xref: /illumos-gate/usr/src/uts/common/io/cxgbe/t4nex/t4_nexus.c (revision 94c3dad2979525d0a82595f3d8350a6116aba8ed)
1 /*
2  * This file and its contents are supplied under the terms of the
3  * Common Development and Distribution License ("CDDL"), version 1.0.
4  * You may only use this file in accordance with the terms of version
5  * 1.0 of the CDDL.
6  *
7  * A full copy of the text of the CDDL should have accompanied this
8  * source. A copy of the CDDL is also available via the Internet at
9  * http://www.illumos.org/license/CDDL.
10  */
11 
12 /*
13  * This file is part of the Chelsio T4 support code.
14  *
15  * Copyright (C) 2010-2013 Chelsio Communications.  All rights reserved.
16  *
17  * This program is distributed in the hope that it will be useful, but WITHOUT
18  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
19  * FITNESS FOR A PARTICULAR PURPOSE.  See the LICENSE file included in this
20  * release for licensing terms and conditions.
21  */
22 
23 #include <sys/ddi.h>
24 #include <sys/sunddi.h>
25 #include <sys/sunndi.h>
26 #include <sys/modctl.h>
27 #include <sys/conf.h>
28 #include <sys/devops.h>
29 #include <sys/pci.h>
30 #include <sys/atomic.h>
31 #include <sys/types.h>
32 #include <sys/file.h>
33 #include <sys/errno.h>
34 #include <sys/open.h>
35 #include <sys/cred.h>
36 #include <sys/stat.h>
37 #include <sys/mkdev.h>
38 #include <sys/queue.h>
39 #include <sys/containerof.h>
40 
41 #include "version.h"
42 #include "common/common.h"
43 #include "common/t4_msg.h"
44 #include "common/t4_regs.h"
45 #include "firmware/t4_fw.h"
46 #include "firmware/t4_cfg.h"
47 #include "firmware/t5_fw.h"
48 #include "firmware/t5_cfg.h"
49 #include "firmware/t6_fw.h"
50 #include "firmware/t6_cfg.h"
51 #include "t4_l2t.h"
52 
53 static int t4_cb_open(dev_t *devp, int flag, int otyp, cred_t *credp);
54 static int t4_cb_close(dev_t dev, int flag, int otyp, cred_t *credp);
55 static int t4_cb_ioctl(dev_t dev, int cmd, intptr_t d, int mode, cred_t *credp,
56     int *rp);
57 struct cb_ops t4_cb_ops = {
58 	.cb_open =		t4_cb_open,
59 	.cb_close =		t4_cb_close,
60 	.cb_strategy =		nodev,
61 	.cb_print =		nodev,
62 	.cb_dump =		nodev,
63 	.cb_read =		nodev,
64 	.cb_write =		nodev,
65 	.cb_ioctl =		t4_cb_ioctl,
66 	.cb_devmap =		nodev,
67 	.cb_mmap =		nodev,
68 	.cb_segmap =		nodev,
69 	.cb_chpoll =		nochpoll,
70 	.cb_prop_op =		ddi_prop_op,
71 	.cb_flag =		D_MP,
72 	.cb_rev =		CB_REV,
73 	.cb_aread =		nodev,
74 	.cb_awrite =		nodev
75 };
76 
77 static int t4_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t op,
78     void *arg, void *result);
79 static int t4_bus_config(dev_info_t *dip, uint_t flags, ddi_bus_config_op_t op,
80     void *arg, dev_info_t **cdipp);
81 static int t4_bus_unconfig(dev_info_t *dip, uint_t flags,
82     ddi_bus_config_op_t op, void *arg);
83 struct bus_ops t4_bus_ops = {
84 	.busops_rev =		BUSO_REV,
85 	.bus_ctl =		t4_bus_ctl,
86 	.bus_prop_op =		ddi_bus_prop_op,
87 	.bus_config =		t4_bus_config,
88 	.bus_unconfig =		t4_bus_unconfig,
89 };
90 
91 static int t4_devo_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg,
92     void **rp);
93 static int t4_devo_probe(dev_info_t *dip);
94 static int t4_devo_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
95 static int t4_devo_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
96 static int t4_devo_quiesce(dev_info_t *dip);
97 struct dev_ops t4_dev_ops = {
98 	.devo_rev =		DEVO_REV,
99 	.devo_getinfo =		t4_devo_getinfo,
100 	.devo_identify =	nulldev,
101 	.devo_probe =		t4_devo_probe,
102 	.devo_attach =		t4_devo_attach,
103 	.devo_detach =		t4_devo_detach,
104 	.devo_reset =		nodev,
105 	.devo_cb_ops =		&t4_cb_ops,
106 	.devo_bus_ops =		&t4_bus_ops,
107 	.devo_quiesce =		&t4_devo_quiesce,
108 };
109 
110 static struct modldrv modldrv = {
111 	.drv_modops =		&mod_driverops,
112 	.drv_linkinfo =		"Chelsio T4 nexus " DRV_VERSION,
113 	.drv_dev_ops =		&t4_dev_ops
114 };
115 
116 static struct modlinkage modlinkage = {
117 	.ml_rev =		MODREV_1,
118 	.ml_linkage =		{&modldrv, NULL},
119 };
120 
121 void *t4_list;
122 
123 struct intrs_and_queues {
124 	int intr_type;		/* DDI_INTR_TYPE_* */
125 	int nirq;		/* Number of vectors */
126 	int intr_fwd;		/* Interrupts forwarded */
127 	int ntxq10g;		/* # of NIC txq's for each 10G port */
128 	int nrxq10g;		/* # of NIC rxq's for each 10G port */
129 	int ntxq1g;		/* # of NIC txq's for each 1G port */
130 	int nrxq1g;		/* # of NIC rxq's for each 1G port */
131 #ifdef TCP_OFFLOAD_ENABLE
132 	int nofldtxq10g;	/* # of TOE txq's for each 10G port */
133 	int nofldrxq10g;	/* # of TOE rxq's for each 10G port */
134 	int nofldtxq1g;		/* # of TOE txq's for each 1G port */
135 	int nofldrxq1g;		/* # of TOE rxq's for each 1G port */
136 #endif
137 };
138 
139 struct fw_info fi[3];
140 
141 static int cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss,
142     mblk_t *m);
143 static int fw_msg_not_handled(struct adapter *, const __be64 *);
144 int t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h);
145 static unsigned int getpf(struct adapter *sc);
146 static int prep_firmware(struct adapter *sc);
147 static int upload_config_file(struct adapter *sc, uint32_t *mt, uint32_t *ma);
148 static int partition_resources(struct adapter *sc);
149 static int adap__pre_init_tweaks(struct adapter *sc);
150 static int get_params__pre_init(struct adapter *sc);
151 static int get_params__post_init(struct adapter *sc);
152 static int set_params__post_init(struct adapter *);
153 static void setup_memwin(struct adapter *sc);
154 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
155     uint32_t *);
156 void memwin_info(struct adapter *, int, uint32_t *, uint32_t *);
157 uint32_t position_memwin(struct adapter *, int, uint32_t);
158 static int prop_lookup_int_array(struct adapter *sc, char *name, int *data,
159     uint_t count);
160 static int prop_lookup_int_array(struct adapter *sc, char *name, int *data,
161     uint_t count);
162 static int init_driver_props(struct adapter *sc, struct driver_properties *p);
163 static int remove_extra_props(struct adapter *sc, int n10g, int n1g);
164 static int cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g,
165     struct intrs_and_queues *iaq);
166 static int add_child_node(struct adapter *sc, int idx);
167 static int remove_child_node(struct adapter *sc, int idx);
168 static kstat_t *setup_kstats(struct adapter *sc);
169 static kstat_t *setup_wc_kstats(struct adapter *);
170 static int update_wc_kstats(kstat_t *, int);
171 #ifdef TCP_OFFLOAD_ENABLE
172 static int toe_capability(struct port_info *pi, int enable);
173 static int activate_uld(struct adapter *sc, int id, struct uld_softc *usc);
174 static int deactivate_uld(struct uld_softc *usc);
175 #endif
176 static kmutex_t t4_adapter_list_lock;
177 static SLIST_HEAD(, adapter) t4_adapter_list;
178 #ifdef TCP_OFFLOAD_ENABLE
179 static kmutex_t t4_uld_list_lock;
180 static SLIST_HEAD(, uld_info) t4_uld_list;
181 #endif
182 
183 int
184 _init(void)
185 {
186 	int rc;
187 
188 	rc = ddi_soft_state_init(&t4_list, sizeof (struct adapter), 0);
189 	if (rc != 0)
190 		return (rc);
191 
192 	rc = mod_install(&modlinkage);
193 	if (rc != 0)
194 		ddi_soft_state_fini(&t4_list);
195 
196 	mutex_init(&t4_adapter_list_lock, NULL, MUTEX_DRIVER, NULL);
197 	SLIST_INIT(&t4_adapter_list);
198 
199 #ifdef TCP_OFFLOAD_ENABLE
200 	mutex_init(&t4_uld_list_lock, NULL, MUTEX_DRIVER, NULL);
201 	SLIST_INIT(&t4_uld_list);
202 #endif
203 
204 	return (rc);
205 }
206 
207 int
208 _fini(void)
209 {
210 	int rc;
211 
212 	rc = mod_remove(&modlinkage);
213 	if (rc != 0)
214 		return (rc);
215 
216 	ddi_soft_state_fini(&t4_list);
217 	return (0);
218 }
219 
220 int
221 _info(struct modinfo *mi)
222 {
223 	return (mod_info(&modlinkage, mi));
224 }
225 
226 /* ARGSUSED */
227 static int
228 t4_devo_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **rp)
229 {
230 	struct adapter *sc;
231 	minor_t minor;
232 
233 	minor = getminor((dev_t)arg);	/* same as instance# in our case */
234 
235 	if (cmd == DDI_INFO_DEVT2DEVINFO) {
236 		sc = ddi_get_soft_state(t4_list, minor);
237 		if (sc == NULL)
238 			return (DDI_FAILURE);
239 
240 		ASSERT(sc->dev == (dev_t)arg);
241 		*rp = (void *)sc->dip;
242 	} else if (cmd == DDI_INFO_DEVT2INSTANCE)
243 		*rp = (void *) (unsigned long) minor;
244 	else
245 		ASSERT(0);
246 
247 	return (DDI_SUCCESS);
248 }
249 
250 static int
251 t4_devo_probe(dev_info_t *dip)
252 {
253 	int rc, id, *reg;
254 	uint_t n, pf;
255 
256 	id = ddi_prop_get_int(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
257 	    "device-id", 0xffff);
258 	if (id == 0xffff)
259 		return (DDI_PROBE_DONTCARE);
260 
261 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
262 	    "reg", &reg, &n);
263 	if (rc != DDI_SUCCESS)
264 		return (DDI_PROBE_DONTCARE);
265 
266 	pf = PCI_REG_FUNC_G(reg[0]);
267 	ddi_prop_free(reg);
268 
269 	/* Prevent driver attachment on any PF except 0 on the FPGA */
270 	if (id == 0xa000 && pf != 0)
271 		return (DDI_PROBE_FAILURE);
272 
273 	return (DDI_PROBE_DONTCARE);
274 }
275 
276 static int
277 t4_devo_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
278 {
279 	struct adapter *sc = NULL;
280 	struct sge *s;
281 	int i, instance, rc = DDI_SUCCESS, rqidx, tqidx, q;
282 	int irq = 0, nxg, n100g, n40g, n25g, n10g, n1g;
283 #ifdef TCP_OFFLOAD_ENABLE
284 	int ofld_rqidx, ofld_tqidx;
285 #endif
286 	char name[16];
287 	struct driver_properties *prp;
288 	struct intrs_and_queues iaq;
289 	ddi_device_acc_attr_t da = {
290 		.devacc_attr_version = DDI_DEVICE_ATTR_V0,
291 		.devacc_attr_endian_flags = DDI_STRUCTURE_LE_ACC,
292 		.devacc_attr_dataorder = DDI_UNORDERED_OK_ACC
293 	};
294 	ddi_device_acc_attr_t da1 = {
295 		.devacc_attr_version = DDI_DEVICE_ATTR_V0,
296 		.devacc_attr_endian_flags = DDI_STRUCTURE_LE_ACC,
297 		.devacc_attr_dataorder = DDI_MERGING_OK_ACC
298 	};
299 
300 	if (cmd != DDI_ATTACH)
301 		return (DDI_FAILURE);
302 
303 	/*
304 	 * Allocate space for soft state.
305 	 */
306 	instance = ddi_get_instance(dip);
307 	rc = ddi_soft_state_zalloc(t4_list, instance);
308 	if (rc != DDI_SUCCESS) {
309 		cxgb_printf(dip, CE_WARN,
310 		    "failed to allocate soft state: %d", rc);
311 		return (DDI_FAILURE);
312 	}
313 
314 	sc = ddi_get_soft_state(t4_list, instance);
315 	sc->dip = dip;
316 	sc->dev = makedevice(ddi_driver_major(dip), instance);
317 	mutex_init(&sc->lock, NULL, MUTEX_DRIVER, NULL);
318 	cv_init(&sc->cv, NULL, CV_DRIVER, NULL);
319 	mutex_init(&sc->sfl_lock, NULL, MUTEX_DRIVER, NULL);
320 
321 	mutex_enter(&t4_adapter_list_lock);
322 	SLIST_INSERT_HEAD(&t4_adapter_list, sc, link);
323 	mutex_exit(&t4_adapter_list_lock);
324 
325 	sc->pf = getpf(sc);
326 	if (sc->pf > 8) {
327 		rc = EINVAL;
328 		cxgb_printf(dip, CE_WARN,
329 		    "failed to determine PCI PF# of device");
330 		goto done;
331 	}
332 	sc->mbox = sc->pf;
333 
334 	/* Initialize the driver properties */
335 	prp = &sc->props;
336 	(void)init_driver_props(sc, prp);
337 
338 	/*
339 	 * Enable access to the PCI config space.
340 	 */
341 	rc = pci_config_setup(dip, &sc->pci_regh);
342 	if (rc != DDI_SUCCESS) {
343 		cxgb_printf(dip, CE_WARN,
344 		    "failed to enable PCI config space access: %d", rc);
345 		goto done;
346 	}
347 
348 	/* TODO: Set max read request to 4K */
349 
350 	/*
351 	 * Enable MMIO access.
352 	 */
353 	rc = ddi_regs_map_setup(dip, 1, &sc->regp, 0, 0, &da, &sc->regh);
354 	if (rc != DDI_SUCCESS) {
355 		cxgb_printf(dip, CE_WARN,
356 		    "failed to map device registers: %d", rc);
357 		goto done;
358 	}
359 
360 	(void) memset(sc->chan_map, 0xff, sizeof (sc->chan_map));
361 
362 	/*
363 	 * Initialize cpl handler.
