xref: /illumos-gate/usr/src/uts/common/io/ib/ibtl/ibtl_impl.c (revision b1d7ec75953cd517f5b7c3d9cb427ff8ec5d7d07)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 /*
26  * ibtl_impl.c
27  *
28  * This file contains the IBTF module's initialization and
29  * IBTF Clients/Modules registration routines.
30  */
31 
32 #include <sys/modctl.h>
33 #include <sys/sunndi.h>
34 #include <sys/ib/ibtl/impl/ibtl.h>
35 #include <sys/ib/ibtl/impl/ibtl_ibnex.h>
36 
37 /*
38  * Globals.
39  */
40 static char ibtf[] = "ibtl_impl";
41 
42 extern ibtl_ibnex_callback_t	ibtl_ibnex_callback_routine;
43 
44 /*
45  * ibtl_clnt_list:
46  *
47  *	Head of the list of IBT Client Instances. The IBT Client List
48  *	is modified by IBTF on an IBT client's ibt_attach/ibt_detach call.
49  *
50  * ibtl_hca_list:
51  *
52  *	Head of the list of HCA devices. The HCA List is modified by IBTF on
53  *	a CI's ibc_attach/ibc_detach call.
54  *	The datap of the list elements points to an ibtl_hca_devinfo_s
55  *	structure.
56  *
57  *				(ibc_attach)
58  *  ibtl_hca_list	-> ibtl_hca_devinfo_t--> ...	-->ibtl_hca_devinfo_t
59  *	[per-hca_dev]		|	^			{nth HCA Dev}
60  *				|	|
61  *				|  ibtl_hca_t (ibt_open_hca)
62  *				|	^  |
63  *				|	|  |
64  *				v	|  V
65  *  ibtl_clnt_list	->	ibtl_clnt_t--> ...--> {n'th Module}
66  *	[per-client_instance]	(ibt_attach)
67  *
68  */
69 
70 /* Global List of IBT Client Instances, and associated mutex. */
71 struct ibtl_clnt_s *ibtl_clnt_list = NULL;
72 kmutex_t ibtl_clnt_list_mutex;
73 
74 /* Lock for the race between the client and CM to free QPs. */
75 kmutex_t ibtl_free_qp_mutex;
76 
77 /* Lock for the race between the client closing the HCA and QPN being freed. */
78 kcondvar_t ibtl_close_hca_cv;
79 
80 /* Global List of HCA Devices, and associated mutex. */
81 struct ibtl_hca_devinfo_s *ibtl_hca_list = NULL;
82 
83 /* Well-known async handlers and associated client private. */
84 ibt_async_handler_t ibtl_cm_async_handler;
85 ibt_async_handler_t ibtl_dm_async_handler;
86 ibt_async_handler_t ibtl_ibma_async_handler;
87 void	*ibtl_cm_clnt_private;
88 void	*ibtl_dm_clnt_private;
89 void	*ibtl_ibma_clnt_private;
90 
91 extern int ib_hw_status;
92 _NOTE(SCHEME_PROTECTS_DATA("Scheme protects data", ib_hw_status))
93 
94 /*
95  * Misc Module Declarations.
96  */
97 extern struct mod_ops mod_miscops;
98 static struct modlmisc modlmisc = {
99 	&mod_miscops,			/* Type of module - misc. */
100 	"IB Transport Layer"		/* Name of the Module. */
101 };
102 
103 static struct modlinkage modlinkage = {
104 	MODREV_1, (void *)&modlmisc, NULL
105 };
106 
107 static void ibtl_kstat_init(ibtl_hca_devinfo_t *);
108 static void ibtl_kstat_fini(ibtl_hca_devinfo_t *);
109 static void ibtl_kstat_stats_create(ibtl_hca_devinfo_t *, uint_t);
110 static void ibtl_kstat_pkeys_create(ibtl_hca_devinfo_t *, uint_t);
111 
112 extern kmutex_t ibtl_part_attr_mutex;
113 
114 /*
115  * IBTF Loadable Module Routines.
116  */
117 
118 int
119 _init(void)
120 {
121 	int rval;
122 
123 	if ((rval = mod_install(&modlinkage)) != 0)
124 		return (rval);
125 
126 	/*
127 	 * initialize IBTL ib2usec table
128 	 */
129 	ibtl_ib2usec_init();
130 
131 	/*
132 	 * Initialize Logging
133 	 */
134 	ibtl_logging_initialization();
135 
136 	/*
137 	 * Initialize the Alloc QP States.
138 	 */
139 	ibtl_init_cep_states();
140 
141 	/*
142 	 * Initialize all Global Link Lists.
143 	 */
144 	mutex_init(&ibtl_clnt_list_mutex, NULL, MUTEX_DEFAULT, NULL);
145 	mutex_init(&ibtl_free_qp_mutex, NULL, MUTEX_DEFAULT, NULL);
146 	cv_init(&ibtl_close_hca_cv, NULL, CV_DEFAULT, NULL);
147 
148 	mutex_init(&ibtl_qp_mutex, NULL, MUTEX_DEFAULT, NULL);
149 	cv_init(&ibtl_qp_cv, NULL, CV_DEFAULT, NULL);
150 
151 	mutex_init(&ibtl_part_attr_mutex, NULL, MUTEX_DEFAULT, NULL);
152 
153 	ibtl_thread_init();
154 
155 	return (rval);
156 }
157 
158 
159 /*
160  * The IBTF Module is never unloaded. Actually there is no need of this
161  * routine, but provided just in case.
162  */
163 int
164 _fini(void)
165 {
166 	int rval;
167 
168 	if ((rval = mod_remove(&modlinkage)) != 0) {
169 		return (rval);
170 	}
171 
172 	ibtl_thread_fini();
173 
174 	mutex_destroy(&ibtl_clnt_list_mutex);
175 	mutex_destroy(&ibtl_free_qp_mutex);
176 	cv_destroy(&ibtl_close_hca_cv);
177 	mutex_destroy(&ibtl_qp_mutex);
178 	cv_destroy(&ibtl_qp_cv);
179 	mutex_destroy(&ibtl_part_attr_mutex);
180 
181 	/*
182 	 * Stop Logging
183 	 */
184 	ibtl_logging_destroy();
185 
186 	return (rval);
187 }
188 
189 
190 int
191 _info(struct modinfo *modinfop)
192 {
193 	/* Return the Module Information. */
194 	return (mod_info(&modlinkage, modinfop));
195 }
196 
197 
198 /*
199  * IBTF Client Registration Routines.
200  */
201 
202 /*
203  * Function:
204  *	ibt_attach
205  * Input:
206  *	modinfop	- Client Module info structure.
207  *	arg		- usually client's dip
208  *	clnt_private	- client's private data pointer.
209  * Output:
210  *	ibt_hdl_p	- pointer to client's specific IBT handle,
211  *			 which is opaque to clients.
212  * Returns:
213  *	IBT_SUCCESS
214  *	IBT_INVALID_PARAM
215  * Called by:
216  *	IBTF Client module during its attach() to register its instance
217  *	to IBTF.
