEMWIN2_APERTURE;
1719 	} else {
1720 		/* For T5, only relative offset inside the PCIe BAR is passed */
1721 		mem_win0_base = MEMWIN0_BASE;
1722 		mem_win1_base = MEMWIN1_BASE;
1723 		mem_win2_base = MEMWIN2_BASE_T5;
1724 		mem_win2_aperture = MEMWIN2_APERTURE_T5;
1725 	}
1726 
1727 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 0),
1728 	    mem_win0_base | V_BIR(0) |
1729 	    V_WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
1730 
1731 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 1),
1732 	    mem_win1_base | V_BIR(0) |
1733 	    V_WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
1734 
1735 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2),
1736 	    mem_win2_base | V_BIR(0) |
1737 	    V_WINDOW(ilog2(mem_win2_aperture) - 10));
1738 
1739 	/* flush */
1740 	(void)t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
1741 }
1742 
1743 /*
1744  * Positions the memory window such that it can be used to access the specified
1745  * address in the chip's address space.  The return value is the offset of addr
1746  * from the start of the window.
1747  */
1748 uint32_t
1749 position_memwin(struct adapter *sc, int n, uint32_t addr)
1750 {
1751 	uint32_t start, pf;
1752 	uint32_t reg;
1753 
1754 	if (addr & 3) {
1755 		cxgb_printf(sc->dip, CE_WARN,
1756 		    "addr (0x%x) is not at a 4B boundary.\n", addr);
1757 		return (EFAULT);
1758 	}
1759 
1760 	if (is_t4(sc->params.chip)) {
1761 		pf = 0;
1762 		start = addr & ~0xf;    /* start must be 16B aligned */
1763 	} else {
1764 		pf = V_PFNUM(sc->pf);
1765 		start = addr & ~0x7f;   /* start must be 128B aligned */
1766 	}
1767 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, n);
1768 
1769 	t4_write_reg(sc, reg, start | pf);
1770 	(void) t4_read_reg(sc, reg);
1771 
1772 	return (addr - start);
1773 }
1774 
1775 
1776 /*
1777  * Reads the named property and fills up the "data" array (which has at least
1778  * "count" elements).  We first try and lookup the property for our dev_t and
1779  * then retry with DDI_DEV_T_ANY if it's not found.
1780  *
1781  * Returns non-zero if the property was found and "data" has been updated.
1782  */
1783 static int
1784 prop_lookup_int_array(struct adapter *sc, char *name, int *data, uint_t count)
1785 {
1786 	dev_info_t *dip = sc->dip;
1787 	dev_t dev = sc->dev;
1788 	int rc, *d;
1789 	uint_t i, n;
1790 
1791 	rc = ddi_prop_lookup_int_array(dev, dip, DDI_PROP_DONTPASS,
1792 	    name, &d, &n);
1793 	if (rc == DDI_PROP_SUCCESS)
1794 		goto found;
1795 
1796 	if (rc != DDI_PROP_NOT_FOUND) {
1797 		cxgb_printf(dip, CE_WARN,
1798 		    "failed to lookup property %s for minor %d: %d.",
1799 		    name, getminor(dev), rc);
1800 		return (0);
1801 	}
1802 
1803 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
1804 	    name, &d, &n);
1805 	if (rc == DDI_PROP_SUCCESS)
1806 		goto found;
1807 
1808 	if (rc != DDI_PROP_NOT_FOUND) {
1809 		cxgb_printf(dip, CE_WARN,
1810 		    "failed to lookup property %s: %d.", name, rc);
1811 		return (0);
1812 	}
1813 
1814 	return (0);
1815 
1816 found:
1817 	if (n > count) {
1818 		cxgb_printf(dip, CE_NOTE,
1819 		    "property %s has too many elements (%d), ignoring extras",
1820 		    name, n);
1821 	}
1822 
1823 	for (i = 0; i < n && i < count; i++)
1824 		data[i] = d[i];
1825 	ddi_prop_free(d);
1826 
1827 	return (1);
1828 }
1829 
1830 static int
1831 prop_lookup_int(struct adapter *sc, char *name, int defval)
1832 {
1833 	int rc;
1834 
1835 	rc = ddi_prop_get_int(sc->dev, sc->dip, DDI_PROP_DONTPASS, name, -1);
1836 	if (rc != -1)
1837 		return (rc);
1838 
1839 	return (ddi_prop_get_int(DDI_DEV_T_ANY, sc->dip, DDI_PROP_DONTPASS,
1840 	    name, defval));
1841 }
1842 
1843 static int
1844 init_driver_props(struct adapter *sc, struct driver_properties *p)
1845 {
1846 	dev_t dev = sc->dev;
1847 	dev_info_t *dip = sc->dip;
1848 	int i, *data;
1849 	uint_t tmr[SGE_NTIMERS] = {5, 10, 20, 50, 100, 200};
1850 	uint_t cnt[SGE_NCOUNTERS] = {1, 8, 16, 32}; /* 63 max */
1851 
1852 	/*
1853 	 * Holdoff timer
1854 	 */
1855 	data = &p->timer_val[0];
1856 	for (i = 0; i < SGE_NTIMERS; i++)
1857 		data[i] = tmr[i];
1858 	(void) prop_lookup_int_array(sc, "holdoff-timer-values", data,
1859 	    SGE_NTIMERS);
1860 	for (i = 0; i < SGE_NTIMERS; i++) {
1861 		int limit = 200U;
1862 		if (data[i] > limit) {
1863 			cxgb_printf(dip, CE_WARN,
1864 			    "holdoff timer %d is too high (%d), lowered to %d.",
1865 			    i, data[i], limit);
1866 			data[i] = limit;
1867 		}
1868 	}
1869 	(void) ddi_prop_update_int_array(dev, dip, "holdoff-timer-values",
1870 	    data, SGE_NTIMERS);
1871 
1872 	/*
1873 	 * Holdoff packet counter
1874 	 */
1875 	data = &p->counter_val[0];
1876 	for (i = 0; i < SGE_NCOUNTERS; i++)
1877 		data[i] = cnt[i];
1878 	(void) prop_lookup_int_array(sc, "holdoff-pkt-counter-values", data,
1879 	    SGE_NCOUNTERS);
1880 	for (i = 0; i < SGE_NCOUNTERS; i++) {
1881 		int limit = M_THRESHOLD_0;
1882 		if (data[i] > limit) {
1883 			cxgb_printf(dip, CE_WARN,
1884 			    "holdoff pkt-counter %d is too high (%d), "
1885 			    "lowered to %d.", i, data[i], limit);
1886 			data[i] = limit;
1887 		}
1888 	}
1889 	(void) ddi_prop_update_int_array(dev, dip, "holdoff-pkt-counter-values",
1890 	    data, SGE_NCOUNTERS);
1891 
1892 	/*
1893 	 * Maximum # of tx and rx queues to use for each
1894 	 * 100G, 40G, 25G, 10G and 1G port.