364 	 */
365 	for (i = 0; i < ARRAY_SIZE(sc->cpl_handler); i++) {
366 		sc->cpl_handler[i] = cpl_not_handled;
367 	}
368 
369 	for (i = 0; i < ARRAY_SIZE(sc->fw_msg_handler); i++) {
370 		sc->fw_msg_handler[i] = fw_msg_not_handled;
371 	}
372 
373 	for (i = 0; i < NCHAN; i++) {
374 		(void) snprintf(name, sizeof (name), "%s-%d",
375 				"reclaim", i);
376 		sc->tq[i] = ddi_taskq_create(sc->dip,
377 		   name, 1, TASKQ_DEFAULTPRI, 0);
378 
379 		if (sc->tq[i] == NULL) {
380 			cxgb_printf(dip, CE_WARN,
381 				   "failed to create task queues");
382 			rc = DDI_FAILURE;
383 			goto done;
384 		}
385 	}
386 
387 	/*
388 	 * Prepare the adapter for operation.
389 	 */
390 	rc = -t4_prep_adapter(sc, false);
391 	if (rc != 0) {
392 		cxgb_printf(dip, CE_WARN, "failed to prepare adapter: %d", rc);
393 		goto done;
394 	}
395 
396 	/*
397 	 * Enable BAR1 access.
398 	 */
399 	sc->doorbells |= DOORBELL_KDB;
400 	rc = ddi_regs_map_setup(dip, 2, &sc->reg1p, 0, 0, &da1, &sc->reg1h);
401 	if (rc != DDI_SUCCESS) {
402 		cxgb_printf(dip, CE_WARN,
403 		    "failed to map BAR1 device registers: %d", rc);
404 		goto done;
405 	} else {
406 		if (is_t5(sc->params.chip)) {
407 			sc->doorbells |= DOORBELL_UDB;
408 			if (prp->wc) {
409 				/*
410 				 * Enable write combining on BAR2.  This is the
411 				 * userspace doorbell BAR and is split into 128B
412 				 * (UDBS_SEG_SIZE) doorbell regions, each associated
413 				 * with an egress queue.  The first 64B has the doorbell
414 				 * and the second 64B can be used to submit a tx work
415 				 * request with an implicit doorbell.
416 				 */
417 				sc->doorbells &= ~DOORBELL_UDB;
418 				sc->doorbells |= (DOORBELL_WCWR |
419 				    DOORBELL_UDBWC);
420 				t4_write_reg(sc, A_SGE_STAT_CFG,
421 				    V_STATSOURCE_T5(7) | V_STATMODE(0));
422 			}
423 		}
424 	}
425 
426 	/*
427 	 * Do this really early.  Note that minor number = instance.
428 	 */
429 	(void) snprintf(name, sizeof (name), "%s,%d", T4_NEXUS_NAME, instance);
430 	rc = ddi_create_minor_node(dip, name, S_IFCHR, instance,
431 	    DDI_NT_NEXUS, 0);
432 	if (rc != DDI_SUCCESS) {
433 		cxgb_printf(dip, CE_WARN,
434 		    "failed to create device node: %d", rc);
435 		rc = DDI_SUCCESS; /* carry on */
436 	}
437 
438 	/* Do this early. Memory window is required for loading config file. */
439 	setup_memwin(sc);
440 
441 	/* Prepare the firmware for operation */
442 	rc = prep_firmware(sc);
443 	if (rc != 0)
444 		goto done; /* error message displayed already */
445 
446 	rc = adap__pre_init_tweaks(sc);
447 	if (rc != 0)
448 		goto done;
449 
450 	rc = get_params__pre_init(sc);
451 	if (rc != 0)
452 		goto done; /* error message displayed already */
453 
454 	t4_sge_init(sc);
455 
456 	if (sc->flags & MASTER_PF) {
457 		/* get basic stuff going */
458 		rc = -t4_fw_initialize(sc, sc->mbox);
459 		if (rc != 0) {
460 			cxgb_printf(sc->dip, CE_WARN,
461 			    "early init failed: %d.\n", rc);
462 			goto done;
463 		}
464 	}
465 
466 	rc = get_params__post_init(sc);
467 	if (rc != 0)
468 		goto done; /* error message displayed already */
469 
470 	rc = set_params__post_init(sc);
471 	if (rc != 0)
472 		goto done; /* error message displayed already */
473 
474 	/*
475 	 * TODO: This is the place to call t4_set_filter_mode()
476 	 */
477 
478 	/* tweak some settings */
479 	t4_write_reg(sc, A_TP_SHIFT_CNT, V_SYNSHIFTMAX(6) | V_RXTSHIFTMAXR1(4) |
480 	    V_RXTSHIFTMAXR2(15) | V_PERSHIFTBACKOFFMAX(8) | V_PERSHIFTMAX(8) |
481 	    V_KEEPALIVEMAXR1(4) | V_KEEPALIVEMAXR2(9));
482 	t4_write_reg(sc, A_ULP_RX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
483 
484 	/*
485 	 * Work-around for bug 2619
486 	 * Set DisableVlan field in TP_RSS_CONFIG_VRT register so that the
487 	 * VLAN tag extraction is disabled.
488 	 */
489 	t4_set_reg_field(sc, A_TP_RSS_CONFIG_VRT, F_DISABLEVLAN, F_DISABLEVLAN);
490 
491 	/* Store filter mode */
492 	t4_read_indirect(sc, A_TP_PIO_ADDR, A_TP_PIO_DATA, &sc->filter_mode, 1,
493 	    A_TP_VLAN_PRI_MAP);
494 
495 	/*
496 	 * First pass over all the ports - allocate VIs and initialize some
497 	 * basic parameters like mac address, port type, etc.  We also figure
498 	 * out whether a port is 10G or 1G and use that information when
499 	 * calculating how many interrupts to attempt to allocate.
500 	 */
501 	n100g = n40g = n25g = n10g = n1g = 0;
502 	for_each_port(sc, i) {
503 		struct port_info *pi;
504 
505 		pi = kmem_zalloc(sizeof (*pi), KM_SLEEP);
506 		sc->port[i] = pi;
507 
508 		/* These must be set before t4_port_init */
509 		pi->adapter = sc;
510 		/* LINTED: E_ASSIGN_NARROW_CONV */
511 		pi->port_id = i;
512 	}
513 
514 	/* Allocate the vi and initialize parameters like mac addr */
515 	rc = -t4_port_init(sc, sc->mbox, sc->pf, 0);
516 	if (rc) {
517 		cxgb_printf(dip, CE_WARN,
518 			    "unable to initialize port: %d", rc);
519 		goto done;
520 	}
521 
522 	for_each_port(sc, i) {
523 		struct port_info *pi = sc->port[i];
524 
525 		mutex_init(&pi->lock, NULL, MUTEX_DRIVER, NULL);
526 		pi->mtu = ETHERMTU;
527 
528 		if (is_100G_port(pi)) {
529 			n100g++;
530 			pi->tmr_idx = prp->tmr_idx_10g;
531 			pi->pktc_idx = prp->pktc_idx_10g;
532 		} else if (is_40G_port(pi)) {
533 			n40g++;
534 			pi->tmr_idx = prp->tmr_idx_10g;
535 			pi->pktc_idx = prp->pktc_idx_10g;
536 		} else if (is_25G_port(pi)) {
537 			n25g++;
538 			pi->tmr_idx = prp->tmr_idx_10g;
539 			pi->pktc_idx = prp->pktc_idx_10g;
540 		} else if (is_10G_port(pi)) {
541 			n10g++;
542 			pi->tmr_idx = prp->tmr_idx_10g;
543 			pi->pktc_idx = prp->pktc_idx_10g;
544 		} else {
545 			n1g++;
546 			pi->tmr_idx = prp->tmr_idx_1g;
547 			pi->pktc_idx = prp->pktc_idx_1g;
548 		}
549 
550 		pi->xact_addr_filt = -1;
551 		t4_mc_init(pi);
552 
553 		setbit(&sc->registered_device_map, i);
554 	}
555 
556 	nxg = n10g + n25g + n40g + n100g;
557 	(void) remove_extra_props(sc, nxg, n1g);
558 
559 	if (sc->registered_device_map == 0) {
560 		cxgb_printf(dip, CE_WARN, "no usable ports");
561 		rc = DDI_FAILURE;
562 		goto done;
563 	}
564 
565 	rc = cfg_itype_and_nqueues(sc, nxg, n1g, &iaq);
566 	if (rc != 0)
567 		goto done; /* error message displayed already */
568 
569 	sc->intr_type = iaq.intr_type;
570 	sc->intr_count = iaq.nirq;
571 
572 	if (sc->props.multi_rings && (sc->intr_type != DDI_INTR_TYPE_MSIX)) {
573 		sc->props.multi_rings = 0;
574 		cxgb_printf(dip, CE_WARN,
575 		    "Multiple rings disabled as interrupt type is not MSI-X");
576 	}
577 
578 	if (sc->props.multi_rings && iaq.intr_fwd) {
579 		sc->props.multi_rings = 0;
580 		cxgb_printf(dip, CE_WARN,
581 		    "Multiple rings disabled as interrupts are forwarded");
582 	}
583 
584 	if (!sc->props.multi_rings) {
585 		iaq.ntxq10g = 1;
586 		iaq.ntxq1g = 1;
587 	}
588 	s = &sc->sge;
589 	s->nrxq = nxg * iaq.nrxq10g + n1g * iaq.nrxq1g;
590 	s->ntxq = nxg * iaq.ntxq10g + n1g * iaq.ntxq1g;
591 	s->neq = s->ntxq + s->nrxq;	/* the fl in an rxq is an eq */
592 #ifdef TCP_OFFLOAD_ENABLE
593 	/* control queues, 1 per port + 1 mgmtq */
594 	s->neq += sc->params.nports + 1;
595 #endif
596 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
597 	if (iaq.intr_fwd != 0)
598 		sc->flags |= INTR_FWD;
599 #ifdef TCP_OFFLOAD_ENABLE
600 	if (is_offload(sc) != 0) {
601 
602 		s->nofldrxq = nxg * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
603 		s->nofldtxq = nxg * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
604 		s->neq += s->nofldtxq + s->nofldrxq;
605 		s->niq += s->nofldrxq;
606 
607 		s->ofld_rxq = kmem_zalloc(s->nofldrxq *
608 		    sizeof (struct sge_ofld_rxq), KM_SLEEP);
609 		s->ofld_txq = kmem_zalloc(s->nofldtxq *
610 		    sizeof (struct sge_wrq), KM_SLEEP);
611 		s->ctrlq = kmem_zalloc(sc->params.nports *
612 		    sizeof (struct sge_wrq), KM_SLEEP);
613 
614 	}
615 #endif
616 	s->rxq = kmem_zalloc(s->nrxq * sizeof (struct sge_rxq), KM_SLEEP);
617 	s->txq = kmem_zalloc(s->ntxq * sizeof (struct sge_txq), KM_SLEEP);
618 	s->iqmap = kmem_zalloc(s->niq * sizeof (struct sge_iq *), KM_SLEEP);
619 	s->eqmap = kmem_zalloc(s->neq * sizeof (struct sge_eq *), KM_SLEEP);
620 
621 	sc->intr_handle = kmem_zalloc(sc->intr_count *
622 	    sizeof (ddi_intr_handle_t), KM_SLEEP);
623 
624 	/*
625 	 * Second pass over the ports.  This time we know the number of rx and
626 	 * tx queues that each port should get.
627 	 */
628 	rqidx = tqidx = 0;
629 #ifdef TCP_OFFLOAD_ENABLE
630 	ofld_rqidx = ofld_tqidx = 0;
631 #endif
632 	for_each_port(sc, i) {
633 		struct port_info *pi = sc->port[i];
634 
635 		if (pi == NULL)
636 			continue;
637 
638 		t4_mc_cb_init(pi);
639 		/* LINTED: E_ASSIGN_NARROW_CONV */
640 		pi->first_rxq = rqidx;
641 		/* LINTED: E_ASSIGN_NARROW_CONV */
642 		pi->nrxq = (is_10XG_port(pi)) ? iaq.nrxq10g
643 		    : iaq.nrxq1g;
644 		/* LINTED: E_ASSIGN_NARROW_CONV */
645 		pi->first_txq = tqidx;
646 		/* LINTED: E_ASSIGN_NARROW_CONV */
647 		pi->ntxq = (is_10XG_port(pi)) ? iaq.ntxq10g
648 		    : iaq.ntxq1g;
649 
650 		rqidx += pi->nrxq;
651 		tqidx += pi->ntxq;
652 
653 #ifdef TCP_OFFLOAD_ENABLE
654 		if (is_offload(sc) != 0) {
655 			/* LINTED: E_ASSIGN_NARROW_CONV */
656 			pi->first_ofld_rxq = ofld_rqidx;
657 			pi->nofldrxq = max(1, pi->nrxq / 4);
658 
659 			/* LINTED: E_ASSIGN_NARROW_CONV */
660 			pi->first_ofld_txq = ofld_tqidx;
661 			pi->nofldtxq = max(1, pi->ntxq / 2);
662 
663 			ofld_rqidx += pi->nofldrxq;
664 			ofld_tqidx += pi->nofldtxq;
665 		}
666 #endif
667 
668 		/*
669 		 * Enable hw checksumming and LSO for all ports by default.
670 		 * They can be disabled using ndd (hw_csum and hw_lso).
671 		 */
672 		pi->features |= (CXGBE_HW_CSUM | CXGBE_HW_LSO);
673 	}
674 
675 #ifdef TCP_OFFLOAD_ENABLE
676 		sc->l2t = t4_init_l2t(sc);
677 #endif
678 
679 	/*
680 	 * Setup Interrupts.
681 	 */
682 
683 	i = 0;
684 	rc = ddi_intr_alloc(dip, sc->intr_handle, sc->intr_type, 0,
685 	    sc->intr_count, &i, DDI_INTR_ALLOC_STRICT);
686 	if (rc != DDI_SUCCESS) {
687 		cxgb_printf(dip, CE_WARN,
688 		    "failed to allocate %d interrupt(s) of type %d: %d, %d",
689 		    sc->intr_count, sc->intr_type, rc, i);
690 		goto done;
691 	}
692 	ASSERT(sc->intr_count == i); /* allocation was STRICT */
693 	(void) ddi_intr_get_cap(sc->intr_handle[0], &sc->intr_cap);
694 	(void) ddi_intr_get_pri(sc->intr_handle[0], &sc->intr_pri);
695 	if (sc->intr_count == 1) {
696 		ASSERT(sc->flags & INTR_FWD);
697 		(void) ddi_intr_add_handler(sc->intr_handle[0], t4_intr_all, sc,
698 		    &s->fwq);
699 	} else {
700 		/* Multiple interrupts.  The first one is always error intr */
701 		(void) ddi_intr_add_handler(sc->intr_handle[0], t4_intr_err, sc,
702 		    NULL);
703 		irq++;
704 
705 		/* The second one is always the firmware event queue */
706 		(void) ddi_intr_add_handler(sc->intr_handle[1], t4_intr, sc,
707 		    &s->fwq);
708 		irq++;
709 		/*
710 		 * Note that if INTR_FWD is set then either the NIC rx
711 		 * queues or (exclusive or) the TOE rx queueus will be taking
712 		 * direct interrupts.