218  * Description:
219  *	Registers the IBTF client module instance and returns an opaque
220  *	handler to the client to be used for future calls to IBTF.
221  *	Adds this client module instance to ibtl_clnt_list list.
222  *	Records well-known async handlers.
223  */
224 ibt_status_t
225 ibt_attach(ibt_clnt_modinfo_t *mod_infop, dev_info_t *arg, void *clnt_private,
226     ibt_clnt_hdl_t *ibt_hdl_p)
227 {
228 	dev_info_t	*pdip;
229 	ibtl_clnt_t	*clntp;
230 
231 	IBTF_DPRINTF_L3(ibtf, "ibt_attach(%p, %p, %p)",
232 	    mod_infop, arg, clnt_private);
233 
234 	if (mod_infop->mi_clnt_name == NULL) {
235 		IBTF_DPRINTF_L1(ibtf, "ibt_attach: "
236 		    "IB client needs to specify its name");
237 		return (IBT_INVALID_PARAM);
238 	}
239 
240 	/*
241 	 * Validate the Transport API version.
242 	 */
243 	if (mod_infop->mi_ibt_version != IBTI_V_CURR) {
244 		IBTF_DPRINTF_L1(ibtf, "ibt_attach: IB client '%s' has an "
245 		    "invalid IB TI Version '%d'", mod_infop->mi_clnt_name,
246 		    mod_infop->mi_ibt_version);
247 		return (IBT_NOT_SUPPORTED);
248 	}
249 
250 	if (mod_infop->mi_async_handler == NULL) {
251 		IBTF_DPRINTF_L2(ibtf, "ibt_attach: Client '%s' has not\n"
252 		    "        provided an Asynchronous Event Handler.\n"
253 		    "        This will be required soon.",
254 		    mod_infop->mi_clnt_name);
255 	}
256 
257 	/*
258 	 * Check out Client's Class information. If it is not of mgmt class,
259 	 * we expect 'arg' to be Not NULL and point to client driver's
260 	 * device info struct.
261 	 */
262 	if ((!IBT_MISCMOD_CLIENTS(mod_infop->mi_clnt_class)) &&
263 	    (arg == NULL)) {
264 		IBTF_DPRINTF_L1(ibtf, "ibt_attach: "
265 		    "arg not set with driver's dip.");
266 		return (IBT_INVALID_PARAM);
267 	}
268 
269 	if (!IBT_MISCMOD_CLIENTS(mod_infop->mi_clnt_class)) {
270 		pdip = ddi_get_parent(arg);
271 		if (pdip == NULL ||
272 		    ibtl_ibnex_valid_hca_parent(pdip) != IBT_SUCCESS) {
273 			IBTF_DPRINTF_L2(ibtf, "ibt_attach: "
274 			    "client %s is not a child of IB nexus driver.",
275 			    ddi_driver_name(arg));
276 			return (IBT_INVALID_PARAM);
277 		}
278 	}
279 
280 	mutex_enter(&ibtl_clnt_list_mutex);
281 	if (mod_infop->mi_clnt_class == IBT_CM) {
282 		if (ibtl_cm_async_handler != NULL) {
283 			IBTF_DPRINTF_L1(ibtf, "ibt_attach: "
284 			    "CM is already attached.");
285 			mutex_exit(&ibtl_clnt_list_mutex);
286 			return (IBT_INVALID_PARAM);
287 		}
288 		ibtl_cm_async_handler = mod_infop->mi_async_handler;
289 		ibtl_cm_clnt_private = clnt_private;
290 	} else if (mod_infop->mi_clnt_class == IBT_DM) {
291 		if (ibtl_dm_async_handler != NULL) {
292 			IBTF_DPRINTF_L1(ibtf, "ibt_attach: "
293 			    "DM is already attached.");
294 			mutex_exit(&ibtl_clnt_list_mutex);
295 			return (IBT_INVALID_PARAM);
296 		}
297 		ibtl_dm_async_handler = mod_infop->mi_async_handler;
298 		ibtl_dm_clnt_private = clnt_private;
299 	} else if (mod_infop->mi_clnt_class == IBT_IBMA) {
300 		if (ibtl_ibma_async_handler != NULL) {
301 			IBTF_DPRINTF_L1(ibtf, "ibt_attach: "
302 			    "IBMF is already attached.");
303 			mutex_exit(&ibtl_clnt_list_mutex);
304 			return (IBT_INVALID_PARAM);
305 		}
306 		ibtl_ibma_async_handler = mod_infop->mi_async_handler;
307 		ibtl_ibma_clnt_private = clnt_private;
308 	}
309 
310 	/* Allocate the memory for per-client-device info structure */
311 	clntp = kmem_zalloc(sizeof (ibtl_clnt_t), KM_SLEEP);
312 
313 	_NOTE(NOW_INVISIBLE_TO_OTHER_THREADS(clntp->clnt_modinfop,
314 	    clntp->clnt_dip, clntp->clnt_name, clntp->clnt_async_cnt,
315 	    clntp->clnt_private))
316 	/* Update the Client info structure */
317 	clntp->clnt_modinfop = mod_infop;	/* IBT Client's Mod Info */
318 	clntp->clnt_private = clnt_private;	/* IBT Client's private */
319 	clntp->clnt_dip = arg;			/* IBT Client's dip */
320 	clntp->clnt_async_cnt = 0;
321 	/* using a count of 7 below guarantees it is NULL terminated */
322 	(void) strncpy(clntp->clnt_name, mod_infop->mi_clnt_name, 7);
323 	_NOTE(NOW_VISIBLE_TO_OTHER_THREADS(clntp->clnt_modinfop,
324 	    clntp->clnt_dip, clntp->clnt_name, clntp->clnt_async_cnt,
325 	    clntp->clnt_private))
326 
327 	/*
328 	 * Update Client Device Instance List.
329 	 */
330 	clntp->clnt_list_link = ibtl_clnt_list;
331 	ibtl_clnt_list = clntp;
332 	mutex_exit(&ibtl_clnt_list_mutex);
333 
334 	/*
335 	 * The ibt_hdl_p is a opaque handle which is the address of
336 	 * ibt_clnt_t structure passed back to the clients.
337 	 * The client will pass on this handle in its future calls to IBTF.
338 	 */
339 	*ibt_hdl_p = clntp;
340 
341 	return (IBT_SUCCESS);
342 }
343 
344 
345 /*
346  * Function:
347  *	ibt_detach
348  * Input:
349  *	ibt_hdl - IBT Handle as returned during ibt_attach call.
350  * Output:
351  *	none
352  * Returns:
353  *	IBT_SUCCESS
354  *	IBT_INVALID_PARAM.
355  * Called by:
356  *	IBTF Client module during its detach() to de-register its instance
357  *	from IBTF.
358  * Description:
359  *	Deregisters the IBTF client module instance from the IBTF.
360  *	All resources and any reference to this ibt_hdl will be removed.
