xref: /illumos-gate/usr/src/uts/common/io/cxgbe/t4nex/t4_nexus.c (revision 3aa6c13072f3d4792a18693e916aed260a496c1f)
1 /*
2  * This file and its contents are supplied under the terms of the
3  * Common Development and Distribution License ("CDDL"), version 1.0.
4  * You may only use this file in accordance with the terms of version
5  * 1.0 of the CDDL.
6  *
7  * A full copy of the text of the CDDL should have accompanied this
8  * source. A copy of the CDDL is also available via the Internet at
9  * http://www.illumos.org/license/CDDL.
10  */
11 
12 /*
13  * This file is part of the Chelsio T4 support code.
14  *
15  * Copyright (C) 2010-2013 Chelsio Communications.  All rights reserved.
16  *
17  * This program is distributed in the hope that it will be useful, but WITHOUT
18  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
19  * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
20  * release for licensing terms and conditions.
21  */
22 
23 #include <sys/ddi.h>
24 #include <sys/sunddi.h>
25 #include <sys/sunndi.h>
26 #include <sys/modctl.h>
27 #include <sys/conf.h>
28 #include <sys/devops.h>
29 #include <sys/pci.h>
30 #include <sys/atomic.h>
31 #include <sys/types.h>
32 #include <sys/file.h>
33 #include <sys/errno.h>
34 #include <sys/open.h>
35 #include <sys/cred.h>
36 #include <sys/stat.h>
37 #include <sys/mkdev.h>
38 #include <sys/queue.h>
39 #include <sys/containerof.h>
40 #include <sys/sensors.h>
41 
42 #include "version.h"
43 #include "common/common.h"
44 #include "common/t4_msg.h"
45 #include "common/t4_regs.h"
46 #include "firmware/t4_fw.h"
47 #include "firmware/t4_cfg.h"
48 #include "firmware/t5_fw.h"
49 #include "firmware/t5_cfg.h"
50 #include "firmware/t6_fw.h"
51 #include "firmware/t6_cfg.h"
52 #include "t4_l2t.h"
53 
54 static int t4_cb_open(dev_t *devp, int flag, int otyp, cred_t *credp);
55 static int t4_cb_close(dev_t dev, int flag, int otyp, cred_t *credp);
56 static int t4_cb_ioctl(dev_t dev, int cmd, intptr_t d, int mode, cred_t *credp,
57     int *rp);
58 struct cb_ops t4_cb_ops = {
59 	.cb_open =		t4_cb_open,
60 	.cb_close =		t4_cb_close,
61 	.cb_strategy =		nodev,
62 	.cb_print =		nodev,
63 	.cb_dump =		nodev,
64 	.cb_read =		nodev,
65 	.cb_write =		nodev,
66 	.cb_ioctl =		t4_cb_ioctl,
67 	.cb_devmap =		nodev,
68 	.cb_mmap =		nodev,
69 	.cb_segmap =		nodev,
70 	.cb_chpoll =		nochpoll,
71 	.cb_prop_op =		ddi_prop_op,
72 	.cb_flag =		D_MP,
73 	.cb_rev =		CB_REV,
74 	.cb_aread =		nodev,
75 	.cb_awrite =		nodev
76 };
77 
78 static int t4_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t op,
79     void *arg, void *result);
80 static int t4_bus_config(dev_info_t *dip, uint_t flags, ddi_bus_config_op_t op,
81     void *arg, dev_info_t **cdipp);
82 static int t4_bus_unconfig(dev_info_t *dip, uint_t flags,
83     ddi_bus_config_op_t op, void *arg);
84 struct bus_ops t4_bus_ops = {
85 	.busops_rev =		BUSO_REV,
86 	.bus_ctl =		t4_bus_ctl,
87 	.bus_prop_op =		ddi_bus_prop_op,
88 	.bus_config =		t4_bus_config,
89 	.bus_unconfig =		t4_bus_unconfig,
90 };
91 
92 static int t4_devo_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg,
93     void **rp);
94 static int t4_devo_probe(dev_info_t *dip);
95 static int t4_devo_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
96 static int t4_devo_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
97 static int t4_devo_quiesce(dev_info_t *dip);
98 struct dev_ops t4_dev_ops = {
99 	.devo_rev =		DEVO_REV,
100 	.devo_getinfo =		t4_devo_getinfo,
101 	.devo_identify =	nulldev,
102 	.devo_probe =		t4_devo_probe,
103 	.devo_attach =		t4_devo_attach,
104 	.devo_detach =		t4_devo_detach,
105 	.devo_reset =		nodev,
106 	.devo_cb_ops =		&t4_cb_ops,
107 	.devo_bus_ops =		&t4_bus_ops,
108 	.devo_quiesce =		&t4_devo_quiesce,
109 };
110 
111 static struct modldrv modldrv = {
112 	.drv_modops =		&mod_driverops,
113 	.drv_linkinfo =		"Chelsio T4 nexus " DRV_VERSION,
114 	.drv_dev_ops =		&t4_dev_ops
115 };
116 
117 static struct modlinkage modlinkage = {
118 	.ml_rev =		MODREV_1,
119 	.ml_linkage =		{&modldrv, NULL},
120 };
121 
122 void *t4_list;
123 
124 struct intrs_and_queues {
125 	int intr_type;		/* DDI_INTR_TYPE_* */
126 	int nirq;		/* Number of vectors */
127 	int intr_fwd;		/* Interrupts forwarded */
128 	int ntxq10g;		/* # of NIC txq's for each 10G port */
129 	int nrxq10g;		/* # of NIC rxq's for each 10G port */
130 	int ntxq1g;		/* # of NIC txq's for each 1G port */
131 	int nrxq1g;		/* # of NIC rxq's for each 1G port */
132 #ifdef TCP_OFFLOAD_ENABLE
133 	int nofldtxq10g;	/* # of TOE txq's for each 10G port */
134 	int nofldrxq10g;	/* # of TOE rxq's for each 10G port */
135 	int nofldtxq1g;		/* # of TOE txq's for each 1G port */
136 	int nofldrxq1g;		/* # of TOE rxq's for each 1G port */
137 #endif
138 };
139 
140 struct fw_info fi[3];
141 
142 static int cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss,
143     mblk_t *m);
144 static int fw_msg_not_handled(struct adapter *, const __be64 *);
145 int t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h);
146 static unsigned int getpf(struct adapter *sc);
147 static int prep_firmware(struct adapter *sc);
148 static int upload_config_file(struct adapter *sc, uint32_t *mt, uint32_t *ma);
149 static int partition_resources(struct adapter *sc);
150 static int adap__pre_init_tweaks(struct adapter *sc);
151 static int get_params__pre_init(struct adapter *sc);
152 static int get_params__post_init(struct adapter *sc);
153 static int set_params__post_init(struct adapter *);
154 static void setup_memwin(struct adapter *sc);
155 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
156     uint32_t *);
157 void memwin_info(struct adapter *, int, uint32_t *, uint32_t *);
158 uint32_t position_memwin(struct adapter *, int, uint32_t);
159 static int prop_lookup_int_array(struct adapter *sc, char *name, int *data,
160     uint_t count);
161 static int prop_lookup_int_array(struct adapter *sc, char *name, int *data,
162     uint_t count);
163 static int init_driver_props(struct adapter *sc, struct driver_properties *p);
164 static int remove_extra_props(struct adapter *sc, int n10g, int n1g);
165 static int cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g,
166     struct intrs_and_queues *iaq);
167 static int add_child_node(struct adapter *sc, int idx);
168 static int remove_child_node(struct adapter *sc, int idx);
169 static kstat_t *setup_kstats(struct adapter *sc);
170 static kstat_t *setup_wc_kstats(struct adapter *);
171 static int update_wc_kstats(kstat_t *, int);
172 #ifdef TCP_OFFLOAD_ENABLE
173 static int toe_capability(struct port_info *pi, int enable);
174 static int activate_uld(struct adapter *sc, int id, struct uld_softc *usc);
175 static int deactivate_uld(struct uld_softc *usc);
176 #endif
177 static kmutex_t t4_adapter_list_lock;
178 static SLIST_HEAD(, adapter) t4_adapter_list;
179 #ifdef TCP_OFFLOAD_ENABLE
180 static kmutex_t t4_uld_list_lock;
181 static SLIST_HEAD(, uld_info) t4_uld_list;
182 #endif
183 
184 static int t4_temperature_read(void *, sensor_ioctl_scalar_t *);
185 static int t4_voltage_read(void *, sensor_ioctl_scalar_t *);
186 static const ksensor_ops_t t4_temp_ops = {
187 	.kso_kind = ksensor_kind_temperature,
188 	.kso_scalar = t4_temperature_read
189 };
190 
191 static const ksensor_ops_t t4_volt_ops = {
192 	.kso_kind = ksensor_kind_voltage,
193 	.kso_scalar = t4_voltage_read
194 };
195 
196 static int t4_ufm_getcaps(ddi_ufm_handle_t *, void *, ddi_ufm_cap_t *);
197 static int t4_ufm_fill_image(ddi_ufm_handle_t *, void *, uint_t,
198     ddi_ufm_image_t *);
199 static int t4_ufm_fill_slot(ddi_ufm_handle_t *, void *, uint_t, uint_t,
200     ddi_ufm_slot_t *);
201 static ddi_ufm_ops_t t4_ufm_ops = {
202 	.ddi_ufm_op_fill_image = t4_ufm_fill_image,
203 	.ddi_ufm_op_fill_slot = t4_ufm_fill_slot,
204 	.ddi_ufm_op_getcaps = t4_ufm_getcaps
205 };
206 
207 int
208 _init(void)
209 {
210 	int rc;
211 
212 	rc = ddi_soft_state_init(&t4_list, sizeof (struct adapter), 0);
213 	if (rc != 0)
214 		return (rc);
215 
216 	rc = mod_install(&modlinkage);
217 	if (rc != 0)
218 		ddi_soft_state_fini(&t4_list);
219 
220 	mutex_init(&t4_adapter_list_lock, NULL, MUTEX_DRIVER, NULL);
221 	SLIST_INIT(&t4_adapter_list);
222 
223 #ifdef TCP_OFFLOAD_ENABLE
224 	mutex_init(&t4_uld_list_lock, NULL, MUTEX_DRIVER, NULL);
225 	SLIST_INIT(&t4_uld_list);
226 #endif
227 
228 	return (rc);
229 }
230 
231 int
232 _fini(void)
233 {
234 	int rc;
235 
236 	rc = mod_remove(&modlinkage);
237 	if (rc != 0)
238 		return (rc);
239 
240 	ddi_soft_state_fini(&t4_list);
241 	return (0);
242 }
243 
244 int
245 _info(struct modinfo *mi)
246 {
247 	return (mod_info(&modlinkage, mi));
248 }
249 
250 /* ARGSUSED */
251 static int
252 t4_devo_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **rp)
253 {
254 	struct adapter *sc;
255 	minor_t minor;
256 
257 	minor = getminor((dev_t)arg);	/* same as instance# in our case */
258 
259 	if (cmd == DDI_INFO_DEVT2DEVINFO) {
260 		sc = ddi_get_soft_state(t4_list, minor);
261 		if (sc == NULL)
262 			return (DDI_FAILURE);
263 
264 		ASSERT(sc->dev == (dev_t)arg);
265 		*rp = (void *)sc->dip;
266 	} else if (cmd == DDI_INFO_DEVT2INSTANCE)
267 		*rp = (void *) (unsigned long) minor;
268 	else
269 		ASSERT(0);
270 
271 	return (DDI_SUCCESS);
272 }
273 
274 static int
275 t4_devo_probe(dev_info_t *dip)
276 {
277 	int rc, id, *reg;
278 	uint_t n, pf;
279 
280 	id = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
281 	    "device-id", 0xffff);
282 	if (id == 0xffff)
283 		return (DDI_PROBE_DONTCARE);
284 
285 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
286 	    "reg", &reg, &n);
287 	if (rc != DDI_SUCCESS)
288 		return (DDI_PROBE_DONTCARE);
289 
290 	pf = PCI_REG_FUNC_G(reg[0]);
291 	ddi_prop_free(reg);
292 
293 	/* Prevent driver attachment on any PF except 0 on the FPGA */
294 	if (id == 0xa000 && pf != 0)
295 		return (DDI_PROBE_FAILURE);
296 
297 	return (DDI_PROBE_DONTCARE);
298 }
299 
300 static int
301 t4_devo_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
302 {
303 	struct adapter *sc = NULL;
304 	struct sge *s;
305 	int i, instance, rc = DDI_SUCCESS, rqidx, tqidx, q;
306 	int irq = 0, nxg, n100g, n40g, n25g, n10g, n1g;
307 #ifdef TCP_OFFLOAD_ENABLE
308 	int ofld_rqidx, ofld_tqidx;
309 #endif
310 	char name[16];
311 	struct driver_properties *prp;
312 	struct intrs_and_queues iaq;
313 	ddi_device_acc_attr_t da = {
314 		.devacc_attr_version = DDI_DEVICE_ATTR_V0,
315 		.devacc_attr_endian_flags = DDI_STRUCTURE_LE_ACC,
316 		.devacc_attr_dataorder = DDI_UNORDERED_OK_ACC
317 	};
318 	ddi_device_acc_attr_t da1 = {
319 		.devacc_attr_version = DDI_DEVICE_ATTR_V0,
320 		.devacc_attr_endian_flags = DDI_STRUCTURE_LE_ACC,
321 		.devacc_attr_dataorder = DDI_MERGING_OK_ACC
322 	};
323 
324 	if (cmd != DDI_ATTACH)
325 		return (DDI_FAILURE);
326 
327 	/*
328 	 * Allocate space for soft state.
329 	 */
330 	instance = ddi_get_instance(dip);
331 	rc = ddi_soft_state_zalloc(t4_list, instance);
332 	if (rc != DDI_SUCCESS) {
333 		cxgb_printf(dip, CE_WARN,
334 		    "failed to allocate soft state: %d", rc);
335 		return (DDI_FAILURE);
336 	}
337 
338 	sc = ddi_get_soft_state(t4_list, instance);
339 	sc->dip = dip;
340 	sc->dev = makedevice(ddi_driver_major(dip), instance);
341 	mutex_init(&sc->lock, NULL, MUTEX_DRIVER, NULL);
342 	cv_init(&sc->cv, NULL, CV_DRIVER, NULL);
343 	mutex_init(&sc->sfl_lock, NULL, MUTEX_DRIVER, NULL);
344 	TAILQ_INIT(&sc->sfl);
345 
346 	mutex_enter(&t4_adapter_list_lock);
347 	SLIST_INSERT_HEAD(&t4_adapter_list, sc, link);
348 	mutex_exit(&t4_adapter_list_lock);
349 
350 	sc->pf = getpf(sc);
351 	if (sc->pf > 8) {
352 		rc = EINVAL;
353 		cxgb_printf(dip, CE_WARN,
354 		    "failed to determine PCI PF# of device");
355 		goto done;
356 	}
357 	sc->mbox = sc->pf;
358 
359 	/* Initialize the driver properties */
360 	prp = &sc->props;
361 	(void)init_driver_props(sc, prp);
362 
363 	/*
364 	 * Enable access to the PCI config space.
365 	 */
366 	rc = pci_config_setup(dip, &sc->pci_regh);
367 	if (rc != DDI_SUCCESS) {
368 		cxgb_printf(dip, CE_WARN,
369 		    "failed to enable PCI config space access: %d", rc);
370 		goto done;
371 	}
372 
373 	/* TODO: Set max read request to 4K */
374 
375 	/*
376 	 * Enable MMIO access.