713 		 *
714 		 * There is no need to check for is_offload(sc) as nofldrxq
715 		 * will be 0 if offload is disabled.
716 		 */
717 		for_each_port(sc, i) {
718 			struct port_info *pi = sc->port[i];
719 			struct sge_rxq *rxq;
720 #ifdef TCP_OFFLOAD_ENABLE
721 			struct sge_ofld_rxq *ofld_rxq;
722 
723 			/*
724 			 * Skip over the NIC queues if they aren't taking direct
725 			 * interrupts.
726 			 */
727 			if ((sc->flags & INTR_FWD) &&
728 			    pi->nofldrxq > pi->nrxq)
729 				goto ofld_queues;
730 #endif
731 			rxq = &s->rxq[pi->first_rxq];
732 			for (q = 0; q < pi->nrxq; q++, rxq++) {
733 				(void) ddi_intr_add_handler(
734 				    sc->intr_handle[irq], t4_intr, sc,
735 				    &rxq->iq);
736 				irq++;
737 			}
738 
739 #ifdef TCP_OFFLOAD_ENABLE
740 			/*
741 			 * Skip over the offload queues if they aren't taking
742 			 * direct interrupts.
743 			 */
744 			if ((sc->flags & INTR_FWD))
745 				continue;
746 ofld_queues:
747 			ofld_rxq = &s->ofld_rxq[pi->first_ofld_rxq];
748 			for (q = 0; q < pi->nofldrxq; q++, ofld_rxq++) {
749 				(void) ddi_intr_add_handler(
750 				    sc->intr_handle[irq], t4_intr, sc,
751 				    &ofld_rxq->iq);
752 				irq++;
753 			}
754 #endif
755 		}
756 
757 	}
758 	sc->flags |= INTR_ALLOCATED;
759 
760 	ASSERT(rc == DDI_SUCCESS);
761 	ddi_report_dev(dip);
762 
763 	/*
764 	 * Hardware/Firmware/etc. Version/Revision IDs.
765 	 */
766 	t4_dump_version_info(sc);
767 
768 	if (n100g) {
769 		cxgb_printf(dip, CE_NOTE,
770 		    "%dx100G (%d rxq, %d txq total) %d %s.",
771 		    n100g, rqidx, tqidx, sc->intr_count,
772 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
773 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
774 		    "fixed interrupt");
775 	} else if (n40g) {
776 		cxgb_printf(dip, CE_NOTE,
777 		    "%dx40G (%d rxq, %d txq total) %d %s.",
778 		    n40g, rqidx, tqidx, sc->intr_count,
779 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
780 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
781 		    "fixed interrupt");
782 	} else if (n25g) {
783 		cxgb_printf(dip, CE_NOTE,
784 		    "%dx25G (%d rxq, %d txq total) %d %s.",
785 		    n25g, rqidx, tqidx, sc->intr_count,
786 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
787 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
788 		    "fixed interrupt");
789 	} else if (n10g && n1g) {
790 		cxgb_printf(dip, CE_NOTE,
791 		    "%dx10G %dx1G (%d rxq, %d txq total) %d %s.",
792 		    n10g, n1g, rqidx, tqidx, sc->intr_count,
793 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
794 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
795 		    "fixed interrupt");
796 	} else {
797 		cxgb_printf(dip, CE_NOTE,
798 		    "%dx%sG (%d rxq, %d txq per port) %d %s.",
799 		    n10g ? n10g : n1g,
800 		    n10g ? "10" : "1",
801 		    n10g ? iaq.nrxq10g : iaq.nrxq1g,
802 		    n10g ? iaq.ntxq10g : iaq.ntxq1g,
803 		    sc->intr_count,
804 		    sc->intr_type == DDI_INTR_TYPE_MSIX ? "MSI-X interrupts" :
805 		    sc->intr_type == DDI_INTR_TYPE_MSI ? "MSI interrupts" :
806 		    "fixed interrupt");
807 	}
808 
809 	sc->ksp = setup_kstats(sc);
810 	sc->ksp_stat = setup_wc_kstats(sc);
811 	sc->params.drv_memwin = MEMWIN_NIC;
812 
813 done:
814 	if (rc != DDI_SUCCESS) {
815 		(void) t4_devo_detach(dip, DDI_DETACH);
816 
817 		/* rc may have errno style errors or DDI errors */
818 		rc = DDI_FAILURE;
819 	}
820 
821 	return (rc);
822 }
823 
824 static int
825 t4_devo_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
826 {
827 	int instance, i;
828 	struct adapter *sc;
829 	struct port_info *pi;
830 	struct sge *s;
831 
832 	if (cmd != DDI_DETACH)
833 		return (DDI_FAILURE);
834 
835 	instance = ddi_get_instance(dip);
836 	sc = ddi_get_soft_state(t4_list, instance);
837 	if (sc == NULL)
838 		return (DDI_SUCCESS);
839 
840 	if (sc->flags & FULL_INIT_DONE) {
841 		t4_intr_disable(sc);
842 		for_each_port(sc, i) {
843 			pi = sc->port[i];
844 			if (pi && pi->flags & PORT_INIT_DONE)
845 				(void) port_full_uninit(pi);
846 		}
847 		(void) adapter_full_uninit(sc);
848 	}
849 
850 	/* Safe to call no matter what */
851 	ddi_prop_remove_all(dip);
852 	ddi_remove_minor_node(dip, NULL);
853 
854 	for (i = 0; i < NCHAN; i++) {
855 		if (sc->tq[i]) {
856 			ddi_taskq_wait(sc->tq[i]);
857 			ddi_taskq_destroy(sc->tq[i]);
858 		}
859 	}
860 
861 	if (sc->ksp != NULL)
862 		kstat_delete(sc->ksp);
863 	if (sc->ksp_stat != NULL)
864 		kstat_delete(sc->ksp_stat);
865 
866 	s = &sc->sge;
867 	if (s->rxq != NULL)
868 		kmem_free(s->rxq, s->nrxq * sizeof (struct sge_rxq));
869 #ifdef TCP_OFFLOAD_ENABLE
870 	if (s->ofld_txq != NULL)
871 		kmem_free(s->ofld_txq, s->nofldtxq * sizeof (struct sge_wrq));
872 	if (s->ofld_rxq != NULL)
873 		kmem_free(s->ofld_rxq,
874 		    s->nofldrxq * sizeof (struct sge_ofld_rxq));
875 	if (s->ctrlq != NULL)
876 		kmem_free(s->ctrlq,
877 		    sc->params.nports * sizeof (struct sge_wrq));
878 #endif
879 	if (s->txq != NULL)
880 		kmem_free(s->txq, s->ntxq * sizeof (struct sge_txq));
881 	if (s->iqmap != NULL)
882 		kmem_free(s->iqmap, s->niq * sizeof (struct sge_iq *));
883 	if (s->eqmap != NULL)
884 		kmem_free(s->eqmap, s->neq * sizeof (struct sge_eq *));
885 
886 	if (s->rxbuf_cache != NULL)
887 		rxbuf_cache_destroy(s->rxbuf_cache);
888 
889 	if (sc->flags & INTR_ALLOCATED) {
890 		for (i = 0; i < sc->intr_count; i++) {
891 			(void) ddi_intr_remove_handler(sc->intr_handle[i]);
892 			(void) ddi_intr_free(sc->intr_handle[i]);
893 		}
894 		sc->flags &= ~INTR_ALLOCATED;
895 	}
896 
897 	if (sc->intr_handle != NULL) {
898 		kmem_free(sc->intr_handle,
899 		    sc->intr_count * sizeof (*sc->intr_handle));
900 	}
901 
902 	for_each_port(sc, i) {
903 		pi = sc->port[i];
904 		if (pi != NULL) {
905 			mutex_destroy(&pi->lock);
906 			kmem_free(pi, sizeof (*pi));
907 			clrbit(&sc->registered_device_map, i);
908 		}
909 	}
910 
911 	if (sc->flags & FW_OK)
912 		(void) t4_fw_bye(sc, sc->mbox);
913 
914 	if (sc->reg1h != NULL)
915 		ddi_regs_map_free(&sc->reg1h);
916 
917 	if (sc->regh != NULL)
918 		ddi_regs_map_free(&sc->regh);
919 
920 	if (sc->pci_regh != NULL)
921 		pci_config_teardown(&sc->pci_regh);
922 
923 	mutex_enter(&t4_adapter_list_lock);
924 	SLIST_REMOVE_HEAD(&t4_adapter_list, link);
925 	mutex_exit(&t4_adapter_list_lock);
926 
927 	mutex_destroy(&sc->lock);
928 	cv_destroy(&sc->cv);
929 	mutex_destroy(&sc->sfl_lock);
930 
931 #ifdef DEBUG
932 	bzero(sc, sizeof (*sc));
933 #endif
934 	ddi_soft_state_free(t4_list, instance);
935 
936 	return (DDI_SUCCESS);
937 }
938 
939 static int
940 t4_devo_quiesce(dev_info_t *dip)
941 {
942 	int instance;
943 	struct adapter *sc;
944 
945 	instance = ddi_get_instance(dip);
946 	sc = ddi_get_soft_state(t4_list, instance);
947 	if (sc == NULL)
948 		return (DDI_SUCCESS);
949 
950 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
951 	t4_intr_disable(sc);
952 	t4_write_reg(sc, A_PL_RST, F_PIORSTMODE | F_PIORST);
953 
954 	return (DDI_SUCCESS);
955 }
956 
957 static int
958 t4_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t op, void *arg,
959     void *result)
960 {
961 	char s[4];
962 	struct port_info *pi;
963 	dev_info_t *child = (dev_info_t *)arg;
964 
965 	switch (op) {
966 	case DDI_CTLOPS_REPORTDEV:
967 		pi = ddi_get_parent_data(rdip);
968 		pi->instance = ddi_get_instance(dip);
969 		pi->child_inst = ddi_get_instance(rdip);
970 		cmn_err(CE_CONT, "?%s%d is port %s on %s%d\n",
971 		    ddi_node_name(rdip), ddi_get_instance(rdip),
972 		    ddi_get_name_addr(rdip), ddi_driver_name(dip),
973 		    ddi_get_instance(dip));
974 		return (DDI_SUCCESS);
975 
976 	case DDI_CTLOPS_INITCHILD:
977 		pi = ddi_get_parent_data(child);
978 		if (pi == NULL)
979 			return (DDI_NOT_WELL_FORMED);
980 		(void) snprintf(s, sizeof (s), "%d", pi->port_id);
981 		ddi_set_name_addr(child, s);
982 		return (DDI_SUCCESS);
983 
984 	case DDI_CTLOPS_UNINITCHILD:
985 		ddi_set_name_addr(child, NULL);
986 		return (DDI_SUCCESS);
987 
988 	case DDI_CTLOPS_ATTACH:
989 	case DDI_CTLOPS_DETACH:
990 		return (DDI_SUCCESS);
991 
992 	default:
993 		return (ddi_ctlops(dip, rdip, op, arg, result));
994 	}
995 }
996 
997 static int
998 t4_bus_config(dev_info_t *dip, uint_t flags, ddi_bus_config_op_t op, void *arg,
999     dev_info_t **cdipp)
1000 {
1001 	int instance, i;
1002 	struct adapter *sc;
1003 
1004 	instance = ddi_get_instance(dip);
1005 	sc = ddi_get_soft_state(t4_list, instance);
1006 
1007 	if (op == BUS_CONFIG_ONE) {
1008 		char *c;
1009 
1010 		/*
1011 		 * arg is something like "cxgb@0" where 0 is the port_id hanging
1012 		 * off this nexus.