361  */
362 ibt_status_t
363 ibt_detach(ibt_clnt_hdl_t ibt_hdl)
364 {
365 	ibtl_clnt_t **clntpp;
366 
367 	IBTF_DPRINTF_L3(ibtf, "ibt_detach(%p)", ibt_hdl);
368 
369 	mutex_enter(&ibtl_clnt_list_mutex);
370 	clntpp = &ibtl_clnt_list;
371 	for (; *clntpp != NULL; clntpp = &(*clntpp)->clnt_list_link)
372 		if (*clntpp == ibt_hdl)
373 			break;
374 	if (*clntpp == NULL) {
375 		IBTF_DPRINTF_L1(ibtf, "ibt_detach: Client @ %p Not Found",
376 		    ibt_hdl);
377 		mutex_exit(&ibtl_clnt_list_mutex);
378 		return (IBT_INVALID_PARAM);
379 	}
380 
381 	/*
382 	 * Check out whether the client has freed all its resources.
383 	 * If not done, then fail the detach.
384 	 *
385 	 * viz. A client has to close all the HCA they have opened,
386 	 * i.e. the HCA List maintained for clients has to be empty.
387 	 * If this list is not empty, then the client has not performed
388 	 * complete clean-up, so fail the detach.
389 	 */
390 	if (ibt_hdl->clnt_hca_list != NULL) {
391 		mutex_exit(&ibtl_clnt_list_mutex);
392 
393 		IBTF_DPRINTF_L2(ibtf, "ibt_detach: "
394 		    "ERROR: Client '%s' has not closed all of its HCAs",
395 		    ibt_hdl->clnt_modinfop->mi_clnt_name);
396 		return (IBT_HCA_RESOURCES_NOT_FREED);
397 	}
398 
399 	if (ibt_hdl->clnt_srv_cnt != 0) {
400 		mutex_exit(&ibtl_clnt_list_mutex);
401 		IBTF_DPRINTF_L2(ibtf, "ibt_detach: client '%s' still has "
402 		    "services or subnet_notices registered",
403 		    ibt_hdl->clnt_modinfop->mi_clnt_name);
404 		return (IBT_HCA_RESOURCES_NOT_FREED);
405 	}
406 
407 	/*
408 	 * Delete the entry of this module from the ibtl_clnt_list List.
409 	 */
410 	*clntpp = ibt_hdl->clnt_list_link;	/* remove us */
411 
412 	/* make sure asyncs complete before freeing */
413 	ibtl_free_clnt_async_check(ibt_hdl);
414 
415 	if (ibt_hdl->clnt_modinfop->mi_clnt_class == IBT_CM) {
416 		ibtl_cm_async_handler = NULL;
417 		ibtl_cm_clnt_private = NULL;
418 	} else if (ibt_hdl->clnt_modinfop->mi_clnt_class == IBT_DM) {
419 		ibtl_dm_async_handler = NULL;
420 		ibtl_dm_clnt_private = NULL;
421 	} else if (ibt_hdl->clnt_modinfop->mi_clnt_class == IBT_IBMA) {
422 		ibtl_ibma_async_handler = NULL;
423 		ibtl_ibma_clnt_private = NULL;
424 	}
425 	mutex_exit(&ibtl_clnt_list_mutex);
426 
427 	/* Free up the memory of per-client info struct. */
428 	kmem_free(ibt_hdl, sizeof (ibtl_clnt_t));
429 
430 	return (IBT_SUCCESS);
431 }
432 
433 static void
434 ibtl_set_ibhw_status()
435 {
436 	ib_hw_status++;
437 }
438 
439 static void
440 ibtl_clear_ibhw_status()
441 {
442 	ib_hw_status--;
443 }
444 
445 /*
446  * Function:
447  *	ibc_init
448  * Input:
449  *	modlp		- Pointer to IBC client module linkage structure
450  * Output:
451  *	None
452  * Returns:
453  *	0 always for now
454  * Called by:
455  *	CI client calls IBTF during its _init() to register HCA with
456  *	Solaris I/O framework.
457  * Description:
458  *	Initializes the CI clients module linkage structure with
459  *	default bus_ops structure
460  */
461 int
462 ibc_init(struct modlinkage *modlp)
463 {
464 	ibtl_ibnex_cb_args_t	cb_args;
465 
466 	mutex_enter(&ibtl_clnt_list_mutex);
467 	cb_args.cb_flag = IBTL_IBNEX_IBC_INIT;
468 	cb_args.cb_modlp = modlp;
469 	if (ibtl_ibnex_callback_routine) {
470 		(void) ((*ibtl_ibnex_callback_routine)(&cb_args));
471 	}
472 	mutex_exit(&ibtl_clnt_list_mutex);
473 	return (0);
474 }
475 
476 
477 /*
478  * Function:
479  *	ibc_fini
480  * Input:
481  *	modlp		- Pointer to IBC client module linkage structure
482  * Output:
483  *	None
484  * Returns:
485  *	None
486  * Called by:
487  *	CI client calls IBTF during its _fini() to remove HCA with
488  *	Solaris I/O framework.
489  * Description:
490  *	Undo what is done during ibc_init
491  */
492 void
493 ibc_fini(struct modlinkage *modlp)
494 {
495 	ibtl_ibnex_cb_args_t	cb_args;
496 
497 	mutex_enter(&ibtl_clnt_list_mutex);
498 	cb_args.cb_flag = IBTL_IBNEX_IBC_FINI;
499 	cb_args.cb_modlp = modlp;
500 	if (ibtl_ibnex_callback_routine) {
501 		(void) ((*ibtl_ibnex_callback_routine)(&cb_args));
502 	}
503 	mutex_exit(&ibtl_clnt_list_mutex);
504 }
505 
506 /*
507  * Function:
508  *	ibc_attach
509  * Input:
510  *	info_p		- IBC HCA Info.
511  * Output:
512  *	ibc_hdl_p	- IBC Client's HCA Handle.
513  * Returns:
514  *	IBC_SUCCESS
515  *	IBC_FAILURE
516  * Called by:
517  *	CI calls IBTF during its attach() to register HCA Device with IBTF.
518  * Description:
519  *	Registers the presence of HCA device by providing the HCA device info
520  *  	structure and provides an opaque HCA handler for future calls to this
521  *  	HCA device.
