xref: /illumos-gate/usr/src/uts/common/sys/ib/mgt/ibcm/ibcm_impl.h (revision 9d3d2ed09c8e9ba0b2ba44fdd1dd300b2c3f9e8e)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #ifndef _SYS_IB_MGT_IBCM_IBCM_IMPL_H
28 #define	_SYS_IB_MGT_IBCM_IBCM_IMPL_H
29 
30 #pragma ident	"%Z%%M%	%I%	%E% SMI"
31 
32 /*
33  * ibcm_impl.h
34  *
35  * This file contains all of the internal data structures and
36  * definitions for IBCM.
37  *
38  * The general state transition processing of CM is achieved by the
39  * following callgraph:
40  *
41  * CM INIT : Register for hca attach and detach callbacks, and other asyncs
42  *
43  * On new HCA attach:	Register with IBMF on all ports of upcoming HCA
44  *			Specify CM callback and callback "per HCA arg"
45  *			Register with SA, allocate AVL trees etc.
46  *
47  * IBMF Callback
48  *  	Validate combination of method and attribute Id in the generic MAD hdr
49  *	-> Call CM Connection state transition function based on attribute ID
50  *	    Create/lookup/delete CM state structure and save it into avl tree
51  *	    Handle duplicate messages and MRA to adjust timers etc.
52  *	    Handle stale connections
53  *	    Allocate reply MADs
54  *		-> Call CM QP/EEC state transition function based on CM message
55  *		     Change QP/EEC state  (to enable recvQ posting by client)
56  *		     Call Client/Server handler callback function
57  *		     Modify QP/EEC attributes
58  *		     Optionally fill up some fields of response MAD
59  *	    Post reply MADs
60  *	    Store reply MADs and reply MAD address, if necessary
61  *	    Initialize timeouts for the message
62  *	    Change CM state
63  *	    Deallocate reply MADs
64  *
65  * NOTES:
66  * 	o There are *NO* explicit CM allocation and deallocation routines for
67  *	CM MADs and state data structures
68  *	o CM timeouts are scheduled using timeout(9f), and cancelled using
69  *	untimeout(9f)
70  *	o svc_id allocation scheme
71  *	A new counter for svcid is maintained in ibcm_hca_info_t
72  *	which is used to allocate svcid. The svcids are incremented
73  *	sequentially and allocated (with wrap around on overflow) with
74  *	these considerations:
75  *		The WellKnown service id's and locally allocated svcid's
76  *		could be maintained in separate lists, thus allowing the
77  *		lists to be kept apart and sorted easily.
78  *		The insertions are done at the end of the list
79  *	o reqid allocation scheme
80  *	The list is a sorted one (as reqid's are allocated sequentially).
81  *	If there is a code required for wrap around, it would search for
82  *	a reqid from the head of the list.
83  *	The insertions are always done at the end of the lists
84  *	o XXX svc_id allocation scheme and req_id allocation scheme will
85  *	be revisited.
86  */
87 
88 #include <sys/sysmacros.h>
89 #include <sys/systm.h>
90 #include <sys/kmem.h>
91 #include <sys/modctl.h>
92 #include <sys/avl.h>
93 #include <sys/taskq.h>
94 #include <sys/vmem.h>
95 #include <sys/note.h>
96 #include <sys/t_lock.h>
97 
98 #include <sys/ib/ibtl/ibvti.h>
99 #include <sys/ib/ibtl/impl/ibtl_cm.h>
100 #include <sys/ib/ibtl/impl/ibtl_util.h>
101 #include <sys/ib/mgt/ibmf/ibmf.h>
102 #include <sys/ib/mgt/ibcm/ibcm_trace.h>
103 
104 #ifdef __cplusplus
105 extern "C" {
106 #endif
107 
108 _NOTE(SCHEME_PROTECTS_DATA("Private", sa_service_record_s))
109 _NOTE(SCHEME_PROTECTS_DATA("Exclusive access to ibmf msg buf based on state",
110 ib_mad_hdr_t))
111 _NOTE(SCHEME_PROTECTS_DATA("Exclusive access to ibmf msg buf based on state",
112 _ibmf_msg))
113 
114 /*
115  * Defines for all CM state machine states, as defined in
116  * section 12.9.7. IBCM_REJ_SENT is a state not defined in
117  * the spec and is added for implementation purposes.
118  */
119 typedef enum ibcm_conn_state_e {
120 	/* Initial states */
121 	IBCM_STATE_IDLE			= 0,
122 	IBCM_STATE_LISTEN,
123 
124 	/* States during connection establishment */
125 	IBCM_STATE_REQ_SENT,
126 	IBCM_STATE_REQ_RCVD,
127 	IBCM_STATE_REP_SENT,
128 	IBCM_STATE_REP_RCVD,
129 	IBCM_STATE_REP_WAIT,
130 	IBCM_STATE_MRA_SENT,
131 	IBCM_STATE_MRA_REP_SENT,
132 	IBCM_STATE_MRA_REP_RCVD,
133 
134 	/* States during connection establishment failures */
135 	IBCM_STATE_TIMED_OUT,
136 	IBCM_STATE_ABORTED,
137 	IBCM_STATE_REJ_SENT,
138 
139 	/* Established state */
140 	IBCM_STATE_TRANSIENT_ESTABLISHED,
141 	IBCM_STATE_ESTABLISHED,
142 
143 	/* States during connection teardown */
144 	IBCM_STATE_TRANSIENT_DREQ_SENT,
145 	IBCM_STATE_DREQ_SENT,
146 	IBCM_STATE_DREQ_RCVD,
147 	IBCM_STATE_DREP_RCVD,
148 	IBCM_STATE_TIMEWAIT,
149 
150 	/* states for UD side of things */
151 	IBCM_STATE_SIDR_REQ_SENT,
152 	IBCM_STATE_SIDR_REQ_RCVD,
153 	IBCM_STATE_SIDR_REP_SENT,
154 	IBCM_STATE_SIDR_REP_RCVD,
155 
156 	/* states common to RC and UD, during state resource deletion */
157 	IBCM_STATE_DELETE
158 } ibcm_conn_state_t;
159 
160 /* Defines the AP states for LAP/APR */
161 typedef enum ibcm_ap_state_e {
162 	IBCM_AP_STATE_IDLE	= 0x0,
163 	IBCM_AP_STATE_LAP_SENT,
164 	IBCM_AP_STATE_LAP_RCVD,
165 	IBCM_AP_STATE_APR_RCVD,
166 	IBCM_AP_STATE_MRA_LAP_RCVD,
167 	IBCM_AP_STATE_MRA_LAP_SENT,
168 	IBCM_AP_STATE_TIMED_OUT
169 } ibcm_ap_state_t;
170 
171 /*
172  * Defines for the CM event types/MAD attribute IDs
173  */
174 typedef enum ibcm_event_type_e {
175 	IBCM_INCOMING_REQ	= 0x0,
176 	IBCM_INCOMING_MRA	= 0x1,
177 	IBCM_INCOMING_REJ	= 0x2,
178 	IBCM_INCOMING_REP	= 0x3,
179 	IBCM_INCOMING_RTU	= 0x4,
180 	IBCM_INCOMING_DREQ	= 0x5,
181 	IBCM_INCOMING_DREP	= 0x6,
182 	IBCM_INCOMING_SIDR_REQ	= 0x7,
183 	IBCM_INCOMING_SIDR_REP	= 0x8,
184 	IBCM_INCOMING_LAP	= 0x9,
185 	IBCM_INCOMING_APR	= 0xA,
186 	IBCM_OUTGOING_REQ	= 0xB,	/* REQ Sent on active CM side */
187 	IBCM_INCOMING_REQ_STALE	= 0xC,	/* lookup by remote HCA and */
188 					/* remote comid */
189 	IBCM_INCOMING_REP_STALE	= 0xD,	/* lookup by passive HCA and QPN */
190 	IBCM_INCOMING_REJ_RCOMID = 0xE	/* lookup by remote com id */
191 } ibcm_event_type_t;
192 
193 /*
194  * IBMF calls back into CM on only the first 11 events defined in
195  * ibcm_event_type_t. CM has pre-defined functions for these 11 events
196  *
197  */
198 #define	IBCM_MAX_EVENTS		11
199 
200 /*
201  * CM message attribute IDs begin at this "base ID". The first 11 event types
202  * in ibcm_event_type_t are CM protocol messages that are posted to IBMF by
203  * adding the "base_id" to the respective event type value. By subtracting
204  * the "base_id" in IBMF callback in CM MAD, the message type is gotten back
205  */
206 #define	IBCM_ATTR_BASE_ID		0x10
207 
208 #define	IBCM_MAX_RETRY_CNT		15
209 #define	IBCM_ATTRID_FIELD_SIZE		4
210 #define	IBCM_TRANID_PRIV_FIELD_SIZE	28
211 
212 #define	IBCM_RNR_RETRY_CNT_MASK		0x7	/* 3 bits */
213 #define	IBCM_MAX_RNR_RETRY_CNT		7
214 
215 #define	IBCM_INITIAL_COMID		1
216 #define	IBCM_INITIAL_REQID		1
217 #define	IBCM_INITIAL_SID		1
218 
219 /*
220  * Maximum number of com ids / req ids that can be active at any given time
221  * MUST ENSURE THAT (INITIAL ID + MAX IDS -1), for any of the IDs does not
222  * exceed the max 32 bit
223  */
224 
225 /* An hca can have max of 2^24 -2  RC connections */
226 #define	IBCM_MAX_COMIDS		(0x01000000 - 2)
227 #define	IBCM_MAX_REQIDS		0xFFFFFFFF
228 #define	IBCM_MAX_LOCAL_SIDS	0xFFFFFFFF
229 
230 typedef uint32_t ib_com_id_t;	/* CM Communication ID */
231 
232 /*
233  * Defines the CM Mode of operation for a connection
234  */
235 typedef enum ibcm_mode_e {
236 	IBCM_ACTIVE_MODE	= 1,	/* Active side CM */
237 	IBCM_PASSIVE_MODE	= 2	/* Passive side CM */
238 } ibcm_mode_t;
239 
240 
241 /* different IBCM return values */
242 typedef enum ibcm_status_e {
243 	IBCM_SUCCESS  		= 0,	/* good status */
244 	IBCM_LOOKUP_EXISTS,		/* statep lookup found existing entry */
245 	IBCM_LOOKUP_NEW,		/* lookup created new statep entry */
246 	IBCM_LOOKUP_FAIL,		/* lookup found no statep entry */
247 	IBCM_SEND_REJ,			/* CM QP state change sent REJ msg */
248 	IBCM_SEND_REP,			/* CM QP state change sent REP msg */
249 	IBCM_SEND_RTU,			/* CM QP state change sent RTU msg */
250 	IBCM_SEND_APR,			/* CM to send APR MAD as response */
251 	IBCM_SEND_SIDR_REP, 		/* client's UD handler returned this */
252 	IBCM_DEFER,			/* client's handler returned this */
253 	IBCM_FAILURE			/* generic IBCM failure */
254 } ibcm_status_t;
255 
256 /*
257  * Struct definition for addressing information that CM maintains for
258  * each of the incoming MADs
259  */
260 typedef	struct	ibcm_mad_addr {
261 	ibmf_global_addr_info_t	grh_hdr;	/* GRH related fields of MAD */
262 	ibmf_addr_info_t	rcvd_addr;	/* Outgoing/Incoming MAD addr */
263 	ibmf_handle_t		ibmf_hdl;	/* IBMF handle */
264 	boolean_t		grh_exists;	/* TRUE if grh exists */
265 	uint8_t			port_num;
266 	struct ibcm_qp_list_s	*cm_qp_entry;	/* IBMF hdl on which MAD rcvd */
267 						/* or on which MAD shall be */
268 						/* sent out */
269 } ibcm_mad_addr_t;
270 
271 _NOTE(READ_ONLY_DATA(ibcm_mad_addr))
272 
273 #define	IBCM_MAD_SIZE		0x100			/* size of MAD */
274 #define	IBCM_MAD_HDR_SIZE	sizeof (ib_mad_hdr_t)	/* size of MAD HDR */
275 #define	IBCM_MSG_SIZE		IBCM_MAD_SIZE-IBCM_MAD_HDR_SIZE
276 
277 typedef enum ibcm_abort_flag_e {
278 	IBCM_ABORT_INIT		= 0,	/* no abort flag is set */
279 	IBCM_ABORT_CLIENT	= 1,	/* client requested connection abort */
280 	IBCM_ABORT_REJ		= 2	/* REJ received with timeout reason */
281 } ibcm_abort_flag_t;
282 
283 typedef	enum ibcm_isync_e {
284 	IBCM_BLOCK	= 0,	/* Block cm operation */
285 	IBCM_UNBLOCK	= 1,	/* Unblock cm operation */
286 	IBCM_FAIL	= 2	/* fail cm operation */
287 } ibcm_isync_t;
288 
289 /*
290  * Define a connection state structure, used by the IBTF CM
291  * to maintain state about connected QPs.