377 	 */
378 	rc = ddi_regs_map_setup(dip, 1, &sc->regp, 0, 0, &da, &sc->regh);
379 	if (rc != DDI_SUCCESS) {
380 		cxgb_printf(dip, CE_WARN,
381 		    "failed to map device registers: %d", rc);
382 		goto done;
383 	}
384 
385 	(void) memset(sc->chan_map, 0xff, sizeof (sc->chan_map));
386 
387 	/*
388 	 * Initialize cpl handler.
389 	 */
390 	for (i = 0; i < ARRAY_SIZE(sc->cpl_handler); i++) {
391 		sc->cpl_handler[i] = cpl_not_handled;
392 	}
393 
394 	for (i = 0; i < ARRAY_SIZE(sc->fw_msg_handler); i++) {
395 		sc->fw_msg_handler[i] = fw_msg_not_handled;
396 	}
397 
398 	for (i = 0; i < NCHAN; i++) {
399 		(void) snprintf(name, sizeof (name), "%s-%d",
400 				"reclaim", i);
401 		sc->tq[i] = ddi_taskq_create(sc->dip,
402 		   name, 1, TASKQ_DEFAULTPRI, 0);
403 
404 		if (sc->tq[i] == NULL) {
405 			cxgb_printf(dip, CE_WARN,
406 				   "failed to create task queues");
407 			rc = DDI_FAILURE;
408 			goto done;
409 		}
410 	}
411 
412 	/*
413 	 * Prepare the adapter for operation.
414 	 */
415 	rc = -t4_prep_adapter(sc, false);
416 	if (rc != 0) {
417 		cxgb_printf(dip, CE_WARN, "failed to prepare adapter: %d", rc);
418 		goto done;
419 	}
420 
421 	/*
422 	 * Enable BAR1 access.
423 	 */
424 	sc->doorbells |= DOORBELL_KDB;
425 	rc = ddi_regs_map_setup(dip, 2, &sc->reg1p, 0, 0, &da1, &sc->reg1h);
426 	if (rc != DDI_SUCCESS) {
427 		cxgb_printf(dip, CE_WARN,
428 		    "failed to map BAR1 device registers: %d", rc);
429 		goto done;
430 	} else {
431 		if (is_t5(sc->params.chip)) {
432 			sc->doorbells |= DOORBELL_UDB;
433 			if (prp->wc) {
434 				/*
435 				 * Enable write combining on BAR2.  This is the
436 				 * userspace doorbell BAR and is split into 128B
437 				 * (UDBS_SEG_SIZE) doorbell regions, each associated
438 				 * with an egress queue.  The first 64B has the doorbell
439 				 * and the second 64B can be used to submit a tx work
440 				 * request with an implicit doorbell.
441 				 */
442 				sc->doorbells &= ~DOORBELL_UDB;
443 				sc->doorbells |= (DOORBELL_WCWR |
444 				    DOORBELL_UDBWC);
445 				t4_write_reg(sc, A_SGE_STAT_CFG,
446 				    V_STATSOURCE_T5(7) | V_STATMODE(0));
447 			}
448 		}
449 	}
450 
451 	/*
452 	 * Do this really early.  Note that minor number = instance.
453 	 */
454 	(void) snprintf(name, sizeof (name), "%s,%d", T4_NEXUS_NAME, instance);
455 	rc = ddi_create_minor_node(dip, name, S_IFCHR, instance,
456 	    DDI_NT_NEXUS, 0);
457 	if (rc != DDI_SUCCESS) {
458 		cxgb_printf(dip, CE_WARN,
459 		    "failed to create device node: %d", rc);
460 		rc = DDI_SUCCESS; /* carry on */
461 	}
462 
463 	/* Do this early. Memory window is required for loading config file. */
464 	setup_memwin(sc);
465 
466 	/* Prepare the firmware for operation */
467 	rc = prep_firmware(sc);
468 	if (rc != 0)
469 		goto done; /* error message displayed already */
470 
471 	rc = adap__pre_init_tweaks(sc);
472 	if (rc != 0)
473 		goto done;
474 
475 	rc = get_params__pre_init(sc);
476 	if (rc != 0)
477 		goto done; /* error message displayed already */
478 
479 	t4_sge_init(sc);
480 
481 	if (sc->flags & MASTER_PF) {
482 		/* get basic stuff going */
483 		rc = -t4_fw_initialize(sc, sc->mbox);
484 		if (rc != 0) {
485 			cxgb_printf(sc->dip, CE_WARN,
486 			    "early init failed: %d.\n", rc);
487 			goto done;
488 		}
489 	}
490 
491 	rc = get_params__post_init(sc);
492 	if (rc != 0)
493 		goto done; /* error message displayed already */
494 
495 	rc = set_params__post_init(sc);
496 	if (rc != 0)
497 		goto done; /* error message displayed already */
498 
499 	/*
500 	 * TODO: This is the place to call t4_set_filter_mode()
501 	 */
502 
503 	/* tweak some settings */
504 	t4_write_reg(sc, A_TP_SHIFT_CNT, V_SYNSHIFTMAX(6) | V_RXTSHIFTMAXR1(4) |
505 	    V_RXTSHIFTMAXR2(15) | V_PERSHIFTBACKOFFMAX(8) | V_PERSHIFTMAX(8) |
506 	    V_KEEPALIVEMAXR1(4) | V_KEEPALIVEMAXR2(9));
507 	t4_write_reg(sc, A_ULP_RX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
508 
509 	/*
510 	 * Work-around for bug 2619
511 	 * Set DisableVlan field in TP_RSS_CONFIG_VRT register so that the
512 	 * VLAN tag extraction is disabled.
513 	 */
514 	t4_set_reg_field(sc, A_TP_RSS_CONFIG_VRT, F_DISABLEVLAN, F_DISABLEVLAN);
515 
516 	/* Store filter mode */
517 	t4_read_indirect(sc, A_TP_PIO_ADDR, A_TP_PIO_DATA, &sc->filter_mode, 1,
518 	    A_TP_VLAN_PRI_MAP);
519 
520 	/*
521 	 * First pass over all the ports - allocate VIs and initialize some
522 	 * basic parameters like mac address, port type, etc.  We also figure
523 	 * out whether a port is 10G or 1G and use that information when
524 	 * calculating how many interrupts to attempt to allocate.
525 	 */
526 	n100g = n40g = n25g = n10g = n1g = 0;
527 	for_each_port(sc, i) {
528 		struct port_info *pi;
529 
530 		pi = kmem_zalloc(sizeof (*pi), KM_SLEEP);
531 		sc->port[i] = pi;
532 
533 		/* These must be set before t4_port_init */
534 		pi->adapter = sc;
535 		/* LINTED: E_ASSIGN_NARROW_CONV */
536 		pi->port_id = i;
537 	}
538 
539 	/* Allocate the vi and initialize parameters like mac addr */
540 	rc = -t4_port_init(sc, sc->mbox, sc->pf, 0);
541 	if (rc) {
542 		cxgb_printf(dip, CE_WARN,
543 			    "unable to initialize port: %d", rc);
544 		goto done;
545 	}
546 
547 	for_each_port(sc, i) {
548 		struct port_info *pi = sc->port[i];
549 
550 		mutex_init(&pi->lock, NULL, MUTEX_DRIVER, NULL);
551 		pi->mtu = ETHERMTU;
552 
553 		if (is_100G_port(pi)) {
554 			n100g++;
555 			pi->tmr_idx = prp->tmr_idx_10g;
556 			pi->pktc_idx = prp->pktc_idx_10g;
557 		} else if (is_40G_port(pi)) {
558 			n40g++;
559 			pi->tmr_idx = prp->tmr_idx_10g;
560 			pi->pktc_idx = prp->pktc_idx_10g;
561 		} else if (is_25G_port(pi)) {
562 			n25g++;
563 			pi->tmr_idx = prp->tmr_idx_10g;
564 			pi->pktc_idx = prp->pktc_idx_10g;
565 		} else if (is_10G_port(pi)) {
566 			n10g++;
567 			pi->tmr_idx = prp->tmr_idx_10g;
568 			pi->pktc_idx = prp->pktc_idx_10g;
569 		} else {
570 			n1g++;
571 			pi->tmr_idx = prp->tmr_idx_1g;
572 			pi->pktc_idx = prp->pktc_idx_1g;
573 		}
574 
575 		pi->xact_addr_filt = -1;
576 		t4_mc_init(pi);
577 
578 		setbit(&sc->registered_device_map, i);
579 	}
580 
581 	nxg = n10g + n25g + n40g + n100g;
582 	(void) remove_extra_props(sc, nxg, n1g);
583 
584 	if (sc->registered_device_map == 0) {
585 		cxgb_printf(dip, CE_WARN, "no usable ports");
586 		rc = DDI_FAILURE;
587 		goto done;
588 	}
589 
590 	rc = cfg_itype_and_nqueues(sc, nxg, n1g, &iaq);
591 	if (rc != 0)
592 		goto done; /* error message displayed already */
593 
594 	sc->intr_type = iaq.intr_type;
595 	sc->intr_count = iaq.nirq;
596 
597 	if (sc->props.multi_rings && (sc->intr_type != DDI_INTR_TYPE_MSIX)) {
598 		sc->props.multi_rings = 0;
599 		cxgb_printf(dip, CE_WARN,
600 		    "Multiple rings disabled as interrupt type is not MSI-X");
601 	}
602 
603 	if (sc->props.multi_rings && iaq.intr_fwd) {
604 		sc->props.multi_rings = 0;
605 		cxgb_printf(dip, CE_WARN,
606 		    "Multiple rings disabled as interrupts are forwarded");
607 	}
608 
609 	if (!sc->props.multi_rings) {
610 		iaq.ntxq10g = 1;
611 		iaq.ntxq1g = 1;
612 	}
613 	s = &sc->sge;
614 	s->nrxq = nxg * iaq.nrxq10g + n1g * iaq.nrxq1g;
615 	s->ntxq = nxg * iaq.ntxq10g + n1g * iaq.ntxq1g;
616 	s->neq = s->ntxq + s->nrxq;	/* the fl in an rxq is an eq */
617 #ifdef TCP_OFFLOAD_ENABLE
618 	/* control queues, 1 per port + 1 mgmtq */
619 	s->neq += sc->params.nports + 1;
620 #endif
621 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
622 	if (iaq.intr_fwd != 0)
623 		sc->flags |= INTR_FWD;
624 #ifdef TCP_OFFLOAD_ENABLE
625 	if (is_offload(sc) != 0) {
626 
627 		s->nofldrxq = nxg * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
628 		s->nofldtxq = nxg * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
629 		s->neq += s->nofldtxq + s->nofldrxq;
630 		s->niq += s->nofldrxq;
631 
632 		s->ofld_rxq = kmem_zalloc(s->nofldrxq *
633 		    sizeof (struct sge_ofld_rxq), KM_SLEEP);
634 		s->ofld_txq = kmem_zalloc(s->nofldtxq *
635 		    sizeof (struct sge_wrq), KM_SLEEP);
636 		s->ctrlq = kmem_zalloc(sc->params.nports *
637 		    sizeof (struct sge_wrq), KM_SLEEP);
638 
639 	}
640 #endif
641 	s->rxq = kmem_zalloc(s->nrxq * sizeof (struct sge_rxq), KM_SLEEP);
642 	s->txq = kmem_zalloc(s->ntxq * sizeof (struct sge_txq), KM_SLEEP);
643 	s->iqmap = kmem_zalloc(s->niq * sizeof (struct sge_iq *), KM_SLEEP);
644 	s->eqmap = kmem_zalloc(s->neq * sizeof (struct sge_eq *), KM_SLEEP);
645 
646 	sc->intr_handle = kmem_zalloc(sc->intr_count *
647 	    sizeof (ddi_intr_handle_t), KM_SLEEP);
648 
649 	/*
650 	 * Second pass over the ports.  This time we know the number of rx and
651 	 * tx queues that each port should get.
652 	 */
653 	rqidx = tqidx = 0;
654 #ifdef TCP_OFFLOAD_ENABLE
655 	ofld_rqidx = ofld_tqidx = 0;
656 #endif
657 	for_each_port(sc, i) {
658 		struct port_info *pi = sc->port[i];
659 
660 		if (pi == NULL)
661 			continue;
662 
663 		t4_mc_cb_init(pi);
664 		/* LINTED: E_ASSIGN_NARROW_CONV */
665 		pi->first_rxq = rqidx;
666 		/* LINTED: E_ASSIGN_NARROW_CONV */
667 		pi->nrxq = (is_10XG_port(pi)) ? iaq.nrxq10g
668 		    : iaq.nrxq1g;
669 		/* LINTED: E_ASSIGN_NARROW_CONV */
670 		pi->first_txq = tqidx;
671 		/* LINTED: E_ASSIGN_NARROW_CONV */
672 		pi->ntxq = (is_10XG_port(pi)) ? iaq.ntxq10g
673 		    : iaq.ntxq1g;
674 
675 		rqidx += pi->nrxq;
676 		tqidx += pi->ntxq;
677 
678 #ifdef TCP_OFFLOAD_ENABLE
679 		if (is_offload(sc) != 0) {
680 			/* LINTED: E_ASSIGN_NARROW_CONV */
681 			pi->first_ofld_rxq = ofld_rqidx;
682 			pi->nofldrxq = max(1, pi->nrxq / 4);
683 
684 			/* LINTED: E_ASSIGN_NARROW_CONV */
685 			pi->first_ofld_txq = ofld_tqidx;
686 			pi->nofldtxq = max(1, pi->ntxq / 2);
687 
688 			ofld_rqidx += pi->nofldrxq;
689 			ofld_tqidx += pi->nofldtxq;
690 		}
691 #endif
692 
693 		/*
694 		 * Enable hw checksumming and LSO for all ports by default.
695 		 * They can be disabled using ndd (hw_csum and hw_lso).
696 		 */
697 		pi->features |= (CXGBE_HW_CSUM | CXGBE_HW_LSO);
698 	}
699 
700 #ifdef TCP_OFFLOAD_ENABLE
701 		sc->l2t = t4_init_l2t(sc);
702 #endif
703 
704 	/*
705 	 * Setup Interrupts.
706 	 */
707 
708 	i = 0;
709 	rc = ddi_intr_alloc(dip, sc->intr_handle, sc->intr_type, 0,
710 	    sc->intr_count, &i, DDI_INTR_ALLOC_STRICT);
711 	if (rc != DDI_SUCCESS) {
712 		cxgb_printf(dip, CE_WARN,
713 		    "failed to allocate %d interrupt(s) of type %d: %d, %d",
714 		    sc->intr_count, sc->intr_type, rc, i);
715 		goto done;
716 	}
717 	ASSERT(sc->intr_count == i); /* allocation was STRICT */
718 	(void) ddi_intr_get_cap(sc->intr_handle[0], &sc->intr_cap);
719 	(void) ddi_intr_get_pri(sc->intr_handle[0], &sc->intr_pri);
720 	if (sc->intr_count == 1) {
721 		ASSERT(sc->flags & INTR_FWD);
722 		(void) ddi_intr_add_handler(sc->intr_handle[0], t4_intr_all, sc,
723 		    &s->fwq);
724 	} else {
725 		/* Multiple interrupts.  The first one is always error intr */
726 		(void) ddi_intr_add_handler(sc->intr_handle[0], t4_intr_err, sc,
727 		    NULL);
728 		irq++;
729 
730 		/* The second one is always the firmware event queue */
731 		(void) ddi_intr_add_handler(sc->intr_handle[1], t4_intr, sc,
732 		    &s->fwq);
733 		irq++;
734 		/*
735 		 * Note that if INTR_FWD is set then either the NIC rx
736 		 * queues or (exclusive or) the TOE rx queueus will be taking
737 		 * direct interrupts.
738 		 *
739 		 * There is no need to check for is_offload(sc) as nofldrxq
740 		 * will be 0 if offload is disabled.