1895 	 */
1896 	p->max_ntxq_10g = prop_lookup_int(sc, "max-ntxq-10G-port", 8);
1897 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-10G-port",
1898 	    p->max_ntxq_10g);
1899 
1900 	p->max_nrxq_10g = prop_lookup_int(sc, "max-nrxq-10G-port", 8);
1901 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-10G-port",
1902 	    p->max_nrxq_10g);
1903 
1904 	p->max_ntxq_1g = prop_lookup_int(sc, "max-ntxq-1G-port", 2);
1905 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-1G-port",
1906 	    p->max_ntxq_1g);
1907 
1908 	p->max_nrxq_1g = prop_lookup_int(sc, "max-nrxq-1G-port", 2);
1909 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-1G-port",
1910 	    p->max_nrxq_1g);
1911 
1912 #ifdef TCP_OFFLOAD_ENABLE
1913 	p->max_nofldtxq_10g = prop_lookup_int(sc, "max-nofldtxq-10G-port", 8);
1914 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-10G-port",
1915 	    p->max_nofldtxq_10g);
1916 
1917 	p->max_nofldrxq_10g = prop_lookup_int(sc, "max-nofldrxq-10G-port", 2);
1918 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-10G-port",
1919 	    p->max_nofldrxq_10g);
1920 
1921 	p->max_nofldtxq_1g = prop_lookup_int(sc, "max-nofldtxq-1G-port", 2);
1922 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-1G-port",
1923 	    p->max_nofldtxq_1g);
1924 
1925 	p->max_nofldrxq_1g = prop_lookup_int(sc, "max-nofldrxq-1G-port", 1);
1926 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-1G-port",
1927 	    p->max_nofldrxq_1g);
1928 #endif
1929 
1930 	/*
1931 	 * Holdoff parameters for 10G and 1G ports.
1932 	 */
1933 	p->tmr_idx_10g = prop_lookup_int(sc, "holdoff-timer-idx-10G", 0);
1934 	(void) ddi_prop_update_int(dev, dip, "holdoff-timer-idx-10G",
1935 	    p->tmr_idx_10g);
1936 
1937 	p->pktc_idx_10g = prop_lookup_int(sc, "holdoff-pktc-idx-10G", 2);
1938 	(void) ddi_prop_update_int(dev, dip, "holdoff-pktc-idx-10G",
1939 	    p->pktc_idx_10g);
1940 
1941 	p->tmr_idx_1g = prop_lookup_int(sc, "holdoff-timer-idx-1G", 0);
1942 	(void) ddi_prop_update_int(dev, dip, "holdoff-timer-idx-1G",
1943 	    p->tmr_idx_1g);
1944 
1945 	p->pktc_idx_1g = prop_lookup_int(sc, "holdoff-pktc-idx-1G", 2);
1946 	(void) ddi_prop_update_int(dev, dip, "holdoff-pktc-idx-1G",
1947 	    p->pktc_idx_1g);
1948 
1949 	/*
1950 	 * Size (number of entries) of each tx and rx queue.
1951 	 */
1952 	i = prop_lookup_int(sc, "qsize-txq", TX_EQ_QSIZE);
1953 	p->qsize_txq = max(i, 128);
1954 	if (p->qsize_txq != i) {
1955 		cxgb_printf(dip, CE_WARN,
1956 		    "using %d instead of %d as the tx queue size",
1957 		    p->qsize_txq, i);
1958 	}
1959 	(void) ddi_prop_update_int(dev, dip, "qsize-txq", p->qsize_txq);
1960 
1961 	i = prop_lookup_int(sc, "qsize-rxq", RX_IQ_QSIZE);
1962 	p->qsize_rxq = max(i, 128);
1963 	while (p->qsize_rxq & 7)
1964 		p->qsize_rxq--;
1965 	if (p->qsize_rxq != i) {
1966 		cxgb_printf(dip, CE_WARN,
1967 		    "using %d instead of %d as the rx queue size",
1968 		    p->qsize_rxq, i);
1969 	}
1970 	(void) ddi_prop_update_int(dev, dip, "qsize-rxq", p->qsize_rxq);
1971 
1972 	/*
1973 	 * Interrupt types allowed.
1974 	 * Bits 0, 1, 2 = INTx, MSI, MSI-X respectively.  See sys/ddi_intr.h
1975 	 */
1976 	p->intr_types = prop_lookup_int(sc, "interrupt-types",
1977 	    DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI | DDI_INTR_TYPE_FIXED);
1978 	(void) ddi_prop_update_int(dev, dip, "interrupt-types", p->intr_types);
1979 
1980 	/*
1981 	 * Forwarded interrupt queues.  Create this property to force the driver
1982 	 * to use forwarded interrupt queues.