1013 		 */
1014 
1015 		c = arg;
1016 		while (*(c + 1))
1017 			c++;
1018 
1019 		/* There should be exactly 1 digit after '@' */
1020 		if (*(c - 1) != '@')
1021 			return (NDI_FAILURE);
1022 
1023 		i = *c - '0';
1024 
1025 		if (add_child_node(sc, i) != 0)
1026 			return (NDI_FAILURE);
1027 
1028 		flags |= NDI_ONLINE_ATTACH;
1029 
1030 	} else if (op == BUS_CONFIG_ALL || op == BUS_CONFIG_DRIVER) {
1031 		/* Allocate and bind all child device nodes */
1032 		for_each_port(sc, i)
1033 		    (void) add_child_node(sc, i);
1034 		flags |= NDI_ONLINE_ATTACH;
1035 	}
1036 
1037 	return (ndi_busop_bus_config(dip, flags, op, arg, cdipp, 0));
1038 }
1039 
1040 static int
1041 t4_bus_unconfig(dev_info_t *dip, uint_t flags, ddi_bus_config_op_t op,
1042     void *arg)
1043 {
1044 	int instance, i, rc;
1045 	struct adapter *sc;
1046 
1047 	instance = ddi_get_instance(dip);
1048 	sc = ddi_get_soft_state(t4_list, instance);
1049 
1050 	if (op == BUS_CONFIG_ONE || op == BUS_UNCONFIG_ALL ||
1051 	    op == BUS_UNCONFIG_DRIVER)
1052 		flags |= NDI_UNCONFIG;
1053 
1054 	rc = ndi_busop_bus_unconfig(dip, flags, op, arg);
1055 	if (rc != 0)
1056 		return (rc);
1057 
1058 	if (op == BUS_UNCONFIG_ONE) {
1059 		char *c;
1060 
1061 		c = arg;
1062 		while (*(c + 1))
1063 			c++;
1064 
1065 		if (*(c - 1) != '@')
1066 			return (NDI_SUCCESS);
1067 
1068 		i = *c - '0';
1069 
1070 		rc = remove_child_node(sc, i);
1071 
1072 	} else if (op == BUS_UNCONFIG_ALL || op == BUS_UNCONFIG_DRIVER) {
1073 
1074 		for_each_port(sc, i)
1075 		    (void) remove_child_node(sc, i);
1076 	}
1077 
1078 	return (rc);
1079 }
1080 
1081 /* ARGSUSED */
1082 static int
1083 t4_cb_open(dev_t *devp, int flag, int otyp, cred_t *credp)
1084 {
1085 	struct adapter *sc;
1086 
1087 	if (otyp != OTYP_CHR)
1088 		return (EINVAL);
1089 
1090 	sc = ddi_get_soft_state(t4_list, getminor(*devp));
1091 	if (sc == NULL)
1092 		return (ENXIO);
1093 
1094 	return (atomic_cas_uint(&sc->open, 0, EBUSY));
1095 }
1096 
1097 /* ARGSUSED */
1098 static int
1099 t4_cb_close(dev_t dev, int flag, int otyp, cred_t *credp)
1100 {
1101 	struct adapter *sc;
1102 
1103 	sc = ddi_get_soft_state(t4_list, getminor(dev));
1104 	if (sc == NULL)
1105 		return (EINVAL);
1106 
1107 	(void) atomic_swap_uint(&sc->open, 0);
1108 	return (0);
1109 }
1110 
1111 /* ARGSUSED */
1112 static int
1113 t4_cb_ioctl(dev_t dev, int cmd, intptr_t d, int mode, cred_t *credp, int *rp)
1114 {
1115 	int instance;
1116 	struct adapter *sc;
1117 	void *data = (void *)d;
1118 
1119 	if (crgetuid(credp) != 0)
1120 		return (EPERM);
1121 
1122 	instance = getminor(dev);
1123 	sc = ddi_get_soft_state(t4_list, instance);
1124 	if (sc == NULL)
1125 		return (EINVAL);
1126 
1127 	return (t4_ioctl(sc, cmd, data, mode));
1128 }
1129 
1130 static unsigned int
1131 getpf(struct adapter *sc)
1132 {
1133 	int rc, *data;
1134 	uint_t n, pf;
1135 
1136 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, sc->dip,
1137 	    DDI_PROP_DONTPASS, "reg", &data, &n);
1138 	if (rc != DDI_SUCCESS) {
1139 		cxgb_printf(sc->dip, CE_WARN,
1140 		    "failed to lookup \"reg\" property: %d", rc);
1141 		return (0xff);
1142 	}
1143 
1144 	pf = PCI_REG_FUNC_G(data[0]);
1145 	ddi_prop_free(data);
1146 
1147 	return (pf);
1148 }
1149 
1150 
1151 static struct fw_info *
1152 find_fw_info(int chip)
1153 {
1154 	u32 i;
1155 
1156 	fi[0].chip = CHELSIO_T4;
1157 	fi[0].fw_hdr.chip = FW_HDR_CHIP_T4;
1158 	fi[0].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T4));
1159 	fi[0].fw_hdr.intfver_nic = FW_INTFVER(T4, NIC);
1160 	fi[0].fw_hdr.intfver_vnic = FW_INTFVER(T4, VNIC);
1161 	fi[0].fw_hdr.intfver_ofld = FW_INTFVER(T4, OFLD);
1162 	fi[0].fw_hdr.intfver_ri = FW_INTFVER(T4, RI);
1163 	fi[0].fw_hdr.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU);
1164 	fi[0].fw_hdr.intfver_iscsi = FW_INTFVER(T4, ISCSI);
1165 	fi[0].fw_hdr.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU);
1166 	fi[0].fw_hdr.intfver_fcoe = FW_INTFVER(T4, FCOE);
1167 
1168 	fi[1].chip = CHELSIO_T5;
1169 	fi[1].fw_hdr.chip = FW_HDR_CHIP_T5;
1170 	fi[1].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T5));
1171 	fi[1].fw_hdr.intfver_nic = FW_INTFVER(T5, NIC);
1172 	fi[1].fw_hdr.intfver_vnic = FW_INTFVER(T5, VNIC);
1173 	fi[1].fw_hdr.intfver_ofld = FW_INTFVER(T5, OFLD);
1174 	fi[1].fw_hdr.intfver_ri = FW_INTFVER(T5, RI);
1175 	fi[1].fw_hdr.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU);
1176 	fi[1].fw_hdr.intfver_iscsi = FW_INTFVER(T5, ISCSI);
1177 	fi[1].fw_hdr.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU);
1178 	fi[1].fw_hdr.intfver_fcoe = FW_INTFVER(T5, FCOE);
1179 
1180 	fi[2].chip = CHELSIO_T6;
1181 	fi[2].fw_hdr.chip = FW_HDR_CHIP_T6;
1182 	fi[2].fw_hdr.fw_ver = cpu_to_be32(FW_VERSION(T6));
1183 	fi[2].fw_hdr.intfver_nic = FW_INTFVER(T6, NIC);
1184 	fi[2].fw_hdr.intfver_vnic = FW_INTFVER(T6, VNIC);
1185 	fi[2].fw_hdr.intfver_ofld = FW_INTFVER(T6, OFLD);
1186 	fi[2].fw_hdr.intfver_ri = FW_INTFVER(T6, RI);
1187 	fi[2].fw_hdr.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU);
1188 	fi[2].fw_hdr.intfver_iscsi = FW_INTFVER(T6, ISCSI);
1189 	fi[2].fw_hdr.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU);
1190 	fi[2].fw_hdr.intfver_fcoe = FW_INTFVER(T6, FCOE);
1191 
1192 	for (i = 0; i < ARRAY_SIZE(fi); i++) {
1193 		if (fi[i].chip == chip)
1194 			return &fi[i];
1195 	}
1196 
1197 	return NULL;
1198 }
1199 
1200 /*
1201  * Install a compatible firmware (if required), establish contact with it,
1202  * become the master, and reset the device.
1203  */
1204 static int
1205 prep_firmware(struct adapter *sc)
1206 {
1207 	int rc;
1208 	int fw_size;
1209 	int reset = 1;
1210 	enum dev_state state;
1211 	unsigned char *fw_data;
1212 	struct fw_info *fw_info;
1213 	struct fw_hdr *card_fw;
1214 
1215 	struct driver_properties *p = &sc->props;
1216 
1217 	/* Contact firmware, request master */
1218 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MUST, &state);
1219 	if (rc < 0) {
1220 		rc = -rc;
1221 		cxgb_printf(sc->dip, CE_WARN,
1222 		    "failed to connect to the firmware: %d.", rc);
1223 		return (rc);
1224 	}
1225 
1226 	if (rc == sc->mbox)
1227 		sc->flags |= MASTER_PF;
1228 
1229 	/* We may need FW version info for later reporting */
1230 	t4_get_version_info(sc);
1231 	fw_info = find_fw_info(CHELSIO_CHIP_VERSION(sc->params.chip));
1232 	/* allocate memory to read the header of the firmware on the
1233 	 * card
1234 	 */
1235 	if (!fw_info) {
1236 		cxgb_printf(sc->dip, CE_WARN,
1237 			    "unable to look up firmware information for chip %d.\n",
1238 			    CHELSIO_CHIP_VERSION(sc->params.chip));
1239 		return EINVAL;
1240 	}
1241 	card_fw = kmem_zalloc(sizeof(*card_fw), KM_SLEEP);
1242 	if(!card_fw) {
1243 		cxgb_printf(sc->dip, CE_WARN,
1244 			    "Memory allocation for card FW header failed\n");
1245 		return ENOMEM;
1246 	}
1247 	switch(CHELSIO_CHIP_VERSION(sc->params.chip)) {
1248 	case CHELSIO_T4:
1249 		fw_data = t4fw_data;
1250 		fw_size = t4fw_size;
1251 		break;
1252 	case CHELSIO_T5:
1253 		fw_data = t5fw_data;
1254 		fw_size = t5fw_size;
1255 		break;
1256 	case CHELSIO_T6:
1257 		fw_data = t6fw_data;
1258 		fw_size = t6fw_size;
1259 		break;
1260 	default:
1261 		cxgb_printf(sc->dip, CE_WARN, "Adapter type not supported\n");
1262 		kmem_free(card_fw, sizeof(*card_fw));
1263 		return EINVAL;
1264 	}
1265 
1266 	rc = -t4_prep_fw(sc, fw_info, fw_data, fw_size, card_fw,
1267 			 p->t4_fw_install, state, &reset);
1268 
1269 	kmem_free(card_fw, sizeof(*card_fw));
1270 
1271 	if (rc != 0) {
1272 		cxgb_printf(sc->dip, CE_WARN,
1273 		    "failed to install firmware: %d", rc);
1274 		return (rc);
1275 	} else {
1276 		/* refresh */
1277 		(void) t4_check_fw_version(sc);
1278 	}
1279 
1280 	/* Reset device */
1281 	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
1282 	if (rc != 0) {
1283 		cxgb_printf(sc->dip, CE_WARN,
1284 		    "firmware reset failed: %d.", rc);
1285 		if (rc != ETIMEDOUT && rc != EIO)
1286 			(void) t4_fw_bye(sc, sc->mbox);
1287 		return (rc);
1288 	}
1289 
1290 	/* Partition adapter resources as specified in the config file. */
1291 	if (sc->flags & MASTER_PF) {
1292 		/* Handle default vs special T4 config file */
1293 
1294 		rc = partition_resources(sc);
1295 		if (rc != 0)
1296 			goto err;	/* error message displayed already */
1297 	}
1298 
1299 	sc->flags |= FW_OK;
1300 	return (0);
1301 err:
1302 	return (rc);
1303 
1304 }
1305 
1306 static const struct memwin t4_memwin[] = {
1307 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1308 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1309 	{ MEMWIN2_BASE, MEMWIN2_APERTURE }
1310 };
1311 
1312 static const struct memwin t5_memwin[] = {
1313 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1314 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1315 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
1316 };
1317 
1318 #define	FW_PARAM_DEV(param) \
1319 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
1320 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
1321 #define	FW_PARAM_PFVF(param) \
1322 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
1323 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
1324 
1325 /*
1326  * Verify that the memory range specified by the memtype/offset/len pair is
1327  * valid and lies entirely within the memtype specified.  The global address of
1328  * the start of the range is returned in addr.
1329  */
1330 int
1331 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
1332 	uint32_t *addr)
1333 {
1334 	uint32_t em, addr_len, maddr, mlen;
1335 
1336 	/* Memory can only be accessed in naturally aligned 4 byte units */
1337 	if (off & 3 || len & 3 || len == 0)
1338 		return (EINVAL);
1339 
1340 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
1341 	switch (mtype) {
1342 		case MEM_EDC0:
1343 			if (!(em & F_EDRAM0_ENABLE))
1344 				return (EINVAL);
1345 			addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
1346 			maddr = G_EDRAM0_BASE(addr_len) << 20;
1347 			mlen = G_EDRAM0_SIZE(addr_len) << 20;
1348 			break;
1349 		case MEM_EDC1:
1350 			if (!(em & F_EDRAM1_ENABLE))
1351 				return (EINVAL);
1352 			addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
1353 			maddr = G_EDRAM1_BASE(addr_len) << 20;
1354 			mlen = G_EDRAM1_SIZE(addr_len) << 20;
1355 			break;
1356 		case MEM_MC:
1357 			if (!(em & F_EXT_MEM_ENABLE))
1358 				return (EINVAL);
1359 			addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
1360 			maddr = G_EXT_MEM_BASE(addr_len) << 20;
1361 			mlen = G_EXT_MEM_SIZE(addr_len) << 20;
1362 			break;
1363 		case MEM_MC1:
1364 			if (is_t4(sc->params.chip) || !(em & F_EXT_MEM1_ENABLE))
1365 				return (EINVAL);
1366 			addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
1367 			maddr = G_EXT_MEM1_BASE(addr_len) << 20;
1368 			mlen = G_EXT_MEM1_SIZE(addr_len) << 20;
1369 			break;
1370 		default:
1371 			return (EINVAL);
1372 	}
1373 
1374 	if (mlen > 0 && off < mlen && off + len <= mlen) {
1375 		*addr = maddr + off;    /* global address */
1376 		return (0);
1377 	}
1378 
1379 	return (EFAULT);
1380 }
1381 
1382 void
1383 memwin_info(struct adapter *sc, int win, uint32_t *base, uint32_t *aperture)
1384 {
1385 	const struct memwin *mw;
1386 
1387 	if (is_t4(sc->params.chip)) {
1388 		mw = &t4_memwin[win];
1389 	} else {
1390 		mw = &t5_memwin[win];
1391 	}
1392 
1393 	if (base != NULL)
1394 		*base = mw->base;
1395 	if (aperture != NULL)
1396 		*aperture = mw->aperture;
1397 }
1398 
1399 /*
1400  * Upload configuration file to card's memory.
1401  */
1402 static int
1403 upload_config_file(struct adapter *sc, uint32_t *mt, uint32_t *ma)
1404 {
1405 	int rc = 0, cflen;
1406 	u_int i, n;
1407 	uint32_t param, val, addr, mtype, maddr;
1408 	uint32_t off, mw_base, mw_aperture;
1409 	const uint32_t *cfdata;
1410 
1411 	/* Figure out where the firmware wants us to upload it. */
1412 	param = FW_PARAM_DEV(CF);
1413 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
1414 	if (rc != 0) {
1415 		/* Firmwares without config file support will fail this way */
1416 		cxgb_printf(sc->dip, CE_WARN,
1417 		    "failed to query config file location: %d.\n", rc);
1418 		return (rc);
1419 	}
1420 	*mt = mtype = G_FW_PARAMS_PARAM_Y(val);
1421 	*ma = maddr = G_FW_PARAMS_PARAM_Z(val) << 16;
1422 
1423 	switch (CHELSIO_CHIP_VERSION(sc->params.chip)) {
1424 	case CHELSIO_T4:
1425 		cflen = t4cfg_size & ~3;
1426 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1427 		cfdata = (const uint32_t *)t4cfg_data;
1428 		break;
1429 	case CHELSIO_T5:
1430 		cflen = t5cfg_size & ~3;
1431 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1432 		cfdata = (const uint32_t *)t5cfg_data;
1433 		break;
1434 	case CHELSIO_T6:
1435 		cflen = t6cfg_size & ~3;
1436 		/* LINTED: E_BAD_PTR_CAST_ALIGN */
1437 		cfdata = (const uint32_t *)t6cfg_data;
1438 		break;
1439 	default:
1440 		cxgb_printf(sc->dip, CE_WARN,
1441 			    "Invalid Adapter detected\n");
1442 		return EINVAL;
1443 	}
1444 
1445 	if (cflen > FLASH_CFG_MAX_SIZE) {
1446 		cxgb_printf(sc->dip, CE_WARN,
1447 		    "config file too long (%d, max allowed is %d).  ",
1448 		    cflen, FLASH_CFG_MAX_SIZE);
1449 		return (EFBIG);
1450 	}
1451 
1452 	rc = validate_mt_off_len(sc, mtype, maddr, cflen, &addr);
1453 	if (rc != 0) {
1454 
1455 		cxgb_printf(sc->dip, CE_WARN,
1456 		    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
1457 		    "Will try to use the config on the card, if any.\n",
1458 		    __func__, mtype, maddr, cflen, rc);
1459 		return (EFAULT);
1460 	}
1461 
1462 	memwin_info(sc, 2, &mw_base, &mw_aperture);
1463 	while (cflen) {
1464 		off = position_memwin(sc, 2, addr);
1465 		n = min(cflen, mw_aperture - off);
1466 		for (i = 0; i < n; i += 4)
1467 			t4_write_reg(sc, mw_base + off + i, *cfdata++);
1468 		cflen -= n;
1469 		addr += n;
1470 	}
1471 
1472 	return (rc);
1473 }
1474 
1475 /*
1476  * Partition chip resources for use between various PFs, VFs, etc.  This is done
1477  * by uploading the firmware configuration file to the adapter and instructing
1478  * the firmware to process it.