522  */
523 ibc_status_t
524 ibc_attach(ibc_clnt_hdl_t *ibc_hdl_p, ibc_hca_info_t *info_p)
525 {
526 	ibtl_hca_devinfo_t	*hca_devp;
527 	uint_t			nports;
528 	ibt_status_t		status;
529 
530 	IBTF_DPRINTF_L2(ibtf, "ibc_attach(%p, %p)", ibc_hdl_p, info_p);
531 
532 	/* Validate the Transport API version */
533 	if (info_p->hca_ci_vers != IBCI_V3) {
534 		IBTF_DPRINTF_L1(ibtf, "ibc_attach: Invalid IB CI Version '%d'",
535 		    info_p->hca_ci_vers);
536 		return (IBC_FAILURE);
537 	}
538 
539 	if (info_p->hca_attr == NULL) {
540 		IBTF_DPRINTF_L1(ibtf, "ibc_attach: "
541 		    "HCA Attributes must be specified.");
542 		return (IBC_FAILURE);
543 	}
544 
545 	nports = info_p->hca_attr->hca_nports;
546 	if (nports == 0) {
547 		IBTF_DPRINTF_L1(ibtf, "ibc_attach: "
548 		    "Number of ports must be valid");
549 		return (IBC_FAILURE);
550 	}
551 
552 	if (info_p->hca_attr->hca_max_port_pkey_tbl_sz == 0) {
553 		IBTF_DPRINTF_L1(ibtf, "ibc_attach: "
554 		    "Number of Partitions must be at least 1");
555 		return (IBC_FAILURE);
556 	}
557 
558 	if ((info_p->hca_attr->hca_flags & IBT_HCA_CURRENT_QP_STATE) == 0) {
559 		IBTF_DPRINTF_L1(ibtf, "ibc_attach: "
560 		    "HCA driver must support QP current state checking");
561 		return (IBC_FAILURE);
562 	}
563 
564 	if ((info_p->hca_attr->hca_flags & IBT_HCA_PORT_UP) == 0) {
565 		IBTF_DPRINTF_L1(ibtf, "ibc_attach: "
566 		    "HCA driver must support PORT_UP async events");
567 		return (IBC_FAILURE);
568 	}
569 
570 	/*
571 	 * Install IB nexus driver (if not installed already)
572 	 */
573 	ibtl_set_ibhw_status();
574 	if (ndi_devi_config_vhci("ib", 0) == NULL) {
575 		IBTF_DPRINTF_L2(ibtf, "ibc_attach: IB nexus attach failed");
576 		ibtl_clear_ibhw_status();
577 		return (IBC_FAILURE);
578 	}
579 
580 	ibtl_thread_init2();
581 
582 	/* Allocate the memory for per-client info structure */
583 	hca_devp = kmem_zalloc(sizeof (ibtl_hca_devinfo_t) +
584 	    (nports - 1) * sizeof (ibtl_async_port_event_t), KM_SLEEP);
585 
586 	mutex_enter(&ibtl_clnt_list_mutex);
587 
588 	/* Update HCA dev info structure */
589 	hca_devp->hd_ibc_hca_hdl = info_p->hca_handle;
590 	hca_devp->hd_ibc_ops	= info_p->hca_ops;
591 	hca_devp->hd_hca_attr	= info_p->hca_attr;
592 	hca_devp->hd_hca_dip	= info_p->hca_dip;
593 
594 	status = ibtl_init_hca_portinfo(hca_devp);
595 	if (status != IBT_SUCCESS) {
596 		mutex_exit(&ibtl_clnt_list_mutex);
597 		IBTF_DPRINTF_L1(ibtf, "ibc_attach: call to ibc_query_hca_ports "
598 		    "failed: status = %d", status);
599 		kmem_free(hca_devp, sizeof (ibtl_hca_devinfo_t) +
600 		    (nports - 1) * sizeof (ibtl_async_port_event_t));
601 		return (IBC_FAILURE);
602 	}
603 
604 	/* Register the with MPxIO as PHCI */
605 	if (ibtl_ibnex_phci_register(info_p->hca_dip) != IBT_SUCCESS) {
606 		mutex_exit(&ibtl_clnt_list_mutex);
607 		IBTF_DPRINTF_L1(ibtf, "ibc_attach: MPxIO register failed");
608 		kmem_free(hca_devp, sizeof (ibtl_hca_devinfo_t) +
609 		    (nports - 1) * sizeof (ibtl_async_port_event_t));
610 		return (IBC_FAILURE);
611 	}
612 
613 	/* Initialize the Client List for this HCA. */
614 	hca_devp->hd_state	= IBTL_HCA_DEV_ATTACHED;
615 
616 	/* lock out asyncs until after we announce the new HCA */
617 	hca_devp->hd_async_busy = 1;
618 
619 	cv_init(&hca_devp->hd_async_task_cv, NULL, CV_DEFAULT, NULL);
620 	cv_init(&hca_devp->hd_async_busy_cv, NULL, CV_DEFAULT, NULL);
621 
622 	/* init portinfo locking variables */
623 	hca_devp->hd_portinfo_locked_port = 0;
624 	cv_init(&hca_devp->hd_portinfo_cv, NULL, CV_DEFAULT, NULL);
625 
626 	ibtl_kstat_init(hca_devp);
627 
628 	mutex_exit(&ibtl_clnt_list_mutex);
629 
630 	/*
631 	 * The ibc_hdl_p points to an opaque handle which is the address
632 	 * of ibt_hca_devinfo_t structure passed back to the CI.
633 	 * The CI will pass on this handle in its future upcalls to IBTF.
634 	 */
635 	*ibc_hdl_p = hca_devp;
636 
637 	return (IBC_SUCCESS);
638 }
639 
640 
641 /*
642  * Function:
643  *	ibc_post_attach
644  * Input:
645  *	ibc_hdl		- IBC Client's HCA Handle.
646  * Returns:
647  *	none
648  * Called by:
649  *	CI calls IBTF during its attach() after a successful ibc_attach().
650  * Description:
651  *	Announces to all known clients the existence of this HCA (by GUID).
652  */
653 void
654 ibc_post_attach(ibc_clnt_hdl_t ibc_hdl)
655 {
656 	IBTF_DPRINTF_L2(ibtf, "ibc_post_attach(%p)", ibc_hdl);
657 
658 	/*
659 	 * Update the HCA Device List.
660 	 */
661 	mutex_enter(&ibtl_clnt_list_mutex);
662 	ibc_hdl->hd_hca_dev_link = ibtl_hca_list;
663 	ibtl_hca_list = ibc_hdl;
664 	mutex_exit(&ibtl_clnt_list_mutex);
665 
666 	/* notify all IBT Client Device Instances of the new HCA Device */
667 	ibtl_announce_new_hca(ibc_hdl);
668 }
669 
670 
671 /*
672  * Function:
673  *	ibc_pre_detach
674  * Input:
675  *	ibc_clnt_hdl	- IBC HCA Handle as returned during ibc_attach call.
676  *  	cmd		- DDI_DETACH/DDI_SUSPEND command.
677  * Output:
678  *	none
679  * Returns:
680  *	IBC_SUCCESS
681  *	IBC_FAILURE.
682  * Called by:
683  *	CI to try to get all IBTF clients to close the HCA device.
684  * Description:
685  *	Attempts to deregister the HCA device entry from the IBTF.
686  *	If all resources are freed by the IBTF clients and this HCA
687  *	is closed, then IBC_SUCCESS is returned.