292  *
293  * mode			: CM connection mode active/passive
294  * state		: CM connection state
295  * ap_state		: CM AP Internal state to manage LAP/APR state machine
296  * state_mutex		: lock for this structure
297  * channel		: Channel associated with this RC state structure
298  * ref_cnt		: Number of active threads that may reference this
299  *			  state structure
300  * svcid		: Service ID
301  * cm_handler		: Client handler callback address
302  * stored_reply_addr	: Address for replying using the stored mad
303  * hcap			: A pointer to the HCA's entry
304  * stored_msg		: Stores the response REP/REJ/RTU MAD
305  * mra_msg		: Stores the response MRA MAD
306  * dreq_msg		: Stores the DREQ MAD
307  * drep_msg		: Stores the DREP MAD
308  * lapr_msg		: Stores the LAP/APR MAD
309  *			  detect duplicate LAP messages
310  * local_comid  	: Local communication id
311  * local_hca_guid	: Local HCA GUID
312  * local_qpn		: Local QPN
313  *
314  * remote_comid 	: Remote communication id
315  * remote_hca_guid	: Remote HCA GUID
316  * remote_qpn		: Remote QPN
317  *
318  * timerid		: Timer id for the timeout either for re-sending the
319  *			  stored mad or deleting the stored mad
320  *			  Ex: A REJ/RTU response for an incoming REP
321  *			      A REP response to an incoming REQ
322  *			      An outgoing REQ on active connection side
323  * timer_value		: Time for any of the above timers in HZ
324  * pkt_life_time	: pkt life time from source to destination
325  * remote_ack_delay	: Remote hca's ack delay in clock_t
326  * rc_alt_pkt_lt	: Life time for new ALT path specified in LAP
327  * stale_clock		: clock used to detect stale vs duplicate REQs
328  * timer_stored_state	: state of connection for timeout() validation
329  * timer_stored_ap_state: CM ap_state for timeout validation
330  * remaining_retry_count: Remaining count for retries ie., posting stored MADs
331  * max_cm_retries	: Max retry count for sending a REQ/REP/DREQ
332  * delete_mra_msg	: Set to TRUE for deletion, if MRA re-send in progress
333  * resend_mad		: B_TRUE, if REQ/REP/RTU/REJ MAD re-send is in progress
334  * resend_mra_mad	: B_TRUE, if a MRA mad re-sens is in progress
335  * cep_retry_cnt	: Retry count for CEP.
336  * stale		: B_TRUE, if connection has become stale
337  * blocking_done	: B_TRUE, if cv_signal been issued to block_client_cv
338  * clnt_hdl		: Clnt_hdl passed in ibt_open_channel
339  * return_data		: RC return args, valid for blocking
340  *			  ibt_open_channel
341  * drep_priv_data;	: The pointer to client specified outgoing private
342  *			  data, from close channel API call
343  * drep_priv_data_len   : The length of DREP private data that client would
344  *			  like to be returned from close channel API call
345  * delete_state_data	: B_TRUE, if CM decides to delete state data, but
346  *			  there is some thread that could access state data
347  *
348  * avl_active_link	: For inserting this state-data into active AVL tree
349  * avl_passive_link	: For inserting this state-data into passive AVL tree
350  * Note : All timer values that are of type "clock_t" below are in usecs
351  */
352 typedef struct ibcm_state_data_s {
353 	/* for AVL tree */
354 	avl_node_t		avl_active_link;
355 	avl_node_t		avl_passive_link;
356 	avl_node_t		avl_passive_comid_link;
357 
358 	/* remote stuff */
359 	ib_guid_t		remote_hca_guid;
360 	ib_com_id_t		remote_comid;
361 	ib_qpn_t		remote_qpn;
362 
363 	/* local stuff */
364 	ib_com_id_t		local_comid;
365 	ib_qpn_t		local_qpn;
366 	ib_guid_t		local_hca_guid;
367 
368 	ibcm_mode_t		mode;
369 	ibcm_conn_state_t	state;
370 	ibcm_ap_state_t		ap_state;
371 	kmutex_t		state_mutex;
372 	ibt_channel_hdl_t	channel;	/* save a copy */
373 
374 	/* ref_cnt so others cannot delete a statep that may be referenced */
375 	int			ref_cnt;
376 
377 	ib_svc_id_t		svcid;
378 	ibt_cm_handler_t	cm_handler;
379 
380 	ibcm_mad_addr_t		stored_reply_addr;
381 
382 	struct ibcm_hca_info_s *hcap;
383 
384 	ibmf_msg_t		*stored_msg;
385 	ibmf_msg_t		*mra_msg;
386 	ibmf_msg_t		*dreq_msg;
387 	ibmf_msg_t		*drep_msg;
388 	ibmf_msg_t		*lapr_msg;
389 
390 	void			*defer_cm_msg;
391 
392 	/* timeout related stuff */
393 	timeout_id_t		timerid;
394 	clock_t			timer_value;
395 	clock_t			pkt_life_time;
396 	clock_t			remote_ack_delay;
397 	clock_t			rc_alt_pkt_lt;
398 
399 	hrtime_t		stale_clock;
400 	hrtime_t		post_time;
401 	hrtime_t		mra_time;
402 
403 	ibcm_conn_state_t	timer_stored_state;
404 	ibcm_ap_state_t		timer_stored_ap_state;
405 	uint8_t			remaining_retry_cnt;
406 	uint8_t			max_cm_retries;
407 	uint8_t			cm_retries;
408 
409 	uint8_t			drep_in_progress;
410 
411 	/* some cep stuff, stored here temporarily during connection est  */
412 	uint8_t			cep_retry_cnt:3;
413 	ibt_srate_t		local_srate;
414 	ibt_srate_t		local_alt_srate;
415 	ib_pkey_t		pkey;
416 	uint8_t			prim_port;
417 	uint8_t			alt_port;
418 	uint32_t		starting_psn;
419 	ib_path_bits_t		prim_src_path_bits;
420 	ib_path_bits_t		alt_src_path_bits;
421 
422 	boolean_t		delete_mra_msg;
423 	boolean_t		stale;
424 	boolean_t		delete_state_data;
425 
426 	boolean_t		open_done;
427 	boolean_t		close_done;
428 	boolean_t		ap_done;
429 
430 	uint8_t			send_mad_flags;
431 	uint8_t			close_flow;
432 	uint8_t			open_flow;
433 	ibcm_abort_flag_t	abort_flag;
434 
435 	struct ibcm_state_data_s	*timeout_next;
436 
437 	ibcm_conn_state_t	timedout_state;
438 
439 	ibcm_isync_t		cep_in_rts;
440 	ibcm_isync_t		clnt_proceed;
441 	ibcm_isync_t		close_nocb_state;
442 
443 	/* Clients' information */
444 	void			*state_cm_private;
445 
446 	/* pointer to service info */
447 	struct ibcm_svc_info_s  *state_svc_infop;
448 
449 	kcondvar_t		block_client_cv;
450 	kcondvar_t		block_mad_cv;
451 
452 	/* Data for recycle function */
453 	struct ibcm_taskq_recycle_arg_s	*recycle_arg;
454 
455 	/* Return data pointers in various cm api calls */
456 	ibt_rc_returns_t	*open_return_data;
457 	ibt_ap_returns_t	*ap_return_data;
458 	uint8_t			*close_priv_data;
459 	ibt_priv_data_len_t	*close_priv_data_len;
460 	uint8_t			*close_ret_status;
461 
462 	/* for queuing of open_rc_channel requests */
463 	struct ibcm_state_data_s	*open_link;
464 
465 	struct ibcm_conn_trace_s	*conn_trace;
466 
467 } ibcm_state_data_t;
468 
469 _NOTE(MUTEX_PROTECTS_DATA(ibcm_state_data_s::state_mutex,
470     ibcm_state_data_s::{state ref_cnt timer_stored_state timer_value
471     timer_stored_ap_state remaining_retry_cnt clnt_proceed cep_in_rts
472     close_nocb_state block_client_cv block_mad_cv timedout_state cm_handler
473     abort_flag mra_msg}))
474 
475 _NOTE(READ_ONLY_DATA(ibcm_state_data_s::{mode channel svcid hcap
476     local_comid local_hca_guid local_qpn remote_comid remote_hca_guid
477     remote_qpn pkt_life_time remote_ack_delay rc_alt_pkt_lt stored_reply_addr
478     max_cm_retries cep_retry_cnt local_srate local_alt_srate pkey
479     prim_port alt_port starting_psn state_svc_infop avl_active_link
480     avl_passive_link avl_passive_comid_link defer_cm_msg recycle_arg
481     conn_trace}))
482 
483 _NOTE(SCHEME_PROTECTS_DATA("Serailized access by block_client_cv",
484     ibcm_state_data_s::{open_return_data ap_return_data close_priv_data
485     close_priv_data_len close_ret_status}))
486 
487 _NOTE(DATA_READABLE_WITHOUT_LOCK(ibcm_state_data_s::{timedout_state
488     cm_handler mra_msg abort_flag}))
489 
490 /*
491  * Definitions for send mad flags. Respective bits in send_mad_flags or
492  * ud_send_mad_flags are set to 1, during MAD transmission, and reset in
493  * ibmf send completion callback or on completion of a blocking ibmf mad post.
494  */
495 #define	IBCM_REP_POST_BUSY	1	/* REP post in progress */
496 #define	IBCM_REJ_POST_BUSY	2	/* REJ post in progress */
497 #define	IBCM_RTU_POST_BUSY	4	/* RTU post in progress */
498 #define	IBCM_MRA_POST_BUSY	8	/* MRA post in progress */
499 #define	IBCM_DREP_POST_BUSY	16	/* DREQ post in progress */
500 #define	IBCM_SREP_POST_BUSY	32	/* SIDR REP post in progress */
501 
502 /* MADs that are retransmitted only because of a timeout */
503 #define	IBCM_REQ_POST_BUSY	64	/* REQ post in progress */
504 
505 
506 /* Incr/Decr ref_cnt by 1 */
507 #define	IBCM_REF_CNT_INCR(s)	(s->ref_cnt++)
508 #define	IBCM_REF_CNT_DECR(s)	\
509 	if ((--(s->ref_cnt) == 0) && (s->delete_state_data == B_TRUE)) { \
510 		ibcm_add_tlist(s);\
511 	} \
512 	ASSERT(s->ref_cnt >= 0);
513 
514 /*
515  * This macro checks if ch_qp/ch_eec handles are both not set for a channel
516  */
517 #define	IBCM_INVALID_CHANNEL(chan)	(chan == NULL)
518 
519 /*
520  * The next macros are used to get/set the statep from the QP
521  * handles, using the CM private data. These call into IBTL.
522  * The WAIT and RELEASE macros deal with related issues that
523  * require use of the same lock within IBTL.
524  */
525 #define	IBCM_GET_CHAN_PRIVATE(ch, s) \
526 	if ((ch) != NULL) { \
527 		s = ibtl_cm_get_chan_private(ch); \
528 	} else \
529 		s = NULL;
530 
531 #define	IBCM_SET_CHAN_PRIVATE(ch, s) \
532 	if ((ch) != NULL) { \
533 		ibtl_cm_set_chan_private(ch, (void *)(s)); \
534 	}
535 
536 #define	IBCM_RELEASE_CHAN_PRIVATE(ch) \
537 	if ((ch) != NULL) { \
538 		ibtl_cm_release_chan_private(ch); \
539 	}
540 
541 #define	IBCM_WAIT_CHAN_PRIVATE(ch) \
542 	ibtl_cm_wait_chan_private(ch);
543 
544 /* In future, if we intend to change it to realtime_timeout, it's easy */
545 #define	IBCM_TIMEOUT(arg1, arg2)	timeout(ibcm_timeout_cb, arg1,\
546 						drv_usectohz(arg2))
547 #define	IBCM_UD_TIMEOUT(arg1, arg2)	timeout(ibcm_sidr_timeout_cb, arg1,\
548 						drv_usectohz(arg2))
549 
550 extern void ibcm_open_enqueue(ibcm_state_data_t *statep);
551 extern void ibcm_open_done(ibcm_state_data_t *statep);
552 extern void ibcm_close_done(ibcm_state_data_t *statep, int send_done);
553 extern void ibcm_close_enter(void);
554 extern void ibcm_close_exit(void);
555 extern void ibcm_lapr_enter(void);
556 extern void ibcm_lapr_exit(void);
557 extern void ibcm_check_for_opens(void);
558 extern void ibcm_run_tlist_thread(void);
559 
560 /*
561  * Structures & defines for SIDR
562  */
563 
564 /*
565  * Define a connection state structure, used for SIDR REQ and REP
566  * (ibcm_ud_state_data_t - struct for SIDR connection)
567  *
568  * ud_state: 		CM connection state (See ibcm_conn_state_t)
569  * ud_req_id:		Request ID
570  * ud_svcid:		Service ID
571  * ud_state_mutex:	CM connection state
572  *
573  * ud_max_cm_retries:	Max retry count for sending a SIDR REQ
574  * ud_ref_cnt:		State ref count for not deleting accidentally
575  * ud_remaining_retry_count: Remaining count for retries ie., posting
576  *			stored MADs
577  * ud_cm_handler:	Server's handler callback address
578  *
579  * ud_nextp:		CM link for IBTF list
580  * ud_hcap:		A pointer to the HCA's entry
581  *
582  * ud_timerid:		Timer id for the timeout either for re-sending the
583  *			stored mad or deleting the stored mad
584  *			Ex: A SIDR REP response for an incoming SIDR REQ
585  *			An outgoing SIDR REQ on active connection side
586  * ud_timer_value:	Time for any of the above timers in HZ
587  * ud_pkt_life_time:	pkt life time from source to destination
588  * ud_stored_reply_addr: Address for replying using the stored mad
589  *
590  * ud_sidr_req_lid:	SIDR REQ sender's port LID
591  * ud_sidr_req_gid:	SIDR REQ sender's port GID
592  * ud_grh_exists:	TRUE if GRH exists in the incoming SIDR REQ
593  *
594  * ud_passive_qpn:	QPN allocated by server for a SIDR REQ
595  * ud_passive_qpn_qkey:	QPN's QKEY allocated by server
596  *
597  * ud_block_client_cv:	CV condition variable on which ibt_ud_get_dqpn() waits,
598  *			if called in blocking mode.