1479  */
1480 static int
1481 partition_resources(struct adapter *sc)
1482 {
1483 	int rc;
1484 	struct fw_caps_config_cmd caps;
1485 	uint32_t mtype, maddr, finicsum, cfcsum;
1486 
1487 	rc = upload_config_file(sc, &mtype, &maddr);
1488 	if (rc != 0) {
1489 		mtype = FW_MEMTYPE_CF_FLASH;
1490 		maddr = t4_flash_cfg_addr(sc);
1491 	}
1492 
1493 	bzero(&caps, sizeof (caps));
1494 	caps.op_to_write = BE_32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1495 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1496 	caps.cfvalid_to_len16 = BE_32(F_FW_CAPS_CONFIG_CMD_CFVALID |
1497 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
1498 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) | FW_LEN16(caps));
1499 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), &caps);
1500 	if (rc != 0) {
1501 		cxgb_printf(sc->dip, CE_WARN,
1502 		    "failed to pre-process config file: %d.\n", rc);
1503 		return (rc);
1504 	}
1505 
1506 	finicsum = ntohl(caps.finicsum);
1507 	cfcsum = ntohl(caps.cfcsum);
1508 	if (finicsum != cfcsum) {
1509 		cxgb_printf(sc->dip, CE_WARN,
1510 		    "WARNING: config file checksum mismatch: %08x %08x\n",
1511 		    finicsum, cfcsum);
1512 	}
1513 	sc->cfcsum = cfcsum;
1514 
1515 	/* TODO: Need to configure this correctly */
1516 	caps.toecaps = htons(FW_CAPS_CONFIG_TOE);
1517 	caps.iscsicaps = 0;
1518 	caps.rdmacaps = 0;
1519 	caps.fcoecaps = 0;
1520 	/* TODO: Disable VNIC cap for now */
1521 	caps.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
1522 
1523 	caps.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1524 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
1525 	caps.cfvalid_to_len16 = htonl(FW_LEN16(caps));
1526 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), NULL);
1527 	if (rc != 0) {
1528 		cxgb_printf(sc->dip, CE_WARN,
1529 		    "failed to process config file: %d.\n", rc);
1530 		return (rc);
1531 	}
1532 
1533 	return (0);
1534 }
1535 
1536 /*
1537  * Tweak configuration based on module parameters, etc.  Most of these have
1538  * defaults assigned to them by Firmware Configuration Files (if we're using
1539  * them) but need to be explicitly set if we're using hard-coded
1540  * initialization.  But even in the case of using Firmware Configuration
1541  * Files, we'd like to expose the ability to change these via module
1542  * parameters so these are essentially common tweaks/settings for
1543  * Configuration Files and hard-coded initialization ...
1544  */
1545 static int
1546 adap__pre_init_tweaks(struct adapter *sc)
1547 {
1548 	int rx_dma_offset = 2; /* Offset of RX packets into DMA buffers */
1549 
1550 	/*
1551 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
1552 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
1553 	 * 64B Cache Line Size ...
1554 	 */
1555 	(void) t4_fixup_host_params_compat(sc, PAGE_SIZE, CACHE_LINE, T5_LAST_REV);
1556 
1557 	t4_set_reg_field(sc, A_SGE_CONTROL,
1558 			 V_PKTSHIFT(M_PKTSHIFT), V_PKTSHIFT(rx_dma_offset));
1559 
1560 	return 0;
1561 }
1562 /*
1563  * Retrieve parameters that are needed (or nice to have) prior to calling
1564  * t4_sge_init and t4_fw_initialize.
1565  */
1566 static int
1567 get_params__pre_init(struct adapter *sc)
1568 {
1569 	int rc;
1570 	uint32_t param[2], val[2];
1571 	struct fw_devlog_cmd cmd;
1572 	struct devlog_params *dlog = &sc->params.devlog;
1573 
1574 	/*
1575 	 * Grab the raw VPD parameters.
1576 	 */
1577 	rc = -t4_get_raw_vpd_params(sc, &sc->params.vpd);
1578 	if (rc != 0) {
1579 		cxgb_printf(sc->dip, CE_WARN,
1580 		    "failed to query VPD parameters (pre_init): %d.\n", rc);
1581 		return (rc);
1582 	}
1583 
1584 	param[0] = FW_PARAM_DEV(PORTVEC);
1585 	param[1] = FW_PARAM_DEV(CCLK);
1586 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
1587 	if (rc != 0) {
1588 		cxgb_printf(sc->dip, CE_WARN,
1589 		    "failed to query parameters (pre_init): %d.\n", rc);
1590 		return (rc);
1591 	}
1592 
1593 	sc->params.portvec = val[0];
1594 	sc->params.nports = 0;
1595 	while (val[0]) {
1596 		sc->params.nports++;
1597 		val[0] &= val[0] - 1;
1598 	}
1599 
1600 	sc->params.vpd.cclk = val[1];
1601 
1602 	/* Read device log parameters. */
1603 	bzero(&cmd, sizeof (cmd));
1604 	cmd.op_to_write = htonl(V_FW_CMD_OP(FW_DEVLOG_CMD) |
1605 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1606 	cmd.retval_len16 = htonl(FW_LEN16(cmd));
1607 	rc = -t4_wr_mbox(sc, sc->mbox, &cmd, sizeof (cmd), &cmd);
1608 	if (rc != 0) {
1609 		cxgb_printf(sc->dip, CE_WARN,
1610 		    "failed to get devlog parameters: %d.\n", rc);
1611 		bzero(dlog, sizeof (*dlog));
1612 		rc = 0;	/* devlog isn't critical for device operation */
1613 	} else {
1614 		val[0] = ntohl(cmd.memtype_devlog_memaddr16_devlog);
1615 		dlog->memtype = G_FW_DEVLOG_CMD_MEMTYPE_DEVLOG(val[0]);
1616 		dlog->start = G_FW_DEVLOG_CMD_MEMADDR16_DEVLOG(val[0]) << 4;
1617 		dlog->size = ntohl(cmd.memsize_devlog);
1618 	}
1619 
1620 	return (rc);
1621 }
1622 
1623 /*
1624  * Retrieve various parameters that are of interest to the driver.  The device
1625  * has been initialized by the firmware at this point.
1626  */
1627 static int
1628 get_params__post_init(struct adapter *sc)
1629 {
1630 	int rc;
1631 	uint32_t param[7], val[7];
1632 	struct fw_caps_config_cmd caps;
1633 
1634 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
1635 	param[1] = FW_PARAM_PFVF(EQ_START);
1636 	param[2] = FW_PARAM_PFVF(FILTER_START);
1637 	param[3] = FW_PARAM_PFVF(FILTER_END);
1638 	param[4] = FW_PARAM_PFVF(L2T_START);
1639 	param[5] = FW_PARAM_PFVF(L2T_END);
1640 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
1641 	if (rc != 0) {
1642 		cxgb_printf(sc->dip, CE_WARN,
1643 		    "failed to query parameters (post_init): %d.\n", rc);
1644 		return (rc);
1645 	}
1646 
1647 	/* LINTED: E_ASSIGN_NARROW_CONV */
1648 	sc->sge.iq_start = val[0];
1649 	sc->sge.eq_start = val[1];
1650 	sc->tids.ftid_base = val[2];
1651 	sc->tids.nftids = val[3] - val[2] + 1;
1652 	sc->vres.l2t.start = val[4];
1653 	sc->vres.l2t.size = val[5] - val[4] + 1;
1654 
1655 	/* get capabilites */
1656 	bzero(&caps, sizeof (caps));
1657 	caps.op_to_write = htonl(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
1658 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
1659 	caps.cfvalid_to_len16 = htonl(FW_LEN16(caps));
1660 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof (caps), &caps);
1661 	if (rc != 0) {
1662 		cxgb_printf(sc->dip, CE_WARN,
1663 		    "failed to get card capabilities: %d.\n", rc);
1664 		return (rc);
1665 	}
1666 
1667 	if (caps.toecaps != 0) {
1668 		/* query offload-related parameters */
1669 		param[0] = FW_PARAM_DEV(NTID);
1670 		param[1] = FW_PARAM_PFVF(SERVER_START);
1671 		param[2] = FW_PARAM_PFVF(SERVER_END);
1672 		param[3] = FW_PARAM_PFVF(TDDP_START);
1673 		param[4] = FW_PARAM_PFVF(TDDP_END);
1674 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
1675 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
1676 		if (rc != 0) {
1677 			cxgb_printf(sc->dip, CE_WARN,
1678 			    "failed to query TOE parameters: %d.\n", rc);
1679 			return (rc);
1680 		}
1681 		sc->tids.ntids = val[0];
1682 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
1683 		sc->tids.stid_base = val[1];
1684 		sc->tids.nstids = val[2] - val[1] + 1;
1685 		sc->vres.ddp.start = val[3];
1686 		sc->vres.ddp.size = val[4] - val[3] + 1;
1687 		sc->params.ofldq_wr_cred = val[5];
1688 		sc->params.offload = 1;
1689 	}
1690 
1691 	/* These are finalized by FW initialization, load their values now */
1692 	val[0] = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
1693 	sc->params.tp.tre = G_TIMERRESOLUTION(val[0]);
1694 	sc->params.tp.dack_re = G_DELAYEDACKRESOLUTION(val[0]);
1695 	t4_read_mtu_tbl(sc, sc->params.mtus, NULL);
1696 
1697 	return (rc);
1698 }
1699 
1700 static int
1701 set_params__post_init(struct adapter *sc)
1702 {
1703 	uint32_t param, val;
1704 
1705 	/* ask for encapsulated CPLs */
1706 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
1707 	val = 1;
1708 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
1709 
1710 	return (0);
1711 }
1712 
1713 /* TODO: verify */
1714 static void
1715 setup_memwin(struct adapter *sc)
1716 {
1717 	pci_regspec_t *data;
1718 	int rc;
1719 	uint_t n;
1720 	uintptr_t bar0;
1721 	uintptr_t mem_win0_base, mem_win1_base, mem_win2_base;
1722 	uintptr_t mem_win2_aperture;
1723 
1724 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, sc->dip,
1725 	    DDI_PROP_DONTPASS, "assigned-addresses", (int **)&data, &n);
1726 	if (rc != DDI_SUCCESS) {
1727 		cxgb_printf(sc->dip, CE_WARN,
1728 		    "failed to lookup \"assigned-addresses\" property: %d", rc);
1729 		return;
1730 	}
1731 	n /= sizeof (*data);
1732 
1733 	bar0 = ((uint64_t)data[0].pci_phys_mid << 32) | data[0].pci_phys_low;
1734 	ddi_prop_free(data);
1735 
1736 	if (is_t4(sc->params.chip)) {
1737 		mem_win0_base = bar0 + MEMWIN0_BASE;
1738 		mem_win1_base = bar0 + MEMWIN1_BASE;
1739 		mem_win2_base = bar0 + MEMWIN2_BASE;
1740 		mem_win2_aperture = MEMWIN2_APERTURE;
1741 	} else {
1742 		/* For T5, only relative offset inside the PCIe BAR is passed */
1743 		mem_win0_base = MEMWIN0_BASE;
1744 		mem_win1_base = MEMWIN1_BASE;
1745 		mem_win2_base = MEMWIN2_BASE_T5;
1746 		mem_win2_aperture = MEMWIN2_APERTURE_T5;
1747 	}
1748 
1749 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 0),
1750 	    mem_win0_base | V_BIR(0) |
1751 	    V_WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
1752 
1753 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 1),
1754 	    mem_win1_base | V_BIR(0) |
1755 	    V_WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
1756 
1757 	t4_write_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2),
1758 	    mem_win2_base | V_BIR(0) |
1759 	    V_WINDOW(ilog2(mem_win2_aperture) - 10));
1760 
1761 	/* flush */
1762 	(void)t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
1763 }
1764 
1765 /*
1766  * Positions the memory window such that it can be used to access the specified
1767  * address in the chip's address space.  The return value is the offset of addr
1768  * from the start of the window.
1769  */
1770 uint32_t
1771 position_memwin(struct adapter *sc, int n, uint32_t addr)
1772 {
1773 	uint32_t start, pf;
1774 	uint32_t reg;
1775 
1776 	if (addr & 3) {
1777 		cxgb_printf(sc->dip, CE_WARN,
1778 		    "addr (0x%x) is not at a 4B boundary.\n", addr);
1779 		return (EFAULT);
1780 	}
1781 
1782 	if (is_t4(sc->params.chip)) {
1783 		pf = 0;
1784 		start = addr & ~0xf;    /* start must be 16B aligned */
1785 	} else {
1786 		pf = V_PFNUM(sc->pf);
1787 		start = addr & ~0x7f;   /* start must be 128B aligned */
1788 	}
1789 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, n);
1790 
1791 	t4_write_reg(sc, reg, start | pf);
1792 	(void) t4_read_reg(sc, reg);
1793 
1794 	return (addr - start);
1795 }
1796 
1797 
1798 /*
1799  * Reads the named property and fills up the "data" array (which has at least
1800  * "count" elements).  We first try and lookup the property for our dev_t and
1801  * then retry with DDI_DEV_T_ANY if it's not found.
1802  *
1803  * Returns non-zero if the property was found and "data" has been updated.