688  */
689 ibc_status_t
690 ibc_pre_detach(ibc_clnt_hdl_t hca_devp, ddi_detach_cmd_t cmd)
691 {
692 	ibtl_hca_devinfo_t **hcapp, *hcap;
693 
694 	IBTF_DPRINTF_L2(ibtf, "ibc_pre_detach(%p, 0x%x)", hca_devp, cmd);
695 
696 	/*
697 	 * Return failure, if command is not DDI_DETACH
698 	 */
699 	switch (cmd) {
700 	case DDI_DETACH:
701 		break;
702 	default:
703 		return (IBC_FAILURE); /* TBD: DDI_FAILURE */
704 	}
705 
706 	/* Make sure this HCA is on the HCA Device List.  */
707 	mutex_enter(&ibtl_clnt_list_mutex);
708 	hcap = ibtl_hca_list;
709 	while (hcap != NULL) {
710 		if (hcap == hca_devp)
711 			break;
712 		hcap = hcap->hd_hca_dev_link;
713 	}
714 	if (hcap == NULL) {
715 		mutex_exit(&ibtl_clnt_list_mutex);
716 		return (IBC_FAILURE);
717 	}
718 
719 	/*
720 	 * Initially set the state to "Detaching".
721 	 */
722 	hca_devp->hd_state = IBTL_HCA_DEV_DETACHING;
723 
724 	/*
725 	 * Try to detach all IBTI clients, and continue only if all
726 	 * of the detaches succeed.
727 	 */
728 	if (ibtl_detach_all_clients(hca_devp)) {
729 		hca_devp->hd_state = IBTL_HCA_DEV_ATTACHED; /* fix hd_state */
730 		mutex_exit(&ibtl_clnt_list_mutex);
731 
732 		return (IBC_FAILURE);
733 	}
734 
735 	/*
736 	 * Check to see if all clients closed this HCA, or not.
737 	 * We only succeed if all clients cooperated.
738 	 */
739 	if (hca_devp->hd_clnt_list != NULL) {
740 		hca_devp->hd_state = IBTL_HCA_DEV_ATTACHED;
741 		mutex_exit(&ibtl_clnt_list_mutex);
742 		IBTF_DPRINTF_L2(ibtf, "ibc_pre_detach: HCA still has attached "
743 		    "clients");
744 		return (IBC_FAILURE);
745 	}
746 
747 	/*
748 	 * mark this device as detached
749 	 */
750 	hca_devp->hd_state = IBTL_HCA_DEV_DETACHED;
751 
752 	/* Delete the entry for this hca_devp from hca_head_list */
753 	hcapp = &ibtl_hca_list;
754 	while (*hcapp != NULL) {
755 		if (*hcapp == hca_devp)
756 			break;
757 		hcapp = &(*hcapp)->hd_hca_dev_link;
758 	}
759 
760 	if (ibtl_ibnex_phci_unregister(hca_devp->hd_hca_dip) != IBT_SUCCESS) {
761 		hca_devp->hd_state = IBTL_HCA_DEV_ATTACHED; /* fix hd_state */
762 		mutex_exit(&ibtl_clnt_list_mutex);
763 		IBTF_DPRINTF_L1(ibtf, "ibc_pre_detach: PHCI unregister failed");
764 		return (IBC_FAILURE);
765 	}
766 
767 	if (*hcapp == NULL) {
768 		hca_devp->hd_state = IBTL_HCA_DEV_ATTACHED; /* fix hd_state */
769 		mutex_exit(&ibtl_clnt_list_mutex);
770 		IBTF_DPRINTF_L1(ibtf, "ibc_pre_detach: HCA not attached");
771 		return (IBC_FAILURE);
772 	}
773 	*hcapp = hca_devp->hd_hca_dev_link;
774 	ibtl_fast_gid_cache_valid = B_FALSE;	/* invalidate fast_gid_cache */
775 	mutex_exit(&ibtl_clnt_list_mutex);
776 
777 	return (IBC_SUCCESS);
778 }
779 
780 /*
781  * Function:
782  *	ibc_detach
783  * Input:
784  *	ibc_clnt_hdl	- IBC HCA Handle as returned during ibc_attach call.
785  * Output:
786  *	none
787  * Returns:
788  *	None
789  * Called by:
790  *	CI to detach the HCA device from IBTF.
791  * Description:
792  *	Do the second step of detaching the HCA, which is required
793  *	after a successful ibc_pre_detach.
794  */
795 void
796 ibc_detach(ibc_clnt_hdl_t hca_devp)
797 {
798 	IBTF_DPRINTF_L2(ibtf, "ibc_detach(%p)", hca_devp);
799 
800 	mutex_enter(&ibtl_clnt_list_mutex);
801 	if (hca_devp->hd_state != IBTL_HCA_DEV_DETACHED) {
802 		mutex_exit(&ibtl_clnt_list_mutex);
803 		IBTF_DPRINTF_L0(ibtf, "ibc_detach: HCA has not successfully "
804 		    "pre-detached");
805 		return;
806 	}
807 
808 	cv_destroy(&hca_devp->hd_async_task_cv);
809 	cv_destroy(&hca_devp->hd_async_busy_cv);
810 	cv_destroy(&hca_devp->hd_portinfo_cv);
811 
812 	kmem_free(hca_devp->hd_portinfop, hca_devp->hd_portinfo_len);
813 	mutex_exit(&ibtl_clnt_list_mutex);
814 
815 	ibtl_kstat_fini(hca_devp);
816 
817 	/* Free up the memory of per-client info struct */
818 	kmem_free(hca_devp, sizeof (ibtl_hca_devinfo_t) +
819 	    (hca_devp->hd_hca_attr->hca_nports - 1) *
820 	    sizeof (ibtl_async_port_event_t));
821 	ibtl_clear_ibhw_status();
822 }
823 
824 /*
825  * Function:
826  *	ibt_ci_data_in()
827  *
828  * Input:
829  *	hca_hdl			HCA Handle.
830  *	flags			IBT_COMPLETE_ALLOC - Finish a deferred alloc.
831  *      object                  Identifies the type object pointed to by
832  *                              ibt_object_handle.
833  *
834  *      ibt_object_handle       The handle of the object to be associated with
835  *				the data in/out
836  *
837  *	data_p			Pointer data passed in to the CI. The buffer
838  *				should be allocated by the caller.
839  *
840  *	data_sz			The size of the buffer pointed to by
841  *				data_p.
842  * Output:
843  *
844  * Returns:
845  *	IBT_SUCCESS
846  *	IBT_NOT_SUPPORTED	Feature not supported.
847  *	IBT_INVALID_PARAM	Invalid object type specified.
848  *	IBT_HCA_HDL_INVALID
849  *	IBT_AH_HDL_INVALID/IBT_UD_DEST_HDL_INVALID
850  *	IBT_CHAN_HDL_INVALID/IBT_QP_HDL_INVALID
851  *	IBT_CQ_HDL_INVALID
852  *	IBT_EEC_HDL_INVALID
853  *	IBT_RDD_HDL_INVALID
854  *	IBT_MW_HDL_INVALID
855  *	IBT_PD_HDL_INVALID
856  *	IBT_SRQ_HDL_INVALID
857  *
858  * Description:
859  *	Exchange CI private data for the specified CI object.