1983 	 */
1984 	if (ddi_prop_exists(dev, dip, DDI_PROP_DONTPASS,
1985 	    "interrupt-forwarding") != 0 ||
1986 	    ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
1987 	    "interrupt-forwarding") != 0) {
1988 		UNIMPLEMENTED();
1989 		(void) ddi_prop_create(dev, dip, DDI_PROP_CANSLEEP,
1990 		    "interrupt-forwarding", NULL, 0);
1991 	}
1992 
1993 	/*
1994 	 * Write combining
1995 	 * 0 to disable, 1 to enable
1996 	 */
1997 	p->wc = prop_lookup_int(sc, "write-combine", 1);
1998 	cxgb_printf(dip, CE_WARN, "write-combine: using of %d", p->wc);
1999 	if (p->wc != 0 && p->wc != 1) {
2000 		cxgb_printf(dip, CE_WARN,
2001 		    "write-combine: using 1 instead of %d", p->wc);
2002 		p->wc = 1;
2003 	}
2004 	(void) ddi_prop_update_int(dev, dip, "write-combine", p->wc);
2005 
2006 	p->t4_fw_install = prop_lookup_int(sc, "t4_fw_install", 1);
2007 	if (p->t4_fw_install != 0 && p->t4_fw_install != 2)
2008 		p->t4_fw_install = 1;
2009 	(void) ddi_prop_update_int(dev, dip, "t4_fw_install", p->t4_fw_install);
2010 
2011 	/* Multiple Rings */
2012 	p->multi_rings = prop_lookup_int(sc, "multi-rings", 1);
2013 	if (p->multi_rings != 0 && p->multi_rings != 1) {
2014 		cxgb_printf(dip, CE_NOTE,
2015 			   "multi-rings: using value 1 instead of %d", p->multi_rings);
2016 		p->multi_rings = 1;
2017 	}
2018 
2019 	(void) ddi_prop_update_int(dev, dip, "multi-rings", p->multi_rings);
2020 
2021 	return (0);
2022 }
2023 
2024 static int
2025 remove_extra_props(struct adapter *sc, int n10g, int n1g)
2026 {
2027 	if (n10g == 0) {
2028 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-ntxq-10G-port");
2029 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-nrxq-10G-port");
2030 		(void) ddi_prop_remove(sc->dev, sc->dip,
2031 		    "holdoff-timer-idx-10G");
2032 		(void) ddi_prop_remove(sc->dev, sc->dip,
2033 		    "holdoff-pktc-idx-10G");
2034 	}
2035 
2036 	if (n1g == 0) {
2037 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-ntxq-1G-port");
2038 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-nrxq-1G-port");
2039 		(void) ddi_prop_remove(sc->dev, sc->dip,
2040 		    "holdoff-timer-idx-1G");
2041 		(void) ddi_prop_remove(sc->dev, sc->dip, "holdoff-pktc-idx-1G");
2042 	}
2043 
2044 	return (0);
2045 }
2046 
2047 static int
2048 cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g,
2049     struct intrs_and_queues *iaq)
2050 {
2051 	struct driver_properties *p = &sc->props;
2052 	int rc, itype, itypes, navail, nc, nrxq10g, nrxq1g, n;
2053 	int nofldrxq10g = 0, nofldrxq1g = 0;
2054 
2055 	bzero(iaq, sizeof (*iaq));
2056 	nc = ncpus;	/* our snapshot of the number of CPUs */
2057 	iaq->ntxq10g = min(nc, p->max_ntxq_10g);
2058 	iaq->ntxq1g = min(nc, p->max_ntxq_1g);
2059 	iaq->nrxq10g = nrxq10g = min(nc, p->max_nrxq_10g);
2060 	iaq->nrxq1g = nrxq1g = min(nc, p->max_nrxq_1g);
2061 #ifdef TCP_OFFLOAD_ENABLE
2062 	iaq->nofldtxq10g = min(nc, p->max_nofldtxq_10g);
2063 	iaq->nofldtxq1g = min(nc, p->max_nofldtxq_1g);
2064 	iaq->nofldrxq10g = nofldrxq10g = min(nc, p->max_nofldrxq_10g);
2065 	iaq->nofldrxq1g = nofldrxq1g = min(nc, p->max_nofldrxq_1g);
2066 #endif
2067 
2068 	rc = ddi_intr_get_supported_types(sc->dip, &itypes);
2069 	if (rc != DDI_SUCCESS) {
2070 		cxgb_printf(sc->dip, CE_WARN,
2071 		    "failed to determine supported interrupt types: %d", rc);
2072 		return (rc);
2073 	}
2074 
2075 	for (itype = DDI_INTR_TYPE_MSIX; itype; itype >>= 1) {
2076 		ASSERT(itype == DDI_INTR_TYPE_MSIX ||
2077 		    itype == DDI_INTR_TYPE_MSI ||
2078 		    itype == DDI_INTR_TYPE_FIXED);
2079 
2080 		if ((itype & itypes & p->intr_types) == 0)
2081 			continue;	/* not supported or not allowed */
2082 
2083 		navail = 0;
2084 		rc = ddi_intr_get_navail(sc->dip, itype, &navail);
2085 		if (rc != DDI_SUCCESS || navail == 0) {
2086 			cxgb_printf(sc->dip, CE_WARN,
2087 			    "failed to get # of interrupts for type %d: %d",
2088 			    itype, rc);
2089 			continue;	/* carry on */
2090 		}
2091 
2092 		iaq->intr_type = itype;
2093 		if (navail == 0)
2094 			continue;
2095 
2096 		/*
2097 		 * Best option: an interrupt vector for errors, one for the
2098 		 * firmware event queue, and one each for each rxq (NIC as well
2099 		 * as offload).
2100 		 */
2101 		iaq->nirq = T4_EXTRA_INTR;
2102 		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2103 		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2104 
2105 		if (iaq->nirq <= navail &&
2106 		    (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq))) {
2107 			iaq->intr_fwd = 0;
2108 			goto allocate;
2109 		}
2110 
2111 		/*
2112 		 * Second best option: an interrupt vector for errors, one for
2113 		 * the firmware event queue, and one each for either NIC or
2114 		 * offload rxq's.