1804  */
1805 static int
1806 prop_lookup_int_array(struct adapter *sc, char *name, int *data, uint_t count)
1807 {
1808 	dev_info_t *dip = sc->dip;
1809 	dev_t dev = sc->dev;
1810 	int rc, *d;
1811 	uint_t i, n;
1812 
1813 	rc = ddi_prop_lookup_int_array(dev, dip, DDI_PROP_DONTPASS,
1814 	    name, &d, &n);
1815 	if (rc == DDI_PROP_SUCCESS)
1816 		goto found;
1817 
1818 	if (rc != DDI_PROP_NOT_FOUND) {
1819 		cxgb_printf(dip, CE_WARN,
1820 		    "failed to lookup property %s for minor %d: %d.",
1821 		    name, getminor(dev), rc);
1822 		return (0);
1823 	}
1824 
1825 	rc = ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
1826 	    name, &d, &n);
1827 	if (rc == DDI_PROP_SUCCESS)
1828 		goto found;
1829 
1830 	if (rc != DDI_PROP_NOT_FOUND) {
1831 		cxgb_printf(dip, CE_WARN,
1832 		    "failed to lookup property %s: %d.", name, rc);
1833 		return (0);
1834 	}
1835 
1836 	return (0);
1837 
1838 found:
1839 	if (n > count) {
1840 		cxgb_printf(dip, CE_NOTE,
1841 		    "property %s has too many elements (%d), ignoring extras",
1842 		    name, n);
1843 	}
1844 
1845 	for (i = 0; i < n && i < count; i++)
1846 		data[i] = d[i];
1847 	ddi_prop_free(d);
1848 
1849 	return (1);
1850 }
1851 
1852 static int
1853 prop_lookup_int(struct adapter *sc, char *name, int defval)
1854 {
1855 	int rc;
1856 
1857 	rc = ddi_prop_get_int(sc->dev, sc->dip, DDI_PROP_DONTPASS, name, -1);
1858 	if (rc != -1)
1859 		return (rc);
1860 
1861 	return (ddi_prop_get_int(DDI_DEV_T_ANY, sc->dip, DDI_PROP_DONTPASS,
1862 	    name, defval));
1863 }
1864 
1865 static int
1866 init_driver_props(struct adapter *sc, struct driver_properties *p)
1867 {
1868 	dev_t dev = sc->dev;
1869 	dev_info_t *dip = sc->dip;
1870 	int i, *data;
1871 	uint_t tmr[SGE_NTIMERS] = {5, 10, 20, 50, 100, 200};
1872 	uint_t cnt[SGE_NCOUNTERS] = {1, 8, 16, 32}; /* 63 max */
1873 
1874 	/*
1875 	 * Holdoff timer
1876 	 */
1877 	data = &p->timer_val[0];
1878 	for (i = 0; i < SGE_NTIMERS; i++)
1879 		data[i] = tmr[i];
1880 	(void) prop_lookup_int_array(sc, "holdoff-timer-values", data,
1881 	    SGE_NTIMERS);
1882 	for (i = 0; i < SGE_NTIMERS; i++) {
1883 		int limit = 200U;
1884 		if (data[i] > limit) {
1885 			cxgb_printf(dip, CE_WARN,
1886 			    "holdoff timer %d is too high (%d), lowered to %d.",
1887 			    i, data[i], limit);
1888 			data[i] = limit;
1889 		}
1890 	}
1891 	(void) ddi_prop_update_int_array(dev, dip, "holdoff-timer-values",
1892 	    data, SGE_NTIMERS);
1893 
1894 	/*
1895 	 * Holdoff packet counter
1896 	 */
1897 	data = &p->counter_val[0];
1898 	for (i = 0; i < SGE_NCOUNTERS; i++)
1899 		data[i] = cnt[i];
1900 	(void) prop_lookup_int_array(sc, "holdoff-pkt-counter-values", data,
1901 	    SGE_NCOUNTERS);
1902 	for (i = 0; i < SGE_NCOUNTERS; i++) {
1903 		int limit = M_THRESHOLD_0;
1904 		if (data[i] > limit) {
1905 			cxgb_printf(dip, CE_WARN,
1906 			    "holdoff pkt-counter %d is too high (%d), "
1907 			    "lowered to %d.", i, data[i], limit);
1908 			data[i] = limit;
1909 		}
1910 	}
1911 	(void) ddi_prop_update_int_array(dev, dip, "holdoff-pkt-counter-values",
1912 	    data, SGE_NCOUNTERS);
1913 
1914 	/*
1915 	 * Maximum # of tx and rx queues to use for each
1916 	 * 100G, 40G, 25G, 10G and 1G port.
1917 	 */
1918 	p->max_ntxq_10g = prop_lookup_int(sc, "max-ntxq-10G-port", 8);
1919 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-10G-port",
1920 	    p->max_ntxq_10g);
1921 
1922 	p->max_nrxq_10g = prop_lookup_int(sc, "max-nrxq-10G-port", 8);
1923 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-10G-port",
1924 	    p->max_nrxq_10g);
1925 
1926 	p->max_ntxq_1g = prop_lookup_int(sc, "max-ntxq-1G-port", 2);
1927 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-1G-port",
1928 	    p->max_ntxq_1g);
1929 
1930 	p->max_nrxq_1g = prop_lookup_int(sc, "max-nrxq-1G-port", 2);
1931 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-1G-port",
1932 	    p->max_nrxq_1g);
1933 
1934 #ifdef TCP_OFFLOAD_ENABLE
1935 	p->max_nofldtxq_10g = prop_lookup_int(sc, "max-nofldtxq-10G-port", 8);
1936 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-10G-port",
1937 	    p->max_nofldtxq_10g);
1938 
1939 	p->max_nofldrxq_10g = prop_lookup_int(sc, "max-nofldrxq-10G-port", 2);
1940 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-10G-port",
1941 	    p->max_nofldrxq_10g);
1942 
1943 	p->max_nofldtxq_1g = prop_lookup_int(sc, "max-nofldtxq-1G-port", 2);
1944 	(void) ddi_prop_update_int(dev, dip, "max-ntxq-1G-port",
1945 	    p->max_nofldtxq_1g);
1946 
1947 	p->max_nofldrxq_1g = prop_lookup_int(sc, "max-nofldrxq-1G-port", 1);
1948 	(void) ddi_prop_update_int(dev, dip, "max-nrxq-1G-port",
1949 	    p->max_nofldrxq_1g);
1950 #endif
1951 
1952 	/*
1953 	 * Holdoff parameters for 10G and 1G ports.
1954 	 */
1955 	p->tmr_idx_10g = prop_lookup_int(sc, "holdoff-timer-idx-10G", 0);
1956 	(void) ddi_prop_update_int(dev, dip, "holdoff-timer-idx-10G",
1957 	    p->tmr_idx_10g);
1958 
1959 	p->pktc_idx_10g = prop_lookup_int(sc, "holdoff-pktc-idx-10G", 2);
1960 	(void) ddi_prop_update_int(dev, dip, "holdoff-pktc-idx-10G",
1961 	    p->pktc_idx_10g);
1962 
1963 	p->tmr_idx_1g = prop_lookup_int(sc, "holdoff-timer-idx-1G", 0);
1964 	(void) ddi_prop_update_int(dev, dip, "holdoff-timer-idx-1G",
1965 	    p->tmr_idx_1g);
1966 
1967 	p->pktc_idx_1g = prop_lookup_int(sc, "holdoff-pktc-idx-1G", 2);
1968 	(void) ddi_prop_update_int(dev, dip, "holdoff-pktc-idx-1G",
1969 	    p->pktc_idx_1g);
1970 
1971 	/*
1972 	 * Size (number of entries) of each tx and rx queue.
1973 	 */
1974 	i = prop_lookup_int(sc, "qsize-txq", TX_EQ_QSIZE);
1975 	p->qsize_txq = max(i, 128);
1976 	if (p->qsize_txq != i) {
1977 		cxgb_printf(dip, CE_WARN,
1978 		    "using %d instead of %d as the tx queue size",
1979 		    p->qsize_txq, i);
1980 	}
1981 	(void) ddi_prop_update_int(dev, dip, "qsize-txq", p->qsize_txq);
1982 
1983 	i = prop_lookup_int(sc, "qsize-rxq", RX_IQ_QSIZE);
1984 	p->qsize_rxq = max(i, 128);
1985 	while (p->qsize_rxq & 7)
1986 		p->qsize_rxq--;
1987 	if (p->qsize_rxq != i) {
1988 		cxgb_printf(dip, CE_WARN,
1989 		    "using %d instead of %d as the rx queue size",
1990 		    p->qsize_rxq, i);
1991 	}
1992 	(void) ddi_prop_update_int(dev, dip, "qsize-rxq", p->qsize_rxq);
1993 
1994 	/*
1995 	 * Interrupt types allowed.
1996 	 * Bits 0, 1, 2 = INTx, MSI, MSI-X respectively.  See sys/ddi_intr.h
1997 	 */
1998 	p->intr_types = prop_lookup_int(sc, "interrupt-types",
1999 	    DDI_INTR_TYPE_MSIX | DDI_INTR_TYPE_MSI | DDI_INTR_TYPE_FIXED);
2000 	(void) ddi_prop_update_int(dev, dip, "interrupt-types", p->intr_types);
2001 
2002 	/*
2003 	 * Forwarded interrupt queues.  Create this property to force the driver
2004 	 * to use forwarded interrupt queues.
2005 	 */
2006 	if (ddi_prop_exists(dev, dip, DDI_PROP_DONTPASS,
2007 	    "interrupt-forwarding") != 0 ||
2008 	    ddi_prop_exists(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
2009 	    "interrupt-forwarding") != 0) {
2010 		UNIMPLEMENTED();
2011 		(void) ddi_prop_create(dev, dip, DDI_PROP_CANSLEEP,
2012 		    "interrupt-forwarding", NULL, 0);
2013 	}
2014 
2015 	/*
2016 	 * Write combining
2017 	 * 0 to disable, 1 to enable
2018 	 */
2019 	p->wc = prop_lookup_int(sc, "write-combine", 1);
2020 	cxgb_printf(dip, CE_WARN, "write-combine: using of %d", p->wc);
2021 	if (p->wc != 0 && p->wc != 1) {
2022 		cxgb_printf(dip, CE_WARN,
2023 		    "write-combine: using 1 instead of %d", p->wc);
2024 		p->wc = 1;
2025 	}
2026 	(void) ddi_prop_update_int(dev, dip, "write-combine", p->wc);
2027 
2028 	p->t4_fw_install = prop_lookup_int(sc, "t4_fw_install", 1);
2029 	if (p->t4_fw_install != 0 && p->t4_fw_install != 2)
2030 		p->t4_fw_install = 1;
2031 	(void) ddi_prop_update_int(dev, dip, "t4_fw_install", p->t4_fw_install);
2032 
2033 	/* Multiple Rings */
2034 	p->multi_rings = prop_lookup_int(sc, "multi-rings", 1);
2035 	if (p->multi_rings != 0 && p->multi_rings != 1) {
2036 		cxgb_printf(dip, CE_NOTE,
2037 			   "multi-rings: using value 1 instead of %d", p->multi_rings);
2038 		p->multi_rings = 1;
2039 	}
2040 
2041 	(void) ddi_prop_update_int(dev, dip, "multi-rings", p->multi_rings);
2042 
2043 	return (0);
2044 }
2045 
2046 static int
2047 remove_extra_props(struct adapter *sc, int n10g, int n1g)
2048 {
2049 	if (n10g == 0) {
2050 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-ntxq-10G-port");
2051 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-nrxq-10G-port");
2052 		(void) ddi_prop_remove(sc->dev, sc->dip,
2053 		    "holdoff-timer-idx-10G");
2054 		(void) ddi_prop_remove(sc->dev, sc->dip,
2055 		    "holdoff-pktc-idx-10G");
2056 	}
2057 
2058 	if (n1g == 0) {
2059 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-ntxq-1G-port");
2060 		(void) ddi_prop_remove(sc->dev, sc->dip, "max-nrxq-1G-port");
2061 		(void) ddi_prop_remove(sc->dev, sc->dip,
2062 		    "holdoff-timer-idx-1G");
2063 		(void) ddi_prop_remove(sc->dev, sc->dip, "holdoff-pktc-idx-1G");
2064 	}
2065 
2066 	return (0);
2067 }
2068 
2069 static int
2070 cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g,
2071     struct intrs_and_queues *iaq)
2072 {
2073 	struct driver_properties *p = &sc->props;
2074 	int rc, itype, itypes, navail, nc, nrxq10g, nrxq1g, n;
2075 	int nofldrxq10g = 0, nofldrxq1g = 0;
2076 
2077 	bzero(iaq, sizeof (*iaq));
2078 	nc = ncpus;	/* our snapshot of the number of CPUs */
2079 	iaq->ntxq10g = min(nc, p->max_ntxq_10g);
2080 	iaq->ntxq1g = min(nc, p->max_ntxq_1g);
2081 	iaq->nrxq10g = nrxq10g = min(nc, p->max_nrxq_10g);
2082 	iaq->nrxq1g = nrxq1g = min(nc, p->max_nrxq_1g);
2083 #ifdef TCP_OFFLOAD_ENABLE
2084 	iaq->nofldtxq10g = min(nc, p->max_nofldtxq_10g);
2085 	iaq->nofldtxq1g = min(nc, p->max_nofldtxq_1g);
2086 	iaq->nofldrxq10g = nofldrxq10g = min(nc, p->max_nofldrxq_10g);
2087 	iaq->nofldrxq1g = nofldrxq1g = min(nc, p->max_nofldrxq_1g);
2088 #endif
2089 
2090 	rc = ddi_intr_get_supported_types(sc->dip, &itypes);
2091 	if (rc != DDI_SUCCESS) {
2092 		cxgb_printf(sc->dip, CE_WARN,
2093 		    "failed to determine supported interrupt types: %d", rc);
2094 		return (rc);
2095 	}
2096 
2097 	for (itype = DDI_INTR_TYPE_MSIX; itype; itype >>= 1) {
2098 		ASSERT(itype == DDI_INTR_TYPE_MSIX ||
2099 		    itype == DDI_INTR_TYPE_MSI ||
2100 		    itype == DDI_INTR_TYPE_FIXED);
2101 
2102 		if ((itype & itypes & p->intr_types) == 0)
2103 			continue;	/* not supported or not allowed */
2104 
2105 		navail = 0;
2106 		rc = ddi_intr_get_navail(sc->dip, itype, &navail);
2107 		if (rc != DDI_SUCCESS || navail == 0) {
2108 			cxgb_printf(sc->dip, CE_WARN,
2109 			    "failed to get # of interrupts for type %d: %d",
2110 			    itype, rc);
2111 			continue;	/* carry on */
2112 		}
2113 
2114 		iaq->intr_type = itype;
2115 		if (navail == 0)
2116 			continue;
2117 
2118 		/*
2119 		 * Best option: an interrupt vector for errors, one for the
2120 		 * firmware event queue, and one each for each rxq (NIC as well
2121 		 * as offload).