860  */
861 ibt_status_t
862 ibt_ci_data_in(ibt_hca_hdl_t hca, ibt_ci_data_flags_t flags,
863     ibt_object_type_t object, void *ibt_object_handle, void *data_p,
864     size_t data_sz)
865 {
866 	ibt_status_t		retval;
867 	void			*ci_obj_hdl;
868 
869 	IBTF_DPRINTF_L3(ibtf, "ibt_ci_data_in(%p, %x, %d, %p, %p, %d)",
870 	    hca, flags, object, ibt_object_handle, data_p, data_sz);
871 
872 	switch (object) {
873 	case IBT_HDL_HCA:
874 		ci_obj_hdl = (void *)
875 		    (IBTL_HCA2CIHCA(((ibt_hca_hdl_t)ibt_object_handle)));
876 		break;
877 
878 	case IBT_HDL_CHANNEL:
879 		ci_obj_hdl = (void *)
880 		    (IBTL_CHAN2CIQP(((ibt_channel_hdl_t)ibt_object_handle)));
881 		break;
882 
883 	case IBT_HDL_CQ:
884 		ci_obj_hdl = (void *)
885 		    (((ibt_cq_hdl_t)(ibt_object_handle))->cq_ibc_cq_hdl);
886 		break;
887 
888 	case IBT_HDL_EEC:
889 		ci_obj_hdl = (void *)
890 		    (((ibt_eec_hdl_t)(ibt_object_handle))->eec_ibc_eec_hdl);
891 		break;
892 
893 	case IBT_HDL_UD_DEST:
894 		ci_obj_hdl = (void *)
895 		    (((ibt_ud_dest_hdl_t)(ibt_object_handle))->ud_ah);
896 		break;
897 
898 	case IBT_HDL_SRQ:
899 		ci_obj_hdl = (void *)
900 		    (((ibt_srq_hdl_t)(ibt_object_handle))->srq_ibc_srq_hdl);
901 		break;
902 
903 	default:
904 		ci_obj_hdl = ibt_object_handle;
905 		break;
906 	}
907 
908 	retval = (IBTL_HCA2CIHCAOPS_P(hca)->ibc_ci_data_in)(IBTL_HCA2CIHCA(hca),
909 	    flags, object, ci_obj_hdl, data_p, data_sz);
910 
911 	if (retval != IBT_SUCCESS) {
912 		IBTF_DPRINTF_L2(ibtf, "ibt_ci_data_in: Failed : %d", retval);
913 	}
914 	return (retval);
915 }
916 
917 /*
918  * Function:
919  *	ibt_ci_data_out()
920  *
921  * Input:
922  *	hca_hdl			HCA Handle.
923  *	flags			IBT_COMPLETE_ALLOC - Finish a deferred alloc.
924  *      object                  Identifies the type object pointed to by
925  *                              ibt_object_handle.
926  *
927  *      ibt_object_handle       The handle of the object to be associated with
928  *				the data in/out
929  *
930  *	data_p			Pointer to a buffer in which to return the CI
931  *				private data. The buffer should be allocated
932  *				by the caller.
933  *
934  *	data_sz			The size of the buffer pointed to by
935  *				data_p.
936  * Output:
937  *
938  * Returns:
939  *	IBT_SUCCESS
940  *	IBT_NOT_SUPPORTED	Feature not supported.
941  *	IBT_INSUFF_RESOURCE	The buffer pointed to by data_p was too
942  *				small to hold the data.
943  *	IBT_INVALID_PARAM	Invalid object type specified.
944  *	IBT_HCA_HDL_INVALID
945  *	IBT_AH_HDL_INVALID/IBT_UD_DEST_HDL_INVALID
946  *	IBT_CHAN_HDL_INVALID/IBT_QP_HDL_INVALID
947  *	IBT_CQ_HDL_INVALID
948  *	IBT_EEC_HDL_INVALID
949  *	IBT_RDD_HDL_INVALID
950  *	IBT_MW_HDL_INVALID
951  *	IBT_PD_HDL_INVALID
952  *	IBT_SRQ_HDL_INVALID
953  *
954  * Description:
955  *	Exchange CI private data for the specified CI object.
956  */
957 ibt_status_t
958 ibt_ci_data_out(ibt_hca_hdl_t hca, ibt_ci_data_flags_t flags,
959     ibt_object_type_t object, void *ibt_object_handle, void *data_p,
960     size_t data_sz)
961 {
962 	ibt_status_t		retval;
963 	void			*ci_obj_hdl;
964 
965 	IBTF_DPRINTF_L3(ibtf, "ibt_ci_data_out(%p, %x, %d, %p, %p, %d)",
966 	    hca, flags, object, ibt_object_handle, data_p, data_sz);
967 
968 	switch (object) {
969 	case  IBT_HDL_HCA:
970 		ci_obj_hdl = (void *)
971 		    (IBTL_HCA2CIHCA(((ibt_hca_hdl_t)ibt_object_handle)));
972 		break;
973 
974 	case IBT_HDL_CHANNEL:
975 		ci_obj_hdl = (void *)
976 		    (IBTL_CHAN2CIQP(((ibt_channel_hdl_t)ibt_object_handle)));
977 		break;
978 
979 	case IBT_HDL_CQ:
980 		ci_obj_hdl = (void *)
981 		    (((ibt_cq_hdl_t)(ibt_object_handle))->cq_ibc_cq_hdl);
982 		break;
983 
984 	case IBT_HDL_EEC:
985 		ci_obj_hdl = (void *)
986 		    (((ibt_eec_hdl_t)(ibt_object_handle))->eec_ibc_eec_hdl);
987 		break;
988 
989 	case IBT_HDL_UD_DEST:
990 		ci_obj_hdl = (void *)
991 		    (((ibt_ud_dest_hdl_t)(ibt_object_handle))->ud_ah);
992 		break;
993 
994 	case IBT_HDL_SRQ:
995 		ci_obj_hdl = (void *)
996 		    (((ibt_srq_hdl_t)(ibt_object_handle))->srq_ibc_srq_hdl);
997 		break;
998 
999 	default:
1000 		ci_obj_hdl = ibt_object_handle;
1001 		break;
1002 	}
1003 
1004 	retval = (IBTL_HCA2CIHCAOPS_P(hca)->ibc_ci_data_out)
1005 	    (IBTL_HCA2CIHCA(hca), flags, object, ci_obj_hdl, data_p, data_sz);
1006 
1007 	if (retval != IBT_SUCCESS) {
1008 		IBTF_DPRINTF_L2(ibtf, "ibt_ci_data_out: Failed : %d", retval);
1009 	}
1010 	return (retval);
1011 }
1012 
1013 
1014 /*
1015  * FMA Support functions.
1016  */
1017 
1018 #define	IBTL_ENA_MASK		0xC0000000
1019 #define	IBTL_ENA_POSSIBLE	0x80000000
1020 #define	IBTL_TYPE_SHIFT		27
1021 
1022 /*
1023  * Function:
1024  *	ibt_get_module_failure()
1025  *
1026  * Input:
1027  *	type			Identifies the failing IB module.