2115 		 */
2116 		iaq->nirq = T4_EXTRA_INTR;
2117 		iaq->nirq += n10g * max(nrxq10g, nofldrxq10g);
2118 		iaq->nirq += n1g * max(nrxq1g, nofldrxq1g);
2119 		if (iaq->nirq <= navail &&
2120 		    (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq))) {
2121 			iaq->intr_fwd = 1;
2122 			goto allocate;
2123 		}
2124 
2125 		/*
2126 		 * Next best option: an interrupt vector for errors, one for the
2127 		 * firmware event queue, and at least one per port.  At this
2128 		 * point we know we'll have to downsize nrxq or nofldrxq to fit
2129 		 * what's available to us.
2130 		 */
2131 		iaq->nirq = T4_EXTRA_INTR;
2132 		iaq->nirq += n10g + n1g;
2133 		if (iaq->nirq <= navail) {
2134 			int leftover = navail - iaq->nirq;
2135 
2136 			if (n10g > 0) {
2137 				int target = max(nrxq10g, nofldrxq10g);
2138 
2139 				n = 1;
2140 				while (n < target && leftover >= n10g) {
2141 					leftover -= n10g;
2142 					iaq->nirq += n10g;
2143 					n++;
2144 				}
2145 				iaq->nrxq10g = min(n, nrxq10g);
2146 #ifdef TCP_OFFLOAD_ENABLE
2147 				iaq->nofldrxq10g = min(n, nofldrxq10g);
2148 #endif
2149 			}
2150 
2151 			if (n1g > 0) {
2152 				int target = max(nrxq1g, nofldrxq1g);
2153 
2154 				n = 1;
2155 				while (n < target && leftover >= n1g) {
2156 					leftover -= n1g;
2157 					iaq->nirq += n1g;
2158 					n++;
2159 				}
2160 				iaq->nrxq1g = min(n, nrxq1g);
2161 #ifdef TCP_OFFLOAD_ENABLE
2162 				iaq->nofldrxq1g = min(n, nofldrxq1g);
2163 #endif
2164 			}
2165 
2166 			/* We have arrived at a minimum value required to enable
2167 			 * per queue irq(either NIC or offload). Thus for non-
2168 			 * offload case, we will get a vector per queue, while
2169 			 * offload case, we will get a vector per offload/NIC q.
2170 			 * Hence enable Interrupt forwarding only for offload
2171 			 * case.
2172 			 */
2173 #ifdef TCP_OFFLOAD_ENABLE
2174 			if (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq)) {
2175 				iaq->intr_fwd = 1;
2176 #else
2177 			if (itype != DDI_INTR_TYPE_MSI) {
2178 #endif
2179 				goto allocate;
2180 			}
2181 		}
2182 
2183 		/*
2184 		 * Least desirable option: one interrupt vector for everything.
2185 		 */
2186 		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2187 #ifdef TCP_OFFLOAD_ENABLE
2188 		iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2189 #endif
2190 		iaq->intr_fwd = 1;
2191 
2192 allocate:
2193 		return (0);
2194 	}
2195 
2196 	cxgb_printf(sc->dip, CE_WARN,
2197 	    "failed to find a usable interrupt type.  supported=%d, allowed=%d",
2198 	    itypes, p->intr_types);
2199 	return (DDI_FAILURE);
2200 }
2201 
2202 static int
2203 add_child_node(struct adapter *sc, int idx)
2204 {
2205 	int rc;
2206 	struct port_info *pi;
2207 
2208 	if (idx < 0 || idx >= sc->params.nports)
2209 		return (EINVAL);
2210 
2211 	pi = sc->port[idx];
2212 	if (pi == NULL)
2213 		return (ENODEV);	/* t4_port_init failed earlier */
2214 
2215 	PORT_LOCK(pi);
2216 	if (pi->dip != NULL) {
2217 		rc = 0;		/* EEXIST really, but then bus_config fails */
2218 		goto done;
2219 	}
2220 
2221 	rc = ndi_devi_alloc(sc->dip, T4_PORT_NAME, DEVI_SID_NODEID, &pi->dip);
2222 	if (rc != DDI_SUCCESS || pi->dip == NULL) {
2223 		rc = ENOMEM;
2224 		goto done;
2225 	}
2226 
2227 	(void) ddi_set_parent_data(pi->dip, pi);
2228 	(void) ndi_devi_bind_driver(pi->dip, 0);
2229 	rc = 0;
2230 done:
2231 	PORT_UNLOCK(pi);
2232 	return (rc);
2233 }
2234 
2235 static int
2236 remove_child_node(struct adapter *sc, int idx)
2237 {
2238 	int rc;
2239 	struct port_info *pi;
2240 
2241 	if (idx < 0 || idx >= sc->params.nports)
2242 		return (EINVAL);
2243 
2244 	pi = sc->port[idx];
2245 	if (pi == NULL)
2246 		return (ENODEV);
2247 
2248 	PORT_LOCK(pi);
2249 	if (pi->dip == NULL) {
2250 		rc = ENODEV;
2251 		goto done;
2252 	}
2253 
2254 	rc = ndi_devi_free(pi->dip);
2255 	if (rc == 0)
2256 		pi->dip = NULL;
2257 done:
2258 	PORT_UNLOCK(pi);
2259 	return (rc);
2260 }
2261 
2262 #define	KS_UINIT(x)	kstat_named_init(&kstatp->x, #x, KSTAT_DATA_ULONG)
2263 #define	KS_CINIT(x)	kstat_named_init(&kstatp->x, #x, KSTAT_DATA_CHAR)
2264 #define	KS_U_SET(x, y)	kstatp->x.value.ul = (y)
2265 #define	KS_C_SET(x, ...)	\