741 		 */
742 		for_each_port(sc, i) {
743 			struct port_info *pi = sc->port[i];
744 			struct sge_rxq *rxq;
745 #ifdef TCP_OFFLOAD_ENABLE
746 			struct sge_ofld_rxq *ofld_rxq;
747 
748 			/*
749 			 * Skip over the NIC queues if they aren't taking direct
750 			 * interrupts.
751 			 */
752 			if ((sc->flags & INTR_FWD) &&
753 			    pi->nofldrxq > pi->nrxq)
754 				goto ofld_queues;
755 #endif
756 			rxq = &s->rxq[pi->first_rxq];
757 			for (q = 0; q < pi->nrxq; q++, rxq++) {
758 				(void) ddi_intr_add_handler(
759 				    sc->intr_handle[irq], t4_intr, sc,
760 				    &rxq->iq);
761 				irq++;
762 			}
763 
764 #ifdef TCP_OFFLOAD_ENABLE
765 			/*
766 			 * Skip over the offload queues if they aren't taking
767 			 * direct interrupts.
768 			 */
769 			if ((sc->flags & INTR_FWD))
770 				continue;
771 ofld_queues:
772 			ofld_rxq = &s->ofld_rxq[pi->first_ofld_rxq];
773 			for (q = 0; q < pi->nofldrxq; q++, ofld_rxq++) {
774 				(void) ddi_intr_add_handler(
775 				    sc->intr_handle[irq], t4_intr, sc,
776 				    &ofld_rxq->iq);
777 				irq++;
778 			}
779 #endif
780 		}
781 
782 	}
783 	sc->flags |= INTR_ALLOCATED;
784 
785 	if ((rc = ksensor_create_scalar_pcidev(dip, SENSOR_KIND_TEMPERATURE,
786 	    &t4_temp_ops, sc, "temp", &sc->temp_sensor)) != 0) {
787 		cxgb_printf(dip, CE_WARN, "failed to create temperature "
788 		    "sensor: %d", rc);
789 		rc = DDI_FAILURE;
790 		goto done;
791 	}
792 
793 	if ((rc = ksensor_create_scalar_pcidev(dip, SENSOR_KIND_VOLTAGE,
794 	    &t4_volt_ops, sc, "vdd", &sc->volt_sensor)) != 0) {
795 		cxgb_printf(dip, CE_WARN, "failed to create voltage "
796 		    "sensor: %d", rc);
797 		rc = DDI_FAILURE;
798 		goto done;
799 	}
800 
801 
802 	if ((rc = ddi_ufm_init(dip, DDI_UFM_CURRENT_VERSION, &t4_ufm_ops,
803 	    &sc->ufm_hdl, sc)) != 0) {
804 		cxgb_printf(dip, CE_WARN, "failed to enable UFM ops: %d", rc);
805 		rc = DDI_FAILURE;
806 		goto done;
807 	}
808 	ddi_ufm_update(sc->ufm_hdl);
809 	ddi_report_dev(dip);
810 
811 	/*
812 	 * Hardware/Firmware/etc. Version/Revision IDs.
813 	 */
814 	t4_dump_version_info(sc);
815 
816 	if (n100g) {
817 		cxgb_printf(dip, CE_NOTE,
818 		    "%dx100G (%d rxq, %d txq total) %d %s.",
819 		    n100g, rqidx, tqidx, sc->intr_count,
820 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
821 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
822 		    "fixed interrupt");
823 	} else if (n40g) {
824 		cxgb_printf(dip, CE_NOTE,
825 		    "%dx40G (%d rxq, %d txq total) %d %s.",
826 		    n40g, rqidx, tqidx, sc->intr_count,
827 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
828 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
829 		    "fixed interrupt");
830 	} else if (n25g) {
831 		cxgb_printf(dip, CE_NOTE,
832 		    "%dx25G (%d rxq, %d txq total) %d %s.",
833 		    n25g, rqidx, tqidx, sc->intr_count,
834 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
835 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
836 		    "fixed interrupt");
837 	} else if (n10g && n1g) {
838 		cxgb_printf(dip, CE_NOTE,
839 		    "%dx10G %dx1G (%d rxq, %d txq total) %d %s.",
840 		    n10g, n1g, rqidx, tqidx, sc->intr_count,
841 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
842 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
843 		    "fixed interrupt");
844 	} else {
845 		cxgb_printf(dip, CE_NOTE,
846 		    "%dx%sG (%d rxq, %d txq per port) %d %s.",
847 		    n10g ? n10g : n1g,
848 		    n10g ? "10" : "1",
849 		    n10g ? iaq.nrxq10g : iaq.nrxq1g,
850 		    n10g ? iaq.ntxq10g : iaq.ntxq1g,
851 		    sc->intr_count,
852 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
853 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
854 		    "fixed interrupt");
855 	}
856 
857 	sc->ksp = setup_kstats(sc);
858 	sc->ksp_stat = setup_wc_kstats(sc);
859 	sc->params.drv_memwin = MEMWIN_NIC;
860 
861 done:
862 	if (rc != DDI_SUCCESS) {
863 		(void) t4_devo_detach(dip, DDI_DETACH);
864 
865 		/* rc may have errno style errors or DDI errors */
866 		rc = DDI_FAILURE;
867 	}
868 
869 	return (rc);
870 }
871 
872 static int
873 t4_devo_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
874 {
875 	int instance, i;
876 	struct adapter *sc;
877 	struct port_info *pi;
878 	struct sge *s;
879 
880 	if (cmd != DDI_DETACH)
881 		return (DDI_FAILURE);
882 
883 	instance = ddi_get_instance(dip);
884 	sc = ddi_get_soft_state(t4_list, instance);
885 	if (sc == NULL)
886 		return (DDI_SUCCESS);
887 
888 	if (sc->flags & FULL_INIT_DONE) {
889 		t4_intr_disable(sc);
890 		for_each_port(sc, i) {
891 			pi = sc->port[i];
892 			if (pi && pi->flags & PORT_INIT_DONE)
893 				(void) port_full_uninit(pi);
894 		}
895 		(void) adapter_full_uninit(sc);
896 	}
897 
898 	/* Safe to call no matter what */
899 	if (sc->ufm_hdl != NULL) {
900 		ddi_ufm_fini(sc->ufm_hdl);
901 		sc->ufm_hdl = NULL;
902 	}
903 	(void) ksensor_remove(dip, KSENSOR_ALL_IDS);
904 	ddi_prop_remove_all(dip);
905 	ddi_remove_minor_node(dip, NULL);
906 
907 	for (i = 0; i < NCHAN; i++) {
908 		if (sc->tq[i]) {
909 			ddi_taskq_wait(sc->tq[i]);
910 			ddi_taskq_destroy(sc->tq[i]);
911 		}
912 	}
913 
914 	if (sc->ksp != NULL)
915 		kstat_delete(sc->ksp);
916 	if (sc->ksp_stat != NULL)
917 		kstat_delete(sc->ksp_stat);
918 
919 	s = &sc->sge;
920 	if (s->rxq != NULL)
921 		kmem_free(s->rxq, s->nrxq * sizeof (struct sge_rxq));
922 #ifdef TCP_OFFLOAD_ENABLE
923 	if (s->ofld_txq != NULL)
924 		kmem_free(s->ofld_txq, s->nofldtxq * sizeof (struct sge_wrq));
925 	if (s->ofld_rxq != NULL)
926 		kmem_free(s->ofld_rxq,
927 		    s->nofldrxq * sizeof (struct sge_ofld_rxq));
928 	if (s->ctrlq != NULL)
929 		kmem_free(s->ctrlq,
930 		    sc->params.nports * sizeof (struct sge_wrq));
931 #endif
932 	if (s->txq != NULL)
933 		kmem_free(s->txq, s->ntxq * sizeof (struct sge_txq));
934 	if (s->iqmap != NULL)
935 		kmem_free(s->iqmap, s->niq * sizeof (struct sge_iq *));
936 	if (s->eqmap != NULL)
937 		kmem_free(s->eqmap, s->neq * sizeof (struct sge_eq *));
938 
939 	if (s->rxbuf_cache != NULL)
940 		rxbuf_cache_destroy(s->rxbuf_cache);
941 
942 	if (sc->flags & INTR_ALLOCATED) {
943 		for (i = 0; i < sc->intr_count; i++) {
944 			(void) ddi_intr_remove_handler(sc->intr_handle[i]);
945 			(void) ddi_intr_free(sc->intr_handle[i]);
946 		}
947 		sc->flags &= ~INTR_ALLOCATED;
948 	}
949 
950 	if (sc->intr_handle != NULL) {
951 		kmem_free(sc->intr_handle,
952 		    sc->intr_count * sizeof (*sc->intr_handle));
953 	}
954 
955 	for_each_port(sc, i) {
956 		pi = sc->port[i];
957 		if (pi != NULL) {
958 			mutex_destroy(&pi->lock);
959 			kmem_free(pi, sizeof (*pi));
960 			clrbit(&sc->registered_device_map, i);
961 		}
962 	}
963 
964 	if (sc->flags & FW_OK)
965 		(void) t4_fw_bye(sc, sc->mbox);
966 
967 	if (sc->reg1h != NULL)
968 		ddi_regs_map_free(&sc->reg1h);
969 
970 	if (sc->regh != NULL)
971 		ddi_regs_map_free(&sc->regh);
972 
973 	if (sc->pci_regh != NULL)
974 		pci_config_teardown(&sc->pci_regh);
975 
976 	mutex_enter(&t4_adapter_list_lock);
977 	SLIST_REMOVE_HEAD(&t4_adapter_list, link);
978 	mutex_exit(&t4_adapter_list_lock);
979 
980 	mutex_destroy(&sc->lock);
981 	cv_destroy(&sc->cv);
982 	mutex_destroy(&sc->sfl_lock);
983 
984 #ifdef DEBUG
985 	bzero(sc, sizeof (*sc));
986 #endif
987 	ddi_soft_state_free(t4_list, instance);
988 
989 	return (DDI_SUCCESS);
990 }
991 
992 static int
993 t4_devo_quiesce(dev_info_t *dip)
994 {
995 	int instance;
996 	struct adapter *sc;
997 
998 	instance = ddi_get_instance(dip);
999 	sc = ddi_get_soft_state(t4_list, instance);
1000 	if (sc == NULL)
1001 		return (DDI_SUCCESS);
1002 
1003 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
1004 	t4_intr_disable(sc);
1005 	t4_write_reg(sc, A_PL_RST, F_PIORSTMODE | F_PIORST);
1006 
1007 	return (DDI_SUCCESS);
1008 }
1009 
1010 static int
1011 t4_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t op, void *arg,
1012     void *result)
1013 {
1014 	char s[4];
1015 	struct port_info *pi;
1016 	dev_info_t *child = (dev_info_t *)arg;
1017 
1018 	switch (op) {
1019 	case DDI_CTLOPS_REPORTDEV:
1020 		pi = ddi_get_parent_data(rdip);
1021 		pi->instance = ddi_get_instance(dip);
1022 		pi->child_inst = ddi_get_instance(rdip);
1023 		cmn_err(CE_CONT, "?%s%d is port %s on %s%d\n",
1024 		    ddi_node_name(rdip), ddi_get_instance(rdip),
1025 		    ddi_get_name_addr(rdip), ddi_driver_name(dip),
1026 		    ddi_get_instance(dip));
1027 		return (DDI_SUCCESS);
1028 
1029 	case DDI_CTLOPS_INITCHILD:
1030 		pi = ddi_get_parent_data(child);
1031 		if (pi == NULL)
1032 			return (DDI_NOT_WELL_FORMED);
1033 		(void) snprintf(s, sizeof (s), "%d", pi->port_id);
1034 		ddi_set_name_addr(child, s);
1035 		return (DDI_SUCCESS);
1036 
1037 	case DDI_CTLOPS_UNINITCHILD:
1038 		ddi_set_name_addr(child, NULL);
1039 		return (DDI_SUCCESS);
1040 
1041 	case DDI_CTLOPS_ATTACH:
1042 	case DDI_CTLOPS_DETACH:
1043 		return (DDI_SUCCESS);
1044 
1045 	default:
1046 		return (ddi_ctlops(dip, rdip, op, arg, result));
1047 	}
1048 }
1049 
1050 static int
1051 t4_bus_config(dev_info_t *dip, uint_t flags, ddi_bus_config_op_t op, void *arg,
1052     dev_info_t **cdipp)
1053 {
1054 	int instance, i;
1055 	struct adapter *sc;
1056 
1057 	instance = ddi_get_instance(dip);
1058 	sc = ddi_get_soft_state(t4_list, instance);
1059 
1060 	if (op == BUS_CONFIG_ONE) {
1061 		char *c;
1062 
1063 		/*
1064 		 * arg is something like "cxgb@0" where 0 is the port_id hanging
1065 		 * off this nexus.