599  * ud_return_data:	UD return args, valid for blocking ibt_ud_get_dqpn
600  * ud_timer_stored_state: State stored for timeout handling
601  * ud_blocking_done	: Tells if cv_wait is needed or not. To handle the
602  *			  case where a cv_signal is received prior to its
603  *			  cv_wait().
604  * Note : All timer values that are of type "clock_t" below are in usec
605  */
606 typedef struct ibcm_ud_state_data_s {
607 	kmutex_t		ud_state_mutex;
608 	ibcm_conn_state_t	ud_state;
609 	ibcm_mode_t		ud_mode;
610 
611 	int			ud_ref_cnt;
612 
613 	uint32_t		ud_req_id;
614 	ib_svc_id_t		ud_svc_id;
615 
616 	uint8_t			ud_max_cm_retries;
617 	uint8_t			ud_remaining_retry_cnt;
618 	ibt_cm_ud_handler_t	ud_cm_handler;
619 
620 	struct ibcm_ud_state_data_s	*ud_nextp;
621 	struct ibcm_hca_info_s *ud_hcap;
622 
623 	/* timeout related stuff */
624 	timeout_id_t		ud_timerid;
625 	clock_t			ud_timer_value;
626 	clock_t			ud_pkt_life_time;
627 	ibcm_mad_addr_t		ud_stored_reply_addr;
628 	ibmf_msg_t		*ud_stored_msg;
629 
630 
631 	/* SIDR REQ side related */
632 	ib_lid_t		ud_sidr_req_lid;
633 	ib_gid_t		ud_sidr_req_gid;
634 	boolean_t		ud_grh_exists;
635 
636 	/* Stored values on server/SIDR REP side for re-transmits */
637 	ib_qpn_t		ud_passive_qpn;
638 	ib_qkey_t		ud_passive_qp_qkey;
639 
640 	/* Clients' information */
641 	void			*ud_state_cm_private;
642 
643 	struct ibcm_ud_state_data_s	*ud_timeout_next;
644 	boolean_t		ud_delete_state_data;
645 	boolean_t		ud_blocking_done;
646 
647 	uint8_t			ud_send_mad_flags;
648 
649 	ibcm_isync_t		ud_clnt_proceed;
650 
651 	/* The following fields are not used by server side connection */
652 	kcondvar_t		ud_block_client_cv;
653 	ibt_ud_returns_t	*ud_return_data;
654 	ibcm_conn_state_t	ud_timer_stored_state;
655 } ibcm_ud_state_data_t;
656 
657 _NOTE(MUTEX_PROTECTS_DATA(ibcm_ud_state_data_s::ud_state_mutex,
658     ibcm_ud_state_data_s::{ud_state ud_ref_cnt ud_timerid
659     ud_delete_state_data ud_blocking_done ud_send_mad_flags ud_clnt_proceed
660     ud_timer_stored_state ud_send_mad_flags ud_clnt_proceed
661     ud_block_client_cv ud_timer_value ud_remaining_retry_cnt}))
662 
663 _NOTE(READ_ONLY_DATA(ibcm_ud_state_data_s::{ud_mode ud_req_id ud_svc_id
664     ud_max_cm_retries ud_pkt_life_time ud_stored_reply_addr ud_stored_msg
665     ud_sidr_req_lid ud_sidr_req_gid ud_grh_exists ud_passive_qpn
666     ud_passive_qp_qkey ud_state_cm_private ud_stored_reply_addr ud_stored_msg}))
667 
668 _NOTE(SCHEME_PROTECTS_DATA("Serailized access by ud_block_client_cv",
669     ibcm_ud_state_data_s::{ud_return_data}))
670 
671 _NOTE(DATA_READABLE_WITHOUT_LOCK(ibcm_ud_state_data_s::{ud_cm_handler}))
672 
673 /*
674  * Structure used to specify the SIDR search parameters
675  */
676 typedef struct ibcm_sidr_srch_s {
677 	ib_lid_t		srch_lid;
678 	ib_gid_t		srch_gid;
679 	boolean_t		srch_grh_exists;
680 	uint32_t		srch_req_id;
681 	ibcm_mode_t		srch_mode;
682 } ibcm_sidr_srch_t;
683 
684 _NOTE(READ_ONLY_DATA(ibcm_sidr_srch_s))
685 
686 /*
687  * Incr/Decr ud_ref_cnt by 1
688  */
689 #define	IBCM_UD_REF_CNT_INCR(s)	((s)->ud_ref_cnt++)
690 #define	IBCM_UD_REF_CNT_DECR(s)	\
691 	if ((--(s->ud_ref_cnt) == 0) && (s->ud_delete_state_data == B_TRUE)) { \
692 		ibcm_add_ud_tlist(s);\
693 	} \
694 	ASSERT(s->ud_ref_cnt >= 0);
695 
696 /*
697  * Structure to store the Service Registration and Service Bind entries.
698  *
699  * Well known service id's are unique on a given HCA, but can be registered
700  * only at some GID's. Hence can be multiple GID's per Service ID. For each
701  * such GID and PKEY combination registered, there will be an ibcm_svc_info_t
702  * entry in the CM global service list.
703  *
704  * Annex A of the spec constrains that there shall be one service provider per
705  * service id, which implies same svc_rc_handler for all such entries
706  * There can be multiple transport types (svc_tran_type) per Service ID. For
707  * each such transport type, there will be an ibcm_svc_info_t entry in the
708  * CM global service list and cm handler can be different
709  *
710  * For locally allocated service id's (maintained by OS), there can be only
711  * one GID, where the service can be registered
712  *
713  * svc_id:		Service ID
714  * svc_num_sids:	Number (Range) of service-ids supported
715  * svc_flags:		Service flags specified at registration time
716  * svc_link:		Global AVL tree of ibcm_svc_info_t structs
717  * svc_rc_handler:	Server handler for RC (only one is valid at a time)
718  * svc_ud_handler:	Server handler for UD (only one is valid at a time)
719  * svc_ref_cnt:		Reference count
720  * svc_to_delete:	If 1, then the entry is marked to be deleted
721  *
722  * sbind_gid:		GID
723  * sbind_pkey:		P_Key
724  * sbind_lease:		Service Lease
725  * sbind_name:		Service Name
726  */
727 typedef struct ibcm_svc_info_s {
728 	avl_node_t		svc_link;
729 	struct ibcm_svc_bind_s	*svc_bind_list;
730 	ibt_cm_handler_t	svc_rc_handler;
731 	ibt_cm_ud_handler_t	svc_ud_handler;
732 	int			svc_ref_cnt;
733 	int			svc_to_delete;
734 	ib_svc_id_t		svc_id;
735 	int			svc_num_sids;
736 	ibt_service_flags_t	svc_flags;
737 } ibcm_svc_info_t;
738 
739 typedef struct ibcm_svc_bind_s {
740 	struct ibcm_svc_bind_s	*sbind_link;
741 	void			*sbind_cm_private;
742 	ib_gid_t		sbind_gid;
743 	ib_guid_t		sbind_hcaguid;
744 	uint64_t		sbind_key[2];
745 				/* sbind_data is assumed to be 8-byte aligned */
746 	uint8_t			sbind_data[IB_SVC_DATA_LEN]; /* ServiceData */
747 	uint32_t		sbind_lease;
748 	ib_pkey_t		sbind_pkey;
749 	uint8_t			sbind_port;
750 	uint8_t			sbind_rewrite_state;
751 	char			sbind_name[IB_SVC_NAME_LEN];
752 } ibcm_svc_bind_t;
753 
754 /*
755  * Service records may be lost by the SM/SA (reboot, change in who
756  * is the master, etc.).  When any of the above occurs, a PORT_UP
757  * async event is supposed to occur, at which point we mark all of
758  * our service record information as stale (REWRITE_NEEDED), and
759  * subsequently make the necessary sa_update calls to get the
760  * SM/SA in sync with all the service records we previously wrote.
761  *
762  * Values for sbind_rewrite_state follow.  This field is protected by
763  * ibcm_svc_info_lock.  ibt_unbind_service has to wait until a service
764  * binding is either idle or needed, sleeping on ibcm_svc_info_cv if
765  * busy (rewrite in progress).
766  */
767 #define	IBCM_REWRITE_IDLE	0
768 #define	IBCM_REWRITE_NEEDED	1
769 #define	IBCM_REWRITE_BUSY	2
770 
771 typedef struct ibcm_port_up_s {
772 	ib_guid_t	pup_hca_guid;
773 	uint8_t		pup_port;
774 } ibcm_port_up_t;
775 
776 /* arg is a pointer to ibcm_port_up_t */
777 extern void ibcm_service_record_rewrite_task(void *);
778 
779 #define	IBCM_SVC_INCR(svcinfop) (svcinfop)->svc_ref_cnt++
780 #define	IBCM_SVC_DECR(svcinfop) \
781 	if (--((svcinfop)->svc_ref_cnt) == 0 && \
782 	    (svcinfop)->svc_to_delete) \
783 		cv_broadcast(&ibcm_svc_info_cv); \
784 	ASSERT(svcinfop->svc_ref_cnt >= 0);
785 
786 _NOTE(READ_ONLY_DATA(ibcm_svc_info_s::{svc_rc_handler svc_ud_handler svc_id
787     svc_num_sids svc_flags}))
788 
789 _NOTE(READ_ONLY_DATA(ibcm_svc_bind_s::{sbind_cm_private sbind_gid sbind_hcaguid
790     sbind_key sbind_data sbind_lease sbind_pkey sbind_port sbind_name}))
791 
792 /* for avl tree search */
793 typedef struct ibcm_svc_lookup_s {
794 	ib_svc_id_t	sid;
795 	int		num_sids;
796 } ibcm_svc_lookup_t;
797 
798 typedef struct ibcm_ar_ref_s {
799 	struct ibcm_ar_ref_s	*ar_ref_link;
800 	ibt_clnt_hdl_t		ar_ibt_hdl;
801 } ibcm_ar_ref_t;
802 
803 typedef struct ibcm_ar_s {
804 	ibt_ar_t		ar;
805 	int			ar_flags;	/* 1 = INITING, 2 = FAILED */
806 	int			ar_waiters;	/* # of waiters */
807 	kcondvar_t		ar_cv;
808 	uint8_t			ar_port;
809 	uint8_t			ar_rewrite_state; /* see sbind_rewrite_state */
810 	ibcm_ar_ref_t		*ar_ibt_hdl_list;
811 	struct ibcm_ar_s	*ar_link;
812 	sa_service_record_t	*ar_srv_recp;
813 	ibmf_saa_handle_t	ar_saa_handle;
814 	struct ibcm_hca_info_s	*ar_hcap;
815 } ibcm_ar_t;
816 
817 /* ar_flags */
818 #define	IBCM_AR_SUCCESS		0
819 #define	IBCM_AR_FAILED		1
820 #define	IBCM_AR_INITING		2
821 
822 
823 /*
824  * These flags are used for adding (if an entry does not exist) or
825  * for just looking one up
826  */
827 typedef enum ibcm_lookup_flag_e {
828 	IBCM_FLAG_LOOKUP		= 0,	/* just lookup */
829 	IBCM_FLAG_ADD			= 1,	/* just add */
830 	IBCM_FLAG_LOOKUP_AND_ADD	= 2	/* lookup first. add if  */
831 						/* lookup failed */
832 } ibcm_lookup_flag_t;
833 
834 typedef enum ibcm_finit_state_e {
835 	IBCM_FINIT_INIT,		/* CM's init is not yet completed */
836 	IBCM_FINIT_IDLE,		/* CM not in either init or fini */
837 	IBCM_FINIT_BUSY,		/* CM busy either in init or fini */
838 	IBCM_FINIT_FAIL,		/* Init failed */
839 	IBCM_FINIT_SUCCESS		/* Fini has succeeded */
840 } ibcm_finit_state_t;
841 
842 /*
843  * Identifies HCA's state. Used in the definition of ibcm_hca_info_t
844  * If HCA is in ACTIVE state only does CM allow any MAD processing.
845  */
846 typedef enum ibcm_hca_state_e {
847 	IBCM_HCA_INIT,
848 	IBCM_HCA_ACTIVE,
849 	IBCM_HCA_NOT_ACTIVE
850 } ibcm_hca_state_t;
851 
852 /* QP information per pkey, stored in port information */
853 typedef struct ibcm_qp_list_s {
854 	ib_pkey_t		qp_pkey;
855 	ibmf_qp_handle_t	qp_cm;
856 	uint32_t		qp_ref_cnt;
857 	struct ibcm_port_info_s *qp_port;
858 	struct ibcm_qp_list_s	*qp_next;
859 } ibcm_qp_list_t;
860 
861 _NOTE(READ_ONLY_DATA(ibcm_qp_list_s::{qp_pkey qp_cm qp_port qp_next}))
862 _NOTE(DATA_READABLE_WITHOUT_LOCK(ibcm_qp_list_s))
863 
864 /*
865  * port information per HCA
866  * port_ibmf_hdl	- contains IBMF handle for that port if valid
867  *			  otherwise is NULL
868  * port_ibmf_saa_hdl	- contains SA Access handle for that port if valid
869  *			  otherwise is NULL
870  */
871 typedef struct ibcm_port_info_s {
872 	ibmf_handle_t		port_ibmf_hdl;
873 	ibmf_saa_handle_t	port_ibmf_saa_hdl;
874 	ib_gid_t		port_sgid0;
875 	uint8_t			port_event_status;
876 	uint8_t			port_saa_open_in_progress;
877 	uint8_t			port_num;
878 	ibmf_register_info_t	port_ibmf_reg;
879 	ibmf_impl_caps_t	port_ibmf_caps;
880 	ibcm_qp_list_t		port_qp1;
881 	ibcm_qp_list_t		*port_qplist;
882 	struct ibcm_hca_info_s	*port_hcap;
883 } ibcm_port_info_t;
884 
885 _NOTE(READ_ONLY_DATA(ibcm_port_info_s::{port_num port_ibmf_caps port_qp1
886     port_hcap}))
887 
888 /* Value to indicate to exit the timeout list processing thread */
889 #define	IBCM_TIMEOUT_THREAD_EXIT	01
890 
891 /*
892  * IBCM code relies on AVL routines already in kernel for faster lookups.