2266 			(void) snprintf(kstatp->x.value.c, 16,  __VA_ARGS__)
2267 
2268 /*
2269  * t4nex:X:config
2270  */
2271 struct t4_kstats {
2272 	kstat_named_t chip_ver;
2273 	kstat_named_t fw_vers;
2274 	kstat_named_t tp_vers;
2275 	kstat_named_t driver_version;
2276 	kstat_named_t serial_number;
2277 	kstat_named_t ec_level;
2278 	kstat_named_t id;
2279 	kstat_named_t bus_type;
2280 	kstat_named_t bus_width;
2281 	kstat_named_t bus_speed;
2282 	kstat_named_t core_clock;
2283 	kstat_named_t port_cnt;
2284 	kstat_named_t port_type;
2285 	kstat_named_t pci_vendor_id;
2286 	kstat_named_t pci_device_id;
2287 };
2288 static kstat_t *
2289 setup_kstats(struct adapter *sc)
2290 {
2291 	kstat_t *ksp;
2292 	struct t4_kstats *kstatp;
2293 	int ndata;
2294 	struct pci_params *p = &sc->params.pci;
2295 	struct vpd_params *v = &sc->params.vpd;
2296 	uint16_t pci_vendor, pci_device;
2297 
2298 	ndata = sizeof (struct t4_kstats) / sizeof (kstat_named_t);
2299 
2300 	ksp = kstat_create(T4_NEXUS_NAME, ddi_get_instance(sc->dip), "config",
2301 	    "nexus", KSTAT_TYPE_NAMED, ndata, 0);
2302 	if (ksp == NULL) {
2303 		cxgb_printf(sc->dip, CE_WARN, "failed to initialize kstats.");
2304 		return (NULL);
2305 	}
2306 
2307 	kstatp = (struct t4_kstats *)ksp->ks_data;
2308 
2309 	KS_UINIT(chip_ver);
2310 	KS_CINIT(fw_vers);
2311 	KS_CINIT(tp_vers);
2312 	KS_CINIT(driver_version);
2313 	KS_CINIT(serial_number);
2314 	KS_CINIT(ec_level);
2315 	KS_CINIT(id);
2316 	KS_CINIT(bus_type);
2317 	KS_CINIT(bus_width);
2318 	KS_CINIT(bus_speed);
2319 	KS_UINIT(core_clock);
2320 	KS_UINIT(port_cnt);
2321 	KS_CINIT(port_type);
2322 	KS_CINIT(pci_vendor_id);
2323 	KS_CINIT(pci_device_id);
2324 
2325 	KS_U_SET(chip_ver, sc->params.chip);
2326 	KS_C_SET(fw_vers, "%d.%d.%d.%d",
2327 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
2328 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
2329 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
2330 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
2331 	KS_C_SET(tp_vers, "%d.%d.%d.%d",
2332 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
2333 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
2334 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
2335 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
2336 	KS_C_SET(driver_version, DRV_VERSION);
2337 	KS_C_SET(serial_number, "%s", v->sn);
2338 	KS_C_SET(ec_level, "%s", v->ec);
2339 	KS_C_SET(id, "%s", v->id);
2340 	KS_C_SET(bus_type, "pci-express");
2341 	KS_C_SET(bus_width, "x%d lanes", p->width);
2342 	KS_C_SET(bus_speed, "%d", p->speed);
2343 	KS_U_SET(core_clock, v->cclk);
2344 	KS_U_SET(port_cnt, sc->params.nports);
2345 
2346 	t4_os_pci_read_cfg2(sc, PCI_CONF_VENID, &pci_vendor);
2347 	KS_C_SET(pci_vendor_id, "0x%x", pci_vendor);
2348 
2349 	t4_os_pci_read_cfg2(sc, PCI_CONF_DEVID, &pci_device);
2350 	KS_C_SET(pci_device_id, "0x%x", pci_device);
2351 
2352 	KS_C_SET(port_type, "%s/%s/%s/%s",
2353 		 print_port_speed(sc->port[0]),
2354 		 print_port_speed(sc->port[1]),
2355 		 print_port_speed(sc->port[2]),
2356 		 print_port_speed(sc->port[3]));
2357 
2358 	/* Do NOT set ksp->ks_update.  These kstats do not change. */
2359 
2360 	/* Install the kstat */
2361 	ksp->ks_private = (void *)sc;
2362 	kstat_install(ksp);
2363 
2364 	return (ksp);
2365 }
2366 
2367 /*
2368  * t4nex:X:stat
2369  */
2370 struct t4_wc_kstats {
2371 	kstat_named_t write_coal_success;
2372 	kstat_named_t write_coal_failure;
2373 };
2374 static kstat_t *
2375 setup_wc_kstats(struct adapter *sc)
2376 {
2377 	kstat_t *ksp;
2378 	struct t4_wc_kstats *kstatp;
2379 	int ndata;
2380 
2381 	ndata = sizeof(struct t4_wc_kstats) / sizeof(kstat_named_t);
2382 	ksp = kstat_create(T4_NEXUS_NAME, ddi_get_instance(sc->dip), "stats",
2383 	    "nexus", KSTAT_TYPE_NAMED, ndata, 0);
2384 	if (ksp == NULL) {
2385 		cxgb_printf(sc->dip, CE_WARN, "failed to initialize kstats.");
2386 		return (NULL);
2387 	}
2388 
2389 	kstatp = (struct t4_wc_kstats *)ksp->ks_data;
2390 
2391 	KS_UINIT(write_coal_success);
2392 	KS_UINIT(write_coal_failure);
2393 
2394 	ksp->ks_update = update_wc_kstats;
2395 	/* Install the kstat */
2396 	ksp->ks_private = (void *)sc;
2397 	kstat_install(ksp);
2398 
2399 	return (ksp);