1066 		 */
1067 
1068 		c = arg;
1069 		while (*(c + 1))
1070 			c++;
1071 
1072 		/* There should be exactly 1 digit after '@' */
1073 		if (*(c - 1) != '@')
1074 			return (NDI_FAILURE);
1075 
1076 		i = *c - '0';
1077 
1078 		if (add_child_node(sc, i) != 0)
1079 			return (NDI_FAILURE);
1080 
1081 		flags |= NDI_ONLINE_ATTACH;
1082 
1083 	} else if (op == BUS_CONFIG_ALL || op == BUS_CONFIG_DRIVER) {
1084 		/* Allocate and bind all child device nodes */
1085 		for_each_port(sc, i)
1086 		    (void) add_child_node(sc, i);
1087 		flags |= NDI_ONLINE_ATTACH;
1088 	}
1089 
1090 	return (ndi_busop_bus_config(dip, flags, op, arg, cdipp, 0));
1091 }
1092 
1093 static int
1094 t4_bus_unconfig(dev_info_t *dip, uint_t flags, ddi_bus_config_op_t op,
1095     void *arg)
1096 {
1097 	int instance, i, rc;
1098 	struct adapter *sc;
1099 
1100 	instance = ddi_get_instance(dip);
1101 	sc = ddi_get_soft_state(t4_list, instance);
1102 
1103 	if (op == BUS_CONFIG_ONE || op == BUS_UNCONFIG_ALL ||
1104 	    op == BUS_UNCONFIG_DRIVER)
1105 		flags |= NDI_UNCONFIG;
1106 
1107 	rc = ndi_busop_bus_unconfig(dip, flags, op, arg);
1108 	if (rc != 0)
1109 		return (rc);
1110 
1111 	if (op == BUS_UNCONFIG_ONE) {
1112 		char *c;
1113 
1114 		c = arg;
1115 		while (*(c + 1))
1116 			c++;
1117 
1118 		if (*(c - 1) != '@')
1119 			return (NDI_SUCCESS);
1120 
1121 		i = *c - '0';
1122 
1123 		rc = remove_child_node(sc, i);
1124 
1125 	} else if (op == BUS_UNCONFIG_ALL || op == BUS_UNCONFIG_DRIVER) {
1126 
1127 		for_each_port(sc, i)
1128 		    (void) remove_child_node(sc, i);
1129 	}
1130 
1131 	return (rc);
1132 }
1133 
1134 /* ARGSUSED */
1135 static int
1136 t4_cb_open(dev_t *devp, int flag, int otyp, cred_t *credp)
1137 {
1138 	struct adapter *sc;
1139 
1140 	if (otyp != OTYP_CHR)
1141 		return (EINVAL);
1142 
1143 	sc = ddi_get_soft_state(t4_list, getminor(*devp));
1144 	if (sc == NULL)
1145 		return (ENXIO);
1146 
1147 	return (atomic_cas_uint(&sc->open, 0, EBUSY));
1148 }
1149 
1150 /* ARGSUSED */
1151 static int
1152 t4_cb_close(dev_t dev, int flag, int otyp, cred_t *credp)
1153 {
1154 	struct adapter *sc;
1155 
1156 	sc = ddi_get_soft_state(t4_list, getminor(dev));
1157 	if (sc == NULL)
1158 		return (EINVAL);
1159 
1160 	(void) atomic_swap_uint(&sc->open, 0);
1161 	return (0);
1162 }
1163 
1164 /* ARGSUSED */
1165 static int
1166 t4_cb_ioctl(dev_t dev, int cmd, intptr_t d, int mode, cred_t *credp, int *rp)
1167 {
1168 	int instance;
1169 	struct adapter *sc;
1170 	void *data = (void *)d;
1171 
1172 	if (crgetuid(credp) != 0)
1173 		return (EPERM);
1174 
1175 	instance = getminor(dev);
1176 	sc = ddi_get_soft_state(t4_list, instance);
1177 	if (sc == NULL)
1178 		return (EINVAL);
1179 
1180 	return (t4_ioctl(sc, cmd, data, mode));
1181 }
1182 
1183 static unsigned int
1184 getpf(struct adapter *sc)
1185 {
1186 	int rc, *data;
1187 	uint_t n, pf;
1188 
1189 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, sc->dip,
1190 	    DDI_PROP_DONTPASS, "reg", &data, &n);
1191 	if (rc != DDI_SUCCESS) {
1192 		cxgb_printf(sc->dip, CE_WARN,
1193 		    "failed to lookup \"reg\" property: %d", rc);
1194 		return (0xff);
1195 	}
1196 
1197 	pf = PCI_REG_FUNC_G(data[0]);
1198 	ddi_prop_free(data);
1199 
1200 	return (pf);
1201 }
1202 
1203 
1204 static struct fw_info *
1205 find_fw_info(int chip)
1206 {
1207 	u32 i;
1208 
1209 	fi[0].chip = CHELSIO_T4;
1210 	fi[0].fw_hdr.chip = FW_HDR_CHIP_T4;
1211 	fi[0].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T4));
1212 	fi[0].fw_hdr.intfver_nic = FW_INTFVER(T4, NIC);
1213 	fi[0].fw_hdr.intfver_vnic = FW_INTFVER(T4, VNIC);
1214 	fi[0].fw_hdr.intfver_ofld = FW_INTFVER(T4, OFLD);
1215 	fi[0].fw_hdr.intfver_ri = FW_INTFVER(T4, RI);
1216 	fi[0].fw_hdr.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU);
1217 	fi[0].fw_hdr.intfver_iscsi = FW_INTFVER(T4, ISCSI);
1218 	fi[0].fw_hdr.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU);
1219 	fi[0].fw_hdr.intfver_fcoe = FW_INTFVER(T4, FCOE);
1220 
1221 	fi[1].chip = CHELSIO_T5;
1222 	fi[1].fw_hdr.chip = FW_HDR_CHIP_T5;
1223 	fi[1].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T5));
1224 	fi[1].fw_hdr.intfver_nic = FW_INTFVER(T5, NIC);
1225 	fi[1].fw_hdr.intfver_vnic = FW_INTFVER(T5, VNIC);
1226 	fi[1].fw_hdr.intfver_ofld = FW_INTFVER(T5, OFLD);
1227 	fi[1].fw_hdr.intfver_ri = FW_INTFVER(T5, RI);
1228 	fi[1].fw_hdr.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU);
1229 	fi[1].fw_hdr.intfver_iscsi = FW_INTFVER(T5, ISCSI);
1230 	fi[1].fw_hdr.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU);
1231 	fi[1].fw_hdr.intfver_fcoe = FW_INTFVER(T5, FCOE);
1232 
1233 	fi[2].chip = CHELSIO_T6;
1234 	fi[2].fw_hdr.chip = FW_HDR_CHIP_T6;
1235 	fi[2].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T6));
1236 	fi[2].fw_hdr.intfver_nic = FW_INTFVER(T6, NIC);
1237 	fi[2].fw_hdr.intfver_vnic = FW_INTFVER(T6, VNIC);
1238 	fi[2].fw_hdr.intfver_ofld = FW_INTFVER(T6, OFLD);
1239 	fi[2].fw_hdr.intfver_ri = FW_INTFVER(T6, RI);
1240 	fi[2].fw_hdr.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU);
1241 	fi[2].fw_hdr.intfver_iscsi = FW_INTFVER(T6, ISCSI);
1242 	fi[2].fw_hdr.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU);
1243 	fi[2].fw_hdr.intfver_fcoe = FW_INTFVER(T6, FCOE);
1244 
1245 	for (i = 0; i < ARRAY_SIZE(fi); i++) {
1246 		if (fi[i].chip == chip)
1247 			return &fi[i];
1248 	}
1249 
1250 	return NULL;
1251 }
1252 
1253 /*
1254  * Install a compatible firmware (if required), establish contact with it,
1255  * become the master, and reset the device.
1256  */
1257 static int
1258 prep_firmware(struct adapter *sc)
1259 {
1260 	int rc;
1261 	int fw_size;
1262 	int reset = 1;
1263 	enum dev_state state;
1264 	unsigned char *fw_data;
1265 	struct fw_info *fw_info;
1266 	struct fw_hdr *card_fw;
1267 
1268 	struct driver_properties *p = &sc->props;
1269 
1270 	/* Contact firmware, request master */
1271 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MUST, &state);
1272 	if (rc < 0) {
1273 		rc = -rc;
1274 		cxgb_printf(sc->dip, CE_WARN,
1275 		    "failed to connect to the firmware: %d.", rc);
1276 		return (rc);
1277 	}
1278 
1279 	if (rc == sc->mbox)
1280 		sc->flags |= MASTER_PF;
1281 
1282 	/* We may need FW version info for later reporting */
1283 	t4_get_version_info(sc);
1284 	fw_info = find_fw_info(CHELSIO_CHIP_VERSION(sc->params.chip));
1285 	/* allocate memory to read the header of the firmware on the
1286 	 * card
1287 	 */
1288 	if (!fw_info) {
1289 		cxgb_printf(sc->dip, CE_WARN,
1290 			    "unable to look up firmware information for chip %d.\n",
1291 			    CHELSIO_CHIP_VERSION(sc->params.chip));
1292 		return EINVAL;
1293 	}
1294 	card_fw = kmem_zalloc(sizeof(*card_fw), KM_SLEEP);
1295 	if(!card_fw) {
1296 		cxgb_printf(sc->dip, CE_WARN,
1297 			    "Memory allocation for card FW header failed\n");
1298 		return ENOMEM;
1299 	}
1300 	switch(CHELSIO_CHIP_VERSION(sc->params.chip)) {
1301 	case CHELSIO_T4:
1302 		fw_data = t4fw_data;
1303 		fw_size = t4fw_size;
1304 		break;
1305 	case CHELSIO_T5:
1306 		fw_data = t5fw_data;
1307 		fw_size = t5fw_size;
1308 		break;
1309 	case CHELSIO_T6:
1310 		fw_data = t6fw_data;
1311 		fw_size = t6fw_size;
1312 		break;
1313 	default:
1314 		cxgb_printf(sc->dip, CE_WARN, "Adapter type not supported\n");
1315 		kmem_free(card_fw, sizeof(*card_fw));
1316 		return EINVAL;
1317 	}
1318 
1319 	rc = -t4_prep_fw(sc, fw_info, fw_data, fw_size, card_fw,
1320 			 p->t4_fw_install, state, &reset);
1321 
1322 	kmem_free(card_fw, sizeof(*card_fw));
1323 
1324 	if (rc != 0) {
1325 		cxgb_printf(sc->dip, CE_WARN,
1326 		    "failed to install firmware: %d", rc);
1327 		return (rc);
1328 	} else {
1329 		/* refresh */
1330 		(void) t4_check_fw_version(sc);
1331 	}
1332 
1333 	/* Reset device */
1334 	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
1335 	if (rc != 0) {
1336 		cxgb_printf(sc->dip, CE_WARN,
1337 		    "firmware reset failed: %d.", rc);
1338 		if (rc != ETIMEDOUT && rc != EIO)
1339 			(void) t4_fw_bye(sc, sc->mbox);
1340 		return (rc);
1341 	}
1342 
1343 	/* Partition adapter resources as specified in the config file. */
1344 	if (sc->flags & MASTER_PF) {
1345 		/* Handle default vs special T4 config file */
1346 
1347 		rc = partition_resources(sc);
1348 		if (rc != 0)
1349 			goto err;	/* error message displayed already */
1350 	}
1351 
1352 	sc->flags |= FW_OK;
1353 	return (0);
1354 err:
1355 	return (rc);
1356 
1357 }
1358 
1359 static const struct memwin t4_memwin[] = {
1360 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1361 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1362 	{ MEMWIN2_BASE, MEMWIN2_APERTURE }
1363 };
1364 
1365 static const struct memwin t5_memwin[] = {
1366 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1367 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1368 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
1369 };
1370 
1371 #define	FW_PARAM_DEV(param) \
1372 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
1373 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
1374 #define	FW_PARAM_PFVF(param) \
1375 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
1376 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
1377 
1378 /*
1379  * Verify that the memory range specified by the memtype/offset/len pair is
1380  * valid and lies entirely within the memtype specified.  The global address of
1381  * the start of the range is returned in addr.
1382  */
1383 int
1384 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
1385 	uint32_t *addr)
1386 {
1387 	uint32_t em, addr_len, maddr, mlen;
1388 
1389 	/* Memory can only be accessed in naturally aligned 4 byte units */
1390 	if (off & 3 || len & 3 || len == 0)
1391 		return (EINVAL);
1392 
1393 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
1394 	switch (mtype) {
1395 		case MEM_EDC0:
1396 			if (!(em & F_EDRAM0_ENABLE))
1397 				return (EINVAL);
1398 			addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
1399 			maddr = G_EDRAM0_BASE(addr_len) << 20;
1400 			mlen = G_EDRAM0_SIZE(addr_len) << 20;
1401 			break;
1402 		case MEM_EDC1:
1403 			if (!(em & F_EDRAM1_ENABLE))
1404 				return (EINVAL);
1405 			addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
1406 			maddr = G_EDRAM1_BASE(addr_len) << 20;
1407 			mlen = G_EDRAM1_SIZE(addr_len) << 20;
1408 			break;
1409 		case MEM_MC:
1410 			if (!(em & F_EXT_MEM_ENABLE))
1411 				return (EINVAL);
1412 			addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
1413 			maddr = G_EXT_MEM_BASE(addr_len) << 20;
1414 			mlen = G_EXT_MEM_SIZE(addr_len) << 20;
1415 			break;
1416 		case MEM_MC1:
1417 			if (is_t4(sc->params.chip) || !(em & F_EXT_MEM1_ENABLE))
1418 				return (EINVAL);
1419 			addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
1420 			maddr = G_EXT_MEM1_BASE(addr_len) << 20;
1421 			mlen = G_EXT_MEM1_SIZE(addr_len) << 20;
1422 			break;
1423 		default:
1424 			return (EINVAL);
1425 	}
1426 
1427 	if (mlen > 0 && off < mlen && off + len <= mlen) {
1428 		*addr = maddr + off;    /* global address */
1429 		return (0);
1430 	}
1431 
1432 	return (EFAULT);
1433 }
1434 
1435 void
1436 memwin_info(struct adapter *sc, int win, uint32_t *base, uint32_t *aperture)
1437 {
1438 	const struct memwin *mw;
1439 
1440 	if (is_t4(sc->params.chip)) {
1441 		mw = &t4_memwin[win];
1442 	} else {
1443 		mw = &t5_memwin[win];
1444 	}
1445 
1446 	if (base != NULL)
1447 		*base = mw->base;
1448 	if (aperture != NULL)
1449 		*aperture = mw->aperture;
1450 }
1451 
1452 /*
1453  * Upload configuration file to card's memory.
1454  */
1455 static int
1456 upload_config_file(struct adapter *sc, uint32_t *mt, uint32_t *ma)
1457 {
1458 	int rc = 0, cflen;
1459 	u_int i, n;
1460 	uint32_t param, val, addr, mtype, maddr;
1461 	uint32_t off, mw_base, mw_aperture;
1462 	const uint32_t *cfdata;
1463 
1464 	/* Figure out where the firmware wants us to upload it. */
1465 	param = FW_PARAM_DEV(CF);
1466 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
1467 	if (rc != 0) {
1468 		/* Firmwares without config file support will fail this way */
1469 		cxgb_printf(sc->dip, CE_WARN,
1470 		    "failed to query config file location: %d.\n", rc);
1471 		return (rc);
1472 	}
1473 	*mt = mtype = G_FW_PARAMS_PARAM_Y(val);
1474 	*ma = maddr = G_FW_PARAMS_PARAM_Z(val) << 16;
1475 
1476 	switch (CHELSIO_CHIP_VERSION(sc->params.chip)) {
1477 	case CHELSIO_T4:
1478 		cflen = t4cfg_size & ~3;
1479 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1480 		cfdata = (const uint32_t *)t4cfg_data;
1481 		break;
1482 	case CHELSIO_T5:
1483 		cflen = t5cfg_size & ~3;
1484 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1485 		cfdata = (const uint32_t *)t5cfg_data;
1486 		break;
1487 	case CHELSIO_T6:
1488 		cflen = t6cfg_size & ~3;
1489 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1490 		cfdata = (const uint32_t *)t6cfg_data;
1491 		break;
1492 	default:
1493 		cxgb_printf(sc->dip, CE_WARN,
1494 			    "Invalid Adapter detected\n");
1495 		return EINVAL;
1496 	}
1497 
1498 	if (cflen > FLASH_CFG_MAX_SIZE) {
1499 		cxgb_printf(sc->dip, CE_WARN,
1500 		    "config file too long (%d, max allowed is %d).  ",
1501 		    cflen, FLASH_CFG_MAX_SIZE);
1502 		return (EFBIG);
1503 	}
1504 
1505 	rc = validate_mt_off_len(sc, mtype, maddr, cflen, &addr);
1506 	if (rc != 0) {
1507 
1508 		cxgb_printf(sc->dip, CE_WARN,
1509 		    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
1510 		    "Will try to use the config on the card, if any.\n",
1511 		    __func__, mtype, maddr, cflen, rc);
1512 		return (EFAULT);
1513 	}
1514 
1515 	memwin_info(sc, 2, &mw_base, &mw_aperture);
1516 	while (cflen) {
1517 		off = position_memwin(sc, 2, addr);
1518 		n = min(cflen, mw_aperture - off);
1519 		for (i = 0; i < n; i += 4)
1520 			t4_write_reg(sc, mw_base + off + i, *cfdata++);
1521 		cflen -= n;
1522 		addr += n;
1523 	}
1524 
1525 	return (rc);
1526 }
1527 
1528 /*
1529  * Partition chip resources for use between various PFs, VFs, etc.  This is done
1530  * by uploading the firmware configuration file to the adapter and instructing
1531  * the firmware to process it.