2122 		 */
2123 		iaq->nirq = T4_EXTRA_INTR;
2124 		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2125 		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2126 
2127 		if (iaq->nirq <= navail &&
2128 		    (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq))) {
2129 			iaq->intr_fwd = 0;
2130 			goto allocate;
2131 		}
2132 
2133 		/*
2134 		 * Second best option: an interrupt vector for errors, one for
2135 		 * the firmware event queue, and one each for either NIC or
2136 		 * offload rxq's.
2137 		 */
2138 		iaq->nirq = T4_EXTRA_INTR;
2139 		iaq->nirq += n10g * max(nrxq10g, nofldrxq10g);
2140 		iaq->nirq += n1g * max(nrxq1g, nofldrxq1g);
2141 		if (iaq->nirq <= navail &&
2142 		    (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq))) {
2143 			iaq->intr_fwd = 1;
2144 			goto allocate;
2145 		}
2146 
2147 		/*
2148 		 * Next best option: an interrupt vector for errors, one for the
2149 		 * firmware event queue, and at least one per port.  At this
2150 		 * point we know we'll have to downsize nrxq or nofldrxq to fit
2151 		 * what's available to us.
2152 		 */
2153 		iaq->nirq = T4_EXTRA_INTR;
2154 		iaq->nirq += n10g + n1g;
2155 		if (iaq->nirq <= navail) {
2156 			int leftover = navail - iaq->nirq;
2157 
2158 			if (n10g > 0) {
2159 				int target = max(nrxq10g, nofldrxq10g);
2160 
2161 				n = 1;
2162 				while (n < target && leftover >= n10g) {
2163 					leftover -= n10g;
2164 					iaq->nirq += n10g;
2165 					n++;
2166 				}
2167 				iaq->nrxq10g = min(n, nrxq10g);
2168 #ifdef TCP_OFFLOAD_ENABLE
2169 				iaq->nofldrxq10g = min(n, nofldrxq10g);
2170 #endif
2171 			}
2172 
2173 			if (n1g > 0) {
2174 				int target = max(nrxq1g, nofldrxq1g);
2175 
2176 				n = 1;
2177 				while (n < target && leftover >= n1g) {
2178 					leftover -= n1g;
2179 					iaq->nirq += n1g;
2180 					n++;
2181 				}
2182 				iaq->nrxq1g = min(n, nrxq1g);
2183 #ifdef TCP_OFFLOAD_ENABLE
2184 				iaq->nofldrxq1g = min(n, nofldrxq1g);
2185 #endif
2186 			}
2187 
2188 			/* We have arrived at a minimum value required to enable
2189 			 * per queue irq(either NIC or offload). Thus for non-
2190 			 * offload case, we will get a vector per queue, while
2191 			 * offload case, we will get a vector per offload/NIC q.
2192 			 * Hence enable Interrupt forwarding only for offload
2193 			 * case.
2194 			 */
2195 #ifdef TCP_OFFLOAD_ENABLE
2196 			if (itype != DDI_INTR_TYPE_MSI || ISP2(iaq->nirq)) {
2197 				iaq->intr_fwd = 1;
2198 #else
2199 			if (itype != DDI_INTR_TYPE_MSI) {
2200 #endif
2201 				goto allocate;
2202 			}
2203 		}
2204 
2205 		/*
2206 		 * Least desirable option: one interrupt vector for everything.
2207 		 */
2208 		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2209 #ifdef TCP_OFFLOAD_ENABLE
2210 		iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2211 #endif
2212 		iaq->intr_fwd = 1;
2213 
2214 allocate:
2215 		return (0);
2216 	}
2217 
2218 	cxgb_printf(sc->dip, CE_WARN,
2219 	    "failed to find a usable interrupt type.  supported=%d, allowed=%d",
2220 	    itypes, p->intr_types);
2221 	return (DDI_FAILURE);
2222 }
2223 
2224 static int
2225 add_child_node(struct adapter *sc, int idx)
2226 {
2227 	int rc;
2228 	struct port_info *pi;
2229 
2230 	if (idx < 0 || idx >= sc->params.nports)
2231 		return (EINVAL);
2232 
2233 	pi = sc->port[idx];
2234 	if (pi == NULL)
2235 		return (ENODEV);	/* t4_port_init failed earlier */
2236 
2237 	PORT_LOCK(pi);
2238 	if (pi->dip != NULL) {
2239 		rc = 0;		/* EEXIST really, but then bus_config fails */
2240 		goto done;
2241 	}
2242 
2243 	rc = ndi_devi_alloc(sc->dip, T4_PORT_NAME, DEVI_SID_NODEID, &pi->dip);
2244 	if (rc != DDI_SUCCESS || pi->dip == NULL) {
2245 		rc = ENOMEM;
2246 		goto done;
2247 	}
2248 
2249 	(void) ddi_set_parent_data(pi->dip, pi);
2250 	(void) ndi_devi_bind_driver(pi->dip, 0);
2251 	rc = 0;
2252 done:
2253 	PORT_UNLOCK(pi);
2254 	return (rc);
2255 }
2256 
2257 static int
2258 remove_child_node(struct adapter *sc, int idx)
2259 {
2260 	int rc;
2261 	struct port_info *pi;
2262 
2263 	if (idx < 0 || idx >= sc->params.nports)
2264 		return (EINVAL);
2265 
2266 	pi = sc->port[idx];
2267 	if (pi == NULL)
2268 		return (ENODEV);
2269 
2270 	PORT_LOCK(pi);
2271 	if (pi->dip == NULL) {
2272 		rc = ENODEV;
2273 		goto done;
2274 	}
2275 
2276 	rc = ndi_devi_free(pi->dip);
2277 	if (rc == 0)
2278 		pi->dip = NULL;
2279 done:
2280 	PORT_UNLOCK(pi);
2281 	return (rc);
2282 }
2283 
2284 #define	KS_UINIT(x)	kstat_named_init(&kstatp->x, #x, KSTAT_DATA_ULONG)
2285 #define	KS_CINIT(x)	kstat_named_init(&kstatp->x, #x, KSTAT_DATA_CHAR)
2286 #define	KS_U_SET(x, y)	kstatp->x.value.ul = (y)
2287 #define	KS_C_SET(x, ...)	\
2288 			(void) snprintf(kstatp->x.value.c, 16,  __VA_ARGS__)
2289 
2290 /*
2291  * t4nex:X:config
2292  */
2293 struct t4_kstats {
2294 	kstat_named_t chip_ver;
2295 	kstat_named_t fw_vers;
2296 	kstat_named_t tp_vers;
2297 	kstat_named_t driver_version;
2298 	kstat_named_t serial_number;
2299 	kstat_named_t ec_level;
2300 	kstat_named_t id;
2301 	kstat_named_t bus_type;
2302 	kstat_named_t bus_width;
2303 	kstat_named_t bus_speed;
2304 	kstat_named_t core_clock;
2305 	kstat_named_t port_cnt;
2306 	kstat_named_t port_type;
2307 	kstat_named_t pci_vendor_id;
2308 	kstat_named_t pci_device_id;
2309 };
2310 static kstat_t *
2311 setup_kstats(struct adapter *sc)
2312 {
2313 	kstat_t *ksp;
2314 	struct t4_kstats *kstatp;
2315 	int ndata;
2316 	struct pci_params *p = &sc->params.pci;
2317 	struct vpd_params *v = &sc->params.vpd;
2318 	uint16_t pci_vendor, pci_device;
2319 
2320 	ndata = sizeof (struct t4_kstats) / sizeof (kstat_named_t);
2321 
2322 	ksp = kstat_create(T4_NEXUS_NAME, ddi_get_instance(sc->dip), "config",
2323 	    "nexus", KSTAT_TYPE_NAMED, ndata, 0);
2324 	if (ksp == NULL) {
2325 		cxgb_printf(sc->dip, CE_WARN, "failed to initialize kstats.");
2326 		return (NULL);
2327 	}
2328 
2329 	kstatp = (struct t4_kstats *)ksp->ks_data;
2330 
2331 	KS_UINIT(chip_ver);
2332 	KS_CINIT(fw_vers);
2333 	KS_CINIT(tp_vers);
2334 	KS_CINIT(driver_version);
2335 	KS_CINIT(serial_number);
2336 	KS_CINIT(ec_level);
2337 	KS_CINIT(id);
2338 	KS_CINIT(bus_type);
2339 	KS_CINIT(bus_width);
2340 	KS_CINIT(bus_speed);
2341 	KS_UINIT(core_clock);
2342 	KS_UINIT(port_cnt);
2343 	KS_CINIT(port_type);
2344 	KS_CINIT(pci_vendor_id);
2345 	KS_CINIT(pci_device_id);
2346 
2347 	KS_U_SET(chip_ver, sc->params.chip);
2348 	KS_C_SET(fw_vers, "%d.%d.%d.%d",
2349 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
2350 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
2351 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
2352 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
2353 	KS_C_SET(tp_vers, "%d.%d.%d.%d",
2354 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
2355 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
2356 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
2357 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
2358 	KS_C_SET(driver_version, DRV_VERSION);
2359 	KS_C_SET(serial_number, "%s", v->sn);
2360 	KS_C_SET(ec_level, "%s", v->ec);
2361 	KS_C_SET(id, "%s", v->id);
2362 	KS_C_SET(bus_type, "pci-express");
2363 	KS_C_SET(bus_width, "x%d lanes", p->width);
2364 	KS_C_SET(bus_speed, "%d", p->speed);
2365 	KS_U_SET(core_clock, v->cclk);
2366 	KS_U_SET(port_cnt, sc->params.nports);
2367 
2368 	t4_os_pci_read_cfg2(sc, PCI_CONF_VENID, &pci_vendor);
2369 	KS_C_SET(pci_vendor_id, "0x%x", pci_vendor);
2370 
2371 	t4_os_pci_read_cfg2(sc, PCI_CONF_DEVID, &pci_device);
2372 	KS_C_SET(pci_device_id, "0x%x", pci_device);
2373 
2374 	KS_C_SET(port_type, "%s/%s/%s/%s",
2375 		 print_port_speed(sc->port[0]),
2376 		 print_port_speed(sc->port[1]),
2377 		 print_port_speed(sc->port[2]),
2378 		 print_port_speed(sc->port[3]));
2379 
2380 	/* Do NOT set ksp->ks_update.  These kstats do not change. */
2381 
2382 	/* Install the kstat */
2383 	ksp->ks_private = (void *)sc;
2384 	kstat_install(ksp);
2385 
2386 	return (ksp);
2387 }
2388 
2389 /*
2390  * t4nex:X:stat
2391  */
2392 struct t4_wc_kstats {
2393 	kstat_named_t write_coal_success;
2394 	kstat_named_t write_coal_failure;
2395 };
2396 static kstat_t *
2397 setup_wc_kstats(struct adapter *sc)
2398 {
2399 	kstat_t *ksp;
2400 	struct t4_wc_kstats *kstatp;
2401 	int ndata;
2402 
2403 	ndata = sizeof(struct t4_wc_kstats) / sizeof(kstat_named_t);
2404 	ksp = kstat_create(T4_NEXUS_NAME, ddi_get_instance(sc->dip), "stats",
2405 	    "nexus", KSTAT_TYPE_NAMED, ndata, 0);
2406 	if (ksp == NULL) {
2407 		cxgb_printf(sc->dip, CE_WARN, "failed to initialize kstats.");
2408 		return (NULL);
2409 	}
2410 
2411 	kstatp = (struct t4_wc_kstats *)ksp->ks_data;
2412 
2413 	KS_UINIT(write_coal_success);
2414 	KS_UINIT(write_coal_failure);
2415 
2416 	ksp->ks_update = update_wc_kstats;
2417 	/* Install the kstat */
2418 	ksp->ks_private = (void *)sc;
2419 	kstat_install(ksp);
2420 
2421 	return (ksp);
2422 }
2423 
2424 static int
2425 update_wc_kstats(kstat_t *ksp, int rw)
2426 {
2427 	struct t4_wc_kstats *kstatp = (struct t4_wc_kstats *)ksp->ks_data;
2428 	struct adapter *sc = ksp->ks_private;
2429 	uint32_t wc_total, wc_success, wc_failure;
2430 
2431 	if (rw == KSTAT_WRITE)
2432 		return (0);
2433 
2434 	if (is_t5(sc->params.chip)) {
2435 		wc_total = t4_read_reg(sc, A_SGE_STAT_TOTAL);
2436 		wc_failure = t4_read_reg(sc, A_SGE_STAT_MATCH);
2437 		wc_success = wc_total - wc_failure;
2438 	} else {
2439 		wc_success = 0;
2440 		wc_failure = 0;
2441 	}
2442 
2443 	KS_U_SET(write_coal_success, wc_success);
2444 	KS_U_SET(write_coal_failure, wc_failure);
2445 
2446 	return (0);
2447 }
2448 
2449 int
2450 adapter_full_init(struct adapter *sc)
2451 {
2452 	int i, rc = 0;
2453 
2454 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2455 
2456 	rc = t4_setup_adapter_queues(sc);
2457 	if (rc != 0)
2458 		goto done;
2459 
2460 	if (sc->intr_cap & DDI_INTR_FLAG_BLOCK)
2461 		(void) ddi_intr_block_enable(sc->intr_handle, sc->intr_count);
2462 	else {
2463 		for (i = 0; i < sc->intr_count; i++)
2464 			(void) ddi_intr_enable(sc->intr_handle[i]);
2465 	}
2466 	t4_intr_enable(sc);
2467 	sc->flags |= FULL_INIT_DONE;
2468 
2469 #ifdef TCP_OFFLOAD_ENABLE
2470 	/* TODO: wrong place to enable TOE capability */
2471 	if (is_offload(sc) != 0) {
2472 		for_each_port(sc, i) {
2473 			struct port_info *pi = sc->port[i];
2474 			rc = toe_capability(pi, 1);
2475 			if (rc != 0) {
2476 				cxgb_printf(pi->dip, CE_WARN,
2477 				    "Failed to activate toe capability: %d",
2478 				    rc);
2479 				rc = 0;		/* not a fatal error */
2480 			}
2481 		}
2482 	}
2483 #endif
2484 
2485 done:
2486 	if (rc != 0)
2487 		(void) adapter_full_uninit(sc);
2488 
2489 	return (rc);
2490 }
2491 
2492 int
2493 adapter_full_uninit(struct adapter *sc)
2494 {
2495 	int i, rc = 0;
2496 
2497 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2498 
2499 	if (sc->intr_cap & DDI_INTR_FLAG_BLOCK)
2500 		(void) ddi_intr_block_disable(sc->intr_handle, sc->intr_count);
2501 	else {
2502 		for (i = 0; i < sc->intr_count; i++)
2503 			(void) ddi_intr_disable(sc->intr_handle[i]);
2504 	}
2505 
2506 	rc = t4_teardown_adapter_queues(sc);
2507 	if (rc != 0)
2508 		return (rc);
2509 
2510 	sc->flags &= ~FULL_INIT_DONE;
2511 
2512 	return (0);
2513 }
2514 
2515 int
2516 port_full_init(struct port_info *pi)
2517 {
2518 	struct adapter *sc = pi->adapter;
2519 	uint16_t *rss;
2520 	struct sge_rxq *rxq;
2521 	int rc, i;
2522 
2523 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
2524 	ASSERT((pi->flags & PORT_INIT_DONE) == 0);
2525 
2526 	/*
2527 	 * Allocate tx/rx/fl queues for this port.