893  * AVL was chosen over mod hashing mechanism based on the its internal
894  * limitations in the kernel (no support for over 100,000 keys).
895  *
896  * IBCM uses two AVL trees on the passive side and one on active side per HCA.
897  * The two trees are need on the passive side because the tree lookup criteria
898  * changes based on the type of message being processed. On passive side it is
899  * based on remote_qpn and remote_hca_guid for only incoming REQ message and for
900  * for all other messages the search criteria is based upon remote_comid.
901  * On active side the lookup criteria remains static based upon local_comid.
902  *
903  * AVL tree insertions are done by grabbing the writer lock (hca_state_rwlock)
904  * and lookups are done by grabbing the reader lock.
905  */
906 
907 /*
908  * CM's per HCA data structure.
909  *
910  * One such entry is added/removed on hca attach/detach notifications to CM
911  * respectively.
912  *
913  * Comids are used for all connections. Req ids are used for SIDR REQ and
914  * SIDR REP messages.  These are  simple counters that wrap around INT_MAX.
915  * NOTE: The starting value for comid, per HCA, is 2.
916  *
917  * hca_state:		HCA's current state (ibcm_hca_state_t) - whether
918  *				IBT_HCA_ACTIVE, IBT_HCA_NOT_ACTIVE,
919  * hca_guid:            Active HCA guid
920  * hca_caps:		HCA capability mask
921  * hca_ack_delay:	HCA ack delay
922  * hca_max_rdma_rd	Max RDMA in Reads
923  * hca_max_rdma_dpt	Max RDMA out Reads
924  * hca_active_tree:	This tree is used for lookups on Active/Passive side
925  *			CM based on communication id ONLY.
926  * hca_passive_tree:	This tree is used to lookup/create ibcm_state_data_t on
927  *			Passive Side CM based on remote_qpn and remote_hca_guid.
928  * hca_passive_comid_tree:
929  *			This tree is used to lookup/create ibcm_state_data_t on
930  *			Passive Side CM based on remote_comid and
931  *			remote_hca_guid.
932  * hca_state_rwlock:	reader/writer Lock for the hca entry
933  *				for hca_active_tree
934  *				for hca_passive_tree
935  *				for hca_next_comid
936  * hca_sidr_list:	List for UD side
937  * hca_sidr_list_lock:	List lock for UD side
938  *				for hca_sidr_list
939  *				for hca_next_reqid
940  * hca_next_reqid:	Next active ReqId
941  * hca_next_comid:	Next active ComID
942  * hca_next:		Pointer to the next HCA
943  * hca_svc_cnt:		A count of services registered on this hca
944  * hca_acc_cnt:		A count of active references to this ibcm_hca_info_t
945  * hca_res_cnt:		A count of client's active resources on this hca
946  * hca_num_ports:	Number of ports that this HCA has
947  * hca_port_info:	Per port information (IBMA/SA access handles etc.)
948  *
949  * Note : The global mutex ibcm_global_hca_mutex declared in CM is used for
950  * accesses to the following fields :
951  * hca_acc_cnt, hca_res_cnt, hca_svc_cnt, hca_state
952  */
953 typedef struct ibcm_hca_info_s {
954 	ibcm_hca_state_t	hca_state;		/* Is HCA attached? */
955 	ib_guid_t		hca_guid;		/* HCA's guid value */
956 	ibt_hca_flags_t		hca_caps;		/* HCA capabilities */
957 	ib_time_t		hca_ack_delay;		/* HCA ack delay */
958 	uint8_t			hca_max_rdma_in_qp;	/* Max RDMA in Reads */
959 	uint8_t			hca_max_rdma_out_qp;	/* Max RDMA out Reads */
960 	vmem_t			*hca_comid_arena;	/* arena for com ids */
961 	vmem_t			*hca_reqid_arena;	/* arena for req ids */
962 	avl_tree_t		hca_active_tree;	/* active node tree */
963 	avl_tree_t		hca_passive_tree;	/* passive node tree */
964 	avl_tree_t		hca_passive_comid_tree;	/* passive comid tree */
965 	krwlock_t		hca_state_rwlock;	/* per HCA lock */
966 	ibcm_ud_state_data_t	*hca_sidr_list;		/* SIDR state list */
967 	krwlock_t		hca_sidr_list_lock;
968 
969 	struct ibcm_hca_info_s	*hca_next;		/* Next HCA entry */
970 
971 	int			hca_svc_cnt;		/* # of */
972 							/* services allocated */
973 	int			hca_acc_cnt;		/* active references */
974 	int			hca_res_cnt;		/* total resources */
975 	uint8_t			hca_num_ports;		/* #ports on this HCA */
976 	ibcm_port_info_t	hca_port_info[1];	/* Per portinfo array */
977 } ibcm_hca_info_t;
978 
979 _NOTE(RWLOCK_PROTECTS_DATA(ibcm_hca_info_s::hca_state_rwlock,
980     ibcm_hca_info_s::{hca_active_tree hca_passive_tree hca_passive_comid_tree}))
981 
982 _NOTE(SCHEME_PROTECTS_DATA("hca_sidr_list_lock protects hca_sidr_list",
983     ibcm_hca_info_s::{hca_sidr_list}))
984 
985 _NOTE(READ_ONLY_DATA(ibcm_hca_info_s::{hca_guid hca_caps hca_ack_delay
986     hca_max_rdma_in_qp hca_max_rdma_out_qp hca_comid_arena hca_reqid_arena
987     hca_passive_tree hca_active_tree hca_passive_comid_tree hca_num_ports }))
988 
989 
990 /*
991  * called to ensure that HCA is in "attached" state and is willing to
992  * process connections etc.
993  */
994 #define	IBCM_ACCESS_HCA_OK(s)	((s)->hca_state == IBCM_HCA_ACTIVE)
995 
996 /*
997  * Passive AVL tree lookup info  (for hca_passive_tree)
998  * CM needs this structure as passive tree lookups are based on
999  * QPN and HCA GUID.
1000  */
1001 typedef	struct ibcm_passive_node_info_s {
1002 	ib_qpn_t	info_qpn;
1003 	ib_guid_t	info_hca_guid;
1004 } ibcm_passive_node_info_t;
1005 
1006 /*
1007  * Passive Com ID AVL tree lookup info  (for hca_passive_comid_tree)
1008  * CM needs this structure as passive comid tree lookups are based on
1009  * Remote Com ID and Remote HCA GUID.
1010  */
1011 typedef struct ibcm_passive_comid_node_info_s {
1012 	ib_com_id_t	info_comid;
1013 	ib_guid_t	info_hca_guid;
1014 } ibcm_passive_comid_node_info_t;
1015 
1016 /* CM proceed task args structure definition */
1017 typedef struct ibcm_proceed_targs_s {
1018 	ibt_cm_event_type_t	event;
1019 	ibt_cm_status_t		status;
1020 	union tst_t {
1021 		struct rc_s {
1022 			ibcm_state_data_t	*statep;
1023 			ibt_cm_proceed_reply_t	rc_cm_event_data;
1024 		} rc;
1025 		struct ud_s {
1026 			ibcm_ud_state_data_t	*ud_statep;
1027 			ib_qpn_t		ud_qpn;
1028 			ib_qkey_t		ud_qkey;
1029 			ibt_redirect_info_t	ud_redirect_info;
1030 		} ud;
1031 	} tst;
1032 	ibt_priv_data_len_t	priv_data_len;
1033 	/* keep priv_data as the last field */
1034 	uint8_t			priv_data[IBT_MAX_PRIV_DATA_SZ];
1035 } ibcm_proceed_targs_t;
1036 
1037 _NOTE(READ_ONLY_DATA(ibcm_proceed_targs_s))
1038 
1039 
1040 /*
1041  * function prototypes for AVL tree compares
1042  */
1043 int	ibcm_active_node_compare(const void *, const void *);
1044 int	ibcm_passive_node_compare(const void *, const void *);
1045 int	ibcm_passive_comid_node_compare(const void *, const void *);
1046 
1047 /*
1048  * function prototypes to allocate IBMF/SA_ACCESS handles
1049  */
1050 ibt_status_t	ibcm_hca_reinit_port(ibcm_hca_info_t *hca_p,
1051 		    uint8_t port_index);
1052 
1053 /* function prototypes to Manage CM's IBMF QP's */
1054 
1055 ibcm_qp_list_t *ibcm_find_qp(ibcm_hca_info_t *hcap, int port_no,
1056 		    ib_pkey_t pkey);
1057 
1058 void		ibcm_release_qp(ibcm_qp_list_t *cm_qp_entry);
1059 
1060 ibcm_status_t	ibcm_free_qp(ibcm_qp_list_t *cm_qp_entry);
1061 
1062 ibcm_status_t	ibcm_free_allqps(ibcm_hca_info_t *hcap, int port_no);
1063 
1064 /*
1065  * function prototypes to allocate and free outgoing CM messages
1066  */
1067 ibt_status_t
1068 ibcm_alloc_out_msg(ibmf_handle_t ibmf_handle, ibmf_msg_t **ibmf_msgpp,
1069     uint8_t method);
1070 ibcm_status_t
1071 ibcm_free_out_msg(ibmf_handle_t ibmf_handle, ibmf_msg_t **ibmf_msgpp);
1072 
1073 /*
1074  * Definition for CM state transition processing function
1075  */
1076 typedef void (*ibcm_state_handler_t)(ibcm_hca_info_t *hcap,
1077 		uint8_t *cm_input_mad, ibcm_mad_addr_t *cm_mad_addr);
1078 
1079 /*
1080  * CM REQ Message structure
1081  *
1082  * Request for communication.