1532  */
1533 static int
1534 partition_resources(struct adapter *sc)
1535 {
1536 	int rc;
1537 	struct fw_caps_config_cmd caps;
1538 	uint32_t mtype, maddr, finicsum, cfcsum;
1539 
1540 	rc = upload_config_file(sc, &mtype, &maddr);
1541 	if (rc != 0) {
1542 		mtype = FW_MEMTYPE_CF_FLASH;
1543 		maddr = t4_flash_cfg_addr(sc);
1544 	}
1545 
1546 	bzero(&caps, sizeof (caps));
1547 	caps.op_to_write = BE_32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1548 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1549 	caps.cfvalid_to_len16 = BE_32(F_FW_CAPS_CONFIG_CMD_CFVALID |
1550 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
1551 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) | FW_LEN16(caps));
1552 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), &caps);
1553 	if (rc != 0) {
1554 		cxgb_printf(sc->dip, CE_WARN,
1555 		    "failed to pre-process config file: %d.\n", rc);
1556 		return (rc);
1557 	}
1558 
1559 	finicsum = ntohl(caps.finicsum);
1560 	cfcsum = ntohl(caps.cfcsum);
1561 	if (finicsum != cfcsum) {
1562 		cxgb_printf(sc->dip, CE_WARN,
1563 		    "WARNING: config file checksum mismatch: %08x %08x\n",
1564 		    finicsum, cfcsum);
1565 	}
1566 	sc->cfcsum = cfcsum;
1567 
1568 	/* TODO: Need to configure this correctly */
1569 	caps.toecaps = htons(FW_CAPS_CONFIG_TOE);
1570 	caps.iscsicaps = 0;
1571 	caps.rdmacaps = 0;
1572 	caps.fcoecaps = 0;
1573 	/* TODO: Disable VNIC cap for now */
1574 	caps.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
1575 
1576 	caps.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1577 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
1578 	caps.cfvalid_to_len16 = htonl(FW_LEN16(caps));
1579 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), NULL);
1580 	if (rc != 0) {
1581 		cxgb_printf(sc->dip, CE_WARN,
1582 		    "failed to process config file: %d.\n", rc);
1583 		return (rc);
1584 	}
1585 
1586 	return (0);
1587 }
1588 
1589 /*
1590  * Tweak configuration based on module parameters, etc.  Most of these have
1591  * defaults assigned to them by Firmware Configuration Files (if we're using
1592  * them) but need to be explicitly set if we're using hard-coded
1593  * initialization.  But even in the case of using Firmware Configuration
1594  * Files, we'd like to expose the ability to change these via module
1595  * parameters so these are essentially common tweaks/settings for
1596  * Configuration Files and hard-coded initialization ...
1597  */
1598 static int
1599 adap__pre_init_tweaks(struct adapter *sc)
1600 {
1601 	int rx_dma_offset = 2; /* Offset of RX packets into DMA buffers */
1602 
1603 	/*
1604 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
1605 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
1606 	 * 64B Cache Line Size ...
1607 	 */
1608 	(void) t4_fixup_host_params_compat(sc, PAGE_SIZE, CACHE_LINE, T5_LAST_REV);
1609 
1610 	t4_set_reg_field(sc, A_SGE_CONTROL,
1611 			 V_PKTSHIFT(M_PKTSHIFT), V_PKTSHIFT(rx_dma_offset));
1612 
1613 	return 0;
1614 }
1615 /*
1616  * Retrieve parameters that are needed (or nice to have) prior to calling
1617  * t4_sge_init and t4_fw_initialize.
1618  */
1619 static int
1620 get_params__pre_init(struct adapter *sc)
1621 {
1622 	int rc;
1623 	uint32_t param[2], val[2];
1624 	struct fw_devlog_cmd cmd;
1625 	struct devlog_params *dlog = &sc->params.devlog;
1626 
1627 	/*
1628 	 * Grab the raw VPD parameters.
1629 	 */
1630 	rc = -t4_get_raw_vpd_params(sc, &sc->params.vpd);
1631 	if (rc != 0) {
1632 		cxgb_printf(sc->dip, CE_WARN,
1633 		    "failed to query VPD parameters (pre_init): %d.\n", rc);
1634 		return (rc);
1635 	}
1636 
1637 	param[0] = FW_PARAM_DEV(PORTVEC);
1638 	param[1] = FW_PARAM_DEV(CCLK);
1639 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
1640 	if (rc != 0) {
1641 		cxgb_printf(sc->dip, CE_WARN,
1642 		    "failed to query parameters (pre_init): %d.\n", rc);
1643 		return (rc);
1644 	}
1645 
1646 	sc->params.portvec = val[0];
1647 	sc->params.nports = 0;
1648 	while (val[0]) {
1649 		sc->params.nports++;
1650 		val[0] &= val[0] - 1;
1651 	}
1652 
1653 	sc->params.vpd.cclk = val[1];
1654 
1655 	/* Read device log parameters. */
1656 	bzero(&cmd, sizeof (cmd));
1657 	cmd.op_to_write = htonl(V_FW_CMD_OP(FW_DEVLOG_CMD) |
1658 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1659 	cmd.retval_len16 = htonl(FW_LEN16(cmd));
1660 	rc = -t4_wr_mbox(sc, sc->mbox, &cmd, sizeof (cmd), &cmd);
1661 	if (rc != 0) {
1662 		cxgb_printf(sc->dip, CE_WARN,
1663 		    "failed to get devlog parameters: %d.\n", rc);
1664 		bzero(dlog, sizeof (*dlog));
1665 		rc = 0;	/* devlog isn't critical for device operation */
1666 	} else {
1667 		val[0] = ntohl(cmd.memtype_devlog_memaddr16_devlog);
1668 		dlog->memtype = G_FW_DEVLOG_CMD_MEMTYPE_DEVLOG(val[0]);
1669 		dlog->start = G_FW_DEVLOG_CMD_MEMADDR16_DEVLOG(val[0]) << 4;
1670 		dlog->size = ntohl(cmd.memsize_devlog);
1671 	}
1672 
1673 	return (rc);
1674 }
1675 
1676 /*
1677  * Retrieve various parameters that are of interest to the driver.  The device
1678  * has been initialized by the firmware at this point.
1679  */
1680 static int
1681 get_params__post_init(struct adapter *sc)
1682 {
1683 	int rc;
1684 	uint32_t param[7], val[7];
1685 	struct fw_caps_config_cmd caps;
1686 
1687 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
1688 	param[1] = FW_PARAM_PFVF(EQ_START);
1689 	param[2] = FW_PARAM_PFVF(FILTER_START);
1690 	param[3] = FW_PARAM_PFVF(FILTER_END);
1691 	param[4] = FW_PARAM_PFVF(L2T_START);
1692 	param[5] = FW_PARAM_PFVF(L2T_END);
1693 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
1694 	if (rc != 0) {
1695 		cxgb_printf(sc->dip, CE_WARN,
1696 		    "failed to query parameters (post_init): %d.\n", rc);
1697 		return (rc);
1698 	}
1699 
1700 	/* LINTED: E_ASSIGN_NARROW_CONV */
1701 	sc->sge.iq_start = val[0];
1702 	sc->sge.eq_start = val[1];
1703 	sc->tids.ftid_base = val[2];
1704 	sc->tids.nftids = val[3] - val[2] + 1;
1705 	sc->vres.l2t.start = val[4];
1706 	sc->vres.l2t.size = val[5] - val[4] + 1;
1707 
1708 	/* get capabilites */
1709 	bzero(&caps, sizeof (caps));
1710 	caps.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1711 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1712 	caps.cfvalid_to_len16 = htonl(FW_LEN16(caps));
1713 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), &caps);
1714 	if (rc != 0) {
1715 		cxgb_printf(sc->dip, CE_WARN,
1716 		    "failed to get card capabilities: %d.\n", rc);
1717 		return (rc);
1718 	}
1719 
1720 	if (caps.toecaps != 0) {
1721 		/* query offload-related parameters */
1722 		param[0] = FW_PARAM_DEV(NTID);
1723 		param[1] = FW_PARAM_PFVF(SERVER_START);
1724 		param[2] = FW_PARAM_PFVF(SERVER_END);
1725 		param[3] = FW_PARAM_PFVF(TDDP_START);
1726 		param[4] = FW_PARAM_PFVF(TDDP_END);
1727 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
1728 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
1729 		if (rc != 0) {
1730 			cxgb_printf(sc->dip, CE_WARN,
1731 			    "failed to query TOE parameters: %d.\n", rc);
1732 			return (rc);
1733 		}
1734 		sc->tids.ntids = val[0];
1735 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
1736 		sc->tids.stid_base = val[1];
1737 		sc->tids.nstids = val[2] - val[1] + 1;
1738 		sc->vres.ddp.start = val[3];
1739 		sc->vres.ddp.size = val[4] - val[3] + 1;
1740 		sc->params.ofldq_wr_cred = val[5];
1741 		sc->params.offload = 1;
1742 	}
1743 
1744 	/* These are finalized by FW initialization, load their values now */
1745 	val[0] = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
1746 	sc->params.tp.tre = G_TIMERRESOLUTION(val[0]);
1747 	sc->params.tp.dack_re = G_DELAYEDACKRESOLUTION(val[0]);
1748 	t4_read_mtu_tbl(sc, sc->params.mtus, NULL);
1749 
1750 	return (rc);
1751 }
1752 
1753 static int
1754 set_params__post_init(struct adapter *sc)
1755 {
1756 	uint32_t param, val;
1757 
1758 	/* ask for encapsulated CPLs */
1759 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
1760 	val = 1;
1761 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
1762 
1763 	return (0);
1764 }
1765 
1766 /* TODO: verify */
1767 static void
1768 setup_memwin(struct adapter *sc)
1769 {
1770 	pci_regspec_t *data;
1771 	int rc;
1772 	uint_t n;
1773 	uintptr_t bar0;
1774 	uintptr_t mem_win0_base, mem_win1_base, mem_win2_base;
1775 	uintptr_t mem_win2_aperture;
1776 
1777 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, sc->dip,
1778 	    DDI_PROP_DONTPASS, "assigned-addresses", (int **)&data, &n);
1779 	if (rc != DDI_SUCCESS) {
1780 		cxgb_printf(sc->dip, CE_WARN,
1781 		    "failed to lookup \"assigned-addresses\" property: %d", rc);
1782 		return;
1783 	}
1784 	n /= sizeof (*data);
1785 
1786 	bar0 = ((uint64_t)data[0].pci_phys_mid << 32) | data[0].pci_phys_low;
1787 	ddi_prop_free(data);
1788 
1789 	if (is_t4(sc->params.chip)) {
1790 		mem_win0_base = bar0 + MEMWIN0_BASE;
1791 		mem_win1_base = bar0 + MEMWIN1_BASE;
1792 		mem_win2_base = bar0 + MEMWIN2_BASE;
1793 		mem_win2_aperture = MEMWIN2_APERTURE;
1794 	} else {
1795 		/* For T5, only relative offset inside the PCIe BAR is passed */
1796 		mem_win0_base = MEMWIN0_BASE;
1797 		mem_win1_base = MEMWIN1_BASE;
1798 		mem_win2_base = MEMWIN2_BASE_T5;
1799 		mem_win2_aperture = MEMWIN2_APERTURE_T5;
1800 	}
1801 
1802 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 0),
1803 	    mem_win0_base | V_BIR(0) |
1804 	    V_WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
1805 
1806 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 1),
1807 	    mem_win1_base | V_BIR(0) |
1808 	    V_WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
1809 
1810 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2),
1811 	    mem_win2_base | V_BIR(0) |
1812 	    V_WINDOW(ilog2(mem_win2_aperture) - 10));
1813 
1814 	/* flush */
1815 	(void)t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
1816 }
1817 
1818 /*
1819  * Positions the memory window such that it can be used to access the specified
1820  * address in the chip's address space.  The return value is the offset of addr
1821  * from the start of the window.
1822  */
1823 uint32_t
1824 position_memwin(struct adapter *sc, int n, uint32_t addr)
1825 {
1826 	uint32_t start, pf;
1827 	uint32_t reg;
1828 
1829 	if (addr & 3) {
1830 		cxgb_printf(sc->dip, CE_WARN,
1831 		    "addr (0x%x) is not at a 4B boundary.\n", addr);
1832 		return (EFAULT);
1833 	}
1834 
1835 	if (is_t4(sc->params.chip)) {
1836 		pf = 0;
1837 		start = addr & ~0xf;    /* start must be 16B aligned */
1838 	} else {
1839 		pf = V_PFNUM(sc->pf);
1840 		start = addr & ~0x7f;   /* start must be 128B aligned */
1841 	}
1842 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, n);
1843 
1844 	t4_write_reg(sc, reg, start | pf);
1845 	(void) t4_read_reg(sc, reg);
1846 
1847 	return (addr - start);
1848 }
1849 
1850 
1851 /*
1852  * Reads the named property and fills up the "data" array (which has at least
1853  * "count" elements).  We first try and lookup the property for our dev_t and
1854  * then retry with DDI_DEV_T_ANY if it's not found.
1855  *
1856  * Returns non-zero if the property was found and "data" has been updated.
1857  */
1858 static int
1859 prop_lookup_int_array(struct adapter *sc, char *name, int *data, uint_t count)
1860 {
1861 	dev_info_t *dip = sc->dip;
1862 	dev_t dev = sc->dev;
1863 	int rc, *d;
1864 	uint_t i, n;
1865 
1866 	rc = ddi_prop_lookup_int_array(dev, dip, DDI_PROP_DONTPASS,
1867 	    name, &d, &n);
1868 	if (rc == DDI_PROP_SUCCESS)
1869 		goto found;
1870 
1871 	if (rc != DDI_PROP_NOT_FOUND) {
1872 		cxgb_printf(dip, CE_WARN,
1873 		    "failed to lookup property %s for minor %d: %d.",
1874 		    name, getminor(dev), rc);
1875 		return (0);
1876 	}
1877 
1878 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
1879 	    name, &d, &n);
1880 	if (rc == DDI_PROP_SUCCESS)
1881 		goto found;
1882 
1883 	if (rc != DDI_PROP_NOT_FOUND) {
1884 		cxgb_printf(dip, CE_WARN,
1885 		    "failed to lookup property %s: %d.", name, rc);
1886 		return (0);
1887 	}
1888 
1889 	return (0);
1890 
1891 found:
1892 	if (n > count) {
1893 		cxgb_printf(dip, CE_NOTE,
1894 		    "property %s has too many elements (%d), ignoring extras",
1895 		    name, n);
1896 	}
1897 
1898 	for (i = 0; i < n && i < count; i++)
1899 		data[i] = d[i];
1900 	ddi_prop_free(d);
1901 
1902 	return (1);
1903 }
1904 
1905 static int
1906 prop_lookup_int(struct adapter *sc, char *name, int defval)
1907 {
1908 	int rc;
1909 
1910 	rc = ddi_prop_get_int(sc->dev, sc->dip, DDI_PROP_DONTPASS, name, -1);
1911 	if (rc != -1)
1912 		return (rc);
1913 
1914 	return (ddi_prop_get_int(DDI_DEV_T_ANY, sc->dip, DDI_PROP_DONTPASS,
1915 	    name, defval));
1916 }
1917 
1918 static int
1919 init_driver_props(struct adapter *sc, struct driver_properties *p)
1920 {
1921 	dev_t dev = sc->dev;
1922 	dev_info_t *dip = sc->dip;
1923 	int i, *data;
1924 	uint_t tmr[SGE_NTIMERS] = {5, 10, 20, 50, 100, 200};
1925 	uint_t cnt[SGE_NCOUNTERS] = {1, 8, 16, 32}; /* 63 max */
1926 
1927 	/*
1928 	 * Holdoff timer
1929 	 */
1930 	data = &p->timer_val[0];
1931 	for (i = 0; i < SGE_NTIMERS; i++)
1932 		data[i] = tmr[i];
1933 	(void) prop_lookup_int_array(sc, "holdoff-timer-values", data,
1934 	    SGE_NTIMERS);
1935 	for (i = 0; i < SGE_NTIMERS; i++) {
1936 		int limit = 200U;
1937 		if (data[i] > limit) {
1938 			cxgb_printf(dip, CE_WARN,
1939 			    "holdoff timer %d is too high (%d), lowered to %d.",
1940 			    i, data[i], limit);
1941 			data[i] = limit;
1942 		}
1943 	}
1944 	(void) ddi_prop_update_int_array(dev, dip, "holdoff-timer-values",
1945 	    data, SGE_NTIMERS);
1946 
1947 	/*
1948 	 * Holdoff packet counter
1949 	 */
1950 	data = &p->counter_val[0];
1951 	for (i = 0; i < SGE_NCOUNTERS; i++)
1952 		data[i] = cnt[i];
1953 	(void) prop_lookup_int_array(sc, "holdoff-pkt-counter-values", data,
1954 	    SGE_NCOUNTERS);
1955 	for (i = 0; i < SGE_NCOUNTERS; i++) {
1956 		int limit = M_THRESHOLD_0;
1957 		if (data[i] > limit) {
1958 			cxgb_printf(dip, CE_WARN,
1959 			    "holdoff pkt-counter %d is too high (%d), "
1960 			    "lowered to %d.", i, data[i], limit);
1961 			data[i] = limit;
1962 		}
1963 	}
1964 	(void) ddi_prop_update_int_array(dev, dip, "holdoff-pkt-counter-values",
1965 	    data, SGE_NCOUNTERS);
1966 
1967 	/*
1968 	 * Maximum # of tx and rx queues to use for each
1969 	 * 100G, 40G, 25G, 10G and 1G port.