2400 }
2401 
2402 static int
2403 update_wc_kstats(kstat_t *ksp, int rw)
2404 {
2405 	struct t4_wc_kstats *kstatp = (struct t4_wc_kstats *)ksp->ks_data;
2406 	struct adapter *sc = ksp->ks_private;
2407 	uint32_t wc_total, wc_success, wc_failure;
2408 
2409 	if (rw == KSTAT_WRITE)
2410 		return (0);
2411 
2412 	if (is_t5(sc->params.chip)) {
2413 		wc_total = t4_read_reg(sc, A_SGE_STAT_TOTAL);
2414 		wc_failure = t4_read_reg(sc, A_SGE_STAT_MATCH);
2415 		wc_success = wc_total - wc_failure;
2416 	} else {
2417 		wc_success = 0;
2418 		wc_failure = 0;
2419 	}
2420 
2421 	KS_U_SET(write_coal_success, wc_success);
2422 	KS_U_SET(write_coal_failure, wc_failure);
2423 
2424 	return (0);
2425 }
2426 
2427 int
2428 adapter_full_init(struct adapter *sc)
2429 {
2430 	int i, rc = 0;
2431 
2432 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2433 
2434 	rc = t4_setup_adapter_queues(sc);
2435 	if (rc != 0)
2436 		goto done;
2437 
2438 	if (sc->intr_cap & DDI_INTR_FLAG_BLOCK)
2439 		(void) ddi_intr_block_enable(sc->intr_handle, sc->intr_count);
2440 	else {
2441 		for (i = 0; i < sc->intr_count; i++)
2442 			(void) ddi_intr_enable(sc->intr_handle[i]);
2443 	}
2444 	t4_intr_enable(sc);
2445 	sc->flags |= FULL_INIT_DONE;
2446 
2447 #ifdef TCP_OFFLOAD_ENABLE
2448 	/* TODO: wrong place to enable TOE capability */
2449 	if (is_offload(sc) != 0) {
2450 		for_each_port(sc, i) {
2451 			struct port_info *pi = sc->port[i];
2452 			rc = toe_capability(pi, 1);
2453 			if (rc != 0) {
2454 				cxgb_printf(pi->dip, CE_WARN,
2455 				    "Failed to activate toe capability: %d",
2456 				    rc);
2457 				rc = 0;		/* not a fatal error */
2458 			}
2459 		}
2460 	}
2461 #endif
2462 
2463 done:
2464 	if (rc != 0)
2465 		(void) adapter_full_uninit(sc);
2466 
2467 	return (rc);
2468 }
2469 
2470 int
2471 adapter_full_uninit(struct adapter *sc)
2472 {
2473 	int i, rc = 0;
2474 
2475 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2476 
2477 	if (sc->intr_cap & DDI_INTR_FLAG_BLOCK)
2478 		(void) ddi_intr_block_disable(sc->intr_handle, sc->intr_count);
2479 	else {
2480 		for (i = 0; i < sc->intr_count; i++)
2481 			(void) ddi_intr_disable(sc->intr_handle[i]);
2482 	}
2483 
2484 	rc = t4_teardown_adapter_queues(sc);
2485 	if (rc != 0)
2486 		return (rc);
2487 
2488 	sc->flags &= ~FULL_INIT_DONE;
2489 
2490 	return (0);
2491 }
2492 
2493 int
2494 port_full_init(struct port_info *pi)
2495 {
2496 	struct adapter *sc = pi->adapter;
2497 	uint16_t *rss;
2498 	struct sge_rxq *rxq;
2499 	int rc, i;
2500 
2501 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2502 	ASSERT((pi->flags & PORT_INIT_DONE) == 0);
2503 
2504 	/*
2505 	 * Allocate tx/rx/fl queues for this port.
2506 	 */
2507 	rc = t4_setup_port_queues(pi);
2508 	if (rc != 0)
2509 		goto done;	/* error message displayed already */
2510 
2511 	/*
2512 	 * Setup RSS for this port.
2513 	 */
2514 	rss = kmem_zalloc(pi->nrxq * sizeof (*rss), KM_SLEEP);
2515 	for_each_rxq(pi, i, rxq) {
2516 		rss[i] = rxq->iq.abs_id;
2517 	}
2518 	rc = -t4_config_rss_range(sc, sc->mbox, pi->viid, 0,
2519 	    pi->rss_size, rss, pi->nrxq);
2520 	kmem_free(rss, pi->nrxq * sizeof (*rss));
2521 	if (rc != 0) {
2522 		cxgb_printf(pi->dip, CE_WARN, "rss_config failed: %d", rc);
2523 		goto done;
2524 	}
2525 
2526 	pi->flags |= PORT_INIT_DONE;
2527 done:
2528 	if (rc != 0)
2529 		(void) port_full_uninit(pi);
2530 
2531 	return (rc);
2532 }
2533 
2534 /*
2535  * Idempotent.
2536  */
2537 int
2538 port_full_uninit(struct port_info *pi)
2539 {
2540 
2541 	ASSERT(pi->flags & PORT_INIT_DONE);
2542 
2543 	(void) t4_teardown_port_queues(pi);
2544 	pi->flags &= ~PORT_INIT_DONE;
2545 
2546 	return (0);
2547 }
2548 
2549 void
2550 enable_port_queues(struct port_info *pi)
2551 {
2552 	struct adapter *sc = pi->adapter;
2553 	int i;
2554 	struct sge_iq *iq;
2555 	struct sge_rxq *rxq;
2556 #ifdef TCP_OFFLOAD_ENABLE
2557 	struct sge_ofld_rxq *ofld_rxq;
2558 #endif
2559 
2560 	ASSERT(pi->flags & PORT_INIT_DONE);
2561 
2562 	/*
2563 	 * TODO: whatever was queued up after we set iq->state to IQS_DISABLED
2564 	 * back in disable_port_queues will be processed now, after an unbounded
2565 	 * delay.  This can't be good.