2528 	 */
2529 	rc = t4_setup_port_queues(pi);
2530 	if (rc != 0)
2531 		goto done;	/* error message displayed already */
2532 
2533 	/*
2534 	 * Setup RSS for this port.
2535 	 */
2536 	rss = kmem_zalloc(pi->nrxq * sizeof (*rss), KM_SLEEP);
2537 	for_each_rxq(pi, i, rxq) {
2538 		rss[i] = rxq->iq.abs_id;
2539 	}
2540 	rc = -t4_config_rss_range(sc, sc->mbox, pi->viid, 0,
2541 	    pi->rss_size, rss, pi->nrxq);
2542 	kmem_free(rss, pi->nrxq * sizeof (*rss));
2543 	if (rc != 0) {
2544 		cxgb_printf(pi->dip, CE_WARN, "rss_config failed: %d", rc);
2545 		goto done;
2546 	}
2547 
2548 	pi->flags |= PORT_INIT_DONE;
2549 done:
2550 	if (rc != 0)
2551 		(void) port_full_uninit(pi);
2552 
2553 	return (rc);
2554 }
2555 
2556 /*
2557  * Idempotent.
2558  */
2559 int
2560 port_full_uninit(struct port_info *pi)
2561 {
2562 
2563 	ASSERT(pi->flags & PORT_INIT_DONE);
2564 
2565 	(void) t4_teardown_port_queues(pi);
2566 	pi->flags &= ~PORT_INIT_DONE;
2567 
2568 	return (0);
2569 }
2570 
2571 void
2572 enable_port_queues(struct port_info *pi)
2573 {
2574 	struct adapter *sc = pi->adapter;
2575 	int i;
2576 	struct sge_iq *iq;
2577 	struct sge_rxq *rxq;
2578 #ifdef TCP_OFFLOAD_ENABLE
2579 	struct sge_ofld_rxq *ofld_rxq;
2580 #endif
2581 
2582 	ASSERT(pi->flags & PORT_INIT_DONE);
2583 
2584 	/*
2585 	 * TODO: whatever was queued up after we set iq->state to IQS_DISABLED
2586 	 * back in disable_port_queues will be processed now, after an unbounded
2587 	 * delay.  This can't be good.
2588 	 */
2589 
2590 #ifdef TCP_OFFLOAD_ENABLE
2591 	for_each_ofld_rxq(pi, i, ofld_rxq) {
2592 		iq = &ofld_rxq->iq;
2593 		if (atomic_cas_uint(&iq->state, IQS_DISABLED, IQS_IDLE) !=
2594 		    IQS_DISABLED)
2595 			panic("%s: iq %p wasn't disabled", __func__,
2596 			    (void *)iq);
2597 		t4_write_reg(sc, MYPF_REG(A_SGE_PF_GTS),
2598 		    V_SEINTARM(iq->intr_params) | V_INGRESSQID(iq->cntxt_id));
2599 	}
2600 #endif
2601 
2602 	for_each_rxq(pi, i, rxq) {
2603 		iq = &rxq->iq;
2604 		if (atomic_cas_uint(&iq->state, IQS_DISABLED, IQS_IDLE) !=
2605 		    IQS_DISABLED)
2606 			panic("%s: iq %p wasn't disabled", __func__,
2607 			    (void *) iq);
2608 		t4_write_reg(sc, MYPF_REG(A_SGE_PF_GTS),
2609 		    V_SEINTARM(iq->intr_params) | V_INGRESSQID(iq->cntxt_id));
2610 	}
2611 }
2612 
2613 void
2614 disable_port_queues(struct port_info *pi)
2615 {
2616 	int i;
2617 	struct adapter *sc = pi->adapter;
2618 	struct sge_rxq *rxq;
2619 #ifdef TCP_OFFLOAD_ENABLE
2620 	struct sge_ofld_rxq *ofld_rxq;
2621 #endif
2622 
2623 	ASSERT(pi->flags & PORT_INIT_DONE);
2624 
2625 	/*
2626 	 * TODO: need proper implementation for all tx queues (ctrl, eth, ofld).
2627 	 */
2628 
2629 #ifdef TCP_OFFLOAD_ENABLE
2630 	for_each_ofld_rxq(pi, i, ofld_rxq) {
2631 		while (atomic_cas_uint(&ofld_rxq->iq.state, IQS_IDLE,
2632 		    IQS_DISABLED) != IQS_IDLE)
2633 			msleep(1);
2634 	}
2635 #endif
2636 
2637 	for_each_rxq(pi, i, rxq) {
2638 		while (atomic_cas_uint(&rxq->iq.state, IQS_IDLE,
2639 		    IQS_DISABLED) != IQS_IDLE)
2640 			msleep(1);
2641 	}
2642 
2643 	mutex_enter(&sc->sfl_lock);
2644 #ifdef TCP_OFFLOAD_ENABLE
2645 	for_each_ofld_rxq(pi, i, ofld_rxq)
2646 	    ofld_rxq->fl.flags |= FL_DOOMED;
2647 #endif
2648 	for_each_rxq(pi, i, rxq)
2649 	    rxq->fl.flags |= FL_DOOMED;
2650 	mutex_exit(&sc->sfl_lock);
2651 	/* TODO: need to wait for all fl's to be removed from sc->sfl */
2652 }
2653 
2654 void
2655 t4_fatal_err(struct adapter *sc)
2656 {
2657 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2658 	t4_intr_disable(sc);
2659 	cxgb_printf(sc->dip, CE_WARN,
2660 	    "encountered fatal error, adapter stopped.");
2661 }
2662 
2663 int
2664 t4_os_find_pci_capability(struct adapter *sc, int cap)
2665 {
2666 	uint16_t stat;
2667 	uint8_t cap_ptr, cap_id;
2668 
2669 	t4_os_pci_read_cfg2(sc, PCI_CONF_STAT, &stat);
2670 	if ((stat & PCI_STAT_CAP) == 0)
2671 		return (0); /* does not implement capabilities */
2672 
2673 	t4_os_pci_read_cfg1(sc, PCI_CONF_CAP_PTR, &cap_ptr);
2674 	while (cap_ptr) {
2675 		t4_os_pci_read_cfg1(sc, cap_ptr + PCI_CAP_ID, &cap_id);
2676 		if (cap_id == cap)
2677 			return (cap_ptr); /* found */
2678 		t4_os_pci_read_cfg1(sc, cap_ptr + PCI_CAP_NEXT_PTR, &cap_ptr);
2679 	}
2680 
2681 	return (0); /* not found */
2682 }
2683 
2684 void
2685 t4_os_portmod_changed(const struct adapter *sc, int idx)
2686 {
2687 	static const char *mod_str[] = {
2688 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
2689 	};
2690 	const struct port_info *pi = sc->port[idx];
2691 
2692 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
2693 		cxgb_printf(pi->dip, CE_NOTE, "transceiver unplugged.");
2694 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
2695 		cxgb_printf(pi->dip, CE_NOTE,
2696 		    "unknown transceiver inserted.\n");
2697 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
2698 		cxgb_printf(pi->dip, CE_NOTE,
2699 		    "unsupported transceiver inserted.\n");
2700 	else if (pi->mod_type > 0 && pi->mod_type < ARRAY_SIZE(mod_str))
2701 		cxgb_printf(pi->dip, CE_NOTE, "%s transceiver inserted.\n",
2702 		    mod_str[pi->mod_type]);
2703 	else
2704 		cxgb_printf(pi->dip, CE_NOTE, "transceiver (type %d) inserted.",
2705 		    pi->mod_type);
2706 }
2707 
2708 /* ARGSUSED */
2709 static int
2710 cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss, mblk_t *m)
2711 {
2712 	if (m != NULL)
2713 		freemsg(m);
2714 	return (0);
2715 }
2716 
2717 int
2718 t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
2719 {
2720 	uint_t *loc, new;
2721 
2722 	if (opcode >= ARRAY_SIZE(sc->cpl_handler))
2723 		return (EINVAL);
2724 
2725 	new = (uint_t)(unsigned long) (h ? h : cpl_not_handled);
2726 	loc = (uint_t *)&sc->cpl_handler[opcode];
2727 	(void) atomic_swap_uint(loc, new);
2728 
2729 	return (0);
2730 }
2731 
2732 static int
2733 fw_msg_not_handled(struct adapter *sc, const __be64 *data)
2734 {
2735 	struct cpl_fw6_msg *cpl;
2736 
2737 	cpl = __containerof((void *)data, struct cpl_fw6_msg, data);
2738 
2739 	cxgb_printf(sc->dip, CE_WARN, "%s fw_msg type %d", __func__, cpl->type);
2740 	return (0);
2741 }
2742 
2743 int
2744 t4_register_fw_msg_handler(struct adapter *sc, int type, fw_msg_handler_t h)
2745 {
2746 	fw_msg_handler_t *loc, new;
2747 
2748 	if (type >= ARRAY_SIZE(sc->fw_msg_handler))
2749 		return (EINVAL);
2750 
2751 	/*
2752 	 * These are dispatched by the handler for FW{4|6}_CPL_MSG using the CPL
2753 	 * handler dispatch table.  Reject any attempt to install a handler for
2754 	 * this subtype.
2755 	 */
2756 	if (type == FW_TYPE_RSSCPL || type == FW6_TYPE_RSSCPL)
2757 		return (EINVAL);
2758 
2759 	new = h ? h : fw_msg_not_handled;
2760 	loc = &sc->fw_msg_handler[type];
2761 	(void)atomic_swap_ptr(loc, (void *)new);
2762 
2763 	return (0);
2764 }
2765 
2766 #ifdef TCP_OFFLOAD_ENABLE
2767 static int
2768 toe_capability(struct port_info *pi, int enable)
2769 {
2770 	int rc;
2771 	struct adapter *sc = pi->adapter;
2772 
2773 	if (!is_offload(sc))
2774 		return (ENODEV);
2775 
2776 	if (enable != 0) {
2777 		if (isset(&sc->offload_map, pi->port_id) != 0)
2778 			return (0);
2779 
2780 		if (sc->offload_map == 0) {
2781 			rc = activate_uld(sc, ULD_TOM, &sc->tom);
2782 			if (rc != 0)
2783 				return (rc);
2784 		}
2785 
2786 		setbit(&sc->offload_map, pi->port_id);
2787 	} else {
2788 		if (!isset(&sc->offload_map, pi->port_id))
2789 			return (0);
2790 
2791 		clrbit(&sc->offload_map, pi->port_id);
2792 
2793 		if (sc->offload_map == 0) {
2794 			rc = deactivate_uld(&sc->tom);
2795 			if (rc != 0) {
2796 				setbit(&sc->offload_map, pi->port_id);
2797 				return (rc);
2798 			}
2799 		}
2800 	}
2801 
2802 	return (0);
2803 }
2804 
2805 /*
2806  * Add an upper layer driver to the global list.
2807  */
2808 int
2809 t4_register_uld(struct uld_info *ui)
2810 {
2811 	int rc = 0;
2812 	struct uld_info *u;
2813 
2814 	mutex_enter(&t4_uld_list_lock);
2815 	SLIST_FOREACH(u, &t4_uld_list, link) {
2816 		if (u->uld_id == ui->uld_id) {
2817 			rc = EEXIST;
2818 			goto done;
2819 		}
2820 	}
2821 
2822 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
2823 	ui->refcount = 0;
2824 done:
2825 	mutex_exit(&t4_uld_list_lock);
2826 	return (rc);
2827 }
2828 
2829 int
2830 t4_unregister_uld(struct uld_info *ui)
2831 {
2832 	int rc = EINVAL;
2833 	struct uld_info *u;
2834 
2835 	mutex_enter(&t4_uld_list_lock);
2836 
2837 	SLIST_FOREACH(u, &t4_uld_list, link) {
2838 		if (u == ui) {
2839 			if (ui->refcount > 0) {
2840 				rc = EBUSY;
2841 				goto done;
2842 			}
2843 
2844 			SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
2845 			rc = 0;
2846 			goto done;
2847 		}
2848 	}
2849 done:
2850 	mutex_exit(&t4_uld_list_lock);
2851 	return (rc);
2852 }
2853 
2854 static int
2855 activate_uld(struct adapter *sc, int id, struct uld_softc *usc)
2856 {
2857 	int rc = EAGAIN;
2858 	struct uld_info *ui;
2859 
2860 	mutex_enter(&t4_uld_list_lock);
2861 
2862 	SLIST_FOREACH(ui, &t4_uld_list, link) {
2863 		if (ui->uld_id == id) {
2864 			rc = ui->attach(sc, &usc->softc);
2865 			if (rc == 0) {
2866 				ASSERT(usc->softc != NULL);
2867 				ui->refcount++;
2868 				usc->uld = ui;
2869 			}
2870 			goto done;
2871 		}
2872 	}
2873 done:
2874 	mutex_exit(&t4_uld_list_lock);
2875 
2876 	return (rc);
2877 }
2878 
2879 static int
2880 deactivate_uld(struct uld_softc *usc)
2881 {
2882 	int rc;
2883 
2884 	mutex_enter(&t4_uld_list_lock);
2885 
2886 	if (usc->uld == NULL || usc->softc == NULL) {
2887 		rc = EINVAL;
2888 		goto done;
2889 	}
2890 
2891 	rc = usc->uld->detach(usc->softc);
2892 	if (rc == 0) {
2893 		ASSERT(usc->uld->refcount > 0);
2894 		usc->uld->refcount--;
2895 		usc->uld = NULL;
2896 		usc->softc = NULL;
2897 	}
2898 done:
2899 	mutex_exit(&t4_uld_list_lock);
2900 
2901 	return (rc);
2902 }
2903 
2904 void
2905 t4_iterate(void (*func)(int, void *), void *arg)
2906 {
2907 	struct adapter *sc;
2908 
2909 	mutex_enter(&t4_adapter_list_lock);
2910 	SLIST_FOREACH(sc, &t4_adapter_list, link) {
2911 		/*
2912 		 * func should not make any assumptions about what state sc is
2913 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
2914 		 */
2915 		func(ddi_get_instance(sc->dip), arg);
2916 	}
2917 	mutex_exit(&t4_adapter_list_lock);
2918 }
2919 
2920 #endif
2921