1970 	 */
1971 	p->max_ntxq_10g = prop_lookup_int(sc, "max-ntxq-10G-port", 8);
1972 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-10G-port",
1973 	    p->max_ntxq_10g);
1974 
1975 	p->max_nrxq_10g = prop_lookup_int(sc, "max-nrxq-10G-port", 8);
1976 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-10G-port",
1977 	    p->max_nrxq_10g);
1978 
1979 	p->max_ntxq_1g = prop_lookup_int(sc, "max-ntxq-1G-port", 2);
1980 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-1G-port",
1981 	    p->max_ntxq_1g);
1982 
1983 	p->max_nrxq_1g = prop_lookup_int(sc, "max-nrxq-1G-port", 2);
1984 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-1G-port",
1985 	    p->max_nrxq_1g);
1986 
1987 #ifdef TCP_OFFLOAD_ENABLE
1988 	p->max_nofldtxq_10g = prop_lookup_int(sc, "max-nofldtxq-10G-port", 8);
1989 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-10G-port",
1990 	    p->max_nofldtxq_10g);
1991 
1992 	p->max_nofldrxq_10g = prop_lookup_int(sc, "max-nofldrxq-10G-port", 2);
1993 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-10G-port",
1994 	    p->max_nofldrxq_10g);
1995 
1996 	p->max_nofldtxq_1g = prop_lookup_int(sc, "max-nofldtxq-1G-port", 2);
1997 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-1G-port",
1998 	    p->max_nofldtxq_1g);
1999 
2000 	p->max_nofldrxq_1g = prop_lookup_int(sc, "max-nofldrxq-1G-port", 1);
2001 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-1G-port",
2002 	    p->max_nofldrxq_1g);
2003 #endif
2004 
2005 	/*
2006 	 * Holdoff parameters for 10G and 1G ports.
2007 	 */
2008 	p->tmr_idx_10g = prop_lookup_int(sc, "holdoff-timer-idx-10G", 0);
2009 	(void) ddi_prop_update_int(dev, dip, "holdoff-timer-idx-10G",
2010 	    p->tmr_idx_10g);
2011 
2012 	p->pktc_idx_10g = prop_lookup_int(sc, "holdoff-pktc-idx-10G", 2);
2013 	(void) ddi_prop_update_int(dev, dip, "holdoff-pktc-idx-10G",
2014 	    p->pktc_idx_10g);
2015 
2016 	p->tmr_idx_1g = prop_lookup_int(sc, "holdoff-timer-idx-1G", 0);
2017 	(void) ddi_prop_update_int(dev, dip, "holdoff-timer-idx-1G",
2018 	    p->tmr_idx_1g);
2019 
2020 	p->pktc_idx_1g = prop_lookup_int(sc, "holdoff-pktc-idx-1G", 2);
2021 	(void) ddi_prop_update_int(dev, dip, "holdoff-pktc-idx-1G",
2022 	    p->pktc_idx_1g);
2023 
2024 	/*
2025 	 * Size (number of entries) of each tx and rx queue.
2026 	 */
2027 	i = prop_lookup_int(sc, "qsize-txq", TX_EQ_QSIZE);
2028 	p->qsize_txq = max(i, 128);
2029 	if (p->qsize_txq != i) {
2030 		cxgb_printf(dip, CE_WARN,
2031 		    "using %d instead of %d as the tx queue size",
2032 		    p->qsize_txq, i);
2033 	}
2034 	(void) ddi_prop_update_int(dev, dip, "qsize-txq", p->qsize_txq);
2035 
2036 	i = prop_lookup_int(sc, "qsize-rxq", RX_IQ_QSIZE);
2037 	p->qsize_rxq = max(i, 128);
2038 	while (p->qsize_rxq & 7)
2039 		p->qsize_rxq--;
2040 	if (p->qsize_rxq != i) {
2041 		cxgb_printf(dip, CE_WARN,
2042 		    "using %d instead of %d as the rx queue size",
2043 		    p->qsize_rxq, i);
2044 	}
2045 	(void) ddi_prop_update_int(dev, dip, "qsize-rxq", p->qsize_rxq);
2046 
2047 	/*
2048 	 * Interrupt types allowed.
2049 	 * Bits 0, 1, 2 = INTx, MSI, MSI-X respectively.  See sys/ddi_intr.h
2050 	 */
2051 	p->intr_types = prop_lookup_int(sc, "interrupt-types",
2052 	    DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI | DDI_INTR_TYPE_FIXED);
2053 	(void) ddi_prop_update_int(dev, dip, "interrupt-types", p->intr_types);
2054 
2055 	/*
2056 	 * Forwarded interrupt queues.  Create this property to force the driver
2057 	 * to use forwarded interrupt queues.
2058 	 */
2059 	if (ddi_prop_exists(dev, dip, DDI_PROP_DONTPASS,
2060 	    "interrupt-forwarding") != 0 ||
2061 	    ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2062 	    "interrupt-forwarding") != 0) {
2063 		UNIMPLEMENTED();
2064 		(void) ddi_prop_create(dev, dip, DDI_PROP_CANSLEEP,
2065 		    "interrupt-forwarding", NULL, 0);
2066 	}
2067 
2068 	/*
2069 	 * Write combining
2070 	 * 0 to disable, 1 to enable
2071 	 */
2072 	p->wc = prop_lookup_int(sc, "write-combine", 1);
2073 	cxgb_printf(dip, CE_WARN, "write-combine: using of %d", p->wc);
2074 	if (p->wc != 0 && p->wc != 1) {
2075 		cxgb_printf(dip, CE_WARN,
2076 		    "write-combine: using 1 instead of %d", p->wc);
2077 		p->wc = 1;
2078 	}
2079 	(void) ddi_prop_update_int(dev, dip, "write-combine", p->wc);
2080 
2081 	p->t4_fw_install = prop_lookup_int(sc, "t4_fw_install", 1);
2082 	if (p->t4_fw_install != 0 && p->t4_fw_install != 2)
2083 		p->t4_fw_install = 1;
2084 	(void) ddi_prop_update_int(dev, dip, "t4_fw_install", p->t4_fw_install);
2085 
2086 	/* Multiple Rings */
2087 	p->multi_rings = prop_lookup_int(sc, "multi-rings", 1);
2088 	if (p->multi_rings != 0 && p->multi_rings != 1) {
2089 		cxgb_printf(dip, CE_NOTE,
2090 			   "multi-rings: using value 1 instead of %d", p->multi_rings);
2091 		p->multi_rings = 1;
2092 	}
2093 
2094 	(void) ddi_prop_update_int(dev, dip, "multi-rings", p->multi_rings);
2095 
2096 	return (0);
2097 }
2098 
2099 static int
2100 remove_extra_props(struct adapter *sc, int n10g, int n1g)
2101 {
2102 	if (n10g == 0) {
2103 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-ntxq-10G-port");
2104 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-nrxq-10G-port");
2105 		(void) ddi_prop_remove(sc->dev, sc->dip,
2106 		    "holdoff-timer-idx-10G");
2107 		(void) ddi_prop_remove(sc->dev, sc->dip,
2108 		    "holdoff-pktc-idx-10G");
2109 	}
2110 
2111 	if (n1g == 0) {
2112 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-ntxq-1G-port");
2113 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-nrxq-1G-port");
2114 		(void) ddi_prop_remove(sc->dev, sc->dip,
2115 		    "holdoff-timer-idx-1G");
2116 		(void) ddi_prop_remove(sc->dev, sc->dip, "holdoff-pktc-idx-1G");
2117 	}
2118 
2119 	return (0);
2120 }
2121 
2122 static int
2123 cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g,
2124     struct intrs_and_queues *iaq)
2125 {
2126 	struct driver_properties *p = &sc->props;
2127 	int rc, itype, itypes, navail, nc, nrxq10g, nrxq1g, n;
2128 	int nofldrxq10g = 0, nofldrxq1g = 0;
2129 
2130 	bzero(iaq, sizeof (*iaq));
2131 	nc = ncpus;	/* our snapshot of the number of CPUs */
2132 	iaq->ntxq10g = min(nc, p->max_ntxq_10g);
2133 	iaq->ntxq1g = min(nc, p->max_ntxq_1g);
2134 	iaq->nrxq10g = nrxq10g = min(nc, p->max_nrxq_10g);
2135 	iaq->nrxq1g = nrxq1g = min(nc, p->max_nrxq_1g);
2136 #ifdef TCP_OFFLOAD_ENABLE
2137 	iaq->nofldtxq10g = min(nc, p->max_nofldtxq_10g);
2138 	iaq->nofldtxq1g = min(nc, p->max_nofldtxq_1g);
2139 	iaq->nofldrxq10g = nofldrxq10g = min(nc, p->max_nofldrxq_10g);
2140 	iaq->nofldrxq1g = nofldrxq1g = min(nc, p->max_nofldrxq_1g);
2141 #endif
2142 
2143 	rc = ddi_intr_get_supported_types(sc->dip, &itypes);
2144 	if (rc != DDI_SUCCESS) {
2145 		cxgb_printf(sc->dip, CE_WARN,
2146 		    "failed to determine supported interrupt types: %d", rc);
2147 		return (rc);
2148 	}
2149 
2150 	for (itype = DDI_INTR_TYPE_MSIX; itype; itype >>= 1) {
2151 		ASSERT(itype == DDI_INTR_TYPE_MSIX ||
2152 		    itype == DDI_INTR_TYPE_MSI ||
2153 		    itype == DDI_INTR_TYPE_FIXED);
2154 
2155 		if ((itype & itypes & p->intr_types) == 0)
2156 			continue;	/* not supported or not allowed */
2157 
2158 		navail = 0;
2159 		rc = ddi_intr_get_navail(sc->dip, itype, &navail);
2160 		if (rc != DDI_SUCCESS || navail == 0) {
2161 			cxgb_printf(sc->dip, CE_WARN,
2162 			    "failed to get # of interrupts for type %d: %d",
2163 			    itype, rc);
2164 			continue;	/* carry on */
2165 		}
2166 
2167 		iaq->intr_type = itype;
2168 		if (navail == 0)
2169 			continue;
2170 
2171 		/*
2172 		 * Best option: an interrupt vector for errors, one for the
2173 		 * firmware event queue, and one each for each rxq (NIC as well
2174 		 * as offload).
2175 		 */
2176 		iaq->nirq = T4_EXTRA_INTR;
2177 		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2178 		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2179 
2180 		if (iaq->nirq <= navail &&
2181 		    (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq))) {
2182 			iaq->intr_fwd = 0;
2183 			goto allocate;
2184 		}
2185 
2186 		/*
2187 		 * Second best option: an interrupt vector for errors, one for
2188 		 * the firmware event queue, and one each for either NIC or
2189 		 * offload rxq's.
2190 		 */
2191 		iaq->nirq = T4_EXTRA_INTR;
2192 		iaq->nirq += n10g * max(nrxq10g, nofldrxq10g);
2193 		iaq->nirq += n1g * max(nrxq1g, nofldrxq1g);
2194 		if (iaq->nirq <= navail &&
2195 		    (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq))) {
2196 			iaq->intr_fwd = 1;
2197 			goto allocate;
2198 		}
2199 
2200 		/*
2201 		 * Next best option: an interrupt vector for errors, one for the
2202 		 * firmware event queue, and at least one per port.  At this
2203 		 * point we know we'll have to downsize nrxq or nofldrxq to fit
2204 		 * what's available to us.
2205 		 */
2206 		iaq->nirq = T4_EXTRA_INTR;
2207 		iaq->nirq += n10g + n1g;
2208 		if (iaq->nirq <= navail) {
2209 			int leftover = navail - iaq->nirq;
2210 
2211 			if (n10g > 0) {
2212 				int target = max(nrxq10g, nofldrxq10g);
2213 
2214 				n = 1;
2215 				while (n < target && leftover >= n10g) {
2216 					leftover -= n10g;
2217 					iaq->nirq += n10g;
2218 					n++;
2219 				}
2220 				iaq->nrxq10g = min(n, nrxq10g);
2221 #ifdef TCP_OFFLOAD_ENABLE
2222 				iaq->nofldrxq10g = min(n, nofldrxq10g);
2223 #endif
2224 			}
2225 
2226 			if (n1g > 0) {
2227 				int target = max(nrxq1g, nofldrxq1g);
2228 
2229 				n = 1;
2230 				while (n < target && leftover >= n1g) {
2231 					leftover -= n1g;
2232 					iaq->nirq += n1g;
2233 					n++;
2234 				}
2235 				iaq->nrxq1g = min(n, nrxq1g);
2236 #ifdef TCP_OFFLOAD_ENABLE
2237 				iaq->nofldrxq1g = min(n, nofldrxq1g);
2238 #endif
2239 			}
2240 
2241 			/* We have arrived at a minimum value required to enable
2242 			 * per queue irq(either NIC or offload). Thus for non-
2243 			 * offload case, we will get a vector per queue, while
2244 			 * offload case, we will get a vector per offload/NIC q.
2245 			 * Hence enable Interrupt forwarding only for offload
2246 			 * case.
2247 			 */
2248 #ifdef TCP_OFFLOAD_ENABLE
2249 			if (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq)) {
2250 				iaq->intr_fwd = 1;
2251 #else
2252 			if (itype != DDI_INTR_TYPE_MSI) {
2253 #endif
2254 				goto allocate;
2255 			}
2256 		}
2257 
2258 		/*
2259 		 * Least desirable option: one interrupt vector for everything.