1083  *
1084  * Things of interest are:-
1085  * ib_qpn_t cannot be used - it is typecast to uint32_t but is 24 bits
1086  * ib_eecn_t cannot be used - it is typecast to uint32_t but is 24 bits
1087  *
1088  * (See Table 85 REQ Message Contents - chapter 12 in IB Spec v1.0a)
1089  *
1090  */
1091 typedef struct ibcm_req_msg_s {
1092 	ib_com_id_t	req_local_comm_id;	/* Local communication id */
1093 						/* 32 bits */
1094 	uint32_t	req_rsvd1;		/* Reserved1 - 32 bits */
1095 	ib_svc_id_t	req_svc_id;		/* Service Id - 64 bits */
1096 	ib_guid_t	req_local_ca_guid;	/* Local CA GUID - 64 bits */
1097 	uint32_t	req_rsvd1p;		/* Reserved1+ - 32 bits */
1098 	ib_qkey_t	req_local_qkey;		/* Local Q_KEY - 32 bits */
1099 	uint32_t	req_local_qpn_plus;	/* QPN_24 RESP_RSRC_8 */
1100 						/* local side QPN - 24 bits */
1101 						/* Offered responder */
1102 						/* resources - 8 bits */
1103 	uint32_t	req_local_eec_no_plus;	/* LOCAL_EECN_24 INIT_DEPTH_8 */
1104 						/* Local side EECN - 24 bits */
1105 						/* Offered initiator */
1106 						/* depth - 8 bits */
1107 	uint32_t	req_remote_eecn_plus;	/* REM_EECN_24 TO_5 TT_2 EE_1 */
1108 						/* Remote side EECN - 24 bits */
1109 						/* Remote CM timeout - 5 bits */
1110 						/* Transport srvtype - 2 bits */
1111 						/* End-to-End flow - 1 bit */
1112 	uint32_t	req_starting_psn_plus;	/* START_PSN_24 TO_5 RETRY_3 */
1113 						/* Starting PSN - 24 bits */
1114 						/* Local CM timeout - 5 bits */
1115 						/* Retry count - 3 bits */
1116 	ib_pkey_t	req_part_key;		/* Partition key - 16 bits */
1117 	uint8_t		req_mtu_plus;		/* PATH_MTU_4 RDC_1 RNR_3 */
1118 						/* Path Pkt MTU - 4 bits */
1119 						/* Does RDC exist? - 1 bits */
1120 						/* RNR retry count - 3 bits */
1121 	uint8_t		req_max_cm_retries_plus; /* MAX_CM_RET_4 SRQ_1 RSV_3 */
1122 						/* Max CM retries - 4 bits */
1123 						/* SRQ Exists - 1 bit */
1124 						/* Reserved2 - 3 bits */
1125 	ib_lid_t	req_primary_l_port_lid;	/* Primary local port LID */
1126 	ib_lid_t	req_primary_r_port_lid;	/* Primary Remote port LID */
1127 	ib_gid_t	req_primary_l_port_gid;	/* Primary local port GID */
1128 	ib_gid_t	req_primary_r_port_gid;	/* Primary remote port GID */
1129 	uint32_t	req_primary_flow_label_plus; /* FLOW_20 RSV_4 SRATE_6 */
1130 						/* Prim. flow label - 20 bits */
1131 						/* Reserved3 - 6 bits */
1132 						/* Primary rate - 6 bits */
1133 	uint8_t		req_primary_traffic_class;
1134 						/* Primary Traffic class */
1135 	uint8_t		req_primary_hop_limit;	/* Prim Hop Limit */
1136 	uint8_t		req_primary_sl_plus;	/* PRIMARY_SL_4 LOCAL_1 RSV_3 */
1137 						/* Primary SL - 4 bits */
1138 						/* Prim. subnet local - 1 bit */
1139 						/* Reserved4 - 3 bits */
1140 	uint8_t		req_primary_localtime_plus; /* LOCAL_TO_5 RSV_3 */
1141 						/* Primary local */
1142 						/* timeout - 5 bits */
1143 						/* Reserved5 - 3 bits */
1144 	ib_lid_t	req_alt_l_port_lid;	/* Alt local port LID */
1145 	ib_lid_t	req_alt_r_port_lid;	/* Alt Remote port LID */
1146 	/* Note: req_alt_l_port_gid/req_alt_r_port_gid are not 8-byte aligned */
1147 	uint8_t		req_alt_l_port_gid[16];	/* Alt local port GID */
1148 	uint8_t		req_alt_r_port_gid[16];	/* Alt remote port GID */
1149 	uint32_t	req_alt_flow_label_plus; /* ALT_FLOW_20 RSV_6 ARATE_6 */
1150 						/* Alt flow label - 20 bits */
1151 						/* Reserved6 - 6 bits */
1152 						/* Alternate rate - 6 bits */
1153 	uint8_t		req_alt_traffic_class;	/* Alt traffic class */
1154 	uint8_t		req_alt_hop_limit;	/* Alt hop limit */
1155 	uint8_t		req_alt_sl_plus;	/* ALT_SL_4 A_LOCAL_1 RSV_3 */
1156 						/* Alternate SL - 4 bits */
1157 						/* Alt subnet local - 1 bit */
1158 						/* Reserved7 - 3 bits */
1159 	uint8_t		req_alt_localtime_plus;	/* ALT_LOCAL_ACK_TO_5 RSV_3 */
1160 						/* Alt Local ACK */
1161 						/* timeout - 5 bits */
1162 						/* Reserved8 - 3 bits */
1163 	uint8_t		req_private_data[IBT_REQ_PRIV_DATA_SZ];
1164 						/* Private data */
1165 } ibcm_req_msg_t;
1166 
1167 
1168 /*
1169  * The following set of defines are short-cuts to CEP_PATH or GRH info
1170  */
1171 #define	IBCM_PRIM_CEP_PATH(s)	(s)->oc_path->pi_prim_cep_path
1172 #define	IBCM_PRIM_ADDS_VECT(s)	(s)->oc_path->pi_prim_cep_path.cep_adds_vect
1173 
1174 #define	IBCM_ALT_CEP_PATH(s)	(s)->oc_path->pi_alt_cep_path
1175 #define	IBCM_ALT_ADDS_VECT(s)	(s)->oc_path->pi_alt_cep_path.cep_adds_vect
1176 
1177 #define	IBCM_UD_CEP_PATH(s)	(s)->us_path_info->ai_cep_path
1178 #define	IBCM_UD_ADDS_VECT(s)	(s)->us_path_info->ai_cep_path.cep_adds_vect
1179 
1180 /*
1181  * The following set of defines are short-cuts to ibt_cm_event_t
1182  */
1183 #define	IBCM_EVT_REQ(e)		(e).cm_event.req
1184 #define	IBCM_EVT_REP(e)		(e).cm_event.rep
1185 
1186 /*
1187  * The following set of defines are short-cuts to qp_attrs or qp_info
1188  */
1189 #define	IBCM_QP_RC(q)		(q).qp_info.qp_transport.rc
1190 #define	IBCM_QP_UD(q)		(q).qp_info.qp_transport.ud
1191 #define	IBCM_QP_UC(q)		(q).qp_info.qp_transport.uc
1192 
1193 #define	IBCM_QPINFO(q)		(q).qp_transport
1194 #define	IBCM_QPINFO_RC(q)	(q).qp_transport.rc
1195 #define	IBCM_QPINFO_RC_PATH(q)	(q).qp_transport.rc.rc_path
1196 #define	IBCM_QPINFO_UC(q)	(q).qp_transport.uc
1197 #define	IBCM_QPINFO_UC_PATH(q)	(q).qp_transport.uc.uc_path
1198 #define	IBCM_QPINFO_UD(q)	(q).qp_transport.ud
1199 
1200 
1201 /* The following set of defines are short-cuts to RC and SIDR MAD HDRs */
1202 
1203 #define	IBCM_OUT_MADP(msgp)	(msgp->im_msgbufs_send.im_bufs_mad_hdr)
1204 #define	IBCM_OUT_HDRP(msgp)	((ib_mad_hdr_t *)IBCM_OUT_MADP(msgp))
1205 #define	IBCM_OUT_MSGP(msgp)	(msgp->im_msgbufs_send.im_bufs_cl_data)
1206 
1207 #define	IBCM_IN_MADP(msgp)	(msgp->im_msgbufs_recv.im_bufs_mad_hdr)
1208 #define	IBCM_IN_HDRP(msgp)	((ib_mad_hdr_t *)IBCM_IN_MADP(msgp))
1209 #define	IBCM_IN_MSGP(msgp)	(msgp->im_msgbufs_recv.im_bufs_cl_data)
1210 
1211 #define	IBCM_REJ_PRIV(msgp)  &(((ibcm_rej_msg_t *) \
1212 	IBCM_OUT_MSGP(statep->stored_msg))->rej_private_data[0])
1213 /*
1214  * CM MRA Message structure
1215  *
1216  * Message Receipt Acknowledgement (MRA).
1217  *
1218  * NOTE: IB hosts and targets are required to be able to receive and
1219  * act upon an MRA, but the ability to send an MRA is optional.
1220  */
1221 typedef struct ibcm_mra_msg_s {
1222 	ib_com_id_t	mra_local_comm_id;	/* Local communication id */
1223 	ib_com_id_t	mra_remote_comm_id;	/* Remote communication id */
1224 	uint8_t		mra_message_type_plus;	/* Message Type - 2 bits */
1225 						/* Reserved1 - 6 bits */
1226 	uint8_t		mra_service_timeout_plus; /* SVC_TO_5 RSV_3 */
1227 						/* Service timeout - 5 bits */
1228 						/* Reserved2 - 3 bits */
1229 	uint8_t		mra_private_data[IBT_MRA_PRIV_DATA_SZ];
1230 						/* Private data */
1231 } ibcm_mra_msg_t;
1232 
1233 /*
1234  * CM REJ Message structure
1235  * REJ indicates that the sender will not continue through the communication
1236  * establishment sequence and the reason why it will not.
1237  *
1238  * NOTE: See ibt_cm_reason_t in common/sys/ib/ib_cm.h for complete list
1239  * of rejection reasons supported.
1240  */
1241 typedef struct ibcm_rej_msg_s {
1242 	ib_com_id_t	rej_local_comm_id;	/* Local communication id */
1243 	ib_com_id_t	rej_remote_comm_id;	/* Remote communication id */
1244 	uint8_t		rej_msg_type_plus;	/* REJ_MSG_TYPE_2 RSV_6 */
1245 						/* Msg being REJed - 2 bits */
1246 						/* Reserved1 - 6 bits */
1247 	uint8_t		rej_reject_info_len_plus; /* REJ_INFO_LEN_7 RSV_1 */
1248 						/* Rej. Info Length - 7 bits */
1249 						/* Reserved2 - 1 bit */
1250 	uint16_t	rej_rejection_reason;	/* Reject err code - 16 bits */
1251 	uint8_t		rej_addl_rej_info[IBT_CM_ADDL_REJ_LEN];
1252 						/* Additional Reject Info */
1253 	uint8_t		rej_private_data[IBT_REJ_PRIV_DATA_SZ];
1254 						/* Private data */
1255 } ibcm_rej_msg_t;
1256 
1257 /*
1258  * CM REP Message structure
1259  *
1260  * REP is returned in response to REQ, indicating that the respondent
1261  * accepts the Service-ID, proposed primary port, and any parameters
1262  * specified in the PrivateData of the REQ.
1263  */
1264 typedef struct ibcm_rep_msg_s {
1265 	ib_com_id_t	rep_local_comm_id;	/* Local communication id */
1266 	ib_com_id_t	rep_remote_comm_id;	/* Remote communication id */
1267 	ib_qkey_t	rep_local_qkey;		/* Local Q_KEY */
1268 	uint32_t	rep_local_qpn_plus;	/* LOCAL_QPN_24 RSV_8 */
1269 						/* Local side QPN - 24 bits */
1270 						/* Reserved1 - 8 bits */
1271 	uint32_t	rep_local_eecn_plus;	/* LOCAL_EECN_24 RSV_8 */
1272 						/* Local side EECN - 24 bits */
1273 						/* Reserved2 - 8 bits */
1274 	uint32_t	rep_starting_psn_plus;	/* STARTING_PSN_24 RSV_8 */
1275 						/* Starting PSN - 24 bits */
1276 						/* Reserved3 - 8 bits */
1277 	uint8_t		rep_resp_resources;	/* Responder resources 8 bits */
1278 	uint8_t		rep_initiator_depth;	/* Initiator depth - 8 bits */
1279 	uint8_t		rep_target_delay_plus;	/* TGT_ACK_DLY_5 FAIL_2 EE_1 */
1280 						/* Target ACK delay - 5 bits */
1281 						/* Failover accepted - 2 bits */
1282 						/* End-to-End flow control - */
1283 						/* 1 bit */
1284 	uint8_t		rep_rnr_retry_cnt_plus;	/* RNR_COUNT_3 SRQ_1 RSV_4 */
1285 						/* RNR retry count - 3 bits */
1286 						/* SRQ Exists - 1 bit */
1287 						/* Reserved4 - 4 bits */
1288 	uint8_t		rep_local_ca_guid[8];	/* Local CA GUID - 64 bits */
1289 	uint8_t		rep_private_data[IBT_REP_PRIV_DATA_SZ];
1290 						/* Private data */
1291 } ibcm_rep_msg_t;
1292 
1293 
1294 /*
1295  * CM RTU Message structure
1296  *
1297  * RTU indicates that the connection is established, and that the
1298  * recipient may begin transmitting.
1299  */
1300 typedef struct ibcm_rtu_msg_s {
1301 	ib_com_id_t	rtu_local_comm_id;	/* Local communication id */
1302 	ib_com_id_t	rtu_remote_comm_id;	/* Remote communication id */
1303 	uint8_t		rtu_private_data[IBT_RTU_PRIV_DATA_SZ];
1304 						/* Private data */
1305 } ibcm_rtu_msg_t;
1306 
1307 
1308 /*
1309  * CM DREQ Message structure
1310  *
1311  * DREQ is sent to initiate the connection release sequence.
1312  */
1313 typedef struct ibcm_dreq_msg_s {
1314 	ib_com_id_t	dreq_local_comm_id;	/* Local communication id */
1315 	ib_com_id_t	dreq_remote_comm_id;	/* Remote communication id */
1316 	uint32_t	dreq_remote_qpn_eecn_plus; /* REM_EECN_24 RSV_8 */
1317 						/* Remote QPN/EECN - 24 bits */
1318 						/* reserved - 8 bits */
1319 	uint8_t		dreq_private_data[IBT_DREQ_PRIV_DATA_SZ];
1320 						/* Private data */
1321 } ibcm_dreq_msg_t;
1322 
1323 
1324 /*
1325  * CM DREP Message structure
1326  *
1327  * DREP is sent in response to DREQ, and signifies that the sender has
1328  * received DREQ.
1329  */
1330 typedef struct ibcm_drep_msg_s {
1331 	ib_com_id_t	drep_local_comm_id;	/* Local communication id */
1332 	ib_com_id_t	drep_remote_comm_id;	/* Remote communication id */
1333 	uint8_t		drep_private_data[IBT_DREP_PRIV_DATA_SZ];
1334 						/* Private Data */
1335 } ibcm_drep_msg_t;
1336 
1337 
1338 /*
1339  * CM LAP Message structure
1340  *
1341  * NOTE: LAP and APR messages are optional. These are needed if CM
1342  * accepts REQ messages and agrees to perform Automatic Path Migration.
1343  *
1344  * This message is used to change the alternate path information for a
1345  * specific connection.
1346  */
1347 typedef struct ibcm_lap_msg_s {
1348 	ib_com_id_t	lap_local_comm_id;	/* Local communication id */
1349 	ib_com_id_t	lap_remote_comm_id;	/* Remote communication id */
1350 	uint32_t	lap_rsvd1;		/* Reserved - 32 bits */
1351 	uint32_t	lap_remote_qpn_eecn_plus; /* REM_EECN_24 TO_5 RSV_3 */
1352 						/* Remote QPN/EECN - 24 bits */
1353 						/* Remote CM response */
1354 						/* timeout - 5 bits */
1355 						/* Reserved1 - 3 bits */
1356 	uint32_t	lap_rsvd2;		/* Reserved2 - 32 bits */
1357 	ib_lid_t	lap_alt_l_port_lid;	/* Alt local port LID */
1358 	ib_lid_t	lap_alt_r_port_lid;	/* Alt Remote port LID */
1359 	ib_gid_t	lap_alt_l_port_gid;	/* Alt local port GID */
1360 	ib_gid_t	lap_alt_r_port_gid;	/* Alt remote port GID */
1361 	uint32_t	lap_alt_flow_label_plus; /* ALT_FLOW_20 RSV_4 TCL_8 */
1362 						/* Alt flow label - 20 bits */
1363 						/* Reserved3 - 4 bits */
1364 						/* Alt traffic class - 8 bits */
1365 	uint8_t		lap_alt_hop_limit;	/* Alt hop limit */
1366 	uint8_t		lap_alt_srate_plus;	/* Reserved4 - 2 bits */
1367 						/* Alt. static rate - 6 bits */
1368 	uint8_t		lap_alt_sl_plus;	/* ALT_SL_4 A_LOCAL_1 RSV_3 */
1369 						/* Alternate SL - 4 bits */
1370 						/* Alt subnet local - 1 bit */
1371 						/* Reserved5 - 3 bits */
1372 	uint8_t		lap_alt_local_acktime_plus; /* ALT_TO_5 RSV_3 */
1373 						/* Alt Local ACK */
1374 						/* timeout - 5 bits */
1375 						/* Reserved6 - 3 bits */
1376 	uint8_t		lap_private_data[IBT_LAP_PRIV_DATA_SZ];
1377 						/* Private data */
1378 } ibcm_lap_msg_t;
1379 
1380 
1381 /*
1382  * CM APR Message structure
1383  *
1384  * APR is sent in response to a LAP request. MRA may be sent to allow
1385  * processing of the LAP.