2566 	 */
2567 
2568 #ifdef TCP_OFFLOAD_ENABLE
2569 	for_each_ofld_rxq(pi, i, ofld_rxq) {
2570 		iq = &ofld_rxq->iq;
2571 		if (atomic_cas_uint(&iq->state, IQS_DISABLED, IQS_IDLE) !=
2572 		    IQS_DISABLED)
2573 			panic("%s: iq %p wasn't disabled", __func__,
2574 			    (void *)iq);
2575 		t4_write_reg(sc, MYPF_REG(A_SGE_PF_GTS),
2576 		    V_SEINTARM(iq->intr_params) | V_INGRESSQID(iq->cntxt_id));
2577 	}
2578 #endif
2579 
2580 	for_each_rxq(pi, i, rxq) {
2581 		iq = &rxq->iq;
2582 		if (atomic_cas_uint(&iq->state, IQS_DISABLED, IQS_IDLE) !=
2583 		    IQS_DISABLED)
2584 			panic("%s: iq %p wasn't disabled", __func__,
2585 			    (void *) iq);
2586 		t4_write_reg(sc, MYPF_REG(A_SGE_PF_GTS),
2587 		    V_SEINTARM(iq->intr_params) | V_INGRESSQID(iq->cntxt_id));
2588 	}
2589 }
2590 
2591 void
2592 disable_port_queues(struct port_info *pi)
2593 {
2594 	int i;
2595 	struct adapter *sc = pi->adapter;
2596 	struct sge_rxq *rxq;
2597 #ifdef TCP_OFFLOAD_ENABLE
2598 	struct sge_ofld_rxq *ofld_rxq;
2599 #endif
2600 
2601 	ASSERT(pi->flags & PORT_INIT_DONE);
2602 
2603 	/*
2604 	 * TODO: need proper implementation for all tx queues (ctrl, eth, ofld).
2605 	 */
2606 
2607 #ifdef TCP_OFFLOAD_ENABLE
2608 	for_each_ofld_rxq(pi, i, ofld_rxq) {
2609 		while (atomic_cas_uint(&ofld_rxq->iq.state, IQS_IDLE,
2610 		    IQS_DISABLED) != IQS_IDLE)
2611 			msleep(1);
2612 	}
2613 #endif
2614 
2615 	for_each_rxq(pi, i, rxq) {
2616 		while (atomic_cas_uint(&rxq->iq.state, IQS_IDLE,
2617 		    IQS_DISABLED) != IQS_IDLE)
2618 			msleep(1);
2619 	}
2620 
2621 	mutex_enter(&sc->sfl_lock);
2622 #ifdef TCP_OFFLOAD_ENABLE
2623 	for_each_ofld_rxq(pi, i, ofld_rxq)
2624 	    ofld_rxq->fl.flags |= FL_DOOMED;
2625 #endif
2626 	for_each_rxq(pi, i, rxq)
2627 	    rxq->fl.flags |= FL_DOOMED;
2628 	mutex_exit(&sc->sfl_lock);
2629 	/* TODO: need to wait for all fl's to be removed from sc->sfl */
2630 }
2631 
2632 void
2633 t4_fatal_err(struct adapter *sc)
2634 {
2635 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2636 	t4_intr_disable(sc);
2637 	cxgb_printf(sc->dip, CE_WARN,
2638 	    "encountered fatal error, adapter stopped.");
2639 }
2640 
2641 int
2642 t4_os_find_pci_capability(struct adapter *sc, int cap)
2643 {
2644 	uint16_t stat;
2645 	uint8_t cap_ptr, cap_id;
2646 
2647 	t4_os_pci_read_cfg2(sc, PCI_CONF_STAT, &stat);
2648 	if ((stat & PCI_STAT_CAP) == 0)
2649 		return (0); /* does not implement capabilities */
2650 
2651 	t4_os_pci_read_cfg1(sc, PCI_CONF_CAP_PTR, &cap_ptr);
2652 	while (cap_ptr) {
2653 		t4_os_pci_read_cfg1(sc, cap_ptr + PCI_CAP_ID, &cap_id);
2654 		if (cap_id == cap)
2655 			return (cap_ptr); /* found */
2656 		t4_os_pci_read_cfg1(sc, cap_ptr + PCI_CAP_NEXT_PTR, &cap_ptr);
2657 	}
2658 
2659 	return (0); /* not found */
2660 }
2661 
2662 void
2663 t4_os_portmod_changed(const struct adapter *sc, int idx)
2664 {
2665 	static const char *mod_str[] = {
2666 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
2667 	};
2668 	const struct port_info *pi = sc->port[idx];
2669 
2670 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
2671 		cxgb_printf(pi->dip, CE_NOTE, "transceiver unplugged.");
2672 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
2673 		cxgb_printf(pi->dip, CE_NOTE,
2674 		    "unknown transceiver inserted.\n");
2675 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
2676 		cxgb_printf(pi->dip, CE_NOTE,
2677 		    "unsupported transceiver inserted.\n");
2678 	else if (pi->mod_type > 0 && pi->mod_type < ARRAY_SIZE(mod_str))
2679 		cxgb_printf(pi->dip, CE_NOTE, "%s transceiver inserted.\n",
2680 		    mod_str[pi->mod_type]);
2681 	else
2682 		cxgb_printf(pi->dip, CE_NOTE, "transceiver (type %d) inserted.",
2683 		    pi->mod_type);
2684 }
2685 
2686 /* ARGSUSED */
2687 static int
2688 cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss, mblk_t *m)
2689 {
2690 	if (m != NULL)
2691 		freemsg(m);
2692 	return (0);
2693 }
2694 
2695 int
2696 t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
2697 {
2698 	uint_t *loc, new;
2699 
2700 	if (opcode >= ARRAY_SIZE(sc->cpl_handler))
2701 		return (EINVAL);
2702 
2703 	new = (uint_t)(unsigned long) (h ? h : cpl_not_handled);
2704 	loc = (uint_t *)&sc->cpl_handler[opcode];
2705 	(void) atomic_swap_uint(loc, new);
2706 
2707 	return (0);
2708 }
2709 
2710 static int
2711 fw_msg_not_handled(struct adapter *sc, const __be64 *data)
2712 {
2713 	struct cpl_fw6_msg *cpl = container_of(data, struct cpl_fw6_msg, data);
2714 
2715 	cxgb_printf(sc->dip, CE_WARN, "%s fw_msg type %d", __func__, cpl->type);
2716 	return (0);
2717 }
2718 
2719 int
2720 t4_register_fw_msg_handler(struct adapter *sc, int type, fw_msg_handler_t h)
2721 {
2722 	fw_msg_handler_t *loc, new;
2723 
2724 	if (type >= ARRAY_SIZE(sc->fw_msg_handler))
2725 		return (EINVAL);
2726 
2727 	/*
2728 	 * These are dispatched by the handler for FW{4|6}_CPL_MSG using the CPL
2729 	 * handler dispatch table.  Reject any attempt to install a handler for
2730 	 * this subtype.