2260 		 */
2261 		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2262 #ifdef TCP_OFFLOAD_ENABLE
2263 		iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2264 #endif
2265 		iaq->intr_fwd = 1;
2266 
2267 allocate:
2268 		return (0);
2269 	}
2270 
2271 	cxgb_printf(sc->dip, CE_WARN,
2272 	    "failed to find a usable interrupt type.  supported=%d, allowed=%d",
2273 	    itypes, p->intr_types);
2274 	return (DDI_FAILURE);
2275 }
2276 
2277 static int
2278 add_child_node(struct adapter *sc, int idx)
2279 {
2280 	int rc;
2281 	struct port_info *pi;
2282 
2283 	if (idx < 0 || idx >= sc->params.nports)
2284 		return (EINVAL);
2285 
2286 	pi = sc->port[idx];
2287 	if (pi == NULL)
2288 		return (ENODEV);	/* t4_port_init failed earlier */
2289 
2290 	PORT_LOCK(pi);
2291 	if (pi->dip != NULL) {
2292 		rc = 0;		/* EEXIST really, but then bus_config fails */
2293 		goto done;
2294 	}
2295 
2296 	rc = ndi_devi_alloc(sc->dip, T4_PORT_NAME, DEVI_SID_NODEID, &pi->dip);
2297 	if (rc != DDI_SUCCESS || pi->dip == NULL) {
2298 		rc = ENOMEM;
2299 		goto done;
2300 	}
2301 
2302 	(void) ddi_set_parent_data(pi->dip, pi);
2303 	(void) ndi_devi_bind_driver(pi->dip, 0);
2304 	rc = 0;
2305 done:
2306 	PORT_UNLOCK(pi);
2307 	return (rc);
2308 }
2309 
2310 static int
2311 remove_child_node(struct adapter *sc, int idx)
2312 {
2313 	int rc;
2314 	struct port_info *pi;
2315 
2316 	if (idx < 0 || idx >= sc->params.nports)
2317 		return (EINVAL);
2318 
2319 	pi = sc->port[idx];
2320 	if (pi == NULL)
2321 		return (ENODEV);
2322 
2323 	PORT_LOCK(pi);
2324 	if (pi->dip == NULL) {
2325 		rc = ENODEV;
2326 		goto done;
2327 	}
2328 
2329 	rc = ndi_devi_free(pi->dip);
2330 	if (rc == 0)
2331 		pi->dip = NULL;
2332 done:
2333 	PORT_UNLOCK(pi);
2334 	return (rc);
2335 }
2336 
2337 #define	KS_UINIT(x)	kstat_named_init(&kstatp->x, #x, KSTAT_DATA_ULONG)
2338 #define	KS_CINIT(x)	kstat_named_init(&kstatp->x, #x, KSTAT_DATA_CHAR)
2339 #define	KS_U_SET(x, y)	kstatp->x.value.ul = (y)
2340 #define	KS_C_SET(x, ...)	\
2341 			(void) snprintf(kstatp->x.value.c, 16,  __VA_ARGS__)
2342 
2343 /*
2344  * t4nex:X:config
2345  */
2346 struct t4_kstats {
2347 	kstat_named_t chip_ver;
2348 	kstat_named_t fw_vers;
2349 	kstat_named_t tp_vers;
2350 	kstat_named_t driver_version;
2351 	kstat_named_t serial_number;
2352 	kstat_named_t ec_level;
2353 	kstat_named_t id;
2354 	kstat_named_t bus_type;
2355 	kstat_named_t bus_width;
2356 	kstat_named_t bus_speed;
2357 	kstat_named_t core_clock;
2358 	kstat_named_t port_cnt;
2359 	kstat_named_t port_type;
2360 	kstat_named_t pci_vendor_id;
2361 	kstat_named_t pci_device_id;
2362 };
2363 static kstat_t *
2364 setup_kstats(struct adapter *sc)
2365 {
2366 	kstat_t *ksp;
2367 	struct t4_kstats *kstatp;
2368 	int ndata;
2369 	struct pci_params *p = &sc->params.pci;
2370 	struct vpd_params *v = &sc->params.vpd;
2371 	uint16_t pci_vendor, pci_device;
2372 
2373 	ndata = sizeof (struct t4_kstats) / sizeof (kstat_named_t);
2374 
2375 	ksp = kstat_create(T4_NEXUS_NAME, ddi_get_instance(sc->dip), "config",
2376 	    "nexus", KSTAT_TYPE_NAMED, ndata, 0);
2377 	if (ksp == NULL) {
2378 		cxgb_printf(sc->dip, CE_WARN, "failed to initialize kstats.");
2379 		return (NULL);
2380 	}
2381 
2382 	kstatp = (struct t4_kstats *)ksp->ks_data;
2383 
2384 	KS_UINIT(chip_ver);
2385 	KS_CINIT(fw_vers);
2386 	KS_CINIT(tp_vers);
2387 	KS_CINIT(driver_version);
2388 	KS_CINIT(serial_number);
2389 	KS_CINIT(ec_level);
2390 	KS_CINIT(id);
2391 	KS_CINIT(bus_type);
2392 	KS_CINIT(bus_width);
2393 	KS_CINIT(bus_speed);
2394 	KS_UINIT(core_clock);
2395 	KS_UINIT(port_cnt);
2396 	KS_CINIT(port_type);
2397 	KS_CINIT(pci_vendor_id);
2398 	KS_CINIT(pci_device_id);
2399 
2400 	KS_U_SET(chip_ver, sc->params.chip);
2401 	KS_C_SET(fw_vers, "%d.%d.%d.%d",
2402 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
2403 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
2404 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
2405 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
2406 	KS_C_SET(tp_vers, "%d.%d.%d.%d",
2407 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
2408 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
2409 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
2410 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
2411 	KS_C_SET(driver_version, DRV_VERSION);
2412 	KS_C_SET(serial_number, "%s", v->sn);
2413 	KS_C_SET(ec_level, "%s", v->ec);
2414 	KS_C_SET(id, "%s", v->id);
2415 	KS_C_SET(bus_type, "pci-express");
2416 	KS_C_SET(bus_width, "x%d lanes", p->width);
2417 	KS_C_SET(bus_speed, "%d", p->speed);
2418 	KS_U_SET(core_clock, v->cclk);
2419 	KS_U_SET(port_cnt, sc->params.nports);
2420 
2421 	t4_os_pci_read_cfg2(sc, PCI_CONF_VENID, &pci_vendor);
2422 	KS_C_SET(pci_vendor_id, "0x%x", pci_vendor);
2423 
2424 	t4_os_pci_read_cfg2(sc, PCI_CONF_DEVID, &pci_device);
2425 	KS_C_SET(pci_device_id, "0x%x", pci_device);
2426 
2427 	KS_C_SET(port_type, "%s/%s/%s/%s",
2428 		 print_port_speed(sc->port[0]),
2429 		 print_port_speed(sc->port[1]),
2430 		 print_port_speed(sc->port[2]),
2431 		 print_port_speed(sc->port[3]));
2432 
2433 	/* Do NOT set ksp->ks_update.  These kstats do not change. */
2434 
2435 	/* Install the kstat */
2436 	ksp->ks_private = (void *)sc;
2437 	kstat_install(ksp);
2438 
2439 	return (ksp);
2440 }
2441 
2442 /*
2443  * t4nex:X:stat
2444  */
2445 struct t4_wc_kstats {
2446 	kstat_named_t write_coal_success;
2447 	kstat_named_t write_coal_failure;
2448 };
2449 static kstat_t *
2450 setup_wc_kstats(struct adapter *sc)
2451 {
2452 	kstat_t *ksp;
2453 	struct t4_wc_kstats *kstatp;
2454 	int ndata;
2455 
2456 	ndata = sizeof(struct t4_wc_kstats) / sizeof(kstat_named_t);
2457 	ksp = kstat_create(T4_NEXUS_NAME, ddi_get_instance(sc->dip), "stats",
2458 	    "nexus", KSTAT_TYPE_NAMED, ndata, 0);
2459 	if (ksp == NULL) {
2460 		cxgb_printf(sc->dip, CE_WARN, "failed to initialize kstats.");
2461 		return (NULL);
2462 	}
2463 
2464 	kstatp = (struct t4_wc_kstats *)ksp->ks_data;
2465 
2466 	KS_UINIT(write_coal_success);
2467 	KS_UINIT(write_coal_failure);
2468 
2469 	ksp->ks_update = update_wc_kstats;
2470 	/* Install the kstat */
2471 	ksp->ks_private = (void *)sc;
2472 	kstat_install(ksp);
2473 
2474 	return (ksp);
2475 }
2476 
2477 static int
2478 update_wc_kstats(kstat_t *ksp, int rw)
2479 {
2480 	struct t4_wc_kstats *kstatp = (struct t4_wc_kstats *)ksp->ks_data;
2481 	struct adapter *sc = ksp->ks_private;
2482 	uint32_t wc_total, wc_success, wc_failure;
2483 
2484 	if (rw == KSTAT_WRITE)
2485 		return (0);
2486 
2487 	if (is_t5(sc->params.chip)) {
2488 		wc_total = t4_read_reg(sc, A_SGE_STAT_TOTAL);
2489 		wc_failure = t4_read_reg(sc, A_SGE_STAT_MATCH);
2490 		wc_success = wc_total - wc_failure;
2491 	} else {
2492 		wc_success = 0;
2493 		wc_failure = 0;
2494 	}
2495 
2496 	KS_U_SET(write_coal_success, wc_success);
2497 	KS_U_SET(write_coal_failure, wc_failure);
2498 
2499 	return (0);
2500 }
2501 
2502 int
2503 adapter_full_init(struct adapter *sc)
2504 {
2505 	int i, rc = 0;
2506 
2507 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2508 
2509 	rc = t4_setup_adapter_queues(sc);
2510 	if (rc != 0)
2511 		goto done;
2512 
2513 	if (sc->intr_cap & DDI_INTR_FLAG_BLOCK)
2514 		(void) ddi_intr_block_enable(sc->intr_handle, sc->intr_count);
2515 	else {
2516 		for (i = 0; i < sc->intr_count; i++)
2517 			(void) ddi_intr_enable(sc->intr_handle[i]);
2518 	}
2519 	t4_intr_enable(sc);
2520 	sc->flags |= FULL_INIT_DONE;
2521 
2522 #ifdef TCP_OFFLOAD_ENABLE
2523 	/* TODO: wrong place to enable TOE capability */
2524 	if (is_offload(sc) != 0) {
2525 		for_each_port(sc, i) {
2526 			struct port_info *pi = sc->port[i];
2527 			rc = toe_capability(pi, 1);
2528 			if (rc != 0) {
2529 				cxgb_printf(pi->dip, CE_WARN,
2530 				    "Failed to activate toe capability: %d",
2531 				    rc);
2532 				rc = 0;		/* not a fatal error */
2533 			}
2534 		}
2535 	}
2536 #endif
2537 
2538 done:
2539 	if (rc != 0)
2540 		(void) adapter_full_uninit(sc);
2541 
2542 	return (rc);
2543 }
2544 
2545 int
2546 adapter_full_uninit(struct adapter *sc)
2547 {
2548 	int i, rc = 0;
2549 
2550 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2551 
2552 	if (sc->intr_cap & DDI_INTR_FLAG_BLOCK)
2553 		(void) ddi_intr_block_disable(sc->intr_handle, sc->intr_count);
2554 	else {
2555 		for (i = 0; i < sc->intr_count; i++)
2556 			(void) ddi_intr_disable(sc->intr_handle[i]);
2557 	}
2558 
2559 	rc = t4_teardown_adapter_queues(sc);
2560 	if (rc != 0)
2561 		return (rc);
2562 
2563 	sc->flags &= ~FULL_INIT_DONE;
2564 
2565 	return (0);
2566 }
2567 
2568 int
2569 port_full_init(struct port_info *pi)
2570 {
2571 	struct adapter *sc = pi->adapter;
2572 	uint16_t *rss;
2573 	struct sge_rxq *rxq;
2574 	int rc, i;
2575 
2576 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2577 	ASSERT((pi->flags & PORT_INIT_DONE) == 0);
2578 
2579 	/*
2580 	 * Allocate tx/rx/fl queues for this port.
2581 	 */
2582 	rc = t4_setup_port_queues(pi);
2583 	if (rc != 0)
2584 		goto done;	/* error message displayed already */
2585 
2586 	/*
2587 	 * Setup RSS for this port.
2588 	 */
2589 	rss = kmem_zalloc(pi->nrxq * sizeof (*rss), KM_SLEEP);
2590 	for_each_rxq(pi, i, rxq) {
2591 		rss[i] = rxq->iq.abs_id;
2592 	}
2593 	rc = -t4_config_rss_range(sc, sc->mbox, pi->viid, 0,
2594 	    pi->rss_size, rss, pi->nrxq);
2595 	kmem_free(rss, pi->nrxq * sizeof (*rss));
2596 	if (rc != 0) {
2597 		cxgb_printf(pi->dip, CE_WARN, "rss_config failed: %d", rc);
2598 		goto done;
2599 	}
2600 
2601 	pi->flags |= PORT_INIT_DONE;
2602 done:
2603 	if (rc != 0)
2604 		(void) port_full_uninit(pi);
2605 
2606 	return (rc);
2607 }
2608 
2609 /*
2610  * Idempotent.
2611  */
2612 int
2613 port_full_uninit(struct port_info *pi)
2614 {
2615 
2616 	ASSERT(pi->flags & PORT_INIT_DONE);
2617 
2618 	(void) t4_teardown_port_queues(pi);
2619 	pi->flags &= ~PORT_INIT_DONE;
2620 
2621 	return (0);
2622 }
2623 
2624 void
2625 enable_port_queues(struct port_info *pi)
2626 {
2627 	struct adapter *sc = pi->adapter;
2628 	int i;
2629 	struct sge_iq *iq;
2630 	struct sge_rxq *rxq;
2631 #ifdef TCP_OFFLOAD_ENABLE
2632 	struct sge_ofld_rxq *ofld_rxq;
2633 #endif
2634 
2635 	ASSERT(pi->flags & PORT_INIT_DONE);
2636 
2637 	/*
2638 	 * TODO: whatever was queued up after we set iq->state to IQS_DISABLED
2639 	 * back in disable_port_queues will be processed now, after an unbounded
2640 	 * delay.  This can't be good.
2641 	 */
2642 
2643 #ifdef TCP_OFFLOAD_ENABLE
2644 	for_each_ofld_rxq(pi, i, ofld_rxq) {
2645 		iq = &ofld_rxq->iq;
2646 		if (atomic_cas_uint(&iq->state, IQS_DISABLED, IQS_IDLE) !=
2647 		    IQS_DISABLED)
2648 			panic("%s: iq %p wasn't disabled", __func__,
2649 			    (void *)iq);
2650 		t4_write_reg(sc, MYPF_REG(A_SGE_PF_GTS),
2651 		    V_SEINTARM(iq->intr_params) | V_INGRESSQID(iq->cntxt_id));
2652 	}
2653 #endif
2654 
2655 	for_each_rxq(pi, i, rxq) {
2656 		iq = &rxq->iq;
2657 		if (atomic_cas_uint(&iq->state, IQS_DISABLED, IQS_IDLE) !=
2658 		    IQS_DISABLED)
2659 			panic("%s: iq %p wasn't disabled", __func__,
2660 			    (void *) iq);
2661 		t4_write_reg(sc, MYPF_REG(A_SGE_PF_GTS),
2662 		    V_SEINTARM(iq->intr_params) | V_INGRESSQID(iq->cntxt_id));
2663 	}
2664 }
2665 
2666 void
2667 disable_port_queues(struct port_info *pi)
2668 {
2669 	int i;
2670 	struct adapter *sc = pi->adapter;
2671 	struct sge_rxq *rxq;
2672 #ifdef TCP_OFFLOAD_ENABLE
2673 	struct sge_ofld_rxq *ofld_rxq;
2674 #endif
2675 
2676 	ASSERT(pi->flags & PORT_INIT_DONE);
2677 
2678 	/*
2679 	 * TODO: need proper implementation for all tx queues (ctrl, eth, ofld).