1386  */
1387 typedef struct ibcm_apr_msg_s {
1388 	ib_com_id_t	apr_local_comm_id;	/* Local communication id */
1389 	ib_com_id_t	apr_remote_comm_id;	/* Remote communication id */
1390 	uint8_t		apr_addl_info_len;	/* Add'l Info Len - 8 bits */
1391 	uint8_t		apr_ap_status;		/* AP status - 8 bits */
1392 	uint16_t	apr_rsvd1;		/* Reserved1 - 16 bits */
1393 	uint8_t		apr_addl_info[IBT_CM_APR_ADDL_LEN];
1394 						/* Additional Information */
1395 	uint8_t		apr_private_data[IBT_APR_PRIV_DATA_SZ];
1396 						/* Private data */
1397 } ibcm_apr_msg_t;
1398 
1399 
1400 /*
1401  * CM SIDR_REQ Message structure
1402  *
1403  * NOTE: SIDR_REQ and SIDR_REP messages are conditionally required.
1404  * These are needed if non-management services are provided on the Channel
1405  * Adapter other than fixed QPNs. Management services include those
1406  * provided thru Subnet Manager Packets or thru General Management Packets.
1407  *
1408  * SIDR_REQ requests that the recipient return the information necessary
1409  * to communicate via UD messages with the entity specified by
1410  * SIDR_REQ:ServiceID
1411  */
1412 typedef struct ibcm_sidr_req_msg_s {
1413 	uint32_t	sidr_req_request_id;		/* Request id */
1414 	ib_pkey_t	sidr_req_pkey;			/* P_Key */
1415 	uint8_t		sidr_req_reserved[2];		/* Reserved */
1416 	ib_svc_id_t	sidr_req_service_id;		/* Service Id */
1417 	uint8_t		sidr_req_private_data[IBT_SIDR_REQ_PRIV_DATA_SZ];
1418 							/* Private Data */
1419 } ibcm_sidr_req_msg_t;
1420 
1421 
1422 /*
1423  * CM SIDR_REP Message structure
1424  *
1425  * SIDR_REP returns the information necessary to communicate via UD
1426  * messages with the entity specified by SIDR_REQ:ServiceID
1427  */
1428 typedef struct ibcm_sidr_rep_msg_s {
1429 	uint32_t	sidr_rep_request_id;		/* Request id */
1430 	uint8_t		sidr_rep_rep_status;		/* Status */
1431 	uint8_t		sidr_rep_add_info_len;		/* Length of Add Info */
1432 	uint8_t		sidr_rep_reserved1[2];		/* Reserved */
1433 	uint32_t	sidr_rep_qpn_plus;		/* QPN_24 RSV_8 */
1434 	/* since the 64-bit SID is not aligned, treat it as a byte array */
1435 	uint8_t		sidr_rep_service_id[8];		/* Service Id */
1436 	ib_qkey_t	sidr_rep_qkey;			/* Q_KEY */
1437 	uint8_t		sidr_rep_class_port_info[IBT_CM_SIDR_CP_LEN];
1438 							/* Class Port Info */
1439 							/* aka., add'l info */
1440 	uint8_t		sidr_rep_private_data[IBT_SIDR_REP_PRIV_DATA_SZ];
1441 							/* Private data */
1442 } ibcm_sidr_rep_msg_t;
1443 
1444 typedef struct ibcm_classportinfo_msg_s {
1445 	uint8_t		BaseVersion;		/* ver. of MAD base format */
1446 	uint8_t		ClassVersion;		/* ver. of MAD class format */
1447 	uint16_t	CapabilityMask;		/* capabilities of this class */
1448 	uint32_t	RespTimeValue_plus;	/* reserved : 27 bits */
1449 						/* resptime value : 5 bits */
1450 	uint64_t	RedirectGID_hi;		/* dest gid of redirect msgs */
1451 	uint64_t	RedirectGID_lo;		/* dest gid of redirect msgs */
1452 	uint32_t	RedirectTC_plus;	/* traffic class: 8 bits */
1453 						/* SL: 4 bits */
1454 						/* Flow label: 20 bits */
1455 	ib_lid_t	RedirectLID;		/* dlid for class services */
1456 	ib_pkey_t	RedirectP_Key;		/* p_key for class services */
1457 	uint32_t	RedirectQP_plus;	/* Reserved: 8 bits */
1458 						/* QPN: 24 bits */
1459 	ib_qkey_t	RedirectQ_Key;		/* q_key for class services */
1460 	uint64_t	TrapGID_hi;		/* dest gid of trap msgs */
1461 	uint64_t	TrapGID_lo;		/* dest gid of trap msgs */
1462 	uint32_t	TrapTC_plus;		/* Trap traffic class, etc., */
1463 	ib_lid_t	TrapLID;		/* dlid for traps */
1464 	ib_pkey_t	TrapP_Key;		/* p_key for traps */
1465 	uint32_t	TrapHL_plus;		/* Trap hop limit,etc., */
1466 	ib_qkey_t	TrapQ_Key;		/* q_key for traps */
1467 } ibcm_classportinfo_msg_t;
1468 
1469 /* All msgs are readonly on receiving side */
1470 _NOTE(READ_ONLY_DATA(ibcm_req_msg_s))
1471 _NOTE(READ_ONLY_DATA(ibcm_rep_msg_s))
1472 _NOTE(READ_ONLY_DATA(ibcm_mra_msg_s))
1473 _NOTE(READ_ONLY_DATA(ibcm_rej_msg_s))
1474 _NOTE(READ_ONLY_DATA(ibcm_lap_msg_s))
1475 _NOTE(READ_ONLY_DATA(ibcm_apr_msg_s))
1476 _NOTE(READ_ONLY_DATA(ibcm_sidr_req_msg_s))
1477 _NOTE(READ_ONLY_DATA(ibcm_sidr_rep_msg_s))
1478 _NOTE(READ_ONLY_DATA(ibcm_rtu_msg_s))
1479 _NOTE(READ_ONLY_DATA(ibcm_dreq_msg_s))
1480 _NOTE(READ_ONLY_DATA(ibcm_drep_msg_s))
1481 _NOTE(READ_ONLY_DATA(ibcm_classportinfo_msg_s))
1482 
1483 /* Prototype definitions for CM implementation functions */
1484 
1485 /*
1486  * The callback from IBMF to CM. This routines calls one of the CM
1487  * state processing functions depending upon mesg/attribute id
1488  *
1489  * ibmf_handle	: IBMF handle on which CM MAD was received
1490  * pktp		: MAD packet
1491  * args		: IBMF receive mad callback arg
1492  */
1493 void	ibcm_recv_cb(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp, void *args);
1494 
1495 /*
1496  * Prototypes for CM state transition handling functions
1497  */
1498 
1499 /*
1500  * The following are the CM state processing functions called on an
1501  * incoming REQ/REP/RTU/MRA/REJ/DREQ/DREP on active/passive sides
1502  * (Also handled are SIDR_REP and SIDR_REQ)
1503  * The brief description of these functions
1504  *	Search based on CM message fields in CM's HCA entry.
1505  *	Create/Delete state structures based on incoming message
1506  *	Handle duplicate messages and state transitions
1507  *	Set and Cancel timeouts
1508  *	Handle stale connections
1509  *	Change CM connection state
1510  *	Call CM CEP state transition functions to update CEP state
1511  *	and set CEP attributes
1512  *
1513  * INPUTS:
1514  *	hcap:		- IBMF callback argument
1515  *	cm_input_mad:	- ibmf message pointer of incoming MAD
1516  *	cm_mad_addr	- CM MAD address
1517  *
1518  * The state transition processing is specified in different functions based
1519  * on incoming message type rather than as one function because, the CM
1520  * processing is different for each of them.
1521  *
1522  * A global call table is initialized with these function addresses
1523  * (is defined in ibcm_impl.c), and invoked from ibcm_recv_cb
1524  * (IBMF's recv callback to CM) based on mesg/attribute id.
1525  */
1526 void	ibcm_process_req_msg(ibcm_hca_info_t *hcap, uint8_t *cm_input_mad,
1527 	    ibcm_mad_addr_t *cm_mad_addr);
1528 void	ibcm_process_rep_msg(ibcm_hca_info_t *hcap, uint8_t *cm_input_mad,
1529 	    ibcm_mad_addr_t *cm_mad_addr);
1530 void	ibcm_process_rtu_msg(ibcm_hca_info_t *hcap, uint8_t *cm_input_mad,
1531 	    ibcm_mad_addr_t *cm_mad_addr);
1532 void	ibcm_process_dreq_msg(ibcm_hca_info_t *hcap, uint8_t *cm_input_mad,
1533 	    ibcm_mad_addr_t *cm_mad_addr);
1534 void	ibcm_process_drep_msg(ibcm_hca_info_t *hcap, uint8_t *cm_input_mad,
1535 	    ibcm_mad_addr_t *cm_mad_addr);
1536 void	ibcm_process_rej_msg(ibcm_hca_info_t *hcap, uint8_t *cm_input_mad,
1537 	    ibcm_mad_addr_t *cm_mad_addr);
1538 void	ibcm_process_mra_msg(ibcm_hca_info_t *hcap, uint8_t *cm_input_mad,
1539 	    ibcm_mad_addr_t *cm_mad_addr);
1540 void	ibcm_process_apr_msg(ibcm_hca_info_t *hcap, uint8_t *cm_input_mad,
1541 	    ibcm_mad_addr_t *cm_mad_addr);
1542 void	ibcm_process_lap_msg(ibcm_hca_info_t *hcap, uint8_t *cm_input_mad,
1543 	    ibcm_mad_addr_t *cm_mad_addr);
1544 void	ibcm_process_sidr_req_msg(ibcm_hca_info_t *hcap,
1545 	    uint8_t *cm_input_mad, ibcm_mad_addr_t *cm_mad_addr);
1546 void	ibcm_process_sidr_rep_msg(ibcm_hca_info_t *hcap,
1547 	    uint8_t *cm_input_mad, ibcm_mad_addr_t *cm_mad_addr);
1548 
1549 typedef enum ibcm_proceed_error_e {
1550 	IBCM_PROCEED_INVALID_NONE	= 0,
1551 	IBCM_PROCEED_INVALID_EVENT,
1552 	IBCM_PROCEED_INVALID_EVENT_STATE,
1553 	IBCM_PROCEED_INVALID_PRIV_SZ,
1554 	IBCM_PROCEED_INVALID_LAP
1555 } ibcm_proceed_error_t;
1556 
1557 /* Encapsulates the information that client returns back from CM callback */
1558 typedef struct ibcm_clnt_reply_info_s {
1559 	ibt_cm_proceed_reply_t	*reply_event;
1560 	void			*priv_data;
1561 	ibt_priv_data_len_t	priv_data_len;
1562 } ibcm_clnt_reply_info_t;
1563 
1564 /* Encapsulates the information that UD client returns back from CM callback */
1565 typedef struct ibcm_ud_clnt_reply_info_s {
1566 	ib_qpn_t		ud_qpn;
1567 	ib_qkey_t		ud_qkey;
1568 	ibt_redirect_info_t	*redirect_infop;
1569 	void			*priv_data;
1570 	ibt_priv_data_len_t	priv_data_len;
1571 } ibcm_ud_clnt_reply_info_t;
1572 
1573 /*
1574  * Prototypes for CM CEP state transition handling functions. These are
1575  * called from CM connection state transition handling functions.