1028  *	ena			'0' or the data for Fault Management
1029  *				Architecture (ENA).
1030  *
1031  * Returns:
1032  *	status			Special IB failure status.
1033  *
1034  * Description:
1035  *	XXX Just stubbed out to return failures with no data for Fault
1036  *	Management Architecture (ENAs) at the moment XXX
1037  */
1038 ibt_status_t
1039 ibt_get_module_failure(ibt_failure_type_t type, uint64_t ena)
1040 {
1041 	ibt_status_t	ret;
1042 
1043 	IBTF_DPRINTF_L3(ibtf, "ibt_get_module_failure(%d, 0x%llX)", type, ena);
1044 
1045 	switch (type) {
1046 	case IBT_FAILURE_CI:
1047 	case IBT_FAILURE_IBMF:
1048 	case IBT_FAILURE_IBCM:
1049 	case IBT_FAILURE_IBDM:
1050 	case IBT_FAILURE_IBTL:
1051 	case IBT_FAILURE_IBSM:
1052 		ret = IBTL_ENA_POSSIBLE | (type << IBTL_TYPE_SHIFT);
1053 		break;
1054 	default:
1055 		ret = IBT_FAILURE;
1056 	}
1057 	IBTF_DPRINTF_L3(ibtf, "ibt_get_module_failure: ret = 0x%lX", ret);
1058 	return (ret);
1059 }
1060 
1061 
1062 /*
1063  * Function:
1064  *	ibc_get_ci_failure()
1065  *
1066  * Input:
1067  *	ena			'0' or the data for Fault Management
1068  *				Architecture (ENA).
1069  *
1070  * Returns:
1071  *	status			Special CI failure status.
1072  *
1073  * Description:
1074  *	Just use the function above to do the job.
1075  */
1076 ibt_status_t
1077 ibc_get_ci_failure(uint64_t ena)
1078 {
1079 	return (ibt_get_module_failure(IBT_FAILURE_CI, ena));
1080 }
1081 
1082 
1083 /*
1084  * ibt_check_failure()
1085  *	Function to test for special case failures.
1086  *
1087  *	status		An ibt_status_t returned from an IBTF function call.
1088  *
1089  *	reserved_p	NULL, or a pointer to where we store the data for
1090  *			Fault Management Architecture (ENA).
1091  *
1092  * Description:
1093  *	XXX Still need to determine the data for Fault Management Architecture
1094  *	(ENA), using 0 for now XXX
1095  */
1096 ibt_failure_type_t
1097 ibt_check_failure(ibt_status_t status, uint64_t *reserved_p)
1098 {
1099 	ibt_failure_type_t type;
1100 
1101 	IBTF_DPRINTF_L3(ibtf, "ibt_check_failure(%X)", status);
1102 
1103 	if ((status & IBTL_ENA_MASK) == IBTL_ENA_POSSIBLE) {
1104 		type = status & ~IBTL_ENA_POSSIBLE >> IBTL_TYPE_SHIFT;
1105 
1106 		/* XXX Need more work here... */
1107 		if (reserved_p != NULL)
1108 			*reserved_p = 0;
1109 	} else {
1110 		type = IBT_FAILURE_STANDARD;
1111 		if (reserved_p != NULL)
1112 			*reserved_p = 0;	/* No FMA Data Available. */
1113 	}
1114 	IBTF_DPRINTF_L3(ibtf, "ibt_check_failure: type = 0x%X", type);
1115 	return (type);
1116 }
1117 
1118 /*
1119  * Initialize and create kstats.
1120  *
1121  * We create the following kstats on all ports of the HCA:
1122  *	<hca_driver_name><instance_number>/port<port_num>/stats
1123  *	<hca_driver_name><instance_number>/port<port_num>/pkeys
1124  */
1125 static void
1126 ibtl_kstat_init(ibtl_hca_devinfo_t *hca_devp)
1127 {
1128 	uint_t			nports = hca_devp->hd_hca_attr->hca_nports;
1129 	ibtl_hca_port_kstat_t	*pks;
1130 	int			i;
1131 
1132 	IBTF_DPRINTF_L3(ibtf, "ibtl_kstat_init(hca_devp = 0x%p)", hca_devp);
1133 
1134 	hca_devp->hd_hca_port_ks_info_len =
1135 	    sizeof (ibtl_hca_port_kstat_t) * nports;
1136 	pks = kmem_zalloc(hca_devp->hd_hca_port_ks_info_len, KM_SLEEP);
1137 	hca_devp->hd_hca_port_ks_info = pks;
1138 
1139 	for (i = 0; i < nports; i++, pks++) {
1140 		pks->pks_hca_devp = hca_devp;
1141 		pks->pks_port_num = i + 1;
1142 		ibtl_kstat_stats_create(hca_devp, i + 1);
1143 		ibtl_kstat_pkeys_create(hca_devp, i + 1);
1144 	}
1145 }
1146 
1147 /*
1148  * Delete kstats on all ports of the HCA.
1149  */
1150 static void
1151 ibtl_kstat_fini(ibtl_hca_devinfo_t *hca_devp)
1152 {
1153 	ibtl_hca_port_kstat_t	*pks;
1154 	int			i;
1155 
1156 	IBTF_DPRINTF_L3(ibtf, "ibtl_kstat_fini(hca_devp = 0x%p)", hca_devp);
1157 
1158 	_NOTE(NOW_INVISIBLE_TO_OTHER_THREADS(*hca_devp))
1159 
1160 	pks = hca_devp->hd_hca_port_ks_info;
1161 	_NOTE(NOW_INVISIBLE_TO_OTHER_THREADS(*pks))
1162 
1163 	if (pks == NULL)
1164 		return;
1165 
1166 	for (i = 0; i < hca_devp->hd_hca_attr->hca_nports; i++, pks++) {
1167 		if (pks->pks_stats_ksp)
1168 			kstat_delete(pks->pks_stats_ksp);
1169 
1170 		if (pks->pks_pkeys_ksp) {
1171 			ASSERT(!MUTEX_HELD(&ibtl_clnt_list_mutex));
1172 			kstat_delete(pks->pks_pkeys_ksp);
1173 		}
1174 	}
1175 
1176 	kmem_free(hca_devp->hd_hca_port_ks_info,
1177 	    hca_devp->hd_hca_port_ks_info_len);
1178 }
1179 
1180 /*
1181  * Update "stats" kstat.
1182  * Called by kstat framework.
1183  */
1184 static int
1185 ibtl_kstat_stats_update(kstat_t *ksp, int rw)
1186 {
1187 	ibtl_hca_port_kstat_t	*pks;
1188 	ibtl_hca_devinfo_t	*hca_devp;
1189 	ibt_hca_portinfo_t	*p;
1190 	struct kstat_named	*data;
1191 
1192 	IBTF_DPRINTF_L4(ibtf, "ibtl_kstat_stats_update(ksp = 0x%p, rw = %d)",
1193 	    ksp, rw);
1194 
1195 	if (rw == KSTAT_WRITE)
1196 		return (EACCES);
1197 
1198 	mutex_enter(&ibtl_clnt_list_mutex);
1199 
1200 	/*
1201 	 * Update the link_state kstat using the value from portinfo cache.