2680 	 */
2681 
2682 #ifdef TCP_OFFLOAD_ENABLE
2683 	for_each_ofld_rxq(pi, i, ofld_rxq) {
2684 		while (atomic_cas_uint(&ofld_rxq->iq.state, IQS_IDLE,
2685 		    IQS_DISABLED) != IQS_IDLE)
2686 			msleep(1);
2687 	}
2688 #endif
2689 
2690 	for_each_rxq(pi, i, rxq) {
2691 		while (atomic_cas_uint(&rxq->iq.state, IQS_IDLE,
2692 		    IQS_DISABLED) != IQS_IDLE)
2693 			msleep(1);
2694 	}
2695 
2696 	mutex_enter(&sc->sfl_lock);
2697 #ifdef TCP_OFFLOAD_ENABLE
2698 	for_each_ofld_rxq(pi, i, ofld_rxq)
2699 	    ofld_rxq->fl.flags |= FL_DOOMED;
2700 #endif
2701 	for_each_rxq(pi, i, rxq)
2702 	    rxq->fl.flags |= FL_DOOMED;
2703 	mutex_exit(&sc->sfl_lock);
2704 	/* TODO: need to wait for all fl's to be removed from sc->sfl */
2705 }
2706 
2707 void
2708 t4_fatal_err(struct adapter *sc)
2709 {
2710 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2711 	t4_intr_disable(sc);
2712 	cxgb_printf(sc->dip, CE_WARN,
2713 	    "encountered fatal error, adapter stopped.");
2714 }
2715 
2716 int
2717 t4_os_find_pci_capability(struct adapter *sc, int cap)
2718 {
2719 	uint16_t stat;
2720 	uint8_t cap_ptr, cap_id;
2721 
2722 	t4_os_pci_read_cfg2(sc, PCI_CONF_STAT, &stat);
2723 	if ((stat & PCI_STAT_CAP) == 0)
2724 		return (0); /* does not implement capabilities */
2725 
2726 	t4_os_pci_read_cfg1(sc, PCI_CONF_CAP_PTR, &cap_ptr);
2727 	while (cap_ptr) {
2728 		t4_os_pci_read_cfg1(sc, cap_ptr + PCI_CAP_ID, &cap_id);
2729 		if (cap_id == cap)
2730 			return (cap_ptr); /* found */
2731 		t4_os_pci_read_cfg1(sc, cap_ptr + PCI_CAP_NEXT_PTR, &cap_ptr);
2732 	}
2733 
2734 	return (0); /* not found */
2735 }
2736 
2737 void
2738 t4_os_portmod_changed(const struct adapter *sc, int idx)
2739 {
2740 	static const char *mod_str[] = {
2741 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
2742 	};
2743 	const struct port_info *pi = sc->port[idx];
2744 
2745 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
2746 		cxgb_printf(pi->dip, CE_NOTE, "transceiver unplugged.");
2747 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
2748 		cxgb_printf(pi->dip, CE_NOTE,
2749 		    "unknown transceiver inserted.\n");
2750 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
2751 		cxgb_printf(pi->dip, CE_NOTE,
2752 		    "unsupported transceiver inserted.\n");
2753 	else if (pi->mod_type > 0 && pi->mod_type < ARRAY_SIZE(mod_str))
2754 		cxgb_printf(pi->dip, CE_NOTE, "%s transceiver inserted.\n",
2755 		    mod_str[pi->mod_type]);
2756 	else
2757 		cxgb_printf(pi->dip, CE_NOTE, "transceiver (type %d) inserted.",
2758 		    pi->mod_type);
2759 }
2760 
2761 /* ARGSUSED */
2762 static int
2763 cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss, mblk_t *m)
2764 {
2765 	if (m != NULL)
2766 		freemsg(m);
2767 	return (0);
2768 }
2769 
2770 int
2771 t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
2772 {
2773 	uint_t *loc, new;
2774 
2775 	if (opcode >= ARRAY_SIZE(sc->cpl_handler))
2776 		return (EINVAL);
2777 
2778 	new = (uint_t)(unsigned long) (h ? h : cpl_not_handled);
2779 	loc = (uint_t *)&sc->cpl_handler[opcode];
2780 	(void) atomic_swap_uint(loc, new);
2781 
2782 	return (0);
2783 }
2784 
2785 static int
2786 fw_msg_not_handled(struct adapter *sc, const __be64 *data)
2787 {
2788 	struct cpl_fw6_msg *cpl;
2789 
2790 	cpl = __containerof((void *)data, struct cpl_fw6_msg, data);
2791 
2792 	cxgb_printf(sc->dip, CE_WARN, "%s fw_msg type %d", __func__, cpl->type);
2793 	return (0);
2794 }
2795 
2796 int
2797 t4_register_fw_msg_handler(struct adapter *sc, int type, fw_msg_handler_t h)
2798 {
2799 	fw_msg_handler_t *loc, new;
2800 
2801 	if (type >= ARRAY_SIZE(sc->fw_msg_handler))
2802 		return (EINVAL);
2803 
2804 	/*
2805 	 * These are dispatched by the handler for FW{4|6}_CPL_MSG using the CPL
2806 	 * handler dispatch table.  Reject any attempt to install a handler for
2807 	 * this subtype.
2808 	 */
2809 	if (type == FW_TYPE_RSSCPL || type == FW6_TYPE_RSSCPL)
2810 		return (EINVAL);
2811 
2812 	new = h ? h : fw_msg_not_handled;
2813 	loc = &sc->fw_msg_handler[type];
2814 	(void)atomic_swap_ptr(loc, (void *)new);
2815 
2816 	return (0);
2817 }
2818 
2819 #ifdef TCP_OFFLOAD_ENABLE
2820 static int
2821 toe_capability(struct port_info *pi, int enable)
2822 {
2823 	int rc;
2824 	struct adapter *sc = pi->adapter;
2825 
2826 	if (!is_offload(sc))
2827 		return (ENODEV);
2828 
2829 	if (enable != 0) {
2830 		if (isset(&sc->offload_map, pi->port_id) != 0)
2831 			return (0);
2832 
2833 		if (sc->offload_map == 0) {
2834 			rc = activate_uld(sc, ULD_TOM, &sc->tom);
2835 			if (rc != 0)
2836 				return (rc);
2837 		}
2838 
2839 		setbit(&sc->offload_map, pi->port_id);
2840 	} else {
2841 		if (!isset(&sc->offload_map, pi->port_id))
2842 			return (0);
2843 
2844 		clrbit(&sc->offload_map, pi->port_id);
2845 
2846 		if (sc->offload_map == 0) {
2847 			rc = deactivate_uld(&sc->tom);
2848 			if (rc != 0) {
2849 				setbit(&sc->offload_map, pi->port_id);
2850 				return (rc);
2851 			}
2852 		}
2853 	}
2854 
2855 	return (0);
2856 }
2857 
2858 /*
2859  * Add an upper layer driver to the global list.
2860  */
2861 int
2862 t4_register_uld(struct uld_info *ui)
2863 {
2864 	int rc = 0;
2865 	struct uld_info *u;
2866 
2867 	mutex_enter(&t4_uld_list_lock);
2868 	SLIST_FOREACH(u, &t4_uld_list, link) {
2869 		if (u->uld_id == ui->uld_id) {
2870 			rc = EEXIST;
2871 			goto done;
2872 		}
2873 	}
2874 
2875 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
2876 	ui->refcount = 0;
2877 done:
2878 	mutex_exit(&t4_uld_list_lock);
2879 	return (rc);
2880 }
2881 
2882 int
2883 t4_unregister_uld(struct uld_info *ui)
2884 {
2885 	int rc = EINVAL;
2886 	struct uld_info *u;
2887 
2888 	mutex_enter(&t4_uld_list_lock);
2889 
2890 	SLIST_FOREACH(u, &t4_uld_list, link) {
2891 		if (u == ui) {
2892 			if (ui->refcount > 0) {
2893 				rc = EBUSY;
2894 				goto done;
2895 			}
2896 
2897 			SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
2898 			rc = 0;
2899 			goto done;
2900 		}
2901 	}
2902 done:
2903 	mutex_exit(&t4_uld_list_lock);
2904 	return (rc);
2905 }
2906 
2907 static int
2908 activate_uld(struct adapter *sc, int id, struct uld_softc *usc)
2909 {
2910 	int rc = EAGAIN;
2911 	struct uld_info *ui;
2912 
2913 	mutex_enter(&t4_uld_list_lock);
2914 
2915 	SLIST_FOREACH(ui, &t4_uld_list, link) {
2916 		if (ui->uld_id == id) {
2917 			rc = ui->attach(sc, &usc->softc);
2918 			if (rc == 0) {
2919 				ASSERT(usc->softc != NULL);
2920 				ui->refcount++;
2921 				usc->uld = ui;
2922 			}
2923 			goto done;
2924 		}
2925 	}
2926 done:
2927 	mutex_exit(&t4_uld_list_lock);
2928 
2929 	return (rc);
2930 }
2931 
2932 static int
2933 deactivate_uld(struct uld_softc *usc)
2934 {
2935 	int rc;
2936 
2937 	mutex_enter(&t4_uld_list_lock);
2938 
2939 	if (usc->uld == NULL || usc->softc == NULL) {
2940 		rc = EINVAL;
2941 		goto done;
2942 	}
2943 
2944 	rc = usc->uld->detach(usc->softc);
2945 	if (rc == 0) {
2946 		ASSERT(usc->uld->refcount > 0);
2947 		usc->uld->refcount--;
2948 		usc->uld = NULL;
2949 		usc->softc = NULL;
2950 	}
2951 done:
2952 	mutex_exit(&t4_uld_list_lock);
2953 
2954 	return (rc);
2955 }
2956 
2957 void
2958 t4_iterate(void (*func)(int, void *), void *arg)
2959 {
2960 	struct adapter *sc;
2961 
2962 	mutex_enter(&t4_adapter_list_lock);
2963 	SLIST_FOREACH(sc, &t4_adapter_list, link) {
2964 		/*
2965 		 * func should not make any assumptions about what state sc is
2966 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
2967 		 */
2968 		func(ddi_get_instance(sc->dip), arg);
2969 	}
2970 	mutex_exit(&t4_adapter_list_lock);
2971 }
2972 
2973 #endif
2974 
2975 static int
2976 t4_sensor_read(struct adapter *sc, uint32_t diag, uint32_t *valp)
2977 {
2978 	int rc;
2979 	struct port_info *pi = sc->port[0];
2980 	uint32_t param, val;
2981 
2982 	rc = begin_synchronized_op(pi, 1, 1);
2983 	if (rc != 0) {
2984 		return (rc);
2985 	}
2986 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2987 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
2988 	    V_FW_PARAMS_PARAM_Y(diag);
2989 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2990 	end_synchronized_op(pi, 1);
2991 
2992 	if (rc != 0) {
2993 		return (rc);
2994 	}
2995 
2996 	if (val == 0) {
2997 		return (EIO);
2998 	}
2999 
3000 	*valp = val;
3001 	return (0);
3002 }
3003 
3004 static int
3005 t4_temperature_read(void *arg, sensor_ioctl_scalar_t *scalar)
3006 {
3007 	int ret;
3008 	struct adapter *sc = arg;
3009 	uint32_t val;
3010 
3011 	ret = t4_sensor_read(sc, FW_PARAM_DEV_DIAG_TMP, &val);
3012 	if (ret != 0) {
3013 		return (ret);
3014 	}
3015 
3016 	/*
3017 	 * The device measures temperature in units of 1 degree Celsius. We
3018 	 * don't know its precision.
3019 	 */
3020 	scalar->sis_unit = SENSOR_UNIT_CELSIUS;
3021 	scalar->sis_gran = 1;
3022 	scalar->sis_prec = 0;
3023 	scalar->sis_value = val;
3024 
3025 	return (0);
3026 }
3027 
3028 static int
3029 t4_voltage_read(void *arg, sensor_ioctl_scalar_t *scalar)
3030 {
3031 	int ret;
3032 	struct adapter *sc = arg;
3033 	uint32_t val;
3034 
3035 	ret = t4_sensor_read(sc, FW_PARAM_DEV_DIAG_VDD, &val);
3036 	if (ret != 0) {
3037 		return (ret);
3038 	}
3039 
3040 	scalar->sis_unit = SENSOR_UNIT_VOLTS;
3041 	scalar->sis_gran = 1000;
3042 	scalar->sis_prec = 0;
3043 	scalar->sis_value = val;
3044 
3045 	return (0);
3046 }
3047 
3048 /*
3049  * While the hardware supports the ability to read and write the flash image,
3050  * this is not currently wired up.
3051  */
3052 static int
3053 t4_ufm_getcaps(ddi_ufm_handle_t *ufmh, void *arg, ddi_ufm_cap_t *caps)
3054 {
3055 	*caps = DDI_UFM_CAP_REPORT;
3056 	return (0);
3057 }
3058 
3059 static int
3060 t4_ufm_fill_image(ddi_ufm_handle_t *ufmh, void *arg, uint_t imgno,
3061     ddi_ufm_image_t *imgp)
3062 {
3063 	if (imgno != 0) {
3064 		return (EINVAL);
3065 	}
3066 
3067 	ddi_ufm_image_set_desc(imgp, "Firmware");
3068 	ddi_ufm_image_set_nslots(imgp, 1);
3069 
3070 	return (0);
3071 }
3072 
3073 static int
3074 t4_ufm_fill_slot_version(nvlist_t *nvl, const char *key, uint32_t vers)
3075 {
3076 	char buf[128];
3077 
3078 	if (vers == 0) {
3079 		return (0);
3080 	}
3081 
3082 	if (snprintf(buf, sizeof (buf), "%u.%u.%u.%u",
3083 	    G_FW_HDR_FW_VER_MAJOR(vers), G_FW_HDR_FW_VER_MINOR(vers),
3084 	    G_FW_HDR_FW_VER_MICRO(vers), G_FW_HDR_FW_VER_BUILD(vers)) >=
3085 	    sizeof (buf)) {
3086 		return (EOVERFLOW);
3087 	}
3088 
3089 	return (nvlist_add_string(nvl, key, buf));
3090 }
3091 
3092 static int
3093 t4_ufm_fill_slot(ddi_ufm_handle_t *ufmh, void *arg, uint_t imgno, uint_t slotno,
3094     ddi_ufm_slot_t *slotp)
3095 {
3096 	int ret;
3097 	struct adapter *sc = arg;
3098 	nvlist_t *misc = NULL;
3099 	char buf[128];
3100 
3101 	if (imgno != 0 || slotno != 0) {
3102 		return (EINVAL);
3103 	}
3104 
3105 	if (snprintf(buf, sizeof (buf), "%u.%u.%u.%u",
3106 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3107 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3108 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3109 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers)) >= sizeof (buf)) {
3110 		return (EOVERFLOW);
3111 	}
3112 
3113 	ddi_ufm_slot_set_version(slotp, buf);
3114 
3115 	(void) nvlist_alloc(&misc, NV_UNIQUE_NAME, KM_SLEEP);
3116 	if ((ret = t4_ufm_fill_slot_version(misc, "TP Microcode",
3117 	    sc->params.tp_vers)) != 0) {
3118 		goto err;
3119 	}
3120 
3121 	if ((ret = t4_ufm_fill_slot_version(misc, "Bootstrap",
3122 	    sc->params.bs_vers)) != 0) {
3123 		goto err;
3124 	}
3125 
3126 	if ((ret = t4_ufm_fill_slot_version(misc, "Expansion ROM",
3127 	    sc->params.er_vers)) != 0) {
3128 		goto err;
3129 	}
3130 
3131 	if ((ret = nvlist_add_uint32(misc, "Serial Configuration",
3132 	    sc->params.scfg_vers)) != 0) {
3133 		goto err;
3134 	}
3135 
3136 	if ((ret = nvlist_add_uint32(misc, "VPD Version",
3137 	    sc->params.vpd_vers)) != 0) {
3138 		goto err;
3139 	}
3140 
3141 	ddi_ufm_slot_set_misc(slotp, misc);
3142 	ddi_ufm_slot_set_attrs(slotp, DDI_UFM_ATTR_ACTIVE |
3143 	    DDI_UFM_ATTR_WRITEABLE | DDI_UFM_ATTR_READABLE);
3144 	return (0);
3145 
3146 err:
3147 	nvlist_free(misc);
3148 	return (ret);
3149 
3150 }
3151