1576  *
1577  * The brief description of these functions :
1578  *	Validate CEP related attributes in the messages
1579  *	Change CEP state
1580  *	Set CEP attributes (modify CEP)
1581  *	Call client/server callback handlers
1582  *	Fill up the response MADs
1583  *
1584  * The arguments are :
1585  *	statep:		Connection state structure
1586  *	cm_req/rep/rtu/rej msg : Received CM message
1587  *	cm_output_mad	: The response CM MAD with some of the fields filled in
1588  *			  The cm output mad is allocated by CM state transition
1589  *			  functions and has generic MAD header
1590  *			  Certain fields like com id, etc., are filled by CM
1591  *			  connection state transition functions that are above
1592  */
1593 
1594 /* QP state transition function called for an incoming REQ on passive side */
1595 ibcm_status_t	ibcm_cep_state_req(ibcm_state_data_t *statep,
1596 		    ibcm_req_msg_t *cm_req_msg, ibt_cm_reason_t *reason,
1597 		    uint8_t *arej_info_len);
1598 
1599 /* Processes QP state machine based on return values from cm handler */
1600 ibcm_status_t	ibcm_process_cep_req_cm_hdlr(ibcm_state_data_t *statep,
1601 		    ibt_cm_status_t cb_status,
1602 		    ibcm_clnt_reply_info_t *clnt_info,
1603 		    ibt_cm_reason_t *reject_reason, uint8_t *arej_len,
1604 		    ibcm_req_msg_t *cm_req_msgp);
1605 
1606 /* Processes CM state machine based on return values from ibcm_cep_state_req */
1607 void		ibcm_handle_cep_req_response(ibcm_state_data_t *statep,
1608 		    ibcm_status_t response, ibt_cm_reason_t reject_reason,
1609 		    uint8_t arej_info_len);
1610 
1611 /* QP state transition function called for an incoming REP on active side */
1612 ibcm_status_t	ibcm_cep_state_rep(ibcm_state_data_t *statep,
1613 		    ibcm_rep_msg_t *cm_rep_msg, ibt_cm_reason_t *reason,
1614 		    uint8_t *arej_info_len);
1615 
1616 /* Processes QP state machine based on return values from cm handler */
1617 ibcm_status_t	ibcm_process_cep_rep_cm_hdlr(ibcm_state_data_t *statep,
1618 		    ibt_cm_status_t cb_status,
1619 		    ibcm_clnt_reply_info_t *clnt_info,
1620 		    ibt_cm_reason_t *reject_reason, uint8_t *arej_len,
1621 		    ibcm_rep_msg_t *cm_rep_msgp);
1622 
1623 /* Processes CM state machine based on return values from ibcm_cep_state_rep */
1624 void		ibcm_handle_cep_rep_response(ibcm_state_data_t *statep,
1625 		    ibcm_status_t response, ibt_cm_reason_t reject_reason,
1626 		    uint8_t arej_info_len, ibcm_rep_msg_t *rep_msgp);
1627 
1628 /* QP state transition function called for an incoming RTU on passive side */
1629 void	ibcm_cep_state_rtu(ibcm_state_data_t *statep,
1630 	    ibcm_rtu_msg_t *cm_rtu_msg);
1631 
1632 /* QP state transition func called for an incoming REJ on active/passive side */
1633 void	ibcm_cep_state_rej(ibcm_state_data_t *statep,
1634 	    ibcm_rej_msg_t *cm_rej_msg, ibcm_conn_state_t rej_state);
1635 
1636 /* QP state transition func for an incoming REJ on active side in est state */
1637 void	ibcm_cep_state_rej_est(ibcm_state_data_t *statep);
1638 
1639 /*
1640  * QP state transition function called for an outgoing RTU on active side,
1641  * after setting CEP to RTS state active/passive side
1642  */
1643 void	ibcm_cep_send_rtu(ibcm_state_data_t *statep);
1644 
1645 
1646 /* QP state transition function called for an incoming LAP */
1647 ibcm_status_t	ibcm_cep_state_lap(ibcm_state_data_t *statep,
1648 		    ibcm_lap_msg_t *lap_msg, ibcm_apr_msg_t *apr_msg);
1649 
1650 /* Processes QP state machine based on return value from cm handler for LAP */
1651 void		ibcm_process_cep_lap_cm_hdlr(ibcm_state_data_t *statep,
1652 		    ibt_cm_status_t cb_status,
1653 		    ibcm_clnt_reply_info_t *clnt_info,
1654 		    ibcm_lap_msg_t *lap_msg, ibcm_apr_msg_t *apr_msg);
1655 
1656 void		ibcm_post_apr_mad(ibcm_state_data_t *statep);
1657 
1658 void		ibcm_cep_state_apr(ibcm_state_data_t *statep,
1659 		    ibcm_lap_msg_t *lap_msg, ibcm_apr_msg_t *apr_msg);
1660 
1661 /* Processes CM state machine based on return value from cm handler */
1662 void		ibcm_handle_cep_dreq_response(ibcm_state_data_t *statep,
1663 		    void *priv_data, ibt_priv_data_len_t  priv_data_len);
1664 
1665 /* Processes CM UD state machine based on return values from cm handler */
1666 void		ibcm_process_sidr_req_cm_hdlr(ibcm_ud_state_data_t *ud_statep,
1667 		    ibt_cm_status_t cb_status,
1668 		    ibcm_ud_clnt_reply_info_t *ud_clnt_info,
1669 		    ibt_sidr_status_t *sidr_status,
1670 		    ibcm_sidr_rep_msg_t *sidr_repp);
1671 
1672 void		ibcm_proceed_via_taskq(void *targs);
1673 void		ibcm_ud_proceed_via_taskq(void *targs);
1674 
1675 /*
1676  * Builds the reply MAD address based on "incoming mad addr" that is
1677  * supplied to it as an arg.
1678  *	Swaps the source and destination lids in ibmf_addr_info_t
1679  *	Swaps the source and destination gids in ib_grh_t
1680  *
1681  * INPUTS:
1682  *	incoming_cm_mad_addr	- Address information in the incoming MAD
1683  *	reply_cm_mad_addr	- Derived address for the reply MAD
1684  *				  The reply MAD address is derived based
1685  *				  address information of incoming CM MAD
1686  */
1687 void	ibcm_build_reply_mad_addr(ibcm_mad_addr_t *incoming_cm_mad_addr,
1688 	    ibcm_mad_addr_t *reply_cm_mad_addr);
1689 
1690 /*  Posts RC CM MAD using IBMF */
1691 void	ibcm_post_rc_mad(ibcm_state_data_t *statep, ibmf_msg_t *msgp,
1692 	    ibmf_msg_cb_t post_cb, void *args);
1693 
1694 /*  Posts UD CM MAD using IBMF */
1695 void	ibcm_post_ud_mad(ibcm_ud_state_data_t *ud_statep, ibmf_msg_t *msgp,
1696 	    ibmf_msg_cb_t ud_post_cb, void *args);
1697 
1698 /*  Posts CM MAD using IBMF */
1699 ibt_status_t	ibcm_post_mad(ibmf_msg_t *msgp, ibcm_mad_addr_t *cm_mad_addr,
1700 	    ibmf_msg_cb_t post_cb, void *args);
1701 
1702 /* Post REJ MAD */
1703 void	ibcm_post_rej_mad(ibcm_state_data_t *statep, ibt_cm_reason_t reason,
1704 	    int who, void *addl_rej_info, uint8_t arej_info_len);
1705 
1706 /* Post REP MAD */
1707 void	ibcm_post_rep_mad(ibcm_state_data_t *statep);
1708 
1709 /* Post RTU MAD */
1710 ibcm_status_t	ibcm_post_rtu_mad(ibcm_state_data_t *statep);
1711 
1712 /* Post DREQ MAD */
1713 void	ibcm_post_dreq_mad(void *statep);
1714 
1715 /* Post LAP MAD */
1716 void	ibcm_post_lap_mad(ibcm_state_data_t *statep);
1717 
1718 
1719 /*
1720  * Posts CM SIDR MAD using IBMF in blocking mode
1721  *
1722  * INPUTS:
1723  *	ud_statep:	UD statep which is posting the mad
1724  *	cm_mad_addr:	Address information for the MAD to be posted
1725  *	status:		SIDR status
1726  */
1727 void	ibcm_post_sidr_rep_mad(ibcm_ud_state_data_t *ud_statep,
1728 	    ibt_sidr_status_t status);
1729 
1730 /* prototypes to resend RC mad and UD MAD */
1731 void	ibcm_resend_rep_mad(ibcm_state_data_t *statep);
1732 void	ibcm_resend_rtu_mad(ibcm_state_data_t *statep);
1733 void	ibcm_resend_rej_mad(ibcm_state_data_t *statep);
1734 void	ibcm_resend_mra_mad(ibcm_state_data_t *statep);
1735 void	ibcm_resend_srep_mad(ibcm_ud_state_data_t *statep);
1736 
1737 
1738 /* Helper function used in connection abort processing */
1739 void	ibcm_process_abort(ibcm_state_data_t	*statep);
1740 
1741 /*
1742  * Prototypes for CM functions that lookup for a connection state structure
1743  */
1744 
1745 /*
1746  * ibcm_lookup_msg:
1747  *
1748  * Retrieves an existing state structure or creates a new one if none found.
1749  * This function is used during passive side of connection establishment for
1750  * INCOMING REQ/REJ/RTU/MRA
1751  * This function is used during active side of connection establishment for
1752  * INCOMING REP/REJ/MRA
1753  * This function is used during active side of connection establishment for
1754  * an outgoing REQ.
1755  *
1756  * NOTE: IBCM_LOOKP_FAIL is only returned if a new entry wasn't created and
1757  * a match wasn't found.
1758  *
1759  * Arguments are:-
1760  *	ibcm_event_type_t	- what type of message
1761  *				  incoming REQ, REP, REJ, MRA, RTU, DREQ, DREP
1762  *	local_comid		- ONLY *NOT* valid for incoming REQ.
1763  *					needed for others
1764  *	remote_qpn		- Remote CM's QP number
1765  *	remote_hca_guid		- ONLY VALID FOR incoming REQ.
1766  *				  Ignored for others
1767  *	hcap			- HCA entry table pointer
1768  *	statep			- "return"ed state pointer
1769  *
1770  * Return Values:
1771  *	IBCM_LOOKUP_NEW		- new statep allocated
1772  *	IBCM_LOOKUP_EXISTS	- found an existing entry
1773  *	IBCM_LOOKUP_FAIL	- failed to find an entry
1774  *	IBCM_MEMORY_FAILURE	- failed to get memory
1775  *					iff flags != IBT_CHAN_BLOCKING
1776  */
1777 ibcm_status_t	ibcm_lookup_msg(ibcm_event_type_t event_type,
1778 		    ib_com_id_t local_comid, ib_qpn_t remote_qpn,
1779 		    ib_guid_t remote_hca_guid, ibcm_hca_info_t *hcap,
1780 		    ibcm_state_data_t **statep);
1781 
1782 
1783 /*
1784  * Routines for CM SIDR state structure list manipulation
1785  * Wherever possible, the list routines of ibtl are used
1786  * for list manipulation
1787  */
1788 
1789 /*
1790  * Finds an entry based on lid, gid and grh exists fields
1791  * lid:		LID of incoming SIDR REQ
1792  * gid:		GID of incoming SIDR REQ
1793  * grh_exists:		TRUE if GRH exists in the incoming SIDR REQ
1794  * hcap:	CM State HCA entry ptr to search for SIDR state structure
1795  * statep:	Returns a valid state structure, if one exists based
1796  *		on lid, gid and grh_exists fields
1797  * flag:	whether to just look OR to look and add if it doesn't exist.
1798  */
1799 ibcm_status_t		ibcm_find_sidr_entry(ibcm_sidr_srch_t *srch_param,
1800 			    ibcm_hca_info_t *hcap,
1801 			    ibcm_ud_state_data_t **statep,
1802 			    ibcm_lookup_flag_t flag);
1803 
1804 ibcm_ud_state_data_t	*ibcm_add_sidr_entry(ibcm_sidr_srch_t *srch_param,
1805 			    ibcm_hca_info_t *hcap);
1806 
1807 /*
1808  * Deletes a given state structure, from both hca state and passive trees
1809  * If ref cnt is zero, deallocates all buffers and memory of state data
1810  */
1811 void	ibcm_delete_state_data(ibcm_state_data_t *statep);
1812 
1813 /*
1814  * Deallocates all the buffers and memory of state data.
1815  * This function must be called, only when ref_cnt is zero.
1816  */
1817 void	ibcm_dealloc_state_data(ibcm_state_data_t *statep);
1818 
1819 /*
1820  * Deletes a given UD state structure, from SIDR list.
1821  * The routine acquires and releases the SIDR list lock.
1822  */
1823 void	ibcm_delete_ud_state_data(ibcm_ud_state_data_t *statep);
1824 void	ibcm_dealloc_ud_state_data(ibcm_ud_state_data_t *statep);
1825 
1826 /*
1827  * Service ID entry create and lookup functions
1828  */
1829 
1830 /*
1831  * Adds/looks-up an ibcm_svc_info_t entry in the CM's global table.
1832  * This global table is defined in ibcm_impl.c.
1833  *
1834  * svc_info_list_lock must be held for RW_READER by caller of
1835  * ibcm_find_svc_entry().
1836  *
1837  * Arguments are:-
1838  *	sid		- service id
1839  *	num_sids	- Number (Range) of service-ids
1840  *
1841  * Return values:
1842  *	Pointer to ibcm_svc_info_t on success, otherwise NULL.
1843  */
1844 int ibcm_svc_compare(const void *p1, const void *p2);
1845 ibcm_svc_info_t *ibcm_create_svc_entry(ib_svc_id_t sid, int num_sids);
1846 ibcm_svc_info_t *ibcm_find_svc_entry(ib_svc_id_t sid);
1847 
1848 /*
1849  * The following are the function prototypes for various id initialization,
1850  * allocation, free and destroy operations. The cm id allocations are based
1851  * on vmem operations
1852  * The service id's are maintained globally per host
1853  * The com id and req id's are maintained per hca
1854  * To maintain compatibility with intel, service ids are allocated on a 32 bit
1855  * range, though spec has 64 bit range for service id's
1856  */
1857 ibcm_status_t	ibcm_init_ids();
1858 void		ibcm_fini_ids();
1859 
1860 ibcm_status_t	ibcm_init_hca_ids(ibcm_hca_info_t *hcap);
1861 void		ibcm_fini_hca_ids(ibcm_hca_info_t *hcap);
1862 
1863 ibcm_status_t	ibcm_alloc_comid(ibcm_hca_info_t *hcap, ib_com_id_t *comid);
1864 void		ibcm_free_comid(ibcm_hca_info_t *hcap, ib_com_id_t comid);
1865 
1866 ibcm_status_t	ibcm_alloc_reqid(ibcm_hca_info_t *hcap, uint32_t *reqid);
1867 void		ibcm_free_reqid(ibcm_hca_info_t *hcap, uint32_t reqid);
1868 
1869 ib_svc_id_t	ibcm_alloc_local_sids(int num_sids);
1870 void		ibcm_free_local_sids(ib_svc_id_t service_id, int num_sids);
1871 
1872 uint64_t	ibcm_generate_tranid(uint8_t event, uint32_t id,
1873 		    uint32_t cm_tran_priv);
1874 
1875 void		ibcm_decode_tranid(uint64_t tran_id, uint32_t *cm_tran_priv);
1876 
1877 ibcm_status_t	ibcm_ar_init(void);
1878 ibcm_status_t	ibcm_ar_fini(void);
1879 
1880 /*
1881  * These functions are called to do timeout processing from CM connection
1882  * state transitions. (Also for SIDR REQ and SIDR REP processing)
1883  *
1884  * Brief description :
1885  *	If retry count is below max retry value, then post the stored response
1886  *	MAD using IBMF in blocking mode, adjusts remaining retry counters.