1202 	 */
1203 	pks = ksp->ks_private;
1204 	hca_devp = pks->pks_hca_devp;
1205 	data = (struct kstat_named *)(ksp->ks_data);
1206 	p = hca_devp->hd_portinfop + pks->pks_port_num - 1;
1207 	data[0].value.ui32 = (uint32_t)p->p_linkstate;
1208 
1209 	mutex_exit(&ibtl_clnt_list_mutex);
1210 
1211 	return (0);
1212 }
1213 
1214 /*
1215  * Create "stats" kstat for the specified HCA port in the form:
1216  *	<hca_driver_name><instance_number>/port<port_num>/stats
1217  *	At preset it contains only one named data of "link_state"
1218  */
1219 static void
1220 ibtl_kstat_stats_create(ibtl_hca_devinfo_t *hca_devp, uint_t port_num)
1221 {
1222 	struct kstat		*ksp;
1223 	struct kstat_named	*named_data;
1224 	char			*drv_name;
1225 	int			drv_instance;
1226 	ibtl_hca_port_kstat_t	*pks;
1227 	char			kname[40];
1228 
1229 	IBTF_DPRINTF_L3(ibtf, "ibtl_kstat_stats_create(hca_devp = 0x%p, "
1230 	    "port_num = 0x%u)", hca_devp, port_num);
1231 
1232 	drv_name = (char *)ddi_driver_name(hca_devp->hd_hca_dip);
1233 	drv_instance = ddi_get_instance(hca_devp->hd_hca_dip);
1234 	(void) snprintf(kname, sizeof (kname), "%s%d/port%d/stats",
1235 	    drv_name, drv_instance, port_num);
1236 
1237 	ksp = kstat_create("ibtf", 0, kname, "ib", KSTAT_TYPE_NAMED, 1, 0);
1238 	if (ksp == NULL) {
1239 		IBTF_DPRINTF_L2(ibtf,
1240 		    "ibtl_kstat_stats_create: kstat_create() failed");
1241 		return;
1242 	}
1243 
1244 	named_data = (struct kstat_named *)(ksp->ks_data);
1245 	kstat_named_init(&named_data[0], "link_state", KSTAT_DATA_UINT32);
1246 
1247 	pks = hca_devp->hd_hca_port_ks_info + port_num - 1;
1248 	pks->pks_stats_ksp = ksp;
1249 
1250 	ksp->ks_private = pks;
1251 	ksp->ks_update = ibtl_kstat_stats_update;
1252 
1253 	/* Install the kstat */
1254 	kstat_install(ksp);
1255 }
1256 
1257 /*
1258  * Update "pkeys" kstat.
1259  *
1260  * Called by kstat framework. Since ks_lock was set to ibtl_clnt_list_mutex
1261  * at the time of the kstat creation, kstat framework will hold this lock
1262  * while calling this function.
1263  */
1264 static int
1265 ibtl_kstat_pkeys_update(kstat_t *ksp, int rw)
1266 {
1267 	ibtl_hca_port_kstat_t	*pks;
1268 	ibtl_hca_devinfo_t	*hca_devp;
1269 	ibt_hca_portinfo_t	*p;
1270 
1271 	IBTF_DPRINTF_L4(ibtf, "ibtl_kstat_pkeys_update(ksp = 0x%p, rw = %d)",
1272 	    ksp, rw);
1273 
1274 #ifndef	__lock_lint
1275 	ASSERT(MUTEX_HELD(&ibtl_clnt_list_mutex));
1276 #endif
1277 
1278 	if (rw == KSTAT_WRITE)
1279 		return (EACCES);
1280 
1281 	_NOTE(NOW_INVISIBLE_TO_OTHER_THREADS(*ksp))
1282 
1283 	pks = ksp->ks_private;
1284 	_NOTE(NOW_INVISIBLE_TO_OTHER_THREADS(*pks))
1285 
1286 	hca_devp = pks->pks_hca_devp;
1287 	_NOTE(NOW_INVISIBLE_TO_OTHER_THREADS(*hca_devp))
1288 
1289 	/*
1290 	 * Point kstat data to the pkey table in the portinfo cache.
1291 	 */
1292 
1293 	p = hca_devp->hd_portinfop + pks->pks_port_num - 1;
1294 
1295 	ksp->ks_data = p->p_pkey_tbl;
1296 	ksp->ks_ndata = p->p_pkey_tbl_sz;
1297 	ksp->ks_data_size = p->p_pkey_tbl_sz * sizeof (ib_pkey_t);
1298 
1299 	return (0);
1300 }
1301 
1302 /*
1303  * Create "pkeys" kstat for the specified HCA port in the form:
1304  *	<hca_driver_name><instance_number>/port<port_num>/pkeys
1305  *
1306  * Currently kstat framework allows only some fixed data types as named
1307  * data components under a named kstat. Due to this limitation it is not
1308  * possible to add "pkeys" as a named data under the "stats" kstat.
1309  */
1310 static void
1311 ibtl_kstat_pkeys_create(ibtl_hca_devinfo_t *hca_devp, uint_t port_num)
1312 {
1313 	struct kstat		*ksp;
1314 	char			*drv_name;
1315 	int			drv_instance;
1316 	char			kname[40];
1317 	ibtl_hca_port_kstat_t	*pks;
1318 
1319 	IBTF_DPRINTF_L3(ibtf, "ibtl_kstat_stats_create(hca_devp = 0x%p, "
1320 	    "port_num = 0x%u)", hca_devp, port_num);
1321 
1322 	drv_name = (char *)ddi_driver_name(hca_devp->hd_hca_dip);
1323 	drv_instance = ddi_get_instance(hca_devp->hd_hca_dip);
1324 	(void) snprintf(kname, sizeof (kname), "%s%d/port%d/pkeys",
1325 	    drv_name, drv_instance, port_num);
1326 
1327 	ksp = kstat_create("ibtf", 0, kname, "ib", KSTAT_TYPE_RAW, 0,
1328 	    KSTAT_FLAG_VAR_SIZE | KSTAT_FLAG_VIRTUAL);
1329 	if (ksp == NULL) {
1330 		IBTF_DPRINTF_L2(ibtf,
1331 		    "ibtl_kstat_pkeys_create: kstat_create() failed");
1332 		return;
1333 	}
1334 
1335 	pks = hca_devp->hd_hca_port_ks_info + port_num - 1;
1336 	pks->pks_pkeys_ksp = ksp;
1337 
1338 	ksp->ks_private = pks;
1339 	ksp->ks_update = ibtl_kstat_pkeys_update;
1340 	ksp->ks_lock = &ibtl_clnt_list_mutex;
1341 
1342 	/*
1343 	 * We just go with the default_kstat_snapshot().
1344 	 * So there is no need to set ks_snapshot field.
1345 	 */
1346 
1347 	/* Install the kstat */
1348 	kstat_install(ksp);
1349 }
1350