2731 	 */
2732 	if (type == FW_TYPE_RSSCPL || type == FW6_TYPE_RSSCPL)
2733 		return (EINVAL);
2734 
2735 	new = h ? h : fw_msg_not_handled;
2736 	loc = &sc->fw_msg_handler[type];
2737 	(void)atomic_swap_ptr(loc, (void *)new);
2738 
2739 	return (0);
2740 }
2741 
2742 #ifdef TCP_OFFLOAD_ENABLE
2743 static int
2744 toe_capability(struct port_info *pi, int enable)
2745 {
2746 	int rc;
2747 	struct adapter *sc = pi->adapter;
2748 
2749 	if (!is_offload(sc))
2750 		return (ENODEV);
2751 
2752 	if (enable != 0) {
2753 		if (isset(&sc->offload_map, pi->port_id) != 0)
2754 			return (0);
2755 
2756 		if (sc->offload_map == 0) {
2757 			rc = activate_uld(sc, ULD_TOM, &sc->tom);
2758 			if (rc != 0)
2759 				return (rc);
2760 		}
2761 
2762 		setbit(&sc->offload_map, pi->port_id);
2763 	} else {
2764 		if (!isset(&sc->offload_map, pi->port_id))
2765 			return (0);
2766 
2767 		clrbit(&sc->offload_map, pi->port_id);
2768 
2769 		if (sc->offload_map == 0) {
2770 			rc = deactivate_uld(&sc->tom);
2771 			if (rc != 0) {
2772 				setbit(&sc->offload_map, pi->port_id);
2773 				return (rc);
2774 			}
2775 		}
2776 	}
2777 
2778 	return (0);
2779 }
2780 
2781 /*
2782  * Add an upper layer driver to the global list.
2783  */
2784 int
2785 t4_register_uld(struct uld_info *ui)
2786 {
2787 	int rc = 0;
2788 	struct uld_info *u;
2789 
2790 	mutex_enter(&t4_uld_list_lock);
2791 	SLIST_FOREACH(u, &t4_uld_list, link) {
2792 		if (u->uld_id == ui->uld_id) {
2793 			rc = EEXIST;
2794 			goto done;
2795 		}
2796 	}
2797 
2798 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
2799 	ui->refcount = 0;
2800 done:
2801 	mutex_exit(&t4_uld_list_lock);
2802 	return (rc);
2803 }
2804 
2805 int
2806 t4_unregister_uld(struct uld_info *ui)
2807 {
2808 	int rc = EINVAL;
2809 	struct uld_info *u;
2810 
2811 	mutex_enter(&t4_uld_list_lock);
2812 
2813 	SLIST_FOREACH(u, &t4_uld_list, link) {
2814 		if (u == ui) {
2815 			if (ui->refcount > 0) {
2816 				rc = EBUSY;
2817 				goto done;
2818 			}
2819 
2820 			SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
2821 			rc = 0;
2822 			goto done;
2823 		}
2824 	}
2825 done:
2826 	mutex_exit(&t4_uld_list_lock);
2827 	return (rc);
2828 }
2829 
2830 static int
2831 activate_uld(struct adapter *sc, int id, struct uld_softc *usc)
2832 {
2833 	int rc = EAGAIN;
2834 	struct uld_info *ui;
2835 
2836 	mutex_enter(&t4_uld_list_lock);
2837 
2838 	SLIST_FOREACH(ui, &t4_uld_list, link) {
2839 		if (ui->uld_id == id) {
2840 			rc = ui->attach(sc, &usc->softc);
2841 			if (rc == 0) {
2842 				ASSERT(usc->softc != NULL);
2843 				ui->refcount++;
2844 				usc->uld = ui;
2845 			}
2846 			goto done;
2847 		}
2848 	}
2849 done:
2850 	mutex_exit(&t4_uld_list_lock);
2851 
2852 	return (rc);
2853 }
2854 
2855 static int
2856 deactivate_uld(struct uld_softc *usc)
2857 {
2858 	int rc;
2859 
2860 	mutex_enter(&t4_uld_list_lock);
2861 
2862 	if (usc->uld == NULL || usc->softc == NULL) {
2863 		rc = EINVAL;
2864 		goto done;
2865 	}
2866 
2867 	rc = usc->uld->detach(usc->softc);
2868 	if (rc == 0) {
2869 		ASSERT(usc->uld->refcount > 0);
2870 		usc->uld->refcount--;
2871 		usc->uld = NULL;
2872 		usc->softc = NULL;
2873 	}
2874 done:
2875 	mutex_exit(&t4_uld_list_lock);
2876 
2877 	return (rc);
2878 }
2879 
2880 void
2881 t4_iterate(void (*func)(int, void *), void *arg)
2882 {
2883 	struct adapter *sc;
2884 
2885 	mutex_enter(&t4_adapter_list_lock);
2886 	SLIST_FOREACH(sc, &t4_adapter_list, link) {
2887 		/*
2888 		 * func should not make any assumptions about what state sc is
2889 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
2890 		 */
2891 		func(ddi_get_instance(sc->dip), arg);
2892 	}
2893 	mutex_exit(&t4_adapter_list_lock);
2894 }
2895 
2896 #endif
2897