1887  *	If retry counter reaches max value, then retry failure handling is
1888  *	done here
1889  *
1890  *	CM will ensure that the state data structure of the associated
1891  *	timeout is valid when this timeout function is called.
1892  *	(See timer_stored_state in ibcm_state_data_t and
1893  *	ud_timer_stored_state in ibcm_ud_state_data_t)
1894  */
1895 void	ibcm_timeout_cb(void *arg);
1896 void	ibcm_sidr_timeout_cb(void *arg);
1897 
1898 /*
1899  * function prototypes for IBMF send completion callbacks on non-blocking
1900  * MAD posts
1901  */
1902 void	ibcm_post_req_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1903 	    void *args);
1904 void	ibcm_post_rep_wait_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1905 	    void *args);	/* MRA Rcvd on active side */
1906 void	ibcm_post_rep_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1907 	    void *args);
1908 void	ibcm_resend_post_rep_complete(ibmf_handle_t ibmf_handle,
1909 	    ibmf_msg_t *msgp, void *args);
1910 void	ibcm_post_mra_rep_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1911 	    void *args);	/* MRA Rcvd on passive side */
1912 void	ibcm_post_rej_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1913 	    void *args);
1914 void	ibcm_post_dreq_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1915 	    void *args);
1916 void	ibcm_post_drep_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1917 	    void *args);
1918 void	ibcm_post_lap_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1919 	    void *args);
1920 void	ibcm_post_apr_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1921 	    void *args);
1922 void	ibcm_post_stored_apr_complete(ibmf_handle_t ibmf_handle,
1923 	    ibmf_msg_t *msgp, void *args);
1924 void	ibcm_post_mra_lap_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1925 	    void *args);	/* MRA Rcvd for LAP on active side */
1926 void	ibcm_post_mra_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1927 	    void *args);	/* for MRA sender */
1928 void	ibcm_post_rtu_complete(ibmf_handle_t ibmf_handle, ibmf_msg_t *msgp,
1929 	    void *args);
1930 
1931 void	ibcm_post_sidr_req_complete(ibmf_handle_t ibmf_handle,
1932 	    ibmf_msg_t *msgp, void *args);
1933 
1934 /*
1935  * ibcm_find_hca_entry:
1936  *	Given a HCA's GUID find out ibcm_hca_info_t entry for that HCA
1937  *	This entry can be then used to access AVL tree/SIDR list etc.
1938  *
1939  *	NOTE: This entry is not removed from the "ibcm_hca_listp".
1940  *	And this function is called with ibcm_hca_list_mutex mutex held.
1941  *
1942  * INPUTS:
1943  *	hca_guid	- HCA's guid
1944  *
1945  * RETURN VALUE:
1946  *	hcap		- if a match is found, else NULL
1947  */
1948 ibcm_hca_info_t	*ibcm_find_hca_entry(ib_guid_t hca_guid);
1949 ibcm_hca_info_t	*ibcm_find_hcap_entry(ib_guid_t hca_guid);
1950 void ibcm_delete_hca_entry(ibcm_hca_info_t *hcap);
1951 
1952 /* Routines that manage the hca's temporary access count */
1953 ibcm_status_t ibcm_inc_hca_acc_cnt(ibcm_hca_info_t *hca);
1954 void ibcm_dec_hca_acc_cnt(ibcm_hca_info_t *hca);
1955 
1956 /* Routines that manage the hca's resource count */
1957 void ibcm_inc_hca_res_cnt(ibcm_hca_info_t *hca);
1958 void ibcm_dec_hca_res_cnt(ibcm_hca_info_t *hca);
1959 
1960 /* Routines that manage the hca's service count */
1961 void ibcm_inc_hca_svc_cnt(ibcm_hca_info_t *hca);
1962 void ibcm_dec_hca_svc_cnt(ibcm_hca_info_t *hca);
1963 
1964 /* Routine to fetch the saa_handle */
1965 ibmf_saa_handle_t ibcm_get_saa_handle(ibcm_hca_info_t *hcap, uint8_t port);
1966 
1967 /* Allow some flow control of RC connection initiations */
1968 void ibcm_flow_inc(void);
1969 void ibcm_flow_dec(hrtime_t delta, char *mad_type);
1970 
1971 /* Allow some flow control of SA requests */
1972 void ibcm_sa_access_enter(void);
1973 void ibcm_sa_access_exit(void);
1974 
1975 /*
1976  * ibcm_cep_to_error_state:
1977  *	Helper function to transition a CEP to ERROR state
1978  *
1979  *	NOTE: This function checks if ch_qp is valid or ch_eec and calls
1980  *	into IBTL to transition the CEP.
1981  *
1982  * INPUTS:
1983  *	statep	- Connection state pointer
1984  *
1985  * RETURN VALUE:
1986  *	IBT_SUCCESS	- if CEP transition succeeded; else error
1987  */
1988 ibt_status_t	ibcm_cep_to_error_state(ibcm_state_data_t *statep);
1989 
1990 /*
1991  * Processes the pending stateps in a linked list. The operations are to
1992  * invoke a cm handler or delete statep
1993  * When the above operations are required on statep from a timeout handler,
1994  * they are linked for later processing by an independent thread
1995  */
1996 void	ibcm_process_tlist();
1997 /* Links RC stateps to an RC timeout processing list */
1998 void	ibcm_add_tlist(ibcm_state_data_t *statep);
1999 
2000 /* Links SIDR/UD stateps to an SIDR/UD timeout processing list */
2001 void	ibcm_add_ud_tlist(ibcm_ud_state_data_t *ud_statep);
2002 
2003 /*
2004  * This call either aborts a pending or completes a in-progress LAP/APR
2005  * operation
2006  */
2007 void	ibcm_sync_lapr_idle(ibcm_state_data_t	*statep);
2008 
2009 void	ibcm_process_rc_recycle(void *recycle_arg);
2010 
2011 /*
2012  * Helper function to handle endianess in case of Service Data.
2013  * Used by ibt_bind_service() and ibt_get_paths().
2014  */
2015 void ibcm_swizzle_from_srv(ibt_srv_data_t *sb_data, uint8_t *service_bytes);
2016 void ibcm_swizzle_to_srv(uint8_t *service_bytes, ibt_srv_data_t *sb_data);
2017 
2018 /* Misc ibcm global variables */
2019 extern char			cmlog[];
2020 extern ibt_clnt_hdl_t		ibcm_ibt_handle;
2021 extern taskq_t			*ibcm_taskq;
2022 extern ibcm_state_handler_t	ibcm_sm_funcs_tbl[];
2023 extern uint8_t			ibcm_timeout_list_flags;
2024 extern ibcm_classportinfo_msg_t	ibcm_clpinfo;
2025 
2026 /* Global lists */
2027 extern avl_tree_t	ibcm_svc_avl_tree;	/* global service id tree */
2028 extern ibcm_state_data_t	*ibcm_timeout_list_hdr, *ibcm_timeout_list_tail;
2029 extern ibcm_ud_state_data_t	*ibcm_ud_timeout_list_hdr,
2030 				*ibcm_ud_timeout_list_tail;
2031 /* Default global retry counts */
2032 extern uint32_t		ibcm_max_retries;
2033 extern uint32_t		ibcm_max_sa_retries;
2034 extern int		ibcm_sa_timeout_delay;	/* in ticks */
2035 
2036 /* Various default global timers */
2037 extern ibt_rnr_nak_time_t	ibcm_default_rnr_nak_time;
2038 
2039 extern clock_t		ibcm_local_processing_time;	/* usecs */
2040 extern clock_t		ibcm_remote_response_time;
2041 extern ib_time_t	ibcm_max_sidr_rep_proctime;
2042 extern ib_time_t	ibcm_max_sidr_rep_store_time;
2043 extern uint32_t		ibcm_adj_btime;
2044 extern uint32_t		ibcm_sw_delay;
2045 
2046 extern ib_time_t	ibcm_max_ib_pkt_lt;
2047 extern ib_time_t	ibcm_max_ib_mad_pkt_lt;
2048 
2049 /* Global locks */
2050 extern kmutex_t		ibcm_svc_info_lock;
2051 extern kmutex_t		ibcm_global_hca_lock;
2052 extern kmutex_t		ibcm_qp_list_lock;
2053 extern kmutex_t		ibcm_timeout_list_lock;
2054 extern kmutex_t		ibcm_recv_mutex;
2055 
2056 /* Global cond variables */
2057 extern kcondvar_t	ibcm_global_hca_cv;
2058 extern kcondvar_t	ibcm_svc_info_cv;
2059 extern kcondvar_t	ibcm_timeout_list_cv;
2060 extern kcondvar_t	ibcm_timeout_thread_done_cv;
2061 
2062 _NOTE(LOCK_ORDER(ibcm_state_data_s::state_mutex ibcm_timeout_list_lock))
2063 _NOTE(LOCK_ORDER(ibcm_ud_state_data_s::ud_state_mutex ibcm_timeout_list_lock))
2064 _NOTE(LOCK_ORDER(ibcm_hca_info_s::hca_state_rwlock
2065     ibcm_state_data_s::state_mutex))
2066 _NOTE(LOCK_ORDER(ibcm_hca_info_s::hca_sidr_list_lock
2067     ibcm_ud_state_data_s::ud_state_mutex))
2068 
2069 _NOTE(READ_ONLY_DATA(ibcm_local_processing_time ibcm_remote_response_time
2070     ibcm_max_sidr_rep_proctime ibcm_max_sidr_rep_store_time ibcm_adj_btime
2071     ibcm_sw_delay ibcm_max_retries ibcm_max_sa_retries))
2072 
2073 /*
2074  * miscellaneous defines for retries, times etc.
2075  */
2076 #define	IBCM_MAX_RETRIES		11	/* Max CM retries for a msg */
2077 #define	IBCM_LOCAL_RESPONSE_TIME	300000	/* Local CM processing time */
2078 						/* in usecs */
2079 #define	IBCM_REMOTE_RESPONSE_TIME	300000	/* Remote CM response time  */
2080 						/* in usecs */
2081 #define	IBCM_MAX_SIDR_PROCESS_TIME	16	/* Time to process SIDR REP */
2082 #define	IBCM_MAX_SIDR_PKT_LIFE_TIME	9	/* Approx pkt lt for UD srver */
2083 
2084 #define	IBCM_MAX_IB_PKT_LT		18	/* 1 second */
2085 #define	IBCM_MAX_IB_MAD_PKT_LT		18	/* 1 second */
2086 
2087 #define	IBCM_MAX_SA_RETRIES		0	/* Max CM retry for SA update */
2088 
2089 /* versions for CM MADs */
2090 #define	IBCM_MAD_BASE_VERSION		1
2091 #define	IBCM_MAD_CLASS_VERSION		2
2092 
2093 /* for Class_Port_Info stuff - see section 16.7.3.1 in Vol1 IB Spec */
2094 #define	IBCM_CPINFO_CAP_RC		0x0200	/* RC is supported */
2095 #define	IBCM_CPINFO_CAP_RD		0x0400	/* RD is supported */
2096 #define	IBCM_CPINFO_CAP_RAW		0x0800	/* Raw Datagrams supported */
2097 #define	IBCM_CPINFO_CAP_UC		0x1000	/* UC supported */
2098 #define	IBCM_CPINFO_CAP_SIDR		0x2000	/* SIDR supported */
2099 
2100 /*
2101  * for debug purposes
2102  */
2103 #ifdef	DEBUG
2104 extern	int ibcm_test_mode;
2105 
2106 void	ibcm_query_qp(ibmf_handle_t ibmf_hdl, ibmf_qp_handle_t ibmf_qp);
2107 void	ibcm_dump_raw_message(uchar_t *);
2108 void	ibcm_dump_srvrec(sa_service_record_t *);
2109 void	ibcm_dump_pathrec(sa_path_record_t *);
2110 void	ibcm_dump_noderec(sa_node_record_t *);
2111 
2112 void	ibcm_query_classport_info(ibt_channel_hdl_t channel);
2113 
2114 #define	IBCM_DUMP_RAW_MSG	ibcm_dump_raw_message
2115 #define	IBCM_DUMP_SERVICE_REC	ibcm_dump_srvrec
2116 #define	IBCM_DUMP_PATH_REC	ibcm_dump_pathrec
2117 #define	IBCM_DUMP_NODE_REC	ibcm_dump_noderec
2118 #else
2119 #define	IBCM_DUMP_RAW_MSG	0 &&
2120 #define	IBCM_DUMP_SERVICE_REC	0 &&
2121 #define	IBCM_DUMP_PATH_REC	0 &&
2122 #define	IBCM_DUMP_NODE_REC	0 &&
2123 #endif
2124 
2125 ibt_status_t ibcm_ibmf_analyze_error(int ibmf_status);
2126 
2127 ibt_status_t ibcm_contact_sa_access(ibmf_saa_handle_t saa_handle,
2128     ibmf_saa_access_args_t *access_args, size_t *length, void **results_p);
2129 
2130 void ibcm_path_cache_init(void);
2131 void ibcm_path_cache_fini(void);
2132 void ibcm_path_cache_purge(void);
2133 
2134 #ifdef	__cplusplus
2135 }
2136 #endif
2137 
2138 
2139 #endif /* _SYS_IB_MGT_IBCM_IBCM_IMPL_H */
2140